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                 if (unlikely(err != 0)) {
  81                         if (err == -EEXIST) {
  82                                 /*
  83                                  * The get_block() function could be called
  84                                  * from multiple callers for an inode.
  85                                  * However, the page having this block must
  86                                  * be locked in this case.
  87                                  */
  88                                 printk(KERN_ERR
  89                                        "nilfs_get_block: a race condition "
  90                                        "while inserting a data block. "
  91                                        "(inode number=%lu, file block "
  92                                        "offset=%llu)\n",
  93                                        inode->i_ino,
  94                                        (unsigned long long)blkoff);
  95                                 BUG();
  96                         } else if (err == -EINVAL) {
  97                                 nilfs_error(inode->i_sb, __func__,
  98                                             "broken bmap (inode=%lu)\n",
  99                                             inode->i_ino);
 100                                 err = -EIO;
 101                         }
 102                         nilfs_transaction_abort(inode->i_sb);
 103                         goto out;
 104                 }
 105                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 106                 /* Error handling should be detailed */
 107                 set_buffer_new(bh_result);
 108                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 109                                                       to proper value */
 110         } else if (ret == -ENOENT) {
 111                 /* not found is not error (e.g. hole); must return without
 112                    the mapped state flag. */
 113                 ;
 114         } else {
 115                 err = ret;
 116         }
 117
 118  out:
 119         return err;
 120 }
 121
 122 /**
 123  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 124  * address_space_operations.
 125  * @file - file struct of the file to be read
 126  * @page - the page to be read
 127  */
 128 static int nilfs_readpage(struct file *file, struct page *page)
 129 {
 130         return mpage_readpage(page, nilfs_get_block);
 131 }
 132
 133 /**
 134  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 135  * address_space_operations.
 136  * @file - file struct of the file to be read
 137  * @mapping - address_space struct used for reading multiple pages
 138  * @pages - the pages to be read
 139  * @nr_pages - number of pages to be read
 140  */
 141 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 142                            struct list_head *pages, unsigned nr_pages)
 143 {
 144         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 145 }
 146
 147 static int nilfs_writepages(struct address_space *mapping,
 148                             struct writeback_control *wbc)
 149 {
 150         struct inode *inode = mapping->host;
 151         int err = 0;
 152
 153         if (wbc->sync_mode == WB_SYNC_ALL)
 154                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 155                                                     wbc->range_start,
 156                                                     wbc->range_end);
 157         return err;
 158 }
 159
 160 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 161 {
 162         struct inode *inode = page->mapping->host;
 163         int err;
 164
 165         redirty_page_for_writepage(wbc, page);
 166         unlock_page(page);
 167
 168         if (wbc->sync_mode == WB_SYNC_ALL) {
 169                 err = nilfs_construct_segment(inode->i_sb);
 170                 if (unlikely(err))
 171                         return err;
 172         } else if (wbc->for_reclaim)
 173                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 174
 175         return 0;
 176 }
 177
 178 static int nilfs_set_page_dirty(struct page *page)
 179 {
 180         int ret = __set_page_dirty_buffers(page);
 181
 182         if (ret) {
 183                 struct inode *inode = page->mapping->host;
 184                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 185                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 186
 187                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 188         }
 189         return ret;
 190 }
 191
 192 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 193                              loff_t pos, unsigned len, unsigned flags,
 194                              struct page **pagep, void **fsdata)
 195
 196 {
 197         struct inode *inode = mapping->host;
 198         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 199
 200         if (unlikely(err))
 201                 return err;
 202
 203         *pagep = NULL;
 204         err = block_write_begin(file, mapping, pos, len, flags, pagep,
 205                                 fsdata, nilfs_get_block);
 206         if (unlikely(err))
 207                 nilfs_transaction_abort(inode->i_sb);
 208         return err;
 209 }
 210
 211 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 212                            loff_t pos, unsigned len, unsigned copied,
 213                            struct page *page, void *fsdata)
 214 {
 215         struct inode *inode = mapping->host;
 216         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 217         unsigned nr_dirty;
 218         int err;
 219
 220         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 221                                                   start + copied);
 222         copied = generic_write_end(file, mapping, pos, len, copied, page,
 223                                    fsdata);
 224         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 225         err = nilfs_transaction_commit(inode->i_sb);
 226         return err ? : copied;
 227 }
 228
 229 static ssize_t
 230 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 231                 loff_t offset, unsigned long nr_segs)
 232 {
 233         struct file *file = iocb->ki_filp;
 234         struct inode *inode = file->f_mapping->host;
 235         ssize_t size;
 236
 237         if (rw == WRITE)
 238                 return 0;
 239
 240         /* Needs synchronization with the cleaner */
 241         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 242                                   offset, nr_segs, nilfs_get_block, NULL);
 243         return size;
 244 }
 245
 246 struct address_space_operations nilfs_aops = {
 247         .writepage              = nilfs_writepage,
 248         .readpage               = nilfs_readpage,
 249         /* .sync_page           = nilfs_sync_page, */
 250         .writepages             = nilfs_writepages,
 251         .set_page_dirty         = nilfs_set_page_dirty,
 252         .readpages              = nilfs_readpages,
 253         .write_begin            = nilfs_write_begin,
 254         .write_end              = nilfs_write_end,
 255         /* .releasepage         = nilfs_releasepage, */
 256         .invalidatepage         = block_invalidatepage,
 257         .direct_IO              = nilfs_direct_IO,
 258 };
 259
 260 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 261 {
 262         struct super_block *sb = dir->i_sb;
 263         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 264         struct inode *inode;
 265         struct nilfs_inode_info *ii;
 266         int err = -ENOMEM;
 267         ino_t ino;
 268
 269         inode = new_inode(sb);
 270         if (unlikely(!inode))
 271                 goto failed;
 272
 273         mapping_set_gfp_mask(inode->i_mapping,
 274                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 275
 276         ii = NILFS_I(inode);
 277         ii->i_state = 1 << NILFS_I_NEW;
 278
 279         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
 280         if (unlikely(err))
 281                 goto failed_ifile_create_inode;
 282         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 283
 284         atomic_inc(&sbi->s_inodes_count);
 285
 286         inode->i_uid = current_fsuid();
 287         if (dir->i_mode & S_ISGID) {
 288                 inode->i_gid = dir->i_gid;
 289                 if (S_ISDIR(mode))
 290                         mode |= S_ISGID;
 291         } else
 292                 inode->i_gid = current_fsgid();
 293
 294         inode->i_mode = mode;
 295         inode->i_ino = ino;
 296         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 297
 298         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 299                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 300                 if (err < 0)
 301                         goto failed_bmap;
 302
 303                 set_bit(NILFS_I_BMAP, &ii->i_state);
 304                 /* No lock is needed; iget() ensures it. */
 305         }
 306
 307         ii->i_flags = NILFS_I(dir)->i_flags;
 308         if (S_ISLNK(mode))
 309                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 310         if (!S_ISDIR(mode))
 311                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 312
 313         /* ii->i_file_acl = 0; */
 314         /* ii->i_dir_acl = 0; */
 315         ii->i_dtime = 0;
 316         ii->i_dir_start_lookup = 0;
 317 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 318         ii->i_acl = NULL;
 319         ii->i_default_acl = NULL;
 320 #endif
 321         ii->i_cno = 0;
 322         nilfs_set_inode_flags(inode);
 323         spin_lock(&sbi->s_next_gen_lock);
 324         inode->i_generation = sbi->s_next_generation++;
 325         spin_unlock(&sbi->s_next_gen_lock);
 326         insert_inode_hash(inode);
 327
 328         err = nilfs_init_acl(inode, dir);
 329         if (unlikely(err))
 330                 goto failed_acl; /* never occur. When supporting
 331                                     nilfs_init_acl(), proper cancellation of
 332                                     above jobs should be considered */
 333
 334         mark_inode_dirty(inode);
 335         return inode;
 336
 337  failed_acl:
 338  failed_bmap:
 339         inode->i_nlink = 0;
 340         iput(inode);  /* raw_inode will be deleted through
 341                          generic_delete_inode() */
 342         goto failed;
 343
 344  failed_ifile_create_inode:
 345         make_bad_inode(inode);
 346         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 347                          called */
 348  failed:
 349         return ERR_PTR(err);
 350 }
 351
 352 void nilfs_free_inode(struct inode *inode)
 353 {
 354         struct super_block *sb = inode->i_sb;
 355         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 356
 357         clear_inode(inode);
 358         /* XXX: check error code? Is there any thing I can do? */
 359         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
 360         atomic_dec(&sbi->s_inodes_count);
 361 }
 362
 363 void nilfs_set_inode_flags(struct inode *inode)
 364 {
 365         unsigned int flags = NILFS_I(inode)->i_flags;
 366
 367         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 368                             S_DIRSYNC);
 369         if (flags & NILFS_SYNC_FL)
 370                 inode->i_flags |= S_SYNC;
 371         if (flags & NILFS_APPEND_FL)
 372                 inode->i_flags |= S_APPEND;
 373         if (flags & NILFS_IMMUTABLE_FL)
 374                 inode->i_flags |= S_IMMUTABLE;
 375 #ifndef NILFS_ATIME_DISABLE
 376         if (flags & NILFS_NOATIME_FL)
 377 #endif
 378                 inode->i_flags |= S_NOATIME;
 379         if (flags & NILFS_DIRSYNC_FL)
 380                 inode->i_flags |= S_DIRSYNC;
 381         mapping_set_gfp_mask(inode->i_mapping,
 382                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 383 }
 384
 385 int nilfs_read_inode_common(struct inode *inode,
 386                             struct nilfs_inode *raw_inode)
 387 {
 388         struct nilfs_inode_info *ii = NILFS_I(inode);
 389         int err;
 390
 391         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 392         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 393         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 394         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 395         inode->i_size = le64_to_cpu(raw_inode->i_size);
 396         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 397         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 398         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 399         inode->i_atime.tv_nsec = 0;
 400         inode->i_ctime.tv_nsec = 0;
 401         inode->i_mtime.tv_nsec = 0;
 402         ii->i_dtime = le64_to_cpu(raw_inode->i_dtime);
 403         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ii->i_dtime))
 404                 return -EINVAL; /* this inode is deleted */
 405
 406         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 407         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 408 #if 0
 409         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 410         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 411                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 412 #endif
 413         ii->i_cno = 0;
 414         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 415
 416         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 417             S_ISLNK(inode->i_mode)) {
 418                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 419                 if (err < 0)
 420                         return err;
 421                 set_bit(NILFS_I_BMAP, &ii->i_state);
 422                 /* No lock is needed; iget() ensures it. */
 423         }
 424         return 0;
 425 }
 426
 427 static int nilfs_read_sketch_inode(struct inode *inode)
 428 {
 429         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 430         int err = 0;
 431
 432         if (sbi->s_snapshot_cno) {
 433                 struct the_nilfs *nilfs = sbi->s_nilfs;
 434                 struct buffer_head *bh_cp;
 435                 struct nilfs_checkpoint *raw_cp;
 436
 437                 err = nilfs_cpfile_get_checkpoint(
 438                         nilfs->ns_cpfile, sbi->s_snapshot_cno, 0, &raw_cp,
 439                         &bh_cp);
 440                 if (likely(!err)) {
 441                         if (!nilfs_checkpoint_sketch(raw_cp))
 442                                 inode->i_size = 0;
 443                         nilfs_cpfile_put_checkpoint(
 444                                 nilfs->ns_cpfile, sbi->s_snapshot_cno, bh_cp);
 445                 }
 446                 inode->i_flags |= S_NOCMTIME;
 447         }
 448         return err;
 449 }
 450
 451 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 452                               struct inode *inode)
 453 {
 454         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 455         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 456         struct buffer_head *bh;
 457         struct nilfs_inode *raw_inode;
 458         int err;
 459
 460         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 461         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 462         if (unlikely(err))
 463                 goto bad_inode;
 464
 465         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 466
 467 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 468         ii->i_acl = NILFS_ACL_NOT_CACHED;
 469         ii->i_default_acl = NILFS_ACL_NOT_CACHED;
 470 #endif
 471         if (nilfs_read_inode_common(inode, raw_inode))
 472                 goto failed_unmap;
 473
 474         if (S_ISREG(inode->i_mode)) {
 475                 inode->i_op = &nilfs_file_inode_operations;
 476                 inode->i_fop = &nilfs_file_operations;
 477                 inode->i_mapping->a_ops = &nilfs_aops;
 478                 if (unlikely(inode->i_ino == NILFS_SKETCH_INO)) {
 479                         err = nilfs_read_sketch_inode(inode);
 480                         if (unlikely(err))
 481                                 goto failed_unmap;
 482                 }
 483         } else if (S_ISDIR(inode->i_mode)) {
 484                 inode->i_op = &nilfs_dir_inode_operations;
 485                 inode->i_fop = &nilfs_dir_operations;
 486                 inode->i_mapping->a_ops = &nilfs_aops;
 487         } else if (S_ISLNK(inode->i_mode)) {
 488                 inode->i_op = &nilfs_symlink_inode_operations;
 489                 inode->i_mapping->a_ops = &nilfs_aops;
 490         } else {
 491                 inode->i_op = &nilfs_special_inode_operations;
 492                 init_special_inode(
 493                         inode, inode->i_mode,
 494                         new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 495         }
 496         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 497         brelse(bh);
 498         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 499         nilfs_set_inode_flags(inode);
 500         return 0;
 501
 502  failed_unmap:
 503         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 504         brelse(bh);
 505
 506  bad_inode:
 507         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 508         return err;
 509 }
 510
 511 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 512 {
 513         struct inode *inode;
 514         int err;
 515
 516         inode = iget_locked(sb, ino);
 517         if (unlikely(!inode))
 518                 return ERR_PTR(-ENOMEM);
 519         if (!(inode->i_state & I_NEW))
 520                 return inode;
 521
 522         err = __nilfs_read_inode(sb, ino, inode);
 523         if (unlikely(err)) {
 524                 iget_failed(inode);
 525                 return ERR_PTR(err);
 526         }
 527         unlock_new_inode(inode);
 528         return inode;
 529 }
 530
 531 void nilfs_write_inode_common(struct inode *inode,
 532                               struct nilfs_inode *raw_inode, int has_bmap)
 533 {
 534         struct nilfs_inode_info *ii = NILFS_I(inode);
 535
 536         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 537         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 538         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 539         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 540         raw_inode->i_size = cpu_to_le64(inode->i_size);
 541         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 542         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 543         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 544
 545         raw_inode->i_dtime = cpu_to_le64(ii->i_dtime);
 546         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 547         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 548
 549         if (has_bmap)
 550                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 551         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 552                 raw_inode->i_device_code =
 553                         cpu_to_le64(new_encode_dev(inode->i_rdev));
 554         /* When extending inode, nilfs->ns_inode_size should be checked
 555            for substitutions of appended fields */
 556 }
 557
 558 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 559 {
 560         ino_t ino = inode->i_ino;
 561         struct nilfs_inode_info *ii = NILFS_I(inode);
 562         struct super_block *sb = inode->i_sb;
 563         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 564         struct nilfs_inode *raw_inode;
 565
 566         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
 567
 568         /* The buffer is guarded with lock_buffer() by the caller */
 569         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 570                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 571         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 572
 573         nilfs_write_inode_common(inode, raw_inode, 0);
 574                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 575                    deadlock of bmap. This delays update of i_bmap to just
 576                    before writing */
 577         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 578 }
 579
 580 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 581
 582 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 583                                 unsigned long from)
 584 {
 585         unsigned long b;
 586         int ret;
 587
 588         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 589                 return;
 590  repeat:
 591         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 592         if (ret == -ENOENT)
 593                 return;
 594         else if (ret < 0)
 595                 goto failed;
 596
 597         if (b < from)
 598                 return;
 599
 600         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 601         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 602         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 603         if (!ret || (ret == -ENOMEM &&
 604                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 605                 goto repeat;
 606
 607  failed:
 608         if (ret == -EINVAL)
 609                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 610                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 611         else
 612                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 613                               "failed to truncate bmap (ino=%lu, err=%d)",
 614                               ii->vfs_inode.i_ino, ret);
 615 }
 616
 617 void nilfs_truncate(struct inode *inode)
 618 {
 619         unsigned long blkoff;
 620         unsigned int blocksize;
 621         struct nilfs_transaction_info ti;
 622         struct super_block *sb = inode->i_sb;
 623         struct nilfs_inode_info *ii = NILFS_I(inode);
 624         int ret;
 625
 626         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 627                 return;
 628         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 629                 return;
 630
 631         blocksize = sb->s_blocksize;
 632         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 633         ret = nilfs_transaction_begin(sb, &ti, 0);
 634         BUG_ON(ret);
 635
 636         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 637
 638         nilfs_truncate_bmap(ii, blkoff);
 639
 640         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 641         if (IS_SYNC(inode))
 642                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 643
 644         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 645         nilfs_transaction_commit(sb);
 646         /* May construct a logical segment and may fail in sync mode.
 647            But truncate has no return value. */
 648 }
 649
 650 void nilfs_delete_inode(struct inode *inode)
 651 {
 652         struct nilfs_transaction_info ti;
 653         struct super_block *sb = inode->i_sb;
 654         struct nilfs_inode_info *ii = NILFS_I(inode);
 655         int err;
 656
 657         if (unlikely(is_bad_inode(inode))) {
 658                 if (inode->i_data.nrpages)
 659                         truncate_inode_pages(&inode->i_data, 0);
 660                 clear_inode(inode);
 661                 return;
 662         }
 663         err = nilfs_transaction_begin(sb, &ti, 0);
 664         BUG_ON(err);
 665         if (inode->i_data.nrpages)
 666                 truncate_inode_pages(&inode->i_data, 0);
 667
 668         nilfs_truncate_bmap(ii, 0);
 669         nilfs_free_inode(inode);
 670         /* nilfs_free_inode() marks inode buffer dirty */
 671         if (IS_SYNC(inode))
 672                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 673         nilfs_transaction_commit(sb);
 674         /* May construct a logical segment and may fail in sync mode.
 675            But delete_inode has no return value. */
 676 }
 677
 678 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 679 {
 680         struct nilfs_transaction_info ti;
 681         struct inode *inode = dentry->d_inode;
 682         struct super_block *sb = inode->i_sb;
 683         int err;
 684
 685         err = inode_change_ok(inode, iattr);
 686         if (err)
 687                 return err;
 688
 689         err = nilfs_transaction_begin(sb, &ti, 0);
 690         if (unlikely(err))
 691                 return err;
 692         err = inode_setattr(inode, iattr);
 693         if (!err && (iattr->ia_valid & ATTR_MODE))
 694                 err = nilfs_acl_chmod(inode);
 695         if (likely(!err))
 696                 err = nilfs_transaction_commit(sb);
 697         else
 698                 nilfs_transaction_abort(sb);
 699
 700         return err;
 701 }
 702
 703 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 704                            struct buffer_head **pbh)
 705 {
 706         struct nilfs_inode_info *ii = NILFS_I(inode);
 707         int err;
 708
 709         spin_lock(&sbi->s_inode_lock);
 710         /* Caller of this function MUST lock s_inode_lock */
 711         if (ii->i_bh == NULL) {
 712                 spin_unlock(&sbi->s_inode_lock);
 713                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
 714                                                   pbh);
 715                 if (unlikely(err))
 716                         return err;
 717                 spin_lock(&sbi->s_inode_lock);
 718                 if (ii->i_bh == NULL)
 719                         ii->i_bh = *pbh;
 720                 else {
 721                         brelse(*pbh);
 722                         *pbh = ii->i_bh;
 723                 }
 724         } else
 725                 *pbh = ii->i_bh;
 726
 727         get_bh(*pbh);
 728         spin_unlock(&sbi->s_inode_lock);
 729         return 0;
 730 }
 731
 732 int nilfs_inode_dirty(struct inode *inode)
 733 {
 734         struct nilfs_inode_info *ii = NILFS_I(inode);
 735         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 736         int ret = 0;
 737
 738         if (!list_empty(&ii->i_dirty)) {
 739                 spin_lock(&sbi->s_inode_lock);
 740                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 741                         test_bit(NILFS_I_BUSY, &ii->i_state);
 742                 spin_unlock(&sbi->s_inode_lock);
 743         }
 744         return ret;
 745 }
 746
 747 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 748                          unsigned nr_dirty)
 749 {
 750         struct nilfs_inode_info *ii = NILFS_I(inode);
 751
 752         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 753
 754         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state) ||
 755             unlikely(inode->i_ino == NILFS_SKETCH_INO))
 756                 return 0;
 757
 758         spin_lock(&sbi->s_inode_lock);
 759         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 760             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 761                 /* Because this routine may race with nilfs_dispose_list(),
 762                    we have to check NILFS_I_QUEUED here, too. */
 763                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 764                         /* This will happen when somebody is freeing
 765                            this inode. */
 766                         nilfs_warning(sbi->s_super, __func__,
 767                                       "cannot get inode (ino=%lu)\n",
 768                                       inode->i_ino);
 769                         spin_unlock(&sbi->s_inode_lock);
 770                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 771                                            freeing inode */
 772                 }
 773                 list_del(&ii->i_dirty);
 774                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 775                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 776         }
 777         spin_unlock(&sbi->s_inode_lock);
 778         return 0;
 779 }
 780
 781 int nilfs_mark_inode_dirty(struct inode *inode)
 782 {
 783         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 784         struct buffer_head *ibh;
 785         int err;
 786
 787         err = nilfs_load_inode_block(sbi, inode, &ibh);
 788         if (unlikely(err)) {
 789                 nilfs_warning(inode->i_sb, __func__,
 790                               "failed to reget inode block.\n");
 791                 return err;
 792         }
 793         lock_buffer(ibh);
 794         nilfs_update_inode(inode, ibh);
 795         unlock_buffer(ibh);
 796         nilfs_mdt_mark_buffer_dirty(ibh);
 797         nilfs_mdt_mark_dirty(sbi->s_ifile);
 798         brelse(ibh);
 799         return 0;
 800 }
 801
 802 /**
 803  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 804  * @inode: inode of the file to be registered.
 805  *
 806  * nilfs_dirty_inode() loads a inode block containing the specified
 807  * @inode and copies data from a nilfs_inode to a corresponding inode
 808  * entry in the inode block. This operation is excluded from the segment
 809  * construction. This function can be called both as a single operation
 810  * and as a part of indivisible file operations.
 811  */
 812 void nilfs_dirty_inode(struct inode *inode)
 813 {
 814         struct nilfs_transaction_info ti;
 815
 816         if (is_bad_inode(inode)) {
 817                 nilfs_warning(inode->i_sb, __func__,
 818                               "tried to mark bad_inode dirty. ignored.\n");
 819                 dump_stack();
 820                 return;
 821         }
 822         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 823         if (likely(inode->i_ino != NILFS_SKETCH_INO))
 824                 nilfs_mark_inode_dirty(inode);
 825         nilfs_transaction_commit(inode->i_sb); /* never fails */
 826 }