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(META_locked, &(mp)->flag)
44 static inline void unlock_metapage(struct metapage *mp)
46 clear_bit(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, uint 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, uint 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(void *foo, struct kmem_cache *cachep, unsigned long flags)
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),
216 0, 0, init_once, NULL);
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 int metapage_read_end_io(struct bio *bio, unsigned int bytes_done,
286 struct page *page = bio->bi_private;
291 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
292 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
296 dec_io(page, last_read_complete);
302 static void remove_from_logsync(struct metapage *mp)
304 struct jfs_log *log = mp->log;
307 * This can race. Recheck that log hasn't been set to null, and after
308 * acquiring logsync lock, recheck lsn
313 LOGSYNC_LOCK(log, flags);
319 list_del(&mp->synclist);
321 LOGSYNC_UNLOCK(log, flags);
324 static void last_write_complete(struct page *page)
329 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
330 mp = page_to_mp(page, offset);
331 if (mp && test_bit(META_io, &mp->flag)) {
333 remove_from_logsync(mp);
334 clear_bit(META_io, &mp->flag);
337 * I'd like to call drop_metapage here, but I don't think it's
338 * safe unless I have the page locked
341 end_page_writeback(page);
344 static int metapage_write_end_io(struct bio *bio, unsigned int bytes_done,
347 struct page *page = bio->bi_private;
349 BUG_ON(!PagePrivate(page));
354 if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) {
355 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
358 dec_io(page, last_write_complete);
363 static int metapage_writepage(struct page *page, struct writeback_control *wbc)
365 struct bio *bio = NULL;
366 unsigned int block_offset; /* block offset of mp within page */
367 struct inode *inode = page->mapping->host;
368 unsigned int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
375 sector_t next_block = 0;
377 unsigned long bio_bytes = 0;
378 unsigned long bio_offset = 0;
381 page_start = (sector_t)page->index <<
382 (PAGE_CACHE_SHIFT - inode->i_blkbits);
383 BUG_ON(!PageLocked(page));
384 BUG_ON(PageWriteback(page));
386 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
387 mp = page_to_mp(page, offset);
389 if (!mp || !test_bit(META_dirty, &mp->flag))
392 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
395 * Make sure this page isn't blocked indefinitely.
396 * If the journal isn't undergoing I/O, push it
398 if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
399 jfs_flush_journal(mp->log, 0);
403 clear_bit(META_dirty, &mp->flag);
404 block_offset = offset >> inode->i_blkbits;
405 lblock = page_start + block_offset;
407 if (xlen && lblock == next_block) {
408 /* Contiguous, in memory & on disk */
409 len = min(xlen, blocks_per_mp);
411 bio_bytes += len << inode->i_blkbits;
412 set_bit(META_io, &mp->flag);
416 if (bio_add_page(bio, page, bio_bytes, bio_offset) <
420 * Increment counter before submitting i/o to keep
421 * count from hitting zero before we're through
426 submit_bio(WRITE, bio);
429 set_page_writeback(page);
432 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
433 pblock = metapage_get_blocks(inode, lblock, &xlen);
435 /* Need better error handling */
436 printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
437 dec_io(page, last_write_complete);
440 set_bit(META_io, &mp->flag);
441 len = min(xlen, (uint) JFS_SBI(inode->i_sb)->nbperpage);
443 bio = bio_alloc(GFP_NOFS, 1);
444 bio->bi_bdev = inode->i_sb->s_bdev;
445 bio->bi_sector = pblock << (inode->i_blkbits - 9);
446 bio->bi_end_io = metapage_write_end_io;
447 bio->bi_private = page;
449 /* Don't call bio_add_page yet, we may add to this vec */
451 bio_bytes = len << inode->i_blkbits;
454 next_block = lblock + len;
457 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
462 submit_bio(WRITE, bio);
465 redirty_page_for_writepage(wbc, page);
471 /* We should never reach here, since we're only adding one vec */
472 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
475 dump_mem("bio", bio, sizeof(*bio));
479 dec_io(page, last_write_complete);
484 static int metapage_readpage(struct file *fp, struct page *page)
486 struct inode *inode = page->mapping->host;
487 struct bio *bio = NULL;
488 unsigned int block_offset;
489 unsigned int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
490 sector_t page_start; /* address of page in fs blocks */
496 BUG_ON(!PageLocked(page));
497 page_start = (sector_t)page->index <<
498 (PAGE_CACHE_SHIFT - inode->i_blkbits);
501 while (block_offset < blocks_per_page) {
502 xlen = blocks_per_page - block_offset;
503 pblock = metapage_get_blocks(inode, page_start + block_offset,
506 if (!PagePrivate(page))
507 insert_metapage(page, NULL);
510 submit_bio(READ, bio);
512 bio = bio_alloc(GFP_NOFS, 1);
513 bio->bi_bdev = inode->i_sb->s_bdev;
514 bio->bi_sector = pblock << (inode->i_blkbits - 9);
515 bio->bi_end_io = metapage_read_end_io;
516 bio->bi_private = page;
517 len = xlen << inode->i_blkbits;
518 offset = block_offset << inode->i_blkbits;
519 if (bio_add_page(bio, page, len, offset) < len)
521 block_offset += xlen;
526 submit_bio(READ, bio);
533 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
535 dec_io(page, last_read_complete);
539 static int metapage_releasepage(struct page *page, gfp_t gfp_mask)
545 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
546 mp = page_to_mp(page, offset);
551 jfs_info("metapage_releasepage: mp = 0x%p", mp);
552 if (mp->count || mp->nohomeok ||
553 test_bit(META_dirty, &mp->flag)) {
554 jfs_info("count = %ld, nohomeok = %d", mp->count,
560 remove_from_logsync(mp);
561 remove_metapage(page, mp);
562 INCREMENT(mpStat.pagefree);
568 static void metapage_invalidatepage(struct page *page, unsigned long offset)
572 BUG_ON(PageWriteback(page));
574 metapage_releasepage(page, 0);
577 const struct address_space_operations jfs_metapage_aops = {
578 .readpage = metapage_readpage,
579 .writepage = metapage_writepage,
580 .sync_page = block_sync_page,
581 .releasepage = metapage_releasepage,
582 .invalidatepage = metapage_invalidatepage,
583 .set_page_dirty = __set_page_dirty_nobuffers,
586 struct metapage *__get_metapage(struct inode *inode, unsigned long lblock,
587 unsigned int size, int absolute,
592 struct address_space *mapping;
593 struct metapage *mp = NULL;
595 unsigned long page_index;
596 unsigned long page_offset;
598 jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
599 inode->i_ino, lblock, absolute);
601 l2bsize = inode->i_blkbits;
602 l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize;
603 page_index = lblock >> l2BlocksPerPage;
604 page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize;
605 if ((page_offset + size) > PAGE_CACHE_SIZE) {
606 jfs_err("MetaData crosses page boundary!!");
607 jfs_err("lblock = %lx, size = %d", lblock, size);
612 mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping;
615 * If an nfs client tries to read an inode that is larger
616 * than any existing inodes, we may try to read past the
617 * end of the inode map
619 if ((lblock << inode->i_blkbits) >= inode->i_size)
621 mapping = inode->i_mapping;
624 if (new && (PSIZE == PAGE_CACHE_SIZE)) {
625 page = grab_cache_page(mapping, page_index);
627 jfs_err("grab_cache_page failed!");
630 SetPageUptodate(page);
632 page = read_mapping_page(mapping, page_index, NULL);
633 if (IS_ERR(page) || !PageUptodate(page)) {
634 jfs_err("read_mapping_page failed!");
640 mp = page_to_mp(page, page_offset);
642 if (mp->logical_size != size) {
643 jfs_error(inode->i_sb,
644 "__get_metapage: mp->logical_size != size");
645 jfs_err("logical_size = %d, size = %d",
646 mp->logical_size, size);
652 if (test_bit(META_discard, &mp->flag)) {
654 jfs_error(inode->i_sb,
655 "__get_metapage: using a "
656 "discarded metapage");
657 discard_metapage(mp);
660 clear_bit(META_discard, &mp->flag);
663 INCREMENT(mpStat.pagealloc);
664 mp = alloc_metapage(GFP_NOFS);
667 mp->xflag = COMMIT_PAGE;
670 mp->logical_size = size;
671 mp->data = page_address(page) + page_offset;
673 if (unlikely(insert_metapage(page, mp))) {
681 jfs_info("zeroing mp = 0x%p", mp);
682 memset(mp->data, 0, PSIZE);
686 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
694 void grab_metapage(struct metapage * mp)
696 jfs_info("grab_metapage: mp = 0x%p", mp);
697 page_cache_get(mp->page);
701 unlock_page(mp->page);
704 void force_metapage(struct metapage *mp)
706 struct page *page = mp->page;
707 jfs_info("force_metapage: mp = 0x%p", mp);
708 set_bit(META_forcewrite, &mp->flag);
709 clear_bit(META_sync, &mp->flag);
710 page_cache_get(page);
712 set_page_dirty(page);
713 write_one_page(page, 1);
714 clear_bit(META_forcewrite, &mp->flag);
715 page_cache_release(page);
718 void hold_metapage(struct metapage *mp)
723 void put_metapage(struct metapage *mp)
725 if (mp->count || mp->nohomeok) {
726 /* Someone else will release this */
727 unlock_page(mp->page);
730 page_cache_get(mp->page);
733 unlock_page(mp->page);
734 release_metapage(mp);
737 void release_metapage(struct metapage * mp)
739 struct page *page = mp->page;
740 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
748 if (--mp->count || mp->nohomeok) {
750 page_cache_release(page);
754 if (test_bit(META_dirty, &mp->flag)) {
755 set_page_dirty(page);
756 if (test_bit(META_sync, &mp->flag)) {
757 clear_bit(META_sync, &mp->flag);
758 write_one_page(page, 1);
759 lock_page(page); /* write_one_page unlocks the page */
761 } else if (mp->lsn) /* discard_metapage doesn't remove it */
762 remove_from_logsync(mp);
764 /* Try to keep metapages from using up too much memory */
765 drop_metapage(page, mp);
768 page_cache_release(page);
771 void __invalidate_metapages(struct inode *ip, s64 addr, int len)
774 int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
775 int BlocksPerPage = 1 << l2BlocksPerPage;
776 /* All callers are interested in block device's mapping */
777 struct address_space *mapping =
778 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
784 * Mark metapages to discard. They will eventually be
785 * released, but should not be written.
787 for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
788 lblock += BlocksPerPage) {
789 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
792 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
793 mp = page_to_mp(page, offset);
796 if (mp->index < addr)
798 if (mp->index >= addr + len)
801 clear_bit(META_dirty, &mp->flag);
802 set_bit(META_discard, &mp->flag);
804 remove_from_logsync(mp);
807 page_cache_release(page);
811 #ifdef CONFIG_JFS_STATISTICS
812 int jfs_mpstat_read(char *buffer, char **start, off_t offset, int length,
813 int *eof, void *data)
818 len += sprintf(buffer,
819 "JFS Metapage statistics\n"
820 "=======================\n"
821 "page allocations = %d\n"
829 *start = buffer + begin;