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 void nilfs_inode_add_blocks(struct inode *inode, int n)
  45 {
  46         struct nilfs_root *root = NILFS_I(inode)->i_root;
  47
  48         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  49         if (root)
  50                 atomic_add(n, &root->blocks_count);
  51 }
  52
  53 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  54 {
  55         struct nilfs_root *root = NILFS_I(inode)->i_root;
  56
  57         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  58         if (root)
  59                 atomic_sub(n, &root->blocks_count);
  60 }
  61
  62 /**
  63  * nilfs_get_block() - get a file block on the filesystem (callback function)
  64  * @inode - inode struct of the target file
  65  * @blkoff - file block number
  66  * @bh_result - buffer head to be mapped on
  67  * @create - indicate whether allocating the block or not when it has not
  68  *      been allocated yet.
  69  *
  70  * This function does not issue actual read request of the specified data
  71  * block. It is done by VFS.
  72  */
  73 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  74                     struct buffer_head *bh_result, int create)
  75 {
  76         struct nilfs_inode_info *ii = NILFS_I(inode);
  77         __u64 blknum = 0;
  78         int err = 0, ret;
  79         struct inode *dat = NILFS_I_NILFS(inode)->ns_dat;
  80         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  81
  82         down_read(&NILFS_MDT(dat)->mi_sem);
  83         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  84         up_read(&NILFS_MDT(dat)->mi_sem);
  85         if (ret >= 0) { /* found */
  86                 map_bh(bh_result, inode->i_sb, blknum);
  87                 if (ret > 0)
  88                         bh_result->b_size = (ret << inode->i_blkbits);
  89                 goto out;
  90         }
  91         /* data block was not found */
  92         if (ret == -ENOENT && create) {
  93                 struct nilfs_transaction_info ti;
  94
  95                 bh_result->b_blocknr = 0;
  96                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  97                 if (unlikely(err))
  98                         goto out;
  99                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 100                                         (unsigned long)bh_result);
 101                 if (unlikely(err != 0)) {
 102                         if (err == -EEXIST) {
 103                                 /*
 104                                  * The get_block() function could be called
 105                                  * from multiple callers for an inode.
 106                                  * However, the page having this block must
 107                                  * be locked in this case.
 108                                  */
 109                                 printk(KERN_WARNING
 110                                        "nilfs_get_block: a race condition "
 111                                        "while inserting a data block. "
 112                                        "(inode number=%lu, file block "
 113                                        "offset=%llu)\n",
 114                                        inode->i_ino,
 115                                        (unsigned long long)blkoff);
 116                                 err = 0;
 117                         }
 118                         nilfs_transaction_abort(inode->i_sb);
 119                         goto out;
 120                 }
 121                 nilfs_mark_inode_dirty(inode);
 122                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 123                 /* Error handling should be detailed */
 124                 set_buffer_new(bh_result);
 125                 set_buffer_delay(bh_result);
 126                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 127                                                       to proper value */
 128         } else if (ret == -ENOENT) {
 129                 /* not found is not error (e.g. hole); must return without
 130                    the mapped state flag. */
 131                 ;
 132         } else {
 133                 err = ret;
 134         }
 135
 136  out:
 137         return err;
 138 }
 139
 140 /**
 141  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 142  * address_space_operations.
 143  * @file - file struct of the file to be read
 144  * @page - the page to be read
 145  */
 146 static int nilfs_readpage(struct file *file, struct page *page)
 147 {
 148         return mpage_readpage(page, nilfs_get_block);
 149 }
 150
 151 /**
 152  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 153  * address_space_operations.
 154  * @file - file struct of the file to be read
 155  * @mapping - address_space struct used for reading multiple pages
 156  * @pages - the pages to be read
 157  * @nr_pages - number of pages to be read
 158  */
 159 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 160                            struct list_head *pages, unsigned nr_pages)
 161 {
 162         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 163 }
 164
 165 static int nilfs_writepages(struct address_space *mapping,
 166                             struct writeback_control *wbc)
 167 {
 168         struct inode *inode = mapping->host;
 169         int err = 0;
 170
 171         if (wbc->sync_mode == WB_SYNC_ALL)
 172                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 173                                                     wbc->range_start,
 174                                                     wbc->range_end);
 175         return err;
 176 }
 177
 178 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 179 {
 180         struct inode *inode = page->mapping->host;
 181         int err;
 182
 183         redirty_page_for_writepage(wbc, page);
 184         unlock_page(page);
 185
 186         if (wbc->sync_mode == WB_SYNC_ALL) {
 187                 err = nilfs_construct_segment(inode->i_sb);
 188                 if (unlikely(err))
 189                         return err;
 190         } else if (wbc->for_reclaim)
 191                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 192
 193         return 0;
 194 }
 195
 196 static int nilfs_set_page_dirty(struct page *page)
 197 {
 198         int ret = __set_page_dirty_buffers(page);
 199
 200         if (ret) {
 201                 struct inode *inode = page->mapping->host;
 202                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 203
 204                 nilfs_set_file_dirty(inode, nr_dirty);
 205         }
 206         return ret;
 207 }
 208
 209 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 210                              loff_t pos, unsigned len, unsigned flags,
 211                              struct page **pagep, void **fsdata)
 212
 213 {
 214         struct inode *inode = mapping->host;
 215         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 216
 217         if (unlikely(err))
 218                 return err;
 219
 220         err = block_write_begin(mapping, pos, len, flags, pagep,
 221                                 nilfs_get_block);
 222         if (unlikely(err)) {
 223                 loff_t isize = mapping->host->i_size;
 224                 if (pos + len > isize)
 225                         vmtruncate(mapping->host, isize);
 226
 227                 nilfs_transaction_abort(inode->i_sb);
 228         }
 229         return err;
 230 }
 231
 232 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 233                            loff_t pos, unsigned len, unsigned copied,
 234                            struct page *page, void *fsdata)
 235 {
 236         struct inode *inode = mapping->host;
 237         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 238         unsigned nr_dirty;
 239         int err;
 240
 241         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 242                                                   start + copied);
 243         copied = generic_write_end(file, mapping, pos, len, copied, page,
 244                                    fsdata);
 245         nilfs_set_file_dirty(inode, nr_dirty);
 246         err = nilfs_transaction_commit(inode->i_sb);
 247         return err ? : copied;
 248 }
 249
 250 static ssize_t
 251 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 252                 loff_t offset, unsigned long nr_segs)
 253 {
 254         struct file *file = iocb->ki_filp;
 255         struct inode *inode = file->f_mapping->host;
 256         ssize_t size;
 257
 258         if (rw == WRITE)
 259                 return 0;
 260
 261         /* Needs synchronization with the cleaner */
 262         size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
 263                                   offset, nr_segs, nilfs_get_block, NULL);
 264
 265         /*
 266          * In case of error extending write may have instantiated a few
 267          * blocks outside i_size. Trim these off again.
 268          */
 269         if (unlikely((rw & WRITE) && size < 0)) {
 270                 loff_t isize = i_size_read(inode);
 271                 loff_t end = offset + iov_length(iov, nr_segs);
 272
 273                 if (end > isize)
 274                         vmtruncate(inode, isize);
 275         }
 276
 277         return size;
 278 }
 279
 280 const struct address_space_operations nilfs_aops = {
 281         .writepage              = nilfs_writepage,
 282         .readpage               = nilfs_readpage,
 283         .sync_page              = block_sync_page,
 284         .writepages             = nilfs_writepages,
 285         .set_page_dirty         = nilfs_set_page_dirty,
 286         .readpages              = nilfs_readpages,
 287         .write_begin            = nilfs_write_begin,
 288         .write_end              = nilfs_write_end,
 289         /* .releasepage         = nilfs_releasepage, */
 290         .invalidatepage         = block_invalidatepage,
 291         .direct_IO              = nilfs_direct_IO,
 292         .is_partially_uptodate  = block_is_partially_uptodate,
 293 };
 294
 295 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 296 {
 297         struct super_block *sb = dir->i_sb;
 298         struct the_nilfs *nilfs = sb->s_fs_info;
 299         struct inode *inode;
 300         struct nilfs_inode_info *ii;
 301         struct nilfs_root *root;
 302         int err = -ENOMEM;
 303         ino_t ino;
 304
 305         inode = new_inode(sb);
 306         if (unlikely(!inode))
 307                 goto failed;
 308
 309         mapping_set_gfp_mask(inode->i_mapping,
 310                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 311
 312         root = NILFS_I(dir)->i_root;
 313         ii = NILFS_I(inode);
 314         ii->i_state = 1 << NILFS_I_NEW;
 315         ii->i_root = root;
 316
 317         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 318         if (unlikely(err))
 319                 goto failed_ifile_create_inode;
 320         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 321
 322         atomic_inc(&root->inodes_count);
 323         inode_init_owner(inode, dir, mode);
 324         inode->i_ino = ino;
 325         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 326
 327         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 328                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 329                 if (err < 0)
 330                         goto failed_bmap;
 331
 332                 set_bit(NILFS_I_BMAP, &ii->i_state);
 333                 /* No lock is needed; iget() ensures it. */
 334         }
 335
 336         ii->i_flags = nilfs_mask_flags(
 337                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 338
 339         /* ii->i_file_acl = 0; */
 340         /* ii->i_dir_acl = 0; */
 341         ii->i_dir_start_lookup = 0;
 342         nilfs_set_inode_flags(inode);
 343         spin_lock(&nilfs->ns_next_gen_lock);
 344         inode->i_generation = nilfs->ns_next_generation++;
 345         spin_unlock(&nilfs->ns_next_gen_lock);
 346         insert_inode_hash(inode);
 347
 348         err = nilfs_init_acl(inode, dir);
 349         if (unlikely(err))
 350                 goto failed_acl; /* never occur. When supporting
 351                                     nilfs_init_acl(), proper cancellation of
 352                                     above jobs should be considered */
 353
 354         return inode;
 355
 356  failed_acl:
 357  failed_bmap:
 358         inode->i_nlink = 0;
 359         iput(inode);  /* raw_inode will be deleted through
 360                          generic_delete_inode() */
 361         goto failed;
 362
 363  failed_ifile_create_inode:
 364         make_bad_inode(inode);
 365         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 366                          called */
 367  failed:
 368         return ERR_PTR(err);
 369 }
 370
 371 void nilfs_set_inode_flags(struct inode *inode)
 372 {
 373         unsigned int flags = NILFS_I(inode)->i_flags;
 374
 375         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 376                             S_DIRSYNC);
 377         if (flags & FS_SYNC_FL)
 378                 inode->i_flags |= S_SYNC;
 379         if (flags & FS_APPEND_FL)
 380                 inode->i_flags |= S_APPEND;
 381         if (flags & FS_IMMUTABLE_FL)
 382                 inode->i_flags |= S_IMMUTABLE;
 383         if (flags & FS_NOATIME_FL)
 384                 inode->i_flags |= S_NOATIME;
 385         if (flags & FS_DIRSYNC_FL)
 386                 inode->i_flags |= S_DIRSYNC;
 387         mapping_set_gfp_mask(inode->i_mapping,
 388                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 389 }
 390
 391 int nilfs_read_inode_common(struct inode *inode,
 392                             struct nilfs_inode *raw_inode)
 393 {
 394         struct nilfs_inode_info *ii = NILFS_I(inode);
 395         int err;
 396
 397         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 398         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 399         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 400         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 401         inode->i_size = le64_to_cpu(raw_inode->i_size);
 402         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 403         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 404         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 405         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 406         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 407         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 408         if (inode->i_nlink == 0 && inode->i_mode == 0)
 409                 return -EINVAL; /* this inode is deleted */
 410
 411         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 412         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 413 #if 0
 414         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 415         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 416                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 417 #endif
 418         ii->i_dir_start_lookup = 0;
 419         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 420
 421         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 422             S_ISLNK(inode->i_mode)) {
 423                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 424                 if (err < 0)
 425                         return err;
 426                 set_bit(NILFS_I_BMAP, &ii->i_state);
 427                 /* No lock is needed; iget() ensures it. */
 428         }
 429         return 0;
 430 }
 431
 432 static int __nilfs_read_inode(struct super_block *sb,
 433                               struct nilfs_root *root, unsigned long ino,
 434                               struct inode *inode)
 435 {
 436         struct the_nilfs *nilfs = sb->s_fs_info;
 437         struct buffer_head *bh;
 438         struct nilfs_inode *raw_inode;
 439         int err;
 440
 441         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 442         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 443         if (unlikely(err))
 444                 goto bad_inode;
 445
 446         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 447
 448         err = nilfs_read_inode_common(inode, raw_inode);
 449         if (err)
 450                 goto failed_unmap;
 451
 452         if (S_ISREG(inode->i_mode)) {
 453                 inode->i_op = &nilfs_file_inode_operations;
 454                 inode->i_fop = &nilfs_file_operations;
 455                 inode->i_mapping->a_ops = &nilfs_aops;
 456         } else if (S_ISDIR(inode->i_mode)) {
 457                 inode->i_op = &nilfs_dir_inode_operations;
 458                 inode->i_fop = &nilfs_dir_operations;
 459                 inode->i_mapping->a_ops = &nilfs_aops;
 460         } else if (S_ISLNK(inode->i_mode)) {
 461                 inode->i_op = &nilfs_symlink_inode_operations;
 462                 inode->i_mapping->a_ops = &nilfs_aops;
 463         } else {
 464                 inode->i_op = &nilfs_special_inode_operations;
 465                 init_special_inode(
 466                         inode, inode->i_mode,
 467                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 468         }
 469         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 470         brelse(bh);
 471         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 472         nilfs_set_inode_flags(inode);
 473         return 0;
 474
 475  failed_unmap:
 476         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 477         brelse(bh);
 478
 479  bad_inode:
 480         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 481         return err;
 482 }
 483
 484 static int nilfs_iget_test(struct inode *inode, void *opaque)
 485 {
 486         struct nilfs_iget_args *args = opaque;
 487         struct nilfs_inode_info *ii;
 488
 489         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 490                 return 0;
 491
 492         ii = NILFS_I(inode);
 493         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 494                 return !args->for_gc;
 495
 496         return args->for_gc && args->cno == ii->i_cno;
 497 }
 498
 499 static int nilfs_iget_set(struct inode *inode, void *opaque)
 500 {
 501         struct nilfs_iget_args *args = opaque;
 502
 503         inode->i_ino = args->ino;
 504         if (args->for_gc) {
 505                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 506                 NILFS_I(inode)->i_cno = args->cno;
 507                 NILFS_I(inode)->i_root = NULL;
 508         } else {
 509                 if (args->root && args->ino == NILFS_ROOT_INO)
 510                         nilfs_get_root(args->root);
 511                 NILFS_I(inode)->i_root = args->root;
 512         }
 513         return 0;
 514 }
 515
 516 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 517                             unsigned long ino)
 518 {
 519         struct nilfs_iget_args args = {
 520                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 521         };
 522
 523         return ilookup5(sb, ino, nilfs_iget_test, &args);
 524 }
 525
 526 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 527                                 unsigned long ino)
 528 {
 529         struct nilfs_iget_args args = {
 530                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 531         };
 532
 533         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 534 }
 535
 536 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 537                          unsigned long ino)
 538 {
 539         struct inode *inode;
 540         int err;
 541
 542         inode = nilfs_iget_locked(sb, root, ino);
 543         if (unlikely(!inode))
 544                 return ERR_PTR(-ENOMEM);
 545         if (!(inode->i_state & I_NEW))
 546                 return inode;
 547
 548         err = __nilfs_read_inode(sb, root, ino, inode);
 549         if (unlikely(err)) {
 550                 iget_failed(inode);
 551                 return ERR_PTR(err);
 552         }
 553         unlock_new_inode(inode);
 554         return inode;
 555 }
 556
 557 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 558                                 __u64 cno)
 559 {
 560         struct nilfs_iget_args args = {
 561                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 562         };
 563         struct inode *inode;
 564         int err;
 565
 566         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 567         if (unlikely(!inode))
 568                 return ERR_PTR(-ENOMEM);
 569         if (!(inode->i_state & I_NEW))
 570                 return inode;
 571
 572         err = nilfs_init_gcinode(inode);
 573         if (unlikely(err)) {
 574                 iget_failed(inode);
 575                 return ERR_PTR(err);
 576         }
 577         unlock_new_inode(inode);
 578         return inode;
 579 }
 580
 581 void nilfs_write_inode_common(struct inode *inode,
 582                               struct nilfs_inode *raw_inode, int has_bmap)
 583 {
 584         struct nilfs_inode_info *ii = NILFS_I(inode);
 585
 586         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 587         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 588         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 589         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 590         raw_inode->i_size = cpu_to_le64(inode->i_size);
 591         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 592         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 593         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 594         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 595         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 596
 597         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 598         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 599
 600         if (has_bmap)
 601                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 602         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 603                 raw_inode->i_device_code =
 604                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 605         /* When extending inode, nilfs->ns_inode_size should be checked
 606            for substitutions of appended fields */
 607 }
 608
 609 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 610 {
 611         ino_t ino = inode->i_ino;
 612         struct nilfs_inode_info *ii = NILFS_I(inode);
 613         struct inode *ifile = ii->i_root->ifile;
 614         struct nilfs_inode *raw_inode;
 615
 616         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 617
 618         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 619                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 620         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 621
 622         nilfs_write_inode_common(inode, raw_inode, 0);
 623                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 624                    deadlock of bmap. This delays update of i_bmap to just
 625                    before writing */
 626         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 627 }
 628
 629 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 630
 631 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 632                                 unsigned long from)
 633 {
 634         unsigned long b;
 635         int ret;
 636
 637         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 638                 return;
 639 repeat:
 640         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 641         if (ret == -ENOENT)
 642                 return;
 643         else if (ret < 0)
 644                 goto failed;
 645
 646         if (b < from)
 647                 return;
 648
 649         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 650         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 651         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 652         if (!ret || (ret == -ENOMEM &&
 653                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 654                 goto repeat;
 655
 656 failed:
 657         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 658                       "failed to truncate bmap (ino=%lu, err=%d)",
 659                       ii->vfs_inode.i_ino, ret);
 660 }
 661
 662 void nilfs_truncate(struct inode *inode)
 663 {
 664         unsigned long blkoff;
 665         unsigned int blocksize;
 666         struct nilfs_transaction_info ti;
 667         struct super_block *sb = inode->i_sb;
 668         struct nilfs_inode_info *ii = NILFS_I(inode);
 669
 670         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 671                 return;
 672         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 673                 return;
 674
 675         blocksize = sb->s_blocksize;
 676         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 677         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 678
 679         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 680
 681         nilfs_truncate_bmap(ii, blkoff);
 682
 683         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 684         if (IS_SYNC(inode))
 685                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 686
 687         nilfs_mark_inode_dirty(inode);
 688         nilfs_set_file_dirty(inode, 0);
 689         nilfs_transaction_commit(sb);
 690         /* May construct a logical segment and may fail in sync mode.
 691            But truncate has no return value. */
 692 }
 693
 694 static void nilfs_clear_inode(struct inode *inode)
 695 {
 696         struct nilfs_inode_info *ii = NILFS_I(inode);
 697         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 698
 699         /*
 700          * Free resources allocated in nilfs_read_inode(), here.
 701          */
 702         BUG_ON(!list_empty(&ii->i_dirty));
 703         brelse(ii->i_bh);
 704         ii->i_bh = NULL;
 705
 706         if (mdi && mdi->mi_palloc_cache)
 707                 nilfs_palloc_destroy_cache(inode);
 708
 709         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 710                 nilfs_bmap_clear(ii->i_bmap);
 711
 712         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 713
 714         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 715                 nilfs_put_root(ii->i_root);
 716 }
 717
 718 void nilfs_evict_inode(struct inode *inode)
 719 {
 720         struct nilfs_transaction_info ti;
 721         struct super_block *sb = inode->i_sb;
 722         struct nilfs_inode_info *ii = NILFS_I(inode);
 723         int ret;
 724
 725         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 726                 if (inode->i_data.nrpages)
 727                         truncate_inode_pages(&inode->i_data, 0);
 728                 end_writeback(inode);
 729                 nilfs_clear_inode(inode);
 730                 return;
 731         }
 732         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 733
 734         if (inode->i_data.nrpages)
 735                 truncate_inode_pages(&inode->i_data, 0);
 736
 737         /* TODO: some of the following operations may fail.  */
 738         nilfs_truncate_bmap(ii, 0);
 739         nilfs_mark_inode_dirty(inode);
 740         end_writeback(inode);
 741
 742         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 743         if (!ret)
 744                 atomic_dec(&ii->i_root->inodes_count);
 745
 746         nilfs_clear_inode(inode);
 747
 748         if (IS_SYNC(inode))
 749                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 750         nilfs_transaction_commit(sb);
 751         /* May construct a logical segment and may fail in sync mode.
 752            But delete_inode has no return value. */
 753 }
 754
 755 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 756 {
 757         struct nilfs_transaction_info ti;
 758         struct inode *inode = dentry->d_inode;
 759         struct super_block *sb = inode->i_sb;
 760         int err;
 761
 762         err = inode_change_ok(inode, iattr);
 763         if (err)
 764                 return err;
 765
 766         err = nilfs_transaction_begin(sb, &ti, 0);
 767         if (unlikely(err))
 768                 return err;
 769
 770         if ((iattr->ia_valid & ATTR_SIZE) &&
 771             iattr->ia_size != i_size_read(inode)) {
 772                 err = vmtruncate(inode, iattr->ia_size);
 773                 if (unlikely(err))
 774                         goto out_err;
 775         }
 776
 777         setattr_copy(inode, iattr);
 778         mark_inode_dirty(inode);
 779
 780         if (iattr->ia_valid & ATTR_MODE) {
 781                 err = nilfs_acl_chmod(inode);
 782                 if (unlikely(err))
 783                         goto out_err;
 784         }
 785
 786         return nilfs_transaction_commit(sb);
 787
 788 out_err:
 789         nilfs_transaction_abort(sb);
 790         return err;
 791 }
 792
 793 int nilfs_permission(struct inode *inode, int mask, unsigned int flags)
 794 {
 795         struct nilfs_root *root;
 796
 797         if (flags & IPERM_FLAG_RCU)
 798                 return -ECHILD;
 799
 800         root = NILFS_I(inode)->i_root;
 801         if ((mask & MAY_WRITE) && root &&
 802             root->cno != NILFS_CPTREE_CURRENT_CNO)
 803                 return -EROFS; /* snapshot is not writable */
 804
 805         return generic_permission(inode, mask, flags, NULL);
 806 }
 807
 808 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 809 {
 810         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 811         struct nilfs_inode_info *ii = NILFS_I(inode);
 812         int err;
 813
 814         spin_lock(&nilfs->ns_inode_lock);
 815         if (ii->i_bh == NULL) {
 816                 spin_unlock(&nilfs->ns_inode_lock);
 817                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 818                                                   inode->i_ino, pbh);
 819                 if (unlikely(err))
 820                         return err;
 821                 spin_lock(&nilfs->ns_inode_lock);
 822                 if (ii->i_bh == NULL)
 823                         ii->i_bh = *pbh;
 824                 else {
 825                         brelse(*pbh);
 826                         *pbh = ii->i_bh;
 827                 }
 828         } else
 829                 *pbh = ii->i_bh;
 830
 831         get_bh(*pbh);
 832         spin_unlock(&nilfs->ns_inode_lock);
 833         return 0;
 834 }
 835
 836 int nilfs_inode_dirty(struct inode *inode)
 837 {
 838         struct nilfs_inode_info *ii = NILFS_I(inode);
 839         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 840         int ret = 0;
 841
 842         if (!list_empty(&ii->i_dirty)) {
 843                 spin_lock(&nilfs->ns_inode_lock);
 844                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 845                         test_bit(NILFS_I_BUSY, &ii->i_state);
 846                 spin_unlock(&nilfs->ns_inode_lock);
 847         }
 848         return ret;
 849 }
 850
 851 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 852 {
 853         struct nilfs_inode_info *ii = NILFS_I(inode);
 854         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 855
 856         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 857
 858         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 859                 return 0;
 860
 861         spin_lock(&nilfs->ns_inode_lock);
 862         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 863             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 864                 /* Because this routine may race with nilfs_dispose_list(),
 865                    we have to check NILFS_I_QUEUED here, too. */
 866                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 867                         /* This will happen when somebody is freeing
 868                            this inode. */
 869                         nilfs_warning(inode->i_sb, __func__,
 870                                       "cannot get inode (ino=%lu)\n",
 871                                       inode->i_ino);
 872                         spin_unlock(&nilfs->ns_inode_lock);
 873                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 874                                            freeing inode */
 875                 }
 876                 list_del(&ii->i_dirty);
 877                 list_add_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 878                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 879         }
 880         spin_unlock(&nilfs->ns_inode_lock);
 881         return 0;
 882 }
 883
 884 int nilfs_mark_inode_dirty(struct inode *inode)
 885 {
 886         struct buffer_head *ibh;
 887         int err;
 888
 889         err = nilfs_load_inode_block(inode, &ibh);
 890         if (unlikely(err)) {
 891                 nilfs_warning(inode->i_sb, __func__,
 892                               "failed to reget inode block.\n");
 893                 return err;
 894         }
 895         nilfs_update_inode(inode, ibh);
 896         nilfs_mdt_mark_buffer_dirty(ibh);
 897         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 898         brelse(ibh);
 899         return 0;
 900 }
 901
 902 /**
 903  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 904  * @inode: inode of the file to be registered.
 905  *
 906  * nilfs_dirty_inode() loads a inode block containing the specified
 907  * @inode and copies data from a nilfs_inode to a corresponding inode
 908  * entry in the inode block. This operation is excluded from the segment
 909  * construction. This function can be called both as a single operation
 910  * and as a part of indivisible file operations.
 911  */
 912 void nilfs_dirty_inode(struct inode *inode)
 913 {
 914         struct nilfs_transaction_info ti;
 915         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 916
 917         if (is_bad_inode(inode)) {
 918                 nilfs_warning(inode->i_sb, __func__,
 919                               "tried to mark bad_inode dirty. ignored.\n");
 920                 dump_stack();
 921                 return;
 922         }
 923         if (mdi) {
 924                 nilfs_mdt_mark_dirty(inode);
 925                 return;
 926         }
 927         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 928         nilfs_mark_inode_dirty(inode);
 929         nilfs_transaction_commit(inode->i_sb); /* never fails */
 930 }
 931
 932 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 933                  __u64 start, __u64 len)
 934 {
 935         struct the_nilfs *nilfs = NILFS_I_NILFS(inode);
 936         __u64 logical = 0, phys = 0, size = 0;
 937         __u32 flags = 0;
 938         loff_t isize;
 939         sector_t blkoff, end_blkoff;
 940         sector_t delalloc_blkoff;
 941         unsigned long delalloc_blklen;
 942         unsigned int blkbits = inode->i_blkbits;
 943         int ret, n;
 944
 945         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 946         if (ret)
 947                 return ret;
 948
 949         mutex_lock(&inode->i_mutex);
 950
 951         isize = i_size_read(inode);
 952
 953         blkoff = start >> blkbits;
 954         end_blkoff = (start + len - 1) >> blkbits;
 955
 956         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
 957                                                         &delalloc_blkoff);
 958
 959         do {
 960                 __u64 blkphy;
 961                 unsigned int maxblocks;
 962
 963                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
 964                         if (size) {
 965                                 /* End of the current extent */
 966                                 ret = fiemap_fill_next_extent(
 967                                         fieinfo, logical, phys, size, flags);
 968                                 if (ret)
 969                                         break;
 970                         }
 971                         if (blkoff > end_blkoff)
 972                                 break;
 973
 974                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
 975                         logical = blkoff << blkbits;
 976                         phys = 0;
 977                         size = delalloc_blklen << blkbits;
 978
 979                         blkoff = delalloc_blkoff + delalloc_blklen;
 980                         delalloc_blklen = nilfs_find_uncommitted_extent(
 981                                 inode, blkoff, &delalloc_blkoff);
 982                         continue;
 983                 }
 984
 985                 /*
 986                  * Limit the number of blocks that we look up so as
 987                  * not to get into the next delayed allocation extent.
 988                  */
 989                 maxblocks = INT_MAX;
 990                 if (delalloc_blklen)
 991                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
 992                                           maxblocks);
 993                 blkphy = 0;
 994
 995                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 996                 n = nilfs_bmap_lookup_contig(
 997                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
 998                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 999
1000                 if (n < 0) {
1001                         int past_eof;
1002
1003                         if (unlikely(n != -ENOENT))
1004                                 break; /* error */
1005
1006                         /* HOLE */
1007                         blkoff++;
1008                         past_eof = ((blkoff << blkbits) >= isize);
1009
1010                         if (size) {
1011                                 /* End of the current extent */
1012
1013                                 if (past_eof)
1014                                         flags |= FIEMAP_EXTENT_LAST;
1015
1016                                 ret = fiemap_fill_next_extent(
1017                                         fieinfo, logical, phys, size, flags);
1018                                 if (ret)
1019                                         break;
1020                                 size = 0;
1021                         }
1022                         if (blkoff > end_blkoff || past_eof)
1023                                 break;
1024                 } else {
1025                         if (size) {
1026                                 if (phys && blkphy << blkbits == phys + size) {
1027                                         /* The current extent goes on */
1028                                         size += n << blkbits;
1029                                 } else {
1030                                         /* Terminate the current extent */
1031                                         ret = fiemap_fill_next_extent(
1032                                                 fieinfo, logical, phys, size,
1033                                                 flags);
1034                                         if (ret || blkoff > end_blkoff)
1035                                                 break;
1036
1037                                         /* Start another extent */
1038                                         flags = FIEMAP_EXTENT_MERGED;
1039                                         logical = blkoff << blkbits;
1040                                         phys = blkphy << blkbits;
1041                                         size = n << blkbits;
1042                                 }
1043                         } else {
1044                                 /* Start a new extent */
1045                                 flags = FIEMAP_EXTENT_MERGED;
1046                                 logical = blkoff << blkbits;
1047                                 phys = blkphy << blkbits;
1048                                 size = n << blkbits;
1049                         }
1050                         blkoff += n;
1051                 }
1052                 cond_resched();
1053         } while (true);
1054
1055         /* If ret is 1 then we just hit the end of the extent array */
1056         if (ret == 1)
1057                 ret = 0;
1058
1059         mutex_unlock(&inode->i_mutex);
1060         return ret;
1061 }