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/module.h>
23 #include <linux/bio.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/buffer_head.h>
27 #include <linux/mempool.h>
28 #include <linux/seq_file.h>
29 #include "jfs_incore.h"
30 #include "jfs_superblock.h"
31 #include "jfs_filsys.h"
32 #include "jfs_metapage.h"
33 #include "jfs_txnmgr.h"
34 #include "jfs_debug.h"
36 #ifdef CONFIG_JFS_STATISTICS
38 uint pagealloc; /* # of page allocations */
39 uint pagefree; /* # of page frees */
40 uint lockwait; /* # of sleeping lock_metapage() calls */
44 #define metapage_locked(mp) test_bit(META_locked, &(mp)->flag)
45 #define trylock_metapage(mp) test_and_set_bit_lock(META_locked, &(mp)->flag)
47 static inline void unlock_metapage(struct metapage *mp)
49 clear_bit_unlock(META_locked, &mp->flag);
53 static inline void __lock_metapage(struct metapage *mp)
55 DECLARE_WAITQUEUE(wait, current);
56 INCREMENT(mpStat.lockwait);
57 add_wait_queue_exclusive(&mp->wait, &wait);
59 set_current_state(TASK_UNINTERRUPTIBLE);
60 if (metapage_locked(mp)) {
61 unlock_page(mp->page);
65 } while (trylock_metapage(mp));
66 __set_current_state(TASK_RUNNING);
67 remove_wait_queue(&mp->wait, &wait);
71 * Must have mp->page locked
73 static inline void lock_metapage(struct metapage *mp)
75 if (trylock_metapage(mp))
79 #define METAPOOL_MIN_PAGES 32
80 static struct kmem_cache *metapage_cache;
81 static mempool_t *metapage_mempool;
83 #define MPS_PER_PAGE (PAGE_CACHE_SIZE >> L2PSIZE)
90 struct metapage *mp[MPS_PER_PAGE];
92 #define mp_anchor(page) ((struct meta_anchor *)page_private(page))
94 static inline struct metapage *page_to_mp(struct page *page, int offset)
96 if (!PagePrivate(page))
98 return mp_anchor(page)->mp[offset >> L2PSIZE];
101 static inline int insert_metapage(struct page *page, struct metapage *mp)
103 struct meta_anchor *a;
105 int l2mp_blocks; /* log2 blocks per metapage */
107 if (PagePrivate(page))
110 a = kzalloc(sizeof(struct meta_anchor), GFP_NOFS);
113 set_page_private(page, (unsigned long)a);
114 SetPagePrivate(page);
119 l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
120 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
128 static inline void remove_metapage(struct page *page, struct metapage *mp)
130 struct meta_anchor *a = mp_anchor(page);
131 int l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
134 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
136 BUG_ON(a->mp[index] != mp);
139 if (--a->mp_count == 0) {
141 set_page_private(page, 0);
142 ClearPagePrivate(page);
147 static inline void inc_io(struct page *page)
149 atomic_inc(&mp_anchor(page)->io_count);
152 static inline void dec_io(struct page *page, void (*handler) (struct page *))
154 if (atomic_dec_and_test(&mp_anchor(page)->io_count))
159 static inline struct metapage *page_to_mp(struct page *page, int offset)
161 return PagePrivate(page) ? (struct metapage *)page_private(page) : NULL;
164 static inline int insert_metapage(struct page *page, struct metapage *mp)
167 set_page_private(page, (unsigned long)mp);
168 SetPagePrivate(page);
174 static inline void remove_metapage(struct page *page, struct metapage *mp)
176 set_page_private(page, 0);
177 ClearPagePrivate(page);
181 #define inc_io(page) do {} while(0)
182 #define dec_io(page, handler) handler(page)
186 static inline struct metapage *alloc_metapage(gfp_t gfp_mask)
188 struct metapage *mp = mempool_alloc(metapage_mempool, gfp_mask);
196 init_waitqueue_head(&mp->wait);
201 static inline void free_metapage(struct metapage *mp)
203 mempool_free(mp, metapage_mempool);
206 int __init metapage_init(void)
209 * Allocate the metapage structures
211 metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage),
213 if (metapage_cache == NULL)
216 metapage_mempool = mempool_create_slab_pool(METAPOOL_MIN_PAGES,
219 if (metapage_mempool == NULL) {
220 kmem_cache_destroy(metapage_cache);
227 void metapage_exit(void)
229 mempool_destroy(metapage_mempool);
230 kmem_cache_destroy(metapage_cache);
233 static inline void drop_metapage(struct page *page, struct metapage *mp)
235 if (mp->count || mp->nohomeok || test_bit(META_dirty, &mp->flag) ||
236 test_bit(META_io, &mp->flag))
238 remove_metapage(page, mp);
239 INCREMENT(mpStat.pagefree);
244 * Metapage address space operations
247 static sector_t metapage_get_blocks(struct inode *inode, sector_t lblock,
253 sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
256 if (lblock >= file_blocks)
258 if (lblock + *len > file_blocks)
259 *len = file_blocks - lblock;
262 rc = xtLookup(inode, (s64)lblock, *len, &xflag, &xaddr, len, 0);
263 if ((rc == 0) && *len)
264 lblock = (sector_t)xaddr;
267 } /* else no mapping */
272 static void last_read_complete(struct page *page)
274 if (!PageError(page))
275 SetPageUptodate(page);
279 static void metapage_read_end_io(struct bio *bio)
281 struct page *page = bio->bi_private;
284 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
288 dec_io(page, last_read_complete);
292 static void remove_from_logsync(struct metapage *mp)
294 struct jfs_log *log = mp->log;
297 * This can race. Recheck that log hasn't been set to null, and after
298 * acquiring logsync lock, recheck lsn
303 LOGSYNC_LOCK(log, flags);
309 list_del(&mp->synclist);
311 LOGSYNC_UNLOCK(log, flags);
314 static void last_write_complete(struct page *page)
319 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
320 mp = page_to_mp(page, offset);
321 if (mp && test_bit(META_io, &mp->flag)) {
323 remove_from_logsync(mp);
324 clear_bit(META_io, &mp->flag);
327 * I'd like to call drop_metapage here, but I don't think it's
328 * safe unless I have the page locked
331 end_page_writeback(page);
334 static void metapage_write_end_io(struct bio *bio)
336 struct page *page = bio->bi_private;
338 BUG_ON(!PagePrivate(page));
341 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
344 dec_io(page, last_write_complete);
348 static int metapage_writepage(struct page *page, struct writeback_control *wbc)
350 struct bio *bio = NULL;
351 int block_offset; /* block offset of mp within page */
352 struct inode *inode = page->mapping->host;
353 int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
361 sector_t next_block = 0;
363 unsigned long bio_bytes = 0;
364 unsigned long bio_offset = 0;
368 page_start = (sector_t)page->index <<
369 (PAGE_CACHE_SHIFT - inode->i_blkbits);
370 BUG_ON(!PageLocked(page));
371 BUG_ON(PageWriteback(page));
372 set_page_writeback(page);
374 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
375 mp = page_to_mp(page, offset);
377 if (!mp || !test_bit(META_dirty, &mp->flag))
380 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
383 * Make sure this page isn't blocked indefinitely.
384 * If the journal isn't undergoing I/O, push it
386 if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
387 jfs_flush_journal(mp->log, 0);
391 clear_bit(META_dirty, &mp->flag);
392 set_bit(META_io, &mp->flag);
393 block_offset = offset >> inode->i_blkbits;
394 lblock = page_start + block_offset;
396 if (xlen && lblock == next_block) {
397 /* Contiguous, in memory & on disk */
398 len = min(xlen, blocks_per_mp);
400 bio_bytes += len << inode->i_blkbits;
404 if (bio_add_page(bio, page, bio_bytes, bio_offset) <
408 * Increment counter before submitting i/o to keep
409 * count from hitting zero before we're through
412 if (!bio->bi_iter.bi_size)
414 submit_bio(WRITE, bio);
419 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
420 pblock = metapage_get_blocks(inode, lblock, &xlen);
422 printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
424 * We already called inc_io(), but can't cancel it
425 * with dec_io() until we're done with the page
430 len = min(xlen, (int)JFS_SBI(inode->i_sb)->nbperpage);
432 bio = bio_alloc(GFP_NOFS, 1);
433 bio->bi_bdev = inode->i_sb->s_bdev;
434 bio->bi_iter.bi_sector = pblock << (inode->i_blkbits - 9);
435 bio->bi_end_io = metapage_write_end_io;
436 bio->bi_private = page;
438 /* Don't call bio_add_page yet, we may add to this vec */
440 bio_bytes = len << inode->i_blkbits;
443 next_block = lblock + len;
446 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
448 if (!bio->bi_iter.bi_size)
451 submit_bio(WRITE, bio);
455 redirty_page_for_writepage(wbc, page);
462 if (nr_underway == 0)
463 end_page_writeback(page);
467 /* We should never reach here, since we're only adding one vec */
468 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
471 print_hex_dump(KERN_ERR, "JFS: dump of bio: ", DUMP_PREFIX_ADDRESS, 16,
472 4, bio, sizeof(*bio), 0);
476 dec_io(page, last_write_complete);
479 dec_io(page, last_write_complete);
483 static int metapage_readpage(struct file *fp, struct page *page)
485 struct inode *inode = page->mapping->host;
486 struct bio *bio = NULL;
488 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
489 sector_t page_start; /* address of page in fs blocks */
495 BUG_ON(!PageLocked(page));
496 page_start = (sector_t)page->index <<
497 (PAGE_CACHE_SHIFT - inode->i_blkbits);
500 while (block_offset < blocks_per_page) {
501 xlen = blocks_per_page - block_offset;
502 pblock = metapage_get_blocks(inode, page_start + block_offset,
505 if (!PagePrivate(page))
506 insert_metapage(page, NULL);
509 submit_bio(READ, bio);
511 bio = bio_alloc(GFP_NOFS, 1);
512 bio->bi_bdev = inode->i_sb->s_bdev;
513 bio->bi_iter.bi_sector =
514 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 int offset,
571 BUG_ON(offset || length < PAGE_CACHE_SIZE);
573 BUG_ON(PageWriteback(page));
575 metapage_releasepage(page, 0);
578 const struct address_space_operations jfs_metapage_aops = {
579 .readpage = metapage_readpage,
580 .writepage = metapage_writepage,
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_mp->logical_size != size\n");
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 "using a discarded metapage\n");
656 discard_metapage(mp);
659 clear_bit(META_discard, &mp->flag);
662 INCREMENT(mpStat.pagealloc);
663 mp = alloc_metapage(GFP_NOFS);
666 mp->xflag = COMMIT_PAGE;
669 mp->logical_size = size;
670 mp->data = page_address(page) + page_offset;
672 if (unlikely(insert_metapage(page, mp))) {
680 jfs_info("zeroing mp = 0x%p", mp);
681 memset(mp->data, 0, PSIZE);
685 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
693 void grab_metapage(struct metapage * mp)
695 jfs_info("grab_metapage: mp = 0x%p", mp);
696 page_cache_get(mp->page);
700 unlock_page(mp->page);
703 void force_metapage(struct metapage *mp)
705 struct page *page = mp->page;
706 jfs_info("force_metapage: mp = 0x%p", mp);
707 set_bit(META_forcewrite, &mp->flag);
708 clear_bit(META_sync, &mp->flag);
709 page_cache_get(page);
711 set_page_dirty(page);
712 write_one_page(page, 1);
713 clear_bit(META_forcewrite, &mp->flag);
714 page_cache_release(page);
717 void hold_metapage(struct metapage *mp)
722 void put_metapage(struct metapage *mp)
724 if (mp->count || mp->nohomeok) {
725 /* Someone else will release this */
726 unlock_page(mp->page);
729 page_cache_get(mp->page);
732 unlock_page(mp->page);
733 release_metapage(mp);
736 void release_metapage(struct metapage * mp)
738 struct page *page = mp->page;
739 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
747 if (--mp->count || mp->nohomeok) {
749 page_cache_release(page);
753 if (test_bit(META_dirty, &mp->flag)) {
754 set_page_dirty(page);
755 if (test_bit(META_sync, &mp->flag)) {
756 clear_bit(META_sync, &mp->flag);
757 write_one_page(page, 1);
758 lock_page(page); /* write_one_page unlocks the page */
760 } else if (mp->lsn) /* discard_metapage doesn't remove it */
761 remove_from_logsync(mp);
763 /* Try to keep metapages from using up too much memory */
764 drop_metapage(page, mp);
767 page_cache_release(page);
770 void __invalidate_metapages(struct inode *ip, s64 addr, int len)
773 int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
774 int BlocksPerPage = 1 << l2BlocksPerPage;
775 /* All callers are interested in block device's mapping */
776 struct address_space *mapping =
777 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
783 * Mark metapages to discard. They will eventually be
784 * released, but should not be written.
786 for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
787 lblock += BlocksPerPage) {
788 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
791 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
792 mp = page_to_mp(page, offset);
795 if (mp->index < addr)
797 if (mp->index >= addr + len)
800 clear_bit(META_dirty, &mp->flag);
801 set_bit(META_discard, &mp->flag);
803 remove_from_logsync(mp);
806 page_cache_release(page);
810 #ifdef CONFIG_JFS_STATISTICS
811 static int jfs_mpstat_proc_show(struct seq_file *m, void *v)
814 "JFS Metapage statistics\n"
815 "=======================\n"
816 "page allocations = %d\n"
825 static int jfs_mpstat_proc_open(struct inode *inode, struct file *file)
827 return single_open(file, jfs_mpstat_proc_show, NULL);
830 const struct file_operations jfs_mpstat_proc_fops = {
831 .owner = THIS_MODULE,
832 .open = jfs_mpstat_proc_open,
835 .release = single_release,