mdt.c

  1// SPDX-License-Identifier: GPL-2.0+
  2/*
  3 * Meta data file for NILFS
  4 *
  5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
  6 *
  7 * Written by Ryusuke Konishi.
  8 */
  9
 10#include <linux/buffer_head.h>
 11#include <linux/mpage.h>
 12#include <linux/mm.h>
 13#include <linux/writeback.h>
 14#include <linux/backing-dev.h>
 15#include <linux/swap.h>
 16#include <linux/slab.h>
 17#include "nilfs.h"
 18#include "btnode.h"
 19#include "segment.h"
 20#include "page.h"
 21#include "mdt.h"
 22#include "alloc.h"		/* nilfs_palloc_destroy_cache() */
 23
 24#include <trace/events/nilfs2.h>
 25
 26#define NILFS_MDT_MAX_RA_BLOCKS		(16 - 1)
 27
 28
 29static int
 30nilfs_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 *))
 34{
 35	struct nilfs_inode_info *ii = NILFS_I(inode);
 36	struct folio *folio = bh->b_folio;
 37	void *from;
 38	int ret;
 39
 40	/* Caller exclude read accesses using page lock */
 41
 42	/* set_buffer_new(bh); */
 43	bh->b_blocknr = 0;
 44
 45	ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
 46	if (unlikely(ret))
 47		return ret;
 48
 49	set_buffer_mapped(bh);
 50
 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);
 54	if (init_block)
 55		init_block(inode, bh, from);
 56	kunmap_local(from);
 57
 58	flush_dcache_folio(folio);
 59
 60	set_buffer_uptodate(bh);
 61	mark_buffer_dirty(bh);
 62	nilfs_mdt_mark_dirty(inode);
 63
 64	trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
 65
 66	return 0;
 67}
 68
 69static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
 70				  struct buffer_head **out_bh,
 71				  void (*init_block)(struct inode *,
 72						     struct buffer_head *,
 73						     void *))
 74{
 75	struct super_block *sb = inode->i_sb;
 76	struct nilfs_transaction_info ti;
 77	struct buffer_head *bh;
 78	int err;
 79
 80	nilfs_transaction_begin(sb, &ti, 0);
 81
 82	err = -ENOMEM;
 83	bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
 84	if (unlikely(!bh))
 85		goto failed_unlock;
 86
 87	err = -EEXIST;
 88	if (buffer_uptodate(bh))
 89		goto failed_bh;
 90
 91	wait_on_buffer(bh);
 92	if (buffer_uptodate(bh))
 93		goto failed_bh;
 94
 95	err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
 96	if (likely(!err)) {
 97		get_bh(bh);
 98		*out_bh = bh;
 99	}
100
101 failed_bh:
102	folio_unlock(bh->b_folio);
103	folio_put(bh->b_folio);
104	brelse(bh);
105
106 failed_unlock:
107	if (likely(!err))
108		err = nilfs_transaction_commit(sb);
109	else
110		nilfs_transaction_abort(sb);
111
112	return err;
113}
114
115static int
116nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf,
117		       struct buffer_head **out_bh)
118{
119	struct buffer_head *bh;
120	__u64 blknum = 0;
121	int ret = -ENOMEM;
122
123	bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
124	if (unlikely(!bh))
125		goto failed;
126
127	ret = -EEXIST; /* internal code */
128	if (buffer_uptodate(bh))
129		goto out;
130
131	if (opf & REQ_RAHEAD) {
132		if (!trylock_buffer(bh)) {
133			ret = -EBUSY;
134			goto failed_bh;
135		}
136	} else /* opf == REQ_OP_READ */
137		lock_buffer(bh);
138
139	if (buffer_uptodate(bh)) {
140		unlock_buffer(bh);
141		goto out;
142	}
143
144	ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
145	if (unlikely(ret)) {
146		unlock_buffer(bh);
147		goto failed_bh;
148	}
149	map_bh(bh, inode->i_sb, (sector_t)blknum);
150
151	bh->b_end_io = end_buffer_read_sync;
152	get_bh(bh);
153	submit_bh(opf, bh);
154	ret = 0;
155
156	trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff,
157				      opf & REQ_OP_MASK);
158 out:
159	get_bh(bh);
160	*out_bh = bh;
161
162 failed_bh:
163	folio_unlock(bh->b_folio);
164	folio_put(bh->b_folio);
165	brelse(bh);
166 failed:
167	return ret;
168}
169
170static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
171				int readahead, struct buffer_head **out_bh)
172{
173	struct buffer_head *first_bh, *bh;
174	unsigned long blkoff;
175	int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
176	int err;
177
178	err = nilfs_mdt_submit_block(inode, block, REQ_OP_READ, &first_bh);
179	if (err == -EEXIST) /* internal code */
180		goto out;
181
182	if (unlikely(err))
183		goto failed;
184
185	if (readahead) {
186		blkoff = block + 1;
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))
191				brelse(bh);
192			else if (err != -EBUSY)
193				break;
194				/* abort readahead if bmap lookup failed */
195			if (!buffer_locked(first_bh))
196				goto out_no_wait;
197		}
198	}
199
200	wait_on_buffer(first_bh);
201
202 out_no_wait:
203	err = -EIO;
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);
208		goto failed_bh;
209	}
210 out:
211	*out_bh = first_bh;
212	return 0;
213
214 failed_bh:
215	brelse(first_bh);
216 failed:
217	return err;
218}
219
220/**
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
227 *
228 * nilfs_mdt_get_block() looks up the specified buffer and tries to create
229 * a new buffer if @create is not zero.  If (and only if) this function
230 * succeeds, it stores a pointer to the retrieved buffer head in the location
231 * pointed to by @out_bh.
232 *
233 * The retrieved buffer may be either an existing one or a newly allocated one.
234 * For a newly created buffer, if the callback function argument @init_block
235 * is non-NULL, the callback will be called with the buffer locked to format
236 * the block.
237 *
238 * Return: 0 on success, or one of the following negative error codes on
239 * failure:
240 * * %-EIO	- I/O error (including metadata corruption).
241 * * %-ENOENT	- The specified block does not exist (hole block).
242 * * %-ENOMEM	- Insufficient memory available.
243 * * %-EROFS	- Read only filesystem (for create mode).
244 */
245int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
246			void (*init_block)(struct inode *,
247					   struct buffer_head *, void *),
248			struct buffer_head **out_bh)
249{
250	int ret;
251
252	/* Should be rewritten with merging nilfs_mdt_read_block() */
253 retry:
254	ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
255	if (!create || ret != -ENOENT)
256		return ret;
257
258	ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
259	if (unlikely(ret == -EEXIST)) {
260		/* create = 0; */  /* limit read-create loop retries */
261		goto retry;
262	}
263	return ret;
264}
265
266/**
267 * nilfs_mdt_find_block - find and get a buffer on meta data file.
268 * @inode: inode of the meta data file
269 * @start: start block offset (inclusive)
270 * @end: end block offset (inclusive)
271 * @blkoff: block offset
272 * @out_bh: place to store a pointer to buffer_head struct
273 *
274 * nilfs_mdt_find_block() looks up an existing block in range of
275 * [@start, @end] and stores pointer to a buffer head of the block to
276 * @out_bh, and block offset to @blkoff, respectively.  @out_bh and
277 * @blkoff are substituted only when zero is returned.
278 *
279 * Return: 0 on success, or one of the following negative error codes on
280 * failure:
281 * * %-EIO	- I/O error (including metadata corruption).
282 * * %-ENOENT	- No block was found in the range.
283 * * %-ENOMEM	- Insufficient memory available.
284 */
285int nilfs_mdt_find_block(struct inode *inode, unsigned long start,
286			 unsigned long end, unsigned long *blkoff,
287			 struct buffer_head **out_bh)
288{
289	__u64 next;
290	int ret;
291
292	if (unlikely(start > end))
293		return -ENOENT;
294
295	ret = nilfs_mdt_read_block(inode, start, true, out_bh);
296	if (!ret) {
297		*blkoff = start;
298		goto out;
299	}
300	if (unlikely(ret != -ENOENT || start == ULONG_MAX))
301		goto out;
302
303	ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
304	if (!ret) {
305		if (next <= end) {
306			ret = nilfs_mdt_read_block(inode, next, true, out_bh);
307			if (!ret)
308				*blkoff = next;
309		} else {
310			ret = -ENOENT;
311		}
312	}
313out:
314	return ret;
315}
316
317/**
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
321 *
322 * Return: 0 on success, or one of the following negative error codes on
323 * failure:
324 * * %-EIO	- I/O error (including metadata corruption).
325 * * %-ENOENT	- Non-existent block.
326 * * %-ENOMEM	- Insufficient memory available.
327 */
328int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
329{
330	struct nilfs_inode_info *ii = NILFS_I(inode);
331	int err;
332
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);
337	}
338	return err;
339}
340
341/**
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
345 *
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.
348 *
349 * Return: 0 on success, or one of the following negative error codes on
350 * failure:
351 * * %-EBUSY	- Page has an active buffer.
352 * * %-ENOENT	- Page cache has no page addressed by the offset.
353 */
354int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
355{
356	pgoff_t index = block >> (PAGE_SHIFT - inode->i_blkbits);
357	struct folio *folio;
358	struct buffer_head *bh;
359	int ret = 0;
360	int still_dirty;
361
362	folio = filemap_lock_folio(inode->i_mapping, index);
363	if (IS_ERR(folio))
364		return -ENOENT;
365
366	folio_wait_writeback(folio);
367
368	bh = folio_buffers(folio);
369	if (bh) {
370		unsigned long first_block = index <<
371				(PAGE_SHIFT - inode->i_blkbits);
372		bh = get_nth_bh(bh, block - first_block);
373		nilfs_forget_buffer(bh);
374	}
375	still_dirty = folio_test_dirty(folio);
376	folio_unlock(folio);
377	folio_put(folio);
378
379	if (still_dirty ||
380	    invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
381		ret = -EBUSY;
382	return ret;
383}
384
385int nilfs_mdt_fetch_dirty(struct inode *inode)
386{
387	struct nilfs_inode_info *ii = NILFS_I(inode);
388
389	if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
390		set_bit(NILFS_I_DIRTY, &ii->i_state);
391		return 1;
392	}
393	return test_bit(NILFS_I_DIRTY, &ii->i_state);
394}
395
396static int nilfs_mdt_write_folio(struct folio *folio,
397		struct writeback_control *wbc)
398{
399	struct inode *inode = folio->mapping->host;
400	struct super_block *sb;
401	int err = 0;
402
403	if (inode && sb_rdonly(inode->i_sb)) {
404		/*
405		 * It means that filesystem was remounted in read-only
406		 * mode because of error or metadata corruption. But we
407		 * have dirty folios that try to be flushed in background.
408		 * So, here we simply discard this dirty folio.
409		 */
410		nilfs_clear_folio_dirty(folio);
411		folio_unlock(folio);
412		return -EROFS;
413	}
414
415	folio_redirty_for_writepage(wbc, folio);
416	folio_unlock(folio);
417
418	if (!inode)
419		return 0;
420
421	sb = inode->i_sb;
422
423	if (wbc->sync_mode == WB_SYNC_ALL)
424		err = nilfs_construct_segment(sb);
425
426	return err;
427}
428
429static int nilfs_mdt_writeback(struct address_space *mapping,
430		struct writeback_control *wbc)
431{
432	struct folio *folio = NULL;
433	int error;
434
435	while ((folio = writeback_iter(mapping, wbc, folio, &error)))
436		error = nilfs_mdt_write_folio(folio, wbc);
437
438	return error;
439}
440
441static const struct address_space_operations def_mdt_aops = {
442	.dirty_folio		= block_dirty_folio,
443	.invalidate_folio	= block_invalidate_folio,
444	.writepages		= nilfs_mdt_writeback,
445	.migrate_folio		= buffer_migrate_folio_norefs,
446};
447
448static const struct inode_operations def_mdt_iops;
449static const struct file_operations def_mdt_fops;
450
451
452int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
453{
454	struct nilfs_mdt_info *mi;
455
456	mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
457	if (!mi)
458		return -ENOMEM;
459
460	init_rwsem(&mi->mi_sem);
461	inode->i_private = mi;
462
463	inode->i_mode = S_IFREG;
464	mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
465
466	inode->i_op = &def_mdt_iops;
467	inode->i_fop = &def_mdt_fops;
468	inode->i_mapping->a_ops = &def_mdt_aops;
469
470	return 0;
471}
472
473/**
474 * nilfs_mdt_clear - do cleanup for the metadata file
475 * @inode: inode of the metadata file
476 */
477void nilfs_mdt_clear(struct inode *inode)
478{
479	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
480	struct nilfs_shadow_map *shadow = mdi->mi_shadow;
481
482	if (mdi->mi_palloc_cache)
483		nilfs_palloc_destroy_cache(inode);
484
485	if (shadow) {
486		struct inode *s_inode = shadow->inode;
487
488		shadow->inode = NULL;
489		iput(s_inode);
490		mdi->mi_shadow = NULL;
491	}
492}
493
494/**
495 * nilfs_mdt_destroy - release resources used by the metadata file
496 * @inode: inode of the metadata file
497 */
498void nilfs_mdt_destroy(struct inode *inode)
499{
500	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
501
502	kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
503	kfree(mdi);
504}
505
506void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
507			      unsigned int header_size)
508{
509	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
510
511	mi->mi_entry_size = entry_size;
512	mi->mi_entries_per_block = i_blocksize(inode) / entry_size;
513	mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
514}
515
516/**
517 * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
518 * @inode: inode of the metadata file
519 * @shadow: shadow mapping
520 *
521 * Return: 0 on success, or a negative error code on failure.
522 */
523int nilfs_mdt_setup_shadow_map(struct inode *inode,
524			       struct nilfs_shadow_map *shadow)
525{
526	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
527	struct inode *s_inode;
528
529	INIT_LIST_HEAD(&shadow->frozen_buffers);
530
531	s_inode = nilfs_iget_for_shadow(inode);
532	if (IS_ERR(s_inode))
533		return PTR_ERR(s_inode);
534
535	shadow->inode = s_inode;
536	mi->mi_shadow = shadow;
537	return 0;
538}
539
540/**
541 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
542 * @inode: inode of the metadata file
543 *
544 * Return: 0 on success, or a negative error code on failure.
545 */
546int nilfs_mdt_save_to_shadow_map(struct inode *inode)
547{
548	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
549	struct nilfs_inode_info *ii = NILFS_I(inode);
550	struct nilfs_shadow_map *shadow = mi->mi_shadow;
551	struct inode *s_inode = shadow->inode;
552	int ret;
553
554	ret = nilfs_copy_dirty_pages(s_inode->i_mapping, inode->i_mapping);
555	if (ret)
556		goto out;
557
558	ret = nilfs_copy_dirty_pages(NILFS_I(s_inode)->i_assoc_inode->i_mapping,
559				     ii->i_assoc_inode->i_mapping);
560	if (ret)
561		goto out;
562
563	nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
564 out:
565	return ret;
566}
567
568int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
569{
570	struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
571	struct buffer_head *bh_frozen;
572	struct folio *folio;
573	int blkbits = inode->i_blkbits;
574
575	folio = filemap_grab_folio(shadow->inode->i_mapping,
576			bh->b_folio->index);
577	if (IS_ERR(folio))
578		return PTR_ERR(folio);
579
580	bh_frozen = folio_buffers(folio);
581	if (!bh_frozen)
582		bh_frozen = create_empty_buffers(folio, 1 << blkbits, 0);
583
584	bh_frozen = get_nth_bh(bh_frozen,
585			       offset_in_folio(folio, bh->b_data) >> blkbits);
586
587	if (!buffer_uptodate(bh_frozen))
588		nilfs_copy_buffer(bh_frozen, bh);
589	if (list_empty(&bh_frozen->b_assoc_buffers)) {
590		list_add_tail(&bh_frozen->b_assoc_buffers,
591			      &shadow->frozen_buffers);
592		set_buffer_nilfs_redirected(bh);
593	} else {
594		brelse(bh_frozen); /* already frozen */
595	}
596
597	folio_unlock(folio);
598	folio_put(folio);
599	return 0;
600}
601
602struct buffer_head *
603nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
604{
605	struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
606	struct buffer_head *bh_frozen = NULL;
607	struct folio *folio;
608	int n;
609
610	folio = filemap_lock_folio(shadow->inode->i_mapping,
611			bh->b_folio->index);
612	if (!IS_ERR(folio)) {
613		bh_frozen = folio_buffers(folio);
614		if (bh_frozen) {
615			n = offset_in_folio(folio, bh->b_data) >>
616				inode->i_blkbits;
617			bh_frozen = get_nth_bh(bh_frozen, n);
618		}
619		folio_unlock(folio);
620		folio_put(folio);
621	}
622	return bh_frozen;
623}
624
625static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
626{
627	struct list_head *head = &shadow->frozen_buffers;
628	struct buffer_head *bh;
629
630	while (!list_empty(head)) {
631		bh = list_first_entry(head, struct buffer_head,
632				      b_assoc_buffers);
633		list_del_init(&bh->b_assoc_buffers);
634		brelse(bh); /* drop ref-count to make it releasable */
635	}
636}
637
638/**
639 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
640 * @inode: inode of the metadata file
641 */
642void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
643{
644	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
645	struct nilfs_inode_info *ii = NILFS_I(inode);
646	struct nilfs_shadow_map *shadow = mi->mi_shadow;
647
648	down_write(&mi->mi_sem);
649
650	if (mi->mi_palloc_cache)
651		nilfs_palloc_clear_cache(inode);
652
653	nilfs_clear_dirty_pages(inode->i_mapping);
654	nilfs_copy_back_pages(inode->i_mapping, shadow->inode->i_mapping);
655
656	nilfs_clear_dirty_pages(ii->i_assoc_inode->i_mapping);
657	nilfs_copy_back_pages(ii->i_assoc_inode->i_mapping,
658			      NILFS_I(shadow->inode)->i_assoc_inode->i_mapping);
659
660	nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
661
662	up_write(&mi->mi_sem);
663}
664
665/**
666 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
667 * @inode: inode of the metadata file
668 */
669void nilfs_mdt_clear_shadow_map(struct inode *inode)
670{
671	struct nilfs_mdt_info *mi = NILFS_MDT(inode);
672	struct nilfs_shadow_map *shadow = mi->mi_shadow;
673	struct inode *shadow_btnc_inode = NILFS_I(shadow->inode)->i_assoc_inode;
674
675	down_write(&mi->mi_sem);
676	nilfs_release_frozen_buffers(shadow);
677	truncate_inode_pages(shadow->inode->i_mapping, 0);
678	truncate_inode_pages(shadow_btnc_inode->i_mapping, 0);
679	up_write(&mi->mi_sem);
680}