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, KM_USER0);
  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, KM_USER0);
  67
  68         set_buffer_uptodate(bh);
  69         nilfs_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  * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
 241  *
 242  * %-EROFS - Read only filesystem (for create mode)
 243  */
 244 int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
 245                         void (*init_block)(struct inode *,
 246                                            struct buffer_head *, void *),
 247                         struct buffer_head **out_bh)
 248 {
 249         int ret;
 250
 251         /* Should be rewritten with merging nilfs_mdt_read_block() */
 252  retry:
 253         ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
 254         if (!create || ret != -ENOENT)
 255                 return ret;
 256
 257         ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
 258         if (unlikely(ret == -EEXIST)) {
 259                 /* create = 0; */  /* limit read-create loop retries */
 260                 goto retry;
 261         }
 262         return ret;
 263 }
 264
 265 /**
 266  * nilfs_mdt_delete_block - make a hole on the meta data file.
 267  * @inode: inode of the meta data file
 268  * @block: block offset
 269  *
 270  * Return Value: On success, zero is returned.
 271  * On error, one of the following negative error code is returned.
 272  *
 273  * %-ENOMEM - Insufficient memory available.
 274  *
 275  * %-EIO - I/O error
 276  *
 277  * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
 278  */
 279 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
 280 {
 281         struct nilfs_inode_info *ii = NILFS_I(inode);
 282         int err;
 283
 284         err = nilfs_bmap_delete(ii->i_bmap, block);
 285         if (!err || err == -ENOENT) {
 286                 nilfs_mdt_mark_dirty(inode);
 287                 nilfs_mdt_forget_block(inode, block);
 288         }
 289         return err;
 290 }
 291
 292 /**
 293  * nilfs_mdt_forget_block - discard dirty state and try to remove the page
 294  * @inode: inode of the meta data file
 295  * @block: block offset
 296  *
 297  * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
 298  * tries to release the page including the buffer from a page cache.
 299  *
 300  * Return Value: On success, 0 is returned. On error, one of the following
 301  * negative error code is returned.
 302  *
 303  * %-EBUSY - page has an active buffer.
 304  *
 305  * %-ENOENT - page cache has no page addressed by the offset.
 306  */
 307 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
 308 {
 309         pgoff_t index = (pgoff_t)block >>
 310                 (PAGE_CACHE_SHIFT - inode->i_blkbits);
 311         struct page *page;
 312         unsigned long first_block;
 313         int ret = 0;
 314         int still_dirty;
 315
 316         page = find_lock_page(inode->i_mapping, index);
 317         if (!page)
 318                 return -ENOENT;
 319
 320         wait_on_page_writeback(page);
 321
 322         first_block = (unsigned long)index <<
 323                 (PAGE_CACHE_SHIFT - inode->i_blkbits);
 324         if (page_has_buffers(page)) {
 325                 struct buffer_head *bh;
 326
 327                 bh = nilfs_page_get_nth_block(page, block - first_block);
 328                 nilfs_forget_buffer(bh);
 329         }
 330         still_dirty = PageDirty(page);
 331         unlock_page(page);
 332         page_cache_release(page);
 333
 334         if (still_dirty ||
 335             invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
 336                 ret = -EBUSY;
 337         return ret;
 338 }
 339
 340 /**
 341  * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
 342  * @inode: inode of the meta data file
 343  * @block: block offset
 344  *
 345  * Return Value: On success, it returns 0. On error, the following negative
 346  * error code is returned.
 347  *
 348  * %-ENOMEM - Insufficient memory available.
 349  *
 350  * %-EIO - I/O error
 351  *
 352  * %-ENOENT - the specified block does not exist (hole block)
 353  *
 354  * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
 355  */
 356 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
 357 {
 358         struct buffer_head *bh;
 359         int err;
 360
 361         err = nilfs_mdt_read_block(inode, block, 0, &bh);
 362         if (unlikely(err))
 363                 return err;
 364         nilfs_mark_buffer_dirty(bh);
 365         nilfs_mdt_mark_dirty(inode);
 366         brelse(bh);
 367         return 0;
 368 }
 369
 370 int nilfs_mdt_fetch_dirty(struct inode *inode)
 371 {
 372         struct nilfs_inode_info *ii = NILFS_I(inode);
 373
 374         if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
 375                 set_bit(NILFS_I_DIRTY, &ii->i_state);
 376                 return 1;
 377         }
 378         return test_bit(NILFS_I_DIRTY, &ii->i_state);
 379 }
 380
 381 static int
 382 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
 383 {
 384         struct inode *inode;
 385         struct super_block *sb;
 386         int err = 0;
 387
 388         redirty_page_for_writepage(wbc, page);
 389         unlock_page(page);
 390
 391         inode = page->mapping->host;
 392         if (!inode)
 393                 return 0;
 394
 395         sb = inode->i_sb;
 396
 397         if (wbc->sync_mode == WB_SYNC_ALL)
 398                 err = nilfs_construct_segment(sb);
 399         else if (wbc->for_reclaim)
 400                 nilfs_flush_segment(sb, inode->i_ino);
 401
 402         return err;
 403 }
 404
 405
 406 static const struct address_space_operations def_mdt_aops = {
 407         .writepage              = nilfs_mdt_write_page,
 408         .sync_page              = block_sync_page,
 409 };
 410
 411 static const struct inode_operations def_mdt_iops;
 412 static const struct file_operations def_mdt_fops;
 413
 414
 415 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
 416 {
 417         struct nilfs_mdt_info *mi;
 418
 419         mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
 420         if (!mi)
 421                 return -ENOMEM;
 422
 423         init_rwsem(&mi->mi_sem);
 424         inode->i_private = mi;
 425
 426         inode->i_mode = S_IFREG;
 427         mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
 428         inode->i_mapping->backing_dev_info = inode->i_sb->s_bdi;
 429
 430         inode->i_op = &def_mdt_iops;
 431         inode->i_fop = &def_mdt_fops;
 432         inode->i_mapping->a_ops = &def_mdt_aops;
 433
 434         return 0;
 435 }
 436
 437 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
 438                               unsigned header_size)
 439 {
 440         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 441
 442         mi->mi_entry_size = entry_size;
 443         mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
 444         mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
 445 }
 446
 447 static const struct address_space_operations shadow_map_aops = {
 448         .sync_page              = block_sync_page,
 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         nilfs_mapping_init_once(&shadow->frozen_data);
 464         nilfs_mapping_init(&shadow->frozen_data, bdi, &shadow_map_aops);
 465         nilfs_mapping_init_once(&shadow->frozen_btnodes);
 466         nilfs_mapping_init(&shadow->frozen_btnodes, bdi, &shadow_map_aops);
 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         int ret = -ENOMEM;
 503
 504         page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
 505         if (!page)
 506                 return ret;
 507
 508         if (!page_has_buffers(page))
 509                 create_empty_buffers(page, 1 << blkbits, 0);
 510
 511         bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
 512         if (bh_frozen) {
 513                 if (!buffer_uptodate(bh_frozen))
 514                         nilfs_copy_buffer(bh_frozen, bh);
 515                 if (list_empty(&bh_frozen->b_assoc_buffers)) {
 516                         list_add_tail(&bh_frozen->b_assoc_buffers,
 517                                       &shadow->frozen_buffers);
 518                         set_buffer_nilfs_redirected(bh);
 519                 } else {
 520                         brelse(bh_frozen); /* already frozen */
 521                 }
 522                 ret = 0;
 523         }
 524         unlock_page(page);
 525         page_cache_release(page);
 526         return ret;
 527 }
 528
 529 struct buffer_head *
 530 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
 531 {
 532         struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
 533         struct buffer_head *bh_frozen = NULL;
 534         struct page *page;
 535         int n;
 536
 537         page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
 538         if (page) {
 539                 if (page_has_buffers(page)) {
 540                         n = bh_offset(bh) >> inode->i_blkbits;
 541                         bh_frozen = nilfs_page_get_nth_block(page, n);
 542                 }
 543                 unlock_page(page);
 544                 page_cache_release(page);
 545         }
 546         return bh_frozen;
 547 }
 548
 549 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
 550 {
 551         struct list_head *head = &shadow->frozen_buffers;
 552         struct buffer_head *bh;
 553
 554         while (!list_empty(head)) {
 555                 bh = list_first_entry(head, struct buffer_head,
 556                                       b_assoc_buffers);
 557                 list_del_init(&bh->b_assoc_buffers);
 558                 brelse(bh); /* drop ref-count to make it releasable */
 559         }
 560 }
 561
 562 /**
 563  * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
 564  * @inode: inode of the metadata file
 565  */
 566 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
 567 {
 568         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 569         struct nilfs_inode_info *ii = NILFS_I(inode);
 570         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 571
 572         down_write(&mi->mi_sem);
 573
 574         if (mi->mi_palloc_cache)
 575                 nilfs_palloc_clear_cache(inode);
 576
 577         nilfs_clear_dirty_pages(inode->i_mapping);
 578         nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
 579
 580         nilfs_clear_dirty_pages(&ii->i_btnode_cache);
 581         nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
 582
 583         nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
 584
 585         up_write(&mi->mi_sem);
 586 }
 587
 588 /**
 589  * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
 590  * @inode: inode of the metadata file
 591  */
 592 void nilfs_mdt_clear_shadow_map(struct inode *inode)
 593 {
 594         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 595         struct nilfs_shadow_map *shadow = mi->mi_shadow;
 596
 597         down_write(&mi->mi_sem);
 598         nilfs_release_frozen_buffers(shadow);
 599         truncate_inode_pages(&shadow->frozen_data, 0);
 600         truncate_inode_pages(&shadow->frozen_btnodes, 0);
 601         up_write(&mi->mi_sem);
 602 }