blob: 47dd815433fd4df6b5e484b3b52cbc6a0cc5294f [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;
80 struct nilfs_sb_info *writer = NULL;
81 struct super_block *sb = inode->i_sb;
82 struct nilfs_transaction_info ti;
83 struct buffer_head *bh;
84 int err;
85
86 if (!sb) {
87 writer = nilfs_get_writer(nilfs);
88 if (!writer) {
89 err = -EROFS;
90 goto out;
91 }
92 sb = writer->s_super;
93 }
94
95 nilfs_transaction_begin(sb, &ti, 0);
96
97 err = -ENOMEM;
98 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
99 if (unlikely(!bh))
100 goto failed_unlock;
101
102 err = -EEXIST;
103 if (buffer_uptodate(bh) || buffer_mapped(bh))
104 goto failed_bh;
105#if 0
106 /* The uptodate flag is not protected by the page lock, but
107 the mapped flag is. Thus, we don't have to wait the buffer. */
108 wait_on_buffer(bh);
109 if (buffer_uptodate(bh))
110 goto failed_bh;
111#endif
112
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 if (writer)
131 nilfs_put_writer(nilfs);
132 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;
141 unsigned long blknum = 0;
142 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 }
164 if (!buffer_mapped(bh)) { /* unused buffer */
165 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff,
166 &blknum);
167 if (unlikely(ret)) {
168 unlock_buffer(bh);
169 goto failed_bh;
170 }
171 bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev;
172 bh->b_blocknr = blknum;
173 set_buffer_mapped(bh);
174 }
175
176 bh->b_end_io = end_buffer_read_sync;
177 get_bh(bh);
178 submit_bh(mode, bh);
179 ret = 0;
180 out:
181 get_bh(bh);
182 *out_bh = bh;
183
184 failed_bh:
185 unlock_page(bh->b_page);
186 page_cache_release(bh->b_page);
187 brelse(bh);
188 failed:
189 return ret;
190}
191
192static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
193 struct buffer_head **out_bh)
194{
195 struct buffer_head *first_bh, *bh;
196 unsigned long blkoff;
197 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
198 int err;
199
200 err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
201 if (err == -EEXIST) /* internal code */
202 goto out;
203
204 if (unlikely(err))
205 goto failed;
206
207 blkoff = block + 1;
208 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
209 err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
210 if (likely(!err || err == -EEXIST))
211 brelse(bh);
212 else if (err != -EBUSY)
213 break; /* abort readahead if bmap lookup failed */
214
215 if (!buffer_locked(first_bh))
216 goto out_no_wait;
217 }
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:
270 ret = nilfs_mdt_read_block(inode, blkoff, out_bh);
271 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);
302 if (likely(!err)) {
303 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
378 err = nilfs_mdt_read_block(inode, block, &bh);
379 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{
401 struct inode *inode = container_of(page->mapping,
402 struct inode, i_data);
403 struct super_block *sb = inode->i_sb;
404 struct nilfs_sb_info *writer = NULL;
405 int err = 0;
406
407 redirty_page_for_writepage(wbc, page);
408 unlock_page(page);
409
410 if (page->mapping->assoc_mapping)
411 return 0; /* Do not request flush for shadow page cache */
412 if (!sb) {
413 writer = nilfs_get_writer(NILFS_MDT(inode)->mi_nilfs);
414 if (!writer)
415 return -EROFS;
416 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)
425 nilfs_put_writer(NILFS_MDT(inode)->mi_nilfs);
426 return err;
427}
428
429
430static struct address_space_operations def_mdt_aops = {
431 .writepage = nilfs_mdt_write_page,
432};
433
434static struct inode_operations def_mdt_iops;
435static struct file_operations def_mdt_fops;
436
437/*
438 * NILFS2 uses pseudo inodes for meta data files such as DAT, cpfile, sufile,
439 * ifile, or gcinodes. This allows the B-tree code and segment constructor
440 * to treat them like regular files, and this helps to simplify the
441 * implementation.
442 * On the other hand, some of the pseudo inodes have an irregular point:
443 * They don't have valid inode->i_sb pointer because their lifetimes are
444 * longer than those of the super block structs; they may continue for
445 * several consecutive mounts/umounts. This would need discussions.
446 */
447struct inode *
448nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
449 ino_t ino, gfp_t gfp_mask)
450{
451 struct inode *inode = nilfs_alloc_inode(sb);
452
453 if (!inode)
454 return NULL;
455 else {
456 struct address_space * const mapping = &inode->i_data;
457 struct nilfs_mdt_info *mi = kzalloc(sizeof(*mi), GFP_NOFS);
458
459 if (!mi) {
460 nilfs_destroy_inode(inode);
461 return NULL;
462 }
463 mi->mi_nilfs = nilfs;
464 init_rwsem(&mi->mi_sem);
465
466 inode->i_sb = sb; /* sb may be NULL for some meta data files */
467 inode->i_blkbits = nilfs->ns_blocksize_bits;
468 inode->i_flags = 0;
469 atomic_set(&inode->i_count, 1);
470 inode->i_nlink = 1;
471 inode->i_ino = ino;
472 inode->i_mode = S_IFREG;
473 inode->i_private = mi;
474
475#ifdef INIT_UNUSED_INODE_FIELDS
476 atomic_set(&inode->i_writecount, 0);
477 inode->i_size = 0;
478 inode->i_blocks = 0;
479 inode->i_bytes = 0;
480 inode->i_generation = 0;
481#ifdef CONFIG_QUOTA
482 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
483#endif
484 inode->i_pipe = NULL;
485 inode->i_bdev = NULL;
486 inode->i_cdev = NULL;
487 inode->i_rdev = 0;
488#ifdef CONFIG_SECURITY
489 inode->i_security = NULL;
490#endif
491 inode->dirtied_when = 0;
492
493 INIT_LIST_HEAD(&inode->i_list);
494 INIT_LIST_HEAD(&inode->i_sb_list);
495 inode->i_state = 0;
496#endif
497
498 spin_lock_init(&inode->i_lock);
499 mutex_init(&inode->i_mutex);
500 init_rwsem(&inode->i_alloc_sem);
501
502 mapping->host = NULL; /* instead of inode */
503 mapping->flags = 0;
504 mapping_set_gfp_mask(mapping, gfp_mask);
505 mapping->assoc_mapping = NULL;
506 mapping->backing_dev_info = nilfs->ns_bdi;
507
508 inode->i_mapping = mapping;
509 }
510
511 return inode;
512}
513
514struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb,
515 ino_t ino, gfp_t gfp_mask)
516{
517 struct inode *inode = nilfs_mdt_new_common(nilfs, sb, ino, gfp_mask);
518
519 if (!inode)
520 return NULL;
521
522 inode->i_op = &def_mdt_iops;
523 inode->i_fop = &def_mdt_fops;
524 inode->i_mapping->a_ops = &def_mdt_aops;
525 return inode;
526}
527
528void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
529 unsigned header_size)
530{
531 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
532
533 mi->mi_entry_size = entry_size;
534 mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
535 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
536}
537
538void nilfs_mdt_set_shadow(struct inode *orig, struct inode *shadow)
539{
540 shadow->i_mapping->assoc_mapping = orig->i_mapping;
541 NILFS_I(shadow)->i_btnode_cache.assoc_mapping =
542 &NILFS_I(orig)->i_btnode_cache;
543}
544
545void nilfs_mdt_clear(struct inode *inode)
546{
547 struct nilfs_inode_info *ii = NILFS_I(inode);
548
549 invalidate_mapping_pages(inode->i_mapping, 0, -1);
550 truncate_inode_pages(inode->i_mapping, 0);
551
552 nilfs_bmap_clear(ii->i_bmap);
553 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
554}
555
556void nilfs_mdt_destroy(struct inode *inode)
557{
558 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
559
560 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
561 kfree(mdi);
562 nilfs_destroy_inode(inode);
563}