fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 static struct kmem_cache *btrfs_delalloc_work_cachep;
  75 struct kmem_cache *btrfs_trans_handle_cachep;
  76 struct kmem_cache *btrfs_transaction_cachep;
  77 struct kmem_cache *btrfs_path_cachep;
  78 struct kmem_cache *btrfs_free_space_cachep;
  79
  80 #define S_SHIFT 12
  81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  89 };
  90
  91 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  92 static int btrfs_truncate(struct inode *inode);
  93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  94 static noinline int cow_file_range(struct inode *inode,
  95                                    struct page *locked_page,
  96                                    u64 start, u64 end, int *page_started,
  97                                    unsigned long *nr_written, int unlock);
  98
  99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 100                                      struct inode *inode,  struct inode *dir,
 101                                      const struct qstr *qstr)
 102 {
 103         int err;
 104
 105         err = btrfs_init_acl(trans, inode, dir);
 106         if (!err)
 107                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 108         return err;
 109 }
 110
 111 /*
 112  * this does all the hard work for inserting an inline extent into
 113  * the btree.  The caller should have done a btrfs_drop_extents so that
 114  * no overlapping inline items exist in the btree
 115  */
 116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 117                                 struct btrfs_root *root, struct inode *inode,
 118                                 u64 start, size_t size, size_t compressed_size,
 119                                 int compress_type,
 120                                 struct page **compressed_pages)
 121 {
 122         struct btrfs_key key;
 123         struct btrfs_path *path;
 124         struct extent_buffer *leaf;
 125         struct page *page = NULL;
 126         char *kaddr;
 127         unsigned long ptr;
 128         struct btrfs_file_extent_item *ei;
 129         int err = 0;
 130         int ret;
 131         size_t cur_size = size;
 132         size_t datasize;
 133         unsigned long offset;
 134
 135         if (compressed_size && compressed_pages)
 136                 cur_size = compressed_size;
 137
 138         path = btrfs_alloc_path();
 139         if (!path)
 140                 return -ENOMEM;
 141
 142         path->leave_spinning = 1;
 143
 144         key.objectid = btrfs_ino(inode);
 145         key.offset = start;
 146         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 147         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 148
 149         inode_add_bytes(inode, size);
 150         ret = btrfs_insert_empty_item(trans, root, path, &key,
 151                                       datasize);
 152         if (ret) {
 153                 err = ret;
 154                 goto fail;
 155         }
 156         leaf = path->nodes[0];
 157         ei = btrfs_item_ptr(leaf, path->slots[0],
 158                             struct btrfs_file_extent_item);
 159         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 160         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 161         btrfs_set_file_extent_encryption(leaf, ei, 0);
 162         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 163         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 164         ptr = btrfs_file_extent_inline_start(ei);
 165
 166         if (compress_type != BTRFS_COMPRESS_NONE) {
 167                 struct page *cpage;
 168                 int i = 0;
 169                 while (compressed_size > 0) {
 170                         cpage = compressed_pages[i];
 171                         cur_size = min_t(unsigned long, compressed_size,
 172                                        PAGE_CACHE_SIZE);
 173
 174                         kaddr = kmap_atomic(cpage);
 175                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 176                         kunmap_atomic(kaddr);
 177
 178                         i++;
 179                         ptr += cur_size;
 180                         compressed_size -= cur_size;
 181                 }
 182                 btrfs_set_file_extent_compression(leaf, ei,
 183                                                   compress_type);
 184         } else {
 185                 page = find_get_page(inode->i_mapping,
 186                                      start >> PAGE_CACHE_SHIFT);
 187                 btrfs_set_file_extent_compression(leaf, ei, 0);
 188                 kaddr = kmap_atomic(page);
 189                 offset = start & (PAGE_CACHE_SIZE - 1);
 190                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 191                 kunmap_atomic(kaddr);
 192                 page_cache_release(page);
 193         }
 194         btrfs_mark_buffer_dirty(leaf);
 195         btrfs_free_path(path);
 196
 197         /*
 198          * we're an inline extent, so nobody can
 199          * extend the file past i_size without locking
 200          * a page we already have locked.
 201          *
 202          * We must do any isize and inode updates
 203          * before we unlock the pages.  Otherwise we
 204          * could end up racing with unlink.
 205          */
 206         BTRFS_I(inode)->disk_i_size = inode->i_size;
 207         ret = btrfs_update_inode(trans, root, inode);
 208
 209         return ret;
 210 fail:
 211         btrfs_free_path(path);
 212         return err;
 213 }
 214
 215
 216 /*
 217  * conditionally insert an inline extent into the file.  This
 218  * does the checks required to make sure the data is small enough
 219  * to fit as an inline extent.
 220  */
 221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 222                                  struct btrfs_root *root,
 223                                  struct inode *inode, u64 start, u64 end,
 224                                  size_t compressed_size, int compress_type,
 225                                  struct page **compressed_pages)
 226 {
 227         u64 isize = i_size_read(inode);
 228         u64 actual_end = min(end + 1, isize);
 229         u64 inline_len = actual_end - start;
 230         u64 aligned_end = (end + root->sectorsize - 1) &
 231                         ~((u64)root->sectorsize - 1);
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 249         if (ret)
 250                 return ret;
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         if (ret && ret != -ENOSPC) {
 258                 btrfs_abort_transaction(trans, root, ret);
 259                 return ret;
 260         } else if (ret == -ENOSPC) {
 261                 return 1;
 262         }
 263
 264         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 265         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 266         return 0;
 267 }
 268
 269 struct async_extent {
 270         u64 start;
 271         u64 ram_size;
 272         u64 compressed_size;
 273         struct page **pages;
 274         unsigned long nr_pages;
 275         int compress_type;
 276         struct list_head list;
 277 };
 278
 279 struct async_cow {
 280         struct inode *inode;
 281         struct btrfs_root *root;
 282         struct page *locked_page;
 283         u64 start;
 284         u64 end;
 285         struct list_head extents;
 286         struct btrfs_work work;
 287 };
 288
 289 static noinline int add_async_extent(struct async_cow *cow,
 290                                      u64 start, u64 ram_size,
 291                                      u64 compressed_size,
 292                                      struct page **pages,
 293                                      unsigned long nr_pages,
 294                                      int compress_type)
 295 {
 296         struct async_extent *async_extent;
 297
 298         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 299         BUG_ON(!async_extent); /* -ENOMEM */
 300         async_extent->start = start;
 301         async_extent->ram_size = ram_size;
 302         async_extent->compressed_size = compressed_size;
 303         async_extent->pages = pages;
 304         async_extent->nr_pages = nr_pages;
 305         async_extent->compress_type = compress_type;
 306         list_add_tail(&async_extent->list, &cow->extents);
 307         return 0;
 308 }
 309
 310 /*
 311  * we create compressed extents in two phases.  The first
 312  * phase compresses a range of pages that have already been
 313  * locked (both pages and state bits are locked).
 314  *
 315  * This is done inside an ordered work queue, and the compression
 316  * is spread across many cpus.  The actual IO submission is step
 317  * two, and the ordered work queue takes care of making sure that
 318  * happens in the same order things were put onto the queue by
 319  * writepages and friends.
 320  *
 321  * If this code finds it can't get good compression, it puts an
 322  * entry onto the work queue to write the uncompressed bytes.  This
 323  * makes sure that both compressed inodes and uncompressed inodes
 324  * are written in the same order that the flusher thread sent them
 325  * down.
 326  */
 327 static noinline int compress_file_range(struct inode *inode,
 328                                         struct page *locked_page,
 329                                         u64 start, u64 end,
 330                                         struct async_cow *async_cow,
 331                                         int *num_added)
 332 {
 333         struct btrfs_root *root = BTRFS_I(inode)->root;
 334         struct btrfs_trans_handle *trans;
 335         u64 num_bytes;
 336         u64 blocksize = root->sectorsize;
 337         u64 actual_end;
 338         u64 isize = i_size_read(inode);
 339         int ret = 0;
 340         struct page **pages = NULL;
 341         unsigned long nr_pages;
 342         unsigned long nr_pages_ret = 0;
 343         unsigned long total_compressed = 0;
 344         unsigned long total_in = 0;
 345         unsigned long max_compressed = 128 * 1024;
 346         unsigned long max_uncompressed = 128 * 1024;
 347         int i;
 348         int will_compress;
 349         int compress_type = root->fs_info->compress_type;
 350
 351         /* if this is a small write inside eof, kick off a defrag */
 352         if ((end - start + 1) < 16 * 1024 &&
 353             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 354                 btrfs_add_inode_defrag(NULL, inode);
 355
 356         actual_end = min_t(u64, isize, end + 1);
 357 again:
 358         will_compress = 0;
 359         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 360         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 361
 362         /*
 363          * we don't want to send crud past the end of i_size through
 364          * compression, that's just a waste of CPU time.  So, if the
 365          * end of the file is before the start of our current
 366          * requested range of bytes, we bail out to the uncompressed
 367          * cleanup code that can deal with all of this.
 368          *
 369          * It isn't really the fastest way to fix things, but this is a
 370          * very uncommon corner.
 371          */
 372         if (actual_end <= start)
 373                 goto cleanup_and_bail_uncompressed;
 374
 375         total_compressed = actual_end - start;
 376
 377         /* we want to make sure that amount of ram required to uncompress
 378          * an extent is reasonable, so we limit the total size in ram
 379          * of a compressed extent to 128k.  This is a crucial number
 380          * because it also controls how easily we can spread reads across
 381          * cpus for decompression.
 382          *
 383          * We also want to make sure the amount of IO required to do
 384          * a random read is reasonably small, so we limit the size of
 385          * a compressed extent to 128k.
 386          */
 387         total_compressed = min(total_compressed, max_uncompressed);
 388         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 389         num_bytes = max(blocksize,  num_bytes);
 390         total_in = 0;
 391         ret = 0;
 392
 393         /*
 394          * we do compression for mount -o compress and when the
 395          * inode has not been flagged as nocompress.  This flag can
 396          * change at any time if we discover bad compression ratios.
 397          */
 398         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 399             (btrfs_test_opt(root, COMPRESS) ||
 400              (BTRFS_I(inode)->force_compress) ||
 401              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 402                 WARN_ON(pages);
 403                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 404                 if (!pages) {
 405                         /* just bail out to the uncompressed code */
 406                         goto cont;
 407                 }
 408
 409                 if (BTRFS_I(inode)->force_compress)
 410                         compress_type = BTRFS_I(inode)->force_compress;
 411
 412                 ret = btrfs_compress_pages(compress_type,
 413                                            inode->i_mapping, start,
 414                                            total_compressed, pages,
 415                                            nr_pages, &nr_pages_ret,
 416                                            &total_in,
 417                                            &total_compressed,
 418                                            max_compressed);
 419
 420                 if (!ret) {
 421                         unsigned long offset = total_compressed &
 422                                 (PAGE_CACHE_SIZE - 1);
 423                         struct page *page = pages[nr_pages_ret - 1];
 424                         char *kaddr;
 425
 426                         /* zero the tail end of the last page, we might be
 427                          * sending it down to disk
 428                          */
 429                         if (offset) {
 430                                 kaddr = kmap_atomic(page);
 431                                 memset(kaddr + offset, 0,
 432                                        PAGE_CACHE_SIZE - offset);
 433                                 kunmap_atomic(kaddr);
 434                         }
 435                         will_compress = 1;
 436                 }
 437         }
 438 cont:
 439         if (start == 0) {
 440                 trans = btrfs_join_transaction(root);
 441                 if (IS_ERR(trans)) {
 442                         ret = PTR_ERR(trans);
 443                         trans = NULL;
 444                         goto cleanup_and_out;
 445                 }
 446                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 447
 448                 /* lets try to make an inline extent */
 449                 if (ret || total_in < (actual_end - start)) {
 450                         /* we didn't compress the entire range, try
 451                          * to make an uncompressed inline extent.
 452                          */
 453                         ret = cow_file_range_inline(trans, root, inode,
 454                                                     start, end, 0, 0, NULL);
 455                 } else {
 456                         /* try making a compressed inline extent */
 457                         ret = cow_file_range_inline(trans, root, inode,
 458                                                     start, end,
 459                                                     total_compressed,
 460                                                     compress_type, pages);
 461                 }
 462                 if (ret <= 0) {
 463                         /*
 464                          * inline extent creation worked or returned error,
 465                          * we don't need to create any more async work items.
 466                          * Unlock and free up our temp pages.
 467                          */
 468                         extent_clear_unlock_delalloc(inode,
 469                              &BTRFS_I(inode)->io_tree,
 470                              start, end, NULL,
 471                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 472                              EXTENT_CLEAR_DELALLOC |
 473                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 474
 475                         btrfs_end_transaction(trans, root);
 476                         goto free_pages_out;
 477                 }
 478                 btrfs_end_transaction(trans, root);
 479         }
 480
 481         if (will_compress) {
 482                 /*
 483                  * we aren't doing an inline extent round the compressed size
 484                  * up to a block size boundary so the allocator does sane
 485                  * things
 486                  */
 487                 total_compressed = (total_compressed + blocksize - 1) &
 488                         ~(blocksize - 1);
 489
 490                 /*
 491                  * one last check to make sure the compression is really a
 492                  * win, compare the page count read with the blocks on disk
 493                  */
 494                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 495                         ~(PAGE_CACHE_SIZE - 1);
 496                 if (total_compressed >= total_in) {
 497                         will_compress = 0;
 498                 } else {
 499                         num_bytes = total_in;
 500                 }
 501         }
 502         if (!will_compress && pages) {
 503                 /*
 504                  * the compression code ran but failed to make things smaller,
 505                  * free any pages it allocated and our page pointer array
 506                  */
 507                 for (i = 0; i < nr_pages_ret; i++) {
 508                         WARN_ON(pages[i]->mapping);
 509                         page_cache_release(pages[i]);
 510                 }
 511                 kfree(pages);
 512                 pages = NULL;
 513                 total_compressed = 0;
 514                 nr_pages_ret = 0;
 515
 516                 /* flag the file so we don't compress in the future */
 517                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 518                     !(BTRFS_I(inode)->force_compress)) {
 519                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 520                 }
 521         }
 522         if (will_compress) {
 523                 *num_added += 1;
 524
 525                 /* the async work queues will take care of doing actual
 526                  * allocation on disk for these compressed pages,
 527                  * and will submit them to the elevator.
 528                  */
 529                 add_async_extent(async_cow, start, num_bytes,
 530                                  total_compressed, pages, nr_pages_ret,
 531                                  compress_type);
 532
 533                 if (start + num_bytes < end) {
 534                         start += num_bytes;
 535                         pages = NULL;
 536                         cond_resched();
 537                         goto again;
 538                 }
 539         } else {
 540 cleanup_and_bail_uncompressed:
 541                 /*
 542                  * No compression, but we still need to write the pages in
 543                  * the file we've been given so far.  redirty the locked
 544                  * page if it corresponds to our extent and set things up
 545                  * for the async work queue to run cow_file_range to do
 546                  * the normal delalloc dance
 547                  */
 548                 if (page_offset(locked_page) >= start &&
 549                     page_offset(locked_page) <= end) {
 550                         __set_page_dirty_nobuffers(locked_page);
 551                         /* unlocked later on in the async handlers */
 552                 }
 553                 add_async_extent(async_cow, start, end - start + 1,
 554                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 555                 *num_added += 1;
 556         }
 557
 558 out:
 559         return ret;
 560
 561 free_pages_out:
 562         for (i = 0; i < nr_pages_ret; i++) {
 563                 WARN_ON(pages[i]->mapping);
 564                 page_cache_release(pages[i]);
 565         }
 566         kfree(pages);
 567
 568         goto out;
 569
 570 cleanup_and_out:
 571         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 572                                      start, end, NULL,
 573                                      EXTENT_CLEAR_UNLOCK_PAGE |
 574                                      EXTENT_CLEAR_DIRTY |
 575                                      EXTENT_CLEAR_DELALLOC |
 576                                      EXTENT_SET_WRITEBACK |
 577                                      EXTENT_END_WRITEBACK);
 578         if (!trans || IS_ERR(trans))
 579                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 580         else
 581                 btrfs_abort_transaction(trans, root, ret);
 582         goto free_pages_out;
 583 }
 584
 585 /*
 586  * phase two of compressed writeback.  This is the ordered portion
 587  * of the code, which only gets called in the order the work was
 588  * queued.  We walk all the async extents created by compress_file_range
 589  * and send them down to the disk.
 590  */
 591 static noinline int submit_compressed_extents(struct inode *inode,
 592                                               struct async_cow *async_cow)
 593 {
 594         struct async_extent *async_extent;
 595         u64 alloc_hint = 0;
 596         struct btrfs_trans_handle *trans;
 597         struct btrfs_key ins;
 598         struct extent_map *em;
 599         struct btrfs_root *root = BTRFS_I(inode)->root;
 600         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 601         struct extent_io_tree *io_tree;
 602         int ret = 0;
 603
 604         if (list_empty(&async_cow->extents))
 605                 return 0;
 606
 607
 608         while (!list_empty(&async_cow->extents)) {
 609                 async_extent = list_entry(async_cow->extents.next,
 610                                           struct async_extent, list);
 611                 list_del(&async_extent->list);
 612
 613                 io_tree = &BTRFS_I(inode)->io_tree;
 614
 615 retry:
 616                 /* did the compression code fall back to uncompressed IO? */
 617                 if (!async_extent->pages) {
 618                         int page_started = 0;
 619                         unsigned long nr_written = 0;
 620
 621                         lock_extent(io_tree, async_extent->start,
 622                                          async_extent->start +
 623                                          async_extent->ram_size - 1);
 624
 625                         /* allocate blocks */
 626                         ret = cow_file_range(inode, async_cow->locked_page,
 627                                              async_extent->start,
 628                                              async_extent->start +
 629                                              async_extent->ram_size - 1,
 630                                              &page_started, &nr_written, 0);
 631
 632                         /* JDM XXX */
 633
 634                         /*
 635                          * if page_started, cow_file_range inserted an
 636                          * inline extent and took care of all the unlocking
 637                          * and IO for us.  Otherwise, we need to submit
 638                          * all those pages down to the drive.
 639                          */
 640                         if (!page_started && !ret)
 641                                 extent_write_locked_range(io_tree,
 642                                                   inode, async_extent->start,
 643                                                   async_extent->start +
 644                                                   async_extent->ram_size - 1,
 645                                                   btrfs_get_extent,
 646                                                   WB_SYNC_ALL);
 647                         kfree(async_extent);
 648                         cond_resched();
 649                         continue;
 650                 }
 651
 652                 lock_extent(io_tree, async_extent->start,
 653                             async_extent->start + async_extent->ram_size - 1);
 654
 655                 trans = btrfs_join_transaction(root);
 656                 if (IS_ERR(trans)) {
 657                         ret = PTR_ERR(trans);
 658                 } else {
 659                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 660                         ret = btrfs_reserve_extent(trans, root,
 661                                            async_extent->compressed_size,
 662                                            async_extent->compressed_size,
 663                                            0, alloc_hint, &ins, 1);
 664                         if (ret && ret != -ENOSPC)
 665                                 btrfs_abort_transaction(trans, root, ret);
 666                         btrfs_end_transaction(trans, root);
 667                 }
 668
 669                 if (ret) {
 670                         int i;
 671                         for (i = 0; i < async_extent->nr_pages; i++) {
 672                                 WARN_ON(async_extent->pages[i]->mapping);
 673                                 page_cache_release(async_extent->pages[i]);
 674                         }
 675                         kfree(async_extent->pages);
 676                         async_extent->nr_pages = 0;
 677                         async_extent->pages = NULL;
 678                         unlock_extent(io_tree, async_extent->start,
 679                                       async_extent->start +
 680                                       async_extent->ram_size - 1);
 681                         if (ret == -ENOSPC)
 682                                 goto retry;
 683                         goto out_free; /* JDM: Requeue? */
 684                 }
 685
 686                 /*
 687                  * here we're doing allocation and writeback of the
 688                  * compressed pages
 689                  */
 690                 btrfs_drop_extent_cache(inode, async_extent->start,
 691                                         async_extent->start +
 692                                         async_extent->ram_size - 1, 0);
 693
 694                 em = alloc_extent_map();
 695                 BUG_ON(!em); /* -ENOMEM */
 696                 em->start = async_extent->start;
 697                 em->len = async_extent->ram_size;
 698                 em->orig_start = em->start;
 699
 700                 em->block_start = ins.objectid;
 701                 em->block_len = ins.offset;
 702                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 703                 em->compress_type = async_extent->compress_type;
 704                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 705                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 706
 707                 while (1) {
 708                         write_lock(&em_tree->lock);
 709                         ret = add_extent_mapping(em_tree, em);
 710                         write_unlock(&em_tree->lock);
 711                         if (ret != -EEXIST) {
 712                                 free_extent_map(em);
 713                                 break;
 714                         }
 715                         btrfs_drop_extent_cache(inode, async_extent->start,
 716                                                 async_extent->start +
 717                                                 async_extent->ram_size - 1, 0);
 718                 }
 719
 720                 ret = btrfs_add_ordered_extent_compress(inode,
 721                                                 async_extent->start,
 722                                                 ins.objectid,
 723                                                 async_extent->ram_size,
 724                                                 ins.offset,
 725                                                 BTRFS_ORDERED_COMPRESSED,
 726                                                 async_extent->compress_type);
 727                 BUG_ON(ret); /* -ENOMEM */
 728
 729                 /*
 730                  * clear dirty, set writeback and unlock the pages.
 731                  */
 732                 extent_clear_unlock_delalloc(inode,
 733                                 &BTRFS_I(inode)->io_tree,
 734                                 async_extent->start,
 735                                 async_extent->start +
 736                                 async_extent->ram_size - 1,
 737                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 738                                 EXTENT_CLEAR_UNLOCK |
 739                                 EXTENT_CLEAR_DELALLOC |
 740                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 741
 742                 ret = btrfs_submit_compressed_write(inode,
 743                                     async_extent->start,
 744                                     async_extent->ram_size,
 745                                     ins.objectid,
 746                                     ins.offset, async_extent->pages,
 747                                     async_extent->nr_pages);
 748
 749                 BUG_ON(ret); /* -ENOMEM */
 750                 alloc_hint = ins.objectid + ins.offset;
 751                 kfree(async_extent);
 752                 cond_resched();
 753         }
 754         ret = 0;
 755 out:
 756         return ret;
 757 out_free:
 758         kfree(async_extent);
 759         goto out;
 760 }
 761
 762 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 763                                       u64 num_bytes)
 764 {
 765         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 766         struct extent_map *em;
 767         u64 alloc_hint = 0;
 768
 769         read_lock(&em_tree->lock);
 770         em = search_extent_mapping(em_tree, start, num_bytes);
 771         if (em) {
 772                 /*
 773                  * if block start isn't an actual block number then find the
 774                  * first block in this inode and use that as a hint.  If that
 775                  * block is also bogus then just don't worry about it.
 776                  */
 777                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 778                         free_extent_map(em);
 779                         em = search_extent_mapping(em_tree, 0, 0);
 780                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 781                                 alloc_hint = em->block_start;
 782                         if (em)
 783                                 free_extent_map(em);
 784                 } else {
 785                         alloc_hint = em->block_start;
 786                         free_extent_map(em);
 787                 }
 788         }
 789         read_unlock(&em_tree->lock);
 790
 791         return alloc_hint;
 792 }
 793
 794 /*
 795  * when extent_io.c finds a delayed allocation range in the file,
 796  * the call backs end up in this code.  The basic idea is to
 797  * allocate extents on disk for the range, and create ordered data structs
 798  * in ram to track those extents.
 799  *
 800  * locked_page is the page that writepage had locked already.  We use
 801  * it to make sure we don't do extra locks or unlocks.
 802  *
 803  * *page_started is set to one if we unlock locked_page and do everything
 804  * required to start IO on it.  It may be clean and already done with
 805  * IO when we return.
 806  */
 807 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 808                                      struct inode *inode,
 809                                      struct btrfs_root *root,
 810                                      struct page *locked_page,
 811                                      u64 start, u64 end, int *page_started,
 812                                      unsigned long *nr_written,
 813                                      int unlock)
 814 {
 815         u64 alloc_hint = 0;
 816         u64 num_bytes;
 817         unsigned long ram_size;
 818         u64 disk_num_bytes;
 819         u64 cur_alloc_size;
 820         u64 blocksize = root->sectorsize;
 821         struct btrfs_key ins;
 822         struct extent_map *em;
 823         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 824         int ret = 0;
 825
 826         BUG_ON(btrfs_is_free_space_inode(inode));
 827
 828         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 829         num_bytes = max(blocksize,  num_bytes);
 830         disk_num_bytes = num_bytes;
 831
 832         /* if this is a small write inside eof, kick off defrag */
 833         if (num_bytes < 64 * 1024 &&
 834             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 835                 btrfs_add_inode_defrag(trans, inode);
 836
 837         if (start == 0) {
 838                 /* lets try to make an inline extent */
 839                 ret = cow_file_range_inline(trans, root, inode,
 840                                             start, end, 0, 0, NULL);
 841                 if (ret == 0) {
 842                         extent_clear_unlock_delalloc(inode,
 843                                      &BTRFS_I(inode)->io_tree,
 844                                      start, end, NULL,
 845                                      EXTENT_CLEAR_UNLOCK_PAGE |
 846                                      EXTENT_CLEAR_UNLOCK |
 847                                      EXTENT_CLEAR_DELALLOC |
 848                                      EXTENT_CLEAR_DIRTY |
 849                                      EXTENT_SET_WRITEBACK |
 850                                      EXTENT_END_WRITEBACK);
 851
 852                         *nr_written = *nr_written +
 853                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 854                         *page_started = 1;
 855                         goto out;
 856                 } else if (ret < 0) {
 857                         btrfs_abort_transaction(trans, root, ret);
 858                         goto out_unlock;
 859                 }
 860         }
 861
 862         BUG_ON(disk_num_bytes >
 863                btrfs_super_total_bytes(root->fs_info->super_copy));
 864
 865         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 866         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 867
 868         while (disk_num_bytes > 0) {
 869                 unsigned long op;
 870
 871                 cur_alloc_size = disk_num_bytes;
 872                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 873                                            root->sectorsize, 0, alloc_hint,
 874                                            &ins, 1);
 875                 if (ret < 0) {
 876                         btrfs_abort_transaction(trans, root, ret);
 877                         goto out_unlock;
 878                 }
 879
 880                 em = alloc_extent_map();
 881                 BUG_ON(!em); /* -ENOMEM */
 882                 em->start = start;
 883                 em->orig_start = em->start;
 884                 ram_size = ins.offset;
 885                 em->len = ins.offset;
 886
 887                 em->block_start = ins.objectid;
 888                 em->block_len = ins.offset;
 889                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 890                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 891
 892                 while (1) {
 893                         write_lock(&em_tree->lock);
 894                         ret = add_extent_mapping(em_tree, em);
 895                         write_unlock(&em_tree->lock);
 896                         if (ret != -EEXIST) {
 897                                 free_extent_map(em);
 898                                 break;
 899                         }
 900                         btrfs_drop_extent_cache(inode, start,
 901                                                 start + ram_size - 1, 0);
 902                 }
 903
 904                 cur_alloc_size = ins.offset;
 905                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 906                                                ram_size, cur_alloc_size, 0);
 907                 BUG_ON(ret); /* -ENOMEM */
 908
 909                 if (root->root_key.objectid ==
 910                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 911                         ret = btrfs_reloc_clone_csums(inode, start,
 912                                                       cur_alloc_size);
 913                         if (ret) {
 914                                 btrfs_abort_transaction(trans, root, ret);
 915                                 goto out_unlock;
 916                         }
 917                 }
 918
 919                 if (disk_num_bytes < cur_alloc_size)
 920                         break;
 921
 922                 /* we're not doing compressed IO, don't unlock the first
 923                  * page (which the caller expects to stay locked), don't
 924                  * clear any dirty bits and don't set any writeback bits
 925                  *
 926                  * Do set the Private2 bit so we know this page was properly
 927                  * setup for writepage
 928                  */
 929                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 930                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 931                         EXTENT_SET_PRIVATE2;
 932
 933                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 934                                              start, start + ram_size - 1,
 935                                              locked_page, op);
 936                 disk_num_bytes -= cur_alloc_size;
 937                 num_bytes -= cur_alloc_size;
 938                 alloc_hint = ins.objectid + ins.offset;
 939                 start += cur_alloc_size;
 940         }
 941 out:
 942         return ret;
 943
 944 out_unlock:
 945         extent_clear_unlock_delalloc(inode,
 946                      &BTRFS_I(inode)->io_tree,
 947                      start, end, locked_page,
 948                      EXTENT_CLEAR_UNLOCK_PAGE |
 949                      EXTENT_CLEAR_UNLOCK |
 950                      EXTENT_CLEAR_DELALLOC |
 951                      EXTENT_CLEAR_DIRTY |
 952                      EXTENT_SET_WRITEBACK |
 953                      EXTENT_END_WRITEBACK);
 954
 955         goto out;
 956 }
 957
 958 static noinline int cow_file_range(struct inode *inode,
 959                                    struct page *locked_page,
 960                                    u64 start, u64 end, int *page_started,
 961                                    unsigned long *nr_written,
 962                                    int unlock)
 963 {
 964         struct btrfs_trans_handle *trans;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         int ret;
 967
 968         trans = btrfs_join_transaction(root);
 969         if (IS_ERR(trans)) {
 970                 extent_clear_unlock_delalloc(inode,
 971                              &BTRFS_I(inode)->io_tree,
 972                              start, end, locked_page,
 973                              EXTENT_CLEAR_UNLOCK_PAGE |
 974                              EXTENT_CLEAR_UNLOCK |
 975                              EXTENT_CLEAR_DELALLOC |
 976                              EXTENT_CLEAR_DIRTY |
 977                              EXTENT_SET_WRITEBACK |
 978                              EXTENT_END_WRITEBACK);
 979                 return PTR_ERR(trans);
 980         }
 981         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 982
 983         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
 984                                page_started, nr_written, unlock);
 985
 986         btrfs_end_transaction(trans, root);
 987
 988         return ret;
 989 }
 990
 991 /*
 992  * work queue call back to started compression on a file and pages
 993  */
 994 static noinline void async_cow_start(struct btrfs_work *work)
 995 {
 996         struct async_cow *async_cow;
 997         int num_added = 0;
 998         async_cow = container_of(work, struct async_cow, work);
 999
1000         compress_file_range(async_cow->inode, async_cow->locked_page,
1001                             async_cow->start, async_cow->end, async_cow,
1002                             &num_added);
1003         if (num_added == 0) {
1004                 btrfs_add_delayed_iput(async_cow->inode);
1005                 async_cow->inode = NULL;
1006         }
1007 }
1008
1009 /*
1010  * work queue call back to submit previously compressed pages
1011  */
1012 static noinline void async_cow_submit(struct btrfs_work *work)
1013 {
1014         struct async_cow *async_cow;
1015         struct btrfs_root *root;
1016         unsigned long nr_pages;
1017
1018         async_cow = container_of(work, struct async_cow, work);
1019
1020         root = async_cow->root;
1021         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1022                 PAGE_CACHE_SHIFT;
1023
1024         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1025             5 * 1024 * 1024 &&
1026             waitqueue_active(&root->fs_info->async_submit_wait))
1027                 wake_up(&root->fs_info->async_submit_wait);
1028
1029         if (async_cow->inode)
1030                 submit_compressed_extents(async_cow->inode, async_cow);
1031 }
1032
1033 static noinline void async_cow_free(struct btrfs_work *work)
1034 {
1035         struct async_cow *async_cow;
1036         async_cow = container_of(work, struct async_cow, work);
1037         if (async_cow->inode)
1038                 btrfs_add_delayed_iput(async_cow->inode);
1039         kfree(async_cow);
1040 }
1041
1042 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1043                                 u64 start, u64 end, int *page_started,
1044                                 unsigned long *nr_written)
1045 {
1046         struct async_cow *async_cow;
1047         struct btrfs_root *root = BTRFS_I(inode)->root;
1048         unsigned long nr_pages;
1049         u64 cur_end;
1050         int limit = 10 * 1024 * 1024;
1051
1052         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1053                          1, 0, NULL, GFP_NOFS);
1054         while (start < end) {
1055                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1056                 BUG_ON(!async_cow); /* -ENOMEM */
1057                 async_cow->inode = igrab(inode);
1058                 async_cow->root = root;
1059                 async_cow->locked_page = locked_page;
1060                 async_cow->start = start;
1061
1062                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1063                         cur_end = end;
1064                 else
1065                         cur_end = min(end, start + 512 * 1024 - 1);
1066
1067                 async_cow->end = cur_end;
1068                 INIT_LIST_HEAD(&async_cow->extents);
1069
1070                 async_cow->work.func = async_cow_start;
1071                 async_cow->work.ordered_func = async_cow_submit;
1072                 async_cow->work.ordered_free = async_cow_free;
1073                 async_cow->work.flags = 0;
1074
1075                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1076                         PAGE_CACHE_SHIFT;
1077                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1078
1079                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1080                                    &async_cow->work);
1081
1082                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1083                         wait_event(root->fs_info->async_submit_wait,
1084                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1085                             limit));
1086                 }
1087
1088                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1089                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1090                         wait_event(root->fs_info->async_submit_wait,
1091                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1092                            0));
1093                 }
1094
1095                 *nr_written += nr_pages;
1096                 start = cur_end + 1;
1097         }
1098         *page_started = 1;
1099         return 0;
1100 }
1101
1102 static noinline int csum_exist_in_range(struct btrfs_root *root,
1103                                         u64 bytenr, u64 num_bytes)
1104 {
1105         int ret;
1106         struct btrfs_ordered_sum *sums;
1107         LIST_HEAD(list);
1108
1109         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1110                                        bytenr + num_bytes - 1, &list, 0);
1111         if (ret == 0 && list_empty(&list))
1112                 return 0;
1113
1114         while (!list_empty(&list)) {
1115                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1116                 list_del(&sums->list);
1117                 kfree(sums);
1118         }
1119         return 1;
1120 }
1121
1122 /*
1123  * when nowcow writeback call back.  This checks for snapshots or COW copies
1124  * of the extents that exist in the file, and COWs the file as required.
1125  *
1126  * If no cow copies or snapshots exist, we write directly to the existing
1127  * blocks on disk
1128  */
1129 static noinline int run_delalloc_nocow(struct inode *inode,
1130                                        struct page *locked_page,
1131                               u64 start, u64 end, int *page_started, int force,
1132                               unsigned long *nr_written)
1133 {
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135         struct btrfs_trans_handle *trans;
1136         struct extent_buffer *leaf;
1137         struct btrfs_path *path;
1138         struct btrfs_file_extent_item *fi;
1139         struct btrfs_key found_key;
1140         u64 cow_start;
1141         u64 cur_offset;
1142         u64 extent_end;
1143         u64 extent_offset;
1144         u64 disk_bytenr;
1145         u64 num_bytes;
1146         int extent_type;
1147         int ret, err;
1148         int type;
1149         int nocow;
1150         int check_prev = 1;
1151         bool nolock;
1152         u64 ino = btrfs_ino(inode);
1153
1154         path = btrfs_alloc_path();
1155         if (!path) {
1156                 extent_clear_unlock_delalloc(inode,
1157                              &BTRFS_I(inode)->io_tree,
1158                              start, end, locked_page,
1159                              EXTENT_CLEAR_UNLOCK_PAGE |
1160                              EXTENT_CLEAR_UNLOCK |
1161                              EXTENT_CLEAR_DELALLOC |
1162                              EXTENT_CLEAR_DIRTY |
1163                              EXTENT_SET_WRITEBACK |
1164                              EXTENT_END_WRITEBACK);
1165                 return -ENOMEM;
1166         }
1167
1168         nolock = btrfs_is_free_space_inode(inode);
1169
1170         if (nolock)
1171                 trans = btrfs_join_transaction_nolock(root);
1172         else
1173                 trans = btrfs_join_transaction(root);
1174
1175         if (IS_ERR(trans)) {
1176                 extent_clear_unlock_delalloc(inode,
1177                              &BTRFS_I(inode)->io_tree,
1178                              start, end, locked_page,
1179                              EXTENT_CLEAR_UNLOCK_PAGE |
1180                              EXTENT_CLEAR_UNLOCK |
1181                              EXTENT_CLEAR_DELALLOC |
1182                              EXTENT_CLEAR_DIRTY |
1183                              EXTENT_SET_WRITEBACK |
1184                              EXTENT_END_WRITEBACK);
1185                 btrfs_free_path(path);
1186                 return PTR_ERR(trans);
1187         }
1188
1189         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1190
1191         cow_start = (u64)-1;
1192         cur_offset = start;
1193         while (1) {
1194                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1195                                                cur_offset, 0);
1196                 if (ret < 0) {
1197                         btrfs_abort_transaction(trans, root, ret);
1198                         goto error;
1199                 }
1200                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1201                         leaf = path->nodes[0];
1202                         btrfs_item_key_to_cpu(leaf, &found_key,
1203                                               path->slots[0] - 1);
1204                         if (found_key.objectid == ino &&
1205                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1206                                 path->slots[0]--;
1207                 }
1208                 check_prev = 0;
1209 next_slot:
1210                 leaf = path->nodes[0];
1211                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1212                         ret = btrfs_next_leaf(root, path);
1213                         if (ret < 0) {
1214                                 btrfs_abort_transaction(trans, root, ret);
1215                                 goto error;
1216                         }
1217                         if (ret > 0)
1218                                 break;
1219                         leaf = path->nodes[0];
1220                 }
1221
1222                 nocow = 0;
1223                 disk_bytenr = 0;
1224                 num_bytes = 0;
1225                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1226
1227                 if (found_key.objectid > ino ||
1228                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1229                     found_key.offset > end)
1230                         break;
1231
1232                 if (found_key.offset > cur_offset) {
1233                         extent_end = found_key.offset;
1234                         extent_type = 0;
1235                         goto out_check;
1236                 }
1237
1238                 fi = btrfs_item_ptr(leaf, path->slots[0],
1239                                     struct btrfs_file_extent_item);
1240                 extent_type = btrfs_file_extent_type(leaf, fi);
1241
1242                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1243                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1244                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1245                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1246                         extent_end = found_key.offset +
1247                                 btrfs_file_extent_num_bytes(leaf, fi);
1248                         if (extent_end <= start) {
1249                                 path->slots[0]++;
1250                                 goto next_slot;
1251                         }
1252                         if (disk_bytenr == 0)
1253                                 goto out_check;
1254                         if (btrfs_file_extent_compression(leaf, fi) ||
1255                             btrfs_file_extent_encryption(leaf, fi) ||
1256                             btrfs_file_extent_other_encoding(leaf, fi))
1257                                 goto out_check;
1258                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1259                                 goto out_check;
1260                         if (btrfs_extent_readonly(root, disk_bytenr))
1261                                 goto out_check;
1262                         if (btrfs_cross_ref_exist(trans, root, ino,
1263                                                   found_key.offset -
1264                                                   extent_offset, disk_bytenr))
1265                                 goto out_check;
1266                         disk_bytenr += extent_offset;
1267                         disk_bytenr += cur_offset - found_key.offset;
1268                         num_bytes = min(end + 1, extent_end) - cur_offset;
1269                         /*
1270                          * force cow if csum exists in the range.
1271                          * this ensure that csum for a given extent are
1272                          * either valid or do not exist.
1273                          */
1274                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1275                                 goto out_check;
1276                         nocow = 1;
1277                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1278                         extent_end = found_key.offset +
1279                                 btrfs_file_extent_inline_len(leaf, fi);
1280                         extent_end = ALIGN(extent_end, root->sectorsize);
1281                 } else {
1282                         BUG_ON(1);
1283                 }
1284 out_check:
1285                 if (extent_end <= start) {
1286                         path->slots[0]++;
1287                         goto next_slot;
1288                 }
1289                 if (!nocow) {
1290                         if (cow_start == (u64)-1)
1291                                 cow_start = cur_offset;
1292                         cur_offset = extent_end;
1293                         if (cur_offset > end)
1294                                 break;
1295                         path->slots[0]++;
1296                         goto next_slot;
1297                 }
1298
1299                 btrfs_release_path(path);
1300                 if (cow_start != (u64)-1) {
1301                         ret = __cow_file_range(trans, inode, root, locked_page,
1302                                                cow_start, found_key.offset - 1,
1303                                                page_started, nr_written, 1);
1304                         if (ret) {
1305                                 btrfs_abort_transaction(trans, root, ret);
1306                                 goto error;
1307                         }
1308                         cow_start = (u64)-1;
1309                 }
1310
1311                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1312                         struct extent_map *em;
1313                         struct extent_map_tree *em_tree;
1314                         em_tree = &BTRFS_I(inode)->extent_tree;
1315                         em = alloc_extent_map();
1316                         BUG_ON(!em); /* -ENOMEM */
1317                         em->start = cur_offset;
1318                         em->orig_start = em->start;
1319                         em->len = num_bytes;
1320                         em->block_len = num_bytes;
1321                         em->block_start = disk_bytenr;
1322                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1323                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1324                         set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1325                         while (1) {
1326                                 write_lock(&em_tree->lock);
1327                                 ret = add_extent_mapping(em_tree, em);
1328                                 write_unlock(&em_tree->lock);
1329                                 if (ret != -EEXIST) {
1330                                         free_extent_map(em);
1331                                         break;
1332                                 }
1333                                 btrfs_drop_extent_cache(inode, em->start,
1334                                                 em->start + em->len - 1, 0);
1335                         }
1336                         type = BTRFS_ORDERED_PREALLOC;
1337                 } else {
1338                         type = BTRFS_ORDERED_NOCOW;
1339                 }
1340
1341                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1342                                                num_bytes, num_bytes, type);
1343                 BUG_ON(ret); /* -ENOMEM */
1344
1345                 if (root->root_key.objectid ==
1346                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1347                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1348                                                       num_bytes);
1349                         if (ret) {
1350                                 btrfs_abort_transaction(trans, root, ret);
1351                                 goto error;
1352                         }
1353                 }
1354
1355                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1356                                 cur_offset, cur_offset + num_bytes - 1,
1357                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1358                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1359                                 EXTENT_SET_PRIVATE2);
1360                 cur_offset = extent_end;
1361                 if (cur_offset > end)
1362                         break;
1363         }
1364         btrfs_release_path(path);
1365
1366         if (cur_offset <= end && cow_start == (u64)-1) {
1367                 cow_start = cur_offset;
1368                 cur_offset = end;
1369         }
1370
1371         if (cow_start != (u64)-1) {
1372                 ret = __cow_file_range(trans, inode, root, locked_page,
1373                                        cow_start, end,
1374                                        page_started, nr_written, 1);
1375                 if (ret) {
1376                         btrfs_abort_transaction(trans, root, ret);
1377                         goto error;
1378                 }
1379         }
1380
1381 error:
1382         err = btrfs_end_transaction(trans, root);
1383         if (!ret)
1384                 ret = err;
1385
1386         if (ret && cur_offset < end)
1387                 extent_clear_unlock_delalloc(inode,
1388                              &BTRFS_I(inode)->io_tree,
1389                              cur_offset, end, locked_page,
1390                              EXTENT_CLEAR_UNLOCK_PAGE |
1391                              EXTENT_CLEAR_UNLOCK |
1392                              EXTENT_CLEAR_DELALLOC |
1393                              EXTENT_CLEAR_DIRTY |
1394                              EXTENT_SET_WRITEBACK |
1395                              EXTENT_END_WRITEBACK);
1396
1397         btrfs_free_path(path);
1398         return ret;
1399 }
1400
1401 /*
1402  * extent_io.c call back to do delayed allocation processing
1403  */
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405                               u64 start, u64 end, int *page_started,
1406                               unsigned long *nr_written)
1407 {
1408         int ret;
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410
1411         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413                                          page_started, 1, nr_written);
1414         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416                                          page_started, 0, nr_written);
1417         } else if (!btrfs_test_opt(root, COMPRESS) &&
1418                    !(BTRFS_I(inode)->force_compress) &&
1419                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420                 ret = cow_file_range(inode, locked_page, start, end,
1421                                       page_started, nr_written, 1);
1422         } else {
1423                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424                         &BTRFS_I(inode)->runtime_flags);
1425                 ret = cow_file_range_async(inode, locked_page, start, end,
1426                                            page_started, nr_written);
1427         }
1428         return ret;
1429 }
1430
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432                                     struct extent_state *orig, u64 split)
1433 {
1434         /* not delalloc, ignore it */
1435         if (!(orig->state & EXTENT_DELALLOC))
1436                 return;
1437
1438         spin_lock(&BTRFS_I(inode)->lock);
1439         BTRFS_I(inode)->outstanding_extents++;
1440         spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 /*
1444  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445  * extents so we can keep track of new extents that are just merged onto old
1446  * extents, such as when we are doing sequential writes, so we can properly
1447  * account for the metadata space we'll need.
1448  */
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450                                     struct extent_state *new,
1451                                     struct extent_state *other)
1452 {
1453         /* not delalloc, ignore it */
1454         if (!(other->state & EXTENT_DELALLOC))
1455                 return;
1456
1457         spin_lock(&BTRFS_I(inode)->lock);
1458         BTRFS_I(inode)->outstanding_extents--;
1459         spin_unlock(&BTRFS_I(inode)->lock);
1460 }
1461
1462 /*
1463  * extent_io.c set_bit_hook, used to track delayed allocation
1464  * bytes in this file, and to maintain the list of inodes that
1465  * have pending delalloc work to be done.
1466  */
1467 static void btrfs_set_bit_hook(struct inode *inode,
1468                                struct extent_state *state, int *bits)
1469 {
1470
1471         /*
1472          * set_bit and clear bit hooks normally require _irqsave/restore
1473          * but in this case, we are only testing for the DELALLOC
1474          * bit, which is only set or cleared with irqs on
1475          */
1476         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1477                 struct btrfs_root *root = BTRFS_I(inode)->root;
1478                 u64 len = state->end + 1 - state->start;
1479                 bool do_list = !btrfs_is_free_space_inode(inode);
1480
1481                 if (*bits & EXTENT_FIRST_DELALLOC) {
1482                         *bits &= ~EXTENT_FIRST_DELALLOC;
1483                 } else {
1484                         spin_lock(&BTRFS_I(inode)->lock);
1485                         BTRFS_I(inode)->outstanding_extents++;
1486                         spin_unlock(&BTRFS_I(inode)->lock);
1487                 }
1488
1489                 spin_lock(&root->fs_info->delalloc_lock);
1490                 BTRFS_I(inode)->delalloc_bytes += len;
1491                 root->fs_info->delalloc_bytes += len;
1492                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1493                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1494                                       &root->fs_info->delalloc_inodes);
1495                 }
1496                 spin_unlock(&root->fs_info->delalloc_lock);
1497         }
1498 }
1499
1500 /*
1501  * extent_io.c clear_bit_hook, see set_bit_hook for why
1502  */
1503 static void btrfs_clear_bit_hook(struct inode *inode,
1504                                  struct extent_state *state, int *bits)
1505 {
1506         /*
1507          * set_bit and clear bit hooks normally require _irqsave/restore
1508          * but in this case, we are only testing for the DELALLOC
1509          * bit, which is only set or cleared with irqs on
1510          */
1511         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1512                 struct btrfs_root *root = BTRFS_I(inode)->root;
1513                 u64 len = state->end + 1 - state->start;
1514                 bool do_list = !btrfs_is_free_space_inode(inode);
1515
1516                 if (*bits & EXTENT_FIRST_DELALLOC) {
1517                         *bits &= ~EXTENT_FIRST_DELALLOC;
1518                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1519                         spin_lock(&BTRFS_I(inode)->lock);
1520                         BTRFS_I(inode)->outstanding_extents--;
1521                         spin_unlock(&BTRFS_I(inode)->lock);
1522                 }
1523
1524                 if (*bits & EXTENT_DO_ACCOUNTING)
1525                         btrfs_delalloc_release_metadata(inode, len);
1526
1527                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1528                     && do_list)
1529                         btrfs_free_reserved_data_space(inode, len);
1530
1531                 spin_lock(&root->fs_info->delalloc_lock);
1532                 root->fs_info->delalloc_bytes -= len;
1533                 BTRFS_I(inode)->delalloc_bytes -= len;
1534
1535                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1536                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1537                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1538                 }
1539                 spin_unlock(&root->fs_info->delalloc_lock);
1540         }
1541 }
1542
1543 /*
1544  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1545  * we don't create bios that span stripes or chunks
1546  */
1547 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1548                          size_t size, struct bio *bio,
1549                          unsigned long bio_flags)
1550 {
1551         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1552         u64 logical = (u64)bio->bi_sector << 9;
1553         u64 length = 0;
1554         u64 map_length;
1555         int ret;
1556
1557         if (bio_flags & EXTENT_BIO_COMPRESSED)
1558                 return 0;
1559
1560         length = bio->bi_size;
1561         map_length = length;
1562         ret = btrfs_map_block(root->fs_info, READ, logical,
1563                               &map_length, NULL, 0);
1564         /* Will always return 0 with map_multi == NULL */
1565         BUG_ON(ret < 0);
1566         if (map_length < length + size)
1567                 return 1;
1568         return 0;
1569 }
1570
1571 /*
1572  * in order to insert checksums into the metadata in large chunks,
1573  * we wait until bio submission time.   All the pages in the bio are
1574  * checksummed and sums are attached onto the ordered extent record.
1575  *
1576  * At IO completion time the cums attached on the ordered extent record
1577  * are inserted into the btree
1578  */
1579 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1580                                     struct bio *bio, int mirror_num,
1581                                     unsigned long bio_flags,
1582                                     u64 bio_offset)
1583 {
1584         struct btrfs_root *root = BTRFS_I(inode)->root;
1585         int ret = 0;
1586
1587         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1588         BUG_ON(ret); /* -ENOMEM */
1589         return 0;
1590 }
1591
1592 /*
1593  * in order to insert checksums into the metadata in large chunks,
1594  * we wait until bio submission time.   All the pages in the bio are
1595  * checksummed and sums are attached onto the ordered extent record.
1596  *
1597  * At IO completion time the cums attached on the ordered extent record
1598  * are inserted into the btree
1599  */
1600 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1601                           int mirror_num, unsigned long bio_flags,
1602                           u64 bio_offset)
1603 {
1604         struct btrfs_root *root = BTRFS_I(inode)->root;
1605         int ret;
1606
1607         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1608         if (ret)
1609                 bio_endio(bio, ret);
1610         return ret;
1611 }
1612
1613 /*
1614  * extent_io.c submission hook. This does the right thing for csum calculation
1615  * on write, or reading the csums from the tree before a read
1616  */
1617 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1618                           int mirror_num, unsigned long bio_flags,
1619                           u64 bio_offset)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         int ret = 0;
1623         int skip_sum;
1624         int metadata = 0;
1625         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1626
1627         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1628
1629         if (btrfs_is_free_space_inode(inode))
1630                 metadata = 2;
1631
1632         if (!(rw & REQ_WRITE)) {
1633                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1634                 if (ret)
1635                         goto out;
1636
1637                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1638                         ret = btrfs_submit_compressed_read(inode, bio,
1639                                                            mirror_num,
1640                                                            bio_flags);
1641                         goto out;
1642                 } else if (!skip_sum) {
1643                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1644                         if (ret)
1645                                 goto out;
1646                 }
1647                 goto mapit;
1648         } else if (async && !skip_sum) {
1649                 /* csum items have already been cloned */
1650                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1651                         goto mapit;
1652                 /* we're doing a write, do the async checksumming */
1653                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1654                                    inode, rw, bio, mirror_num,
1655                                    bio_flags, bio_offset,
1656                                    __btrfs_submit_bio_start,
1657                                    __btrfs_submit_bio_done);
1658                 goto out;
1659         } else if (!skip_sum) {
1660                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1661                 if (ret)
1662                         goto out;
1663         }
1664
1665 mapit:
1666         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1667
1668 out:
1669         if (ret < 0)
1670                 bio_endio(bio, ret);
1671         return ret;
1672 }
1673
1674 /*
1675  * given a list of ordered sums record them in the inode.  This happens
1676  * at IO completion time based on sums calculated at bio submission time.
1677  */
1678 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1679                              struct inode *inode, u64 file_offset,
1680                              struct list_head *list)
1681 {
1682         struct btrfs_ordered_sum *sum;
1683
1684         list_for_each_entry(sum, list, list) {
1685                 btrfs_csum_file_blocks(trans,
1686                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1687         }
1688         return 0;
1689 }
1690
1691 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1692                               struct extent_state **cached_state)
1693 {
1694         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1695         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1696                                    cached_state, GFP_NOFS);
1697 }
1698
1699 /* see btrfs_writepage_start_hook for details on why this is required */
1700 struct btrfs_writepage_fixup {
1701         struct page *page;
1702         struct btrfs_work work;
1703 };
1704
1705 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1706 {
1707         struct btrfs_writepage_fixup *fixup;
1708         struct btrfs_ordered_extent *ordered;
1709         struct extent_state *cached_state = NULL;
1710         struct page *page;
1711         struct inode *inode;
1712         u64 page_start;
1713         u64 page_end;
1714         int ret;
1715
1716         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1717         page = fixup->page;
1718 again:
1719         lock_page(page);
1720         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1721                 ClearPageChecked(page);
1722                 goto out_page;
1723         }
1724
1725         inode = page->mapping->host;
1726         page_start = page_offset(page);
1727         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1728
1729         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1730                          &cached_state);
1731
1732         /* already ordered? We're done */
1733         if (PagePrivate2(page))
1734                 goto out;
1735
1736         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1737         if (ordered) {
1738                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1739                                      page_end, &cached_state, GFP_NOFS);
1740                 unlock_page(page);
1741                 btrfs_start_ordered_extent(inode, ordered, 1);
1742                 btrfs_put_ordered_extent(ordered);
1743                 goto again;
1744         }
1745
1746         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1747         if (ret) {
1748                 mapping_set_error(page->mapping, ret);
1749                 end_extent_writepage(page, ret, page_start, page_end);
1750                 ClearPageChecked(page);
1751                 goto out;
1752          }
1753
1754         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1755         ClearPageChecked(page);
1756         set_page_dirty(page);
1757 out:
1758         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1759                              &cached_state, GFP_NOFS);
1760 out_page:
1761         unlock_page(page);
1762         page_cache_release(page);
1763         kfree(fixup);
1764 }
1765
1766 /*
1767  * There are a few paths in the higher layers of the kernel that directly
1768  * set the page dirty bit without asking the filesystem if it is a
1769  * good idea.  This causes problems because we want to make sure COW
1770  * properly happens and the data=ordered rules are followed.
1771  *
1772  * In our case any range that doesn't have the ORDERED bit set
1773  * hasn't been properly setup for IO.  We kick off an async process
1774  * to fix it up.  The async helper will wait for ordered extents, set
1775  * the delalloc bit and make it safe to write the page.
1776  */
1777 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1778 {
1779         struct inode *inode = page->mapping->host;
1780         struct btrfs_writepage_fixup *fixup;
1781         struct btrfs_root *root = BTRFS_I(inode)->root;
1782
1783         /* this page is properly in the ordered list */
1784         if (TestClearPagePrivate2(page))
1785                 return 0;
1786
1787         if (PageChecked(page))
1788                 return -EAGAIN;
1789
1790         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1791         if (!fixup)
1792                 return -EAGAIN;
1793
1794         SetPageChecked(page);
1795         page_cache_get(page);
1796         fixup->work.func = btrfs_writepage_fixup_worker;
1797         fixup->page = page;
1798         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1799         return -EBUSY;
1800 }
1801
1802 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1803                                        struct inode *inode, u64 file_pos,
1804                                        u64 disk_bytenr, u64 disk_num_bytes,
1805                                        u64 num_bytes, u64 ram_bytes,
1806                                        u8 compression, u8 encryption,
1807                                        u16 other_encoding, int extent_type)
1808 {
1809         struct btrfs_root *root = BTRFS_I(inode)->root;
1810         struct btrfs_file_extent_item *fi;
1811         struct btrfs_path *path;
1812         struct extent_buffer *leaf;
1813         struct btrfs_key ins;
1814         int ret;
1815
1816         path = btrfs_alloc_path();
1817         if (!path)
1818                 return -ENOMEM;
1819
1820         path->leave_spinning = 1;
1821
1822         /*
1823          * we may be replacing one extent in the tree with another.
1824          * The new extent is pinned in the extent map, and we don't want
1825          * to drop it from the cache until it is completely in the btree.
1826          *
1827          * So, tell btrfs_drop_extents to leave this extent in the cache.
1828          * the caller is expected to unpin it and allow it to be merged
1829          * with the others.
1830          */
1831         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1832                                  file_pos + num_bytes, 0);
1833         if (ret)
1834                 goto out;
1835
1836         ins.objectid = btrfs_ino(inode);
1837         ins.offset = file_pos;
1838         ins.type = BTRFS_EXTENT_DATA_KEY;
1839         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1840         if (ret)
1841                 goto out;
1842         leaf = path->nodes[0];
1843         fi = btrfs_item_ptr(leaf, path->slots[0],
1844                             struct btrfs_file_extent_item);
1845         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1846         btrfs_set_file_extent_type(leaf, fi, extent_type);
1847         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1848         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1849         btrfs_set_file_extent_offset(leaf, fi, 0);
1850         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1851         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1852         btrfs_set_file_extent_compression(leaf, fi, compression);
1853         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1854         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1855
1856         btrfs_mark_buffer_dirty(leaf);
1857         btrfs_release_path(path);
1858
1859         inode_add_bytes(inode, num_bytes);
1860
1861         ins.objectid = disk_bytenr;
1862         ins.offset = disk_num_bytes;
1863         ins.type = BTRFS_EXTENT_ITEM_KEY;
1864         ret = btrfs_alloc_reserved_file_extent(trans, root,
1865                                         root->root_key.objectid,
1866                                         btrfs_ino(inode), file_pos, &ins);
1867 out:
1868         btrfs_free_path(path);
1869
1870         return ret;
1871 }
1872
1873 /*
1874  * helper function for btrfs_finish_ordered_io, this
1875  * just reads in some of the csum leaves to prime them into ram
1876  * before we start the transaction.  It limits the amount of btree
1877  * reads required while inside the transaction.
1878  */
1879 /* as ordered data IO finishes, this gets called so we can finish
1880  * an ordered extent if the range of bytes in the file it covers are
1881  * fully written.
1882  */
1883 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1884 {
1885         struct inode *inode = ordered_extent->inode;
1886         struct btrfs_root *root = BTRFS_I(inode)->root;
1887         struct btrfs_trans_handle *trans = NULL;
1888         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1889         struct extent_state *cached_state = NULL;
1890         int compress_type = 0;
1891         int ret;
1892         bool nolock;
1893
1894         nolock = btrfs_is_free_space_inode(inode);
1895
1896         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1897                 ret = -EIO;
1898                 goto out;
1899         }
1900
1901         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1902                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1903                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1904                 if (!ret) {
1905                         if (nolock)
1906                                 trans = btrfs_join_transaction_nolock(root);
1907                         else
1908                                 trans = btrfs_join_transaction(root);
1909                         if (IS_ERR(trans)) {
1910                                 ret = PTR_ERR(trans);
1911                                 trans = NULL;
1912                                 goto out;
1913                         }
1914                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1915                         ret = btrfs_update_inode_fallback(trans, root, inode);
1916                         if (ret) /* -ENOMEM or corruption */
1917                                 btrfs_abort_transaction(trans, root, ret);
1918                 }
1919                 goto out;
1920         }
1921
1922         lock_extent_bits(io_tree, ordered_extent->file_offset,
1923                          ordered_extent->file_offset + ordered_extent->len - 1,
1924                          0, &cached_state);
1925
1926         if (nolock)
1927                 trans = btrfs_join_transaction_nolock(root);
1928         else
1929                 trans = btrfs_join_transaction(root);
1930         if (IS_ERR(trans)) {
1931                 ret = PTR_ERR(trans);
1932                 trans = NULL;
1933                 goto out_unlock;
1934         }
1935         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1936
1937         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1938                 compress_type = ordered_extent->compress_type;
1939         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1940                 BUG_ON(compress_type);
1941                 ret = btrfs_mark_extent_written(trans, inode,
1942                                                 ordered_extent->file_offset,
1943                                                 ordered_extent->file_offset +
1944                                                 ordered_extent->len);
1945         } else {
1946                 BUG_ON(root == root->fs_info->tree_root);
1947                 ret = insert_reserved_file_extent(trans, inode,
1948                                                 ordered_extent->file_offset,
1949                                                 ordered_extent->start,
1950                                                 ordered_extent->disk_len,
1951                                                 ordered_extent->len,
1952                                                 ordered_extent->len,
1953                                                 compress_type, 0, 0,
1954                                                 BTRFS_FILE_EXTENT_REG);
1955         }
1956         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1957                            ordered_extent->file_offset, ordered_extent->len,
1958                            trans->transid);
1959         if (ret < 0) {
1960                 btrfs_abort_transaction(trans, root, ret);
1961                 goto out_unlock;
1962         }
1963
1964         add_pending_csums(trans, inode, ordered_extent->file_offset,
1965                           &ordered_extent->list);
1966
1967         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1968         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1969                 ret = btrfs_update_inode_fallback(trans, root, inode);
1970                 if (ret) { /* -ENOMEM or corruption */
1971                         btrfs_abort_transaction(trans, root, ret);
1972                         goto out_unlock;
1973                 }
1974         } else {
1975                 btrfs_set_inode_last_trans(trans, inode);
1976         }
1977         ret = 0;
1978 out_unlock:
1979         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1980                              ordered_extent->file_offset +
1981                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1982 out:
1983         if (root != root->fs_info->tree_root)
1984                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1985         if (trans)
1986                 btrfs_end_transaction(trans, root);
1987
1988         if (ret)
1989                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1990                                       ordered_extent->file_offset +
1991                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1992
1993         /*
1994          * This needs to be done to make sure anybody waiting knows we are done
1995          * updating everything for this ordered extent.
1996          */
1997         btrfs_remove_ordered_extent(inode, ordered_extent);
1998
1999         /* once for us */
2000         btrfs_put_ordered_extent(ordered_extent);
2001         /* once for the tree */
2002         btrfs_put_ordered_extent(ordered_extent);
2003
2004         return ret;
2005 }
2006
2007 static void finish_ordered_fn(struct btrfs_work *work)
2008 {
2009         struct btrfs_ordered_extent *ordered_extent;
2010         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2011         btrfs_finish_ordered_io(ordered_extent);
2012 }
2013
2014 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2015                                 struct extent_state *state, int uptodate)
2016 {
2017         struct inode *inode = page->mapping->host;
2018         struct btrfs_root *root = BTRFS_I(inode)->root;
2019         struct btrfs_ordered_extent *ordered_extent = NULL;
2020         struct btrfs_workers *workers;
2021
2022         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2023
2024         ClearPagePrivate2(page);
2025         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2026                                             end - start + 1, uptodate))
2027                 return 0;
2028
2029         ordered_extent->work.func = finish_ordered_fn;
2030         ordered_extent->work.flags = 0;
2031
2032         if (btrfs_is_free_space_inode(inode))
2033                 workers = &root->fs_info->endio_freespace_worker;
2034         else
2035                 workers = &root->fs_info->endio_write_workers;
2036         btrfs_queue_worker(workers, &ordered_extent->work);
2037
2038         return 0;
2039 }
2040
2041 /*
2042  * when reads are done, we need to check csums to verify the data is correct
2043  * if there's a match, we allow the bio to finish.  If not, the code in
2044  * extent_io.c will try to find good copies for us.
2045  */
2046 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2047                                struct extent_state *state, int mirror)
2048 {
2049         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2050         struct inode *inode = page->mapping->host;
2051         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2052         char *kaddr;
2053         u64 private = ~(u32)0;
2054         int ret;
2055         struct btrfs_root *root = BTRFS_I(inode)->root;
2056         u32 csum = ~(u32)0;
2057
2058         if (PageChecked(page)) {
2059                 ClearPageChecked(page);
2060                 goto good;
2061         }
2062
2063         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2064                 goto good;
2065
2066         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2067             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2068                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2069                                   GFP_NOFS);
2070                 return 0;
2071         }
2072
2073         if (state && state->start == start) {
2074                 private = state->private;
2075                 ret = 0;
2076         } else {
2077                 ret = get_state_private(io_tree, start, &private);
2078         }
2079         kaddr = kmap_atomic(page);
2080         if (ret)
2081                 goto zeroit;
2082
2083         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2084         btrfs_csum_final(csum, (char *)&csum);
2085         if (csum != private)
2086                 goto zeroit;
2087
2088         kunmap_atomic(kaddr);
2089 good:
2090         return 0;
2091
2092 zeroit:
2093         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2094                        "private %llu\n",
2095                        (unsigned long long)btrfs_ino(page->mapping->host),
2096                        (unsigned long long)start, csum,
2097                        (unsigned long long)private);
2098         memset(kaddr + offset, 1, end - start + 1);
2099         flush_dcache_page(page);
2100         kunmap_atomic(kaddr);
2101         if (private == 0)
2102                 return 0;
2103         return -EIO;
2104 }
2105
2106 struct delayed_iput {
2107         struct list_head list;
2108         struct inode *inode;
2109 };
2110
2111 /* JDM: If this is fs-wide, why can't we add a pointer to
2112  * btrfs_inode instead and avoid the allocation? */
2113 void btrfs_add_delayed_iput(struct inode *inode)
2114 {
2115         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2116         struct delayed_iput *delayed;
2117
2118         if (atomic_add_unless(&inode->i_count, -1, 1))
2119                 return;
2120
2121         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2122         delayed->inode = inode;
2123
2124         spin_lock(&fs_info->delayed_iput_lock);
2125         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2126         spin_unlock(&fs_info->delayed_iput_lock);
2127 }
2128
2129 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2130 {
2131         LIST_HEAD(list);
2132         struct btrfs_fs_info *fs_info = root->fs_info;
2133         struct delayed_iput *delayed;
2134         int empty;
2135
2136         spin_lock(&fs_info->delayed_iput_lock);
2137         empty = list_empty(&fs_info->delayed_iputs);
2138         spin_unlock(&fs_info->delayed_iput_lock);
2139         if (empty)
2140                 return;
2141
2142         spin_lock(&fs_info->delayed_iput_lock);
2143         list_splice_init(&fs_info->delayed_iputs, &list);
2144         spin_unlock(&fs_info->delayed_iput_lock);
2145
2146         while (!list_empty(&list)) {
2147                 delayed = list_entry(list.next, struct delayed_iput, list);
2148                 list_del(&delayed->list);
2149                 iput(delayed->inode);
2150                 kfree(delayed);
2151         }
2152 }
2153
2154 enum btrfs_orphan_cleanup_state {
2155         ORPHAN_CLEANUP_STARTED  = 1,
2156         ORPHAN_CLEANUP_DONE     = 2,
2157 };
2158
2159 /*
2160  * This is called in transaction commit time. If there are no orphan
2161  * files in the subvolume, it removes orphan item and frees block_rsv
2162  * structure.
2163  */
2164 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2165                               struct btrfs_root *root)
2166 {
2167         struct btrfs_block_rsv *block_rsv;
2168         int ret;
2169
2170         if (atomic_read(&root->orphan_inodes) ||
2171             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2172                 return;
2173
2174         spin_lock(&root->orphan_lock);
2175         if (atomic_read(&root->orphan_inodes)) {
2176                 spin_unlock(&root->orphan_lock);
2177                 return;
2178         }
2179
2180         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2181                 spin_unlock(&root->orphan_lock);
2182                 return;
2183         }
2184
2185         block_rsv = root->orphan_block_rsv;
2186         root->orphan_block_rsv = NULL;
2187         spin_unlock(&root->orphan_lock);
2188
2189         if (root->orphan_item_inserted &&
2190             btrfs_root_refs(&root->root_item) > 0) {
2191                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2192                                             root->root_key.objectid);
2193                 BUG_ON(ret);
2194                 root->orphan_item_inserted = 0;
2195         }
2196
2197         if (block_rsv) {
2198                 WARN_ON(block_rsv->size > 0);
2199                 btrfs_free_block_rsv(root, block_rsv);
2200         }
2201 }
2202
2203 /*
2204  * This creates an orphan entry for the given inode in case something goes
2205  * wrong in the middle of an unlink/truncate.
2206  *
2207  * NOTE: caller of this function should reserve 5 units of metadata for
2208  *       this function.
2209  */
2210 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2211 {
2212         struct btrfs_root *root = BTRFS_I(inode)->root;
2213         struct btrfs_block_rsv *block_rsv = NULL;
2214         int reserve = 0;
2215         int insert = 0;
2216         int ret;
2217
2218         if (!root->orphan_block_rsv) {
2219                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2220                 if (!block_rsv)
2221                         return -ENOMEM;
2222         }
2223
2224         spin_lock(&root->orphan_lock);
2225         if (!root->orphan_block_rsv) {
2226                 root->orphan_block_rsv = block_rsv;
2227         } else if (block_rsv) {
2228                 btrfs_free_block_rsv(root, block_rsv);
2229                 block_rsv = NULL;
2230         }
2231
2232         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2233                               &BTRFS_I(inode)->runtime_flags)) {
2234 #if 0
2235                 /*
2236                  * For proper ENOSPC handling, we should do orphan
2237                  * cleanup when mounting. But this introduces backward
2238                  * compatibility issue.
2239                  */
2240                 if (!xchg(&root->orphan_item_inserted, 1))
2241                         insert = 2;
2242                 else
2243                         insert = 1;
2244 #endif
2245                 insert = 1;
2246                 atomic_inc(&root->orphan_inodes);
2247         }
2248
2249         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2250                               &BTRFS_I(inode)->runtime_flags))
2251                 reserve = 1;
2252         spin_unlock(&root->orphan_lock);
2253
2254         /* grab metadata reservation from transaction handle */
2255         if (reserve) {
2256                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2257                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2258         }
2259
2260         /* insert an orphan item to track this unlinked/truncated file */
2261         if (insert >= 1) {
2262                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2263                 if (ret && ret != -EEXIST) {
2264                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2265                                   &BTRFS_I(inode)->runtime_flags);
2266                         btrfs_abort_transaction(trans, root, ret);
2267                         return ret;
2268                 }
2269                 ret = 0;
2270         }
2271
2272         /* insert an orphan item to track subvolume contains orphan files */
2273         if (insert >= 2) {
2274                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2275                                                root->root_key.objectid);
2276                 if (ret && ret != -EEXIST) {
2277                         btrfs_abort_transaction(trans, root, ret);
2278                         return ret;
2279                 }
2280         }
2281         return 0;
2282 }
2283
2284 /*
2285  * We have done the truncate/delete so we can go ahead and remove the orphan
2286  * item for this particular inode.
2287  */
2288 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2289 {
2290         struct btrfs_root *root = BTRFS_I(inode)->root;
2291         int delete_item = 0;
2292         int release_rsv = 0;
2293         int ret = 0;
2294
2295         spin_lock(&root->orphan_lock);
2296         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2297                                &BTRFS_I(inode)->runtime_flags))
2298                 delete_item = 1;
2299
2300         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2301                                &BTRFS_I(inode)->runtime_flags))
2302                 release_rsv = 1;
2303         spin_unlock(&root->orphan_lock);
2304
2305         if (trans && delete_item) {
2306                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2307                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2308         }
2309
2310         if (release_rsv) {
2311                 btrfs_orphan_release_metadata(inode);
2312                 atomic_dec(&root->orphan_inodes);
2313         }
2314
2315         return 0;
2316 }
2317
2318 /*
2319  * this cleans up any orphans that may be left on the list from the last use
2320  * of this root.
2321  */
2322 int btrfs_orphan_cleanup(struct btrfs_root *root)
2323 {
2324         struct btrfs_path *path;
2325         struct extent_buffer *leaf;
2326         struct btrfs_key key, found_key;
2327         struct btrfs_trans_handle *trans;
2328         struct inode *inode;
2329         u64 last_objectid = 0;
2330         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2331
2332         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2333                 return 0;
2334
2335         path = btrfs_alloc_path();
2336         if (!path) {
2337                 ret = -ENOMEM;
2338                 goto out;
2339         }
2340         path->reada = -1;
2341
2342         key.objectid = BTRFS_ORPHAN_OBJECTID;
2343         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2344         key.offset = (u64)-1;
2345
2346         while (1) {
2347                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2348                 if (ret < 0)
2349                         goto out;
2350
2351                 /*
2352                  * if ret == 0 means we found what we were searching for, which
2353                  * is weird, but possible, so only screw with path if we didn't
2354                  * find the key and see if we have stuff that matches
2355                  */
2356                 if (ret > 0) {
2357                         ret = 0;
2358                         if (path->slots[0] == 0)
2359                                 break;
2360                         path->slots[0]--;
2361                 }
2362
2363                 /* pull out the item */
2364                 leaf = path->nodes[0];
2365                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2366
2367                 /* make sure the item matches what we want */
2368                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2369                         break;
2370                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2371                         break;
2372
2373                 /* release the path since we're done with it */
2374                 btrfs_release_path(path);
2375
2376                 /*
2377                  * this is where we are basically btrfs_lookup, without the
2378                  * crossing root thing.  we store the inode number in the
2379                  * offset of the orphan item.
2380                  */
2381
2382                 if (found_key.offset == last_objectid) {
2383                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2384                                "stopping orphan cleanup\n");
2385                         ret = -EINVAL;
2386                         goto out;
2387                 }
2388
2389                 last_objectid = found_key.offset;
2390
2391                 found_key.objectid = found_key.offset;
2392                 found_key.type = BTRFS_INODE_ITEM_KEY;
2393                 found_key.offset = 0;
2394                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2395                 ret = PTR_RET(inode);
2396                 if (ret && ret != -ESTALE)
2397                         goto out;
2398
2399                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2400                         struct btrfs_root *dead_root;
2401                         struct btrfs_fs_info *fs_info = root->fs_info;
2402                         int is_dead_root = 0;
2403
2404                         /*
2405                          * this is an orphan in the tree root. Currently these
2406                          * could come from 2 sources:
2407                          *  a) a snapshot deletion in progress
2408                          *  b) a free space cache inode
2409                          * We need to distinguish those two, as the snapshot
2410                          * orphan must not get deleted.
2411                          * find_dead_roots already ran before us, so if this
2412                          * is a snapshot deletion, we should find the root
2413                          * in the dead_roots list
2414                          */
2415                         spin_lock(&fs_info->trans_lock);
2416                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2417                                             root_list) {
2418                                 if (dead_root->root_key.objectid ==
2419                                     found_key.objectid) {
2420                                         is_dead_root = 1;
2421                                         break;
2422                                 }
2423                         }
2424                         spin_unlock(&fs_info->trans_lock);
2425                         if (is_dead_root) {
2426                                 /* prevent this orphan from being found again */
2427                                 key.offset = found_key.objectid - 1;
2428                                 continue;
2429                         }
2430                 }
2431                 /*
2432                  * Inode is already gone but the orphan item is still there,
2433                  * kill the orphan item.
2434                  */
2435                 if (ret == -ESTALE) {
2436                         trans = btrfs_start_transaction(root, 1);
2437                         if (IS_ERR(trans)) {
2438                                 ret = PTR_ERR(trans);
2439                                 goto out;
2440                         }
2441                         printk(KERN_ERR "auto deleting %Lu\n",
2442                                found_key.objectid);
2443                         ret = btrfs_del_orphan_item(trans, root,
2444                                                     found_key.objectid);
2445                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2446                         btrfs_end_transaction(trans, root);
2447                         continue;
2448                 }
2449
2450                 /*
2451                  * add this inode to the orphan list so btrfs_orphan_del does
2452                  * the proper thing when we hit it
2453                  */
2454                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2455                         &BTRFS_I(inode)->runtime_flags);
2456
2457                 /* if we have links, this was a truncate, lets do that */
2458                 if (inode->i_nlink) {
2459                         if (!S_ISREG(inode->i_mode)) {
2460                                 WARN_ON(1);
2461                                 iput(inode);
2462                                 continue;
2463                         }
2464                         nr_truncate++;
2465                         ret = btrfs_truncate(inode);
2466                 } else {
2467                         nr_unlink++;
2468                 }
2469
2470                 /* this will do delete_inode and everything for us */
2471                 iput(inode);
2472                 if (ret)
2473                         goto out;
2474         }
2475         /* release the path since we're done with it */
2476         btrfs_release_path(path);
2477
2478         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2479
2480         if (root->orphan_block_rsv)
2481                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2482                                         (u64)-1);
2483
2484         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2485                 trans = btrfs_join_transaction(root);
2486                 if (!IS_ERR(trans))
2487                         btrfs_end_transaction(trans, root);
2488         }
2489
2490         if (nr_unlink)
2491                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2492         if (nr_truncate)
2493                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2494
2495 out:
2496         if (ret)
2497                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2498         btrfs_free_path(path);
2499         return ret;
2500 }
2501
2502 /*
2503  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2504  * don't find any xattrs, we know there can't be any acls.
2505  *
2506  * slot is the slot the inode is in, objectid is the objectid of the inode
2507  */
2508 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2509                                           int slot, u64 objectid)
2510 {
2511         u32 nritems = btrfs_header_nritems(leaf);
2512         struct btrfs_key found_key;
2513         int scanned = 0;
2514
2515         slot++;
2516         while (slot < nritems) {
2517                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2518
2519                 /* we found a different objectid, there must not be acls */
2520                 if (found_key.objectid != objectid)
2521                         return 0;
2522
2523                 /* we found an xattr, assume we've got an acl */
2524                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2525                         return 1;
2526
2527                 /*
2528                  * we found a key greater than an xattr key, there can't
2529                  * be any acls later on
2530                  */
2531                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2532                         return 0;
2533
2534                 slot++;
2535                 scanned++;
2536
2537                 /*
2538                  * it goes inode, inode backrefs, xattrs, extents,
2539                  * so if there are a ton of hard links to an inode there can
2540                  * be a lot of backrefs.  Don't waste time searching too hard,
2541                  * this is just an optimization
2542                  */
2543                 if (scanned >= 8)
2544                         break;
2545         }
2546         /* we hit the end of the leaf before we found an xattr or
2547          * something larger than an xattr.  We have to assume the inode
2548          * has acls
2549          */
2550         return 1;
2551 }
2552
2553 /*
2554  * read an inode from the btree into the in-memory inode
2555  */
2556 static void btrfs_read_locked_inode(struct inode *inode)
2557 {
2558         struct btrfs_path *path;
2559         struct extent_buffer *leaf;
2560         struct btrfs_inode_item *inode_item;
2561         struct btrfs_timespec *tspec;
2562         struct btrfs_root *root = BTRFS_I(inode)->root;
2563         struct btrfs_key location;
2564         int maybe_acls;
2565         u32 rdev;
2566         int ret;
2567         bool filled = false;
2568
2569         ret = btrfs_fill_inode(inode, &rdev);
2570         if (!ret)
2571                 filled = true;
2572
2573         path = btrfs_alloc_path();
2574         if (!path)
2575                 goto make_bad;
2576
2577         path->leave_spinning = 1;
2578         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2579
2580         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2581         if (ret)
2582                 goto make_bad;
2583
2584         leaf = path->nodes[0];
2585
2586         if (filled)
2587                 goto cache_acl;
2588
2589         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2590                                     struct btrfs_inode_item);
2591         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2592         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2593         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2594         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2595         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2596
2597         tspec = btrfs_inode_atime(inode_item);
2598         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2599         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2600
2601         tspec = btrfs_inode_mtime(inode_item);
2602         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2603         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2604
2605         tspec = btrfs_inode_ctime(inode_item);
2606         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2607         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2608
2609         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2610         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2611         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2612
2613         /*
2614          * If we were modified in the current generation and evicted from memory
2615          * and then re-read we need to do a full sync since we don't have any
2616          * idea about which extents were modified before we were evicted from
2617          * cache.
2618          */
2619         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2620                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2621                         &BTRFS_I(inode)->runtime_flags);
2622
2623         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2624         inode->i_generation = BTRFS_I(inode)->generation;
2625         inode->i_rdev = 0;
2626         rdev = btrfs_inode_rdev(leaf, inode_item);
2627
2628         BTRFS_I(inode)->index_cnt = (u64)-1;
2629         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2630 cache_acl:
2631         /*
2632          * try to precache a NULL acl entry for files that don't have
2633          * any xattrs or acls
2634          */
2635         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2636                                            btrfs_ino(inode));
2637         if (!maybe_acls)
2638                 cache_no_acl(inode);
2639
2640         btrfs_free_path(path);
2641
2642         switch (inode->i_mode & S_IFMT) {
2643         case S_IFREG:
2644                 inode->i_mapping->a_ops = &btrfs_aops;
2645                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2646                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2647                 inode->i_fop = &btrfs_file_operations;
2648                 inode->i_op = &btrfs_file_inode_operations;
2649                 break;
2650         case S_IFDIR:
2651                 inode->i_fop = &btrfs_dir_file_operations;
2652                 if (root == root->fs_info->tree_root)
2653                         inode->i_op = &btrfs_dir_ro_inode_operations;
2654                 else
2655                         inode->i_op = &btrfs_dir_inode_operations;
2656                 break;
2657         case S_IFLNK:
2658                 inode->i_op = &btrfs_symlink_inode_operations;
2659                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2660                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2661                 break;
2662         default:
2663                 inode->i_op = &btrfs_special_inode_operations;
2664                 init_special_inode(inode, inode->i_mode, rdev);
2665                 break;
2666         }
2667
2668         btrfs_update_iflags(inode);
2669         return;
2670
2671 make_bad:
2672         btrfs_free_path(path);
2673         make_bad_inode(inode);
2674 }
2675
2676 /*
2677  * given a leaf and an inode, copy the inode fields into the leaf
2678  */
2679 static void fill_inode_item(struct btrfs_trans_handle *trans,
2680                             struct extent_buffer *leaf,
2681                             struct btrfs_inode_item *item,
2682                             struct inode *inode)
2683 {
2684         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2685         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2686         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2687         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2688         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2689
2690         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2691                                inode->i_atime.tv_sec);
2692         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2693                                 inode->i_atime.tv_nsec);
2694
2695         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2696                                inode->i_mtime.tv_sec);
2697         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2698                                 inode->i_mtime.tv_nsec);
2699
2700         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2701                                inode->i_ctime.tv_sec);
2702         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2703                                 inode->i_ctime.tv_nsec);
2704
2705         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2706         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2707         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2708         btrfs_set_inode_transid(leaf, item, trans->transid);
2709         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2710         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2711         btrfs_set_inode_block_group(leaf, item, 0);
2712 }
2713
2714 /*
2715  * copy everything in the in-memory inode into the btree.
2716  */
2717 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2718                                 struct btrfs_root *root, struct inode *inode)
2719 {
2720         struct btrfs_inode_item *inode_item;
2721         struct btrfs_path *path;
2722         struct extent_buffer *leaf;
2723         int ret;
2724
2725         path = btrfs_alloc_path();
2726         if (!path)
2727                 return -ENOMEM;
2728
2729         path->leave_spinning = 1;
2730         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2731                                  1);
2732         if (ret) {
2733                 if (ret > 0)
2734                         ret = -ENOENT;
2735                 goto failed;
2736         }
2737
2738         btrfs_unlock_up_safe(path, 1);
2739         leaf = path->nodes[0];
2740         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2741                                     struct btrfs_inode_item);
2742
2743         fill_inode_item(trans, leaf, inode_item, inode);
2744         btrfs_mark_buffer_dirty(leaf);
2745         btrfs_set_inode_last_trans(trans, inode);
2746         ret = 0;
2747 failed:
2748         btrfs_free_path(path);
2749         return ret;
2750 }
2751
2752 /*
2753  * copy everything in the in-memory inode into the btree.
2754  */
2755 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2756                                 struct btrfs_root *root, struct inode *inode)
2757 {
2758         int ret;
2759
2760         /*
2761          * If the inode is a free space inode, we can deadlock during commit
2762          * if we put it into the delayed code.
2763          *
2764          * The data relocation inode should also be directly updated
2765          * without delay
2766          */
2767         if (!btrfs_is_free_space_inode(inode)
2768             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2769                 btrfs_update_root_times(trans, root);
2770
2771                 ret = btrfs_delayed_update_inode(trans, root, inode);
2772                 if (!ret)
2773                         btrfs_set_inode_last_trans(trans, inode);
2774                 return ret;
2775         }
2776
2777         return btrfs_update_inode_item(trans, root, inode);
2778 }
2779
2780 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2781                                          struct btrfs_root *root,
2782                                          struct inode *inode)
2783 {
2784         int ret;
2785
2786         ret = btrfs_update_inode(trans, root, inode);
2787         if (ret == -ENOSPC)
2788                 return btrfs_update_inode_item(trans, root, inode);
2789         return ret;
2790 }
2791
2792 /*
2793  * unlink helper that gets used here in inode.c and in the tree logging
2794  * recovery code.  It remove a link in a directory with a given name, and
2795  * also drops the back refs in the inode to the directory
2796  */
2797 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2798                                 struct btrfs_root *root,
2799                                 struct inode *dir, struct inode *inode,
2800                                 const char *name, int name_len)
2801 {
2802         struct btrfs_path *path;
2803         int ret = 0;
2804         struct extent_buffer *leaf;
2805         struct btrfs_dir_item *di;
2806         struct btrfs_key key;
2807         u64 index;
2808         u64 ino = btrfs_ino(inode);
2809         u64 dir_ino = btrfs_ino(dir);
2810
2811         path = btrfs_alloc_path();
2812         if (!path) {
2813                 ret = -ENOMEM;
2814                 goto out;
2815         }
2816
2817         path->leave_spinning = 1;
2818         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2819                                     name, name_len, -1);
2820         if (IS_ERR(di)) {
2821                 ret = PTR_ERR(di);
2822                 goto err;
2823         }
2824         if (!di) {
2825                 ret = -ENOENT;
2826                 goto err;
2827         }
2828         leaf = path->nodes[0];
2829         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2830         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2831         if (ret)
2832                 goto err;
2833         btrfs_release_path(path);
2834
2835         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2836                                   dir_ino, &index);
2837         if (ret) {
2838                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2839                        "inode %llu parent %llu\n", name_len, name,
2840                        (unsigned long long)ino, (unsigned long long)dir_ino);
2841                 btrfs_abort_transaction(trans, root, ret);
2842                 goto err;
2843         }
2844
2845         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2846         if (ret) {
2847                 btrfs_abort_transaction(trans, root, ret);
2848                 goto err;
2849         }
2850
2851         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2852                                          inode, dir_ino);
2853         if (ret != 0 && ret != -ENOENT) {
2854                 btrfs_abort_transaction(trans, root, ret);
2855                 goto err;
2856         }
2857
2858         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2859                                            dir, index);
2860         if (ret == -ENOENT)
2861                 ret = 0;
2862 err:
2863         btrfs_free_path(path);
2864         if (ret)
2865                 goto out;
2866
2867         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2868         inode_inc_iversion(inode);
2869         inode_inc_iversion(dir);
2870         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2871         ret = btrfs_update_inode(trans, root, dir);
2872 out:
2873         return ret;
2874 }
2875
2876 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2877                        struct btrfs_root *root,
2878                        struct inode *dir, struct inode *inode,
2879                        const char *name, int name_len)
2880 {
2881         int ret;
2882         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2883         if (!ret) {
2884                 btrfs_drop_nlink(inode);
2885                 ret = btrfs_update_inode(trans, root, inode);
2886         }
2887         return ret;
2888 }
2889
2890
2891 /* helper to check if there is any shared block in the path */
2892 static int check_path_shared(struct btrfs_root *root,
2893                              struct btrfs_path *path)
2894 {
2895         struct extent_buffer *eb;
2896         int level;
2897         u64 refs = 1;
2898
2899         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2900                 int ret;
2901
2902                 if (!path->nodes[level])
2903                         break;
2904                 eb = path->nodes[level];
2905                 if (!btrfs_block_can_be_shared(root, eb))
2906                         continue;
2907                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2908                                                &refs, NULL);
2909                 if (refs > 1)
2910                         return 1;
2911         }
2912         return 0;
2913 }
2914
2915 /*
2916  * helper to start transaction for unlink and rmdir.
2917  *
2918  * unlink and rmdir are special in btrfs, they do not always free space.
2919  * so in enospc case, we should make sure they will free space before
2920  * allowing them to use the global metadata reservation.
2921  */
2922 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2923                                                        struct dentry *dentry)
2924 {
2925         struct btrfs_trans_handle *trans;
2926         struct btrfs_root *root = BTRFS_I(dir)->root;
2927         struct btrfs_path *path;
2928         struct btrfs_dir_item *di;
2929         struct inode *inode = dentry->d_inode;
2930         u64 index;
2931         int check_link = 1;
2932         int err = -ENOSPC;
2933         int ret;
2934         u64 ino = btrfs_ino(inode);
2935         u64 dir_ino = btrfs_ino(dir);
2936
2937         /*
2938          * 1 for the possible orphan item
2939          * 1 for the dir item
2940          * 1 for the dir index
2941          * 1 for the inode ref
2942          * 1 for the inode ref in the tree log
2943          * 2 for the dir entries in the log
2944          * 1 for the inode
2945          */
2946         trans = btrfs_start_transaction(root, 8);
2947         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2948                 return trans;
2949
2950         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2951                 return ERR_PTR(-ENOSPC);
2952
2953         /* check if there is someone else holds reference */
2954         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2955                 return ERR_PTR(-ENOSPC);
2956
2957         if (atomic_read(&inode->i_count) > 2)
2958                 return ERR_PTR(-ENOSPC);
2959
2960         if (xchg(&root->fs_info->enospc_unlink, 1))
2961                 return ERR_PTR(-ENOSPC);
2962
2963         path = btrfs_alloc_path();
2964         if (!path) {
2965                 root->fs_info->enospc_unlink = 0;
2966                 return ERR_PTR(-ENOMEM);
2967         }
2968
2969         /* 1 for the orphan item */
2970         trans = btrfs_start_transaction(root, 1);
2971         if (IS_ERR(trans)) {
2972                 btrfs_free_path(path);
2973                 root->fs_info->enospc_unlink = 0;
2974                 return trans;
2975         }
2976
2977         path->skip_locking = 1;
2978         path->search_commit_root = 1;
2979
2980         ret = btrfs_lookup_inode(trans, root, path,
2981                                 &BTRFS_I(dir)->location, 0);
2982         if (ret < 0) {
2983                 err = ret;
2984                 goto out;
2985         }
2986         if (ret == 0) {
2987                 if (check_path_shared(root, path))
2988                         goto out;
2989         } else {
2990                 check_link = 0;
2991         }
2992         btrfs_release_path(path);
2993
2994         ret = btrfs_lookup_inode(trans, root, path,
2995                                 &BTRFS_I(inode)->location, 0);
2996         if (ret < 0) {
2997                 err = ret;
2998                 goto out;
2999         }
3000         if (ret == 0) {
3001                 if (check_path_shared(root, path))
3002                         goto out;
3003         } else {
3004                 check_link = 0;
3005         }
3006         btrfs_release_path(path);
3007
3008         if (ret == 0 && S_ISREG(inode->i_mode)) {
3009                 ret = btrfs_lookup_file_extent(trans, root, path,
3010                                                ino, (u64)-1, 0);
3011                 if (ret < 0) {
3012                         err = ret;
3013                         goto out;
3014                 }
3015                 BUG_ON(ret == 0); /* Corruption */
3016                 if (check_path_shared(root, path))
3017                         goto out;
3018                 btrfs_release_path(path);
3019         }
3020
3021         if (!check_link) {
3022                 err = 0;
3023                 goto out;
3024         }
3025
3026         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3027                                 dentry->d_name.name, dentry->d_name.len, 0);
3028         if (IS_ERR(di)) {
3029                 err = PTR_ERR(di);
3030                 goto out;
3031         }
3032         if (di) {
3033                 if (check_path_shared(root, path))
3034                         goto out;
3035         } else {
3036                 err = 0;
3037                 goto out;
3038         }
3039         btrfs_release_path(path);
3040
3041         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3042                                         dentry->d_name.len, ino, dir_ino, 0,
3043                                         &index);
3044         if (ret) {
3045                 err = ret;
3046                 goto out;
3047         }
3048
3049         if (check_path_shared(root, path))
3050                 goto out;
3051
3052         btrfs_release_path(path);
3053
3054         /*
3055          * This is a commit root search, if we can lookup inode item and other
3056          * relative items in the commit root, it means the transaction of
3057          * dir/file creation has been committed, and the dir index item that we
3058          * delay to insert has also been inserted into the commit root. So
3059          * we needn't worry about the delayed insertion of the dir index item
3060          * here.
3061          */
3062         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3063                                 dentry->d_name.name, dentry->d_name.len, 0);
3064         if (IS_ERR(di)) {
3065                 err = PTR_ERR(di);
3066                 goto out;
3067         }
3068         BUG_ON(ret == -ENOENT);
3069         if (check_path_shared(root, path))
3070                 goto out;
3071
3072         err = 0;
3073 out:
3074         btrfs_free_path(path);
3075         /* Migrate the orphan reservation over */
3076         if (!err)
3077                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3078                                 &root->fs_info->global_block_rsv,
3079                                 trans->bytes_reserved);
3080
3081         if (err) {
3082                 btrfs_end_transaction(trans, root);
3083                 root->fs_info->enospc_unlink = 0;
3084                 return ERR_PTR(err);
3085         }
3086
3087         trans->block_rsv = &root->fs_info->global_block_rsv;
3088         return trans;
3089 }
3090
3091 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3092                                struct btrfs_root *root)
3093 {
3094         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3095                 btrfs_block_rsv_release(root, trans->block_rsv,
3096                                         trans->bytes_reserved);
3097                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3098                 BUG_ON(!root->fs_info->enospc_unlink);
3099                 root->fs_info->enospc_unlink = 0;
3100         }
3101         btrfs_end_transaction(trans, root);
3102 }
3103
3104 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3105 {
3106         struct btrfs_root *root = BTRFS_I(dir)->root;
3107         struct btrfs_trans_handle *trans;
3108         struct inode *inode = dentry->d_inode;
3109         int ret;
3110
3111         trans = __unlink_start_trans(dir, dentry);
3112         if (IS_ERR(trans))
3113                 return PTR_ERR(trans);
3114
3115         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3116
3117         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3118                                  dentry->d_name.name, dentry->d_name.len);
3119         if (ret)
3120                 goto out;
3121
3122         if (inode->i_nlink == 0) {
3123                 ret = btrfs_orphan_add(trans, inode);
3124                 if (ret)
3125                         goto out;
3126         }
3127
3128 out:
3129         __unlink_end_trans(trans, root);
3130         btrfs_btree_balance_dirty(root);
3131         return ret;
3132 }
3133
3134 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3135                         struct btrfs_root *root,
3136                         struct inode *dir, u64 objectid,
3137                         const char *name, int name_len)
3138 {
3139         struct btrfs_path *path;
3140         struct extent_buffer *leaf;
3141         struct btrfs_dir_item *di;
3142         struct btrfs_key key;
3143         u64 index;
3144         int ret;
3145         u64 dir_ino = btrfs_ino(dir);
3146
3147         path = btrfs_alloc_path();
3148         if (!path)
3149                 return -ENOMEM;
3150
3151         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3152                                    name, name_len, -1);
3153         if (IS_ERR_OR_NULL(di)) {
3154                 if (!di)
3155                         ret = -ENOENT;
3156                 else
3157                         ret = PTR_ERR(di);
3158                 goto out;
3159         }
3160
3161         leaf = path->nodes[0];
3162         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3163         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3164         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3165         if (ret) {
3166                 btrfs_abort_transaction(trans, root, ret);
3167                 goto out;
3168         }
3169         btrfs_release_path(path);
3170
3171         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3172                                  objectid, root->root_key.objectid,
3173                                  dir_ino, &index, name, name_len);
3174         if (ret < 0) {
3175                 if (ret != -ENOENT) {
3176                         btrfs_abort_transaction(trans, root, ret);
3177                         goto out;
3178                 }
3179                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3180                                                  name, name_len);
3181                 if (IS_ERR_OR_NULL(di)) {
3182                         if (!di)
3183                                 ret = -ENOENT;
3184                         else
3185                                 ret = PTR_ERR(di);
3186                         btrfs_abort_transaction(trans, root, ret);
3187                         goto out;
3188                 }
3189
3190                 leaf = path->nodes[0];
3191                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3192                 btrfs_release_path(path);
3193                 index = key.offset;
3194         }
3195         btrfs_release_path(path);
3196
3197         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3198         if (ret) {
3199                 btrfs_abort_transaction(trans, root, ret);
3200                 goto out;
3201         }
3202
3203         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3204         inode_inc_iversion(dir);
3205         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3206         ret = btrfs_update_inode_fallback(trans, root, dir);
3207         if (ret)
3208                 btrfs_abort_transaction(trans, root, ret);
3209 out:
3210         btrfs_free_path(path);
3211         return ret;
3212 }
3213
3214 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3215 {
3216         struct inode *inode = dentry->d_inode;
3217         int err = 0;
3218         struct btrfs_root *root = BTRFS_I(dir)->root;
3219         struct btrfs_trans_handle *trans;
3220
3221         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3222                 return -ENOTEMPTY;
3223         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3224                 return -EPERM;
3225
3226         trans = __unlink_start_trans(dir, dentry);
3227         if (IS_ERR(trans))
3228                 return PTR_ERR(trans);
3229
3230         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3231                 err = btrfs_unlink_subvol(trans, root, dir,
3232                                           BTRFS_I(inode)->location.objectid,
3233                                           dentry->d_name.name,
3234                                           dentry->d_name.len);
3235                 goto out;
3236         }
3237
3238         err = btrfs_orphan_add(trans, inode);
3239         if (err)
3240                 goto out;
3241
3242         /* now the directory is empty */
3243         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3244                                  dentry->d_name.name, dentry->d_name.len);
3245         if (!err)
3246                 btrfs_i_size_write(inode, 0);
3247 out:
3248         __unlink_end_trans(trans, root);
3249         btrfs_btree_balance_dirty(root);
3250
3251         return err;
3252 }
3253
3254 /*
3255  * this can truncate away extent items, csum items and directory items.
3256  * It starts at a high offset and removes keys until it can't find
3257  * any higher than new_size
3258  *
3259  * csum items that cross the new i_size are truncated to the new size
3260  * as well.
3261  *
3262  * min_type is the minimum key type to truncate down to.  If set to 0, this
3263  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3264  */
3265 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3266                                struct btrfs_root *root,
3267                                struct inode *inode,
3268                                u64 new_size, u32 min_type)
3269 {
3270         struct btrfs_path *path;
3271         struct extent_buffer *leaf;
3272         struct btrfs_file_extent_item *fi;
3273         struct btrfs_key key;
3274         struct btrfs_key found_key;
3275         u64 extent_start = 0;
3276         u64 extent_num_bytes = 0;
3277         u64 extent_offset = 0;
3278         u64 item_end = 0;
3279         u64 mask = root->sectorsize - 1;
3280         u32 found_type = (u8)-1;
3281         int found_extent;
3282         int del_item;
3283         int pending_del_nr = 0;
3284         int pending_del_slot = 0;
3285         int extent_type = -1;
3286         int ret;
3287         int err = 0;
3288         u64 ino = btrfs_ino(inode);
3289
3290         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3291
3292         path = btrfs_alloc_path();
3293         if (!path)
3294                 return -ENOMEM;
3295         path->reada = -1;
3296
3297         /*
3298          * We want to drop from the next block forward in case this new size is
3299          * not block aligned since we will be keeping the last block of the
3300          * extent just the way it is.
3301          */
3302         if (root->ref_cows || root == root->fs_info->tree_root)
3303                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3304
3305         /*
3306          * This function is also used to drop the items in the log tree before
3307          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3308          * it is used to drop the loged items. So we shouldn't kill the delayed
3309          * items.
3310          */
3311         if (min_type == 0 && root == BTRFS_I(inode)->root)
3312                 btrfs_kill_delayed_inode_items(inode);
3313
3314         key.objectid = ino;
3315         key.offset = (u64)-1;
3316         key.type = (u8)-1;
3317
3318 search_again:
3319         path->leave_spinning = 1;
3320         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3321         if (ret < 0) {
3322                 err = ret;
3323                 goto out;
3324         }
3325
3326         if (ret > 0) {
3327                 /* there are no items in the tree for us to truncate, we're
3328                  * done
3329                  */
3330                 if (path->slots[0] == 0)
3331                         goto out;
3332                 path->slots[0]--;
3333         }
3334
3335         while (1) {
3336                 fi = NULL;
3337                 leaf = path->nodes[0];
3338                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3339                 found_type = btrfs_key_type(&found_key);
3340
3341                 if (found_key.objectid != ino)
3342                         break;
3343
3344                 if (found_type < min_type)
3345                         break;
3346
3347                 item_end = found_key.offset;
3348                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3349                         fi = btrfs_item_ptr(leaf, path->slots[0],
3350                                             struct btrfs_file_extent_item);
3351                         extent_type = btrfs_file_extent_type(leaf, fi);
3352                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3353                                 item_end +=
3354                                     btrfs_file_extent_num_bytes(leaf, fi);
3355                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3356                                 item_end += btrfs_file_extent_inline_len(leaf,
3357                                                                          fi);
3358                         }
3359                         item_end--;
3360                 }
3361                 if (found_type > min_type) {
3362                         del_item = 1;
3363                 } else {
3364                         if (item_end < new_size)
3365                                 break;
3366                         if (found_key.offset >= new_size)
3367                                 del_item = 1;
3368                         else
3369                                 del_item = 0;
3370                 }
3371                 found_extent = 0;
3372                 /* FIXME, shrink the extent if the ref count is only 1 */
3373                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3374                         goto delete;
3375
3376                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3377                         u64 num_dec;
3378                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3379                         if (!del_item) {
3380                                 u64 orig_num_bytes =
3381                                         btrfs_file_extent_num_bytes(leaf, fi);
3382                                 extent_num_bytes = new_size -
3383                                         found_key.offset + root->sectorsize - 1;
3384                                 extent_num_bytes = extent_num_bytes &
3385                                         ~((u64)root->sectorsize - 1);
3386                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3387                                                          extent_num_bytes);
3388                                 num_dec = (orig_num_bytes -
3389                                            extent_num_bytes);
3390                                 if (root->ref_cows && extent_start != 0)
3391                                         inode_sub_bytes(inode, num_dec);
3392                                 btrfs_mark_buffer_dirty(leaf);
3393                         } else {
3394                                 extent_num_bytes =
3395                                         btrfs_file_extent_disk_num_bytes(leaf,
3396                                                                          fi);
3397                                 extent_offset = found_key.offset -
3398                                         btrfs_file_extent_offset(leaf, fi);
3399
3400                                 /* FIXME blocksize != 4096 */
3401                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3402                                 if (extent_start != 0) {
3403                                         found_extent = 1;
3404                                         if (root->ref_cows)
3405                                                 inode_sub_bytes(inode, num_dec);
3406                                 }
3407                         }
3408                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3409                         /*
3410                          * we can't truncate inline items that have had
3411                          * special encodings
3412                          */
3413                         if (!del_item &&
3414                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3415                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3416                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3417                                 u32 size = new_size - found_key.offset;
3418
3419                                 if (root->ref_cows) {
3420                                         inode_sub_bytes(inode, item_end + 1 -
3421                                                         new_size);
3422                                 }
3423                                 size =
3424                                     btrfs_file_extent_calc_inline_size(size);
3425                                 btrfs_truncate_item(trans, root, path,
3426                                                     size, 1);
3427                         } else if (root->ref_cows) {
3428                                 inode_sub_bytes(inode, item_end + 1 -
3429                                                 found_key.offset);
3430                         }
3431                 }
3432 delete:
3433                 if (del_item) {
3434                         if (!pending_del_nr) {
3435                                 /* no pending yet, add ourselves */
3436                                 pending_del_slot = path->slots[0];
3437                                 pending_del_nr = 1;
3438                         } else if (pending_del_nr &&
3439                                    path->slots[0] + 1 == pending_del_slot) {
3440                                 /* hop on the pending chunk */
3441                                 pending_del_nr++;
3442                                 pending_del_slot = path->slots[0];
3443                         } else {
3444                                 BUG();
3445                         }
3446                 } else {
3447                         break;
3448                 }
3449                 if (found_extent && (root->ref_cows ||
3450                                      root == root->fs_info->tree_root)) {
3451                         btrfs_set_path_blocking(path);
3452                         ret = btrfs_free_extent(trans, root, extent_start,
3453                                                 extent_num_bytes, 0,
3454                                                 btrfs_header_owner(leaf),
3455                                                 ino, extent_offset, 0);
3456                         BUG_ON(ret);
3457                 }
3458
3459                 if (found_type == BTRFS_INODE_ITEM_KEY)
3460                         break;
3461
3462                 if (path->slots[0] == 0 ||
3463                     path->slots[0] != pending_del_slot) {
3464                         if (pending_del_nr) {
3465                                 ret = btrfs_del_items(trans, root, path,
3466                                                 pending_del_slot,
3467                                                 pending_del_nr);
3468                                 if (ret) {
3469                                         btrfs_abort_transaction(trans,
3470                                                                 root, ret);
3471                                         goto error;
3472                                 }
3473                                 pending_del_nr = 0;
3474                         }
3475                         btrfs_release_path(path);
3476                         goto search_again;
3477                 } else {
3478                         path->slots[0]--;
3479                 }
3480         }
3481 out:
3482         if (pending_del_nr) {
3483                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3484                                       pending_del_nr);
3485                 if (ret)
3486                         btrfs_abort_transaction(trans, root, ret);
3487         }
3488 error:
3489         btrfs_free_path(path);
3490         return err;
3491 }
3492
3493 /*
3494  * btrfs_truncate_page - read, zero a chunk and write a page
3495  * @inode - inode that we're zeroing
3496  * @from - the offset to start zeroing
3497  * @len - the length to zero, 0 to zero the entire range respective to the
3498  *      offset
3499  * @front - zero up to the offset instead of from the offset on
3500  *
3501  * This will find the page for the "from" offset and cow the page and zero the
3502  * part we want to zero.  This is used with truncate and hole punching.
3503  */
3504 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3505                         int front)
3506 {
3507         struct address_space *mapping = inode->i_mapping;
3508         struct btrfs_root *root = BTRFS_I(inode)->root;
3509         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3510         struct btrfs_ordered_extent *ordered;
3511         struct extent_state *cached_state = NULL;
3512         char *kaddr;
3513         u32 blocksize = root->sectorsize;
3514         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3515         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3516         struct page *page;
3517         gfp_t mask = btrfs_alloc_write_mask(mapping);
3518         int ret = 0;
3519         u64 page_start;
3520         u64 page_end;
3521
3522         if ((offset & (blocksize - 1)) == 0 &&
3523             (!len || ((len & (blocksize - 1)) == 0)))
3524                 goto out;
3525         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3526         if (ret)
3527                 goto out;
3528
3529 again:
3530         page = find_or_create_page(mapping, index, mask);
3531         if (!page) {
3532                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3533                 ret = -ENOMEM;
3534                 goto out;
3535         }
3536
3537         page_start = page_offset(page);
3538         page_end = page_start + PAGE_CACHE_SIZE - 1;
3539
3540         if (!PageUptodate(page)) {
3541                 ret = btrfs_readpage(NULL, page);
3542                 lock_page(page);
3543                 if (page->mapping != mapping) {
3544                         unlock_page(page);
3545                         page_cache_release(page);
3546                         goto again;
3547                 }
3548                 if (!PageUptodate(page)) {
3549                         ret = -EIO;
3550                         goto out_unlock;
3551                 }
3552         }
3553         wait_on_page_writeback(page);
3554
3555         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3556         set_page_extent_mapped(page);
3557
3558         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3559         if (ordered) {
3560                 unlock_extent_cached(io_tree, page_start, page_end,
3561                                      &cached_state, GFP_NOFS);
3562                 unlock_page(page);
3563                 page_cache_release(page);
3564                 btrfs_start_ordered_extent(inode, ordered, 1);
3565                 btrfs_put_ordered_extent(ordered);
3566                 goto again;
3567         }
3568
3569         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3570                           EXTENT_DIRTY | EXTENT_DELALLOC |
3571                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3572                           0, 0, &cached_state, GFP_NOFS);
3573
3574         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3575                                         &cached_state);
3576         if (ret) {
3577                 unlock_extent_cached(io_tree, page_start, page_end,
3578                                      &cached_state, GFP_NOFS);
3579                 goto out_unlock;
3580         }
3581
3582         if (offset != PAGE_CACHE_SIZE) {
3583                 if (!len)
3584                         len = PAGE_CACHE_SIZE - offset;
3585                 kaddr = kmap(page);
3586                 if (front)
3587                         memset(kaddr, 0, offset);
3588                 else
3589                         memset(kaddr + offset, 0, len);
3590                 flush_dcache_page(page);
3591                 kunmap(page);
3592         }
3593         ClearPageChecked(page);
3594         set_page_dirty(page);
3595         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3596                              GFP_NOFS);
3597
3598 out_unlock:
3599         if (ret)
3600                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3601         unlock_page(page);
3602         page_cache_release(page);
3603 out:
3604         return ret;
3605 }
3606
3607 /*
3608  * This function puts in dummy file extents for the area we're creating a hole
3609  * for.  So if we are truncating this file to a larger size we need to insert
3610  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3611  * the range between oldsize and size
3612  */
3613 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3614 {
3615         struct btrfs_trans_handle *trans;
3616         struct btrfs_root *root = BTRFS_I(inode)->root;
3617         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3618         struct extent_map *em = NULL;
3619         struct extent_state *cached_state = NULL;
3620         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3621         u64 mask = root->sectorsize - 1;
3622         u64 hole_start = (oldsize + mask) & ~mask;
3623         u64 block_end = (size + mask) & ~mask;
3624         u64 last_byte;
3625         u64 cur_offset;
3626         u64 hole_size;
3627         int err = 0;
3628
3629         if (size <= hole_start)
3630                 return 0;
3631
3632         while (1) {
3633                 struct btrfs_ordered_extent *ordered;
3634                 btrfs_wait_ordered_range(inode, hole_start,
3635                                          block_end - hole_start);
3636                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3637                                  &cached_state);
3638                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3639                 if (!ordered)
3640                         break;
3641                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3642                                      &cached_state, GFP_NOFS);
3643                 btrfs_put_ordered_extent(ordered);
3644         }
3645
3646         cur_offset = hole_start;
3647         while (1) {
3648                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3649                                 block_end - cur_offset, 0);
3650                 if (IS_ERR(em)) {
3651                         err = PTR_ERR(em);
3652                         break;
3653                 }
3654                 last_byte = min(extent_map_end(em), block_end);
3655                 last_byte = (last_byte + mask) & ~mask;
3656                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3657                         struct extent_map *hole_em;
3658                         hole_size = last_byte - cur_offset;
3659
3660                         trans = btrfs_start_transaction(root, 3);
3661                         if (IS_ERR(trans)) {
3662                                 err = PTR_ERR(trans);
3663                                 break;
3664                         }
3665
3666                         err = btrfs_drop_extents(trans, root, inode,
3667                                                  cur_offset,
3668                                                  cur_offset + hole_size, 1);
3669                         if (err) {
3670                                 btrfs_abort_transaction(trans, root, err);
3671                                 btrfs_end_transaction(trans, root);
3672                                 break;
3673                         }
3674
3675                         err = btrfs_insert_file_extent(trans, root,
3676                                         btrfs_ino(inode), cur_offset, 0,
3677                                         0, hole_size, 0, hole_size,
3678                                         0, 0, 0);
3679                         if (err) {
3680                                 btrfs_abort_transaction(trans, root, err);
3681                                 btrfs_end_transaction(trans, root);
3682                                 break;
3683                         }
3684
3685                         btrfs_drop_extent_cache(inode, cur_offset,
3686                                                 cur_offset + hole_size - 1, 0);
3687                         hole_em = alloc_extent_map();
3688                         if (!hole_em) {
3689                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3690                                         &BTRFS_I(inode)->runtime_flags);
3691                                 goto next;
3692                         }
3693                         hole_em->start = cur_offset;
3694                         hole_em->len = hole_size;
3695                         hole_em->orig_start = cur_offset;
3696
3697                         hole_em->block_start = EXTENT_MAP_HOLE;
3698                         hole_em->block_len = 0;
3699                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3700                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3701                         hole_em->generation = trans->transid;
3702
3703                         while (1) {
3704                                 write_lock(&em_tree->lock);
3705                                 err = add_extent_mapping(em_tree, hole_em);
3706                                 if (!err)
3707                                         list_move(&hole_em->list,
3708                                                   &em_tree->modified_extents);
3709                                 write_unlock(&em_tree->lock);
3710                                 if (err != -EEXIST)
3711                                         break;
3712                                 btrfs_drop_extent_cache(inode, cur_offset,
3713                                                         cur_offset +
3714                                                         hole_size - 1, 0);
3715                         }
3716                         free_extent_map(hole_em);
3717 next:
3718                         btrfs_update_inode(trans, root, inode);
3719                         btrfs_end_transaction(trans, root);
3720                 }
3721                 free_extent_map(em);
3722                 em = NULL;
3723                 cur_offset = last_byte;
3724                 if (cur_offset >= block_end)
3725                         break;
3726         }
3727
3728         free_extent_map(em);
3729         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3730                              GFP_NOFS);
3731         return err;
3732 }
3733
3734 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3735 {
3736         struct btrfs_root *root = BTRFS_I(inode)->root;
3737         struct btrfs_trans_handle *trans;
3738         loff_t oldsize = i_size_read(inode);
3739         int ret;
3740
3741         if (newsize == oldsize)
3742                 return 0;
3743
3744         if (newsize > oldsize) {
3745                 truncate_pagecache(inode, oldsize, newsize);
3746                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3747                 if (ret)
3748                         return ret;
3749
3750                 trans = btrfs_start_transaction(root, 1);
3751                 if (IS_ERR(trans))
3752                         return PTR_ERR(trans);
3753
3754                 i_size_write(inode, newsize);
3755                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3756                 ret = btrfs_update_inode(trans, root, inode);
3757                 btrfs_end_transaction(trans, root);
3758         } else {
3759
3760                 /*
3761                  * We're truncating a file that used to have good data down to
3762                  * zero. Make sure it gets into the ordered flush list so that
3763                  * any new writes get down to disk quickly.
3764                  */
3765                 if (newsize == 0)
3766                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3767                                 &BTRFS_I(inode)->runtime_flags);
3768
3769                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3770                 truncate_setsize(inode, newsize);
3771                 ret = btrfs_truncate(inode);
3772         }
3773
3774         return ret;
3775 }
3776
3777 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3778 {
3779         struct inode *inode = dentry->d_inode;
3780         struct btrfs_root *root = BTRFS_I(inode)->root;
3781         int err;
3782
3783         if (btrfs_root_readonly(root))
3784                 return -EROFS;
3785
3786         err = inode_change_ok(inode, attr);
3787         if (err)
3788                 return err;
3789
3790         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3791                 err = btrfs_setsize(inode, attr->ia_size);
3792                 if (err)
3793                         return err;
3794         }
3795
3796         if (attr->ia_valid) {
3797                 setattr_copy(inode, attr);
3798                 inode_inc_iversion(inode);
3799                 err = btrfs_dirty_inode(inode);
3800
3801                 if (!err && attr->ia_valid & ATTR_MODE)
3802                         err = btrfs_acl_chmod(inode);
3803         }
3804
3805         return err;
3806 }
3807
3808 void btrfs_evict_inode(struct inode *inode)
3809 {
3810         struct btrfs_trans_handle *trans;
3811         struct btrfs_root *root = BTRFS_I(inode)->root;
3812         struct btrfs_block_rsv *rsv, *global_rsv;
3813         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3814         int ret;
3815
3816         trace_btrfs_inode_evict(inode);
3817
3818         truncate_inode_pages(&inode->i_data, 0);
3819         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3820                                btrfs_is_free_space_inode(inode)))
3821                 goto no_delete;
3822
3823         if (is_bad_inode(inode)) {
3824                 btrfs_orphan_del(NULL, inode);
3825                 goto no_delete;
3826         }
3827         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3828         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3829
3830         if (root->fs_info->log_root_recovering) {
3831                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3832                                  &BTRFS_I(inode)->runtime_flags));
3833                 goto no_delete;
3834         }
3835
3836         if (inode->i_nlink > 0) {
3837                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3838                 goto no_delete;
3839         }
3840
3841         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3842         if (!rsv) {
3843                 btrfs_orphan_del(NULL, inode);
3844                 goto no_delete;
3845         }
3846         rsv->size = min_size;
3847         rsv->failfast = 1;
3848         global_rsv = &root->fs_info->global_block_rsv;
3849
3850         btrfs_i_size_write(inode, 0);
3851
3852         /*
3853          * This is a bit simpler than btrfs_truncate since we've already
3854          * reserved our space for our orphan item in the unlink, so we just
3855          * need to reserve some slack space in case we add bytes and update
3856          * inode item when doing the truncate.
3857          */
3858         while (1) {
3859                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3860                                              BTRFS_RESERVE_FLUSH_LIMIT);
3861
3862                 /*
3863                  * Try and steal from the global reserve since we will
3864                  * likely not use this space anyway, we want to try as
3865                  * hard as possible to get this to work.
3866                  */
3867                 if (ret)
3868                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3869
3870                 if (ret) {
3871                         printk(KERN_WARNING "Could not get space for a "
3872                                "delete, will truncate on mount %d\n", ret);
3873                         btrfs_orphan_del(NULL, inode);
3874                         btrfs_free_block_rsv(root, rsv);
3875                         goto no_delete;
3876                 }
3877
3878                 trans = btrfs_start_transaction_lflush(root, 1);
3879                 if (IS_ERR(trans)) {
3880                         btrfs_orphan_del(NULL, inode);
3881                         btrfs_free_block_rsv(root, rsv);
3882                         goto no_delete;
3883                 }
3884
3885                 trans->block_rsv = rsv;
3886
3887                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3888                 if (ret != -ENOSPC)
3889                         break;
3890
3891                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3892                 ret = btrfs_update_inode(trans, root, inode);
3893                 BUG_ON(ret);
3894
3895                 btrfs_end_transaction(trans, root);
3896                 trans = NULL;
3897                 btrfs_btree_balance_dirty(root);
3898         }
3899
3900         btrfs_free_block_rsv(root, rsv);
3901
3902         if (ret == 0) {
3903                 trans->block_rsv = root->orphan_block_rsv;
3904                 ret = btrfs_orphan_del(trans, inode);
3905                 BUG_ON(ret);
3906         }
3907
3908         trans->block_rsv = &root->fs_info->trans_block_rsv;
3909         if (!(root == root->fs_info->tree_root ||
3910               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3911                 btrfs_return_ino(root, btrfs_ino(inode));
3912
3913         btrfs_end_transaction(trans, root);
3914         btrfs_btree_balance_dirty(root);
3915 no_delete:
3916         clear_inode(inode);
3917         return;
3918 }
3919
3920 /*
3921  * this returns the key found in the dir entry in the location pointer.
3922  * If no dir entries were found, location->objectid is 0.
3923  */
3924 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3925                                struct btrfs_key *location)
3926 {
3927         const char *name = dentry->d_name.name;
3928         int namelen = dentry->d_name.len;
3929         struct btrfs_dir_item *di;
3930         struct btrfs_path *path;
3931         struct btrfs_root *root = BTRFS_I(dir)->root;
3932         int ret = 0;
3933
3934         path = btrfs_alloc_path();
3935         if (!path)
3936                 return -ENOMEM;
3937
3938         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3939                                     namelen, 0);
3940         if (IS_ERR(di))
3941                 ret = PTR_ERR(di);
3942
3943         if (IS_ERR_OR_NULL(di))
3944                 goto out_err;
3945
3946         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3947 out:
3948         btrfs_free_path(path);
3949         return ret;
3950 out_err:
3951         location->objectid = 0;
3952         goto out;
3953 }
3954
3955 /*
3956  * when we hit a tree root in a directory, the btrfs part of the inode
3957  * needs to be changed to reflect the root directory of the tree root.  This
3958  * is kind of like crossing a mount point.
3959  */
3960 static int fixup_tree_root_location(struct btrfs_root *root,
3961                                     struct inode *dir,
3962                                     struct dentry *dentry,
3963                                     struct btrfs_key *location,
3964                                     struct btrfs_root **sub_root)
3965 {
3966         struct btrfs_path *path;
3967         struct btrfs_root *new_root;
3968         struct btrfs_root_ref *ref;
3969         struct extent_buffer *leaf;
3970         int ret;
3971         int err = 0;
3972
3973         path = btrfs_alloc_path();
3974         if (!path) {
3975                 err = -ENOMEM;
3976                 goto out;
3977         }
3978
3979         err = -ENOENT;
3980         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3981                                   BTRFS_I(dir)->root->root_key.objectid,
3982                                   location->objectid);
3983         if (ret) {
3984                 if (ret < 0)
3985                         err = ret;
3986                 goto out;
3987         }
3988
3989         leaf = path->nodes[0];
3990         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3991         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3992             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3993                 goto out;
3994
3995         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3996                                    (unsigned long)(ref + 1),
3997                                    dentry->d_name.len);
3998         if (ret)
3999                 goto out;
4000
4001         btrfs_release_path(path);
4002
4003         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4004         if (IS_ERR(new_root)) {
4005                 err = PTR_ERR(new_root);
4006                 goto out;
4007         }
4008
4009         if (btrfs_root_refs(&new_root->root_item) == 0) {
4010                 err = -ENOENT;
4011                 goto out;
4012         }
4013
4014         *sub_root = new_root;
4015         location->objectid = btrfs_root_dirid(&new_root->root_item);
4016         location->type = BTRFS_INODE_ITEM_KEY;
4017         location->offset = 0;
4018         err = 0;
4019 out:
4020         btrfs_free_path(path);
4021         return err;
4022 }
4023
4024 static void inode_tree_add(struct inode *inode)
4025 {
4026         struct btrfs_root *root = BTRFS_I(inode)->root;
4027         struct btrfs_inode *entry;
4028         struct rb_node **p;
4029         struct rb_node *parent;
4030         u64 ino = btrfs_ino(inode);
4031 again:
4032         p = &root->inode_tree.rb_node;
4033         parent = NULL;
4034
4035         if (inode_unhashed(inode))
4036                 return;
4037
4038         spin_lock(&root->inode_lock);
4039         while (*p) {
4040                 parent = *p;
4041                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4042
4043                 if (ino < btrfs_ino(&entry->vfs_inode))
4044                         p = &parent->rb_left;
4045                 else if (ino > btrfs_ino(&entry->vfs_inode))
4046                         p = &parent->rb_right;
4047                 else {
4048                         WARN_ON(!(entry->vfs_inode.i_state &
4049                                   (I_WILL_FREE | I_FREEING)));
4050                         rb_erase(parent, &root->inode_tree);
4051                         RB_CLEAR_NODE(parent);
4052                         spin_unlock(&root->inode_lock);
4053                         goto again;
4054                 }
4055         }
4056         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4057         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4058         spin_unlock(&root->inode_lock);
4059 }
4060
4061 static void inode_tree_del(struct inode *inode)
4062 {
4063         struct btrfs_root *root = BTRFS_I(inode)->root;
4064         int empty = 0;
4065
4066         spin_lock(&root->inode_lock);
4067         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4068                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4069                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4070                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4071         }
4072         spin_unlock(&root->inode_lock);
4073
4074         /*
4075          * Free space cache has inodes in the tree root, but the tree root has a
4076          * root_refs of 0, so this could end up dropping the tree root as a
4077          * snapshot, so we need the extra !root->fs_info->tree_root check to
4078          * make sure we don't drop it.
4079          */
4080         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4081             root != root->fs_info->tree_root) {
4082                 synchronize_srcu(&root->fs_info->subvol_srcu);
4083                 spin_lock(&root->inode_lock);
4084                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4085                 spin_unlock(&root->inode_lock);
4086                 if (empty)
4087                         btrfs_add_dead_root(root);
4088         }
4089 }
4090
4091 void btrfs_invalidate_inodes(struct btrfs_root *root)
4092 {
4093         struct rb_node *node;
4094         struct rb_node *prev;
4095         struct btrfs_inode *entry;
4096         struct inode *inode;
4097         u64 objectid = 0;
4098
4099         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4100
4101         spin_lock(&root->inode_lock);
4102 again:
4103         node = root->inode_tree.rb_node;
4104         prev = NULL;
4105         while (node) {
4106                 prev = node;
4107                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4108
4109                 if (objectid < btrfs_ino(&entry->vfs_inode))
4110                         node = node->rb_left;
4111                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4112                         node = node->rb_right;
4113                 else
4114                         break;
4115         }
4116         if (!node) {
4117                 while (prev) {
4118                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4119                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4120                                 node = prev;
4121                                 break;
4122                         }
4123                         prev = rb_next(prev);
4124                 }
4125         }
4126         while (node) {
4127                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4128                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4129                 inode = igrab(&entry->vfs_inode);
4130                 if (inode) {
4131                         spin_unlock(&root->inode_lock);
4132                         if (atomic_read(&inode->i_count) > 1)
4133                                 d_prune_aliases(inode);
4134                         /*
4135                          * btrfs_drop_inode will have it removed from
4136                          * the inode cache when its usage count
4137                          * hits zero.
4138                          */
4139                         iput(inode);
4140                         cond_resched();
4141                         spin_lock(&root->inode_lock);
4142                         goto again;
4143                 }
4144
4145                 if (cond_resched_lock(&root->inode_lock))
4146                         goto again;
4147
4148                 node = rb_next(node);
4149         }
4150         spin_unlock(&root->inode_lock);
4151 }
4152
4153 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4154 {
4155         struct btrfs_iget_args *args = p;
4156         inode->i_ino = args->ino;
4157         BTRFS_I(inode)->root = args->root;
4158         return 0;
4159 }
4160
4161 static int btrfs_find_actor(struct inode *inode, void *opaque)
4162 {
4163         struct btrfs_iget_args *args = opaque;
4164         return args->ino == btrfs_ino(inode) &&
4165                 args->root == BTRFS_I(inode)->root;
4166 }
4167
4168 static struct inode *btrfs_iget_locked(struct super_block *s,
4169                                        u64 objectid,
4170                                        struct btrfs_root *root)
4171 {
4172         struct inode *inode;
4173         struct btrfs_iget_args args;
4174         args.ino = objectid;
4175         args.root = root;
4176
4177         inode = iget5_locked(s, objectid, btrfs_find_actor,
4178                              btrfs_init_locked_inode,
4179                              (void *)&args);
4180         return inode;
4181 }
4182
4183 /* Get an inode object given its location and corresponding root.
4184  * Returns in *is_new if the inode was read from disk
4185  */
4186 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4187                          struct btrfs_root *root, int *new)
4188 {
4189         struct inode *inode;
4190
4191         inode = btrfs_iget_locked(s, location->objectid, root);
4192         if (!inode)
4193                 return ERR_PTR(-ENOMEM);
4194
4195         if (inode->i_state & I_NEW) {
4196                 BTRFS_I(inode)->root = root;
4197                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4198                 btrfs_read_locked_inode(inode);
4199                 if (!is_bad_inode(inode)) {
4200                         inode_tree_add(inode);
4201                         unlock_new_inode(inode);
4202                         if (new)
4203                                 *new = 1;
4204                 } else {
4205                         unlock_new_inode(inode);
4206                         iput(inode);
4207                         inode = ERR_PTR(-ESTALE);
4208                 }
4209         }
4210
4211         return inode;
4212 }
4213
4214 static struct inode *new_simple_dir(struct super_block *s,
4215                                     struct btrfs_key *key,
4216                                     struct btrfs_root *root)
4217 {
4218         struct inode *inode = new_inode(s);
4219
4220         if (!inode)
4221                 return ERR_PTR(-ENOMEM);
4222
4223         BTRFS_I(inode)->root = root;
4224         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4225         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4226
4227         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4228         inode->i_op = &btrfs_dir_ro_inode_operations;
4229         inode->i_fop = &simple_dir_operations;
4230         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4231         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4232
4233         return inode;
4234 }
4235
4236 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4237 {
4238         struct inode *inode;
4239         struct btrfs_root *root = BTRFS_I(dir)->root;
4240         struct btrfs_root *sub_root = root;
4241         struct btrfs_key location;
4242         int index;
4243         int ret = 0;
4244
4245         if (dentry->d_name.len > BTRFS_NAME_LEN)
4246                 return ERR_PTR(-ENAMETOOLONG);
4247
4248         if (unlikely(d_need_lookup(dentry))) {
4249                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4250                 kfree(dentry->d_fsdata);
4251                 dentry->d_fsdata = NULL;
4252                 /* This thing is hashed, drop it for now */
4253                 d_drop(dentry);
4254         } else {
4255                 ret = btrfs_inode_by_name(dir, dentry, &location);
4256         }
4257
4258         if (ret < 0)
4259                 return ERR_PTR(ret);
4260
4261         if (location.objectid == 0)
4262                 return NULL;
4263
4264         if (location.type == BTRFS_INODE_ITEM_KEY) {
4265                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4266                 return inode;
4267         }
4268
4269         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4270
4271         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4272         ret = fixup_tree_root_location(root, dir, dentry,
4273                                        &location, &sub_root);
4274         if (ret < 0) {
4275                 if (ret != -ENOENT)
4276                         inode = ERR_PTR(ret);
4277                 else
4278                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4279         } else {
4280                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4281         }
4282         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4283
4284         if (!IS_ERR(inode) && root != sub_root) {
4285                 down_read(&root->fs_info->cleanup_work_sem);
4286                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4287                         ret = btrfs_orphan_cleanup(sub_root);
4288                 up_read(&root->fs_info->cleanup_work_sem);
4289                 if (ret)
4290                         inode = ERR_PTR(ret);
4291         }
4292
4293         return inode;
4294 }
4295
4296 static int btrfs_dentry_delete(const struct dentry *dentry)
4297 {
4298         struct btrfs_root *root;
4299         struct inode *inode = dentry->d_inode;
4300
4301         if (!inode && !IS_ROOT(dentry))
4302                 inode = dentry->d_parent->d_inode;
4303
4304         if (inode) {
4305                 root = BTRFS_I(inode)->root;
4306                 if (btrfs_root_refs(&root->root_item) == 0)
4307                         return 1;
4308
4309                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4310                         return 1;
4311         }
4312         return 0;
4313 }
4314
4315 static void btrfs_dentry_release(struct dentry *dentry)
4316 {
4317         if (dentry->d_fsdata)
4318                 kfree(dentry->d_fsdata);
4319 }
4320
4321 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4322                                    unsigned int flags)
4323 {
4324         struct dentry *ret;
4325
4326         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4327         if (unlikely(d_need_lookup(dentry))) {
4328                 spin_lock(&dentry->d_lock);
4329                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4330                 spin_unlock(&dentry->d_lock);
4331         }
4332         return ret;
4333 }
4334
4335 unsigned char btrfs_filetype_table[] = {
4336         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4337 };
4338
4339 static int btrfs_real_readdir(struct file *filp, void *dirent,
4340                               filldir_t filldir)
4341 {
4342         struct inode *inode = filp->f_dentry->d_inode;
4343         struct btrfs_root *root = BTRFS_I(inode)->root;
4344         struct btrfs_item *item;
4345         struct btrfs_dir_item *di;
4346         struct btrfs_key key;
4347         struct btrfs_key found_key;
4348         struct btrfs_path *path;
4349         struct list_head ins_list;
4350         struct list_head del_list;
4351         int ret;
4352         struct extent_buffer *leaf;
4353         int slot;
4354         unsigned char d_type;
4355         int over = 0;
4356         u32 di_cur;
4357         u32 di_total;
4358         u32 di_len;
4359         int key_type = BTRFS_DIR_INDEX_KEY;
4360         char tmp_name[32];
4361         char *name_ptr;
4362         int name_len;
4363         int is_curr = 0;        /* filp->f_pos points to the current index? */
4364
4365         /* FIXME, use a real flag for deciding about the key type */
4366         if (root->fs_info->tree_root == root)
4367                 key_type = BTRFS_DIR_ITEM_KEY;
4368
4369         /* special case for "." */
4370         if (filp->f_pos == 0) {
4371                 over = filldir(dirent, ".", 1,
4372                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4373                 if (over)
4374                         return 0;
4375                 filp->f_pos = 1;
4376         }
4377         /* special case for .., just use the back ref */
4378         if (filp->f_pos == 1) {
4379                 u64 pino = parent_ino(filp->f_path.dentry);
4380                 over = filldir(dirent, "..", 2,
4381                                filp->f_pos, pino, DT_DIR);
4382                 if (over)
4383                         return 0;
4384                 filp->f_pos = 2;
4385         }
4386         path = btrfs_alloc_path();
4387         if (!path)
4388                 return -ENOMEM;
4389
4390         path->reada = 1;
4391
4392         if (key_type == BTRFS_DIR_INDEX_KEY) {
4393                 INIT_LIST_HEAD(&ins_list);
4394                 INIT_LIST_HEAD(&del_list);
4395                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4396         }
4397
4398         btrfs_set_key_type(&key, key_type);
4399         key.offset = filp->f_pos;
4400         key.objectid = btrfs_ino(inode);
4401
4402         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4403         if (ret < 0)
4404                 goto err;
4405
4406         while (1) {
4407                 leaf = path->nodes[0];
4408                 slot = path->slots[0];
4409                 if (slot >= btrfs_header_nritems(leaf)) {
4410                         ret = btrfs_next_leaf(root, path);
4411                         if (ret < 0)
4412                                 goto err;
4413                         else if (ret > 0)
4414                                 break;
4415                         continue;
4416                 }
4417
4418                 item = btrfs_item_nr(leaf, slot);
4419                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4420
4421                 if (found_key.objectid != key.objectid)
4422                         break;
4423                 if (btrfs_key_type(&found_key) != key_type)
4424                         break;
4425                 if (found_key.offset < filp->f_pos)
4426                         goto next;
4427                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4428                     btrfs_should_delete_dir_index(&del_list,
4429                                                   found_key.offset))
4430                         goto next;
4431
4432                 filp->f_pos = found_key.offset;
4433                 is_curr = 1;
4434
4435                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4436                 di_cur = 0;
4437                 di_total = btrfs_item_size(leaf, item);
4438
4439                 while (di_cur < di_total) {
4440                         struct btrfs_key location;
4441
4442                         if (verify_dir_item(root, leaf, di))
4443                                 break;
4444
4445                         name_len = btrfs_dir_name_len(leaf, di);
4446                         if (name_len <= sizeof(tmp_name)) {
4447                                 name_ptr = tmp_name;
4448                         } else {
4449                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4450                                 if (!name_ptr) {
4451                                         ret = -ENOMEM;
4452                                         goto err;
4453                                 }
4454                         }
4455                         read_extent_buffer(leaf, name_ptr,
4456                                            (unsigned long)(di + 1), name_len);
4457
4458                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4459                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4460
4461
4462                         /* is this a reference to our own snapshot? If so
4463                          * skip it.
4464                          *
4465                          * In contrast to old kernels, we insert the snapshot's
4466                          * dir item and dir index after it has been created, so
4467                          * we won't find a reference to our own snapshot. We
4468                          * still keep the following code for backward
4469                          * compatibility.
4470                          */
4471                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4472                             location.objectid == root->root_key.objectid) {
4473                                 over = 0;
4474                                 goto skip;
4475                         }
4476                         over = filldir(dirent, name_ptr, name_len,
4477                                        found_key.offset, location.objectid,
4478                                        d_type);
4479
4480 skip:
4481                         if (name_ptr != tmp_name)
4482                                 kfree(name_ptr);
4483
4484                         if (over)
4485                                 goto nopos;
4486                         di_len = btrfs_dir_name_len(leaf, di) +
4487                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4488                         di_cur += di_len;
4489                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4490                 }
4491 next:
4492                 path->slots[0]++;
4493         }
4494
4495         if (key_type == BTRFS_DIR_INDEX_KEY) {
4496                 if (is_curr)
4497                         filp->f_pos++;
4498                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4499                                                       &ins_list);
4500                 if (ret)
4501                         goto nopos;
4502         }
4503
4504         /* Reached end of directory/root. Bump pos past the last item. */
4505         if (key_type == BTRFS_DIR_INDEX_KEY)
4506                 /*
4507                  * 32-bit glibc will use getdents64, but then strtol -
4508                  * so the last number we can serve is this.
4509                  */
4510                 filp->f_pos = 0x7fffffff;
4511         else
4512                 filp->f_pos++;
4513 nopos:
4514         ret = 0;
4515 err:
4516         if (key_type == BTRFS_DIR_INDEX_KEY)
4517                 btrfs_put_delayed_items(&ins_list, &del_list);
4518         btrfs_free_path(path);
4519         return ret;
4520 }
4521
4522 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4523 {
4524         struct btrfs_root *root = BTRFS_I(inode)->root;
4525         struct btrfs_trans_handle *trans;
4526         int ret = 0;
4527         bool nolock = false;
4528
4529         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4530                 return 0;
4531
4532         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4533                 nolock = true;
4534
4535         if (wbc->sync_mode == WB_SYNC_ALL) {
4536                 if (nolock)
4537                         trans = btrfs_join_transaction_nolock(root);
4538                 else
4539                         trans = btrfs_join_transaction(root);
4540                 if (IS_ERR(trans))
4541                         return PTR_ERR(trans);
4542                 ret = btrfs_commit_transaction(trans, root);
4543         }
4544         return ret;
4545 }
4546
4547 /*
4548  * This is somewhat expensive, updating the tree every time the
4549  * inode changes.  But, it is most likely to find the inode in cache.
4550  * FIXME, needs more benchmarking...there are no reasons other than performance
4551  * to keep or drop this code.
4552  */
4553 int btrfs_dirty_inode(struct inode *inode)
4554 {
4555         struct btrfs_root *root = BTRFS_I(inode)->root;
4556         struct btrfs_trans_handle *trans;
4557         int ret;
4558
4559         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4560                 return 0;
4561
4562         trans = btrfs_join_transaction(root);
4563         if (IS_ERR(trans))
4564                 return PTR_ERR(trans);
4565
4566         ret = btrfs_update_inode(trans, root, inode);
4567         if (ret && ret == -ENOSPC) {
4568                 /* whoops, lets try again with the full transaction */
4569                 btrfs_end_transaction(trans, root);
4570                 trans = btrfs_start_transaction(root, 1);
4571                 if (IS_ERR(trans))
4572                         return PTR_ERR(trans);
4573
4574                 ret = btrfs_update_inode(trans, root, inode);
4575         }
4576         btrfs_end_transaction(trans, root);
4577         if (BTRFS_I(inode)->delayed_node)
4578                 btrfs_balance_delayed_items(root);
4579
4580         return ret;
4581 }
4582
4583 /*
4584  * This is a copy of file_update_time.  We need this so we can return error on
4585  * ENOSPC for updating the inode in the case of file write and mmap writes.
4586  */
4587 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4588                              int flags)
4589 {
4590         struct btrfs_root *root = BTRFS_I(inode)->root;
4591
4592         if (btrfs_root_readonly(root))
4593                 return -EROFS;
4594
4595         if (flags & S_VERSION)
4596                 inode_inc_iversion(inode);
4597         if (flags & S_CTIME)
4598                 inode->i_ctime = *now;
4599         if (flags & S_MTIME)
4600                 inode->i_mtime = *now;
4601         if (flags & S_ATIME)
4602                 inode->i_atime = *now;
4603         return btrfs_dirty_inode(inode);
4604 }
4605
4606 /*
4607  * find the highest existing sequence number in a directory
4608  * and then set the in-memory index_cnt variable to reflect
4609  * free sequence numbers
4610  */
4611 static int btrfs_set_inode_index_count(struct inode *inode)
4612 {
4613         struct btrfs_root *root = BTRFS_I(inode)->root;
4614         struct btrfs_key key, found_key;
4615         struct btrfs_path *path;
4616         struct extent_buffer *leaf;
4617         int ret;
4618
4619         key.objectid = btrfs_ino(inode);
4620         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4621         key.offset = (u64)-1;
4622
4623         path = btrfs_alloc_path();
4624         if (!path)
4625                 return -ENOMEM;
4626
4627         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4628         if (ret < 0)
4629                 goto out;
4630         /* FIXME: we should be able to handle this */
4631         if (ret == 0)
4632                 goto out;
4633         ret = 0;
4634
4635         /*
4636          * MAGIC NUMBER EXPLANATION:
4637          * since we search a directory based on f_pos we have to start at 2
4638          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4639          * else has to start at 2
4640          */
4641         if (path->slots[0] == 0) {
4642                 BTRFS_I(inode)->index_cnt = 2;
4643                 goto out;
4644         }
4645
4646         path->slots[0]--;
4647
4648         leaf = path->nodes[0];
4649         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4650
4651         if (found_key.objectid != btrfs_ino(inode) ||
4652             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4653                 BTRFS_I(inode)->index_cnt = 2;
4654                 goto out;
4655         }
4656
4657         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4658 out:
4659         btrfs_free_path(path);
4660         return ret;
4661 }
4662
4663 /*
4664  * helper to find a free sequence number in a given directory.  This current
4665  * code is very simple, later versions will do smarter things in the btree
4666  */
4667 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4668 {
4669         int ret = 0;
4670
4671         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4672                 ret = btrfs_inode_delayed_dir_index_count(dir);
4673                 if (ret) {
4674                         ret = btrfs_set_inode_index_count(dir);
4675                         if (ret)
4676                                 return ret;
4677                 }
4678         }
4679
4680         *index = BTRFS_I(dir)->index_cnt;
4681         BTRFS_I(dir)->index_cnt++;
4682
4683         return ret;
4684 }
4685
4686 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4687                                      struct btrfs_root *root,
4688                                      struct inode *dir,
4689                                      const char *name, int name_len,
4690                                      u64 ref_objectid, u64 objectid,
4691                                      umode_t mode, u64 *index)
4692 {
4693         struct inode *inode;
4694         struct btrfs_inode_item *inode_item;
4695         struct btrfs_key *location;
4696         struct btrfs_path *path;
4697         struct btrfs_inode_ref *ref;
4698         struct btrfs_key key[2];
4699         u32 sizes[2];
4700         unsigned long ptr;
4701         int ret;
4702         int owner;
4703
4704         path = btrfs_alloc_path();
4705         if (!path)
4706                 return ERR_PTR(-ENOMEM);
4707
4708         inode = new_inode(root->fs_info->sb);
4709         if (!inode) {
4710                 btrfs_free_path(path);
4711                 return ERR_PTR(-ENOMEM);
4712         }
4713
4714         /*
4715          * we have to initialize this early, so we can reclaim the inode
4716          * number if we fail afterwards in this function.
4717          */
4718         inode->i_ino = objectid;
4719
4720         if (dir) {
4721                 trace_btrfs_inode_request(dir);
4722
4723                 ret = btrfs_set_inode_index(dir, index);
4724                 if (ret) {
4725                         btrfs_free_path(path);
4726                         iput(inode);
4727                         return ERR_PTR(ret);
4728                 }
4729         }
4730         /*
4731          * index_cnt is ignored for everything but a dir,
4732          * btrfs_get_inode_index_count has an explanation for the magic
4733          * number
4734          */
4735         BTRFS_I(inode)->index_cnt = 2;
4736         BTRFS_I(inode)->root = root;
4737         BTRFS_I(inode)->generation = trans->transid;
4738         inode->i_generation = BTRFS_I(inode)->generation;
4739
4740         /*
4741          * We could have gotten an inode number from somebody who was fsynced
4742          * and then removed in this same transaction, so let's just set full
4743          * sync since it will be a full sync anyway and this will blow away the
4744          * old info in the log.
4745          */
4746         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4747
4748         if (S_ISDIR(mode))
4749                 owner = 0;
4750         else
4751                 owner = 1;
4752
4753         key[0].objectid = objectid;
4754         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4755         key[0].offset = 0;
4756
4757         /*
4758          * Start new inodes with an inode_ref. This is slightly more
4759          * efficient for small numbers of hard links since they will
4760          * be packed into one item. Extended refs will kick in if we
4761          * add more hard links than can fit in the ref item.
4762          */
4763         key[1].objectid = objectid;
4764         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4765         key[1].offset = ref_objectid;
4766
4767         sizes[0] = sizeof(struct btrfs_inode_item);
4768         sizes[1] = name_len + sizeof(*ref);
4769
4770         path->leave_spinning = 1;
4771         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4772         if (ret != 0)
4773                 goto fail;
4774
4775         inode_init_owner(inode, dir, mode);
4776         inode_set_bytes(inode, 0);
4777         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4778         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4779                                   struct btrfs_inode_item);
4780         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4781                              sizeof(*inode_item));
4782         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4783
4784         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4785                              struct btrfs_inode_ref);
4786         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4787         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4788         ptr = (unsigned long)(ref + 1);
4789         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4790
4791         btrfs_mark_buffer_dirty(path->nodes[0]);
4792         btrfs_free_path(path);
4793
4794         location = &BTRFS_I(inode)->location;
4795         location->objectid = objectid;
4796         location->offset = 0;
4797         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4798
4799         btrfs_inherit_iflags(inode, dir);
4800
4801         if (S_ISREG(mode)) {
4802                 if (btrfs_test_opt(root, NODATASUM))
4803                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4804                 if (btrfs_test_opt(root, NODATACOW) ||
4805                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4806                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4807         }
4808
4809         insert_inode_hash(inode);
4810         inode_tree_add(inode);
4811
4812         trace_btrfs_inode_new(inode);
4813         btrfs_set_inode_last_trans(trans, inode);
4814
4815         btrfs_update_root_times(trans, root);
4816
4817         return inode;
4818 fail:
4819         if (dir)
4820                 BTRFS_I(dir)->index_cnt--;
4821         btrfs_free_path(path);
4822         iput(inode);
4823         return ERR_PTR(ret);
4824 }
4825
4826 static inline u8 btrfs_inode_type(struct inode *inode)
4827 {
4828         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4829 }
4830
4831 /*
4832  * utility function to add 'inode' into 'parent_inode' with
4833  * a give name and a given sequence number.
4834  * if 'add_backref' is true, also insert a backref from the
4835  * inode to the parent directory.
4836  */
4837 int btrfs_add_link(struct btrfs_trans_handle *trans,
4838                    struct inode *parent_inode, struct inode *inode,
4839                    const char *name, int name_len, int add_backref, u64 index)
4840 {
4841         int ret = 0;
4842         struct btrfs_key key;
4843         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4844         u64 ino = btrfs_ino(inode);
4845         u64 parent_ino = btrfs_ino(parent_inode);
4846
4847         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4848                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4849         } else {
4850                 key.objectid = ino;
4851                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4852                 key.offset = 0;
4853         }
4854
4855         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4856                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4857                                          key.objectid, root->root_key.objectid,
4858                                          parent_ino, index, name, name_len);
4859         } else if (add_backref) {
4860                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4861                                              parent_ino, index);
4862         }
4863
4864         /* Nothing to clean up yet */
4865         if (ret)
4866                 return ret;
4867
4868         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4869                                     parent_inode, &key,
4870                                     btrfs_inode_type(inode), index);
4871         if (ret == -EEXIST)
4872                 goto fail_dir_item;
4873         else if (ret) {
4874                 btrfs_abort_transaction(trans, root, ret);
4875                 return ret;
4876         }
4877
4878         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4879                            name_len * 2);
4880         inode_inc_iversion(parent_inode);
4881         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4882         ret = btrfs_update_inode(trans, root, parent_inode);
4883         if (ret)
4884                 btrfs_abort_transaction(trans, root, ret);
4885         return ret;
4886
4887 fail_dir_item:
4888         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4889                 u64 local_index;
4890                 int err;
4891                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4892                                  key.objectid, root->root_key.objectid,
4893                                  parent_ino, &local_index, name, name_len);
4894
4895         } else if (add_backref) {
4896                 u64 local_index;
4897                 int err;
4898
4899                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4900                                           ino, parent_ino, &local_index);
4901         }
4902         return ret;
4903 }
4904
4905 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4906                             struct inode *dir, struct dentry *dentry,
4907                             struct inode *inode, int backref, u64 index)
4908 {
4909         int err = btrfs_add_link(trans, dir, inode,
4910                                  dentry->d_name.name, dentry->d_name.len,
4911                                  backref, index);
4912         if (err > 0)
4913                 err = -EEXIST;
4914         return err;
4915 }
4916
4917 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4918                         umode_t mode, dev_t rdev)
4919 {
4920         struct btrfs_trans_handle *trans;
4921         struct btrfs_root *root = BTRFS_I(dir)->root;
4922         struct inode *inode = NULL;
4923         int err;
4924         int drop_inode = 0;
4925         u64 objectid;
4926         u64 index = 0;
4927
4928         if (!new_valid_dev(rdev))
4929                 return -EINVAL;
4930
4931         /*
4932          * 2 for inode item and ref
4933          * 2 for dir items
4934          * 1 for xattr if selinux is on
4935          */
4936         trans = btrfs_start_transaction(root, 5);
4937         if (IS_ERR(trans))
4938                 return PTR_ERR(trans);
4939
4940         err = btrfs_find_free_ino(root, &objectid);
4941         if (err)
4942                 goto out_unlock;
4943
4944         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4945                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4946                                 mode, &index);
4947         if (IS_ERR(inode)) {
4948                 err = PTR_ERR(inode);
4949                 goto out_unlock;
4950         }
4951
4952         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4953         if (err) {
4954                 drop_inode = 1;
4955                 goto out_unlock;
4956         }
4957
4958         err = btrfs_update_inode(trans, root, inode);
4959         if (err) {
4960                 drop_inode = 1;
4961                 goto out_unlock;
4962         }
4963
4964         /*
4965         * If the active LSM wants to access the inode during
4966         * d_instantiate it needs these. Smack checks to see
4967         * if the filesystem supports xattrs by looking at the
4968         * ops vector.
4969         */
4970
4971         inode->i_op = &btrfs_special_inode_operations;
4972         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4973         if (err)
4974                 drop_inode = 1;
4975         else {
4976                 init_special_inode(inode, inode->i_mode, rdev);
4977                 btrfs_update_inode(trans, root, inode);
4978                 d_instantiate(dentry, inode);
4979         }
4980 out_unlock:
4981         btrfs_end_transaction(trans, root);
4982         btrfs_btree_balance_dirty(root);
4983         if (drop_inode) {
4984                 inode_dec_link_count(inode);
4985                 iput(inode);
4986         }
4987         return err;
4988 }
4989
4990 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4991                         umode_t mode, bool excl)
4992 {
4993         struct btrfs_trans_handle *trans;
4994         struct btrfs_root *root = BTRFS_I(dir)->root;
4995         struct inode *inode = NULL;
4996         int drop_inode_on_err = 0;
4997         int err;
4998         u64 objectid;
4999         u64 index = 0;
5000
5001         /*
5002          * 2 for inode item and ref
5003          * 2 for dir items
5004          * 1 for xattr if selinux is on
5005          */
5006         trans = btrfs_start_transaction(root, 5);
5007         if (IS_ERR(trans))
5008                 return PTR_ERR(trans);
5009
5010         err = btrfs_find_free_ino(root, &objectid);
5011         if (err)
5012                 goto out_unlock;
5013
5014         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5015                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5016                                 mode, &index);
5017         if (IS_ERR(inode)) {
5018                 err = PTR_ERR(inode);
5019                 goto out_unlock;
5020         }
5021         drop_inode_on_err = 1;
5022
5023         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5024         if (err)
5025                 goto out_unlock;
5026
5027         /*
5028         * If the active LSM wants to access the inode during
5029         * d_instantiate it needs these. Smack checks to see
5030         * if the filesystem supports xattrs by looking at the
5031         * ops vector.
5032         */
5033         inode->i_fop = &btrfs_file_operations;
5034         inode->i_op = &btrfs_file_inode_operations;
5035
5036         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5037         if (err)
5038                 goto out_unlock;
5039
5040         inode->i_mapping->a_ops = &btrfs_aops;
5041         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5042         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5043         d_instantiate(dentry, inode);
5044
5045 out_unlock:
5046         btrfs_end_transaction(trans, root);
5047         if (err && drop_inode_on_err) {
5048                 inode_dec_link_count(inode);
5049                 iput(inode);
5050         }
5051         btrfs_btree_balance_dirty(root);
5052         return err;
5053 }
5054
5055 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5056                       struct dentry *dentry)
5057 {
5058         struct btrfs_trans_handle *trans;
5059         struct btrfs_root *root = BTRFS_I(dir)->root;
5060         struct inode *inode = old_dentry->d_inode;
5061         u64 index;
5062         int err;
5063         int drop_inode = 0;
5064
5065         /* do not allow sys_link's with other subvols of the same device */
5066         if (root->objectid != BTRFS_I(inode)->root->objectid)
5067                 return -EXDEV;
5068
5069         if (inode->i_nlink >= BTRFS_LINK_MAX)
5070                 return -EMLINK;
5071
5072         err = btrfs_set_inode_index(dir, &index);
5073         if (err)
5074                 goto fail;
5075
5076         /*
5077          * 2 items for inode and inode ref
5078          * 2 items for dir items
5079          * 1 item for parent inode
5080          */
5081         trans = btrfs_start_transaction(root, 5);
5082         if (IS_ERR(trans)) {
5083                 err = PTR_ERR(trans);
5084                 goto fail;
5085         }
5086
5087         btrfs_inc_nlink(inode);
5088         inode_inc_iversion(inode);
5089         inode->i_ctime = CURRENT_TIME;
5090         ihold(inode);
5091         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5092
5093         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5094
5095         if (err) {
5096                 drop_inode = 1;
5097         } else {
5098                 struct dentry *parent = dentry->d_parent;
5099                 err = btrfs_update_inode(trans, root, inode);
5100                 if (err)
5101                         goto fail;
5102                 d_instantiate(dentry, inode);
5103                 btrfs_log_new_name(trans, inode, NULL, parent);
5104         }
5105
5106         btrfs_end_transaction(trans, root);
5107 fail:
5108         if (drop_inode) {
5109                 inode_dec_link_count(inode);
5110                 iput(inode);
5111         }
5112         btrfs_btree_balance_dirty(root);
5113         return err;
5114 }
5115
5116 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5117 {
5118         struct inode *inode = NULL;
5119         struct btrfs_trans_handle *trans;
5120         struct btrfs_root *root = BTRFS_I(dir)->root;
5121         int err = 0;
5122         int drop_on_err = 0;
5123         u64 objectid = 0;
5124         u64 index = 0;
5125
5126         /*
5127          * 2 items for inode and ref
5128          * 2 items for dir items
5129          * 1 for xattr if selinux is on
5130          */
5131         trans = btrfs_start_transaction(root, 5);
5132         if (IS_ERR(trans))
5133                 return PTR_ERR(trans);
5134
5135         err = btrfs_find_free_ino(root, &objectid);
5136         if (err)
5137                 goto out_fail;
5138
5139         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5140                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5141                                 S_IFDIR | mode, &index);
5142         if (IS_ERR(inode)) {
5143                 err = PTR_ERR(inode);
5144                 goto out_fail;
5145         }
5146
5147         drop_on_err = 1;
5148
5149         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5150         if (err)
5151                 goto out_fail;
5152
5153         inode->i_op = &btrfs_dir_inode_operations;
5154         inode->i_fop = &btrfs_dir_file_operations;
5155
5156         btrfs_i_size_write(inode, 0);
5157         err = btrfs_update_inode(trans, root, inode);
5158         if (err)
5159                 goto out_fail;
5160
5161         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5162                              dentry->d_name.len, 0, index);
5163         if (err)
5164                 goto out_fail;
5165
5166         d_instantiate(dentry, inode);
5167         drop_on_err = 0;
5168
5169 out_fail:
5170         btrfs_end_transaction(trans, root);
5171         if (drop_on_err)
5172                 iput(inode);
5173         btrfs_btree_balance_dirty(root);
5174         return err;
5175 }
5176
5177 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5178  * and an extent that you want to insert, deal with overlap and insert
5179  * the new extent into the tree.
5180  */
5181 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5182                                 struct extent_map *existing,
5183                                 struct extent_map *em,
5184                                 u64 map_start, u64 map_len)
5185 {
5186         u64 start_diff;
5187
5188         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5189         start_diff = map_start - em->start;
5190         em->start = map_start;
5191         em->len = map_len;
5192         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5193             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5194                 em->block_start += start_diff;
5195                 em->block_len -= start_diff;
5196         }
5197         return add_extent_mapping(em_tree, em);
5198 }
5199
5200 static noinline int uncompress_inline(struct btrfs_path *path,
5201                                       struct inode *inode, struct page *page,
5202                                       size_t pg_offset, u64 extent_offset,
5203                                       struct btrfs_file_extent_item *item)
5204 {
5205         int ret;
5206         struct extent_buffer *leaf = path->nodes[0];
5207         char *tmp;
5208         size_t max_size;
5209         unsigned long inline_size;
5210         unsigned long ptr;
5211         int compress_type;
5212
5213         WARN_ON(pg_offset != 0);
5214         compress_type = btrfs_file_extent_compression(leaf, item);
5215         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5216         inline_size = btrfs_file_extent_inline_item_len(leaf,
5217                                         btrfs_item_nr(leaf, path->slots[0]));
5218         tmp = kmalloc(inline_size, GFP_NOFS);
5219         if (!tmp)
5220                 return -ENOMEM;
5221         ptr = btrfs_file_extent_inline_start(item);
5222
5223         read_extent_buffer(leaf, tmp, ptr, inline_size);
5224
5225         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5226         ret = btrfs_decompress(compress_type, tmp, page,
5227                                extent_offset, inline_size, max_size);
5228         if (ret) {
5229                 char *kaddr = kmap_atomic(page);
5230                 unsigned long copy_size = min_t(u64,
5231                                   PAGE_CACHE_SIZE - pg_offset,
5232                                   max_size - extent_offset);
5233                 memset(kaddr + pg_offset, 0, copy_size);
5234                 kunmap_atomic(kaddr);
5235         }
5236         kfree(tmp);
5237         return 0;
5238 }
5239
5240 /*
5241  * a bit scary, this does extent mapping from logical file offset to the disk.
5242  * the ugly parts come from merging extents from the disk with the in-ram
5243  * representation.  This gets more complex because of the data=ordered code,
5244  * where the in-ram extents might be locked pending data=ordered completion.
5245  *
5246  * This also copies inline extents directly into the page.
5247  */
5248
5249 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5250                                     size_t pg_offset, u64 start, u64 len,
5251                                     int create)
5252 {
5253         int ret;
5254         int err = 0;
5255         u64 bytenr;
5256         u64 extent_start = 0;
5257         u64 extent_end = 0;
5258         u64 objectid = btrfs_ino(inode);
5259         u32 found_type;
5260         struct btrfs_path *path = NULL;
5261         struct btrfs_root *root = BTRFS_I(inode)->root;
5262         struct btrfs_file_extent_item *item;
5263         struct extent_buffer *leaf;
5264         struct btrfs_key found_key;
5265         struct extent_map *em = NULL;
5266         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5267         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5268         struct btrfs_trans_handle *trans = NULL;
5269         int compress_type;
5270
5271 again:
5272         read_lock(&em_tree->lock);
5273         em = lookup_extent_mapping(em_tree, start, len);
5274         if (em)
5275                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5276         read_unlock(&em_tree->lock);
5277
5278         if (em) {
5279                 if (em->start > start || em->start + em->len <= start)
5280                         free_extent_map(em);
5281                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5282                         free_extent_map(em);
5283                 else
5284                         goto out;
5285         }
5286         em = alloc_extent_map();
5287         if (!em) {
5288                 err = -ENOMEM;
5289                 goto out;
5290         }
5291         em->bdev = root->fs_info->fs_devices->latest_bdev;
5292         em->start = EXTENT_MAP_HOLE;
5293         em->orig_start = EXTENT_MAP_HOLE;
5294         em->len = (u64)-1;
5295         em->block_len = (u64)-1;
5296
5297         if (!path) {
5298                 path = btrfs_alloc_path();
5299                 if (!path) {
5300                         err = -ENOMEM;
5301                         goto out;
5302                 }
5303                 /*
5304                  * Chances are we'll be called again, so go ahead and do
5305                  * readahead
5306                  */
5307                 path->reada = 1;
5308         }
5309
5310         ret = btrfs_lookup_file_extent(trans, root, path,
5311                                        objectid, start, trans != NULL);
5312         if (ret < 0) {
5313                 err = ret;
5314                 goto out;
5315         }
5316
5317         if (ret != 0) {
5318                 if (path->slots[0] == 0)
5319                         goto not_found;
5320                 path->slots[0]--;
5321         }
5322
5323         leaf = path->nodes[0];
5324         item = btrfs_item_ptr(leaf, path->slots[0],
5325                               struct btrfs_file_extent_item);
5326         /* are we inside the extent that was found? */
5327         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5328         found_type = btrfs_key_type(&found_key);
5329         if (found_key.objectid != objectid ||
5330             found_type != BTRFS_EXTENT_DATA_KEY) {
5331                 goto not_found;
5332         }
5333
5334         found_type = btrfs_file_extent_type(leaf, item);
5335         extent_start = found_key.offset;
5336         compress_type = btrfs_file_extent_compression(leaf, item);
5337         if (found_type == BTRFS_FILE_EXTENT_REG ||
5338             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5339                 extent_end = extent_start +
5340                        btrfs_file_extent_num_bytes(leaf, item);
5341         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5342                 size_t size;
5343                 size = btrfs_file_extent_inline_len(leaf, item);
5344                 extent_end = (extent_start + size + root->sectorsize - 1) &
5345                         ~((u64)root->sectorsize - 1);
5346         }
5347
5348         if (start >= extent_end) {
5349                 path->slots[0]++;
5350                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5351                         ret = btrfs_next_leaf(root, path);
5352                         if (ret < 0) {
5353                                 err = ret;
5354                                 goto out;
5355                         }
5356                         if (ret > 0)
5357                                 goto not_found;
5358                         leaf = path->nodes[0];
5359                 }
5360                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5361                 if (found_key.objectid != objectid ||
5362                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5363                         goto not_found;
5364                 if (start + len <= found_key.offset)
5365                         goto not_found;
5366                 em->start = start;
5367                 em->len = found_key.offset - start;
5368                 goto not_found_em;
5369         }
5370
5371         if (found_type == BTRFS_FILE_EXTENT_REG ||
5372             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5373                 em->start = extent_start;
5374                 em->len = extent_end - extent_start;
5375                 em->orig_start = extent_start -
5376                                  btrfs_file_extent_offset(leaf, item);
5377                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5378                 if (bytenr == 0) {
5379                         em->block_start = EXTENT_MAP_HOLE;
5380                         goto insert;
5381                 }
5382                 if (compress_type != BTRFS_COMPRESS_NONE) {
5383                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5384                         em->compress_type = compress_type;
5385                         em->block_start = bytenr;
5386                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5387                                                                          item);
5388                 } else {
5389                         bytenr += btrfs_file_extent_offset(leaf, item);
5390                         em->block_start = bytenr;
5391                         em->block_len = em->len;
5392                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5393                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5394                 }
5395                 goto insert;
5396         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5397                 unsigned long ptr;
5398                 char *map;
5399                 size_t size;
5400                 size_t extent_offset;
5401                 size_t copy_size;
5402
5403                 em->block_start = EXTENT_MAP_INLINE;
5404                 if (!page || create) {
5405                         em->start = extent_start;
5406                         em->len = extent_end - extent_start;
5407                         goto out;
5408                 }
5409
5410                 size = btrfs_file_extent_inline_len(leaf, item);
5411                 extent_offset = page_offset(page) + pg_offset - extent_start;
5412                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5413                                 size - extent_offset);
5414                 em->start = extent_start + extent_offset;
5415                 em->len = (copy_size + root->sectorsize - 1) &
5416                         ~((u64)root->sectorsize - 1);
5417                 em->orig_start = EXTENT_MAP_INLINE;
5418                 if (compress_type) {
5419                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5420                         em->compress_type = compress_type;
5421                 }
5422                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5423                 if (create == 0 && !PageUptodate(page)) {
5424                         if (btrfs_file_extent_compression(leaf, item) !=
5425                             BTRFS_COMPRESS_NONE) {
5426                                 ret = uncompress_inline(path, inode, page,
5427                                                         pg_offset,
5428                                                         extent_offset, item);
5429                                 BUG_ON(ret); /* -ENOMEM */
5430                         } else {
5431                                 map = kmap(page);
5432                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5433                                                    copy_size);
5434                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5435                                         memset(map + pg_offset + copy_size, 0,
5436                                                PAGE_CACHE_SIZE - pg_offset -
5437                                                copy_size);
5438                                 }
5439                                 kunmap(page);
5440                         }
5441                         flush_dcache_page(page);
5442                 } else if (create && PageUptodate(page)) {
5443                         BUG();
5444                         if (!trans) {
5445                                 kunmap(page);
5446                                 free_extent_map(em);
5447                                 em = NULL;
5448
5449                                 btrfs_release_path(path);
5450                                 trans = btrfs_join_transaction(root);
5451
5452                                 if (IS_ERR(trans))
5453                                         return ERR_CAST(trans);
5454                                 goto again;
5455                         }
5456                         map = kmap(page);
5457                         write_extent_buffer(leaf, map + pg_offset, ptr,
5458                                             copy_size);
5459                         kunmap(page);
5460                         btrfs_mark_buffer_dirty(leaf);
5461                 }
5462                 set_extent_uptodate(io_tree, em->start,
5463                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5464                 goto insert;
5465         } else {
5466                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5467         }
5468 not_found:
5469         em->start = start;
5470         em->len = len;
5471 not_found_em:
5472         em->block_start = EXTENT_MAP_HOLE;
5473         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5474 insert:
5475         btrfs_release_path(path);
5476         if (em->start > start || extent_map_end(em) <= start) {
5477                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5478                        "[%llu %llu]\n", (unsigned long long)em->start,
5479                        (unsigned long long)em->len,
5480                        (unsigned long long)start,
5481                        (unsigned long long)len);
5482                 err = -EIO;
5483                 goto out;
5484         }
5485
5486         err = 0;
5487         write_lock(&em_tree->lock);
5488         ret = add_extent_mapping(em_tree, em);
5489         /* it is possible that someone inserted the extent into the tree
5490          * while we had the lock dropped.  It is also possible that
5491          * an overlapping map exists in the tree
5492          */
5493         if (ret == -EEXIST) {
5494                 struct extent_map *existing;
5495
5496                 ret = 0;
5497
5498                 existing = lookup_extent_mapping(em_tree, start, len);
5499                 if (existing && (existing->start > start ||
5500                     existing->start + existing->len <= start)) {
5501                         free_extent_map(existing);
5502                         existing = NULL;
5503                 }
5504                 if (!existing) {
5505                         existing = lookup_extent_mapping(em_tree, em->start,
5506                                                          em->len);
5507                         if (existing) {
5508                                 err = merge_extent_mapping(em_tree, existing,
5509                                                            em, start,
5510                                                            root->sectorsize);
5511                                 free_extent_map(existing);
5512                                 if (err) {
5513                                         free_extent_map(em);
5514                                         em = NULL;
5515                                 }
5516                         } else {
5517                                 err = -EIO;
5518                                 free_extent_map(em);
5519                                 em = NULL;
5520                         }
5521                 } else {
5522                         free_extent_map(em);
5523                         em = existing;
5524                         err = 0;
5525                 }
5526         }
5527         write_unlock(&em_tree->lock);
5528 out:
5529
5530         if (em)
5531                 trace_btrfs_get_extent(root, em);
5532
5533         if (path)
5534                 btrfs_free_path(path);
5535         if (trans) {
5536                 ret = btrfs_end_transaction(trans, root);
5537                 if (!err)
5538                         err = ret;
5539         }
5540         if (err) {
5541                 free_extent_map(em);
5542                 return ERR_PTR(err);
5543         }
5544         BUG_ON(!em); /* Error is always set */
5545         return em;
5546 }
5547
5548 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5549                                            size_t pg_offset, u64 start, u64 len,
5550                                            int create)
5551 {
5552         struct extent_map *em;
5553         struct extent_map *hole_em = NULL;
5554         u64 range_start = start;
5555         u64 end;
5556         u64 found;
5557         u64 found_end;
5558         int err = 0;
5559
5560         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5561         if (IS_ERR(em))
5562                 return em;
5563         if (em) {
5564                 /*
5565                  * if our em maps to a hole, there might
5566                  * actually be delalloc bytes behind it
5567                  */
5568                 if (em->block_start != EXTENT_MAP_HOLE)
5569                         return em;
5570                 else
5571                         hole_em = em;
5572         }
5573
5574         /* check to see if we've wrapped (len == -1 or similar) */
5575         end = start + len;
5576         if (end < start)
5577                 end = (u64)-1;
5578         else
5579                 end -= 1;
5580
5581         em = NULL;
5582
5583         /* ok, we didn't find anything, lets look for delalloc */
5584         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5585                                  end, len, EXTENT_DELALLOC, 1);
5586         found_end = range_start + found;
5587         if (found_end < range_start)
5588                 found_end = (u64)-1;
5589
5590         /*
5591          * we didn't find anything useful, return
5592          * the original results from get_extent()
5593          */
5594         if (range_start > end || found_end <= start) {
5595                 em = hole_em;
5596                 hole_em = NULL;
5597                 goto out;
5598         }
5599
5600         /* adjust the range_start to make sure it doesn't
5601          * go backwards from the start they passed in
5602          */
5603         range_start = max(start,range_start);
5604         found = found_end - range_start;
5605
5606         if (found > 0) {
5607                 u64 hole_start = start;
5608                 u64 hole_len = len;
5609
5610                 em = alloc_extent_map();
5611                 if (!em) {
5612                         err = -ENOMEM;
5613                         goto out;
5614                 }
5615                 /*
5616                  * when btrfs_get_extent can't find anything it
5617                  * returns one huge hole
5618                  *
5619                  * make sure what it found really fits our range, and
5620                  * adjust to make sure it is based on the start from
5621                  * the caller
5622                  */
5623                 if (hole_em) {
5624                         u64 calc_end = extent_map_end(hole_em);
5625
5626                         if (calc_end <= start || (hole_em->start > end)) {
5627                                 free_extent_map(hole_em);
5628                                 hole_em = NULL;
5629                         } else {
5630                                 hole_start = max(hole_em->start, start);
5631                                 hole_len = calc_end - hole_start;
5632                         }
5633                 }
5634                 em->bdev = NULL;
5635                 if (hole_em && range_start > hole_start) {
5636                         /* our hole starts before our delalloc, so we
5637                          * have to return just the parts of the hole
5638                          * that go until  the delalloc starts
5639                          */
5640                         em->len = min(hole_len,
5641                                       range_start - hole_start);
5642                         em->start = hole_start;
5643                         em->orig_start = hole_start;
5644                         /*
5645                          * don't adjust block start at all,
5646                          * it is fixed at EXTENT_MAP_HOLE
5647                          */
5648                         em->block_start = hole_em->block_start;
5649                         em->block_len = hole_len;
5650                 } else {
5651                         em->start = range_start;
5652                         em->len = found;
5653                         em->orig_start = range_start;
5654                         em->block_start = EXTENT_MAP_DELALLOC;
5655                         em->block_len = found;
5656                 }
5657         } else if (hole_em) {
5658                 return hole_em;
5659         }
5660 out:
5661
5662         free_extent_map(hole_em);
5663         if (err) {
5664                 free_extent_map(em);
5665                 return ERR_PTR(err);
5666         }
5667         return em;
5668 }
5669
5670 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5671                                                   struct extent_map *em,
5672                                                   u64 start, u64 len)
5673 {
5674         struct btrfs_root *root = BTRFS_I(inode)->root;
5675         struct btrfs_trans_handle *trans;
5676         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5677         struct btrfs_key ins;
5678         u64 alloc_hint;
5679         int ret;
5680         bool insert = false;
5681
5682         /*
5683          * Ok if the extent map we looked up is a hole and is for the exact
5684          * range we want, there is no reason to allocate a new one, however if
5685          * it is not right then we need to free this one and drop the cache for
5686          * our range.
5687          */
5688         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5689             em->len != len) {
5690                 free_extent_map(em);
5691                 em = NULL;
5692                 insert = true;
5693                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5694         }
5695
5696         trans = btrfs_join_transaction(root);
5697         if (IS_ERR(trans))
5698                 return ERR_CAST(trans);
5699
5700         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5701
5702         alloc_hint = get_extent_allocation_hint(inode, start, len);
5703         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5704                                    alloc_hint, &ins, 1);
5705         if (ret) {
5706                 em = ERR_PTR(ret);
5707                 goto out;
5708         }
5709
5710         if (!em) {
5711                 em = alloc_extent_map();
5712                 if (!em) {
5713                         em = ERR_PTR(-ENOMEM);
5714                         goto out;
5715                 }
5716         }
5717
5718         em->start = start;
5719         em->orig_start = em->start;
5720         em->len = ins.offset;
5721
5722         em->block_start = ins.objectid;
5723         em->block_len = ins.offset;
5724         em->bdev = root->fs_info->fs_devices->latest_bdev;
5725
5726         /*
5727          * We need to do this because if we're using the original em we searched
5728          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5729          */
5730         em->flags = 0;
5731         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5732
5733         while (insert) {
5734                 write_lock(&em_tree->lock);
5735                 ret = add_extent_mapping(em_tree, em);
5736                 write_unlock(&em_tree->lock);
5737                 if (ret != -EEXIST)
5738                         break;
5739                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5740         }
5741
5742         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5743                                            ins.offset, ins.offset, 0);
5744         if (ret) {
5745                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5746                 em = ERR_PTR(ret);
5747         }
5748 out:
5749         btrfs_end_transaction(trans, root);
5750         return em;
5751 }
5752
5753 /*
5754  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5755  * block must be cow'd
5756  */
5757 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5758                                       struct inode *inode, u64 offset, u64 len)
5759 {
5760         struct btrfs_path *path;
5761         int ret;
5762         struct extent_buffer *leaf;
5763         struct btrfs_root *root = BTRFS_I(inode)->root;
5764         struct btrfs_file_extent_item *fi;
5765         struct btrfs_key key;
5766         u64 disk_bytenr;
5767         u64 backref_offset;
5768         u64 extent_end;
5769         u64 num_bytes;
5770         int slot;
5771         int found_type;
5772
5773         path = btrfs_alloc_path();
5774         if (!path)
5775                 return -ENOMEM;
5776
5777         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5778                                        offset, 0);
5779         if (ret < 0)
5780                 goto out;
5781
5782         slot = path->slots[0];
5783         if (ret == 1) {
5784                 if (slot == 0) {
5785                         /* can't find the item, must cow */
5786                         ret = 0;
5787                         goto out;
5788                 }
5789                 slot--;
5790         }
5791         ret = 0;
5792         leaf = path->nodes[0];
5793         btrfs_item_key_to_cpu(leaf, &key, slot);
5794         if (key.objectid != btrfs_ino(inode) ||
5795             key.type != BTRFS_EXTENT_DATA_KEY) {
5796                 /* not our file or wrong item type, must cow */
5797                 goto out;
5798         }
5799
5800         if (key.offset > offset) {
5801                 /* Wrong offset, must cow */
5802                 goto out;
5803         }
5804
5805         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5806         found_type = btrfs_file_extent_type(leaf, fi);
5807         if (found_type != BTRFS_FILE_EXTENT_REG &&
5808             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5809                 /* not a regular extent, must cow */
5810                 goto out;
5811         }
5812         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5813         backref_offset = btrfs_file_extent_offset(leaf, fi);
5814
5815         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5816         if (extent_end < offset + len) {
5817                 /* extent doesn't include our full range, must cow */
5818                 goto out;
5819         }
5820
5821         if (btrfs_extent_readonly(root, disk_bytenr))
5822                 goto out;
5823
5824         /*
5825          * look for other files referencing this extent, if we
5826          * find any we must cow
5827          */
5828         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5829                                   key.offset - backref_offset, disk_bytenr))
5830                 goto out;
5831
5832         /*
5833          * adjust disk_bytenr and num_bytes to cover just the bytes
5834          * in this extent we are about to write.  If there
5835          * are any csums in that range we have to cow in order
5836          * to keep the csums correct
5837          */
5838         disk_bytenr += backref_offset;
5839         disk_bytenr += offset - key.offset;
5840         num_bytes = min(offset + len, extent_end) - offset;
5841         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5842                                 goto out;
5843         /*
5844          * all of the above have passed, it is safe to overwrite this extent
5845          * without cow
5846          */
5847         ret = 1;
5848 out:
5849         btrfs_free_path(path);
5850         return ret;
5851 }
5852
5853 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5854                               struct extent_state **cached_state, int writing)
5855 {
5856         struct btrfs_ordered_extent *ordered;
5857         int ret = 0;
5858
5859         while (1) {
5860                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5861                                  0, cached_state);
5862                 /*
5863                  * We're concerned with the entire range that we're going to be
5864                  * doing DIO to, so we need to make sure theres no ordered
5865                  * extents in this range.
5866                  */
5867                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5868                                                      lockend - lockstart + 1);
5869
5870                 /*
5871                  * We need to make sure there are no buffered pages in this
5872                  * range either, we could have raced between the invalidate in
5873                  * generic_file_direct_write and locking the extent.  The
5874                  * invalidate needs to happen so that reads after a write do not
5875                  * get stale data.
5876                  */
5877                 if (!ordered && (!writing ||
5878                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5879                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5880                                     *cached_state)))
5881                         break;
5882
5883                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5884                                      cached_state, GFP_NOFS);
5885
5886                 if (ordered) {
5887                         btrfs_start_ordered_extent(inode, ordered, 1);
5888                         btrfs_put_ordered_extent(ordered);
5889                 } else {
5890                         /* Screw you mmap */
5891                         ret = filemap_write_and_wait_range(inode->i_mapping,
5892                                                            lockstart,
5893                                                            lockend);
5894                         if (ret)
5895                                 break;
5896
5897                         /*
5898                          * If we found a page that couldn't be invalidated just
5899                          * fall back to buffered.
5900                          */
5901                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5902                                         lockstart >> PAGE_CACHE_SHIFT,
5903                                         lockend >> PAGE_CACHE_SHIFT);
5904                         if (ret)
5905                                 break;
5906                 }
5907
5908                 cond_resched();
5909         }
5910
5911         return ret;
5912 }
5913
5914 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5915                                            u64 len, u64 orig_start,
5916                                            u64 block_start, u64 block_len,
5917                                            int type)
5918 {
5919         struct extent_map_tree *em_tree;
5920         struct extent_map *em;
5921         struct btrfs_root *root = BTRFS_I(inode)->root;
5922         int ret;
5923
5924         em_tree = &BTRFS_I(inode)->extent_tree;
5925         em = alloc_extent_map();
5926         if (!em)
5927                 return ERR_PTR(-ENOMEM);
5928
5929         em->start = start;
5930         em->orig_start = orig_start;
5931         em->len = len;
5932         em->block_len = block_len;
5933         em->block_start = block_start;
5934         em->bdev = root->fs_info->fs_devices->latest_bdev;
5935         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5936         if (type == BTRFS_ORDERED_PREALLOC)
5937                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5938
5939         do {
5940                 btrfs_drop_extent_cache(inode, em->start,
5941                                 em->start + em->len - 1, 0);
5942                 write_lock(&em_tree->lock);
5943                 ret = add_extent_mapping(em_tree, em);
5944                 write_unlock(&em_tree->lock);
5945         } while (ret == -EEXIST);
5946
5947         if (ret) {
5948                 free_extent_map(em);
5949                 return ERR_PTR(ret);
5950         }
5951
5952         return em;
5953 }
5954
5955
5956 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5957                                    struct buffer_head *bh_result, int create)
5958 {
5959         struct extent_map *em;
5960         struct btrfs_root *root = BTRFS_I(inode)->root;
5961         struct extent_state *cached_state = NULL;
5962         u64 start = iblock << inode->i_blkbits;
5963         u64 lockstart, lockend;
5964         u64 len = bh_result->b_size;
5965         struct btrfs_trans_handle *trans;
5966         int unlock_bits = EXTENT_LOCKED;
5967         int ret;
5968
5969         if (create) {
5970                 ret = btrfs_delalloc_reserve_space(inode, len);
5971                 if (ret)
5972                         return ret;
5973                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5974         } else {
5975                 len = min_t(u64, len, root->sectorsize);
5976         }
5977
5978         lockstart = start;
5979         lockend = start + len - 1;
5980
5981         /*
5982          * If this errors out it's because we couldn't invalidate pagecache for
5983          * this range and we need to fallback to buffered.
5984          */
5985         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5986                 return -ENOTBLK;
5987
5988         if (create) {
5989                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5990                                      lockend, EXTENT_DELALLOC, NULL,
5991                                      &cached_state, GFP_NOFS);
5992                 if (ret)
5993                         goto unlock_err;
5994         }
5995
5996         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5997         if (IS_ERR(em)) {
5998                 ret = PTR_ERR(em);
5999                 goto unlock_err;
6000         }
6001
6002         /*
6003          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6004          * io.  INLINE is special, and we could probably kludge it in here, but
6005          * it's still buffered so for safety lets just fall back to the generic
6006          * buffered path.
6007          *
6008          * For COMPRESSED we _have_ to read the entire extent in so we can
6009          * decompress it, so there will be buffering required no matter what we
6010          * do, so go ahead and fallback to buffered.
6011          *
6012          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6013          * to buffered IO.  Don't blame me, this is the price we pay for using
6014          * the generic code.
6015          */
6016         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6017             em->block_start == EXTENT_MAP_INLINE) {
6018                 free_extent_map(em);
6019                 ret = -ENOTBLK;
6020                 goto unlock_err;
6021         }
6022
6023         /* Just a good old fashioned hole, return */
6024         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6025                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6026                 free_extent_map(em);
6027                 ret = 0;
6028                 goto unlock_err;
6029         }
6030
6031         /*
6032          * We don't allocate a new extent in the following cases
6033          *
6034          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6035          * existing extent.
6036          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6037          * just use the extent.
6038          *
6039          */
6040         if (!create) {
6041                 len = min(len, em->len - (start - em->start));
6042                 lockstart = start + len;
6043                 goto unlock;
6044         }
6045
6046         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6047             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6048              em->block_start != EXTENT_MAP_HOLE)) {
6049                 int type;
6050                 int ret;
6051                 u64 block_start;
6052
6053                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6054                         type = BTRFS_ORDERED_PREALLOC;
6055                 else
6056                         type = BTRFS_ORDERED_NOCOW;
6057                 len = min(len, em->len - (start - em->start));
6058                 block_start = em->block_start + (start - em->start);
6059
6060                 /*
6061                  * we're not going to log anything, but we do need
6062                  * to make sure the current transaction stays open
6063                  * while we look for nocow cross refs
6064                  */
6065                 trans = btrfs_join_transaction(root);
6066                 if (IS_ERR(trans))
6067                         goto must_cow;
6068
6069                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6070                         u64 orig_start = em->start;
6071
6072                         if (type == BTRFS_ORDERED_PREALLOC) {
6073                                 free_extent_map(em);
6074                                 em = create_pinned_em(inode, start, len,
6075                                                        orig_start,
6076                                                        block_start, len, type);
6077                                 if (IS_ERR(em)) {
6078                                         btrfs_end_transaction(trans, root);
6079                                         goto unlock_err;
6080                                 }
6081                         }
6082
6083                         ret = btrfs_add_ordered_extent_dio(inode, start,
6084                                            block_start, len, len, type);
6085                         btrfs_end_transaction(trans, root);
6086                         if (ret) {
6087                                 free_extent_map(em);
6088                                 goto unlock_err;
6089                         }
6090                         goto unlock;
6091                 }
6092                 btrfs_end_transaction(trans, root);
6093         }
6094 must_cow:
6095         /*
6096          * this will cow the extent, reset the len in case we changed
6097          * it above
6098          */
6099         len = bh_result->b_size;
6100         em = btrfs_new_extent_direct(inode, em, start, len);
6101         if (IS_ERR(em)) {
6102                 ret = PTR_ERR(em);
6103                 goto unlock_err;
6104         }
6105         len = min(len, em->len - (start - em->start));
6106 unlock:
6107         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6108                 inode->i_blkbits;
6109         bh_result->b_size = len;
6110         bh_result->b_bdev = em->bdev;
6111         set_buffer_mapped(bh_result);
6112         if (create) {
6113                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6114                         set_buffer_new(bh_result);
6115
6116                 /*
6117                  * Need to update the i_size under the extent lock so buffered
6118                  * readers will get the updated i_size when we unlock.
6119                  */
6120                 if (start + len > i_size_read(inode))
6121                         i_size_write(inode, start + len);
6122         }
6123
6124         /*
6125          * In the case of write we need to clear and unlock the entire range,
6126          * in the case of read we need to unlock only the end area that we
6127          * aren't using if there is any left over space.
6128          */
6129         if (lockstart < lockend) {
6130                 if (create && len < lockend - lockstart) {
6131                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6132                                          lockstart + len - 1,
6133                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6134                                          &cached_state, GFP_NOFS);
6135                         /*
6136                          * Beside unlock, we also need to cleanup reserved space
6137                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6138                          */
6139                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6140                                          lockstart + len, lockend,
6141                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6142                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6143                 } else {
6144                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6145                                          lockend, unlock_bits, 1, 0,
6146                                          &cached_state, GFP_NOFS);
6147                 }
6148         } else {
6149                 free_extent_state(cached_state);
6150         }
6151
6152         free_extent_map(em);
6153
6154         return 0;
6155
6156 unlock_err:
6157         if (create)
6158                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6159
6160         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6161                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6162         return ret;
6163 }
6164
6165 struct btrfs_dio_private {
6166         struct inode *inode;
6167         u64 logical_offset;
6168         u64 disk_bytenr;
6169         u64 bytes;
6170         void *private;
6171
6172         /* number of bios pending for this dio */
6173         atomic_t pending_bios;
6174
6175         /* IO errors */
6176         int errors;
6177
6178         struct bio *orig_bio;
6179 };
6180
6181 static void btrfs_endio_direct_read(struct bio *bio, int err)
6182 {
6183         struct btrfs_dio_private *dip = bio->bi_private;
6184         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6185         struct bio_vec *bvec = bio->bi_io_vec;
6186         struct inode *inode = dip->inode;
6187         struct btrfs_root *root = BTRFS_I(inode)->root;
6188         u64 start;
6189
6190         start = dip->logical_offset;
6191         do {
6192                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6193                         struct page *page = bvec->bv_page;
6194                         char *kaddr;
6195                         u32 csum = ~(u32)0;
6196                         u64 private = ~(u32)0;
6197                         unsigned long flags;
6198
6199                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6200                                               start, &private))
6201                                 goto failed;
6202                         local_irq_save(flags);
6203                         kaddr = kmap_atomic(page);
6204                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6205                                                csum, bvec->bv_len);
6206                         btrfs_csum_final(csum, (char *)&csum);
6207                         kunmap_atomic(kaddr);
6208                         local_irq_restore(flags);
6209
6210                         flush_dcache_page(bvec->bv_page);
6211                         if (csum != private) {
6212 failed:
6213                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6214                                       " %llu csum %u private %u\n",
6215                                       (unsigned long long)btrfs_ino(inode),
6216                                       (unsigned long long)start,
6217                                       csum, (unsigned)private);
6218                                 err = -EIO;
6219                         }
6220                 }
6221
6222                 start += bvec->bv_len;
6223                 bvec++;
6224         } while (bvec <= bvec_end);
6225
6226         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6227                       dip->logical_offset + dip->bytes - 1);
6228         bio->bi_private = dip->private;
6229
6230         kfree(dip);
6231
6232         /* If we had a csum failure make sure to clear the uptodate flag */
6233         if (err)
6234                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6235         dio_end_io(bio, err);
6236 }
6237
6238 static void btrfs_endio_direct_write(struct bio *bio, int err)
6239 {
6240         struct btrfs_dio_private *dip = bio->bi_private;
6241         struct inode *inode = dip->inode;
6242         struct btrfs_root *root = BTRFS_I(inode)->root;
6243         struct btrfs_ordered_extent *ordered = NULL;
6244         u64 ordered_offset = dip->logical_offset;
6245         u64 ordered_bytes = dip->bytes;
6246         int ret;
6247
6248         if (err)
6249                 goto out_done;
6250 again:
6251         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6252                                                    &ordered_offset,
6253                                                    ordered_bytes, !err);
6254         if (!ret)
6255                 goto out_test;
6256
6257         ordered->work.func = finish_ordered_fn;
6258         ordered->work.flags = 0;
6259         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6260                            &ordered->work);
6261 out_test:
6262         /*
6263          * our bio might span multiple ordered extents.  If we haven't
6264          * completed the accounting for the whole dio, go back and try again
6265          */
6266         if (ordered_offset < dip->logical_offset + dip->bytes) {
6267                 ordered_bytes = dip->logical_offset + dip->bytes -
6268                         ordered_offset;
6269                 ordered = NULL;
6270                 goto again;
6271         }
6272 out_done:
6273         bio->bi_private = dip->private;
6274
6275         kfree(dip);
6276
6277         /* If we had an error make sure to clear the uptodate flag */
6278         if (err)
6279                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6280         dio_end_io(bio, err);
6281 }
6282
6283 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6284                                     struct bio *bio, int mirror_num,
6285                                     unsigned long bio_flags, u64 offset)
6286 {
6287         int ret;
6288         struct btrfs_root *root = BTRFS_I(inode)->root;
6289         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6290         BUG_ON(ret); /* -ENOMEM */
6291         return 0;
6292 }
6293
6294 static void btrfs_end_dio_bio(struct bio *bio, int err)
6295 {
6296         struct btrfs_dio_private *dip = bio->bi_private;
6297
6298         if (err) {
6299                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6300                       "sector %#Lx len %u err no %d\n",
6301                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6302                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6303                 dip->errors = 1;
6304
6305                 /*
6306                  * before atomic variable goto zero, we must make sure
6307                  * dip->errors is perceived to be set.
6308                  */
6309                 smp_mb__before_atomic_dec();
6310         }
6311
6312         /* if there are more bios still pending for this dio, just exit */
6313         if (!atomic_dec_and_test(&dip->pending_bios))
6314                 goto out;
6315
6316         if (dip->errors)
6317                 bio_io_error(dip->orig_bio);
6318         else {
6319                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6320                 bio_endio(dip->orig_bio, 0);
6321         }
6322 out:
6323         bio_put(bio);
6324 }
6325
6326 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6327                                        u64 first_sector, gfp_t gfp_flags)
6328 {
6329         int nr_vecs = bio_get_nr_vecs(bdev);
6330         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6331 }
6332
6333 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6334                                          int rw, u64 file_offset, int skip_sum,
6335                                          int async_submit)
6336 {
6337         int write = rw & REQ_WRITE;
6338         struct btrfs_root *root = BTRFS_I(inode)->root;
6339         int ret;
6340
6341         if (async_submit)
6342                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6343
6344         bio_get(bio);
6345
6346         if (!write) {
6347                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6348                 if (ret)
6349                         goto err;
6350         }
6351
6352         if (skip_sum)
6353                 goto map;
6354
6355         if (write && async_submit) {
6356                 ret = btrfs_wq_submit_bio(root->fs_info,
6357                                    inode, rw, bio, 0, 0,
6358                                    file_offset,
6359                                    __btrfs_submit_bio_start_direct_io,
6360                                    __btrfs_submit_bio_done);
6361                 goto err;
6362         } else if (write) {
6363                 /*
6364                  * If we aren't doing async submit, calculate the csum of the
6365                  * bio now.
6366                  */
6367                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6368                 if (ret)
6369                         goto err;
6370         } else if (!skip_sum) {
6371                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6372                 if (ret)
6373                         goto err;
6374         }
6375
6376 map:
6377         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6378 err:
6379         bio_put(bio);
6380         return ret;
6381 }
6382
6383 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6384                                     int skip_sum)
6385 {
6386         struct inode *inode = dip->inode;
6387         struct btrfs_root *root = BTRFS_I(inode)->root;
6388         struct bio *bio;
6389         struct bio *orig_bio = dip->orig_bio;
6390         struct bio_vec *bvec = orig_bio->bi_io_vec;
6391         u64 start_sector = orig_bio->bi_sector;
6392         u64 file_offset = dip->logical_offset;
6393         u64 submit_len = 0;
6394         u64 map_length;
6395         int nr_pages = 0;
6396         int ret = 0;
6397         int async_submit = 0;
6398
6399         map_length = orig_bio->bi_size;
6400         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6401                               &map_length, NULL, 0);
6402         if (ret) {
6403                 bio_put(orig_bio);
6404                 return -EIO;
6405         }
6406
6407         if (map_length >= orig_bio->bi_size) {
6408                 bio = orig_bio;
6409                 goto submit;
6410         }
6411
6412         async_submit = 1;
6413         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6414         if (!bio)
6415                 return -ENOMEM;
6416         bio->bi_private = dip;
6417         bio->bi_end_io = btrfs_end_dio_bio;
6418         atomic_inc(&dip->pending_bios);
6419
6420         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6421                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6422                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6423                                  bvec->bv_offset) < bvec->bv_len)) {
6424                         /*
6425                          * inc the count before we submit the bio so
6426                          * we know the end IO handler won't happen before
6427                          * we inc the count. Otherwise, the dip might get freed
6428                          * before we're done setting it up
6429                          */
6430                         atomic_inc(&dip->pending_bios);
6431                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6432                                                      file_offset, skip_sum,
6433                                                      async_submit);
6434                         if (ret) {
6435                                 bio_put(bio);
6436                                 atomic_dec(&dip->pending_bios);
6437                                 goto out_err;
6438                         }
6439
6440                         start_sector += submit_len >> 9;
6441                         file_offset += submit_len;
6442
6443                         submit_len = 0;
6444                         nr_pages = 0;
6445
6446                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6447                                                   start_sector, GFP_NOFS);
6448                         if (!bio)
6449                                 goto out_err;
6450                         bio->bi_private = dip;
6451                         bio->bi_end_io = btrfs_end_dio_bio;
6452
6453                         map_length = orig_bio->bi_size;
6454                         ret = btrfs_map_block(root->fs_info, READ,
6455                                               start_sector << 9,
6456                                               &map_length, NULL, 0);
6457                         if (ret) {
6458                                 bio_put(bio);
6459                                 goto out_err;
6460                         }
6461                 } else {
6462                         submit_len += bvec->bv_len;
6463                         nr_pages ++;
6464                         bvec++;
6465                 }
6466         }
6467
6468 submit:
6469         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6470                                      async_submit);
6471         if (!ret)
6472                 return 0;
6473
6474         bio_put(bio);
6475 out_err:
6476         dip->errors = 1;
6477         /*
6478          * before atomic variable goto zero, we must
6479          * make sure dip->errors is perceived to be set.
6480          */
6481         smp_mb__before_atomic_dec();
6482         if (atomic_dec_and_test(&dip->pending_bios))
6483                 bio_io_error(dip->orig_bio);
6484
6485         /* bio_end_io() will handle error, so we needn't return it */
6486         return 0;
6487 }
6488
6489 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6490                                 loff_t file_offset)
6491 {
6492         struct btrfs_root *root = BTRFS_I(inode)->root;
6493         struct btrfs_dio_private *dip;
6494         struct bio_vec *bvec = bio->bi_io_vec;
6495         int skip_sum;
6496         int write = rw & REQ_WRITE;
6497         int ret = 0;
6498
6499         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6500
6501         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6502         if (!dip) {
6503                 ret = -ENOMEM;
6504                 goto free_ordered;
6505         }
6506
6507         dip->private = bio->bi_private;
6508         dip->inode = inode;
6509         dip->logical_offset = file_offset;
6510
6511         dip->bytes = 0;
6512         do {
6513                 dip->bytes += bvec->bv_len;
6514                 bvec++;
6515         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6516
6517         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6518         bio->bi_private = dip;
6519         dip->errors = 0;
6520         dip->orig_bio = bio;
6521         atomic_set(&dip->pending_bios, 0);
6522
6523         if (write)
6524                 bio->bi_end_io = btrfs_endio_direct_write;
6525         else
6526                 bio->bi_end_io = btrfs_endio_direct_read;
6527
6528         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6529         if (!ret)
6530                 return;
6531 free_ordered:
6532         /*
6533          * If this is a write, we need to clean up the reserved space and kill
6534          * the ordered extent.
6535          */
6536         if (write) {
6537                 struct btrfs_ordered_extent *ordered;
6538                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6539                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6540                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6541                         btrfs_free_reserved_extent(root, ordered->start,
6542                                                    ordered->disk_len);
6543                 btrfs_put_ordered_extent(ordered);
6544                 btrfs_put_ordered_extent(ordered);
6545         }
6546         bio_endio(bio, ret);
6547 }
6548
6549 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6550                         const struct iovec *iov, loff_t offset,
6551                         unsigned long nr_segs)
6552 {
6553         int seg;
6554         int i;
6555         size_t size;
6556         unsigned long addr;
6557         unsigned blocksize_mask = root->sectorsize - 1;
6558         ssize_t retval = -EINVAL;
6559         loff_t end = offset;
6560
6561         if (offset & blocksize_mask)
6562                 goto out;
6563
6564         /* Check the memory alignment.  Blocks cannot straddle pages */
6565         for (seg = 0; seg < nr_segs; seg++) {
6566                 addr = (unsigned long)iov[seg].iov_base;
6567                 size = iov[seg].iov_len;
6568                 end += size;
6569                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6570                         goto out;
6571
6572                 /* If this is a write we don't need to check anymore */
6573                 if (rw & WRITE)
6574                         continue;
6575
6576                 /*
6577                  * Check to make sure we don't have duplicate iov_base's in this
6578                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6579                  * when reading back.
6580                  */
6581                 for (i = seg + 1; i < nr_segs; i++) {
6582                         if (iov[seg].iov_base == iov[i].iov_base)
6583                                 goto out;
6584                 }
6585         }
6586         retval = 0;
6587 out:
6588         return retval;
6589 }
6590
6591 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6592                         const struct iovec *iov, loff_t offset,
6593                         unsigned long nr_segs)
6594 {
6595         struct file *file = iocb->ki_filp;
6596         struct inode *inode = file->f_mapping->host;
6597
6598         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6599                             offset, nr_segs))
6600                 return 0;
6601
6602         return __blockdev_direct_IO(rw, iocb, inode,
6603                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6604                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6605                    btrfs_submit_direct, 0);
6606 }
6607
6608 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
6609
6610 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6611                 __u64 start, __u64 len)
6612 {
6613         int     ret;
6614
6615         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6616         if (ret)
6617                 return ret;
6618
6619         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6620 }
6621
6622 int btrfs_readpage(struct file *file, struct page *page)
6623 {
6624         struct extent_io_tree *tree;
6625         tree = &BTRFS_I(page->mapping->host)->io_tree;
6626         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6627 }
6628
6629 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6630 {
6631         struct extent_io_tree *tree;
6632
6633
6634         if (current->flags & PF_MEMALLOC) {
6635                 redirty_page_for_writepage(wbc, page);
6636                 unlock_page(page);
6637                 return 0;
6638         }
6639         tree = &BTRFS_I(page->mapping->host)->io_tree;
6640         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6641 }
6642
6643 int btrfs_writepages(struct address_space *mapping,
6644                      struct writeback_control *wbc)
6645 {
6646         struct extent_io_tree *tree;
6647
6648         tree = &BTRFS_I(mapping->host)->io_tree;
6649         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6650 }
6651
6652 static int
6653 btrfs_readpages(struct file *file, struct address_space *mapping,
6654                 struct list_head *pages, unsigned nr_pages)
6655 {
6656         struct extent_io_tree *tree;
6657         tree = &BTRFS_I(mapping->host)->io_tree;
6658         return extent_readpages(tree, mapping, pages, nr_pages,
6659                                 btrfs_get_extent);
6660 }
6661 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6662 {
6663         struct extent_io_tree *tree;
6664         struct extent_map_tree *map;
6665         int ret;
6666
6667         tree = &BTRFS_I(page->mapping->host)->io_tree;
6668         map = &BTRFS_I(page->mapping->host)->extent_tree;
6669         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6670         if (ret == 1) {
6671                 ClearPagePrivate(page);
6672                 set_page_private(page, 0);
6673                 page_cache_release(page);
6674         }
6675         return ret;
6676 }
6677
6678 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6679 {
6680         if (PageWriteback(page) || PageDirty(page))
6681                 return 0;
6682         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6683 }
6684
6685 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6686 {
6687         struct inode *inode = page->mapping->host;
6688         struct extent_io_tree *tree;
6689         struct btrfs_ordered_extent *ordered;
6690         struct extent_state *cached_state = NULL;
6691         u64 page_start = page_offset(page);
6692         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6693
6694         /*
6695          * we have the page locked, so new writeback can't start,
6696          * and the dirty bit won't be cleared while we are here.
6697          *
6698          * Wait for IO on this page so that we can safely clear
6699          * the PagePrivate2 bit and do ordered accounting
6700          */
6701         wait_on_page_writeback(page);
6702
6703         tree = &BTRFS_I(inode)->io_tree;
6704         if (offset) {
6705                 btrfs_releasepage(page, GFP_NOFS);
6706                 return;
6707         }
6708         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6709         ordered = btrfs_lookup_ordered_extent(inode,
6710                                            page_offset(page));
6711         if (ordered) {
6712                 /*
6713                  * IO on this page will never be started, so we need
6714                  * to account for any ordered extents now
6715                  */
6716                 clear_extent_bit(tree, page_start, page_end,
6717                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6718                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6719                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6720                 /*
6721                  * whoever cleared the private bit is responsible
6722                  * for the finish_ordered_io
6723                  */
6724                 if (TestClearPagePrivate2(page) &&
6725                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6726                                                    PAGE_CACHE_SIZE, 1)) {
6727                         btrfs_finish_ordered_io(ordered);
6728                 }
6729                 btrfs_put_ordered_extent(ordered);
6730                 cached_state = NULL;
6731                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6732         }
6733         clear_extent_bit(tree, page_start, page_end,
6734                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6735                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6736                  &cached_state, GFP_NOFS);
6737         __btrfs_releasepage(page, GFP_NOFS);
6738
6739         ClearPageChecked(page);
6740         if (PagePrivate(page)) {
6741                 ClearPagePrivate(page);
6742                 set_page_private(page, 0);
6743                 page_cache_release(page);
6744         }
6745 }
6746
6747 /*
6748  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6749  * called from a page fault handler when a page is first dirtied. Hence we must
6750  * be careful to check for EOF conditions here. We set the page up correctly
6751  * for a written page which means we get ENOSPC checking when writing into
6752  * holes and correct delalloc and unwritten extent mapping on filesystems that
6753  * support these features.
6754  *
6755  * We are not allowed to take the i_mutex here so we have to play games to
6756  * protect against truncate races as the page could now be beyond EOF.  Because
6757  * vmtruncate() writes the inode size before removing pages, once we have the
6758  * page lock we can determine safely if the page is beyond EOF. If it is not
6759  * beyond EOF, then the page is guaranteed safe against truncation until we
6760  * unlock the page.
6761  */
6762 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6763 {
6764         struct page *page = vmf->page;
6765         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6766         struct btrfs_root *root = BTRFS_I(inode)->root;
6767         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6768         struct btrfs_ordered_extent *ordered;
6769         struct extent_state *cached_state = NULL;
6770         char *kaddr;
6771         unsigned long zero_start;
6772         loff_t size;
6773         int ret;
6774         int reserved = 0;
6775         u64 page_start;
6776         u64 page_end;
6777
6778         sb_start_pagefault(inode->i_sb);
6779         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6780         if (!ret) {
6781                 ret = file_update_time(vma->vm_file);
6782                 reserved = 1;
6783         }
6784         if (ret) {
6785                 if (ret == -ENOMEM)
6786                         ret = VM_FAULT_OOM;
6787                 else /* -ENOSPC, -EIO, etc */
6788                         ret = VM_FAULT_SIGBUS;
6789                 if (reserved)
6790                         goto out;
6791                 goto out_noreserve;
6792         }
6793
6794         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6795 again:
6796         lock_page(page);
6797         size = i_size_read(inode);
6798         page_start = page_offset(page);
6799         page_end = page_start + PAGE_CACHE_SIZE - 1;
6800
6801         if ((page->mapping != inode->i_mapping) ||
6802             (page_start >= size)) {
6803                 /* page got truncated out from underneath us */
6804                 goto out_unlock;
6805         }
6806         wait_on_page_writeback(page);
6807
6808         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6809         set_page_extent_mapped(page);
6810
6811         /*
6812          * we can't set the delalloc bits if there are pending ordered
6813          * extents.  Drop our locks and wait for them to finish
6814          */
6815         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6816         if (ordered) {
6817                 unlock_extent_cached(io_tree, page_start, page_end,
6818                                      &cached_state, GFP_NOFS);
6819                 unlock_page(page);
6820                 btrfs_start_ordered_extent(inode, ordered, 1);
6821                 btrfs_put_ordered_extent(ordered);
6822                 goto again;
6823         }
6824
6825         /*
6826          * XXX - page_mkwrite gets called every time the page is dirtied, even
6827          * if it was already dirty, so for space accounting reasons we need to
6828          * clear any delalloc bits for the range we are fixing to save.  There
6829          * is probably a better way to do this, but for now keep consistent with
6830          * prepare_pages in the normal write path.
6831          */
6832         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6833                           EXTENT_DIRTY | EXTENT_DELALLOC |
6834                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6835                           0, 0, &cached_state, GFP_NOFS);
6836
6837         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6838                                         &cached_state);
6839         if (ret) {
6840                 unlock_extent_cached(io_tree, page_start, page_end,
6841                                      &cached_state, GFP_NOFS);
6842                 ret = VM_FAULT_SIGBUS;
6843                 goto out_unlock;
6844         }
6845         ret = 0;
6846
6847         /* page is wholly or partially inside EOF */
6848         if (page_start + PAGE_CACHE_SIZE > size)
6849                 zero_start = size & ~PAGE_CACHE_MASK;
6850         else
6851                 zero_start = PAGE_CACHE_SIZE;
6852
6853         if (zero_start != PAGE_CACHE_SIZE) {
6854                 kaddr = kmap(page);
6855                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6856                 flush_dcache_page(page);
6857                 kunmap(page);
6858         }
6859         ClearPageChecked(page);
6860         set_page_dirty(page);
6861         SetPageUptodate(page);
6862
6863         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6864         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6865         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6866
6867         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6868
6869 out_unlock:
6870         if (!ret) {
6871                 sb_end_pagefault(inode->i_sb);
6872                 return VM_FAULT_LOCKED;
6873         }
6874         unlock_page(page);
6875 out:
6876         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6877 out_noreserve:
6878         sb_end_pagefault(inode->i_sb);
6879         return ret;
6880 }
6881
6882 static int btrfs_truncate(struct inode *inode)
6883 {
6884         struct btrfs_root *root = BTRFS_I(inode)->root;
6885         struct btrfs_block_rsv *rsv;
6886         int ret;
6887         int err = 0;
6888         struct btrfs_trans_handle *trans;
6889         u64 mask = root->sectorsize - 1;
6890         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6891
6892         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6893         if (ret)
6894                 return ret;
6895
6896         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6897         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6898
6899         /*
6900          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6901          * 3 things going on here
6902          *
6903          * 1) We need to reserve space for our orphan item and the space to
6904          * delete our orphan item.  Lord knows we don't want to have a dangling
6905          * orphan item because we didn't reserve space to remove it.
6906          *
6907          * 2) We need to reserve space to update our inode.
6908          *
6909          * 3) We need to have something to cache all the space that is going to
6910          * be free'd up by the truncate operation, but also have some slack
6911          * space reserved in case it uses space during the truncate (thank you
6912          * very much snapshotting).
6913          *
6914          * And we need these to all be seperate.  The fact is we can use alot of
6915          * space doing the truncate, and we have no earthly idea how much space
6916          * we will use, so we need the truncate reservation to be seperate so it
6917          * doesn't end up using space reserved for updating the inode or
6918          * removing the orphan item.  We also need to be able to stop the
6919          * transaction and start a new one, which means we need to be able to
6920          * update the inode several times, and we have no idea of knowing how
6921          * many times that will be, so we can't just reserve 1 item for the
6922          * entirety of the opration, so that has to be done seperately as well.
6923          * Then there is the orphan item, which does indeed need to be held on
6924          * to for the whole operation, and we need nobody to touch this reserved
6925          * space except the orphan code.
6926          *
6927          * So that leaves us with
6928          *
6929          * 1) root->orphan_block_rsv - for the orphan deletion.
6930          * 2) rsv - for the truncate reservation, which we will steal from the
6931          * transaction reservation.
6932          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6933          * updating the inode.
6934          */
6935         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6936         if (!rsv)
6937                 return -ENOMEM;
6938         rsv->size = min_size;
6939         rsv->failfast = 1;
6940
6941         /*
6942          * 1 for the truncate slack space
6943          * 1 for the orphan item we're going to add
6944          * 1 for the orphan item deletion
6945          * 1 for updating the inode.
6946          */
6947         trans = btrfs_start_transaction(root, 4);
6948         if (IS_ERR(trans)) {
6949                 err = PTR_ERR(trans);
6950                 goto out;
6951         }
6952
6953         /* Migrate the slack space for the truncate to our reserve */
6954         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6955                                       min_size);
6956         BUG_ON(ret);
6957
6958         ret = btrfs_orphan_add(trans, inode);
6959         if (ret) {
6960                 btrfs_end_transaction(trans, root);
6961                 goto out;
6962         }
6963
6964         /*
6965          * setattr is responsible for setting the ordered_data_close flag,
6966          * but that is only tested during the last file release.  That
6967          * could happen well after the next commit, leaving a great big
6968          * window where new writes may get lost if someone chooses to write
6969          * to this file after truncating to zero
6970          *
6971          * The inode doesn't have any dirty data here, and so if we commit
6972          * this is a noop.  If someone immediately starts writing to the inode
6973          * it is very likely we'll catch some of their writes in this
6974          * transaction, and the commit will find this file on the ordered
6975          * data list with good things to send down.
6976          *
6977          * This is a best effort solution, there is still a window where
6978          * using truncate to replace the contents of the file will
6979          * end up with a zero length file after a crash.
6980          */
6981         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6982                                            &BTRFS_I(inode)->runtime_flags))
6983                 btrfs_add_ordered_operation(trans, root, inode);
6984
6985         /*
6986          * So if we truncate and then write and fsync we normally would just
6987          * write the extents that changed, which is a problem if we need to
6988          * first truncate that entire inode.  So set this flag so we write out
6989          * all of the extents in the inode to the sync log so we're completely
6990          * safe.
6991          */
6992         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6993         trans->block_rsv = rsv;
6994
6995         while (1) {
6996                 ret = btrfs_truncate_inode_items(trans, root, inode,
6997                                                  inode->i_size,
6998                                                  BTRFS_EXTENT_DATA_KEY);
6999                 if (ret != -ENOSPC) {
7000                         err = ret;
7001                         break;
7002                 }
7003
7004                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7005                 ret = btrfs_update_inode(trans, root, inode);
7006                 if (ret) {
7007                         err = ret;
7008                         break;
7009                 }
7010
7011                 btrfs_end_transaction(trans, root);
7012                 btrfs_btree_balance_dirty(root);
7013
7014                 trans = btrfs_start_transaction(root, 2);
7015                 if (IS_ERR(trans)) {
7016                         ret = err = PTR_ERR(trans);
7017                         trans = NULL;
7018                         break;
7019                 }
7020
7021                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7022                                               rsv, min_size);
7023                 BUG_ON(ret);    /* shouldn't happen */
7024                 trans->block_rsv = rsv;
7025         }
7026
7027         if (ret == 0 && inode->i_nlink > 0) {
7028                 trans->block_rsv = root->orphan_block_rsv;
7029                 ret = btrfs_orphan_del(trans, inode);
7030                 if (ret)
7031                         err = ret;
7032         } else if (ret && inode->i_nlink > 0) {
7033                 /*
7034                  * Failed to do the truncate, remove us from the in memory
7035                  * orphan list.
7036                  */
7037                 ret = btrfs_orphan_del(NULL, inode);
7038         }
7039
7040         if (trans) {
7041                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7042                 ret = btrfs_update_inode(trans, root, inode);
7043                 if (ret && !err)
7044                         err = ret;
7045
7046                 ret = btrfs_end_transaction(trans, root);
7047                 btrfs_btree_balance_dirty(root);
7048         }
7049
7050 out:
7051         btrfs_free_block_rsv(root, rsv);
7052
7053         if (ret && !err)
7054                 err = ret;
7055
7056         return err;
7057 }
7058
7059 /*
7060  * create a new subvolume directory/inode (helper for the ioctl).
7061  */
7062 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7063                              struct btrfs_root *new_root, u64 new_dirid)
7064 {
7065         struct inode *inode;
7066         int err;
7067         u64 index = 0;
7068
7069         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7070                                 new_dirid, new_dirid,
7071                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7072                                 &index);
7073         if (IS_ERR(inode))
7074                 return PTR_ERR(inode);
7075         inode->i_op = &btrfs_dir_inode_operations;
7076         inode->i_fop = &btrfs_dir_file_operations;
7077
7078         set_nlink(inode, 1);
7079         btrfs_i_size_write(inode, 0);
7080
7081         err = btrfs_update_inode(trans, new_root, inode);
7082
7083         iput(inode);
7084         return err;
7085 }
7086
7087 struct inode *btrfs_alloc_inode(struct super_block *sb)
7088 {
7089         struct btrfs_inode *ei;
7090         struct inode *inode;
7091
7092         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7093         if (!ei)
7094                 return NULL;
7095
7096         ei->root = NULL;
7097         ei->generation = 0;
7098         ei->last_trans = 0;
7099         ei->last_sub_trans = 0;
7100         ei->logged_trans = 0;
7101         ei->delalloc_bytes = 0;
7102         ei->disk_i_size = 0;
7103         ei->flags = 0;
7104         ei->csum_bytes = 0;
7105         ei->index_cnt = (u64)-1;
7106         ei->last_unlink_trans = 0;
7107         ei->last_log_commit = 0;
7108
7109         spin_lock_init(&ei->lock);
7110         ei->outstanding_extents = 0;
7111         ei->reserved_extents = 0;
7112
7113         ei->runtime_flags = 0;
7114         ei->force_compress = BTRFS_COMPRESS_NONE;
7115
7116         ei->delayed_node = NULL;
7117
7118         inode = &ei->vfs_inode;
7119         extent_map_tree_init(&ei->extent_tree);
7120         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7121         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7122         ei->io_tree.track_uptodate = 1;
7123         ei->io_failure_tree.track_uptodate = 1;
7124         atomic_set(&ei->sync_writers, 0);
7125         mutex_init(&ei->log_mutex);
7126         mutex_init(&ei->delalloc_mutex);
7127         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7128         INIT_LIST_HEAD(&ei->delalloc_inodes);
7129         INIT_LIST_HEAD(&ei->ordered_operations);
7130         RB_CLEAR_NODE(&ei->rb_node);
7131
7132         return inode;
7133 }
7134
7135 static void btrfs_i_callback(struct rcu_head *head)
7136 {
7137         struct inode *inode = container_of(head, struct inode, i_rcu);
7138         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7139 }
7140
7141 void btrfs_destroy_inode(struct inode *inode)
7142 {
7143         struct btrfs_ordered_extent *ordered;
7144         struct btrfs_root *root = BTRFS_I(inode)->root;
7145
7146         WARN_ON(!hlist_empty(&inode->i_dentry));
7147         WARN_ON(inode->i_data.nrpages);
7148         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7149         WARN_ON(BTRFS_I(inode)->reserved_extents);
7150         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7151         WARN_ON(BTRFS_I(inode)->csum_bytes);
7152
7153         /*
7154          * This can happen where we create an inode, but somebody else also
7155          * created the same inode and we need to destroy the one we already
7156          * created.
7157          */
7158         if (!root)
7159                 goto free;
7160
7161         /*
7162          * Make sure we're properly removed from the ordered operation
7163          * lists.
7164          */
7165         smp_mb();
7166         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7167                 spin_lock(&root->fs_info->ordered_extent_lock);
7168                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7169                 spin_unlock(&root->fs_info->ordered_extent_lock);
7170         }
7171
7172         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7173                      &BTRFS_I(inode)->runtime_flags)) {
7174                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7175                        (unsigned long long)btrfs_ino(inode));
7176                 atomic_dec(&root->orphan_inodes);
7177         }
7178
7179         while (1) {
7180                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7181                 if (!ordered)
7182                         break;
7183                 else {
7184                         printk(KERN_ERR "btrfs found ordered "
7185                                "extent %llu %llu on inode cleanup\n",
7186                                (unsigned long long)ordered->file_offset,
7187                                (unsigned long long)ordered->len);
7188                         btrfs_remove_ordered_extent(inode, ordered);
7189                         btrfs_put_ordered_extent(ordered);
7190                         btrfs_put_ordered_extent(ordered);
7191                 }
7192         }
7193         inode_tree_del(inode);
7194         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7195 free:
7196         btrfs_remove_delayed_node(inode);
7197         call_rcu(&inode->i_rcu, btrfs_i_callback);
7198 }
7199
7200 int btrfs_drop_inode(struct inode *inode)
7201 {
7202         struct btrfs_root *root = BTRFS_I(inode)->root;
7203
7204         if (btrfs_root_refs(&root->root_item) == 0 &&
7205             !btrfs_is_free_space_inode(inode))
7206                 return 1;
7207         else
7208                 return generic_drop_inode(inode);
7209 }
7210
7211 static void init_once(void *foo)
7212 {
7213         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7214
7215         inode_init_once(&ei->vfs_inode);
7216 }
7217
7218 void btrfs_destroy_cachep(void)
7219 {
7220         /*
7221          * Make sure all delayed rcu free inodes are flushed before we
7222          * destroy cache.
7223          */
7224         rcu_barrier();
7225         if (btrfs_inode_cachep)
7226                 kmem_cache_destroy(btrfs_inode_cachep);
7227         if (btrfs_trans_handle_cachep)
7228                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7229         if (btrfs_transaction_cachep)
7230                 kmem_cache_destroy(btrfs_transaction_cachep);
7231         if (btrfs_path_cachep)
7232                 kmem_cache_destroy(btrfs_path_cachep);
7233         if (btrfs_free_space_cachep)
7234                 kmem_cache_destroy(btrfs_free_space_cachep);
7235         if (btrfs_delalloc_work_cachep)
7236                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7237 }
7238
7239 int btrfs_init_cachep(void)
7240 {
7241         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7242                         sizeof(struct btrfs_inode), 0,
7243                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7244         if (!btrfs_inode_cachep)
7245                 goto fail;
7246
7247         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7248                         sizeof(struct btrfs_trans_handle), 0,
7249                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7250         if (!btrfs_trans_handle_cachep)
7251                 goto fail;
7252
7253         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7254                         sizeof(struct btrfs_transaction), 0,
7255                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7256         if (!btrfs_transaction_cachep)
7257                 goto fail;
7258
7259         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7260                         sizeof(struct btrfs_path), 0,
7261                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7262         if (!btrfs_path_cachep)
7263                 goto fail;
7264
7265         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7266                         sizeof(struct btrfs_free_space), 0,
7267                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7268         if (!btrfs_free_space_cachep)
7269                 goto fail;
7270
7271         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7272                         sizeof(struct btrfs_delalloc_work), 0,
7273                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7274                         NULL);
7275         if (!btrfs_delalloc_work_cachep)
7276                 goto fail;
7277
7278         return 0;
7279 fail:
7280         btrfs_destroy_cachep();
7281         return -ENOMEM;
7282 }
7283
7284 static int btrfs_getattr(struct vfsmount *mnt,
7285                          struct dentry *dentry, struct kstat *stat)
7286 {
7287         struct inode *inode = dentry->d_inode;
7288         u32 blocksize = inode->i_sb->s_blocksize;
7289
7290         generic_fillattr(inode, stat);
7291         stat->dev = BTRFS_I(inode)->root->anon_dev;
7292         stat->blksize = PAGE_CACHE_SIZE;
7293         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7294                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7295         return 0;
7296 }
7297
7298 /*
7299  * If a file is moved, it will inherit the cow and compression flags of the new
7300  * directory.
7301  */
7302 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7303 {
7304         struct btrfs_inode *b_dir = BTRFS_I(dir);
7305         struct btrfs_inode *b_inode = BTRFS_I(inode);
7306
7307         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7308                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7309         else
7310                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7311
7312         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7313                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7314                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7315         } else {
7316                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7317                                     BTRFS_INODE_NOCOMPRESS);
7318         }
7319 }
7320
7321 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7322                            struct inode *new_dir, struct dentry *new_dentry)
7323 {
7324         struct btrfs_trans_handle *trans;
7325         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7326         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7327         struct inode *new_inode = new_dentry->d_inode;
7328         struct inode *old_inode = old_dentry->d_inode;
7329         struct timespec ctime = CURRENT_TIME;
7330         u64 index = 0;
7331         u64 root_objectid;
7332         int ret;
7333         u64 old_ino = btrfs_ino(old_inode);
7334
7335         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7336                 return -EPERM;
7337
7338         /* we only allow rename subvolume link between subvolumes */
7339         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7340                 return -EXDEV;
7341
7342         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7343             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7344                 return -ENOTEMPTY;
7345
7346         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7347             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7348                 return -ENOTEMPTY;
7349         /*
7350          * we're using rename to replace one file with another.
7351          * and the replacement file is large.  Start IO on it now so
7352          * we don't add too much work to the end of the transaction
7353          */
7354         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7355             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7356                 filemap_flush(old_inode->i_mapping);
7357
7358         /* close the racy window with snapshot create/destroy ioctl */
7359         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7360                 down_read(&root->fs_info->subvol_sem);
7361         /*
7362          * We want to reserve the absolute worst case amount of items.  So if
7363          * both inodes are subvols and we need to unlink them then that would
7364          * require 4 item modifications, but if they are both normal inodes it
7365          * would require 5 item modifications, so we'll assume their normal
7366          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7367          * should cover the worst case number of items we'll modify.
7368          */
7369         trans = btrfs_start_transaction(root, 20);
7370         if (IS_ERR(trans)) {
7371                 ret = PTR_ERR(trans);
7372                 goto out_notrans;
7373         }
7374
7375         if (dest != root)
7376                 btrfs_record_root_in_trans(trans, dest);
7377
7378         ret = btrfs_set_inode_index(new_dir, &index);
7379         if (ret)
7380                 goto out_fail;
7381
7382         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7383                 /* force full log commit if subvolume involved. */
7384                 root->fs_info->last_trans_log_full_commit = trans->transid;
7385         } else {
7386                 ret = btrfs_insert_inode_ref(trans, dest,
7387                                              new_dentry->d_name.name,
7388                                              new_dentry->d_name.len,
7389                                              old_ino,
7390                                              btrfs_ino(new_dir), index);
7391                 if (ret)
7392                         goto out_fail;
7393                 /*
7394                  * this is an ugly little race, but the rename is required
7395                  * to make sure that if we crash, the inode is either at the
7396                  * old name or the new one.  pinning the log transaction lets
7397                  * us make sure we don't allow a log commit to come in after
7398                  * we unlink the name but before we add the new name back in.
7399                  */
7400                 btrfs_pin_log_trans(root);
7401         }
7402         /*
7403          * make sure the inode gets flushed if it is replacing
7404          * something.
7405          */
7406         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7407                 btrfs_add_ordered_operation(trans, root, old_inode);
7408
7409         inode_inc_iversion(old_dir);
7410         inode_inc_iversion(new_dir);
7411         inode_inc_iversion(old_inode);
7412         old_dir->i_ctime = old_dir->i_mtime = ctime;
7413         new_dir->i_ctime = new_dir->i_mtime = ctime;
7414         old_inode->i_ctime = ctime;
7415
7416         if (old_dentry->d_parent != new_dentry->d_parent)
7417                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7418
7419         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7420                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7421                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7422                                         old_dentry->d_name.name,
7423                                         old_dentry->d_name.len);
7424         } else {
7425                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7426                                         old_dentry->d_inode,
7427                                         old_dentry->d_name.name,
7428                                         old_dentry->d_name.len);
7429                 if (!ret)
7430                         ret = btrfs_update_inode(trans, root, old_inode);
7431         }
7432         if (ret) {
7433                 btrfs_abort_transaction(trans, root, ret);
7434                 goto out_fail;
7435         }
7436
7437         if (new_inode) {
7438                 inode_inc_iversion(new_inode);
7439                 new_inode->i_ctime = CURRENT_TIME;
7440                 if (unlikely(btrfs_ino(new_inode) ==
7441                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7442                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7443                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7444                                                 root_objectid,
7445                                                 new_dentry->d_name.name,
7446                                                 new_dentry->d_name.len);
7447                         BUG_ON(new_inode->i_nlink == 0);
7448                 } else {
7449                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7450                                                  new_dentry->d_inode,
7451                                                  new_dentry->d_name.name,
7452                                                  new_dentry->d_name.len);
7453                 }
7454                 if (!ret && new_inode->i_nlink == 0) {
7455                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7456                         BUG_ON(ret);
7457                 }
7458                 if (ret) {
7459                         btrfs_abort_transaction(trans, root, ret);
7460                         goto out_fail;
7461                 }
7462         }
7463
7464         fixup_inode_flags(new_dir, old_inode);
7465
7466         ret = btrfs_add_link(trans, new_dir, old_inode,
7467                              new_dentry->d_name.name,
7468                              new_dentry->d_name.len, 0, index);
7469         if (ret) {
7470                 btrfs_abort_transaction(trans, root, ret);
7471                 goto out_fail;
7472         }
7473
7474         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7475                 struct dentry *parent = new_dentry->d_parent;
7476                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7477                 btrfs_end_log_trans(root);
7478         }
7479 out_fail:
7480         btrfs_end_transaction(trans, root);
7481 out_notrans:
7482         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7483                 up_read(&root->fs_info->subvol_sem);
7484
7485         return ret;
7486 }
7487
7488 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7489 {
7490         struct btrfs_delalloc_work *delalloc_work;
7491
7492         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7493                                      work);
7494         if (delalloc_work->wait)
7495                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7496         else
7497                 filemap_flush(delalloc_work->inode->i_mapping);
7498
7499         if (delalloc_work->delay_iput)
7500                 btrfs_add_delayed_iput(delalloc_work->inode);
7501         else
7502                 iput(delalloc_work->inode);
7503         complete(&delalloc_work->completion);
7504 }
7505
7506 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7507                                                     int wait, int delay_iput)
7508 {
7509         struct btrfs_delalloc_work *work;
7510
7511         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7512         if (!work)
7513                 return NULL;
7514
7515         init_completion(&work->completion);
7516         INIT_LIST_HEAD(&work->list);
7517         work->inode = inode;
7518         work->wait = wait;
7519         work->delay_iput = delay_iput;
7520         work->work.func = btrfs_run_delalloc_work;
7521
7522         return work;
7523 }
7524
7525 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7526 {
7527         wait_for_completion(&work->completion);
7528         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7529 }
7530
7531 /*
7532  * some fairly slow code that needs optimization. This walks the list
7533  * of all the inodes with pending delalloc and forces them to disk.
7534  */
7535 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7536 {
7537         struct list_head *head = &root->fs_info->delalloc_inodes;
7538         struct btrfs_inode *binode;
7539         struct inode *inode;
7540         struct btrfs_delalloc_work *work, *next;
7541         struct list_head works;
7542         int ret = 0;
7543
7544         if (root->fs_info->sb->s_flags & MS_RDONLY)
7545                 return -EROFS;
7546
7547         INIT_LIST_HEAD(&works);
7548
7549         spin_lock(&root->fs_info->delalloc_lock);
7550         while (!list_empty(head)) {
7551                 binode = list_entry(head->next, struct btrfs_inode,
7552                                     delalloc_inodes);
7553                 inode = igrab(&binode->vfs_inode);
7554                 if (!inode)
7555                         list_del_init(&binode->delalloc_inodes);
7556                 spin_unlock(&root->fs_info->delalloc_lock);
7557                 if (inode) {
7558                         work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7559                         if (!work) {
7560                                 ret = -ENOMEM;
7561                                 goto out;
7562                         }
7563                         list_add_tail(&work->list, &works);
7564                         btrfs_queue_worker(&root->fs_info->flush_workers,
7565                                            &work->work);
7566                 }
7567                 cond_resched();
7568                 spin_lock(&root->fs_info->delalloc_lock);
7569         }
7570         spin_unlock(&root->fs_info->delalloc_lock);
7571
7572         /* the filemap_flush will queue IO into the worker threads, but
7573          * we have to make sure the IO is actually started and that
7574          * ordered extents get created before we return
7575          */
7576         atomic_inc(&root->fs_info->async_submit_draining);
7577         while (atomic_read(&root->fs_info->nr_async_submits) ||
7578               atomic_read(&root->fs_info->async_delalloc_pages)) {
7579                 wait_event(root->fs_info->async_submit_wait,
7580                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7581                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7582         }
7583         atomic_dec(&root->fs_info->async_submit_draining);
7584 out:
7585         list_for_each_entry_safe(work, next, &works, list) {
7586                 list_del_init(&work->list);
7587                 btrfs_wait_and_free_delalloc_work(work);
7588         }
7589         return ret;
7590 }
7591
7592 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7593                          const char *symname)
7594 {
7595         struct btrfs_trans_handle *trans;
7596         struct btrfs_root *root = BTRFS_I(dir)->root;
7597         struct btrfs_path *path;
7598         struct btrfs_key key;
7599         struct inode *inode = NULL;
7600         int err;
7601         int drop_inode = 0;
7602         u64 objectid;
7603         u64 index = 0 ;
7604         int name_len;
7605         int datasize;
7606         unsigned long ptr;
7607         struct btrfs_file_extent_item *ei;
7608         struct extent_buffer *leaf;
7609
7610         name_len = strlen(symname) + 1;
7611         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7612                 return -ENAMETOOLONG;
7613
7614         /*
7615          * 2 items for inode item and ref
7616          * 2 items for dir items
7617          * 1 item for xattr if selinux is on
7618          */
7619         trans = btrfs_start_transaction(root, 5);
7620         if (IS_ERR(trans))
7621                 return PTR_ERR(trans);
7622
7623         err = btrfs_find_free_ino(root, &objectid);
7624         if (err)
7625                 goto out_unlock;
7626
7627         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7628                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7629                                 S_IFLNK|S_IRWXUGO, &index);
7630         if (IS_ERR(inode)) {
7631                 err = PTR_ERR(inode);
7632                 goto out_unlock;
7633         }
7634
7635         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7636         if (err) {
7637                 drop_inode = 1;
7638                 goto out_unlock;
7639         }
7640
7641         /*
7642         * If the active LSM wants to access the inode during
7643         * d_instantiate it needs these. Smack checks to see
7644         * if the filesystem supports xattrs by looking at the
7645         * ops vector.
7646         */
7647         inode->i_fop = &btrfs_file_operations;
7648         inode->i_op = &btrfs_file_inode_operations;
7649
7650         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7651         if (err)
7652                 drop_inode = 1;
7653         else {
7654                 inode->i_mapping->a_ops = &btrfs_aops;
7655                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7656                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7657         }
7658         if (drop_inode)
7659                 goto out_unlock;
7660
7661         path = btrfs_alloc_path();
7662         if (!path) {
7663                 err = -ENOMEM;
7664                 drop_inode = 1;
7665                 goto out_unlock;
7666         }
7667         key.objectid = btrfs_ino(inode);
7668         key.offset = 0;
7669         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7670         datasize = btrfs_file_extent_calc_inline_size(name_len);
7671         err = btrfs_insert_empty_item(trans, root, path, &key,
7672                                       datasize);
7673         if (err) {
7674                 drop_inode = 1;
7675                 btrfs_free_path(path);
7676                 goto out_unlock;
7677         }
7678         leaf = path->nodes[0];
7679         ei = btrfs_item_ptr(leaf, path->slots[0],
7680                             struct btrfs_file_extent_item);
7681         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7682         btrfs_set_file_extent_type(leaf, ei,
7683                                    BTRFS_FILE_EXTENT_INLINE);
7684         btrfs_set_file_extent_encryption(leaf, ei, 0);
7685         btrfs_set_file_extent_compression(leaf, ei, 0);
7686         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7687         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7688
7689         ptr = btrfs_file_extent_inline_start(ei);
7690         write_extent_buffer(leaf, symname, ptr, name_len);
7691         btrfs_mark_buffer_dirty(leaf);
7692         btrfs_free_path(path);
7693
7694         inode->i_op = &btrfs_symlink_inode_operations;
7695         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7696         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7697         inode_set_bytes(inode, name_len);
7698         btrfs_i_size_write(inode, name_len - 1);
7699         err = btrfs_update_inode(trans, root, inode);
7700         if (err)
7701                 drop_inode = 1;
7702
7703 out_unlock:
7704         if (!err)
7705                 d_instantiate(dentry, inode);
7706         btrfs_end_transaction(trans, root);
7707         if (drop_inode) {
7708                 inode_dec_link_count(inode);
7709                 iput(inode);
7710         }
7711         btrfs_btree_balance_dirty(root);
7712         return err;
7713 }
7714
7715 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7716                                        u64 start, u64 num_bytes, u64 min_size,
7717                                        loff_t actual_len, u64 *alloc_hint,
7718                                        struct btrfs_trans_handle *trans)
7719 {
7720         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7721         struct extent_map *em;
7722         struct btrfs_root *root = BTRFS_I(inode)->root;
7723         struct btrfs_key ins;
7724         u64 cur_offset = start;
7725         u64 i_size;
7726         int ret = 0;
7727         bool own_trans = true;
7728
7729         if (trans)
7730                 own_trans = false;
7731         while (num_bytes > 0) {
7732                 if (own_trans) {
7733                         trans = btrfs_start_transaction(root, 3);
7734                         if (IS_ERR(trans)) {
7735                                 ret = PTR_ERR(trans);
7736                                 break;
7737                         }
7738                 }
7739
7740                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7741                                            0, *alloc_hint, &ins, 1);
7742                 if (ret) {
7743                         if (own_trans)
7744                                 btrfs_end_transaction(trans, root);
7745                         break;
7746                 }
7747
7748                 ret = insert_reserved_file_extent(trans, inode,
7749                                                   cur_offset, ins.objectid,
7750                                                   ins.offset, ins.offset,
7751                                                   ins.offset, 0, 0, 0,
7752                                                   BTRFS_FILE_EXTENT_PREALLOC);
7753                 if (ret) {
7754                         btrfs_abort_transaction(trans, root, ret);
7755                         if (own_trans)
7756                                 btrfs_end_transaction(trans, root);
7757                         break;
7758                 }
7759                 btrfs_drop_extent_cache(inode, cur_offset,
7760                                         cur_offset + ins.offset -1, 0);
7761
7762                 em = alloc_extent_map();
7763                 if (!em) {
7764                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7765                                 &BTRFS_I(inode)->runtime_flags);
7766                         goto next;
7767                 }
7768
7769                 em->start = cur_offset;
7770                 em->orig_start = cur_offset;
7771                 em->len = ins.offset;
7772                 em->block_start = ins.objectid;
7773                 em->block_len = ins.offset;
7774                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7775                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7776                 em->generation = trans->transid;
7777
7778                 while (1) {
7779                         write_lock(&em_tree->lock);
7780                         ret = add_extent_mapping(em_tree, em);
7781                         if (!ret)
7782                                 list_move(&em->list,
7783                                           &em_tree->modified_extents);
7784                         write_unlock(&em_tree->lock);
7785                         if (ret != -EEXIST)
7786                                 break;
7787                         btrfs_drop_extent_cache(inode, cur_offset,
7788                                                 cur_offset + ins.offset - 1,
7789                                                 0);
7790                 }
7791                 free_extent_map(em);
7792 next:
7793                 num_bytes -= ins.offset;
7794                 cur_offset += ins.offset;
7795                 *alloc_hint = ins.objectid + ins.offset;
7796
7797                 inode_inc_iversion(inode);
7798                 inode->i_ctime = CURRENT_TIME;
7799                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7800                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7801                     (actual_len > inode->i_size) &&
7802                     (cur_offset > inode->i_size)) {
7803                         if (cur_offset > actual_len)
7804                                 i_size = actual_len;
7805                         else
7806                                 i_size = cur_offset;
7807                         i_size_write(inode, i_size);
7808                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7809                 }
7810
7811                 ret = btrfs_update_inode(trans, root, inode);
7812
7813                 if (ret) {
7814                         btrfs_abort_transaction(trans, root, ret);
7815                         if (own_trans)
7816                                 btrfs_end_transaction(trans, root);
7817                         break;
7818                 }
7819
7820                 if (own_trans)
7821                         btrfs_end_transaction(trans, root);
7822         }
7823         return ret;
7824 }
7825
7826 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7827                               u64 start, u64 num_bytes, u64 min_size,
7828                               loff_t actual_len, u64 *alloc_hint)
7829 {
7830         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7831                                            min_size, actual_len, alloc_hint,
7832                                            NULL);
7833 }
7834
7835 int btrfs_prealloc_file_range_trans(struct inode *inode,
7836                                     struct btrfs_trans_handle *trans, int mode,
7837                                     u64 start, u64 num_bytes, u64 min_size,
7838                                     loff_t actual_len, u64 *alloc_hint)
7839 {
7840         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7841                                            min_size, actual_len, alloc_hint, trans);
7842 }
7843
7844 static int btrfs_set_page_dirty(struct page *page)
7845 {
7846         return __set_page_dirty_nobuffers(page);
7847 }
7848
7849 static int btrfs_permission(struct inode *inode, int mask)
7850 {
7851         struct btrfs_root *root = BTRFS_I(inode)->root;
7852         umode_t mode = inode->i_mode;
7853
7854         if (mask & MAY_WRITE &&
7855             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7856                 if (btrfs_root_readonly(root))
7857                         return -EROFS;
7858                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7859                         return -EACCES;
7860         }
7861         return generic_permission(inode, mask);
7862 }
7863
7864 static const struct inode_operations btrfs_dir_inode_operations = {
7865         .getattr        = btrfs_getattr,
7866         .lookup         = btrfs_lookup,
7867         .create         = btrfs_create,
7868         .unlink         = btrfs_unlink,
7869         .link           = btrfs_link,
7870         .mkdir          = btrfs_mkdir,
7871         .rmdir          = btrfs_rmdir,
7872         .rename         = btrfs_rename,
7873         .symlink        = btrfs_symlink,
7874         .setattr        = btrfs_setattr,
7875         .mknod          = btrfs_mknod,
7876         .setxattr       = btrfs_setxattr,
7877         .getxattr       = btrfs_getxattr,
7878         .listxattr      = btrfs_listxattr,
7879         .removexattr    = btrfs_removexattr,
7880         .permission     = btrfs_permission,
7881         .get_acl        = btrfs_get_acl,
7882 };
7883 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7884         .lookup         = btrfs_lookup,
7885         .permission     = btrfs_permission,
7886         .get_acl        = btrfs_get_acl,
7887 };
7888
7889 static const struct file_operations btrfs_dir_file_operations = {
7890         .llseek         = generic_file_llseek,
7891         .read           = generic_read_dir,
7892         .readdir        = btrfs_real_readdir,
7893         .unlocked_ioctl = btrfs_ioctl,
7894 #ifdef CONFIG_COMPAT
7895         .compat_ioctl   = btrfs_ioctl,
7896 #endif
7897         .release        = btrfs_release_file,
7898         .fsync          = btrfs_sync_file,
7899 };
7900
7901 static struct extent_io_ops btrfs_extent_io_ops = {
7902         .fill_delalloc = run_delalloc_range,
7903         .submit_bio_hook = btrfs_submit_bio_hook,
7904         .merge_bio_hook = btrfs_merge_bio_hook,
7905         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7906         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7907         .writepage_start_hook = btrfs_writepage_start_hook,
7908         .set_bit_hook = btrfs_set_bit_hook,
7909         .clear_bit_hook = btrfs_clear_bit_hook,
7910         .merge_extent_hook = btrfs_merge_extent_hook,
7911         .split_extent_hook = btrfs_split_extent_hook,
7912 };
7913
7914 /*
7915  * btrfs doesn't support the bmap operation because swapfiles
7916  * use bmap to make a mapping of extents in the file.  They assume
7917  * these extents won't change over the life of the file and they
7918  * use the bmap result to do IO directly to the drive.
7919  *
7920  * the btrfs bmap call would return logical addresses that aren't
7921  * suitable for IO and they also will change frequently as COW
7922  * operations happen.  So, swapfile + btrfs == corruption.
7923  *
7924  * For now we're avoiding this by dropping bmap.
7925  */
7926 static const struct address_space_operations btrfs_aops = {
7927         .readpage       = btrfs_readpage,
7928         .writepage      = btrfs_writepage,
7929         .writepages     = btrfs_writepages,
7930         .readpages      = btrfs_readpages,
7931         .direct_IO      = btrfs_direct_IO,
7932         .invalidatepage = btrfs_invalidatepage,
7933         .releasepage    = btrfs_releasepage,
7934         .set_page_dirty = btrfs_set_page_dirty,
7935         .error_remove_page = generic_error_remove_page,
7936 };
7937
7938 static const struct address_space_operations btrfs_symlink_aops = {
7939         .readpage       = btrfs_readpage,
7940         .writepage      = btrfs_writepage,
7941         .invalidatepage = btrfs_invalidatepage,
7942         .releasepage    = btrfs_releasepage,
7943 };
7944
7945 static const struct inode_operations btrfs_file_inode_operations = {
7946         .getattr        = btrfs_getattr,
7947         .setattr        = btrfs_setattr,
7948         .setxattr       = btrfs_setxattr,
7949         .getxattr       = btrfs_getxattr,
7950         .listxattr      = btrfs_listxattr,
7951         .removexattr    = btrfs_removexattr,
7952         .permission     = btrfs_permission,
7953         .fiemap         = btrfs_fiemap,
7954         .get_acl        = btrfs_get_acl,
7955         .update_time    = btrfs_update_time,
7956 };
7957 static const struct inode_operations btrfs_special_inode_operations = {
7958         .getattr        = btrfs_getattr,
7959         .setattr        = btrfs_setattr,
7960         .permission     = btrfs_permission,
7961         .setxattr       = btrfs_setxattr,
7962         .getxattr       = btrfs_getxattr,
7963         .listxattr      = btrfs_listxattr,
7964         .removexattr    = btrfs_removexattr,
7965         .get_acl        = btrfs_get_acl,
7966         .update_time    = btrfs_update_time,
7967 };
7968 static const struct inode_operations btrfs_symlink_inode_operations = {
7969         .readlink       = generic_readlink,
7970         .follow_link    = page_follow_link_light,
7971         .put_link       = page_put_link,
7972         .getattr        = btrfs_getattr,
7973         .setattr        = btrfs_setattr,
7974         .permission     = btrfs_permission,
7975         .setxattr       = btrfs_setxattr,
7976         .getxattr       = btrfs_getxattr,
7977         .listxattr      = btrfs_listxattr,
7978         .removexattr    = btrfs_removexattr,
7979         .get_acl        = btrfs_get_acl,
7980         .update_time    = btrfs_update_time,
7981 };
7982
7983 const struct dentry_operations btrfs_dentry_operations = {
7984         .d_delete       = btrfs_dentry_delete,
7985         .d_release      = btrfs_dentry_release,
7986 };