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;
187 if (flags & SLAB_CTOR_CONSTRUCTOR) {
194 set_bit(META_free, &mp->flag);
195 init_waitqueue_head(&mp->wait);
199 static inline struct metapage *alloc_metapage(gfp_t gfp_mask)
201 return mempool_alloc(metapage_mempool, gfp_mask);
204 static inline void free_metapage(struct metapage *mp)
207 set_bit(META_free, &mp->flag);
209 mempool_free(mp, metapage_mempool);
212 int __init metapage_init(void)
215 * Allocate the metapage structures
217 metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage),
218 0, 0, init_once, NULL);
219 if (metapage_cache == NULL)
222 metapage_mempool = mempool_create_slab_pool(METAPOOL_MIN_PAGES,
225 if (metapage_mempool == NULL) {
226 kmem_cache_destroy(metapage_cache);
233 void metapage_exit(void)
235 mempool_destroy(metapage_mempool);
236 kmem_cache_destroy(metapage_cache);
239 static inline void drop_metapage(struct page *page, struct metapage *mp)
241 if (mp->count || mp->nohomeok || test_bit(META_dirty, &mp->flag) ||
242 test_bit(META_io, &mp->flag))
244 remove_metapage(page, mp);
245 INCREMENT(mpStat.pagefree);
250 * Metapage address space operations
253 static sector_t metapage_get_blocks(struct inode *inode, sector_t lblock,
259 sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
262 if (lblock >= file_blocks)
264 if (lblock + *len > file_blocks)
265 *len = file_blocks - lblock;
268 rc = xtLookup(inode, (s64)lblock, *len, &xflag, &xaddr, len, 0);
269 if ((rc == 0) && *len)
270 lblock = (sector_t)xaddr;
273 } /* else no mapping */
278 static void last_read_complete(struct page *page)
280 if (!PageError(page))
281 SetPageUptodate(page);
285 static int metapage_read_end_io(struct bio *bio, unsigned int bytes_done,
288 struct page *page = bio->bi_private;
293 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
294 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
298 dec_io(page, last_read_complete);
304 static void remove_from_logsync(struct metapage *mp)
306 struct jfs_log *log = mp->log;
309 * This can race. Recheck that log hasn't been set to null, and after
310 * acquiring logsync lock, recheck lsn
315 LOGSYNC_LOCK(log, flags);
321 list_del(&mp->synclist);
323 LOGSYNC_UNLOCK(log, flags);
326 static void last_write_complete(struct page *page)
331 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
332 mp = page_to_mp(page, offset);
333 if (mp && test_bit(META_io, &mp->flag)) {
335 remove_from_logsync(mp);
336 clear_bit(META_io, &mp->flag);
339 * I'd like to call drop_metapage here, but I don't think it's
340 * safe unless I have the page locked
343 end_page_writeback(page);
346 static int metapage_write_end_io(struct bio *bio, unsigned int bytes_done,
349 struct page *page = bio->bi_private;
351 BUG_ON(!PagePrivate(page));
356 if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) {
357 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
360 dec_io(page, last_write_complete);
365 static int metapage_writepage(struct page *page, struct writeback_control *wbc)
367 struct bio *bio = NULL;
368 unsigned int block_offset; /* block offset of mp within page */
369 struct inode *inode = page->mapping->host;
370 unsigned int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
377 sector_t next_block = 0;
379 unsigned long bio_bytes = 0;
380 unsigned long bio_offset = 0;
383 page_start = (sector_t)page->index <<
384 (PAGE_CACHE_SHIFT - inode->i_blkbits);
385 BUG_ON(!PageLocked(page));
386 BUG_ON(PageWriteback(page));
388 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
389 mp = page_to_mp(page, offset);
391 if (!mp || !test_bit(META_dirty, &mp->flag))
394 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
397 * Make sure this page isn't blocked indefinitely.
398 * If the journal isn't undergoing I/O, push it
400 if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
401 jfs_flush_journal(mp->log, 0);
405 clear_bit(META_dirty, &mp->flag);
406 block_offset = offset >> inode->i_blkbits;
407 lblock = page_start + block_offset;
409 if (xlen && lblock == next_block) {
410 /* Contiguous, in memory & on disk */
411 len = min(xlen, blocks_per_mp);
413 bio_bytes += len << inode->i_blkbits;
414 set_bit(META_io, &mp->flag);
418 if (bio_add_page(bio, page, bio_bytes, bio_offset) <
422 * Increment counter before submitting i/o to keep
423 * count from hitting zero before we're through
428 submit_bio(WRITE, bio);
431 set_page_writeback(page);
434 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
435 pblock = metapage_get_blocks(inode, lblock, &xlen);
437 /* Need better error handling */
438 printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
439 dec_io(page, last_write_complete);
442 set_bit(META_io, &mp->flag);
443 len = min(xlen, (uint) JFS_SBI(inode->i_sb)->nbperpage);
445 bio = bio_alloc(GFP_NOFS, 1);
446 bio->bi_bdev = inode->i_sb->s_bdev;
447 bio->bi_sector = pblock << (inode->i_blkbits - 9);
448 bio->bi_end_io = metapage_write_end_io;
449 bio->bi_private = page;
451 /* Don't call bio_add_page yet, we may add to this vec */
453 bio_bytes = len << inode->i_blkbits;
456 next_block = lblock + len;
459 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
464 submit_bio(WRITE, bio);
467 redirty_page_for_writepage(wbc, page);
473 /* We should never reach here, since we're only adding one vec */
474 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
477 dump_mem("bio", bio, sizeof(*bio));
481 dec_io(page, last_write_complete);
486 static int metapage_readpage(struct file *fp, struct page *page)
488 struct inode *inode = page->mapping->host;
489 struct bio *bio = NULL;
490 unsigned int block_offset;
491 unsigned int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
492 sector_t page_start; /* address of page in fs blocks */
498 BUG_ON(!PageLocked(page));
499 page_start = (sector_t)page->index <<
500 (PAGE_CACHE_SHIFT - inode->i_blkbits);
503 while (block_offset < blocks_per_page) {
504 xlen = blocks_per_page - block_offset;
505 pblock = metapage_get_blocks(inode, page_start + block_offset,
508 if (!PagePrivate(page))
509 insert_metapage(page, NULL);
512 submit_bio(READ, bio);
514 bio = bio_alloc(GFP_NOFS, 1);
515 bio->bi_bdev = inode->i_sb->s_bdev;
516 bio->bi_sector = pblock << (inode->i_blkbits - 9);
517 bio->bi_end_io = metapage_read_end_io;
518 bio->bi_private = page;
519 len = xlen << inode->i_blkbits;
520 offset = block_offset << inode->i_blkbits;
521 if (bio_add_page(bio, page, len, offset) < len)
523 block_offset += xlen;
528 submit_bio(READ, bio);
535 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
537 dec_io(page, last_read_complete);
541 static int metapage_releasepage(struct page *page, gfp_t gfp_mask)
547 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
548 mp = page_to_mp(page, offset);
553 jfs_info("metapage_releasepage: mp = 0x%p", mp);
554 if (mp->count || mp->nohomeok ||
555 test_bit(META_dirty, &mp->flag)) {
556 jfs_info("count = %ld, nohomeok = %d", mp->count,
562 remove_from_logsync(mp);
563 remove_metapage(page, mp);
564 INCREMENT(mpStat.pagefree);
570 static void metapage_invalidatepage(struct page *page, unsigned long offset)
574 BUG_ON(PageWriteback(page));
576 metapage_releasepage(page, 0);
579 const struct address_space_operations jfs_metapage_aops = {
580 .readpage = metapage_readpage,
581 .writepage = metapage_writepage,
582 .sync_page = block_sync_page,
583 .releasepage = metapage_releasepage,
584 .invalidatepage = metapage_invalidatepage,
585 .set_page_dirty = __set_page_dirty_nobuffers,
588 struct metapage *__get_metapage(struct inode *inode, unsigned long lblock,
589 unsigned int size, int absolute,
594 struct address_space *mapping;
595 struct metapage *mp = NULL;
597 unsigned long page_index;
598 unsigned long page_offset;
600 jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
601 inode->i_ino, lblock, absolute);
603 l2bsize = inode->i_blkbits;
604 l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize;
605 page_index = lblock >> l2BlocksPerPage;
606 page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize;
607 if ((page_offset + size) > PAGE_CACHE_SIZE) {
608 jfs_err("MetaData crosses page boundary!!");
609 jfs_err("lblock = %lx, size = %d", lblock, size);
614 mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping;
617 * If an nfs client tries to read an inode that is larger
618 * than any existing inodes, we may try to read past the
619 * end of the inode map
621 if ((lblock << inode->i_blkbits) >= inode->i_size)
623 mapping = inode->i_mapping;
626 if (new && (PSIZE == PAGE_CACHE_SIZE)) {
627 page = grab_cache_page(mapping, page_index);
629 jfs_err("grab_cache_page failed!");
632 SetPageUptodate(page);
634 page = read_mapping_page(mapping, page_index, NULL);
635 if (IS_ERR(page) || !PageUptodate(page)) {
636 jfs_err("read_mapping_page failed!");
642 mp = page_to_mp(page, page_offset);
644 if (mp->logical_size != size) {
645 jfs_error(inode->i_sb,
646 "__get_metapage: mp->logical_size != size");
647 jfs_err("logical_size = %d, size = %d",
648 mp->logical_size, size);
654 if (test_bit(META_discard, &mp->flag)) {
656 jfs_error(inode->i_sb,
657 "__get_metapage: using a "
658 "discarded metapage");
659 discard_metapage(mp);
662 clear_bit(META_discard, &mp->flag);
665 INCREMENT(mpStat.pagealloc);
666 mp = alloc_metapage(GFP_NOFS);
669 mp->xflag = COMMIT_PAGE;
672 mp->logical_size = size;
673 mp->data = page_address(page) + page_offset;
675 if (unlikely(insert_metapage(page, mp))) {
683 jfs_info("zeroing mp = 0x%p", mp);
684 memset(mp->data, 0, PSIZE);
688 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
696 void grab_metapage(struct metapage * mp)
698 jfs_info("grab_metapage: mp = 0x%p", mp);
699 page_cache_get(mp->page);
703 unlock_page(mp->page);
706 void force_metapage(struct metapage *mp)
708 struct page *page = mp->page;
709 jfs_info("force_metapage: mp = 0x%p", mp);
710 set_bit(META_forcewrite, &mp->flag);
711 clear_bit(META_sync, &mp->flag);
712 page_cache_get(page);
714 set_page_dirty(page);
715 write_one_page(page, 1);
716 clear_bit(META_forcewrite, &mp->flag);
717 page_cache_release(page);
720 void hold_metapage(struct metapage *mp)
725 void put_metapage(struct metapage *mp)
727 if (mp->count || mp->nohomeok) {
728 /* Someone else will release this */
729 unlock_page(mp->page);
732 page_cache_get(mp->page);
735 unlock_page(mp->page);
736 release_metapage(mp);
739 void release_metapage(struct metapage * mp)
741 struct page *page = mp->page;
742 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
750 if (--mp->count || mp->nohomeok) {
752 page_cache_release(page);
756 if (test_bit(META_dirty, &mp->flag)) {
757 set_page_dirty(page);
758 if (test_bit(META_sync, &mp->flag)) {
759 clear_bit(META_sync, &mp->flag);
760 write_one_page(page, 1);
761 lock_page(page); /* write_one_page unlocks the page */
763 } else if (mp->lsn) /* discard_metapage doesn't remove it */
764 remove_from_logsync(mp);
766 /* Try to keep metapages from using up too much memory */
767 drop_metapage(page, mp);
770 page_cache_release(page);
773 void __invalidate_metapages(struct inode *ip, s64 addr, int len)
776 int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
777 int BlocksPerPage = 1 << l2BlocksPerPage;
778 /* All callers are interested in block device's mapping */
779 struct address_space *mapping =
780 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
786 * Mark metapages to discard. They will eventually be
787 * released, but should not be written.
789 for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
790 lblock += BlocksPerPage) {
791 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
794 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
795 mp = page_to_mp(page, offset);
798 if (mp->index < addr)
800 if (mp->index >= addr + len)
803 clear_bit(META_dirty, &mp->flag);
804 set_bit(META_discard, &mp->flag);
806 remove_from_logsync(mp);
809 page_cache_release(page);
813 #ifdef CONFIG_JFS_STATISTICS
814 int jfs_mpstat_read(char *buffer, char **start, off_t offset, int length,
815 int *eof, void *data)
820 len += sprintf(buffer,
821 "JFS Metapage statistics\n"
822 "=======================\n"
823 "page allocations = %d\n"
831 *start = buffer + begin;