fs/nilfs2/inode.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * inode.c - 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 "nilfs.h"
  18 #include "btnode.h"
  19 #include "segment.h"
  20 #include "page.h"
  21 #include "mdt.h"
  22 #include "cpfile.h"
  23 #include "ifile.h"
  24
  25 /**
  26  * struct nilfs_iget_args - arguments used during comparison between inodes
  27  * @ino: inode number
  28  * @cno: checkpoint number
  29  * @root: pointer on NILFS root object (mounted checkpoint)
  30  * @for_gc: inode for GC flag
  31  * @for_btnc: inode for B-tree node cache flag
  32  * @for_shadow: inode for shadowed page cache flag
  33  */
  34 struct nilfs_iget_args {
  35         u64 ino;
  36         __u64 cno;
  37         struct nilfs_root *root;
  38         bool for_gc;
  39         bool for_btnc;
  40         bool for_shadow;
  41 };
  42
  43 static int nilfs_iget_test(struct inode *inode, void *opaque);
  44
  45 void nilfs_inode_add_blocks(struct inode *inode, int n)
  46 {
  47         struct nilfs_root *root = NILFS_I(inode)->i_root;
  48
  49         inode_add_bytes(inode, i_blocksize(inode) * n);
  50         if (root)
  51                 atomic64_add(n, &root->blocks_count);
  52 }
  53
  54 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  55 {
  56         struct nilfs_root *root = NILFS_I(inode)->i_root;
  57
  58         inode_sub_bytes(inode, i_blocksize(inode) * n);
  59         if (root)
  60                 atomic64_sub(n, &root->blocks_count);
  61 }
  62
  63 /**
  64  * nilfs_get_block() - get a file block on the filesystem (callback function)
  65  * @inode - inode struct of the target file
  66  * @blkoff - file block number
  67  * @bh_result - buffer head to be mapped on
  68  * @create - indicate whether allocating the block or not when it has not
  69  *      been allocated yet.
  70  *
  71  * This function does not issue actual read request of the specified data
  72  * block. It is done by VFS.
  73  */
  74 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  75                     struct buffer_head *bh_result, int create)
  76 {
  77         struct nilfs_inode_info *ii = NILFS_I(inode);
  78         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  79         __u64 blknum = 0;
  80         int err = 0, ret;
  81         unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
  82
  83         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  84         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  85         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  86         if (ret >= 0) { /* found */
  87                 map_bh(bh_result, inode->i_sb, blknum);
  88                 if (ret > 0)
  89                         bh_result->b_size = (ret << inode->i_blkbits);
  90                 goto out;
  91         }
  92         /* data block was not found */
  93         if (ret == -ENOENT && create) {
  94                 struct nilfs_transaction_info ti;
  95
  96                 bh_result->b_blocknr = 0;
  97                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  98                 if (unlikely(err))
  99                         goto out;
 100                 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 101                                         (unsigned long)bh_result);
 102                 if (unlikely(err != 0)) {
 103                         if (err == -EEXIST) {
 104                                 /*
 105                                  * The get_block() function could be called
 106                                  * from multiple callers for an inode.
 107                                  * However, the page having this block must
 108                                  * be locked in this case.
 109                                  */
 110                                 nilfs_msg(inode->i_sb, KERN_WARNING,
 111                                           "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
 112                                           __func__, inode->i_ino,
 113                                           (unsigned long long)blkoff);
 114                                 err = 0;
 115                         }
 116                         nilfs_transaction_abort(inode->i_sb);
 117                         goto out;
 118                 }
 119                 nilfs_mark_inode_dirty_sync(inode);
 120                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 121                 /* Error handling should be detailed */
 122                 set_buffer_new(bh_result);
 123                 set_buffer_delay(bh_result);
 124                 map_bh(bh_result, inode->i_sb, 0);
 125                 /* Disk block number must be changed to proper value */
 126
 127         } else if (ret == -ENOENT) {
 128                 /*
 129                  * not found is not error (e.g. hole); must return without
 130                  * the mapped state flag.
 131                  */
 132                 ;
 133         } else {
 134                 err = ret;
 135         }
 136
 137  out:
 138         return err;
 139 }
 140
 141 /**
 142  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 143  * address_space_operations.
 144  * @file - file struct of the file to be read
 145  * @page - the page to be read
 146  */
 147 static int nilfs_readpage(struct file *file, struct page *page)
 148 {
 149         return mpage_readpage(page, nilfs_get_block);
 150 }
 151
 152 /**
 153  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 154  * address_space_operations.
 155  * @file - file struct of the file to be read
 156  * @mapping - address_space struct used for reading multiple pages
 157  * @pages - the pages to be read
 158  * @nr_pages - number of pages to be read
 159  */
 160 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 161                            struct list_head *pages, unsigned int nr_pages)
 162 {
 163         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 164 }
 165
 166 static int nilfs_writepages(struct address_space *mapping,
 167                             struct writeback_control *wbc)
 168 {
 169         struct inode *inode = mapping->host;
 170         int err = 0;
 171
 172         if (sb_rdonly(inode->i_sb)) {
 173                 nilfs_clear_dirty_pages(mapping, false);
 174                 return -EROFS;
 175         }
 176
 177         if (wbc->sync_mode == WB_SYNC_ALL)
 178                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 179                                                     wbc->range_start,
 180                                                     wbc->range_end);
 181         return err;
 182 }
 183
 184 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 185 {
 186         struct inode *inode = page->mapping->host;
 187         int err;
 188
 189         if (sb_rdonly(inode->i_sb)) {
 190                 /*
 191                  * It means that filesystem was remounted in read-only
 192                  * mode because of error or metadata corruption. But we
 193                  * have dirty pages that try to be flushed in background.
 194                  * So, here we simply discard this dirty page.
 195                  */
 196                 nilfs_clear_dirty_page(page, false);
 197                 unlock_page(page);
 198                 return -EROFS;
 199         }
 200
 201         redirty_page_for_writepage(wbc, page);
 202         unlock_page(page);
 203
 204         if (wbc->sync_mode == WB_SYNC_ALL) {
 205                 err = nilfs_construct_segment(inode->i_sb);
 206                 if (unlikely(err))
 207                         return err;
 208         } else if (wbc->for_reclaim)
 209                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 210
 211         return 0;
 212 }
 213
 214 static int nilfs_set_page_dirty(struct page *page)
 215 {
 216         struct inode *inode = page->mapping->host;
 217         int ret = __set_page_dirty_nobuffers(page);
 218
 219         if (page_has_buffers(page)) {
 220                 unsigned int nr_dirty = 0;
 221                 struct buffer_head *bh, *head;
 222
 223                 /*
 224                  * This page is locked by callers, and no other thread
 225                  * concurrently marks its buffers dirty since they are
 226                  * only dirtied through routines in fs/buffer.c in
 227                  * which call sites of mark_buffer_dirty are protected
 228                  * by page lock.
 229                  */
 230                 bh = head = page_buffers(page);
 231                 do {
 232                         /* Do not mark hole blocks dirty */
 233                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 234                                 continue;
 235
 236                         set_buffer_dirty(bh);
 237                         nr_dirty++;
 238                 } while (bh = bh->b_this_page, bh != head);
 239
 240                 if (nr_dirty)
 241                         nilfs_set_file_dirty(inode, nr_dirty);
 242         } else if (ret) {
 243                 unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 244
 245                 nilfs_set_file_dirty(inode, nr_dirty);
 246         }
 247         return ret;
 248 }
 249
 250 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 251 {
 252         struct inode *inode = mapping->host;
 253
 254         if (to > inode->i_size) {
 255                 truncate_pagecache(inode, inode->i_size);
 256                 nilfs_truncate(inode);
 257         }
 258 }
 259
 260 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 261                              loff_t pos, unsigned len, unsigned flags,
 262                              struct page **pagep, void **fsdata)
 263
 264 {
 265         struct inode *inode = mapping->host;
 266         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 267
 268         if (unlikely(err))
 269                 return err;
 270
 271         err = block_write_begin(mapping, pos, len, flags, pagep,
 272                                 nilfs_get_block);
 273         if (unlikely(err)) {
 274                 nilfs_write_failed(mapping, pos + len);
 275                 nilfs_transaction_abort(inode->i_sb);
 276         }
 277         return err;
 278 }
 279
 280 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 281                            loff_t pos, unsigned len, unsigned copied,
 282                            struct page *page, void *fsdata)
 283 {
 284         struct inode *inode = mapping->host;
 285         unsigned int start = pos & (PAGE_SIZE - 1);
 286         unsigned int nr_dirty;
 287         int err;
 288
 289         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 290                                                   start + copied);
 291         copied = generic_write_end(file, mapping, pos, len, copied, page,
 292                                    fsdata);
 293         nilfs_set_file_dirty(inode, nr_dirty);
 294         err = nilfs_transaction_commit(inode->i_sb);
 295         return err ? : copied;
 296 }
 297
 298 static ssize_t
 299 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 300 {
 301         struct inode *inode = file_inode(iocb->ki_filp);
 302
 303         if (iov_iter_rw(iter) == WRITE)
 304                 return 0;
 305
 306         /* Needs synchronization with the cleaner */
 307         return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 308 }
 309
 310 const struct address_space_operations nilfs_aops = {
 311         .writepage              = nilfs_writepage,
 312         .readpage               = nilfs_readpage,
 313         .writepages             = nilfs_writepages,
 314         .set_page_dirty         = nilfs_set_page_dirty,
 315         .readpages              = nilfs_readpages,
 316         .write_begin            = nilfs_write_begin,
 317         .write_end              = nilfs_write_end,
 318         /* .releasepage         = nilfs_releasepage, */
 319         .invalidatepage         = block_invalidatepage,
 320         .direct_IO              = nilfs_direct_IO,
 321         .is_partially_uptodate  = block_is_partially_uptodate,
 322 };
 323
 324 static int nilfs_insert_inode_locked(struct inode *inode,
 325                                      struct nilfs_root *root,
 326                                      unsigned long ino)
 327 {
 328         struct nilfs_iget_args args = {
 329                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 330                 .for_btnc = false, .for_shadow = false
 331         };
 332
 333         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 334 }
 335
 336 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 337 {
 338         struct super_block *sb = dir->i_sb;
 339         struct the_nilfs *nilfs = sb->s_fs_info;
 340         struct inode *inode;
 341         struct nilfs_inode_info *ii;
 342         struct nilfs_root *root;
 343         struct buffer_head *bh;
 344         int err = -ENOMEM;
 345         ino_t ino;
 346
 347         inode = new_inode(sb);
 348         if (unlikely(!inode))
 349                 goto failed;
 350
 351         mapping_set_gfp_mask(inode->i_mapping,
 352                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 353
 354         root = NILFS_I(dir)->i_root;
 355         ii = NILFS_I(inode);
 356         ii->i_state = BIT(NILFS_I_NEW);
 357         ii->i_root = root;
 358
 359         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 360         if (unlikely(err))
 361                 goto failed_ifile_create_inode;
 362         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 363
 364         if (unlikely(ino < NILFS_USER_INO)) {
 365                 nilfs_msg(sb, KERN_WARNING,
 366                           "inode bitmap is inconsistent for reserved inodes");
 367                 do {
 368                         brelse(bh);
 369                         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 370                         if (unlikely(err))
 371                                 goto failed_ifile_create_inode;
 372                 } while (ino < NILFS_USER_INO);
 373
 374                 nilfs_msg(sb, KERN_INFO,
 375                           "repaired inode bitmap for reserved inodes");
 376         }
 377         ii->i_bh = bh;
 378
 379         atomic64_inc(&root->inodes_count);
 380         inode_init_owner(inode, dir, mode);
 381         inode->i_ino = ino;
 382         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 383
 384         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 385                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 386                 if (err < 0)
 387                         goto failed_after_creation;
 388
 389                 set_bit(NILFS_I_BMAP, &ii->i_state);
 390                 /* No lock is needed; iget() ensures it. */
 391         }
 392
 393         ii->i_flags = nilfs_mask_flags(
 394                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 395
 396         /* ii->i_file_acl = 0; */
 397         /* ii->i_dir_acl = 0; */
 398         ii->i_dir_start_lookup = 0;
 399         nilfs_set_inode_flags(inode);
 400         spin_lock(&nilfs->ns_next_gen_lock);
 401         inode->i_generation = nilfs->ns_next_generation++;
 402         spin_unlock(&nilfs->ns_next_gen_lock);
 403         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 404                 err = -EIO;
 405                 goto failed_after_creation;
 406         }
 407
 408         err = nilfs_init_acl(inode, dir);
 409         if (unlikely(err))
 410                 /*
 411                  * Never occur.  When supporting nilfs_init_acl(),
 412                  * proper cancellation of above jobs should be considered.
 413                  */
 414                 goto failed_after_creation;
 415
 416         return inode;
 417
 418  failed_after_creation:
 419         clear_nlink(inode);
 420         unlock_new_inode(inode);
 421         iput(inode);  /*
 422                        * raw_inode will be deleted through
 423                        * nilfs_evict_inode().
 424                        */
 425         goto failed;
 426
 427  failed_ifile_create_inode:
 428         make_bad_inode(inode);
 429         iput(inode);
 430  failed:
 431         return ERR_PTR(err);
 432 }
 433
 434 void nilfs_set_inode_flags(struct inode *inode)
 435 {
 436         unsigned int flags = NILFS_I(inode)->i_flags;
 437         unsigned int new_fl = 0;
 438
 439         if (flags & FS_SYNC_FL)
 440                 new_fl |= S_SYNC;
 441         if (flags & FS_APPEND_FL)
 442                 new_fl |= S_APPEND;
 443         if (flags & FS_IMMUTABLE_FL)
 444                 new_fl |= S_IMMUTABLE;
 445         if (flags & FS_NOATIME_FL)
 446                 new_fl |= S_NOATIME;
 447         if (flags & FS_DIRSYNC_FL)
 448                 new_fl |= S_DIRSYNC;
 449         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 450                         S_NOATIME | S_DIRSYNC);
 451 }
 452
 453 int nilfs_read_inode_common(struct inode *inode,
 454                             struct nilfs_inode *raw_inode)
 455 {
 456         struct nilfs_inode_info *ii = NILFS_I(inode);
 457         int err;
 458
 459         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 460         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 461         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 462         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 463         inode->i_size = le64_to_cpu(raw_inode->i_size);
 464         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 465         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 466         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 467         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 468         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 469         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 470         if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
 471                 return -EIO; /* this inode is for metadata and corrupted */
 472         if (inode->i_nlink == 0)
 473                 return -ESTALE; /* this inode is deleted */
 474
 475         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 476         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 477 #if 0
 478         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 479         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 480                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 481 #endif
 482         ii->i_dir_start_lookup = 0;
 483         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 484
 485         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 486             S_ISLNK(inode->i_mode)) {
 487                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 488                 if (err < 0)
 489                         return err;
 490                 set_bit(NILFS_I_BMAP, &ii->i_state);
 491                 /* No lock is needed; iget() ensures it. */
 492         }
 493         return 0;
 494 }
 495
 496 static int __nilfs_read_inode(struct super_block *sb,
 497                               struct nilfs_root *root, unsigned long ino,
 498                               struct inode *inode)
 499 {
 500         struct the_nilfs *nilfs = sb->s_fs_info;
 501         struct buffer_head *bh;
 502         struct nilfs_inode *raw_inode;
 503         int err;
 504
 505         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 506         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 507         if (unlikely(err))
 508                 goto bad_inode;
 509
 510         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 511
 512         err = nilfs_read_inode_common(inode, raw_inode);
 513         if (err)
 514                 goto failed_unmap;
 515
 516         if (S_ISREG(inode->i_mode)) {
 517                 inode->i_op = &nilfs_file_inode_operations;
 518                 inode->i_fop = &nilfs_file_operations;
 519                 inode->i_mapping->a_ops = &nilfs_aops;
 520         } else if (S_ISDIR(inode->i_mode)) {
 521                 inode->i_op = &nilfs_dir_inode_operations;
 522                 inode->i_fop = &nilfs_dir_operations;
 523                 inode->i_mapping->a_ops = &nilfs_aops;
 524         } else if (S_ISLNK(inode->i_mode)) {
 525                 inode->i_op = &nilfs_symlink_inode_operations;
 526                 inode_nohighmem(inode);
 527                 inode->i_mapping->a_ops = &nilfs_aops;
 528         } else {
 529                 inode->i_op = &nilfs_special_inode_operations;
 530                 init_special_inode(
 531                         inode, inode->i_mode,
 532                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 533         }
 534         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 535         brelse(bh);
 536         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 537         nilfs_set_inode_flags(inode);
 538         mapping_set_gfp_mask(inode->i_mapping,
 539                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 540         return 0;
 541
 542  failed_unmap:
 543         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 544         brelse(bh);
 545
 546  bad_inode:
 547         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 548         return err;
 549 }
 550
 551 static int nilfs_iget_test(struct inode *inode, void *opaque)
 552 {
 553         struct nilfs_iget_args *args = opaque;
 554         struct nilfs_inode_info *ii;
 555
 556         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 557                 return 0;
 558
 559         ii = NILFS_I(inode);
 560         if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
 561                 if (!args->for_btnc)
 562                         return 0;
 563         } else if (args->for_btnc) {
 564                 return 0;
 565         }
 566         if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
 567                 if (!args->for_shadow)
 568                         return 0;
 569         } else if (args->for_shadow) {
 570                 return 0;
 571         }
 572
 573         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 574                 return !args->for_gc;
 575
 576         return args->for_gc && args->cno == ii->i_cno;
 577 }
 578
 579 static int nilfs_iget_set(struct inode *inode, void *opaque)
 580 {
 581         struct nilfs_iget_args *args = opaque;
 582
 583         inode->i_ino = args->ino;
 584         NILFS_I(inode)->i_cno = args->cno;
 585         NILFS_I(inode)->i_root = args->root;
 586         if (args->root && args->ino == NILFS_ROOT_INO)
 587                 nilfs_get_root(args->root);
 588
 589         if (args->for_gc)
 590                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 591         if (args->for_btnc)
 592                 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
 593         if (args->for_shadow)
 594                 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
 595         return 0;
 596 }
 597
 598 struct inode *nilfs_ilookup(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 ilookup5(sb, ino, nilfs_iget_test, &args);
 607 }
 608
 609 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 610                                 unsigned long ino)
 611 {
 612         struct nilfs_iget_args args = {
 613                 .ino = ino, .root = root, .cno = 0, .for_gc = false,
 614                 .for_btnc = false, .for_shadow = false
 615         };
 616
 617         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 618 }
 619
 620 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 621                          unsigned long ino)
 622 {
 623         struct inode *inode;
 624         int err;
 625
 626         inode = nilfs_iget_locked(sb, root, ino);
 627         if (unlikely(!inode))
 628                 return ERR_PTR(-ENOMEM);
 629         if (!(inode->i_state & I_NEW))
 630                 return inode;
 631
 632         err = __nilfs_read_inode(sb, root, ino, inode);
 633         if (unlikely(err)) {
 634                 iget_failed(inode);
 635                 return ERR_PTR(err);
 636         }
 637         unlock_new_inode(inode);
 638         return inode;
 639 }
 640
 641 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 642                                 __u64 cno)
 643 {
 644         struct nilfs_iget_args args = {
 645                 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
 646                 .for_btnc = false, .for_shadow = false
 647         };
 648         struct inode *inode;
 649         int err;
 650
 651         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 652         if (unlikely(!inode))
 653                 return ERR_PTR(-ENOMEM);
 654         if (!(inode->i_state & I_NEW))
 655                 return inode;
 656
 657         err = nilfs_init_gcinode(inode);
 658         if (unlikely(err)) {
 659                 iget_failed(inode);
 660                 return ERR_PTR(err);
 661         }
 662         unlock_new_inode(inode);
 663         return inode;
 664 }
 665
 666 /**
 667  * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 668  * @inode: inode object
 669  *
 670  * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 671  * or does nothing if the inode already has it.  This function allocates
 672  * an additional inode to maintain page cache of B-tree nodes one-on-one.
 673  *
 674  * Return Value: On success, 0 is returned. On errors, one of the following
 675  * negative error code is returned.
 676  *
 677  * %-ENOMEM - Insufficient memory available.
 678  */
 679 int nilfs_attach_btree_node_cache(struct inode *inode)
 680 {
 681         struct nilfs_inode_info *ii = NILFS_I(inode);
 682         struct inode *btnc_inode;
 683         struct nilfs_iget_args args;
 684
 685         if (ii->i_assoc_inode)
 686                 return 0;
 687
 688         args.ino = inode->i_ino;
 689         args.root = ii->i_root;
 690         args.cno = ii->i_cno;
 691         args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
 692         args.for_btnc = true;
 693         args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
 694
 695         btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 696                                   nilfs_iget_set, &args);
 697         if (unlikely(!btnc_inode))
 698                 return -ENOMEM;
 699         if (btnc_inode->i_state & I_NEW) {
 700                 nilfs_init_btnc_inode(btnc_inode);
 701                 unlock_new_inode(btnc_inode);
 702         }
 703         NILFS_I(btnc_inode)->i_assoc_inode = inode;
 704         NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 705         ii->i_assoc_inode = btnc_inode;
 706
 707         return 0;
 708 }
 709
 710 /**
 711  * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 712  * @inode: inode object
 713  *
 714  * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 715  * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 716  */
 717 void nilfs_detach_btree_node_cache(struct inode *inode)
 718 {
 719         struct nilfs_inode_info *ii = NILFS_I(inode);
 720         struct inode *btnc_inode = ii->i_assoc_inode;
 721
 722         if (btnc_inode) {
 723                 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 724                 ii->i_assoc_inode = NULL;
 725                 iput(btnc_inode);
 726         }
 727 }
 728
 729 /**
 730  * nilfs_iget_for_shadow - obtain inode for shadow mapping
 731  * @inode: inode object that uses shadow mapping
 732  *
 733  * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 734  * caches for shadow mapping.  The page cache for data pages is set up
 735  * in one inode and the one for b-tree node pages is set up in the
 736  * other inode, which is attached to the former inode.
 737  *
 738  * Return Value: On success, a pointer to the inode for data pages is
 739  * returned. On errors, one of the following negative error code is returned
 740  * in a pointer type.
 741  *
 742  * %-ENOMEM - Insufficient memory available.
 743  */
 744 struct inode *nilfs_iget_for_shadow(struct inode *inode)
 745 {
 746         struct nilfs_iget_args args = {
 747                 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
 748                 .for_btnc = false, .for_shadow = true
 749         };
 750         struct inode *s_inode;
 751         int err;
 752
 753         s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 754                                nilfs_iget_set, &args);
 755         if (unlikely(!s_inode))
 756                 return ERR_PTR(-ENOMEM);
 757         if (!(s_inode->i_state & I_NEW))
 758                 return inode;
 759
 760         NILFS_I(s_inode)->i_flags = 0;
 761         memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 762         mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 763
 764         err = nilfs_attach_btree_node_cache(s_inode);
 765         if (unlikely(err)) {
 766                 iget_failed(s_inode);
 767                 return ERR_PTR(err);
 768         }
 769         unlock_new_inode(s_inode);
 770         return s_inode;
 771 }
 772
 773 void nilfs_write_inode_common(struct inode *inode,
 774                               struct nilfs_inode *raw_inode, int has_bmap)
 775 {
 776         struct nilfs_inode_info *ii = NILFS_I(inode);
 777
 778         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 779         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 780         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 781         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 782         raw_inode->i_size = cpu_to_le64(inode->i_size);
 783         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 784         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 785         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 786         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 787         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 788
 789         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 790         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 791
 792         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 793                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 794
 795                 /* zero-fill unused portion in the case of super root block */
 796                 raw_inode->i_xattr = 0;
 797                 raw_inode->i_pad = 0;
 798                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 799                        nilfs->ns_inode_size - sizeof(*raw_inode));
 800         }
 801
 802         if (has_bmap)
 803                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 804         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 805                 raw_inode->i_device_code =
 806                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 807         /*
 808          * When extending inode, nilfs->ns_inode_size should be checked
 809          * for substitutions of appended fields.
 810          */
 811 }
 812
 813 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 814 {
 815         ino_t ino = inode->i_ino;
 816         struct nilfs_inode_info *ii = NILFS_I(inode);
 817         struct inode *ifile = ii->i_root->ifile;
 818         struct nilfs_inode *raw_inode;
 819
 820         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 821
 822         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 823                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 824         if (flags & I_DIRTY_DATASYNC)
 825                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 826
 827         nilfs_write_inode_common(inode, raw_inode, 0);
 828                 /*
 829                  * XXX: call with has_bmap = 0 is a workaround to avoid
 830                  * deadlock of bmap.  This delays update of i_bmap to just
 831                  * before writing.
 832                  */
 833
 834         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 835 }
 836
 837 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 838
 839 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 840                                 unsigned long from)
 841 {
 842         __u64 b;
 843         int ret;
 844
 845         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 846                 return;
 847 repeat:
 848         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 849         if (ret == -ENOENT)
 850                 return;
 851         else if (ret < 0)
 852                 goto failed;
 853
 854         if (b < from)
 855                 return;
 856
 857         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 858         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 859         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 860         if (!ret || (ret == -ENOMEM &&
 861                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 862                 goto repeat;
 863
 864 failed:
 865         nilfs_msg(ii->vfs_inode.i_sb, KERN_WARNING,
 866                   "error %d truncating bmap (ino=%lu)", ret,
 867                   ii->vfs_inode.i_ino);
 868 }
 869
 870 void nilfs_truncate(struct inode *inode)
 871 {
 872         unsigned long blkoff;
 873         unsigned int blocksize;
 874         struct nilfs_transaction_info ti;
 875         struct super_block *sb = inode->i_sb;
 876         struct nilfs_inode_info *ii = NILFS_I(inode);
 877
 878         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 879                 return;
 880         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 881                 return;
 882
 883         blocksize = sb->s_blocksize;
 884         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 885         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 886
 887         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 888
 889         nilfs_truncate_bmap(ii, blkoff);
 890
 891         inode->i_mtime = inode->i_ctime = current_time(inode);
 892         if (IS_SYNC(inode))
 893                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 894
 895         nilfs_mark_inode_dirty(inode);
 896         nilfs_set_file_dirty(inode, 0);
 897         nilfs_transaction_commit(sb);
 898         /*
 899          * May construct a logical segment and may fail in sync mode.
 900          * But truncate has no return value.
 901          */
 902 }
 903
 904 static void nilfs_clear_inode(struct inode *inode)
 905 {
 906         struct nilfs_inode_info *ii = NILFS_I(inode);
 907
 908         /*
 909          * Free resources allocated in nilfs_read_inode(), here.
 910          */
 911         BUG_ON(!list_empty(&ii->i_dirty));
 912         brelse(ii->i_bh);
 913         ii->i_bh = NULL;
 914
 915         if (nilfs_is_metadata_file_inode(inode))
 916                 nilfs_mdt_clear(inode);
 917
 918         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 919                 nilfs_bmap_clear(ii->i_bmap);
 920
 921         if (!test_bit(NILFS_I_BTNC, &ii->i_state))
 922                 nilfs_detach_btree_node_cache(inode);
 923
 924         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 925                 nilfs_put_root(ii->i_root);
 926 }
 927
 928 void nilfs_evict_inode(struct inode *inode)
 929 {
 930         struct nilfs_transaction_info ti;
 931         struct super_block *sb = inode->i_sb;
 932         struct nilfs_inode_info *ii = NILFS_I(inode);
 933         struct the_nilfs *nilfs;
 934         int ret;
 935
 936         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 937                 truncate_inode_pages_final(&inode->i_data);
 938                 clear_inode(inode);
 939                 nilfs_clear_inode(inode);
 940                 return;
 941         }
 942         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 943
 944         truncate_inode_pages_final(&inode->i_data);
 945
 946         nilfs = sb->s_fs_info;
 947         if (unlikely(sb_rdonly(sb) || !nilfs->ns_writer)) {
 948                 /*
 949                  * If this inode is about to be disposed after the file system
 950                  * has been degraded to read-only due to file system corruption
 951                  * or after the writer has been detached, do not make any
 952                  * changes that cause writes, just clear it.
 953                  * Do this check after read-locking ns_segctor_sem by
 954                  * nilfs_transaction_begin() in order to avoid a race with
 955                  * the writer detach operation.
 956                  */
 957                 clear_inode(inode);
 958                 nilfs_clear_inode(inode);
 959                 nilfs_transaction_abort(sb);
 960                 return;
 961         }
 962
 963         /* TODO: some of the following operations may fail.  */
 964         nilfs_truncate_bmap(ii, 0);
 965         nilfs_mark_inode_dirty(inode);
 966         clear_inode(inode);
 967
 968         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 969         if (!ret)
 970                 atomic64_dec(&ii->i_root->inodes_count);
 971
 972         nilfs_clear_inode(inode);
 973
 974         if (IS_SYNC(inode))
 975                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 976         nilfs_transaction_commit(sb);
 977         /*
 978          * May construct a logical segment and may fail in sync mode.
 979          * But delete_inode has no return value.
 980          */
 981 }
 982
 983 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 984 {
 985         struct nilfs_transaction_info ti;
 986         struct inode *inode = d_inode(dentry);
 987         struct super_block *sb = inode->i_sb;
 988         int err;
 989
 990         err = setattr_prepare(dentry, iattr);
 991         if (err)
 992                 return err;
 993
 994         err = nilfs_transaction_begin(sb, &ti, 0);
 995         if (unlikely(err))
 996                 return err;
 997
 998         if ((iattr->ia_valid & ATTR_SIZE) &&
 999             iattr->ia_size != i_size_read(inode)) {
1000                 inode_dio_wait(inode);
1001                 truncate_setsize(inode, iattr->ia_size);
1002                 nilfs_truncate(inode);
1003         }
1004
1005         setattr_copy(inode, iattr);
1006         mark_inode_dirty(inode);
1007
1008         if (iattr->ia_valid & ATTR_MODE) {
1009                 err = nilfs_acl_chmod(inode);
1010                 if (unlikely(err))
1011                         goto out_err;
1012         }
1013
1014         return nilfs_transaction_commit(sb);
1015
1016 out_err:
1017         nilfs_transaction_abort(sb);
1018         return err;
1019 }
1020
1021 int nilfs_permission(struct inode *inode, int mask)
1022 {
1023         struct nilfs_root *root = NILFS_I(inode)->i_root;
1024
1025         if ((mask & MAY_WRITE) && root &&
1026             root->cno != NILFS_CPTREE_CURRENT_CNO)
1027                 return -EROFS; /* snapshot is not writable */
1028
1029         return generic_permission(inode, mask);
1030 }
1031
1032 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1033 {
1034         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1035         struct nilfs_inode_info *ii = NILFS_I(inode);
1036         int err;
1037
1038         spin_lock(&nilfs->ns_inode_lock);
1039         if (ii->i_bh == NULL || unlikely(!buffer_uptodate(ii->i_bh))) {
1040                 spin_unlock(&nilfs->ns_inode_lock);
1041                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1042                                                   inode->i_ino, pbh);
1043                 if (unlikely(err))
1044                         return err;
1045                 spin_lock(&nilfs->ns_inode_lock);
1046                 if (ii->i_bh == NULL)
1047                         ii->i_bh = *pbh;
1048                 else if (unlikely(!buffer_uptodate(ii->i_bh))) {
1049                         __brelse(ii->i_bh);
1050                         ii->i_bh = *pbh;
1051                 } else {
1052                         brelse(*pbh);
1053                         *pbh = ii->i_bh;
1054                 }
1055         } else
1056                 *pbh = ii->i_bh;
1057
1058         get_bh(*pbh);
1059         spin_unlock(&nilfs->ns_inode_lock);
1060         return 0;
1061 }
1062
1063 int nilfs_inode_dirty(struct inode *inode)
1064 {
1065         struct nilfs_inode_info *ii = NILFS_I(inode);
1066         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1067         int ret = 0;
1068
1069         if (!list_empty(&ii->i_dirty)) {
1070                 spin_lock(&nilfs->ns_inode_lock);
1071                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1072                         test_bit(NILFS_I_BUSY, &ii->i_state);
1073                 spin_unlock(&nilfs->ns_inode_lock);
1074         }
1075         return ret;
1076 }
1077
1078 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1079 {
1080         struct nilfs_inode_info *ii = NILFS_I(inode);
1081         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1082
1083         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1084
1085         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1086                 return 0;
1087
1088         spin_lock(&nilfs->ns_inode_lock);
1089         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1090             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1091                 /*
1092                  * Because this routine may race with nilfs_dispose_list(),
1093                  * we have to check NILFS_I_QUEUED here, too.
1094                  */
1095                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1096                         /*
1097                          * This will happen when somebody is freeing
1098                          * this inode.
1099                          */
1100                         nilfs_msg(inode->i_sb, KERN_WARNING,
1101                                   "cannot set file dirty (ino=%lu): the file is being freed",
1102                                   inode->i_ino);
1103                         spin_unlock(&nilfs->ns_inode_lock);
1104                         return -EINVAL; /*
1105                                          * NILFS_I_DIRTY may remain for
1106                                          * freeing inode.
1107                                          */
1108                 }
1109                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1110                 set_bit(NILFS_I_QUEUED, &ii->i_state);
1111         }
1112         spin_unlock(&nilfs->ns_inode_lock);
1113         return 0;
1114 }
1115
1116 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1117 {
1118         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1119         struct buffer_head *ibh;
1120         int err;
1121
1122         /*
1123          * Do not dirty inodes after the log writer has been detached
1124          * and its nilfs_root struct has been freed.
1125          */
1126         if (unlikely(nilfs_purging(nilfs)))
1127                 return 0;
1128
1129         err = nilfs_load_inode_block(inode, &ibh);
1130         if (unlikely(err)) {
1131                 nilfs_msg(inode->i_sb, KERN_WARNING,
1132                           "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1133                           inode->i_ino, err);
1134                 return err;
1135         }
1136         nilfs_update_inode(inode, ibh, flags);
1137         mark_buffer_dirty(ibh);
1138         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1139         brelse(ibh);
1140         return 0;
1141 }
1142
1143 /**
1144  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1145  * @inode: inode of the file to be registered.
1146  *
1147  * nilfs_dirty_inode() loads a inode block containing the specified
1148  * @inode and copies data from a nilfs_inode to a corresponding inode
1149  * entry in the inode block. This operation is excluded from the segment
1150  * construction. This function can be called both as a single operation
1151  * and as a part of indivisible file operations.
1152  */
1153 void nilfs_dirty_inode(struct inode *inode, int flags)
1154 {
1155         struct nilfs_transaction_info ti;
1156         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1157
1158         if (is_bad_inode(inode)) {
1159                 nilfs_msg(inode->i_sb, KERN_WARNING,
1160                           "tried to mark bad_inode dirty. ignored.");
1161                 dump_stack();
1162                 return;
1163         }
1164         if (mdi) {
1165                 nilfs_mdt_mark_dirty(inode);
1166                 return;
1167         }
1168         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1169         __nilfs_mark_inode_dirty(inode, flags);
1170         nilfs_transaction_commit(inode->i_sb); /* never fails */
1171 }
1172
1173 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1174                  __u64 start, __u64 len)
1175 {
1176         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1177         __u64 logical = 0, phys = 0, size = 0;
1178         __u32 flags = 0;
1179         loff_t isize;
1180         sector_t blkoff, end_blkoff;
1181         sector_t delalloc_blkoff;
1182         unsigned long delalloc_blklen;
1183         unsigned int blkbits = inode->i_blkbits;
1184         int ret, n;
1185
1186         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1187         if (ret)
1188                 return ret;
1189
1190         inode_lock(inode);
1191
1192         isize = i_size_read(inode);
1193
1194         blkoff = start >> blkbits;
1195         end_blkoff = (start + len - 1) >> blkbits;
1196
1197         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1198                                                         &delalloc_blkoff);
1199
1200         do {
1201                 __u64 blkphy;
1202                 unsigned int maxblocks;
1203
1204                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1205                         if (size) {
1206                                 /* End of the current extent */
1207                                 ret = fiemap_fill_next_extent(
1208                                         fieinfo, logical, phys, size, flags);
1209                                 if (ret)
1210                                         break;
1211                         }
1212                         if (blkoff > end_blkoff)
1213                                 break;
1214
1215                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1216                         logical = blkoff << blkbits;
1217                         phys = 0;
1218                         size = delalloc_blklen << blkbits;
1219
1220                         blkoff = delalloc_blkoff + delalloc_blklen;
1221                         delalloc_blklen = nilfs_find_uncommitted_extent(
1222                                 inode, blkoff, &delalloc_blkoff);
1223                         continue;
1224                 }
1225
1226                 /*
1227                  * Limit the number of blocks that we look up so as
1228                  * not to get into the next delayed allocation extent.
1229                  */
1230                 maxblocks = INT_MAX;
1231                 if (delalloc_blklen)
1232                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1233                                           maxblocks);
1234                 blkphy = 0;
1235
1236                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1237                 n = nilfs_bmap_lookup_contig(
1238                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1239                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1240
1241                 if (n < 0) {
1242                         int past_eof;
1243
1244                         if (unlikely(n != -ENOENT))
1245                                 break; /* error */
1246
1247                         /* HOLE */
1248                         blkoff++;
1249                         past_eof = ((blkoff << blkbits) >= isize);
1250
1251                         if (size) {
1252                                 /* End of the current extent */
1253
1254                                 if (past_eof)
1255                                         flags |= FIEMAP_EXTENT_LAST;
1256
1257                                 ret = fiemap_fill_next_extent(
1258                                         fieinfo, logical, phys, size, flags);
1259                                 if (ret)
1260                                         break;
1261                                 size = 0;
1262                         }
1263                         if (blkoff > end_blkoff || past_eof)
1264                                 break;
1265                 } else {
1266                         if (size) {
1267                                 if (phys && blkphy << blkbits == phys + size) {
1268                                         /* The current extent goes on */
1269                                         size += n << blkbits;
1270                                 } else {
1271                                         /* Terminate the current extent */
1272                                         ret = fiemap_fill_next_extent(
1273                                                 fieinfo, logical, phys, size,
1274                                                 flags);
1275                                         if (ret || blkoff > end_blkoff)
1276                                                 break;
1277
1278                                         /* Start another extent */
1279                                         flags = FIEMAP_EXTENT_MERGED;
1280                                         logical = blkoff << blkbits;
1281                                         phys = blkphy << blkbits;
1282                                         size = n << blkbits;
1283                                 }
1284                         } else {
1285                                 /* Start a new extent */
1286                                 flags = FIEMAP_EXTENT_MERGED;
1287                                 logical = blkoff << blkbits;
1288                                 phys = blkphy << blkbits;
1289                                 size = n << blkbits;
1290                         }
1291                         blkoff += n;
1292                 }
1293                 cond_resched();
1294         } while (true);
1295
1296         /* If ret is 1 then we just hit the end of the extent array */
1297         if (ret == 1)
1298                 ret = 0;
1299
1300         inode_unlock(inode);
1301         return ret;
1302 }