fs/nilfs2/inode.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * NILFS inode operations.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  *
   9  */
  10
  11 #include <linux/buffer_head.h>
  12 #include <linux/gfp.h>
  13 #include <linux/mpage.h>
  14 #include <linux/pagemap.h>
  15 #include <linux/writeback.h>
  16 #include <linux/uio.h>
  17 #include <linux/fiemap.h>
  18 #include "nilfs.h"
  19 #include "btnode.h"
  20 #include "segment.h"
  21 #include "page.h"
  22 #include "mdt.h"
  23 #include "cpfile.h"
  24 #include "ifile.h"
  25
  26 /**
  27  * struct nilfs_iget_args - arguments used during comparison between inodes
  28  * @ino: inode number
  29  * @cno: checkpoint number
  30  * @root: pointer on NILFS root object (mounted checkpoint)
  31  * @for_gc: inode for GC flag
  32  * @for_btnc: inode for B-tree node cache flag
  33  * @for_shadow: inode for shadowed page cache flag
  34  */
  35 struct nilfs_iget_args {
  36         u64 ino;
  37         __u64 cno;
  38         struct nilfs_root *root;
  39         bool for_gc;
  40         bool for_btnc;
  41         bool for_shadow;
  42 };
  43
  44 static int nilfs_iget_test(struct inode *inode, void *opaque);
  45
  46 void nilfs_inode_add_blocks(struct inode *inode, int n)
  47 {
  48         struct nilfs_root *root = NILFS_I(inode)->i_root;
  49
  50         inode_add_bytes(inode, i_blocksize(inode) * n);
  51         if (root)
  52                 atomic64_add(n, &root->blocks_count);
  53 }
  54
  55 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  56 {
  57         struct nilfs_root *root = NILFS_I(inode)->i_root;
  58
  59         inode_sub_bytes(inode, i_blocksize(inode) * n);
  60         if (root)
  61                 atomic64_sub(n, &root->blocks_count);
  62 }
  63
  64 /**
  65  * nilfs_get_block() - get a file block on the filesystem (callback function)
  66  * @inode: inode struct of the target file
  67  * @blkoff: file block number
  68  * @bh_result: buffer head to be mapped on
  69  * @create: indicate whether allocating the block or not when it has not
  70  *      been allocated yet.
  71  *
  72  * This function does not issue actual read request of the specified data
  73  * block. It is done by VFS.
  74  */
  75 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  76                     struct buffer_head *bh_result, int create)
  77 {
  78         struct nilfs_inode_info *ii = NILFS_I(inode);
  79         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  80         __u64 blknum = 0;
  81         int err = 0, ret;
  82         unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
  83
  84         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  85         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  86         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  87         if (ret >= 0) { /* found */
  88                 map_bh(bh_result, inode->i_sb, blknum);
  89                 if (ret > 0)
  90                         bh_result->b_size = (ret << inode->i_blkbits);
  91                 goto out;
  92         }
  93         /* data block was not found */
  94         if (ret == -ENOENT && create) {
  95                 struct nilfs_transaction_info ti;
  96
  97                 bh_result->b_blocknr = 0;
  98                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  99                 if (unlikely(err))
 100                         goto out;
 101                 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 102                                         (unsigned long)bh_result);
 103                 if (unlikely(err != 0)) {
 104                         if (err == -EEXIST) {
 105                                 /*
 106                                  * The get_block() function could be called
 107                                  * from multiple callers for an inode.
 108                                  * However, the page having this block must
 109                                  * be locked in this case.
 110                                  */
 111                                 nilfs_warn(inode->i_sb,
 112                                            "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
 113                                            __func__, inode->i_ino,
 114                                            (unsigned long long)blkoff);
 115                                 err = 0;
 116                         }
 117                         nilfs_transaction_abort(inode->i_sb);
 118                         goto out;
 119                 }
 120                 nilfs_mark_inode_dirty_sync(inode);
 121                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 122                 /* Error handling should be detailed */
 123                 set_buffer_new(bh_result);
 124                 set_buffer_delay(bh_result);
 125                 map_bh(bh_result, inode->i_sb, 0);
 126                 /* Disk block number must be changed to proper value */
 127
 128         } else if (ret == -ENOENT) {
 129                 /*
 130                  * not found is not error (e.g. hole); must return without
 131                  * the mapped state flag.
 132                  */
 133                 ;
 134         } else {
 135                 err = ret;
 136         }
 137
 138  out:
 139         return err;
 140 }
 141
 142 /**
 143  * nilfs_read_folio() - implement read_folio() method of nilfs_aops {}
 144  * address_space_operations.
 145  * @file: file struct of the file to be read
 146  * @folio: the folio to be read
 147  */
 148 static int nilfs_read_folio(struct file *file, struct folio *folio)
 149 {
 150         return mpage_read_folio(folio, nilfs_get_block);
 151 }
 152
 153 static void nilfs_readahead(struct readahead_control *rac)
 154 {
 155         mpage_readahead(rac, nilfs_get_block);
 156 }
 157
 158 static int nilfs_writepages(struct address_space *mapping,
 159                             struct writeback_control *wbc)
 160 {
 161         struct inode *inode = mapping->host;
 162         int err = 0;
 163
 164         if (sb_rdonly(inode->i_sb)) {
 165                 nilfs_clear_dirty_pages(mapping, false);
 166                 return -EROFS;
 167         }
 168
 169         if (wbc->sync_mode == WB_SYNC_ALL)
 170                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 171                                                     wbc->range_start,
 172                                                     wbc->range_end);
 173         return err;
 174 }
 175
 176 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 177 {
 178         struct inode *inode = page->mapping->host;
 179         int err;
 180
 181         if (sb_rdonly(inode->i_sb)) {
 182                 /*
 183                  * It means that filesystem was remounted in read-only
 184                  * mode because of error or metadata corruption. But we
 185                  * have dirty pages that try to be flushed in background.
 186                  * So, here we simply discard this dirty page.
 187                  */
 188                 nilfs_clear_dirty_page(page, false);
 189                 unlock_page(page);
 190                 return -EROFS;
 191         }
 192
 193         redirty_page_for_writepage(wbc, page);
 194         unlock_page(page);
 195
 196         if (wbc->sync_mode == WB_SYNC_ALL) {
 197                 err = nilfs_construct_segment(inode->i_sb);
 198                 if (unlikely(err))
 199                         return err;
 200         } else if (wbc->for_reclaim)
 201                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 202
 203         return 0;
 204 }
 205
 206 static bool nilfs_dirty_folio(struct address_space *mapping,
 207                 struct folio *folio)
 208 {
 209         struct inode *inode = mapping->host;
 210         struct buffer_head *head;
 211         unsigned int nr_dirty = 0;
 212         bool ret = filemap_dirty_folio(mapping, folio);
 213
 214         /*
 215          * The page may not be locked, eg if called from try_to_unmap_one()
 216          */
 217         spin_lock(&mapping->private_lock);
 218         head = folio_buffers(folio);
 219         if (head) {
 220                 struct buffer_head *bh = head;
 221
 222                 do {
 223                         /* Do not mark hole blocks dirty */
 224                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 225                                 continue;
 226
 227                         set_buffer_dirty(bh);
 228                         nr_dirty++;
 229                 } while (bh = bh->b_this_page, bh != head);
 230         } else if (ret) {
 231                 nr_dirty = 1 << (folio_shift(folio) - inode->i_blkbits);
 232         }
 233         spin_unlock(&mapping->private_lock);
 234
 235         if (nr_dirty)
 236                 nilfs_set_file_dirty(inode, nr_dirty);
 237         return ret;
 238 }
 239
 240 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 241 {
 242         struct inode *inode = mapping->host;
 243
 244         if (to > inode->i_size) {
 245                 truncate_pagecache(inode, inode->i_size);
 246                 nilfs_truncate(inode);
 247         }
 248 }
 249
 250 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 251                              loff_t pos, unsigned len,
 252                              struct page **pagep, void **fsdata)
 253
 254 {
 255         struct inode *inode = mapping->host;
 256         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 257
 258         if (unlikely(err))
 259                 return err;
 260
 261         err = block_write_begin(mapping, pos, len, pagep, nilfs_get_block);
 262         if (unlikely(err)) {
 263                 nilfs_write_failed(mapping, pos + len);
 264                 nilfs_transaction_abort(inode->i_sb);
 265         }
 266         return err;
 267 }
 268
 269 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 270                            loff_t pos, unsigned len, unsigned copied,
 271                            struct page *page, void *fsdata)
 272 {
 273         struct inode *inode = mapping->host;
 274         unsigned int start = pos & (PAGE_SIZE - 1);
 275         unsigned int nr_dirty;
 276         int err;
 277
 278         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 279                                                   start + copied);
 280         copied = generic_write_end(file, mapping, pos, len, copied, page,
 281                                    fsdata);
 282         nilfs_set_file_dirty(inode, nr_dirty);
 283         err = nilfs_transaction_commit(inode->i_sb);
 284         return err ? : copied;
 285 }
 286
 287 static ssize_t
 288 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 289 {
 290         struct inode *inode = file_inode(iocb->ki_filp);
 291
 292         if (iov_iter_rw(iter) == WRITE)
 293                 return 0;
 294
 295         /* Needs synchronization with the cleaner */
 296         return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 297 }
 298
 299 const struct address_space_operations nilfs_aops = {
 300         .writepage              = nilfs_writepage,
 301         .read_folio             = nilfs_read_folio,
 302         .writepages             = nilfs_writepages,
 303         .dirty_folio            = nilfs_dirty_folio,
 304         .readahead              = nilfs_readahead,
 305         .write_begin            = nilfs_write_begin,
 306         .write_end              = nilfs_write_end,
 307         .invalidate_folio       = block_invalidate_folio,
 308         .direct_IO              = nilfs_direct_IO,
 309         .is_partially_uptodate  = block_is_partially_uptodate,
 310 };
 311
 312 static int nilfs_insert_inode_locked(struct inode *inode,
 313                                      struct nilfs_root *root,
 314                                      unsigned long ino)
 315 {
 316         struct nilfs_iget_args args = {
 317                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 318                 .for_btnc = false, .for_shadow = false
 319         };
 320
 321         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 322 }
 323
 324 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 325 {
 326         struct super_block *sb = dir->i_sb;
 327         struct the_nilfs *nilfs = sb->s_fs_info;
 328         struct inode *inode;
 329         struct nilfs_inode_info *ii;
 330         struct nilfs_root *root;
 331         int err = -ENOMEM;
 332         ino_t ino;
 333
 334         inode = new_inode(sb);
 335         if (unlikely(!inode))
 336                 goto failed;
 337
 338         mapping_set_gfp_mask(inode->i_mapping,
 339                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 340
 341         root = NILFS_I(dir)->i_root;
 342         ii = NILFS_I(inode);
 343         ii->i_state = BIT(NILFS_I_NEW);
 344         ii->i_root = root;
 345
 346         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 347         if (unlikely(err))
 348                 goto failed_ifile_create_inode;
 349         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 350
 351         atomic64_inc(&root->inodes_count);
 352         inode_init_owner(&init_user_ns, inode, dir, mode);
 353         inode->i_ino = ino;
 354         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 355
 356         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 357                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 358                 if (err < 0)
 359                         goto failed_after_creation;
 360
 361                 set_bit(NILFS_I_BMAP, &ii->i_state);
 362                 /* No lock is needed; iget() ensures it. */
 363         }
 364
 365         ii->i_flags = nilfs_mask_flags(
 366                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 367
 368         /* ii->i_file_acl = 0; */
 369         /* ii->i_dir_acl = 0; */
 370         ii->i_dir_start_lookup = 0;
 371         nilfs_set_inode_flags(inode);
 372         spin_lock(&nilfs->ns_next_gen_lock);
 373         inode->i_generation = nilfs->ns_next_generation++;
 374         spin_unlock(&nilfs->ns_next_gen_lock);
 375         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 376                 err = -EIO;
 377                 goto failed_after_creation;
 378         }
 379
 380         err = nilfs_init_acl(inode, dir);
 381         if (unlikely(err))
 382                 /*
 383                  * Never occur.  When supporting nilfs_init_acl(),
 384                  * proper cancellation of above jobs should be considered.
 385                  */
 386                 goto failed_after_creation;
 387
 388         return inode;
 389
 390  failed_after_creation:
 391         clear_nlink(inode);
 392         if (inode->i_state & I_NEW)
 393                 unlock_new_inode(inode);
 394         iput(inode);  /*
 395                        * raw_inode will be deleted through
 396                        * nilfs_evict_inode().
 397                        */
 398         goto failed;
 399
 400  failed_ifile_create_inode:
 401         make_bad_inode(inode);
 402         iput(inode);
 403  failed:
 404         return ERR_PTR(err);
 405 }
 406
 407 void nilfs_set_inode_flags(struct inode *inode)
 408 {
 409         unsigned int flags = NILFS_I(inode)->i_flags;
 410         unsigned int new_fl = 0;
 411
 412         if (flags & FS_SYNC_FL)
 413                 new_fl |= S_SYNC;
 414         if (flags & FS_APPEND_FL)
 415                 new_fl |= S_APPEND;
 416         if (flags & FS_IMMUTABLE_FL)
 417                 new_fl |= S_IMMUTABLE;
 418         if (flags & FS_NOATIME_FL)
 419                 new_fl |= S_NOATIME;
 420         if (flags & FS_DIRSYNC_FL)
 421                 new_fl |= S_DIRSYNC;
 422         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 423                         S_NOATIME | S_DIRSYNC);
 424 }
 425
 426 int nilfs_read_inode_common(struct inode *inode,
 427                             struct nilfs_inode *raw_inode)
 428 {
 429         struct nilfs_inode_info *ii = NILFS_I(inode);
 430         int err;
 431
 432         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 433         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 434         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 435         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 436         inode->i_size = le64_to_cpu(raw_inode->i_size);
 437         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 438         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 439         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 440         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 441         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 442         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 443         if (inode->i_nlink == 0)
 444                 return -ESTALE; /* this inode is deleted */
 445
 446         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 447         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 448 #if 0
 449         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 450         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 451                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 452 #endif
 453         ii->i_dir_start_lookup = 0;
 454         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 455
 456         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 457             S_ISLNK(inode->i_mode)) {
 458                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 459                 if (err < 0)
 460                         return err;
 461                 set_bit(NILFS_I_BMAP, &ii->i_state);
 462                 /* No lock is needed; iget() ensures it. */
 463         }
 464         return 0;
 465 }
 466
 467 static int __nilfs_read_inode(struct super_block *sb,
 468                               struct nilfs_root *root, unsigned long ino,
 469                               struct inode *inode)
 470 {
 471         struct the_nilfs *nilfs = sb->s_fs_info;
 472         struct buffer_head *bh;
 473         struct nilfs_inode *raw_inode;
 474         int err;
 475
 476         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 477         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 478         if (unlikely(err))
 479                 goto bad_inode;
 480
 481         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 482
 483         err = nilfs_read_inode_common(inode, raw_inode);
 484         if (err)
 485                 goto failed_unmap;
 486
 487         if (S_ISREG(inode->i_mode)) {
 488                 inode->i_op = &nilfs_file_inode_operations;
 489                 inode->i_fop = &nilfs_file_operations;
 490                 inode->i_mapping->a_ops = &nilfs_aops;
 491         } else if (S_ISDIR(inode->i_mode)) {
 492                 inode->i_op = &nilfs_dir_inode_operations;
 493                 inode->i_fop = &nilfs_dir_operations;
 494                 inode->i_mapping->a_ops = &nilfs_aops;
 495         } else if (S_ISLNK(inode->i_mode)) {
 496                 inode->i_op = &nilfs_symlink_inode_operations;
 497                 inode_nohighmem(inode);
 498                 inode->i_mapping->a_ops = &nilfs_aops;
 499         } else {
 500                 inode->i_op = &nilfs_special_inode_operations;
 501                 init_special_inode(
 502                         inode, inode->i_mode,
 503                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 504         }
 505         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 506         brelse(bh);
 507         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 508         nilfs_set_inode_flags(inode);
 509         mapping_set_gfp_mask(inode->i_mapping,
 510                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 511         return 0;
 512
 513  failed_unmap:
 514         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 515         brelse(bh);
 516
 517  bad_inode:
 518         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 519         return err;
 520 }
 521
 522 static int nilfs_iget_test(struct inode *inode, void *opaque)
 523 {
 524         struct nilfs_iget_args *args = opaque;
 525         struct nilfs_inode_info *ii;
 526
 527         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 528                 return 0;
 529
 530         ii = NILFS_I(inode);
 531         if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
 532                 if (!args->for_btnc)
 533                         return 0;
 534         } else if (args->for_btnc) {
 535                 return 0;
 536         }
 537         if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
 538                 if (!args->for_shadow)
 539                         return 0;
 540         } else if (args->for_shadow) {
 541                 return 0;
 542         }
 543
 544         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 545                 return !args->for_gc;
 546
 547         return args->for_gc && args->cno == ii->i_cno;
 548 }
 549
 550 static int nilfs_iget_set(struct inode *inode, void *opaque)
 551 {
 552         struct nilfs_iget_args *args = opaque;
 553
 554         inode->i_ino = args->ino;
 555         NILFS_I(inode)->i_cno = args->cno;
 556         NILFS_I(inode)->i_root = args->root;
 557         if (args->root && args->ino == NILFS_ROOT_INO)
 558                 nilfs_get_root(args->root);
 559
 560         if (args->for_gc)
 561                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 562         if (args->for_btnc)
 563                 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
 564         if (args->for_shadow)
 565                 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
 566         return 0;
 567 }
 568
 569 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 570                             unsigned long ino)
 571 {
 572         struct nilfs_iget_args args = {
 573                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 574                 .for_btnc = false, .for_shadow = false
 575         };
 576
 577         return ilookup5(sb, ino, nilfs_iget_test, &args);
 578 }
 579
 580 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 581                                 unsigned long ino)
 582 {
 583         struct nilfs_iget_args args = {
 584                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 585                 .for_btnc = false, .for_shadow = false
 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 = true,
 617                 .for_btnc = false, .for_shadow = false
 618         };
 619         struct inode *inode;
 620         int err;
 621
 622         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 623         if (unlikely(!inode))
 624                 return ERR_PTR(-ENOMEM);
 625         if (!(inode->i_state & I_NEW))
 626                 return inode;
 627
 628         err = nilfs_init_gcinode(inode);
 629         if (unlikely(err)) {
 630                 iget_failed(inode);
 631                 return ERR_PTR(err);
 632         }
 633         unlock_new_inode(inode);
 634         return inode;
 635 }
 636
 637 /**
 638  * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 639  * @inode: inode object
 640  *
 641  * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 642  * or does nothing if the inode already has it.  This function allocates
 643  * an additional inode to maintain page cache of B-tree nodes one-on-one.
 644  *
 645  * Return Value: On success, 0 is returned. On errors, one of the following
 646  * negative error code is returned.
 647  *
 648  * %-ENOMEM - Insufficient memory available.
 649  */
 650 int nilfs_attach_btree_node_cache(struct inode *inode)
 651 {
 652         struct nilfs_inode_info *ii = NILFS_I(inode);
 653         struct inode *btnc_inode;
 654         struct nilfs_iget_args args;
 655
 656         if (ii->i_assoc_inode)
 657                 return 0;
 658
 659         args.ino = inode->i_ino;
 660         args.root = ii->i_root;
 661         args.cno = ii->i_cno;
 662         args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
 663         args.for_btnc = true;
 664         args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
 665
 666         btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 667                                   nilfs_iget_set, &args);
 668         if (unlikely(!btnc_inode))
 669                 return -ENOMEM;
 670         if (btnc_inode->i_state & I_NEW) {
 671                 nilfs_init_btnc_inode(btnc_inode);
 672                 unlock_new_inode(btnc_inode);
 673         }
 674         NILFS_I(btnc_inode)->i_assoc_inode = inode;
 675         NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 676         ii->i_assoc_inode = btnc_inode;
 677
 678         return 0;
 679 }
 680
 681 /**
 682  * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 683  * @inode: inode object
 684  *
 685  * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 686  * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 687  */
 688 void nilfs_detach_btree_node_cache(struct inode *inode)
 689 {
 690         struct nilfs_inode_info *ii = NILFS_I(inode);
 691         struct inode *btnc_inode = ii->i_assoc_inode;
 692
 693         if (btnc_inode) {
 694                 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 695                 ii->i_assoc_inode = NULL;
 696                 iput(btnc_inode);
 697         }
 698 }
 699
 700 /**
 701  * nilfs_iget_for_shadow - obtain inode for shadow mapping
 702  * @inode: inode object that uses shadow mapping
 703  *
 704  * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 705  * caches for shadow mapping.  The page cache for data pages is set up
 706  * in one inode and the one for b-tree node pages is set up in the
 707  * other inode, which is attached to the former inode.
 708  *
 709  * Return Value: On success, a pointer to the inode for data pages is
 710  * returned. On errors, one of the following negative error code is returned
 711  * in a pointer type.
 712  *
 713  * %-ENOMEM - Insufficient memory available.
 714  */
 715 struct inode *nilfs_iget_for_shadow(struct inode *inode)
 716 {
 717         struct nilfs_iget_args args = {
 718                 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
 719                 .for_btnc = false, .for_shadow = true
 720         };
 721         struct inode *s_inode;
 722         int err;
 723
 724         s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 725                                nilfs_iget_set, &args);
 726         if (unlikely(!s_inode))
 727                 return ERR_PTR(-ENOMEM);
 728         if (!(s_inode->i_state & I_NEW))
 729                 return inode;
 730
 731         NILFS_I(s_inode)->i_flags = 0;
 732         memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 733         mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 734
 735         err = nilfs_attach_btree_node_cache(s_inode);
 736         if (unlikely(err)) {
 737                 iget_failed(s_inode);
 738                 return ERR_PTR(err);
 739         }
 740         unlock_new_inode(s_inode);
 741         return s_inode;
 742 }
 743
 744 void nilfs_write_inode_common(struct inode *inode,
 745                               struct nilfs_inode *raw_inode, int has_bmap)
 746 {
 747         struct nilfs_inode_info *ii = NILFS_I(inode);
 748
 749         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 750         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 751         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 752         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 753         raw_inode->i_size = cpu_to_le64(inode->i_size);
 754         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 755         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 756         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 757         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 758         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 759
 760         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 761         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 762
 763         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 764                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 765
 766                 /* zero-fill unused portion in the case of super root block */
 767                 raw_inode->i_xattr = 0;
 768                 raw_inode->i_pad = 0;
 769                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 770                        nilfs->ns_inode_size - sizeof(*raw_inode));
 771         }
 772
 773         if (has_bmap)
 774                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 775         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 776                 raw_inode->i_device_code =
 777                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 778         /*
 779          * When extending inode, nilfs->ns_inode_size should be checked
 780          * for substitutions of appended fields.
 781          */
 782 }
 783
 784 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 785 {
 786         ino_t ino = inode->i_ino;
 787         struct nilfs_inode_info *ii = NILFS_I(inode);
 788         struct inode *ifile = ii->i_root->ifile;
 789         struct nilfs_inode *raw_inode;
 790
 791         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 792
 793         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 794                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 795         if (flags & I_DIRTY_DATASYNC)
 796                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 797
 798         nilfs_write_inode_common(inode, raw_inode, 0);
 799                 /*
 800                  * XXX: call with has_bmap = 0 is a workaround to avoid
 801                  * deadlock of bmap.  This delays update of i_bmap to just
 802                  * before writing.
 803                  */
 804
 805         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 806 }
 807
 808 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 809
 810 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 811                                 unsigned long from)
 812 {
 813         __u64 b;
 814         int ret;
 815
 816         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 817                 return;
 818 repeat:
 819         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 820         if (ret == -ENOENT)
 821                 return;
 822         else if (ret < 0)
 823                 goto failed;
 824
 825         if (b < from)
 826                 return;
 827
 828         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 829         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 830         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 831         if (!ret || (ret == -ENOMEM &&
 832                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 833                 goto repeat;
 834
 835 failed:
 836         nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
 837                    ret, ii->vfs_inode.i_ino);
 838 }
 839
 840 void nilfs_truncate(struct inode *inode)
 841 {
 842         unsigned long blkoff;
 843         unsigned int blocksize;
 844         struct nilfs_transaction_info ti;
 845         struct super_block *sb = inode->i_sb;
 846         struct nilfs_inode_info *ii = NILFS_I(inode);
 847
 848         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 849                 return;
 850         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 851                 return;
 852
 853         blocksize = sb->s_blocksize;
 854         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 855         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 856
 857         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 858
 859         nilfs_truncate_bmap(ii, blkoff);
 860
 861         inode->i_mtime = inode->i_ctime = current_time(inode);
 862         if (IS_SYNC(inode))
 863                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 864
 865         nilfs_mark_inode_dirty(inode);
 866         nilfs_set_file_dirty(inode, 0);
 867         nilfs_transaction_commit(sb);
 868         /*
 869          * May construct a logical segment and may fail in sync mode.
 870          * But truncate has no return value.
 871          */
 872 }
 873
 874 static void nilfs_clear_inode(struct inode *inode)
 875 {
 876         struct nilfs_inode_info *ii = NILFS_I(inode);
 877
 878         /*
 879          * Free resources allocated in nilfs_read_inode(), here.
 880          */
 881         BUG_ON(!list_empty(&ii->i_dirty));
 882         brelse(ii->i_bh);
 883         ii->i_bh = NULL;
 884
 885         if (nilfs_is_metadata_file_inode(inode))
 886                 nilfs_mdt_clear(inode);
 887
 888         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 889                 nilfs_bmap_clear(ii->i_bmap);
 890
 891         if (!test_bit(NILFS_I_BTNC, &ii->i_state))
 892                 nilfs_detach_btree_node_cache(inode);
 893
 894         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 895                 nilfs_put_root(ii->i_root);
 896 }
 897
 898 void nilfs_evict_inode(struct inode *inode)
 899 {
 900         struct nilfs_transaction_info ti;
 901         struct super_block *sb = inode->i_sb;
 902         struct nilfs_inode_info *ii = NILFS_I(inode);
 903         int ret;
 904
 905         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 906                 truncate_inode_pages_final(&inode->i_data);
 907                 clear_inode(inode);
 908                 nilfs_clear_inode(inode);
 909                 return;
 910         }
 911         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 912
 913         truncate_inode_pages_final(&inode->i_data);
 914
 915         /* TODO: some of the following operations may fail.  */
 916         nilfs_truncate_bmap(ii, 0);
 917         nilfs_mark_inode_dirty(inode);
 918         clear_inode(inode);
 919
 920         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 921         if (!ret)
 922                 atomic64_dec(&ii->i_root->inodes_count);
 923
 924         nilfs_clear_inode(inode);
 925
 926         if (IS_SYNC(inode))
 927                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 928         nilfs_transaction_commit(sb);
 929         /*
 930          * May construct a logical segment and may fail in sync mode.
 931          * But delete_inode has no return value.
 932          */
 933 }
 934
 935 int nilfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 936                   struct iattr *iattr)
 937 {
 938         struct nilfs_transaction_info ti;
 939         struct inode *inode = d_inode(dentry);
 940         struct super_block *sb = inode->i_sb;
 941         int err;
 942
 943         err = setattr_prepare(&init_user_ns, dentry, iattr);
 944         if (err)
 945                 return err;
 946
 947         err = nilfs_transaction_begin(sb, &ti, 0);
 948         if (unlikely(err))
 949                 return err;
 950
 951         if ((iattr->ia_valid & ATTR_SIZE) &&
 952             iattr->ia_size != i_size_read(inode)) {
 953                 inode_dio_wait(inode);
 954                 truncate_setsize(inode, iattr->ia_size);
 955                 nilfs_truncate(inode);
 956         }
 957
 958         setattr_copy(&init_user_ns, inode, iattr);
 959         mark_inode_dirty(inode);
 960
 961         if (iattr->ia_valid & ATTR_MODE) {
 962                 err = nilfs_acl_chmod(inode);
 963                 if (unlikely(err))
 964                         goto out_err;
 965         }
 966
 967         return nilfs_transaction_commit(sb);
 968
 969 out_err:
 970         nilfs_transaction_abort(sb);
 971         return err;
 972 }
 973
 974 int nilfs_permission(struct user_namespace *mnt_userns, struct inode *inode,
 975                      int mask)
 976 {
 977         struct nilfs_root *root = NILFS_I(inode)->i_root;
 978
 979         if ((mask & MAY_WRITE) && root &&
 980             root->cno != NILFS_CPTREE_CURRENT_CNO)
 981                 return -EROFS; /* snapshot is not writable */
 982
 983         return generic_permission(&init_user_ns, inode, mask);
 984 }
 985
 986 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 987 {
 988         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 989         struct nilfs_inode_info *ii = NILFS_I(inode);
 990         int err;
 991
 992         spin_lock(&nilfs->ns_inode_lock);
 993         if (ii->i_bh == NULL) {
 994                 spin_unlock(&nilfs->ns_inode_lock);
 995                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 996                                                   inode->i_ino, pbh);
 997                 if (unlikely(err))
 998                         return err;
 999                 spin_lock(&nilfs->ns_inode_lock);
1000                 if (ii->i_bh == NULL)
1001                         ii->i_bh = *pbh;
1002                 else {
1003                         brelse(*pbh);
1004                         *pbh = ii->i_bh;
1005                 }
1006         } else
1007                 *pbh = ii->i_bh;
1008
1009         get_bh(*pbh);
1010         spin_unlock(&nilfs->ns_inode_lock);
1011         return 0;
1012 }
1013
1014 int nilfs_inode_dirty(struct inode *inode)
1015 {
1016         struct nilfs_inode_info *ii = NILFS_I(inode);
1017         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1018         int ret = 0;
1019
1020         if (!list_empty(&ii->i_dirty)) {
1021                 spin_lock(&nilfs->ns_inode_lock);
1022                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1023                         test_bit(NILFS_I_BUSY, &ii->i_state);
1024                 spin_unlock(&nilfs->ns_inode_lock);
1025         }
1026         return ret;
1027 }
1028
1029 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1030 {
1031         struct nilfs_inode_info *ii = NILFS_I(inode);
1032         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1033
1034         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1035
1036         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1037                 return 0;
1038
1039         spin_lock(&nilfs->ns_inode_lock);
1040         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1041             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1042                 /*
1043                  * Because this routine may race with nilfs_dispose_list(),
1044                  * we have to check NILFS_I_QUEUED here, too.
1045                  */
1046                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1047                         /*
1048                          * This will happen when somebody is freeing
1049                          * this inode.
1050                          */
1051                         nilfs_warn(inode->i_sb,
1052                                    "cannot set file dirty (ino=%lu): the file is being freed",
1053                                    inode->i_ino);
1054                         spin_unlock(&nilfs->ns_inode_lock);
1055                         return -EINVAL; /*
1056                                          * NILFS_I_DIRTY may remain for
1057                                          * freeing inode.
1058                                          */
1059                 }
1060                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1061                 set_bit(NILFS_I_QUEUED, &ii->i_state);
1062         }
1063         spin_unlock(&nilfs->ns_inode_lock);
1064         return 0;
1065 }
1066
1067 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1068 {
1069         struct buffer_head *ibh;
1070         int err;
1071
1072         err = nilfs_load_inode_block(inode, &ibh);
1073         if (unlikely(err)) {
1074                 nilfs_warn(inode->i_sb,
1075                            "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1076                            inode->i_ino, err);
1077                 return err;
1078         }
1079         nilfs_update_inode(inode, ibh, flags);
1080         mark_buffer_dirty(ibh);
1081         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1082         brelse(ibh);
1083         return 0;
1084 }
1085
1086 /**
1087  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1088  * @inode: inode of the file to be registered.
1089  * @flags: flags to determine the dirty state of the inode
1090  *
1091  * nilfs_dirty_inode() loads a inode block containing the specified
1092  * @inode and copies data from a nilfs_inode to a corresponding inode
1093  * entry in the inode block. This operation is excluded from the segment
1094  * construction. This function can be called both as a single operation
1095  * and as a part of indivisible file operations.
1096  */
1097 void nilfs_dirty_inode(struct inode *inode, int flags)
1098 {
1099         struct nilfs_transaction_info ti;
1100         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1101
1102         if (is_bad_inode(inode)) {
1103                 nilfs_warn(inode->i_sb,
1104                            "tried to mark bad_inode dirty. ignored.");
1105                 dump_stack();
1106                 return;
1107         }
1108         if (mdi) {
1109                 nilfs_mdt_mark_dirty(inode);
1110                 return;
1111         }
1112         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1113         __nilfs_mark_inode_dirty(inode, flags);
1114         nilfs_transaction_commit(inode->i_sb); /* never fails */
1115 }
1116
1117 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1118                  __u64 start, __u64 len)
1119 {
1120         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1121         __u64 logical = 0, phys = 0, size = 0;
1122         __u32 flags = 0;
1123         loff_t isize;
1124         sector_t blkoff, end_blkoff;
1125         sector_t delalloc_blkoff;
1126         unsigned long delalloc_blklen;
1127         unsigned int blkbits = inode->i_blkbits;
1128         int ret, n;
1129
1130         ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1131         if (ret)
1132                 return ret;
1133
1134         inode_lock(inode);
1135
1136         isize = i_size_read(inode);
1137
1138         blkoff = start >> blkbits;
1139         end_blkoff = (start + len - 1) >> blkbits;
1140
1141         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1142                                                         &delalloc_blkoff);
1143
1144         do {
1145                 __u64 blkphy;
1146                 unsigned int maxblocks;
1147
1148                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1149                         if (size) {
1150                                 /* End of the current extent */
1151                                 ret = fiemap_fill_next_extent(
1152                                         fieinfo, logical, phys, size, flags);
1153                                 if (ret)
1154                                         break;
1155                         }
1156                         if (blkoff > end_blkoff)
1157                                 break;
1158
1159                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1160                         logical = blkoff << blkbits;
1161                         phys = 0;
1162                         size = delalloc_blklen << blkbits;
1163
1164                         blkoff = delalloc_blkoff + delalloc_blklen;
1165                         delalloc_blklen = nilfs_find_uncommitted_extent(
1166                                 inode, blkoff, &delalloc_blkoff);
1167                         continue;
1168                 }
1169
1170                 /*
1171                  * Limit the number of blocks that we look up so as
1172                  * not to get into the next delayed allocation extent.
1173                  */
1174                 maxblocks = INT_MAX;
1175                 if (delalloc_blklen)
1176                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1177                                           maxblocks);
1178                 blkphy = 0;
1179
1180                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1181                 n = nilfs_bmap_lookup_contig(
1182                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1183                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1184
1185                 if (n < 0) {
1186                         int past_eof;
1187
1188                         if (unlikely(n != -ENOENT))
1189                                 break; /* error */
1190
1191                         /* HOLE */
1192                         blkoff++;
1193                         past_eof = ((blkoff << blkbits) >= isize);
1194
1195                         if (size) {
1196                                 /* End of the current extent */
1197
1198                                 if (past_eof)
1199                                         flags |= FIEMAP_EXTENT_LAST;
1200
1201                                 ret = fiemap_fill_next_extent(
1202                                         fieinfo, logical, phys, size, flags);
1203                                 if (ret)
1204                                         break;
1205                                 size = 0;
1206                         }
1207                         if (blkoff > end_blkoff || past_eof)
1208                                 break;
1209                 } else {
1210                         if (size) {
1211                                 if (phys && blkphy << blkbits == phys + size) {
1212                                         /* The current extent goes on */
1213                                         size += n << blkbits;
1214                                 } else {
1215                                         /* Terminate the current extent */
1216                                         ret = fiemap_fill_next_extent(
1217                                                 fieinfo, logical, phys, size,
1218                                                 flags);
1219                                         if (ret || blkoff > end_blkoff)
1220                                                 break;
1221
1222                                         /* Start another extent */
1223                                         flags = FIEMAP_EXTENT_MERGED;
1224                                         logical = blkoff << blkbits;
1225                                         phys = blkphy << blkbits;
1226                                         size = n << blkbits;
1227                                 }
1228                         } else {
1229                                 /* Start a new extent */
1230                                 flags = FIEMAP_EXTENT_MERGED;
1231                                 logical = blkoff << blkbits;
1232                                 phys = blkphy << blkbits;
1233                                 size = n << blkbits;
1234                         }
1235                         blkoff += n;
1236                 }
1237                 cond_resched();
1238         } while (true);
1239
1240         /* If ret is 1 then we just hit the end of the extent array */
1241         if (ret == 1)
1242                 ret = 0;
1243
1244         inode_unlock(inode);
1245         return ret;
1246 }