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_delete_block - make a hole on the meta data file.
 265  * @inode: inode of the meta data file
 266  * @block: block offset
 267  *
 268  * Return Value: On success, zero is returned.
 269  * On error, one of the following negative error code is returned.
 270  *
 271  * %-ENOMEM - Insufficient memory available.
 272  *
 273  * %-EIO - I/O error
 274  */
 275 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
 276 {
 277         struct nilfs_inode_info *ii = NILFS_I(inode);
 278         int err;
 279
 280         err = nilfs_bmap_delete(ii->i_bmap, block);
 281         if (!err || err == -ENOENT) {
 282                 nilfs_mdt_mark_dirty(inode);
 283                 nilfs_mdt_forget_block(inode, block);
 284         }
 285         return err;
 286 }
 287
 288 /**
 289  * nilfs_mdt_forget_block - discard dirty state and try to remove the page
 290  * @inode: inode of the meta data file
 291  * @block: block offset
 292  *
 293  * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
 294  * tries to release the page including the buffer from a page cache.
 295  *
 296  * Return Value: On success, 0 is returned. On error, one of the following
 297  * negative error code is returned.
 298  *
 299  * %-EBUSY - page has an active buffer.
 300  *
 301  * %-ENOENT - page cache has no page addressed by the offset.
 302  */
 303 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
 304 {
 305         pgoff_t index = (pgoff_t)block >>
 306                 (PAGE_CACHE_SHIFT - inode->i_blkbits);
 307         struct page *page;
 308         unsigned long first_block;
 309         int ret = 0;
 310         int still_dirty;
 311
 312         page = find_lock_page(inode->i_mapping, index);
 313         if (!page)
 314                 return -ENOENT;
 315
 316         wait_on_page_writeback(page);
 317
 318         first_block = (unsigned long)index <<
 319                 (PAGE_CACHE_SHIFT - inode->i_blkbits);
 320         if (page_has_buffers(page)) {
 321                 struct buffer_head *bh;
 322
 323                 bh = nilfs_page_get_nth_block(page, block - first_block);
 324                 nilfs_forget_buffer(bh);
 325         }
 326         still_dirty = PageDirty(page);
 327         unlock_page(page);
 328         page_cache_release(page);
 329
 330         if (still_dirty ||
 331             invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
 332                 ret = -EBUSY;
 333         return ret;
 334 }
 335
 336 /**
 337  * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
 338  * @inode: inode of the meta data file
 339  * @block: block offset
 340  *
 341  * Return Value: On success, it returns 0. On error, the following negative
 342  * error code is returned.
 343  *
 344  * %-ENOMEM - Insufficient memory available.
 345  *
 346  * %-EIO - I/O error
 347  *
 348  * %-ENOENT - the specified block does not exist (hole block)
 349  */
 350 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
 351 {
 352         struct buffer_head *bh;
 353         int err;
 354
 355         err = nilfs_mdt_read_block(inode, block, 0, &bh);
 356         if (unlikely(err))
 357                 return err;
 358         mark_buffer_dirty(bh);
 359         nilfs_mdt_mark_dirty(inode);
 360         brelse(bh);
 361         return 0;
 362 }
 363
 364 int nilfs_mdt_fetch_dirty(struct inode *inode)
 365 {
 366         struct nilfs_inode_info *ii = NILFS_I(inode);
 367
 368         if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
 369                 set_bit(NILFS_I_DIRTY, &ii->i_state);
 370                 return 1;
 371         }
 372         return test_bit(NILFS_I_DIRTY, &ii->i_state);
 373 }
 374
 375 static int
 376 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
 377 {
 378         struct inode *inode = page->mapping->host;
 379         struct super_block *sb;
 380         int err = 0;
 381
 382         if (inode && (inode->i_sb->s_flags & MS_RDONLY)) {
 383                 /*
 384                  * It means that filesystem was remounted in read-only
 385                  * mode because of error or metadata corruption. But we
 386                  * have dirty pages that try to be flushed in background.
 387                  * So, here we simply discard this dirty page.
 388                  */
 389                 nilfs_clear_dirty_page(page, false);
 390                 unlock_page(page);
 391                 return -EROFS;
 392         }
 393
 394         redirty_page_for_writepage(wbc, page);
 395         unlock_page(page);
 396
 397         if (!inode)
 398                 return 0;
 399
 400         sb = inode->i_sb;
 401
 402         if (wbc->sync_mode == WB_SYNC_ALL)
 403                 err = nilfs_construct_segment(sb);
 404         else if (wbc->for_reclaim)
 405                 nilfs_flush_segment(sb, inode->i_ino);
 406
 407         return err;
 408 }
 409
 410
 411 static const struct address_space_operations def_mdt_aops = {
 412         .writepage              = nilfs_mdt_write_page,
 413 };
 414
 415 static const struct inode_operations def_mdt_iops;
 416 static const struct file_operations def_mdt_fops;
 417
 418
 419 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
 420 {
 421         struct nilfs_mdt_info *mi;
 422
 423         mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
 424         if (!mi)
 425                 return -ENOMEM;
 426
 427         init_rwsem(&mi->mi_sem);
 428         inode->i_private = mi;
 429
 430         inode->i_mode = S_IFREG;
 431         mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
 432         inode->i_mapping->backing_dev_info = inode->i_sb->s_bdi;
 433
 434         inode->i_op = &def_mdt_iops;
 435         inode->i_fop = &def_mdt_fops;
 436         inode->i_mapping->a_ops = &def_mdt_aops;
 437
 438         return 0;
 439 }
 440
 441 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
 442                               unsigned header_size)
 443 {
 444         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 445
 446         mi->mi_entry_size = entry_size;
 447         mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
 448         mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
 449 }
 450
 451 /**
 452  * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
 453  * @inode: inode of the metadata file
 454  * @shadow: shadow mapping
 455  */
 456 int nilfs_mdt_setup_shadow_map(struct inode *inode,
 457                                struct nilfs_shadow_map *shadow)
 458 {
 459         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 460         struct backing_dev_info *bdi = inode->i_sb->s_bdi;
 461
 462         INIT_LIST_HEAD(&shadow->frozen_buffers);
 463         address_space_init_once(&shadow->frozen_data);
 464         nilfs_mapping_init(&shadow->frozen_data, inode, bdi);
 465         address_space_init_once(&shadow->frozen_btnodes);
 466         nilfs_mapping_init(&shadow->frozen_btnodes, inode, bdi);
 467         mi->mi_shadow = shadow;
 468         return 0;
 469 }
 470
 471 /**
 472  * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
 473  * @inode: inode of the metadata file
 474  */
 475 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
 476 {
 477         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 478         struct nilfs_inode_info *ii = NILFS_I(inode);
 479         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 480         int ret;
 481
 482         ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
 483         if (ret)
 484                 goto out;
 485
 486         ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
 487                                      &ii->i_btnode_cache);
 488         if (ret)
 489                 goto out;
 490
 491         nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
 492  out:
 493         return ret;
 494 }
 495
 496 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
 497 {
 498         struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
 499         struct buffer_head *bh_frozen;
 500         struct page *page;
 501         int blkbits = inode->i_blkbits;
 502
 503         page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
 504         if (!page)
 505                 return -ENOMEM;
 506
 507         if (!page_has_buffers(page))
 508                 create_empty_buffers(page, 1 << blkbits, 0);
 509
 510         bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
 511
 512         if (!buffer_uptodate(bh_frozen))
 513                 nilfs_copy_buffer(bh_frozen, bh);
 514         if (list_empty(&bh_frozen->b_assoc_buffers)) {
 515                 list_add_tail(&bh_frozen->b_assoc_buffers,
 516                               &shadow->frozen_buffers);
 517                 set_buffer_nilfs_redirected(bh);
 518         } else {
 519                 brelse(bh_frozen); /* already frozen */
 520         }
 521
 522         unlock_page(page);
 523         page_cache_release(page);
 524         return 0;
 525 }
 526
 527 struct buffer_head *
 528 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
 529 {
 530         struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
 531         struct buffer_head *bh_frozen = NULL;
 532         struct page *page;
 533         int n;
 534
 535         page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
 536         if (page) {
 537                 if (page_has_buffers(page)) {
 538                         n = bh_offset(bh) >> inode->i_blkbits;
 539                         bh_frozen = nilfs_page_get_nth_block(page, n);
 540                 }
 541                 unlock_page(page);
 542                 page_cache_release(page);
 543         }
 544         return bh_frozen;
 545 }
 546
 547 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
 548 {
 549         struct list_head *head = &shadow->frozen_buffers;
 550         struct buffer_head *bh;
 551
 552         while (!list_empty(head)) {
 553                 bh = list_first_entry(head, struct buffer_head,
 554                                       b_assoc_buffers);
 555                 list_del_init(&bh->b_assoc_buffers);
 556                 brelse(bh); /* drop ref-count to make it releasable */
 557         }
 558 }
 559
 560 /**
 561  * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
 562  * @inode: inode of the metadata file
 563  */
 564 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
 565 {
 566         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 567         struct nilfs_inode_info *ii = NILFS_I(inode);
 568         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 569
 570         down_write(&mi->mi_sem);
 571
 572         if (mi->mi_palloc_cache)
 573                 nilfs_palloc_clear_cache(inode);
 574
 575         nilfs_clear_dirty_pages(inode->i_mapping, true);
 576         nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
 577
 578         nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
 579         nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
 580
 581         nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
 582
 583         up_write(&mi->mi_sem);
 584 }
 585
 586 /**
 587  * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
 588  * @inode: inode of the metadata file
 589  */
 590 void nilfs_mdt_clear_shadow_map(struct inode *inode)
 591 {
 592         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 593         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 594
 595         down_write(&mi->mi_sem);
 596         nilfs_release_frozen_buffers(shadow);
 597         truncate_inode_pages(&shadow->frozen_data, 0);
 598         truncate_inode_pages(&shadow->frozen_btnodes, 0);
 599         up_write(&mi->mi_sem);
 600 }