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, i_blocksize(inode) * 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, i_blocksize(inode) * 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         struct buffer_head *bh;
 348         int err = -ENOMEM;
 349         ino_t ino;
 350
 351         inode = new_inode(sb);
 352         if (unlikely(!inode))
 353                 goto failed;
 354
 355         mapping_set_gfp_mask(inode->i_mapping,
 356                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 357
 358         root = NILFS_I(dir)->i_root;
 359         ii = NILFS_I(inode);
 360         ii->i_state = BIT(NILFS_I_NEW);
 361         ii->i_root = root;
 362
 363         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 364         if (unlikely(err))
 365                 goto failed_ifile_create_inode;
 366         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 367
 368         if (unlikely(ino < NILFS_USER_INO)) {
 369                 nilfs_msg(sb, KERN_WARNING,
 370                           "inode bitmap is inconsistent for reserved inodes");
 371                 do {
 372                         brelse(bh);
 373                         err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 374                         if (unlikely(err))
 375                                 goto failed_ifile_create_inode;
 376                 } while (ino < NILFS_USER_INO);
 377
 378                 nilfs_msg(sb, KERN_INFO,
 379                           "repaired inode bitmap for reserved inodes");
 380         }
 381         ii->i_bh = bh;
 382
 383         atomic64_inc(&root->inodes_count);
 384         inode_init_owner(inode, dir, mode);
 385         inode->i_ino = ino;
 386         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 387
 388         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 389                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 390                 if (err < 0)
 391                         goto failed_after_creation;
 392
 393                 set_bit(NILFS_I_BMAP, &ii->i_state);
 394                 /* No lock is needed; iget() ensures it. */
 395         }
 396
 397         ii->i_flags = nilfs_mask_flags(
 398                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 399
 400         /* ii->i_file_acl = 0; */
 401         /* ii->i_dir_acl = 0; */
 402         ii->i_dir_start_lookup = 0;
 403         nilfs_set_inode_flags(inode);
 404         spin_lock(&nilfs->ns_next_gen_lock);
 405         inode->i_generation = nilfs->ns_next_generation++;
 406         spin_unlock(&nilfs->ns_next_gen_lock);
 407         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 408                 err = -EIO;
 409                 goto failed_after_creation;
 410         }
 411
 412         err = nilfs_init_acl(inode, dir);
 413         if (unlikely(err))
 414                 /*
 415                  * Never occur.  When supporting nilfs_init_acl(),
 416                  * proper cancellation of above jobs should be considered.
 417                  */
 418                 goto failed_after_creation;
 419
 420         return inode;
 421
 422  failed_after_creation:
 423         clear_nlink(inode);
 424         unlock_new_inode(inode);
 425         iput(inode);  /*
 426                        * raw_inode will be deleted through
 427                        * nilfs_evict_inode().
 428                        */
 429         goto failed;
 430
 431  failed_ifile_create_inode:
 432         make_bad_inode(inode);
 433         iput(inode);
 434  failed:
 435         return ERR_PTR(err);
 436 }
 437
 438 void nilfs_set_inode_flags(struct inode *inode)
 439 {
 440         unsigned int flags = NILFS_I(inode)->i_flags;
 441         unsigned int new_fl = 0;
 442
 443         if (flags & FS_SYNC_FL)
 444                 new_fl |= S_SYNC;
 445         if (flags & FS_APPEND_FL)
 446                 new_fl |= S_APPEND;
 447         if (flags & FS_IMMUTABLE_FL)
 448                 new_fl |= S_IMMUTABLE;
 449         if (flags & FS_NOATIME_FL)
 450                 new_fl |= S_NOATIME;
 451         if (flags & FS_DIRSYNC_FL)
 452                 new_fl |= S_DIRSYNC;
 453         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 454                         S_NOATIME | S_DIRSYNC);
 455 }
 456
 457 int nilfs_read_inode_common(struct inode *inode,
 458                             struct nilfs_inode *raw_inode)
 459 {
 460         struct nilfs_inode_info *ii = NILFS_I(inode);
 461         int err;
 462
 463         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 464         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 465         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 466         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 467         inode->i_size = le64_to_cpu(raw_inode->i_size);
 468         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 469         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 470         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 471         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 472         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 473         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 474         if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
 475                 return -EIO; /* this inode is for metadata and corrupted */
 476         if (inode->i_nlink == 0)
 477                 return -ESTALE; /* this inode is deleted */
 478
 479         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 480         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 481 #if 0
 482         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 483         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 484                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 485 #endif
 486         ii->i_dir_start_lookup = 0;
 487         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 488
 489         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 490             S_ISLNK(inode->i_mode)) {
 491                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 492                 if (err < 0)
 493                         return err;
 494                 set_bit(NILFS_I_BMAP, &ii->i_state);
 495                 /* No lock is needed; iget() ensures it. */
 496         }
 497         return 0;
 498 }
 499
 500 static int __nilfs_read_inode(struct super_block *sb,
 501                               struct nilfs_root *root, unsigned long ino,
 502                               struct inode *inode)
 503 {
 504         struct the_nilfs *nilfs = sb->s_fs_info;
 505         struct buffer_head *bh;
 506         struct nilfs_inode *raw_inode;
 507         int err;
 508
 509         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 510         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 511         if (unlikely(err))
 512                 goto bad_inode;
 513
 514         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 515
 516         err = nilfs_read_inode_common(inode, raw_inode);
 517         if (err)
 518                 goto failed_unmap;
 519
 520         if (S_ISREG(inode->i_mode)) {
 521                 inode->i_op = &nilfs_file_inode_operations;
 522                 inode->i_fop = &nilfs_file_operations;
 523                 inode->i_mapping->a_ops = &nilfs_aops;
 524         } else if (S_ISDIR(inode->i_mode)) {
 525                 inode->i_op = &nilfs_dir_inode_operations;
 526                 inode->i_fop = &nilfs_dir_operations;
 527                 inode->i_mapping->a_ops = &nilfs_aops;
 528         } else if (S_ISLNK(inode->i_mode)) {
 529                 inode->i_op = &nilfs_symlink_inode_operations;
 530                 inode_nohighmem(inode);
 531                 inode->i_mapping->a_ops = &nilfs_aops;
 532         } else {
 533                 inode->i_op = &nilfs_special_inode_operations;
 534                 init_special_inode(
 535                         inode, inode->i_mode,
 536                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 537         }
 538         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 539         brelse(bh);
 540         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 541         nilfs_set_inode_flags(inode);
 542         mapping_set_gfp_mask(inode->i_mapping,
 543                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 544         return 0;
 545
 546  failed_unmap:
 547         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 548         brelse(bh);
 549
 550  bad_inode:
 551         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 552         return err;
 553 }
 554
 555 static int nilfs_iget_test(struct inode *inode, void *opaque)
 556 {
 557         struct nilfs_iget_args *args = opaque;
 558         struct nilfs_inode_info *ii;
 559
 560         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 561                 return 0;
 562
 563         ii = NILFS_I(inode);
 564         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 565                 return !args->for_gc;
 566
 567         return args->for_gc && args->cno == ii->i_cno;
 568 }
 569
 570 static int nilfs_iget_set(struct inode *inode, void *opaque)
 571 {
 572         struct nilfs_iget_args *args = opaque;
 573
 574         inode->i_ino = args->ino;
 575         if (args->for_gc) {
 576                 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 577                 NILFS_I(inode)->i_cno = args->cno;
 578                 NILFS_I(inode)->i_root = NULL;
 579         } else {
 580                 if (args->root && args->ino == NILFS_ROOT_INO)
 581                         nilfs_get_root(args->root);
 582                 NILFS_I(inode)->i_root = args->root;
 583         }
 584         return 0;
 585 }
 586
 587 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 588                             unsigned long ino)
 589 {
 590         struct nilfs_iget_args args = {
 591                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 592         };
 593
 594         return ilookup5(sb, ino, nilfs_iget_test, &args);
 595 }
 596
 597 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 598                                 unsigned long ino)
 599 {
 600         struct nilfs_iget_args args = {
 601                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 602         };
 603
 604         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 605 }
 606
 607 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 608                          unsigned long ino)
 609 {
 610         struct inode *inode;
 611         int err;
 612
 613         inode = nilfs_iget_locked(sb, root, ino);
 614         if (unlikely(!inode))
 615                 return ERR_PTR(-ENOMEM);
 616         if (!(inode->i_state & I_NEW))
 617                 return inode;
 618
 619         err = __nilfs_read_inode(sb, root, ino, inode);
 620         if (unlikely(err)) {
 621                 iget_failed(inode);
 622                 return ERR_PTR(err);
 623         }
 624         unlock_new_inode(inode);
 625         return inode;
 626 }
 627
 628 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 629                                 __u64 cno)
 630 {
 631         struct nilfs_iget_args args = {
 632                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 633         };
 634         struct inode *inode;
 635         int err;
 636
 637         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 638         if (unlikely(!inode))
 639                 return ERR_PTR(-ENOMEM);
 640         if (!(inode->i_state & I_NEW))
 641                 return inode;
 642
 643         err = nilfs_init_gcinode(inode);
 644         if (unlikely(err)) {
 645                 iget_failed(inode);
 646                 return ERR_PTR(err);
 647         }
 648         unlock_new_inode(inode);
 649         return inode;
 650 }
 651
 652 void nilfs_write_inode_common(struct inode *inode,
 653                               struct nilfs_inode *raw_inode, int has_bmap)
 654 {
 655         struct nilfs_inode_info *ii = NILFS_I(inode);
 656
 657         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 658         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 659         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 660         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 661         raw_inode->i_size = cpu_to_le64(inode->i_size);
 662         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 663         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 664         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 665         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 666         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 667
 668         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 669         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 670
 671         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 672                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 673
 674                 /* zero-fill unused portion in the case of super root block */
 675                 raw_inode->i_xattr = 0;
 676                 raw_inode->i_pad = 0;
 677                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 678                        nilfs->ns_inode_size - sizeof(*raw_inode));
 679         }
 680
 681         if (has_bmap)
 682                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 683         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 684                 raw_inode->i_device_code =
 685                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 686         /*
 687          * When extending inode, nilfs->ns_inode_size should be checked
 688          * for substitutions of appended fields.
 689          */
 690 }
 691
 692 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 693 {
 694         ino_t ino = inode->i_ino;
 695         struct nilfs_inode_info *ii = NILFS_I(inode);
 696         struct inode *ifile = ii->i_root->ifile;
 697         struct nilfs_inode *raw_inode;
 698
 699         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 700
 701         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 702                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 703         if (flags & I_DIRTY_DATASYNC)
 704                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 705
 706         nilfs_write_inode_common(inode, raw_inode, 0);
 707                 /*
 708                  * XXX: call with has_bmap = 0 is a workaround to avoid
 709                  * deadlock of bmap.  This delays update of i_bmap to just
 710                  * before writing.
 711                  */
 712
 713         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 714 }
 715
 716 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 717
 718 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 719                                 unsigned long from)
 720 {
 721         __u64 b;
 722         int ret;
 723
 724         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 725                 return;
 726 repeat:
 727         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 728         if (ret == -ENOENT)
 729                 return;
 730         else if (ret < 0)
 731                 goto failed;
 732
 733         if (b < from)
 734                 return;
 735
 736         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 737         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 738         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 739         if (!ret || (ret == -ENOMEM &&
 740                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 741                 goto repeat;
 742
 743 failed:
 744         nilfs_msg(ii->vfs_inode.i_sb, KERN_WARNING,
 745                   "error %d truncating bmap (ino=%lu)", ret,
 746                   ii->vfs_inode.i_ino);
 747 }
 748
 749 void nilfs_truncate(struct inode *inode)
 750 {
 751         unsigned long blkoff;
 752         unsigned int blocksize;
 753         struct nilfs_transaction_info ti;
 754         struct super_block *sb = inode->i_sb;
 755         struct nilfs_inode_info *ii = NILFS_I(inode);
 756
 757         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 758                 return;
 759         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 760                 return;
 761
 762         blocksize = sb->s_blocksize;
 763         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 764         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 765
 766         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 767
 768         nilfs_truncate_bmap(ii, blkoff);
 769
 770         inode->i_mtime = inode->i_ctime = current_time(inode);
 771         if (IS_SYNC(inode))
 772                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 773
 774         nilfs_mark_inode_dirty(inode);
 775         nilfs_set_file_dirty(inode, 0);
 776         nilfs_transaction_commit(sb);
 777         /*
 778          * May construct a logical segment and may fail in sync mode.
 779          * But truncate has no return value.
 780          */
 781 }
 782
 783 static void nilfs_clear_inode(struct inode *inode)
 784 {
 785         struct nilfs_inode_info *ii = NILFS_I(inode);
 786
 787         /*
 788          * Free resources allocated in nilfs_read_inode(), here.
 789          */
 790         BUG_ON(!list_empty(&ii->i_dirty));
 791         brelse(ii->i_bh);
 792         ii->i_bh = NULL;
 793
 794         if (nilfs_is_metadata_file_inode(inode))
 795                 nilfs_mdt_clear(inode);
 796
 797         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 798                 nilfs_bmap_clear(ii->i_bmap);
 799
 800         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 801
 802         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 803                 nilfs_put_root(ii->i_root);
 804 }
 805
 806 void nilfs_evict_inode(struct inode *inode)
 807 {
 808         struct nilfs_transaction_info ti;
 809         struct super_block *sb = inode->i_sb;
 810         struct nilfs_inode_info *ii = NILFS_I(inode);
 811         int ret;
 812
 813         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 814                 truncate_inode_pages_final(&inode->i_data);
 815                 clear_inode(inode);
 816                 nilfs_clear_inode(inode);
 817                 return;
 818         }
 819         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 820
 821         truncate_inode_pages_final(&inode->i_data);
 822
 823         /* TODO: some of the following operations may fail.  */
 824         nilfs_truncate_bmap(ii, 0);
 825         nilfs_mark_inode_dirty(inode);
 826         clear_inode(inode);
 827
 828         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 829         if (!ret)
 830                 atomic64_dec(&ii->i_root->inodes_count);
 831
 832         nilfs_clear_inode(inode);
 833
 834         if (IS_SYNC(inode))
 835                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 836         nilfs_transaction_commit(sb);
 837         /*
 838          * May construct a logical segment and may fail in sync mode.
 839          * But delete_inode has no return value.
 840          */
 841 }
 842
 843 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 844 {
 845         struct nilfs_transaction_info ti;
 846         struct inode *inode = d_inode(dentry);
 847         struct super_block *sb = inode->i_sb;
 848         int err;
 849
 850         err = setattr_prepare(dentry, iattr);
 851         if (err)
 852                 return err;
 853
 854         err = nilfs_transaction_begin(sb, &ti, 0);
 855         if (unlikely(err))
 856                 return err;
 857
 858         if ((iattr->ia_valid & ATTR_SIZE) &&
 859             iattr->ia_size != i_size_read(inode)) {
 860                 inode_dio_wait(inode);
 861                 truncate_setsize(inode, iattr->ia_size);
 862                 nilfs_truncate(inode);
 863         }
 864
 865         setattr_copy(inode, iattr);
 866         mark_inode_dirty(inode);
 867
 868         if (iattr->ia_valid & ATTR_MODE) {
 869                 err = nilfs_acl_chmod(inode);
 870                 if (unlikely(err))
 871                         goto out_err;
 872         }
 873
 874         return nilfs_transaction_commit(sb);
 875
 876 out_err:
 877         nilfs_transaction_abort(sb);
 878         return err;
 879 }
 880
 881 int nilfs_permission(struct inode *inode, int mask)
 882 {
 883         struct nilfs_root *root = NILFS_I(inode)->i_root;
 884
 885         if ((mask & MAY_WRITE) && root &&
 886             root->cno != NILFS_CPTREE_CURRENT_CNO)
 887                 return -EROFS; /* snapshot is not writable */
 888
 889         return generic_permission(inode, mask);
 890 }
 891
 892 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 893 {
 894         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 895         struct nilfs_inode_info *ii = NILFS_I(inode);
 896         int err;
 897
 898         spin_lock(&nilfs->ns_inode_lock);
 899         if (ii->i_bh == NULL) {
 900                 spin_unlock(&nilfs->ns_inode_lock);
 901                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 902                                                   inode->i_ino, pbh);
 903                 if (unlikely(err))
 904                         return err;
 905                 spin_lock(&nilfs->ns_inode_lock);
 906                 if (ii->i_bh == NULL)
 907                         ii->i_bh = *pbh;
 908                 else {
 909                         brelse(*pbh);
 910                         *pbh = ii->i_bh;
 911                 }
 912         } else
 913                 *pbh = ii->i_bh;
 914
 915         get_bh(*pbh);
 916         spin_unlock(&nilfs->ns_inode_lock);
 917         return 0;
 918 }
 919
 920 int nilfs_inode_dirty(struct inode *inode)
 921 {
 922         struct nilfs_inode_info *ii = NILFS_I(inode);
 923         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 924         int ret = 0;
 925
 926         if (!list_empty(&ii->i_dirty)) {
 927                 spin_lock(&nilfs->ns_inode_lock);
 928                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 929                         test_bit(NILFS_I_BUSY, &ii->i_state);
 930                 spin_unlock(&nilfs->ns_inode_lock);
 931         }
 932         return ret;
 933 }
 934
 935 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
 936 {
 937         struct nilfs_inode_info *ii = NILFS_I(inode);
 938         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 939
 940         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 941
 942         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 943                 return 0;
 944
 945         spin_lock(&nilfs->ns_inode_lock);
 946         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 947             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 948                 /*
 949                  * Because this routine may race with nilfs_dispose_list(),
 950                  * we have to check NILFS_I_QUEUED here, too.
 951                  */
 952                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 953                         /*
 954                          * This will happen when somebody is freeing
 955                          * this inode.
 956                          */
 957                         nilfs_msg(inode->i_sb, KERN_WARNING,
 958                                   "cannot set file dirty (ino=%lu): the file is being freed",
 959                                   inode->i_ino);
 960                         spin_unlock(&nilfs->ns_inode_lock);
 961                         return -EINVAL; /*
 962                                          * NILFS_I_DIRTY may remain for
 963                                          * freeing inode.
 964                                          */
 965                 }
 966                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 967                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 968         }
 969         spin_unlock(&nilfs->ns_inode_lock);
 970         return 0;
 971 }
 972
 973 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
 974 {
 975         struct buffer_head *ibh;
 976         int err;
 977
 978         err = nilfs_load_inode_block(inode, &ibh);
 979         if (unlikely(err)) {
 980                 nilfs_msg(inode->i_sb, KERN_WARNING,
 981                           "cannot mark inode dirty (ino=%lu): error %d loading inode block",
 982                           inode->i_ino, err);
 983                 return err;
 984         }
 985         nilfs_update_inode(inode, ibh, flags);
 986         mark_buffer_dirty(ibh);
 987         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 988         brelse(ibh);
 989         return 0;
 990 }
 991
 992 /**
 993  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 994  * @inode: inode of the file to be registered.
 995  *
 996  * nilfs_dirty_inode() loads a inode block containing the specified
 997  * @inode and copies data from a nilfs_inode to a corresponding inode
 998  * entry in the inode block. This operation is excluded from the segment
 999  * construction. This function can be called both as a single operation
1000  * and as a part of indivisible file operations.
1001  */
1002 void nilfs_dirty_inode(struct inode *inode, int flags)
1003 {
1004         struct nilfs_transaction_info ti;
1005         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1006
1007         if (is_bad_inode(inode)) {
1008                 nilfs_msg(inode->i_sb, KERN_WARNING,
1009                           "tried to mark bad_inode dirty. ignored.");
1010                 dump_stack();
1011                 return;
1012         }
1013         if (mdi) {
1014                 nilfs_mdt_mark_dirty(inode);
1015                 return;
1016         }
1017         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1018         __nilfs_mark_inode_dirty(inode, flags);
1019         nilfs_transaction_commit(inode->i_sb); /* never fails */
1020 }
1021
1022 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1023                  __u64 start, __u64 len)
1024 {
1025         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1026         __u64 logical = 0, phys = 0, size = 0;
1027         __u32 flags = 0;
1028         loff_t isize;
1029         sector_t blkoff, end_blkoff;
1030         sector_t delalloc_blkoff;
1031         unsigned long delalloc_blklen;
1032         unsigned int blkbits = inode->i_blkbits;
1033         int ret, n;
1034
1035         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1036         if (ret)
1037                 return ret;
1038
1039         inode_lock(inode);
1040
1041         isize = i_size_read(inode);
1042
1043         blkoff = start >> blkbits;
1044         end_blkoff = (start + len - 1) >> blkbits;
1045
1046         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1047                                                         &delalloc_blkoff);
1048
1049         do {
1050                 __u64 blkphy;
1051                 unsigned int maxblocks;
1052
1053                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1054                         if (size) {
1055                                 /* End of the current extent */
1056                                 ret = fiemap_fill_next_extent(
1057                                         fieinfo, logical, phys, size, flags);
1058                                 if (ret)
1059                                         break;
1060                         }
1061                         if (blkoff > end_blkoff)
1062                                 break;
1063
1064                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1065                         logical = blkoff << blkbits;
1066                         phys = 0;
1067                         size = delalloc_blklen << blkbits;
1068
1069                         blkoff = delalloc_blkoff + delalloc_blklen;
1070                         delalloc_blklen = nilfs_find_uncommitted_extent(
1071                                 inode, blkoff, &delalloc_blkoff);
1072                         continue;
1073                 }
1074
1075                 /*
1076                  * Limit the number of blocks that we look up so as
1077                  * not to get into the next delayed allocation extent.
1078                  */
1079                 maxblocks = INT_MAX;
1080                 if (delalloc_blklen)
1081                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1082                                           maxblocks);
1083                 blkphy = 0;
1084
1085                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1086                 n = nilfs_bmap_lookup_contig(
1087                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1088                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1089
1090                 if (n < 0) {
1091                         int past_eof;
1092
1093                         if (unlikely(n != -ENOENT))
1094                                 break; /* error */
1095
1096                         /* HOLE */
1097                         blkoff++;
1098                         past_eof = ((blkoff << blkbits) >= isize);
1099
1100                         if (size) {
1101                                 /* End of the current extent */
1102
1103                                 if (past_eof)
1104                                         flags |= FIEMAP_EXTENT_LAST;
1105
1106                                 ret = fiemap_fill_next_extent(
1107                                         fieinfo, logical, phys, size, flags);
1108                                 if (ret)
1109                                         break;
1110                                 size = 0;
1111                         }
1112                         if (blkoff > end_blkoff || past_eof)
1113                                 break;
1114                 } else {
1115                         if (size) {
1116                                 if (phys && blkphy << blkbits == phys + size) {
1117                                         /* The current extent goes on */
1118                                         size += n << blkbits;
1119                                 } else {
1120                                         /* Terminate the current extent */
1121                                         ret = fiemap_fill_next_extent(
1122                                                 fieinfo, logical, phys, size,
1123                                                 flags);
1124                                         if (ret || blkoff > end_blkoff)
1125                                                 break;
1126
1127                                         /* Start another extent */
1128                                         flags = FIEMAP_EXTENT_MERGED;
1129                                         logical = blkoff << blkbits;
1130                                         phys = blkphy << blkbits;
1131                                         size = n << blkbits;
1132                                 }
1133                         } else {
1134                                 /* Start a new extent */
1135                                 flags = FIEMAP_EXTENT_MERGED;
1136                                 logical = blkoff << blkbits;
1137                                 phys = blkphy << blkbits;
1138                                 size = n << blkbits;
1139                         }
1140                         blkoff += n;
1141                 }
1142                 cond_resched();
1143         } while (true);
1144
1145         /* If ret is 1 then we just hit the end of the extent array */
1146         if (ret == 1)
1147                 ret = 0;
1148
1149         inode_unlock(inode);
1150         return ret;
1151 }