1 // SPDX-License-Identifier: GPL-2.0+
3 * Meta data file for NILFS
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
10 #include <linux/buffer_head.h>
11 #include <linux/mpage.h>
13 #include <linux/writeback.h>
14 #include <linux/backing-dev.h>
15 #include <linux/swap.h>
16 #include <linux/slab.h>
22 #include "alloc.h" /* nilfs_palloc_destroy_cache() */
24 #include <trace/events/nilfs2.h>
26 #define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
30 nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
31 struct buffer_head *bh,
32 void (*init_block)(struct inode *,
33 struct buffer_head *, void *))
35 struct nilfs_inode_info *ii = NILFS_I(inode);
36 struct folio *folio = bh->b_folio;
40 /* Caller exclude read accesses using page lock */
42 /* set_buffer_new(bh); */
45 ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
49 set_buffer_mapped(bh);
51 /* Initialize block (block size > PAGE_SIZE not yet supported) */
52 from = kmap_local_folio(folio, offset_in_folio(folio, bh->b_data));
53 memset(from, 0, bh->b_size);
55 init_block(inode, bh, from);
58 flush_dcache_folio(folio);
60 set_buffer_uptodate(bh);
61 mark_buffer_dirty(bh);
62 nilfs_mdt_mark_dirty(inode);
64 trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
69 static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
70 struct buffer_head **out_bh,
71 void (*init_block)(struct inode *,
75 struct super_block *sb = inode->i_sb;
76 struct nilfs_transaction_info ti;
77 struct buffer_head *bh;
80 nilfs_transaction_begin(sb, &ti, 0);
83 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
88 if (buffer_uptodate(bh))
92 if (buffer_uptodate(bh))
95 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
102 folio_unlock(bh->b_folio);
103 folio_put(bh->b_folio);
108 err = nilfs_transaction_commit(sb);
110 nilfs_transaction_abort(sb);
116 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf,
117 struct buffer_head **out_bh)
119 struct buffer_head *bh;
123 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
127 ret = -EEXIST; /* internal code */
128 if (buffer_uptodate(bh))
131 if (opf & REQ_RAHEAD) {
132 if (!trylock_buffer(bh)) {
136 } else /* opf == REQ_OP_READ */
139 if (buffer_uptodate(bh)) {
144 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
149 map_bh(bh, inode->i_sb, (sector_t)blknum);
151 bh->b_end_io = end_buffer_read_sync;
156 trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff,
163 folio_unlock(bh->b_folio);
164 folio_put(bh->b_folio);
170 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
171 int readahead, struct buffer_head **out_bh)
173 struct buffer_head *first_bh, *bh;
174 unsigned long blkoff;
175 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
178 err = nilfs_mdt_submit_block(inode, block, REQ_OP_READ, &first_bh);
179 if (err == -EEXIST) /* internal code */
187 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
188 err = nilfs_mdt_submit_block(inode, blkoff,
189 REQ_OP_READ | REQ_RAHEAD, &bh);
190 if (likely(!err || err == -EEXIST))
192 else if (err != -EBUSY)
194 /* abort readahead if bmap lookup failed */
195 if (!buffer_locked(first_bh))
200 wait_on_buffer(first_bh);
204 if (!buffer_uptodate(first_bh)) {
205 nilfs_err(inode->i_sb,
206 "I/O error reading meta-data file (ino=%lu, block-offset=%lu)",
207 inode->i_ino, block);
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 = block >> (PAGE_SHIFT - inode->i_blkbits);
363 struct buffer_head *bh;
367 folio = filemap_lock_folio(inode->i_mapping, index);
371 folio_wait_writeback(folio);
373 bh = folio_buffers(folio);
375 unsigned long first_block = index <<
376 (PAGE_SHIFT - inode->i_blkbits);
377 bh = get_nth_bh(bh, block - first_block);
378 nilfs_forget_buffer(bh);
380 still_dirty = folio_test_dirty(folio);
385 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
390 int nilfs_mdt_fetch_dirty(struct inode *inode)
392 struct nilfs_inode_info *ii = NILFS_I(inode);
394 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
395 set_bit(NILFS_I_DIRTY, &ii->i_state);
398 return test_bit(NILFS_I_DIRTY, &ii->i_state);
401 static int nilfs_mdt_write_folio(struct folio *folio,
402 struct writeback_control *wbc)
404 struct inode *inode = folio->mapping->host;
405 struct super_block *sb;
408 if (inode && sb_rdonly(inode->i_sb)) {
410 * It means that filesystem was remounted in read-only
411 * mode because of error or metadata corruption. But we
412 * have dirty folios that try to be flushed in background.
413 * So, here we simply discard this dirty folio.
415 nilfs_clear_folio_dirty(folio);
420 folio_redirty_for_writepage(wbc, folio);
428 if (wbc->sync_mode == WB_SYNC_ALL)
429 err = nilfs_construct_segment(sb);
430 else if (wbc->for_reclaim)
431 nilfs_flush_segment(sb, inode->i_ino);
436 static int nilfs_mdt_writeback(struct address_space *mapping,
437 struct writeback_control *wbc)
439 struct folio *folio = NULL;
442 while ((folio = writeback_iter(mapping, wbc, folio, &error)))
443 error = nilfs_mdt_write_folio(folio, wbc);
448 static const struct address_space_operations def_mdt_aops = {
449 .dirty_folio = block_dirty_folio,
450 .invalidate_folio = block_invalidate_folio,
451 .writepages = nilfs_mdt_writeback,
452 .migrate_folio = buffer_migrate_folio_norefs,
455 static const struct inode_operations def_mdt_iops;
456 static const struct file_operations def_mdt_fops;
459 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
461 struct nilfs_mdt_info *mi;
463 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
467 init_rwsem(&mi->mi_sem);
468 inode->i_private = mi;
470 inode->i_mode = S_IFREG;
471 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
473 inode->i_op = &def_mdt_iops;
474 inode->i_fop = &def_mdt_fops;
475 inode->i_mapping->a_ops = &def_mdt_aops;
481 * nilfs_mdt_clear - do cleanup for the metadata file
482 * @inode: inode of the metadata file
484 void nilfs_mdt_clear(struct inode *inode)
486 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
487 struct nilfs_shadow_map *shadow = mdi->mi_shadow;
489 if (mdi->mi_palloc_cache)
490 nilfs_palloc_destroy_cache(inode);
493 struct inode *s_inode = shadow->inode;
495 shadow->inode = NULL;
497 mdi->mi_shadow = NULL;
502 * nilfs_mdt_destroy - release resources used by the metadata file
503 * @inode: inode of the metadata file
505 void nilfs_mdt_destroy(struct inode *inode)
507 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
509 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
513 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
514 unsigned int header_size)
516 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
518 mi->mi_entry_size = entry_size;
519 mi->mi_entries_per_block = i_blocksize(inode) / entry_size;
520 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
524 * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
525 * @inode: inode of the metadata file
526 * @shadow: shadow mapping
528 int nilfs_mdt_setup_shadow_map(struct inode *inode,
529 struct nilfs_shadow_map *shadow)
531 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
532 struct inode *s_inode;
534 INIT_LIST_HEAD(&shadow->frozen_buffers);
536 s_inode = nilfs_iget_for_shadow(inode);
538 return PTR_ERR(s_inode);
540 shadow->inode = s_inode;
541 mi->mi_shadow = shadow;
546 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
547 * @inode: inode of the metadata file
549 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
551 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
552 struct nilfs_inode_info *ii = NILFS_I(inode);
553 struct nilfs_shadow_map *shadow = mi->mi_shadow;
554 struct inode *s_inode = shadow->inode;
557 ret = nilfs_copy_dirty_pages(s_inode->i_mapping, inode->i_mapping);
561 ret = nilfs_copy_dirty_pages(NILFS_I(s_inode)->i_assoc_inode->i_mapping,
562 ii->i_assoc_inode->i_mapping);
566 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
571 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
573 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
574 struct buffer_head *bh_frozen;
576 int blkbits = inode->i_blkbits;
578 folio = filemap_grab_folio(shadow->inode->i_mapping,
581 return PTR_ERR(folio);
583 bh_frozen = folio_buffers(folio);
585 bh_frozen = create_empty_buffers(folio, 1 << blkbits, 0);
587 bh_frozen = get_nth_bh(bh_frozen,
588 offset_in_folio(folio, bh->b_data) >> blkbits);
590 if (!buffer_uptodate(bh_frozen))
591 nilfs_copy_buffer(bh_frozen, bh);
592 if (list_empty(&bh_frozen->b_assoc_buffers)) {
593 list_add_tail(&bh_frozen->b_assoc_buffers,
594 &shadow->frozen_buffers);
595 set_buffer_nilfs_redirected(bh);
597 brelse(bh_frozen); /* already frozen */
606 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
608 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
609 struct buffer_head *bh_frozen = NULL;
613 folio = filemap_lock_folio(shadow->inode->i_mapping,
615 if (!IS_ERR(folio)) {
616 bh_frozen = folio_buffers(folio);
618 n = offset_in_folio(folio, bh->b_data) >>
620 bh_frozen = get_nth_bh(bh_frozen, n);
628 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
630 struct list_head *head = &shadow->frozen_buffers;
631 struct buffer_head *bh;
633 while (!list_empty(head)) {
634 bh = list_first_entry(head, struct buffer_head,
636 list_del_init(&bh->b_assoc_buffers);
637 brelse(bh); /* drop ref-count to make it releasable */
642 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
643 * @inode: inode of the metadata file
645 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
647 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
648 struct nilfs_inode_info *ii = NILFS_I(inode);
649 struct nilfs_shadow_map *shadow = mi->mi_shadow;
651 down_write(&mi->mi_sem);
653 if (mi->mi_palloc_cache)
654 nilfs_palloc_clear_cache(inode);
656 nilfs_clear_dirty_pages(inode->i_mapping);
657 nilfs_copy_back_pages(inode->i_mapping, shadow->inode->i_mapping);
659 nilfs_clear_dirty_pages(ii->i_assoc_inode->i_mapping);
660 nilfs_copy_back_pages(ii->i_assoc_inode->i_mapping,
661 NILFS_I(shadow->inode)->i_assoc_inode->i_mapping);
663 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
665 up_write(&mi->mi_sem);
669 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
670 * @inode: inode of the metadata file
672 void nilfs_mdt_clear_shadow_map(struct inode *inode)
674 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
675 struct nilfs_shadow_map *shadow = mi->mi_shadow;
676 struct inode *shadow_btnc_inode = NILFS_I(shadow->inode)->i_assoc_inode;
678 down_write(&mi->mi_sem);
679 nilfs_release_frozen_buffers(shadow);
680 truncate_inode_pages(shadow->inode->i_mapping, 0);
681 truncate_inode_pages(shadow_btnc_inode->i_mapping, 0);
682 up_write(&mi->mi_sem);