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                         }
 100                         nilfs_transaction_abort(inode->i_sb);
 101                         goto out;
 102                 }
 103                 nilfs_mark_inode_dirty(inode);
 104                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 105                 /* Error handling should be detailed */
 106                 set_buffer_new(bh_result);
 107                 set_buffer_delay(bh_result);
 108                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 109                                                       to proper value */
 110         } else if (ret == -ENOENT) {
 111                 /* not found is not error (e.g. hole); must return without
 112                    the mapped state flag. */
 113                 ;
 114         } else {
 115                 err = ret;
 116         }
 117
 118  out:
 119         return err;
 120 }
 121
 122 /**
 123  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 124  * address_space_operations.
 125  * @file - file struct of the file to be read
 126  * @page - the page to be read
 127  */
 128 static int nilfs_readpage(struct file *file, struct page *page)
 129 {
 130         return mpage_readpage(page, nilfs_get_block);
 131 }
 132
 133 /**
 134  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 135  * address_space_operations.
 136  * @file - file struct of the file to be read
 137  * @mapping - address_space struct used for reading multiple pages
 138  * @pages - the pages to be read
 139  * @nr_pages - number of pages to be read
 140  */
 141 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 142                            struct list_head *pages, unsigned nr_pages)
 143 {
 144         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 145 }
 146
 147 static int nilfs_writepages(struct address_space *mapping,
 148                             struct writeback_control *wbc)
 149 {
 150         struct inode *inode = mapping->host;
 151         int err = 0;
 152
 153         if (wbc->sync_mode == WB_SYNC_ALL)
 154                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 155                                                     wbc->range_start,
 156                                                     wbc->range_end);
 157         return err;
 158 }
 159
 160 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 161 {
 162         struct inode *inode = page->mapping->host;
 163         int err;
 164
 165         redirty_page_for_writepage(wbc, page);
 166         unlock_page(page);
 167
 168         if (wbc->sync_mode == WB_SYNC_ALL) {
 169                 err = nilfs_construct_segment(inode->i_sb);
 170                 if (unlikely(err))
 171                         return err;
 172         } else if (wbc->for_reclaim)
 173                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 174
 175         return 0;
 176 }
 177
 178 static int nilfs_set_page_dirty(struct page *page)
 179 {
 180         int ret = __set_page_dirty_buffers(page);
 181
 182         if (ret) {
 183                 struct inode *inode = page->mapping->host;
 184                 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 185                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 186
 187                 nilfs_set_file_dirty(sbi, inode, nr_dirty);
 188         }
 189         return ret;
 190 }
 191
 192 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 193                              loff_t pos, unsigned len, unsigned flags,
 194                              struct page **pagep, void **fsdata)
 195
 196 {
 197         struct inode *inode = mapping->host;
 198         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 199
 200         if (unlikely(err))
 201                 return err;
 202
 203         err = block_write_begin(mapping, pos, len, flags, pagep,
 204                                 nilfs_get_block);
 205         if (unlikely(err)) {
 206                 loff_t isize = mapping->host->i_size;
 207                 if (pos + len > isize)
 208                         vmtruncate(mapping->host, isize);
 209
 210                 nilfs_transaction_abort(inode->i_sb);
 211         }
 212         return err;
 213 }
 214
 215 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 216                            loff_t pos, unsigned len, unsigned copied,
 217                            struct page *page, void *fsdata)
 218 {
 219         struct inode *inode = mapping->host;
 220         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 221         unsigned nr_dirty;
 222         int err;
 223
 224         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 225                                                   start + copied);
 226         copied = generic_write_end(file, mapping, pos, len, copied, page,
 227                                    fsdata);
 228         nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
 229         err = nilfs_transaction_commit(inode->i_sb);
 230         return err ? : copied;
 231 }
 232
 233 static ssize_t
 234 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 235                 loff_t offset, unsigned long nr_segs)
 236 {
 237         struct file *file = iocb->ki_filp;
 238         struct inode *inode = file->f_mapping->host;
 239         ssize_t size;
 240
 241         if (rw == WRITE)
 242                 return 0;
 243
 244         /* Needs synchronization with the cleaner */
 245         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 246                                   offset, nr_segs, nilfs_get_block, NULL);
 247
 248         /*
 249          * In case of error extending write may have instantiated a few
 250          * blocks outside i_size. Trim these off again.
 251          */
 252         if (unlikely((rw & WRITE) && size < 0)) {
 253                 loff_t isize = i_size_read(inode);
 254                 loff_t end = offset + iov_length(iov, nr_segs);
 255
 256                 if (end > isize)
 257                         vmtruncate(inode, isize);
 258         }
 259
 260         return size;
 261 }
 262
 263 const struct address_space_operations nilfs_aops = {
 264         .writepage              = nilfs_writepage,
 265         .readpage               = nilfs_readpage,
 266         .sync_page              = block_sync_page,
 267         .writepages             = nilfs_writepages,
 268         .set_page_dirty         = nilfs_set_page_dirty,
 269         .readpages              = nilfs_readpages,
 270         .write_begin            = nilfs_write_begin,
 271         .write_end              = nilfs_write_end,
 272         /* .releasepage         = nilfs_releasepage, */
 273         .invalidatepage         = block_invalidatepage,
 274         .direct_IO              = nilfs_direct_IO,
 275         .is_partially_uptodate  = block_is_partially_uptodate,
 276 };
 277
 278 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 279 {
 280         struct super_block *sb = dir->i_sb;
 281         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 282         struct inode *inode;
 283         struct nilfs_inode_info *ii;
 284         struct nilfs_root *root;
 285         int err = -ENOMEM;
 286         ino_t ino;
 287
 288         inode = new_inode(sb);
 289         if (unlikely(!inode))
 290                 goto failed;
 291
 292         mapping_set_gfp_mask(inode->i_mapping,
 293                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 294
 295         root = NILFS_I(dir)->i_root;
 296         ii = NILFS_I(inode);
 297         ii->i_state = 1 << NILFS_I_NEW;
 298         ii->i_root = root;
 299
 300         err = nilfs_ifile_create_inode(root->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(&root->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_set_inode_flags(struct inode *inode)
 358 {
 359         unsigned int flags = NILFS_I(inode)->i_flags;
 360
 361         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 362                             S_DIRSYNC);
 363         if (flags & NILFS_SYNC_FL)
 364                 inode->i_flags |= S_SYNC;
 365         if (flags & NILFS_APPEND_FL)
 366                 inode->i_flags |= S_APPEND;
 367         if (flags & NILFS_IMMUTABLE_FL)
 368                 inode->i_flags |= S_IMMUTABLE;
 369 #ifndef NILFS_ATIME_DISABLE
 370         if (flags & NILFS_NOATIME_FL)
 371 #endif
 372                 inode->i_flags |= S_NOATIME;
 373         if (flags & NILFS_DIRSYNC_FL)
 374                 inode->i_flags |= S_DIRSYNC;
 375         mapping_set_gfp_mask(inode->i_mapping,
 376                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 377 }
 378
 379 int nilfs_read_inode_common(struct inode *inode,
 380                             struct nilfs_inode *raw_inode)
 381 {
 382         struct nilfs_inode_info *ii = NILFS_I(inode);
 383         int err;
 384
 385         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 386         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 387         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 388         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 389         inode->i_size = le64_to_cpu(raw_inode->i_size);
 390         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 391         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 392         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 393         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 394         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 395         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 396         if (inode->i_nlink == 0 && inode->i_mode == 0)
 397                 return -EINVAL; /* this inode is deleted */
 398
 399         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 400         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 401 #if 0
 402         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 403         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 404                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 405 #endif
 406         ii->i_dir_start_lookup = 0;
 407         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 408
 409         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 410             S_ISLNK(inode->i_mode)) {
 411                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 412                 if (err < 0)
 413                         return err;
 414                 set_bit(NILFS_I_BMAP, &ii->i_state);
 415                 /* No lock is needed; iget() ensures it. */
 416         }
 417         return 0;
 418 }
 419
 420 static int __nilfs_read_inode(struct super_block *sb,
 421                               struct nilfs_root *root, unsigned long ino,
 422                               struct inode *inode)
 423 {
 424         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 425         struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
 426         struct buffer_head *bh;
 427         struct nilfs_inode *raw_inode;
 428         int err;
 429
 430         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
 431         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 432         if (unlikely(err))
 433                 goto bad_inode;
 434
 435         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 436
 437         err = nilfs_read_inode_common(inode, raw_inode);
 438         if (err)
 439                 goto failed_unmap;
 440
 441         if (S_ISREG(inode->i_mode)) {
 442                 inode->i_op = &nilfs_file_inode_operations;
 443                 inode->i_fop = &nilfs_file_operations;
 444                 inode->i_mapping->a_ops = &nilfs_aops;
 445         } else if (S_ISDIR(inode->i_mode)) {
 446                 inode->i_op = &nilfs_dir_inode_operations;
 447                 inode->i_fop = &nilfs_dir_operations;
 448                 inode->i_mapping->a_ops = &nilfs_aops;
 449         } else if (S_ISLNK(inode->i_mode)) {
 450                 inode->i_op = &nilfs_symlink_inode_operations;
 451                 inode->i_mapping->a_ops = &nilfs_aops;
 452         } else {
 453                 inode->i_op = &nilfs_special_inode_operations;
 454                 init_special_inode(
 455                         inode, inode->i_mode,
 456                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 457         }
 458         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 459         brelse(bh);
 460         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 461         nilfs_set_inode_flags(inode);
 462         return 0;
 463
 464  failed_unmap:
 465         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 466         brelse(bh);
 467
 468  bad_inode:
 469         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
 470         return err;
 471 }
 472
 473 static int nilfs_iget_test(struct inode *inode, void *opaque)
 474 {
 475         struct nilfs_iget_args *args = opaque;
 476         struct nilfs_inode_info *ii;
 477
 478         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 479                 return 0;
 480
 481         ii = NILFS_I(inode);
 482         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 483                 return !args->for_gc;
 484
 485         return args->for_gc && args->cno == ii->i_cno;
 486 }
 487
 488 static int nilfs_iget_set(struct inode *inode, void *opaque)
 489 {
 490         struct nilfs_iget_args *args = opaque;
 491
 492         inode->i_ino = args->ino;
 493         if (args->for_gc) {
 494                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 495                 NILFS_I(inode)->i_cno = args->cno;
 496                 NILFS_I(inode)->i_root = NULL;
 497         } else {
 498                 if (args->root && args->ino == NILFS_ROOT_INO)
 499                         nilfs_get_root(args->root);
 500                 NILFS_I(inode)->i_root = args->root;
 501         }
 502         return 0;
 503 }
 504
 505 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 506                             unsigned long ino)
 507 {
 508         struct nilfs_iget_args args = {
 509                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 510         };
 511
 512         return ilookup5(sb, ino, nilfs_iget_test, &args);
 513 }
 514
 515 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 516                                 unsigned long ino)
 517 {
 518         struct nilfs_iget_args args = {
 519                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 520         };
 521
 522         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 523 }
 524
 525 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 526                          unsigned long ino)
 527 {
 528         struct inode *inode;
 529         int err;
 530
 531         inode = nilfs_iget_locked(sb, root, ino);
 532         if (unlikely(!inode))
 533                 return ERR_PTR(-ENOMEM);
 534         if (!(inode->i_state & I_NEW))
 535                 return inode;
 536
 537         err = __nilfs_read_inode(sb, root, ino, inode);
 538         if (unlikely(err)) {
 539                 iget_failed(inode);
 540                 return ERR_PTR(err);
 541         }
 542         unlock_new_inode(inode);
 543         return inode;
 544 }
 545
 546 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 547                                 __u64 cno)
 548 {
 549         struct nilfs_iget_args args = {
 550                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 551         };
 552         struct inode *inode;
 553         int err;
 554
 555         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 556         if (unlikely(!inode))
 557                 return ERR_PTR(-ENOMEM);
 558         if (!(inode->i_state & I_NEW))
 559                 return inode;
 560
 561         err = nilfs_init_gcinode(inode);
 562         if (unlikely(err)) {
 563                 iget_failed(inode);
 564                 return ERR_PTR(err);
 565         }
 566         unlock_new_inode(inode);
 567         return inode;
 568 }
 569
 570 void nilfs_write_inode_common(struct inode *inode,
 571                               struct nilfs_inode *raw_inode, int has_bmap)
 572 {
 573         struct nilfs_inode_info *ii = NILFS_I(inode);
 574
 575         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 576         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 577         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 578         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 579         raw_inode->i_size = cpu_to_le64(inode->i_size);
 580         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 581         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 582         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 583         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 584         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 585
 586         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 587         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 588
 589         if (has_bmap)
 590                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 591         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 592                 raw_inode->i_device_code =
 593                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 594         /* When extending inode, nilfs->ns_inode_size should be checked
 595            for substitutions of appended fields */
 596 }
 597
 598 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 599 {
 600         ino_t ino = inode->i_ino;
 601         struct nilfs_inode_info *ii = NILFS_I(inode);
 602         struct inode *ifile = ii->i_root->ifile;
 603         struct nilfs_inode *raw_inode;
 604
 605         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 606
 607         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 608                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 609         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 610
 611         nilfs_write_inode_common(inode, raw_inode, 0);
 612                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 613                    deadlock of bmap. This delays update of i_bmap to just
 614                    before writing */
 615         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 616 }
 617
 618 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 619
 620 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 621                                 unsigned long from)
 622 {
 623         unsigned long b;
 624         int ret;
 625
 626         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 627                 return;
 628 repeat:
 629         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 630         if (ret == -ENOENT)
 631                 return;
 632         else if (ret < 0)
 633                 goto failed;
 634
 635         if (b < from)
 636                 return;
 637
 638         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 639         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 640         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 641         if (!ret || (ret == -ENOMEM &&
 642                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 643                 goto repeat;
 644
 645 failed:
 646         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 647                       "failed to truncate bmap (ino=%lu, err=%d)",
 648                       ii->vfs_inode.i_ino, ret);
 649 }
 650
 651 void nilfs_truncate(struct inode *inode)
 652 {
 653         unsigned long blkoff;
 654         unsigned int blocksize;
 655         struct nilfs_transaction_info ti;
 656         struct super_block *sb = inode->i_sb;
 657         struct nilfs_inode_info *ii = NILFS_I(inode);
 658
 659         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 660                 return;
 661         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 662                 return;
 663
 664         blocksize = sb->s_blocksize;
 665         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 666         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 667
 668         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 669
 670         nilfs_truncate_bmap(ii, blkoff);
 671
 672         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 673         if (IS_SYNC(inode))
 674                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 675
 676         nilfs_mark_inode_dirty(inode);
 677         nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
 678         nilfs_transaction_commit(sb);
 679         /* May construct a logical segment and may fail in sync mode.
 680            But truncate has no return value. */
 681 }
 682
 683 static void nilfs_clear_inode(struct inode *inode)
 684 {
 685         struct nilfs_inode_info *ii = NILFS_I(inode);
 686         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 687
 688         /*
 689          * Free resources allocated in nilfs_read_inode(), here.
 690          */
 691         BUG_ON(!list_empty(&ii->i_dirty));
 692         brelse(ii->i_bh);
 693         ii->i_bh = NULL;
 694
 695         if (mdi && mdi->mi_palloc_cache)
 696                 nilfs_palloc_destroy_cache(inode);
 697
 698         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 699                 nilfs_bmap_clear(ii->i_bmap);
 700
 701         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 702
 703         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 704                 nilfs_put_root(ii->i_root);
 705 }
 706
 707 void nilfs_evict_inode(struct inode *inode)
 708 {
 709         struct nilfs_transaction_info ti;
 710         struct super_block *sb = inode->i_sb;
 711         struct nilfs_inode_info *ii = NILFS_I(inode);
 712
 713         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 714                 if (inode->i_data.nrpages)
 715                         truncate_inode_pages(&inode->i_data, 0);
 716                 end_writeback(inode);
 717                 nilfs_clear_inode(inode);
 718                 return;
 719         }
 720         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 721
 722         if (inode->i_data.nrpages)
 723                 truncate_inode_pages(&inode->i_data, 0);
 724
 725         /* TODO: some of the following operations may fail.  */
 726         nilfs_truncate_bmap(ii, 0);
 727         nilfs_mark_inode_dirty(inode);
 728         end_writeback(inode);
 729
 730         nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 731         atomic_dec(&ii->i_root->inodes_count);
 732
 733         nilfs_clear_inode(inode);
 734
 735         if (IS_SYNC(inode))
 736                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 737         nilfs_transaction_commit(sb);
 738         /* May construct a logical segment and may fail in sync mode.
 739            But delete_inode has no return value. */
 740 }
 741
 742 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 743 {
 744         struct nilfs_transaction_info ti;
 745         struct inode *inode = dentry->d_inode;
 746         struct super_block *sb = inode->i_sb;
 747         int err;
 748
 749         err = inode_change_ok(inode, iattr);
 750         if (err)
 751                 return err;
 752
 753         err = nilfs_transaction_begin(sb, &ti, 0);
 754         if (unlikely(err))
 755                 return err;
 756
 757         if ((iattr->ia_valid & ATTR_SIZE) &&
 758             iattr->ia_size != i_size_read(inode)) {
 759                 err = vmtruncate(inode, iattr->ia_size);
 760                 if (unlikely(err))
 761                         goto out_err;
 762         }
 763
 764         setattr_copy(inode, iattr);
 765         mark_inode_dirty(inode);
 766
 767         if (iattr->ia_valid & ATTR_MODE) {
 768                 err = nilfs_acl_chmod(inode);
 769                 if (unlikely(err))
 770                         goto out_err;
 771         }
 772
 773         return nilfs_transaction_commit(sb);
 774
 775 out_err:
 776         nilfs_transaction_abort(sb);
 777         return err;
 778 }
 779
 780 int nilfs_permission(struct inode *inode, int mask, unsigned int flags)
 781 {
 782         struct nilfs_root *root;
 783
 784         if (flags & IPERM_FLAG_RCU)
 785                 return -ECHILD;
 786
 787         root = NILFS_I(inode)->i_root;
 788         if ((mask & MAY_WRITE) && root &&
 789             root->cno != NILFS_CPTREE_CURRENT_CNO)
 790                 return -EROFS; /* snapshot is not writable */
 791
 792         return generic_permission(inode, mask, flags, NULL);
 793 }
 794
 795 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
 796                            struct buffer_head **pbh)
 797 {
 798         struct nilfs_inode_info *ii = NILFS_I(inode);
 799         int err;
 800
 801         spin_lock(&sbi->s_inode_lock);
 802         if (ii->i_bh == NULL) {
 803                 spin_unlock(&sbi->s_inode_lock);
 804                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 805                                                   inode->i_ino, pbh);
 806                 if (unlikely(err))
 807                         return err;
 808                 spin_lock(&sbi->s_inode_lock);
 809                 if (ii->i_bh == NULL)
 810                         ii->i_bh = *pbh;
 811                 else {
 812                         brelse(*pbh);
 813                         *pbh = ii->i_bh;
 814                 }
 815         } else
 816                 *pbh = ii->i_bh;
 817
 818         get_bh(*pbh);
 819         spin_unlock(&sbi->s_inode_lock);
 820         return 0;
 821 }
 822
 823 int nilfs_inode_dirty(struct inode *inode)
 824 {
 825         struct nilfs_inode_info *ii = NILFS_I(inode);
 826         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 827         int ret = 0;
 828
 829         if (!list_empty(&ii->i_dirty)) {
 830                 spin_lock(&sbi->s_inode_lock);
 831                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 832                         test_bit(NILFS_I_BUSY, &ii->i_state);
 833                 spin_unlock(&sbi->s_inode_lock);
 834         }
 835         return ret;
 836 }
 837
 838 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
 839                          unsigned nr_dirty)
 840 {
 841         struct nilfs_inode_info *ii = NILFS_I(inode);
 842
 843         atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
 844
 845         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 846                 return 0;
 847
 848         spin_lock(&sbi->s_inode_lock);
 849         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 850             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 851                 /* Because this routine may race with nilfs_dispose_list(),
 852                    we have to check NILFS_I_QUEUED here, too. */
 853                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 854                         /* This will happen when somebody is freeing
 855                            this inode. */
 856                         nilfs_warning(sbi->s_super, __func__,
 857                                       "cannot get inode (ino=%lu)\n",
 858                                       inode->i_ino);
 859                         spin_unlock(&sbi->s_inode_lock);
 860                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 861                                            freeing inode */
 862                 }
 863                 list_del(&ii->i_dirty);
 864                 list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
 865                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 866         }
 867         spin_unlock(&sbi->s_inode_lock);
 868         return 0;
 869 }
 870
 871 int nilfs_mark_inode_dirty(struct inode *inode)
 872 {
 873         struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
 874         struct buffer_head *ibh;
 875         int err;
 876
 877         err = nilfs_load_inode_block(sbi, inode, &ibh);
 878         if (unlikely(err)) {
 879                 nilfs_warning(inode->i_sb, __func__,
 880                               "failed to reget inode block.\n");
 881                 return err;
 882         }
 883         nilfs_update_inode(inode, ibh);
 884         nilfs_mdt_mark_buffer_dirty(ibh);
 885         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 886         brelse(ibh);
 887         return 0;
 888 }
 889
 890 /**
 891  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 892  * @inode: inode of the file to be registered.
 893  *
 894  * nilfs_dirty_inode() loads a inode block containing the specified
 895  * @inode and copies data from a nilfs_inode to a corresponding inode
 896  * entry in the inode block. This operation is excluded from the segment
 897  * construction. This function can be called both as a single operation
 898  * and as a part of indivisible file operations.
 899  */
 900 void nilfs_dirty_inode(struct inode *inode)
 901 {
 902         struct nilfs_transaction_info ti;
 903         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 904
 905         if (is_bad_inode(inode)) {
 906                 nilfs_warning(inode->i_sb, __func__,
 907                               "tried to mark bad_inode dirty. ignored.\n");
 908                 dump_stack();
 909                 return;
 910         }
 911         if (mdi) {
 912                 nilfs_mdt_mark_dirty(inode);
 913                 return;
 914         }
 915         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 916         nilfs_mark_inode_dirty(inode);
 917         nilfs_transaction_commit(inode->i_sb); /* never fails */
 918 }
 919
 920 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 921                  __u64 start, __u64 len)
 922 {
 923         struct the_nilfs *nilfs = NILFS_I_NILFS(inode);
 924         __u64 logical = 0, phys = 0, size = 0;
 925         __u32 flags = 0;
 926         loff_t isize;
 927         sector_t blkoff, end_blkoff;
 928         sector_t delalloc_blkoff;
 929         unsigned long delalloc_blklen;
 930         unsigned int blkbits = inode->i_blkbits;
 931         int ret, n;
 932
 933         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 934         if (ret)
 935                 return ret;
 936
 937         mutex_lock(&inode->i_mutex);
 938
 939         isize = i_size_read(inode);
 940
 941         blkoff = start >> blkbits;
 942         end_blkoff = (start + len - 1) >> blkbits;
 943
 944         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
 945                                                         &delalloc_blkoff);
 946
 947         do {
 948                 __u64 blkphy;
 949                 unsigned int maxblocks;
 950
 951                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
 952                         if (size) {
 953                                 /* End of the current extent */
 954                                 ret = fiemap_fill_next_extent(
 955                                         fieinfo, logical, phys, size, flags);
 956                                 if (ret)
 957                                         break;
 958                         }
 959                         if (blkoff > end_blkoff)
 960                                 break;
 961
 962                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
 963                         logical = blkoff << blkbits;
 964                         phys = 0;
 965                         size = delalloc_blklen << blkbits;
 966
 967                         blkoff = delalloc_blkoff + delalloc_blklen;
 968                         delalloc_blklen = nilfs_find_uncommitted_extent(
 969                                 inode, blkoff, &delalloc_blkoff);
 970                         continue;
 971                 }
 972
 973                 /*
 974                  * Limit the number of blocks that we look up so as
 975                  * not to get into the next delayed allocation extent.
 976                  */
 977                 maxblocks = INT_MAX;
 978                 if (delalloc_blklen)
 979                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
 980                                           maxblocks);
 981                 blkphy = 0;
 982
 983                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 984                 n = nilfs_bmap_lookup_contig(
 985                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
 986                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 987
 988                 if (n < 0) {
 989                         int past_eof;
 990
 991                         if (unlikely(n != -ENOENT))
 992                                 break; /* error */
 993
 994                         /* HOLE */
 995                         blkoff++;
 996                         past_eof = ((blkoff << blkbits) >= isize);
 997
 998                         if (size) {
 999                                 /* End of the current extent */
1000
1001                                 if (past_eof)
1002                                         flags |= FIEMAP_EXTENT_LAST;
1003
1004                                 ret = fiemap_fill_next_extent(
1005                                         fieinfo, logical, phys, size, flags);
1006                                 if (ret)
1007                                         break;
1008                                 size = 0;
1009                         }
1010                         if (blkoff > end_blkoff || past_eof)
1011                                 break;
1012                 } else {
1013                         if (size) {
1014                                 if (phys && blkphy << blkbits == phys + size) {
1015                                         /* The current extent goes on */
1016                                         size += n << blkbits;
1017                                 } else {
1018                                         /* Terminate the current extent */
1019                                         ret = fiemap_fill_next_extent(
1020                                                 fieinfo, logical, phys, size,
1021                                                 flags);
1022                                         if (ret || blkoff > end_blkoff)
1023                                                 break;
1024
1025                                         /* Start another extent */
1026                                         flags = FIEMAP_EXTENT_MERGED;
1027                                         logical = blkoff << blkbits;
1028                                         phys = blkphy << blkbits;
1029                                         size = n << blkbits;
1030                                 }
1031                         } else {
1032                                 /* Start a new extent */
1033                                 flags = FIEMAP_EXTENT_MERGED;
1034                                 logical = blkoff << blkbits;
1035                                 phys = blkphy << blkbits;
1036                                 size = n << blkbits;
1037                         }
1038                         blkoff += n;
1039                 }
1040                 cond_resched();
1041         } while (true);
1042
1043         /* If ret is 1 then we just hit the end of the extent array */
1044         if (ret == 1)
1045                 ret = 0;
1046
1047         mutex_unlock(&inode->i_mutex);
1048         return ret;
1049 }