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 void *kaddr;
37 int ret;
38
39 /* Caller exclude read accesses using page lock */
40
41 /* set_buffer_new(bh); */
42 bh->b_blocknr = 0;
43
44 ret = nilfs_bmap_insert(bmap: ii->i_bmap, key: block, rec: (unsigned long)bh);
45 if (unlikely(ret))
46 return ret;
47
48 set_buffer_mapped(bh);
49
50 kaddr = kmap_atomic(page: bh->b_page);
51 memset(kaddr + bh_offset(bh), 0, i_blocksize(inode));
52 if (init_block)
53 init_block(inode, bh, kaddr);
54 flush_dcache_page(page: bh->b_page);
55 kunmap_atomic(kaddr);
56
57 set_buffer_uptodate(bh);
58 mark_buffer_dirty(bh);
59 nilfs_mdt_mark_dirty(inode);
60
61 trace_nilfs2_mdt_insert_new_block(inode, ino: inode->i_ino, block);
62
63 return 0;
64}
65
66static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
67 struct buffer_head **out_bh,
68 void (*init_block)(struct inode *,
69 struct buffer_head *,
70 void *))
71{
72 struct super_block *sb = inode->i_sb;
73 struct nilfs_transaction_info ti;
74 struct buffer_head *bh;
75 int err;
76
77 nilfs_transaction_begin(sb, &ti, 0);
78
79 err = -ENOMEM;
80 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
81 if (unlikely(!bh))
82 goto failed_unlock;
83
84 err = -EEXIST;
85 if (buffer_uptodate(bh))
86 goto failed_bh;
87
88 wait_on_buffer(bh);
89 if (buffer_uptodate(bh))
90 goto failed_bh;
91
92 bh->b_bdev = sb->s_bdev;
93 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
94 if (likely(!err)) {
95 get_bh(bh);
96 *out_bh = bh;
97 }
98
99 failed_bh:
100 unlock_page(page: bh->b_page);
101 put_page(page: bh->b_page);
102 brelse(bh);
103
104 failed_unlock:
105 if (likely(!err))
106 err = nilfs_transaction_commit(sb);
107 else
108 nilfs_transaction_abort(sb);
109
110 return err;
111}
112
113static int
114nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf,
115 struct buffer_head **out_bh)
116{
117 struct buffer_head *bh;
118 __u64 blknum = 0;
119 int ret = -ENOMEM;
120
121 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
122 if (unlikely(!bh))
123 goto failed;
124
125 ret = -EEXIST; /* internal code */
126 if (buffer_uptodate(bh))
127 goto out;
128
129 if (opf & REQ_RAHEAD) {
130 if (!trylock_buffer(bh)) {
131 ret = -EBUSY;
132 goto failed_bh;
133 }
134 } else /* opf == REQ_OP_READ */
135 lock_buffer(bh);
136
137 if (buffer_uptodate(bh)) {
138 unlock_buffer(bh);
139 goto out;
140 }
141
142 ret = nilfs_bmap_lookup(bmap: NILFS_I(inode)->i_bmap, key: blkoff, ptr: &blknum);
143 if (unlikely(ret)) {
144 unlock_buffer(bh);
145 goto failed_bh;
146 }
147 map_bh(bh, sb: inode->i_sb, block: (sector_t)blknum);
148
149 bh->b_end_io = end_buffer_read_sync;
150 get_bh(bh);
151 submit_bh(opf, bh);
152 ret = 0;
153
154 trace_nilfs2_mdt_submit_block(inode, ino: inode->i_ino, blkoff,
155 mode: opf & REQ_OP_MASK);
156 out:
157 get_bh(bh);
158 *out_bh = bh;
159
160 failed_bh:
161 unlock_page(page: bh->b_page);
162 put_page(page: bh->b_page);
163 brelse(bh);
164 failed:
165 return ret;
166}
167
168static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
169 int readahead, struct buffer_head **out_bh)
170{
171 struct buffer_head *first_bh, *bh;
172 unsigned long blkoff;
173 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
174 int err;
175
176 err = nilfs_mdt_submit_block(inode, blkoff: block, opf: REQ_OP_READ, out_bh: &first_bh);
177 if (err == -EEXIST) /* internal code */
178 goto out;
179
180 if (unlikely(err))
181 goto failed;
182
183 if (readahead) {
184 blkoff = block + 1;
185 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
186 err = nilfs_mdt_submit_block(inode, blkoff,
187 opf: REQ_OP_READ | REQ_RAHEAD, out_bh: &bh);
188 if (likely(!err || err == -EEXIST))
189 brelse(bh);
190 else if (err != -EBUSY)
191 break;
192 /* abort readahead if bmap lookup failed */
193 if (!buffer_locked(bh: first_bh))
194 goto out_no_wait;
195 }
196 }
197
198 wait_on_buffer(bh: first_bh);
199
200 out_no_wait:
201 err = -EIO;
202 if (!buffer_uptodate(bh: 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);
206 goto failed_bh;
207 }
208 out:
209 *out_bh = first_bh;
210 return 0;
211
212 failed_bh:
213 brelse(bh: first_bh);
214 failed:
215 return err;
216}
217
218/**
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
225 *
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.
230 *
231 * Return Value: On success, it returns 0. On error, the following negative
232 * error code is returned.
233 *
234 * %-ENOMEM - Insufficient memory available.
235 *
236 * %-EIO - I/O error
237 *
238 * %-ENOENT - the specified block does not exist (hole block)
239 *
240 * %-EROFS - Read only filesystem (for create mode)
241 */
242int 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)
246{
247 int ret;
248
249 /* Should be rewritten with merging nilfs_mdt_read_block() */
250 retry:
251 ret = nilfs_mdt_read_block(inode, block: blkoff, readahead: !create, out_bh);
252 if (!create || ret != -ENOENT)
253 return ret;
254
255 ret = nilfs_mdt_create_block(inode, block: blkoff, out_bh, init_block);
256 if (unlikely(ret == -EEXIST)) {
257 /* create = 0; */ /* limit read-create loop retries */
258 goto retry;
259 }
260 return ret;
261}
262
263/**
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
270 *
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.
275 *
276 * Return Value: On success, it returns 0. On error, the following negative
277 * error code is returned.
278 *
279 * %-ENOMEM - Insufficient memory available.
280 *
281 * %-EIO - I/O error
282 *
283 * %-ENOENT - no block was found in the range
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, block: start, readahead: 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(bmap: NILFS_I(inode)->i_bmap, start: start + 1, keyp: &next);
304 if (!ret) {
305 if (next <= end) {
306 ret = nilfs_mdt_read_block(inode, block: next, readahead: 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 Value: On success, zero is returned.
323 * On error, one of the following negative error code is returned.
324 *
325 * %-ENOMEM - Insufficient memory available.
326 *
327 * %-EIO - I/O error
328 */
329int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
330{
331 struct nilfs_inode_info *ii = NILFS_I(inode);
332 int err;
333
334 err = nilfs_bmap_delete(bmap: ii->i_bmap, key: block);
335 if (!err || err == -ENOENT) {
336 nilfs_mdt_mark_dirty(inode);
337 nilfs_mdt_forget_block(inode, block);
338 }
339 return err;
340}
341
342/**
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
346 *
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.
349 *
350 * Return Value: On success, 0 is returned. On error, one of the following
351 * negative error code is returned.
352 *
353 * %-EBUSY - page has an active buffer.
354 *
355 * %-ENOENT - page cache has no page addressed by the offset.
356 */
357int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
358{
359 pgoff_t index = block >> (PAGE_SHIFT - inode->i_blkbits);
360 struct folio *folio;
361 struct buffer_head *bh;
362 int ret = 0;
363 int still_dirty;
364
365 folio = filemap_lock_folio(mapping: inode->i_mapping, index);
366 if (IS_ERR(ptr: folio))
367 return -ENOENT;
368
369 folio_wait_writeback(folio);
370
371 bh = folio_buffers(folio);
372 if (bh) {
373 unsigned long first_block = index <<
374 (PAGE_SHIFT - inode->i_blkbits);
375 bh = get_nth_bh(bh, count: block - first_block);
376 nilfs_forget_buffer(bh);
377 }
378 still_dirty = folio_test_dirty(folio);
379 folio_unlock(folio);
380 folio_put(folio);
381
382 if (still_dirty ||
383 invalidate_inode_pages2_range(mapping: inode->i_mapping, start: index, end: index) != 0)
384 ret = -EBUSY;
385 return ret;
386}
387
388int nilfs_mdt_fetch_dirty(struct inode *inode)
389{
390 struct nilfs_inode_info *ii = NILFS_I(inode);
391
392 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
393 set_bit(nr: NILFS_I_DIRTY, addr: &ii->i_state);
394 return 1;
395 }
396 return test_bit(NILFS_I_DIRTY, &ii->i_state);
397}
398
399static int
400nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
401{
402 struct inode *inode = page->mapping->host;
403 struct super_block *sb;
404 int err = 0;
405
406 if (inode && sb_rdonly(sb: inode->i_sb)) {
407 /*
408 * It means that filesystem was remounted in read-only
409 * mode because of error or metadata corruption. But we
410 * have dirty pages that try to be flushed in background.
411 * So, here we simply discard this dirty page.
412 */
413 nilfs_clear_dirty_page(page, false);
414 unlock_page(page);
415 return -EROFS;
416 }
417
418 redirty_page_for_writepage(wbc, page);
419 unlock_page(page);
420
421 if (!inode)
422 return 0;
423
424 sb = inode->i_sb;
425
426 if (wbc->sync_mode == WB_SYNC_ALL)
427 err = nilfs_construct_segment(sb);
428 else if (wbc->for_reclaim)
429 nilfs_flush_segment(sb, inode->i_ino);
430
431 return err;
432}
433
434
435static const struct address_space_operations def_mdt_aops = {
436 .dirty_folio = block_dirty_folio,
437 .invalidate_folio = block_invalidate_folio,
438 .writepage = nilfs_mdt_write_page,
439};
440
441static const struct inode_operations def_mdt_iops;
442static const struct file_operations def_mdt_fops;
443
444
445int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
446{
447 struct nilfs_mdt_info *mi;
448
449 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
450 if (!mi)
451 return -ENOMEM;
452
453 init_rwsem(&mi->mi_sem);
454 inode->i_private = mi;
455
456 inode->i_mode = S_IFREG;
457 mapping_set_gfp_mask(m: inode->i_mapping, mask: gfp_mask);
458
459 inode->i_op = &def_mdt_iops;
460 inode->i_fop = &def_mdt_fops;
461 inode->i_mapping->a_ops = &def_mdt_aops;
462
463 return 0;
464}
465
466/**
467 * nilfs_mdt_clear - do cleanup for the metadata file
468 * @inode: inode of the metadata file
469 */
470void nilfs_mdt_clear(struct inode *inode)
471{
472 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
473 struct nilfs_shadow_map *shadow = mdi->mi_shadow;
474
475 if (mdi->mi_palloc_cache)
476 nilfs_palloc_destroy_cache(inode);
477
478 if (shadow) {
479 struct inode *s_inode = shadow->inode;
480
481 shadow->inode = NULL;
482 iput(s_inode);
483 mdi->mi_shadow = NULL;
484 }
485}
486
487/**
488 * nilfs_mdt_destroy - release resources used by the metadata file
489 * @inode: inode of the metadata file
490 */
491void nilfs_mdt_destroy(struct inode *inode)
492{
493 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
494
495 kfree(objp: mdi->mi_bgl); /* kfree(NULL) is safe */
496 kfree(objp: mdi);
497}
498
499void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
500 unsigned int header_size)
501{
502 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
503
504 mi->mi_entry_size = entry_size;
505 mi->mi_entries_per_block = i_blocksize(node: inode) / entry_size;
506 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
507}
508
509/**
510 * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
511 * @inode: inode of the metadata file
512 * @shadow: shadow mapping
513 */
514int nilfs_mdt_setup_shadow_map(struct inode *inode,
515 struct nilfs_shadow_map *shadow)
516{
517 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
518 struct inode *s_inode;
519
520 INIT_LIST_HEAD(list: &shadow->frozen_buffers);
521
522 s_inode = nilfs_iget_for_shadow(inode);
523 if (IS_ERR(ptr: s_inode))
524 return PTR_ERR(ptr: s_inode);
525
526 shadow->inode = s_inode;
527 mi->mi_shadow = shadow;
528 return 0;
529}
530
531/**
532 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
533 * @inode: inode of the metadata file
534 */
535int nilfs_mdt_save_to_shadow_map(struct inode *inode)
536{
537 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
538 struct nilfs_inode_info *ii = NILFS_I(inode);
539 struct nilfs_shadow_map *shadow = mi->mi_shadow;
540 struct inode *s_inode = shadow->inode;
541 int ret;
542
543 ret = nilfs_copy_dirty_pages(s_inode->i_mapping, inode->i_mapping);
544 if (ret)
545 goto out;
546
547 ret = nilfs_copy_dirty_pages(NILFS_I(inode: s_inode)->i_assoc_inode->i_mapping,
548 ii->i_assoc_inode->i_mapping);
549 if (ret)
550 goto out;
551
552 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
553 out:
554 return ret;
555}
556
557int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
558{
559 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
560 struct buffer_head *bh_frozen;
561 struct folio *folio;
562 int blkbits = inode->i_blkbits;
563
564 folio = filemap_grab_folio(mapping: shadow->inode->i_mapping,
565 index: bh->b_folio->index);
566 if (IS_ERR(ptr: folio))
567 return PTR_ERR(ptr: folio);
568
569 bh_frozen = folio_buffers(folio);
570 if (!bh_frozen)
571 bh_frozen = create_empty_buffers(folio, blocksize: 1 << blkbits, b_state: 0);
572
573 bh_frozen = get_nth_bh(bh: bh_frozen, count: bh_offset(bh) >> blkbits);
574
575 if (!buffer_uptodate(bh: bh_frozen))
576 nilfs_copy_buffer(bh_frozen, bh);
577 if (list_empty(head: &bh_frozen->b_assoc_buffers)) {
578 list_add_tail(new: &bh_frozen->b_assoc_buffers,
579 head: &shadow->frozen_buffers);
580 set_buffer_nilfs_redirected(bh);
581 } else {
582 brelse(bh: bh_frozen); /* already frozen */
583 }
584
585 folio_unlock(folio);
586 folio_put(folio);
587 return 0;
588}
589
590struct buffer_head *
591nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
592{
593 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
594 struct buffer_head *bh_frozen = NULL;
595 struct folio *folio;
596 int n;
597
598 folio = filemap_lock_folio(mapping: shadow->inode->i_mapping,
599 index: bh->b_folio->index);
600 if (!IS_ERR(ptr: folio)) {
601 bh_frozen = folio_buffers(folio);
602 if (bh_frozen) {
603 n = bh_offset(bh) >> inode->i_blkbits;
604 bh_frozen = get_nth_bh(bh: bh_frozen, count: n);
605 }
606 folio_unlock(folio);
607 folio_put(folio);
608 }
609 return bh_frozen;
610}
611
612static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
613{
614 struct list_head *head = &shadow->frozen_buffers;
615 struct buffer_head *bh;
616
617 while (!list_empty(head)) {
618 bh = list_first_entry(head, struct buffer_head,
619 b_assoc_buffers);
620 list_del_init(entry: &bh->b_assoc_buffers);
621 brelse(bh); /* drop ref-count to make it releasable */
622 }
623}
624
625/**
626 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
627 * @inode: inode of the metadata file
628 */
629void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
630{
631 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
632 struct nilfs_inode_info *ii = NILFS_I(inode);
633 struct nilfs_shadow_map *shadow = mi->mi_shadow;
634
635 down_write(sem: &mi->mi_sem);
636
637 if (mi->mi_palloc_cache)
638 nilfs_palloc_clear_cache(inode);
639
640 nilfs_clear_dirty_pages(inode->i_mapping, true);
641 nilfs_copy_back_pages(inode->i_mapping, shadow->inode->i_mapping);
642
643 nilfs_clear_dirty_pages(ii->i_assoc_inode->i_mapping, true);
644 nilfs_copy_back_pages(ii->i_assoc_inode->i_mapping,
645 NILFS_I(inode: shadow->inode)->i_assoc_inode->i_mapping);
646
647 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
648
649 up_write(sem: &mi->mi_sem);
650}
651
652/**
653 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
654 * @inode: inode of the metadata file
655 */
656void nilfs_mdt_clear_shadow_map(struct inode *inode)
657{
658 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
659 struct nilfs_shadow_map *shadow = mi->mi_shadow;
660 struct inode *shadow_btnc_inode = NILFS_I(inode: shadow->inode)->i_assoc_inode;
661
662 down_write(sem: &mi->mi_sem);
663 nilfs_release_frozen_buffers(shadow);
664 truncate_inode_pages(shadow->inode->i_mapping, 0);
665 truncate_inode_pages(shadow_btnc_inode->i_mapping, 0);
666 up_write(sem: &mi->mi_sem);
667}
668

source code of linux/fs/nilfs2/mdt.c