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_readpage() - implement readpage() method of nilfs_aops {}
 144  * address_space_operations.
 145  * @file - file struct of the file to be read
 146  * @page - the page to be read
 147  */
 148 static int nilfs_readpage(struct file *file, struct page *page)
 149 {
 150         return mpage_readpage(page, 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, unsigned flags,
 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, flags, pagep,
 262                                 nilfs_get_block);
 263         if (unlikely(err)) {
 264                 nilfs_write_failed(mapping, pos + len);
 265                 nilfs_transaction_abort(inode->i_sb);
 266         }
 267         return err;
 268 }
 269
 270 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 271                            loff_t pos, unsigned len, unsigned copied,
 272                            struct page *page, void *fsdata)
 273 {
 274         struct inode *inode = mapping->host;
 275         unsigned int start = pos & (PAGE_SIZE - 1);
 276         unsigned int nr_dirty;
 277         int err;
 278
 279         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 280                                                   start + copied);
 281         copied = generic_write_end(file, mapping, pos, len, copied, page,
 282                                    fsdata);
 283         nilfs_set_file_dirty(inode, nr_dirty);
 284         err = nilfs_transaction_commit(inode->i_sb);
 285         return err ? : copied;
 286 }
 287
 288 static ssize_t
 289 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 290 {
 291         struct inode *inode = file_inode(iocb->ki_filp);
 292
 293         if (iov_iter_rw(iter) == WRITE)
 294                 return 0;
 295
 296         /* Needs synchronization with the cleaner */
 297         return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 298 }
 299
 300 const struct address_space_operations nilfs_aops = {
 301         .writepage              = nilfs_writepage,
 302         .readpage               = nilfs_readpage,
 303         .writepages             = nilfs_writepages,
 304         .dirty_folio            = nilfs_dirty_folio,
 305         .readahead              = nilfs_readahead,
 306         .write_begin            = nilfs_write_begin,
 307         .write_end              = nilfs_write_end,
 308         /* .releasepage         = nilfs_releasepage, */
 309         .invalidate_folio       = block_invalidate_folio,
 310         .direct_IO              = nilfs_direct_IO,
 311         .is_partially_uptodate  = block_is_partially_uptodate,
 312 };
 313
 314 static int nilfs_insert_inode_locked(struct inode *inode,
 315                                      struct nilfs_root *root,
 316                                      unsigned long ino)
 317 {
 318         struct nilfs_iget_args args = {
 319                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 320                 .for_btnc = false, .for_shadow = false
 321         };
 322
 323         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 324 }
 325
 326 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 327 {
 328         struct super_block *sb = dir->i_sb;
 329         struct the_nilfs *nilfs = sb->s_fs_info;
 330         struct inode *inode;
 331         struct nilfs_inode_info *ii;
 332         struct nilfs_root *root;
 333         int err = -ENOMEM;
 334         ino_t ino;
 335
 336         inode = new_inode(sb);
 337         if (unlikely(!inode))
 338                 goto failed;
 339
 340         mapping_set_gfp_mask(inode->i_mapping,
 341                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 342
 343         root = NILFS_I(dir)->i_root;
 344         ii = NILFS_I(inode);
 345         ii->i_state = BIT(NILFS_I_NEW);
 346         ii->i_root = root;
 347
 348         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 349         if (unlikely(err))
 350                 goto failed_ifile_create_inode;
 351         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 352
 353         atomic64_inc(&root->inodes_count);
 354         inode_init_owner(&init_user_ns, inode, dir, mode);
 355         inode->i_ino = ino;
 356         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 357
 358         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 359                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 360                 if (err < 0)
 361                         goto failed_after_creation;
 362
 363                 set_bit(NILFS_I_BMAP, &ii->i_state);
 364                 /* No lock is needed; iget() ensures it. */
 365         }
 366
 367         ii->i_flags = nilfs_mask_flags(
 368                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 369
 370         /* ii->i_file_acl = 0; */
 371         /* ii->i_dir_acl = 0; */
 372         ii->i_dir_start_lookup = 0;
 373         nilfs_set_inode_flags(inode);
 374         spin_lock(&nilfs->ns_next_gen_lock);
 375         inode->i_generation = nilfs->ns_next_generation++;
 376         spin_unlock(&nilfs->ns_next_gen_lock);
 377         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 378                 err = -EIO;
 379                 goto failed_after_creation;
 380         }
 381
 382         err = nilfs_init_acl(inode, dir);
 383         if (unlikely(err))
 384                 /*
 385                  * Never occur.  When supporting nilfs_init_acl(),
 386                  * proper cancellation of above jobs should be considered.
 387                  */
 388                 goto failed_after_creation;
 389
 390         return inode;
 391
 392  failed_after_creation:
 393         clear_nlink(inode);
 394         if (inode->i_state & I_NEW)
 395                 unlock_new_inode(inode);
 396         iput(inode);  /*
 397                        * raw_inode will be deleted through
 398                        * nilfs_evict_inode().
 399                        */
 400         goto failed;
 401
 402  failed_ifile_create_inode:
 403         make_bad_inode(inode);
 404         iput(inode);
 405  failed:
 406         return ERR_PTR(err);
 407 }
 408
 409 void nilfs_set_inode_flags(struct inode *inode)
 410 {
 411         unsigned int flags = NILFS_I(inode)->i_flags;
 412         unsigned int new_fl = 0;
 413
 414         if (flags & FS_SYNC_FL)
 415                 new_fl |= S_SYNC;
 416         if (flags & FS_APPEND_FL)
 417                 new_fl |= S_APPEND;
 418         if (flags & FS_IMMUTABLE_FL)
 419                 new_fl |= S_IMMUTABLE;
 420         if (flags & FS_NOATIME_FL)
 421                 new_fl |= S_NOATIME;
 422         if (flags & FS_DIRSYNC_FL)
 423                 new_fl |= S_DIRSYNC;
 424         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 425                         S_NOATIME | S_DIRSYNC);
 426 }
 427
 428 int nilfs_read_inode_common(struct inode *inode,
 429                             struct nilfs_inode *raw_inode)
 430 {
 431         struct nilfs_inode_info *ii = NILFS_I(inode);
 432         int err;
 433
 434         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 435         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 436         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 437         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 438         inode->i_size = le64_to_cpu(raw_inode->i_size);
 439         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 440         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 441         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 442         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 443         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 444         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 445         if (inode->i_nlink == 0)
 446                 return -ESTALE; /* this inode is deleted */
 447
 448         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 449         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 450 #if 0
 451         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 452         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 453                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 454 #endif
 455         ii->i_dir_start_lookup = 0;
 456         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 457
 458         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 459             S_ISLNK(inode->i_mode)) {
 460                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 461                 if (err < 0)
 462                         return err;
 463                 set_bit(NILFS_I_BMAP, &ii->i_state);
 464                 /* No lock is needed; iget() ensures it. */
 465         }
 466         return 0;
 467 }
 468
 469 static int __nilfs_read_inode(struct super_block *sb,
 470                               struct nilfs_root *root, unsigned long ino,
 471                               struct inode *inode)
 472 {
 473         struct the_nilfs *nilfs = sb->s_fs_info;
 474         struct buffer_head *bh;
 475         struct nilfs_inode *raw_inode;
 476         int err;
 477
 478         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 479         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 480         if (unlikely(err))
 481                 goto bad_inode;
 482
 483         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 484
 485         err = nilfs_read_inode_common(inode, raw_inode);
 486         if (err)
 487                 goto failed_unmap;
 488
 489         if (S_ISREG(inode->i_mode)) {
 490                 inode->i_op = &nilfs_file_inode_operations;
 491                 inode->i_fop = &nilfs_file_operations;
 492                 inode->i_mapping->a_ops = &nilfs_aops;
 493         } else if (S_ISDIR(inode->i_mode)) {
 494                 inode->i_op = &nilfs_dir_inode_operations;
 495                 inode->i_fop = &nilfs_dir_operations;
 496                 inode->i_mapping->a_ops = &nilfs_aops;
 497         } else if (S_ISLNK(inode->i_mode)) {
 498                 inode->i_op = &nilfs_symlink_inode_operations;
 499                 inode_nohighmem(inode);
 500                 inode->i_mapping->a_ops = &nilfs_aops;
 501         } else {
 502                 inode->i_op = &nilfs_special_inode_operations;
 503                 init_special_inode(
 504                         inode, inode->i_mode,
 505                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 506         }
 507         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 508         brelse(bh);
 509         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 510         nilfs_set_inode_flags(inode);
 511         mapping_set_gfp_mask(inode->i_mapping,
 512                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 513         return 0;
 514
 515  failed_unmap:
 516         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 517         brelse(bh);
 518
 519  bad_inode:
 520         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 521         return err;
 522 }
 523
 524 static int nilfs_iget_test(struct inode *inode, void *opaque)
 525 {
 526         struct nilfs_iget_args *args = opaque;
 527         struct nilfs_inode_info *ii;
 528
 529         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 530                 return 0;
 531
 532         ii = NILFS_I(inode);
 533         if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
 534                 if (!args->for_btnc)
 535                         return 0;
 536         } else if (args->for_btnc) {
 537                 return 0;
 538         }
 539         if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
 540                 if (!args->for_shadow)
 541                         return 0;
 542         } else if (args->for_shadow) {
 543                 return 0;
 544         }
 545
 546         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 547                 return !args->for_gc;
 548
 549         return args->for_gc && args->cno == ii->i_cno;
 550 }
 551
 552 static int nilfs_iget_set(struct inode *inode, void *opaque)
 553 {
 554         struct nilfs_iget_args *args = opaque;
 555
 556         inode->i_ino = args->ino;
 557         NILFS_I(inode)->i_cno = args->cno;
 558         NILFS_I(inode)->i_root = args->root;
 559         if (args->root && args->ino == NILFS_ROOT_INO)
 560                 nilfs_get_root(args->root);
 561
 562         if (args->for_gc)
 563                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 564         if (args->for_btnc)
 565                 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
 566         if (args->for_shadow)
 567                 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
 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 = false,
 576                 .for_btnc = false, .for_shadow = false
 577         };
 578
 579         return ilookup5(sb, ino, nilfs_iget_test, &args);
 580 }
 581
 582 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 583                                 unsigned long ino)
 584 {
 585         struct nilfs_iget_args args = {
 586                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 587                 .for_btnc = false, .for_shadow = false
 588         };
 589
 590         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 591 }
 592
 593 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 594                          unsigned long ino)
 595 {
 596         struct inode *inode;
 597         int err;
 598
 599         inode = nilfs_iget_locked(sb, root, ino);
 600         if (unlikely(!inode))
 601                 return ERR_PTR(-ENOMEM);
 602         if (!(inode->i_state & I_NEW))
 603                 return inode;
 604
 605         err = __nilfs_read_inode(sb, root, ino, inode);
 606         if (unlikely(err)) {
 607                 iget_failed(inode);
 608                 return ERR_PTR(err);
 609         }
 610         unlock_new_inode(inode);
 611         return inode;
 612 }
 613
 614 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 615                                 __u64 cno)
 616 {
 617         struct nilfs_iget_args args = {
 618                 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
 619                 .for_btnc = false, .for_shadow = false
 620         };
 621         struct inode *inode;
 622         int err;
 623
 624         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 625         if (unlikely(!inode))
 626                 return ERR_PTR(-ENOMEM);
 627         if (!(inode->i_state & I_NEW))
 628                 return inode;
 629
 630         err = nilfs_init_gcinode(inode);
 631         if (unlikely(err)) {
 632                 iget_failed(inode);
 633                 return ERR_PTR(err);
 634         }
 635         unlock_new_inode(inode);
 636         return inode;
 637 }
 638
 639 /**
 640  * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 641  * @inode: inode object
 642  *
 643  * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 644  * or does nothing if the inode already has it.  This function allocates
 645  * an additional inode to maintain page cache of B-tree nodes one-on-one.
 646  *
 647  * Return Value: On success, 0 is returned. On errors, one of the following
 648  * negative error code is returned.
 649  *
 650  * %-ENOMEM - Insufficient memory available.
 651  */
 652 int nilfs_attach_btree_node_cache(struct inode *inode)
 653 {
 654         struct nilfs_inode_info *ii = NILFS_I(inode);
 655         struct inode *btnc_inode;
 656         struct nilfs_iget_args args;
 657
 658         if (ii->i_assoc_inode)
 659                 return 0;
 660
 661         args.ino = inode->i_ino;
 662         args.root = ii->i_root;
 663         args.cno = ii->i_cno;
 664         args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
 665         args.for_btnc = true;
 666         args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
 667
 668         btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 669                                   nilfs_iget_set, &args);
 670         if (unlikely(!btnc_inode))
 671                 return -ENOMEM;
 672         if (btnc_inode->i_state & I_NEW) {
 673                 nilfs_init_btnc_inode(btnc_inode);
 674                 unlock_new_inode(btnc_inode);
 675         }
 676         NILFS_I(btnc_inode)->i_assoc_inode = inode;
 677         NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 678         ii->i_assoc_inode = btnc_inode;
 679
 680         return 0;
 681 }
 682
 683 /**
 684  * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 685  * @inode: inode object
 686  *
 687  * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 688  * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 689  */
 690 void nilfs_detach_btree_node_cache(struct inode *inode)
 691 {
 692         struct nilfs_inode_info *ii = NILFS_I(inode);
 693         struct inode *btnc_inode = ii->i_assoc_inode;
 694
 695         if (btnc_inode) {
 696                 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 697                 ii->i_assoc_inode = NULL;
 698                 iput(btnc_inode);
 699         }
 700 }
 701
 702 /**
 703  * nilfs_iget_for_shadow - obtain inode for shadow mapping
 704  * @inode: inode object that uses shadow mapping
 705  *
 706  * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 707  * caches for shadow mapping.  The page cache for data pages is set up
 708  * in one inode and the one for b-tree node pages is set up in the
 709  * other inode, which is attached to the former inode.
 710  *
 711  * Return Value: On success, a pointer to the inode for data pages is
 712  * returned. On errors, one of the following negative error code is returned
 713  * in a pointer type.
 714  *
 715  * %-ENOMEM - Insufficient memory available.
 716  */
 717 struct inode *nilfs_iget_for_shadow(struct inode *inode)
 718 {
 719         struct nilfs_iget_args args = {
 720                 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
 721                 .for_btnc = false, .for_shadow = true
 722         };
 723         struct inode *s_inode;
 724         int err;
 725
 726         s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 727                                nilfs_iget_set, &args);
 728         if (unlikely(!s_inode))
 729                 return ERR_PTR(-ENOMEM);
 730         if (!(s_inode->i_state & I_NEW))
 731                 return inode;
 732
 733         NILFS_I(s_inode)->i_flags = 0;
 734         memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 735         mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 736
 737         err = nilfs_attach_btree_node_cache(s_inode);
 738         if (unlikely(err)) {
 739                 iget_failed(s_inode);
 740                 return ERR_PTR(err);
 741         }
 742         unlock_new_inode(s_inode);
 743         return s_inode;
 744 }
 745
 746 void nilfs_write_inode_common(struct inode *inode,
 747                               struct nilfs_inode *raw_inode, int has_bmap)
 748 {
 749         struct nilfs_inode_info *ii = NILFS_I(inode);
 750
 751         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 752         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 753         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 754         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 755         raw_inode->i_size = cpu_to_le64(inode->i_size);
 756         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 757         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 758         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 759         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 760         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 761
 762         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 763         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 764
 765         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 766                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 767
 768                 /* zero-fill unused portion in the case of super root block */
 769                 raw_inode->i_xattr = 0;
 770                 raw_inode->i_pad = 0;
 771                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 772                        nilfs->ns_inode_size - sizeof(*raw_inode));
 773         }
 774
 775         if (has_bmap)
 776                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 777         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 778                 raw_inode->i_device_code =
 779                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 780         /*
 781          * When extending inode, nilfs->ns_inode_size should be checked
 782          * for substitutions of appended fields.
 783          */
 784 }
 785
 786 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 787 {
 788         ino_t ino = inode->i_ino;
 789         struct nilfs_inode_info *ii = NILFS_I(inode);
 790         struct inode *ifile = ii->i_root->ifile;
 791         struct nilfs_inode *raw_inode;
 792
 793         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 794
 795         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 796                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 797         if (flags & I_DIRTY_DATASYNC)
 798                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 799
 800         nilfs_write_inode_common(inode, raw_inode, 0);
 801                 /*
 802                  * XXX: call with has_bmap = 0 is a workaround to avoid
 803                  * deadlock of bmap.  This delays update of i_bmap to just
 804                  * before writing.
 805                  */
 806
 807         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 808 }
 809
 810 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 811
 812 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 813                                 unsigned long from)
 814 {
 815         __u64 b;
 816         int ret;
 817
 818         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 819                 return;
 820 repeat:
 821         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 822         if (ret == -ENOENT)
 823                 return;
 824         else if (ret < 0)
 825                 goto failed;
 826
 827         if (b < from)
 828                 return;
 829
 830         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 831         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 832         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 833         if (!ret || (ret == -ENOMEM &&
 834                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 835                 goto repeat;
 836
 837 failed:
 838         nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
 839                    ret, ii->vfs_inode.i_ino);
 840 }
 841
 842 void nilfs_truncate(struct inode *inode)
 843 {
 844         unsigned long blkoff;
 845         unsigned int blocksize;
 846         struct nilfs_transaction_info ti;
 847         struct super_block *sb = inode->i_sb;
 848         struct nilfs_inode_info *ii = NILFS_I(inode);
 849
 850         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 851                 return;
 852         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 853                 return;
 854
 855         blocksize = sb->s_blocksize;
 856         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 857         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 858
 859         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 860
 861         nilfs_truncate_bmap(ii, blkoff);
 862
 863         inode->i_mtime = inode->i_ctime = current_time(inode);
 864         if (IS_SYNC(inode))
 865                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 866
 867         nilfs_mark_inode_dirty(inode);
 868         nilfs_set_file_dirty(inode, 0);
 869         nilfs_transaction_commit(sb);
 870         /*
 871          * May construct a logical segment and may fail in sync mode.
 872          * But truncate has no return value.
 873          */
 874 }
 875
 876 static void nilfs_clear_inode(struct inode *inode)
 877 {
 878         struct nilfs_inode_info *ii = NILFS_I(inode);
 879
 880         /*
 881          * Free resources allocated in nilfs_read_inode(), here.
 882          */
 883         BUG_ON(!list_empty(&ii->i_dirty));
 884         brelse(ii->i_bh);
 885         ii->i_bh = NULL;
 886
 887         if (nilfs_is_metadata_file_inode(inode))
 888                 nilfs_mdt_clear(inode);
 889
 890         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 891                 nilfs_bmap_clear(ii->i_bmap);
 892
 893         if (!test_bit(NILFS_I_BTNC, &ii->i_state))
 894                 nilfs_detach_btree_node_cache(inode);
 895
 896         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 897                 nilfs_put_root(ii->i_root);
 898 }
 899
 900 void nilfs_evict_inode(struct inode *inode)
 901 {
 902         struct nilfs_transaction_info ti;
 903         struct super_block *sb = inode->i_sb;
 904         struct nilfs_inode_info *ii = NILFS_I(inode);
 905         int ret;
 906
 907         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 908                 truncate_inode_pages_final(&inode->i_data);
 909                 clear_inode(inode);
 910                 nilfs_clear_inode(inode);
 911                 return;
 912         }
 913         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 914
 915         truncate_inode_pages_final(&inode->i_data);
 916
 917         /* TODO: some of the following operations may fail.  */
 918         nilfs_truncate_bmap(ii, 0);
 919         nilfs_mark_inode_dirty(inode);
 920         clear_inode(inode);
 921
 922         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 923         if (!ret)
 924                 atomic64_dec(&ii->i_root->inodes_count);
 925
 926         nilfs_clear_inode(inode);
 927
 928         if (IS_SYNC(inode))
 929                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 930         nilfs_transaction_commit(sb);
 931         /*
 932          * May construct a logical segment and may fail in sync mode.
 933          * But delete_inode has no return value.
 934          */
 935 }
 936
 937 int nilfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 938                   struct iattr *iattr)
 939 {
 940         struct nilfs_transaction_info ti;
 941         struct inode *inode = d_inode(dentry);
 942         struct super_block *sb = inode->i_sb;
 943         int err;
 944
 945         err = setattr_prepare(&init_user_ns, dentry, iattr);
 946         if (err)
 947                 return err;
 948
 949         err = nilfs_transaction_begin(sb, &ti, 0);
 950         if (unlikely(err))
 951                 return err;
 952
 953         if ((iattr->ia_valid & ATTR_SIZE) &&
 954             iattr->ia_size != i_size_read(inode)) {
 955                 inode_dio_wait(inode);
 956                 truncate_setsize(inode, iattr->ia_size);
 957                 nilfs_truncate(inode);
 958         }
 959
 960         setattr_copy(&init_user_ns, inode, iattr);
 961         mark_inode_dirty(inode);
 962
 963         if (iattr->ia_valid & ATTR_MODE) {
 964                 err = nilfs_acl_chmod(inode);
 965                 if (unlikely(err))
 966                         goto out_err;
 967         }
 968
 969         return nilfs_transaction_commit(sb);
 970
 971 out_err:
 972         nilfs_transaction_abort(sb);
 973         return err;
 974 }
 975
 976 int nilfs_permission(struct user_namespace *mnt_userns, struct inode *inode,
 977                      int mask)
 978 {
 979         struct nilfs_root *root = NILFS_I(inode)->i_root;
 980
 981         if ((mask & MAY_WRITE) && root &&
 982             root->cno != NILFS_CPTREE_CURRENT_CNO)
 983                 return -EROFS; /* snapshot is not writable */
 984
 985         return generic_permission(&init_user_ns, inode, mask);
 986 }
 987
 988 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 989 {
 990         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 991         struct nilfs_inode_info *ii = NILFS_I(inode);
 992         int err;
 993
 994         spin_lock(&nilfs->ns_inode_lock);
 995         if (ii->i_bh == NULL) {
 996                 spin_unlock(&nilfs->ns_inode_lock);
 997                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 998                                                   inode->i_ino, pbh);
 999                 if (unlikely(err))
1000                         return err;
1001                 spin_lock(&nilfs->ns_inode_lock);
1002                 if (ii->i_bh == NULL)
1003                         ii->i_bh = *pbh;
1004                 else {
1005                         brelse(*pbh);
1006                         *pbh = ii->i_bh;
1007                 }
1008         } else
1009                 *pbh = ii->i_bh;
1010
1011         get_bh(*pbh);
1012         spin_unlock(&nilfs->ns_inode_lock);
1013         return 0;
1014 }
1015
1016 int nilfs_inode_dirty(struct inode *inode)
1017 {
1018         struct nilfs_inode_info *ii = NILFS_I(inode);
1019         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1020         int ret = 0;
1021
1022         if (!list_empty(&ii->i_dirty)) {
1023                 spin_lock(&nilfs->ns_inode_lock);
1024                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1025                         test_bit(NILFS_I_BUSY, &ii->i_state);
1026                 spin_unlock(&nilfs->ns_inode_lock);
1027         }
1028         return ret;
1029 }
1030
1031 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1032 {
1033         struct nilfs_inode_info *ii = NILFS_I(inode);
1034         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1035
1036         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1037
1038         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1039                 return 0;
1040
1041         spin_lock(&nilfs->ns_inode_lock);
1042         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1043             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1044                 /*
1045                  * Because this routine may race with nilfs_dispose_list(),
1046                  * we have to check NILFS_I_QUEUED here, too.
1047                  */
1048                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1049                         /*
1050                          * This will happen when somebody is freeing
1051                          * this inode.
1052                          */
1053                         nilfs_warn(inode->i_sb,
1054                                    "cannot set file dirty (ino=%lu): the file is being freed",
1055                                    inode->i_ino);
1056                         spin_unlock(&nilfs->ns_inode_lock);
1057                         return -EINVAL; /*
1058                                          * NILFS_I_DIRTY may remain for
1059                                          * freeing inode.
1060                                          */
1061                 }
1062                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1063                 set_bit(NILFS_I_QUEUED, &ii->i_state);
1064         }
1065         spin_unlock(&nilfs->ns_inode_lock);
1066         return 0;
1067 }
1068
1069 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1070 {
1071         struct buffer_head *ibh;
1072         int err;
1073
1074         err = nilfs_load_inode_block(inode, &ibh);
1075         if (unlikely(err)) {
1076                 nilfs_warn(inode->i_sb,
1077                            "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1078                            inode->i_ino, err);
1079                 return err;
1080         }
1081         nilfs_update_inode(inode, ibh, flags);
1082         mark_buffer_dirty(ibh);
1083         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1084         brelse(ibh);
1085         return 0;
1086 }
1087
1088 /**
1089  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1090  * @inode: inode of the file to be registered.
1091  *
1092  * nilfs_dirty_inode() loads a inode block containing the specified
1093  * @inode and copies data from a nilfs_inode to a corresponding inode
1094  * entry in the inode block. This operation is excluded from the segment
1095  * construction. This function can be called both as a single operation
1096  * and as a part of indivisible file operations.
1097  */
1098 void nilfs_dirty_inode(struct inode *inode, int flags)
1099 {
1100         struct nilfs_transaction_info ti;
1101         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1102
1103         if (is_bad_inode(inode)) {
1104                 nilfs_warn(inode->i_sb,
1105                            "tried to mark bad_inode dirty. ignored.");
1106                 dump_stack();
1107                 return;
1108         }
1109         if (mdi) {
1110                 nilfs_mdt_mark_dirty(inode);
1111                 return;
1112         }
1113         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1114         __nilfs_mark_inode_dirty(inode, flags);
1115         nilfs_transaction_commit(inode->i_sb); /* never fails */
1116 }
1117
1118 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1119                  __u64 start, __u64 len)
1120 {
1121         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1122         __u64 logical = 0, phys = 0, size = 0;
1123         __u32 flags = 0;
1124         loff_t isize;
1125         sector_t blkoff, end_blkoff;
1126         sector_t delalloc_blkoff;
1127         unsigned long delalloc_blklen;
1128         unsigned int blkbits = inode->i_blkbits;
1129         int ret, n;
1130
1131         ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1132         if (ret)
1133                 return ret;
1134
1135         inode_lock(inode);
1136
1137         isize = i_size_read(inode);
1138
1139         blkoff = start >> blkbits;
1140         end_blkoff = (start + len - 1) >> blkbits;
1141
1142         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1143                                                         &delalloc_blkoff);
1144
1145         do {
1146                 __u64 blkphy;
1147                 unsigned int maxblocks;
1148
1149                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1150                         if (size) {
1151                                 /* End of the current extent */
1152                                 ret = fiemap_fill_next_extent(
1153                                         fieinfo, logical, phys, size, flags);
1154                                 if (ret)
1155                                         break;
1156                         }
1157                         if (blkoff > end_blkoff)
1158                                 break;
1159
1160                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1161                         logical = blkoff << blkbits;
1162                         phys = 0;
1163                         size = delalloc_blklen << blkbits;
1164
1165                         blkoff = delalloc_blkoff + delalloc_blklen;
1166                         delalloc_blklen = nilfs_find_uncommitted_extent(
1167                                 inode, blkoff, &delalloc_blkoff);
1168                         continue;
1169                 }
1170
1171                 /*
1172                  * Limit the number of blocks that we look up so as
1173                  * not to get into the next delayed allocation extent.
1174                  */
1175                 maxblocks = INT_MAX;
1176                 if (delalloc_blklen)
1177                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1178                                           maxblocks);
1179                 blkphy = 0;
1180
1181                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1182                 n = nilfs_bmap_lookup_contig(
1183                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1184                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1185
1186                 if (n < 0) {
1187                         int past_eof;
1188
1189                         if (unlikely(n != -ENOENT))
1190                                 break; /* error */
1191
1192                         /* HOLE */
1193                         blkoff++;
1194                         past_eof = ((blkoff << blkbits) >= isize);
1195
1196                         if (size) {
1197                                 /* End of the current extent */
1198
1199                                 if (past_eof)
1200                                         flags |= FIEMAP_EXTENT_LAST;
1201
1202                                 ret = fiemap_fill_next_extent(
1203                                         fieinfo, logical, phys, size, flags);
1204                                 if (ret)
1205                                         break;
1206                                 size = 0;
1207                         }
1208                         if (blkoff > end_blkoff || past_eof)
1209                                 break;
1210                 } else {
1211                         if (size) {
1212                                 if (phys && blkphy << blkbits == phys + size) {
1213                                         /* The current extent goes on */
1214                                         size += n << blkbits;
1215                                 } else {
1216                                         /* Terminate the current extent */
1217                                         ret = fiemap_fill_next_extent(
1218                                                 fieinfo, logical, phys, size,
1219                                                 flags);
1220                                         if (ret || blkoff > end_blkoff)
1221                                                 break;
1222
1223                                         /* Start another extent */
1224                                         flags = FIEMAP_EXTENT_MERGED;
1225                                         logical = blkoff << blkbits;
1226                                         phys = blkphy << blkbits;
1227                                         size = n << blkbits;
1228                                 }
1229                         } else {
1230                                 /* Start a new extent */
1231                                 flags = FIEMAP_EXTENT_MERGED;
1232                                 logical = blkoff << blkbits;
1233                                 phys = blkphy << blkbits;
1234                                 size = n << blkbits;
1235                         }
1236                         blkoff += n;
1237                 }
1238                 cond_resched();
1239         } while (true);
1240
1241         /* If ret is 1 then we just hit the end of the extent array */
1242         if (ret == 1)
1243                 ret = 0;
1244
1245         inode_unlock(inode);
1246         return ret;
1247 }