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  * Written by Ryusuke Konishi.
  17  *
  18  */
  19
  20 #include <linux/buffer_head.h>
  21 #include <linux/gfp.h>
  22 #include <linux/mpage.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/writeback.h>
  25 #include <linux/uio.h>
  26 #include "nilfs.h"
  27 #include "btnode.h"
  28 #include "segment.h"
  29 #include "page.h"
  30 #include "mdt.h"
  31 #include "cpfile.h"
  32 #include "ifile.h"
  33
  34 /**
  35  * struct nilfs_iget_args - arguments used during comparison between inodes
  36  * @ino: inode number
  37  * @cno: checkpoint number
  38  * @root: pointer on NILFS root object (mounted checkpoint)
  39  * @for_gc: inode for GC flag
  40  */
  41 struct nilfs_iget_args {
  42         u64 ino;
  43         __u64 cno;
  44         struct nilfs_root *root;
  45         int for_gc;
  46 };
  47
  48 static int nilfs_iget_test(struct inode *inode, void *opaque);
  49
  50 void nilfs_inode_add_blocks(struct inode *inode, int n)
  51 {
  52         struct nilfs_root *root = NILFS_I(inode)->i_root;
  53
  54         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  55         if (root)
  56                 atomic64_add(n, &root->blocks_count);
  57 }
  58
  59 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  60 {
  61         struct nilfs_root *root = NILFS_I(inode)->i_root;
  62
  63         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  64         if (root)
  65                 atomic64_sub(n, &root->blocks_count);
  66 }
  67
  68 /**
  69  * nilfs_get_block() - get a file block on the filesystem (callback function)
  70  * @inode - inode struct of the target file
  71  * @blkoff - file block number
  72  * @bh_result - buffer head to be mapped on
  73  * @create - indicate whether allocating the block or not when it has not
  74  *      been allocated yet.
  75  *
  76  * This function does not issue actual read request of the specified data
  77  * block. It is done by VFS.
  78  */
  79 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  80                     struct buffer_head *bh_result, int create)
  81 {
  82         struct nilfs_inode_info *ii = NILFS_I(inode);
  83         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  84         __u64 blknum = 0;
  85         int err = 0, ret;
  86         unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
  87
  88         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  89         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  90         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  91         if (ret >= 0) { /* found */
  92                 map_bh(bh_result, inode->i_sb, blknum);
  93                 if (ret > 0)
  94                         bh_result->b_size = (ret << inode->i_blkbits);
  95                 goto out;
  96         }
  97         /* data block was not found */
  98         if (ret == -ENOENT && create) {
  99                 struct nilfs_transaction_info ti;
 100
 101                 bh_result->b_blocknr = 0;
 102                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 103                 if (unlikely(err))
 104                         goto out;
 105                 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 106                                         (unsigned long)bh_result);
 107                 if (unlikely(err != 0)) {
 108                         if (err == -EEXIST) {
 109                                 /*
 110                                  * The get_block() function could be called
 111                                  * from multiple callers for an inode.
 112                                  * However, the page having this block must
 113                                  * be locked in this case.
 114                                  */
 115                                 nilfs_msg(inode->i_sb, KERN_WARNING,
 116                                           "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
 117                                           __func__, inode->i_ino,
 118                                           (unsigned long long)blkoff);
 119                                 err = 0;
 120                         }
 121                         nilfs_transaction_abort(inode->i_sb);
 122                         goto out;
 123                 }
 124                 nilfs_mark_inode_dirty_sync(inode);
 125                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 126                 /* Error handling should be detailed */
 127                 set_buffer_new(bh_result);
 128                 set_buffer_delay(bh_result);
 129                 map_bh(bh_result, inode->i_sb, 0);
 130                 /* Disk block number must be changed to proper value */
 131
 132         } else if (ret == -ENOENT) {
 133                 /*
 134                  * not found is not error (e.g. hole); must return without
 135                  * the mapped state flag.
 136                  */
 137                 ;
 138         } else {
 139                 err = ret;
 140         }
 141
 142  out:
 143         return err;
 144 }
 145
 146 /**
 147  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 148  * address_space_operations.
 149  * @file - file struct of the file to be read
 150  * @page - the page to be read
 151  */
 152 static int nilfs_readpage(struct file *file, struct page *page)
 153 {
 154         return mpage_readpage(page, nilfs_get_block);
 155 }
 156
 157 /**
 158  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 159  * address_space_operations.
 160  * @file - file struct of the file to be read
 161  * @mapping - address_space struct used for reading multiple pages
 162  * @pages - the pages to be read
 163  * @nr_pages - number of pages to be read
 164  */
 165 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 166                            struct list_head *pages, unsigned int nr_pages)
 167 {
 168         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 169 }
 170
 171 static int nilfs_writepages(struct address_space *mapping,
 172                             struct writeback_control *wbc)
 173 {
 174         struct inode *inode = mapping->host;
 175         int err = 0;
 176
 177         if (inode->i_sb->s_flags & MS_RDONLY) {
 178                 nilfs_clear_dirty_pages(mapping, false);
 179                 return -EROFS;
 180         }
 181
 182         if (wbc->sync_mode == WB_SYNC_ALL)
 183                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 184                                                     wbc->range_start,
 185                                                     wbc->range_end);
 186         return err;
 187 }
 188
 189 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 190 {
 191         struct inode *inode = page->mapping->host;
 192         int err;
 193
 194         if (inode->i_sb->s_flags & MS_RDONLY) {
 195                 /*
 196                  * It means that filesystem was remounted in read-only
 197                  * mode because of error or metadata corruption. But we
 198                  * have dirty pages that try to be flushed in background.
 199                  * So, here we simply discard this dirty page.
 200                  */
 201                 nilfs_clear_dirty_page(page, false);
 202                 unlock_page(page);
 203                 return -EROFS;
 204         }
 205
 206         redirty_page_for_writepage(wbc, page);
 207         unlock_page(page);
 208
 209         if (wbc->sync_mode == WB_SYNC_ALL) {
 210                 err = nilfs_construct_segment(inode->i_sb);
 211                 if (unlikely(err))
 212                         return err;
 213         } else if (wbc->for_reclaim)
 214                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 215
 216         return 0;
 217 }
 218
 219 static int nilfs_set_page_dirty(struct page *page)
 220 {
 221         struct inode *inode = page->mapping->host;
 222         int ret = __set_page_dirty_nobuffers(page);
 223
 224         if (page_has_buffers(page)) {
 225                 unsigned int nr_dirty = 0;
 226                 struct buffer_head *bh, *head;
 227
 228                 /*
 229                  * This page is locked by callers, and no other thread
 230                  * concurrently marks its buffers dirty since they are
 231                  * only dirtied through routines in fs/buffer.c in
 232                  * which call sites of mark_buffer_dirty are protected
 233                  * by page lock.
 234                  */
 235                 bh = head = page_buffers(page);
 236                 do {
 237                         /* Do not mark hole blocks dirty */
 238                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 239                                 continue;
 240
 241                         set_buffer_dirty(bh);
 242                         nr_dirty++;
 243                 } while (bh = bh->b_this_page, bh != head);
 244
 245                 if (nr_dirty)
 246                         nilfs_set_file_dirty(inode, nr_dirty);
 247         } else if (ret) {
 248                 unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 249
 250                 nilfs_set_file_dirty(inode, nr_dirty);
 251         }
 252         return ret;
 253 }
 254
 255 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 256 {
 257         struct inode *inode = mapping->host;
 258
 259         if (to > inode->i_size) {
 260                 truncate_pagecache(inode, inode->i_size);
 261                 nilfs_truncate(inode);
 262         }
 263 }
 264
 265 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 266                              loff_t pos, unsigned len, unsigned flags,
 267                              struct page **pagep, void **fsdata)
 268
 269 {
 270         struct inode *inode = mapping->host;
 271         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 272
 273         if (unlikely(err))
 274                 return err;
 275
 276         err = block_write_begin(mapping, pos, len, flags, pagep,
 277                                 nilfs_get_block);
 278         if (unlikely(err)) {
 279                 nilfs_write_failed(mapping, pos + len);
 280                 nilfs_transaction_abort(inode->i_sb);
 281         }
 282         return err;
 283 }
 284
 285 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 286                            loff_t pos, unsigned len, unsigned copied,
 287                            struct page *page, void *fsdata)
 288 {
 289         struct inode *inode = mapping->host;
 290         unsigned int start = pos & (PAGE_SIZE - 1);
 291         unsigned int nr_dirty;
 292         int err;
 293
 294         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 295                                                   start + copied);
 296         copied = generic_write_end(file, mapping, pos, len, copied, page,
 297                                    fsdata);
 298         nilfs_set_file_dirty(inode, nr_dirty);
 299         err = nilfs_transaction_commit(inode->i_sb);
 300         return err ? : copied;
 301 }
 302
 303 static ssize_t
 304 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 305 {
 306         struct inode *inode = file_inode(iocb->ki_filp);
 307
 308         if (iov_iter_rw(iter) == WRITE)
 309                 return 0;
 310
 311         /* Needs synchronization with the cleaner */
 312         return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 313 }
 314
 315 const struct address_space_operations nilfs_aops = {
 316         .writepage              = nilfs_writepage,
 317         .readpage               = nilfs_readpage,
 318         .writepages             = nilfs_writepages,
 319         .set_page_dirty         = nilfs_set_page_dirty,
 320         .readpages              = nilfs_readpages,
 321         .write_begin            = nilfs_write_begin,
 322         .write_end              = nilfs_write_end,
 323         /* .releasepage         = nilfs_releasepage, */
 324         .invalidatepage         = block_invalidatepage,
 325         .direct_IO              = nilfs_direct_IO,
 326         .is_partially_uptodate  = block_is_partially_uptodate,
 327 };
 328
 329 static int nilfs_insert_inode_locked(struct inode *inode,
 330                                      struct nilfs_root *root,
 331                                      unsigned long ino)
 332 {
 333         struct nilfs_iget_args args = {
 334                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 335         };
 336
 337         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 338 }
 339
 340 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 341 {
 342         struct super_block *sb = dir->i_sb;
 343         struct the_nilfs *nilfs = sb->s_fs_info;
 344         struct inode *inode;
 345         struct nilfs_inode_info *ii;
 346         struct nilfs_root *root;
 347         int err = -ENOMEM;
 348         ino_t ino;
 349
 350         inode = new_inode(sb);
 351         if (unlikely(!inode))
 352                 goto failed;
 353
 354         mapping_set_gfp_mask(inode->i_mapping,
 355                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 356
 357         root = NILFS_I(dir)->i_root;
 358         ii = NILFS_I(inode);
 359         ii->i_state = BIT(NILFS_I_NEW);
 360         ii->i_root = root;
 361
 362         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 363         if (unlikely(err))
 364                 goto failed_ifile_create_inode;
 365         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 366
 367         atomic64_inc(&root->inodes_count);
 368         inode_init_owner(inode, dir, mode);
 369         inode->i_ino = ino;
 370         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 371
 372         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 373                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 374                 if (err < 0)
 375                         goto failed_after_creation;
 376
 377                 set_bit(NILFS_I_BMAP, &ii->i_state);
 378                 /* No lock is needed; iget() ensures it. */
 379         }
 380
 381         ii->i_flags = nilfs_mask_flags(
 382                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 383
 384         /* ii->i_file_acl = 0; */
 385         /* ii->i_dir_acl = 0; */
 386         ii->i_dir_start_lookup = 0;
 387         nilfs_set_inode_flags(inode);
 388         spin_lock(&nilfs->ns_next_gen_lock);
 389         inode->i_generation = nilfs->ns_next_generation++;
 390         spin_unlock(&nilfs->ns_next_gen_lock);
 391         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 392                 err = -EIO;
 393                 goto failed_after_creation;
 394         }
 395
 396         err = nilfs_init_acl(inode, dir);
 397         if (unlikely(err))
 398                 /*
 399                  * Never occur.  When supporting nilfs_init_acl(),
 400                  * proper cancellation of above jobs should be considered.
 401                  */
 402                 goto failed_after_creation;
 403
 404         return inode;
 405
 406  failed_after_creation:
 407         clear_nlink(inode);
 408         unlock_new_inode(inode);
 409         iput(inode);  /*
 410                        * raw_inode will be deleted through
 411                        * nilfs_evict_inode().
 412                        */
 413         goto failed;
 414
 415  failed_ifile_create_inode:
 416         make_bad_inode(inode);
 417         iput(inode);
 418  failed:
 419         return ERR_PTR(err);
 420 }
 421
 422 void nilfs_set_inode_flags(struct inode *inode)
 423 {
 424         unsigned int flags = NILFS_I(inode)->i_flags;
 425         unsigned int new_fl = 0;
 426
 427         if (flags & FS_SYNC_FL)
 428                 new_fl |= S_SYNC;
 429         if (flags & FS_APPEND_FL)
 430                 new_fl |= S_APPEND;
 431         if (flags & FS_IMMUTABLE_FL)
 432                 new_fl |= S_IMMUTABLE;
 433         if (flags & FS_NOATIME_FL)
 434                 new_fl |= S_NOATIME;
 435         if (flags & FS_DIRSYNC_FL)
 436                 new_fl |= S_DIRSYNC;
 437         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 438                         S_NOATIME | S_DIRSYNC);
 439 }
 440
 441 int nilfs_read_inode_common(struct inode *inode,
 442                             struct nilfs_inode *raw_inode)
 443 {
 444         struct nilfs_inode_info *ii = NILFS_I(inode);
 445         int err;
 446
 447         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 448         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 449         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 450         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 451         inode->i_size = le64_to_cpu(raw_inode->i_size);
 452         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 453         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 454         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 455         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 456         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 457         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 458         if (inode->i_nlink == 0)
 459                 return -ESTALE; /* this inode is deleted */
 460
 461         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 462         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 463 #if 0
 464         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 465         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 466                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 467 #endif
 468         ii->i_dir_start_lookup = 0;
 469         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 470
 471         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 472             S_ISLNK(inode->i_mode)) {
 473                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 474                 if (err < 0)
 475                         return err;
 476                 set_bit(NILFS_I_BMAP, &ii->i_state);
 477                 /* No lock is needed; iget() ensures it. */
 478         }
 479         return 0;
 480 }
 481
 482 static int __nilfs_read_inode(struct super_block *sb,
 483                               struct nilfs_root *root, unsigned long ino,
 484                               struct inode *inode)
 485 {
 486         struct the_nilfs *nilfs = sb->s_fs_info;
 487         struct buffer_head *bh;
 488         struct nilfs_inode *raw_inode;
 489         int err;
 490
 491         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 492         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 493         if (unlikely(err))
 494                 goto bad_inode;
 495
 496         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 497
 498         err = nilfs_read_inode_common(inode, raw_inode);
 499         if (err)
 500                 goto failed_unmap;
 501
 502         if (S_ISREG(inode->i_mode)) {
 503                 inode->i_op = &nilfs_file_inode_operations;
 504                 inode->i_fop = &nilfs_file_operations;
 505                 inode->i_mapping->a_ops = &nilfs_aops;
 506         } else if (S_ISDIR(inode->i_mode)) {
 507                 inode->i_op = &nilfs_dir_inode_operations;
 508                 inode->i_fop = &nilfs_dir_operations;
 509                 inode->i_mapping->a_ops = &nilfs_aops;
 510         } else if (S_ISLNK(inode->i_mode)) {
 511                 inode->i_op = &nilfs_symlink_inode_operations;
 512                 inode_nohighmem(inode);
 513                 inode->i_mapping->a_ops = &nilfs_aops;
 514         } else {
 515                 inode->i_op = &nilfs_special_inode_operations;
 516                 init_special_inode(
 517                         inode, inode->i_mode,
 518                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 519         }
 520         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 521         brelse(bh);
 522         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 523         nilfs_set_inode_flags(inode);
 524         mapping_set_gfp_mask(inode->i_mapping,
 525                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 526         return 0;
 527
 528  failed_unmap:
 529         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 530         brelse(bh);
 531
 532  bad_inode:
 533         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 534         return err;
 535 }
 536
 537 static int nilfs_iget_test(struct inode *inode, void *opaque)
 538 {
 539         struct nilfs_iget_args *args = opaque;
 540         struct nilfs_inode_info *ii;
 541
 542         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 543                 return 0;
 544
 545         ii = NILFS_I(inode);
 546         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 547                 return !args->for_gc;
 548
 549         return args->for_gc && args->cno == ii->i_cno;
 550 }
 551
 552 static int nilfs_iget_set(struct inode *inode, void *opaque)
 553 {
 554         struct nilfs_iget_args *args = opaque;
 555
 556         inode->i_ino = args->ino;
 557         if (args->for_gc) {
 558                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 559                 NILFS_I(inode)->i_cno = args->cno;
 560                 NILFS_I(inode)->i_root = NULL;
 561         } else {
 562                 if (args->root && args->ino == NILFS_ROOT_INO)
 563                         nilfs_get_root(args->root);
 564                 NILFS_I(inode)->i_root = args->root;
 565         }
 566         return 0;
 567 }
 568
 569 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 570                             unsigned long ino)
 571 {
 572         struct nilfs_iget_args args = {
 573                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 574         };
 575
 576         return ilookup5(sb, ino, nilfs_iget_test, &args);
 577 }
 578
 579 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 580                                 unsigned long ino)
 581 {
 582         struct nilfs_iget_args args = {
 583                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 584         };
 585
 586         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 587 }
 588
 589 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 590                          unsigned long ino)
 591 {
 592         struct inode *inode;
 593         int err;
 594
 595         inode = nilfs_iget_locked(sb, root, ino);
 596         if (unlikely(!inode))
 597                 return ERR_PTR(-ENOMEM);
 598         if (!(inode->i_state & I_NEW))
 599                 return inode;
 600
 601         err = __nilfs_read_inode(sb, root, ino, inode);
 602         if (unlikely(err)) {
 603                 iget_failed(inode);
 604                 return ERR_PTR(err);
 605         }
 606         unlock_new_inode(inode);
 607         return inode;
 608 }
 609
 610 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 611                                 __u64 cno)
 612 {
 613         struct nilfs_iget_args args = {
 614                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 615         };
 616         struct inode *inode;
 617         int err;
 618
 619         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 620         if (unlikely(!inode))
 621                 return ERR_PTR(-ENOMEM);
 622         if (!(inode->i_state & I_NEW))
 623                 return inode;
 624
 625         err = nilfs_init_gcinode(inode);
 626         if (unlikely(err)) {
 627                 iget_failed(inode);
 628                 return ERR_PTR(err);
 629         }
 630         unlock_new_inode(inode);
 631         return inode;
 632 }
 633
 634 void nilfs_write_inode_common(struct inode *inode,
 635                               struct nilfs_inode *raw_inode, int has_bmap)
 636 {
 637         struct nilfs_inode_info *ii = NILFS_I(inode);
 638
 639         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 640         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 641         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 642         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 643         raw_inode->i_size = cpu_to_le64(inode->i_size);
 644         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 645         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 646         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 647         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 648         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 649
 650         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 651         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 652
 653         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 654                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 655
 656                 /* zero-fill unused portion in the case of super root block */
 657                 raw_inode->i_xattr = 0;
 658                 raw_inode->i_pad = 0;
 659                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 660                        nilfs->ns_inode_size - sizeof(*raw_inode));
 661         }
 662
 663         if (has_bmap)
 664                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 665         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 666                 raw_inode->i_device_code =
 667                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 668         /*
 669          * When extending inode, nilfs->ns_inode_size should be checked
 670          * for substitutions of appended fields.
 671          */
 672 }
 673
 674 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 675 {
 676         ino_t ino = inode->i_ino;
 677         struct nilfs_inode_info *ii = NILFS_I(inode);
 678         struct inode *ifile = ii->i_root->ifile;
 679         struct nilfs_inode *raw_inode;
 680
 681         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 682
 683         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 684                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 685         if (flags & I_DIRTY_DATASYNC)
 686                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 687
 688         nilfs_write_inode_common(inode, raw_inode, 0);
 689                 /*
 690                  * XXX: call with has_bmap = 0 is a workaround to avoid
 691                  * deadlock of bmap.  This delays update of i_bmap to just
 692                  * before writing.
 693                  */
 694
 695         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 696 }
 697
 698 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 699
 700 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 701                                 unsigned long from)
 702 {
 703         __u64 b;
 704         int ret;
 705
 706         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 707                 return;
 708 repeat:
 709         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 710         if (ret == -ENOENT)
 711                 return;
 712         else if (ret < 0)
 713                 goto failed;
 714
 715         if (b < from)
 716                 return;
 717
 718         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 719         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 720         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 721         if (!ret || (ret == -ENOMEM &&
 722                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 723                 goto repeat;
 724
 725 failed:
 726         nilfs_msg(ii->vfs_inode.i_sb, KERN_WARNING,
 727                   "error %d truncating bmap (ino=%lu)", ret,
 728                   ii->vfs_inode.i_ino);
 729 }
 730
 731 void nilfs_truncate(struct inode *inode)
 732 {
 733         unsigned long blkoff;
 734         unsigned int blocksize;
 735         struct nilfs_transaction_info ti;
 736         struct super_block *sb = inode->i_sb;
 737         struct nilfs_inode_info *ii = NILFS_I(inode);
 738
 739         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 740                 return;
 741         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 742                 return;
 743
 744         blocksize = sb->s_blocksize;
 745         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 746         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 747
 748         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 749
 750         nilfs_truncate_bmap(ii, blkoff);
 751
 752         inode->i_mtime = inode->i_ctime = current_time(inode);
 753         if (IS_SYNC(inode))
 754                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 755
 756         nilfs_mark_inode_dirty(inode);
 757         nilfs_set_file_dirty(inode, 0);
 758         nilfs_transaction_commit(sb);
 759         /*
 760          * May construct a logical segment and may fail in sync mode.
 761          * But truncate has no return value.
 762          */
 763 }
 764
 765 static void nilfs_clear_inode(struct inode *inode)
 766 {
 767         struct nilfs_inode_info *ii = NILFS_I(inode);
 768
 769         /*
 770          * Free resources allocated in nilfs_read_inode(), here.
 771          */
 772         BUG_ON(!list_empty(&ii->i_dirty));
 773         brelse(ii->i_bh);
 774         ii->i_bh = NULL;
 775
 776         if (nilfs_is_metadata_file_inode(inode))
 777                 nilfs_mdt_clear(inode);
 778
 779         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 780                 nilfs_bmap_clear(ii->i_bmap);
 781
 782         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 783
 784         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 785                 nilfs_put_root(ii->i_root);
 786 }
 787
 788 void nilfs_evict_inode(struct inode *inode)
 789 {
 790         struct nilfs_transaction_info ti;
 791         struct super_block *sb = inode->i_sb;
 792         struct nilfs_inode_info *ii = NILFS_I(inode);
 793         int ret;
 794
 795         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 796                 truncate_inode_pages_final(&inode->i_data);
 797                 clear_inode(inode);
 798                 nilfs_clear_inode(inode);
 799                 return;
 800         }
 801         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 802
 803         truncate_inode_pages_final(&inode->i_data);
 804
 805         /* TODO: some of the following operations may fail.  */
 806         nilfs_truncate_bmap(ii, 0);
 807         nilfs_mark_inode_dirty(inode);
 808         clear_inode(inode);
 809
 810         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 811         if (!ret)
 812                 atomic64_dec(&ii->i_root->inodes_count);
 813
 814         nilfs_clear_inode(inode);
 815
 816         if (IS_SYNC(inode))
 817                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 818         nilfs_transaction_commit(sb);
 819         /*
 820          * May construct a logical segment and may fail in sync mode.
 821          * But delete_inode has no return value.
 822          */
 823 }
 824
 825 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 826 {
 827         struct nilfs_transaction_info ti;
 828         struct inode *inode = d_inode(dentry);
 829         struct super_block *sb = inode->i_sb;
 830         int err;
 831
 832         err = setattr_prepare(dentry, iattr);
 833         if (err)
 834                 return err;
 835
 836         err = nilfs_transaction_begin(sb, &ti, 0);
 837         if (unlikely(err))
 838                 return err;
 839
 840         if ((iattr->ia_valid & ATTR_SIZE) &&
 841             iattr->ia_size != i_size_read(inode)) {
 842                 inode_dio_wait(inode);
 843                 truncate_setsize(inode, iattr->ia_size);
 844                 nilfs_truncate(inode);
 845         }
 846
 847         setattr_copy(inode, iattr);
 848         mark_inode_dirty(inode);
 849
 850         if (iattr->ia_valid & ATTR_MODE) {
 851                 err = nilfs_acl_chmod(inode);
 852                 if (unlikely(err))
 853                         goto out_err;
 854         }
 855
 856         return nilfs_transaction_commit(sb);
 857
 858 out_err:
 859         nilfs_transaction_abort(sb);
 860         return err;
 861 }
 862
 863 int nilfs_permission(struct inode *inode, int mask)
 864 {
 865         struct nilfs_root *root = NILFS_I(inode)->i_root;
 866
 867         if ((mask & MAY_WRITE) && root &&
 868             root->cno != NILFS_CPTREE_CURRENT_CNO)
 869                 return -EROFS; /* snapshot is not writable */
 870
 871         return generic_permission(inode, mask);
 872 }
 873
 874 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 875 {
 876         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 877         struct nilfs_inode_info *ii = NILFS_I(inode);
 878         int err;
 879
 880         spin_lock(&nilfs->ns_inode_lock);
 881         if (ii->i_bh == NULL) {
 882                 spin_unlock(&nilfs->ns_inode_lock);
 883                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 884                                                   inode->i_ino, pbh);
 885                 if (unlikely(err))
 886                         return err;
 887                 spin_lock(&nilfs->ns_inode_lock);
 888                 if (ii->i_bh == NULL)
 889                         ii->i_bh = *pbh;
 890                 else {
 891                         brelse(*pbh);
 892                         *pbh = ii->i_bh;
 893                 }
 894         } else
 895                 *pbh = ii->i_bh;
 896
 897         get_bh(*pbh);
 898         spin_unlock(&nilfs->ns_inode_lock);
 899         return 0;
 900 }
 901
 902 int nilfs_inode_dirty(struct inode *inode)
 903 {
 904         struct nilfs_inode_info *ii = NILFS_I(inode);
 905         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 906         int ret = 0;
 907
 908         if (!list_empty(&ii->i_dirty)) {
 909                 spin_lock(&nilfs->ns_inode_lock);
 910                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 911                         test_bit(NILFS_I_BUSY, &ii->i_state);
 912                 spin_unlock(&nilfs->ns_inode_lock);
 913         }
 914         return ret;
 915 }
 916
 917 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
 918 {
 919         struct nilfs_inode_info *ii = NILFS_I(inode);
 920         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 921
 922         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 923
 924         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 925                 return 0;
 926
 927         spin_lock(&nilfs->ns_inode_lock);
 928         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 929             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 930                 /*
 931                  * Because this routine may race with nilfs_dispose_list(),
 932                  * we have to check NILFS_I_QUEUED here, too.
 933                  */
 934                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 935                         /*
 936                          * This will happen when somebody is freeing
 937                          * this inode.
 938                          */
 939                         nilfs_msg(inode->i_sb, KERN_WARNING,
 940                                   "cannot set file dirty (ino=%lu): the file is being freed",
 941                                   inode->i_ino);
 942                         spin_unlock(&nilfs->ns_inode_lock);
 943                         return -EINVAL; /*
 944                                          * NILFS_I_DIRTY may remain for
 945                                          * freeing inode.
 946                                          */
 947                 }
 948                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 949                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 950         }
 951         spin_unlock(&nilfs->ns_inode_lock);
 952         return 0;
 953 }
 954
 955 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
 956 {
 957         struct buffer_head *ibh;
 958         int err;
 959
 960         err = nilfs_load_inode_block(inode, &ibh);
 961         if (unlikely(err)) {
 962                 nilfs_msg(inode->i_sb, KERN_WARNING,
 963                           "cannot mark inode dirty (ino=%lu): error %d loading inode block",
 964                           inode->i_ino, err);
 965                 return err;
 966         }
 967         nilfs_update_inode(inode, ibh, flags);
 968         mark_buffer_dirty(ibh);
 969         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 970         brelse(ibh);
 971         return 0;
 972 }
 973
 974 /**
 975  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 976  * @inode: inode of the file to be registered.
 977  *
 978  * nilfs_dirty_inode() loads a inode block containing the specified
 979  * @inode and copies data from a nilfs_inode to a corresponding inode
 980  * entry in the inode block. This operation is excluded from the segment
 981  * construction. This function can be called both as a single operation
 982  * and as a part of indivisible file operations.
 983  */
 984 void nilfs_dirty_inode(struct inode *inode, int flags)
 985 {
 986         struct nilfs_transaction_info ti;
 987         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 988
 989         if (is_bad_inode(inode)) {
 990                 nilfs_msg(inode->i_sb, KERN_WARNING,
 991                           "tried to mark bad_inode dirty. ignored.");
 992                 dump_stack();
 993                 return;
 994         }
 995         if (mdi) {
 996                 nilfs_mdt_mark_dirty(inode);
 997                 return;
 998         }
 999         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1000         __nilfs_mark_inode_dirty(inode, flags);
1001         nilfs_transaction_commit(inode->i_sb); /* never fails */
1002 }
1003
1004 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1005                  __u64 start, __u64 len)
1006 {
1007         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1008         __u64 logical = 0, phys = 0, size = 0;
1009         __u32 flags = 0;
1010         loff_t isize;
1011         sector_t blkoff, end_blkoff;
1012         sector_t delalloc_blkoff;
1013         unsigned long delalloc_blklen;
1014         unsigned int blkbits = inode->i_blkbits;
1015         int ret, n;
1016
1017         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1018         if (ret)
1019                 return ret;
1020
1021         inode_lock(inode);
1022
1023         isize = i_size_read(inode);
1024
1025         blkoff = start >> blkbits;
1026         end_blkoff = (start + len - 1) >> blkbits;
1027
1028         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1029                                                         &delalloc_blkoff);
1030
1031         do {
1032                 __u64 blkphy;
1033                 unsigned int maxblocks;
1034
1035                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1036                         if (size) {
1037                                 /* End of the current extent */
1038                                 ret = fiemap_fill_next_extent(
1039                                         fieinfo, logical, phys, size, flags);
1040                                 if (ret)
1041                                         break;
1042                         }
1043                         if (blkoff > end_blkoff)
1044                                 break;
1045
1046                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1047                         logical = blkoff << blkbits;
1048                         phys = 0;
1049                         size = delalloc_blklen << blkbits;
1050
1051                         blkoff = delalloc_blkoff + delalloc_blklen;
1052                         delalloc_blklen = nilfs_find_uncommitted_extent(
1053                                 inode, blkoff, &delalloc_blkoff);
1054                         continue;
1055                 }
1056
1057                 /*
1058                  * Limit the number of blocks that we look up so as
1059                  * not to get into the next delayed allocation extent.
1060                  */
1061                 maxblocks = INT_MAX;
1062                 if (delalloc_blklen)
1063                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1064                                           maxblocks);
1065                 blkphy = 0;
1066
1067                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1068                 n = nilfs_bmap_lookup_contig(
1069                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1070                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1071
1072                 if (n < 0) {
1073                         int past_eof;
1074
1075                         if (unlikely(n != -ENOENT))
1076                                 break; /* error */
1077
1078                         /* HOLE */
1079                         blkoff++;
1080                         past_eof = ((blkoff << blkbits) >= isize);
1081
1082                         if (size) {
1083                                 /* End of the current extent */
1084
1085                                 if (past_eof)
1086                                         flags |= FIEMAP_EXTENT_LAST;
1087
1088                                 ret = fiemap_fill_next_extent(
1089                                         fieinfo, logical, phys, size, flags);
1090                                 if (ret)
1091                                         break;
1092                                 size = 0;
1093                         }
1094                         if (blkoff > end_blkoff || past_eof)
1095                                 break;
1096                 } else {
1097                         if (size) {
1098                                 if (phys && blkphy << blkbits == phys + size) {
1099                                         /* The current extent goes on */
1100                                         size += n << blkbits;
1101                                 } else {
1102                                         /* Terminate the current extent */
1103                                         ret = fiemap_fill_next_extent(
1104                                                 fieinfo, logical, phys, size,
1105                                                 flags);
1106                                         if (ret || blkoff > end_blkoff)
1107                                                 break;
1108
1109                                         /* Start another extent */
1110                                         flags = FIEMAP_EXTENT_MERGED;
1111                                         logical = blkoff << blkbits;
1112                                         phys = blkphy << blkbits;
1113                                         size = n << blkbits;
1114                                 }
1115                         } else {
1116                                 /* Start a new extent */
1117                                 flags = FIEMAP_EXTENT_MERGED;
1118                                 logical = blkoff << blkbits;
1119                                 phys = blkphy << blkbits;
1120                                 size = n << blkbits;
1121                         }
1122                         blkoff += n;
1123                 }
1124                 cond_resched();
1125         } while (true);
1126
1127         /* If ret is 1 then we just hit the end of the extent array */
1128         if (ret == 1)
1129                 ret = 0;
1130
1131         inode_unlock(inode);
1132         return ret;
1133 }