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         int for_gc;
  41 };
  42
  43 /**
  44  * nilfs_get_block() - get a file block on the filesystem (callback function)
  45  * @inode - inode struct of the target file
  46  * @blkoff - file block number
  47  * @bh_result - buffer head to be mapped on
  48  * @create - indicate whether allocating the block or not when it has not
  49  *      been allocated yet.
  50  *
  51  * This function does not issue actual read request of the specified data
  52  * block. It is done by VFS.
  53  */
  54 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  55                     struct buffer_head *bh_result, int create)
  56 {
  57         struct nilfs_inode_info *ii = NILFS_I(inode);
  58         __u64 blknum = 0;
  59         int err = 0, ret;
  60         struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
  61         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  62
  63         down_read(&NILFS_MDT(dat)->mi_sem);
  64         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  65         up_read(&NILFS_MDT(dat)->mi_sem);
  66         if (ret >= 0) { /* found */
  67                 map_bh(bh_result, inode->i_sb, blknum);
  68                 if (ret > 0)
  69                         bh_result->b_size = (ret << inode->i_blkbits);
  70                 goto out;
  71         }
  72         /* data block was not found */
  73         if (ret == -ENOENT && create) {
  74                 struct nilfs_transaction_info ti;
  75
  76                 bh_result->b_blocknr = 0;
  77                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  78                 if (unlikely(err))
  79                         goto out;
  80                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
  81                                         (unsigned long)bh_result);
  82                 if (unlikely(err != 0)) {
  83                         if (err == -EEXIST) {
  84                                 /*
  85                                  * The get_block() function could be called
  86                                  * from multiple callers for an inode.
  87                                  * However, the page having this block must
  88                                  * be locked in this case.
  89                                  */
  90                                 printk(KERN_WARNING
  91                                        "nilfs_get_block: a race condition "
  92                                        "while inserting a data block. "
  93                                        "(inode number=%lu, file block "
  94                                        "offset=%llu)\n",
  95                                        inode->i_ino,
  96                                        (unsigned long long)blkoff);
  97                                 err = 0;
  98                         } else if (err == -EINVAL) {
  99                                 nilfs_error(inode->i_sb, __func__,
 100                                             "broken bmap (inode=%lu)\n",
 101                                             inode->i_ino);
 102                                 err = -EIO;
 103                         }
 104                         nilfs_transaction_abort(inode->i_sb);
 105                         goto out;
 106                 }
 107                 nilfs_mark_inode_dirty(inode);
 108                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 109                 /* Error handling should be detailed */
 110                 set_buffer_new(bh_result);
 111                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 112                                                       to proper value */
 113         } else if (ret == -ENOENT) {
 114                 /* not found is not error (e.g. hole); must return without
 115                    the mapped state flag. */
 116                 ;
 117         } else {
 118                 err = ret;
 119         }
 120
 121  out:
 122         return err;
 123 }
 124
 125 /**
 126  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 127  * address_space_operations.
 128  * @file - file struct of the file to be read
 129  * @page - the page to be read
 130  */
 131 static int nilfs_readpage(struct file *file, struct page *page)
 132 {
 133         return mpage_readpage(page, nilfs_get_block);
 134 }
 135
 136 /**
 137  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 138  * address_space_operations.
 139  * @file - file struct of the file to be read
 140  * @mapping - address_space struct used for reading multiple pages
 141  * @pages - the pages to be read
 142  * @nr_pages - number of pages to be read
 143  */
 144 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 145                            struct list_head *pages, unsigned nr_pages)
 146 {
 147         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 148 }
 149
 150 static int nilfs_writepages(struct address_space *mapping,
 151                             struct writeback_control *wbc)
 152 {
 153         struct inode *inode = mapping->host;
 154         int err = 0;
 155
 156         if (wbc->sync_mode == WB_SYNC_ALL)
 157                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 158                                                     wbc->range_start,
 159                                                     wbc->range_end);
 160         return err;
 161 }
 162
 163 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 164 {
 165         struct inode *inode = page->mapping->host;
 166         int err;
 167
 168         redirty_page_for_writepage(wbc, page);
 169         unlock_page(page);
 170
 171         if (wbc->sync_mode == WB_SYNC_ALL) {
 172                 err = nilfs_construct_segment(inode->i_sb);
 173                 if (unlikely(err))
 174                         return err;
 175         } else if (wbc->for_reclaim)
 176                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 177
 178         return 0;
 179 }
 180
 181 static int nilfs_set_page_dirty(struct page *page)
 182 {
 183         int ret = __set_page_dirty_buffers(page);
 184
 185         if (ret) {
 186                 struct inode *inode = page->mapping->host;
 187                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 188                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 189
 190                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 191         }
 192         return ret;
 193 }
 194
 195 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 196                              loff_t pos, unsigned len, unsigned flags,
 197                              struct page **pagep, void **fsdata)
 198
 199 {
 200         struct inode *inode = mapping->host;
 201         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 202
 203         if (unlikely(err))
 204                 return err;
 205
 206         err = block_write_begin(mapping, pos, len, flags, pagep,
 207                                 nilfs_get_block);
 208         if (unlikely(err)) {
 209                 loff_t isize = mapping->host->i_size;
 210                 if (pos + len > isize)
 211                         vmtruncate(mapping->host, isize);
 212
 213                 nilfs_transaction_abort(inode->i_sb);
 214         }
 215         return err;
 216 }
 217
 218 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 219                            loff_t pos, unsigned len, unsigned copied,
 220                            struct page *page, void *fsdata)
 221 {
 222         struct inode *inode = mapping->host;
 223         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 224         unsigned nr_dirty;
 225         int err;
 226
 227         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 228                                                   start + copied);
 229         copied = generic_write_end(file, mapping, pos, len, copied, page,
 230                                    fsdata);
 231         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 232         err = nilfs_transaction_commit(inode->i_sb);
 233         return err ? : copied;
 234 }
 235
 236 static ssize_t
 237 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 238                 loff_t offset, unsigned long nr_segs)
 239 {
 240         struct file *file = iocb->ki_filp;
 241         struct inode *inode = file->f_mapping->host;
 242         ssize_t size;
 243
 244         if (rw == WRITE)
 245                 return 0;
 246
 247         /* Needs synchronization with the cleaner */
 248         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 249                                   offset, nr_segs, nilfs_get_block, NULL);
 250
 251         /*
 252          * In case of error extending write may have instantiated a few
 253          * blocks outside i_size. Trim these off again.
 254          */
 255         if (unlikely((rw & WRITE) && size < 0)) {
 256                 loff_t isize = i_size_read(inode);
 257                 loff_t end = offset + iov_length(iov, nr_segs);
 258
 259                 if (end > isize)
 260                         vmtruncate(inode, isize);
 261         }
 262
 263         return size;
 264 }
 265
 266 const struct address_space_operations nilfs_aops = {
 267         .writepage              = nilfs_writepage,
 268         .readpage               = nilfs_readpage,
 269         .sync_page              = block_sync_page,
 270         .writepages             = nilfs_writepages,
 271         .set_page_dirty         = nilfs_set_page_dirty,
 272         .readpages              = nilfs_readpages,
 273         .write_begin            = nilfs_write_begin,
 274         .write_end              = nilfs_write_end,
 275         /* .releasepage         = nilfs_releasepage, */
 276         .invalidatepage         = block_invalidatepage,
 277         .direct_IO              = nilfs_direct_IO,
 278         .is_partially_uptodate  = block_is_partially_uptodate,
 279 };
 280
 281 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 282 {
 283         struct super_block *sb = dir->i_sb;
 284         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 285         struct inode *inode;
 286         struct nilfs_inode_info *ii;
 287         int err = -ENOMEM;
 288         ino_t ino;
 289
 290         inode = new_inode(sb);
 291         if (unlikely(!inode))
 292                 goto failed;
 293
 294         mapping_set_gfp_mask(inode->i_mapping,
 295                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 296
 297         ii = NILFS_I(inode);
 298         ii->i_state = 1 << NILFS_I_NEW;
 299
 300         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
 301         if (unlikely(err))
 302                 goto failed_ifile_create_inode;
 303         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 304
 305         atomic_inc(&sbi->s_inodes_count);
 306         inode_init_owner(inode, dir, mode);
 307         inode->i_ino = ino;
 308         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 309
 310         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 311                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 312                 if (err < 0)
 313                         goto failed_bmap;
 314
 315                 set_bit(NILFS_I_BMAP, &ii->i_state);
 316                 /* No lock is needed; iget() ensures it. */
 317         }
 318
 319         ii->i_flags = NILFS_I(dir)->i_flags;
 320         if (S_ISLNK(mode))
 321                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 322         if (!S_ISDIR(mode))
 323                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 324
 325         /* ii->i_file_acl = 0; */
 326         /* ii->i_dir_acl = 0; */
 327         ii->i_dir_start_lookup = 0;
 328         nilfs_set_inode_flags(inode);
 329         spin_lock(&sbi->s_next_gen_lock);
 330         inode->i_generation = sbi->s_next_generation++;
 331         spin_unlock(&sbi->s_next_gen_lock);
 332         insert_inode_hash(inode);
 333
 334         err = nilfs_init_acl(inode, dir);
 335         if (unlikely(err))
 336                 goto failed_acl; /* never occur. When supporting
 337                                     nilfs_init_acl(), proper cancellation of
 338                                     above jobs should be considered */
 339
 340         return inode;
 341
 342  failed_acl:
 343  failed_bmap:
 344         inode->i_nlink = 0;
 345         iput(inode);  /* raw_inode will be deleted through
 346                          generic_delete_inode() */
 347         goto failed;
 348
 349  failed_ifile_create_inode:
 350         make_bad_inode(inode);
 351         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 352                          called */
 353  failed:
 354         return ERR_PTR(err);
 355 }
 356
 357 void nilfs_free_inode(struct inode *inode)
 358 {
 359         struct super_block *sb = inode->i_sb;
 360         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 361
 362         /* XXX: check error code? Is there any thing I can do? */
 363         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
 364         atomic_dec(&sbi->s_inodes_count);
 365 }
 366
 367 void nilfs_set_inode_flags(struct inode *inode)
 368 {
 369         unsigned int flags = NILFS_I(inode)->i_flags;
 370
 371         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 372                             S_DIRSYNC);
 373         if (flags & NILFS_SYNC_FL)
 374                 inode->i_flags |= S_SYNC;
 375         if (flags & NILFS_APPEND_FL)
 376                 inode->i_flags |= S_APPEND;
 377         if (flags & NILFS_IMMUTABLE_FL)
 378                 inode->i_flags |= S_IMMUTABLE;
 379 #ifndef NILFS_ATIME_DISABLE
 380         if (flags & NILFS_NOATIME_FL)
 381 #endif
 382                 inode->i_flags |= S_NOATIME;
 383         if (flags & NILFS_DIRSYNC_FL)
 384                 inode->i_flags |= S_DIRSYNC;
 385         mapping_set_gfp_mask(inode->i_mapping,
 386                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 387 }
 388
 389 int nilfs_read_inode_common(struct inode *inode,
 390                             struct nilfs_inode *raw_inode)
 391 {
 392         struct nilfs_inode_info *ii = NILFS_I(inode);
 393         int err;
 394
 395         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 396         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 397         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 398         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 399         inode->i_size = le64_to_cpu(raw_inode->i_size);
 400         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 401         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 402         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 403         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 404         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 405         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 406         if (inode->i_nlink == 0 && inode->i_mode == 0)
 407                 return -EINVAL; /* this inode is deleted */
 408
 409         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 410         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 411 #if 0
 412         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 413         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 414                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 415 #endif
 416         ii->i_dir_start_lookup = 0;
 417         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 418
 419         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 420             S_ISLNK(inode->i_mode)) {
 421                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 422                 if (err < 0)
 423                         return err;
 424                 set_bit(NILFS_I_BMAP, &ii->i_state);
 425                 /* No lock is needed; iget() ensures it. */
 426         }
 427         return 0;
 428 }
 429
 430 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 431                               struct inode *inode)
 432 {
 433         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 434         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 435         struct buffer_head *bh;
 436         struct nilfs_inode *raw_inode;
 437         int err;
 438
 439         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 440         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 441         if (unlikely(err))
 442                 goto bad_inode;
 443
 444         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 445
 446         err = nilfs_read_inode_common(inode, raw_inode);
 447         if (err)
 448                 goto failed_unmap;
 449
 450         if (S_ISREG(inode->i_mode)) {
 451                 inode->i_op = &nilfs_file_inode_operations;
 452                 inode->i_fop = &nilfs_file_operations;
 453                 inode->i_mapping->a_ops = &nilfs_aops;
 454         } else if (S_ISDIR(inode->i_mode)) {
 455                 inode->i_op = &nilfs_dir_inode_operations;
 456                 inode->i_fop = &nilfs_dir_operations;
 457                 inode->i_mapping->a_ops = &nilfs_aops;
 458         } else if (S_ISLNK(inode->i_mode)) {
 459                 inode->i_op = &nilfs_symlink_inode_operations;
 460                 inode->i_mapping->a_ops = &nilfs_aops;
 461         } else {
 462                 inode->i_op = &nilfs_special_inode_operations;
 463                 init_special_inode(
 464                         inode, inode->i_mode,
 465                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 466         }
 467         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 468         brelse(bh);
 469         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 470         nilfs_set_inode_flags(inode);
 471         return 0;
 472
 473  failed_unmap:
 474         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 475         brelse(bh);
 476
 477  bad_inode:
 478         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 479         return err;
 480 }
 481
 482 static int nilfs_iget_test(struct inode *inode, void *opaque)
 483 {
 484         struct nilfs_iget_args *args = opaque;
 485         struct nilfs_inode_info *ii;
 486
 487         if (args->ino != inode->i_ino)
 488                 return 0;
 489
 490         ii = NILFS_I(inode);
 491         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 492                 return !args->for_gc;
 493
 494         return args->for_gc && args->cno == ii->i_cno;
 495 }
 496
 497 static int nilfs_iget_set(struct inode *inode, void *opaque)
 498 {
 499         struct nilfs_iget_args *args = opaque;
 500
 501         inode->i_ino = args->ino;
 502         if (args->for_gc) {
 503                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 504                 NILFS_I(inode)->i_cno = args->cno;
 505         }
 506         return 0;
 507 }
 508
 509 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 510 {
 511         struct nilfs_iget_args args = { .ino = ino, .cno = 0, .for_gc = 0 };
 512         struct inode *inode;
 513         int err;
 514
 515         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 516         if (unlikely(!inode))
 517                 return ERR_PTR(-ENOMEM);
 518         if (!(inode->i_state & I_NEW))
 519                 return inode;
 520
 521         err = __nilfs_read_inode(sb, ino, inode);
 522         if (unlikely(err)) {
 523                 iget_failed(inode);
 524                 return ERR_PTR(err);
 525         }
 526         unlock_new_inode(inode);
 527         return inode;
 528 }
 529
 530 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 531                                 __u64 cno)
 532 {
 533         struct nilfs_iget_args args = { .ino = ino, .cno = cno, .for_gc = 1 };
 534         struct inode *inode;
 535         int err;
 536
 537         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 538         if (unlikely(!inode))
 539                 return ERR_PTR(-ENOMEM);
 540         if (!(inode->i_state & I_NEW))
 541                 return inode;
 542
 543         err = nilfs_init_gcinode(inode);
 544         if (unlikely(err)) {
 545                 iget_failed(inode);
 546                 return ERR_PTR(err);
 547         }
 548         unlock_new_inode(inode);
 549         return inode;
 550 }
 551
 552 void nilfs_write_inode_common(struct inode *inode,
 553                               struct nilfs_inode *raw_inode, int has_bmap)
 554 {
 555         struct nilfs_inode_info *ii = NILFS_I(inode);
 556
 557         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 558         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 559         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 560         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 561         raw_inode->i_size = cpu_to_le64(inode->i_size);
 562         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 563         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 564         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 565         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 566         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 567
 568         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 569         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 570
 571         if (has_bmap)
 572                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 573         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 574                 raw_inode->i_device_code =
 575                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 576         /* When extending inode, nilfs->ns_inode_size should be checked
 577            for substitutions of appended fields */
 578 }
 579
 580 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 581 {
 582         ino_t ino = inode->i_ino;
 583         struct nilfs_inode_info *ii = NILFS_I(inode);
 584         struct super_block *sb = inode->i_sb;
 585         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 586         struct nilfs_inode *raw_inode;
 587
 588         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
 589
 590         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 591                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 592         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 593
 594         nilfs_write_inode_common(inode, raw_inode, 0);
 595                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 596                    deadlock of bmap. This delays update of i_bmap to just
 597                    before writing */
 598         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 599 }
 600
 601 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 602
 603 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 604                                 unsigned long from)
 605 {
 606         unsigned long b;
 607         int ret;
 608
 609         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 610                 return;
 611  repeat:
 612         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 613         if (ret == -ENOENT)
 614                 return;
 615         else if (ret < 0)
 616                 goto failed;
 617
 618         if (b < from)
 619                 return;
 620
 621         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 622         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 623         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 624         if (!ret || (ret == -ENOMEM &&
 625                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 626                 goto repeat;
 627
 628  failed:
 629         if (ret == -EINVAL)
 630                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 631                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 632         else
 633                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 634                               "failed to truncate bmap (ino=%lu, err=%d)",
 635                               ii->vfs_inode.i_ino, ret);
 636 }
 637
 638 void nilfs_truncate(struct inode *inode)
 639 {
 640         unsigned long blkoff;
 641         unsigned int blocksize;
 642         struct nilfs_transaction_info ti;
 643         struct super_block *sb = inode->i_sb;
 644         struct nilfs_inode_info *ii = NILFS_I(inode);
 645
 646         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 647                 return;
 648         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 649                 return;
 650
 651         blocksize = sb->s_blocksize;
 652         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 653         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 654
 655         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 656
 657         nilfs_truncate_bmap(ii, blkoff);
 658
 659         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 660         if (IS_SYNC(inode))
 661                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 662
 663         nilfs_mark_inode_dirty(inode);
 664         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 665         nilfs_transaction_commit(sb);
 666         /* May construct a logical segment and may fail in sync mode.
 667            But truncate has no return value. */
 668 }
 669
 670 static void nilfs_clear_inode(struct inode *inode)
 671 {
 672         struct nilfs_inode_info *ii = NILFS_I(inode);
 673
 674         /*
 675          * Free resources allocated in nilfs_read_inode(), here.
 676          */
 677         BUG_ON(!list_empty(&ii->i_dirty));
 678         brelse(ii->i_bh);
 679         ii->i_bh = NULL;
 680
 681         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 682                 nilfs_bmap_clear(ii->i_bmap);
 683
 684         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 685 }
 686
 687 void nilfs_evict_inode(struct inode *inode)
 688 {
 689         struct nilfs_transaction_info ti;
 690         struct super_block *sb = inode->i_sb;
 691         struct nilfs_inode_info *ii = NILFS_I(inode);
 692
 693         if (inode->i_nlink || unlikely(is_bad_inode(inode))) {
 694                 if (inode->i_data.nrpages)
 695                         truncate_inode_pages(&inode->i_data, 0);
 696                 end_writeback(inode);
 697                 nilfs_clear_inode(inode);
 698                 return;
 699         }
 700         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 701
 702         if (inode->i_data.nrpages)
 703                 truncate_inode_pages(&inode->i_data, 0);
 704
 705         nilfs_truncate_bmap(ii, 0);
 706         nilfs_mark_inode_dirty(inode);
 707         end_writeback(inode);
 708         nilfs_clear_inode(inode);
 709         nilfs_free_inode(inode);
 710         /* nilfs_free_inode() marks inode buffer dirty */
 711         if (IS_SYNC(inode))
 712                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 713         nilfs_transaction_commit(sb);
 714         /* May construct a logical segment and may fail in sync mode.
 715            But delete_inode has no return value. */
 716 }
 717
 718 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 719 {
 720         struct nilfs_transaction_info ti;
 721         struct inode *inode = dentry->d_inode;
 722         struct super_block *sb = inode->i_sb;
 723         int err;
 724
 725         err = inode_change_ok(inode, iattr);
 726         if (err)
 727                 return err;
 728
 729         err = nilfs_transaction_begin(sb, &ti, 0);
 730         if (unlikely(err))
 731                 return err;
 732
 733         if ((iattr->ia_valid & ATTR_SIZE) &&
 734             iattr->ia_size != i_size_read(inode)) {
 735                 err = vmtruncate(inode, iattr->ia_size);
 736                 if (unlikely(err))
 737                         goto out_err;
 738         }
 739
 740         setattr_copy(inode, iattr);
 741         mark_inode_dirty(inode);
 742
 743         if (iattr->ia_valid & ATTR_MODE) {
 744                 err = nilfs_acl_chmod(inode);
 745                 if (unlikely(err))
 746                         goto out_err;
 747         }
 748
 749         return nilfs_transaction_commit(sb);
 750
 751 out_err:
 752         nilfs_transaction_abort(sb);
 753         return err;
 754 }
 755
 756 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 757                            struct buffer_head **pbh)
 758 {
 759         struct nilfs_inode_info *ii = NILFS_I(inode);
 760         int err;
 761
 762         spin_lock(&sbi->s_inode_lock);
 763         if (ii->i_bh == NULL) {
 764                 spin_unlock(&sbi->s_inode_lock);
 765                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
 766                                                   pbh);
 767                 if (unlikely(err))
 768                         return err;
 769                 spin_lock(&sbi->s_inode_lock);
 770                 if (ii->i_bh == NULL)
 771                         ii->i_bh = *pbh;
 772                 else {
 773                         brelse(*pbh);
 774                         *pbh = ii->i_bh;
 775                 }
 776         } else
 777                 *pbh = ii->i_bh;
 778
 779         get_bh(*pbh);
 780         spin_unlock(&sbi->s_inode_lock);
 781         return 0;
 782 }
 783
 784 int nilfs_inode_dirty(struct inode *inode)
 785 {
 786         struct nilfs_inode_info *ii = NILFS_I(inode);
 787         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 788         int ret = 0;
 789
 790         if (!list_empty(&ii->i_dirty)) {
 791                 spin_lock(&sbi->s_inode_lock);
 792                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 793                         test_bit(NILFS_I_BUSY, &ii->i_state);
 794                 spin_unlock(&sbi->s_inode_lock);
 795         }
 796         return ret;
 797 }
 798
 799 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 800                          unsigned nr_dirty)
 801 {
 802         struct nilfs_inode_info *ii = NILFS_I(inode);
 803
 804         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 805
 806         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 807                 return 0;
 808
 809         spin_lock(&sbi->s_inode_lock);
 810         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 811             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 812                 /* Because this routine may race with nilfs_dispose_list(),
 813                    we have to check NILFS_I_QUEUED here, too. */
 814                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 815                         /* This will happen when somebody is freeing
 816                            this inode. */
 817                         nilfs_warning(sbi->s_super, __func__,
 818                                       "cannot get inode (ino=%lu)\n",
 819                                       inode->i_ino);
 820                         spin_unlock(&sbi->s_inode_lock);
 821                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 822                                            freeing inode */
 823                 }
 824                 list_del(&ii->i_dirty);
 825                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 826                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 827         }
 828         spin_unlock(&sbi->s_inode_lock);
 829         return 0;
 830 }
 831
 832 int nilfs_mark_inode_dirty(struct inode *inode)
 833 {
 834         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 835         struct buffer_head *ibh;
 836         int err;
 837
 838         err = nilfs_load_inode_block(sbi, inode, &ibh);
 839         if (unlikely(err)) {
 840                 nilfs_warning(inode->i_sb, __func__,
 841                               "failed to reget inode block.\n");
 842                 return err;
 843         }
 844         nilfs_update_inode(inode, ibh);
 845         nilfs_mdt_mark_buffer_dirty(ibh);
 846         nilfs_mdt_mark_dirty(sbi->s_ifile);
 847         brelse(ibh);
 848         return 0;
 849 }
 850
 851 /**
 852  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 853  * @inode: inode of the file to be registered.
 854  *
 855  * nilfs_dirty_inode() loads a inode block containing the specified
 856  * @inode and copies data from a nilfs_inode to a corresponding inode
 857  * entry in the inode block. This operation is excluded from the segment
 858  * construction. This function can be called both as a single operation
 859  * and as a part of indivisible file operations.
 860  */
 861 void nilfs_dirty_inode(struct inode *inode)
 862 {
 863         struct nilfs_transaction_info ti;
 864         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 865
 866         if (is_bad_inode(inode)) {
 867                 nilfs_warning(inode->i_sb, __func__,
 868                               "tried to mark bad_inode dirty. ignored.\n");
 869                 dump_stack();
 870                 return;
 871         }
 872         if (mdi) {
 873                 nilfs_mdt_mark_dirty(inode);
 874                 return;
 875         }
 876         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 877         nilfs_mark_inode_dirty(inode);
 878         nilfs_transaction_commit(inode->i_sb); /* never fails */
 879 }