fs/nilfs2/inode.c

   1 /*
   2  * inode.c - NILFS inode operations.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/buffer_head.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mpage.h>
  27 #include <linux/writeback.h>
  28 #include <linux/aio.h>
  29 #include "nilfs.h"
  30 #include "btnode.h"
  31 #include "segment.h"
  32 #include "page.h"
  33 #include "mdt.h"
  34 #include "cpfile.h"
  35 #include "ifile.h"
  36
  37 /**
  38  * struct nilfs_iget_args - arguments used during comparison between inodes
  39  * @ino: inode number
  40  * @cno: checkpoint number
  41  * @root: pointer on NILFS root object (mounted checkpoint)
  42  * @for_gc: inode for GC flag
  43  */
  44 struct nilfs_iget_args {
  45         u64 ino;
  46         __u64 cno;
  47         struct nilfs_root *root;
  48         int for_gc;
  49 };
  50
  51 void nilfs_inode_add_blocks(struct inode *inode, int n)
  52 {
  53         struct nilfs_root *root = NILFS_I(inode)->i_root;
  54
  55         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  56         if (root)
  57                 atomic64_add(n, &root->blocks_count);
  58 }
  59
  60 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  61 {
  62         struct nilfs_root *root = NILFS_I(inode)->i_root;
  63
  64         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  65         if (root)
  66                 atomic64_sub(n, &root->blocks_count);
  67 }
  68
  69 /**
  70  * nilfs_get_block() - get a file block on the filesystem (callback function)
  71  * @inode - inode struct of the target file
  72  * @blkoff - file block number
  73  * @bh_result - buffer head to be mapped on
  74  * @create - indicate whether allocating the block or not when it has not
  75  *      been allocated yet.
  76  *
  77  * This function does not issue actual read request of the specified data
  78  * block. It is done by VFS.
  79  */
  80 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  81                     struct buffer_head *bh_result, int create)
  82 {
  83         struct nilfs_inode_info *ii = NILFS_I(inode);
  84         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  85         __u64 blknum = 0;
  86         int err = 0, ret;
  87         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  88
  89         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  90         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  91         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  92         if (ret >= 0) { /* found */
  93                 map_bh(bh_result, inode->i_sb, blknum);
  94                 if (ret > 0)
  95                         bh_result->b_size = (ret << inode->i_blkbits);
  96                 goto out;
  97         }
  98         /* data block was not found */
  99         if (ret == -ENOENT && create) {
 100                 struct nilfs_transaction_info ti;
 101
 102                 bh_result->b_blocknr = 0;
 103                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 104                 if (unlikely(err))
 105                         goto out;
 106                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 107                                         (unsigned long)bh_result);
 108                 if (unlikely(err != 0)) {
 109                         if (err == -EEXIST) {
 110                                 /*
 111                                  * The get_block() function could be called
 112                                  * from multiple callers for an inode.
 113                                  * However, the page having this block must
 114                                  * be locked in this case.
 115                                  */
 116                                 printk(KERN_WARNING
 117                                        "nilfs_get_block: a race condition "
 118                                        "while inserting a data block. "
 119                                        "(inode number=%lu, file block "
 120                                        "offset=%llu)\n",
 121                                        inode->i_ino,
 122                                        (unsigned long long)blkoff);
 123                                 err = 0;
 124                         }
 125                         nilfs_transaction_abort(inode->i_sb);
 126                         goto out;
 127                 }
 128                 nilfs_mark_inode_dirty(inode);
 129                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 130                 /* Error handling should be detailed */
 131                 set_buffer_new(bh_result);
 132                 set_buffer_delay(bh_result);
 133                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 134                                                       to proper value */
 135         } else if (ret == -ENOENT) {
 136                 /* not found is not error (e.g. hole); must return without
 137                    the mapped state flag. */
 138                 ;
 139         } else {
 140                 err = ret;
 141         }
 142
 143  out:
 144         return err;
 145 }
 146
 147 /**
 148  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 149  * address_space_operations.
 150  * @file - file struct of the file to be read
 151  * @page - the page to be read
 152  */
 153 static int nilfs_readpage(struct file *file, struct page *page)
 154 {
 155         return mpage_readpage(page, nilfs_get_block);
 156 }
 157
 158 /**
 159  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 160  * address_space_operations.
 161  * @file - file struct of the file to be read
 162  * @mapping - address_space struct used for reading multiple pages
 163  * @pages - the pages to be read
 164  * @nr_pages - number of pages to be read
 165  */
 166 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 167                            struct list_head *pages, unsigned nr_pages)
 168 {
 169         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 170 }
 171
 172 static int nilfs_writepages(struct address_space *mapping,
 173                             struct writeback_control *wbc)
 174 {
 175         struct inode *inode = mapping->host;
 176         int err = 0;
 177
 178         if (inode->i_sb->s_flags & MS_RDONLY) {
 179                 nilfs_clear_dirty_pages(mapping, false);
 180                 return -EROFS;
 181         }
 182
 183         if (wbc->sync_mode == WB_SYNC_ALL)
 184                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 185                                                     wbc->range_start,
 186                                                     wbc->range_end);
 187         return err;
 188 }
 189
 190 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 191 {
 192         struct inode *inode = page->mapping->host;
 193         int err;
 194
 195         if (inode->i_sb->s_flags & MS_RDONLY) {
 196                 /*
 197                  * It means that filesystem was remounted in read-only
 198                  * mode because of error or metadata corruption. But we
 199                  * have dirty pages that try to be flushed in background.
 200                  * So, here we simply discard this dirty page.
 201                  */
 202                 nilfs_clear_dirty_page(page, false);
 203                 unlock_page(page);
 204                 return -EROFS;
 205         }
 206
 207         redirty_page_for_writepage(wbc, page);
 208         unlock_page(page);
 209
 210         if (wbc->sync_mode == WB_SYNC_ALL) {
 211                 err = nilfs_construct_segment(inode->i_sb);
 212                 if (unlikely(err))
 213                         return err;
 214         } else if (wbc->for_reclaim)
 215                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 216
 217         return 0;
 218 }
 219
 220 static int nilfs_set_page_dirty(struct page *page)
 221 {
 222         int ret = __set_page_dirty_nobuffers(page);
 223
 224         if (page_has_buffers(page)) {
 225                 struct inode *inode = page->mapping->host;
 226                 unsigned nr_dirty = 0;
 227                 struct buffer_head *bh, *head;
 228
 229                 /*
 230                  * This page is locked by callers, and no other thread
 231                  * concurrently marks its buffers dirty since they are
 232                  * only dirtied through routines in fs/buffer.c in
 233                  * which call sites of mark_buffer_dirty are protected
 234                  * by page lock.
 235                  */
 236                 bh = head = page_buffers(page);
 237                 do {
 238                         /* Do not mark hole blocks dirty */
 239                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 240                                 continue;
 241
 242                         set_buffer_dirty(bh);
 243                         nr_dirty++;
 244                 } while (bh = bh->b_this_page, bh != head);
 245
 246                 if (nr_dirty)
 247                         nilfs_set_file_dirty(inode, nr_dirty);
 248         }
 249         return ret;
 250 }
 251
 252 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 253 {
 254         struct inode *inode = mapping->host;
 255
 256         if (to > inode->i_size) {
 257                 truncate_pagecache(inode, inode->i_size);
 258                 nilfs_truncate(inode);
 259         }
 260 }
 261
 262 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 263                              loff_t pos, unsigned len, unsigned flags,
 264                              struct page **pagep, void **fsdata)
 265
 266 {
 267         struct inode *inode = mapping->host;
 268         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 269
 270         if (unlikely(err))
 271                 return err;
 272
 273         err = block_write_begin(mapping, pos, len, flags, pagep,
 274                                 nilfs_get_block);
 275         if (unlikely(err)) {
 276                 nilfs_write_failed(mapping, pos + len);
 277                 nilfs_transaction_abort(inode->i_sb);
 278         }
 279         return err;
 280 }
 281
 282 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 283                            loff_t pos, unsigned len, unsigned copied,
 284                            struct page *page, void *fsdata)
 285 {
 286         struct inode *inode = mapping->host;
 287         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 288         unsigned nr_dirty;
 289         int err;
 290
 291         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 292                                                   start + copied);
 293         copied = generic_write_end(file, mapping, pos, len, copied, page,
 294                                    fsdata);
 295         nilfs_set_file_dirty(inode, nr_dirty);
 296         err = nilfs_transaction_commit(inode->i_sb);
 297         return err ? : copied;
 298 }
 299
 300 static ssize_t
 301 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 302                 loff_t offset, unsigned long nr_segs)
 303 {
 304         struct file *file = iocb->ki_filp;
 305         struct address_space *mapping = file->f_mapping;
 306         struct inode *inode = file->f_mapping->host;
 307         ssize_t size;
 308
 309         if (rw == WRITE)
 310                 return 0;
 311
 312         /* Needs synchronization with the cleaner */
 313         size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 314                                   nilfs_get_block);
 315
 316         /*
 317          * In case of error extending write may have instantiated a few
 318          * blocks outside i_size. Trim these off again.
 319          */
 320         if (unlikely((rw & WRITE) && size < 0)) {
 321                 loff_t isize = i_size_read(inode);
 322                 loff_t end = offset + iov_length(iov, nr_segs);
 323
 324                 if (end > isize)
 325                         nilfs_write_failed(mapping, end);
 326         }
 327
 328         return size;
 329 }
 330
 331 const struct address_space_operations nilfs_aops = {
 332         .writepage              = nilfs_writepage,
 333         .readpage               = nilfs_readpage,
 334         .writepages             = nilfs_writepages,
 335         .set_page_dirty         = nilfs_set_page_dirty,
 336         .readpages              = nilfs_readpages,
 337         .write_begin            = nilfs_write_begin,
 338         .write_end              = nilfs_write_end,
 339         /* .releasepage         = nilfs_releasepage, */
 340         .invalidatepage         = block_invalidatepage,
 341         .direct_IO              = nilfs_direct_IO,
 342         .is_partially_uptodate  = block_is_partially_uptodate,
 343 };
 344
 345 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 346 {
 347         struct super_block *sb = dir->i_sb;
 348         struct the_nilfs *nilfs = sb->s_fs_info;
 349         struct inode *inode;
 350         struct nilfs_inode_info *ii;
 351         struct nilfs_root *root;
 352         int err = -ENOMEM;
 353         ino_t ino;
 354
 355         inode = new_inode(sb);
 356         if (unlikely(!inode))
 357                 goto failed;
 358
 359         mapping_set_gfp_mask(inode->i_mapping,
 360                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 361
 362         root = NILFS_I(dir)->i_root;
 363         ii = NILFS_I(inode);
 364         ii->i_state = 1 << NILFS_I_NEW;
 365         ii->i_root = root;
 366
 367         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 368         if (unlikely(err))
 369                 goto failed_ifile_create_inode;
 370         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 371
 372         atomic64_inc(&root->inodes_count);
 373         inode_init_owner(inode, dir, mode);
 374         inode->i_ino = ino;
 375         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 376
 377         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 378                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 379                 if (err < 0)
 380                         goto failed_bmap;
 381
 382                 set_bit(NILFS_I_BMAP, &ii->i_state);
 383                 /* No lock is needed; iget() ensures it. */
 384         }
 385
 386         ii->i_flags = nilfs_mask_flags(
 387                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 388
 389         /* ii->i_file_acl = 0; */
 390         /* ii->i_dir_acl = 0; */
 391         ii->i_dir_start_lookup = 0;
 392         nilfs_set_inode_flags(inode);
 393         spin_lock(&nilfs->ns_next_gen_lock);
 394         inode->i_generation = nilfs->ns_next_generation++;
 395         spin_unlock(&nilfs->ns_next_gen_lock);
 396         insert_inode_hash(inode);
 397
 398         err = nilfs_init_acl(inode, dir);
 399         if (unlikely(err))
 400                 goto failed_acl; /* never occur. When supporting
 401                                     nilfs_init_acl(), proper cancellation of
 402                                     above jobs should be considered */
 403
 404         return inode;
 405
 406  failed_acl:
 407  failed_bmap:
 408         clear_nlink(inode);
 409         iput(inode);  /* raw_inode will be deleted through
 410                          generic_delete_inode() */
 411         goto failed;
 412
 413  failed_ifile_create_inode:
 414         make_bad_inode(inode);
 415         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 416                          called */
 417  failed:
 418         return ERR_PTR(err);
 419 }
 420
 421 void nilfs_set_inode_flags(struct inode *inode)
 422 {
 423         unsigned int flags = NILFS_I(inode)->i_flags;
 424
 425         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 426                             S_DIRSYNC);
 427         if (flags & FS_SYNC_FL)
 428                 inode->i_flags |= S_SYNC;
 429         if (flags & FS_APPEND_FL)
 430                 inode->i_flags |= S_APPEND;
 431         if (flags & FS_IMMUTABLE_FL)
 432                 inode->i_flags |= S_IMMUTABLE;
 433         if (flags & FS_NOATIME_FL)
 434                 inode->i_flags |= S_NOATIME;
 435         if (flags & FS_DIRSYNC_FL)
 436                 inode->i_flags |= S_DIRSYNC;
 437         mapping_set_gfp_mask(inode->i_mapping,
 438                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 439 }
 440
 441 int nilfs_read_inode_common(struct inode *inode,
 442                             struct nilfs_inode *raw_inode)
 443 {
 444         struct nilfs_inode_info *ii = NILFS_I(inode);
 445         int err;
 446
 447         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 448         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 449         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 450         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 451         inode->i_size = le64_to_cpu(raw_inode->i_size);
 452         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 453         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 454         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 455         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 456         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 457         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 458         if (inode->i_nlink == 0 && inode->i_mode == 0)
 459                 return -EINVAL; /* this inode is deleted */
 460
 461         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 462         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 463 #if 0
 464         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 465         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 466                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 467 #endif
 468         ii->i_dir_start_lookup = 0;
 469         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 470
 471         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 472             S_ISLNK(inode->i_mode)) {
 473                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 474                 if (err < 0)
 475                         return err;
 476                 set_bit(NILFS_I_BMAP, &ii->i_state);
 477                 /* No lock is needed; iget() ensures it. */
 478         }
 479         return 0;
 480 }
 481
 482 static int __nilfs_read_inode(struct super_block *sb,
 483                               struct nilfs_root *root, unsigned long ino,
 484                               struct inode *inode)
 485 {
 486         struct the_nilfs *nilfs = sb->s_fs_info;
 487         struct buffer_head *bh;
 488         struct nilfs_inode *raw_inode;
 489         int err;
 490
 491         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 492         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 493         if (unlikely(err))
 494                 goto bad_inode;
 495
 496         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 497
 498         err = nilfs_read_inode_common(inode, raw_inode);
 499         if (err)
 500                 goto failed_unmap;
 501
 502         if (S_ISREG(inode->i_mode)) {
 503                 inode->i_op = &nilfs_file_inode_operations;
 504                 inode->i_fop = &nilfs_file_operations;
 505                 inode->i_mapping->a_ops = &nilfs_aops;
 506         } else if (S_ISDIR(inode->i_mode)) {
 507                 inode->i_op = &nilfs_dir_inode_operations;
 508                 inode->i_fop = &nilfs_dir_operations;
 509                 inode->i_mapping->a_ops = &nilfs_aops;
 510         } else if (S_ISLNK(inode->i_mode)) {
 511                 inode->i_op = &nilfs_symlink_inode_operations;
 512                 inode->i_mapping->a_ops = &nilfs_aops;
 513         } else {
 514                 inode->i_op = &nilfs_special_inode_operations;
 515                 init_special_inode(
 516                         inode, inode->i_mode,
 517                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 518         }
 519         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 520         brelse(bh);
 521         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 522         nilfs_set_inode_flags(inode);
 523         return 0;
 524
 525  failed_unmap:
 526         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 527         brelse(bh);
 528
 529  bad_inode:
 530         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 531         return err;
 532 }
 533
 534 static int nilfs_iget_test(struct inode *inode, void *opaque)
 535 {
 536         struct nilfs_iget_args *args = opaque;
 537         struct nilfs_inode_info *ii;
 538
 539         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 540                 return 0;
 541
 542         ii = NILFS_I(inode);
 543         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 544                 return !args->for_gc;
 545
 546         return args->for_gc && args->cno == ii->i_cno;
 547 }
 548
 549 static int nilfs_iget_set(struct inode *inode, void *opaque)
 550 {
 551         struct nilfs_iget_args *args = opaque;
 552
 553         inode->i_ino = args->ino;
 554         if (args->for_gc) {
 555                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 556                 NILFS_I(inode)->i_cno = args->cno;
 557                 NILFS_I(inode)->i_root = NULL;
 558         } else {
 559                 if (args->root && args->ino == NILFS_ROOT_INO)
 560                         nilfs_get_root(args->root);
 561                 NILFS_I(inode)->i_root = args->root;
 562         }
 563         return 0;
 564 }
 565
 566 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 567                             unsigned long ino)
 568 {
 569         struct nilfs_iget_args args = {
 570                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 571         };
 572
 573         return ilookup5(sb, ino, nilfs_iget_test, &args);
 574 }
 575
 576 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 577                                 unsigned long ino)
 578 {
 579         struct nilfs_iget_args args = {
 580                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 581         };
 582
 583         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 584 }
 585
 586 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 587                          unsigned long ino)
 588 {
 589         struct inode *inode;
 590         int err;
 591
 592         inode = nilfs_iget_locked(sb, root, ino);
 593         if (unlikely(!inode))
 594                 return ERR_PTR(-ENOMEM);
 595         if (!(inode->i_state & I_NEW))
 596                 return inode;
 597
 598         err = __nilfs_read_inode(sb, root, ino, inode);
 599         if (unlikely(err)) {
 600                 iget_failed(inode);
 601                 return ERR_PTR(err);
 602         }
 603         unlock_new_inode(inode);
 604         return inode;
 605 }
 606
 607 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 608                                 __u64 cno)
 609 {
 610         struct nilfs_iget_args args = {
 611                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 612         };
 613         struct inode *inode;
 614         int err;
 615
 616         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 617         if (unlikely(!inode))
 618                 return ERR_PTR(-ENOMEM);
 619         if (!(inode->i_state & I_NEW))
 620                 return inode;
 621
 622         err = nilfs_init_gcinode(inode);
 623         if (unlikely(err)) {
 624                 iget_failed(inode);
 625                 return ERR_PTR(err);
 626         }
 627         unlock_new_inode(inode);
 628         return inode;
 629 }
 630
 631 void nilfs_write_inode_common(struct inode *inode,
 632                               struct nilfs_inode *raw_inode, int has_bmap)
 633 {
 634         struct nilfs_inode_info *ii = NILFS_I(inode);
 635
 636         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 637         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 638         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 639         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 640         raw_inode->i_size = cpu_to_le64(inode->i_size);
 641         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 642         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 643         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 644         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 645         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 646
 647         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 648         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 649
 650         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 651                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 652
 653                 /* zero-fill unused portion in the case of super root block */
 654                 raw_inode->i_xattr = 0;
 655                 raw_inode->i_pad = 0;
 656                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 657                        nilfs->ns_inode_size - sizeof(*raw_inode));
 658         }
 659
 660         if (has_bmap)
 661                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 662         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 663                 raw_inode->i_device_code =
 664                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 665         /* When extending inode, nilfs->ns_inode_size should be checked
 666            for substitutions of appended fields */
 667 }
 668
 669 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 670 {
 671         ino_t ino = inode->i_ino;
 672         struct nilfs_inode_info *ii = NILFS_I(inode);
 673         struct inode *ifile = ii->i_root->ifile;
 674         struct nilfs_inode *raw_inode;
 675
 676         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 677
 678         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 679                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 680         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 681
 682         nilfs_write_inode_common(inode, raw_inode, 0);
 683                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 684                    deadlock of bmap. This delays update of i_bmap to just
 685                    before writing */
 686         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 687 }
 688
 689 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 690
 691 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 692                                 unsigned long from)
 693 {
 694         unsigned long b;
 695         int ret;
 696
 697         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 698                 return;
 699 repeat:
 700         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 701         if (ret == -ENOENT)
 702                 return;
 703         else if (ret < 0)
 704                 goto failed;
 705
 706         if (b < from)
 707                 return;
 708
 709         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 710         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 711         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 712         if (!ret || (ret == -ENOMEM &&
 713                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 714                 goto repeat;
 715
 716 failed:
 717         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 718                       "failed to truncate bmap (ino=%lu, err=%d)",
 719                       ii->vfs_inode.i_ino, ret);
 720 }
 721
 722 void nilfs_truncate(struct inode *inode)
 723 {
 724         unsigned long blkoff;
 725         unsigned int blocksize;
 726         struct nilfs_transaction_info ti;
 727         struct super_block *sb = inode->i_sb;
 728         struct nilfs_inode_info *ii = NILFS_I(inode);
 729
 730         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 731                 return;
 732         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 733                 return;
 734
 735         blocksize = sb->s_blocksize;
 736         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 737         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 738
 739         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 740
 741         nilfs_truncate_bmap(ii, blkoff);
 742
 743         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 744         if (IS_SYNC(inode))
 745                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 746
 747         nilfs_mark_inode_dirty(inode);
 748         nilfs_set_file_dirty(inode, 0);
 749         nilfs_transaction_commit(sb);
 750         /* May construct a logical segment and may fail in sync mode.
 751            But truncate has no return value. */
 752 }
 753
 754 static void nilfs_clear_inode(struct inode *inode)
 755 {
 756         struct nilfs_inode_info *ii = NILFS_I(inode);
 757         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 758
 759         /*
 760          * Free resources allocated in nilfs_read_inode(), here.
 761          */
 762         BUG_ON(!list_empty(&ii->i_dirty));
 763         brelse(ii->i_bh);
 764         ii->i_bh = NULL;
 765
 766         if (mdi && mdi->mi_palloc_cache)
 767                 nilfs_palloc_destroy_cache(inode);
 768
 769         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 770                 nilfs_bmap_clear(ii->i_bmap);
 771
 772         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 773
 774         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 775                 nilfs_put_root(ii->i_root);
 776 }
 777
 778 void nilfs_evict_inode(struct inode *inode)
 779 {
 780         struct nilfs_transaction_info ti;
 781         struct super_block *sb = inode->i_sb;
 782         struct nilfs_inode_info *ii = NILFS_I(inode);
 783         int ret;
 784
 785         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 786                 if (inode->i_data.nrpages)
 787                         truncate_inode_pages(&inode->i_data, 0);
 788                 clear_inode(inode);
 789                 nilfs_clear_inode(inode);
 790                 return;
 791         }
 792         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 793
 794         if (inode->i_data.nrpages)
 795                 truncate_inode_pages(&inode->i_data, 0);
 796
 797         /* TODO: some of the following operations may fail.  */
 798         nilfs_truncate_bmap(ii, 0);
 799         nilfs_mark_inode_dirty(inode);
 800         clear_inode(inode);
 801
 802         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 803         if (!ret)
 804                 atomic64_dec(&ii->i_root->inodes_count);
 805
 806         nilfs_clear_inode(inode);
 807
 808         if (IS_SYNC(inode))
 809                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 810         nilfs_transaction_commit(sb);
 811         /* May construct a logical segment and may fail in sync mode.
 812            But delete_inode has no return value. */
 813 }
 814
 815 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 816 {
 817         struct nilfs_transaction_info ti;
 818         struct inode *inode = dentry->d_inode;
 819         struct super_block *sb = inode->i_sb;
 820         int err;
 821
 822         err = inode_change_ok(inode, iattr);
 823         if (err)
 824                 return err;
 825
 826         err = nilfs_transaction_begin(sb, &ti, 0);
 827         if (unlikely(err))
 828                 return err;
 829
 830         if ((iattr->ia_valid & ATTR_SIZE) &&
 831             iattr->ia_size != i_size_read(inode)) {
 832                 inode_dio_wait(inode);
 833                 truncate_setsize(inode, iattr->ia_size);
 834                 nilfs_truncate(inode);
 835         }
 836
 837         setattr_copy(inode, iattr);
 838         mark_inode_dirty(inode);
 839
 840         if (iattr->ia_valid & ATTR_MODE) {
 841                 err = nilfs_acl_chmod(inode);
 842                 if (unlikely(err))
 843                         goto out_err;
 844         }
 845
 846         return nilfs_transaction_commit(sb);
 847
 848 out_err:
 849         nilfs_transaction_abort(sb);
 850         return err;
 851 }
 852
 853 int nilfs_permission(struct inode *inode, int mask)
 854 {
 855         struct nilfs_root *root = NILFS_I(inode)->i_root;
 856         if ((mask & MAY_WRITE) && root &&
 857             root->cno != NILFS_CPTREE_CURRENT_CNO)
 858                 return -EROFS; /* snapshot is not writable */
 859
 860         return generic_permission(inode, mask);
 861 }
 862
 863 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 864 {
 865         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 866         struct nilfs_inode_info *ii = NILFS_I(inode);
 867         int err;
 868
 869         spin_lock(&nilfs->ns_inode_lock);
 870         if (ii->i_bh == NULL) {
 871                 spin_unlock(&nilfs->ns_inode_lock);
 872                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 873                                                   inode->i_ino, pbh);
 874                 if (unlikely(err))
 875                         return err;
 876                 spin_lock(&nilfs->ns_inode_lock);
 877                 if (ii->i_bh == NULL)
 878                         ii->i_bh = *pbh;
 879                 else {
 880                         brelse(*pbh);
 881                         *pbh = ii->i_bh;
 882                 }
 883         } else
 884                 *pbh = ii->i_bh;
 885
 886         get_bh(*pbh);
 887         spin_unlock(&nilfs->ns_inode_lock);
 888         return 0;
 889 }
 890
 891 int nilfs_inode_dirty(struct inode *inode)
 892 {
 893         struct nilfs_inode_info *ii = NILFS_I(inode);
 894         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 895         int ret = 0;
 896
 897         if (!list_empty(&ii->i_dirty)) {
 898                 spin_lock(&nilfs->ns_inode_lock);
 899                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 900                         test_bit(NILFS_I_BUSY, &ii->i_state);
 901                 spin_unlock(&nilfs->ns_inode_lock);
 902         }
 903         return ret;
 904 }
 905
 906 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 907 {
 908         struct nilfs_inode_info *ii = NILFS_I(inode);
 909         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 910
 911         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 912
 913         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 914                 return 0;
 915
 916         spin_lock(&nilfs->ns_inode_lock);
 917         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 918             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 919                 /* Because this routine may race with nilfs_dispose_list(),
 920                    we have to check NILFS_I_QUEUED here, too. */
 921                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 922                         /* This will happen when somebody is freeing
 923                            this inode. */
 924                         nilfs_warning(inode->i_sb, __func__,
 925                                       "cannot get inode (ino=%lu)\n",
 926                                       inode->i_ino);
 927                         spin_unlock(&nilfs->ns_inode_lock);
 928                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 929                                            freeing inode */
 930                 }
 931                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 932                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 933         }
 934         spin_unlock(&nilfs->ns_inode_lock);
 935         return 0;
 936 }
 937
 938 int nilfs_mark_inode_dirty(struct inode *inode)
 939 {
 940         struct buffer_head *ibh;
 941         int err;
 942
 943         err = nilfs_load_inode_block(inode, &ibh);
 944         if (unlikely(err)) {
 945                 nilfs_warning(inode->i_sb, __func__,
 946                               "failed to reget inode block.\n");
 947                 return err;
 948         }
 949         nilfs_update_inode(inode, ibh);
 950         mark_buffer_dirty(ibh);
 951         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 952         brelse(ibh);
 953         return 0;
 954 }
 955
 956 /**
 957  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 958  * @inode: inode of the file to be registered.
 959  *
 960  * nilfs_dirty_inode() loads a inode block containing the specified
 961  * @inode and copies data from a nilfs_inode to a corresponding inode
 962  * entry in the inode block. This operation is excluded from the segment
 963  * construction. This function can be called both as a single operation
 964  * and as a part of indivisible file operations.
 965  */
 966 void nilfs_dirty_inode(struct inode *inode, int flags)
 967 {
 968         struct nilfs_transaction_info ti;
 969         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 970
 971         if (is_bad_inode(inode)) {
 972                 nilfs_warning(inode->i_sb, __func__,
 973                               "tried to mark bad_inode dirty. ignored.\n");
 974                 dump_stack();
 975                 return;
 976         }
 977         if (mdi) {
 978                 nilfs_mdt_mark_dirty(inode);
 979                 return;
 980         }
 981         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 982         nilfs_mark_inode_dirty(inode);
 983         nilfs_transaction_commit(inode->i_sb); /* never fails */
 984 }
 985
 986 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 987                  __u64 start, __u64 len)
 988 {
 989         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 990         __u64 logical = 0, phys = 0, size = 0;
 991         __u32 flags = 0;
 992         loff_t isize;
 993         sector_t blkoff, end_blkoff;
 994         sector_t delalloc_blkoff;
 995         unsigned long delalloc_blklen;
 996         unsigned int blkbits = inode->i_blkbits;
 997         int ret, n;
 998
 999         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1000         if (ret)
1001                 return ret;
1002
1003         mutex_lock(&inode->i_mutex);
1004
1005         isize = i_size_read(inode);
1006
1007         blkoff = start >> blkbits;
1008         end_blkoff = (start + len - 1) >> blkbits;
1009
1010         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1011                                                         &delalloc_blkoff);
1012
1013         do {
1014                 __u64 blkphy;
1015                 unsigned int maxblocks;
1016
1017                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1018                         if (size) {
1019                                 /* End of the current extent */
1020                                 ret = fiemap_fill_next_extent(
1021                                         fieinfo, logical, phys, size, flags);
1022                                 if (ret)
1023                                         break;
1024                         }
1025                         if (blkoff > end_blkoff)
1026                                 break;
1027
1028                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1029                         logical = blkoff << blkbits;
1030                         phys = 0;
1031                         size = delalloc_blklen << blkbits;
1032
1033                         blkoff = delalloc_blkoff + delalloc_blklen;
1034                         delalloc_blklen = nilfs_find_uncommitted_extent(
1035                                 inode, blkoff, &delalloc_blkoff);
1036                         continue;
1037                 }
1038
1039                 /*
1040                  * Limit the number of blocks that we look up so as
1041                  * not to get into the next delayed allocation extent.
1042                  */
1043                 maxblocks = INT_MAX;
1044                 if (delalloc_blklen)
1045                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1046                                           maxblocks);
1047                 blkphy = 0;
1048
1049                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1050                 n = nilfs_bmap_lookup_contig(
1051                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1052                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1053
1054                 if (n < 0) {
1055                         int past_eof;
1056
1057                         if (unlikely(n != -ENOENT))
1058                                 break; /* error */
1059
1060                         /* HOLE */
1061                         blkoff++;
1062                         past_eof = ((blkoff << blkbits) >= isize);
1063
1064                         if (size) {
1065                                 /* End of the current extent */
1066
1067                                 if (past_eof)
1068                                         flags |= FIEMAP_EXTENT_LAST;
1069
1070                                 ret = fiemap_fill_next_extent(
1071                                         fieinfo, logical, phys, size, flags);
1072                                 if (ret)
1073                                         break;
1074                                 size = 0;
1075                         }
1076                         if (blkoff > end_blkoff || past_eof)
1077                                 break;
1078                 } else {
1079                         if (size) {
1080                                 if (phys && blkphy << blkbits == phys + size) {
1081                                         /* The current extent goes on */
1082                                         size += n << blkbits;
1083                                 } else {
1084                                         /* Terminate the current extent */
1085                                         ret = fiemap_fill_next_extent(
1086                                                 fieinfo, logical, phys, size,
1087                                                 flags);
1088                                         if (ret || blkoff > end_blkoff)
1089                                                 break;
1090
1091                                         /* Start another extent */
1092                                         flags = FIEMAP_EXTENT_MERGED;
1093                                         logical = blkoff << blkbits;
1094                                         phys = blkphy << blkbits;
1095                                         size = n << blkbits;
1096                                 }
1097                         } else {
1098                                 /* Start a new extent */
1099                                 flags = FIEMAP_EXTENT_MERGED;
1100                                 logical = blkoff << blkbits;
1101                                 phys = blkphy << blkbits;
1102                                 size = n << blkbits;
1103                         }
1104                         blkoff += n;
1105                 }
1106                 cond_resched();
1107         } while (true);
1108
1109         /* If ret is 1 then we just hit the end of the extent array */
1110         if (ret == 1)
1111                 ret = 0;
1112
1113         mutex_unlock(&inode->i_mutex);
1114         return ret;
1115 }