1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20#include <linux/fs.h>
21#include <linux/mm.h>
22#include <linux/module.h>
23#include <linux/bio.h>
24#include <linux/init.h>
25#include <linux/buffer_head.h>
26#include <linux/mempool.h>
27#include <linux/seq_file.h>
28#include "jfs_incore.h"
29#include "jfs_superblock.h"
30#include "jfs_filsys.h"
31#include "jfs_metapage.h"
32#include "jfs_txnmgr.h"
33#include "jfs_debug.h"
34
35#ifdef CONFIG_JFS_STATISTICS
36static struct {
37 uint pagealloc;
38 uint pagefree;
39 uint lockwait;
40} mpStat;
41#endif
42
43#define metapage_locked(mp) test_bit(META_locked, &(mp)->flag)
44#define trylock_metapage(mp) test_and_set_bit_lock(META_locked, &(mp)->flag)
45
46static inline void unlock_metapage(struct metapage *mp)
47{
48 clear_bit_unlock(META_locked, &mp->flag);
49 wake_up(&mp->wait);
50}
51
52static inline void __lock_metapage(struct metapage *mp)
53{
54 DECLARE_WAITQUEUE(wait, current);
55 INCREMENT(mpStat.lockwait);
56 add_wait_queue_exclusive(&mp->wait, &wait);
57 do {
58 set_current_state(TASK_UNINTERRUPTIBLE);
59 if (metapage_locked(mp)) {
60 unlock_page(mp->page);
61 io_schedule();
62 lock_page(mp->page);
63 }
64 } while (trylock_metapage(mp));
65 __set_current_state(TASK_RUNNING);
66 remove_wait_queue(&mp->wait, &wait);
67}
68
69
70
71
72static inline void lock_metapage(struct metapage *mp)
73{
74 if (trylock_metapage(mp))
75 __lock_metapage(mp);
76}
77
78#define METAPOOL_MIN_PAGES 32
79static struct kmem_cache *metapage_cache;
80static mempool_t *metapage_mempool;
81
82#define MPS_PER_PAGE (PAGE_CACHE_SIZE >> L2PSIZE)
83
84#if MPS_PER_PAGE > 1
85
86struct meta_anchor {
87 int mp_count;
88 atomic_t io_count;
89 struct metapage *mp[MPS_PER_PAGE];
90};
91#define mp_anchor(page) ((struct meta_anchor *)page_private(page))
92
93static inline struct metapage *page_to_mp(struct page *page, int offset)
94{
95 if (!PagePrivate(page))
96 return NULL;
97 return mp_anchor(page)->mp[offset >> L2PSIZE];
98}
99
100static inline int insert_metapage(struct page *page, struct metapage *mp)
101{
102 struct meta_anchor *a;
103 int index;
104 int l2mp_blocks;
105
106 if (PagePrivate(page))
107 a = mp_anchor(page);
108 else {
109 a = kzalloc(sizeof(struct meta_anchor), GFP_NOFS);
110 if (!a)
111 return -ENOMEM;
112 set_page_private(page, (unsigned long)a);
113 SetPagePrivate(page);
114 kmap(page);
115 }
116
117 if (mp) {
118 l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
119 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
120 a->mp_count++;
121 a->mp[index] = mp;
122 }
123
124 return 0;
125}
126
127static inline void remove_metapage(struct page *page, struct metapage *mp)
128{
129 struct meta_anchor *a = mp_anchor(page);
130 int l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
131 int index;
132
133 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
134
135 BUG_ON(a->mp[index] != mp);
136
137 a->mp[index] = NULL;
138 if (--a->mp_count == 0) {
139 kfree(a);
140 set_page_private(page, 0);
141 ClearPagePrivate(page);
142 kunmap(page);
143 }
144}
145
146static inline void inc_io(struct page *page)
147{
148 atomic_inc(&mp_anchor(page)->io_count);
149}
150
151static inline void dec_io(struct page *page, void (*handler) (struct page *))
152{
153 if (atomic_dec_and_test(&mp_anchor(page)->io_count))
154 handler(page);
155}
156
157#else
158static inline struct metapage *page_to_mp(struct page *page, int offset)
159{
160 return PagePrivate(page) ? (struct metapage *)page_private(page) : NULL;
161}
162
163static inline int insert_metapage(struct page *page, struct metapage *mp)
164{
165 if (mp) {
166 set_page_private(page, (unsigned long)mp);
167 SetPagePrivate(page);
168 kmap(page);
169 }
170 return 0;
171}
172
173static inline void remove_metapage(struct page *page, struct metapage *mp)
174{
175 set_page_private(page, 0);
176 ClearPagePrivate(page);
177 kunmap(page);
178}
179
180#define inc_io(page) do {} while(0)
181#define dec_io(page, handler) handler(page)
182
183#endif
184
185static void init_once(void *foo)
186{
187 struct metapage *mp = (struct metapage *)foo;
188
189 mp->lid = 0;
190 mp->lsn = 0;
191 mp->flag = 0;
192 mp->data = NULL;
193 mp->clsn = 0;
194 mp->log = NULL;
195 set_bit(META_free, &mp->flag);
196 init_waitqueue_head(&mp->wait);
197}
198
199static inline struct metapage *alloc_metapage(gfp_t gfp_mask)
200{
201 return mempool_alloc(metapage_mempool, gfp_mask);
202}
203
204static inline void free_metapage(struct metapage *mp)
205{
206 mp->flag = 0;
207 set_bit(META_free, &mp->flag);
208
209 mempool_free(mp, metapage_mempool);
210}
211
212int __init metapage_init(void)
213{
214
215
216
217 metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage),
218 0, 0, init_once);
219 if (metapage_cache == NULL)
220 return -ENOMEM;
221
222 metapage_mempool = mempool_create_slab_pool(METAPOOL_MIN_PAGES,
223 metapage_cache);
224
225 if (metapage_mempool == NULL) {
226 kmem_cache_destroy(metapage_cache);
227 return -ENOMEM;
228 }
229
230 return 0;
231}
232
233void metapage_exit(void)
234{
235 mempool_destroy(metapage_mempool);
236 kmem_cache_destroy(metapage_cache);
237}
238
239static inline void drop_metapage(struct page *page, struct metapage *mp)
240{
241 if (mp->count || mp->nohomeok || test_bit(META_dirty, &mp->flag) ||
242 test_bit(META_io, &mp->flag))
243 return;
244 remove_metapage(page, mp);
245 INCREMENT(mpStat.pagefree);
246 free_metapage(mp);
247}
248
249
250
251
252
253static sector_t metapage_get_blocks(struct inode *inode, sector_t lblock,
254 int *len)
255{
256 int rc = 0;
257 int xflag;
258 s64 xaddr;
259 sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
260 inode->i_blkbits;
261
262 if (lblock >= file_blocks)
263 return 0;
264 if (lblock + *len > file_blocks)
265 *len = file_blocks - lblock;
266
267 if (inode->i_ino) {
268 rc = xtLookup(inode, (s64)lblock, *len, &xflag, &xaddr, len, 0);
269 if ((rc == 0) && *len)
270 lblock = (sector_t)xaddr;
271 else
272 lblock = 0;
273 }
274
275 return lblock;
276}
277
278static void last_read_complete(struct page *page)
279{
280 if (!PageError(page))
281 SetPageUptodate(page);
282 unlock_page(page);
283}
284
285static void metapage_read_end_io(struct bio *bio, int err)
286{
287 struct page *page = bio->bi_private;
288
289 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
290 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
291 SetPageError(page);
292 }
293
294 dec_io(page, last_read_complete);
295 bio_put(bio);
296}
297
298static void remove_from_logsync(struct metapage *mp)
299{
300 struct jfs_log *log = mp->log;
301 unsigned long flags;
302
303
304
305
306 if (!log)
307 return;
308
309 LOGSYNC_LOCK(log, flags);
310 if (mp->lsn) {
311 mp->log = NULL;
312 mp->lsn = 0;
313 mp->clsn = 0;
314 log->count--;
315 list_del(&mp->synclist);
316 }
317 LOGSYNC_UNLOCK(log, flags);
318}
319
320static void last_write_complete(struct page *page)
321{
322 struct metapage *mp;
323 unsigned int offset;
324
325 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
326 mp = page_to_mp(page, offset);
327 if (mp && test_bit(META_io, &mp->flag)) {
328 if (mp->lsn)
329 remove_from_logsync(mp);
330 clear_bit(META_io, &mp->flag);
331 }
332
333
334
335
336 }
337 end_page_writeback(page);
338}
339
340static void metapage_write_end_io(struct bio *bio, int err)
341{
342 struct page *page = bio->bi_private;
343
344 BUG_ON(!PagePrivate(page));
345
346 if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) {
347 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
348 SetPageError(page);
349 }
350 dec_io(page, last_write_complete);
351 bio_put(bio);
352}
353
354static int metapage_writepage(struct page *page, struct writeback_control *wbc)
355{
356 struct bio *bio = NULL;
357 int block_offset;
358 struct inode *inode = page->mapping->host;
359 int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
360 int len;
361 int xlen;
362 struct metapage *mp;
363 int redirty = 0;
364 sector_t lblock;
365 int nr_underway = 0;
366 sector_t pblock;
367 sector_t next_block = 0;
368 sector_t page_start;
369 unsigned long bio_bytes = 0;
370 unsigned long bio_offset = 0;
371 int offset;
372 int bad_blocks = 0;
373
374 page_start = (sector_t)page->index <<
375 (PAGE_CACHE_SHIFT - inode->i_blkbits);
376 BUG_ON(!PageLocked(page));
377 BUG_ON(PageWriteback(page));
378 set_page_writeback(page);
379
380 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
381 mp = page_to_mp(page, offset);
382
383 if (!mp || !test_bit(META_dirty, &mp->flag))
384 continue;
385
386 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
387 redirty = 1;
388
389
390
391
392 if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
393 jfs_flush_journal(mp->log, 0);
394 continue;
395 }
396
397 clear_bit(META_dirty, &mp->flag);
398 set_bit(META_io, &mp->flag);
399 block_offset = offset >> inode->i_blkbits;
400 lblock = page_start + block_offset;
401 if (bio) {
402 if (xlen && lblock == next_block) {
403
404 len = min(xlen, blocks_per_mp);
405 xlen -= len;
406 bio_bytes += len << inode->i_blkbits;
407 continue;
408 }
409
410 if (bio_add_page(bio, page, bio_bytes, bio_offset) <
411 bio_bytes)
412 goto add_failed;
413
414
415
416
417 inc_io(page);
418 if (!bio->bi_size)
419 goto dump_bio;
420 submit_bio(WRITE, bio);
421 nr_underway++;
422 bio = NULL;
423 } else
424 inc_io(page);
425 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
426 pblock = metapage_get_blocks(inode, lblock, &xlen);
427 if (!pblock) {
428 printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
429
430
431
432
433 bad_blocks++;
434 continue;
435 }
436 len = min(xlen, (int)JFS_SBI(inode->i_sb)->nbperpage);
437
438 bio = bio_alloc(GFP_NOFS, 1);
439 bio->bi_bdev = inode->i_sb->s_bdev;
440 bio->bi_sector = pblock << (inode->i_blkbits - 9);
441 bio->bi_end_io = metapage_write_end_io;
442 bio->bi_private = page;
443
444
445 bio_offset = offset;
446 bio_bytes = len << inode->i_blkbits;
447
448 xlen -= len;
449 next_block = lblock + len;
450 }
451 if (bio) {
452 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
453 goto add_failed;
454 if (!bio->bi_size)
455 goto dump_bio;
456
457 submit_bio(WRITE, bio);
458 nr_underway++;
459 }
460 if (redirty)
461 redirty_page_for_writepage(wbc, page);
462
463 unlock_page(page);
464
465 if (bad_blocks)
466 goto err_out;
467
468 if (nr_underway == 0)
469 end_page_writeback(page);
470
471 return 0;
472add_failed:
473
474 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
475 goto skip;
476dump_bio:
477 print_hex_dump(KERN_ERR, "JFS: dump of bio: ", DUMP_PREFIX_ADDRESS, 16,
478 4, bio, sizeof(*bio), 0);
479skip:
480 bio_put(bio);
481 unlock_page(page);
482 dec_io(page, last_write_complete);
483err_out:
484 while (bad_blocks--)
485 dec_io(page, last_write_complete);
486 return -EIO;
487}
488
489static int metapage_readpage(struct file *fp, struct page *page)
490{
491 struct inode *inode = page->mapping->host;
492 struct bio *bio = NULL;
493 int block_offset;
494 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
495 sector_t page_start;
496 sector_t pblock;
497 int xlen;
498 unsigned int len;
499 int offset;
500
501 BUG_ON(!PageLocked(page));
502 page_start = (sector_t)page->index <<
503 (PAGE_CACHE_SHIFT - inode->i_blkbits);
504
505 block_offset = 0;
506 while (block_offset < blocks_per_page) {
507 xlen = blocks_per_page - block_offset;
508 pblock = metapage_get_blocks(inode, page_start + block_offset,
509 &xlen);
510 if (pblock) {
511 if (!PagePrivate(page))
512 insert_metapage(page, NULL);
513 inc_io(page);
514 if (bio)
515 submit_bio(READ, bio);
516
517 bio = bio_alloc(GFP_NOFS, 1);
518 bio->bi_bdev = inode->i_sb->s_bdev;
519 bio->bi_sector = pblock << (inode->i_blkbits - 9);
520 bio->bi_end_io = metapage_read_end_io;
521 bio->bi_private = page;
522 len = xlen << inode->i_blkbits;
523 offset = block_offset << inode->i_blkbits;
524 if (bio_add_page(bio, page, len, offset) < len)
525 goto add_failed;
526 block_offset += xlen;
527 } else
528 block_offset++;
529 }
530 if (bio)
531 submit_bio(READ, bio);
532 else
533 unlock_page(page);
534
535 return 0;
536
537add_failed:
538 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
539 bio_put(bio);
540 dec_io(page, last_read_complete);
541 return -EIO;
542}
543
544static int metapage_releasepage(struct page *page, gfp_t gfp_mask)
545{
546 struct metapage *mp;
547 int ret = 1;
548 int offset;
549
550 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
551 mp = page_to_mp(page, offset);
552
553 if (!mp)
554 continue;
555
556 jfs_info("metapage_releasepage: mp = 0x%p", mp);
557 if (mp->count || mp->nohomeok ||
558 test_bit(META_dirty, &mp->flag)) {
559 jfs_info("count = %ld, nohomeok = %d", mp->count,
560 mp->nohomeok);
561 ret = 0;
562 continue;
563 }
564 if (mp->lsn)
565 remove_from_logsync(mp);
566 remove_metapage(page, mp);
567 INCREMENT(mpStat.pagefree);
568 free_metapage(mp);
569 }
570 return ret;
571}
572
573static void metapage_invalidatepage(struct page *page, unsigned long offset)
574{
575 BUG_ON(offset);
576
577 BUG_ON(PageWriteback(page));
578
579 metapage_releasepage(page, 0);
580}
581
582const struct address_space_operations jfs_metapage_aops = {
583 .readpage = metapage_readpage,
584 .writepage = metapage_writepage,
585 .sync_page = block_sync_page,
586 .releasepage = metapage_releasepage,
587 .invalidatepage = metapage_invalidatepage,
588 .set_page_dirty = __set_page_dirty_nobuffers,
589};
590
591struct metapage *__get_metapage(struct inode *inode, unsigned long lblock,
592 unsigned int size, int absolute,
593 unsigned long new)
594{
595 int l2BlocksPerPage;
596 int l2bsize;
597 struct address_space *mapping;
598 struct metapage *mp = NULL;
599 struct page *page;
600 unsigned long page_index;
601 unsigned long page_offset;
602
603 jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
604 inode->i_ino, lblock, absolute);
605
606 l2bsize = inode->i_blkbits;
607 l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize;
608 page_index = lblock >> l2BlocksPerPage;
609 page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize;
610 if ((page_offset + size) > PAGE_CACHE_SIZE) {
611 jfs_err("MetaData crosses page boundary!!");
612 jfs_err("lblock = %lx, size = %d", lblock, size);
613 dump_stack();
614 return NULL;
615 }
616 if (absolute)
617 mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping;
618 else {
619
620
621
622
623
624 if ((lblock << inode->i_blkbits) >= inode->i_size)
625 return NULL;
626 mapping = inode->i_mapping;
627 }
628
629 if (new && (PSIZE == PAGE_CACHE_SIZE)) {
630 page = grab_cache_page(mapping, page_index);
631 if (!page) {
632 jfs_err("grab_cache_page failed!");
633 return NULL;
634 }
635 SetPageUptodate(page);
636 } else {
637 page = read_mapping_page(mapping, page_index, NULL);
638 if (IS_ERR(page) || !PageUptodate(page)) {
639 jfs_err("read_mapping_page failed!");
640 return NULL;
641 }
642 lock_page(page);
643 }
644
645 mp = page_to_mp(page, page_offset);
646 if (mp) {
647 if (mp->logical_size != size) {
648 jfs_error(inode->i_sb,
649 "__get_metapage: mp->logical_size != size");
650 jfs_err("logical_size = %d, size = %d",
651 mp->logical_size, size);
652 dump_stack();
653 goto unlock;
654 }
655 mp->count++;
656 lock_metapage(mp);
657 if (test_bit(META_discard, &mp->flag)) {
658 if (!new) {
659 jfs_error(inode->i_sb,
660 "__get_metapage: using a "
661 "discarded metapage");
662 discard_metapage(mp);
663 goto unlock;
664 }
665 clear_bit(META_discard, &mp->flag);
666 }
667 } else {
668 INCREMENT(mpStat.pagealloc);
669 mp = alloc_metapage(GFP_NOFS);
670 mp->page = page;
671 mp->flag = 0;
672 mp->xflag = COMMIT_PAGE;
673 mp->count = 1;
674 mp->nohomeok = 0;
675 mp->logical_size = size;
676 mp->data = page_address(page) + page_offset;
677 mp->index = lblock;
678 if (unlikely(insert_metapage(page, mp))) {
679 free_metapage(mp);
680 goto unlock;
681 }
682 lock_metapage(mp);
683 }
684
685 if (new) {
686 jfs_info("zeroing mp = 0x%p", mp);
687 memset(mp->data, 0, PSIZE);
688 }
689
690 unlock_page(page);
691 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
692 return mp;
693
694unlock:
695 unlock_page(page);
696 return NULL;
697}
698
699void grab_metapage(struct metapage * mp)
700{
701 jfs_info("grab_metapage: mp = 0x%p", mp);
702 page_cache_get(mp->page);
703 lock_page(mp->page);
704 mp->count++;
705 lock_metapage(mp);
706 unlock_page(mp->page);
707}
708
709void force_metapage(struct metapage *mp)
710{
711 struct page *page = mp->page;
712 jfs_info("force_metapage: mp = 0x%p", mp);
713 set_bit(META_forcewrite, &mp->flag);
714 clear_bit(META_sync, &mp->flag);
715 page_cache_get(page);
716 lock_page(page);
717 set_page_dirty(page);
718 write_one_page(page, 1);
719 clear_bit(META_forcewrite, &mp->flag);
720 page_cache_release(page);
721}
722
723void hold_metapage(struct metapage *mp)
724{
725 lock_page(mp->page);
726}
727
728void put_metapage(struct metapage *mp)
729{
730 if (mp->count || mp->nohomeok) {
731
732 unlock_page(mp->page);
733 return;
734 }
735 page_cache_get(mp->page);
736 mp->count++;
737 lock_metapage(mp);
738 unlock_page(mp->page);
739 release_metapage(mp);
740}
741
742void release_metapage(struct metapage * mp)
743{
744 struct page *page = mp->page;
745 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
746
747 BUG_ON(!page);
748
749 lock_page(page);
750 unlock_metapage(mp);
751
752 assert(mp->count);
753 if (--mp->count || mp->nohomeok) {
754 unlock_page(page);
755 page_cache_release(page);
756 return;
757 }
758
759 if (test_bit(META_dirty, &mp->flag)) {
760 set_page_dirty(page);
761 if (test_bit(META_sync, &mp->flag)) {
762 clear_bit(META_sync, &mp->flag);
763 write_one_page(page, 1);
764 lock_page(page);
765 }
766 } else if (mp->lsn)
767 remove_from_logsync(mp);
768
769
770 drop_metapage(page, mp);
771
772 unlock_page(page);
773 page_cache_release(page);
774}
775
776void __invalidate_metapages(struct inode *ip, s64 addr, int len)
777{
778 sector_t lblock;
779 int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
780 int BlocksPerPage = 1 << l2BlocksPerPage;
781
782 struct address_space *mapping =
783 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
784 struct metapage *mp;
785 struct page *page;
786 unsigned int offset;
787
788
789
790
791
792 for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
793 lblock += BlocksPerPage) {
794 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
795 if (!page)
796 continue;
797 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
798 mp = page_to_mp(page, offset);
799 if (!mp)
800 continue;
801 if (mp->index < addr)
802 continue;
803 if (mp->index >= addr + len)
804 break;
805
806 clear_bit(META_dirty, &mp->flag);
807 set_bit(META_discard, &mp->flag);
808 if (mp->lsn)
809 remove_from_logsync(mp);
810 }
811 unlock_page(page);
812 page_cache_release(page);
813 }
814}
815
816#ifdef CONFIG_JFS_STATISTICS
817static int jfs_mpstat_proc_show(struct seq_file *m, void *v)
818{
819 seq_printf(m,
820 "JFS Metapage statistics\n"
821 "=======================\n"
822 "page allocations = %d\n"
823 "page frees = %d\n"
824 "lock waits = %d\n",
825 mpStat.pagealloc,
826 mpStat.pagefree,
827 mpStat.lockwait);
828 return 0;
829}
830
831static int jfs_mpstat_proc_open(struct inode *inode, struct file *file)
832{
833 return single_open(file, jfs_mpstat_proc_show, NULL);
834}
835
836const struct file_operations jfs_mpstat_proc_fops = {
837 .owner = THIS_MODULE,
838 .open = jfs_mpstat_proc_open,
839 .read = seq_read,
840 .llseek = seq_lseek,
841 .release = single_release,
842};
843#endif
844
