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         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  78         __u64 blknum = 0;
  79         int err = 0, ret;
  80         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  81
  82         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  83         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  84         up_read(&NILFS_MDT(nilfs->ns_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_nobuffers(page);
 199
 200         if (page_has_buffers(page)) {
 201                 struct inode *inode = page->mapping->host;
 202                 unsigned nr_dirty = 0;
 203                 struct buffer_head *bh, *head;
 204
 205                 /*
 206                  * This page is locked by callers, and no other thread
 207                  * concurrently marks its buffers dirty since they are
 208                  * only dirtied through routines in fs/buffer.c in
 209                  * which call sites of mark_buffer_dirty are protected
 210                  * by page lock.
 211                  */
 212                 bh = head = page_buffers(page);
 213                 do {
 214                         /* Do not mark hole blocks dirty */
 215                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 216                                 continue;
 217
 218                         set_buffer_dirty(bh);
 219                         nr_dirty++;
 220                 } while (bh = bh->b_this_page, bh != head);
 221
 222                 if (nr_dirty)
 223                         nilfs_set_file_dirty(inode, nr_dirty);
 224         }
 225         return ret;
 226 }
 227
 228 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 229                              loff_t pos, unsigned len, unsigned flags,
 230                              struct page **pagep, void **fsdata)
 231
 232 {
 233         struct inode *inode = mapping->host;
 234         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 235
 236         if (unlikely(err))
 237                 return err;
 238
 239         err = block_write_begin(mapping, pos, len, flags, pagep,
 240                                 nilfs_get_block);
 241         if (unlikely(err)) {
 242                 loff_t isize = mapping->host->i_size;
 243                 if (pos + len > isize)
 244                         vmtruncate(mapping->host, isize);
 245
 246                 nilfs_transaction_abort(inode->i_sb);
 247         }
 248         return err;
 249 }
 250
 251 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 252                            loff_t pos, unsigned len, unsigned copied,
 253                            struct page *page, void *fsdata)
 254 {
 255         struct inode *inode = mapping->host;
 256         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 257         unsigned nr_dirty;
 258         int err;
 259
 260         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 261                                                   start + copied);
 262         copied = generic_write_end(file, mapping, pos, len, copied, page,
 263                                    fsdata);
 264         nilfs_set_file_dirty(inode, nr_dirty);
 265         err = nilfs_transaction_commit(inode->i_sb);
 266         return err ? : copied;
 267 }
 268
 269 static ssize_t
 270 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 271                 loff_t offset, unsigned long nr_segs)
 272 {
 273         struct file *file = iocb->ki_filp;
 274         struct inode *inode = file->f_mapping->host;
 275         ssize_t size;
 276
 277         if (rw == WRITE)
 278                 return 0;
 279
 280         /* Needs synchronization with the cleaner */
 281         size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 282                                   nilfs_get_block);
 283
 284         /*
 285          * In case of error extending write may have instantiated a few
 286          * blocks outside i_size. Trim these off again.
 287          */
 288         if (unlikely((rw & WRITE) && size < 0)) {
 289                 loff_t isize = i_size_read(inode);
 290                 loff_t end = offset + iov_length(iov, nr_segs);
 291
 292                 if (end > isize)
 293                         vmtruncate(inode, isize);
 294         }
 295
 296         return size;
 297 }
 298
 299 const struct address_space_operations nilfs_aops = {
 300         .writepage              = nilfs_writepage,
 301         .readpage               = nilfs_readpage,
 302         .writepages             = nilfs_writepages,
 303         .set_page_dirty         = nilfs_set_page_dirty,
 304         .readpages              = nilfs_readpages,
 305         .write_begin            = nilfs_write_begin,
 306         .write_end              = nilfs_write_end,
 307         /* .releasepage         = nilfs_releasepage, */
 308         .invalidatepage         = block_invalidatepage,
 309         .direct_IO              = nilfs_direct_IO,
 310         .is_partially_uptodate  = block_is_partially_uptodate,
 311 };
 312
 313 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 314 {
 315         struct super_block *sb = dir->i_sb;
 316         struct the_nilfs *nilfs = sb->s_fs_info;
 317         struct inode *inode;
 318         struct nilfs_inode_info *ii;
 319         struct nilfs_root *root;
 320         int err = -ENOMEM;
 321         ino_t ino;
 322
 323         inode = new_inode(sb);
 324         if (unlikely(!inode))
 325                 goto failed;
 326
 327         mapping_set_gfp_mask(inode->i_mapping,
 328                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 329
 330         root = NILFS_I(dir)->i_root;
 331         ii = NILFS_I(inode);
 332         ii->i_state = 1 << NILFS_I_NEW;
 333         ii->i_root = root;
 334
 335         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 336         if (unlikely(err))
 337                 goto failed_ifile_create_inode;
 338         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 339
 340         atomic_inc(&root->inodes_count);
 341         inode_init_owner(inode, dir, mode);
 342         inode->i_ino = ino;
 343         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 344
 345         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 346                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 347                 if (err < 0)
 348                         goto failed_bmap;
 349
 350                 set_bit(NILFS_I_BMAP, &ii->i_state);
 351                 /* No lock is needed; iget() ensures it. */
 352         }
 353
 354         ii->i_flags = nilfs_mask_flags(
 355                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 356
 357         /* ii->i_file_acl = 0; */
 358         /* ii->i_dir_acl = 0; */
 359         ii->i_dir_start_lookup = 0;
 360         nilfs_set_inode_flags(inode);
 361         spin_lock(&nilfs->ns_next_gen_lock);
 362         inode->i_generation = nilfs->ns_next_generation++;
 363         spin_unlock(&nilfs->ns_next_gen_lock);
 364         insert_inode_hash(inode);
 365
 366         err = nilfs_init_acl(inode, dir);
 367         if (unlikely(err))
 368                 goto failed_acl; /* never occur. When supporting
 369                                     nilfs_init_acl(), proper cancellation of
 370                                     above jobs should be considered */
 371
 372         return inode;
 373
 374  failed_acl:
 375  failed_bmap:
 376         clear_nlink(inode);
 377         iput(inode);  /* raw_inode will be deleted through
 378                          generic_delete_inode() */
 379         goto failed;
 380
 381  failed_ifile_create_inode:
 382         make_bad_inode(inode);
 383         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 384                          called */
 385  failed:
 386         return ERR_PTR(err);
 387 }
 388
 389 void nilfs_set_inode_flags(struct inode *inode)
 390 {
 391         unsigned int flags = NILFS_I(inode)->i_flags;
 392
 393         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 394                             S_DIRSYNC);
 395         if (flags & FS_SYNC_FL)
 396                 inode->i_flags |= S_SYNC;
 397         if (flags & FS_APPEND_FL)
 398                 inode->i_flags |= S_APPEND;
 399         if (flags & FS_IMMUTABLE_FL)
 400                 inode->i_flags |= S_IMMUTABLE;
 401         if (flags & FS_NOATIME_FL)
 402                 inode->i_flags |= S_NOATIME;
 403         if (flags & FS_DIRSYNC_FL)
 404                 inode->i_flags |= S_DIRSYNC;
 405         mapping_set_gfp_mask(inode->i_mapping,
 406                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 407 }
 408
 409 int nilfs_read_inode_common(struct inode *inode,
 410                             struct nilfs_inode *raw_inode)
 411 {
 412         struct nilfs_inode_info *ii = NILFS_I(inode);
 413         int err;
 414
 415         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 416         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 417         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 418         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 419         inode->i_size = le64_to_cpu(raw_inode->i_size);
 420         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 421         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 422         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 423         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 424         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 425         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 426         if (inode->i_nlink == 0 && inode->i_mode == 0)
 427                 return -EINVAL; /* this inode is deleted */
 428
 429         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 430         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 431 #if 0
 432         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 433         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 434                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 435 #endif
 436         ii->i_dir_start_lookup = 0;
 437         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 438
 439         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 440             S_ISLNK(inode->i_mode)) {
 441                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 442                 if (err < 0)
 443                         return err;
 444                 set_bit(NILFS_I_BMAP, &ii->i_state);
 445                 /* No lock is needed; iget() ensures it. */
 446         }
 447         return 0;
 448 }
 449
 450 static int __nilfs_read_inode(struct super_block *sb,
 451                               struct nilfs_root *root, unsigned long ino,
 452                               struct inode *inode)
 453 {
 454         struct the_nilfs *nilfs = sb->s_fs_info;
 455         struct buffer_head *bh;
 456         struct nilfs_inode *raw_inode;
 457         int err;
 458
 459         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 460         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 461         if (unlikely(err))
 462                 goto bad_inode;
 463
 464         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 465
 466         err = nilfs_read_inode_common(inode, raw_inode);
 467         if (err)
 468                 goto failed_unmap;
 469
 470         if (S_ISREG(inode->i_mode)) {
 471                 inode->i_op = &nilfs_file_inode_operations;
 472                 inode->i_fop = &nilfs_file_operations;
 473                 inode->i_mapping->a_ops = &nilfs_aops;
 474         } else if (S_ISDIR(inode->i_mode)) {
 475                 inode->i_op = &nilfs_dir_inode_operations;
 476                 inode->i_fop = &nilfs_dir_operations;
 477                 inode->i_mapping->a_ops = &nilfs_aops;
 478         } else if (S_ISLNK(inode->i_mode)) {
 479                 inode->i_op = &nilfs_symlink_inode_operations;
 480                 inode->i_mapping->a_ops = &nilfs_aops;
 481         } else {
 482                 inode->i_op = &nilfs_special_inode_operations;
 483                 init_special_inode(
 484                         inode, inode->i_mode,
 485                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 486         }
 487         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 488         brelse(bh);
 489         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 490         nilfs_set_inode_flags(inode);
 491         return 0;
 492
 493  failed_unmap:
 494         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 495         brelse(bh);
 496
 497  bad_inode:
 498         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 499         return err;
 500 }
 501
 502 static int nilfs_iget_test(struct inode *inode, void *opaque)
 503 {
 504         struct nilfs_iget_args *args = opaque;
 505         struct nilfs_inode_info *ii;
 506
 507         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 508                 return 0;
 509
 510         ii = NILFS_I(inode);
 511         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 512                 return !args->for_gc;
 513
 514         return args->for_gc && args->cno == ii->i_cno;
 515 }
 516
 517 static int nilfs_iget_set(struct inode *inode, void *opaque)
 518 {
 519         struct nilfs_iget_args *args = opaque;
 520
 521         inode->i_ino = args->ino;
 522         if (args->for_gc) {
 523                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 524                 NILFS_I(inode)->i_cno = args->cno;
 525                 NILFS_I(inode)->i_root = NULL;
 526         } else {
 527                 if (args->root && args->ino == NILFS_ROOT_INO)
 528                         nilfs_get_root(args->root);
 529                 NILFS_I(inode)->i_root = args->root;
 530         }
 531         return 0;
 532 }
 533
 534 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 535                             unsigned long ino)
 536 {
 537         struct nilfs_iget_args args = {
 538                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 539         };
 540
 541         return ilookup5(sb, ino, nilfs_iget_test, &args);
 542 }
 543
 544 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 545                                 unsigned long ino)
 546 {
 547         struct nilfs_iget_args args = {
 548                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 549         };
 550
 551         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 552 }
 553
 554 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 555                          unsigned long ino)
 556 {
 557         struct inode *inode;
 558         int err;
 559
 560         inode = nilfs_iget_locked(sb, root, ino);
 561         if (unlikely(!inode))
 562                 return ERR_PTR(-ENOMEM);
 563         if (!(inode->i_state & I_NEW))
 564                 return inode;
 565
 566         err = __nilfs_read_inode(sb, root, ino, inode);
 567         if (unlikely(err)) {
 568                 iget_failed(inode);
 569                 return ERR_PTR(err);
 570         }
 571         unlock_new_inode(inode);
 572         return inode;
 573 }
 574
 575 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 576                                 __u64 cno)
 577 {
 578         struct nilfs_iget_args args = {
 579                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 580         };
 581         struct inode *inode;
 582         int err;
 583
 584         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 585         if (unlikely(!inode))
 586                 return ERR_PTR(-ENOMEM);
 587         if (!(inode->i_state & I_NEW))
 588                 return inode;
 589
 590         err = nilfs_init_gcinode(inode);
 591         if (unlikely(err)) {
 592                 iget_failed(inode);
 593                 return ERR_PTR(err);
 594         }
 595         unlock_new_inode(inode);
 596         return inode;
 597 }
 598
 599 void nilfs_write_inode_common(struct inode *inode,
 600                               struct nilfs_inode *raw_inode, int has_bmap)
 601 {
 602         struct nilfs_inode_info *ii = NILFS_I(inode);
 603
 604         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 605         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 606         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 607         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 608         raw_inode->i_size = cpu_to_le64(inode->i_size);
 609         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 610         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 611         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 612         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 613         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 614
 615         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 616         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 617
 618         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 619                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 620
 621                 /* zero-fill unused portion in the case of super root block */
 622                 raw_inode->i_xattr = 0;
 623                 raw_inode->i_pad = 0;
 624                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 625                        nilfs->ns_inode_size - sizeof(*raw_inode));
 626         }
 627
 628         if (has_bmap)
 629                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 630         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 631                 raw_inode->i_device_code =
 632                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 633         /* When extending inode, nilfs->ns_inode_size should be checked
 634            for substitutions of appended fields */
 635 }
 636
 637 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 638 {
 639         ino_t ino = inode->i_ino;
 640         struct nilfs_inode_info *ii = NILFS_I(inode);
 641         struct inode *ifile = ii->i_root->ifile;
 642         struct nilfs_inode *raw_inode;
 643
 644         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 645
 646         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 647                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 648         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 649
 650         nilfs_write_inode_common(inode, raw_inode, 0);
 651                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 652                    deadlock of bmap. This delays update of i_bmap to just
 653                    before writing */
 654         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 655 }
 656
 657 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 658
 659 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 660                                 unsigned long from)
 661 {
 662         unsigned long b;
 663         int ret;
 664
 665         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 666                 return;
 667 repeat:
 668         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 669         if (ret == -ENOENT)
 670                 return;
 671         else if (ret < 0)
 672                 goto failed;
 673
 674         if (b < from)
 675                 return;
 676
 677         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 678         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 679         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 680         if (!ret || (ret == -ENOMEM &&
 681                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 682                 goto repeat;
 683
 684 failed:
 685         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 686                       "failed to truncate bmap (ino=%lu, err=%d)",
 687                       ii->vfs_inode.i_ino, ret);
 688 }
 689
 690 void nilfs_truncate(struct inode *inode)
 691 {
 692         unsigned long blkoff;
 693         unsigned int blocksize;
 694         struct nilfs_transaction_info ti;
 695         struct super_block *sb = inode->i_sb;
 696         struct nilfs_inode_info *ii = NILFS_I(inode);
 697
 698         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 699                 return;
 700         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 701                 return;
 702
 703         blocksize = sb->s_blocksize;
 704         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 705         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 706
 707         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 708
 709         nilfs_truncate_bmap(ii, blkoff);
 710
 711         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 712         if (IS_SYNC(inode))
 713                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 714
 715         nilfs_mark_inode_dirty(inode);
 716         nilfs_set_file_dirty(inode, 0);
 717         nilfs_transaction_commit(sb);
 718         /* May construct a logical segment and may fail in sync mode.
 719            But truncate has no return value. */
 720 }
 721
 722 static void nilfs_clear_inode(struct inode *inode)
 723 {
 724         struct nilfs_inode_info *ii = NILFS_I(inode);
 725         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 726
 727         /*
 728          * Free resources allocated in nilfs_read_inode(), here.
 729          */
 730         BUG_ON(!list_empty(&ii->i_dirty));
 731         brelse(ii->i_bh);
 732         ii->i_bh = NULL;
 733
 734         if (mdi && mdi->mi_palloc_cache)
 735                 nilfs_palloc_destroy_cache(inode);
 736
 737         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 738                 nilfs_bmap_clear(ii->i_bmap);
 739
 740         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 741
 742         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 743                 nilfs_put_root(ii->i_root);
 744 }
 745
 746 void nilfs_evict_inode(struct inode *inode)
 747 {
 748         struct nilfs_transaction_info ti;
 749         struct super_block *sb = inode->i_sb;
 750         struct nilfs_inode_info *ii = NILFS_I(inode);
 751         int ret;
 752
 753         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 754                 if (inode->i_data.nrpages)
 755                         truncate_inode_pages(&inode->i_data, 0);
 756                 end_writeback(inode);
 757                 nilfs_clear_inode(inode);
 758                 return;
 759         }
 760         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 761
 762         if (inode->i_data.nrpages)
 763                 truncate_inode_pages(&inode->i_data, 0);
 764
 765         /* TODO: some of the following operations may fail.  */
 766         nilfs_truncate_bmap(ii, 0);
 767         nilfs_mark_inode_dirty(inode);
 768         end_writeback(inode);
 769
 770         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 771         if (!ret)
 772                 atomic_dec(&ii->i_root->inodes_count);
 773
 774         nilfs_clear_inode(inode);
 775
 776         if (IS_SYNC(inode))
 777                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 778         nilfs_transaction_commit(sb);
 779         /* May construct a logical segment and may fail in sync mode.
 780            But delete_inode has no return value. */
 781 }
 782
 783 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 784 {
 785         struct nilfs_transaction_info ti;
 786         struct inode *inode = dentry->d_inode;
 787         struct super_block *sb = inode->i_sb;
 788         int err;
 789
 790         err = inode_change_ok(inode, iattr);
 791         if (err)
 792                 return err;
 793
 794         err = nilfs_transaction_begin(sb, &ti, 0);
 795         if (unlikely(err))
 796                 return err;
 797
 798         if ((iattr->ia_valid & ATTR_SIZE) &&
 799             iattr->ia_size != i_size_read(inode)) {
 800                 inode_dio_wait(inode);
 801
 802                 err = vmtruncate(inode, iattr->ia_size);
 803                 if (unlikely(err))
 804                         goto out_err;
 805         }
 806
 807         setattr_copy(inode, iattr);
 808         mark_inode_dirty(inode);
 809
 810         if (iattr->ia_valid & ATTR_MODE) {
 811                 err = nilfs_acl_chmod(inode);
 812                 if (unlikely(err))
 813                         goto out_err;
 814         }
 815
 816         return nilfs_transaction_commit(sb);
 817
 818 out_err:
 819         nilfs_transaction_abort(sb);
 820         return err;
 821 }
 822
 823 int nilfs_permission(struct inode *inode, int mask)
 824 {
 825         struct nilfs_root *root = NILFS_I(inode)->i_root;
 826         if ((mask & MAY_WRITE) && root &&
 827             root->cno != NILFS_CPTREE_CURRENT_CNO)
 828                 return -EROFS; /* snapshot is not writable */
 829
 830         return generic_permission(inode, mask);
 831 }
 832
 833 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 834 {
 835         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 836         struct nilfs_inode_info *ii = NILFS_I(inode);
 837         int err;
 838
 839         spin_lock(&nilfs->ns_inode_lock);
 840         if (ii->i_bh == NULL) {
 841                 spin_unlock(&nilfs->ns_inode_lock);
 842                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 843                                                   inode->i_ino, pbh);
 844                 if (unlikely(err))
 845                         return err;
 846                 spin_lock(&nilfs->ns_inode_lock);
 847                 if (ii->i_bh == NULL)
 848                         ii->i_bh = *pbh;
 849                 else {
 850                         brelse(*pbh);
 851                         *pbh = ii->i_bh;
 852                 }
 853         } else
 854                 *pbh = ii->i_bh;
 855
 856         get_bh(*pbh);
 857         spin_unlock(&nilfs->ns_inode_lock);
 858         return 0;
 859 }
 860
 861 int nilfs_inode_dirty(struct inode *inode)
 862 {
 863         struct nilfs_inode_info *ii = NILFS_I(inode);
 864         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 865         int ret = 0;
 866
 867         if (!list_empty(&ii->i_dirty)) {
 868                 spin_lock(&nilfs->ns_inode_lock);
 869                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 870                         test_bit(NILFS_I_BUSY, &ii->i_state);
 871                 spin_unlock(&nilfs->ns_inode_lock);
 872         }
 873         return ret;
 874 }
 875
 876 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 877 {
 878         struct nilfs_inode_info *ii = NILFS_I(inode);
 879         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 880
 881         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 882
 883         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 884                 return 0;
 885
 886         spin_lock(&nilfs->ns_inode_lock);
 887         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 888             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 889                 /* Because this routine may race with nilfs_dispose_list(),
 890                    we have to check NILFS_I_QUEUED here, too. */
 891                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 892                         /* This will happen when somebody is freeing
 893                            this inode. */
 894                         nilfs_warning(inode->i_sb, __func__,
 895                                       "cannot get inode (ino=%lu)\n",
 896                                       inode->i_ino);
 897                         spin_unlock(&nilfs->ns_inode_lock);
 898                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 899                                            freeing inode */
 900                 }
 901                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 902                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 903         }
 904         spin_unlock(&nilfs->ns_inode_lock);
 905         return 0;
 906 }
 907
 908 int nilfs_mark_inode_dirty(struct inode *inode)
 909 {
 910         struct buffer_head *ibh;
 911         int err;
 912
 913         err = nilfs_load_inode_block(inode, &ibh);
 914         if (unlikely(err)) {
 915                 nilfs_warning(inode->i_sb, __func__,
 916                               "failed to reget inode block.\n");
 917                 return err;
 918         }
 919         nilfs_update_inode(inode, ibh);
 920         mark_buffer_dirty(ibh);
 921         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 922         brelse(ibh);
 923         return 0;
 924 }
 925
 926 /**
 927  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 928  * @inode: inode of the file to be registered.
 929  *
 930  * nilfs_dirty_inode() loads a inode block containing the specified
 931  * @inode and copies data from a nilfs_inode to a corresponding inode
 932  * entry in the inode block. This operation is excluded from the segment
 933  * construction. This function can be called both as a single operation
 934  * and as a part of indivisible file operations.
 935  */
 936 void nilfs_dirty_inode(struct inode *inode, int flags)
 937 {
 938         struct nilfs_transaction_info ti;
 939         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 940
 941         if (is_bad_inode(inode)) {
 942                 nilfs_warning(inode->i_sb, __func__,
 943                               "tried to mark bad_inode dirty. ignored.\n");
 944                 dump_stack();
 945                 return;
 946         }
 947         if (mdi) {
 948                 nilfs_mdt_mark_dirty(inode);
 949                 return;
 950         }
 951         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 952         nilfs_mark_inode_dirty(inode);
 953         nilfs_transaction_commit(inode->i_sb); /* never fails */
 954 }
 955
 956 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 957                  __u64 start, __u64 len)
 958 {
 959         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 960         __u64 logical = 0, phys = 0, size = 0;
 961         __u32 flags = 0;
 962         loff_t isize;
 963         sector_t blkoff, end_blkoff;
 964         sector_t delalloc_blkoff;
 965         unsigned long delalloc_blklen;
 966         unsigned int blkbits = inode->i_blkbits;
 967         int ret, n;
 968
 969         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 970         if (ret)
 971                 return ret;
 972
 973         mutex_lock(&inode->i_mutex);
 974
 975         isize = i_size_read(inode);
 976
 977         blkoff = start >> blkbits;
 978         end_blkoff = (start + len - 1) >> blkbits;
 979
 980         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
 981                                                         &delalloc_blkoff);
 982
 983         do {
 984                 __u64 blkphy;
 985                 unsigned int maxblocks;
 986
 987                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
 988                         if (size) {
 989                                 /* End of the current extent */
 990                                 ret = fiemap_fill_next_extent(
 991                                         fieinfo, logical, phys, size, flags);
 992                                 if (ret)
 993                                         break;
 994                         }
 995                         if (blkoff > end_blkoff)
 996                                 break;
 997
 998                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
 999                         logical = blkoff << blkbits;
1000                         phys = 0;
1001                         size = delalloc_blklen << blkbits;
1002
1003                         blkoff = delalloc_blkoff + delalloc_blklen;
1004                         delalloc_blklen = nilfs_find_uncommitted_extent(
1005                                 inode, blkoff, &delalloc_blkoff);
1006                         continue;
1007                 }
1008
1009                 /*
1010                  * Limit the number of blocks that we look up so as
1011                  * not to get into the next delayed allocation extent.
1012                  */
1013                 maxblocks = INT_MAX;
1014                 if (delalloc_blklen)
1015                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1016                                           maxblocks);
1017                 blkphy = 0;
1018
1019                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1020                 n = nilfs_bmap_lookup_contig(
1021                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1022                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1023
1024                 if (n < 0) {
1025                         int past_eof;
1026
1027                         if (unlikely(n != -ENOENT))
1028                                 break; /* error */
1029
1030                         /* HOLE */
1031                         blkoff++;
1032                         past_eof = ((blkoff << blkbits) >= isize);
1033
1034                         if (size) {
1035                                 /* End of the current extent */
1036
1037                                 if (past_eof)
1038                                         flags |= FIEMAP_EXTENT_LAST;
1039
1040                                 ret = fiemap_fill_next_extent(
1041                                         fieinfo, logical, phys, size, flags);
1042                                 if (ret)
1043                                         break;
1044                                 size = 0;
1045                         }
1046                         if (blkoff > end_blkoff || past_eof)
1047                                 break;
1048                 } else {
1049                         if (size) {
1050                                 if (phys && blkphy << blkbits == phys + size) {
1051                                         /* The current extent goes on */
1052                                         size += n << blkbits;
1053                                 } else {
1054                                         /* Terminate the current extent */
1055                                         ret = fiemap_fill_next_extent(
1056                                                 fieinfo, logical, phys, size,
1057                                                 flags);
1058                                         if (ret || blkoff > end_blkoff)
1059                                                 break;
1060
1061                                         /* Start another extent */
1062                                         flags = FIEMAP_EXTENT_MERGED;
1063                                         logical = blkoff << blkbits;
1064                                         phys = blkphy << blkbits;
1065                                         size = n << blkbits;
1066                                 }
1067                         } else {
1068                                 /* Start a new extent */
1069                                 flags = FIEMAP_EXTENT_MERGED;
1070                                 logical = blkoff << blkbits;
1071                                 phys = blkphy << blkbits;
1072                                 size = n << blkbits;
1073                         }
1074                         blkoff += n;
1075                 }
1076                 cond_resched();
1077         } while (true);
1078
1079         /* If ret is 1 then we just hit the end of the extent array */
1080         if (ret == 1)
1081                 ret = 0;
1082
1083         mutex_unlock(&inode->i_mutex);
1084         return ret;
1085 }