blob: 700870a92bc4a1a2499e2ffef381613f8bad6e1a [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/*
3 * mdt.c - meta data file for NILFS
4 *
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
Cong Wang7b9c0972011-11-25 23:14:33 +080050 kaddr = kmap_atomic(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);
Cong Wang7b9c0972011-11-25 23:14:33 +080055 kunmap_atomic(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:
100 unlock_page(bh->b_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300101 put_page(bh->b_page);
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
114nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
Mike Christie2a222ca2016-06-05 14:31:43 -0500115 int mode, int mode_flags, 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
Mike Christie2a222ca2016-06-05 14:31:43 -0500129 if (mode_flags & REQ_RAHEAD) {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700130 if (!trylock_buffer(bh)) {
131 ret = -EBUSY;
132 goto failed_bh;
133 }
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700134 } else /* mode == 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);
Mike Christie2a222ca2016-06-05 14:31:43 -0500151 submit_bh(mode, mode_flags, bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700152 ret = 0;
Hitoshi Mitakea9cd2072015-11-06 16:32:08 -0800153
154 trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff, mode);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700155 out:
156 get_bh(bh);
157 *out_bh = bh;
158
159 failed_bh:
160 unlock_page(bh->b_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300161 put_page(bh->b_page);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700162 brelse(bh);
163 failed:
164 return ret;
165}
166
167static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900168 int readahead, struct buffer_head **out_bh)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700169{
170 struct buffer_head *first_bh, *bh;
171 unsigned long blkoff;
172 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
173 int err;
174
Mike Christie2a222ca2016-06-05 14:31:43 -0500175 err = nilfs_mdt_submit_block(inode, block, REQ_OP_READ, 0, &first_bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700176 if (err == -EEXIST) /* internal code */
177 goto out;
178
179 if (unlikely(err))
180 goto failed;
181
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900182 if (readahead) {
183 blkoff = block + 1;
184 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
Mike Christie2a222ca2016-06-05 14:31:43 -0500185 err = nilfs_mdt_submit_block(inode, blkoff, REQ_OP_READ,
186 REQ_RAHEAD, &bh);
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900187 if (likely(!err || err == -EEXIST))
188 brelse(bh);
189 else if (err != -EBUSY)
190 break;
191 /* abort readahead if bmap lookup failed */
192 if (!buffer_locked(first_bh))
193 goto out_no_wait;
194 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700195 }
196
197 wait_on_buffer(first_bh);
198
199 out_no_wait:
200 err = -EIO;
Ryusuke Konishi39a9dcc2016-08-02 14:05:17 -0700201 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);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700205 goto failed_bh;
Ryusuke Konishi39a9dcc2016-08-02 14:05:17 -0700206 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700207 out:
208 *out_bh = first_bh;
209 return 0;
210
211 failed_bh:
212 brelse(first_bh);
213 failed:
214 return err;
215}
216
217/**
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
224 *
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.
229 *
230 * Return Value: On success, it returns 0. On error, the following negative
231 * error code is returned.
232 *
233 * %-ENOMEM - Insufficient memory available.
234 *
235 * %-EIO - I/O error
236 *
237 * %-ENOENT - the specified block does not exist (hole block)
238 *
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700239 * %-EROFS - Read only filesystem (for create mode)
240 */
241int 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)
245{
246 int ret;
247
248 /* Should be rewritten with merging nilfs_mdt_read_block() */
249 retry:
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900250 ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700251 if (!create || ret != -ENOENT)
252 return ret;
253
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 */
257 goto retry;
258 }
259 return ret;
260}
261
262/**
Ryusuke Konishifa339152015-04-16 12:46:39 -0700263 * 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
269 *
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.
274 *
275 * Return Value: On success, it returns 0. On error, the following negative
276 * error code is returned.
277 *
278 * %-ENOMEM - Insufficient memory available.
279 *
280 * %-EIO - I/O error
281 *
282 * %-ENOENT - no block was found in the range
283 */
284int nilfs_mdt_find_block(struct inode *inode, unsigned long start,
285 unsigned long end, unsigned long *blkoff,
286 struct buffer_head **out_bh)
287{
288 __u64 next;
289 int ret;
290
291 if (unlikely(start > end))
292 return -ENOENT;
293
294 ret = nilfs_mdt_read_block(inode, start, true, out_bh);
295 if (!ret) {
296 *blkoff = start;
297 goto out;
298 }
299 if (unlikely(ret != -ENOENT || start == ULONG_MAX))
300 goto out;
301
302 ret = nilfs_bmap_seek_key(NILFS_I(inode)->i_bmap, start + 1, &next);
303 if (!ret) {
304 if (next <= end) {
305 ret = nilfs_mdt_read_block(inode, next, true, out_bh);
306 if (!ret)
307 *blkoff = next;
308 } else {
309 ret = -ENOENT;
310 }
311 }
312out:
313 return ret;
314}
315
316/**
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700317 * nilfs_mdt_delete_block - make a hole on the meta data file.
318 * @inode: inode of the meta data file
319 * @block: block offset
320 *
321 * Return Value: On success, zero is returned.
322 * On error, one of the following negative error code is returned.
323 *
324 * %-ENOMEM - Insufficient memory available.
325 *
326 * %-EIO - I/O error
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700327 */
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);
Ryusuke Konishi84338232009-05-05 21:52:06 +0900334 if (!err || err == -ENOENT) {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700335 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 Value: On success, 0 is returned. On error, one of the following
350 * negative error code is returned.
351 *
352 * %-EBUSY - page has an active buffer.
353 *
354 * %-ENOENT - page cache has no page addressed by the offset.
355 */
356int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
357{
358 pgoff_t index = (pgoff_t)block >>
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300359 (PAGE_SHIFT - inode->i_blkbits);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700360 struct page *page;
361 unsigned long first_block;
362 int ret = 0;
363 int still_dirty;
364
365 page = find_lock_page(inode->i_mapping, index);
366 if (!page)
367 return -ENOENT;
368
369 wait_on_page_writeback(page);
370
371 first_block = (unsigned long)index <<
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300372 (PAGE_SHIFT - inode->i_blkbits);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700373 if (page_has_buffers(page)) {
374 struct buffer_head *bh;
375
376 bh = nilfs_page_get_nth_block(page, block - first_block);
377 nilfs_forget_buffer(bh);
378 }
379 still_dirty = PageDirty(page);
380 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300381 put_page(page);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700382
383 if (still_dirty ||
384 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
385 ret = -EBUSY;
386 return ret;
387}
388
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700389int nilfs_mdt_fetch_dirty(struct inode *inode)
390{
391 struct nilfs_inode_info *ii = NILFS_I(inode);
392
393 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
394 set_bit(NILFS_I_DIRTY, &ii->i_state);
395 return 1;
396 }
397 return test_bit(NILFS_I_DIRTY, &ii->i_state);
398}
399
400static int
401nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
402{
Vyacheslav Dubeyko8c26c4e2013-04-30 15:27:48 -0700403 struct inode *inode = page->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
411 * have dirty pages that try to be flushed in background.
412 * So, here we simply discard this dirty page.
413 */
414 nilfs_clear_dirty_page(page, false);
415 unlock_page(page);
416 return -EROFS;
417 }
418
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700419 redirty_page_for_writepage(wbc, page);
420 unlock_page(page);
421
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 = {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700437 .writepage = nilfs_mdt_write_page,
438};
439
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700440static const struct inode_operations def_mdt_iops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700441static const struct file_operations def_mdt_fops;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700442
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900443
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900444int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900445{
446 struct nilfs_mdt_info *mi;
447
448 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
449 if (!mi)
450 return -ENOMEM;
451
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900452 init_rwsem(&mi->mi_sem);
453 inode->i_private = mi;
454
455 inode->i_mode = S_IFREG;
456 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900457
458 inode->i_op = &def_mdt_iops;
459 inode->i_fop = &def_mdt_fops;
460 inode->i_mapping->a_ops = &def_mdt_aops;
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900461
462 return 0;
463}
464
Ryusuke Konishi2d199612016-05-23 16:23:20 -0700465/**
466 * nilfs_mdt_clear - do cleanup for the metadata file
467 * @inode: inode of the metadata file
468 */
469void nilfs_mdt_clear(struct inode *inode)
470{
471 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
472
473 if (mdi->mi_palloc_cache)
474 nilfs_palloc_destroy_cache(inode);
475}
476
477/**
478 * nilfs_mdt_destroy - release resources used by the metadata file
479 * @inode: inode of the metadata file
480 */
481void nilfs_mdt_destroy(struct inode *inode)
482{
483 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
484
485 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
486 kfree(mdi);
487}
488
Ryusuke Konishi0c6c44c2016-05-23 16:23:39 -0700489void nilfs_mdt_set_entry_size(struct inode *inode, unsigned int entry_size,
490 unsigned int header_size)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700491{
492 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
493
494 mi->mi_entry_size = entry_size;
Fabian Frederick93407472017-02-27 14:28:32 -0800495 mi->mi_entries_per_block = i_blocksize(inode) / entry_size;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700496 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
497}
498
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900499/**
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
503 */
504int nilfs_mdt_setup_shadow_map(struct inode *inode,
505 struct nilfs_shadow_map *shadow)
506{
507 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900508
509 INIT_LIST_HEAD(&shadow->frozen_buffers);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100510 address_space_init_once(&shadow->frozen_data);
Christoph Hellwigb83ae6d2015-01-14 10:42:37 +0100511 nilfs_mapping_init(&shadow->frozen_data, inode);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100512 address_space_init_once(&shadow->frozen_btnodes);
Christoph Hellwigb83ae6d2015-01-14 10:42:37 +0100513 nilfs_mapping_init(&shadow->frozen_btnodes, inode);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900514 mi->mi_shadow = shadow;
515 return 0;
516}
517
518/**
519 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
520 * @inode: inode of the metadata file
521 */
522int nilfs_mdt_save_to_shadow_map(struct inode *inode)
523{
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;
527 int ret;
528
529 ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
530 if (ret)
531 goto out;
532
533 ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
534 &ii->i_btnode_cache);
535 if (ret)
536 goto out;
537
538 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
539 out:
540 return ret;
541}
542
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900543int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
544{
545 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
546 struct buffer_head *bh_frozen;
547 struct page *page;
548 int blkbits = inode->i_blkbits;
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900549
550 page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
551 if (!page)
Ryusuke Konishia7a84472010-12-27 00:03:02 +0900552 return -ENOMEM;
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900553
554 if (!page_has_buffers(page))
555 create_empty_buffers(page, 1 << blkbits, 0);
556
557 bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
Ryusuke Konishia7a84472010-12-27 00:03:02 +0900558
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);
565 } else {
566 brelse(bh_frozen); /* already frozen */
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900567 }
Ryusuke Konishia7a84472010-12-27 00:03:02 +0900568
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900569 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300570 put_page(page);
Ryusuke Konishia7a84472010-12-27 00:03:02 +0900571 return 0;
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900572}
573
574struct buffer_head *
575nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
576{
577 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
578 struct buffer_head *bh_frozen = NULL;
579 struct page *page;
580 int n;
581
582 page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
583 if (page) {
584 if (page_has_buffers(page)) {
585 n = bh_offset(bh) >> inode->i_blkbits;
586 bh_frozen = nilfs_page_get_nth_block(page, n);
587 }
588 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300589 put_page(page);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900590 }
591 return bh_frozen;
592}
593
594static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
595{
596 struct list_head *head = &shadow->frozen_buffers;
597 struct buffer_head *bh;
598
599 while (!list_empty(head)) {
600 bh = list_first_entry(head, struct buffer_head,
601 b_assoc_buffers);
602 list_del_init(&bh->b_assoc_buffers);
603 brelse(bh); /* drop ref-count to make it releasable */
604 }
605}
606
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900607/**
608 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
609 * @inode: inode of the metadata file
610 */
611void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
612{
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;
616
617 down_write(&mi->mi_sem);
618
619 if (mi->mi_palloc_cache)
620 nilfs_palloc_clear_cache(inode);
621
Vyacheslav Dubeyko8c26c4e2013-04-30 15:27:48 -0700622 nilfs_clear_dirty_pages(inode->i_mapping, true);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900623 nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
624
Vyacheslav Dubeyko8c26c4e2013-04-30 15:27:48 -0700625 nilfs_clear_dirty_pages(&ii->i_btnode_cache, true);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900626 nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
627
628 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
629
630 up_write(&mi->mi_sem);
631}
632
633/**
634 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
635 * @inode: inode of the metadata file
636 */
637void nilfs_mdt_clear_shadow_map(struct inode *inode)
638{
639 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
640 struct nilfs_shadow_map *shadow = mi->mi_shadow;
641
642 down_write(&mi->mi_sem);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900643 nilfs_release_frozen_buffers(shadow);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900644 truncate_inode_pages(&shadow->frozen_data, 0);
645 truncate_inode_pages(&shadow->frozen_btnodes, 0);
646 up_write(&mi->mi_sem);
647}