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 "nilfs.h"
  28 #include "segment.h"
  29 #include "page.h"
  30 #include "mdt.h"
  31 #include "cpfile.h"
  32 #include "ifile.h"
  33
  34
  35 /**
  36  * nilfs_get_block() - get a file block on the filesystem (callback function)
  37  * @inode - inode struct of the target file
  38  * @blkoff - file block number
  39  * @bh_result - buffer head to be mapped on
  40  * @create - indicate whether allocating the block or not when it has not
  41  *      been allocated yet.
  42  *
  43  * This function does not issue actual read request of the specified data
  44  * block. It is done by VFS.
  45  * Bulk read for direct-io is not supported yet. (should be supported)
  46  */
  47 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  48                     struct buffer_head *bh_result, int create)
  49 {
  50         struct nilfs_inode_info *ii = NILFS_I(inode);
  51         unsigned long blknum = 0;
  52         int err = 0, ret;
  53         struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
  54
  55         /* This exclusion control is a workaround; should be revised */
  56         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
  57         ret = nilfs_bmap_lookup(ii->i_bmap, (unsigned long)blkoff, &blknum);
  58         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
  59         if (ret == 0) { /* found */
  60                 map_bh(bh_result, inode->i_sb, blknum);
  61                 goto out;
  62         }
  63         if (unlikely(ret == 1)) {
  64                 printk(KERN_ERR "nilfs_get_block: bmap_lookup returns "
  65                        "buffer_head pointer (blkoff=%llu, blknum=%lu)\n",
  66                        (unsigned long long)blkoff, blknum);
  67                 BUG();
  68         }
  69         /* data block was not found */
  70         if (ret == -ENOENT && create) {
  71                 struct nilfs_transaction_info ti;
  72
  73                 bh_result->b_blocknr = 0;
  74                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  75                 if (unlikely(err))
  76                         goto out;
  77                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
  78                                         (unsigned long)bh_result);
  79                 nilfs_transaction_end(inode->i_sb, !err);
  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                         goto out;
 103                 }
 104                 /* Error handling should be detailed */
 105                 set_buffer_new(bh_result);
 106                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 107                                                       to proper value */
 108         } else if (ret == -ENOENT) {
 109                 /* not found is not error (e.g. hole); must return without
 110                    the mapped state flag. */
 111                 ;
 112         } else {
 113                 err = ret;
 114         }
 115
 116  out:
 117         return err;
 118 }
 119
 120 /**
 121  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 122  * address_space_operations.
 123  * @file - file struct of the file to be read
 124  * @page - the page to be read
 125  */
 126 static int nilfs_readpage(struct file *file, struct page *page)
 127 {
 128         return mpage_readpage(page, nilfs_get_block);
 129 }
 130
 131 /**
 132  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 133  * address_space_operations.
 134  * @file - file struct of the file to be read
 135  * @mapping - address_space struct used for reading multiple pages
 136  * @pages - the pages to be read
 137  * @nr_pages - number of pages to be read
 138  */
 139 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 140                            struct list_head *pages, unsigned nr_pages)
 141 {
 142         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 143 }
 144
 145 static int nilfs_writepages(struct address_space *mapping,
 146                             struct writeback_control *wbc)
 147 {
 148         /* This empty method is required not to call generic_writepages() */
 149         return 0;
 150 }
 151
 152 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 153 {
 154         struct inode *inode = page->mapping->host;
 155         int err;
 156
 157         redirty_page_for_writepage(wbc, page);
 158         unlock_page(page);
 159
 160         if (wbc->sync_mode == WB_SYNC_ALL) {
 161                 err = nilfs_construct_segment(inode->i_sb);
 162                 if (unlikely(err))
 163                         return err;
 164         } else if (wbc->for_reclaim)
 165                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 166
 167         return 0;
 168 }
 169
 170 static int nilfs_set_page_dirty(struct page *page)
 171 {
 172         int ret = __set_page_dirty_buffers(page);
 173
 174         if (ret) {
 175                 struct inode *inode = page->mapping->host;
 176                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 177                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 178
 179                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 180         }
 181         return ret;
 182 }
 183
 184 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 185                              loff_t pos, unsigned len, unsigned flags,
 186                              struct page **pagep, void **fsdata)
 187
 188 {
 189         struct inode *inode = mapping->host;
 190         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 191
 192         if (unlikely(err))
 193                 return err;
 194
 195         *pagep = NULL;
 196         err = block_write_begin(file, mapping, pos, len, flags, pagep,
 197                                 fsdata, nilfs_get_block);
 198         if (unlikely(err))
 199                 nilfs_transaction_end(inode->i_sb, 0);
 200         return err;
 201 }
 202
 203 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 204                            loff_t pos, unsigned len, unsigned copied,
 205                            struct page *page, void *fsdata)
 206 {
 207         struct inode *inode = mapping->host;
 208         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 209         unsigned nr_dirty;
 210         int err;
 211
 212         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 213                                                   start + copied);
 214         copied = generic_write_end(file, mapping, pos, len, copied, page,
 215                                    fsdata);
 216         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 217         err = nilfs_transaction_end(inode->i_sb, 1);
 218         return err ? : copied;
 219 }
 220
 221 static ssize_t
 222 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 223                 loff_t offset, unsigned long nr_segs)
 224 {
 225         struct file *file = iocb->ki_filp;
 226         struct inode *inode = file->f_mapping->host;
 227         ssize_t size;
 228         int err;
 229
 230         err = nilfs_construct_dsync_segment(inode->i_sb, inode);
 231         if (unlikely(err))
 232                 return err;
 233
 234         if (rw == WRITE)
 235                 return 0;
 236
 237         /* Needs synchronization with the cleaner */
 238         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 239                                   offset, nr_segs, nilfs_get_block, NULL);
 240         return size;
 241 }
 242
 243 struct address_space_operations nilfs_aops = {
 244         .writepage              = nilfs_writepage,
 245         .readpage               = nilfs_readpage,
 246         /* .sync_page           = nilfs_sync_page, */
 247         .writepages             = nilfs_writepages,
 248         .set_page_dirty         = nilfs_set_page_dirty,
 249         .readpages              = nilfs_readpages,
 250         .write_begin            = nilfs_write_begin,
 251         .write_end              = nilfs_write_end,
 252         /* .releasepage         = nilfs_releasepage, */
 253         .invalidatepage         = block_invalidatepage,
 254         .direct_IO              = nilfs_direct_IO,
 255 };
 256
 257 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 258 {
 259         struct super_block *sb = dir->i_sb;
 260         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 261         struct inode *inode;
 262         struct nilfs_inode_info *ii;
 263         int err = -ENOMEM;
 264         ino_t ino;
 265
 266         inode = new_inode(sb);
 267         if (unlikely(!inode))
 268                 goto failed;
 269
 270         mapping_set_gfp_mask(inode->i_mapping,
 271                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 272
 273         ii = NILFS_I(inode);
 274         ii->i_state = 1 << NILFS_I_NEW;
 275
 276         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
 277         if (unlikely(err))
 278                 goto failed_ifile_create_inode;
 279         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 280
 281         atomic_inc(&sbi->s_inodes_count);
 282
 283         inode->i_uid = current_fsuid();
 284         if (dir->i_mode & S_ISGID) {
 285                 inode->i_gid = dir->i_gid;
 286                 if (S_ISDIR(mode))
 287                         mode |= S_ISGID;
 288         } else
 289                 inode->i_gid = current_fsgid();
 290
 291         inode->i_mode = mode;
 292         inode->i_ino = ino;
 293         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 294
 295         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 296                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 297                 if (err < 0)
 298                         goto failed_bmap;
 299
 300                 set_bit(NILFS_I_BMAP, &ii->i_state);
 301                 /* No lock is needed; iget() ensures it. */
 302         }
 303
 304         ii->i_flags = NILFS_I(dir)->i_flags;
 305         if (S_ISLNK(mode))
 306                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 307         if (!S_ISDIR(mode))
 308                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 309
 310         /* ii->i_file_acl = 0; */
 311         /* ii->i_dir_acl = 0; */
 312         ii->i_dtime = 0;
 313         ii->i_dir_start_lookup = 0;
 314 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 315         ii->i_acl = NULL;
 316         ii->i_default_acl = NULL;
 317 #endif
 318         ii->i_cno = 0;
 319         nilfs_set_inode_flags(inode);
 320         spin_lock(&sbi->s_next_gen_lock);
 321         inode->i_generation = sbi->s_next_generation++;
 322         spin_unlock(&sbi->s_next_gen_lock);
 323         insert_inode_hash(inode);
 324
 325         err = nilfs_init_acl(inode, dir);
 326         if (unlikely(err))
 327                 goto failed_acl; /* never occur. When supporting
 328                                     nilfs_init_acl(), proper cancellation of
 329                                     above jobs should be considered */
 330
 331         mark_inode_dirty(inode);
 332         return inode;
 333
 334  failed_acl:
 335  failed_bmap:
 336         inode->i_nlink = 0;
 337         iput(inode);  /* raw_inode will be deleted through
 338                          generic_delete_inode() */
 339         goto failed;
 340
 341  failed_ifile_create_inode:
 342         make_bad_inode(inode);
 343         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 344                          called */
 345  failed:
 346         return ERR_PTR(err);
 347 }
 348
 349 void nilfs_free_inode(struct inode *inode)
 350 {
 351         struct super_block *sb = inode->i_sb;
 352         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 353
 354         clear_inode(inode);
 355         /* XXX: check error code? Is there any thing I can do? */
 356         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
 357         atomic_dec(&sbi->s_inodes_count);
 358 }
 359
 360 void nilfs_set_inode_flags(struct inode *inode)
 361 {
 362         unsigned int flags = NILFS_I(inode)->i_flags;
 363
 364         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 365                             S_DIRSYNC);
 366         if (flags & NILFS_SYNC_FL)
 367                 inode->i_flags |= S_SYNC;
 368         if (flags & NILFS_APPEND_FL)
 369                 inode->i_flags |= S_APPEND;
 370         if (flags & NILFS_IMMUTABLE_FL)
 371                 inode->i_flags |= S_IMMUTABLE;
 372 #ifndef NILFS_ATIME_DISABLE
 373         if (flags & NILFS_NOATIME_FL)
 374 #endif
 375                 inode->i_flags |= S_NOATIME;
 376         if (flags & NILFS_DIRSYNC_FL)
 377                 inode->i_flags |= S_DIRSYNC;
 378         mapping_set_gfp_mask(inode->i_mapping,
 379                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 380 }
 381
 382 int nilfs_read_inode_common(struct inode *inode,
 383                             struct nilfs_inode *raw_inode)
 384 {
 385         struct nilfs_inode_info *ii = NILFS_I(inode);
 386         int err;
 387
 388         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 389         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 390         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 391         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 392         inode->i_size = le64_to_cpu(raw_inode->i_size);
 393         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 394         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 395         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 396         inode->i_atime.tv_nsec = 0;
 397         inode->i_ctime.tv_nsec = 0;
 398         inode->i_mtime.tv_nsec = 0;
 399         ii->i_dtime = le64_to_cpu(raw_inode->i_dtime);
 400         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ii->i_dtime))
 401                 return -EINVAL; /* this inode is deleted */
 402
 403         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 404         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 405 #if 0
 406         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 407         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 408                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 409 #endif
 410         ii->i_cno = 0;
 411         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 412
 413         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 414             S_ISLNK(inode->i_mode)) {
 415                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 416                 if (err < 0)
 417                         return err;
 418                 set_bit(NILFS_I_BMAP, &ii->i_state);
 419                 /* No lock is needed; iget() ensures it. */
 420         }
 421         return 0;
 422 }
 423
 424 static int nilfs_read_sketch_inode(struct inode *inode)
 425 {
 426         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 427         int err = 0;
 428
 429         if (sbi->s_snapshot_cno) {
 430                 struct the_nilfs *nilfs = sbi->s_nilfs;
 431                 struct buffer_head *bh_cp;
 432                 struct nilfs_checkpoint *raw_cp;
 433
 434                 err = nilfs_cpfile_get_checkpoint(
 435                         nilfs->ns_cpfile, sbi->s_snapshot_cno, 0, &raw_cp,
 436                         &bh_cp);
 437                 if (likely(!err)) {
 438                         if (!nilfs_checkpoint_sketch(raw_cp))
 439                                 inode->i_size = 0;
 440                         nilfs_cpfile_put_checkpoint(
 441                                 nilfs->ns_cpfile, sbi->s_snapshot_cno, bh_cp);
 442                 }
 443                 inode->i_flags |= S_NOCMTIME;
 444         }
 445         return err;
 446 }
 447
 448 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 449                               struct inode *inode)
 450 {
 451         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 452         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 453         struct buffer_head *bh;
 454         struct nilfs_inode *raw_inode;
 455         int err;
 456
 457         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 458         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 459         if (unlikely(err))
 460                 goto bad_inode;
 461
 462         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 463
 464 #ifdef CONFIG_NILFS_FS_POSIX_ACL
 465         ii->i_acl = NILFS_ACL_NOT_CACHED;
 466         ii->i_default_acl = NILFS_ACL_NOT_CACHED;
 467 #endif
 468         if (nilfs_read_inode_common(inode, raw_inode))
 469                 goto failed_unmap;
 470
 471         if (S_ISREG(inode->i_mode)) {
 472                 inode->i_op = &nilfs_file_inode_operations;
 473                 inode->i_fop = &nilfs_file_operations;
 474                 inode->i_mapping->a_ops = &nilfs_aops;
 475                 if (unlikely(inode->i_ino == NILFS_SKETCH_INO)) {
 476                         err = nilfs_read_sketch_inode(inode);
 477                         if (unlikely(err))
 478                                 goto failed_unmap;
 479                 }
 480         } else if (S_ISDIR(inode->i_mode)) {
 481                 inode->i_op = &nilfs_dir_inode_operations;
 482                 inode->i_fop = &nilfs_dir_operations;
 483                 inode->i_mapping->a_ops = &nilfs_aops;
 484         } else if (S_ISLNK(inode->i_mode)) {
 485                 inode->i_op = &nilfs_symlink_inode_operations;
 486                 inode->i_mapping->a_ops = &nilfs_aops;
 487         } else {
 488                 inode->i_op = &nilfs_special_inode_operations;
 489                 init_special_inode(
 490                         inode, inode->i_mode,
 491                         new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 492         }
 493         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 494         brelse(bh);
 495         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 496         nilfs_set_inode_flags(inode);
 497         return 0;
 498
 499  failed_unmap:
 500         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 501         brelse(bh);
 502
 503  bad_inode:
 504         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 505         return err;
 506 }
 507
 508 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 509 {
 510         struct inode *inode;
 511         int err;
 512
 513         inode = iget_locked(sb, ino);
 514         if (unlikely(!inode))
 515                 return ERR_PTR(-ENOMEM);
 516         if (!(inode->i_state & I_NEW))
 517                 return inode;
 518
 519         err = __nilfs_read_inode(sb, ino, inode);
 520         if (unlikely(err)) {
 521                 iget_failed(inode);
 522                 return ERR_PTR(err);
 523         }
 524         unlock_new_inode(inode);
 525         return inode;
 526 }
 527
 528 void nilfs_write_inode_common(struct inode *inode,
 529                               struct nilfs_inode *raw_inode, int has_bmap)
 530 {
 531         struct nilfs_inode_info *ii = NILFS_I(inode);
 532
 533         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 534         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 535         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 536         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 537         raw_inode->i_size = cpu_to_le64(inode->i_size);
 538         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 539         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 540         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 541
 542         raw_inode->i_dtime = cpu_to_le64(ii->i_dtime);
 543         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 544         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 545
 546         if (has_bmap)
 547                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 548         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 549                 raw_inode->i_device_code =
 550                         cpu_to_le64(new_encode_dev(inode->i_rdev));
 551         /* When extending inode, nilfs->ns_inode_size should be checked
 552            for substitutions of appended fields */
 553 }
 554
 555 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 556 {
 557         ino_t ino = inode->i_ino;
 558         struct nilfs_inode_info *ii = NILFS_I(inode);
 559         struct super_block *sb = inode->i_sb;
 560         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 561         struct nilfs_inode *raw_inode;
 562
 563         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
 564
 565         /* The buffer is guarded with lock_buffer() by the caller */
 566         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 567                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 568         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 569
 570         nilfs_write_inode_common(inode, raw_inode, 0);
 571                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 572                    deadlock of bmap. This delays update of i_bmap to just
 573                    before writing */
 574         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 575 }
 576
 577 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 578
 579 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 580                                 unsigned long from)
 581 {
 582         unsigned long b;
 583         int ret;
 584
 585         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 586                 return;
 587  repeat:
 588         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 589         if (ret == -ENOENT)
 590                 return;
 591         else if (ret < 0)
 592                 goto failed;
 593
 594         if (b < from)
 595                 return;
 596
 597         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 598         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 599         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 600         if (!ret || (ret == -ENOMEM &&
 601                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 602                 goto repeat;
 603
 604  failed:
 605         if (ret == -EINVAL)
 606                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 607                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 608         else
 609                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 610                               "failed to truncate bmap (ino=%lu, err=%d)",
 611                               ii->vfs_inode.i_ino, ret);
 612 }
 613
 614 void nilfs_truncate(struct inode *inode)
 615 {
 616         unsigned long blkoff;
 617         unsigned int blocksize;
 618         struct nilfs_transaction_info ti;
 619         struct super_block *sb = inode->i_sb;
 620         struct nilfs_inode_info *ii = NILFS_I(inode);
 621         int ret;
 622
 623         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 624                 return;
 625         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 626                 return;
 627
 628         blocksize = sb->s_blocksize;
 629         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 630         ret = nilfs_transaction_begin(sb, &ti, 0);
 631         BUG_ON(ret);
 632
 633         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 634
 635         nilfs_truncate_bmap(ii, blkoff);
 636
 637         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 638         if (IS_SYNC(inode))
 639                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 640
 641         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 642         nilfs_transaction_end(sb, 1);
 643         /* May construct a logical segment and may fail in sync mode.
 644            But truncate has no return value. */
 645 }
 646
 647 void nilfs_delete_inode(struct inode *inode)
 648 {
 649         struct nilfs_transaction_info ti;
 650         struct super_block *sb = inode->i_sb;
 651         struct nilfs_inode_info *ii = NILFS_I(inode);
 652         int err;
 653
 654         if (unlikely(is_bad_inode(inode))) {
 655                 if (inode->i_data.nrpages)
 656                         truncate_inode_pages(&inode->i_data, 0);
 657                 clear_inode(inode);
 658                 return;
 659         }
 660         err = nilfs_transaction_begin(sb, &ti, 0);
 661         BUG_ON(err);
 662         if (inode->i_data.nrpages)
 663                 truncate_inode_pages(&inode->i_data, 0);
 664
 665         nilfs_truncate_bmap(ii, 0);
 666         nilfs_free_inode(inode);
 667         /* nilfs_free_inode() marks inode buffer dirty */
 668         if (IS_SYNC(inode))
 669                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 670         nilfs_transaction_end(sb, 1);
 671         /* May construct a logical segment and may fail in sync mode.
 672            But delete_inode has no return value. */
 673 }
 674
 675 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 676 {
 677         struct nilfs_transaction_info ti;
 678         struct inode *inode = dentry->d_inode;
 679         struct super_block *sb = inode->i_sb;
 680         int err, err2;
 681
 682         err = inode_change_ok(inode, iattr);
 683         if (err)
 684                 return err;
 685
 686         err = nilfs_transaction_begin(sb, &ti, 0);
 687         if (unlikely(err))
 688                 return err;
 689         err = inode_setattr(inode, iattr);
 690         if (!err && (iattr->ia_valid & ATTR_MODE))
 691                 err = nilfs_acl_chmod(inode);
 692         err2 = nilfs_transaction_end(sb, 1);
 693         return err ? : err2;
 694 }
 695
 696 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 697                            struct buffer_head **pbh)
 698 {
 699         struct nilfs_inode_info *ii = NILFS_I(inode);
 700         int err;
 701
 702         spin_lock(&sbi->s_inode_lock);
 703         /* Caller of this function MUST lock s_inode_lock */
 704         if (ii->i_bh == NULL) {
 705                 spin_unlock(&sbi->s_inode_lock);
 706                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
 707                                                   pbh);
 708                 if (unlikely(err))
 709                         return err;
 710                 spin_lock(&sbi->s_inode_lock);
 711                 if (ii->i_bh == NULL)
 712                         ii->i_bh = *pbh;
 713                 else {
 714                         brelse(*pbh);
 715                         *pbh = ii->i_bh;
 716                 }
 717         } else
 718                 *pbh = ii->i_bh;
 719
 720         get_bh(*pbh);
 721         spin_unlock(&sbi->s_inode_lock);
 722         return 0;
 723 }
 724
 725 int nilfs_inode_dirty(struct inode *inode)
 726 {
 727         struct nilfs_inode_info *ii = NILFS_I(inode);
 728         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 729         int ret = 0;
 730
 731         if (!list_empty(&ii->i_dirty)) {
 732                 spin_lock(&sbi->s_inode_lock);
 733                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 734                         test_bit(NILFS_I_BUSY, &ii->i_state);
 735                 spin_unlock(&sbi->s_inode_lock);
 736         }
 737         return ret;
 738 }
 739
 740 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 741                          unsigned nr_dirty)
 742 {
 743         struct nilfs_inode_info *ii = NILFS_I(inode);
 744
 745         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 746
 747         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state) ||
 748             unlikely(inode->i_ino == NILFS_SKETCH_INO))
 749                 return 0;
 750
 751         spin_lock(&sbi->s_inode_lock);
 752         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 753             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 754                 /* Because this routine may race with nilfs_dispose_list(),
 755                    we have to check NILFS_I_QUEUED here, too. */
 756                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 757                         /* This will happen when somebody is freeing
 758                            this inode. */
 759                         nilfs_warning(sbi->s_super, __func__,
 760                                       "cannot get inode (ino=%lu)\n",
 761                                       inode->i_ino);
 762                         spin_unlock(&sbi->s_inode_lock);
 763                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 764                                            freeing inode */
 765                 }
 766                 list_del(&ii->i_dirty);
 767                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 768                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 769         }
 770         spin_unlock(&sbi->s_inode_lock);
 771         return 0;
 772 }
 773
 774 int nilfs_mark_inode_dirty(struct inode *inode)
 775 {
 776         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 777         struct buffer_head *ibh;
 778         int err;
 779
 780         err = nilfs_load_inode_block(sbi, inode, &ibh);
 781         if (unlikely(err)) {
 782                 nilfs_warning(inode->i_sb, __func__,
 783                               "failed to reget inode block.\n");
 784                 return err;
 785         }
 786         lock_buffer(ibh);
 787         nilfs_update_inode(inode, ibh);
 788         unlock_buffer(ibh);
 789         nilfs_mdt_mark_buffer_dirty(ibh);
 790         nilfs_mdt_mark_dirty(sbi->s_ifile);
 791         brelse(ibh);
 792         return 0;
 793 }
 794
 795 /**
 796  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 797  * @inode: inode of the file to be registered.
 798  *
 799  * nilfs_dirty_inode() loads a inode block containing the specified
 800  * @inode and copies data from a nilfs_inode to a corresponding inode
 801  * entry in the inode block. This operation is excluded from the segment
 802  * construction. This function can be called both as a single operation
 803  * and as a part of indivisible file operations.
 804  */
 805 void nilfs_dirty_inode(struct inode *inode)
 806 {
 807         struct nilfs_transaction_info ti;
 808
 809         if (is_bad_inode(inode)) {
 810                 nilfs_warning(inode->i_sb, __func__,
 811                               "tried to mark bad_inode dirty. ignored.\n");
 812                 dump_stack();
 813                 return;
 814         }
 815         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 816         if (likely(inode->i_ino != NILFS_SKETCH_INO))
 817                 nilfs_mark_inode_dirty(inode);
 818         nilfs_transaction_end(inode->i_sb, 1); /* never fails */
 819 }