fs/nilfs2/inode.c

   1 /*
   2  * inode.c - NILFS inode operations.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * Written by Ryusuke Konishi.
  17  *
  18  */
  19
  20 #include <linux/buffer_head.h>
  21 #include <linux/gfp.h>
  22 #include <linux/mpage.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/writeback.h>
  25 #include <linux/uio.h>
  26 #include "nilfs.h"
  27 #include "btnode.h"
  28 #include "segment.h"
  29 #include "page.h"
  30 #include "mdt.h"
  31 #include "cpfile.h"
  32 #include "ifile.h"
  33
  34 /**
  35  * struct nilfs_iget_args - arguments used during comparison between inodes
  36  * @ino: inode number
  37  * @cno: checkpoint number
  38  * @root: pointer on NILFS root object (mounted checkpoint)
  39  * @for_gc: inode for GC flag
  40  * @for_btnc: inode for B-tree node cache flag
  41  * @for_shadow: inode for shadowed page cache flag
  42  */
  43 struct nilfs_iget_args {
  44         u64 ino;
  45         __u64 cno;
  46         struct nilfs_root *root;
  47         bool for_gc;
  48         bool for_btnc;
  49         bool for_shadow;
  50 };
  51
  52 static int nilfs_iget_test(struct inode *inode, void *opaque);
  53
  54 void nilfs_inode_add_blocks(struct inode *inode, int n)
  55 {
  56         struct nilfs_root *root = NILFS_I(inode)->i_root;
  57
  58         inode_add_bytes(inode, i_blocksize(inode) * n);
  59         if (root)
  60                 atomic64_add(n, &root->blocks_count);
  61 }
  62
  63 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  64 {
  65         struct nilfs_root *root = NILFS_I(inode)->i_root;
  66
  67         inode_sub_bytes(inode, i_blocksize(inode) * n);
  68         if (root)
  69                 atomic64_sub(n, &root->blocks_count);
  70 }
  71
  72 /**
  73  * nilfs_get_block() - get a file block on the filesystem (callback function)
  74  * @inode - inode struct of the target file
  75  * @blkoff - file block number
  76  * @bh_result - buffer head to be mapped on
  77  * @create - indicate whether allocating the block or not when it has not
  78  *      been allocated yet.
  79  *
  80  * This function does not issue actual read request of the specified data
  81  * block. It is done by VFS.
  82  */
  83 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  84                     struct buffer_head *bh_result, int create)
  85 {
  86         struct nilfs_inode_info *ii = NILFS_I(inode);
  87         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  88         __u64 blknum = 0;
  89         int err = 0, ret;
  90         unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
  91
  92         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  93         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  94         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  95         if (ret >= 0) { /* found */
  96                 map_bh(bh_result, inode->i_sb, blknum);
  97                 if (ret > 0)
  98                         bh_result->b_size = (ret << inode->i_blkbits);
  99                 goto out;
 100         }
 101         /* data block was not found */
 102         if (ret == -ENOENT && create) {
 103                 struct nilfs_transaction_info ti;
 104
 105                 bh_result->b_blocknr = 0;
 106                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 107                 if (unlikely(err))
 108                         goto out;
 109                 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 110                                         (unsigned long)bh_result);
 111                 if (unlikely(err != 0)) {
 112                         if (err == -EEXIST) {
 113                                 /*
 114                                  * The get_block() function could be called
 115                                  * from multiple callers for an inode.
 116                                  * However, the page having this block must
 117                                  * be locked in this case.
 118                                  */
 119                                 nilfs_msg(inode->i_sb, KERN_WARNING,
 120                                           "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
 121                                           __func__, inode->i_ino,
 122                                           (unsigned long long)blkoff);
 123                                 err = 0;
 124                         }
 125                         nilfs_transaction_abort(inode->i_sb);
 126                         goto out;
 127                 }
 128                 nilfs_mark_inode_dirty_sync(inode);
 129                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 130                 /* Error handling should be detailed */
 131                 set_buffer_new(bh_result);
 132                 set_buffer_delay(bh_result);
 133                 map_bh(bh_result, inode->i_sb, 0);
 134                 /* Disk block number must be changed to proper value */
 135
 136         } else if (ret == -ENOENT) {
 137                 /*
 138                  * not found is not error (e.g. hole); must return without
 139                  * the mapped state flag.
 140                  */
 141                 ;
 142         } else {
 143                 err = ret;
 144         }
 145
 146  out:
 147         return err;
 148 }
 149
 150 /**
 151  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 152  * address_space_operations.
 153  * @file - file struct of the file to be read
 154  * @page - the page to be read
 155  */
 156 static int nilfs_readpage(struct file *file, struct page *page)
 157 {
 158         return mpage_readpage(page, nilfs_get_block);
 159 }
 160
 161 /**
 162  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 163  * address_space_operations.
 164  * @file - file struct of the file to be read
 165  * @mapping - address_space struct used for reading multiple pages
 166  * @pages - the pages to be read
 167  * @nr_pages - number of pages to be read
 168  */
 169 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 170                            struct list_head *pages, unsigned int nr_pages)
 171 {
 172         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 173 }
 174
 175 static int nilfs_writepages(struct address_space *mapping,
 176                             struct writeback_control *wbc)
 177 {
 178         struct inode *inode = mapping->host;
 179         int err = 0;
 180
 181         if (sb_rdonly(inode->i_sb)) {
 182                 nilfs_clear_dirty_pages(mapping, false);
 183                 return -EROFS;
 184         }
 185
 186         if (wbc->sync_mode == WB_SYNC_ALL)
 187                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 188                                                     wbc->range_start,
 189                                                     wbc->range_end);
 190         return err;
 191 }
 192
 193 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 194 {
 195         struct inode *inode = page->mapping->host;
 196         int err;
 197
 198         if (sb_rdonly(inode->i_sb)) {
 199                 /*
 200                  * It means that filesystem was remounted in read-only
 201                  * mode because of error or metadata corruption. But we
 202                  * have dirty pages that try to be flushed in background.
 203                  * So, here we simply discard this dirty page.
 204                  */
 205                 nilfs_clear_dirty_page(page, false);
 206                 unlock_page(page);
 207                 return -EROFS;
 208         }
 209
 210         redirty_page_for_writepage(wbc, page);
 211         unlock_page(page);
 212
 213         if (wbc->sync_mode == WB_SYNC_ALL) {
 214                 err = nilfs_construct_segment(inode->i_sb);
 215                 if (unlikely(err))
 216                         return err;
 217         } else if (wbc->for_reclaim)
 218                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 219
 220         return 0;
 221 }
 222
 223 static int nilfs_set_page_dirty(struct page *page)
 224 {
 225         struct inode *inode = page->mapping->host;
 226         int ret = __set_page_dirty_nobuffers(page);
 227
 228         if (page_has_buffers(page)) {
 229                 unsigned int nr_dirty = 0;
 230                 struct buffer_head *bh, *head;
 231
 232                 /*
 233                  * This page is locked by callers, and no other thread
 234                  * concurrently marks its buffers dirty since they are
 235                  * only dirtied through routines in fs/buffer.c in
 236                  * which call sites of mark_buffer_dirty are protected
 237                  * by page lock.
 238                  */
 239                 bh = head = page_buffers(page);
 240                 do {
 241                         /* Do not mark hole blocks dirty */
 242                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 243                                 continue;
 244
 245                         set_buffer_dirty(bh);
 246                         nr_dirty++;
 247                 } while (bh = bh->b_this_page, bh != head);
 248
 249                 if (nr_dirty)
 250                         nilfs_set_file_dirty(inode, nr_dirty);
 251         } else if (ret) {
 252                 unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 253
 254                 nilfs_set_file_dirty(inode, nr_dirty);
 255         }
 256         return ret;
 257 }
 258
 259 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 260 {
 261         struct inode *inode = mapping->host;
 262
 263         if (to > inode->i_size) {
 264                 truncate_pagecache(inode, inode->i_size);
 265                 nilfs_truncate(inode);
 266         }
 267 }
 268
 269 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 270                              loff_t pos, unsigned len, unsigned flags,
 271                              struct page **pagep, void **fsdata)
 272
 273 {
 274         struct inode *inode = mapping->host;
 275         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 276
 277         if (unlikely(err))
 278                 return err;
 279
 280         err = block_write_begin(mapping, pos, len, flags, pagep,
 281                                 nilfs_get_block);
 282         if (unlikely(err)) {
 283                 nilfs_write_failed(mapping, pos + len);
 284                 nilfs_transaction_abort(inode->i_sb);
 285         }
 286         return err;
 287 }
 288
 289 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 290                            loff_t pos, unsigned len, unsigned copied,
 291                            struct page *page, void *fsdata)
 292 {
 293         struct inode *inode = mapping->host;
 294         unsigned int start = pos & (PAGE_SIZE - 1);
 295         unsigned int nr_dirty;
 296         int err;
 297
 298         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 299                                                   start + copied);
 300         copied = generic_write_end(file, mapping, pos, len, copied, page,
 301                                    fsdata);
 302         nilfs_set_file_dirty(inode, nr_dirty);
 303         err = nilfs_transaction_commit(inode->i_sb);
 304         return err ? : copied;
 305 }
 306
 307 static ssize_t
 308 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 309 {
 310         struct inode *inode = file_inode(iocb->ki_filp);
 311
 312         if (iov_iter_rw(iter) == WRITE)
 313                 return 0;
 314
 315         /* Needs synchronization with the cleaner */
 316         return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 317 }
 318
 319 const struct address_space_operations nilfs_aops = {
 320         .writepage              = nilfs_writepage,
 321         .readpage               = nilfs_readpage,
 322         .writepages             = nilfs_writepages,
 323         .set_page_dirty         = nilfs_set_page_dirty,
 324         .readpages              = nilfs_readpages,
 325         .write_begin            = nilfs_write_begin,
 326         .write_end              = nilfs_write_end,
 327         /* .releasepage         = nilfs_releasepage, */
 328         .invalidatepage         = block_invalidatepage,
 329         .direct_IO              = nilfs_direct_IO,
 330         .is_partially_uptodate  = block_is_partially_uptodate,
 331 };
 332
 333 static int nilfs_insert_inode_locked(struct inode *inode,
 334                                      struct nilfs_root *root,
 335                                      unsigned long ino)
 336 {
 337         struct nilfs_iget_args args = {
 338                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 339                 .for_btnc = false, .for_shadow = false
 340         };
 341
 342         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 343 }
 344
 345 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 346 {
 347         struct super_block *sb = dir->i_sb;
 348         struct the_nilfs *nilfs = sb->s_fs_info;
 349         struct inode *inode;
 350         struct nilfs_inode_info *ii;
 351         struct nilfs_root *root;
 352         struct buffer_head *bh;
 353         int err = -ENOMEM;
 354         ino_t ino;
 355
 356         inode = new_inode(sb);
 357         if (unlikely(!inode))
 358                 goto failed;
 359
 360         mapping_set_gfp_mask(inode->i_mapping,
 361                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 362
 363         root = NILFS_I(dir)->i_root;
 364         ii = NILFS_I(inode);
 365         ii->i_state = BIT(NILFS_I_NEW);
 366         ii->i_root = root;
 367
 368         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 369         if (unlikely(err))
 370                 goto failed_ifile_create_inode;
 371         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 372
 373         if (unlikely(ino < NILFS_USER_INO)) {
 374                 nilfs_msg(sb, KERN_WARNING,
 375                           "inode bitmap is inconsistent for reserved inodes");
 376                 do {
 377                         brelse(bh);
 378                         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 379                         if (unlikely(err))
 380                                 goto failed_ifile_create_inode;
 381                 } while (ino < NILFS_USER_INO);
 382
 383                 nilfs_msg(sb, KERN_INFO,
 384                           "repaired inode bitmap for reserved inodes");
 385         }
 386         ii->i_bh = bh;
 387
 388         atomic64_inc(&root->inodes_count);
 389         inode_init_owner(inode, dir, mode);
 390         inode->i_ino = ino;
 391         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 392
 393         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 394                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 395                 if (err < 0)
 396                         goto failed_after_creation;
 397
 398                 set_bit(NILFS_I_BMAP, &ii->i_state);
 399                 /* No lock is needed; iget() ensures it. */
 400         }
 401
 402         ii->i_flags = nilfs_mask_flags(
 403                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 404
 405         /* ii->i_file_acl = 0; */
 406         /* ii->i_dir_acl = 0; */
 407         ii->i_dir_start_lookup = 0;
 408         nilfs_set_inode_flags(inode);
 409         spin_lock(&nilfs->ns_next_gen_lock);
 410         inode->i_generation = nilfs->ns_next_generation++;
 411         spin_unlock(&nilfs->ns_next_gen_lock);
 412         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 413                 err = -EIO;
 414                 goto failed_after_creation;
 415         }
 416
 417         err = nilfs_init_acl(inode, dir);
 418         if (unlikely(err))
 419                 /*
 420                  * Never occur.  When supporting nilfs_init_acl(),
 421                  * proper cancellation of above jobs should be considered.
 422                  */
 423                 goto failed_after_creation;
 424
 425         return inode;
 426
 427  failed_after_creation:
 428         clear_nlink(inode);
 429         unlock_new_inode(inode);
 430         iput(inode);  /*
 431                        * raw_inode will be deleted through
 432                        * nilfs_evict_inode().
 433                        */
 434         goto failed;
 435
 436  failed_ifile_create_inode:
 437         make_bad_inode(inode);
 438         iput(inode);
 439  failed:
 440         return ERR_PTR(err);
 441 }
 442
 443 void nilfs_set_inode_flags(struct inode *inode)
 444 {
 445         unsigned int flags = NILFS_I(inode)->i_flags;
 446         unsigned int new_fl = 0;
 447
 448         if (flags & FS_SYNC_FL)
 449                 new_fl |= S_SYNC;
 450         if (flags & FS_APPEND_FL)
 451                 new_fl |= S_APPEND;
 452         if (flags & FS_IMMUTABLE_FL)
 453                 new_fl |= S_IMMUTABLE;
 454         if (flags & FS_NOATIME_FL)
 455                 new_fl |= S_NOATIME;
 456         if (flags & FS_DIRSYNC_FL)
 457                 new_fl |= S_DIRSYNC;
 458         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 459                         S_NOATIME | S_DIRSYNC);
 460 }
 461
 462 int nilfs_read_inode_common(struct inode *inode,
 463                             struct nilfs_inode *raw_inode)
 464 {
 465         struct nilfs_inode_info *ii = NILFS_I(inode);
 466         int err;
 467
 468         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 469         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 470         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 471         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 472         inode->i_size = le64_to_cpu(raw_inode->i_size);
 473         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 474         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 475         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 476         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 477         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 478         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 479         if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
 480                 return -EIO; /* this inode is for metadata and corrupted */
 481         if (inode->i_nlink == 0)
 482                 return -ESTALE; /* this inode is deleted */
 483
 484         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 485         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 486 #if 0
 487         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 488         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 489                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 490 #endif
 491         ii->i_dir_start_lookup = 0;
 492         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 493
 494         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 495             S_ISLNK(inode->i_mode)) {
 496                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 497                 if (err < 0)
 498                         return err;
 499                 set_bit(NILFS_I_BMAP, &ii->i_state);
 500                 /* No lock is needed; iget() ensures it. */
 501         }
 502         return 0;
 503 }
 504
 505 static int __nilfs_read_inode(struct super_block *sb,
 506                               struct nilfs_root *root, unsigned long ino,
 507                               struct inode *inode)
 508 {
 509         struct the_nilfs *nilfs = sb->s_fs_info;
 510         struct buffer_head *bh;
 511         struct nilfs_inode *raw_inode;
 512         int err;
 513
 514         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 515         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 516         if (unlikely(err))
 517                 goto bad_inode;
 518
 519         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 520
 521         err = nilfs_read_inode_common(inode, raw_inode);
 522         if (err)
 523                 goto failed_unmap;
 524
 525         if (S_ISREG(inode->i_mode)) {
 526                 inode->i_op = &nilfs_file_inode_operations;
 527                 inode->i_fop = &nilfs_file_operations;
 528                 inode->i_mapping->a_ops = &nilfs_aops;
 529         } else if (S_ISDIR(inode->i_mode)) {
 530                 inode->i_op = &nilfs_dir_inode_operations;
 531                 inode->i_fop = &nilfs_dir_operations;
 532                 inode->i_mapping->a_ops = &nilfs_aops;
 533         } else if (S_ISLNK(inode->i_mode)) {
 534                 inode->i_op = &nilfs_symlink_inode_operations;
 535                 inode_nohighmem(inode);
 536                 inode->i_mapping->a_ops = &nilfs_aops;
 537         } else {
 538                 inode->i_op = &nilfs_special_inode_operations;
 539                 init_special_inode(
 540                         inode, inode->i_mode,
 541                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 542         }
 543         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 544         brelse(bh);
 545         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 546         nilfs_set_inode_flags(inode);
 547         mapping_set_gfp_mask(inode->i_mapping,
 548                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 549         return 0;
 550
 551  failed_unmap:
 552         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 553         brelse(bh);
 554
 555  bad_inode:
 556         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 557         return err;
 558 }
 559
 560 static int nilfs_iget_test(struct inode *inode, void *opaque)
 561 {
 562         struct nilfs_iget_args *args = opaque;
 563         struct nilfs_inode_info *ii;
 564
 565         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 566                 return 0;
 567
 568         ii = NILFS_I(inode);
 569         if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
 570                 if (!args->for_btnc)
 571                         return 0;
 572         } else if (args->for_btnc) {
 573                 return 0;
 574         }
 575         if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
 576                 if (!args->for_shadow)
 577                         return 0;
 578         } else if (args->for_shadow) {
 579                 return 0;
 580         }
 581
 582         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 583                 return !args->for_gc;
 584
 585         return args->for_gc && args->cno == ii->i_cno;
 586 }
 587
 588 static int nilfs_iget_set(struct inode *inode, void *opaque)
 589 {
 590         struct nilfs_iget_args *args = opaque;
 591
 592         inode->i_ino = args->ino;
 593         NILFS_I(inode)->i_cno = args->cno;
 594         NILFS_I(inode)->i_root = args->root;
 595         if (args->root && args->ino == NILFS_ROOT_INO)
 596                 nilfs_get_root(args->root);
 597
 598         if (args->for_gc)
 599                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 600         if (args->for_btnc)
 601                 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
 602         if (args->for_shadow)
 603                 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
 604         return 0;
 605 }
 606
 607 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 608                             unsigned long ino)
 609 {
 610         struct nilfs_iget_args args = {
 611                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 612                 .for_btnc = false, .for_shadow = false
 613         };
 614
 615         return ilookup5(sb, ino, nilfs_iget_test, &args);
 616 }
 617
 618 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 619                                 unsigned long ino)
 620 {
 621         struct nilfs_iget_args args = {
 622                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 623                 .for_btnc = false, .for_shadow = false
 624         };
 625
 626         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 627 }
 628
 629 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 630                          unsigned long ino)
 631 {
 632         struct inode *inode;
 633         int err;
 634
 635         inode = nilfs_iget_locked(sb, root, ino);
 636         if (unlikely(!inode))
 637                 return ERR_PTR(-ENOMEM);
 638         if (!(inode->i_state & I_NEW))
 639                 return inode;
 640
 641         err = __nilfs_read_inode(sb, root, ino, inode);
 642         if (unlikely(err)) {
 643                 iget_failed(inode);
 644                 return ERR_PTR(err);
 645         }
 646         unlock_new_inode(inode);
 647         return inode;
 648 }
 649
 650 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 651                                 __u64 cno)
 652 {
 653         struct nilfs_iget_args args = {
 654                 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
 655                 .for_btnc = false, .for_shadow = false
 656         };
 657         struct inode *inode;
 658         int err;
 659
 660         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 661         if (unlikely(!inode))
 662                 return ERR_PTR(-ENOMEM);
 663         if (!(inode->i_state & I_NEW))
 664                 return inode;
 665
 666         err = nilfs_init_gcinode(inode);
 667         if (unlikely(err)) {
 668                 iget_failed(inode);
 669                 return ERR_PTR(err);
 670         }
 671         unlock_new_inode(inode);
 672         return inode;
 673 }
 674
 675 /**
 676  * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 677  * @inode: inode object
 678  *
 679  * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 680  * or does nothing if the inode already has it.  This function allocates
 681  * an additional inode to maintain page cache of B-tree nodes one-on-one.
 682  *
 683  * Return Value: On success, 0 is returned. On errors, one of the following
 684  * negative error code is returned.
 685  *
 686  * %-ENOMEM - Insufficient memory available.
 687  */
 688 int nilfs_attach_btree_node_cache(struct inode *inode)
 689 {
 690         struct nilfs_inode_info *ii = NILFS_I(inode);
 691         struct inode *btnc_inode;
 692         struct nilfs_iget_args args;
 693
 694         if (ii->i_assoc_inode)
 695                 return 0;
 696
 697         args.ino = inode->i_ino;
 698         args.root = ii->i_root;
 699         args.cno = ii->i_cno;
 700         args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
 701         args.for_btnc = true;
 702         args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
 703
 704         btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 705                                   nilfs_iget_set, &args);
 706         if (unlikely(!btnc_inode))
 707                 return -ENOMEM;
 708         if (btnc_inode->i_state & I_NEW) {
 709                 nilfs_init_btnc_inode(btnc_inode);
 710                 unlock_new_inode(btnc_inode);
 711         }
 712         NILFS_I(btnc_inode)->i_assoc_inode = inode;
 713         NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 714         ii->i_assoc_inode = btnc_inode;
 715
 716         return 0;
 717 }
 718
 719 /**
 720  * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 721  * @inode: inode object
 722  *
 723  * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 724  * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 725  */
 726 void nilfs_detach_btree_node_cache(struct inode *inode)
 727 {
 728         struct nilfs_inode_info *ii = NILFS_I(inode);
 729         struct inode *btnc_inode = ii->i_assoc_inode;
 730
 731         if (btnc_inode) {
 732                 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 733                 ii->i_assoc_inode = NULL;
 734                 iput(btnc_inode);
 735         }
 736 }
 737
 738 /**
 739  * nilfs_iget_for_shadow - obtain inode for shadow mapping
 740  * @inode: inode object that uses shadow mapping
 741  *
 742  * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 743  * caches for shadow mapping.  The page cache for data pages is set up
 744  * in one inode and the one for b-tree node pages is set up in the
 745  * other inode, which is attached to the former inode.
 746  *
 747  * Return Value: On success, a pointer to the inode for data pages is
 748  * returned. On errors, one of the following negative error code is returned
 749  * in a pointer type.
 750  *
 751  * %-ENOMEM - Insufficient memory available.
 752  */
 753 struct inode *nilfs_iget_for_shadow(struct inode *inode)
 754 {
 755         struct nilfs_iget_args args = {
 756                 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
 757                 .for_btnc = false, .for_shadow = true
 758         };
 759         struct inode *s_inode;
 760         int err;
 761
 762         s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 763                                nilfs_iget_set, &args);
 764         if (unlikely(!s_inode))
 765                 return ERR_PTR(-ENOMEM);
 766         if (!(s_inode->i_state & I_NEW))
 767                 return inode;
 768
 769         NILFS_I(s_inode)->i_flags = 0;
 770         memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 771         mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 772
 773         err = nilfs_attach_btree_node_cache(s_inode);
 774         if (unlikely(err)) {
 775                 iget_failed(s_inode);
 776                 return ERR_PTR(err);
 777         }
 778         unlock_new_inode(s_inode);
 779         return s_inode;
 780 }
 781
 782 void nilfs_write_inode_common(struct inode *inode,
 783                               struct nilfs_inode *raw_inode, int has_bmap)
 784 {
 785         struct nilfs_inode_info *ii = NILFS_I(inode);
 786
 787         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 788         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 789         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 790         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 791         raw_inode->i_size = cpu_to_le64(inode->i_size);
 792         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 793         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 794         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 795         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 796         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 797
 798         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 799         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 800
 801         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 802                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 803
 804                 /* zero-fill unused portion in the case of super root block */
 805                 raw_inode->i_xattr = 0;
 806                 raw_inode->i_pad = 0;
 807                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 808                        nilfs->ns_inode_size - sizeof(*raw_inode));
 809         }
 810
 811         if (has_bmap)
 812                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 813         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 814                 raw_inode->i_device_code =
 815                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 816         /*
 817          * When extending inode, nilfs->ns_inode_size should be checked
 818          * for substitutions of appended fields.
 819          */
 820 }
 821
 822 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 823 {
 824         ino_t ino = inode->i_ino;
 825         struct nilfs_inode_info *ii = NILFS_I(inode);
 826         struct inode *ifile = ii->i_root->ifile;
 827         struct nilfs_inode *raw_inode;
 828
 829         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 830
 831         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 832                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 833         if (flags & I_DIRTY_DATASYNC)
 834                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 835
 836         nilfs_write_inode_common(inode, raw_inode, 0);
 837                 /*
 838                  * XXX: call with has_bmap = 0 is a workaround to avoid
 839                  * deadlock of bmap.  This delays update of i_bmap to just
 840                  * before writing.
 841                  */
 842
 843         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 844 }
 845
 846 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 847
 848 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 849                                 unsigned long from)
 850 {
 851         __u64 b;
 852         int ret;
 853
 854         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 855                 return;
 856 repeat:
 857         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 858         if (ret == -ENOENT)
 859                 return;
 860         else if (ret < 0)
 861                 goto failed;
 862
 863         if (b < from)
 864                 return;
 865
 866         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 867         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 868         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 869         if (!ret || (ret == -ENOMEM &&
 870                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 871                 goto repeat;
 872
 873 failed:
 874         nilfs_msg(ii->vfs_inode.i_sb, KERN_WARNING,
 875                   "error %d truncating bmap (ino=%lu)", ret,
 876                   ii->vfs_inode.i_ino);
 877 }
 878
 879 void nilfs_truncate(struct inode *inode)
 880 {
 881         unsigned long blkoff;
 882         unsigned int blocksize;
 883         struct nilfs_transaction_info ti;
 884         struct super_block *sb = inode->i_sb;
 885         struct nilfs_inode_info *ii = NILFS_I(inode);
 886
 887         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 888                 return;
 889         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 890                 return;
 891
 892         blocksize = sb->s_blocksize;
 893         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 894         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 895
 896         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 897
 898         nilfs_truncate_bmap(ii, blkoff);
 899
 900         inode->i_mtime = inode->i_ctime = current_time(inode);
 901         if (IS_SYNC(inode))
 902                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 903
 904         nilfs_mark_inode_dirty(inode);
 905         nilfs_set_file_dirty(inode, 0);
 906         nilfs_transaction_commit(sb);
 907         /*
 908          * May construct a logical segment and may fail in sync mode.
 909          * But truncate has no return value.
 910          */
 911 }
 912
 913 static void nilfs_clear_inode(struct inode *inode)
 914 {
 915         struct nilfs_inode_info *ii = NILFS_I(inode);
 916
 917         /*
 918          * Free resources allocated in nilfs_read_inode(), here.
 919          */
 920         BUG_ON(!list_empty(&ii->i_dirty));
 921         brelse(ii->i_bh);
 922         ii->i_bh = NULL;
 923
 924         if (nilfs_is_metadata_file_inode(inode))
 925                 nilfs_mdt_clear(inode);
 926
 927         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 928                 nilfs_bmap_clear(ii->i_bmap);
 929
 930         if (!test_bit(NILFS_I_BTNC, &ii->i_state))
 931                 nilfs_detach_btree_node_cache(inode);
 932
 933         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 934                 nilfs_put_root(ii->i_root);
 935 }
 936
 937 void nilfs_evict_inode(struct inode *inode)
 938 {
 939         struct nilfs_transaction_info ti;
 940         struct super_block *sb = inode->i_sb;
 941         struct nilfs_inode_info *ii = NILFS_I(inode);
 942         struct the_nilfs *nilfs;
 943         int ret;
 944
 945         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 946                 truncate_inode_pages_final(&inode->i_data);
 947                 clear_inode(inode);
 948                 nilfs_clear_inode(inode);
 949                 return;
 950         }
 951         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 952
 953         truncate_inode_pages_final(&inode->i_data);
 954
 955         nilfs = sb->s_fs_info;
 956         if (unlikely(sb_rdonly(sb) || !nilfs->ns_writer)) {
 957                 /*
 958                  * If this inode is about to be disposed after the file system
 959                  * has been degraded to read-only due to file system corruption
 960                  * or after the writer has been detached, do not make any
 961                  * changes that cause writes, just clear it.
 962                  * Do this check after read-locking ns_segctor_sem by
 963                  * nilfs_transaction_begin() in order to avoid a race with
 964                  * the writer detach operation.
 965                  */
 966                 clear_inode(inode);
 967                 nilfs_clear_inode(inode);
 968                 nilfs_transaction_abort(sb);
 969                 return;
 970         }
 971
 972         /* TODO: some of the following operations may fail.  */
 973         nilfs_truncate_bmap(ii, 0);
 974         nilfs_mark_inode_dirty(inode);
 975         clear_inode(inode);
 976
 977         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 978         if (!ret)
 979                 atomic64_dec(&ii->i_root->inodes_count);
 980
 981         nilfs_clear_inode(inode);
 982
 983         if (IS_SYNC(inode))
 984                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 985         nilfs_transaction_commit(sb);
 986         /*
 987          * May construct a logical segment and may fail in sync mode.
 988          * But delete_inode has no return value.
 989          */
 990 }
 991
 992 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 993 {
 994         struct nilfs_transaction_info ti;
 995         struct inode *inode = d_inode(dentry);
 996         struct super_block *sb = inode->i_sb;
 997         int err;
 998
 999         err = setattr_prepare(dentry, iattr);
1000         if (err)
1001                 return err;
1002
1003         err = nilfs_transaction_begin(sb, &ti, 0);
1004         if (unlikely(err))
1005                 return err;
1006
1007         if ((iattr->ia_valid & ATTR_SIZE) &&
1008             iattr->ia_size != i_size_read(inode)) {
1009                 inode_dio_wait(inode);
1010                 truncate_setsize(inode, iattr->ia_size);
1011                 nilfs_truncate(inode);
1012         }
1013
1014         setattr_copy(inode, iattr);
1015         mark_inode_dirty(inode);
1016
1017         if (iattr->ia_valid & ATTR_MODE) {
1018                 err = nilfs_acl_chmod(inode);
1019                 if (unlikely(err))
1020                         goto out_err;
1021         }
1022
1023         return nilfs_transaction_commit(sb);
1024
1025 out_err:
1026         nilfs_transaction_abort(sb);
1027         return err;
1028 }
1029
1030 int nilfs_permission(struct inode *inode, int mask)
1031 {
1032         struct nilfs_root *root = NILFS_I(inode)->i_root;
1033
1034         if ((mask & MAY_WRITE) && root &&
1035             root->cno != NILFS_CPTREE_CURRENT_CNO)
1036                 return -EROFS; /* snapshot is not writable */
1037
1038         return generic_permission(inode, mask);
1039 }
1040
1041 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1042 {
1043         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1044         struct nilfs_inode_info *ii = NILFS_I(inode);
1045         int err;
1046
1047         spin_lock(&nilfs->ns_inode_lock);
1048         if (ii->i_bh == NULL || unlikely(!buffer_uptodate(ii->i_bh))) {
1049                 spin_unlock(&nilfs->ns_inode_lock);
1050                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1051                                                   inode->i_ino, pbh);
1052                 if (unlikely(err))
1053                         return err;
1054                 spin_lock(&nilfs->ns_inode_lock);
1055                 if (ii->i_bh == NULL)
1056                         ii->i_bh = *pbh;
1057                 else if (unlikely(!buffer_uptodate(ii->i_bh))) {
1058                         __brelse(ii->i_bh);
1059                         ii->i_bh = *pbh;
1060                 } else {
1061                         brelse(*pbh);
1062                         *pbh = ii->i_bh;
1063                 }
1064         } else
1065                 *pbh = ii->i_bh;
1066
1067         get_bh(*pbh);
1068         spin_unlock(&nilfs->ns_inode_lock);
1069         return 0;
1070 }
1071
1072 int nilfs_inode_dirty(struct inode *inode)
1073 {
1074         struct nilfs_inode_info *ii = NILFS_I(inode);
1075         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1076         int ret = 0;
1077
1078         if (!list_empty(&ii->i_dirty)) {
1079                 spin_lock(&nilfs->ns_inode_lock);
1080                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1081                         test_bit(NILFS_I_BUSY, &ii->i_state);
1082                 spin_unlock(&nilfs->ns_inode_lock);
1083         }
1084         return ret;
1085 }
1086
1087 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1088 {
1089         struct nilfs_inode_info *ii = NILFS_I(inode);
1090         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1091
1092         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1093
1094         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1095                 return 0;
1096
1097         spin_lock(&nilfs->ns_inode_lock);
1098         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1099             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1100                 /*
1101                  * Because this routine may race with nilfs_dispose_list(),
1102                  * we have to check NILFS_I_QUEUED here, too.
1103                  */
1104                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1105                         /*
1106                          * This will happen when somebody is freeing
1107                          * this inode.
1108                          */
1109                         nilfs_msg(inode->i_sb, KERN_WARNING,
1110                                   "cannot set file dirty (ino=%lu): the file is being freed",
1111                                   inode->i_ino);
1112                         spin_unlock(&nilfs->ns_inode_lock);
1113                         return -EINVAL; /*
1114                                          * NILFS_I_DIRTY may remain for
1115                                          * freeing inode.
1116                                          */
1117                 }
1118                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1119                 set_bit(NILFS_I_QUEUED, &ii->i_state);
1120         }
1121         spin_unlock(&nilfs->ns_inode_lock);
1122         return 0;
1123 }
1124
1125 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1126 {
1127         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1128         struct buffer_head *ibh;
1129         int err;
1130
1131         /*
1132          * Do not dirty inodes after the log writer has been detached
1133          * and its nilfs_root struct has been freed.
1134          */
1135         if (unlikely(nilfs_purging(nilfs)))
1136                 return 0;
1137
1138         err = nilfs_load_inode_block(inode, &ibh);
1139         if (unlikely(err)) {
1140                 nilfs_msg(inode->i_sb, KERN_WARNING,
1141                           "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1142                           inode->i_ino, err);
1143                 return err;
1144         }
1145         nilfs_update_inode(inode, ibh, flags);
1146         mark_buffer_dirty(ibh);
1147         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1148         brelse(ibh);
1149         return 0;
1150 }
1151
1152 /**
1153  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1154  * @inode: inode of the file to be registered.
1155  *
1156  * nilfs_dirty_inode() loads a inode block containing the specified
1157  * @inode and copies data from a nilfs_inode to a corresponding inode
1158  * entry in the inode block. This operation is excluded from the segment
1159  * construction. This function can be called both as a single operation
1160  * and as a part of indivisible file operations.
1161  */
1162 void nilfs_dirty_inode(struct inode *inode, int flags)
1163 {
1164         struct nilfs_transaction_info ti;
1165         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1166
1167         if (is_bad_inode(inode)) {
1168                 nilfs_msg(inode->i_sb, KERN_WARNING,
1169                           "tried to mark bad_inode dirty. ignored.");
1170                 dump_stack();
1171                 return;
1172         }
1173         if (mdi) {
1174                 nilfs_mdt_mark_dirty(inode);
1175                 return;
1176         }
1177         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1178         __nilfs_mark_inode_dirty(inode, flags);
1179         nilfs_transaction_commit(inode->i_sb); /* never fails */
1180 }
1181
1182 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1183                  __u64 start, __u64 len)
1184 {
1185         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1186         __u64 logical = 0, phys = 0, size = 0;
1187         __u32 flags = 0;
1188         loff_t isize;
1189         sector_t blkoff, end_blkoff;
1190         sector_t delalloc_blkoff;
1191         unsigned long delalloc_blklen;
1192         unsigned int blkbits = inode->i_blkbits;
1193         int ret, n;
1194
1195         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1196         if (ret)
1197                 return ret;
1198
1199         inode_lock(inode);
1200
1201         isize = i_size_read(inode);
1202
1203         blkoff = start >> blkbits;
1204         end_blkoff = (start + len - 1) >> blkbits;
1205
1206         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1207                                                         &delalloc_blkoff);
1208
1209         do {
1210                 __u64 blkphy;
1211                 unsigned int maxblocks;
1212
1213                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1214                         if (size) {
1215                                 /* End of the current extent */
1216                                 ret = fiemap_fill_next_extent(
1217                                         fieinfo, logical, phys, size, flags);
1218                                 if (ret)
1219                                         break;
1220                         }
1221                         if (blkoff > end_blkoff)
1222                                 break;
1223
1224                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1225                         logical = blkoff << blkbits;
1226                         phys = 0;
1227                         size = delalloc_blklen << blkbits;
1228
1229                         blkoff = delalloc_blkoff + delalloc_blklen;
1230                         delalloc_blklen = nilfs_find_uncommitted_extent(
1231                                 inode, blkoff, &delalloc_blkoff);
1232                         continue;
1233                 }
1234
1235                 /*
1236                  * Limit the number of blocks that we look up so as
1237                  * not to get into the next delayed allocation extent.
1238                  */
1239                 maxblocks = INT_MAX;
1240                 if (delalloc_blklen)
1241                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1242                                           maxblocks);
1243                 blkphy = 0;
1244
1245                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1246                 n = nilfs_bmap_lookup_contig(
1247                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1248                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1249
1250                 if (n < 0) {
1251                         int past_eof;
1252
1253                         if (unlikely(n != -ENOENT))
1254                                 break; /* error */
1255
1256                         /* HOLE */
1257                         blkoff++;
1258                         past_eof = ((blkoff << blkbits) >= isize);
1259
1260                         if (size) {
1261                                 /* End of the current extent */
1262
1263                                 if (past_eof)
1264                                         flags |= FIEMAP_EXTENT_LAST;
1265
1266                                 ret = fiemap_fill_next_extent(
1267                                         fieinfo, logical, phys, size, flags);
1268                                 if (ret)
1269                                         break;
1270                                 size = 0;
1271                         }
1272                         if (blkoff > end_blkoff || past_eof)
1273                                 break;
1274                 } else {
1275                         if (size) {
1276                                 if (phys && blkphy << blkbits == phys + size) {
1277                                         /* The current extent goes on */
1278                                         size += n << blkbits;
1279                                 } else {
1280                                         /* Terminate the current extent */
1281                                         ret = fiemap_fill_next_extent(
1282                                                 fieinfo, logical, phys, size,
1283                                                 flags);
1284                                         if (ret || blkoff > end_blkoff)
1285                                                 break;
1286
1287                                         /* Start another extent */
1288                                         flags = FIEMAP_EXTENT_MERGED;
1289                                         logical = blkoff << blkbits;
1290                                         phys = blkphy << blkbits;
1291                                         size = n << blkbits;
1292                                 }
1293                         } else {
1294                                 /* Start a new extent */
1295                                 flags = FIEMAP_EXTENT_MERGED;
1296                                 logical = blkoff << blkbits;
1297                                 phys = blkphy << blkbits;
1298                                 size = n << blkbits;
1299                         }
1300                         blkoff += n;
1301                 }
1302                 cond_resched();
1303         } while (true);
1304
1305         /* If ret is 1 then we just hit the end of the extent array */
1306         if (ret == 1)
1307                 ret = 0;
1308
1309         inode_unlock(inode);
1310         return ret;
1311 }