blob: ceb7dc0b5badd3578d51d6216c719e100cba3a51 [file] [log] [blame]
Ryusuke Konishiae980432018-09-04 15:46:30 -07001// SPDX-License-Identifier: GPL-2.0+
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -07002/*
Ryusuke Konishi94ee1d92021-11-08 18:35:01 -08003 * Meta data file for NILFS
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -07004 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
Ryusuke Konishi4b420ab2016-05-23 16:23:09 -07007 * Written by Ryusuke Konishi.
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -07008 */
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/slab.h>
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070017#include "nilfs.h"
Ryusuke Konishi05d0e942010-07-10 20:52:09 +090018#include "btnode.h"
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070019#include "segment.h"
20#include "page.h"
21#include "mdt.h"
Ryusuke Konishi2d199612016-05-23 16:23:20 -070022#include "alloc.h" /* nilfs_palloc_destroy_cache() */
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070023
Hitoshi Mitakea9cd2072015-11-06 16:32:08 -080024#include <trace/events/nilfs2.h>
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070025
26#define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
27
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070028
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(ii->i_bmap, block, (unsigned long)bh);
45 if (unlikely(ret))
46 return ret;
47
48 set_buffer_mapped(bh);
49
Ryusuke Konishia3baca52024-01-22 23:01:51 +090050 kaddr = kmap_local_page(bh->b_page);
Fabian Frederick93407472017-02-27 14:28:32 -080051 memset(kaddr + bh_offset(bh), 0, i_blocksize(inode));
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070052 if (init_block)
53 init_block(inode, bh, kaddr);
54 flush_dcache_page(bh->b_page);
Ryusuke Konishia3baca52024-01-22 23:01:51 +090055 kunmap_local(kaddr);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070056
57 set_buffer_uptodate(bh);
Ryusuke Konishi5fc7b142011-05-05 12:56:51 +090058 mark_buffer_dirty(bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070059 nilfs_mdt_mark_dirty(inode);
Hitoshi Mitakea9cd2072015-11-06 16:32:08 -080060
61 trace_nilfs2_mdt_insert_new_block(inode, inode->i_ino, block);
62
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070063 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{
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070072 struct super_block *sb = inode->i_sb;
73 struct nilfs_transaction_info ti;
74 struct buffer_head *bh;
75 int err;
76
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070077 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;
Ryusuke Konishi14351102009-09-06 17:49:49 +090085 if (buffer_uptodate(bh))
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070086 goto failed_bh;
Ryusuke Konishi14351102009-09-06 17:49:49 +090087
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070088 wait_on_buffer(bh);
89 if (buffer_uptodate(bh))
90 goto failed_bh;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070091
Ryusuke Konishi090fd5b2010-09-05 16:17:35 +090092 bh->b_bdev = sb->s_bdev;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070093 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:
Matthew Wilcox (Oracle)83d96382023-11-14 17:44:28 +0900100 folio_unlock(bh->b_folio);
101 folio_put(bh->b_folio);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700102 brelse(bh);
103
104 failed_unlock:
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700105 if (likely(!err))
106 err = nilfs_transaction_commit(sb);
107 else
108 nilfs_transaction_abort(sb);
Ryusuke Konishi090fd5b2010-09-05 16:17:35 +0900109
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700110 return err;
111}
112
113static int
Bart Van Asscheed451252022-07-14 11:07:24 -0700114nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf,
115 struct buffer_head **out_bh)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700116{
117 struct buffer_head *bh;
Ryusuke Konishi0f3fe332009-08-15 02:29:28 +0900118 __u64 blknum = 0;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700119 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
Bart Van Asscheed451252022-07-14 11:07:24 -0700129 if (opf & REQ_RAHEAD) {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700130 if (!trylock_buffer(bh)) {
131 ret = -EBUSY;
132 goto failed_bh;
133 }
Bart Van Asscheed451252022-07-14 11:07:24 -0700134 } else /* opf == REQ_OP_READ */
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700135 lock_buffer(bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700136
137 if (buffer_uptodate(bh)) {
138 unlock_buffer(bh);
139 goto out;
140 }
Ryusuke Konishi14351102009-09-06 17:49:49 +0900141
142 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
143 if (unlikely(ret)) {
144 unlock_buffer(bh);
145 goto failed_bh;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700146 }
Ryusuke Konishic6e07182010-09-05 00:23:50 +0900147 map_bh(bh, inode->i_sb, (sector_t)blknum);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700148
149 bh->b_end_io = end_buffer_read_sync;
150 get_bh(bh);
Bart Van Asscheed451252022-07-14 11:07:24 -0700151 submit_bh(opf, bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700152 ret = 0;
Hitoshi Mitakea9cd2072015-11-06 16:32:08 -0800153
Bart Van Asscheed451252022-07-14 11:07:24 -0700154 trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff,
155 opf & REQ_OP_MASK);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700156 out:
157 get_bh(bh);
158 *out_bh = bh;
159
160 failed_bh:
Matthew Wilcox (Oracle)319a12c2023-11-14 17:44:29 +0900161 folio_unlock(bh->b_folio);
162 folio_put(bh->b_folio);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700163 brelse(bh);
164 failed:
165 return ret;
166}
167
168static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900169 int readahead, struct buffer_head **out_bh)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700170{
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
Bart Van Asscheed451252022-07-14 11:07:24 -0700176 err = nilfs_mdt_submit_block(inode, block, REQ_OP_READ, &first_bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700177 if (err == -EEXIST) /* internal code */
178 goto out;
179
180 if (unlikely(err))
181 goto failed;
182
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900183 if (readahead) {
184 blkoff = block + 1;
185 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
Bart Van Asscheed451252022-07-14 11:07:24 -0700186 err = nilfs_mdt_submit_block(inode, blkoff,
187 REQ_OP_READ | REQ_RAHEAD, &bh);
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900188 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(first_bh))
194 goto out_no_wait;
195 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700196 }
197
198 wait_on_buffer(first_bh);
199
200 out_no_wait:
201 err = -EIO;
Ryusuke Konishi39a9dcc2016-08-02 14:05:17 -0700202 if (!buffer_uptodate(first_bh)) {
Joe Perchesa1d07472020-08-11 18:35:49 -0700203 nilfs_err(inode->i_sb,
Ryusuke Konishi39a9dcc2016-08-02 14:05:17 -0700204 "I/O error reading meta-data file (ino=%lu, block-offset=%lu)",
205 inode->i_ino, block);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700206 goto failed_bh;
Ryusuke Konishi39a9dcc2016-08-02 14:05:17 -0700207 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700208 out:
209 *out_bh = first_bh;
210 return 0;
211
212 failed_bh:
213 brelse(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 *
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700240 * %-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:
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900251 ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700252 if (!create || ret != -ENOENT)
253 return ret;
254
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 */
258 goto retry;
259 }
260 return ret;
261}
262
263/**
Ryusuke Konishifa339152015-04-16 12:46:39 -0700264 * 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, 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/**
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700318 * 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
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700328 */
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(ii->i_bmap, block);
Ryusuke Konishi84338232009-05-05 21:52:06 +0900335 if (!err || err == -ENOENT) {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700336 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{
Matthew Wilcox (Oracle)1a846bf32023-10-16 21:11:00 +0100359 pgoff_t index = block >> (PAGE_SHIFT - inode->i_blkbits);
360 struct folio *folio;
361 struct buffer_head *bh;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700362 int ret = 0;
363 int still_dirty;
364
Matthew Wilcox (Oracle)1a846bf32023-10-16 21:11:00 +0100365 folio = filemap_lock_folio(inode->i_mapping, index);
366 if (IS_ERR(folio))
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700367 return -ENOENT;
368
Matthew Wilcox (Oracle)1a846bf32023-10-16 21:11:00 +0100369 folio_wait_writeback(folio);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700370
Matthew Wilcox (Oracle)1a846bf32023-10-16 21:11:00 +0100371 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, block - first_block);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700376 nilfs_forget_buffer(bh);
377 }
Matthew Wilcox (Oracle)1a846bf32023-10-16 21:11:00 +0100378 still_dirty = folio_test_dirty(folio);
379 folio_unlock(folio);
380 folio_put(folio);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700381
382 if (still_dirty ||
383 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
384 ret = -EBUSY;
385 return ret;
386}
387
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700388int 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(NILFS_I_DIRTY, &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{
Matthew Wilcox (Oracle)021cff92023-11-14 17:44:23 +0900402 struct folio *folio = page_folio(page);
403 struct inode *inode = folio->mapping->host;
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900404 struct super_block *sb;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700405 int err = 0;
406
David Howellsbc98a422017-07-17 08:45:34 +0100407 if (inode && sb_rdonly(inode->i_sb)) {
Vyacheslav Dubeyko8c26c4e2013-04-30 15:27:48 -0700408 /*
409 * It means that filesystem was remounted in read-only
410 * mode because of error or metadata corruption. But we
Matthew Wilcox (Oracle)021cff92023-11-14 17:44:23 +0900411 * have dirty folios that try to be flushed in background.
412 * So, here we simply discard this dirty folio.
Vyacheslav Dubeyko8c26c4e2013-04-30 15:27:48 -0700413 */
Ryusuke Konishi299910d2024-08-16 18:01:28 +0900414 nilfs_clear_folio_dirty(folio);
Matthew Wilcox (Oracle)021cff92023-11-14 17:44:23 +0900415 folio_unlock(folio);
Vyacheslav Dubeyko8c26c4e2013-04-30 15:27:48 -0700416 return -EROFS;
417 }
418
Matthew Wilcox (Oracle)021cff92023-11-14 17:44:23 +0900419 folio_redirty_for_writepage(wbc, folio);
420 folio_unlock(folio);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700421
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900422 if (!inode)
423 return 0;
424
425 sb = inode->i_sb;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700426
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);
431
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700432 return err;
433}
434
435
Alexey Dobriyan7f094102009-09-21 17:01:10 -0700436static const struct address_space_operations def_mdt_aops = {
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000437 .dirty_folio = block_dirty_folio,
438 .invalidate_folio = block_invalidate_folio,
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700439 .writepage = nilfs_mdt_write_page,
440};
441
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700442static const struct inode_operations def_mdt_iops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700443static const struct file_operations def_mdt_fops;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700444
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900445
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900446int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900447{
448 struct nilfs_mdt_info *mi;
449
450 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
451 if (!mi)
452 return -ENOMEM;
453
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900454 init_rwsem(&mi->mi_sem);
455 inode->i_private = mi;
456
457 inode->i_mode = S_IFREG;
458 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900459
460 inode->i_op = &def_mdt_iops;
461 inode->i_fop = &def_mdt_fops;
462 inode->i_mapping->a_ops = &def_mdt_aops;
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900463
464 return 0;
465}
466
Ryusuke Konishi2d199612016-05-23 16:23:20 -0700467/**
468 * nilfs_mdt_clear - do cleanup for the metadata file
469 * @inode: inode of the metadata file
470 */
471void nilfs_mdt_clear(struct inode *inode)
472{
473 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700474 struct nilfs_shadow_map *shadow = mdi->mi_shadow;
Ryusuke Konishi2d199612016-05-23 16:23:20 -0700475
476 if (mdi->mi_palloc_cache)
477 nilfs_palloc_destroy_cache(inode);
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700478
479 if (shadow) {
480 struct inode *s_inode = shadow->inode;
481
482 shadow->inode = NULL;
483 iput(s_inode);
484 mdi->mi_shadow = NULL;
485 }
Ryusuke Konishi2d199612016-05-23 16:23:20 -0700486}
487
488/**
489 * nilfs_mdt_destroy - release resources used by the metadata file
490 * @inode: inode of the metadata file
491 */
492void nilfs_mdt_destroy(struct inode *inode)
493{
494 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
495
496 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
497 kfree(mdi);
498}
499
Ryusuke Konishi0c6c44c2016-05-23 16:23:39 -0700500void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
501 unsigned int header_size)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700502{
503 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
504
505 mi->mi_entry_size = entry_size;
Fabian Frederick93407472017-02-27 14:28:32 -0800506 mi->mi_entries_per_block = i_blocksize(inode) / entry_size;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700507 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
508}
509
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900510/**
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
514 */
515int nilfs_mdt_setup_shadow_map(struct inode *inode,
516 struct nilfs_shadow_map *shadow)
517{
518 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700519 struct inode *s_inode;
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900520
521 INIT_LIST_HEAD(&shadow->frozen_buffers);
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700522
523 s_inode = nilfs_iget_for_shadow(inode);
524 if (IS_ERR(s_inode))
525 return PTR_ERR(s_inode);
526
527 shadow->inode = s_inode;
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900528 mi->mi_shadow = shadow;
529 return 0;
530}
531
532/**
533 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
534 * @inode: inode of the metadata file
535 */
536int nilfs_mdt_save_to_shadow_map(struct inode *inode)
537{
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;
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700541 struct inode *s_inode = shadow->inode;
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900542 int ret;
543
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700544 ret = nilfs_copy_dirty_pages(s_inode->i_mapping, inode->i_mapping);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900545 if (ret)
546 goto out;
547
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700548 ret = nilfs_copy_dirty_pages(NILFS_I(s_inode)->i_assoc_inode->i_mapping,
Ryusuke Konishie897be12022-04-01 11:28:18 -0700549 ii->i_assoc_inode->i_mapping);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900550 if (ret)
551 goto out;
552
553 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
554 out:
555 return ret;
556}
557
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900558int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
559{
560 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
561 struct buffer_head *bh_frozen;
Matthew Wilcox (Oracle)6c346be2023-10-16 21:10:57 +0100562 struct folio *folio;
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900563 int blkbits = inode->i_blkbits;
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900564
Matthew Wilcox (Oracle)6c346be2023-10-16 21:10:57 +0100565 folio = filemap_grab_folio(shadow->inode->i_mapping,
566 bh->b_folio->index);
567 if (IS_ERR(folio))
568 return PTR_ERR(folio);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900569
Matthew Wilcox (Oracle)6c346be2023-10-16 21:10:57 +0100570 bh_frozen = folio_buffers(folio);
571 if (!bh_frozen)
Matthew Wilcox (Oracle)0a888102023-10-16 21:11:14 +0100572 bh_frozen = create_empty_buffers(folio, 1 << blkbits, 0);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900573
Matthew Wilcox (Oracle)6c346be2023-10-16 21:10:57 +0100574 bh_frozen = get_nth_bh(bh_frozen, bh_offset(bh) >> blkbits);
Ryusuke Konishia7a84472010-12-27 00:03:02 +0900575
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);
582 } else {
583 brelse(bh_frozen); /* already frozen */
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900584 }
Ryusuke Konishia7a84472010-12-27 00:03:02 +0900585
Matthew Wilcox (Oracle)6c346be2023-10-16 21:10:57 +0100586 folio_unlock(folio);
587 folio_put(folio);
Ryusuke Konishia7a84472010-12-27 00:03:02 +0900588 return 0;
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900589}
590
591struct buffer_head *
592nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
593{
594 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
595 struct buffer_head *bh_frozen = NULL;
Matthew Wilcox (Oracle)664c87b2023-10-16 21:11:01 +0100596 struct folio *folio;
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900597 int n;
598
Matthew Wilcox (Oracle)664c87b2023-10-16 21:11:01 +0100599 folio = filemap_lock_folio(shadow->inode->i_mapping,
600 bh->b_folio->index);
601 if (!IS_ERR(folio)) {
602 bh_frozen = folio_buffers(folio);
603 if (bh_frozen) {
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900604 n = bh_offset(bh) >> inode->i_blkbits;
Matthew Wilcox (Oracle)664c87b2023-10-16 21:11:01 +0100605 bh_frozen = get_nth_bh(bh_frozen, n);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900606 }
Matthew Wilcox (Oracle)664c87b2023-10-16 21:11:01 +0100607 folio_unlock(folio);
608 folio_put(folio);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900609 }
610 return bh_frozen;
611}
612
613static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
614{
615 struct list_head *head = &shadow->frozen_buffers;
616 struct buffer_head *bh;
617
618 while (!list_empty(head)) {
619 bh = list_first_entry(head, struct buffer_head,
620 b_assoc_buffers);
621 list_del_init(&bh->b_assoc_buffers);
622 brelse(bh); /* drop ref-count to make it releasable */
623 }
624}
625
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900626/**
627 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
628 * @inode: inode of the metadata file
629 */
630void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
631{
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;
635
636 down_write(&mi->mi_sem);
637
638 if (mi->mi_palloc_cache)
639 nilfs_palloc_clear_cache(inode);
640
Ryusuke Konishi299910d2024-08-16 18:01:28 +0900641 nilfs_clear_dirty_pages(inode->i_mapping);
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700642 nilfs_copy_back_pages(inode->i_mapping, shadow->inode->i_mapping);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900643
Ryusuke Konishi299910d2024-08-16 18:01:28 +0900644 nilfs_clear_dirty_pages(ii->i_assoc_inode->i_mapping);
Ryusuke Konishie897be12022-04-01 11:28:18 -0700645 nilfs_copy_back_pages(ii->i_assoc_inode->i_mapping,
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700646 NILFS_I(shadow->inode)->i_assoc_inode->i_mapping);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900647
648 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
649
650 up_write(&mi->mi_sem);
651}
652
653/**
654 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
655 * @inode: inode of the metadata file
656 */
657void nilfs_mdt_clear_shadow_map(struct inode *inode)
658{
659 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
660 struct nilfs_shadow_map *shadow = mi->mi_shadow;
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700661 struct inode *shadow_btnc_inode = NILFS_I(shadow->inode)->i_assoc_inode;
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900662
663 down_write(&mi->mi_sem);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900664 nilfs_release_frozen_buffers(shadow);
Ryusuke Konishi6e211932022-04-01 11:28:21 -0700665 truncate_inode_pages(shadow->inode->i_mapping, 0);
666 truncate_inode_pages(shadow_btnc_inode->i_mapping, 0);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900667 up_write(&mi->mi_sem);
668}