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/pagemap.h>
  28 #include <linux/writeback.h>
  29 #include <linux/uio.h>
  30 #include "nilfs.h"
  31 #include "btnode.h"
  32 #include "segment.h"
  33 #include "page.h"
  34 #include "mdt.h"
  35 #include "cpfile.h"
  36 #include "ifile.h"
  37
  38 /**
  39  * struct nilfs_iget_args - arguments used during comparison between inodes
  40  * @ino: inode number
  41  * @cno: checkpoint number
  42  * @root: pointer on NILFS root object (mounted checkpoint)
  43  * @for_gc: inode for GC flag
  44  */
  45 struct nilfs_iget_args {
  46         u64 ino;
  47         __u64 cno;
  48         struct nilfs_root *root;
  49         int for_gc;
  50 };
  51
  52 static int nilfs_iget_test(struct inode *inode, void *opaque);
  53
  54 void nilfs_inode_add_blocks(struct inode *inode, int n)
  55 {
  56         struct nilfs_root *root = NILFS_I(inode)->i_root;
  57
  58         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  59         if (root)
  60                 atomic64_add(n, &root->blocks_count);
  61 }
  62
  63 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  64 {
  65         struct nilfs_root *root = NILFS_I(inode)->i_root;
  66
  67         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  68         if (root)
  69                 atomic64_sub(n, &root->blocks_count);
  70 }
  71
  72 /**
  73  * nilfs_get_block() - get a file block on the filesystem (callback function)
  74  * @inode - inode struct of the target file
  75  * @blkoff - file block number
  76  * @bh_result - buffer head to be mapped on
  77  * @create - indicate whether allocating the block or not when it has not
  78  *      been allocated yet.
  79  *
  80  * This function does not issue actual read request of the specified data
  81  * block. It is done by VFS.
  82  */
  83 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  84                     struct buffer_head *bh_result, int create)
  85 {
  86         struct nilfs_inode_info *ii = NILFS_I(inode);
  87         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  88         __u64 blknum = 0;
  89         int err = 0, ret;
  90         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  91
  92         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  93         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  94         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  95         if (ret >= 0) { /* found */
  96                 map_bh(bh_result, inode->i_sb, blknum);
  97                 if (ret > 0)
  98                         bh_result->b_size = (ret << inode->i_blkbits);
  99                 goto out;
 100         }
 101         /* data block was not found */
 102         if (ret == -ENOENT && create) {
 103                 struct nilfs_transaction_info ti;
 104
 105                 bh_result->b_blocknr = 0;
 106                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 107                 if (unlikely(err))
 108                         goto out;
 109                 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 110                                         (unsigned long)bh_result);
 111                 if (unlikely(err != 0)) {
 112                         if (err == -EEXIST) {
 113                                 /*
 114                                  * The get_block() function could be called
 115                                  * from multiple callers for an inode.
 116                                  * However, the page having this block must
 117                                  * be locked in this case.
 118                                  */
 119                                 printk(KERN_WARNING
 120                                        "nilfs_get_block: a race condition "
 121                                        "while inserting a data block. "
 122                                        "(inode number=%lu, file block "
 123                                        "offset=%llu)\n",
 124                                        inode->i_ino,
 125                                        (unsigned long long)blkoff);
 126                                 err = 0;
 127                         }
 128                         nilfs_transaction_abort(inode->i_sb);
 129                         goto out;
 130                 }
 131                 nilfs_mark_inode_dirty_sync(inode);
 132                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 133                 /* Error handling should be detailed */
 134                 set_buffer_new(bh_result);
 135                 set_buffer_delay(bh_result);
 136                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 137                                                       to proper value */
 138         } else if (ret == -ENOENT) {
 139                 /* not found is not error (e.g. hole); must return without
 140                    the mapped state flag. */
 141                 ;
 142         } else {
 143                 err = ret;
 144         }
 145
 146  out:
 147         return err;
 148 }
 149
 150 /**
 151  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 152  * address_space_operations.
 153  * @file - file struct of the file to be read
 154  * @page - the page to be read
 155  */
 156 static int nilfs_readpage(struct file *file, struct page *page)
 157 {
 158         return mpage_readpage(page, nilfs_get_block);
 159 }
 160
 161 /**
 162  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 163  * address_space_operations.
 164  * @file - file struct of the file to be read
 165  * @mapping - address_space struct used for reading multiple pages
 166  * @pages - the pages to be read
 167  * @nr_pages - number of pages to be read
 168  */
 169 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 170                            struct list_head *pages, unsigned nr_pages)
 171 {
 172         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 173 }
 174
 175 static int nilfs_writepages(struct address_space *mapping,
 176                             struct writeback_control *wbc)
 177 {
 178         struct inode *inode = mapping->host;
 179         int err = 0;
 180
 181         if (inode->i_sb->s_flags & MS_RDONLY) {
 182                 nilfs_clear_dirty_pages(mapping, false);
 183                 return -EROFS;
 184         }
 185
 186         if (wbc->sync_mode == WB_SYNC_ALL)
 187                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 188                                                     wbc->range_start,
 189                                                     wbc->range_end);
 190         return err;
 191 }
 192
 193 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 194 {
 195         struct inode *inode = page->mapping->host;
 196         int err;
 197
 198         if (inode->i_sb->s_flags & MS_RDONLY) {
 199                 /*
 200                  * It means that filesystem was remounted in read-only
 201                  * mode because of error or metadata corruption. But we
 202                  * have dirty pages that try to be flushed in background.
 203                  * So, here we simply discard this dirty page.
 204                  */
 205                 nilfs_clear_dirty_page(page, false);
 206                 unlock_page(page);
 207                 return -EROFS;
 208         }
 209
 210         redirty_page_for_writepage(wbc, page);
 211         unlock_page(page);
 212
 213         if (wbc->sync_mode == WB_SYNC_ALL) {
 214                 err = nilfs_construct_segment(inode->i_sb);
 215                 if (unlikely(err))
 216                         return err;
 217         } else if (wbc->for_reclaim)
 218                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 219
 220         return 0;
 221 }
 222
 223 static int nilfs_set_page_dirty(struct page *page)
 224 {
 225         struct inode *inode = page->mapping->host;
 226         int ret = __set_page_dirty_nobuffers(page);
 227
 228         if (page_has_buffers(page)) {
 229                 unsigned nr_dirty = 0;
 230                 struct buffer_head *bh, *head;
 231
 232                 /*
 233                  * This page is locked by callers, and no other thread
 234                  * concurrently marks its buffers dirty since they are
 235                  * only dirtied through routines in fs/buffer.c in
 236                  * which call sites of mark_buffer_dirty are protected
 237                  * by page lock.
 238                  */
 239                 bh = head = page_buffers(page);
 240                 do {
 241                         /* Do not mark hole blocks dirty */
 242                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 243                                 continue;
 244
 245                         set_buffer_dirty(bh);
 246                         nr_dirty++;
 247                 } while (bh = bh->b_this_page, bh != head);
 248
 249                 if (nr_dirty)
 250                         nilfs_set_file_dirty(inode, nr_dirty);
 251         } else if (ret) {
 252                 unsigned nr_dirty = 1 << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 253
 254                 nilfs_set_file_dirty(inode, nr_dirty);
 255         }
 256         return ret;
 257 }
 258
 259 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 260 {
 261         struct inode *inode = mapping->host;
 262
 263         if (to > inode->i_size) {
 264                 truncate_pagecache(inode, inode->i_size);
 265                 nilfs_truncate(inode);
 266         }
 267 }
 268
 269 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 270                              loff_t pos, unsigned len, unsigned flags,
 271                              struct page **pagep, void **fsdata)
 272
 273 {
 274         struct inode *inode = mapping->host;
 275         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 276
 277         if (unlikely(err))
 278                 return err;
 279
 280         err = block_write_begin(mapping, pos, len, flags, pagep,
 281                                 nilfs_get_block);
 282         if (unlikely(err)) {
 283                 nilfs_write_failed(mapping, pos + len);
 284                 nilfs_transaction_abort(inode->i_sb);
 285         }
 286         return err;
 287 }
 288
 289 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 290                            loff_t pos, unsigned len, unsigned copied,
 291                            struct page *page, void *fsdata)
 292 {
 293         struct inode *inode = mapping->host;
 294         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 295         unsigned nr_dirty;
 296         int err;
 297
 298         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 299                                                   start + copied);
 300         copied = generic_write_end(file, mapping, pos, len, copied, page,
 301                                    fsdata);
 302         nilfs_set_file_dirty(inode, nr_dirty);
 303         err = nilfs_transaction_commit(inode->i_sb);
 304         return err ? : copied;
 305 }
 306
 307 static ssize_t
 308 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
 309 {
 310         struct inode *inode = file_inode(iocb->ki_filp);
 311
 312         if (iov_iter_rw(iter) == WRITE)
 313                 return 0;
 314
 315         /* Needs synchronization with the cleaner */
 316         return blockdev_direct_IO(iocb, inode, iter, offset, nilfs_get_block);
 317 }
 318
 319 const struct address_space_operations nilfs_aops = {
 320         .writepage              = nilfs_writepage,
 321         .readpage               = nilfs_readpage,
 322         .writepages             = nilfs_writepages,
 323         .set_page_dirty         = nilfs_set_page_dirty,
 324         .readpages              = nilfs_readpages,
 325         .write_begin            = nilfs_write_begin,
 326         .write_end              = nilfs_write_end,
 327         /* .releasepage         = nilfs_releasepage, */
 328         .invalidatepage         = block_invalidatepage,
 329         .direct_IO              = nilfs_direct_IO,
 330         .is_partially_uptodate  = block_is_partially_uptodate,
 331 };
 332
 333 static int nilfs_insert_inode_locked(struct inode *inode,
 334                                      struct nilfs_root *root,
 335                                      unsigned long ino)
 336 {
 337         struct nilfs_iget_args args = {
 338                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 339         };
 340
 341         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 342 }
 343
 344 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 345 {
 346         struct super_block *sb = dir->i_sb;
 347         struct the_nilfs *nilfs = sb->s_fs_info;
 348         struct inode *inode;
 349         struct nilfs_inode_info *ii;
 350         struct nilfs_root *root;
 351         int err = -ENOMEM;
 352         ino_t ino;
 353
 354         inode = new_inode(sb);
 355         if (unlikely(!inode))
 356                 goto failed;
 357
 358         mapping_set_gfp_mask(inode->i_mapping,
 359                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 360
 361         root = NILFS_I(dir)->i_root;
 362         ii = NILFS_I(inode);
 363         ii->i_state = 1 << NILFS_I_NEW;
 364         ii->i_root = root;
 365
 366         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 367         if (unlikely(err))
 368                 goto failed_ifile_create_inode;
 369         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 370
 371         atomic64_inc(&root->inodes_count);
 372         inode_init_owner(inode, dir, mode);
 373         inode->i_ino = ino;
 374         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 375
 376         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 377                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 378                 if (err < 0)
 379                         goto failed_after_creation;
 380
 381                 set_bit(NILFS_I_BMAP, &ii->i_state);
 382                 /* No lock is needed; iget() ensures it. */
 383         }
 384
 385         ii->i_flags = nilfs_mask_flags(
 386                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 387
 388         /* ii->i_file_acl = 0; */
 389         /* ii->i_dir_acl = 0; */
 390         ii->i_dir_start_lookup = 0;
 391         nilfs_set_inode_flags(inode);
 392         spin_lock(&nilfs->ns_next_gen_lock);
 393         inode->i_generation = nilfs->ns_next_generation++;
 394         spin_unlock(&nilfs->ns_next_gen_lock);
 395         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 396                 err = -EIO;
 397                 goto failed_after_creation;
 398         }
 399
 400         err = nilfs_init_acl(inode, dir);
 401         if (unlikely(err))
 402                 goto failed_after_creation; /* never occur. When supporting
 403                                     nilfs_init_acl(), proper cancellation of
 404                                     above jobs should be considered */
 405
 406         return inode;
 407
 408  failed_after_creation:
 409         clear_nlink(inode);
 410         unlock_new_inode(inode);
 411         iput(inode);  /* raw_inode will be deleted through
 412                          nilfs_evict_inode() */
 413         goto failed;
 414
 415  failed_ifile_create_inode:
 416         make_bad_inode(inode);
 417         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 418                          called */
 419  failed:
 420         return ERR_PTR(err);
 421 }
 422
 423 void nilfs_set_inode_flags(struct inode *inode)
 424 {
 425         unsigned int flags = NILFS_I(inode)->i_flags;
 426         unsigned int new_fl = 0;
 427
 428         if (flags & FS_SYNC_FL)
 429                 new_fl |= S_SYNC;
 430         if (flags & FS_APPEND_FL)
 431                 new_fl |= S_APPEND;
 432         if (flags & FS_IMMUTABLE_FL)
 433                 new_fl |= S_IMMUTABLE;
 434         if (flags & FS_NOATIME_FL)
 435                 new_fl |= S_NOATIME;
 436         if (flags & FS_DIRSYNC_FL)
 437                 new_fl |= S_DIRSYNC;
 438         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 439                         S_NOATIME | S_DIRSYNC);
 440 }
 441
 442 int nilfs_read_inode_common(struct inode *inode,
 443                             struct nilfs_inode *raw_inode)
 444 {
 445         struct nilfs_inode_info *ii = NILFS_I(inode);
 446         int err;
 447
 448         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 449         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 450         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 451         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 452         inode->i_size = le64_to_cpu(raw_inode->i_size);
 453         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 454         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 455         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 456         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 457         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 458         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 459         if (inode->i_nlink == 0)
 460                 return -ESTALE; /* this inode is deleted */
 461
 462         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 463         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 464 #if 0
 465         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 466         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 467                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 468 #endif
 469         ii->i_dir_start_lookup = 0;
 470         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 471
 472         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 473             S_ISLNK(inode->i_mode)) {
 474                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 475                 if (err < 0)
 476                         return err;
 477                 set_bit(NILFS_I_BMAP, &ii->i_state);
 478                 /* No lock is needed; iget() ensures it. */
 479         }
 480         return 0;
 481 }
 482
 483 static int __nilfs_read_inode(struct super_block *sb,
 484                               struct nilfs_root *root, unsigned long ino,
 485                               struct inode *inode)
 486 {
 487         struct the_nilfs *nilfs = sb->s_fs_info;
 488         struct buffer_head *bh;
 489         struct nilfs_inode *raw_inode;
 490         int err;
 491
 492         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 493         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 494         if (unlikely(err))
 495                 goto bad_inode;
 496
 497         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 498
 499         err = nilfs_read_inode_common(inode, raw_inode);
 500         if (err)
 501                 goto failed_unmap;
 502
 503         if (S_ISREG(inode->i_mode)) {
 504                 inode->i_op = &nilfs_file_inode_operations;
 505                 inode->i_fop = &nilfs_file_operations;
 506                 inode->i_mapping->a_ops = &nilfs_aops;
 507         } else if (S_ISDIR(inode->i_mode)) {
 508                 inode->i_op = &nilfs_dir_inode_operations;
 509                 inode->i_fop = &nilfs_dir_operations;
 510                 inode->i_mapping->a_ops = &nilfs_aops;
 511         } else if (S_ISLNK(inode->i_mode)) {
 512                 inode->i_op = &nilfs_symlink_inode_operations;
 513                 inode_nohighmem(inode);
 514                 inode->i_mapping->a_ops = &nilfs_aops;
 515         } else {
 516                 inode->i_op = &nilfs_special_inode_operations;
 517                 init_special_inode(
 518                         inode, inode->i_mode,
 519                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 520         }
 521         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 522         brelse(bh);
 523         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 524         nilfs_set_inode_flags(inode);
 525         mapping_set_gfp_mask(inode->i_mapping,
 526                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 527         return 0;
 528
 529  failed_unmap:
 530         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 531         brelse(bh);
 532
 533  bad_inode:
 534         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 535         return err;
 536 }
 537
 538 static int nilfs_iget_test(struct inode *inode, void *opaque)
 539 {
 540         struct nilfs_iget_args *args = opaque;
 541         struct nilfs_inode_info *ii;
 542
 543         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 544                 return 0;
 545
 546         ii = NILFS_I(inode);
 547         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 548                 return !args->for_gc;
 549
 550         return args->for_gc && args->cno == ii->i_cno;
 551 }
 552
 553 static int nilfs_iget_set(struct inode *inode, void *opaque)
 554 {
 555         struct nilfs_iget_args *args = opaque;
 556
 557         inode->i_ino = args->ino;
 558         if (args->for_gc) {
 559                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 560                 NILFS_I(inode)->i_cno = args->cno;
 561                 NILFS_I(inode)->i_root = NULL;
 562         } else {
 563                 if (args->root && args->ino == NILFS_ROOT_INO)
 564                         nilfs_get_root(args->root);
 565                 NILFS_I(inode)->i_root = args->root;
 566         }
 567         return 0;
 568 }
 569
 570 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 571                             unsigned long ino)
 572 {
 573         struct nilfs_iget_args args = {
 574                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 575         };
 576
 577         return ilookup5(sb, ino, nilfs_iget_test, &args);
 578 }
 579
 580 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 581                                 unsigned long ino)
 582 {
 583         struct nilfs_iget_args args = {
 584                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 585         };
 586
 587         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 588 }
 589
 590 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 591                          unsigned long ino)
 592 {
 593         struct inode *inode;
 594         int err;
 595
 596         inode = nilfs_iget_locked(sb, root, ino);
 597         if (unlikely(!inode))
 598                 return ERR_PTR(-ENOMEM);
 599         if (!(inode->i_state & I_NEW))
 600                 return inode;
 601
 602         err = __nilfs_read_inode(sb, root, ino, inode);
 603         if (unlikely(err)) {
 604                 iget_failed(inode);
 605                 return ERR_PTR(err);
 606         }
 607         unlock_new_inode(inode);
 608         return inode;
 609 }
 610
 611 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 612                                 __u64 cno)
 613 {
 614         struct nilfs_iget_args args = {
 615                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 616         };
 617         struct inode *inode;
 618         int err;
 619
 620         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 621         if (unlikely(!inode))
 622                 return ERR_PTR(-ENOMEM);
 623         if (!(inode->i_state & I_NEW))
 624                 return inode;
 625
 626         err = nilfs_init_gcinode(inode);
 627         if (unlikely(err)) {
 628                 iget_failed(inode);
 629                 return ERR_PTR(err);
 630         }
 631         unlock_new_inode(inode);
 632         return inode;
 633 }
 634
 635 void nilfs_write_inode_common(struct inode *inode,
 636                               struct nilfs_inode *raw_inode, int has_bmap)
 637 {
 638         struct nilfs_inode_info *ii = NILFS_I(inode);
 639
 640         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 641         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 642         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 643         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 644         raw_inode->i_size = cpu_to_le64(inode->i_size);
 645         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 646         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 647         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 648         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 649         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 650
 651         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 652         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 653
 654         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 655                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 656
 657                 /* zero-fill unused portion in the case of super root block */
 658                 raw_inode->i_xattr = 0;
 659                 raw_inode->i_pad = 0;
 660                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 661                        nilfs->ns_inode_size - sizeof(*raw_inode));
 662         }
 663
 664         if (has_bmap)
 665                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 666         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 667                 raw_inode->i_device_code =
 668                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 669         /* When extending inode, nilfs->ns_inode_size should be checked
 670            for substitutions of appended fields */
 671 }
 672
 673 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 674 {
 675         ino_t ino = inode->i_ino;
 676         struct nilfs_inode_info *ii = NILFS_I(inode);
 677         struct inode *ifile = ii->i_root->ifile;
 678         struct nilfs_inode *raw_inode;
 679
 680         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 681
 682         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 683                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 684         if (flags & I_DIRTY_DATASYNC)
 685                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 686
 687         nilfs_write_inode_common(inode, raw_inode, 0);
 688                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 689                    deadlock of bmap. This delays update of i_bmap to just
 690                    before writing */
 691         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 692 }
 693
 694 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 695
 696 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 697                                 unsigned long from)
 698 {
 699         __u64 b;
 700         int ret;
 701
 702         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 703                 return;
 704 repeat:
 705         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 706         if (ret == -ENOENT)
 707                 return;
 708         else if (ret < 0)
 709                 goto failed;
 710
 711         if (b < from)
 712                 return;
 713
 714         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 715         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 716         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 717         if (!ret || (ret == -ENOMEM &&
 718                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 719                 goto repeat;
 720
 721 failed:
 722         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 723                       "failed to truncate bmap (ino=%lu, err=%d)",
 724                       ii->vfs_inode.i_ino, ret);
 725 }
 726
 727 void nilfs_truncate(struct inode *inode)
 728 {
 729         unsigned long blkoff;
 730         unsigned int blocksize;
 731         struct nilfs_transaction_info ti;
 732         struct super_block *sb = inode->i_sb;
 733         struct nilfs_inode_info *ii = NILFS_I(inode);
 734
 735         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 736                 return;
 737         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 738                 return;
 739
 740         blocksize = sb->s_blocksize;
 741         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 742         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 743
 744         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 745
 746         nilfs_truncate_bmap(ii, blkoff);
 747
 748         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 749         if (IS_SYNC(inode))
 750                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 751
 752         nilfs_mark_inode_dirty(inode);
 753         nilfs_set_file_dirty(inode, 0);
 754         nilfs_transaction_commit(sb);
 755         /* May construct a logical segment and may fail in sync mode.
 756            But truncate has no return value. */
 757 }
 758
 759 static void nilfs_clear_inode(struct inode *inode)
 760 {
 761         struct nilfs_inode_info *ii = NILFS_I(inode);
 762         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 763
 764         /*
 765          * Free resources allocated in nilfs_read_inode(), here.
 766          */
 767         BUG_ON(!list_empty(&ii->i_dirty));
 768         brelse(ii->i_bh);
 769         ii->i_bh = NULL;
 770
 771         if (mdi && mdi->mi_palloc_cache)
 772                 nilfs_palloc_destroy_cache(inode);
 773
 774         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 775                 nilfs_bmap_clear(ii->i_bmap);
 776
 777         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 778
 779         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 780                 nilfs_put_root(ii->i_root);
 781 }
 782
 783 void nilfs_evict_inode(struct inode *inode)
 784 {
 785         struct nilfs_transaction_info ti;
 786         struct super_block *sb = inode->i_sb;
 787         struct nilfs_inode_info *ii = NILFS_I(inode);
 788         int ret;
 789
 790         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 791                 truncate_inode_pages_final(&inode->i_data);
 792                 clear_inode(inode);
 793                 nilfs_clear_inode(inode);
 794                 return;
 795         }
 796         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 797
 798         truncate_inode_pages_final(&inode->i_data);
 799
 800         /* TODO: some of the following operations may fail.  */
 801         nilfs_truncate_bmap(ii, 0);
 802         nilfs_mark_inode_dirty(inode);
 803         clear_inode(inode);
 804
 805         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 806         if (!ret)
 807                 atomic64_dec(&ii->i_root->inodes_count);
 808
 809         nilfs_clear_inode(inode);
 810
 811         if (IS_SYNC(inode))
 812                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 813         nilfs_transaction_commit(sb);
 814         /* May construct a logical segment and may fail in sync mode.
 815            But delete_inode has no return value. */
 816 }
 817
 818 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 819 {
 820         struct nilfs_transaction_info ti;
 821         struct inode *inode = d_inode(dentry);
 822         struct super_block *sb = inode->i_sb;
 823         int err;
 824
 825         err = inode_change_ok(inode, iattr);
 826         if (err)
 827                 return err;
 828
 829         err = nilfs_transaction_begin(sb, &ti, 0);
 830         if (unlikely(err))
 831                 return err;
 832
 833         if ((iattr->ia_valid & ATTR_SIZE) &&
 834             iattr->ia_size != i_size_read(inode)) {
 835                 inode_dio_wait(inode);
 836                 truncate_setsize(inode, iattr->ia_size);
 837                 nilfs_truncate(inode);
 838         }
 839
 840         setattr_copy(inode, iattr);
 841         mark_inode_dirty(inode);
 842
 843         if (iattr->ia_valid & ATTR_MODE) {
 844                 err = nilfs_acl_chmod(inode);
 845                 if (unlikely(err))
 846                         goto out_err;
 847         }
 848
 849         return nilfs_transaction_commit(sb);
 850
 851 out_err:
 852         nilfs_transaction_abort(sb);
 853         return err;
 854 }
 855
 856 int nilfs_permission(struct inode *inode, int mask)
 857 {
 858         struct nilfs_root *root = NILFS_I(inode)->i_root;
 859         if ((mask & MAY_WRITE) && root &&
 860             root->cno != NILFS_CPTREE_CURRENT_CNO)
 861                 return -EROFS; /* snapshot is not writable */
 862
 863         return generic_permission(inode, mask);
 864 }
 865
 866 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 867 {
 868         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 869         struct nilfs_inode_info *ii = NILFS_I(inode);
 870         int err;
 871
 872         spin_lock(&nilfs->ns_inode_lock);
 873         if (ii->i_bh == NULL) {
 874                 spin_unlock(&nilfs->ns_inode_lock);
 875                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 876                                                   inode->i_ino, pbh);
 877                 if (unlikely(err))
 878                         return err;
 879                 spin_lock(&nilfs->ns_inode_lock);
 880                 if (ii->i_bh == NULL)
 881                         ii->i_bh = *pbh;
 882                 else {
 883                         brelse(*pbh);
 884                         *pbh = ii->i_bh;
 885                 }
 886         } else
 887                 *pbh = ii->i_bh;
 888
 889         get_bh(*pbh);
 890         spin_unlock(&nilfs->ns_inode_lock);
 891         return 0;
 892 }
 893
 894 int nilfs_inode_dirty(struct inode *inode)
 895 {
 896         struct nilfs_inode_info *ii = NILFS_I(inode);
 897         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 898         int ret = 0;
 899
 900         if (!list_empty(&ii->i_dirty)) {
 901                 spin_lock(&nilfs->ns_inode_lock);
 902                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 903                         test_bit(NILFS_I_BUSY, &ii->i_state);
 904                 spin_unlock(&nilfs->ns_inode_lock);
 905         }
 906         return ret;
 907 }
 908
 909 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 910 {
 911         struct nilfs_inode_info *ii = NILFS_I(inode);
 912         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 913
 914         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 915
 916         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 917                 return 0;
 918
 919         spin_lock(&nilfs->ns_inode_lock);
 920         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 921             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 922                 /* Because this routine may race with nilfs_dispose_list(),
 923                    we have to check NILFS_I_QUEUED here, too. */
 924                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 925                         /* This will happen when somebody is freeing
 926                            this inode. */
 927                         nilfs_warning(inode->i_sb, __func__,
 928                                       "cannot get inode (ino=%lu)\n",
 929                                       inode->i_ino);
 930                         spin_unlock(&nilfs->ns_inode_lock);
 931                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 932                                            freeing inode */
 933                 }
 934                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 935                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 936         }
 937         spin_unlock(&nilfs->ns_inode_lock);
 938         return 0;
 939 }
 940
 941 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
 942 {
 943         struct buffer_head *ibh;
 944         int err;
 945
 946         err = nilfs_load_inode_block(inode, &ibh);
 947         if (unlikely(err)) {
 948                 nilfs_warning(inode->i_sb, __func__,
 949                               "failed to reget inode block.\n");
 950                 return err;
 951         }
 952         nilfs_update_inode(inode, ibh, flags);
 953         mark_buffer_dirty(ibh);
 954         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 955         brelse(ibh);
 956         return 0;
 957 }
 958
 959 /**
 960  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 961  * @inode: inode of the file to be registered.
 962  *
 963  * nilfs_dirty_inode() loads a inode block containing the specified
 964  * @inode and copies data from a nilfs_inode to a corresponding inode
 965  * entry in the inode block. This operation is excluded from the segment
 966  * construction. This function can be called both as a single operation
 967  * and as a part of indivisible file operations.
 968  */
 969 void nilfs_dirty_inode(struct inode *inode, int flags)
 970 {
 971         struct nilfs_transaction_info ti;
 972         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 973
 974         if (is_bad_inode(inode)) {
 975                 nilfs_warning(inode->i_sb, __func__,
 976                               "tried to mark bad_inode dirty. ignored.\n");
 977                 dump_stack();
 978                 return;
 979         }
 980         if (mdi) {
 981                 nilfs_mdt_mark_dirty(inode);
 982                 return;
 983         }
 984         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 985         __nilfs_mark_inode_dirty(inode, flags);
 986         nilfs_transaction_commit(inode->i_sb); /* never fails */
 987 }
 988
 989 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 990                  __u64 start, __u64 len)
 991 {
 992         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 993         __u64 logical = 0, phys = 0, size = 0;
 994         __u32 flags = 0;
 995         loff_t isize;
 996         sector_t blkoff, end_blkoff;
 997         sector_t delalloc_blkoff;
 998         unsigned long delalloc_blklen;
 999         unsigned int blkbits = inode->i_blkbits;
1000         int ret, n;
1001
1002         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1003         if (ret)
1004                 return ret;
1005
1006         inode_lock(inode);
1007
1008         isize = i_size_read(inode);
1009
1010         blkoff = start >> blkbits;
1011         end_blkoff = (start + len - 1) >> blkbits;
1012
1013         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1014                                                         &delalloc_blkoff);
1015
1016         do {
1017                 __u64 blkphy;
1018                 unsigned int maxblocks;
1019
1020                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1021                         if (size) {
1022                                 /* End of the current extent */
1023                                 ret = fiemap_fill_next_extent(
1024                                         fieinfo, logical, phys, size, flags);
1025                                 if (ret)
1026                                         break;
1027                         }
1028                         if (blkoff > end_blkoff)
1029                                 break;
1030
1031                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1032                         logical = blkoff << blkbits;
1033                         phys = 0;
1034                         size = delalloc_blklen << blkbits;
1035
1036                         blkoff = delalloc_blkoff + delalloc_blklen;
1037                         delalloc_blklen = nilfs_find_uncommitted_extent(
1038                                 inode, blkoff, &delalloc_blkoff);
1039                         continue;
1040                 }
1041
1042                 /*
1043                  * Limit the number of blocks that we look up so as
1044                  * not to get into the next delayed allocation extent.
1045                  */
1046                 maxblocks = INT_MAX;
1047                 if (delalloc_blklen)
1048                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1049                                           maxblocks);
1050                 blkphy = 0;
1051
1052                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1053                 n = nilfs_bmap_lookup_contig(
1054                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1055                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1056
1057                 if (n < 0) {
1058                         int past_eof;
1059
1060                         if (unlikely(n != -ENOENT))
1061                                 break; /* error */
1062
1063                         /* HOLE */
1064                         blkoff++;
1065                         past_eof = ((blkoff << blkbits) >= isize);
1066
1067                         if (size) {
1068                                 /* End of the current extent */
1069
1070                                 if (past_eof)
1071                                         flags |= FIEMAP_EXTENT_LAST;
1072
1073                                 ret = fiemap_fill_next_extent(
1074                                         fieinfo, logical, phys, size, flags);
1075                                 if (ret)
1076                                         break;
1077                                 size = 0;
1078                         }
1079                         if (blkoff > end_blkoff || past_eof)
1080                                 break;
1081                 } else {
1082                         if (size) {
1083                                 if (phys && blkphy << blkbits == phys + size) {
1084                                         /* The current extent goes on */
1085                                         size += n << blkbits;
1086                                 } else {
1087                                         /* Terminate the current extent */
1088                                         ret = fiemap_fill_next_extent(
1089                                                 fieinfo, logical, phys, size,
1090                                                 flags);
1091                                         if (ret || blkoff > end_blkoff)
1092                                                 break;
1093
1094                                         /* Start another extent */
1095                                         flags = FIEMAP_EXTENT_MERGED;
1096                                         logical = blkoff << blkbits;
1097                                         phys = blkphy << blkbits;
1098                                         size = n << blkbits;
1099                                 }
1100                         } else {
1101                                 /* Start a new extent */
1102                                 flags = FIEMAP_EXTENT_MERGED;
1103                                 logical = blkoff << blkbits;
1104                                 phys = blkphy << blkbits;
1105                                 size = n << blkbits;
1106                         }
1107                         blkoff += n;
1108                 }
1109                 cond_resched();
1110         } while (true);
1111
1112         /* If ret is 1 then we just hit the end of the extent array */
1113         if (ret == 1)
1114                 ret = 0;
1115
1116         inode_unlock(inode);
1117         return ret;
1118 }