1 // SPDX-License-Identifier: GPL-2.0+
3 * mdt.c - 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 unlock_page(bh->b_page);
101 put_page(bh->b_page);
106 err = nilfs_transaction_commit(sb);
108 nilfs_transaction_abort(sb);
114 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
115 int mode, int mode_flags, 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 (mode_flags & REQ_RAHEAD) {
130 if (!trylock_buffer(bh)) {
134 } else /* mode == 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;
151 submit_bh(mode, mode_flags, bh);
154 trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff, mode);
160 unlock_page(bh->b_page);
161 put_page(bh->b_page);
167 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
168 int readahead, struct buffer_head **out_bh)
170 struct buffer_head *first_bh, *bh;
171 unsigned long blkoff;
172 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
175 err = nilfs_mdt_submit_block(inode, block, REQ_OP_READ, 0, &first_bh);
176 if (err == -EEXIST) /* internal code */
184 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
185 err = nilfs_mdt_submit_block(inode, blkoff, REQ_OP_READ,
187 if (likely(!err || err == -EEXIST))
189 else if (err != -EBUSY)
191 /* abort readahead if bmap lookup failed */
192 if (!buffer_locked(first_bh))
197 wait_on_buffer(first_bh);
201 if (!buffer_uptodate(first_bh)) {
202 nilfs_msg(inode->i_sb, KERN_ERR,
203 "I/O error reading meta-data file (ino=%lu, block-offset=%lu)",
204 inode->i_ino, block);
218 * nilfs_mdt_get_block - read or create a buffer on meta data file.
219 * @inode: inode of the meta data file
220 * @blkoff: block offset
221 * @create: create flag
222 * @init_block: initializer used for newly allocated block
223 * @out_bh: output of a pointer to the buffer_head
225 * nilfs_mdt_get_block() looks up the specified buffer and tries to create
226 * a new buffer if @create is not zero. On success, the returned buffer is
227 * assured to be either existing or formatted using a buffer lock on success.
228 * @out_bh is substituted only when zero is returned.
230 * Return Value: On success, it returns 0. On error, the following negative
231 * error code is returned.
233 * %-ENOMEM - Insufficient memory available.
237 * %-ENOENT - the specified block does not exist (hole block)
239 * %-EROFS - Read only filesystem (for create mode)
241 int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
242 void (*init_block)(struct inode *,
243 struct buffer_head *, void *),
244 struct buffer_head **out_bh)
248 /* Should be rewritten with merging nilfs_mdt_read_block() */
250 ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
251 if (!create || ret != -ENOENT)
254 ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
255 if (unlikely(ret == -EEXIST)) {
256 /* create = 0; */ /* limit read-create loop retries */
263 * nilfs_mdt_find_block - find and get a buffer on meta data file.
264 * @inode: inode of the meta data file
265 * @start: start block offset (inclusive)
266 * @end: end block offset (inclusive)
267 * @blkoff: block offset
268 * @out_bh: place to store a pointer to buffer_head struct
270 * nilfs_mdt_find_block() looks up an existing block in range of
271 * [@start, @end] and stores pointer to a buffer head of the block to
272 * @out_bh, and block offset to @blkoff, respectively. @out_bh and
273 * @blkoff are substituted only when zero is returned.
275 * Return Value: On success, it returns 0. On error, the following negative
276 * error code is returned.
278 * %-ENOMEM - Insufficient memory available.
282 * %-ENOENT - no block was found in the range
284 int nilfs_mdt_find_block(struct inode *inode, unsigned long start,
285 unsigned long end, unsigned long *blkoff,
286 struct buffer_head **out_bh)
291 if (unlikely(start > end))
294 ret = nilfs_mdt_read_block(inode, start, true, out_bh);
299 if (unlikely(ret != -ENOENT || start == ULONG_MAX))
302 ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
305 ret = nilfs_mdt_read_block(inode, next, true, out_bh);
317 * nilfs_mdt_delete_block - make a hole on the meta data file.
318 * @inode: inode of the meta data file
319 * @block: block offset
321 * Return Value: On success, zero is returned.
322 * On error, one of the following negative error code is returned.
324 * %-ENOMEM - Insufficient memory available.
328 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
330 struct nilfs_inode_info *ii = NILFS_I(inode);
333 err = nilfs_bmap_delete(ii->i_bmap, block);
334 if (!err || err == -ENOENT) {
335 nilfs_mdt_mark_dirty(inode);
336 nilfs_mdt_forget_block(inode, block);
342 * nilfs_mdt_forget_block - discard dirty state and try to remove the page
343 * @inode: inode of the meta data file
344 * @block: block offset
346 * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
347 * tries to release the page including the buffer from a page cache.
349 * Return Value: On success, 0 is returned. On error, one of the following
350 * negative error code is returned.
352 * %-EBUSY - page has an active buffer.
354 * %-ENOENT - page cache has no page addressed by the offset.
356 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
358 pgoff_t index = (pgoff_t)block >>
359 (PAGE_SHIFT - inode->i_blkbits);
361 unsigned long first_block;
365 page = find_lock_page(inode->i_mapping, index);
369 wait_on_page_writeback(page);
371 first_block = (unsigned long)index <<
372 (PAGE_SHIFT - inode->i_blkbits);
373 if (page_has_buffers(page)) {
374 struct buffer_head *bh;
376 bh = nilfs_page_get_nth_block(page, block - first_block);
377 nilfs_forget_buffer(bh);
379 still_dirty = PageDirty(page);
384 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
389 int nilfs_mdt_fetch_dirty(struct inode *inode)
391 struct nilfs_inode_info *ii = NILFS_I(inode);
393 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
394 set_bit(NILFS_I_DIRTY, &ii->i_state);
397 return test_bit(NILFS_I_DIRTY, &ii->i_state);
401 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
403 struct inode *inode = page->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 pages that try to be flushed in background.
412 * So, here we simply discard this dirty page.
414 nilfs_clear_dirty_page(page, false);
419 redirty_page_for_writepage(wbc, page);
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 .writepage = nilfs_mdt_write_page,
440 static const struct inode_operations def_mdt_iops;
441 static const struct file_operations def_mdt_fops;
444 int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
446 struct nilfs_mdt_info *mi;
448 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
452 init_rwsem(&mi->mi_sem);
453 inode->i_private = mi;
455 inode->i_mode = S_IFREG;
456 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
458 inode->i_op = &def_mdt_iops;
459 inode->i_fop = &def_mdt_fops;
460 inode->i_mapping->a_ops = &def_mdt_aops;
466 * nilfs_mdt_clear - do cleanup for the metadata file
467 * @inode: inode of the metadata file
469 void nilfs_mdt_clear(struct inode *inode)
471 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
473 if (mdi->mi_palloc_cache)
474 nilfs_palloc_destroy_cache(inode);
478 * nilfs_mdt_destroy - release resources used by the metadata file
479 * @inode: inode of the metadata file
481 void nilfs_mdt_destroy(struct inode *inode)
483 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
485 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
489 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
490 unsigned int header_size)
492 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
494 mi->mi_entry_size = entry_size;
495 mi->mi_entries_per_block = i_blocksize(inode) / entry_size;
496 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
500 * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
501 * @inode: inode of the metadata file
502 * @shadow: shadow mapping
504 int nilfs_mdt_setup_shadow_map(struct inode *inode,
505 struct nilfs_shadow_map *shadow)
507 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
509 INIT_LIST_HEAD(&shadow->frozen_buffers);
510 address_space_init_once(&shadow->frozen_data);
511 nilfs_mapping_init(&shadow->frozen_data, inode);
512 address_space_init_once(&shadow->frozen_btnodes);
513 nilfs_mapping_init(&shadow->frozen_btnodes, inode);
514 mi->mi_shadow = shadow;
519 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
520 * @inode: inode of the metadata file
522 int nilfs_mdt_save_to_shadow_map(struct inode *inode)
524 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
525 struct nilfs_inode_info *ii = NILFS_I(inode);
526 struct nilfs_shadow_map *shadow = mi->mi_shadow;
529 ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
533 ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
534 &ii->i_btnode_cache);
538 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
543 int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
545 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
546 struct buffer_head *bh_frozen;
548 int blkbits = inode->i_blkbits;
550 page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
554 if (!page_has_buffers(page))
555 create_empty_buffers(page, 1 << blkbits, 0);
557 bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
559 if (!buffer_uptodate(bh_frozen))
560 nilfs_copy_buffer(bh_frozen, bh);
561 if (list_empty(&bh_frozen->b_assoc_buffers)) {
562 list_add_tail(&bh_frozen->b_assoc_buffers,
563 &shadow->frozen_buffers);
564 set_buffer_nilfs_redirected(bh);
566 brelse(bh_frozen); /* already frozen */
575 nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
577 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
578 struct buffer_head *bh_frozen = NULL;
582 page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
584 if (page_has_buffers(page)) {
585 n = bh_offset(bh) >> inode->i_blkbits;
586 bh_frozen = nilfs_page_get_nth_block(page, n);
594 static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
596 struct list_head *head = &shadow->frozen_buffers;
597 struct buffer_head *bh;
599 while (!list_empty(head)) {
600 bh = list_first_entry(head, struct buffer_head,
602 list_del_init(&bh->b_assoc_buffers);
603 brelse(bh); /* drop ref-count to make it releasable */
608 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
609 * @inode: inode of the metadata file
611 void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
613 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
614 struct nilfs_inode_info *ii = NILFS_I(inode);
615 struct nilfs_shadow_map *shadow = mi->mi_shadow;
617 down_write(&mi->mi_sem);
619 if (mi->mi_palloc_cache)
620 nilfs_palloc_clear_cache(inode);
622 nilfs_clear_dirty_pages(inode->i_mapping, true);
623 nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
625 nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
626 nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
628 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
630 up_write(&mi->mi_sem);
634 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
635 * @inode: inode of the metadata file
637 void nilfs_mdt_clear_shadow_map(struct inode *inode)
639 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
640 struct nilfs_shadow_map *shadow = mi->mi_shadow;
642 down_write(&mi->mi_sem);
643 nilfs_release_frozen_buffers(shadow);
644 truncate_inode_pages(&shadow->frozen_data, 0);
645 truncate_inode_pages(&shadow->frozen_btnodes, 0);
646 up_write(&mi->mi_sem);