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 * Written by Ryusuke Konishi <ryusuke@osrg.net>
19 #include <linux/buffer_head.h>
20 #include <linux/mpage.h>
22 #include <linux/writeback.h>
23 #include <linux/backing-dev.h>
24 #include <linux/swap.h>
25 #include <linux/slab.h>
32 #include <trace/events/nilfs2.h>
34 #define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
38 nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
39 struct buffer_head *bh,
40 void (*init_block)(struct inode *,
41 struct buffer_head *, void *))
43 struct nilfs_inode_info *ii = NILFS_I(inode);
47 /* Caller exclude read accesses using page lock */
49 /* set_buffer_new(bh); */
52 ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
56 set_buffer_mapped(bh);
58 kaddr = kmap_atomic(bh->b_page);
59 memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
61 init_block(inode, bh, kaddr);
62 flush_dcache_page(bh->b_page);
65 set_buffer_uptodate(bh);
66 mark_buffer_dirty(bh);
67 nilfs_mdt_mark_dirty(inode);
69 trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
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 *,
80 struct super_block *sb = inode->i_sb;
81 struct nilfs_transaction_info ti;
82 struct buffer_head *bh;
85 nilfs_transaction_begin(sb, &ti, 0);
88 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
93 if (buffer_uptodate(bh))
97 if (buffer_uptodate(bh))
100 bh->b_bdev = sb->s_bdev;
101 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
108 unlock_page(bh->b_page);
109 put_page(bh->b_page);
114 err = nilfs_transaction_commit(sb);
116 nilfs_transaction_abort(sb);
122 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
123 int mode, struct buffer_head **out_bh)
125 struct buffer_head *bh;
129 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
133 ret = -EEXIST; /* internal code */
134 if (buffer_uptodate(bh))
138 if (!trylock_buffer(bh)) {
142 } else /* mode == READ */
145 if (buffer_uptodate(bh)) {
150 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
155 map_bh(bh, inode->i_sb, (sector_t)blknum);
157 bh->b_end_io = end_buffer_read_sync;
162 trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff, mode);
168 unlock_page(bh->b_page);
169 put_page(bh->b_page);
175 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
176 int readahead, struct buffer_head **out_bh)
178 struct buffer_head *first_bh, *bh;
179 unsigned long blkoff;
180 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
183 err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
184 if (err == -EEXIST) /* internal code */
192 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
193 err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
194 if (likely(!err || err == -EEXIST))
196 else if (err != -EBUSY)
198 /* abort readahead if bmap lookup failed */
199 if (!buffer_locked(first_bh))
204 wait_on_buffer(first_bh);
208 if (!buffer_uptodate(first_bh))
221 * nilfs_mdt_get_block - read or create a buffer on meta data file.
222 * @inode: inode of the meta data file
223 * @blkoff: block offset
224 * @create: create flag
225 * @init_block: initializer used for newly allocated block
226 * @out_bh: output of a pointer to the buffer_head
228 * nilfs_mdt_get_block() looks up the specified buffer and tries to create
229 * a new buffer if @create is not zero. On success, the returned buffer is
230 * assured to be either existing or formatted using a buffer lock on success.
231 * @out_bh is substituted only when zero is returned.
233 * Return Value: On success, it returns 0. On error, the following negative
234 * error code is returned.
236 * %-ENOMEM - Insufficient memory available.
240 * %-ENOENT - the specified block does not exist (hole block)
242 * %-EROFS - Read only filesystem (for create mode)
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)
251 /* Should be rewritten with merging nilfs_mdt_read_block() */
253 ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
254 if (!create || ret != -ENOENT)
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 */
266 * nilfs_mdt_find_block - find and get a buffer on meta data file.
267 * @inode: inode of the meta data file
268 * @start: start block offset (inclusive)
269 * @end: end block offset (inclusive)
270 * @blkoff: block offset
271 * @out_bh: place to store a pointer to buffer_head struct
273 * nilfs_mdt_find_block() looks up an existing block in range of
274 * [@start, @end] and stores pointer to a buffer head of the block to
275 * @out_bh, and block offset to @blkoff, respectively. @out_bh and
276 * @blkoff are substituted only when zero is returned.
278 * Return Value: On success, it returns 0. On error, the following negative
279 * error code is returned.
281 * %-ENOMEM - Insufficient memory available.
285 * %-ENOENT - no block was found in the range
287 int nilfs_mdt_find_block(struct inode *inode, unsigned long start,
288 unsigned long end, unsigned long *blkoff,
289 struct buffer_head **out_bh)
294 if (unlikely(start > end))
297 ret = nilfs_mdt_read_block(inode, start, true, out_bh);
302 if (unlikely(ret != -ENOENT || start == ULONG_MAX))
305 ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
308 ret = nilfs_mdt_read_block(inode, next, true, out_bh);
320 * nilfs_mdt_delete_block - make a hole on the meta data file.
321 * @inode: inode of the meta data file
322 * @block: block offset
324 * Return Value: On success, zero is returned.
325 * On error, one of the following negative error code is returned.
327 * %-ENOMEM - Insufficient memory available.
331 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
333 struct nilfs_inode_info *ii = NILFS_I(inode);
336 err = nilfs_bmap_delete(ii->i_bmap, block);
337 if (!err || err == -ENOENT) {
338 nilfs_mdt_mark_dirty(inode);
339 nilfs_mdt_forget_block(inode, block);
345 * nilfs_mdt_forget_block - discard dirty state and try to remove the page
346 * @inode: inode of the meta data file
347 * @block: block offset
349 * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
350 * tries to release the page including the buffer from a page cache.
352 * Return Value: On success, 0 is returned. On error, one of the following
353 * negative error code is returned.
355 * %-EBUSY - page has an active buffer.
357 * %-ENOENT - page cache has no page addressed by the offset.
359 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
361 pgoff_t index = (pgoff_t)block >>
362 (PAGE_SHIFT - inode->i_blkbits);
364 unsigned long first_block;
368 page = find_lock_page(inode->i_mapping, index);
372 wait_on_page_writeback(page);
374 first_block = (unsigned long)index <<
375 (PAGE_SHIFT - inode->i_blkbits);
376 if (page_has_buffers(page)) {
377 struct buffer_head *bh;
379 bh = nilfs_page_get_nth_block(page, block - first_block);
380 nilfs_forget_buffer(bh);
382 still_dirty = PageDirty(page);
387 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
393 * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
394 * @inode: inode of the meta data file
395 * @block: block offset
397 * Return Value: On success, it returns 0. On error, the following negative
398 * error code is returned.
400 * %-ENOMEM - Insufficient memory available.
404 * %-ENOENT - the specified block does not exist (hole block)
406 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
408 struct buffer_head *bh;
411 err = nilfs_mdt_read_block(inode, block, 0, &bh);
414 mark_buffer_dirty(bh);
415 nilfs_mdt_mark_dirty(inode);
420 int nilfs_mdt_fetch_dirty(struct inode *inode)
422 struct nilfs_inode_info *ii = NILFS_I(inode);
424 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
425 set_bit(NILFS_I_DIRTY, &ii->i_state);
428 return test_bit(NILFS_I_DIRTY, &ii->i_state);
432 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
434 struct inode *inode = page->mapping->host;
435 struct super_block *sb;
438 if (inode && (inode->i_sb->s_flags & MS_RDONLY)) {
440 * It means that filesystem was remounted in read-only
441 * mode because of error or metadata corruption. But we
442 * have dirty pages that try to be flushed in background.
443 * So, here we simply discard this dirty page.
445 nilfs_clear_dirty_page(page, false);
450 redirty_page_for_writepage(wbc, page);
458 if (wbc->sync_mode == WB_SYNC_ALL)
459 err = nilfs_construct_segment(sb);
460 else if (wbc->for_reclaim)
461 nilfs_flush_segment(sb, inode->i_ino);
467 static const struct address_space_operations def_mdt_aops = {
468 .writepage = nilfs_mdt_write_page,
471 static const struct inode_operations def_mdt_iops;
472 static const struct file_operations def_mdt_fops;
475 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
477 struct nilfs_mdt_info *mi;
479 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
483 init_rwsem(&mi->mi_sem);
484 inode->i_private = mi;
486 inode->i_mode = S_IFREG;
487 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
489 inode->i_op = &def_mdt_iops;
490 inode->i_fop = &def_mdt_fops;
491 inode->i_mapping->a_ops = &def_mdt_aops;
496 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
497 unsigned header_size)
499 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
501 mi->mi_entry_size = entry_size;
502 mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
503 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
507 * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
508 * @inode: inode of the metadata file
509 * @shadow: shadow mapping
511 int nilfs_mdt_setup_shadow_map(struct inode *inode,
512 struct nilfs_shadow_map *shadow)
514 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
516 INIT_LIST_HEAD(&shadow->frozen_buffers);
517 address_space_init_once(&shadow->frozen_data);
518 nilfs_mapping_init(&shadow->frozen_data, inode);
519 address_space_init_once(&shadow->frozen_btnodes);
520 nilfs_mapping_init(&shadow->frozen_btnodes, inode);
521 mi->mi_shadow = shadow;
526 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
527 * @inode: inode of the metadata file
529 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
531 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
532 struct nilfs_inode_info *ii = NILFS_I(inode);
533 struct nilfs_shadow_map *shadow = mi->mi_shadow;
536 ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
540 ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
541 &ii->i_btnode_cache);
545 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
550 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
552 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
553 struct buffer_head *bh_frozen;
555 int blkbits = inode->i_blkbits;
557 page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
561 if (!page_has_buffers(page))
562 create_empty_buffers(page, 1 << blkbits, 0);
564 bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
566 if (!buffer_uptodate(bh_frozen))
567 nilfs_copy_buffer(bh_frozen, bh);
568 if (list_empty(&bh_frozen->b_assoc_buffers)) {
569 list_add_tail(&bh_frozen->b_assoc_buffers,
570 &shadow->frozen_buffers);
571 set_buffer_nilfs_redirected(bh);
573 brelse(bh_frozen); /* already frozen */
582 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
584 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
585 struct buffer_head *bh_frozen = NULL;
589 page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
591 if (page_has_buffers(page)) {
592 n = bh_offset(bh) >> inode->i_blkbits;
593 bh_frozen = nilfs_page_get_nth_block(page, n);
601 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
603 struct list_head *head = &shadow->frozen_buffers;
604 struct buffer_head *bh;
606 while (!list_empty(head)) {
607 bh = list_first_entry(head, struct buffer_head,
609 list_del_init(&bh->b_assoc_buffers);
610 brelse(bh); /* drop ref-count to make it releasable */
615 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
616 * @inode: inode of the metadata file
618 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
620 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
621 struct nilfs_inode_info *ii = NILFS_I(inode);
622 struct nilfs_shadow_map *shadow = mi->mi_shadow;
624 down_write(&mi->mi_sem);
626 if (mi->mi_palloc_cache)
627 nilfs_palloc_clear_cache(inode);
629 nilfs_clear_dirty_pages(inode->i_mapping, true);
630 nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
632 nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
633 nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
635 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
637 up_write(&mi->mi_sem);
641 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
642 * @inode: inode of the metadata file
644 void nilfs_mdt_clear_shadow_map(struct inode *inode)
646 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
647 struct nilfs_shadow_map *shadow = mi->mi_shadow;
649 down_write(&mi->mi_sem);
650 nilfs_release_frozen_buffers(shadow);
651 truncate_inode_pages(&shadow->frozen_data, 0);
652 truncate_inode_pages(&shadow->frozen_btnodes, 0);
653 up_write(&mi->mi_sem);