blob: 44326cfe1fa9415d3fc52a1c84317ab1ca932a36 [file] [log] [blame]
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -07001/*
2 * mdt.c - meta data file for NILFS
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
21 */
22
23#include <linux/buffer_head.h>
24#include <linux/mpage.h>
25#include <linux/mm.h>
26#include <linux/writeback.h>
27#include <linux/backing-dev.h>
28#include <linux/swap.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090029#include <linux/slab.h>
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070030#include "nilfs.h"
Ryusuke Konishi05d0e942010-07-10 20:52:09 +090031#include "btnode.h"
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070032#include "segment.h"
33#include "page.h"
34#include "mdt.h"
35
36
37#define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
38
39#define INIT_UNUSED_INODE_FIELDS
40
41static int
42nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
43 struct buffer_head *bh,
44 void (*init_block)(struct inode *,
45 struct buffer_head *, void *))
46{
47 struct nilfs_inode_info *ii = NILFS_I(inode);
48 void *kaddr;
49 int ret;
50
51 /* Caller exclude read accesses using page lock */
52
53 /* set_buffer_new(bh); */
54 bh->b_blocknr = 0;
55
56 ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
57 if (unlikely(ret))
58 return ret;
59
60 set_buffer_mapped(bh);
61
62 kaddr = kmap_atomic(bh->b_page, KM_USER0);
63 memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
64 if (init_block)
65 init_block(inode, bh, kaddr);
66 flush_dcache_page(bh->b_page);
67 kunmap_atomic(kaddr, KM_USER0);
68
69 set_buffer_uptodate(bh);
70 nilfs_mark_buffer_dirty(bh);
71 nilfs_mdt_mark_dirty(inode);
72 return 0;
73}
74
75static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
76 struct buffer_head **out_bh,
77 void (*init_block)(struct inode *,
78 struct buffer_head *,
79 void *))
80{
81 struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070082 struct super_block *sb = inode->i_sb;
83 struct nilfs_transaction_info ti;
84 struct buffer_head *bh;
85 int err;
86
87 if (!sb) {
Ryusuke Konishi201913e2009-04-28 21:04:59 +090088 /*
89 * Make sure this function is not called from any
90 * read-only context.
91 */
92 if (!nilfs->ns_writer) {
93 WARN_ON(1);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070094 err = -EROFS;
95 goto out;
96 }
Ryusuke Konishi201913e2009-04-28 21:04:59 +090097 sb = nilfs->ns_writer->s_super;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070098 }
99
100 nilfs_transaction_begin(sb, &ti, 0);
101
102 err = -ENOMEM;
103 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
104 if (unlikely(!bh))
105 goto failed_unlock;
106
107 err = -EEXIST;
Ryusuke Konishi14351102009-09-06 17:49:49 +0900108 if (buffer_uptodate(bh))
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700109 goto failed_bh;
Ryusuke Konishi14351102009-09-06 17:49:49 +0900110
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700111 wait_on_buffer(bh);
112 if (buffer_uptodate(bh))
113 goto failed_bh;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700114
115 bh->b_bdev = nilfs->ns_bdev;
116 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
117 if (likely(!err)) {
118 get_bh(bh);
119 *out_bh = bh;
120 }
121
122 failed_bh:
123 unlock_page(bh->b_page);
124 page_cache_release(bh->b_page);
125 brelse(bh);
126
127 failed_unlock:
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700128 if (likely(!err))
129 err = nilfs_transaction_commit(sb);
130 else
131 nilfs_transaction_abort(sb);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700132 out:
133 return err;
134}
135
136static int
137nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
138 int mode, struct buffer_head **out_bh)
139{
140 struct buffer_head *bh;
Ryusuke Konishi0f3fe332009-08-15 02:29:28 +0900141 __u64 blknum = 0;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700142 int ret = -ENOMEM;
143
144 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
145 if (unlikely(!bh))
146 goto failed;
147
148 ret = -EEXIST; /* internal code */
149 if (buffer_uptodate(bh))
150 goto out;
151
152 if (mode == READA) {
153 if (!trylock_buffer(bh)) {
154 ret = -EBUSY;
155 goto failed_bh;
156 }
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700157 } else /* mode == READ */
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700158 lock_buffer(bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700159
160 if (buffer_uptodate(bh)) {
161 unlock_buffer(bh);
162 goto out;
163 }
Ryusuke Konishi14351102009-09-06 17:49:49 +0900164
165 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
166 if (unlikely(ret)) {
167 unlock_buffer(bh);
168 goto failed_bh;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700169 }
Ryusuke Konishi14351102009-09-06 17:49:49 +0900170 bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev;
Ryusuke Konishi0f3fe332009-08-15 02:29:28 +0900171 bh->b_blocknr = (sector_t)blknum;
Ryusuke Konishi14351102009-09-06 17:49:49 +0900172 set_buffer_mapped(bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700173
174 bh->b_end_io = end_buffer_read_sync;
175 get_bh(bh);
176 submit_bh(mode, bh);
177 ret = 0;
178 out:
179 get_bh(bh);
180 *out_bh = bh;
181
182 failed_bh:
183 unlock_page(bh->b_page);
184 page_cache_release(bh->b_page);
185 brelse(bh);
186 failed:
187 return ret;
188}
189
190static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900191 int readahead, struct buffer_head **out_bh)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700192{
193 struct buffer_head *first_bh, *bh;
194 unsigned long blkoff;
195 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
196 int err;
197
198 err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
199 if (err == -EEXIST) /* internal code */
200 goto out;
201
202 if (unlikely(err))
203 goto failed;
204
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900205 if (readahead) {
206 blkoff = block + 1;
207 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
208 err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
209 if (likely(!err || err == -EEXIST))
210 brelse(bh);
211 else if (err != -EBUSY)
212 break;
213 /* abort readahead if bmap lookup failed */
214 if (!buffer_locked(first_bh))
215 goto out_no_wait;
216 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700217 }
218
219 wait_on_buffer(first_bh);
220
221 out_no_wait:
222 err = -EIO;
223 if (!buffer_uptodate(first_bh))
224 goto failed_bh;
225 out:
226 *out_bh = first_bh;
227 return 0;
228
229 failed_bh:
230 brelse(first_bh);
231 failed:
232 return err;
233}
234
235/**
236 * nilfs_mdt_get_block - read or create a buffer on meta data file.
237 * @inode: inode of the meta data file
238 * @blkoff: block offset
239 * @create: create flag
240 * @init_block: initializer used for newly allocated block
241 * @out_bh: output of a pointer to the buffer_head
242 *
243 * nilfs_mdt_get_block() looks up the specified buffer and tries to create
244 * a new buffer if @create is not zero. On success, the returned buffer is
245 * assured to be either existing or formatted using a buffer lock on success.
246 * @out_bh is substituted only when zero is returned.
247 *
248 * Return Value: On success, it returns 0. On error, the following negative
249 * error code is returned.
250 *
251 * %-ENOMEM - Insufficient memory available.
252 *
253 * %-EIO - I/O error
254 *
255 * %-ENOENT - the specified block does not exist (hole block)
256 *
257 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
258 *
259 * %-EROFS - Read only filesystem (for create mode)
260 */
261int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
262 void (*init_block)(struct inode *,
263 struct buffer_head *, void *),
264 struct buffer_head **out_bh)
265{
266 int ret;
267
268 /* Should be rewritten with merging nilfs_mdt_read_block() */
269 retry:
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900270 ret = nilfs_mdt_read_block(inode, blkoff, !create, out_bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700271 if (!create || ret != -ENOENT)
272 return ret;
273
274 ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
275 if (unlikely(ret == -EEXIST)) {
276 /* create = 0; */ /* limit read-create loop retries */
277 goto retry;
278 }
279 return ret;
280}
281
282/**
283 * nilfs_mdt_delete_block - make a hole on the meta data file.
284 * @inode: inode of the meta data file
285 * @block: block offset
286 *
287 * Return Value: On success, zero is returned.
288 * On error, one of the following negative error code is returned.
289 *
290 * %-ENOMEM - Insufficient memory available.
291 *
292 * %-EIO - I/O error
293 *
294 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
295 */
296int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
297{
298 struct nilfs_inode_info *ii = NILFS_I(inode);
299 int err;
300
301 err = nilfs_bmap_delete(ii->i_bmap, block);
Ryusuke Konishi84338232009-05-05 21:52:06 +0900302 if (!err || err == -ENOENT) {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700303 nilfs_mdt_mark_dirty(inode);
304 nilfs_mdt_forget_block(inode, block);
305 }
306 return err;
307}
308
309/**
310 * nilfs_mdt_forget_block - discard dirty state and try to remove the page
311 * @inode: inode of the meta data file
312 * @block: block offset
313 *
314 * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
315 * tries to release the page including the buffer from a page cache.
316 *
317 * Return Value: On success, 0 is returned. On error, one of the following
318 * negative error code is returned.
319 *
320 * %-EBUSY - page has an active buffer.
321 *
322 * %-ENOENT - page cache has no page addressed by the offset.
323 */
324int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
325{
326 pgoff_t index = (pgoff_t)block >>
327 (PAGE_CACHE_SHIFT - inode->i_blkbits);
328 struct page *page;
329 unsigned long first_block;
330 int ret = 0;
331 int still_dirty;
332
333 page = find_lock_page(inode->i_mapping, index);
334 if (!page)
335 return -ENOENT;
336
337 wait_on_page_writeback(page);
338
339 first_block = (unsigned long)index <<
340 (PAGE_CACHE_SHIFT - inode->i_blkbits);
341 if (page_has_buffers(page)) {
342 struct buffer_head *bh;
343
344 bh = nilfs_page_get_nth_block(page, block - first_block);
345 nilfs_forget_buffer(bh);
346 }
347 still_dirty = PageDirty(page);
348 unlock_page(page);
349 page_cache_release(page);
350
351 if (still_dirty ||
352 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
353 ret = -EBUSY;
354 return ret;
355}
356
357/**
358 * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
359 * @inode: inode of the meta data file
360 * @block: block offset
361 *
362 * Return Value: On success, it returns 0. On error, the following negative
363 * error code is returned.
364 *
365 * %-ENOMEM - Insufficient memory available.
366 *
367 * %-EIO - I/O error
368 *
369 * %-ENOENT - the specified block does not exist (hole block)
370 *
371 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
372 */
373int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
374{
375 struct buffer_head *bh;
376 int err;
377
Ryusuke Konishib34a6502009-11-14 00:09:47 +0900378 err = nilfs_mdt_read_block(inode, block, 0, &bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700379 if (unlikely(err))
380 return err;
381 nilfs_mark_buffer_dirty(bh);
382 nilfs_mdt_mark_dirty(inode);
383 brelse(bh);
384 return 0;
385}
386
387int nilfs_mdt_fetch_dirty(struct inode *inode)
388{
389 struct nilfs_inode_info *ii = NILFS_I(inode);
390
391 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
392 set_bit(NILFS_I_DIRTY, &ii->i_state);
393 return 1;
394 }
395 return test_bit(NILFS_I_DIRTY, &ii->i_state);
396}
397
398static int
399nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
400{
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900401 struct inode *inode;
402 struct super_block *sb;
403 struct the_nilfs *nilfs;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700404 struct nilfs_sb_info *writer = NULL;
405 int err = 0;
406
407 redirty_page_for_writepage(wbc, page);
408 unlock_page(page);
409
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900410 inode = page->mapping->host;
411 if (!inode)
412 return 0;
413
414 sb = inode->i_sb;
415 nilfs = NILFS_MDT(inode)->mi_nilfs;
416
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700417 if (!sb) {
Ryusuke Konishi027d6402009-08-02 22:45:33 +0900418 down_read(&nilfs->ns_writer_sem);
419 writer = nilfs->ns_writer;
Ryusuke Konishi01a261e2009-08-02 17:45:55 +0900420 if (!writer) {
Ryusuke Konishi027d6402009-08-02 22:45:33 +0900421 up_read(&nilfs->ns_writer_sem);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700422 return -EROFS;
Ryusuke Konishi01a261e2009-08-02 17:45:55 +0900423 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700424 sb = writer->s_super;
425 }
426
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
432 if (writer)
Ryusuke Konishi027d6402009-08-02 22:45:33 +0900433 up_read(&nilfs->ns_writer_sem);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700434 return err;
435}
436
437
Alexey Dobriyan7f094102009-09-21 17:01:10 -0700438static const struct address_space_operations def_mdt_aops = {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700439 .writepage = nilfs_mdt_write_page,
Ryusuke Konishifa032742009-05-27 22:44:34 +0900440 .sync_page = block_sync_page,
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700441};
442
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700443static const struct inode_operations def_mdt_iops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700444static const struct file_operations def_mdt_fops;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700445
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900446
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900447int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900448{
449 struct nilfs_mdt_info *mi;
450
451 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
452 if (!mi)
453 return -ENOMEM;
454
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900455 mi->mi_nilfs = NILFS_I_NILFS(inode);
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900456 init_rwsem(&mi->mi_sem);
457 inode->i_private = mi;
458
459 inode->i_mode = S_IFREG;
460 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900461 inode->i_mapping->backing_dev_info = inode->i_sb->s_bdi;
462
463 inode->i_op = &def_mdt_iops;
464 inode->i_fop = &def_mdt_fops;
465 inode->i_mapping->a_ops = &def_mdt_aops;
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900466
467 return 0;
468}
469
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700470/*
471 * NILFS2 uses pseudo inodes for meta data files such as DAT, cpfile, sufile,
472 * ifile, or gcinodes. This allows the B-tree code and segment constructor
473 * to treat them like regular files, and this helps to simplify the
474 * implementation.
475 * On the other hand, some of the pseudo inodes have an irregular point:
476 * They don't have valid inode->i_sb pointer because their lifetimes are
477 * longer than those of the super block structs; they may continue for
478 * several consecutive mounts/umounts. This would need discussions.
479 */
Ryusuke Konishi5731e192009-11-12 22:42:04 +0900480/**
481 * nilfs_mdt_new_common - allocate a pseudo inode for metadata file
482 * @nilfs: nilfs object
483 * @sb: super block instance the metadata file belongs to
484 * @ino: inode number
Ryusuke Konishi5731e192009-11-12 22:42:04 +0900485 */
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700486struct inode *
487nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900488 ino_t ino)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700489{
Ryusuke Konishia53b4752009-05-27 22:11:46 +0900490 struct inode *inode = nilfs_alloc_inode_common(nilfs);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700491
492 if (!inode)
493 return NULL;
494 else {
495 struct address_space * const mapping = &inode->i_data;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700496
497 inode->i_sb = sb; /* sb may be NULL for some meta data files */
498 inode->i_blkbits = nilfs->ns_blocksize_bits;
499 inode->i_flags = 0;
500 atomic_set(&inode->i_count, 1);
501 inode->i_nlink = 1;
502 inode->i_ino = ino;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700503
504#ifdef INIT_UNUSED_INODE_FIELDS
505 atomic_set(&inode->i_writecount, 0);
506 inode->i_size = 0;
507 inode->i_blocks = 0;
508 inode->i_bytes = 0;
509 inode->i_generation = 0;
510#ifdef CONFIG_QUOTA
511 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
512#endif
513 inode->i_pipe = NULL;
514 inode->i_bdev = NULL;
515 inode->i_cdev = NULL;
516 inode->i_rdev = 0;
517#ifdef CONFIG_SECURITY
518 inode->i_security = NULL;
519#endif
520 inode->dirtied_when = 0;
521
522 INIT_LIST_HEAD(&inode->i_list);
523 INIT_LIST_HEAD(&inode->i_sb_list);
524 inode->i_state = 0;
525#endif
526
527 spin_lock_init(&inode->i_lock);
528 mutex_init(&inode->i_mutex);
529 init_rwsem(&inode->i_alloc_sem);
530
531 mapping->host = NULL; /* instead of inode */
532 mapping->flags = 0;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700533 mapping->assoc_mapping = NULL;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700534
535 inode->i_mapping = mapping;
536 }
537
538 return inode;
539}
540
541struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb,
Ryusuke Konishi5731e192009-11-12 22:42:04 +0900542 ino_t ino, size_t objsz)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700543{
Ryusuke Konishi5731e192009-11-12 22:42:04 +0900544 struct inode *inode;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700545
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900546 inode = nilfs_mdt_new_common(nilfs, sb, ino);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700547 if (!inode)
548 return NULL;
549
Ryusuke Konishif1e89c82010-09-05 12:20:59 +0900550 if (nilfs_mdt_init(inode, NILFS_MDT_GFP, objsz) < 0) {
Ryusuke Konishi5e19a992010-08-21 22:01:51 +0900551 nilfs_destroy_inode(inode);
552 return NULL;
553 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700554 return inode;
555}
556
557void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
558 unsigned header_size)
559{
560 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
561
562 mi->mi_entry_size = entry_size;
563 mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
564 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
565}
566
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900567static const struct address_space_operations shadow_map_aops = {
568 .sync_page = block_sync_page,
569};
570
571/**
572 * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
573 * @inode: inode of the metadata file
574 * @shadow: shadow mapping
575 */
576int nilfs_mdt_setup_shadow_map(struct inode *inode,
577 struct nilfs_shadow_map *shadow)
578{
579 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
580 struct backing_dev_info *bdi = NILFS_I_NILFS(inode)->ns_bdi;
581
582 INIT_LIST_HEAD(&shadow->frozen_buffers);
583 nilfs_mapping_init_once(&shadow->frozen_data);
584 nilfs_mapping_init(&shadow->frozen_data, bdi, &shadow_map_aops);
585 nilfs_mapping_init_once(&shadow->frozen_btnodes);
586 nilfs_mapping_init(&shadow->frozen_btnodes, bdi, &shadow_map_aops);
587 mi->mi_shadow = shadow;
588 return 0;
589}
590
591/**
592 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
593 * @inode: inode of the metadata file
594 */
595int nilfs_mdt_save_to_shadow_map(struct inode *inode)
596{
597 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
598 struct nilfs_inode_info *ii = NILFS_I(inode);
599 struct nilfs_shadow_map *shadow = mi->mi_shadow;
600 int ret;
601
602 ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
603 if (ret)
604 goto out;
605
606 ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
607 &ii->i_btnode_cache);
608 if (ret)
609 goto out;
610
611 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
612 out:
613 return ret;
614}
615
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900616int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
617{
618 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
619 struct buffer_head *bh_frozen;
620 struct page *page;
621 int blkbits = inode->i_blkbits;
622 int ret = -ENOMEM;
623
624 page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
625 if (!page)
626 return ret;
627
628 if (!page_has_buffers(page))
629 create_empty_buffers(page, 1 << blkbits, 0);
630
631 bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
632 if (bh_frozen) {
633 if (!buffer_uptodate(bh_frozen))
634 nilfs_copy_buffer(bh_frozen, bh);
635 if (list_empty(&bh_frozen->b_assoc_buffers)) {
636 list_add_tail(&bh_frozen->b_assoc_buffers,
637 &shadow->frozen_buffers);
638 set_buffer_nilfs_redirected(bh);
639 } else {
640 brelse(bh_frozen); /* already frozen */
641 }
642 ret = 0;
643 }
644 unlock_page(page);
645 page_cache_release(page);
646 return ret;
647}
648
649struct buffer_head *
650nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
651{
652 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
653 struct buffer_head *bh_frozen = NULL;
654 struct page *page;
655 int n;
656
657 page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
658 if (page) {
659 if (page_has_buffers(page)) {
660 n = bh_offset(bh) >> inode->i_blkbits;
661 bh_frozen = nilfs_page_get_nth_block(page, n);
662 }
663 unlock_page(page);
664 page_cache_release(page);
665 }
666 return bh_frozen;
667}
668
669static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
670{
671 struct list_head *head = &shadow->frozen_buffers;
672 struct buffer_head *bh;
673
674 while (!list_empty(head)) {
675 bh = list_first_entry(head, struct buffer_head,
676 b_assoc_buffers);
677 list_del_init(&bh->b_assoc_buffers);
678 brelse(bh); /* drop ref-count to make it releasable */
679 }
680}
681
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900682/**
683 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
684 * @inode: inode of the metadata file
685 */
686void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
687{
688 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
689 struct nilfs_inode_info *ii = NILFS_I(inode);
690 struct nilfs_shadow_map *shadow = mi->mi_shadow;
691
692 down_write(&mi->mi_sem);
693
694 if (mi->mi_palloc_cache)
695 nilfs_palloc_clear_cache(inode);
696
697 nilfs_clear_dirty_pages(inode->i_mapping);
698 nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
699
700 nilfs_clear_dirty_pages(&ii->i_btnode_cache);
701 nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
702
703 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
704
705 up_write(&mi->mi_sem);
706}
707
708/**
709 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
710 * @inode: inode of the metadata file
711 */
712void nilfs_mdt_clear_shadow_map(struct inode *inode)
713{
714 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
715 struct nilfs_shadow_map *shadow = mi->mi_shadow;
716
717 down_write(&mi->mi_sem);
Ryusuke Konishib1f6a4f2010-08-31 11:40:34 +0900718 nilfs_release_frozen_buffers(shadow);
Ryusuke Konishiebdfed42010-09-06 12:05:43 +0900719 truncate_inode_pages(&shadow->frozen_data, 0);
720 truncate_inode_pages(&shadow->frozen_btnodes, 0);
721 up_write(&mi->mi_sem);
722}
723
Ryusuke Konishifd66c0d2009-11-13 02:25:41 +0900724static void nilfs_mdt_clear(struct inode *inode)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700725{
726 struct nilfs_inode_info *ii = NILFS_I(inode);
727
728 invalidate_mapping_pages(inode->i_mapping, 0, -1);
729 truncate_inode_pages(inode->i_mapping, 0);
730
Ryusuke Konishifd66c0d2009-11-13 02:25:41 +0900731 if (test_bit(NILFS_I_BMAP, &ii->i_state))
732 nilfs_bmap_clear(ii->i_bmap);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700733 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
734}
735
736void nilfs_mdt_destroy(struct inode *inode)
737{
738 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
739
Ryusuke Konishidb38d5a2009-11-14 15:54:27 +0900740 if (mdi->mi_palloc_cache)
741 nilfs_palloc_destroy_cache(inode);
Ryusuke Konishifd66c0d2009-11-13 02:25:41 +0900742 nilfs_mdt_clear(inode);
743
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700744 nilfs_destroy_inode(inode);
745}