2 * mdt.c - meta data file for NILFS
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
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.
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.
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
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
23 #include <linux/buffer_head.h>
24 #include <linux/mpage.h>
26 #include <linux/writeback.h>
27 #include <linux/backing-dev.h>
28 #include <linux/swap.h>
35 #define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
37 #define INIT_UNUSED_INODE_FIELDS
40 nilfs_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 *))
45 struct nilfs_inode_info *ii = NILFS_I(inode);
49 /* Caller exclude read accesses using page lock */
51 /* set_buffer_new(bh); */
54 ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
58 set_buffer_mapped(bh);
60 kaddr = kmap_atomic(bh->b_page, KM_USER0);
61 memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
63 init_block(inode, bh, kaddr);
64 flush_dcache_page(bh->b_page);
65 kunmap_atomic(kaddr, KM_USER0);
67 set_buffer_uptodate(bh);
68 nilfs_mark_buffer_dirty(bh);
69 nilfs_mdt_mark_dirty(inode);
73 static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
74 struct buffer_head **out_bh,
75 void (*init_block)(struct inode *,
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;
87 writer = nilfs_get_writer(nilfs);
95 nilfs_transaction_begin(sb, &ti, 0);
98 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
103 if (buffer_uptodate(bh) || buffer_mapped(bh))
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. */
109 if (buffer_uptodate(bh))
113 bh->b_bdev = nilfs->ns_bdev;
114 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
121 unlock_page(bh->b_page);
122 page_cache_release(bh->b_page);
127 err = nilfs_transaction_commit(sb);
129 nilfs_transaction_abort(sb);
131 nilfs_put_writer(nilfs);
137 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
138 int mode, struct buffer_head **out_bh)
140 struct buffer_head *bh;
141 unsigned long blknum = 0;
144 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
148 ret = -EEXIST; /* internal code */
149 if (buffer_uptodate(bh))
153 if (!trylock_buffer(bh)) {
158 BUG_ON(mode != READ);
162 if (buffer_uptodate(bh)) {
166 if (!buffer_mapped(bh)) { /* unused buffer */
167 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff,
173 bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev;
174 bh->b_blocknr = blknum;
175 set_buffer_mapped(bh);
178 bh->b_end_io = end_buffer_read_sync;
187 unlock_page(bh->b_page);
188 page_cache_release(bh->b_page);
194 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
195 struct buffer_head **out_bh)
197 struct buffer_head *first_bh, *bh;
198 unsigned long blkoff;
199 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
202 err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
203 if (err == -EEXIST) /* internal code */
210 for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
211 err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
212 if (likely(!err || err == -EEXIST))
214 else if (err != -EBUSY)
215 break; /* abort readahead if bmap lookup failed */
217 if (!buffer_locked(first_bh))
221 wait_on_buffer(first_bh);
225 if (!buffer_uptodate(first_bh))
238 * nilfs_mdt_get_block - read or create a buffer on meta data file.
239 * @inode: inode of the meta data file
240 * @blkoff: block offset
241 * @create: create flag
242 * @init_block: initializer used for newly allocated block
243 * @out_bh: output of a pointer to the buffer_head
245 * nilfs_mdt_get_block() looks up the specified buffer and tries to create
246 * a new buffer if @create is not zero. On success, the returned buffer is
247 * assured to be either existing or formatted using a buffer lock on success.
248 * @out_bh is substituted only when zero is returned.
250 * Return Value: On success, it returns 0. On error, the following negative
251 * error code is returned.
253 * %-ENOMEM - Insufficient memory available.
257 * %-ENOENT - the specified block does not exist (hole block)
259 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
261 * %-EROFS - Read only filesystem (for create mode)
263 int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
264 void (*init_block)(struct inode *,
265 struct buffer_head *, void *),
266 struct buffer_head **out_bh)
270 /* Should be rewritten with merging nilfs_mdt_read_block() */
272 ret = nilfs_mdt_read_block(inode, blkoff, out_bh);
273 if (!create || ret != -ENOENT)
276 ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
277 if (unlikely(ret == -EEXIST)) {
278 /* create = 0; */ /* limit read-create loop retries */
285 * nilfs_mdt_delete_block - make a hole on the meta data file.
286 * @inode: inode of the meta data file
287 * @block: block offset
289 * Return Value: On success, zero is returned.
290 * On error, one of the following negative error code is returned.
292 * %-ENOMEM - Insufficient memory available.
296 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
298 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
300 struct nilfs_inode_info *ii = NILFS_I(inode);
303 err = nilfs_bmap_delete(ii->i_bmap, block);
305 nilfs_mdt_mark_dirty(inode);
306 nilfs_mdt_forget_block(inode, block);
312 * nilfs_mdt_forget_block - discard dirty state and try to remove the page
313 * @inode: inode of the meta data file
314 * @block: block offset
316 * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
317 * tries to release the page including the buffer from a page cache.
319 * Return Value: On success, 0 is returned. On error, one of the following
320 * negative error code is returned.
322 * %-EBUSY - page has an active buffer.
324 * %-ENOENT - page cache has no page addressed by the offset.
326 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
328 pgoff_t index = (pgoff_t)block >>
329 (PAGE_CACHE_SHIFT - inode->i_blkbits);
331 unsigned long first_block;
335 page = find_lock_page(inode->i_mapping, index);
339 wait_on_page_writeback(page);
341 first_block = (unsigned long)index <<
342 (PAGE_CACHE_SHIFT - inode->i_blkbits);
343 if (page_has_buffers(page)) {
344 struct buffer_head *bh;
346 bh = nilfs_page_get_nth_block(page, block - first_block);
347 nilfs_forget_buffer(bh);
349 still_dirty = PageDirty(page);
351 page_cache_release(page);
354 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
360 * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
361 * @inode: inode of the meta data file
362 * @block: block offset
364 * Return Value: On success, it returns 0. On error, the following negative
365 * error code is returned.
367 * %-ENOMEM - Insufficient memory available.
371 * %-ENOENT - the specified block does not exist (hole block)
373 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
375 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
377 struct buffer_head *bh;
380 err = nilfs_mdt_read_block(inode, block, &bh);
383 nilfs_mark_buffer_dirty(bh);
384 nilfs_mdt_mark_dirty(inode);
389 int nilfs_mdt_fetch_dirty(struct inode *inode)
391 struct nilfs_inode_info *ii = NILFS_I(inode);
393 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
394 set_bit(NILFS_I_DIRTY, &ii->i_state);
397 return test_bit(NILFS_I_DIRTY, &ii->i_state);
401 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
403 struct inode *inode = container_of(page->mapping,
404 struct inode, i_data);
405 struct super_block *sb = inode->i_sb;
406 struct nilfs_sb_info *writer = NULL;
409 redirty_page_for_writepage(wbc, page);
412 if (page->mapping->assoc_mapping)
413 return 0; /* Do not request flush for shadow page cache */
415 writer = nilfs_get_writer(NILFS_MDT(inode)->mi_nilfs);
418 sb = writer->s_super;
421 if (wbc->sync_mode == WB_SYNC_ALL)
422 err = nilfs_construct_segment(sb);
423 else if (wbc->for_reclaim)
424 nilfs_flush_segment(sb, inode->i_ino);
427 nilfs_put_writer(NILFS_MDT(inode)->mi_nilfs);
432 static struct address_space_operations def_mdt_aops = {
433 .writepage = nilfs_mdt_write_page,
436 static struct inode_operations def_mdt_iops;
437 static struct file_operations def_mdt_fops;
440 * NILFS2 uses pseudo inodes for meta data files such as DAT, cpfile, sufile,
441 * ifile, or gcinodes. This allows the B-tree code and segment constructor
442 * to treat them like regular files, and this helps to simplify the
444 * On the other hand, some of the pseudo inodes have an irregular point:
445 * They don't have valid inode->i_sb pointer because their lifetimes are
446 * longer than those of the super block structs; they may continue for
447 * several consecutive mounts/umounts. This would need discussions.
450 nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
451 ino_t ino, gfp_t gfp_mask)
453 struct inode *inode = nilfs_alloc_inode(sb);
458 struct address_space * const mapping = &inode->i_data;
459 struct nilfs_mdt_info *mi = kzalloc(sizeof(*mi), GFP_NOFS);
462 nilfs_destroy_inode(inode);
465 mi->mi_nilfs = nilfs;
466 init_rwsem(&mi->mi_sem);
468 inode->i_sb = sb; /* sb may be NULL for some meta data files */
469 inode->i_blkbits = nilfs->ns_blocksize_bits;
471 atomic_set(&inode->i_count, 1);
474 inode->i_mode = S_IFREG;
475 inode->i_private = mi;
477 #ifdef INIT_UNUSED_INODE_FIELDS
478 atomic_set(&inode->i_writecount, 0);
482 inode->i_generation = 0;
484 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
486 inode->i_pipe = NULL;
487 inode->i_bdev = NULL;
488 inode->i_cdev = NULL;
490 #ifdef CONFIG_SECURITY
491 inode->i_security = NULL;
493 inode->dirtied_when = 0;
495 INIT_LIST_HEAD(&inode->i_list);
496 INIT_LIST_HEAD(&inode->i_sb_list);
500 spin_lock_init(&inode->i_lock);
501 mutex_init(&inode->i_mutex);
502 init_rwsem(&inode->i_alloc_sem);
504 mapping->host = NULL; /* instead of inode */
506 mapping_set_gfp_mask(mapping, gfp_mask);
507 mapping->assoc_mapping = NULL;
508 mapping->backing_dev_info = nilfs->ns_bdi;
510 inode->i_mapping = mapping;
516 struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb,
517 ino_t ino, gfp_t gfp_mask)
519 struct inode *inode = nilfs_mdt_new_common(nilfs, sb, ino, gfp_mask);
524 inode->i_op = &def_mdt_iops;
525 inode->i_fop = &def_mdt_fops;
526 inode->i_mapping->a_ops = &def_mdt_aops;
530 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
531 unsigned header_size)
533 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
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);
540 void nilfs_mdt_set_shadow(struct inode *orig, struct inode *shadow)
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;
547 void nilfs_mdt_clear(struct inode *inode)
549 struct nilfs_inode_info *ii = NILFS_I(inode);
551 invalidate_mapping_pages(inode->i_mapping, 0, -1);
552 truncate_inode_pages(inode->i_mapping, 0);
554 nilfs_bmap_clear(ii->i_bmap);
555 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
558 void nilfs_mdt_destroy(struct inode *inode)
560 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
562 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
564 nilfs_destroy_inode(inode);