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);
39 /* Caller exclude read accesses using page lock */
41 /* set_buffer_new(bh); */
44 ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
48 set_buffer_mapped(bh);
50 kaddr = kmap_atomic(bh->b_page);
51 memset(kaddr + bh_offset(bh), 0, i_blocksize(inode));
53 init_block(inode, bh, kaddr);
54 flush_dcache_page(bh->b_page);
57 set_buffer_uptodate(bh);
58 mark_buffer_dirty(bh);
59 nilfs_mdt_mark_dirty(inode);
61 trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
66 static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
67 struct buffer_head **out_bh,
68 void (*init_block)(struct inode *,
72 struct super_block *sb = inode->i_sb;
73 struct nilfs_transaction_info ti;
74 struct buffer_head *bh;
77 nilfs_transaction_begin(sb, &ti, 0);
80 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
85 if (buffer_uptodate(bh))
89 if (buffer_uptodate(bh))
92 bh->b_bdev = sb->s_bdev;
93 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
100 folio_unlock(bh->b_folio);
101 folio_put(bh->b_folio);
106 err = nilfs_transaction_commit(sb);
108 nilfs_transaction_abort(sb);
114 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf,
115 struct buffer_head **out_bh)
117 struct buffer_head *bh;
121 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
125 ret = -EEXIST; /* internal code */
126 if (buffer_uptodate(bh))
129 if (opf & REQ_RAHEAD) {
130 if (!trylock_buffer(bh)) {
134 } else /* opf == REQ_OP_READ */
137 if (buffer_uptodate(bh)) {
142 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
147 map_bh(bh, inode->i_sb, (sector_t)blknum);
149 bh->b_end_io = end_buffer_read_sync;
154 trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff,
161 folio_unlock(bh->b_folio);
162 folio_put(bh->b_folio);
168 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
169 int readahead, struct buffer_head **out_bh)
171 struct buffer_head *first_bh, *bh;
172 unsigned long blkoff;
173 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
176 err = nilfs_mdt_submit_block(inode, block, REQ_OP_READ, &first_bh);
177 if (err == -EEXIST) /* internal code */
185 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
186 err = nilfs_mdt_submit_block(inode, blkoff,
187 REQ_OP_READ | REQ_RAHEAD, &bh);
188 if (likely(!err || err == -EEXIST))
190 else if (err != -EBUSY)
192 /* abort readahead if bmap lookup failed */
193 if (!buffer_locked(first_bh))
198 wait_on_buffer(first_bh);
202 if (!buffer_uptodate(first_bh)) {
203 nilfs_err(inode->i_sb,
204 "I/O error reading meta-data file (ino=%lu, block-offset=%lu)",
205 inode->i_ino, block);
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
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.
231 * Return Value: On success, it returns 0. On error, the following negative
232 * error code is returned.
234 * %-ENOMEM - Insufficient memory available.
238 * %-ENOENT - the specified block does not exist (hole block)
240 * %-EROFS - Read only filesystem (for create mode)
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)
249 /* Should be rewritten with merging nilfs_mdt_read_block() */
251 ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
252 if (!create || ret != -ENOENT)
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 */
264 * nilfs_mdt_find_block - find and get a buffer on meta data file.
265 * @inode: inode of the meta data file
266 * @start: start block offset (inclusive)
267 * @end: end block offset (inclusive)
268 * @blkoff: block offset
269 * @out_bh: place to store a pointer to buffer_head struct
271 * nilfs_mdt_find_block() looks up an existing block in range of
272 * [@start, @end] and stores pointer to a buffer head of the block to
273 * @out_bh, and block offset to @blkoff, respectively. @out_bh and
274 * @blkoff are substituted only when zero is returned.
276 * Return Value: On success, it returns 0. On error, the following negative
277 * error code is returned.
279 * %-ENOMEM - Insufficient memory available.
283 * %-ENOENT - no block was found in the range
285 int nilfs_mdt_find_block(struct inode *inode, unsigned long start,
286 unsigned long end, unsigned long *blkoff,
287 struct buffer_head **out_bh)
292 if (unlikely(start > end))
295 ret = nilfs_mdt_read_block(inode, start, true, out_bh);
300 if (unlikely(ret != -ENOENT || start == ULONG_MAX))
303 ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
306 ret = nilfs_mdt_read_block(inode, next, true, out_bh);
318 * nilfs_mdt_delete_block - make a hole on the meta data file.
319 * @inode: inode of the meta data file
320 * @block: block offset
322 * Return Value: On success, zero is returned.
323 * On error, one of the following negative error code is returned.
325 * %-ENOMEM - Insufficient memory available.
329 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
331 struct nilfs_inode_info *ii = NILFS_I(inode);
334 err = nilfs_bmap_delete(ii->i_bmap, block);
335 if (!err || err == -ENOENT) {
336 nilfs_mdt_mark_dirty(inode);
337 nilfs_mdt_forget_block(inode, block);
343 * nilfs_mdt_forget_block - discard dirty state and try to remove the page
344 * @inode: inode of the meta data file
345 * @block: block offset
347 * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
348 * tries to release the page including the buffer from a page cache.
350 * Return Value: On success, 0 is returned. On error, one of the following
351 * negative error code is returned.
353 * %-EBUSY - page has an active buffer.
355 * %-ENOENT - page cache has no page addressed by the offset.
357 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
359 pgoff_t index = block >> (PAGE_SHIFT - inode->i_blkbits);
361 struct buffer_head *bh;
365 folio = filemap_lock_folio(inode->i_mapping, index);
369 folio_wait_writeback(folio);
371 bh = folio_buffers(folio);
373 unsigned long first_block = index <<
374 (PAGE_SHIFT - inode->i_blkbits);
375 bh = get_nth_bh(bh, block - first_block);
376 nilfs_forget_buffer(bh);
378 still_dirty = folio_test_dirty(folio);
383 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
388 int nilfs_mdt_fetch_dirty(struct inode *inode)
390 struct nilfs_inode_info *ii = NILFS_I(inode);
392 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
393 set_bit(NILFS_I_DIRTY, &ii->i_state);
396 return test_bit(NILFS_I_DIRTY, &ii->i_state);
400 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
402 struct folio *folio = page_folio(page);
403 struct inode *inode = folio->mapping->host;
404 struct super_block *sb;
407 if (inode && sb_rdonly(inode->i_sb)) {
409 * It means that filesystem was remounted in read-only
410 * mode because of error or metadata corruption. But we
411 * have dirty folios that try to be flushed in background.
412 * So, here we simply discard this dirty folio.
414 nilfs_clear_folio_dirty(folio, false);
419 folio_redirty_for_writepage(wbc, folio);
427 if (wbc->sync_mode == WB_SYNC_ALL)
428 err = nilfs_construct_segment(sb);
429 else if (wbc->for_reclaim)
430 nilfs_flush_segment(sb, inode->i_ino);
436 static const struct address_space_operations def_mdt_aops = {
437 .dirty_folio = block_dirty_folio,
438 .invalidate_folio = block_invalidate_folio,
439 .writepage = nilfs_mdt_write_page,
442 static const struct inode_operations def_mdt_iops;
443 static const struct file_operations def_mdt_fops;
446 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
448 struct nilfs_mdt_info *mi;
450 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
454 init_rwsem(&mi->mi_sem);
455 inode->i_private = mi;
457 inode->i_mode = S_IFREG;
458 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
460 inode->i_op = &def_mdt_iops;
461 inode->i_fop = &def_mdt_fops;
462 inode->i_mapping->a_ops = &def_mdt_aops;
468 * nilfs_mdt_clear - do cleanup for the metadata file
469 * @inode: inode of the metadata file
471 void nilfs_mdt_clear(struct inode *inode)
473 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
474 struct nilfs_shadow_map *shadow = mdi->mi_shadow;
476 if (mdi->mi_palloc_cache)
477 nilfs_palloc_destroy_cache(inode);
480 struct inode *s_inode = shadow->inode;
482 shadow->inode = NULL;
484 mdi->mi_shadow = NULL;
489 * nilfs_mdt_destroy - release resources used by the metadata file
490 * @inode: inode of the metadata file
492 void nilfs_mdt_destroy(struct inode *inode)
494 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
496 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
500 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
501 unsigned int header_size)
503 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
505 mi->mi_entry_size = entry_size;
506 mi->mi_entries_per_block = i_blocksize(inode) / 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);
519 struct inode *s_inode;
521 INIT_LIST_HEAD(&shadow->frozen_buffers);
523 s_inode = nilfs_iget_for_shadow(inode);
525 return PTR_ERR(s_inode);
527 shadow->inode = s_inode;
528 mi->mi_shadow = shadow;
533 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
534 * @inode: inode of the metadata file
536 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
538 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
539 struct nilfs_inode_info *ii = NILFS_I(inode);
540 struct nilfs_shadow_map *shadow = mi->mi_shadow;
541 struct inode *s_inode = shadow->inode;
544 ret = nilfs_copy_dirty_pages(s_inode->i_mapping, inode->i_mapping);
548 ret = nilfs_copy_dirty_pages(NILFS_I(s_inode)->i_assoc_inode->i_mapping,
549 ii->i_assoc_inode->i_mapping);
553 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
558 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
560 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
561 struct buffer_head *bh_frozen;
563 int blkbits = inode->i_blkbits;
565 folio = filemap_grab_folio(shadow->inode->i_mapping,
568 return PTR_ERR(folio);
570 bh_frozen = folio_buffers(folio);
572 bh_frozen = create_empty_buffers(folio, 1 << blkbits, 0);
574 bh_frozen = get_nth_bh(bh_frozen, bh_offset(bh) >> blkbits);
576 if (!buffer_uptodate(bh_frozen))
577 nilfs_copy_buffer(bh_frozen, bh);
578 if (list_empty(&bh_frozen->b_assoc_buffers)) {
579 list_add_tail(&bh_frozen->b_assoc_buffers,
580 &shadow->frozen_buffers);
581 set_buffer_nilfs_redirected(bh);
583 brelse(bh_frozen); /* already frozen */
592 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
594 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
595 struct buffer_head *bh_frozen = NULL;
599 folio = filemap_lock_folio(shadow->inode->i_mapping,
601 if (!IS_ERR(folio)) {
602 bh_frozen = folio_buffers(folio);
604 n = bh_offset(bh) >> inode->i_blkbits;
605 bh_frozen = get_nth_bh(bh_frozen, n);
613 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
615 struct list_head *head = &shadow->frozen_buffers;
616 struct buffer_head *bh;
618 while (!list_empty(head)) {
619 bh = list_first_entry(head, struct buffer_head,
621 list_del_init(&bh->b_assoc_buffers);
622 brelse(bh); /* drop ref-count to make it releasable */
627 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
628 * @inode: inode of the metadata file
630 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
632 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
633 struct nilfs_inode_info *ii = NILFS_I(inode);
634 struct nilfs_shadow_map *shadow = mi->mi_shadow;
636 down_write(&mi->mi_sem);
638 if (mi->mi_palloc_cache)
639 nilfs_palloc_clear_cache(inode);
641 nilfs_clear_dirty_pages(inode->i_mapping, true);
642 nilfs_copy_back_pages(inode->i_mapping, shadow->inode->i_mapping);
644 nilfs_clear_dirty_pages(ii->i_assoc_inode->i_mapping, true);
645 nilfs_copy_back_pages(ii->i_assoc_inode->i_mapping,
646 NILFS_I(shadow->inode)->i_assoc_inode->i_mapping);
648 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
650 up_write(&mi->mi_sem);
654 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
655 * @inode: inode of the metadata file
657 void nilfs_mdt_clear_shadow_map(struct inode *inode)
659 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
660 struct nilfs_shadow_map *shadow = mi->mi_shadow;
661 struct inode *shadow_btnc_inode = NILFS_I(shadow->inode)->i_assoc_inode;
663 down_write(&mi->mi_sem);
664 nilfs_release_frozen_buffers(shadow);
665 truncate_inode_pages(shadow->inode->i_mapping, 0);
666 truncate_inode_pages(shadow_btnc_inode->i_mapping, 0);
667 up_write(&mi->mi_sem);