2 * Copyright (C) International Business Machines Corp., 2000-2005
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 #include <linux/bio.h>
23 #include <linux/init.h>
24 #include <linux/buffer_head.h>
25 #include <linux/mempool.h>
26 #include "jfs_incore.h"
27 #include "jfs_superblock.h"
28 #include "jfs_filsys.h"
29 #include "jfs_metapage.h"
30 #include "jfs_txnmgr.h"
31 #include "jfs_debug.h"
33 #ifdef CONFIG_JFS_STATISTICS
35 uint pagealloc; /* # of page allocations */
36 uint pagefree; /* # of page frees */
37 uint lockwait; /* # of sleeping lock_metapage() calls */
41 #define metapage_locked(mp) test_bit(META_locked, &(mp)->flag)
42 #define trylock_metapage(mp) test_and_set_bit_lock(META_locked, &(mp)->flag)
44 static inline void unlock_metapage(struct metapage *mp)
46 clear_bit_unlock(META_locked, &mp->flag);
50 static inline void __lock_metapage(struct metapage *mp)
52 DECLARE_WAITQUEUE(wait, current);
53 INCREMENT(mpStat.lockwait);
54 add_wait_queue_exclusive(&mp->wait, &wait);
56 set_current_state(TASK_UNINTERRUPTIBLE);
57 if (metapage_locked(mp)) {
58 unlock_page(mp->page);
62 } while (trylock_metapage(mp));
63 __set_current_state(TASK_RUNNING);
64 remove_wait_queue(&mp->wait, &wait);
68 * Must have mp->page locked
70 static inline void lock_metapage(struct metapage *mp)
72 if (trylock_metapage(mp))
76 #define METAPOOL_MIN_PAGES 32
77 static struct kmem_cache *metapage_cache;
78 static mempool_t *metapage_mempool;
80 #define MPS_PER_PAGE (PAGE_CACHE_SIZE >> L2PSIZE)
87 struct metapage *mp[MPS_PER_PAGE];
89 #define mp_anchor(page) ((struct meta_anchor *)page_private(page))
91 static inline struct metapage *page_to_mp(struct page *page, int offset)
93 if (!PagePrivate(page))
95 return mp_anchor(page)->mp[offset >> L2PSIZE];
98 static inline int insert_metapage(struct page *page, struct metapage *mp)
100 struct meta_anchor *a;
102 int l2mp_blocks; /* log2 blocks per metapage */
104 if (PagePrivate(page))
107 a = kzalloc(sizeof(struct meta_anchor), GFP_NOFS);
110 set_page_private(page, (unsigned long)a);
111 SetPagePrivate(page);
116 l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
117 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
125 static inline void remove_metapage(struct page *page, struct metapage *mp)
127 struct meta_anchor *a = mp_anchor(page);
128 int l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
131 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
133 BUG_ON(a->mp[index] != mp);
136 if (--a->mp_count == 0) {
138 set_page_private(page, 0);
139 ClearPagePrivate(page);
144 static inline void inc_io(struct page *page)
146 atomic_inc(&mp_anchor(page)->io_count);
149 static inline void dec_io(struct page *page, void (*handler) (struct page *))
151 if (atomic_dec_and_test(&mp_anchor(page)->io_count))
156 static inline struct metapage *page_to_mp(struct page *page, int offset)
158 return PagePrivate(page) ? (struct metapage *)page_private(page) : NULL;
161 static inline int insert_metapage(struct page *page, struct metapage *mp)
164 set_page_private(page, (unsigned long)mp);
165 SetPagePrivate(page);
171 static inline void remove_metapage(struct page *page, struct metapage *mp)
173 set_page_private(page, 0);
174 ClearPagePrivate(page);
178 #define inc_io(page) do {} while(0)
179 #define dec_io(page, handler) handler(page)
183 static void init_once(struct kmem_cache *cachep, void *foo)
185 struct metapage *mp = (struct metapage *)foo;
193 set_bit(META_free, &mp->flag);
194 init_waitqueue_head(&mp->wait);
197 static inline struct metapage *alloc_metapage(gfp_t gfp_mask)
199 return mempool_alloc(metapage_mempool, gfp_mask);
202 static inline void free_metapage(struct metapage *mp)
205 set_bit(META_free, &mp->flag);
207 mempool_free(mp, metapage_mempool);
210 int __init metapage_init(void)
213 * Allocate the metapage structures
215 metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage),
217 if (metapage_cache == NULL)
220 metapage_mempool = mempool_create_slab_pool(METAPOOL_MIN_PAGES,
223 if (metapage_mempool == NULL) {
224 kmem_cache_destroy(metapage_cache);
231 void metapage_exit(void)
233 mempool_destroy(metapage_mempool);
234 kmem_cache_destroy(metapage_cache);
237 static inline void drop_metapage(struct page *page, struct metapage *mp)
239 if (mp->count || mp->nohomeok || test_bit(META_dirty, &mp->flag) ||
240 test_bit(META_io, &mp->flag))
242 remove_metapage(page, mp);
243 INCREMENT(mpStat.pagefree);
248 * Metapage address space operations
251 static sector_t metapage_get_blocks(struct inode *inode, sector_t lblock,
257 sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
260 if (lblock >= file_blocks)
262 if (lblock + *len > file_blocks)
263 *len = file_blocks - lblock;
266 rc = xtLookup(inode, (s64)lblock, *len, &xflag, &xaddr, len, 0);
267 if ((rc == 0) && *len)
268 lblock = (sector_t)xaddr;
271 } /* else no mapping */
276 static void last_read_complete(struct page *page)
278 if (!PageError(page))
279 SetPageUptodate(page);
283 static void metapage_read_end_io(struct bio *bio, int err)
285 struct page *page = bio->bi_private;
287 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
288 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
292 dec_io(page, last_read_complete);
296 static void remove_from_logsync(struct metapage *mp)
298 struct jfs_log *log = mp->log;
301 * This can race. Recheck that log hasn't been set to null, and after
302 * acquiring logsync lock, recheck lsn
307 LOGSYNC_LOCK(log, flags);
313 list_del(&mp->synclist);
315 LOGSYNC_UNLOCK(log, flags);
318 static void last_write_complete(struct page *page)
323 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
324 mp = page_to_mp(page, offset);
325 if (mp && test_bit(META_io, &mp->flag)) {
327 remove_from_logsync(mp);
328 clear_bit(META_io, &mp->flag);
331 * I'd like to call drop_metapage here, but I don't think it's
332 * safe unless I have the page locked
335 end_page_writeback(page);
338 static void metapage_write_end_io(struct bio *bio, int err)
340 struct page *page = bio->bi_private;
342 BUG_ON(!PagePrivate(page));
344 if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) {
345 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
348 dec_io(page, last_write_complete);
352 static int metapage_writepage(struct page *page, struct writeback_control *wbc)
354 struct bio *bio = NULL;
355 int block_offset; /* block offset of mp within page */
356 struct inode *inode = page->mapping->host;
357 int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
365 sector_t next_block = 0;
367 unsigned long bio_bytes = 0;
368 unsigned long bio_offset = 0;
371 page_start = (sector_t)page->index <<
372 (PAGE_CACHE_SHIFT - inode->i_blkbits);
373 BUG_ON(!PageLocked(page));
374 BUG_ON(PageWriteback(page));
375 set_page_writeback(page);
377 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
378 mp = page_to_mp(page, offset);
380 if (!mp || !test_bit(META_dirty, &mp->flag))
383 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
386 * Make sure this page isn't blocked indefinitely.
387 * If the journal isn't undergoing I/O, push it
389 if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
390 jfs_flush_journal(mp->log, 0);
394 clear_bit(META_dirty, &mp->flag);
395 block_offset = offset >> inode->i_blkbits;
396 lblock = page_start + block_offset;
398 if (xlen && lblock == next_block) {
399 /* Contiguous, in memory & on disk */
400 len = min(xlen, blocks_per_mp);
402 bio_bytes += len << inode->i_blkbits;
403 set_bit(META_io, &mp->flag);
407 if (bio_add_page(bio, page, bio_bytes, bio_offset) <
411 * Increment counter before submitting i/o to keep
412 * count from hitting zero before we're through
417 submit_bio(WRITE, bio);
422 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
423 pblock = metapage_get_blocks(inode, lblock, &xlen);
425 /* Need better error handling */
426 printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
427 dec_io(page, last_write_complete);
430 set_bit(META_io, &mp->flag);
431 len = min(xlen, (int)JFS_SBI(inode->i_sb)->nbperpage);
433 bio = bio_alloc(GFP_NOFS, 1);
434 bio->bi_bdev = inode->i_sb->s_bdev;
435 bio->bi_sector = pblock << (inode->i_blkbits - 9);
436 bio->bi_end_io = metapage_write_end_io;
437 bio->bi_private = page;
439 /* Don't call bio_add_page yet, we may add to this vec */
441 bio_bytes = len << inode->i_blkbits;
444 next_block = lblock + len;
447 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
452 submit_bio(WRITE, bio);
456 redirty_page_for_writepage(wbc, page);
460 if (nr_underway == 0)
461 end_page_writeback(page);
465 /* We should never reach here, since we're only adding one vec */
466 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
469 print_hex_dump(KERN_ERR, "JFS: dump of bio: ", DUMP_PREFIX_ADDRESS, 16,
470 4, bio, sizeof(*bio), 0);
474 dec_io(page, last_write_complete);
479 static int metapage_readpage(struct file *fp, struct page *page)
481 struct inode *inode = page->mapping->host;
482 struct bio *bio = NULL;
484 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
485 sector_t page_start; /* address of page in fs blocks */
491 BUG_ON(!PageLocked(page));
492 page_start = (sector_t)page->index <<
493 (PAGE_CACHE_SHIFT - inode->i_blkbits);
496 while (block_offset < blocks_per_page) {
497 xlen = blocks_per_page - block_offset;
498 pblock = metapage_get_blocks(inode, page_start + block_offset,
501 if (!PagePrivate(page))
502 insert_metapage(page, NULL);
505 submit_bio(READ, bio);
507 bio = bio_alloc(GFP_NOFS, 1);
508 bio->bi_bdev = inode->i_sb->s_bdev;
509 bio->bi_sector = pblock << (inode->i_blkbits - 9);
510 bio->bi_end_io = metapage_read_end_io;
511 bio->bi_private = page;
512 len = xlen << inode->i_blkbits;
513 offset = block_offset << inode->i_blkbits;
514 if (bio_add_page(bio, page, len, offset) < len)
516 block_offset += xlen;
521 submit_bio(READ, bio);
528 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
530 dec_io(page, last_read_complete);
534 static int metapage_releasepage(struct page *page, gfp_t gfp_mask)
540 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
541 mp = page_to_mp(page, offset);
546 jfs_info("metapage_releasepage: mp = 0x%p", mp);
547 if (mp->count || mp->nohomeok ||
548 test_bit(META_dirty, &mp->flag)) {
549 jfs_info("count = %ld, nohomeok = %d", mp->count,
555 remove_from_logsync(mp);
556 remove_metapage(page, mp);
557 INCREMENT(mpStat.pagefree);
563 static void metapage_invalidatepage(struct page *page, unsigned long offset)
567 BUG_ON(PageWriteback(page));
569 metapage_releasepage(page, 0);
572 const struct address_space_operations jfs_metapage_aops = {
573 .readpage = metapage_readpage,
574 .writepage = metapage_writepage,
575 .sync_page = block_sync_page,
576 .releasepage = metapage_releasepage,
577 .invalidatepage = metapage_invalidatepage,
578 .set_page_dirty = __set_page_dirty_nobuffers,
581 struct metapage *__get_metapage(struct inode *inode, unsigned long lblock,
582 unsigned int size, int absolute,
587 struct address_space *mapping;
588 struct metapage *mp = NULL;
590 unsigned long page_index;
591 unsigned long page_offset;
593 jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
594 inode->i_ino, lblock, absolute);
596 l2bsize = inode->i_blkbits;
597 l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize;
598 page_index = lblock >> l2BlocksPerPage;
599 page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize;
600 if ((page_offset + size) > PAGE_CACHE_SIZE) {
601 jfs_err("MetaData crosses page boundary!!");
602 jfs_err("lblock = %lx, size = %d", lblock, size);
607 mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping;
610 * If an nfs client tries to read an inode that is larger
611 * than any existing inodes, we may try to read past the
612 * end of the inode map
614 if ((lblock << inode->i_blkbits) >= inode->i_size)
616 mapping = inode->i_mapping;
619 if (new && (PSIZE == PAGE_CACHE_SIZE)) {
620 page = grab_cache_page(mapping, page_index);
622 jfs_err("grab_cache_page failed!");
625 SetPageUptodate(page);
627 page = read_mapping_page(mapping, page_index, NULL);
628 if (IS_ERR(page) || !PageUptodate(page)) {
629 jfs_err("read_mapping_page failed!");
635 mp = page_to_mp(page, page_offset);
637 if (mp->logical_size != size) {
638 jfs_error(inode->i_sb,
639 "__get_metapage: mp->logical_size != size");
640 jfs_err("logical_size = %d, size = %d",
641 mp->logical_size, size);
647 if (test_bit(META_discard, &mp->flag)) {
649 jfs_error(inode->i_sb,
650 "__get_metapage: using a "
651 "discarded metapage");
652 discard_metapage(mp);
655 clear_bit(META_discard, &mp->flag);
658 INCREMENT(mpStat.pagealloc);
659 mp = alloc_metapage(GFP_NOFS);
662 mp->xflag = COMMIT_PAGE;
665 mp->logical_size = size;
666 mp->data = page_address(page) + page_offset;
668 if (unlikely(insert_metapage(page, mp))) {
676 jfs_info("zeroing mp = 0x%p", mp);
677 memset(mp->data, 0, PSIZE);
681 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
689 void grab_metapage(struct metapage * mp)
691 jfs_info("grab_metapage: mp = 0x%p", mp);
692 page_cache_get(mp->page);
696 unlock_page(mp->page);
699 void force_metapage(struct metapage *mp)
701 struct page *page = mp->page;
702 jfs_info("force_metapage: mp = 0x%p", mp);
703 set_bit(META_forcewrite, &mp->flag);
704 clear_bit(META_sync, &mp->flag);
705 page_cache_get(page);
707 set_page_dirty(page);
708 write_one_page(page, 1);
709 clear_bit(META_forcewrite, &mp->flag);
710 page_cache_release(page);
713 void hold_metapage(struct metapage *mp)
718 void put_metapage(struct metapage *mp)
720 if (mp->count || mp->nohomeok) {
721 /* Someone else will release this */
722 unlock_page(mp->page);
725 page_cache_get(mp->page);
728 unlock_page(mp->page);
729 release_metapage(mp);
732 void release_metapage(struct metapage * mp)
734 struct page *page = mp->page;
735 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
743 if (--mp->count || mp->nohomeok) {
745 page_cache_release(page);
749 if (test_bit(META_dirty, &mp->flag)) {
750 set_page_dirty(page);
751 if (test_bit(META_sync, &mp->flag)) {
752 clear_bit(META_sync, &mp->flag);
753 write_one_page(page, 1);
754 lock_page(page); /* write_one_page unlocks the page */
756 } else if (mp->lsn) /* discard_metapage doesn't remove it */
757 remove_from_logsync(mp);
759 /* Try to keep metapages from using up too much memory */
760 drop_metapage(page, mp);
763 page_cache_release(page);
766 void __invalidate_metapages(struct inode *ip, s64 addr, int len)
769 int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
770 int BlocksPerPage = 1 << l2BlocksPerPage;
771 /* All callers are interested in block device's mapping */
772 struct address_space *mapping =
773 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
779 * Mark metapages to discard. They will eventually be
780 * released, but should not be written.
782 for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
783 lblock += BlocksPerPage) {
784 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
787 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
788 mp = page_to_mp(page, offset);
791 if (mp->index < addr)
793 if (mp->index >= addr + len)
796 clear_bit(META_dirty, &mp->flag);
797 set_bit(META_discard, &mp->flag);
799 remove_from_logsync(mp);
802 page_cache_release(page);
806 #ifdef CONFIG_JFS_STATISTICS
807 int jfs_mpstat_read(char *buffer, char **start, off_t offset, int length,
808 int *eof, void *data)
813 len += sprintf(buffer,
814 "JFS Metapage statistics\n"
815 "=======================\n"
816 "page allocations = %d\n"
824 *start = buffer + begin;