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