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/mpage.h>
  26 #include <linux/writeback.h>
  27 #include <linux/uio.h>
  28 #include "nilfs.h"
  29 #include "segment.h"
  30 #include "page.h"
  31 #include "mdt.h"
  32 #include "cpfile.h"
  33 #include "ifile.h"
  34
  35
  36 /**
  37  * nilfs_get_block() - get a file block on the filesystem (callback function)
  38  * @inode - inode struct of the target file
  39  * @blkoff - file block number
  40  * @bh_result - buffer head to be mapped on
  41  * @create - indicate whether allocating the block or not when it has not
  42  *      been allocated yet.
  43  *
  44  * This function does not issue actual read request of the specified data
  45  * block. It is done by VFS.
  46  * Bulk read for direct-io is not supported yet. (should be supported)
  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         unsigned long blknum = 0;
  53         int err = 0, ret;
  54         struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
  55
  56         /* This exclusion control is a workaround; should be revised */
  57         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
  58         ret = nilfs_bmap_lookup(ii->i_bmap, (unsigned long)blkoff, &blknum);
  59         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
  60         if (ret == 0) { /* found */
  61                 map_bh(bh_result, inode->i_sb, blknum);
  62                 goto out;
  63         }
  64         /* data block was not found */
  65         if (ret == -ENOENT && create) {
  66                 struct nilfs_transaction_info ti;
  67
  68                 bh_result->b_blocknr = 0;
  69                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  70                 if (unlikely(err))
  71                         goto out;
  72                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
  73                                         (unsigned long)bh_result);
  74                 if (unlikely(err != 0)) {
  75                         if (err == -EEXIST) {
  76                                 /*
  77                                  * The get_block() function could be called
  78                                  * from multiple callers for an inode.
  79                                  * However, the page having this block must
  80                                  * be locked in this case.
  81                                  */
  82                                 printk(KERN_WARNING
  83                                        "nilfs_get_block: a race condition "
  84                                        "while inserting a data block. "
  85                                        "(inode number=%lu, file block "
  86                                        "offset=%llu)\n",
  87                                        inode->i_ino,
  88                                        (unsigned long long)blkoff);
  89                                 err = 0;
  90                         } else if (err == -EINVAL) {
  91                                 nilfs_error(inode->i_sb, __func__,
  92                                             "broken bmap (inode=%lu)\n",
  93                                             inode->i_ino);
  94                                 err = -EIO;
  95                         }
  96                         nilfs_transaction_abort(inode->i_sb);
  97                         goto out;
  98                 }
  99                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 100                 /* Error handling should be detailed */
 101                 set_buffer_new(bh_result);
 102                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 103                                                       to proper value */
 104         } else if (ret == -ENOENT) {
 105                 /* not found is not error (e.g. hole); must return without
 106                    the mapped state flag. */
 107                 ;
 108         } else {
 109                 err = ret;
 110         }
 111
 112  out:
 113         return err;
 114 }
 115
 116 /**
 117  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 118  * address_space_operations.
 119  * @file - file struct of the file to be read
 120  * @page - the page to be read
 121  */
 122 static int nilfs_readpage(struct file *file, struct page *page)
 123 {
 124         return mpage_readpage(page, nilfs_get_block);
 125 }
 126
 127 /**
 128  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 129  * address_space_operations.
 130  * @file - file struct of the file to be read
 131  * @mapping - address_space struct used for reading multiple pages
 132  * @pages - the pages to be read
 133  * @nr_pages - number of pages to be read
 134  */
 135 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 136                            struct list_head *pages, unsigned nr_pages)
 137 {
 138         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 139 }
 140
 141 static int nilfs_writepages(struct address_space *mapping,
 142                             struct writeback_control *wbc)
 143 {
 144         struct inode *inode = mapping->host;
 145         int err = 0;
 146
 147         if (wbc->sync_mode == WB_SYNC_ALL)
 148                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 149                                                     wbc->range_start,
 150                                                     wbc->range_end);
 151         return err;
 152 }
 153
 154 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 155 {
 156         struct inode *inode = page->mapping->host;
 157         int err;
 158
 159         redirty_page_for_writepage(wbc, page);
 160         unlock_page(page);
 161
 162         if (wbc->sync_mode == WB_SYNC_ALL) {
 163                 err = nilfs_construct_segment(inode->i_sb);
 164                 if (unlikely(err))
 165                         return err;
 166         } else if (wbc->for_reclaim)
 167                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 168
 169         return 0;
 170 }
 171
 172 static int nilfs_set_page_dirty(struct page *page)
 173 {
 174         int ret = __set_page_dirty_buffers(page);
 175
 176         if (ret) {
 177                 struct inode *inode = page->mapping->host;
 178                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 179                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 180
 181                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 182         }
 183         return ret;
 184 }
 185
 186 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 187                              loff_t pos, unsigned len, unsigned flags,
 188                              struct page **pagep, void **fsdata)
 189
 190 {
 191         struct inode *inode = mapping->host;
 192         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 193
 194         if (unlikely(err))
 195                 return err;
 196
 197         *pagep = NULL;
 198         err = block_write_begin(file, mapping, pos, len, flags, pagep,
 199                                 fsdata, nilfs_get_block);
 200         if (unlikely(err))
 201                 nilfs_transaction_abort(inode->i_sb);
 202         return err;
 203 }
 204
 205 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 206                            loff_t pos, unsigned len, unsigned copied,
 207                            struct page *page, void *fsdata)
 208 {
 209         struct inode *inode = mapping->host;
 210         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 211         unsigned nr_dirty;
 212         int err;
 213
 214         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 215                                                   start + copied);
 216         copied = generic_write_end(file, mapping, pos, len, copied, page,
 217                                    fsdata);
 218         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 219         err = nilfs_transaction_commit(inode->i_sb);
 220         return err ? : copied;
 221 }
 222
 223 static ssize_t
 224 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 225                 loff_t offset, unsigned long nr_segs)
 226 {
 227         struct file *file = iocb->ki_filp;
 228         struct inode *inode = file->f_mapping->host;
 229         ssize_t size;
 230
 231         if (rw == WRITE)
 232                 return 0;
 233
 234         /* Needs synchronization with the cleaner */
 235         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 236                                   offset, nr_segs, nilfs_get_block, NULL);
 237         return size;
 238 }
 239
 240 struct address_space_operations nilfs_aops = {
 241         .writepage              = nilfs_writepage,
 242         .readpage               = nilfs_readpage,
 243         /* .sync_page           = nilfs_sync_page, */
 244         .writepages             = nilfs_writepages,
 245         .set_page_dirty         = nilfs_set_page_dirty,
 246         .readpages              = nilfs_readpages,
 247         .write_begin            = nilfs_write_begin,
 248         .write_end              = nilfs_write_end,
 249         /* .releasepage         = nilfs_releasepage, */
 250         .invalidatepage         = block_invalidatepage,
 251         .direct_IO              = nilfs_direct_IO,
 252 };
 253
 254 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 255 {
 256         struct super_block *sb = dir->i_sb;
 257         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 258         struct inode *inode;
 259         struct nilfs_inode_info *ii;
 260         int err = -ENOMEM;
 261         ino_t ino;
 262
 263         inode = new_inode(sb);
 264         if (unlikely(!inode))
 265                 goto failed;
 266
 267         mapping_set_gfp_mask(inode->i_mapping,
 268                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 269
 270         ii = NILFS_I(inode);
 271         ii->i_state = 1 << NILFS_I_NEW;
 272
 273         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
 274         if (unlikely(err))
 275                 goto failed_ifile_create_inode;
 276         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 277
 278         atomic_inc(&sbi->s_inodes_count);
 279
 280         inode->i_uid = current_fsuid();
 281         if (dir->i_mode & S_ISGID) {
 282                 inode->i_gid = dir->i_gid;
 283                 if (S_ISDIR(mode))
 284                         mode |= S_ISGID;
 285         } else
 286                 inode->i_gid = current_fsgid();
 287
 288         inode->i_mode = mode;
 289         inode->i_ino = ino;
 290         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 291
 292         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 293                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 294                 if (err < 0)
 295                         goto failed_bmap;
 296
 297                 set_bit(NILFS_I_BMAP, &ii->i_state);
 298                 /* No lock is needed; iget() ensures it. */
 299         }
 300
 301         ii->i_flags = NILFS_I(dir)->i_flags;
 302         if (S_ISLNK(mode))
 303                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 304         if (!S_ISDIR(mode))
 305                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 306
 307         /* ii->i_file_acl = 0; */
 308         /* ii->i_dir_acl = 0; */
 309         ii->i_dtime = 0;
 310         ii->i_dir_start_lookup = 0;
 311 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 312         ii->i_acl = NULL;
 313         ii->i_default_acl = NULL;
 314 #endif
 315         ii->i_cno = 0;
 316         nilfs_set_inode_flags(inode);
 317         spin_lock(&sbi->s_next_gen_lock);
 318         inode->i_generation = sbi->s_next_generation++;
 319         spin_unlock(&sbi->s_next_gen_lock);
 320         insert_inode_hash(inode);
 321
 322         err = nilfs_init_acl(inode, dir);
 323         if (unlikely(err))
 324                 goto failed_acl; /* never occur. When supporting
 325                                     nilfs_init_acl(), proper cancellation of
 326                                     above jobs should be considered */
 327
 328         mark_inode_dirty(inode);
 329         return inode;
 330
 331  failed_acl:
 332  failed_bmap:
 333         inode->i_nlink = 0;
 334         iput(inode);  /* raw_inode will be deleted through
 335                          generic_delete_inode() */
 336         goto failed;
 337
 338  failed_ifile_create_inode:
 339         make_bad_inode(inode);
 340         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 341                          called */
 342  failed:
 343         return ERR_PTR(err);
 344 }
 345
 346 void nilfs_free_inode(struct inode *inode)
 347 {
 348         struct super_block *sb = inode->i_sb;
 349         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 350
 351         clear_inode(inode);
 352         /* XXX: check error code? Is there any thing I can do? */
 353         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
 354         atomic_dec(&sbi->s_inodes_count);
 355 }
 356
 357 void nilfs_set_inode_flags(struct inode *inode)
 358 {
 359         unsigned int flags = NILFS_I(inode)->i_flags;
 360
 361         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 362                             S_DIRSYNC);
 363         if (flags & NILFS_SYNC_FL)
 364                 inode->i_flags |= S_SYNC;
 365         if (flags & NILFS_APPEND_FL)
 366                 inode->i_flags |= S_APPEND;
 367         if (flags & NILFS_IMMUTABLE_FL)
 368                 inode->i_flags |= S_IMMUTABLE;
 369 #ifndef NILFS_ATIME_DISABLE
 370         if (flags & NILFS_NOATIME_FL)
 371 #endif
 372                 inode->i_flags |= S_NOATIME;
 373         if (flags & NILFS_DIRSYNC_FL)
 374                 inode->i_flags |= S_DIRSYNC;
 375         mapping_set_gfp_mask(inode->i_mapping,
 376                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 377 }
 378
 379 int nilfs_read_inode_common(struct inode *inode,
 380                             struct nilfs_inode *raw_inode)
 381 {
 382         struct nilfs_inode_info *ii = NILFS_I(inode);
 383         int err;
 384
 385         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 386         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 387         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 388         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 389         inode->i_size = le64_to_cpu(raw_inode->i_size);
 390         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 391         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 392         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 393         inode->i_atime.tv_nsec = 0;
 394         inode->i_ctime.tv_nsec = 0;
 395         inode->i_mtime.tv_nsec = 0;
 396         ii->i_dtime = le64_to_cpu(raw_inode->i_dtime);
 397         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ii->i_dtime))
 398                 return -EINVAL; /* this inode is deleted */
 399
 400         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 401         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 402 #if 0
 403         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 404         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 405                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 406 #endif
 407         ii->i_cno = 0;
 408         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 409
 410         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 411             S_ISLNK(inode->i_mode)) {
 412                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 413                 if (err < 0)
 414                         return err;
 415                 set_bit(NILFS_I_BMAP, &ii->i_state);
 416                 /* No lock is needed; iget() ensures it. */
 417         }
 418         return 0;
 419 }
 420
 421 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 422                               struct inode *inode)
 423 {
 424         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 425         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 426         struct buffer_head *bh;
 427         struct nilfs_inode *raw_inode;
 428         int err;
 429
 430         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 431         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 432         if (unlikely(err))
 433                 goto bad_inode;
 434
 435         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 436
 437 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 438         ii->i_acl = NILFS_ACL_NOT_CACHED;
 439         ii->i_default_acl = NILFS_ACL_NOT_CACHED;
 440 #endif
 441         if (nilfs_read_inode_common(inode, raw_inode))
 442                 goto failed_unmap;
 443
 444         if (S_ISREG(inode->i_mode)) {
 445                 inode->i_op = &nilfs_file_inode_operations;
 446                 inode->i_fop = &nilfs_file_operations;
 447                 inode->i_mapping->a_ops = &nilfs_aops;
 448         } else if (S_ISDIR(inode->i_mode)) {
 449                 inode->i_op = &nilfs_dir_inode_operations;
 450                 inode->i_fop = &nilfs_dir_operations;
 451                 inode->i_mapping->a_ops = &nilfs_aops;
 452         } else if (S_ISLNK(inode->i_mode)) {
 453                 inode->i_op = &nilfs_symlink_inode_operations;
 454                 inode->i_mapping->a_ops = &nilfs_aops;
 455         } else {
 456                 inode->i_op = &nilfs_special_inode_operations;
 457                 init_special_inode(
 458                         inode, inode->i_mode,
 459                         new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 460         }
 461         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 462         brelse(bh);
 463         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 464         nilfs_set_inode_flags(inode);
 465         return 0;
 466
 467  failed_unmap:
 468         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 469         brelse(bh);
 470
 471  bad_inode:
 472         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 473         return err;
 474 }
 475
 476 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 477 {
 478         struct inode *inode;
 479         int err;
 480
 481         inode = iget_locked(sb, ino);
 482         if (unlikely(!inode))
 483                 return ERR_PTR(-ENOMEM);
 484         if (!(inode->i_state & I_NEW))
 485                 return inode;
 486
 487         err = __nilfs_read_inode(sb, ino, inode);
 488         if (unlikely(err)) {
 489                 iget_failed(inode);
 490                 return ERR_PTR(err);
 491         }
 492         unlock_new_inode(inode);
 493         return inode;
 494 }
 495
 496 void nilfs_write_inode_common(struct inode *inode,
 497                               struct nilfs_inode *raw_inode, int has_bmap)
 498 {
 499         struct nilfs_inode_info *ii = NILFS_I(inode);
 500
 501         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 502         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 503         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 504         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 505         raw_inode->i_size = cpu_to_le64(inode->i_size);
 506         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 507         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 508         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 509
 510         raw_inode->i_dtime = cpu_to_le64(ii->i_dtime);
 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(new_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         /* The buffer is guarded with lock_buffer() by the caller */
 534         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 535                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 536         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 537
 538         nilfs_write_inode_common(inode, raw_inode, 0);
 539                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 540                    deadlock of bmap. This delays update of i_bmap to just
 541                    before writing */
 542         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 543 }
 544
 545 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 546
 547 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 548                                 unsigned long from)
 549 {
 550         unsigned long b;
 551         int ret;
 552
 553         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 554                 return;
 555  repeat:
 556         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 557         if (ret == -ENOENT)
 558                 return;
 559         else if (ret < 0)
 560                 goto failed;
 561
 562         if (b < from)
 563                 return;
 564
 565         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 566         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 567         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 568         if (!ret || (ret == -ENOMEM &&
 569                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 570                 goto repeat;
 571
 572  failed:
 573         if (ret == -EINVAL)
 574                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 575                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 576         else
 577                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 578                               "failed to truncate bmap (ino=%lu, err=%d)",
 579                               ii->vfs_inode.i_ino, ret);
 580 }
 581
 582 void nilfs_truncate(struct inode *inode)
 583 {
 584         unsigned long blkoff;
 585         unsigned int blocksize;
 586         struct nilfs_transaction_info ti;
 587         struct super_block *sb = inode->i_sb;
 588         struct nilfs_inode_info *ii = NILFS_I(inode);
 589
 590         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 591                 return;
 592         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 593                 return;
 594
 595         blocksize = sb->s_blocksize;
 596         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 597         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 598
 599         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 600
 601         nilfs_truncate_bmap(ii, blkoff);
 602
 603         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 604         if (IS_SYNC(inode))
 605                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 606
 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_free_inode(inode);
 632         /* nilfs_free_inode() marks inode buffer dirty */
 633         if (IS_SYNC(inode))
 634                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 635         nilfs_transaction_commit(sb);
 636         /* May construct a logical segment and may fail in sync mode.
 637            But delete_inode has no return value. */
 638 }
 639
 640 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 641 {
 642         struct nilfs_transaction_info ti;
 643         struct inode *inode = dentry->d_inode;
 644         struct super_block *sb = inode->i_sb;
 645         int err;
 646
 647         err = inode_change_ok(inode, iattr);
 648         if (err)
 649                 return err;
 650
 651         err = nilfs_transaction_begin(sb, &ti, 0);
 652         if (unlikely(err))
 653                 return err;
 654         err = inode_setattr(inode, iattr);
 655         if (!err && (iattr->ia_valid & ATTR_MODE))
 656                 err = nilfs_acl_chmod(inode);
 657         if (likely(!err))
 658                 err = nilfs_transaction_commit(sb);
 659         else
 660                 nilfs_transaction_abort(sb);
 661
 662         return err;
 663 }
 664
 665 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 666                            struct buffer_head **pbh)
 667 {
 668         struct nilfs_inode_info *ii = NILFS_I(inode);
 669         int err;
 670
 671         spin_lock(&sbi->s_inode_lock);
 672         /* Caller of this function MUST lock 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         lock_buffer(ibh);
 755         nilfs_update_inode(inode, ibh);
 756         unlock_buffer(ibh);
 757         nilfs_mdt_mark_buffer_dirty(ibh);
 758         nilfs_mdt_mark_dirty(sbi->s_ifile);
 759         brelse(ibh);
 760         return 0;
 761 }
 762
 763 /**
 764  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 765  * @inode: inode of the file to be registered.
 766  *
 767  * nilfs_dirty_inode() loads a inode block containing the specified
 768  * @inode and copies data from a nilfs_inode to a corresponding inode
 769  * entry in the inode block. This operation is excluded from the segment
 770  * construction. This function can be called both as a single operation
 771  * and as a part of indivisible file operations.
 772  */
 773 void nilfs_dirty_inode(struct inode *inode)
 774 {
 775         struct nilfs_transaction_info ti;
 776
 777         if (is_bad_inode(inode)) {
 778                 nilfs_warning(inode->i_sb, __func__,
 779                               "tried to mark bad_inode dirty. ignored.\n");
 780                 dump_stack();
 781                 return;
 782         }
 783         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 784         nilfs_mark_inode_dirty(inode);
 785         nilfs_transaction_commit(inode->i_sb); /* never fails */
 786 }