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
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 super_block *sb = inode->i_sb;
81 struct nilfs_transaction_info ti;
82 struct buffer_head *bh;
87 * Make sure this function is not called from any
90 if (!nilfs->ns_writer) {
95 sb = nilfs->ns_writer->s_super;
98 nilfs_transaction_begin(sb, &ti, 0);
101 bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
106 if (buffer_uptodate(bh))
110 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);
135 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
136 int mode, struct buffer_head **out_bh)
138 struct buffer_head *bh;
142 bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
146 ret = -EEXIST; /* internal code */
147 if (buffer_uptodate(bh))
151 if (!trylock_buffer(bh)) {
155 } else /* mode == READ */
158 if (buffer_uptodate(bh)) {
163 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff, &blknum);
168 bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev;
169 bh->b_blocknr = (sector_t)blknum;
170 set_buffer_mapped(bh);
172 bh->b_end_io = end_buffer_read_sync;
181 unlock_page(bh->b_page);
182 page_cache_release(bh->b_page);
188 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
189 struct buffer_head **out_bh)
191 struct buffer_head *first_bh, *bh;
192 unsigned long blkoff;
193 int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
196 err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
197 if (err == -EEXIST) /* internal code */
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))
208 else if (err != -EBUSY)
209 break; /* abort readahead if bmap lookup failed */
211 if (!buffer_locked(first_bh))
215 wait_on_buffer(first_bh);
219 if (!buffer_uptodate(first_bh))
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
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.
244 * Return Value: On success, it returns 0. On error, the following negative
245 * error code is returned.
247 * %-ENOMEM - Insufficient memory available.
251 * %-ENOENT - the specified block does not exist (hole block)
253 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
255 * %-EROFS - Read only filesystem (for create mode)
257 int 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)
264 /* Should be rewritten with merging nilfs_mdt_read_block() */
266 ret = nilfs_mdt_read_block(inode, blkoff, out_bh);
267 if (!create || ret != -ENOENT)
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 */
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
283 * Return Value: On success, zero is returned.
284 * On error, one of the following negative error code is returned.
286 * %-ENOMEM - Insufficient memory available.
290 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
292 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
294 struct nilfs_inode_info *ii = NILFS_I(inode);
297 err = nilfs_bmap_delete(ii->i_bmap, block);
298 if (!err || err == -ENOENT) {
299 nilfs_mdt_mark_dirty(inode);
300 nilfs_mdt_forget_block(inode, block);
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
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.
313 * Return Value: On success, 0 is returned. On error, one of the following
314 * negative error code is returned.
316 * %-EBUSY - page has an active buffer.
318 * %-ENOENT - page cache has no page addressed by the offset.
320 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
322 pgoff_t index = (pgoff_t)block >>
323 (PAGE_CACHE_SHIFT - inode->i_blkbits);
325 unsigned long first_block;
329 page = find_lock_page(inode->i_mapping, index);
333 wait_on_page_writeback(page);
335 first_block = (unsigned long)index <<
336 (PAGE_CACHE_SHIFT - inode->i_blkbits);
337 if (page_has_buffers(page)) {
338 struct buffer_head *bh;
340 bh = nilfs_page_get_nth_block(page, block - first_block);
341 nilfs_forget_buffer(bh);
343 still_dirty = PageDirty(page);
345 page_cache_release(page);
348 invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
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
358 * Return Value: On success, it returns 0. On error, the following negative
359 * error code is returned.
361 * %-ENOMEM - Insufficient memory available.
365 * %-ENOENT - the specified block does not exist (hole block)
367 * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
369 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
371 struct buffer_head *bh;
374 err = nilfs_mdt_read_block(inode, block, &bh);
377 nilfs_mark_buffer_dirty(bh);
378 nilfs_mdt_mark_dirty(inode);
383 int nilfs_mdt_fetch_dirty(struct inode *inode)
385 struct nilfs_inode_info *ii = NILFS_I(inode);
387 if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
388 set_bit(NILFS_I_DIRTY, &ii->i_state);
391 return test_bit(NILFS_I_DIRTY, &ii->i_state);
395 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
397 struct inode *inode = container_of(page->mapping,
398 struct inode, i_data);
399 struct super_block *sb = inode->i_sb;
400 struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
401 struct nilfs_sb_info *writer = NULL;
404 redirty_page_for_writepage(wbc, page);
407 if (page->mapping->assoc_mapping)
408 return 0; /* Do not request flush for shadow page cache */
410 down_read(&nilfs->ns_writer_sem);
411 writer = nilfs->ns_writer;
413 up_read(&nilfs->ns_writer_sem);
416 sb = writer->s_super;
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);
425 up_read(&nilfs->ns_writer_sem);
430 static struct address_space_operations def_mdt_aops = {
431 .writepage = nilfs_mdt_write_page,
432 .sync_page = block_sync_page,
435 static struct inode_operations def_mdt_iops;
436 static struct file_operations def_mdt_fops;
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
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.
449 nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
450 ino_t ino, gfp_t gfp_mask)
452 struct inode *inode = nilfs_alloc_inode_common(nilfs);
457 struct address_space * const mapping = &inode->i_data;
458 struct nilfs_mdt_info *mi = kzalloc(sizeof(*mi), GFP_NOFS);
461 nilfs_destroy_inode(inode);
464 mi->mi_nilfs = nilfs;
465 init_rwsem(&mi->mi_sem);
467 inode->i_sb = sb; /* sb may be NULL for some meta data files */
468 inode->i_blkbits = nilfs->ns_blocksize_bits;
470 atomic_set(&inode->i_count, 1);
473 inode->i_mode = S_IFREG;
474 inode->i_private = mi;
476 #ifdef INIT_UNUSED_INODE_FIELDS
477 atomic_set(&inode->i_writecount, 0);
481 inode->i_generation = 0;
483 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
485 inode->i_pipe = NULL;
486 inode->i_bdev = NULL;
487 inode->i_cdev = NULL;
489 #ifdef CONFIG_SECURITY
490 inode->i_security = NULL;
492 inode->dirtied_when = 0;
494 INIT_LIST_HEAD(&inode->i_list);
495 INIT_LIST_HEAD(&inode->i_sb_list);
499 spin_lock_init(&inode->i_lock);
500 mutex_init(&inode->i_mutex);
501 init_rwsem(&inode->i_alloc_sem);
503 mapping->host = NULL; /* instead of inode */
505 mapping_set_gfp_mask(mapping, gfp_mask);
506 mapping->assoc_mapping = NULL;
507 mapping->backing_dev_info = nilfs->ns_bdi;
509 inode->i_mapping = mapping;
515 struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb,
518 struct inode *inode = nilfs_mdt_new_common(nilfs, sb, ino,
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);