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         struct buffer_head *bh;
 332         int err = -ENOMEM;
 333         ino_t ino;
 334
 335         inode = new_inode(sb);
 336         if (unlikely(!inode))
 337                 goto failed;
 338
 339         mapping_set_gfp_mask(inode->i_mapping,
 340                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 341
 342         root = NILFS_I(dir)->i_root;
 343         ii = NILFS_I(inode);
 344         ii->i_state = BIT(NILFS_I_NEW);
 345         ii->i_root = root;
 346
 347         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 348         if (unlikely(err))
 349                 goto failed_ifile_create_inode;
 350         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 351
 352         if (unlikely(ino < NILFS_USER_INO)) {
 353                 nilfs_warn(sb,
 354                            "inode bitmap is inconsistent for reserved inodes");
 355                 do {
 356                         brelse(bh);
 357                         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 358                         if (unlikely(err))
 359                                 goto failed_ifile_create_inode;
 360                 } while (ino < NILFS_USER_INO);
 361
 362                 nilfs_info(sb, "repaired inode bitmap for reserved inodes");
 363         }
 364         ii->i_bh = bh;
 365
 366         atomic64_inc(&root->inodes_count);
 367         inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
 368         inode->i_ino = ino;
 369         simple_inode_init_ts(inode);
 370
 371         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 372                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 373                 if (err < 0)
 374                         goto failed_after_creation;
 375
 376                 set_bit(NILFS_I_BMAP, &ii->i_state);
 377                 /* No lock is needed; iget() ensures it. */
 378         }
 379
 380         ii->i_flags = nilfs_mask_flags(
 381                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 382
 383         /* ii->i_file_acl = 0; */
 384         /* ii->i_dir_acl = 0; */
 385         ii->i_dir_start_lookup = 0;
 386         nilfs_set_inode_flags(inode);
 387         spin_lock(&nilfs->ns_next_gen_lock);
 388         inode->i_generation = nilfs->ns_next_generation++;
 389         spin_unlock(&nilfs->ns_next_gen_lock);
 390         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 391                 err = -EIO;
 392                 goto failed_after_creation;
 393         }
 394
 395         err = nilfs_init_acl(inode, dir);
 396         if (unlikely(err))
 397                 /*
 398                  * Never occur.  When supporting nilfs_init_acl(),
 399                  * proper cancellation of above jobs should be considered.
 400                  */
 401                 goto failed_after_creation;
 402
 403         return inode;
 404
 405  failed_after_creation:
 406         clear_nlink(inode);
 407         if (inode->i_state & I_NEW)
 408                 unlock_new_inode(inode);
 409         iput(inode);  /*
 410                        * raw_inode will be deleted through
 411                        * nilfs_evict_inode().
 412                        */
 413         goto failed;
 414
 415  failed_ifile_create_inode:
 416         make_bad_inode(inode);
 417         iput(inode);
 418  failed:
 419         return ERR_PTR(err);
 420 }
 421
 422 void nilfs_set_inode_flags(struct inode *inode)
 423 {
 424         unsigned int flags = NILFS_I(inode)->i_flags;
 425         unsigned int new_fl = 0;
 426
 427         if (flags & FS_SYNC_FL)
 428                 new_fl |= S_SYNC;
 429         if (flags & FS_APPEND_FL)
 430                 new_fl |= S_APPEND;
 431         if (flags & FS_IMMUTABLE_FL)
 432                 new_fl |= S_IMMUTABLE;
 433         if (flags & FS_NOATIME_FL)
 434                 new_fl |= S_NOATIME;
 435         if (flags & FS_DIRSYNC_FL)
 436                 new_fl |= S_DIRSYNC;
 437         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 438                         S_NOATIME | S_DIRSYNC);
 439 }
 440
 441 int nilfs_read_inode_common(struct inode *inode,
 442                             struct nilfs_inode *raw_inode)
 443 {
 444         struct nilfs_inode_info *ii = NILFS_I(inode);
 445         int err;
 446
 447         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 448         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 449         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 450         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 451         inode->i_size = le64_to_cpu(raw_inode->i_size);
 452         inode_set_atime(inode, le64_to_cpu(raw_inode->i_mtime),
 453                         le32_to_cpu(raw_inode->i_mtime_nsec));
 454         inode_set_ctime(inode, le64_to_cpu(raw_inode->i_ctime),
 455                         le32_to_cpu(raw_inode->i_ctime_nsec));
 456         inode_set_mtime(inode, le64_to_cpu(raw_inode->i_mtime),
 457                         le32_to_cpu(raw_inode->i_mtime_nsec));
 458         if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
 459                 return -EIO; /* this inode is for metadata and corrupted */
 460         if (inode->i_nlink == 0)
 461                 return -ESTALE; /* this inode is deleted */
 462
 463         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 464         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 465 #if 0
 466         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 467         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 468                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 469 #endif
 470         ii->i_dir_start_lookup = 0;
 471         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 472
 473         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 474             S_ISLNK(inode->i_mode)) {
 475                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 476                 if (err < 0)
 477                         return err;
 478                 set_bit(NILFS_I_BMAP, &ii->i_state);
 479                 /* No lock is needed; iget() ensures it. */
 480         }
 481         return 0;
 482 }
 483
 484 static int __nilfs_read_inode(struct super_block *sb,
 485                               struct nilfs_root *root, unsigned long ino,
 486                               struct inode *inode)
 487 {
 488         struct the_nilfs *nilfs = sb->s_fs_info;
 489         struct buffer_head *bh;
 490         struct nilfs_inode *raw_inode;
 491         int err;
 492
 493         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 494         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 495         if (unlikely(err))
 496                 goto bad_inode;
 497
 498         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 499
 500         err = nilfs_read_inode_common(inode, raw_inode);
 501         if (err)
 502                 goto failed_unmap;
 503
 504         if (S_ISREG(inode->i_mode)) {
 505                 inode->i_op = &nilfs_file_inode_operations;
 506                 inode->i_fop = &nilfs_file_operations;
 507                 inode->i_mapping->a_ops = &nilfs_aops;
 508         } else if (S_ISDIR(inode->i_mode)) {
 509                 inode->i_op = &nilfs_dir_inode_operations;
 510                 inode->i_fop = &nilfs_dir_operations;
 511                 inode->i_mapping->a_ops = &nilfs_aops;
 512         } else if (S_ISLNK(inode->i_mode)) {
 513                 inode->i_op = &nilfs_symlink_inode_operations;
 514                 inode_nohighmem(inode);
 515                 inode->i_mapping->a_ops = &nilfs_aops;
 516         } else {
 517                 inode->i_op = &nilfs_special_inode_operations;
 518                 init_special_inode(
 519                         inode, inode->i_mode,
 520                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 521         }
 522         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 523         brelse(bh);
 524         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 525         nilfs_set_inode_flags(inode);
 526         mapping_set_gfp_mask(inode->i_mapping,
 527                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 528         return 0;
 529
 530  failed_unmap:
 531         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 532         brelse(bh);
 533
 534  bad_inode:
 535         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 536         return err;
 537 }
 538
 539 static int nilfs_iget_test(struct inode *inode, void *opaque)
 540 {
 541         struct nilfs_iget_args *args = opaque;
 542         struct nilfs_inode_info *ii;
 543
 544         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 545                 return 0;
 546
 547         ii = NILFS_I(inode);
 548         if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
 549                 if (!args->for_btnc)
 550                         return 0;
 551         } else if (args->for_btnc) {
 552                 return 0;
 553         }
 554         if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
 555                 if (!args->for_shadow)
 556                         return 0;
 557         } else if (args->for_shadow) {
 558                 return 0;
 559         }
 560
 561         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 562                 return !args->for_gc;
 563
 564         return args->for_gc && args->cno == ii->i_cno;
 565 }
 566
 567 static int nilfs_iget_set(struct inode *inode, void *opaque)
 568 {
 569         struct nilfs_iget_args *args = opaque;
 570
 571         inode->i_ino = args->ino;
 572         NILFS_I(inode)->i_cno = args->cno;
 573         NILFS_I(inode)->i_root = args->root;
 574         if (args->root && args->ino == NILFS_ROOT_INO)
 575                 nilfs_get_root(args->root);
 576
 577         if (args->for_gc)
 578                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 579         if (args->for_btnc)
 580                 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
 581         if (args->for_shadow)
 582                 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
 583         return 0;
 584 }
 585
 586 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 587                             unsigned long ino)
 588 {
 589         struct nilfs_iget_args args = {
 590                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 591                 .for_btnc = false, .for_shadow = false
 592         };
 593
 594         return ilookup5(sb, ino, nilfs_iget_test, &args);
 595 }
 596
 597 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 598                                 unsigned long ino)
 599 {
 600         struct nilfs_iget_args args = {
 601                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 602                 .for_btnc = false, .for_shadow = false
 603         };
 604
 605         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 606 }
 607
 608 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 609                          unsigned long ino)
 610 {
 611         struct inode *inode;
 612         int err;
 613
 614         inode = nilfs_iget_locked(sb, root, ino);
 615         if (unlikely(!inode))
 616                 return ERR_PTR(-ENOMEM);
 617         if (!(inode->i_state & I_NEW))
 618                 return inode;
 619
 620         err = __nilfs_read_inode(sb, root, ino, inode);
 621         if (unlikely(err)) {
 622                 iget_failed(inode);
 623                 return ERR_PTR(err);
 624         }
 625         unlock_new_inode(inode);
 626         return inode;
 627 }
 628
 629 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 630                                 __u64 cno)
 631 {
 632         struct nilfs_iget_args args = {
 633                 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
 634                 .for_btnc = false, .for_shadow = false
 635         };
 636         struct inode *inode;
 637         int err;
 638
 639         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 640         if (unlikely(!inode))
 641                 return ERR_PTR(-ENOMEM);
 642         if (!(inode->i_state & I_NEW))
 643                 return inode;
 644
 645         err = nilfs_init_gcinode(inode);
 646         if (unlikely(err)) {
 647                 iget_failed(inode);
 648                 return ERR_PTR(err);
 649         }
 650         unlock_new_inode(inode);
 651         return inode;
 652 }
 653
 654 /**
 655  * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 656  * @inode: inode object
 657  *
 658  * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 659  * or does nothing if the inode already has it.  This function allocates
 660  * an additional inode to maintain page cache of B-tree nodes one-on-one.
 661  *
 662  * Return Value: On success, 0 is returned. On errors, one of the following
 663  * negative error code is returned.
 664  *
 665  * %-ENOMEM - Insufficient memory available.
 666  */
 667 int nilfs_attach_btree_node_cache(struct inode *inode)
 668 {
 669         struct nilfs_inode_info *ii = NILFS_I(inode);
 670         struct inode *btnc_inode;
 671         struct nilfs_iget_args args;
 672
 673         if (ii->i_assoc_inode)
 674                 return 0;
 675
 676         args.ino = inode->i_ino;
 677         args.root = ii->i_root;
 678         args.cno = ii->i_cno;
 679         args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
 680         args.for_btnc = true;
 681         args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
 682
 683         btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 684                                   nilfs_iget_set, &args);
 685         if (unlikely(!btnc_inode))
 686                 return -ENOMEM;
 687         if (btnc_inode->i_state & I_NEW) {
 688                 nilfs_init_btnc_inode(btnc_inode);
 689                 unlock_new_inode(btnc_inode);
 690         }
 691         NILFS_I(btnc_inode)->i_assoc_inode = inode;
 692         NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 693         ii->i_assoc_inode = btnc_inode;
 694
 695         return 0;
 696 }
 697
 698 /**
 699  * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 700  * @inode: inode object
 701  *
 702  * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 703  * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 704  */
 705 void nilfs_detach_btree_node_cache(struct inode *inode)
 706 {
 707         struct nilfs_inode_info *ii = NILFS_I(inode);
 708         struct inode *btnc_inode = ii->i_assoc_inode;
 709
 710         if (btnc_inode) {
 711                 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 712                 ii->i_assoc_inode = NULL;
 713                 iput(btnc_inode);
 714         }
 715 }
 716
 717 /**
 718  * nilfs_iget_for_shadow - obtain inode for shadow mapping
 719  * @inode: inode object that uses shadow mapping
 720  *
 721  * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 722  * caches for shadow mapping.  The page cache for data pages is set up
 723  * in one inode and the one for b-tree node pages is set up in the
 724  * other inode, which is attached to the former inode.
 725  *
 726  * Return Value: On success, a pointer to the inode for data pages is
 727  * returned. On errors, one of the following negative error code is returned
 728  * in a pointer type.
 729  *
 730  * %-ENOMEM - Insufficient memory available.
 731  */
 732 struct inode *nilfs_iget_for_shadow(struct inode *inode)
 733 {
 734         struct nilfs_iget_args args = {
 735                 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
 736                 .for_btnc = false, .for_shadow = true
 737         };
 738         struct inode *s_inode;
 739         int err;
 740
 741         s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 742                                nilfs_iget_set, &args);
 743         if (unlikely(!s_inode))
 744                 return ERR_PTR(-ENOMEM);
 745         if (!(s_inode->i_state & I_NEW))
 746                 return inode;
 747
 748         NILFS_I(s_inode)->i_flags = 0;
 749         memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 750         mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 751
 752         err = nilfs_attach_btree_node_cache(s_inode);
 753         if (unlikely(err)) {
 754                 iget_failed(s_inode);
 755                 return ERR_PTR(err);
 756         }
 757         unlock_new_inode(s_inode);
 758         return s_inode;
 759 }
 760
 761 void nilfs_write_inode_common(struct inode *inode,
 762                               struct nilfs_inode *raw_inode, int has_bmap)
 763 {
 764         struct nilfs_inode_info *ii = NILFS_I(inode);
 765
 766         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 767         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 768         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 769         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 770         raw_inode->i_size = cpu_to_le64(inode->i_size);
 771         raw_inode->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
 772         raw_inode->i_mtime = cpu_to_le64(inode_get_mtime_sec(inode));
 773         raw_inode->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
 774         raw_inode->i_mtime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
 775         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 776
 777         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 778         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 779
 780         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 781                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 782
 783                 /* zero-fill unused portion in the case of super root block */
 784                 raw_inode->i_xattr = 0;
 785                 raw_inode->i_pad = 0;
 786                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 787                        nilfs->ns_inode_size - sizeof(*raw_inode));
 788         }
 789
 790         if (has_bmap)
 791                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 792         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 793                 raw_inode->i_device_code =
 794                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 795         /*
 796          * When extending inode, nilfs->ns_inode_size should be checked
 797          * for substitutions of appended fields.
 798          */
 799 }
 800
 801 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 802 {
 803         ino_t ino = inode->i_ino;
 804         struct nilfs_inode_info *ii = NILFS_I(inode);
 805         struct inode *ifile = ii->i_root->ifile;
 806         struct nilfs_inode *raw_inode;
 807
 808         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 809
 810         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 811                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 812         if (flags & I_DIRTY_DATASYNC)
 813                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 814
 815         nilfs_write_inode_common(inode, raw_inode, 0);
 816                 /*
 817                  * XXX: call with has_bmap = 0 is a workaround to avoid
 818                  * deadlock of bmap.  This delays update of i_bmap to just
 819                  * before writing.
 820                  */
 821
 822         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 823 }
 824
 825 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 826
 827 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 828                                 unsigned long from)
 829 {
 830         __u64 b;
 831         int ret;
 832
 833         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 834                 return;
 835 repeat:
 836         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 837         if (ret == -ENOENT)
 838                 return;
 839         else if (ret < 0)
 840                 goto failed;
 841
 842         if (b < from)
 843                 return;
 844
 845         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 846         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 847         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 848         if (!ret || (ret == -ENOMEM &&
 849                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 850                 goto repeat;
 851
 852 failed:
 853         nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
 854                    ret, ii->vfs_inode.i_ino);
 855 }
 856
 857 void nilfs_truncate(struct inode *inode)
 858 {
 859         unsigned long blkoff;
 860         unsigned int blocksize;
 861         struct nilfs_transaction_info ti;
 862         struct super_block *sb = inode->i_sb;
 863         struct nilfs_inode_info *ii = NILFS_I(inode);
 864
 865         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 866                 return;
 867         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 868                 return;
 869
 870         blocksize = sb->s_blocksize;
 871         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 872         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 873
 874         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 875
 876         nilfs_truncate_bmap(ii, blkoff);
 877
 878         inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
 879         if (IS_SYNC(inode))
 880                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 881
 882         nilfs_mark_inode_dirty(inode);
 883         nilfs_set_file_dirty(inode, 0);
 884         nilfs_transaction_commit(sb);
 885         /*
 886          * May construct a logical segment and may fail in sync mode.
 887          * But truncate has no return value.
 888          */
 889 }
 890
 891 static void nilfs_clear_inode(struct inode *inode)
 892 {
 893         struct nilfs_inode_info *ii = NILFS_I(inode);
 894
 895         /*
 896          * Free resources allocated in nilfs_read_inode(), here.
 897          */
 898         BUG_ON(!list_empty(&ii->i_dirty));
 899         brelse(ii->i_bh);
 900         ii->i_bh = NULL;
 901
 902         if (nilfs_is_metadata_file_inode(inode))
 903                 nilfs_mdt_clear(inode);
 904
 905         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 906                 nilfs_bmap_clear(ii->i_bmap);
 907
 908         if (!test_bit(NILFS_I_BTNC, &ii->i_state))
 909                 nilfs_detach_btree_node_cache(inode);
 910
 911         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 912                 nilfs_put_root(ii->i_root);
 913 }
 914
 915 void nilfs_evict_inode(struct inode *inode)
 916 {
 917         struct nilfs_transaction_info ti;
 918         struct super_block *sb = inode->i_sb;
 919         struct nilfs_inode_info *ii = NILFS_I(inode);
 920         struct the_nilfs *nilfs;
 921         int ret;
 922
 923         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 924                 truncate_inode_pages_final(&inode->i_data);
 925                 clear_inode(inode);
 926                 nilfs_clear_inode(inode);
 927                 return;
 928         }
 929         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 930
 931         truncate_inode_pages_final(&inode->i_data);
 932
 933         nilfs = sb->s_fs_info;
 934         if (unlikely(sb_rdonly(sb) || !nilfs->ns_writer)) {
 935                 /*
 936                  * If this inode is about to be disposed after the file system
 937                  * has been degraded to read-only due to file system corruption
 938                  * or after the writer has been detached, do not make any
 939                  * changes that cause writes, just clear it.
 940                  * Do this check after read-locking ns_segctor_sem by
 941                  * nilfs_transaction_begin() in order to avoid a race with
 942                  * the writer detach operation.
 943                  */
 944                 clear_inode(inode);
 945                 nilfs_clear_inode(inode);
 946                 nilfs_transaction_abort(sb);
 947                 return;
 948         }
 949
 950         /* TODO: some of the following operations may fail.  */
 951         nilfs_truncate_bmap(ii, 0);
 952         nilfs_mark_inode_dirty(inode);
 953         clear_inode(inode);
 954
 955         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 956         if (!ret)
 957                 atomic64_dec(&ii->i_root->inodes_count);
 958
 959         nilfs_clear_inode(inode);
 960
 961         if (IS_SYNC(inode))
 962                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 963         nilfs_transaction_commit(sb);
 964         /*
 965          * May construct a logical segment and may fail in sync mode.
 966          * But delete_inode has no return value.
 967          */
 968 }
 969
 970 int nilfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 971                   struct iattr *iattr)
 972 {
 973         struct nilfs_transaction_info ti;
 974         struct inode *inode = d_inode(dentry);
 975         struct super_block *sb = inode->i_sb;
 976         int err;
 977
 978         err = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
 979         if (err)
 980                 return err;
 981
 982         err = nilfs_transaction_begin(sb, &ti, 0);
 983         if (unlikely(err))
 984                 return err;
 985
 986         if ((iattr->ia_valid & ATTR_SIZE) &&
 987             iattr->ia_size != i_size_read(inode)) {
 988                 inode_dio_wait(inode);
 989                 truncate_setsize(inode, iattr->ia_size);
 990                 nilfs_truncate(inode);
 991         }
 992
 993         setattr_copy(&nop_mnt_idmap, inode, iattr);
 994         mark_inode_dirty(inode);
 995
 996         if (iattr->ia_valid & ATTR_MODE) {
 997                 err = nilfs_acl_chmod(inode);
 998                 if (unlikely(err))
 999                         goto out_err;
1000         }
1001
1002         return nilfs_transaction_commit(sb);
1003
1004 out_err:
1005         nilfs_transaction_abort(sb);
1006         return err;
1007 }
1008
1009 int nilfs_permission(struct mnt_idmap *idmap, struct inode *inode,
1010                      int mask)
1011 {
1012         struct nilfs_root *root = NILFS_I(inode)->i_root;
1013
1014         if ((mask & MAY_WRITE) && root &&
1015             root->cno != NILFS_CPTREE_CURRENT_CNO)
1016                 return -EROFS; /* snapshot is not writable */
1017
1018         return generic_permission(&nop_mnt_idmap, inode, mask);
1019 }
1020
1021 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1022 {
1023         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1024         struct nilfs_inode_info *ii = NILFS_I(inode);
1025         int err;
1026
1027         spin_lock(&nilfs->ns_inode_lock);
1028         if (ii->i_bh == NULL || unlikely(!buffer_uptodate(ii->i_bh))) {
1029                 spin_unlock(&nilfs->ns_inode_lock);
1030                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1031                                                   inode->i_ino, pbh);
1032                 if (unlikely(err))
1033                         return err;
1034                 spin_lock(&nilfs->ns_inode_lock);
1035                 if (ii->i_bh == NULL)
1036                         ii->i_bh = *pbh;
1037                 else if (unlikely(!buffer_uptodate(ii->i_bh))) {
1038                         __brelse(ii->i_bh);
1039                         ii->i_bh = *pbh;
1040                 } else {
1041                         brelse(*pbh);
1042                         *pbh = ii->i_bh;
1043                 }
1044         } else
1045                 *pbh = ii->i_bh;
1046
1047         get_bh(*pbh);
1048         spin_unlock(&nilfs->ns_inode_lock);
1049         return 0;
1050 }
1051
1052 int nilfs_inode_dirty(struct inode *inode)
1053 {
1054         struct nilfs_inode_info *ii = NILFS_I(inode);
1055         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1056         int ret = 0;
1057
1058         if (!list_empty(&ii->i_dirty)) {
1059                 spin_lock(&nilfs->ns_inode_lock);
1060                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1061                         test_bit(NILFS_I_BUSY, &ii->i_state);
1062                 spin_unlock(&nilfs->ns_inode_lock);
1063         }
1064         return ret;
1065 }
1066
1067 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1068 {
1069         struct nilfs_inode_info *ii = NILFS_I(inode);
1070         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1071
1072         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1073
1074         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1075                 return 0;
1076
1077         spin_lock(&nilfs->ns_inode_lock);
1078         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1079             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1080                 /*
1081                  * Because this routine may race with nilfs_dispose_list(),
1082                  * we have to check NILFS_I_QUEUED here, too.
1083                  */
1084                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1085                         /*
1086                          * This will happen when somebody is freeing
1087                          * this inode.
1088                          */
1089                         nilfs_warn(inode->i_sb,
1090                                    "cannot set file dirty (ino=%lu): the file is being freed",
1091                                    inode->i_ino);
1092                         spin_unlock(&nilfs->ns_inode_lock);
1093                         return -EINVAL; /*
1094                                          * NILFS_I_DIRTY may remain for
1095                                          * freeing inode.
1096                                          */
1097                 }
1098                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1099                 set_bit(NILFS_I_QUEUED, &ii->i_state);
1100         }
1101         spin_unlock(&nilfs->ns_inode_lock);
1102         return 0;
1103 }
1104
1105 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1106 {
1107         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1108         struct buffer_head *ibh;
1109         int err;
1110
1111         /*
1112          * Do not dirty inodes after the log writer has been detached
1113          * and its nilfs_root struct has been freed.
1114          */
1115         if (unlikely(nilfs_purging(nilfs)))
1116                 return 0;
1117
1118         err = nilfs_load_inode_block(inode, &ibh);
1119         if (unlikely(err)) {
1120                 nilfs_warn(inode->i_sb,
1121                            "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1122                            inode->i_ino, err);
1123                 return err;
1124         }
1125         nilfs_update_inode(inode, ibh, flags);
1126         mark_buffer_dirty(ibh);
1127         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1128         brelse(ibh);
1129         return 0;
1130 }
1131
1132 /**
1133  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1134  * @inode: inode of the file to be registered.
1135  * @flags: flags to determine the dirty state of the inode
1136  *
1137  * nilfs_dirty_inode() loads a inode block containing the specified
1138  * @inode and copies data from a nilfs_inode to a corresponding inode
1139  * entry in the inode block. This operation is excluded from the segment
1140  * construction. This function can be called both as a single operation
1141  * and as a part of indivisible file operations.
1142  */
1143 void nilfs_dirty_inode(struct inode *inode, int flags)
1144 {
1145         struct nilfs_transaction_info ti;
1146         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1147
1148         if (is_bad_inode(inode)) {
1149                 nilfs_warn(inode->i_sb,
1150                            "tried to mark bad_inode dirty. ignored.");
1151                 dump_stack();
1152                 return;
1153         }
1154         if (mdi) {
1155                 nilfs_mdt_mark_dirty(inode);
1156                 return;
1157         }
1158         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1159         __nilfs_mark_inode_dirty(inode, flags);
1160         nilfs_transaction_commit(inode->i_sb); /* never fails */
1161 }
1162
1163 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1164                  __u64 start, __u64 len)
1165 {
1166         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1167         __u64 logical = 0, phys = 0, size = 0;
1168         __u32 flags = 0;
1169         loff_t isize;
1170         sector_t blkoff, end_blkoff;
1171         sector_t delalloc_blkoff;
1172         unsigned long delalloc_blklen;
1173         unsigned int blkbits = inode->i_blkbits;
1174         int ret, n;
1175
1176         ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1177         if (ret)
1178                 return ret;
1179
1180         inode_lock(inode);
1181
1182         isize = i_size_read(inode);
1183
1184         blkoff = start >> blkbits;
1185         end_blkoff = (start + len - 1) >> blkbits;
1186
1187         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1188                                                         &delalloc_blkoff);
1189
1190         do {
1191                 __u64 blkphy;
1192                 unsigned int maxblocks;
1193
1194                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1195                         if (size) {
1196                                 /* End of the current extent */
1197                                 ret = fiemap_fill_next_extent(
1198                                         fieinfo, logical, phys, size, flags);
1199                                 if (ret)
1200                                         break;
1201                         }
1202                         if (blkoff > end_blkoff)
1203                                 break;
1204
1205                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1206                         logical = blkoff << blkbits;
1207                         phys = 0;
1208                         size = delalloc_blklen << blkbits;
1209
1210                         blkoff = delalloc_blkoff + delalloc_blklen;
1211                         delalloc_blklen = nilfs_find_uncommitted_extent(
1212                                 inode, blkoff, &delalloc_blkoff);
1213                         continue;
1214                 }
1215
1216                 /*
1217                  * Limit the number of blocks that we look up so as
1218                  * not to get into the next delayed allocation extent.
1219                  */
1220                 maxblocks = INT_MAX;
1221                 if (delalloc_blklen)
1222                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1223                                           maxblocks);
1224                 blkphy = 0;
1225
1226                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1227                 n = nilfs_bmap_lookup_contig(
1228                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1229                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1230
1231                 if (n < 0) {
1232                         int past_eof;
1233
1234                         if (unlikely(n != -ENOENT))
1235                                 break; /* error */
1236
1237                         /* HOLE */
1238                         blkoff++;
1239                         past_eof = ((blkoff << blkbits) >= isize);
1240
1241                         if (size) {
1242                                 /* End of the current extent */
1243
1244                                 if (past_eof)
1245                                         flags |= FIEMAP_EXTENT_LAST;
1246
1247                                 ret = fiemap_fill_next_extent(
1248                                         fieinfo, logical, phys, size, flags);
1249                                 if (ret)
1250                                         break;
1251                                 size = 0;
1252                         }
1253                         if (blkoff > end_blkoff || past_eof)
1254                                 break;
1255                 } else {
1256                         if (size) {
1257                                 if (phys && blkphy << blkbits == phys + size) {
1258                                         /* The current extent goes on */
1259                                         size += n << blkbits;
1260                                 } else {
1261                                         /* Terminate the current extent */
1262                                         ret = fiemap_fill_next_extent(
1263                                                 fieinfo, logical, phys, size,
1264                                                 flags);
1265                                         if (ret || blkoff > end_blkoff)
1266                                                 break;
1267
1268                                         /* Start another extent */
1269                                         flags = FIEMAP_EXTENT_MERGED;
1270                                         logical = blkoff << blkbits;
1271                                         phys = blkphy << blkbits;
1272                                         size = n << blkbits;
1273                                 }
1274                         } else {
1275                                 /* Start a new extent */
1276                                 flags = FIEMAP_EXTENT_MERGED;
1277                                 logical = blkoff << blkbits;
1278                                 phys = blkphy << blkbits;
1279                                 size = n << blkbits;
1280                         }
1281                         blkoff += n;
1282                 }
1283                 cond_resched();
1284         } while (true);
1285
1286         /* If ret is 1 then we just hit the end of the extent array */
1287         if (ret == 1)
1288                 ret = 0;
1289
1290         inode_unlock(inode);
1291         return ret;
1292 }