1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) International Business Machines Corp., 2000-2005
4 * Portions Copyright (C) Christoph Hellwig, 2001-2002
9 #include <linux/module.h>
10 #include <linux/bio.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mempool.h>
15 #include <linux/seq_file.h>
16 #include "jfs_incore.h"
17 #include "jfs_superblock.h"
18 #include "jfs_filsys.h"
19 #include "jfs_metapage.h"
20 #include "jfs_txnmgr.h"
21 #include "jfs_debug.h"
23 #ifdef CONFIG_JFS_STATISTICS
25 uint pagealloc
; /* # of page allocations */
26 uint pagefree
; /* # of page frees */
27 uint lockwait
; /* # of sleeping lock_metapage() calls */
31 #define metapage_locked(mp) test_bit(META_locked, &(mp)->flag)
32 #define trylock_metapage(mp) test_and_set_bit_lock(META_locked, &(mp)->flag)
34 static inline void unlock_metapage(struct metapage
*mp
)
36 clear_bit_unlock(META_locked
, &mp
->flag
);
40 static inline void __lock_metapage(struct metapage
*mp
)
42 DECLARE_WAITQUEUE(wait
, current
);
43 INCREMENT(mpStat
.lockwait
);
44 add_wait_queue_exclusive(&mp
->wait
, &wait
);
46 set_current_state(TASK_UNINTERRUPTIBLE
);
47 if (metapage_locked(mp
)) {
48 unlock_page(mp
->page
);
52 } while (trylock_metapage(mp
));
53 __set_current_state(TASK_RUNNING
);
54 remove_wait_queue(&mp
->wait
, &wait
);
58 * Must have mp->page locked
60 static inline void lock_metapage(struct metapage
*mp
)
62 if (trylock_metapage(mp
))
66 #define METAPOOL_MIN_PAGES 32
67 static struct kmem_cache
*metapage_cache
;
68 static mempool_t
*metapage_mempool
;
70 #define MPS_PER_PAGE (PAGE_SIZE >> L2PSIZE)
77 struct metapage
*mp
[MPS_PER_PAGE
];
79 #define mp_anchor(page) ((struct meta_anchor *)page_private(page))
81 static inline struct metapage
*page_to_mp(struct page
*page
, int offset
)
83 if (!PagePrivate(page
))
85 return mp_anchor(page
)->mp
[offset
>> L2PSIZE
];
88 static inline int insert_metapage(struct page
*page
, struct metapage
*mp
)
90 struct meta_anchor
*a
;
92 int l2mp_blocks
; /* log2 blocks per metapage */
94 if (PagePrivate(page
))
97 a
= kzalloc(sizeof(struct meta_anchor
), GFP_NOFS
);
100 set_page_private(page
, (unsigned long)a
);
101 SetPagePrivate(page
);
106 l2mp_blocks
= L2PSIZE
- page
->mapping
->host
->i_blkbits
;
107 index
= (mp
->index
>> l2mp_blocks
) & (MPS_PER_PAGE
- 1);
115 static inline void remove_metapage(struct page
*page
, struct metapage
*mp
)
117 struct meta_anchor
*a
= mp_anchor(page
);
118 int l2mp_blocks
= L2PSIZE
- page
->mapping
->host
->i_blkbits
;
121 index
= (mp
->index
>> l2mp_blocks
) & (MPS_PER_PAGE
- 1);
123 BUG_ON(a
->mp
[index
] != mp
);
126 if (--a
->mp_count
== 0) {
128 set_page_private(page
, 0);
129 ClearPagePrivate(page
);
134 static inline void inc_io(struct page
*page
)
136 atomic_inc(&mp_anchor(page
)->io_count
);
139 static inline void dec_io(struct page
*page
, void (*handler
) (struct page
*))
141 if (atomic_dec_and_test(&mp_anchor(page
)->io_count
))
146 static inline struct metapage
*page_to_mp(struct page
*page
, int offset
)
148 return PagePrivate(page
) ? (struct metapage
*)page_private(page
) : NULL
;
151 static inline int insert_metapage(struct page
*page
, struct metapage
*mp
)
154 set_page_private(page
, (unsigned long)mp
);
155 SetPagePrivate(page
);
161 static inline void remove_metapage(struct page
*page
, struct metapage
*mp
)
163 set_page_private(page
, 0);
164 ClearPagePrivate(page
);
168 #define inc_io(page) do {} while(0)
169 #define dec_io(page, handler) handler(page)
173 static inline struct metapage
*alloc_metapage(gfp_t gfp_mask
)
175 struct metapage
*mp
= mempool_alloc(metapage_mempool
, gfp_mask
);
183 init_waitqueue_head(&mp
->wait
);
188 static inline void free_metapage(struct metapage
*mp
)
190 mempool_free(mp
, metapage_mempool
);
193 int __init
metapage_init(void)
196 * Allocate the metapage structures
198 metapage_cache
= kmem_cache_create("jfs_mp", sizeof(struct metapage
),
200 if (metapage_cache
== NULL
)
203 metapage_mempool
= mempool_create_slab_pool(METAPOOL_MIN_PAGES
,
206 if (metapage_mempool
== NULL
) {
207 kmem_cache_destroy(metapage_cache
);
214 void metapage_exit(void)
216 mempool_destroy(metapage_mempool
);
217 kmem_cache_destroy(metapage_cache
);
220 static inline void drop_metapage(struct page
*page
, struct metapage
*mp
)
222 if (mp
->count
|| mp
->nohomeok
|| test_bit(META_dirty
, &mp
->flag
) ||
223 test_bit(META_io
, &mp
->flag
))
225 remove_metapage(page
, mp
);
226 INCREMENT(mpStat
.pagefree
);
231 * Metapage address space operations
234 static sector_t
metapage_get_blocks(struct inode
*inode
, sector_t lblock
,
240 sector_t file_blocks
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
243 if (lblock
>= file_blocks
)
245 if (lblock
+ *len
> file_blocks
)
246 *len
= file_blocks
- lblock
;
249 rc
= xtLookup(inode
, (s64
)lblock
, *len
, &xflag
, &xaddr
, len
, 0);
250 if ((rc
== 0) && *len
)
251 lblock
= (sector_t
)xaddr
;
254 } /* else no mapping */
259 static void last_read_complete(struct page
*page
)
261 if (!PageError(page
))
262 SetPageUptodate(page
);
266 static void metapage_read_end_io(struct bio
*bio
)
268 struct page
*page
= bio
->bi_private
;
270 if (bio
->bi_status
) {
271 printk(KERN_ERR
"metapage_read_end_io: I/O error\n");
275 dec_io(page
, last_read_complete
);
279 static void remove_from_logsync(struct metapage
*mp
)
281 struct jfs_log
*log
= mp
->log
;
284 * This can race. Recheck that log hasn't been set to null, and after
285 * acquiring logsync lock, recheck lsn
290 LOGSYNC_LOCK(log
, flags
);
296 list_del(&mp
->synclist
);
298 LOGSYNC_UNLOCK(log
, flags
);
301 static void last_write_complete(struct page
*page
)
306 for (offset
= 0; offset
< PAGE_SIZE
; offset
+= PSIZE
) {
307 mp
= page_to_mp(page
, offset
);
308 if (mp
&& test_bit(META_io
, &mp
->flag
)) {
310 remove_from_logsync(mp
);
311 clear_bit(META_io
, &mp
->flag
);
314 * I'd like to call drop_metapage here, but I don't think it's
315 * safe unless I have the page locked
318 end_page_writeback(page
);
321 static void metapage_write_end_io(struct bio
*bio
)
323 struct page
*page
= bio
->bi_private
;
325 BUG_ON(!PagePrivate(page
));
327 if (bio
->bi_status
) {
328 printk(KERN_ERR
"metapage_write_end_io: I/O error\n");
331 dec_io(page
, last_write_complete
);
335 static int metapage_writepage(struct page
*page
, struct writeback_control
*wbc
)
337 struct bio
*bio
= NULL
;
338 int block_offset
; /* block offset of mp within page */
339 struct inode
*inode
= page
->mapping
->host
;
340 int blocks_per_mp
= JFS_SBI(inode
->i_sb
)->nbperpage
;
348 sector_t next_block
= 0;
350 unsigned long bio_bytes
= 0;
351 unsigned long bio_offset
= 0;
355 page_start
= (sector_t
)page
->index
<<
356 (PAGE_SHIFT
- inode
->i_blkbits
);
357 BUG_ON(!PageLocked(page
));
358 BUG_ON(PageWriteback(page
));
359 set_page_writeback(page
);
361 for (offset
= 0; offset
< PAGE_SIZE
; offset
+= PSIZE
) {
362 mp
= page_to_mp(page
, offset
);
364 if (!mp
|| !test_bit(META_dirty
, &mp
->flag
))
367 if (mp
->nohomeok
&& !test_bit(META_forcewrite
, &mp
->flag
)) {
370 * Make sure this page isn't blocked indefinitely.
371 * If the journal isn't undergoing I/O, push it
373 if (mp
->log
&& !(mp
->log
->cflag
& logGC_PAGEOUT
))
374 jfs_flush_journal(mp
->log
, 0);
378 clear_bit(META_dirty
, &mp
->flag
);
379 set_bit(META_io
, &mp
->flag
);
380 block_offset
= offset
>> inode
->i_blkbits
;
381 lblock
= page_start
+ block_offset
;
383 if (xlen
&& lblock
== next_block
) {
384 /* Contiguous, in memory & on disk */
385 len
= min(xlen
, blocks_per_mp
);
387 bio_bytes
+= len
<< inode
->i_blkbits
;
391 if (bio_add_page(bio
, page
, bio_bytes
, bio_offset
) <
395 * Increment counter before submitting i/o to keep
396 * count from hitting zero before we're through
399 if (!bio
->bi_iter
.bi_size
)
406 xlen
= (PAGE_SIZE
- offset
) >> inode
->i_blkbits
;
407 pblock
= metapage_get_blocks(inode
, lblock
, &xlen
);
409 printk(KERN_ERR
"JFS: metapage_get_blocks failed\n");
411 * We already called inc_io(), but can't cancel it
412 * with dec_io() until we're done with the page
417 len
= min(xlen
, (int)JFS_SBI(inode
->i_sb
)->nbperpage
);
419 bio
= bio_alloc(GFP_NOFS
, 1);
420 bio_set_dev(bio
, inode
->i_sb
->s_bdev
);
421 bio
->bi_iter
.bi_sector
= pblock
<< (inode
->i_blkbits
- 9);
422 bio
->bi_end_io
= metapage_write_end_io
;
423 bio
->bi_private
= page
;
424 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
426 /* Don't call bio_add_page yet, we may add to this vec */
428 bio_bytes
= len
<< inode
->i_blkbits
;
431 next_block
= lblock
+ len
;
434 if (bio_add_page(bio
, page
, bio_bytes
, bio_offset
) < bio_bytes
)
436 if (!bio
->bi_iter
.bi_size
)
443 redirty_page_for_writepage(wbc
, page
);
450 if (nr_underway
== 0)
451 end_page_writeback(page
);
455 /* We should never reach here, since we're only adding one vec */
456 printk(KERN_ERR
"JFS: bio_add_page failed unexpectedly\n");
459 print_hex_dump(KERN_ERR
, "JFS: dump of bio: ", DUMP_PREFIX_ADDRESS
, 16,
460 4, bio
, sizeof(*bio
), 0);
464 dec_io(page
, last_write_complete
);
467 dec_io(page
, last_write_complete
);
471 static int metapage_readpage(struct file
*fp
, struct page
*page
)
473 struct inode
*inode
= page
->mapping
->host
;
474 struct bio
*bio
= NULL
;
476 int blocks_per_page
= PAGE_SIZE
>> inode
->i_blkbits
;
477 sector_t page_start
; /* address of page in fs blocks */
483 BUG_ON(!PageLocked(page
));
484 page_start
= (sector_t
)page
->index
<<
485 (PAGE_SHIFT
- inode
->i_blkbits
);
488 while (block_offset
< blocks_per_page
) {
489 xlen
= blocks_per_page
- block_offset
;
490 pblock
= metapage_get_blocks(inode
, page_start
+ block_offset
,
493 if (!PagePrivate(page
))
494 insert_metapage(page
, NULL
);
499 bio
= bio_alloc(GFP_NOFS
, 1);
500 bio_set_dev(bio
, inode
->i_sb
->s_bdev
);
501 bio
->bi_iter
.bi_sector
=
502 pblock
<< (inode
->i_blkbits
- 9);
503 bio
->bi_end_io
= metapage_read_end_io
;
504 bio
->bi_private
= page
;
505 bio_set_op_attrs(bio
, REQ_OP_READ
, 0);
506 len
= xlen
<< inode
->i_blkbits
;
507 offset
= block_offset
<< inode
->i_blkbits
;
508 if (bio_add_page(bio
, page
, len
, offset
) < len
)
510 block_offset
+= xlen
;
522 printk(KERN_ERR
"JFS: bio_add_page failed unexpectedly\n");
524 dec_io(page
, last_read_complete
);
528 static int metapage_releasepage(struct page
*page
, gfp_t gfp_mask
)
534 for (offset
= 0; offset
< PAGE_SIZE
; offset
+= PSIZE
) {
535 mp
= page_to_mp(page
, offset
);
540 jfs_info("metapage_releasepage: mp = 0x%p", mp
);
541 if (mp
->count
|| mp
->nohomeok
||
542 test_bit(META_dirty
, &mp
->flag
)) {
543 jfs_info("count = %ld, nohomeok = %d", mp
->count
,
549 remove_from_logsync(mp
);
550 remove_metapage(page
, mp
);
551 INCREMENT(mpStat
.pagefree
);
557 static void metapage_invalidatepage(struct page
*page
, unsigned int offset
,
560 BUG_ON(offset
|| length
< PAGE_SIZE
);
562 BUG_ON(PageWriteback(page
));
564 metapage_releasepage(page
, 0);
567 const struct address_space_operations jfs_metapage_aops
= {
568 .readpage
= metapage_readpage
,
569 .writepage
= metapage_writepage
,
570 .releasepage
= metapage_releasepage
,
571 .invalidatepage
= metapage_invalidatepage
,
572 .set_page_dirty
= __set_page_dirty_nobuffers
,
575 struct metapage
*__get_metapage(struct inode
*inode
, unsigned long lblock
,
576 unsigned int size
, int absolute
,
581 struct address_space
*mapping
;
582 struct metapage
*mp
= NULL
;
584 unsigned long page_index
;
585 unsigned long page_offset
;
587 jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
588 inode
->i_ino
, lblock
, absolute
);
590 l2bsize
= inode
->i_blkbits
;
591 l2BlocksPerPage
= PAGE_SHIFT
- l2bsize
;
592 page_index
= lblock
>> l2BlocksPerPage
;
593 page_offset
= (lblock
- (page_index
<< l2BlocksPerPage
)) << l2bsize
;
594 if ((page_offset
+ size
) > PAGE_SIZE
) {
595 jfs_err("MetaData crosses page boundary!!");
596 jfs_err("lblock = %lx, size = %d", lblock
, size
);
601 mapping
= JFS_SBI(inode
->i_sb
)->direct_inode
->i_mapping
;
604 * If an nfs client tries to read an inode that is larger
605 * than any existing inodes, we may try to read past the
606 * end of the inode map
608 if ((lblock
<< inode
->i_blkbits
) >= inode
->i_size
)
610 mapping
= inode
->i_mapping
;
613 if (new && (PSIZE
== PAGE_SIZE
)) {
614 page
= grab_cache_page(mapping
, page_index
);
616 jfs_err("grab_cache_page failed!");
619 SetPageUptodate(page
);
621 page
= read_mapping_page(mapping
, page_index
, NULL
);
622 if (IS_ERR(page
) || !PageUptodate(page
)) {
623 jfs_err("read_mapping_page failed!");
629 mp
= page_to_mp(page
, page_offset
);
631 if (mp
->logical_size
!= size
) {
632 jfs_error(inode
->i_sb
,
633 "get_mp->logical_size != size\n");
634 jfs_err("logical_size = %d, size = %d",
635 mp
->logical_size
, size
);
641 if (test_bit(META_discard
, &mp
->flag
)) {
643 jfs_error(inode
->i_sb
,
644 "using a discarded metapage\n");
645 discard_metapage(mp
);
648 clear_bit(META_discard
, &mp
->flag
);
651 INCREMENT(mpStat
.pagealloc
);
652 mp
= alloc_metapage(GFP_NOFS
);
656 mp
->sb
= inode
->i_sb
;
658 mp
->xflag
= COMMIT_PAGE
;
661 mp
->logical_size
= size
;
662 mp
->data
= page_address(page
) + page_offset
;
664 if (unlikely(insert_metapage(page
, mp
))) {
672 jfs_info("zeroing mp = 0x%p", mp
);
673 memset(mp
->data
, 0, PSIZE
);
677 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp
, mp
->data
);
685 void grab_metapage(struct metapage
* mp
)
687 jfs_info("grab_metapage: mp = 0x%p", mp
);
692 unlock_page(mp
->page
);
695 void force_metapage(struct metapage
*mp
)
697 struct page
*page
= mp
->page
;
698 jfs_info("force_metapage: mp = 0x%p", mp
);
699 set_bit(META_forcewrite
, &mp
->flag
);
700 clear_bit(META_sync
, &mp
->flag
);
703 set_page_dirty(page
);
704 if (write_one_page(page
))
705 jfs_error(mp
->sb
, "write_one_page() failed\n");
706 clear_bit(META_forcewrite
, &mp
->flag
);
710 void hold_metapage(struct metapage
*mp
)
715 void put_metapage(struct metapage
*mp
)
717 if (mp
->count
|| mp
->nohomeok
) {
718 /* Someone else will release this */
719 unlock_page(mp
->page
);
725 unlock_page(mp
->page
);
726 release_metapage(mp
);
729 void release_metapage(struct metapage
* mp
)
731 struct page
*page
= mp
->page
;
732 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp
, mp
->flag
);
740 if (--mp
->count
|| mp
->nohomeok
) {
746 if (test_bit(META_dirty
, &mp
->flag
)) {
747 set_page_dirty(page
);
748 if (test_bit(META_sync
, &mp
->flag
)) {
749 clear_bit(META_sync
, &mp
->flag
);
750 if (write_one_page(page
))
751 jfs_error(mp
->sb
, "write_one_page() failed\n");
752 lock_page(page
); /* write_one_page unlocks the page */
754 } else if (mp
->lsn
) /* discard_metapage doesn't remove it */
755 remove_from_logsync(mp
);
757 /* Try to keep metapages from using up too much memory */
758 drop_metapage(page
, mp
);
764 void __invalidate_metapages(struct inode
*ip
, s64 addr
, int len
)
767 int l2BlocksPerPage
= PAGE_SHIFT
- ip
->i_blkbits
;
768 int BlocksPerPage
= 1 << l2BlocksPerPage
;
769 /* All callers are interested in block device's mapping */
770 struct address_space
*mapping
=
771 JFS_SBI(ip
->i_sb
)->direct_inode
->i_mapping
;
777 * Mark metapages to discard. They will eventually be
778 * released, but should not be written.
780 for (lblock
= addr
& ~(BlocksPerPage
- 1); lblock
< addr
+ len
;
781 lblock
+= BlocksPerPage
) {
782 page
= find_lock_page(mapping
, lblock
>> l2BlocksPerPage
);
785 for (offset
= 0; offset
< PAGE_SIZE
; offset
+= PSIZE
) {
786 mp
= page_to_mp(page
, offset
);
789 if (mp
->index
< addr
)
791 if (mp
->index
>= addr
+ len
)
794 clear_bit(META_dirty
, &mp
->flag
);
795 set_bit(META_discard
, &mp
->flag
);
797 remove_from_logsync(mp
);
804 #ifdef CONFIG_JFS_STATISTICS
805 int jfs_mpstat_proc_show(struct seq_file
*m
, void *v
)
808 "JFS Metapage statistics\n"
809 "=======================\n"
810 "page allocations = %d\n"