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/gfp.h>
  26 #include <linux/mpage.h>
  27 #include <linux/writeback.h>
  28 #include <linux/uio.h>
  29 #include "nilfs.h"
  30 #include "btnode.h"
  31 #include "segment.h"
  32 #include "page.h"
  33 #include "mdt.h"
  34 #include "cpfile.h"
  35 #include "ifile.h"
  36
  37 struct nilfs_iget_args {
  38         u64 ino;
  39         __u64 cno;
  40         struct nilfs_root *root;
  41         int for_gc;
  42 };
  43
  44 /**
  45  * nilfs_get_block() - get a file block on the filesystem (callback function)
  46  * @inode - inode struct of the target file
  47  * @blkoff - file block number
  48  * @bh_result - buffer head to be mapped on
  49  * @create - indicate whether allocating the block or not when it has not
  50  *      been allocated yet.
  51  *
  52  * This function does not issue actual read request of the specified data
  53  * block. It is done by VFS.
  54  */
  55 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  56                     struct buffer_head *bh_result, int create)
  57 {
  58         struct nilfs_inode_info *ii = NILFS_I(inode);
  59         __u64 blknum = 0;
  60         int err = 0, ret;
  61         struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
  62         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  63
  64         down_read(&NILFS_MDT(dat)->mi_sem);
  65         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  66         up_read(&NILFS_MDT(dat)->mi_sem);
  67         if (ret >= 0) { /* found */
  68                 map_bh(bh_result, inode->i_sb, blknum);
  69                 if (ret > 0)
  70                         bh_result->b_size = (ret << inode->i_blkbits);
  71                 goto out;
  72         }
  73         /* data block was not found */
  74         if (ret == -ENOENT && create) {
  75                 struct nilfs_transaction_info ti;
  76
  77                 bh_result->b_blocknr = 0;
  78                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  79                 if (unlikely(err))
  80                         goto out;
  81                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
  82                                         (unsigned long)bh_result);
  83                 if (unlikely(err != 0)) {
  84                         if (err == -EEXIST) {
  85                                 /*
  86                                  * The get_block() function could be called
  87                                  * from multiple callers for an inode.
  88                                  * However, the page having this block must
  89                                  * be locked in this case.
  90                                  */
  91                                 printk(KERN_WARNING
  92                                        "nilfs_get_block: a race condition "
  93                                        "while inserting a data block. "
  94                                        "(inode number=%lu, file block "
  95                                        "offset=%llu)\n",
  96                                        inode->i_ino,
  97                                        (unsigned long long)blkoff);
  98                                 err = 0;
  99                         } else if (err == -EINVAL) {
 100                                 nilfs_error(inode->i_sb, __func__,
 101                                             "broken bmap (inode=%lu)\n",
 102                                             inode->i_ino);
 103                                 err = -EIO;
 104                         }
 105                         nilfs_transaction_abort(inode->i_sb);
 106                         goto out;
 107                 }
 108                 nilfs_mark_inode_dirty(inode);
 109                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 110                 /* Error handling should be detailed */
 111                 set_buffer_new(bh_result);
 112                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 113                                                       to proper value */
 114         } else if (ret == -ENOENT) {
 115                 /* not found is not error (e.g. hole); must return without
 116                    the mapped state flag. */
 117                 ;
 118         } else {
 119                 err = ret;
 120         }
 121
 122  out:
 123         return err;
 124 }
 125
 126 /**
 127  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 128  * address_space_operations.
 129  * @file - file struct of the file to be read
 130  * @page - the page to be read
 131  */
 132 static int nilfs_readpage(struct file *file, struct page *page)
 133 {
 134         return mpage_readpage(page, nilfs_get_block);
 135 }
 136
 137 /**
 138  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 139  * address_space_operations.
 140  * @file - file struct of the file to be read
 141  * @mapping - address_space struct used for reading multiple pages
 142  * @pages - the pages to be read
 143  * @nr_pages - number of pages to be read
 144  */
 145 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 146                            struct list_head *pages, unsigned nr_pages)
 147 {
 148         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 149 }
 150
 151 static int nilfs_writepages(struct address_space *mapping,
 152                             struct writeback_control *wbc)
 153 {
 154         struct inode *inode = mapping->host;
 155         int err = 0;
 156
 157         if (wbc->sync_mode == WB_SYNC_ALL)
 158                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 159                                                     wbc->range_start,
 160                                                     wbc->range_end);
 161         return err;
 162 }
 163
 164 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 165 {
 166         struct inode *inode = page->mapping->host;
 167         int err;
 168
 169         redirty_page_for_writepage(wbc, page);
 170         unlock_page(page);
 171
 172         if (wbc->sync_mode == WB_SYNC_ALL) {
 173                 err = nilfs_construct_segment(inode->i_sb);
 174                 if (unlikely(err))
 175                         return err;
 176         } else if (wbc->for_reclaim)
 177                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 178
 179         return 0;
 180 }
 181
 182 static int nilfs_set_page_dirty(struct page *page)
 183 {
 184         int ret = __set_page_dirty_buffers(page);
 185
 186         if (ret) {
 187                 struct inode *inode = page->mapping->host;
 188                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 189                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 190
 191                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 192         }
 193         return ret;
 194 }
 195
 196 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 197                              loff_t pos, unsigned len, unsigned flags,
 198                              struct page **pagep, void **fsdata)
 199
 200 {
 201         struct inode *inode = mapping->host;
 202         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 203
 204         if (unlikely(err))
 205                 return err;
 206
 207         err = block_write_begin(mapping, pos, len, flags, pagep,
 208                                 nilfs_get_block);
 209         if (unlikely(err)) {
 210                 loff_t isize = mapping->host->i_size;
 211                 if (pos + len > isize)
 212                         vmtruncate(mapping->host, isize);
 213
 214                 nilfs_transaction_abort(inode->i_sb);
 215         }
 216         return err;
 217 }
 218
 219 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 220                            loff_t pos, unsigned len, unsigned copied,
 221                            struct page *page, void *fsdata)
 222 {
 223         struct inode *inode = mapping->host;
 224         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 225         unsigned nr_dirty;
 226         int err;
 227
 228         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 229                                                   start + copied);
 230         copied = generic_write_end(file, mapping, pos, len, copied, page,
 231                                    fsdata);
 232         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 233         err = nilfs_transaction_commit(inode->i_sb);
 234         return err ? : copied;
 235 }
 236
 237 static ssize_t
 238 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 239                 loff_t offset, unsigned long nr_segs)
 240 {
 241         struct file *file = iocb->ki_filp;
 242         struct inode *inode = file->f_mapping->host;
 243         ssize_t size;
 244
 245         if (rw == WRITE)
 246                 return 0;
 247
 248         /* Needs synchronization with the cleaner */
 249         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 250                                   offset, nr_segs, nilfs_get_block, NULL);
 251
 252         /*
 253          * In case of error extending write may have instantiated a few
 254          * blocks outside i_size. Trim these off again.
 255          */
 256         if (unlikely((rw & WRITE) && size < 0)) {
 257                 loff_t isize = i_size_read(inode);
 258                 loff_t end = offset + iov_length(iov, nr_segs);
 259
 260                 if (end > isize)
 261                         vmtruncate(inode, isize);
 262         }
 263
 264         return size;
 265 }
 266
 267 const struct address_space_operations nilfs_aops = {
 268         .writepage              = nilfs_writepage,
 269         .readpage               = nilfs_readpage,
 270         .sync_page              = block_sync_page,
 271         .writepages             = nilfs_writepages,
 272         .set_page_dirty         = nilfs_set_page_dirty,
 273         .readpages              = nilfs_readpages,
 274         .write_begin            = nilfs_write_begin,
 275         .write_end              = nilfs_write_end,
 276         /* .releasepage         = nilfs_releasepage, */
 277         .invalidatepage         = block_invalidatepage,
 278         .direct_IO              = nilfs_direct_IO,
 279         .is_partially_uptodate  = block_is_partially_uptodate,
 280 };
 281
 282 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 283 {
 284         struct super_block *sb = dir->i_sb;
 285         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 286         struct inode *inode;
 287         struct nilfs_inode_info *ii;
 288         struct nilfs_root *root;
 289         int err = -ENOMEM;
 290         ino_t ino;
 291
 292         inode = new_inode(sb);
 293         if (unlikely(!inode))
 294                 goto failed;
 295
 296         mapping_set_gfp_mask(inode->i_mapping,
 297                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 298
 299         root = NILFS_I(dir)->i_root;
 300         ii = NILFS_I(inode);
 301         ii->i_state = 1 << NILFS_I_NEW;
 302         ii->i_root = root;
 303
 304         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 305         if (unlikely(err))
 306                 goto failed_ifile_create_inode;
 307         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 308
 309         atomic_inc(&root->inodes_count);
 310         inode_init_owner(inode, dir, mode);
 311         inode->i_ino = ino;
 312         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 313
 314         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 315                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 316                 if (err < 0)
 317                         goto failed_bmap;
 318
 319                 set_bit(NILFS_I_BMAP, &ii->i_state);
 320                 /* No lock is needed; iget() ensures it. */
 321         }
 322
 323         ii->i_flags = NILFS_I(dir)->i_flags;
 324         if (S_ISLNK(mode))
 325                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 326         if (!S_ISDIR(mode))
 327                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 328
 329         /* ii->i_file_acl = 0; */
 330         /* ii->i_dir_acl = 0; */
 331         ii->i_dir_start_lookup = 0;
 332         nilfs_set_inode_flags(inode);
 333         spin_lock(&sbi->s_next_gen_lock);
 334         inode->i_generation = sbi->s_next_generation++;
 335         spin_unlock(&sbi->s_next_gen_lock);
 336         insert_inode_hash(inode);
 337
 338         err = nilfs_init_acl(inode, dir);
 339         if (unlikely(err))
 340                 goto failed_acl; /* never occur. When supporting
 341                                     nilfs_init_acl(), proper cancellation of
 342                                     above jobs should be considered */
 343
 344         return inode;
 345
 346  failed_acl:
 347  failed_bmap:
 348         inode->i_nlink = 0;
 349         iput(inode);  /* raw_inode will be deleted through
 350                          generic_delete_inode() */
 351         goto failed;
 352
 353  failed_ifile_create_inode:
 354         make_bad_inode(inode);
 355         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 356                          called */
 357  failed:
 358         return ERR_PTR(err);
 359 }
 360
 361 void nilfs_set_inode_flags(struct inode *inode)
 362 {
 363         unsigned int flags = NILFS_I(inode)->i_flags;
 364
 365         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 366                             S_DIRSYNC);
 367         if (flags & NILFS_SYNC_FL)
 368                 inode->i_flags |= S_SYNC;
 369         if (flags & NILFS_APPEND_FL)
 370                 inode->i_flags |= S_APPEND;
 371         if (flags & NILFS_IMMUTABLE_FL)
 372                 inode->i_flags |= S_IMMUTABLE;
 373 #ifndef NILFS_ATIME_DISABLE
 374         if (flags & NILFS_NOATIME_FL)
 375 #endif
 376                 inode->i_flags |= S_NOATIME;
 377         if (flags & NILFS_DIRSYNC_FL)
 378                 inode->i_flags |= S_DIRSYNC;
 379         mapping_set_gfp_mask(inode->i_mapping,
 380                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 381 }
 382
 383 int nilfs_read_inode_common(struct inode *inode,
 384                             struct nilfs_inode *raw_inode)
 385 {
 386         struct nilfs_inode_info *ii = NILFS_I(inode);
 387         int err;
 388
 389         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 390         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 391         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 392         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 393         inode->i_size = le64_to_cpu(raw_inode->i_size);
 394         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 395         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 396         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 397         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 398         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 399         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 400         if (inode->i_nlink == 0 && inode->i_mode == 0)
 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_dir_start_lookup = 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_inode(struct super_block *sb,
 425                               struct nilfs_root *root, unsigned long ino,
 426                               struct inode *inode)
 427 {
 428         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 429         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 430         struct buffer_head *bh;
 431         struct nilfs_inode *raw_inode;
 432         int err;
 433
 434         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 435         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 436         if (unlikely(err))
 437                 goto bad_inode;
 438
 439         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 440
 441         err = nilfs_read_inode_common(inode, raw_inode);
 442         if (err)
 443                 goto failed_unmap;
 444
 445         if (S_ISREG(inode->i_mode)) {
 446                 inode->i_op = &nilfs_file_inode_operations;
 447                 inode->i_fop = &nilfs_file_operations;
 448                 inode->i_mapping->a_ops = &nilfs_aops;
 449         } else if (S_ISDIR(inode->i_mode)) {
 450                 inode->i_op = &nilfs_dir_inode_operations;
 451                 inode->i_fop = &nilfs_dir_operations;
 452                 inode->i_mapping->a_ops = &nilfs_aops;
 453         } else if (S_ISLNK(inode->i_mode)) {
 454                 inode->i_op = &nilfs_symlink_inode_operations;
 455                 inode->i_mapping->a_ops = &nilfs_aops;
 456         } else {
 457                 inode->i_op = &nilfs_special_inode_operations;
 458                 init_special_inode(
 459                         inode, inode->i_mode,
 460                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 461         }
 462         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 463         brelse(bh);
 464         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 465         nilfs_set_inode_flags(inode);
 466         return 0;
 467
 468  failed_unmap:
 469         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 470         brelse(bh);
 471
 472  bad_inode:
 473         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 474         return err;
 475 }
 476
 477 static int nilfs_iget_test(struct inode *inode, void *opaque)
 478 {
 479         struct nilfs_iget_args *args = opaque;
 480         struct nilfs_inode_info *ii;
 481
 482         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 483                 return 0;
 484
 485         ii = NILFS_I(inode);
 486         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 487                 return !args->for_gc;
 488
 489         return args->for_gc && args->cno == ii->i_cno;
 490 }
 491
 492 static int nilfs_iget_set(struct inode *inode, void *opaque)
 493 {
 494         struct nilfs_iget_args *args = opaque;
 495
 496         inode->i_ino = args->ino;
 497         if (args->for_gc) {
 498                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 499                 NILFS_I(inode)->i_cno = args->cno;
 500                 NILFS_I(inode)->i_root = NULL;
 501         } else {
 502                 if (args->root && args->ino == NILFS_ROOT_INO)
 503                         nilfs_get_root(args->root);
 504                 NILFS_I(inode)->i_root = args->root;
 505         }
 506         return 0;
 507 }
 508
 509 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 510                                 unsigned long ino)
 511 {
 512         struct nilfs_iget_args args = {
 513                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 514         };
 515
 516         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 517 }
 518
 519 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 520                          unsigned long ino)
 521 {
 522         struct inode *inode;
 523         int err;
 524
 525         inode = nilfs_iget_locked(sb, root, ino);
 526         if (unlikely(!inode))
 527                 return ERR_PTR(-ENOMEM);
 528         if (!(inode->i_state & I_NEW))
 529                 return inode;
 530
 531         err = __nilfs_read_inode(sb, root, ino, inode);
 532         if (unlikely(err)) {
 533                 iget_failed(inode);
 534                 return ERR_PTR(err);
 535         }
 536         unlock_new_inode(inode);
 537         return inode;
 538 }
 539
 540 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 541                                 __u64 cno)
 542 {
 543         struct nilfs_iget_args args = {
 544                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 545         };
 546         struct inode *inode;
 547         int err;
 548
 549         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 550         if (unlikely(!inode))
 551                 return ERR_PTR(-ENOMEM);
 552         if (!(inode->i_state & I_NEW))
 553                 return inode;
 554
 555         err = nilfs_init_gcinode(inode);
 556         if (unlikely(err)) {
 557                 iget_failed(inode);
 558                 return ERR_PTR(err);
 559         }
 560         unlock_new_inode(inode);
 561         return inode;
 562 }
 563
 564 void nilfs_write_inode_common(struct inode *inode,
 565                               struct nilfs_inode *raw_inode, int has_bmap)
 566 {
 567         struct nilfs_inode_info *ii = NILFS_I(inode);
 568
 569         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 570         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 571         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 572         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 573         raw_inode->i_size = cpu_to_le64(inode->i_size);
 574         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 575         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 576         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 577         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 578         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 579
 580         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 581         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 582
 583         if (has_bmap)
 584                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 585         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 586                 raw_inode->i_device_code =
 587                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 588         /* When extending inode, nilfs->ns_inode_size should be checked
 589            for substitutions of appended fields */
 590 }
 591
 592 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 593 {
 594         ino_t ino = inode->i_ino;
 595         struct nilfs_inode_info *ii = NILFS_I(inode);
 596         struct inode *ifile = ii->i_root->ifile;
 597         struct nilfs_inode *raw_inode;
 598
 599         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 600
 601         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 602                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 603         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 604
 605         nilfs_write_inode_common(inode, raw_inode, 0);
 606                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 607                    deadlock of bmap. This delays update of i_bmap to just
 608                    before writing */
 609         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 610 }
 611
 612 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 613
 614 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 615                                 unsigned long from)
 616 {
 617         unsigned long b;
 618         int ret;
 619
 620         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 621                 return;
 622  repeat:
 623         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 624         if (ret == -ENOENT)
 625                 return;
 626         else if (ret < 0)
 627                 goto failed;
 628
 629         if (b < from)
 630                 return;
 631
 632         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 633         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 634         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 635         if (!ret || (ret == -ENOMEM &&
 636                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 637                 goto repeat;
 638
 639  failed:
 640         if (ret == -EINVAL)
 641                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 642                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 643         else
 644                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 645                               "failed to truncate bmap (ino=%lu, err=%d)",
 646                               ii->vfs_inode.i_ino, ret);
 647 }
 648
 649 void nilfs_truncate(struct inode *inode)
 650 {
 651         unsigned long blkoff;
 652         unsigned int blocksize;
 653         struct nilfs_transaction_info ti;
 654         struct super_block *sb = inode->i_sb;
 655         struct nilfs_inode_info *ii = NILFS_I(inode);
 656
 657         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 658                 return;
 659         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 660                 return;
 661
 662         blocksize = sb->s_blocksize;
 663         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 664         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 665
 666         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 667
 668         nilfs_truncate_bmap(ii, blkoff);
 669
 670         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 671         if (IS_SYNC(inode))
 672                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 673
 674         nilfs_mark_inode_dirty(inode);
 675         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 676         nilfs_transaction_commit(sb);
 677         /* May construct a logical segment and may fail in sync mode.
 678            But truncate has no return value. */
 679 }
 680
 681 static void nilfs_clear_inode(struct inode *inode)
 682 {
 683         struct nilfs_inode_info *ii = NILFS_I(inode);
 684         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 685
 686         /*
 687          * Free resources allocated in nilfs_read_inode(), here.
 688          */
 689         BUG_ON(!list_empty(&ii->i_dirty));
 690         brelse(ii->i_bh);
 691         ii->i_bh = NULL;
 692
 693         if (mdi && mdi->mi_palloc_cache)
 694                 nilfs_palloc_destroy_cache(inode);
 695
 696         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 697                 nilfs_bmap_clear(ii->i_bmap);
 698
 699         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 700
 701         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 702                 nilfs_put_root(ii->i_root);
 703 }
 704
 705 void nilfs_evict_inode(struct inode *inode)
 706 {
 707         struct nilfs_transaction_info ti;
 708         struct super_block *sb = inode->i_sb;
 709         struct nilfs_inode_info *ii = NILFS_I(inode);
 710
 711         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 712                 if (inode->i_data.nrpages)
 713                         truncate_inode_pages(&inode->i_data, 0);
 714                 end_writeback(inode);
 715                 nilfs_clear_inode(inode);
 716                 return;
 717         }
 718         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 719
 720         if (inode->i_data.nrpages)
 721                 truncate_inode_pages(&inode->i_data, 0);
 722
 723         /* TODO: some of the following operations may fail.  */
 724         nilfs_truncate_bmap(ii, 0);
 725         nilfs_mark_inode_dirty(inode);
 726         end_writeback(inode);
 727
 728         nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 729         atomic_dec(&ii->i_root->inodes_count);
 730
 731         nilfs_clear_inode(inode);
 732
 733         if (IS_SYNC(inode))
 734                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 735         nilfs_transaction_commit(sb);
 736         /* May construct a logical segment and may fail in sync mode.
 737            But delete_inode has no return value. */
 738 }
 739
 740 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 741 {
 742         struct nilfs_transaction_info ti;
 743         struct inode *inode = dentry->d_inode;
 744         struct super_block *sb = inode->i_sb;
 745         int err;
 746
 747         err = inode_change_ok(inode, iattr);
 748         if (err)
 749                 return err;
 750
 751         err = nilfs_transaction_begin(sb, &ti, 0);
 752         if (unlikely(err))
 753                 return err;
 754
 755         if ((iattr->ia_valid & ATTR_SIZE) &&
 756             iattr->ia_size != i_size_read(inode)) {
 757                 err = vmtruncate(inode, iattr->ia_size);
 758                 if (unlikely(err))
 759                         goto out_err;
 760         }
 761
 762         setattr_copy(inode, iattr);
 763         mark_inode_dirty(inode);
 764
 765         if (iattr->ia_valid & ATTR_MODE) {
 766                 err = nilfs_acl_chmod(inode);
 767                 if (unlikely(err))
 768                         goto out_err;
 769         }
 770
 771         return nilfs_transaction_commit(sb);
 772
 773 out_err:
 774         nilfs_transaction_abort(sb);
 775         return err;
 776 }
 777
 778 int nilfs_permission(struct inode *inode, int mask)
 779 {
 780         struct nilfs_root *root = NILFS_I(inode)->i_root;
 781
 782         if ((mask & MAY_WRITE) && root &&
 783             root->cno != NILFS_CPTREE_CURRENT_CNO)
 784                 return -EROFS; /* snapshot is not writable */
 785
 786         return generic_permission(inode, mask, NULL);
 787 }
 788
 789 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 790                            struct buffer_head **pbh)
 791 {
 792         struct nilfs_inode_info *ii = NILFS_I(inode);
 793         int err;
 794
 795         spin_lock(&sbi->s_inode_lock);
 796         if (ii->i_bh == NULL) {
 797                 spin_unlock(&sbi->s_inode_lock);
 798                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 799                                                   inode->i_ino, pbh);
 800                 if (unlikely(err))
 801                         return err;
 802                 spin_lock(&sbi->s_inode_lock);
 803                 if (ii->i_bh == NULL)
 804                         ii->i_bh = *pbh;
 805                 else {
 806                         brelse(*pbh);
 807                         *pbh = ii->i_bh;
 808                 }
 809         } else
 810                 *pbh = ii->i_bh;
 811
 812         get_bh(*pbh);
 813         spin_unlock(&sbi->s_inode_lock);
 814         return 0;
 815 }
 816
 817 int nilfs_inode_dirty(struct inode *inode)
 818 {
 819         struct nilfs_inode_info *ii = NILFS_I(inode);
 820         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 821         int ret = 0;
 822
 823         if (!list_empty(&ii->i_dirty)) {
 824                 spin_lock(&sbi->s_inode_lock);
 825                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 826                         test_bit(NILFS_I_BUSY, &ii->i_state);
 827                 spin_unlock(&sbi->s_inode_lock);
 828         }
 829         return ret;
 830 }
 831
 832 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 833                          unsigned nr_dirty)
 834 {
 835         struct nilfs_inode_info *ii = NILFS_I(inode);
 836
 837         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 838
 839         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 840                 return 0;
 841
 842         spin_lock(&sbi->s_inode_lock);
 843         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 844             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 845                 /* Because this routine may race with nilfs_dispose_list(),
 846                    we have to check NILFS_I_QUEUED here, too. */
 847                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 848                         /* This will happen when somebody is freeing
 849                            this inode. */
 850                         nilfs_warning(sbi->s_super, __func__,
 851                                       "cannot get inode (ino=%lu)\n",
 852                                       inode->i_ino);
 853                         spin_unlock(&sbi->s_inode_lock);
 854                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 855                                            freeing inode */
 856                 }
 857                 list_del(&ii->i_dirty);
 858                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 859                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 860         }
 861         spin_unlock(&sbi->s_inode_lock);
 862         return 0;
 863 }
 864
 865 int nilfs_mark_inode_dirty(struct inode *inode)
 866 {
 867         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 868         struct buffer_head *ibh;
 869         int err;
 870
 871         err = nilfs_load_inode_block(sbi, inode, &ibh);
 872         if (unlikely(err)) {
 873                 nilfs_warning(inode->i_sb, __func__,
 874                               "failed to reget inode block.\n");
 875                 return err;
 876         }
 877         nilfs_update_inode(inode, ibh);
 878         nilfs_mdt_mark_buffer_dirty(ibh);
 879         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 880         brelse(ibh);
 881         return 0;
 882 }
 883
 884 /**
 885  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 886  * @inode: inode of the file to be registered.
 887  *
 888  * nilfs_dirty_inode() loads a inode block containing the specified
 889  * @inode and copies data from a nilfs_inode to a corresponding inode
 890  * entry in the inode block. This operation is excluded from the segment
 891  * construction. This function can be called both as a single operation
 892  * and as a part of indivisible file operations.
 893  */
 894 void nilfs_dirty_inode(struct inode *inode)
 895 {
 896         struct nilfs_transaction_info ti;
 897         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 898
 899         if (is_bad_inode(inode)) {
 900                 nilfs_warning(inode->i_sb, __func__,
 901                               "tried to mark bad_inode dirty. ignored.\n");
 902                 dump_stack();
 903                 return;
 904         }
 905         if (mdi) {
 906                 nilfs_mdt_mark_dirty(inode);
 907                 return;
 908         }
 909         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 910         nilfs_mark_inode_dirty(inode);
 911         nilfs_transaction_commit(inode->i_sb); /* never fails */
 912 }