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 folio *folio = page_folio(page);
 179         struct inode *inode = folio->mapping->host;
 180         int err;
 181
 182         if (sb_rdonly(inode->i_sb)) {
 183                 /*
 184                  * It means that filesystem was remounted in read-only
 185                  * mode because of error or metadata corruption. But we
 186                  * have dirty pages that try to be flushed in background.
 187                  * So, here we simply discard this dirty page.
 188                  */
 189                 nilfs_clear_folio_dirty(folio, false);
 190                 folio_unlock(folio);
 191                 return -EROFS;
 192         }
 193
 194         folio_redirty_for_writepage(wbc, folio);
 195         folio_unlock(folio);
 196
 197         if (wbc->sync_mode == WB_SYNC_ALL) {
 198                 err = nilfs_construct_segment(inode->i_sb);
 199                 if (unlikely(err))
 200                         return err;
 201         } else if (wbc->for_reclaim)
 202                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 203
 204         return 0;
 205 }
 206
 207 static bool nilfs_dirty_folio(struct address_space *mapping,
 208                 struct folio *folio)
 209 {
 210         struct inode *inode = mapping->host;
 211         struct buffer_head *head;
 212         unsigned int nr_dirty = 0;
 213         bool ret = filemap_dirty_folio(mapping, folio);
 214
 215         /*
 216          * The page may not be locked, eg if called from try_to_unmap_one()
 217          */
 218         spin_lock(&mapping->i_private_lock);
 219         head = folio_buffers(folio);
 220         if (head) {
 221                 struct buffer_head *bh = head;
 222
 223                 do {
 224                         /* Do not mark hole blocks dirty */
 225                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 226                                 continue;
 227
 228                         set_buffer_dirty(bh);
 229                         nr_dirty++;
 230                 } while (bh = bh->b_this_page, bh != head);
 231         } else if (ret) {
 232                 nr_dirty = 1 << (folio_shift(folio) - inode->i_blkbits);
 233         }
 234         spin_unlock(&mapping->i_private_lock);
 235
 236         if (nr_dirty)
 237                 nilfs_set_file_dirty(inode, nr_dirty);
 238         return ret;
 239 }
 240
 241 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 242 {
 243         struct inode *inode = mapping->host;
 244
 245         if (to > inode->i_size) {
 246                 truncate_pagecache(inode, inode->i_size);
 247                 nilfs_truncate(inode);
 248         }
 249 }
 250
 251 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 252                              loff_t pos, unsigned len,
 253                              struct page **pagep, void **fsdata)
 254
 255 {
 256         struct inode *inode = mapping->host;
 257         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 258
 259         if (unlikely(err))
 260                 return err;
 261
 262         err = block_write_begin(mapping, pos, len, pagep, 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         .read_folio             = nilfs_read_folio,
 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         .invalidate_folio       = block_invalidate_folio,
 309         .direct_IO              = nilfs_direct_IO,
 310         .is_partially_uptodate  = block_is_partially_uptodate,
 311 };
 312
 313 static int nilfs_insert_inode_locked(struct inode *inode,
 314                                      struct nilfs_root *root,
 315                                      unsigned long ino)
 316 {
 317         struct nilfs_iget_args args = {
 318                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 319                 .for_btnc = false, .for_shadow = false
 320         };
 321
 322         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 323 }
 324
 325 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 326 {
 327         struct super_block *sb = dir->i_sb;
 328         struct the_nilfs *nilfs = sb->s_fs_info;
 329         struct inode *inode;
 330         struct nilfs_inode_info *ii;
 331         struct nilfs_root *root;
 332         struct buffer_head *bh;
 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, &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         if (unlikely(ino < NILFS_USER_INO)) {
 354                 nilfs_warn(sb,
 355                            "inode bitmap is inconsistent for reserved inodes");
 356                 do {
 357                         brelse(bh);
 358                         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 359                         if (unlikely(err))
 360                                 goto failed_ifile_create_inode;
 361                 } while (ino < NILFS_USER_INO);
 362
 363                 nilfs_info(sb, "repaired inode bitmap for reserved inodes");
 364         }
 365         ii->i_bh = bh;
 366
 367         atomic64_inc(&root->inodes_count);
 368         inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
 369         inode->i_ino = ino;
 370         simple_inode_init_ts(inode);
 371
 372         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 373                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 374                 if (err < 0)
 375                         goto failed_after_creation;
 376
 377                 set_bit(NILFS_I_BMAP, &ii->i_state);
 378                 /* No lock is needed; iget() ensures it. */
 379         }
 380
 381         ii->i_flags = nilfs_mask_flags(
 382                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 383
 384         /* ii->i_file_acl = 0; */
 385         /* ii->i_dir_acl = 0; */
 386         ii->i_dir_start_lookup = 0;
 387         nilfs_set_inode_flags(inode);
 388         spin_lock(&nilfs->ns_next_gen_lock);
 389         inode->i_generation = nilfs->ns_next_generation++;
 390         spin_unlock(&nilfs->ns_next_gen_lock);
 391         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 392                 err = -EIO;
 393                 goto failed_after_creation;
 394         }
 395
 396         err = nilfs_init_acl(inode, dir);
 397         if (unlikely(err))
 398                 /*
 399                  * Never occur.  When supporting nilfs_init_acl(),
 400                  * proper cancellation of above jobs should be considered.
 401                  */
 402                 goto failed_after_creation;
 403
 404         return inode;
 405
 406  failed_after_creation:
 407         clear_nlink(inode);
 408         if (inode->i_state & I_NEW)
 409                 unlock_new_inode(inode);
 410         iput(inode);  /*
 411                        * raw_inode will be deleted through
 412                        * nilfs_evict_inode().
 413                        */
 414         goto failed;
 415
 416  failed_ifile_create_inode:
 417         make_bad_inode(inode);
 418         iput(inode);
 419  failed:
 420         return ERR_PTR(err);
 421 }
 422
 423 void nilfs_set_inode_flags(struct inode *inode)
 424 {
 425         unsigned int flags = NILFS_I(inode)->i_flags;
 426         unsigned int new_fl = 0;
 427
 428         if (flags & FS_SYNC_FL)
 429                 new_fl |= S_SYNC;
 430         if (flags & FS_APPEND_FL)
 431                 new_fl |= S_APPEND;
 432         if (flags & FS_IMMUTABLE_FL)
 433                 new_fl |= S_IMMUTABLE;
 434         if (flags & FS_NOATIME_FL)
 435                 new_fl |= S_NOATIME;
 436         if (flags & FS_DIRSYNC_FL)
 437                 new_fl |= S_DIRSYNC;
 438         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 439                         S_NOATIME | S_DIRSYNC);
 440 }
 441
 442 int nilfs_read_inode_common(struct inode *inode,
 443                             struct nilfs_inode *raw_inode)
 444 {
 445         struct nilfs_inode_info *ii = NILFS_I(inode);
 446         int err;
 447
 448         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 449         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 450         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 451         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 452         inode->i_size = le64_to_cpu(raw_inode->i_size);
 453         inode_set_atime(inode, le64_to_cpu(raw_inode->i_mtime),
 454                         le32_to_cpu(raw_inode->i_mtime_nsec));
 455         inode_set_ctime(inode, le64_to_cpu(raw_inode->i_ctime),
 456                         le32_to_cpu(raw_inode->i_ctime_nsec));
 457         inode_set_mtime(inode, le64_to_cpu(raw_inode->i_mtime),
 458                         le32_to_cpu(raw_inode->i_mtime_nsec));
 459         if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
 460                 return -EIO; /* this inode is for metadata and corrupted */
 461         if (inode->i_nlink == 0)
 462                 return -ESTALE; /* this inode is deleted */
 463
 464         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 465         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 466 #if 0
 467         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 468         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 469                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 470 #endif
 471         ii->i_dir_start_lookup = 0;
 472         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 473
 474         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 475             S_ISLNK(inode->i_mode)) {
 476                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 477                 if (err < 0)
 478                         return err;
 479                 set_bit(NILFS_I_BMAP, &ii->i_state);
 480                 /* No lock is needed; iget() ensures it. */
 481         }
 482         return 0;
 483 }
 484
 485 static int __nilfs_read_inode(struct super_block *sb,
 486                               struct nilfs_root *root, unsigned long ino,
 487                               struct inode *inode)
 488 {
 489         struct the_nilfs *nilfs = sb->s_fs_info;
 490         struct buffer_head *bh;
 491         struct nilfs_inode *raw_inode;
 492         int err;
 493
 494         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 495         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 496         if (unlikely(err))
 497                 goto bad_inode;
 498
 499         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 500
 501         err = nilfs_read_inode_common(inode, raw_inode);
 502         if (err)
 503                 goto failed_unmap;
 504
 505         if (S_ISREG(inode->i_mode)) {
 506                 inode->i_op = &nilfs_file_inode_operations;
 507                 inode->i_fop = &nilfs_file_operations;
 508                 inode->i_mapping->a_ops = &nilfs_aops;
 509         } else if (S_ISDIR(inode->i_mode)) {
 510                 inode->i_op = &nilfs_dir_inode_operations;
 511                 inode->i_fop = &nilfs_dir_operations;
 512                 inode->i_mapping->a_ops = &nilfs_aops;
 513         } else if (S_ISLNK(inode->i_mode)) {
 514                 inode->i_op = &nilfs_symlink_inode_operations;
 515                 inode_nohighmem(inode);
 516                 inode->i_mapping->a_ops = &nilfs_aops;
 517         } else {
 518                 inode->i_op = &nilfs_special_inode_operations;
 519                 init_special_inode(
 520                         inode, inode->i_mode,
 521                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 522         }
 523         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 524         brelse(bh);
 525         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 526         nilfs_set_inode_flags(inode);
 527         mapping_set_gfp_mask(inode->i_mapping,
 528                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 529         return 0;
 530
 531  failed_unmap:
 532         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 533         brelse(bh);
 534
 535  bad_inode:
 536         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 537         return err;
 538 }
 539
 540 static int nilfs_iget_test(struct inode *inode, void *opaque)
 541 {
 542         struct nilfs_iget_args *args = opaque;
 543         struct nilfs_inode_info *ii;
 544
 545         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 546                 return 0;
 547
 548         ii = NILFS_I(inode);
 549         if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
 550                 if (!args->for_btnc)
 551                         return 0;
 552         } else if (args->for_btnc) {
 553                 return 0;
 554         }
 555         if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
 556                 if (!args->for_shadow)
 557                         return 0;
 558         } else if (args->for_shadow) {
 559                 return 0;
 560         }
 561
 562         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 563                 return !args->for_gc;
 564
 565         return args->for_gc && args->cno == ii->i_cno;
 566 }
 567
 568 static int nilfs_iget_set(struct inode *inode, void *opaque)
 569 {
 570         struct nilfs_iget_args *args = opaque;
 571
 572         inode->i_ino = args->ino;
 573         NILFS_I(inode)->i_cno = args->cno;
 574         NILFS_I(inode)->i_root = args->root;
 575         if (args->root && args->ino == NILFS_ROOT_INO)
 576                 nilfs_get_root(args->root);
 577
 578         if (args->for_gc)
 579                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 580         if (args->for_btnc)
 581                 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
 582         if (args->for_shadow)
 583                 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
 584         return 0;
 585 }
 586
 587 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 588                             unsigned long ino)
 589 {
 590         struct nilfs_iget_args args = {
 591                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 592                 .for_btnc = false, .for_shadow = false
 593         };
 594
 595         return ilookup5(sb, ino, nilfs_iget_test, &args);
 596 }
 597
 598 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 599                                 unsigned long ino)
 600 {
 601         struct nilfs_iget_args args = {
 602                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 603                 .for_btnc = false, .for_shadow = false
 604         };
 605
 606         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 607 }
 608
 609 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 610                          unsigned long ino)
 611 {
 612         struct inode *inode;
 613         int err;
 614
 615         inode = nilfs_iget_locked(sb, root, ino);
 616         if (unlikely(!inode))
 617                 return ERR_PTR(-ENOMEM);
 618         if (!(inode->i_state & I_NEW))
 619                 return inode;
 620
 621         err = __nilfs_read_inode(sb, root, ino, inode);
 622         if (unlikely(err)) {
 623                 iget_failed(inode);
 624                 return ERR_PTR(err);
 625         }
 626         unlock_new_inode(inode);
 627         return inode;
 628 }
 629
 630 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 631                                 __u64 cno)
 632 {
 633         struct nilfs_iget_args args = {
 634                 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
 635                 .for_btnc = false, .for_shadow = false
 636         };
 637         struct inode *inode;
 638         int err;
 639
 640         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 641         if (unlikely(!inode))
 642                 return ERR_PTR(-ENOMEM);
 643         if (!(inode->i_state & I_NEW))
 644                 return inode;
 645
 646         err = nilfs_init_gcinode(inode);
 647         if (unlikely(err)) {
 648                 iget_failed(inode);
 649                 return ERR_PTR(err);
 650         }
 651         unlock_new_inode(inode);
 652         return inode;
 653 }
 654
 655 /**
 656  * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 657  * @inode: inode object
 658  *
 659  * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 660  * or does nothing if the inode already has it.  This function allocates
 661  * an additional inode to maintain page cache of B-tree nodes one-on-one.
 662  *
 663  * Return Value: On success, 0 is returned. On errors, one of the following
 664  * negative error code is returned.
 665  *
 666  * %-ENOMEM - Insufficient memory available.
 667  */
 668 int nilfs_attach_btree_node_cache(struct inode *inode)
 669 {
 670         struct nilfs_inode_info *ii = NILFS_I(inode);
 671         struct inode *btnc_inode;
 672         struct nilfs_iget_args args;
 673
 674         if (ii->i_assoc_inode)
 675                 return 0;
 676
 677         args.ino = inode->i_ino;
 678         args.root = ii->i_root;
 679         args.cno = ii->i_cno;
 680         args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
 681         args.for_btnc = true;
 682         args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
 683
 684         btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 685                                   nilfs_iget_set, &args);
 686         if (unlikely(!btnc_inode))
 687                 return -ENOMEM;
 688         if (btnc_inode->i_state & I_NEW) {
 689                 nilfs_init_btnc_inode(btnc_inode);
 690                 unlock_new_inode(btnc_inode);
 691         }
 692         NILFS_I(btnc_inode)->i_assoc_inode = inode;
 693         NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 694         ii->i_assoc_inode = btnc_inode;
 695
 696         return 0;
 697 }
 698
 699 /**
 700  * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 701  * @inode: inode object
 702  *
 703  * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 704  * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 705  */
 706 void nilfs_detach_btree_node_cache(struct inode *inode)
 707 {
 708         struct nilfs_inode_info *ii = NILFS_I(inode);
 709         struct inode *btnc_inode = ii->i_assoc_inode;
 710
 711         if (btnc_inode) {
 712                 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 713                 ii->i_assoc_inode = NULL;
 714                 iput(btnc_inode);
 715         }
 716 }
 717
 718 /**
 719  * nilfs_iget_for_shadow - obtain inode for shadow mapping
 720  * @inode: inode object that uses shadow mapping
 721  *
 722  * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 723  * caches for shadow mapping.  The page cache for data pages is set up
 724  * in one inode and the one for b-tree node pages is set up in the
 725  * other inode, which is attached to the former inode.
 726  *
 727  * Return Value: On success, a pointer to the inode for data pages is
 728  * returned. On errors, one of the following negative error code is returned
 729  * in a pointer type.
 730  *
 731  * %-ENOMEM - Insufficient memory available.
 732  */
 733 struct inode *nilfs_iget_for_shadow(struct inode *inode)
 734 {
 735         struct nilfs_iget_args args = {
 736                 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
 737                 .for_btnc = false, .for_shadow = true
 738         };
 739         struct inode *s_inode;
 740         int err;
 741
 742         s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 743                                nilfs_iget_set, &args);
 744         if (unlikely(!s_inode))
 745                 return ERR_PTR(-ENOMEM);
 746         if (!(s_inode->i_state & I_NEW))
 747                 return inode;
 748
 749         NILFS_I(s_inode)->i_flags = 0;
 750         memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 751         mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 752
 753         err = nilfs_attach_btree_node_cache(s_inode);
 754         if (unlikely(err)) {
 755                 iget_failed(s_inode);
 756                 return ERR_PTR(err);
 757         }
 758         unlock_new_inode(s_inode);
 759         return s_inode;
 760 }
 761
 762 void nilfs_write_inode_common(struct inode *inode,
 763                               struct nilfs_inode *raw_inode, int has_bmap)
 764 {
 765         struct nilfs_inode_info *ii = NILFS_I(inode);
 766
 767         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 768         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 769         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 770         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 771         raw_inode->i_size = cpu_to_le64(inode->i_size);
 772         raw_inode->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
 773         raw_inode->i_mtime = cpu_to_le64(inode_get_mtime_sec(inode));
 774         raw_inode->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
 775         raw_inode->i_mtime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
 776         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 777
 778         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 779         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 780
 781         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 782                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 783
 784                 /* zero-fill unused portion in the case of super root block */
 785                 raw_inode->i_xattr = 0;
 786                 raw_inode->i_pad = 0;
 787                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 788                        nilfs->ns_inode_size - sizeof(*raw_inode));
 789         }
 790
 791         if (has_bmap)
 792                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 793         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 794                 raw_inode->i_device_code =
 795                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 796         /*
 797          * When extending inode, nilfs->ns_inode_size should be checked
 798          * for substitutions of appended fields.
 799          */
 800 }
 801
 802 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 803 {
 804         ino_t ino = inode->i_ino;
 805         struct nilfs_inode_info *ii = NILFS_I(inode);
 806         struct inode *ifile = ii->i_root->ifile;
 807         struct nilfs_inode *raw_inode;
 808
 809         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 810
 811         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 812                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 813         if (flags & I_DIRTY_DATASYNC)
 814                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 815
 816         nilfs_write_inode_common(inode, raw_inode, 0);
 817                 /*
 818                  * XXX: call with has_bmap = 0 is a workaround to avoid
 819                  * deadlock of bmap.  This delays update of i_bmap to just
 820                  * before writing.
 821                  */
 822
 823         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 824 }
 825
 826 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 827
 828 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 829                                 unsigned long from)
 830 {
 831         __u64 b;
 832         int ret;
 833
 834         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 835                 return;
 836 repeat:
 837         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 838         if (ret == -ENOENT)
 839                 return;
 840         else if (ret < 0)
 841                 goto failed;
 842
 843         if (b < from)
 844                 return;
 845
 846         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 847         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 848         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 849         if (!ret || (ret == -ENOMEM &&
 850                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 851                 goto repeat;
 852
 853 failed:
 854         nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
 855                    ret, ii->vfs_inode.i_ino);
 856 }
 857
 858 void nilfs_truncate(struct inode *inode)
 859 {
 860         unsigned long blkoff;
 861         unsigned int blocksize;
 862         struct nilfs_transaction_info ti;
 863         struct super_block *sb = inode->i_sb;
 864         struct nilfs_inode_info *ii = NILFS_I(inode);
 865
 866         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 867                 return;
 868         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 869                 return;
 870
 871         blocksize = sb->s_blocksize;
 872         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 873         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 874
 875         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 876
 877         nilfs_truncate_bmap(ii, blkoff);
 878
 879         inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
 880         if (IS_SYNC(inode))
 881                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 882
 883         nilfs_mark_inode_dirty(inode);
 884         nilfs_set_file_dirty(inode, 0);
 885         nilfs_transaction_commit(sb);
 886         /*
 887          * May construct a logical segment and may fail in sync mode.
 888          * But truncate has no return value.
 889          */
 890 }
 891
 892 static void nilfs_clear_inode(struct inode *inode)
 893 {
 894         struct nilfs_inode_info *ii = NILFS_I(inode);
 895
 896         /*
 897          * Free resources allocated in nilfs_read_inode(), here.
 898          */
 899         BUG_ON(!list_empty(&ii->i_dirty));
 900         brelse(ii->i_bh);
 901         ii->i_bh = NULL;
 902
 903         if (nilfs_is_metadata_file_inode(inode))
 904                 nilfs_mdt_clear(inode);
 905
 906         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 907                 nilfs_bmap_clear(ii->i_bmap);
 908
 909         if (!test_bit(NILFS_I_BTNC, &ii->i_state))
 910                 nilfs_detach_btree_node_cache(inode);
 911
 912         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 913                 nilfs_put_root(ii->i_root);
 914 }
 915
 916 void nilfs_evict_inode(struct inode *inode)
 917 {
 918         struct nilfs_transaction_info ti;
 919         struct super_block *sb = inode->i_sb;
 920         struct nilfs_inode_info *ii = NILFS_I(inode);
 921         struct the_nilfs *nilfs;
 922         int ret;
 923
 924         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 925                 truncate_inode_pages_final(&inode->i_data);
 926                 clear_inode(inode);
 927                 nilfs_clear_inode(inode);
 928                 return;
 929         }
 930         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 931
 932         truncate_inode_pages_final(&inode->i_data);
 933
 934         nilfs = sb->s_fs_info;
 935         if (unlikely(sb_rdonly(sb) || !nilfs->ns_writer)) {
 936                 /*
 937                  * If this inode is about to be disposed after the file system
 938                  * has been degraded to read-only due to file system corruption
 939                  * or after the writer has been detached, do not make any
 940                  * changes that cause writes, just clear it.
 941                  * Do this check after read-locking ns_segctor_sem by
 942                  * nilfs_transaction_begin() in order to avoid a race with
 943                  * the writer detach operation.
 944                  */
 945                 clear_inode(inode);
 946                 nilfs_clear_inode(inode);
 947                 nilfs_transaction_abort(sb);
 948                 return;
 949         }
 950
 951         /* TODO: some of the following operations may fail.  */
 952         nilfs_truncate_bmap(ii, 0);
 953         nilfs_mark_inode_dirty(inode);
 954         clear_inode(inode);
 955
 956         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 957         if (!ret)
 958                 atomic64_dec(&ii->i_root->inodes_count);
 959
 960         nilfs_clear_inode(inode);
 961
 962         if (IS_SYNC(inode))
 963                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 964         nilfs_transaction_commit(sb);
 965         /*
 966          * May construct a logical segment and may fail in sync mode.
 967          * But delete_inode has no return value.
 968          */
 969 }
 970
 971 int nilfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 972                   struct iattr *iattr)
 973 {
 974         struct nilfs_transaction_info ti;
 975         struct inode *inode = d_inode(dentry);
 976         struct super_block *sb = inode->i_sb;
 977         int err;
 978
 979         err = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
 980         if (err)
 981                 return err;
 982
 983         err = nilfs_transaction_begin(sb, &ti, 0);
 984         if (unlikely(err))
 985                 return err;
 986
 987         if ((iattr->ia_valid & ATTR_SIZE) &&
 988             iattr->ia_size != i_size_read(inode)) {
 989                 inode_dio_wait(inode);
 990                 truncate_setsize(inode, iattr->ia_size);
 991                 nilfs_truncate(inode);
 992         }
 993
 994         setattr_copy(&nop_mnt_idmap, inode, iattr);
 995         mark_inode_dirty(inode);
 996
 997         if (iattr->ia_valid & ATTR_MODE) {
 998                 err = nilfs_acl_chmod(inode);
 999                 if (unlikely(err))
1000                         goto out_err;
1001         }
1002
1003         return nilfs_transaction_commit(sb);
1004
1005 out_err:
1006         nilfs_transaction_abort(sb);
1007         return err;
1008 }
1009
1010 int nilfs_permission(struct mnt_idmap *idmap, struct inode *inode,
1011                      int mask)
1012 {
1013         struct nilfs_root *root = NILFS_I(inode)->i_root;
1014
1015         if ((mask & MAY_WRITE) && root &&
1016             root->cno != NILFS_CPTREE_CURRENT_CNO)
1017                 return -EROFS; /* snapshot is not writable */
1018
1019         return generic_permission(&nop_mnt_idmap, inode, mask);
1020 }
1021
1022 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1023 {
1024         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1025         struct nilfs_inode_info *ii = NILFS_I(inode);
1026         int err;
1027
1028         spin_lock(&nilfs->ns_inode_lock);
1029         if (ii->i_bh == NULL || unlikely(!buffer_uptodate(ii->i_bh))) {
1030                 spin_unlock(&nilfs->ns_inode_lock);
1031                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1032                                                   inode->i_ino, pbh);
1033                 if (unlikely(err))
1034                         return err;
1035                 spin_lock(&nilfs->ns_inode_lock);
1036                 if (ii->i_bh == NULL)
1037                         ii->i_bh = *pbh;
1038                 else if (unlikely(!buffer_uptodate(ii->i_bh))) {
1039                         __brelse(ii->i_bh);
1040                         ii->i_bh = *pbh;
1041                 } else {
1042                         brelse(*pbh);
1043                         *pbh = ii->i_bh;
1044                 }
1045         } else
1046                 *pbh = ii->i_bh;
1047
1048         get_bh(*pbh);
1049         spin_unlock(&nilfs->ns_inode_lock);
1050         return 0;
1051 }
1052
1053 int nilfs_inode_dirty(struct inode *inode)
1054 {
1055         struct nilfs_inode_info *ii = NILFS_I(inode);
1056         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1057         int ret = 0;
1058
1059         if (!list_empty(&ii->i_dirty)) {
1060                 spin_lock(&nilfs->ns_inode_lock);
1061                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1062                         test_bit(NILFS_I_BUSY, &ii->i_state);
1063                 spin_unlock(&nilfs->ns_inode_lock);
1064         }
1065         return ret;
1066 }
1067
1068 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1069 {
1070         struct nilfs_inode_info *ii = NILFS_I(inode);
1071         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1072
1073         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1074
1075         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1076                 return 0;
1077
1078         spin_lock(&nilfs->ns_inode_lock);
1079         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1080             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1081                 /*
1082                  * Because this routine may race with nilfs_dispose_list(),
1083                  * we have to check NILFS_I_QUEUED here, too.
1084                  */
1085                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1086                         /*
1087                          * This will happen when somebody is freeing
1088                          * this inode.
1089                          */
1090                         nilfs_warn(inode->i_sb,
1091                                    "cannot set file dirty (ino=%lu): the file is being freed",
1092                                    inode->i_ino);
1093                         spin_unlock(&nilfs->ns_inode_lock);
1094                         return -EINVAL; /*
1095                                          * NILFS_I_DIRTY may remain for
1096                                          * freeing inode.
1097                                          */
1098                 }
1099                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1100                 set_bit(NILFS_I_QUEUED, &ii->i_state);
1101         }
1102         spin_unlock(&nilfs->ns_inode_lock);
1103         return 0;
1104 }
1105
1106 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1107 {
1108         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1109         struct buffer_head *ibh;
1110         int err;
1111
1112         /*
1113          * Do not dirty inodes after the log writer has been detached
1114          * and its nilfs_root struct has been freed.
1115          */
1116         if (unlikely(nilfs_purging(nilfs)))
1117                 return 0;
1118
1119         err = nilfs_load_inode_block(inode, &ibh);
1120         if (unlikely(err)) {
1121                 nilfs_warn(inode->i_sb,
1122                            "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1123                            inode->i_ino, err);
1124                 return err;
1125         }
1126         nilfs_update_inode(inode, ibh, flags);
1127         mark_buffer_dirty(ibh);
1128         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1129         brelse(ibh);
1130         return 0;
1131 }
1132
1133 /**
1134  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1135  * @inode: inode of the file to be registered.
1136  * @flags: flags to determine the dirty state of the inode
1137  *
1138  * nilfs_dirty_inode() loads a inode block containing the specified
1139  * @inode and copies data from a nilfs_inode to a corresponding inode
1140  * entry in the inode block. This operation is excluded from the segment
1141  * construction. This function can be called both as a single operation
1142  * and as a part of indivisible file operations.
1143  */
1144 void nilfs_dirty_inode(struct inode *inode, int flags)
1145 {
1146         struct nilfs_transaction_info ti;
1147         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1148
1149         if (is_bad_inode(inode)) {
1150                 nilfs_warn(inode->i_sb,
1151                            "tried to mark bad_inode dirty. ignored.");
1152                 dump_stack();
1153                 return;
1154         }
1155         if (mdi) {
1156                 nilfs_mdt_mark_dirty(inode);
1157                 return;
1158         }
1159         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1160         __nilfs_mark_inode_dirty(inode, flags);
1161         nilfs_transaction_commit(inode->i_sb); /* never fails */
1162 }
1163
1164 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1165                  __u64 start, __u64 len)
1166 {
1167         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1168         __u64 logical = 0, phys = 0, size = 0;
1169         __u32 flags = 0;
1170         loff_t isize;
1171         sector_t blkoff, end_blkoff;
1172         sector_t delalloc_blkoff;
1173         unsigned long delalloc_blklen;
1174         unsigned int blkbits = inode->i_blkbits;
1175         int ret, n;
1176
1177         ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1178         if (ret)
1179                 return ret;
1180
1181         inode_lock(inode);
1182
1183         isize = i_size_read(inode);
1184
1185         blkoff = start >> blkbits;
1186         end_blkoff = (start + len - 1) >> blkbits;
1187
1188         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1189                                                         &delalloc_blkoff);
1190
1191         do {
1192                 __u64 blkphy;
1193                 unsigned int maxblocks;
1194
1195                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1196                         if (size) {
1197                                 /* End of the current extent */
1198                                 ret = fiemap_fill_next_extent(
1199                                         fieinfo, logical, phys, size, flags);
1200                                 if (ret)
1201                                         break;
1202                         }
1203                         if (blkoff > end_blkoff)
1204                                 break;
1205
1206                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1207                         logical = blkoff << blkbits;
1208                         phys = 0;
1209                         size = delalloc_blklen << blkbits;
1210
1211                         blkoff = delalloc_blkoff + delalloc_blklen;
1212                         delalloc_blklen = nilfs_find_uncommitted_extent(
1213                                 inode, blkoff, &delalloc_blkoff);
1214                         continue;
1215                 }
1216
1217                 /*
1218                  * Limit the number of blocks that we look up so as
1219                  * not to get into the next delayed allocation extent.
1220                  */
1221                 maxblocks = INT_MAX;
1222                 if (delalloc_blklen)
1223                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1224                                           maxblocks);
1225                 blkphy = 0;
1226
1227                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1228                 n = nilfs_bmap_lookup_contig(
1229                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1230                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1231
1232                 if (n < 0) {
1233                         int past_eof;
1234
1235                         if (unlikely(n != -ENOENT))
1236                                 break; /* error */
1237
1238                         /* HOLE */
1239                         blkoff++;
1240                         past_eof = ((blkoff << blkbits) >= isize);
1241
1242                         if (size) {
1243                                 /* End of the current extent */
1244
1245                                 if (past_eof)
1246                                         flags |= FIEMAP_EXTENT_LAST;
1247
1248                                 ret = fiemap_fill_next_extent(
1249                                         fieinfo, logical, phys, size, flags);
1250                                 if (ret)
1251                                         break;
1252                                 size = 0;
1253                         }
1254                         if (blkoff > end_blkoff || past_eof)
1255                                 break;
1256                 } else {
1257                         if (size) {
1258                                 if (phys && blkphy << blkbits == phys + size) {
1259                                         /* The current extent goes on */
1260                                         size += n << blkbits;
1261                                 } else {
1262                                         /* Terminate the current extent */
1263                                         ret = fiemap_fill_next_extent(
1264                                                 fieinfo, logical, phys, size,
1265                                                 flags);
1266                                         if (ret || blkoff > end_blkoff)
1267                                                 break;
1268
1269                                         /* Start another extent */
1270                                         flags = FIEMAP_EXTENT_MERGED;
1271                                         logical = blkoff << blkbits;
1272                                         phys = blkphy << blkbits;
1273                                         size = n << blkbits;
1274                                 }
1275                         } else {
1276                                 /* Start a new extent */
1277                                 flags = FIEMAP_EXTENT_MERGED;
1278                                 logical = blkoff << blkbits;
1279                                 phys = blkphy << blkbits;
1280                                 size = n << blkbits;
1281                         }
1282                         blkoff += n;
1283                 }
1284                 cond_resched();
1285         } while (true);
1286
1287         /* If ret is 1 then we just hit the end of the extent array */
1288         if (ret == 1)
1289                 ret = 0;
1290
1291         inode_unlock(inode);
1292         return ret;
1293 }