fs/nilfs2/mdt.c

   1 /*
   2  * mdt.c - meta data file for NILFS
   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 #include <linux/buffer_head.h>
  24 #include <linux/mpage.h>
  25 #include <linux/mm.h>
  26 #include <linux/writeback.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/swap.h>
  29 #include <linux/slab.h>
  30 #include "nilfs.h"
  31 #include "btnode.h"
  32 #include "segment.h"
  33 #include "page.h"
  34 #include "mdt.h"
  35
  36 #include <trace/events/nilfs2.h>
  37
  38 #define NILFS_MDT_MAX_RA_BLOCKS         (16 - 1)
  39
  40
  41 static int
  42 nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
  43                            struct buffer_head *bh,
  44                            void (*init_block)(struct inode *,
  45                                               struct buffer_head *, void *))
  46 {
  47         struct nilfs_inode_info *ii = NILFS_I(inode);
  48         void *kaddr;
  49         int ret;
  50
  51         /* Caller exclude read accesses using page lock */
  52
  53         /* set_buffer_new(bh); */
  54         bh->b_blocknr = 0;
  55
  56         ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
  57         if (unlikely(ret))
  58                 return ret;
  59
  60         set_buffer_mapped(bh);
  61
  62         kaddr = kmap_atomic(bh->b_page);
  63         memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
  64         if (init_block)
  65                 init_block(inode, bh, kaddr);
  66         flush_dcache_page(bh->b_page);
  67         kunmap_atomic(kaddr);
  68
  69         set_buffer_uptodate(bh);
  70         mark_buffer_dirty(bh);
  71         nilfs_mdt_mark_dirty(inode);
  72
  73         trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
  74
  75         return 0;
  76 }
  77
  78 static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
  79                                   struct buffer_head **out_bh,
  80                                   void (*init_block)(struct inode *,
  81                                                      struct buffer_head *,
  82                                                      void *))
  83 {
  84         struct super_block *sb = inode->i_sb;
  85         struct nilfs_transaction_info ti;
  86         struct buffer_head *bh;
  87         int err;
  88
  89         nilfs_transaction_begin(sb, &ti, 0);
  90
  91         err = -ENOMEM;
  92         bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
  93         if (unlikely(!bh))
  94                 goto failed_unlock;
  95
  96         err = -EEXIST;
  97         if (buffer_uptodate(bh))
  98                 goto failed_bh;
  99
 100         wait_on_buffer(bh);
 101         if (buffer_uptodate(bh))
 102                 goto failed_bh;
 103
 104         bh->b_bdev = sb->s_bdev;
 105         err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
 106         if (likely(!err)) {
 107                 get_bh(bh);
 108                 *out_bh = bh;
 109         }
 110
 111  failed_bh:
 112         unlock_page(bh->b_page);
 113         put_page(bh->b_page);
 114         brelse(bh);
 115
 116  failed_unlock:
 117         if (likely(!err))
 118                 err = nilfs_transaction_commit(sb);
 119         else
 120                 nilfs_transaction_abort(sb);
 121
 122         return err;
 123 }
 124
 125 static int
 126 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
 127                        int mode, struct buffer_head **out_bh)
 128 {
 129         struct buffer_head *bh;
 130         __u64 blknum = 0;
 131         int ret = -ENOMEM;
 132
 133         bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
 134         if (unlikely(!bh))
 135                 goto failed;
 136
 137         ret = -EEXIST; /* internal code */
 138         if (buffer_uptodate(bh))
 139                 goto out;
 140
 141         if (mode == READA) {
 142                 if (!trylock_buffer(bh)) {
 143                         ret = -EBUSY;
 144                         goto failed_bh;
 145                 }
 146         } else /* mode == READ */
 147                 lock_buffer(bh);
 148
 149         if (buffer_uptodate(bh)) {
 150                 unlock_buffer(bh);
 151                 goto out;
 152         }
 153
 154         ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
 155         if (unlikely(ret)) {
 156                 unlock_buffer(bh);
 157                 goto failed_bh;
 158         }
 159         map_bh(bh, inode->i_sb, (sector_t)blknum);
 160
 161         bh->b_end_io = end_buffer_read_sync;
 162         get_bh(bh);
 163         submit_bh(mode, bh);
 164         ret = 0;
 165
 166         trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff, mode);
 167  out:
 168         get_bh(bh);
 169         *out_bh = bh;
 170
 171  failed_bh:
 172         unlock_page(bh->b_page);
 173         put_page(bh->b_page);
 174         brelse(bh);
 175  failed:
 176         return ret;
 177 }
 178
 179 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
 180                                 int readahead, struct buffer_head **out_bh)
 181 {
 182         struct buffer_head *first_bh, *bh;
 183         unsigned long blkoff;
 184         int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
 185         int err;
 186
 187         err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
 188         if (err == -EEXIST) /* internal code */
 189                 goto out;
 190
 191         if (unlikely(err))
 192                 goto failed;
 193
 194         if (readahead) {
 195                 blkoff = block + 1;
 196                 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
 197                         err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
 198                         if (likely(!err || err == -EEXIST))
 199                                 brelse(bh);
 200                         else if (err != -EBUSY)
 201                                 break;
 202                                 /* abort readahead if bmap lookup failed */
 203                         if (!buffer_locked(first_bh))
 204                                 goto out_no_wait;
 205                 }
 206         }
 207
 208         wait_on_buffer(first_bh);
 209
 210  out_no_wait:
 211         err = -EIO;
 212         if (!buffer_uptodate(first_bh))
 213                 goto failed_bh;
 214  out:
 215         *out_bh = first_bh;
 216         return 0;
 217
 218  failed_bh:
 219         brelse(first_bh);
 220  failed:
 221         return err;
 222 }
 223
 224 /**
 225  * nilfs_mdt_get_block - read or create a buffer on meta data file.
 226  * @inode: inode of the meta data file
 227  * @blkoff: block offset
 228  * @create: create flag
 229  * @init_block: initializer used for newly allocated block
 230  * @out_bh: output of a pointer to the buffer_head
 231  *
 232  * nilfs_mdt_get_block() looks up the specified buffer and tries to create
 233  * a new buffer if @create is not zero.  On success, the returned buffer is
 234  * assured to be either existing or formatted using a buffer lock on success.
 235  * @out_bh is substituted only when zero is returned.
 236  *
 237  * Return Value: On success, it returns 0. On error, the following negative
 238  * error code is returned.
 239  *
 240  * %-ENOMEM - Insufficient memory available.
 241  *
 242  * %-EIO - I/O error
 243  *
 244  * %-ENOENT - the specified block does not exist (hole block)
 245  *
 246  * %-EROFS - Read only filesystem (for create mode)
 247  */
 248 int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
 249                         void (*init_block)(struct inode *,
 250                                            struct buffer_head *, void *),
 251                         struct buffer_head **out_bh)
 252 {
 253         int ret;
 254
 255         /* Should be rewritten with merging nilfs_mdt_read_block() */
 256  retry:
 257         ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
 258         if (!create || ret != -ENOENT)
 259                 return ret;
 260
 261         ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
 262         if (unlikely(ret == -EEXIST)) {
 263                 /* create = 0; */  /* limit read-create loop retries */
 264                 goto retry;
 265         }
 266         return ret;
 267 }
 268
 269 /**
 270  * nilfs_mdt_find_block - find and get a buffer on meta data file.
 271  * @inode: inode of the meta data file
 272  * @start: start block offset (inclusive)
 273  * @end: end block offset (inclusive)
 274  * @blkoff: block offset
 275  * @out_bh: place to store a pointer to buffer_head struct
 276  *
 277  * nilfs_mdt_find_block() looks up an existing block in range of
 278  * [@start, @end] and stores pointer to a buffer head of the block to
 279  * @out_bh, and block offset to @blkoff, respectively.  @out_bh and
 280  * @blkoff are substituted only when zero is returned.
 281  *
 282  * Return Value: On success, it returns 0. On error, the following negative
 283  * error code is returned.
 284  *
 285  * %-ENOMEM - Insufficient memory available.
 286  *
 287  * %-EIO - I/O error
 288  *
 289  * %-ENOENT - no block was found in the range
 290  */
 291 int nilfs_mdt_find_block(struct inode *inode, unsigned long start,
 292                          unsigned long end, unsigned long *blkoff,
 293                          struct buffer_head **out_bh)
 294 {
 295         __u64 next;
 296         int ret;
 297
 298         if (unlikely(start > end))
 299                 return -ENOENT;
 300
 301         ret = nilfs_mdt_read_block(inode, start, true, out_bh);
 302         if (!ret) {
 303                 *blkoff = start;
 304                 goto out;
 305         }
 306         if (unlikely(ret != -ENOENT || start == ULONG_MAX))
 307                 goto out;
 308
 309         ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
 310         if (!ret) {
 311                 if (next <= end) {
 312                         ret = nilfs_mdt_read_block(inode, next, true, out_bh);
 313                         if (!ret)
 314                                 *blkoff = next;
 315                 } else {
 316                         ret = -ENOENT;
 317                 }
 318         }
 319 out:
 320         return ret;
 321 }
 322
 323 /**
 324  * nilfs_mdt_delete_block - make a hole on the meta data file.
 325  * @inode: inode of the meta data file
 326  * @block: block offset
 327  *
 328  * Return Value: On success, zero is returned.
 329  * On error, one of the following negative error code is returned.
 330  *
 331  * %-ENOMEM - Insufficient memory available.
 332  *
 333  * %-EIO - I/O error
 334  */
 335 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
 336 {
 337         struct nilfs_inode_info *ii = NILFS_I(inode);
 338         int err;
 339
 340         err = nilfs_bmap_delete(ii->i_bmap, block);
 341         if (!err || err == -ENOENT) {
 342                 nilfs_mdt_mark_dirty(inode);
 343                 nilfs_mdt_forget_block(inode, block);
 344         }
 345         return err;
 346 }
 347
 348 /**
 349  * nilfs_mdt_forget_block - discard dirty state and try to remove the page
 350  * @inode: inode of the meta data file
 351  * @block: block offset
 352  *
 353  * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
 354  * tries to release the page including the buffer from a page cache.
 355  *
 356  * Return Value: On success, 0 is returned. On error, one of the following
 357  * negative error code is returned.
 358  *
 359  * %-EBUSY - page has an active buffer.
 360  *
 361  * %-ENOENT - page cache has no page addressed by the offset.
 362  */
 363 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
 364 {
 365         pgoff_t index = (pgoff_t)block >>
 366                 (PAGE_SHIFT - inode->i_blkbits);
 367         struct page *page;
 368         unsigned long first_block;
 369         int ret = 0;
 370         int still_dirty;
 371
 372         page = find_lock_page(inode->i_mapping, index);
 373         if (!page)
 374                 return -ENOENT;
 375
 376         wait_on_page_writeback(page);
 377
 378         first_block = (unsigned long)index <<
 379                 (PAGE_SHIFT - inode->i_blkbits);
 380         if (page_has_buffers(page)) {
 381                 struct buffer_head *bh;
 382
 383                 bh = nilfs_page_get_nth_block(page, block - first_block);
 384                 nilfs_forget_buffer(bh);
 385         }
 386         still_dirty = PageDirty(page);
 387         unlock_page(page);
 388         put_page(page);
 389
 390         if (still_dirty ||
 391             invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
 392                 ret = -EBUSY;
 393         return ret;
 394 }
 395
 396 /**
 397  * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
 398  * @inode: inode of the meta data file
 399  * @block: block offset
 400  *
 401  * Return Value: On success, it returns 0. On error, the following negative
 402  * error code is returned.
 403  *
 404  * %-ENOMEM - Insufficient memory available.
 405  *
 406  * %-EIO - I/O error
 407  *
 408  * %-ENOENT - the specified block does not exist (hole block)
 409  */
 410 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
 411 {
 412         struct buffer_head *bh;
 413         int err;
 414
 415         err = nilfs_mdt_read_block(inode, block, 0, &bh);
 416         if (unlikely(err))
 417                 return err;
 418         mark_buffer_dirty(bh);
 419         nilfs_mdt_mark_dirty(inode);
 420         brelse(bh);
 421         return 0;
 422 }
 423
 424 int nilfs_mdt_fetch_dirty(struct inode *inode)
 425 {
 426         struct nilfs_inode_info *ii = NILFS_I(inode);
 427
 428         if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
 429                 set_bit(NILFS_I_DIRTY, &ii->i_state);
 430                 return 1;
 431         }
 432         return test_bit(NILFS_I_DIRTY, &ii->i_state);
 433 }
 434
 435 static int
 436 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
 437 {
 438         struct inode *inode = page->mapping->host;
 439         struct super_block *sb;
 440         int err = 0;
 441
 442         if (inode && (inode->i_sb->s_flags & MS_RDONLY)) {
 443                 /*
 444                  * It means that filesystem was remounted in read-only
 445                  * mode because of error or metadata corruption. But we
 446                  * have dirty pages that try to be flushed in background.
 447                  * So, here we simply discard this dirty page.
 448                  */
 449                 nilfs_clear_dirty_page(page, false);
 450                 unlock_page(page);
 451                 return -EROFS;
 452         }
 453
 454         redirty_page_for_writepage(wbc, page);
 455         unlock_page(page);
 456
 457         if (!inode)
 458                 return 0;
 459
 460         sb = inode->i_sb;
 461
 462         if (wbc->sync_mode == WB_SYNC_ALL)
 463                 err = nilfs_construct_segment(sb);
 464         else if (wbc->for_reclaim)
 465                 nilfs_flush_segment(sb, inode->i_ino);
 466
 467         return err;
 468 }
 469
 470
 471 static const struct address_space_operations def_mdt_aops = {
 472         .writepage              = nilfs_mdt_write_page,
 473 };
 474
 475 static const struct inode_operations def_mdt_iops;
 476 static const struct file_operations def_mdt_fops;
 477
 478
 479 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
 480 {
 481         struct nilfs_mdt_info *mi;
 482
 483         mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
 484         if (!mi)
 485                 return -ENOMEM;
 486
 487         init_rwsem(&mi->mi_sem);
 488         inode->i_private = mi;
 489
 490         inode->i_mode = S_IFREG;
 491         mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
 492
 493         inode->i_op = &def_mdt_iops;
 494         inode->i_fop = &def_mdt_fops;
 495         inode->i_mapping->a_ops = &def_mdt_aops;
 496
 497         return 0;
 498 }
 499
 500 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
 501                               unsigned header_size)
 502 {
 503         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 504
 505         mi->mi_entry_size = entry_size;
 506         mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
 507         mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
 508 }
 509
 510 /**
 511  * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
 512  * @inode: inode of the metadata file
 513  * @shadow: shadow mapping
 514  */
 515 int nilfs_mdt_setup_shadow_map(struct inode *inode,
 516                                struct nilfs_shadow_map *shadow)
 517 {
 518         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 519
 520         INIT_LIST_HEAD(&shadow->frozen_buffers);
 521         address_space_init_once(&shadow->frozen_data);
 522         nilfs_mapping_init(&shadow->frozen_data, inode);
 523         address_space_init_once(&shadow->frozen_btnodes);
 524         nilfs_mapping_init(&shadow->frozen_btnodes, inode);
 525         mi->mi_shadow = shadow;
 526         return 0;
 527 }
 528
 529 /**
 530  * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
 531  * @inode: inode of the metadata file
 532  */
 533 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
 534 {
 535         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 536         struct nilfs_inode_info *ii = NILFS_I(inode);
 537         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 538         int ret;
 539
 540         ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
 541         if (ret)
 542                 goto out;
 543
 544         ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
 545                                      &ii->i_btnode_cache);
 546         if (ret)
 547                 goto out;
 548
 549         nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
 550  out:
 551         return ret;
 552 }
 553
 554 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
 555 {
 556         struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
 557         struct buffer_head *bh_frozen;
 558         struct page *page;
 559         int blkbits = inode->i_blkbits;
 560
 561         page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
 562         if (!page)
 563                 return -ENOMEM;
 564
 565         if (!page_has_buffers(page))
 566                 create_empty_buffers(page, 1 << blkbits, 0);
 567
 568         bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
 569
 570         if (!buffer_uptodate(bh_frozen))
 571                 nilfs_copy_buffer(bh_frozen, bh);
 572         if (list_empty(&bh_frozen->b_assoc_buffers)) {
 573                 list_add_tail(&bh_frozen->b_assoc_buffers,
 574                               &shadow->frozen_buffers);
 575                 set_buffer_nilfs_redirected(bh);
 576         } else {
 577                 brelse(bh_frozen); /* already frozen */
 578         }
 579
 580         unlock_page(page);
 581         put_page(page);
 582         return 0;
 583 }
 584
 585 struct buffer_head *
 586 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
 587 {
 588         struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
 589         struct buffer_head *bh_frozen = NULL;
 590         struct page *page;
 591         int n;
 592
 593         page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
 594         if (page) {
 595                 if (page_has_buffers(page)) {
 596                         n = bh_offset(bh) >> inode->i_blkbits;
 597                         bh_frozen = nilfs_page_get_nth_block(page, n);
 598                 }
 599                 unlock_page(page);
 600                 put_page(page);
 601         }
 602         return bh_frozen;
 603 }
 604
 605 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
 606 {
 607         struct list_head *head = &shadow->frozen_buffers;
 608         struct buffer_head *bh;
 609
 610         while (!list_empty(head)) {
 611                 bh = list_first_entry(head, struct buffer_head,
 612                                       b_assoc_buffers);
 613                 list_del_init(&bh->b_assoc_buffers);
 614                 brelse(bh); /* drop ref-count to make it releasable */
 615         }
 616 }
 617
 618 /**
 619  * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
 620  * @inode: inode of the metadata file
 621  */
 622 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
 623 {
 624         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 625         struct nilfs_inode_info *ii = NILFS_I(inode);
 626         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 627
 628         down_write(&mi->mi_sem);
 629
 630         if (mi->mi_palloc_cache)
 631                 nilfs_palloc_clear_cache(inode);
 632
 633         nilfs_clear_dirty_pages(inode->i_mapping, true);
 634         nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
 635
 636         nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
 637         nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
 638
 639         nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
 640
 641         up_write(&mi->mi_sem);
 642 }
 643
 644 /**
 645  * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
 646  * @inode: inode of the metadata file
 647  */
 648 void nilfs_mdt_clear_shadow_map(struct inode *inode)
 649 {
 650         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 651         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 652
 653         down_write(&mi->mi_sem);
 654         nilfs_release_frozen_buffers(shadow);
 655         truncate_inode_pages(&shadow->frozen_data, 0);
 656         truncate_inode_pages(&shadow->frozen_btnodes, 0);
 657         up_write(&mi->mi_sem);
 658 }