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  * Written by Ryusuke Konishi.
  17  */
  18
  19 #include <linux/buffer_head.h>
  20 #include <linux/mpage.h>
  21 #include <linux/mm.h>
  22 #include <linux/writeback.h>
  23 #include <linux/backing-dev.h>
  24 #include <linux/swap.h>
  25 #include <linux/slab.h>
  26 #include "nilfs.h"
  27 #include "btnode.h"
  28 #include "segment.h"
  29 #include "page.h"
  30 #include "mdt.h"
  31 #include "alloc.h"              /* nilfs_palloc_destroy_cache() */
  32
  33 #include <trace/events/nilfs2.h>
  34
  35 #define NILFS_MDT_MAX_RA_BLOCKS         (16 - 1)
  36
  37
  38 static int
  39 nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
  40                            struct buffer_head *bh,
  41                            void (*init_block)(struct inode *,
  42                                               struct buffer_head *, void *))
  43 {
  44         struct nilfs_inode_info *ii = NILFS_I(inode);
  45         void *kaddr;
  46         int ret;
  47
  48         /* Caller exclude read accesses using page lock */
  49
  50         /* set_buffer_new(bh); */
  51         bh->b_blocknr = 0;
  52
  53         ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
  54         if (unlikely(ret))
  55                 return ret;
  56
  57         set_buffer_mapped(bh);
  58
  59         kaddr = kmap_atomic(bh->b_page);
  60         memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
  61         if (init_block)
  62                 init_block(inode, bh, kaddr);
  63         flush_dcache_page(bh->b_page);
  64         kunmap_atomic(kaddr);
  65
  66         set_buffer_uptodate(bh);
  67         mark_buffer_dirty(bh);
  68         nilfs_mdt_mark_dirty(inode);
  69
  70         trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
  71
  72         return 0;
  73 }
  74
  75 static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
  76                                   struct buffer_head **out_bh,
  77                                   void (*init_block)(struct inode *,
  78                                                      struct buffer_head *,
  79                                                      void *))
  80 {
  81         struct super_block *sb = inode->i_sb;
  82         struct nilfs_transaction_info ti;
  83         struct buffer_head *bh;
  84         int err;
  85
  86         nilfs_transaction_begin(sb, &ti, 0);
  87
  88         err = -ENOMEM;
  89         bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
  90         if (unlikely(!bh))
  91                 goto failed_unlock;
  92
  93         err = -EEXIST;
  94         if (buffer_uptodate(bh))
  95                 goto failed_bh;
  96
  97         wait_on_buffer(bh);
  98         if (buffer_uptodate(bh))
  99                 goto failed_bh;
 100
 101         bh->b_bdev = sb->s_bdev;
 102         err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
 103         if (likely(!err)) {
 104                 get_bh(bh);
 105                 *out_bh = bh;
 106         }
 107
 108  failed_bh:
 109         unlock_page(bh->b_page);
 110         put_page(bh->b_page);
 111         brelse(bh);
 112
 113  failed_unlock:
 114         if (likely(!err))
 115                 err = nilfs_transaction_commit(sb);
 116         else
 117                 nilfs_transaction_abort(sb);
 118
 119         return err;
 120 }
 121
 122 static int
 123 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
 124                        int mode, struct buffer_head **out_bh)
 125 {
 126         struct buffer_head *bh;
 127         __u64 blknum = 0;
 128         int ret = -ENOMEM;
 129
 130         bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
 131         if (unlikely(!bh))
 132                 goto failed;
 133
 134         ret = -EEXIST; /* internal code */
 135         if (buffer_uptodate(bh))
 136                 goto out;
 137
 138         if (mode == READA) {
 139                 if (!trylock_buffer(bh)) {
 140                         ret = -EBUSY;
 141                         goto failed_bh;
 142                 }
 143         } else /* mode == READ */
 144                 lock_buffer(bh);
 145
 146         if (buffer_uptodate(bh)) {
 147                 unlock_buffer(bh);
 148                 goto out;
 149         }
 150
 151         ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
 152         if (unlikely(ret)) {
 153                 unlock_buffer(bh);
 154                 goto failed_bh;
 155         }
 156         map_bh(bh, inode->i_sb, (sector_t)blknum);
 157
 158         bh->b_end_io = end_buffer_read_sync;
 159         get_bh(bh);
 160         submit_bh(mode, bh);
 161         ret = 0;
 162
 163         trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff, mode);
 164  out:
 165         get_bh(bh);
 166         *out_bh = bh;
 167
 168  failed_bh:
 169         unlock_page(bh->b_page);
 170         put_page(bh->b_page);
 171         brelse(bh);
 172  failed:
 173         return ret;
 174 }
 175
 176 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
 177                                 int readahead, struct buffer_head **out_bh)
 178 {
 179         struct buffer_head *first_bh, *bh;
 180         unsigned long blkoff;
 181         int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
 182         int err;
 183
 184         err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
 185         if (err == -EEXIST) /* internal code */
 186                 goto out;
 187
 188         if (unlikely(err))
 189                 goto failed;
 190
 191         if (readahead) {
 192                 blkoff = block + 1;
 193                 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
 194                         err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
 195                         if (likely(!err || err == -EEXIST))
 196                                 brelse(bh);
 197                         else if (err != -EBUSY)
 198                                 break;
 199                                 /* abort readahead if bmap lookup failed */
 200                         if (!buffer_locked(first_bh))
 201                                 goto out_no_wait;
 202                 }
 203         }
 204
 205         wait_on_buffer(first_bh);
 206
 207  out_no_wait:
 208         err = -EIO;
 209         if (!buffer_uptodate(first_bh))
 210                 goto failed_bh;
 211  out:
 212         *out_bh = first_bh;
 213         return 0;
 214
 215  failed_bh:
 216         brelse(first_bh);
 217  failed:
 218         return err;
 219 }
 220
 221 /**
 222  * nilfs_mdt_get_block - read or create a buffer on meta data file.
 223  * @inode: inode of the meta data file
 224  * @blkoff: block offset
 225  * @create: create flag
 226  * @init_block: initializer used for newly allocated block
 227  * @out_bh: output of a pointer to the buffer_head
 228  *
 229  * nilfs_mdt_get_block() looks up the specified buffer and tries to create
 230  * a new buffer if @create is not zero.  On success, the returned buffer is
 231  * assured to be either existing or formatted using a buffer lock on success.
 232  * @out_bh is substituted only when zero is returned.
 233  *
 234  * Return Value: On success, it returns 0. On error, the following negative
 235  * error code is returned.
 236  *
 237  * %-ENOMEM - Insufficient memory available.
 238  *
 239  * %-EIO - I/O error
 240  *
 241  * %-ENOENT - the specified block does not exist (hole block)
 242  *
 243  * %-EROFS - Read only filesystem (for create mode)
 244  */
 245 int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
 246                         void (*init_block)(struct inode *,
 247                                            struct buffer_head *, void *),
 248                         struct buffer_head **out_bh)
 249 {
 250         int ret;
 251
 252         /* Should be rewritten with merging nilfs_mdt_read_block() */
 253  retry:
 254         ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
 255         if (!create || ret != -ENOENT)
 256                 return ret;
 257
 258         ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
 259         if (unlikely(ret == -EEXIST)) {
 260                 /* create = 0; */  /* limit read-create loop retries */
 261                 goto retry;
 262         }
 263         return ret;
 264 }
 265
 266 /**
 267  * nilfs_mdt_find_block - find and get a buffer on meta data file.
 268  * @inode: inode of the meta data file
 269  * @start: start block offset (inclusive)
 270  * @end: end block offset (inclusive)
 271  * @blkoff: block offset
 272  * @out_bh: place to store a pointer to buffer_head struct
 273  *
 274  * nilfs_mdt_find_block() looks up an existing block in range of
 275  * [@start, @end] and stores pointer to a buffer head of the block to
 276  * @out_bh, and block offset to @blkoff, respectively.  @out_bh and
 277  * @blkoff are substituted only when zero is returned.
 278  *
 279  * Return Value: On success, it returns 0. On error, the following negative
 280  * error code is returned.
 281  *
 282  * %-ENOMEM - Insufficient memory available.
 283  *
 284  * %-EIO - I/O error
 285  *
 286  * %-ENOENT - no block was found in the range
 287  */
 288 int nilfs_mdt_find_block(struct inode *inode, unsigned long start,
 289                          unsigned long end, unsigned long *blkoff,
 290                          struct buffer_head **out_bh)
 291 {
 292         __u64 next;
 293         int ret;
 294
 295         if (unlikely(start > end))
 296                 return -ENOENT;
 297
 298         ret = nilfs_mdt_read_block(inode, start, true, out_bh);
 299         if (!ret) {
 300                 *blkoff = start;
 301                 goto out;
 302         }
 303         if (unlikely(ret != -ENOENT || start == ULONG_MAX))
 304                 goto out;
 305
 306         ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
 307         if (!ret) {
 308                 if (next <= end) {
 309                         ret = nilfs_mdt_read_block(inode, next, true, out_bh);
 310                         if (!ret)
 311                                 *blkoff = next;
 312                 } else {
 313                         ret = -ENOENT;
 314                 }
 315         }
 316 out:
 317         return ret;
 318 }
 319
 320 /**
 321  * nilfs_mdt_delete_block - make a hole on the meta data file.
 322  * @inode: inode of the meta data file
 323  * @block: block offset
 324  *
 325  * Return Value: On success, zero is returned.
 326  * On error, one of the following negative error code is returned.
 327  *
 328  * %-ENOMEM - Insufficient memory available.
 329  *
 330  * %-EIO - I/O error
 331  */
 332 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
 333 {
 334         struct nilfs_inode_info *ii = NILFS_I(inode);
 335         int err;
 336
 337         err = nilfs_bmap_delete(ii->i_bmap, block);
 338         if (!err || err == -ENOENT) {
 339                 nilfs_mdt_mark_dirty(inode);
 340                 nilfs_mdt_forget_block(inode, block);
 341         }
 342         return err;
 343 }
 344
 345 /**
 346  * nilfs_mdt_forget_block - discard dirty state and try to remove the page
 347  * @inode: inode of the meta data file
 348  * @block: block offset
 349  *
 350  * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
 351  * tries to release the page including the buffer from a page cache.
 352  *
 353  * Return Value: On success, 0 is returned. On error, one of the following
 354  * negative error code is returned.
 355  *
 356  * %-EBUSY - page has an active buffer.
 357  *
 358  * %-ENOENT - page cache has no page addressed by the offset.
 359  */
 360 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
 361 {
 362         pgoff_t index = (pgoff_t)block >>
 363                 (PAGE_SHIFT - inode->i_blkbits);
 364         struct page *page;
 365         unsigned long first_block;
 366         int ret = 0;
 367         int still_dirty;
 368
 369         page = find_lock_page(inode->i_mapping, index);
 370         if (!page)
 371                 return -ENOENT;
 372
 373         wait_on_page_writeback(page);
 374
 375         first_block = (unsigned long)index <<
 376                 (PAGE_SHIFT - inode->i_blkbits);
 377         if (page_has_buffers(page)) {
 378                 struct buffer_head *bh;
 379
 380                 bh = nilfs_page_get_nth_block(page, block - first_block);
 381                 nilfs_forget_buffer(bh);
 382         }
 383         still_dirty = PageDirty(page);
 384         unlock_page(page);
 385         put_page(page);
 386
 387         if (still_dirty ||
 388             invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
 389                 ret = -EBUSY;
 390         return ret;
 391 }
 392
 393 int nilfs_mdt_fetch_dirty(struct inode *inode)
 394 {
 395         struct nilfs_inode_info *ii = NILFS_I(inode);
 396
 397         if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
 398                 set_bit(NILFS_I_DIRTY, &ii->i_state);
 399                 return 1;
 400         }
 401         return test_bit(NILFS_I_DIRTY, &ii->i_state);
 402 }
 403
 404 static int
 405 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
 406 {
 407         struct inode *inode = page->mapping->host;
 408         struct super_block *sb;
 409         int err = 0;
 410
 411         if (inode && (inode->i_sb->s_flags & MS_RDONLY)) {
 412                 /*
 413                  * It means that filesystem was remounted in read-only
 414                  * mode because of error or metadata corruption. But we
 415                  * have dirty pages that try to be flushed in background.
 416                  * So, here we simply discard this dirty page.
 417                  */
 418                 nilfs_clear_dirty_page(page, false);
 419                 unlock_page(page);
 420                 return -EROFS;
 421         }
 422
 423         redirty_page_for_writepage(wbc, page);
 424         unlock_page(page);
 425
 426         if (!inode)
 427                 return 0;
 428
 429         sb = inode->i_sb;
 430
 431         if (wbc->sync_mode == WB_SYNC_ALL)
 432                 err = nilfs_construct_segment(sb);
 433         else if (wbc->for_reclaim)
 434                 nilfs_flush_segment(sb, inode->i_ino);
 435
 436         return err;
 437 }
 438
 439
 440 static const struct address_space_operations def_mdt_aops = {
 441         .writepage              = nilfs_mdt_write_page,
 442 };
 443
 444 static const struct inode_operations def_mdt_iops;
 445 static const struct file_operations def_mdt_fops;
 446
 447
 448 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
 449 {
 450         struct nilfs_mdt_info *mi;
 451
 452         mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
 453         if (!mi)
 454                 return -ENOMEM;
 455
 456         init_rwsem(&mi->mi_sem);
 457         inode->i_private = mi;
 458
 459         inode->i_mode = S_IFREG;
 460         mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
 461
 462         inode->i_op = &def_mdt_iops;
 463         inode->i_fop = &def_mdt_fops;
 464         inode->i_mapping->a_ops = &def_mdt_aops;
 465
 466         return 0;
 467 }
 468
 469 /**
 470  * nilfs_mdt_clear - do cleanup for the metadata file
 471  * @inode: inode of the metadata file
 472  */
 473 void nilfs_mdt_clear(struct inode *inode)
 474 {
 475         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 476
 477         if (mdi->mi_palloc_cache)
 478                 nilfs_palloc_destroy_cache(inode);
 479 }
 480
 481 /**
 482  * nilfs_mdt_destroy - release resources used by the metadata file
 483  * @inode: inode of the metadata file
 484  */
 485 void nilfs_mdt_destroy(struct inode *inode)
 486 {
 487         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 488
 489         kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
 490         kfree(mdi);
 491 }
 492
 493 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
 494                               unsigned int header_size)
 495 {
 496         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 497
 498         mi->mi_entry_size = entry_size;
 499         mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
 500         mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
 501 }
 502
 503 /**
 504  * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
 505  * @inode: inode of the metadata file
 506  * @shadow: shadow mapping
 507  */
 508 int nilfs_mdt_setup_shadow_map(struct inode *inode,
 509                                struct nilfs_shadow_map *shadow)
 510 {
 511         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 512
 513         INIT_LIST_HEAD(&shadow->frozen_buffers);
 514         address_space_init_once(&shadow->frozen_data);
 515         nilfs_mapping_init(&shadow->frozen_data, inode);
 516         address_space_init_once(&shadow->frozen_btnodes);
 517         nilfs_mapping_init(&shadow->frozen_btnodes, inode);
 518         mi->mi_shadow = shadow;
 519         return 0;
 520 }
 521
 522 /**
 523  * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
 524  * @inode: inode of the metadata file
 525  */
 526 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
 527 {
 528         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 529         struct nilfs_inode_info *ii = NILFS_I(inode);
 530         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 531         int ret;
 532
 533         ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
 534         if (ret)
 535                 goto out;
 536
 537         ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
 538                                      &ii->i_btnode_cache);
 539         if (ret)
 540                 goto out;
 541
 542         nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
 543  out:
 544         return ret;
 545 }
 546
 547 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
 548 {
 549         struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
 550         struct buffer_head *bh_frozen;
 551         struct page *page;
 552         int blkbits = inode->i_blkbits;
 553
 554         page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
 555         if (!page)
 556                 return -ENOMEM;
 557
 558         if (!page_has_buffers(page))
 559                 create_empty_buffers(page, 1 << blkbits, 0);
 560
 561         bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
 562
 563         if (!buffer_uptodate(bh_frozen))
 564                 nilfs_copy_buffer(bh_frozen, bh);
 565         if (list_empty(&bh_frozen->b_assoc_buffers)) {
 566                 list_add_tail(&bh_frozen->b_assoc_buffers,
 567                               &shadow->frozen_buffers);
 568                 set_buffer_nilfs_redirected(bh);
 569         } else {
 570                 brelse(bh_frozen); /* already frozen */
 571         }
 572
 573         unlock_page(page);
 574         put_page(page);
 575         return 0;
 576 }
 577
 578 struct buffer_head *
 579 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
 580 {
 581         struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
 582         struct buffer_head *bh_frozen = NULL;
 583         struct page *page;
 584         int n;
 585
 586         page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
 587         if (page) {
 588                 if (page_has_buffers(page)) {
 589                         n = bh_offset(bh) >> inode->i_blkbits;
 590                         bh_frozen = nilfs_page_get_nth_block(page, n);
 591                 }
 592                 unlock_page(page);
 593                 put_page(page);
 594         }
 595         return bh_frozen;
 596 }
 597
 598 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
 599 {
 600         struct list_head *head = &shadow->frozen_buffers;
 601         struct buffer_head *bh;
 602
 603         while (!list_empty(head)) {
 604                 bh = list_first_entry(head, struct buffer_head,
 605                                       b_assoc_buffers);
 606                 list_del_init(&bh->b_assoc_buffers);
 607                 brelse(bh); /* drop ref-count to make it releasable */
 608         }
 609 }
 610
 611 /**
 612  * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
 613  * @inode: inode of the metadata file
 614  */
 615 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
 616 {
 617         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 618         struct nilfs_inode_info *ii = NILFS_I(inode);
 619         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 620
 621         down_write(&mi->mi_sem);
 622
 623         if (mi->mi_palloc_cache)
 624                 nilfs_palloc_clear_cache(inode);
 625
 626         nilfs_clear_dirty_pages(inode->i_mapping, true);
 627         nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
 628
 629         nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
 630         nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
 631
 632         nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
 633
 634         up_write(&mi->mi_sem);
 635 }
 636
 637 /**
 638  * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
 639  * @inode: inode of the metadata file
 640  */
 641 void nilfs_mdt_clear_shadow_map(struct inode *inode)
 642 {
 643         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 644         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 645
 646         down_write(&mi->mi_sem);
 647         nilfs_release_frozen_buffers(shadow);
 648         truncate_inode_pages(&shadow->frozen_data, 0);
 649         truncate_inode_pages(&shadow->frozen_btnodes, 0);
 650         up_write(&mi->mi_sem);
 651 }