blob: f6326112d647537c5ea0419d36ec4b73d466fbbd [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>
29#include "nilfs.h"
30#include "segment.h"
31#include "page.h"
32#include "mdt.h"
33
34
35#define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
36
37#define INIT_UNUSED_INODE_FIELDS
38
39static int
40nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
41 struct buffer_head *bh,
42 void (*init_block)(struct inode *,
43 struct buffer_head *, void *))
44{
45 struct nilfs_inode_info *ii = NILFS_I(inode);
46 void *kaddr;
47 int ret;
48
49 /* Caller exclude read accesses using page lock */
50
51 /* set_buffer_new(bh); */
52 bh->b_blocknr = 0;
53
54 ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
55 if (unlikely(ret))
56 return ret;
57
58 set_buffer_mapped(bh);
59
60 kaddr = kmap_atomic(bh->b_page, KM_USER0);
61 memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
62 if (init_block)
63 init_block(inode, bh, kaddr);
64 flush_dcache_page(bh->b_page);
65 kunmap_atomic(kaddr, KM_USER0);
66
67 set_buffer_uptodate(bh);
68 nilfs_mark_buffer_dirty(bh);
69 nilfs_mdt_mark_dirty(inode);
70 return 0;
71}
72
73static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
74 struct buffer_head **out_bh,
75 void (*init_block)(struct inode *,
76 struct buffer_head *,
77 void *))
78{
79 struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070080 struct super_block *sb = inode->i_sb;
81 struct nilfs_transaction_info ti;
82 struct buffer_head *bh;
83 int err;
84
85 if (!sb) {
Ryusuke Konishi201913e2009-04-28 21:04:59 +090086 /*
87 * Make sure this function is not called from any
88 * read-only context.
89 */
90 if (!nilfs->ns_writer) {
91 WARN_ON(1);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070092 err = -EROFS;
93 goto out;
94 }
Ryusuke Konishi201913e2009-04-28 21:04:59 +090095 sb = nilfs->ns_writer->s_super;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -070096 }
97
98 nilfs_transaction_begin(sb, &ti, 0);
99
100 err = -ENOMEM;
101 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
102 if (unlikely(!bh))
103 goto failed_unlock;
104
105 err = -EEXIST;
Ryusuke Konishi14351102009-09-06 17:49:49 +0900106 if (buffer_uptodate(bh))
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700107 goto failed_bh;
Ryusuke Konishi14351102009-09-06 17:49:49 +0900108
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700109 wait_on_buffer(bh);
110 if (buffer_uptodate(bh))
111 goto failed_bh;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700112
113 bh->b_bdev = nilfs->ns_bdev;
114 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
115 if (likely(!err)) {
116 get_bh(bh);
117 *out_bh = bh;
118 }
119
120 failed_bh:
121 unlock_page(bh->b_page);
122 page_cache_release(bh->b_page);
123 brelse(bh);
124
125 failed_unlock:
Ryusuke Konishi47420c72009-04-06 19:01:45 -0700126 if (likely(!err))
127 err = nilfs_transaction_commit(sb);
128 else
129 nilfs_transaction_abort(sb);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700130 out:
131 return err;
132}
133
134static int
135nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
136 int mode, struct buffer_head **out_bh)
137{
138 struct buffer_head *bh;
Ryusuke Konishi0f3fe332009-08-15 02:29:28 +0900139 __u64 blknum = 0;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700140 int ret = -ENOMEM;
141
142 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
143 if (unlikely(!bh))
144 goto failed;
145
146 ret = -EEXIST; /* internal code */
147 if (buffer_uptodate(bh))
148 goto out;
149
150 if (mode == READA) {
151 if (!trylock_buffer(bh)) {
152 ret = -EBUSY;
153 goto failed_bh;
154 }
Ryusuke Konishi1f5abe72009-04-06 19:01:55 -0700155 } else /* mode == READ */
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700156 lock_buffer(bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700157
158 if (buffer_uptodate(bh)) {
159 unlock_buffer(bh);
160 goto out;
161 }
Ryusuke Konishi14351102009-09-06 17:49:49 +0900162
163 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
164 if (unlikely(ret)) {
165 unlock_buffer(bh);
166 goto failed_bh;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700167 }
Ryusuke Konishi14351102009-09-06 17:49:49 +0900168 bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev;
Ryusuke Konishi0f3fe332009-08-15 02:29:28 +0900169 bh->b_blocknr = (sector_t)blknum;
Ryusuke Konishi14351102009-09-06 17:49:49 +0900170 set_buffer_mapped(bh);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700171
172 bh->b_end_io = end_buffer_read_sync;
173 get_bh(bh);
174 submit_bh(mode, bh);
175 ret = 0;
176 out:
177 get_bh(bh);
178 *out_bh = bh;
179
180 failed_bh:
181 unlock_page(bh->b_page);
182 page_cache_release(bh->b_page);
183 brelse(bh);
184 failed:
185 return ret;
186}
187
188static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
189 struct buffer_head **out_bh)
190{
191 struct buffer_head *first_bh, *bh;
192 unsigned long blkoff;
193 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
194 int err;
195
196 err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
197 if (err == -EEXIST) /* internal code */
198 goto out;
199
200 if (unlikely(err))
201 goto failed;
202
203 blkoff = block + 1;
204 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
205 err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
206 if (likely(!err || err == -EEXIST))
207 brelse(bh);
208 else if (err != -EBUSY)
209 break; /* abort readahead if bmap lookup failed */
210
211 if (!buffer_locked(first_bh))
212 goto out_no_wait;
213 }
214
215 wait_on_buffer(first_bh);
216
217 out_no_wait:
218 err = -EIO;
219 if (!buffer_uptodate(first_bh))
220 goto failed_bh;
221 out:
222 *out_bh = first_bh;
223 return 0;
224
225 failed_bh:
226 brelse(first_bh);
227 failed:
228 return err;
229}
230
231/**
232 * nilfs_mdt_get_block - read or create a buffer on meta data file.
233 * @inode: inode of the meta data file
234 * @blkoff: block offset
235 * @create: create flag
236 * @init_block: initializer used for newly allocated block
237 * @out_bh: output of a pointer to the buffer_head
238 *
239 * nilfs_mdt_get_block() looks up the specified buffer and tries to create
240 * a new buffer if @create is not zero. On success, the returned buffer is
241 * assured to be either existing or formatted using a buffer lock on success.
242 * @out_bh is substituted only when zero is returned.
243 *
244 * Return Value: On success, it returns 0. On error, the following negative
245 * error code is returned.
246 *
247 * %-ENOMEM - Insufficient memory available.
248 *
249 * %-EIO - I/O error
250 *
251 * %-ENOENT - the specified block does not exist (hole block)
252 *
253 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
254 *
255 * %-EROFS - Read only filesystem (for create mode)
256 */
257int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
258 void (*init_block)(struct inode *,
259 struct buffer_head *, void *),
260 struct buffer_head **out_bh)
261{
262 int ret;
263
264 /* Should be rewritten with merging nilfs_mdt_read_block() */
265 retry:
266 ret = nilfs_mdt_read_block(inode, blkoff, out_bh);
267 if (!create || ret != -ENOENT)
268 return ret;
269
270 ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
271 if (unlikely(ret == -EEXIST)) {
272 /* create = 0; */ /* limit read-create loop retries */
273 goto retry;
274 }
275 return ret;
276}
277
278/**
279 * nilfs_mdt_delete_block - make a hole on the meta data file.
280 * @inode: inode of the meta data file
281 * @block: block offset
282 *
283 * Return Value: On success, zero is returned.
284 * On error, one of the following negative error code is returned.
285 *
286 * %-ENOMEM - Insufficient memory available.
287 *
288 * %-EIO - I/O error
289 *
290 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
291 */
292int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
293{
294 struct nilfs_inode_info *ii = NILFS_I(inode);
295 int err;
296
297 err = nilfs_bmap_delete(ii->i_bmap, block);
Ryusuke Konishi84338232009-05-05 21:52:06 +0900298 if (!err || err == -ENOENT) {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700299 nilfs_mdt_mark_dirty(inode);
300 nilfs_mdt_forget_block(inode, block);
301 }
302 return err;
303}
304
305/**
306 * nilfs_mdt_forget_block - discard dirty state and try to remove the page
307 * @inode: inode of the meta data file
308 * @block: block offset
309 *
310 * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
311 * tries to release the page including the buffer from a page cache.
312 *
313 * Return Value: On success, 0 is returned. On error, one of the following
314 * negative error code is returned.
315 *
316 * %-EBUSY - page has an active buffer.
317 *
318 * %-ENOENT - page cache has no page addressed by the offset.
319 */
320int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
321{
322 pgoff_t index = (pgoff_t)block >>
323 (PAGE_CACHE_SHIFT - inode->i_blkbits);
324 struct page *page;
325 unsigned long first_block;
326 int ret = 0;
327 int still_dirty;
328
329 page = find_lock_page(inode->i_mapping, index);
330 if (!page)
331 return -ENOENT;
332
333 wait_on_page_writeback(page);
334
335 first_block = (unsigned long)index <<
336 (PAGE_CACHE_SHIFT - inode->i_blkbits);
337 if (page_has_buffers(page)) {
338 struct buffer_head *bh;
339
340 bh = nilfs_page_get_nth_block(page, block - first_block);
341 nilfs_forget_buffer(bh);
342 }
343 still_dirty = PageDirty(page);
344 unlock_page(page);
345 page_cache_release(page);
346
347 if (still_dirty ||
348 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
349 ret = -EBUSY;
350 return ret;
351}
352
353/**
354 * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
355 * @inode: inode of the meta data file
356 * @block: block offset
357 *
358 * Return Value: On success, it returns 0. On error, the following negative
359 * error code is returned.
360 *
361 * %-ENOMEM - Insufficient memory available.
362 *
363 * %-EIO - I/O error
364 *
365 * %-ENOENT - the specified block does not exist (hole block)
366 *
367 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
368 */
369int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
370{
371 struct buffer_head *bh;
372 int err;
373
374 err = nilfs_mdt_read_block(inode, block, &bh);
375 if (unlikely(err))
376 return err;
377 nilfs_mark_buffer_dirty(bh);
378 nilfs_mdt_mark_dirty(inode);
379 brelse(bh);
380 return 0;
381}
382
383int nilfs_mdt_fetch_dirty(struct inode *inode)
384{
385 struct nilfs_inode_info *ii = NILFS_I(inode);
386
387 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
388 set_bit(NILFS_I_DIRTY, &ii->i_state);
389 return 1;
390 }
391 return test_bit(NILFS_I_DIRTY, &ii->i_state);
392}
393
394static int
395nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
396{
397 struct inode *inode = container_of(page->mapping,
398 struct inode, i_data);
399 struct super_block *sb = inode->i_sb;
Ryusuke Konishi027d6402009-08-02 22:45:33 +0900400 struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700401 struct nilfs_sb_info *writer = NULL;
402 int err = 0;
403
404 redirty_page_for_writepage(wbc, page);
405 unlock_page(page);
406
407 if (page->mapping->assoc_mapping)
408 return 0; /* Do not request flush for shadow page cache */
409 if (!sb) {
Ryusuke Konishi027d6402009-08-02 22:45:33 +0900410 down_read(&nilfs->ns_writer_sem);
411 writer = nilfs->ns_writer;
Ryusuke Konishi01a261e2009-08-02 17:45:55 +0900412 if (!writer) {
Ryusuke Konishi027d6402009-08-02 22:45:33 +0900413 up_read(&nilfs->ns_writer_sem);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700414 return -EROFS;
Ryusuke Konishi01a261e2009-08-02 17:45:55 +0900415 }
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700416 sb = writer->s_super;
417 }
418
419 if (wbc->sync_mode == WB_SYNC_ALL)
420 err = nilfs_construct_segment(sb);
421 else if (wbc->for_reclaim)
422 nilfs_flush_segment(sb, inode->i_ino);
423
424 if (writer)
Ryusuke Konishi027d6402009-08-02 22:45:33 +0900425 up_read(&nilfs->ns_writer_sem);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700426 return err;
427}
428
429
Alexey Dobriyan7f094102009-09-21 17:01:10 -0700430static const struct address_space_operations def_mdt_aops = {
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700431 .writepage = nilfs_mdt_write_page,
Ryusuke Konishifa032742009-05-27 22:44:34 +0900432 .sync_page = block_sync_page,
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700433};
434
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700435static const struct inode_operations def_mdt_iops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700436static const struct file_operations def_mdt_fops;
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700437
438/*
439 * NILFS2 uses pseudo inodes for meta data files such as DAT, cpfile, sufile,
440 * ifile, or gcinodes. This allows the B-tree code and segment constructor
441 * to treat them like regular files, and this helps to simplify the
442 * implementation.
443 * On the other hand, some of the pseudo inodes have an irregular point:
444 * They don't have valid inode->i_sb pointer because their lifetimes are
445 * longer than those of the super block structs; they may continue for
446 * several consecutive mounts/umounts. This would need discussions.
447 */
448struct inode *
449nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
450 ino_t ino, gfp_t gfp_mask)
451{
Ryusuke Konishia53b4752009-05-27 22:11:46 +0900452 struct inode *inode = nilfs_alloc_inode_common(nilfs);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700453
454 if (!inode)
455 return NULL;
456 else {
457 struct address_space * const mapping = &inode->i_data;
458 struct nilfs_mdt_info *mi = kzalloc(sizeof(*mi), GFP_NOFS);
459
460 if (!mi) {
461 nilfs_destroy_inode(inode);
462 return NULL;
463 }
464 mi->mi_nilfs = nilfs;
465 init_rwsem(&mi->mi_sem);
466
467 inode->i_sb = sb; /* sb may be NULL for some meta data files */
468 inode->i_blkbits = nilfs->ns_blocksize_bits;
469 inode->i_flags = 0;
470 atomic_set(&inode->i_count, 1);
471 inode->i_nlink = 1;
472 inode->i_ino = ino;
473 inode->i_mode = S_IFREG;
474 inode->i_private = mi;
475
476#ifdef INIT_UNUSED_INODE_FIELDS
477 atomic_set(&inode->i_writecount, 0);
478 inode->i_size = 0;
479 inode->i_blocks = 0;
480 inode->i_bytes = 0;
481 inode->i_generation = 0;
482#ifdef CONFIG_QUOTA
483 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
484#endif
485 inode->i_pipe = NULL;
486 inode->i_bdev = NULL;
487 inode->i_cdev = NULL;
488 inode->i_rdev = 0;
489#ifdef CONFIG_SECURITY
490 inode->i_security = NULL;
491#endif
492 inode->dirtied_when = 0;
493
494 INIT_LIST_HEAD(&inode->i_list);
495 INIT_LIST_HEAD(&inode->i_sb_list);
496 inode->i_state = 0;
497#endif
498
499 spin_lock_init(&inode->i_lock);
500 mutex_init(&inode->i_mutex);
501 init_rwsem(&inode->i_alloc_sem);
502
503 mapping->host = NULL; /* instead of inode */
504 mapping->flags = 0;
505 mapping_set_gfp_mask(mapping, gfp_mask);
506 mapping->assoc_mapping = NULL;
507 mapping->backing_dev_info = nilfs->ns_bdi;
508
509 inode->i_mapping = mapping;
510 }
511
512 return inode;
513}
514
515struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb,
Ryusuke Konishi7a102b02009-08-15 13:47:09 +0900516 ino_t ino)
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700517{
Ryusuke Konishi7a102b02009-08-15 13:47:09 +0900518 struct inode *inode = nilfs_mdt_new_common(nilfs, sb, ino,
519 NILFS_MDT_GFP);
Ryusuke Konishi5eb563f2009-04-06 19:01:28 -0700520
521 if (!inode)
522 return NULL;
523
524 inode->i_op = &def_mdt_iops;
525 inode->i_fop = &def_mdt_fops;
526 inode->i_mapping->a_ops = &def_mdt_aops;
527 return inode;
528}
529
530void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
531 unsigned header_size)
532{
533 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
534
535 mi->mi_entry_size = entry_size;
536 mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
537 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
538}
539
540void nilfs_mdt_set_shadow(struct inode *orig, struct inode *shadow)
541{
542 shadow->i_mapping->assoc_mapping = orig->i_mapping;
543 NILFS_I(shadow)->i_btnode_cache.assoc_mapping =
544 &NILFS_I(orig)->i_btnode_cache;
545}
546
547void nilfs_mdt_clear(struct inode *inode)
548{
549 struct nilfs_inode_info *ii = NILFS_I(inode);
550
551 invalidate_mapping_pages(inode->i_mapping, 0, -1);
552 truncate_inode_pages(inode->i_mapping, 0);
553
554 nilfs_bmap_clear(ii->i_bmap);
555 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
556}
557
558void nilfs_mdt_destroy(struct inode *inode)
559{
560 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
561
562 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
563 kfree(mdi);
564 nilfs_destroy_inode(inode);
565}