git.oblomov.eu Git - linux-2.6/blob - 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_sketch_inode(struct inode *inode)
 422 {
 423         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 424         int err = 0;
 425
 426         if (sbi->s_snapshot_cno) {
 427                 struct the_nilfs *nilfs = sbi->s_nilfs;
 428                 struct buffer_head *bh_cp;
 429                 struct nilfs_checkpoint *raw_cp;
 430
 431                 err = nilfs_cpfile_get_checkpoint(
 432                         nilfs->ns_cpfile, sbi->s_snapshot_cno, 0, &raw_cp,
 433                         &bh_cp);
 434                 if (likely(!err)) {
 435                         if (!nilfs_checkpoint_sketch(raw_cp))
 436                                 inode->i_size = 0;
 437                         nilfs_cpfile_put_checkpoint(
 438                                 nilfs->ns_cpfile, sbi->s_snapshot_cno, bh_cp);
 439                 }
 440                 inode->i_flags |= S_NOCMTIME;
 441         }
 442         return err;
 443 }
 444
 445 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 446                               struct inode *inode)
 447 {
 448         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 449         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 450         struct buffer_head *bh;
 451         struct nilfs_inode *raw_inode;
 452         int err;
 453
 454         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 455         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 456         if (unlikely(err))
 457                 goto bad_inode;
 458
 459         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 460
 461 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 462         ii->i_acl = NILFS_ACL_NOT_CACHED;
 463         ii->i_default_acl = NILFS_ACL_NOT_CACHED;
 464 #endif
 465         if (nilfs_read_inode_common(inode, raw_inode))
 466                 goto failed_unmap;
 467
 468         if (S_ISREG(inode->i_mode)) {
 469                 inode->i_op = &nilfs_file_inode_operations;
 470                 inode->i_fop = &nilfs_file_operations;
 471                 inode->i_mapping->a_ops = &nilfs_aops;
 472                 if (unlikely(inode->i_ino == NILFS_SKETCH_INO)) {
 473                         err = nilfs_read_sketch_inode(inode);
 474                         if (unlikely(err))
 475                                 goto failed_unmap;
 476                 }
 477         } else if (S_ISDIR(inode->i_mode)) {
 478                 inode->i_op = &nilfs_dir_inode_operations;
 479                 inode->i_fop = &nilfs_dir_operations;
 480                 inode->i_mapping->a_ops = &nilfs_aops;
 481         } else if (S_ISLNK(inode->i_mode)) {
 482                 inode->i_op = &nilfs_symlink_inode_operations;
 483                 inode->i_mapping->a_ops = &nilfs_aops;
 484         } else {
 485                 inode->i_op = &nilfs_special_inode_operations;
 486                 init_special_inode(
 487                         inode, inode->i_mode,
 488                         new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 489         }
 490         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 491         brelse(bh);
 492         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 493         nilfs_set_inode_flags(inode);
 494         return 0;
 495
 496  failed_unmap:
 497         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 498         brelse(bh);
 499
 500  bad_inode:
 501         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 502         return err;
 503 }
 504
 505 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 506 {
 507         struct inode *inode;
 508         int err;
 509
 510         inode = iget_locked(sb, ino);
 511         if (unlikely(!inode))
 512                 return ERR_PTR(-ENOMEM);
 513         if (!(inode->i_state & I_NEW))
 514                 return inode;
 515
 516         err = __nilfs_read_inode(sb, ino, inode);
 517         if (unlikely(err)) {
 518                 iget_failed(inode);
 519                 return ERR_PTR(err);
 520         }
 521         unlock_new_inode(inode);
 522         return inode;
 523 }
 524
 525 void nilfs_write_inode_common(struct inode *inode,
 526                               struct nilfs_inode *raw_inode, int has_bmap)
 527 {
 528         struct nilfs_inode_info *ii = NILFS_I(inode);
 529
 530         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 531         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 532         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 533         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 534         raw_inode->i_size = cpu_to_le64(inode->i_size);
 535         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 536         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 537         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 538
 539         raw_inode->i_dtime = cpu_to_le64(ii->i_dtime);
 540         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 541         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 542
 543         if (has_bmap)
 544                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 545         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 546                 raw_inode->i_device_code =
 547                         cpu_to_le64(new_encode_dev(inode->i_rdev));
 548         /* When extending inode, nilfs->ns_inode_size should be checked
 549            for substitutions of appended fields */
 550 }
 551
 552 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 553 {
 554         ino_t ino = inode->i_ino;
 555         struct nilfs_inode_info *ii = NILFS_I(inode);
 556         struct super_block *sb = inode->i_sb;
 557         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 558         struct nilfs_inode *raw_inode;
 559
 560         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
 561
 562         /* The buffer is guarded with lock_buffer() by the caller */
 563         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 564                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 565         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 566
 567         nilfs_write_inode_common(inode, raw_inode, 0);
 568                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 569                    deadlock of bmap. This delays update of i_bmap to just
 570                    before writing */
 571         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 572 }
 573
 574 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 575
 576 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 577                                 unsigned long from)
 578 {
 579         unsigned long b;
 580         int ret;
 581
 582         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 583                 return;
 584  repeat:
 585         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 586         if (ret == -ENOENT)
 587                 return;
 588         else if (ret < 0)
 589                 goto failed;
 590
 591         if (b < from)
 592                 return;
 593
 594         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 595         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 596         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 597         if (!ret || (ret == -ENOMEM &&
 598                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 599                 goto repeat;
 600
 601  failed:
 602         if (ret == -EINVAL)
 603                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 604                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 605         else
 606                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 607                               "failed to truncate bmap (ino=%lu, err=%d)",
 608                               ii->vfs_inode.i_ino, ret);
 609 }
 610
 611 void nilfs_truncate(struct inode *inode)
 612 {
 613         unsigned long blkoff;
 614         unsigned int blocksize;
 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 (!test_bit(NILFS_I_BMAP, &ii->i_state))
 620                 return;
 621         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 622                 return;
 623
 624         blocksize = sb->s_blocksize;
 625         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 626         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 627
 628         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 629
 630         nilfs_truncate_bmap(ii, blkoff);
 631
 632         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 633         if (IS_SYNC(inode))
 634                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 635
 636         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 637         nilfs_transaction_commit(sb);
 638         /* May construct a logical segment and may fail in sync mode.
 639            But truncate has no return value. */
 640 }
 641
 642 void nilfs_delete_inode(struct inode *inode)
 643 {
 644         struct nilfs_transaction_info ti;
 645         struct super_block *sb = inode->i_sb;
 646         struct nilfs_inode_info *ii = NILFS_I(inode);
 647
 648         if (unlikely(is_bad_inode(inode))) {
 649                 if (inode->i_data.nrpages)
 650                         truncate_inode_pages(&inode->i_data, 0);
 651                 clear_inode(inode);
 652                 return;
 653         }
 654         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 655
 656         if (inode->i_data.nrpages)
 657                 truncate_inode_pages(&inode->i_data, 0);
 658
 659         nilfs_truncate_bmap(ii, 0);
 660         nilfs_free_inode(inode);
 661         /* nilfs_free_inode() marks inode buffer dirty */
 662         if (IS_SYNC(inode))
 663                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 664         nilfs_transaction_commit(sb);
 665         /* May construct a logical segment and may fail in sync mode.
 666            But delete_inode has no return value. */
 667 }
 668
 669 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 670 {
 671         struct nilfs_transaction_info ti;
 672         struct inode *inode = dentry->d_inode;
 673         struct super_block *sb = inode->i_sb;
 674         int err;
 675
 676         err = inode_change_ok(inode, iattr);
 677         if (err)
 678                 return err;
 679
 680         err = nilfs_transaction_begin(sb, &ti, 0);
 681         if (unlikely(err))
 682                 return err;
 683         err = inode_setattr(inode, iattr);
 684         if (!err && (iattr->ia_valid & ATTR_MODE))
 685                 err = nilfs_acl_chmod(inode);
 686         if (likely(!err))
 687                 err = nilfs_transaction_commit(sb);
 688         else
 689                 nilfs_transaction_abort(sb);
 690
 691         return err;
 692 }
 693
 694 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 695                            struct buffer_head **pbh)
 696 {
 697         struct nilfs_inode_info *ii = NILFS_I(inode);
 698         int err;
 699
 700         spin_lock(&sbi->s_inode_lock);
 701         /* Caller of this function MUST lock s_inode_lock */
 702         if (ii->i_bh == NULL) {
 703                 spin_unlock(&sbi->s_inode_lock);
 704                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
 705                                                   pbh);
 706                 if (unlikely(err))
 707                         return err;
 708                 spin_lock(&sbi->s_inode_lock);
 709                 if (ii->i_bh == NULL)
 710                         ii->i_bh = *pbh;
 711                 else {
 712                         brelse(*pbh);
 713                         *pbh = ii->i_bh;
 714                 }
 715         } else
 716                 *pbh = ii->i_bh;
 717
 718         get_bh(*pbh);
 719         spin_unlock(&sbi->s_inode_lock);
 720         return 0;
 721 }
 722
 723 int nilfs_inode_dirty(struct inode *inode)
 724 {
 725         struct nilfs_inode_info *ii = NILFS_I(inode);
 726         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 727         int ret = 0;
 728
 729         if (!list_empty(&ii->i_dirty)) {
 730                 spin_lock(&sbi->s_inode_lock);
 731                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 732                         test_bit(NILFS_I_BUSY, &ii->i_state);
 733                 spin_unlock(&sbi->s_inode_lock);
 734         }
 735         return ret;
 736 }
 737
 738 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 739                          unsigned nr_dirty)
 740 {
 741         struct nilfs_inode_info *ii = NILFS_I(inode);
 742
 743         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 744
 745         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state) ||
 746             unlikely(inode->i_ino == NILFS_SKETCH_INO))
 747                 return 0;
 748
 749         spin_lock(&sbi->s_inode_lock);
 750         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 751             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 752                 /* Because this routine may race with nilfs_dispose_list(),
 753                    we have to check NILFS_I_QUEUED here, too. */
 754                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 755                         /* This will happen when somebody is freeing
 756                            this inode. */
 757                         nilfs_warning(sbi->s_super, __func__,
 758                                       "cannot get inode (ino=%lu)\n",
 759                                       inode->i_ino);
 760                         spin_unlock(&sbi->s_inode_lock);
 761                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 762                                            freeing inode */
 763                 }
 764                 list_del(&ii->i_dirty);
 765                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 766                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 767         }
 768         spin_unlock(&sbi->s_inode_lock);
 769         return 0;
 770 }
 771
 772 int nilfs_mark_inode_dirty(struct inode *inode)
 773 {
 774         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 775         struct buffer_head *ibh;
 776         int err;
 777
 778         err = nilfs_load_inode_block(sbi, inode, &ibh);
 779         if (unlikely(err)) {
 780                 nilfs_warning(inode->i_sb, __func__,
 781                               "failed to reget inode block.\n");
 782                 return err;
 783         }
 784         lock_buffer(ibh);
 785         nilfs_update_inode(inode, ibh);
 786         unlock_buffer(ibh);
 787         nilfs_mdt_mark_buffer_dirty(ibh);
 788         nilfs_mdt_mark_dirty(sbi->s_ifile);
 789         brelse(ibh);
 790         return 0;
 791 }
 792
 793 /**
 794  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 795  * @inode: inode of the file to be registered.
 796  *
 797  * nilfs_dirty_inode() loads a inode block containing the specified
 798  * @inode and copies data from a nilfs_inode to a corresponding inode
 799  * entry in the inode block. This operation is excluded from the segment
 800  * construction. This function can be called both as a single operation
 801  * and as a part of indivisible file operations.
 802  */
 803 void nilfs_dirty_inode(struct inode *inode)
 804 {
 805         struct nilfs_transaction_info ti;
 806
 807         if (is_bad_inode(inode)) {
 808                 nilfs_warning(inode->i_sb, __func__,
 809                               "tried to mark bad_inode dirty. ignored.\n");
 810                 dump_stack();
 811                 return;
 812         }
 813         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 814         if (likely(inode->i_ino != NILFS_SKETCH_INO))
 815                 nilfs_mark_inode_dirty(inode);
 816         nilfs_transaction_commit(inode->i_sb); /* never fails */
 817 }