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_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
188 SLAB_CTOR_CONSTRUCTOR) {
195 set_bit(META_free, &mp->flag);
196 init_waitqueue_head(&mp->wait);
200 static inline struct metapage *alloc_metapage(gfp_t gfp_mask)
202 return mempool_alloc(metapage_mempool, gfp_mask);
205 static inline void free_metapage(struct metapage *mp)
208 set_bit(META_free, &mp->flag);
210 mempool_free(mp, metapage_mempool);
213 int __init metapage_init(void)
216 * Allocate the metapage structures
218 metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage),
219 0, 0, init_once, NULL);
220 if (metapage_cache == NULL)
223 metapage_mempool = mempool_create_slab_pool(METAPOOL_MIN_PAGES,
226 if (metapage_mempool == NULL) {
227 kmem_cache_destroy(metapage_cache);
234 void metapage_exit(void)
236 mempool_destroy(metapage_mempool);
237 kmem_cache_destroy(metapage_cache);
240 static inline void drop_metapage(struct page *page, struct metapage *mp)
242 if (mp->count || mp->nohomeok || test_bit(META_dirty, &mp->flag) ||
243 test_bit(META_io, &mp->flag))
245 remove_metapage(page, mp);
246 INCREMENT(mpStat.pagefree);
251 * Metapage address space operations
254 static sector_t metapage_get_blocks(struct inode *inode, sector_t lblock,
260 sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
263 if (lblock >= file_blocks)
265 if (lblock + *len > file_blocks)
266 *len = file_blocks - lblock;
269 rc = xtLookup(inode, (s64)lblock, *len, &xflag, &xaddr, len, 0);
270 if ((rc == 0) && *len)
271 lblock = (sector_t)xaddr;
274 } /* else no mapping */
279 static void last_read_complete(struct page *page)
281 if (!PageError(page))
282 SetPageUptodate(page);
286 static int metapage_read_end_io(struct bio *bio, unsigned int bytes_done,
289 struct page *page = bio->bi_private;
294 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
295 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
299 dec_io(page, last_read_complete);
305 static void remove_from_logsync(struct metapage *mp)
307 struct jfs_log *log = mp->log;
310 * This can race. Recheck that log hasn't been set to null, and after
311 * acquiring logsync lock, recheck lsn
316 LOGSYNC_LOCK(log, flags);
322 list_del(&mp->synclist);
324 LOGSYNC_UNLOCK(log, flags);
327 static void last_write_complete(struct page *page)
332 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
333 mp = page_to_mp(page, offset);
334 if (mp && test_bit(META_io, &mp->flag)) {
336 remove_from_logsync(mp);
337 clear_bit(META_io, &mp->flag);
340 * I'd like to call drop_metapage here, but I don't think it's
341 * safe unless I have the page locked
344 end_page_writeback(page);
347 static int metapage_write_end_io(struct bio *bio, unsigned int bytes_done,
350 struct page *page = bio->bi_private;
352 BUG_ON(!PagePrivate(page));
357 if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) {
358 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
361 dec_io(page, last_write_complete);
366 static int metapage_writepage(struct page *page, struct writeback_control *wbc)
368 struct bio *bio = NULL;
369 unsigned int block_offset; /* block offset of mp within page */
370 struct inode *inode = page->mapping->host;
371 unsigned int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
378 sector_t next_block = 0;
380 unsigned long bio_bytes = 0;
381 unsigned long bio_offset = 0;
384 page_start = (sector_t)page->index <<
385 (PAGE_CACHE_SHIFT - inode->i_blkbits);
386 BUG_ON(!PageLocked(page));
387 BUG_ON(PageWriteback(page));
389 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
390 mp = page_to_mp(page, offset);
392 if (!mp || !test_bit(META_dirty, &mp->flag))
395 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
398 * Make sure this page isn't blocked indefinitely.
399 * If the journal isn't undergoing I/O, push it
401 if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
402 jfs_flush_journal(mp->log, 0);
406 clear_bit(META_dirty, &mp->flag);
407 block_offset = offset >> inode->i_blkbits;
408 lblock = page_start + block_offset;
410 if (xlen && lblock == next_block) {
411 /* Contiguous, in memory & on disk */
412 len = min(xlen, blocks_per_mp);
414 bio_bytes += len << inode->i_blkbits;
415 set_bit(META_io, &mp->flag);
419 if (bio_add_page(bio, page, bio_bytes, bio_offset) <
423 * Increment counter before submitting i/o to keep
424 * count from hitting zero before we're through
429 submit_bio(WRITE, bio);
432 set_page_writeback(page);
435 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
436 pblock = metapage_get_blocks(inode, lblock, &xlen);
438 /* Need better error handling */
439 printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
440 dec_io(page, last_write_complete);
443 set_bit(META_io, &mp->flag);
444 len = min(xlen, (uint) JFS_SBI(inode->i_sb)->nbperpage);
446 bio = bio_alloc(GFP_NOFS, 1);
447 bio->bi_bdev = inode->i_sb->s_bdev;
448 bio->bi_sector = pblock << (inode->i_blkbits - 9);
449 bio->bi_end_io = metapage_write_end_io;
450 bio->bi_private = page;
452 /* Don't call bio_add_page yet, we may add to this vec */
454 bio_bytes = len << inode->i_blkbits;
457 next_block = lblock + len;
460 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
465 submit_bio(WRITE, bio);
468 redirty_page_for_writepage(wbc, page);
474 /* We should never reach here, since we're only adding one vec */
475 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
478 dump_mem("bio", bio, sizeof(*bio));
482 dec_io(page, last_write_complete);
487 static int metapage_readpage(struct file *fp, struct page *page)
489 struct inode *inode = page->mapping->host;
490 struct bio *bio = NULL;
491 unsigned int block_offset;
492 unsigned int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
493 sector_t page_start; /* address of page in fs blocks */
499 BUG_ON(!PageLocked(page));
500 page_start = (sector_t)page->index <<
501 (PAGE_CACHE_SHIFT - inode->i_blkbits);
504 while (block_offset < blocks_per_page) {
505 xlen = blocks_per_page - block_offset;
506 pblock = metapage_get_blocks(inode, page_start + block_offset,
509 if (!PagePrivate(page))
510 insert_metapage(page, NULL);
513 submit_bio(READ, bio);
515 bio = bio_alloc(GFP_NOFS, 1);
516 bio->bi_bdev = inode->i_sb->s_bdev;
517 bio->bi_sector = pblock << (inode->i_blkbits - 9);
518 bio->bi_end_io = metapage_read_end_io;
519 bio->bi_private = page;
520 len = xlen << inode->i_blkbits;
521 offset = block_offset << inode->i_blkbits;
522 if (bio_add_page(bio, page, len, offset) < len)
524 block_offset += xlen;
529 submit_bio(READ, bio);
536 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
538 dec_io(page, last_read_complete);
542 static int metapage_releasepage(struct page *page, gfp_t gfp_mask)
548 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
549 mp = page_to_mp(page, offset);
554 jfs_info("metapage_releasepage: mp = 0x%p", mp);
555 if (mp->count || mp->nohomeok ||
556 test_bit(META_dirty, &mp->flag)) {
557 jfs_info("count = %ld, nohomeok = %d", mp->count,
563 remove_from_logsync(mp);
564 remove_metapage(page, mp);
565 INCREMENT(mpStat.pagefree);
571 static void metapage_invalidatepage(struct page *page, unsigned long offset)
575 BUG_ON(PageWriteback(page));
577 metapage_releasepage(page, 0);
580 const struct address_space_operations jfs_metapage_aops = {
581 .readpage = metapage_readpage,
582 .writepage = metapage_writepage,
583 .sync_page = block_sync_page,
584 .releasepage = metapage_releasepage,
585 .invalidatepage = metapage_invalidatepage,
586 .set_page_dirty = __set_page_dirty_nobuffers,
589 struct metapage *__get_metapage(struct inode *inode, unsigned long lblock,
590 unsigned int size, int absolute,
595 struct address_space *mapping;
596 struct metapage *mp = NULL;
598 unsigned long page_index;
599 unsigned long page_offset;
601 jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
602 inode->i_ino, lblock, absolute);
604 l2bsize = inode->i_blkbits;
605 l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize;
606 page_index = lblock >> l2BlocksPerPage;
607 page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize;
608 if ((page_offset + size) > PAGE_CACHE_SIZE) {
609 jfs_err("MetaData crosses page boundary!!");
610 jfs_err("lblock = %lx, size = %d", lblock, size);
615 mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping;
618 * If an nfs client tries to read an inode that is larger
619 * than any existing inodes, we may try to read past the
620 * end of the inode map
622 if ((lblock << inode->i_blkbits) >= inode->i_size)
624 mapping = inode->i_mapping;
627 if (new && (PSIZE == PAGE_CACHE_SIZE)) {
628 page = grab_cache_page(mapping, page_index);
630 jfs_err("grab_cache_page failed!");
633 SetPageUptodate(page);
635 page = read_mapping_page(mapping, page_index, NULL);
636 if (IS_ERR(page) || !PageUptodate(page)) {
637 jfs_err("read_mapping_page failed!");
643 mp = page_to_mp(page, page_offset);
645 if (mp->logical_size != size) {
646 jfs_error(inode->i_sb,
647 "__get_metapage: mp->logical_size != size");
648 jfs_err("logical_size = %d, size = %d",
649 mp->logical_size, size);
655 if (test_bit(META_discard, &mp->flag)) {
657 jfs_error(inode->i_sb,
658 "__get_metapage: using a "
659 "discarded metapage");
660 discard_metapage(mp);
663 clear_bit(META_discard, &mp->flag);
666 INCREMENT(mpStat.pagealloc);
667 mp = alloc_metapage(GFP_NOFS);
670 mp->xflag = COMMIT_PAGE;
673 mp->logical_size = size;
674 mp->data = page_address(page) + page_offset;
676 if (unlikely(insert_metapage(page, mp))) {
684 jfs_info("zeroing mp = 0x%p", mp);
685 memset(mp->data, 0, PSIZE);
689 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
697 void grab_metapage(struct metapage * mp)
699 jfs_info("grab_metapage: mp = 0x%p", mp);
700 page_cache_get(mp->page);
704 unlock_page(mp->page);
707 void force_metapage(struct metapage *mp)
709 struct page *page = mp->page;
710 jfs_info("force_metapage: mp = 0x%p", mp);
711 set_bit(META_forcewrite, &mp->flag);
712 clear_bit(META_sync, &mp->flag);
713 page_cache_get(page);
715 set_page_dirty(page);
716 write_one_page(page, 1);
717 clear_bit(META_forcewrite, &mp->flag);
718 page_cache_release(page);
721 void hold_metapage(struct metapage *mp)
726 void put_metapage(struct metapage *mp)
728 if (mp->count || mp->nohomeok) {
729 /* Someone else will release this */
730 unlock_page(mp->page);
733 page_cache_get(mp->page);
736 unlock_page(mp->page);
737 release_metapage(mp);
740 void release_metapage(struct metapage * mp)
742 struct page *page = mp->page;
743 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
751 if (--mp->count || mp->nohomeok) {
753 page_cache_release(page);
757 if (test_bit(META_dirty, &mp->flag)) {
758 set_page_dirty(page);
759 if (test_bit(META_sync, &mp->flag)) {
760 clear_bit(META_sync, &mp->flag);
761 write_one_page(page, 1);
762 lock_page(page); /* write_one_page unlocks the page */
764 } else if (mp->lsn) /* discard_metapage doesn't remove it */
765 remove_from_logsync(mp);
767 /* Try to keep metapages from using up too much memory */
768 drop_metapage(page, mp);
771 page_cache_release(page);
774 void __invalidate_metapages(struct inode *ip, s64 addr, int len)
777 int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
778 int BlocksPerPage = 1 << l2BlocksPerPage;
779 /* All callers are interested in block device's mapping */
780 struct address_space *mapping =
781 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
787 * Mark metapages to discard. They will eventually be
788 * released, but should not be written.
790 for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
791 lblock += BlocksPerPage) {
792 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
795 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
796 mp = page_to_mp(page, offset);
799 if (mp->index < addr)
801 if (mp->index >= addr + len)
804 clear_bit(META_dirty, &mp->flag);
805 set_bit(META_discard, &mp->flag);
807 remove_from_logsync(mp);
810 page_cache_release(page);
814 #ifdef CONFIG_JFS_STATISTICS
815 int jfs_mpstat_read(char *buffer, char **start, off_t offset, int length,
816 int *eof, void *data)
821 len += sprintf(buffer,
822 "JFS Metapage statistics\n"
823 "=======================\n"
824 "page allocations = %d\n"
832 *start = buffer + begin;