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/pagemap.h>
  28 #include <linux/writeback.h>
  29 #include <linux/aio.h>
  30 #include "nilfs.h"
  31 #include "btnode.h"
  32 #include "segment.h"
  33 #include "page.h"
  34 #include "mdt.h"
  35 #include "cpfile.h"
  36 #include "ifile.h"
  37
  38 /**
  39  * struct nilfs_iget_args - arguments used during comparison between inodes
  40  * @ino: inode number
  41  * @cno: checkpoint number
  42  * @root: pointer on NILFS root object (mounted checkpoint)
  43  * @for_gc: inode for GC flag
  44  */
  45 struct nilfs_iget_args {
  46         u64 ino;
  47         __u64 cno;
  48         struct nilfs_root *root;
  49         int for_gc;
  50 };
  51
  52 void nilfs_inode_add_blocks(struct inode *inode, int n)
  53 {
  54         struct nilfs_root *root = NILFS_I(inode)->i_root;
  55
  56         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  57         if (root)
  58                 atomic64_add(n, &root->blocks_count);
  59 }
  60
  61 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  62 {
  63         struct nilfs_root *root = NILFS_I(inode)->i_root;
  64
  65         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  66         if (root)
  67                 atomic64_sub(n, &root->blocks_count);
  68 }
  69
  70 /**
  71  * nilfs_get_block() - get a file block on the filesystem (callback function)
  72  * @inode - inode struct of the target file
  73  * @blkoff - file block number
  74  * @bh_result - buffer head to be mapped on
  75  * @create - indicate whether allocating the block or not when it has not
  76  *      been allocated yet.
  77  *
  78  * This function does not issue actual read request of the specified data
  79  * block. It is done by VFS.
  80  */
  81 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  82                     struct buffer_head *bh_result, int create)
  83 {
  84         struct nilfs_inode_info *ii = NILFS_I(inode);
  85         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  86         __u64 blknum = 0;
  87         int err = 0, ret;
  88         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  89
  90         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  91         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  92         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  93         if (ret >= 0) { /* found */
  94                 map_bh(bh_result, inode->i_sb, blknum);
  95                 if (ret > 0)
  96                         bh_result->b_size = (ret << inode->i_blkbits);
  97                 goto out;
  98         }
  99         /* data block was not found */
 100         if (ret == -ENOENT && create) {
 101                 struct nilfs_transaction_info ti;
 102
 103                 bh_result->b_blocknr = 0;
 104                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 105                 if (unlikely(err))
 106                         goto out;
 107                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 108                                         (unsigned long)bh_result);
 109                 if (unlikely(err != 0)) {
 110                         if (err == -EEXIST) {
 111                                 /*
 112                                  * The get_block() function could be called
 113                                  * from multiple callers for an inode.
 114                                  * However, the page having this block must
 115                                  * be locked in this case.
 116                                  */
 117                                 printk(KERN_WARNING
 118                                        "nilfs_get_block: a race condition "
 119                                        "while inserting a data block. "
 120                                        "(inode number=%lu, file block "
 121                                        "offset=%llu)\n",
 122                                        inode->i_ino,
 123                                        (unsigned long long)blkoff);
 124                                 err = 0;
 125                         }
 126                         nilfs_transaction_abort(inode->i_sb);
 127                         goto out;
 128                 }
 129                 nilfs_mark_inode_dirty(inode);
 130                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 131                 /* Error handling should be detailed */
 132                 set_buffer_new(bh_result);
 133                 set_buffer_delay(bh_result);
 134                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 135                                                       to proper value */
 136         } else if (ret == -ENOENT) {
 137                 /* not found is not error (e.g. hole); must return without
 138                    the mapped state flag. */
 139                 ;
 140         } else {
 141                 err = ret;
 142         }
 143
 144  out:
 145         return err;
 146 }
 147
 148 /**
 149  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 150  * address_space_operations.
 151  * @file - file struct of the file to be read
 152  * @page - the page to be read
 153  */
 154 static int nilfs_readpage(struct file *file, struct page *page)
 155 {
 156         return mpage_readpage(page, nilfs_get_block);
 157 }
 158
 159 /**
 160  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 161  * address_space_operations.
 162  * @file - file struct of the file to be read
 163  * @mapping - address_space struct used for reading multiple pages
 164  * @pages - the pages to be read
 165  * @nr_pages - number of pages to be read
 166  */
 167 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 168                            struct list_head *pages, unsigned nr_pages)
 169 {
 170         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 171 }
 172
 173 static int nilfs_writepages(struct address_space *mapping,
 174                             struct writeback_control *wbc)
 175 {
 176         struct inode *inode = mapping->host;
 177         int err = 0;
 178
 179         if (inode->i_sb->s_flags & MS_RDONLY) {
 180                 nilfs_clear_dirty_pages(mapping, false);
 181                 return -EROFS;
 182         }
 183
 184         if (wbc->sync_mode == WB_SYNC_ALL)
 185                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 186                                                     wbc->range_start,
 187                                                     wbc->range_end);
 188         return err;
 189 }
 190
 191 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 192 {
 193         struct inode *inode = page->mapping->host;
 194         int err;
 195
 196         if (inode->i_sb->s_flags & MS_RDONLY) {
 197                 /*
 198                  * It means that filesystem was remounted in read-only
 199                  * mode because of error or metadata corruption. But we
 200                  * have dirty pages that try to be flushed in background.
 201                  * So, here we simply discard this dirty page.
 202                  */
 203                 nilfs_clear_dirty_page(page, false);
 204                 unlock_page(page);
 205                 return -EROFS;
 206         }
 207
 208         redirty_page_for_writepage(wbc, page);
 209         unlock_page(page);
 210
 211         if (wbc->sync_mode == WB_SYNC_ALL) {
 212                 err = nilfs_construct_segment(inode->i_sb);
 213                 if (unlikely(err))
 214                         return err;
 215         } else if (wbc->for_reclaim)
 216                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 217
 218         return 0;
 219 }
 220
 221 static int nilfs_set_page_dirty(struct page *page)
 222 {
 223         struct inode *inode = page->mapping->host;
 224         int ret = __set_page_dirty_nobuffers(page);
 225
 226         if (page_has_buffers(page)) {
 227                 unsigned nr_dirty = 0;
 228                 struct buffer_head *bh, *head;
 229
 230                 /*
 231                  * This page is locked by callers, and no other thread
 232                  * concurrently marks its buffers dirty since they are
 233                  * only dirtied through routines in fs/buffer.c in
 234                  * which call sites of mark_buffer_dirty are protected
 235                  * by page lock.
 236                  */
 237                 bh = head = page_buffers(page);
 238                 do {
 239                         /* Do not mark hole blocks dirty */
 240                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 241                                 continue;
 242
 243                         set_buffer_dirty(bh);
 244                         nr_dirty++;
 245                 } while (bh = bh->b_this_page, bh != head);
 246
 247                 if (nr_dirty)
 248                         nilfs_set_file_dirty(inode, nr_dirty);
 249         } else if (ret) {
 250                 unsigned nr_dirty = 1 << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 251
 252                 nilfs_set_file_dirty(inode, nr_dirty);
 253         }
 254         return ret;
 255 }
 256
 257 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 258 {
 259         struct inode *inode = mapping->host;
 260
 261         if (to > inode->i_size) {
 262                 truncate_pagecache(inode, inode->i_size);
 263                 nilfs_truncate(inode);
 264         }
 265 }
 266
 267 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 268                              loff_t pos, unsigned len, unsigned flags,
 269                              struct page **pagep, void **fsdata)
 270
 271 {
 272         struct inode *inode = mapping->host;
 273         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 274
 275         if (unlikely(err))
 276                 return err;
 277
 278         err = block_write_begin(mapping, pos, len, flags, pagep,
 279                                 nilfs_get_block);
 280         if (unlikely(err)) {
 281                 nilfs_write_failed(mapping, pos + len);
 282                 nilfs_transaction_abort(inode->i_sb);
 283         }
 284         return err;
 285 }
 286
 287 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 288                            loff_t pos, unsigned len, unsigned copied,
 289                            struct page *page, void *fsdata)
 290 {
 291         struct inode *inode = mapping->host;
 292         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 293         unsigned nr_dirty;
 294         int err;
 295
 296         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 297                                                   start + copied);
 298         copied = generic_write_end(file, mapping, pos, len, copied, page,
 299                                    fsdata);
 300         nilfs_set_file_dirty(inode, nr_dirty);
 301         err = nilfs_transaction_commit(inode->i_sb);
 302         return err ? : copied;
 303 }
 304
 305 static ssize_t
 306 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 307                 loff_t offset, unsigned long nr_segs)
 308 {
 309         struct file *file = iocb->ki_filp;
 310         struct address_space *mapping = file->f_mapping;
 311         struct inode *inode = file->f_mapping->host;
 312         ssize_t size;
 313
 314         if (rw == WRITE)
 315                 return 0;
 316
 317         /* Needs synchronization with the cleaner */
 318         size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 319                                   nilfs_get_block);
 320
 321         /*
 322          * In case of error extending write may have instantiated a few
 323          * blocks outside i_size. Trim these off again.
 324          */
 325         if (unlikely((rw & WRITE) && size < 0)) {
 326                 loff_t isize = i_size_read(inode);
 327                 loff_t end = offset + iov_length(iov, nr_segs);
 328
 329                 if (end > isize)
 330                         nilfs_write_failed(mapping, end);
 331         }
 332
 333         return size;
 334 }
 335
 336 const struct address_space_operations nilfs_aops = {
 337         .writepage              = nilfs_writepage,
 338         .readpage               = nilfs_readpage,
 339         .writepages             = nilfs_writepages,
 340         .set_page_dirty         = nilfs_set_page_dirty,
 341         .readpages              = nilfs_readpages,
 342         .write_begin            = nilfs_write_begin,
 343         .write_end              = nilfs_write_end,
 344         /* .releasepage         = nilfs_releasepage, */
 345         .invalidatepage         = block_invalidatepage,
 346         .direct_IO              = nilfs_direct_IO,
 347         .is_partially_uptodate  = block_is_partially_uptodate,
 348 };
 349
 350 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 351 {
 352         struct super_block *sb = dir->i_sb;
 353         struct the_nilfs *nilfs = sb->s_fs_info;
 354         struct inode *inode;
 355         struct nilfs_inode_info *ii;
 356         struct nilfs_root *root;
 357         int err = -ENOMEM;
 358         ino_t ino;
 359
 360         inode = new_inode(sb);
 361         if (unlikely(!inode))
 362                 goto failed;
 363
 364         mapping_set_gfp_mask(inode->i_mapping,
 365                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 366
 367         root = NILFS_I(dir)->i_root;
 368         ii = NILFS_I(inode);
 369         ii->i_state = 1 << NILFS_I_NEW;
 370         ii->i_root = root;
 371
 372         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 373         if (unlikely(err))
 374                 goto failed_ifile_create_inode;
 375         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 376
 377         atomic64_inc(&root->inodes_count);
 378         inode_init_owner(inode, dir, mode);
 379         inode->i_ino = ino;
 380         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 381
 382         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 383                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 384                 if (err < 0)
 385                         goto failed_bmap;
 386
 387                 set_bit(NILFS_I_BMAP, &ii->i_state);
 388                 /* No lock is needed; iget() ensures it. */
 389         }
 390
 391         ii->i_flags = nilfs_mask_flags(
 392                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 393
 394         /* ii->i_file_acl = 0; */
 395         /* ii->i_dir_acl = 0; */
 396         ii->i_dir_start_lookup = 0;
 397         nilfs_set_inode_flags(inode);
 398         spin_lock(&nilfs->ns_next_gen_lock);
 399         inode->i_generation = nilfs->ns_next_generation++;
 400         spin_unlock(&nilfs->ns_next_gen_lock);
 401         insert_inode_hash(inode);
 402
 403         err = nilfs_init_acl(inode, dir);
 404         if (unlikely(err))
 405                 goto failed_acl; /* never occur. When supporting
 406                                     nilfs_init_acl(), proper cancellation of
 407                                     above jobs should be considered */
 408
 409         return inode;
 410
 411  failed_acl:
 412  failed_bmap:
 413         clear_nlink(inode);
 414         iput(inode);  /* raw_inode will be deleted through
 415                          generic_delete_inode() */
 416         goto failed;
 417
 418  failed_ifile_create_inode:
 419         make_bad_inode(inode);
 420         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 421                          called */
 422  failed:
 423         return ERR_PTR(err);
 424 }
 425
 426 void nilfs_set_inode_flags(struct inode *inode)
 427 {
 428         unsigned int flags = NILFS_I(inode)->i_flags;
 429
 430         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 431                             S_DIRSYNC);
 432         if (flags & FS_SYNC_FL)
 433                 inode->i_flags |= S_SYNC;
 434         if (flags & FS_APPEND_FL)
 435                 inode->i_flags |= S_APPEND;
 436         if (flags & FS_IMMUTABLE_FL)
 437                 inode->i_flags |= S_IMMUTABLE;
 438         if (flags & FS_NOATIME_FL)
 439                 inode->i_flags |= S_NOATIME;
 440         if (flags & FS_DIRSYNC_FL)
 441                 inode->i_flags |= S_DIRSYNC;
 442         mapping_set_gfp_mask(inode->i_mapping,
 443                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 444 }
 445
 446 int nilfs_read_inode_common(struct inode *inode,
 447                             struct nilfs_inode *raw_inode)
 448 {
 449         struct nilfs_inode_info *ii = NILFS_I(inode);
 450         int err;
 451
 452         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 453         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 454         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 455         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 456         inode->i_size = le64_to_cpu(raw_inode->i_size);
 457         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 458         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 459         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 460         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 461         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 462         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 463         if (inode->i_nlink == 0 && inode->i_mode == 0)
 464                 return -EINVAL; /* this inode is deleted */
 465
 466         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 467         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 468 #if 0
 469         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 470         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 471                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 472 #endif
 473         ii->i_dir_start_lookup = 0;
 474         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 475
 476         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 477             S_ISLNK(inode->i_mode)) {
 478                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 479                 if (err < 0)
 480                         return err;
 481                 set_bit(NILFS_I_BMAP, &ii->i_state);
 482                 /* No lock is needed; iget() ensures it. */
 483         }
 484         return 0;
 485 }
 486
 487 static int __nilfs_read_inode(struct super_block *sb,
 488                               struct nilfs_root *root, unsigned long ino,
 489                               struct inode *inode)
 490 {
 491         struct the_nilfs *nilfs = sb->s_fs_info;
 492         struct buffer_head *bh;
 493         struct nilfs_inode *raw_inode;
 494         int err;
 495
 496         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 497         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 498         if (unlikely(err))
 499                 goto bad_inode;
 500
 501         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 502
 503         err = nilfs_read_inode_common(inode, raw_inode);
 504         if (err)
 505                 goto failed_unmap;
 506
 507         if (S_ISREG(inode->i_mode)) {
 508                 inode->i_op = &nilfs_file_inode_operations;
 509                 inode->i_fop = &nilfs_file_operations;
 510                 inode->i_mapping->a_ops = &nilfs_aops;
 511         } else if (S_ISDIR(inode->i_mode)) {
 512                 inode->i_op = &nilfs_dir_inode_operations;
 513                 inode->i_fop = &nilfs_dir_operations;
 514                 inode->i_mapping->a_ops = &nilfs_aops;
 515         } else if (S_ISLNK(inode->i_mode)) {
 516                 inode->i_op = &nilfs_symlink_inode_operations;
 517                 inode->i_mapping->a_ops = &nilfs_aops;
 518         } else {
 519                 inode->i_op = &nilfs_special_inode_operations;
 520                 init_special_inode(
 521                         inode, inode->i_mode,
 522                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 523         }
 524         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 525         brelse(bh);
 526         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 527         nilfs_set_inode_flags(inode);
 528         return 0;
 529
 530  failed_unmap:
 531         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 532         brelse(bh);
 533
 534  bad_inode:
 535         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 536         return err;
 537 }
 538
 539 static int nilfs_iget_test(struct inode *inode, void *opaque)
 540 {
 541         struct nilfs_iget_args *args = opaque;
 542         struct nilfs_inode_info *ii;
 543
 544         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 545                 return 0;
 546
 547         ii = NILFS_I(inode);
 548         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 549                 return !args->for_gc;
 550
 551         return args->for_gc && args->cno == ii->i_cno;
 552 }
 553
 554 static int nilfs_iget_set(struct inode *inode, void *opaque)
 555 {
 556         struct nilfs_iget_args *args = opaque;
 557
 558         inode->i_ino = args->ino;
 559         if (args->for_gc) {
 560                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 561                 NILFS_I(inode)->i_cno = args->cno;
 562                 NILFS_I(inode)->i_root = NULL;
 563         } else {
 564                 if (args->root && args->ino == NILFS_ROOT_INO)
 565                         nilfs_get_root(args->root);
 566                 NILFS_I(inode)->i_root = args->root;
 567         }
 568         return 0;
 569 }
 570
 571 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 572                             unsigned long ino)
 573 {
 574         struct nilfs_iget_args args = {
 575                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 576         };
 577
 578         return ilookup5(sb, ino, nilfs_iget_test, &args);
 579 }
 580
 581 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 582                                 unsigned long ino)
 583 {
 584         struct nilfs_iget_args args = {
 585                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 586         };
 587
 588         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 589 }
 590
 591 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 592                          unsigned long ino)
 593 {
 594         struct inode *inode;
 595         int err;
 596
 597         inode = nilfs_iget_locked(sb, root, ino);
 598         if (unlikely(!inode))
 599                 return ERR_PTR(-ENOMEM);
 600         if (!(inode->i_state & I_NEW))
 601                 return inode;
 602
 603         err = __nilfs_read_inode(sb, root, ino, inode);
 604         if (unlikely(err)) {
 605                 iget_failed(inode);
 606                 return ERR_PTR(err);
 607         }
 608         unlock_new_inode(inode);
 609         return inode;
 610 }
 611
 612 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 613                                 __u64 cno)
 614 {
 615         struct nilfs_iget_args args = {
 616                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 617         };
 618         struct inode *inode;
 619         int err;
 620
 621         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 622         if (unlikely(!inode))
 623                 return ERR_PTR(-ENOMEM);
 624         if (!(inode->i_state & I_NEW))
 625                 return inode;
 626
 627         err = nilfs_init_gcinode(inode);
 628         if (unlikely(err)) {
 629                 iget_failed(inode);
 630                 return ERR_PTR(err);
 631         }
 632         unlock_new_inode(inode);
 633         return inode;
 634 }
 635
 636 void nilfs_write_inode_common(struct inode *inode,
 637                               struct nilfs_inode *raw_inode, int has_bmap)
 638 {
 639         struct nilfs_inode_info *ii = NILFS_I(inode);
 640
 641         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 642         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 643         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 644         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 645         raw_inode->i_size = cpu_to_le64(inode->i_size);
 646         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 647         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 648         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 649         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 650         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 651
 652         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 653         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 654
 655         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 656                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 657
 658                 /* zero-fill unused portion in the case of super root block */
 659                 raw_inode->i_xattr = 0;
 660                 raw_inode->i_pad = 0;
 661                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 662                        nilfs->ns_inode_size - sizeof(*raw_inode));
 663         }
 664
 665         if (has_bmap)
 666                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 667         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 668                 raw_inode->i_device_code =
 669                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 670         /* When extending inode, nilfs->ns_inode_size should be checked
 671            for substitutions of appended fields */
 672 }
 673
 674 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 675 {
 676         ino_t ino = inode->i_ino;
 677         struct nilfs_inode_info *ii = NILFS_I(inode);
 678         struct inode *ifile = ii->i_root->ifile;
 679         struct nilfs_inode *raw_inode;
 680
 681         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 682
 683         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 684                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 685         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 686
 687         nilfs_write_inode_common(inode, raw_inode, 0);
 688                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 689                    deadlock of bmap. This delays update of i_bmap to just
 690                    before writing */
 691         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 692 }
 693
 694 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 695
 696 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 697                                 unsigned long from)
 698 {
 699         unsigned long b;
 700         int ret;
 701
 702         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 703                 return;
 704 repeat:
 705         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 706         if (ret == -ENOENT)
 707                 return;
 708         else if (ret < 0)
 709                 goto failed;
 710
 711         if (b < from)
 712                 return;
 713
 714         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 715         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 716         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 717         if (!ret || (ret == -ENOMEM &&
 718                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 719                 goto repeat;
 720
 721 failed:
 722         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 723                       "failed to truncate bmap (ino=%lu, err=%d)",
 724                       ii->vfs_inode.i_ino, ret);
 725 }
 726
 727 void nilfs_truncate(struct inode *inode)
 728 {
 729         unsigned long blkoff;
 730         unsigned int blocksize;
 731         struct nilfs_transaction_info ti;
 732         struct super_block *sb = inode->i_sb;
 733         struct nilfs_inode_info *ii = NILFS_I(inode);
 734
 735         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 736                 return;
 737         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 738                 return;
 739
 740         blocksize = sb->s_blocksize;
 741         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 742         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 743
 744         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 745
 746         nilfs_truncate_bmap(ii, blkoff);
 747
 748         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 749         if (IS_SYNC(inode))
 750                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 751
 752         nilfs_mark_inode_dirty(inode);
 753         nilfs_set_file_dirty(inode, 0);
 754         nilfs_transaction_commit(sb);
 755         /* May construct a logical segment and may fail in sync mode.
 756            But truncate has no return value. */
 757 }
 758
 759 static void nilfs_clear_inode(struct inode *inode)
 760 {
 761         struct nilfs_inode_info *ii = NILFS_I(inode);
 762         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 763
 764         /*
 765          * Free resources allocated in nilfs_read_inode(), here.
 766          */
 767         BUG_ON(!list_empty(&ii->i_dirty));
 768         brelse(ii->i_bh);
 769         ii->i_bh = NULL;
 770
 771         if (mdi && mdi->mi_palloc_cache)
 772                 nilfs_palloc_destroy_cache(inode);
 773
 774         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 775                 nilfs_bmap_clear(ii->i_bmap);
 776
 777         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 778
 779         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 780                 nilfs_put_root(ii->i_root);
 781 }
 782
 783 void nilfs_evict_inode(struct inode *inode)
 784 {
 785         struct nilfs_transaction_info ti;
 786         struct super_block *sb = inode->i_sb;
 787         struct nilfs_inode_info *ii = NILFS_I(inode);
 788         int ret;
 789
 790         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 791                 if (inode->i_data.nrpages)
 792                         truncate_inode_pages(&inode->i_data, 0);
 793                 clear_inode(inode);
 794                 nilfs_clear_inode(inode);
 795                 return;
 796         }
 797         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 798
 799         if (inode->i_data.nrpages)
 800                 truncate_inode_pages(&inode->i_data, 0);
 801
 802         /* TODO: some of the following operations may fail.  */
 803         nilfs_truncate_bmap(ii, 0);
 804         nilfs_mark_inode_dirty(inode);
 805         clear_inode(inode);
 806
 807         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 808         if (!ret)
 809                 atomic64_dec(&ii->i_root->inodes_count);
 810
 811         nilfs_clear_inode(inode);
 812
 813         if (IS_SYNC(inode))
 814                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 815         nilfs_transaction_commit(sb);
 816         /* May construct a logical segment and may fail in sync mode.
 817            But delete_inode has no return value. */
 818 }
 819
 820 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 821 {
 822         struct nilfs_transaction_info ti;
 823         struct inode *inode = dentry->d_inode;
 824         struct super_block *sb = inode->i_sb;
 825         int err;
 826
 827         err = inode_change_ok(inode, iattr);
 828         if (err)
 829                 return err;
 830
 831         err = nilfs_transaction_begin(sb, &ti, 0);
 832         if (unlikely(err))
 833                 return err;
 834
 835         if ((iattr->ia_valid & ATTR_SIZE) &&
 836             iattr->ia_size != i_size_read(inode)) {
 837                 inode_dio_wait(inode);
 838                 truncate_setsize(inode, iattr->ia_size);
 839                 nilfs_truncate(inode);
 840         }
 841
 842         setattr_copy(inode, iattr);
 843         mark_inode_dirty(inode);
 844
 845         if (iattr->ia_valid & ATTR_MODE) {
 846                 err = nilfs_acl_chmod(inode);
 847                 if (unlikely(err))
 848                         goto out_err;
 849         }
 850
 851         return nilfs_transaction_commit(sb);
 852
 853 out_err:
 854         nilfs_transaction_abort(sb);
 855         return err;
 856 }
 857
 858 int nilfs_permission(struct inode *inode, int mask)
 859 {
 860         struct nilfs_root *root = NILFS_I(inode)->i_root;
 861         if ((mask & MAY_WRITE) && root &&
 862             root->cno != NILFS_CPTREE_CURRENT_CNO)
 863                 return -EROFS; /* snapshot is not writable */
 864
 865         return generic_permission(inode, mask);
 866 }
 867
 868 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 869 {
 870         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 871         struct nilfs_inode_info *ii = NILFS_I(inode);
 872         int err;
 873
 874         spin_lock(&nilfs->ns_inode_lock);
 875         if (ii->i_bh == NULL) {
 876                 spin_unlock(&nilfs->ns_inode_lock);
 877                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 878                                                   inode->i_ino, pbh);
 879                 if (unlikely(err))
 880                         return err;
 881                 spin_lock(&nilfs->ns_inode_lock);
 882                 if (ii->i_bh == NULL)
 883                         ii->i_bh = *pbh;
 884                 else {
 885                         brelse(*pbh);
 886                         *pbh = ii->i_bh;
 887                 }
 888         } else
 889                 *pbh = ii->i_bh;
 890
 891         get_bh(*pbh);
 892         spin_unlock(&nilfs->ns_inode_lock);
 893         return 0;
 894 }
 895
 896 int nilfs_inode_dirty(struct inode *inode)
 897 {
 898         struct nilfs_inode_info *ii = NILFS_I(inode);
 899         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 900         int ret = 0;
 901
 902         if (!list_empty(&ii->i_dirty)) {
 903                 spin_lock(&nilfs->ns_inode_lock);
 904                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 905                         test_bit(NILFS_I_BUSY, &ii->i_state);
 906                 spin_unlock(&nilfs->ns_inode_lock);
 907         }
 908         return ret;
 909 }
 910
 911 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 912 {
 913         struct nilfs_inode_info *ii = NILFS_I(inode);
 914         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 915
 916         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 917
 918         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 919                 return 0;
 920
 921         spin_lock(&nilfs->ns_inode_lock);
 922         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 923             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 924                 /* Because this routine may race with nilfs_dispose_list(),
 925                    we have to check NILFS_I_QUEUED here, too. */
 926                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 927                         /* This will happen when somebody is freeing
 928                            this inode. */
 929                         nilfs_warning(inode->i_sb, __func__,
 930                                       "cannot get inode (ino=%lu)\n",
 931                                       inode->i_ino);
 932                         spin_unlock(&nilfs->ns_inode_lock);
 933                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 934                                            freeing inode */
 935                 }
 936                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 937                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 938         }
 939         spin_unlock(&nilfs->ns_inode_lock);
 940         return 0;
 941 }
 942
 943 int nilfs_mark_inode_dirty(struct inode *inode)
 944 {
 945         struct buffer_head *ibh;
 946         int err;
 947
 948         err = nilfs_load_inode_block(inode, &ibh);
 949         if (unlikely(err)) {
 950                 nilfs_warning(inode->i_sb, __func__,
 951                               "failed to reget inode block.\n");
 952                 return err;
 953         }
 954         nilfs_update_inode(inode, ibh);
 955         mark_buffer_dirty(ibh);
 956         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 957         brelse(ibh);
 958         return 0;
 959 }
 960
 961 /**
 962  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 963  * @inode: inode of the file to be registered.
 964  *
 965  * nilfs_dirty_inode() loads a inode block containing the specified
 966  * @inode and copies data from a nilfs_inode to a corresponding inode
 967  * entry in the inode block. This operation is excluded from the segment
 968  * construction. This function can be called both as a single operation
 969  * and as a part of indivisible file operations.
 970  */
 971 void nilfs_dirty_inode(struct inode *inode, int flags)
 972 {
 973         struct nilfs_transaction_info ti;
 974         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 975
 976         if (is_bad_inode(inode)) {
 977                 nilfs_warning(inode->i_sb, __func__,
 978                               "tried to mark bad_inode dirty. ignored.\n");
 979                 dump_stack();
 980                 return;
 981         }
 982         if (mdi) {
 983                 nilfs_mdt_mark_dirty(inode);
 984                 return;
 985         }
 986         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 987         nilfs_mark_inode_dirty(inode);
 988         nilfs_transaction_commit(inode->i_sb); /* never fails */
 989 }
 990
 991 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 992                  __u64 start, __u64 len)
 993 {
 994         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 995         __u64 logical = 0, phys = 0, size = 0;
 996         __u32 flags = 0;
 997         loff_t isize;
 998         sector_t blkoff, end_blkoff;
 999         sector_t delalloc_blkoff;
1000         unsigned long delalloc_blklen;
1001         unsigned int blkbits = inode->i_blkbits;
1002         int ret, n;
1003
1004         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1005         if (ret)
1006                 return ret;
1007
1008         mutex_lock(&inode->i_mutex);
1009
1010         isize = i_size_read(inode);
1011
1012         blkoff = start >> blkbits;
1013         end_blkoff = (start + len - 1) >> blkbits;
1014
1015         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1016                                                         &delalloc_blkoff);
1017
1018         do {
1019                 __u64 blkphy;
1020                 unsigned int maxblocks;
1021
1022                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1023                         if (size) {
1024                                 /* End of the current extent */
1025                                 ret = fiemap_fill_next_extent(
1026                                         fieinfo, logical, phys, size, flags);
1027                                 if (ret)
1028                                         break;
1029                         }
1030                         if (blkoff > end_blkoff)
1031                                 break;
1032
1033                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1034                         logical = blkoff << blkbits;
1035                         phys = 0;
1036                         size = delalloc_blklen << blkbits;
1037
1038                         blkoff = delalloc_blkoff + delalloc_blklen;
1039                         delalloc_blklen = nilfs_find_uncommitted_extent(
1040                                 inode, blkoff, &delalloc_blkoff);
1041                         continue;
1042                 }
1043
1044                 /*
1045                  * Limit the number of blocks that we look up so as
1046                  * not to get into the next delayed allocation extent.
1047                  */
1048                 maxblocks = INT_MAX;
1049                 if (delalloc_blklen)
1050                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1051                                           maxblocks);
1052                 blkphy = 0;
1053
1054                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1055                 n = nilfs_bmap_lookup_contig(
1056                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1057                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1058
1059                 if (n < 0) {
1060                         int past_eof;
1061
1062                         if (unlikely(n != -ENOENT))
1063                                 break; /* error */
1064
1065                         /* HOLE */
1066                         blkoff++;
1067                         past_eof = ((blkoff << blkbits) >= isize);
1068
1069                         if (size) {
1070                                 /* End of the current extent */
1071
1072                                 if (past_eof)
1073                                         flags |= FIEMAP_EXTENT_LAST;
1074
1075                                 ret = fiemap_fill_next_extent(
1076                                         fieinfo, logical, phys, size, flags);
1077                                 if (ret)
1078                                         break;
1079                                 size = 0;
1080                         }
1081                         if (blkoff > end_blkoff || past_eof)
1082                                 break;
1083                 } else {
1084                         if (size) {
1085                                 if (phys && blkphy << blkbits == phys + size) {
1086                                         /* The current extent goes on */
1087                                         size += n << blkbits;
1088                                 } else {
1089                                         /* Terminate the current extent */
1090                                         ret = fiemap_fill_next_extent(
1091                                                 fieinfo, logical, phys, size,
1092                                                 flags);
1093                                         if (ret || blkoff > end_blkoff)
1094                                                 break;
1095
1096                                         /* Start another extent */
1097                                         flags = FIEMAP_EXTENT_MERGED;
1098                                         logical = blkoff << blkbits;
1099                                         phys = blkphy << blkbits;
1100                                         size = n << blkbits;
1101                                 }
1102                         } else {
1103                                 /* Start a new extent */
1104                                 flags = FIEMAP_EXTENT_MERGED;
1105                                 logical = blkoff << blkbits;
1106                                 phys = blkphy << blkbits;
1107                                 size = n << blkbits;
1108                         }
1109                         blkoff += n;
1110                 }
1111                 cond_resched();
1112         } while (true);
1113
1114         /* If ret is 1 then we just hit the end of the extent array */
1115         if (ret == 1)
1116                 ret = 0;
1117
1118         mutex_unlock(&inode->i_mutex);
1119         return ret;
1120 }