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         if (unlikely(ret == 1)) {
  65                 printk(KERN_ERR "nilfs_get_block: bmap_lookup returns "
  66                        "buffer_head pointer (blkoff=%llu, blknum=%lu)\n",
  67                        (unsigned long long)blkoff, blknum);
  68                 BUG();
  69         }
  70         /* data block was not found */
  71         if (ret == -ENOENT && create) {
  72                 struct nilfs_transaction_info ti;
  73
  74                 bh_result->b_blocknr = 0;
  75                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  76                 if (unlikely(err))
  77                         goto out;
  78                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
  79                                         (unsigned long)bh_result);
  80                 nilfs_transaction_end(inode->i_sb, !err);
  81                 if (unlikely(err != 0)) {
  82                         if (err == -EEXIST) {
  83                                 /*
  84                                  * The get_block() function could be called
  85                                  * from multiple callers for an inode.
  86                                  * However, the page having this block must
  87                                  * be locked in this case.
  88                                  */
  89                                 printk(KERN_ERR
  90                                        "nilfs_get_block: a race condition "
  91                                        "while inserting a data block. "
  92                                        "(inode number=%lu, file block "
  93                                        "offset=%llu)\n",
  94                                        inode->i_ino,
  95                                        (unsigned long long)blkoff);
  96                                 BUG();
  97                         } else if (err == -EINVAL) {
  98                                 nilfs_error(inode->i_sb, __func__,
  99                                             "broken bmap (inode=%lu)\n",
 100                                             inode->i_ino);
 101                                 err = -EIO;
 102                         }
 103                         goto out;
 104                 }
 105                 /* Error handling should be detailed */
 106                 set_buffer_new(bh_result);
 107                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 108                                                       to proper value */
 109         } else if (ret == -ENOENT) {
 110                 /* not found is not error (e.g. hole); must return without
 111                    the mapped state flag. */
 112                 ;
 113         } else {
 114                 err = ret;
 115         }
 116
 117  out:
 118         return err;
 119 }
 120
 121 /**
 122  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 123  * address_space_operations.
 124  * @file - file struct of the file to be read
 125  * @page - the page to be read
 126  */
 127 static int nilfs_readpage(struct file *file, struct page *page)
 128 {
 129         return mpage_readpage(page, nilfs_get_block);
 130 }
 131
 132 /**
 133  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 134  * address_space_operations.
 135  * @file - file struct of the file to be read
 136  * @mapping - address_space struct used for reading multiple pages
 137  * @pages - the pages to be read
 138  * @nr_pages - number of pages to be read
 139  */
 140 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 141                            struct list_head *pages, unsigned nr_pages)
 142 {
 143         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 144 }
 145
 146 static int nilfs_writepages(struct address_space *mapping,
 147                             struct writeback_control *wbc)
 148 {
 149         struct inode *inode = mapping->host;
 150         int err = 0;
 151
 152         if (wbc->sync_mode == WB_SYNC_ALL)
 153                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 154                                                     wbc->range_start,
 155                                                     wbc->range_end);
 156         return err;
 157 }
 158
 159 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 160 {
 161         struct inode *inode = page->mapping->host;
 162         int err;
 163
 164         redirty_page_for_writepage(wbc, page);
 165         unlock_page(page);
 166
 167         if (wbc->sync_mode == WB_SYNC_ALL) {
 168                 err = nilfs_construct_segment(inode->i_sb);
 169                 if (unlikely(err))
 170                         return err;
 171         } else if (wbc->for_reclaim)
 172                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 173
 174         return 0;
 175 }
 176
 177 static int nilfs_set_page_dirty(struct page *page)
 178 {
 179         int ret = __set_page_dirty_buffers(page);
 180
 181         if (ret) {
 182                 struct inode *inode = page->mapping->host;
 183                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 184                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 185
 186                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 187         }
 188         return ret;
 189 }
 190
 191 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 192                              loff_t pos, unsigned len, unsigned flags,
 193                              struct page **pagep, void **fsdata)
 194
 195 {
 196         struct inode *inode = mapping->host;
 197         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 198
 199         if (unlikely(err))
 200                 return err;
 201
 202         *pagep = NULL;
 203         err = block_write_begin(file, mapping, pos, len, flags, pagep,
 204                                 fsdata, nilfs_get_block);
 205         if (unlikely(err))
 206                 nilfs_transaction_end(inode->i_sb, 0);
 207         return err;
 208 }
 209
 210 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 211                            loff_t pos, unsigned len, unsigned copied,
 212                            struct page *page, void *fsdata)
 213 {
 214         struct inode *inode = mapping->host;
 215         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 216         unsigned nr_dirty;
 217         int err;
 218
 219         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 220                                                   start + copied);
 221         copied = generic_write_end(file, mapping, pos, len, copied, page,
 222                                    fsdata);
 223         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 224         err = nilfs_transaction_end(inode->i_sb, 1);
 225         return err ? : copied;
 226 }
 227
 228 static ssize_t
 229 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 230                 loff_t offset, unsigned long nr_segs)
 231 {
 232         struct file *file = iocb->ki_filp;
 233         struct inode *inode = file->f_mapping->host;
 234         ssize_t size;
 235
 236         if (rw == WRITE)
 237                 return 0;
 238
 239         /* Needs synchronization with the cleaner */
 240         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 241                                   offset, nr_segs, nilfs_get_block, NULL);
 242         return size;
 243 }
 244
 245 struct address_space_operations nilfs_aops = {
 246         .writepage              = nilfs_writepage,
 247         .readpage               = nilfs_readpage,
 248         /* .sync_page           = nilfs_sync_page, */
 249         .writepages             = nilfs_writepages,
 250         .set_page_dirty         = nilfs_set_page_dirty,
 251         .readpages              = nilfs_readpages,
 252         .write_begin            = nilfs_write_begin,
 253         .write_end              = nilfs_write_end,
 254         /* .releasepage         = nilfs_releasepage, */
 255         .invalidatepage         = block_invalidatepage,
 256         .direct_IO              = nilfs_direct_IO,
 257 };
 258
 259 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 260 {
 261         struct super_block *sb = dir->i_sb;
 262         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 263         struct inode *inode;
 264         struct nilfs_inode_info *ii;
 265         int err = -ENOMEM;
 266         ino_t ino;
 267
 268         inode = new_inode(sb);
 269         if (unlikely(!inode))
 270                 goto failed;
 271
 272         mapping_set_gfp_mask(inode->i_mapping,
 273                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 274
 275         ii = NILFS_I(inode);
 276         ii->i_state = 1 << NILFS_I_NEW;
 277
 278         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
 279         if (unlikely(err))
 280                 goto failed_ifile_create_inode;
 281         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 282
 283         atomic_inc(&sbi->s_inodes_count);
 284
 285         inode->i_uid = current_fsuid();
 286         if (dir->i_mode & S_ISGID) {
 287                 inode->i_gid = dir->i_gid;
 288                 if (S_ISDIR(mode))
 289                         mode |= S_ISGID;
 290         } else
 291                 inode->i_gid = current_fsgid();
 292
 293         inode->i_mode = mode;
 294         inode->i_ino = ino;
 295         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 296
 297         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 298                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 299                 if (err < 0)
 300                         goto failed_bmap;
 301
 302                 set_bit(NILFS_I_BMAP, &ii->i_state);
 303                 /* No lock is needed; iget() ensures it. */
 304         }
 305
 306         ii->i_flags = NILFS_I(dir)->i_flags;
 307         if (S_ISLNK(mode))
 308                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 309         if (!S_ISDIR(mode))
 310                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 311
 312         /* ii->i_file_acl = 0; */
 313         /* ii->i_dir_acl = 0; */
 314         ii->i_dtime = 0;
 315         ii->i_dir_start_lookup = 0;
 316 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 317         ii->i_acl = NULL;
 318         ii->i_default_acl = NULL;
 319 #endif
 320         ii->i_cno = 0;
 321         nilfs_set_inode_flags(inode);
 322         spin_lock(&sbi->s_next_gen_lock);
 323         inode->i_generation = sbi->s_next_generation++;
 324         spin_unlock(&sbi->s_next_gen_lock);
 325         insert_inode_hash(inode);
 326
 327         err = nilfs_init_acl(inode, dir);
 328         if (unlikely(err))
 329                 goto failed_acl; /* never occur. When supporting
 330                                     nilfs_init_acl(), proper cancellation of
 331                                     above jobs should be considered */
 332
 333         mark_inode_dirty(inode);
 334         return inode;
 335
 336  failed_acl:
 337  failed_bmap:
 338         inode->i_nlink = 0;
 339         iput(inode);  /* raw_inode will be deleted through
 340                          generic_delete_inode() */
 341         goto failed;
 342
 343  failed_ifile_create_inode:
 344         make_bad_inode(inode);
 345         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 346                          called */
 347  failed:
 348         return ERR_PTR(err);
 349 }
 350
 351 void nilfs_free_inode(struct inode *inode)
 352 {
 353         struct super_block *sb = inode->i_sb;
 354         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 355
 356         clear_inode(inode);
 357         /* XXX: check error code? Is there any thing I can do? */
 358         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
 359         atomic_dec(&sbi->s_inodes_count);
 360 }
 361
 362 void nilfs_set_inode_flags(struct inode *inode)
 363 {
 364         unsigned int flags = NILFS_I(inode)->i_flags;
 365
 366         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 367                             S_DIRSYNC);
 368         if (flags & NILFS_SYNC_FL)
 369                 inode->i_flags |= S_SYNC;
 370         if (flags & NILFS_APPEND_FL)
 371                 inode->i_flags |= S_APPEND;
 372         if (flags & NILFS_IMMUTABLE_FL)
 373                 inode->i_flags |= S_IMMUTABLE;
 374 #ifndef NILFS_ATIME_DISABLE
 375         if (flags & NILFS_NOATIME_FL)
 376 #endif
 377                 inode->i_flags |= S_NOATIME;
 378         if (flags & NILFS_DIRSYNC_FL)
 379                 inode->i_flags |= S_DIRSYNC;
 380         mapping_set_gfp_mask(inode->i_mapping,
 381                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 382 }
 383
 384 int nilfs_read_inode_common(struct inode *inode,
 385                             struct nilfs_inode *raw_inode)
 386 {
 387         struct nilfs_inode_info *ii = NILFS_I(inode);
 388         int err;
 389
 390         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 391         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 392         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 393         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 394         inode->i_size = le64_to_cpu(raw_inode->i_size);
 395         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 396         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 397         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 398         inode->i_atime.tv_nsec = 0;
 399         inode->i_ctime.tv_nsec = 0;
 400         inode->i_mtime.tv_nsec = 0;
 401         ii->i_dtime = le64_to_cpu(raw_inode->i_dtime);
 402         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ii->i_dtime))
 403                 return -EINVAL; /* this inode is deleted */
 404
 405         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 406         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 407 #if 0
 408         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 409         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 410                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 411 #endif
 412         ii->i_cno = 0;
 413         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 414
 415         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 416             S_ISLNK(inode->i_mode)) {
 417                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 418                 if (err < 0)
 419                         return err;
 420                 set_bit(NILFS_I_BMAP, &ii->i_state);
 421                 /* No lock is needed; iget() ensures it. */
 422         }
 423         return 0;
 424 }
 425
 426 static int nilfs_read_sketch_inode(struct inode *inode)
 427 {
 428         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 429         int err = 0;
 430
 431         if (sbi->s_snapshot_cno) {
 432                 struct the_nilfs *nilfs = sbi->s_nilfs;
 433                 struct buffer_head *bh_cp;
 434                 struct nilfs_checkpoint *raw_cp;
 435
 436                 err = nilfs_cpfile_get_checkpoint(
 437                         nilfs->ns_cpfile, sbi->s_snapshot_cno, 0, &raw_cp,
 438                         &bh_cp);
 439                 if (likely(!err)) {
 440                         if (!nilfs_checkpoint_sketch(raw_cp))
 441                                 inode->i_size = 0;
 442                         nilfs_cpfile_put_checkpoint(
 443                                 nilfs->ns_cpfile, sbi->s_snapshot_cno, bh_cp);
 444                 }
 445                 inode->i_flags |= S_NOCMTIME;
 446         }
 447         return err;
 448 }
 449
 450 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 451                               struct inode *inode)
 452 {
 453         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 454         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 455         struct buffer_head *bh;
 456         struct nilfs_inode *raw_inode;
 457         int err;
 458
 459         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 460         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 461         if (unlikely(err))
 462                 goto bad_inode;
 463
 464         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 465
 466 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 467         ii->i_acl = NILFS_ACL_NOT_CACHED;
 468         ii->i_default_acl = NILFS_ACL_NOT_CACHED;
 469 #endif
 470         if (nilfs_read_inode_common(inode, raw_inode))
 471                 goto failed_unmap;
 472
 473         if (S_ISREG(inode->i_mode)) {
 474                 inode->i_op = &nilfs_file_inode_operations;
 475                 inode->i_fop = &nilfs_file_operations;
 476                 inode->i_mapping->a_ops = &nilfs_aops;
 477                 if (unlikely(inode->i_ino == NILFS_SKETCH_INO)) {
 478                         err = nilfs_read_sketch_inode(inode);
 479                         if (unlikely(err))
 480                                 goto failed_unmap;
 481                 }
 482         } else if (S_ISDIR(inode->i_mode)) {
 483                 inode->i_op = &nilfs_dir_inode_operations;
 484                 inode->i_fop = &nilfs_dir_operations;
 485                 inode->i_mapping->a_ops = &nilfs_aops;
 486         } else if (S_ISLNK(inode->i_mode)) {
 487                 inode->i_op = &nilfs_symlink_inode_operations;
 488                 inode->i_mapping->a_ops = &nilfs_aops;
 489         } else {
 490                 inode->i_op = &nilfs_special_inode_operations;
 491                 init_special_inode(
 492                         inode, inode->i_mode,
 493                         new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 494         }
 495         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 496         brelse(bh);
 497         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 498         nilfs_set_inode_flags(inode);
 499         return 0;
 500
 501  failed_unmap:
 502         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 503         brelse(bh);
 504
 505  bad_inode:
 506         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 507         return err;
 508 }
 509
 510 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 511 {
 512         struct inode *inode;
 513         int err;
 514
 515         inode = iget_locked(sb, ino);
 516         if (unlikely(!inode))
 517                 return ERR_PTR(-ENOMEM);
 518         if (!(inode->i_state & I_NEW))
 519                 return inode;
 520
 521         err = __nilfs_read_inode(sb, ino, inode);
 522         if (unlikely(err)) {
 523                 iget_failed(inode);
 524                 return ERR_PTR(err);
 525         }
 526         unlock_new_inode(inode);
 527         return inode;
 528 }
 529
 530 void nilfs_write_inode_common(struct inode *inode,
 531                               struct nilfs_inode *raw_inode, int has_bmap)
 532 {
 533         struct nilfs_inode_info *ii = NILFS_I(inode);
 534
 535         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 536         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 537         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 538         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 539         raw_inode->i_size = cpu_to_le64(inode->i_size);
 540         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 541         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 542         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 543
 544         raw_inode->i_dtime = cpu_to_le64(ii->i_dtime);
 545         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 546         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 547
 548         if (has_bmap)
 549                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 550         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 551                 raw_inode->i_device_code =
 552                         cpu_to_le64(new_encode_dev(inode->i_rdev));
 553         /* When extending inode, nilfs->ns_inode_size should be checked
 554            for substitutions of appended fields */
 555 }
 556
 557 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 558 {
 559         ino_t ino = inode->i_ino;
 560         struct nilfs_inode_info *ii = NILFS_I(inode);
 561         struct super_block *sb = inode->i_sb;
 562         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 563         struct nilfs_inode *raw_inode;
 564
 565         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
 566
 567         /* The buffer is guarded with lock_buffer() by the caller */
 568         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 569                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 570         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 571
 572         nilfs_write_inode_common(inode, raw_inode, 0);
 573                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 574                    deadlock of bmap. This delays update of i_bmap to just
 575                    before writing */
 576         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 577 }
 578
 579 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 580
 581 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 582                                 unsigned long from)
 583 {
 584         unsigned long b;
 585         int ret;
 586
 587         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 588                 return;
 589  repeat:
 590         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 591         if (ret == -ENOENT)
 592                 return;
 593         else if (ret < 0)
 594                 goto failed;
 595
 596         if (b < from)
 597                 return;
 598
 599         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 600         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 601         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 602         if (!ret || (ret == -ENOMEM &&
 603                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 604                 goto repeat;
 605
 606  failed:
 607         if (ret == -EINVAL)
 608                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 609                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 610         else
 611                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 612                               "failed to truncate bmap (ino=%lu, err=%d)",
 613                               ii->vfs_inode.i_ino, ret);
 614 }
 615
 616 void nilfs_truncate(struct inode *inode)
 617 {
 618         unsigned long blkoff;
 619         unsigned int blocksize;
 620         struct nilfs_transaction_info ti;
 621         struct super_block *sb = inode->i_sb;
 622         struct nilfs_inode_info *ii = NILFS_I(inode);
 623         int ret;
 624
 625         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 626                 return;
 627         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 628                 return;
 629
 630         blocksize = sb->s_blocksize;
 631         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 632         ret = nilfs_transaction_begin(sb, &ti, 0);
 633         BUG_ON(ret);
 634
 635         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 636
 637         nilfs_truncate_bmap(ii, blkoff);
 638
 639         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 640         if (IS_SYNC(inode))
 641                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 642
 643         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 644         nilfs_transaction_end(sb, 1);
 645         /* May construct a logical segment and may fail in sync mode.
 646            But truncate has no return value. */
 647 }
 648
 649 void nilfs_delete_inode(struct inode *inode)
 650 {
 651         struct nilfs_transaction_info ti;
 652         struct super_block *sb = inode->i_sb;
 653         struct nilfs_inode_info *ii = NILFS_I(inode);
 654         int err;
 655
 656         if (unlikely(is_bad_inode(inode))) {
 657                 if (inode->i_data.nrpages)
 658                         truncate_inode_pages(&inode->i_data, 0);
 659                 clear_inode(inode);
 660                 return;
 661         }
 662         err = nilfs_transaction_begin(sb, &ti, 0);
 663         BUG_ON(err);
 664         if (inode->i_data.nrpages)
 665                 truncate_inode_pages(&inode->i_data, 0);
 666
 667         nilfs_truncate_bmap(ii, 0);
 668         nilfs_free_inode(inode);
 669         /* nilfs_free_inode() marks inode buffer dirty */
 670         if (IS_SYNC(inode))
 671                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 672         nilfs_transaction_end(sb, 1);
 673         /* May construct a logical segment and may fail in sync mode.
 674            But delete_inode has no return value. */
 675 }
 676
 677 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 678 {
 679         struct nilfs_transaction_info ti;
 680         struct inode *inode = dentry->d_inode;
 681         struct super_block *sb = inode->i_sb;
 682         int err, err2;
 683
 684         err = inode_change_ok(inode, iattr);
 685         if (err)
 686                 return err;
 687
 688         err = nilfs_transaction_begin(sb, &ti, 0);
 689         if (unlikely(err))
 690                 return err;
 691         err = inode_setattr(inode, iattr);
 692         if (!err && (iattr->ia_valid & ATTR_MODE))
 693                 err = nilfs_acl_chmod(inode);
 694         err2 = nilfs_transaction_end(sb, 1);
 695         return err ? : err2;
 696 }
 697
 698 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 699                            struct buffer_head **pbh)
 700 {
 701         struct nilfs_inode_info *ii = NILFS_I(inode);
 702         int err;
 703
 704         spin_lock(&sbi->s_inode_lock);
 705         /* Caller of this function MUST lock s_inode_lock */
 706         if (ii->i_bh == NULL) {
 707                 spin_unlock(&sbi->s_inode_lock);
 708                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
 709                                                   pbh);
 710                 if (unlikely(err))
 711                         return err;
 712                 spin_lock(&sbi->s_inode_lock);
 713                 if (ii->i_bh == NULL)
 714                         ii->i_bh = *pbh;
 715                 else {
 716                         brelse(*pbh);
 717                         *pbh = ii->i_bh;
 718                 }
 719         } else
 720                 *pbh = ii->i_bh;
 721
 722         get_bh(*pbh);
 723         spin_unlock(&sbi->s_inode_lock);
 724         return 0;
 725 }
 726
 727 int nilfs_inode_dirty(struct inode *inode)
 728 {
 729         struct nilfs_inode_info *ii = NILFS_I(inode);
 730         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 731         int ret = 0;
 732
 733         if (!list_empty(&ii->i_dirty)) {
 734                 spin_lock(&sbi->s_inode_lock);
 735                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 736                         test_bit(NILFS_I_BUSY, &ii->i_state);
 737                 spin_unlock(&sbi->s_inode_lock);
 738         }
 739         return ret;
 740 }
 741
 742 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 743                          unsigned nr_dirty)
 744 {
 745         struct nilfs_inode_info *ii = NILFS_I(inode);
 746
 747         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 748
 749         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state) ||
 750             unlikely(inode->i_ino == NILFS_SKETCH_INO))
 751                 return 0;
 752
 753         spin_lock(&sbi->s_inode_lock);
 754         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 755             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 756                 /* Because this routine may race with nilfs_dispose_list(),
 757                    we have to check NILFS_I_QUEUED here, too. */
 758                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 759                         /* This will happen when somebody is freeing
 760                            this inode. */
 761                         nilfs_warning(sbi->s_super, __func__,
 762                                       "cannot get inode (ino=%lu)\n",
 763                                       inode->i_ino);
 764                         spin_unlock(&sbi->s_inode_lock);
 765                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 766                                            freeing inode */
 767                 }
 768                 list_del(&ii->i_dirty);
 769                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 770                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 771         }
 772         spin_unlock(&sbi->s_inode_lock);
 773         return 0;
 774 }
 775
 776 int nilfs_mark_inode_dirty(struct inode *inode)
 777 {
 778         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 779         struct buffer_head *ibh;
 780         int err;
 781
 782         err = nilfs_load_inode_block(sbi, inode, &ibh);
 783         if (unlikely(err)) {
 784                 nilfs_warning(inode->i_sb, __func__,
 785                               "failed to reget inode block.\n");
 786                 return err;
 787         }
 788         lock_buffer(ibh);
 789         nilfs_update_inode(inode, ibh);
 790         unlock_buffer(ibh);
 791         nilfs_mdt_mark_buffer_dirty(ibh);
 792         nilfs_mdt_mark_dirty(sbi->s_ifile);
 793         brelse(ibh);
 794         return 0;
 795 }
 796
 797 /**
 798  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 799  * @inode: inode of the file to be registered.
 800  *
 801  * nilfs_dirty_inode() loads a inode block containing the specified
 802  * @inode and copies data from a nilfs_inode to a corresponding inode
 803  * entry in the inode block. This operation is excluded from the segment
 804  * construction. This function can be called both as a single operation
 805  * and as a part of indivisible file operations.
 806  */
 807 void nilfs_dirty_inode(struct inode *inode)
 808 {
 809         struct nilfs_transaction_info ti;
 810
 811         if (is_bad_inode(inode)) {
 812                 nilfs_warning(inode->i_sb, __func__,
 813                               "tried to mark bad_inode dirty. ignored.\n");
 814                 dump_stack();
 815                 return;
 816         }
 817         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 818         if (likely(inode->i_ino != NILFS_SKETCH_INO))
 819                 nilfs_mark_inode_dirty(inode);
 820         nilfs_transaction_end(inode->i_sb, 1); /* never fails */
 821 }