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
  38 /**
  39  * nilfs_get_block() - get a file block on the filesystem (callback function)
  40  * @inode - inode struct of the target file
  41  * @blkoff - file block number
  42  * @bh_result - buffer head to be mapped on
  43  * @create - indicate whether allocating the block or not when it has not
  44  *      been allocated yet.
  45  *
  46  * This function does not issue actual read request of the specified data
  47  * block. It is done by VFS.
  48  */
  49 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  50                     struct buffer_head *bh_result, int create)
  51 {
  52         struct nilfs_inode_info *ii = NILFS_I(inode);
  53         __u64 blknum = 0;
  54         int err = 0, ret;
  55         struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
  56         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  57
  58         down_read(&NILFS_MDT(dat)->mi_sem);
  59         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  60         up_read(&NILFS_MDT(dat)->mi_sem);
  61         if (ret >= 0) { /* found */
  62                 map_bh(bh_result, inode->i_sb, blknum);
  63                 if (ret > 0)
  64                         bh_result->b_size = (ret << inode->i_blkbits);
  65                 goto out;
  66         }
  67         /* data block was not found */
  68         if (ret == -ENOENT && create) {
  69                 struct nilfs_transaction_info ti;
  70
  71                 bh_result->b_blocknr = 0;
  72                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  73                 if (unlikely(err))
  74                         goto out;
  75                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
  76                                         (unsigned long)bh_result);
  77                 if (unlikely(err != 0)) {
  78                         if (err == -EEXIST) {
  79                                 /*
  80                                  * The get_block() function could be called
  81                                  * from multiple callers for an inode.
  82                                  * However, the page having this block must
  83                                  * be locked in this case.
  84                                  */
  85                                 printk(KERN_WARNING
  86                                        "nilfs_get_block: a race condition "
  87                                        "while inserting a data block. "
  88                                        "(inode number=%lu, file block "
  89                                        "offset=%llu)\n",
  90                                        inode->i_ino,
  91                                        (unsigned long long)blkoff);
  92                                 err = 0;
  93                         } else if (err == -EINVAL) {
  94                                 nilfs_error(inode->i_sb, __func__,
  95                                             "broken bmap (inode=%lu)\n",
  96                                             inode->i_ino);
  97                                 err = -EIO;
  98                         }
  99                         nilfs_transaction_abort(inode->i_sb);
 100                         goto out;
 101                 }
 102                 nilfs_mark_inode_dirty(inode);
 103                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 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         struct inode *inode = mapping->host;
 149         int err = 0;
 150
 151         if (wbc->sync_mode == WB_SYNC_ALL)
 152                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 153                                                     wbc->range_start,
 154                                                     wbc->range_end);
 155         return err;
 156 }
 157
 158 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 159 {
 160         struct inode *inode = page->mapping->host;
 161         int err;
 162
 163         redirty_page_for_writepage(wbc, page);
 164         unlock_page(page);
 165
 166         if (wbc->sync_mode == WB_SYNC_ALL) {
 167                 err = nilfs_construct_segment(inode->i_sb);
 168                 if (unlikely(err))
 169                         return err;
 170         } else if (wbc->for_reclaim)
 171                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 172
 173         return 0;
 174 }
 175
 176 static int nilfs_set_page_dirty(struct page *page)
 177 {
 178         int ret = __set_page_dirty_buffers(page);
 179
 180         if (ret) {
 181                 struct inode *inode = page->mapping->host;
 182                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 183                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 184
 185                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 186         }
 187         return ret;
 188 }
 189
 190 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 191                              loff_t pos, unsigned len, unsigned flags,
 192                              struct page **pagep, void **fsdata)
 193
 194 {
 195         struct inode *inode = mapping->host;
 196         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 197
 198         if (unlikely(err))
 199                 return err;
 200
 201         err = block_write_begin(mapping, pos, len, flags, pagep,
 202                                 nilfs_get_block);
 203         if (unlikely(err)) {
 204                 loff_t isize = mapping->host->i_size;
 205                 if (pos + len > isize)
 206                         vmtruncate(mapping->host, isize);
 207
 208                 nilfs_transaction_abort(inode->i_sb);
 209         }
 210         return err;
 211 }
 212
 213 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 214                            loff_t pos, unsigned len, unsigned copied,
 215                            struct page *page, void *fsdata)
 216 {
 217         struct inode *inode = mapping->host;
 218         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 219         unsigned nr_dirty;
 220         int err;
 221
 222         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 223                                                   start + copied);
 224         copied = generic_write_end(file, mapping, pos, len, copied, page,
 225                                    fsdata);
 226         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 227         err = nilfs_transaction_commit(inode->i_sb);
 228         return err ? : copied;
 229 }
 230
 231 static ssize_t
 232 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 233                 loff_t offset, unsigned long nr_segs)
 234 {
 235         struct file *file = iocb->ki_filp;
 236         struct inode *inode = file->f_mapping->host;
 237         ssize_t size;
 238
 239         if (rw == WRITE)
 240                 return 0;
 241
 242         /* Needs synchronization with the cleaner */
 243         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 244                                   offset, nr_segs, nilfs_get_block, NULL);
 245
 246         /*
 247          * In case of error extending write may have instantiated a few
 248          * blocks outside i_size. Trim these off again.
 249          */
 250         if (unlikely((rw & WRITE) && size < 0)) {
 251                 loff_t isize = i_size_read(inode);
 252                 loff_t end = offset + iov_length(iov, nr_segs);
 253
 254                 if (end > isize)
 255                         vmtruncate(inode, isize);
 256         }
 257
 258         return size;
 259 }
 260
 261 const struct address_space_operations nilfs_aops = {
 262         .writepage              = nilfs_writepage,
 263         .readpage               = nilfs_readpage,
 264         .sync_page              = block_sync_page,
 265         .writepages             = nilfs_writepages,
 266         .set_page_dirty         = nilfs_set_page_dirty,
 267         .readpages              = nilfs_readpages,
 268         .write_begin            = nilfs_write_begin,
 269         .write_end              = nilfs_write_end,
 270         /* .releasepage         = nilfs_releasepage, */
 271         .invalidatepage         = block_invalidatepage,
 272         .direct_IO              = nilfs_direct_IO,
 273         .is_partially_uptodate  = block_is_partially_uptodate,
 274 };
 275
 276 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 277 {
 278         struct super_block *sb = dir->i_sb;
 279         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 280         struct inode *inode;
 281         struct nilfs_inode_info *ii;
 282         int err = -ENOMEM;
 283         ino_t ino;
 284
 285         inode = new_inode(sb);
 286         if (unlikely(!inode))
 287                 goto failed;
 288
 289         mapping_set_gfp_mask(inode->i_mapping,
 290                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 291
 292         ii = NILFS_I(inode);
 293         ii->i_state = 1 << NILFS_I_NEW;
 294
 295         err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
 296         if (unlikely(err))
 297                 goto failed_ifile_create_inode;
 298         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 299
 300         atomic_inc(&sbi->s_inodes_count);
 301         inode_init_owner(inode, dir, mode);
 302         inode->i_ino = ino;
 303         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 304
 305         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 306                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 307                 if (err < 0)
 308                         goto failed_bmap;
 309
 310                 set_bit(NILFS_I_BMAP, &ii->i_state);
 311                 /* No lock is needed; iget() ensures it. */
 312         }
 313
 314         ii->i_flags = NILFS_I(dir)->i_flags;
 315         if (S_ISLNK(mode))
 316                 ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
 317         if (!S_ISDIR(mode))
 318                 ii->i_flags &= ~NILFS_DIRSYNC_FL;
 319
 320         /* ii->i_file_acl = 0; */
 321         /* ii->i_dir_acl = 0; */
 322         ii->i_dir_start_lookup = 0;
 323         ii->i_cno = 0;
 324         nilfs_set_inode_flags(inode);
 325         spin_lock(&sbi->s_next_gen_lock);
 326         inode->i_generation = sbi->s_next_generation++;
 327         spin_unlock(&sbi->s_next_gen_lock);
 328         insert_inode_hash(inode);
 329
 330         err = nilfs_init_acl(inode, dir);
 331         if (unlikely(err))
 332                 goto failed_acl; /* never occur. When supporting
 333                                     nilfs_init_acl(), proper cancellation of
 334                                     above jobs should be considered */
 335
 336         return inode;
 337
 338  failed_acl:
 339  failed_bmap:
 340         inode->i_nlink = 0;
 341         iput(inode);  /* raw_inode will be deleted through
 342                          generic_delete_inode() */
 343         goto failed;
 344
 345  failed_ifile_create_inode:
 346         make_bad_inode(inode);
 347         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 348                          called */
 349  failed:
 350         return ERR_PTR(err);
 351 }
 352
 353 void nilfs_free_inode(struct inode *inode)
 354 {
 355         struct super_block *sb = inode->i_sb;
 356         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 357
 358         /* XXX: check error code? Is there any thing I can do? */
 359         (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
 360         atomic_dec(&sbi->s_inodes_count);
 361 }
 362
 363 void nilfs_set_inode_flags(struct inode *inode)
 364 {
 365         unsigned int flags = NILFS_I(inode)->i_flags;
 366
 367         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 368                             S_DIRSYNC);
 369         if (flags & NILFS_SYNC_FL)
 370                 inode->i_flags |= S_SYNC;
 371         if (flags & NILFS_APPEND_FL)
 372                 inode->i_flags |= S_APPEND;
 373         if (flags & NILFS_IMMUTABLE_FL)
 374                 inode->i_flags |= S_IMMUTABLE;
 375 #ifndef NILFS_ATIME_DISABLE
 376         if (flags & NILFS_NOATIME_FL)
 377 #endif
 378                 inode->i_flags |= S_NOATIME;
 379         if (flags & NILFS_DIRSYNC_FL)
 380                 inode->i_flags |= S_DIRSYNC;
 381         mapping_set_gfp_mask(inode->i_mapping,
 382                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 383 }
 384
 385 int nilfs_read_inode_common(struct inode *inode,
 386                             struct nilfs_inode *raw_inode)
 387 {
 388         struct nilfs_inode_info *ii = NILFS_I(inode);
 389         int err;
 390
 391         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 392         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 393         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 394         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 395         inode->i_size = le64_to_cpu(raw_inode->i_size);
 396         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 397         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 398         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 399         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 400         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 401         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 402         if (inode->i_nlink == 0 && inode->i_mode == 0)
 403                 return -EINVAL; /* this inode is deleted */
 404
 405         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 406         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 407 #if 0
 408         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 409         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 410                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 411 #endif
 412         ii->i_dir_start_lookup = 0;
 413         ii->i_cno = 0;
 414         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 415
 416         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 417             S_ISLNK(inode->i_mode)) {
 418                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 419                 if (err < 0)
 420                         return err;
 421                 set_bit(NILFS_I_BMAP, &ii->i_state);
 422                 /* No lock is needed; iget() ensures it. */
 423         }
 424         return 0;
 425 }
 426
 427 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
 428                               struct inode *inode)
 429 {
 430         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 431         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 432         struct buffer_head *bh;
 433         struct nilfs_inode *raw_inode;
 434         int err;
 435
 436         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 437         err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
 438         if (unlikely(err))
 439                 goto bad_inode;
 440
 441         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
 442
 443         err = nilfs_read_inode_common(inode, raw_inode);
 444         if (err)
 445                 goto failed_unmap;
 446
 447         if (S_ISREG(inode->i_mode)) {
 448                 inode->i_op = &nilfs_file_inode_operations;
 449                 inode->i_fop = &nilfs_file_operations;
 450                 inode->i_mapping->a_ops = &nilfs_aops;
 451         } else if (S_ISDIR(inode->i_mode)) {
 452                 inode->i_op = &nilfs_dir_inode_operations;
 453                 inode->i_fop = &nilfs_dir_operations;
 454                 inode->i_mapping->a_ops = &nilfs_aops;
 455         } else if (S_ISLNK(inode->i_mode)) {
 456                 inode->i_op = &nilfs_symlink_inode_operations;
 457                 inode->i_mapping->a_ops = &nilfs_aops;
 458         } else {
 459                 inode->i_op = &nilfs_special_inode_operations;
 460                 init_special_inode(
 461                         inode, inode->i_mode,
 462                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 463         }
 464         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 465         brelse(bh);
 466         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 467         nilfs_set_inode_flags(inode);
 468         return 0;
 469
 470  failed_unmap:
 471         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
 472         brelse(bh);
 473
 474  bad_inode:
 475         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 476         return err;
 477 }
 478
 479 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
 480 {
 481         struct inode *inode;
 482         int err;
 483
 484         inode = iget_locked(sb, ino);
 485         if (unlikely(!inode))
 486                 return ERR_PTR(-ENOMEM);
 487         if (!(inode->i_state & I_NEW))
 488                 return inode;
 489
 490         err = __nilfs_read_inode(sb, ino, inode);
 491         if (unlikely(err)) {
 492                 iget_failed(inode);
 493                 return ERR_PTR(err);
 494         }
 495         unlock_new_inode(inode);
 496         return inode;
 497 }
 498
 499 void nilfs_write_inode_common(struct inode *inode,
 500                               struct nilfs_inode *raw_inode, int has_bmap)
 501 {
 502         struct nilfs_inode_info *ii = NILFS_I(inode);
 503
 504         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 505         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 506         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 507         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 508         raw_inode->i_size = cpu_to_le64(inode->i_size);
 509         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 510         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 511         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 512         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 513         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 514
 515         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 516         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 517
 518         if (has_bmap)
 519                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 520         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 521                 raw_inode->i_device_code =
 522                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 523         /* When extending inode, nilfs->ns_inode_size should be checked
 524            for substitutions of appended fields */
 525 }
 526
 527 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 528 {
 529         ino_t ino = inode->i_ino;
 530         struct nilfs_inode_info *ii = NILFS_I(inode);
 531         struct super_block *sb = inode->i_sb;
 532         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 533         struct nilfs_inode *raw_inode;
 534
 535         raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
 536
 537         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 538                 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
 539         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 540
 541         nilfs_write_inode_common(inode, raw_inode, 0);
 542                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 543                    deadlock of bmap. This delays update of i_bmap to just
 544                    before writing */
 545         nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
 546 }
 547
 548 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 549
 550 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 551                                 unsigned long from)
 552 {
 553         unsigned long b;
 554         int ret;
 555
 556         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 557                 return;
 558  repeat:
 559         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 560         if (ret == -ENOENT)
 561                 return;
 562         else if (ret < 0)
 563                 goto failed;
 564
 565         if (b < from)
 566                 return;
 567
 568         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 569         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 570         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 571         if (!ret || (ret == -ENOMEM &&
 572                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 573                 goto repeat;
 574
 575  failed:
 576         if (ret == -EINVAL)
 577                 nilfs_error(ii->vfs_inode.i_sb, __func__,
 578                             "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
 579         else
 580                 nilfs_warning(ii->vfs_inode.i_sb, __func__,
 581                               "failed to truncate bmap (ino=%lu, err=%d)",
 582                               ii->vfs_inode.i_ino, ret);
 583 }
 584
 585 void nilfs_truncate(struct inode *inode)
 586 {
 587         unsigned long blkoff;
 588         unsigned int blocksize;
 589         struct nilfs_transaction_info ti;
 590         struct super_block *sb = inode->i_sb;
 591         struct nilfs_inode_info *ii = NILFS_I(inode);
 592
 593         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 594                 return;
 595         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 596                 return;
 597
 598         blocksize = sb->s_blocksize;
 599         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 600         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 601
 602         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 603
 604         nilfs_truncate_bmap(ii, blkoff);
 605
 606         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 607         if (IS_SYNC(inode))
 608                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 609
 610         nilfs_mark_inode_dirty(inode);
 611         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 612         nilfs_transaction_commit(sb);
 613         /* May construct a logical segment and may fail in sync mode.
 614            But truncate has no return value. */
 615 }
 616
 617 static void nilfs_clear_inode(struct inode *inode)
 618 {
 619         struct nilfs_inode_info *ii = NILFS_I(inode);
 620
 621         /*
 622          * Free resources allocated in nilfs_read_inode(), here.
 623          */
 624         BUG_ON(!list_empty(&ii->i_dirty));
 625         brelse(ii->i_bh);
 626         ii->i_bh = NULL;
 627
 628         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 629                 nilfs_bmap_clear(ii->i_bmap);
 630
 631         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 632 }
 633
 634 void nilfs_evict_inode(struct inode *inode)
 635 {
 636         struct nilfs_transaction_info ti;
 637         struct super_block *sb = inode->i_sb;
 638         struct nilfs_inode_info *ii = NILFS_I(inode);
 639
 640         if (inode->i_nlink || unlikely(is_bad_inode(inode))) {
 641                 if (inode->i_data.nrpages)
 642                         truncate_inode_pages(&inode->i_data, 0);
 643                 end_writeback(inode);
 644                 nilfs_clear_inode(inode);
 645                 return;
 646         }
 647         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 648
 649         if (inode->i_data.nrpages)
 650                 truncate_inode_pages(&inode->i_data, 0);
 651
 652         nilfs_truncate_bmap(ii, 0);
 653         nilfs_mark_inode_dirty(inode);
 654         end_writeback(inode);
 655         nilfs_clear_inode(inode);
 656         nilfs_free_inode(inode);
 657         /* nilfs_free_inode() marks inode buffer dirty */
 658         if (IS_SYNC(inode))
 659                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 660         nilfs_transaction_commit(sb);
 661         /* May construct a logical segment and may fail in sync mode.
 662            But delete_inode has no return value. */
 663 }
 664
 665 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 666 {
 667         struct nilfs_transaction_info ti;
 668         struct inode *inode = dentry->d_inode;
 669         struct super_block *sb = inode->i_sb;
 670         int err;
 671
 672         err = inode_change_ok(inode, iattr);
 673         if (err)
 674                 return err;
 675
 676         err = nilfs_transaction_begin(sb, &ti, 0);
 677         if (unlikely(err))
 678                 return err;
 679
 680         if ((iattr->ia_valid & ATTR_SIZE) &&
 681             iattr->ia_size != i_size_read(inode)) {
 682                 err = vmtruncate(inode, iattr->ia_size);
 683                 if (unlikely(err))
 684                         goto out_err;
 685         }
 686
 687         setattr_copy(inode, iattr);
 688         mark_inode_dirty(inode);
 689
 690         if (iattr->ia_valid & ATTR_MODE) {
 691                 err = nilfs_acl_chmod(inode);
 692                 if (unlikely(err))
 693                         goto out_err;
 694         }
 695
 696         return nilfs_transaction_commit(sb);
 697
 698 out_err:
 699         nilfs_transaction_abort(sb);
 700         return err;
 701 }
 702
 703 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 704                            struct buffer_head **pbh)
 705 {
 706         struct nilfs_inode_info *ii = NILFS_I(inode);
 707         int err;
 708
 709         spin_lock(&sbi->s_inode_lock);
 710         if (ii->i_bh == NULL) {
 711                 spin_unlock(&sbi->s_inode_lock);
 712                 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
 713                                                   pbh);
 714                 if (unlikely(err))
 715                         return err;
 716                 spin_lock(&sbi->s_inode_lock);
 717                 if (ii->i_bh == NULL)
 718                         ii->i_bh = *pbh;
 719                 else {
 720                         brelse(*pbh);
 721                         *pbh = ii->i_bh;
 722                 }
 723         } else
 724                 *pbh = ii->i_bh;
 725
 726         get_bh(*pbh);
 727         spin_unlock(&sbi->s_inode_lock);
 728         return 0;
 729 }
 730
 731 int nilfs_inode_dirty(struct inode *inode)
 732 {
 733         struct nilfs_inode_info *ii = NILFS_I(inode);
 734         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 735         int ret = 0;
 736
 737         if (!list_empty(&ii->i_dirty)) {
 738                 spin_lock(&sbi->s_inode_lock);
 739                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 740                         test_bit(NILFS_I_BUSY, &ii->i_state);
 741                 spin_unlock(&sbi->s_inode_lock);
 742         }
 743         return ret;
 744 }
 745
 746 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 747                          unsigned nr_dirty)
 748 {
 749         struct nilfs_inode_info *ii = NILFS_I(inode);
 750
 751         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 752
 753         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 754                 return 0;
 755
 756         spin_lock(&sbi->s_inode_lock);
 757         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 758             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 759                 /* Because this routine may race with nilfs_dispose_list(),
 760                    we have to check NILFS_I_QUEUED here, too. */
 761                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 762                         /* This will happen when somebody is freeing
 763                            this inode. */
 764                         nilfs_warning(sbi->s_super, __func__,
 765                                       "cannot get inode (ino=%lu)\n",
 766                                       inode->i_ino);
 767                         spin_unlock(&sbi->s_inode_lock);
 768                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 769                                            freeing inode */
 770                 }
 771                 list_del(&ii->i_dirty);
 772                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 773                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 774         }
 775         spin_unlock(&sbi->s_inode_lock);
 776         return 0;
 777 }
 778
 779 int nilfs_mark_inode_dirty(struct inode *inode)
 780 {
 781         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 782         struct buffer_head *ibh;
 783         int err;
 784
 785         err = nilfs_load_inode_block(sbi, inode, &ibh);
 786         if (unlikely(err)) {
 787                 nilfs_warning(inode->i_sb, __func__,
 788                               "failed to reget inode block.\n");
 789                 return err;
 790         }
 791         nilfs_update_inode(inode, ibh);
 792         nilfs_mdt_mark_buffer_dirty(ibh);
 793         nilfs_mdt_mark_dirty(sbi->s_ifile);
 794         brelse(ibh);
 795         return 0;
 796 }
 797
 798 /**
 799  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 800  * @inode: inode of the file to be registered.
 801  *
 802  * nilfs_dirty_inode() loads a inode block containing the specified
 803  * @inode and copies data from a nilfs_inode to a corresponding inode
 804  * entry in the inode block. This operation is excluded from the segment
 805  * construction. This function can be called both as a single operation
 806  * and as a part of indivisible file operations.
 807  */
 808 void nilfs_dirty_inode(struct inode *inode)
 809 {
 810         struct nilfs_transaction_info ti;
 811
 812         if (is_bad_inode(inode)) {
 813                 nilfs_warning(inode->i_sb, __func__,
 814                               "tried to mark bad_inode dirty. ignored.\n");
 815                 dump_stack();
 816                 return;
 817         }
 818         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 819         nilfs_mark_inode_dirty(inode);
 820         nilfs_transaction_commit(inode->i_sb); /* never fails */
 821 }