2 * mdt.c - meta data file for NILFS
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
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.
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.
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
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
23 #include <linux/buffer_head.h>
24 #include <linux/mpage.h>
26 #include <linux/writeback.h>
27 #include <linux/backing-dev.h>
28 #include <linux/swap.h>
29 #include <linux/slab.h>
36 #include <trace/events/nilfs2.h>
38 #define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
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 *))
47 struct nilfs_inode_info *ii = NILFS_I(inode);
51 /* Caller exclude read accesses using page lock */
53 /* set_buffer_new(bh); */
56 ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
60 set_buffer_mapped(bh);
62 kaddr = kmap_atomic(bh->b_page);
63 memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
65 init_block(inode, bh, kaddr);
66 flush_dcache_page(bh->b_page);
69 set_buffer_uptodate(bh);
70 mark_buffer_dirty(bh);
71 nilfs_mdt_mark_dirty(inode);
73 trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
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 *,
84 struct super_block *sb = inode->i_sb;
85 struct nilfs_transaction_info ti;
86 struct buffer_head *bh;
89 nilfs_transaction_begin(sb, &ti, 0);
92 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
97 if (buffer_uptodate(bh))
101 if (buffer_uptodate(bh))
104 bh->b_bdev = sb->s_bdev;
105 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
112 unlock_page(bh->b_page);
113 put_page(bh->b_page);
118 err = nilfs_transaction_commit(sb);
120 nilfs_transaction_abort(sb);
126 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
127 int mode, struct buffer_head **out_bh)
129 struct buffer_head *bh;
133 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
137 ret = -EEXIST; /* internal code */
138 if (buffer_uptodate(bh))
142 if (!trylock_buffer(bh)) {
146 } else /* mode == READ */
149 if (buffer_uptodate(bh)) {
154 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
159 map_bh(bh, inode->i_sb, (sector_t)blknum);
161 bh->b_end_io = end_buffer_read_sync;
166 trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff, mode);
172 unlock_page(bh->b_page);
173 put_page(bh->b_page);
179 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
180 int readahead, struct buffer_head **out_bh)
182 struct buffer_head *first_bh, *bh;
183 unsigned long blkoff;
184 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
187 err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
188 if (err == -EEXIST) /* internal code */
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))
200 else if (err != -EBUSY)
202 /* abort readahead if bmap lookup failed */
203 if (!buffer_locked(first_bh))
208 wait_on_buffer(first_bh);
212 if (!buffer_uptodate(first_bh))
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
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.
237 * Return Value: On success, it returns 0. On error, the following negative
238 * error code is returned.
240 * %-ENOMEM - Insufficient memory available.
244 * %-ENOENT - the specified block does not exist (hole block)
246 * %-EROFS - Read only filesystem (for create mode)
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)
255 /* Should be rewritten with merging nilfs_mdt_read_block() */
257 ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
258 if (!create || ret != -ENOENT)
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 */
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
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.
282 * Return Value: On success, it returns 0. On error, the following negative
283 * error code is returned.
285 * %-ENOMEM - Insufficient memory available.
289 * %-ENOENT - no block was found in the range
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)
298 if (unlikely(start > end))
301 ret = nilfs_mdt_read_block(inode, start, true, out_bh);
306 if (unlikely(ret != -ENOENT || start == ULONG_MAX))
309 ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
312 ret = nilfs_mdt_read_block(inode, next, true, out_bh);
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
328 * Return Value: On success, zero is returned.
329 * On error, one of the following negative error code is returned.
331 * %-ENOMEM - Insufficient memory available.
335 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
337 struct nilfs_inode_info *ii = NILFS_I(inode);
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);
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
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.
356 * Return Value: On success, 0 is returned. On error, one of the following
357 * negative error code is returned.
359 * %-EBUSY - page has an active buffer.
361 * %-ENOENT - page cache has no page addressed by the offset.
363 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
365 pgoff_t index = (pgoff_t)block >>
366 (PAGE_SHIFT - inode->i_blkbits);
368 unsigned long first_block;
372 page = find_lock_page(inode->i_mapping, index);
376 wait_on_page_writeback(page);
378 first_block = (unsigned long)index <<
379 (PAGE_SHIFT - inode->i_blkbits);
380 if (page_has_buffers(page)) {
381 struct buffer_head *bh;
383 bh = nilfs_page_get_nth_block(page, block - first_block);
384 nilfs_forget_buffer(bh);
386 still_dirty = PageDirty(page);
391 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
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
401 * Return Value: On success, it returns 0. On error, the following negative
402 * error code is returned.
404 * %-ENOMEM - Insufficient memory available.
408 * %-ENOENT - the specified block does not exist (hole block)
410 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
412 struct buffer_head *bh;
415 err = nilfs_mdt_read_block(inode, block, 0, &bh);
418 mark_buffer_dirty(bh);
419 nilfs_mdt_mark_dirty(inode);
424 int nilfs_mdt_fetch_dirty(struct inode *inode)
426 struct nilfs_inode_info *ii = NILFS_I(inode);
428 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
429 set_bit(NILFS_I_DIRTY, &ii->i_state);
432 return test_bit(NILFS_I_DIRTY, &ii->i_state);
436 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
438 struct inode *inode = page->mapping->host;
439 struct super_block *sb;
442 if (inode && (inode->i_sb->s_flags & MS_RDONLY)) {
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.
449 nilfs_clear_dirty_page(page, false);
454 redirty_page_for_writepage(wbc, page);
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);
471 static const struct address_space_operations def_mdt_aops = {
472 .writepage = nilfs_mdt_write_page,
475 static const struct inode_operations def_mdt_iops;
476 static const struct file_operations def_mdt_fops;
479 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
481 struct nilfs_mdt_info *mi;
483 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
487 init_rwsem(&mi->mi_sem);
488 inode->i_private = mi;
490 inode->i_mode = S_IFREG;
491 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
493 inode->i_op = &def_mdt_iops;
494 inode->i_fop = &def_mdt_fops;
495 inode->i_mapping->a_ops = &def_mdt_aops;
500 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
501 unsigned header_size)
503 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
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);
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
515 int nilfs_mdt_setup_shadow_map(struct inode *inode,
516 struct nilfs_shadow_map *shadow)
518 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
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;
530 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
531 * @inode: inode of the metadata file
533 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
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;
540 ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
544 ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
545 &ii->i_btnode_cache);
549 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
554 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
556 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
557 struct buffer_head *bh_frozen;
559 int blkbits = inode->i_blkbits;
561 page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
565 if (!page_has_buffers(page))
566 create_empty_buffers(page, 1 << blkbits, 0);
568 bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
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);
577 brelse(bh_frozen); /* already frozen */
586 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
588 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
589 struct buffer_head *bh_frozen = NULL;
593 page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
595 if (page_has_buffers(page)) {
596 n = bh_offset(bh) >> inode->i_blkbits;
597 bh_frozen = nilfs_page_get_nth_block(page, n);
605 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
607 struct list_head *head = &shadow->frozen_buffers;
608 struct buffer_head *bh;
610 while (!list_empty(head)) {
611 bh = list_first_entry(head, struct buffer_head,
613 list_del_init(&bh->b_assoc_buffers);
614 brelse(bh); /* drop ref-count to make it releasable */
619 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
620 * @inode: inode of the metadata file
622 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
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;
628 down_write(&mi->mi_sem);
630 if (mi->mi_palloc_cache)
631 nilfs_palloc_clear_cache(inode);
633 nilfs_clear_dirty_pages(inode->i_mapping, true);
634 nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
636 nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
637 nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
639 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
641 up_write(&mi->mi_sem);
645 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
646 * @inode: inode of the metadata file
648 void nilfs_mdt_clear_shadow_map(struct inode *inode)
650 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
651 struct nilfs_shadow_map *shadow = mi->mi_shadow;
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);