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  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/buffer_head.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mpage.h>
  27 #include <linux/writeback.h>
  28 #include <linux/uio.h>
  29 #include "nilfs.h"
  30 #include "segment.h"
  31 #include "page.h"
  32 #include "mdt.h"
  33 #include "cpfile.h"
  34 #include "ifile.h"
  35
  36
  37 /**
  38  * nilfs_get_block() - get a file block on the filesystem (callback function)
  39  * @inode - inode struct of the target file
  40  * @blkoff - file block number
  41  * @bh_result - buffer head to be mapped on
  42  * @create - indicate whether allocating the block or not when it has not
  43  *      been allocated yet.
  44  *
  45  * This function does not issue actual read request of the specified data
  46  * block. It is done by VFS.
  47  */
  48 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  49                     struct buffer_head *bh_result, int create)
  50 {
  51         struct nilfs_inode_info *ii = NILFS_I(inode);
  52         __u64 blknum = 0;
  53         int err = 0, ret;
  54         struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
  55         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  56
  57         down_read(&NILFS_MDT(dat)->mi_sem);
  58         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  59         up_read(&NILFS_MDT(dat)->mi_sem);
  60         if (ret >= 0) { /* found */
  61                 map_bh(bh_result, inode->i_sb, blknum);
  62                 if (ret > 0)
  63                         bh_result->b_size = (ret << inode->i_blkbits);
  64                 goto out;
  65         }
  66         /* data block was not found */
  67         if (ret == -ENOENT && create) {
  68                 struct nilfs_transaction_info ti;
  69
  70                 bh_result->b_blocknr = 0;
  71                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  72                 if (unlikely(err))
  73                         goto out;
  74                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
  75                                         (unsigned long)bh_result);
  76                 if (unlikely(err != 0)) {
  77                         if (err == -EEXIST) {
  78                                 /*
  79                                  * The get_block() function could be called
  80                                  * from multiple callers for an inode.
  81                                  * However, the page having this block must
  82                                  * be locked in this case.
  83                                  */
  84                                 printk(KERN_WARNING
  85                                        "nilfs_get_block: a race condition "
  86                                        "while inserting a data block. "
  87                                        "(inode number=%lu, file block "
  88                                        "offset=%llu)\n",
  89                                        inode->i_ino,
  90                                        (unsigned long long)blkoff);
  91                                 err = 0;
  92                         } else if (err == -EINVAL) {
  93                                 nilfs_error(inode->i_sb, __func__,
  94                                             "broken bmap (inode=%lu)\n",
  95                                             inode->i_ino);
  96                                 err = -EIO;
  97                         }
  98                         nilfs_transaction_abort(inode->i_sb);
  99                         goto out;
 100                 }
 101                 nilfs_mark_inode_dirty(inode);
 102                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 103                 /* Error handling should be detailed */
 104                 set_buffer_new(bh_result);
 105                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 106                                                       to proper value */
 107         } else if (ret == -ENOENT) {
 108                 /* not found is not error (e.g. hole); must return without
 109                    the mapped state flag. */
 110                 ;
 111         } else {
 112                 err = ret;
 113         }
 114
 115  out:
 116         return err;
 117 }
 118
 119 /**
 120  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 121  * address_space_operations.
 122  * @file - file struct of the file to be read
 123  * @page - the page to be read
 124  */
 125 static int nilfs_readpage(struct file *file, struct page *page)
 126 {
 127         return mpage_readpage(page, nilfs_get_block);
 128 }
 129
 130 /**
 131  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 132  * address_space_operations.
 133  * @file - file struct of the file to be read
 134  * @mapping - address_space struct used for reading multiple pages
 135  * @pages - the pages to be read
 136  * @nr_pages - number of pages to be read
 137  */
 138 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 139                            struct list_head *pages, unsigned nr_pages)
 140 {
 141         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 142 }
 143
 144 static int nilfs_writepages(struct address_space *mapping,
 145                             struct writeback_control *wbc)
 146 {
 147         struct inode *inode = mapping->host;
 148         int err = 0;
 149
 150         if (wbc->sync_mode == WB_SYNC_ALL)
 151                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 152                                                     wbc->range_start,
 153                                                     wbc->range_end);
 154         return err;
 155 }
 156
 157 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 158 {
 159         struct inode *inode = page->mapping->host;
 160         int err;
 161
 162         redirty_page_for_writepage(wbc, page);
 163         unlock_page(page);
 164
 165         if (wbc->sync_mode == WB_SYNC_ALL) {
 166                 err = nilfs_construct_segment(inode->i_sb);
 167                 if (unlikely(err))
 168                         return err;
 169         } else if (wbc->for_reclaim)
 170                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 171
 172         return 0;
 173 }
 174
 175 static int nilfs_set_page_dirty(struct page *page)
 176 {
 177         int ret = __set_page_dirty_buffers(page);
 178
 179         if (ret) {
 180                 struct inode *inode = page->mapping->host;
 181                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 182                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 183
 184                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 185         }
 186         return ret;
 187 }
 188
 189 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 190                              loff_t pos, unsigned len, unsigned flags,
 191                              struct page **pagep, void **fsdata)
 192
 193 {
 194         struct inode *inode = mapping->host;
 195         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 196
 197         if (unlikely(err))
 198                 return err;
 199
 200         *pagep = NULL;
 201         err = block_write_begin(file, mapping, pos, len, flags, pagep,
 202                                 fsdata, nilfs_get_block);
 203         if (unlikely(err))
 204                 nilfs_transaction_abort(inode->i_sb);
 205         return err;
 206 }
 207
 208 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 209                            loff_t pos, unsigned len, unsigned copied,
 210                            struct page *page, void *fsdata)
 211 {
 212         struct inode *inode = mapping->host;
 213         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 214         unsigned nr_dirty;
 215         int err;
 216
 217         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 218                                                   start + copied);
 219         copied = generic_write_end(file, mapping, pos, len, copied, page,
 220                                    fsdata);
 221         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 222         err = nilfs_transaction_commit(inode->i_sb);
 223         return err ? : copied;
 224 }
 225
 226 static ssize_t
 227 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 228                 loff_t offset, unsigned long nr_segs)
 229 {
 230         struct file *file = iocb->ki_filp;
 231         struct inode *inode = file->f_mapping->host;
 232         ssize_t size;
 233
 234         if (rw == WRITE)
 235                 return 0;
 236
 237         /* Needs synchronization with the cleaner */
 238         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 239                                   offset, nr_segs, nilfs_get_block, NULL);
 240
 241         /*
 242          * In case of error extending write may have instantiated a few
 243          * blocks outside i_size. Trim these off again.
 244          */
 245         if (unlikely((rw & WRITE) && size < 0)) {
 246                 loff_t isize = i_size_read(inode);
 247                 loff_t end = offset + iov_length(iov, nr_segs);
 248
 249                 if (end > isize)
 250                         vmtruncate(inode, isize);
 251         }
 252
 253         return size;
 254 }
 255
 256 const struct address_space_operations nilfs_aops = {
 257         .writepage              = nilfs_writepage,
 258         .readpage               = nilfs_readpage,
 259         .sync_page              = block_sync_page,
 260         .writepages             = nilfs_writepages,
 261         .set_page_dirty         = nilfs_set_page_dirty,
 262         .readpages              = nilfs_readpages,
 263         .write_begin            = nilfs_write_begin,
 264         .write_end              = nilfs_write_end,
 265         /* .releasepage         = nilfs_releasepage, */
 266         .invalidatepage         = block_invalidatepage,
 267         .direct_IO              = nilfs_direct_IO,
 268         .is_partially_uptodate  = block_is_partially_uptodate,
 269 };
 270
 271 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 272 {
 273         struct super_block *sb = dir->i_sb;
 274         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 275         struct inode *inode;
 276         struct nilfs_inode_info *ii;
 277         int err = -ENOMEM;
 278         ino_t ino;
 279
 280         inode = new_inode(sb);
 281         if (unlikely(!inode))
 282                 goto failed;
 283
 284         mapping_set_gfp_mask(inode->i_mapping,
 285                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 286
 287         ii = NILFS_I(inode);
 288         ii->i_state = 1 << NILFS_I_NEW;
 289
 290         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
 291         if (unlikely(err))
 292                 goto failed_ifile_create_inode;
 293         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 294
 295         atomic_inc(&sbi->s_inodes_count);
 296         inode_init_owner(inode, dir, mode);
 297         inode->i_ino = ino;
 298         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 299
 300         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 301                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 302                 if (err < 0)
 303                         goto failed_bmap;
 304
 305                 set_bit(NILFS_I_BMAP, &ii->i_state);
 306                 /* No lock is needed; iget() ensures it. */
 307         }
 308
 309         ii->i_flags = NILFS_I(dir)->i_flags;
 310         if (S_ISLNK(mode))
 311                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 312         if (!S_ISDIR(mode))
 313                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 314
 315         /* ii->i_file_acl = 0; */
 316         /* ii->i_dir_acl = 0; */
 317         ii->i_dir_start_lookup = 0;
 318         ii->i_cno = 0;
 319         nilfs_set_inode_flags(inode);
 320         spin_lock(&sbi->s_next_gen_lock);
 321         inode->i_generation = sbi->s_next_generation++;
 322         spin_unlock(&sbi->s_next_gen_lock);
 323         insert_inode_hash(inode);
 324
 325         err = nilfs_init_acl(inode, dir);
 326         if (unlikely(err))
 327                 goto failed_acl; /* never occur. When supporting
 328                                     nilfs_init_acl(), proper cancellation of
 329                                     above jobs should be considered */
 330
 331         return inode;
 332
 333  failed_acl:
 334  failed_bmap:
 335         inode->i_nlink = 0;
 336         iput(inode);  /* raw_inode will be deleted through
 337                          generic_delete_inode() */
 338         goto failed;
 339
 340  failed_ifile_create_inode:
 341         make_bad_inode(inode);
 342         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 343                          called */
 344  failed:
 345         return ERR_PTR(err);
 346 }
 347
 348 void nilfs_free_inode(struct inode *inode)
 349 {
 350         struct super_block *sb = inode->i_sb;
 351         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 352
 353         clear_inode(inode);
 354         /* XXX: check error code? Is there any thing I can do? */
 355         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
 356         atomic_dec(&sbi->s_inodes_count);
 357 }
 358
 359 void nilfs_set_inode_flags(struct inode *inode)
 360 {
 361         unsigned int flags = NILFS_I(inode)->i_flags;
 362
 363         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 364                             S_DIRSYNC);
 365         if (flags & NILFS_SYNC_FL)
 366                 inode->i_flags |= S_SYNC;
 367         if (flags & NILFS_APPEND_FL)
 368                 inode->i_flags |= S_APPEND;
 369         if (flags & NILFS_IMMUTABLE_FL)
 370                 inode->i_flags |= S_IMMUTABLE;
 371 #ifndef NILFS_ATIME_DISABLE
 372         if (flags & NILFS_NOATIME_FL)
 373 #endif
 374                 inode->i_flags |= S_NOATIME;
 375         if (flags & NILFS_DIRSYNC_FL)
 376                 inode->i_flags |= S_DIRSYNC;
 377         mapping_set_gfp_mask(inode->i_mapping,
 378                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 379 }
 380
 381 int nilfs_read_inode_common(struct inode *inode,
 382                             struct nilfs_inode *raw_inode)
 383 {
 384         struct nilfs_inode_info *ii = NILFS_I(inode);
 385         int err;
 386
 387         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 388         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 389         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 390         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 391         inode->i_size = le64_to_cpu(raw_inode->i_size);
 392         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 393         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 394         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 395         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 396         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 397         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 398         if (inode->i_nlink == 0 && inode->i_mode == 0)
 399                 return -EINVAL; /* this inode is deleted */
 400
 401         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 402         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 403 #if 0
 404         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 405         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 406                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 407 #endif
 408         ii->i_dir_start_lookup = 0;
 409         ii->i_cno = 0;
 410         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 411
 412         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 413             S_ISLNK(inode->i_mode)) {
 414                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 415                 if (err < 0)
 416                         return err;
 417                 set_bit(NILFS_I_BMAP, &ii->i_state);
 418                 /* No lock is needed; iget() ensures it. */
 419         }
 420         return 0;
 421 }
 422
 423 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 424                               struct inode *inode)
 425 {
 426         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 427         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 428         struct buffer_head *bh;
 429         struct nilfs_inode *raw_inode;
 430         int err;
 431
 432         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 433         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 434         if (unlikely(err))
 435                 goto bad_inode;
 436
 437         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 438
 439         err = nilfs_read_inode_common(inode, raw_inode);
 440         if (err)
 441                 goto failed_unmap;
 442
 443         if (S_ISREG(inode->i_mode)) {
 444                 inode->i_op = &nilfs_file_inode_operations;
 445                 inode->i_fop = &nilfs_file_operations;
 446                 inode->i_mapping->a_ops = &nilfs_aops;
 447         } else if (S_ISDIR(inode->i_mode)) {
 448                 inode->i_op = &nilfs_dir_inode_operations;
 449                 inode->i_fop = &nilfs_dir_operations;
 450                 inode->i_mapping->a_ops = &nilfs_aops;
 451         } else if (S_ISLNK(inode->i_mode)) {
 452                 inode->i_op = &nilfs_symlink_inode_operations;
 453                 inode->i_mapping->a_ops = &nilfs_aops;
 454         } else {
 455                 inode->i_op = &nilfs_special_inode_operations;
 456                 init_special_inode(
 457                         inode, inode->i_mode,
 458                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 459         }
 460         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 461         brelse(bh);
 462         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 463         nilfs_set_inode_flags(inode);
 464         return 0;
 465
 466  failed_unmap:
 467         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 468         brelse(bh);
 469
 470  bad_inode:
 471         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 472         return err;
 473 }
 474
 475 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 476 {
 477         struct inode *inode;
 478         int err;
 479
 480         inode = iget_locked(sb, ino);
 481         if (unlikely(!inode))
 482                 return ERR_PTR(-ENOMEM);
 483         if (!(inode->i_state & I_NEW))
 484                 return inode;
 485
 486         err = __nilfs_read_inode(sb, ino, inode);
 487         if (unlikely(err)) {
 488                 iget_failed(inode);
 489                 return ERR_PTR(err);
 490         }
 491         unlock_new_inode(inode);
 492         return inode;
 493 }
 494
 495 void nilfs_write_inode_common(struct inode *inode,
 496                               struct nilfs_inode *raw_inode, int has_bmap)
 497 {
 498         struct nilfs_inode_info *ii = NILFS_I(inode);
 499
 500         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 501         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 502         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 503         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 504         raw_inode->i_size = cpu_to_le64(inode->i_size);
 505         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 506         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 507         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 508         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 509         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 510
 511         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 512         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 513
 514         if (has_bmap)
 515                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 516         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 517                 raw_inode->i_device_code =
 518                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 519         /* When extending inode, nilfs->ns_inode_size should be checked
 520            for substitutions of appended fields */
 521 }
 522
 523 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 524 {
 525         ino_t ino = inode->i_ino;
 526         struct nilfs_inode_info *ii = NILFS_I(inode);
 527         struct super_block *sb = inode->i_sb;
 528         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 529         struct nilfs_inode *raw_inode;
 530
 531         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
 532
 533         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 534                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 535         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 536
 537         nilfs_write_inode_common(inode, raw_inode, 0);
 538                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 539                    deadlock of bmap. This delays update of i_bmap to just
 540                    before writing */
 541         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 542 }
 543
 544 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 545
 546 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 547                                 unsigned long from)
 548 {
 549         unsigned long b;
 550         int ret;
 551
 552         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 553                 return;
 554  repeat:
 555         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 556         if (ret == -ENOENT)
 557                 return;
 558         else if (ret < 0)
 559                 goto failed;
 560
 561         if (b < from)
 562                 return;
 563
 564         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 565         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 566         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 567         if (!ret || (ret == -ENOMEM &&
 568                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 569                 goto repeat;
 570
 571  failed:
 572         if (ret == -EINVAL)
 573                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 574                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 575         else
 576                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 577                               "failed to truncate bmap (ino=%lu, err=%d)",
 578                               ii->vfs_inode.i_ino, ret);
 579 }
 580
 581 void nilfs_truncate(struct inode *inode)
 582 {
 583         unsigned long blkoff;
 584         unsigned int blocksize;
 585         struct nilfs_transaction_info ti;
 586         struct super_block *sb = inode->i_sb;
 587         struct nilfs_inode_info *ii = NILFS_I(inode);
 588
 589         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 590                 return;
 591         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 592                 return;
 593
 594         blocksize = sb->s_blocksize;
 595         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 596         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 597
 598         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 599
 600         nilfs_truncate_bmap(ii, blkoff);
 601
 602         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 603         if (IS_SYNC(inode))
 604                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 605
 606         nilfs_mark_inode_dirty(inode);
 607         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 608         nilfs_transaction_commit(sb);
 609         /* May construct a logical segment and may fail in sync mode.
 610            But truncate has no return value. */
 611 }
 612
 613 void nilfs_delete_inode(struct inode *inode)
 614 {
 615         struct nilfs_transaction_info ti;
 616         struct super_block *sb = inode->i_sb;
 617         struct nilfs_inode_info *ii = NILFS_I(inode);
 618
 619         if (unlikely(is_bad_inode(inode))) {
 620                 if (inode->i_data.nrpages)
 621                         truncate_inode_pages(&inode->i_data, 0);
 622                 clear_inode(inode);
 623                 return;
 624         }
 625         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 626
 627         if (inode->i_data.nrpages)
 628                 truncate_inode_pages(&inode->i_data, 0);
 629
 630         nilfs_truncate_bmap(ii, 0);
 631         nilfs_mark_inode_dirty(inode);
 632         nilfs_free_inode(inode);
 633         /* nilfs_free_inode() marks inode buffer dirty */
 634         if (IS_SYNC(inode))
 635                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 636         nilfs_transaction_commit(sb);
 637         /* May construct a logical segment and may fail in sync mode.
 638            But delete_inode has no return value. */
 639 }
 640
 641 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 642 {
 643         struct nilfs_transaction_info ti;
 644         struct inode *inode = dentry->d_inode;
 645         struct super_block *sb = inode->i_sb;
 646         int err;
 647
 648         err = inode_change_ok(inode, iattr);
 649         if (err)
 650                 return err;
 651
 652         err = nilfs_transaction_begin(sb, &ti, 0);
 653         if (unlikely(err))
 654                 return err;
 655         err = inode_setattr(inode, iattr);
 656         if (!err && (iattr->ia_valid & ATTR_MODE))
 657                 err = nilfs_acl_chmod(inode);
 658         if (likely(!err))
 659                 err = nilfs_transaction_commit(sb);
 660         else
 661                 nilfs_transaction_abort(sb);
 662
 663         return err;
 664 }
 665
 666 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 667                            struct buffer_head **pbh)
 668 {
 669         struct nilfs_inode_info *ii = NILFS_I(inode);
 670         int err;
 671
 672         spin_lock(&sbi->s_inode_lock);
 673         if (ii->i_bh == NULL) {
 674                 spin_unlock(&sbi->s_inode_lock);
 675                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
 676                                                   pbh);
 677                 if (unlikely(err))
 678                         return err;
 679                 spin_lock(&sbi->s_inode_lock);
 680                 if (ii->i_bh == NULL)
 681                         ii->i_bh = *pbh;
 682                 else {
 683                         brelse(*pbh);
 684                         *pbh = ii->i_bh;
 685                 }
 686         } else
 687                 *pbh = ii->i_bh;
 688
 689         get_bh(*pbh);
 690         spin_unlock(&sbi->s_inode_lock);
 691         return 0;
 692 }
 693
 694 int nilfs_inode_dirty(struct inode *inode)
 695 {
 696         struct nilfs_inode_info *ii = NILFS_I(inode);
 697         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 698         int ret = 0;
 699
 700         if (!list_empty(&ii->i_dirty)) {
 701                 spin_lock(&sbi->s_inode_lock);
 702                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 703                         test_bit(NILFS_I_BUSY, &ii->i_state);
 704                 spin_unlock(&sbi->s_inode_lock);
 705         }
 706         return ret;
 707 }
 708
 709 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 710                          unsigned nr_dirty)
 711 {
 712         struct nilfs_inode_info *ii = NILFS_I(inode);
 713
 714         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 715
 716         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 717                 return 0;
 718
 719         spin_lock(&sbi->s_inode_lock);
 720         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 721             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 722                 /* Because this routine may race with nilfs_dispose_list(),
 723                    we have to check NILFS_I_QUEUED here, too. */
 724                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 725                         /* This will happen when somebody is freeing
 726                            this inode. */
 727                         nilfs_warning(sbi->s_super, __func__,
 728                                       "cannot get inode (ino=%lu)\n",
 729                                       inode->i_ino);
 730                         spin_unlock(&sbi->s_inode_lock);
 731                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 732                                            freeing inode */
 733                 }
 734                 list_del(&ii->i_dirty);
 735                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 736                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 737         }
 738         spin_unlock(&sbi->s_inode_lock);
 739         return 0;
 740 }
 741
 742 int nilfs_mark_inode_dirty(struct inode *inode)
 743 {
 744         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 745         struct buffer_head *ibh;
 746         int err;
 747
 748         err = nilfs_load_inode_block(sbi, inode, &ibh);
 749         if (unlikely(err)) {
 750                 nilfs_warning(inode->i_sb, __func__,
 751                               "failed to reget inode block.\n");
 752                 return err;
 753         }
 754         nilfs_update_inode(inode, ibh);
 755         nilfs_mdt_mark_buffer_dirty(ibh);
 756         nilfs_mdt_mark_dirty(sbi->s_ifile);
 757         brelse(ibh);
 758         return 0;
 759 }
 760
 761 /**
 762  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 763  * @inode: inode of the file to be registered.
 764  *
 765  * nilfs_dirty_inode() loads a inode block containing the specified
 766  * @inode and copies data from a nilfs_inode to a corresponding inode
 767  * entry in the inode block. This operation is excluded from the segment
 768  * construction. This function can be called both as a single operation
 769  * and as a part of indivisible file operations.
 770  */
 771 void nilfs_dirty_inode(struct inode *inode)
 772 {
 773         struct nilfs_transaction_info ti;
 774
 775         if (is_bad_inode(inode)) {
 776                 nilfs_warning(inode->i_sb, __func__,
 777                               "tried to mark bad_inode dirty. ignored.\n");
 778                 dump_stack();
 779                 return;
 780         }
 781         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 782         nilfs_mark_inode_dirty(inode);
 783         nilfs_transaction_commit(inode->i_sb); /* never fails */
 784 }