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/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"
35
36#ifdef CONFIG_JFS_STATISTICS
37static struct {
38 uint pagealloc;
39 uint pagefree;
40 uint lockwait;
41} mpStat;
42#endif
43
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)
46
47static inline void unlock_metapage(struct metapage *mp)
48{
49 clear_bit_unlock(META_locked, &mp->flag);
50 wake_up(&mp->wait);
51}
52
53static inline void __lock_metapage(struct metapage *mp)
54{
55 DECLARE_WAITQUEUE(wait, current);
56 INCREMENT(mpStat.lockwait);
57 add_wait_queue_exclusive(&mp->wait, &wait);
58 do {
59 set_current_state(TASK_UNINTERRUPTIBLE);
60 if (metapage_locked(mp)) {
61 unlock_page(mp->page);
62 io_schedule();
63 lock_page(mp->page);
64 }
65 } while (trylock_metapage(mp));
66 __set_current_state(TASK_RUNNING);
67 remove_wait_queue(&mp->wait, &wait);
68}
69
70
71
72
73static inline void lock_metapage(struct metapage *mp)
74{
75 if (trylock_metapage(mp))
76 __lock_metapage(mp);
77}
78
79#define METAPOOL_MIN_PAGES 32
80static struct kmem_cache *metapage_cache;
81static mempool_t *metapage_mempool;
82
83#define MPS_PER_PAGE (PAGE_CACHE_SIZE >> L2PSIZE)
84
85#if MPS_PER_PAGE > 1
86
87struct meta_anchor {
88 int mp_count;
89 atomic_t io_count;
90 struct metapage *mp[MPS_PER_PAGE];
91};
92#define mp_anchor(page) ((struct meta_anchor *)page_private(page))
93
94static inline struct metapage *page_to_mp(struct page *page, int offset)
95{
96 if (!PagePrivate(page))
97 return NULL;
98 return mp_anchor(page)->mp[offset >> L2PSIZE];
99}
100
101static inline int insert_metapage(struct page *page, struct metapage *mp)
102{
103 struct meta_anchor *a;
104 int index;
105 int l2mp_blocks;
106
107 if (PagePrivate(page))
108 a = mp_anchor(page);
109 else {
110 a = kzalloc(sizeof(struct meta_anchor), GFP_NOFS);
111 if (!a)
112 return -ENOMEM;
113 set_page_private(page, (unsigned long)a);
114 SetPagePrivate(page);
115 kmap(page);
116 }
117
118 if (mp) {
119 l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
120 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
121 a->mp_count++;
122 a->mp[index] = mp;
123 }
124
125 return 0;
126}
127
128static inline void remove_metapage(struct page *page, struct metapage *mp)
129{
130 struct meta_anchor *a = mp_anchor(page);
131 int l2mp_blocks = L2PSIZE - page->mapping->host->i_blkbits;
132 int index;
133
134 index = (mp->index >> l2mp_blocks) & (MPS_PER_PAGE - 1);
135
136 BUG_ON(a->mp[index] != mp);
137
138 a->mp[index] = NULL;
139 if (--a->mp_count == 0) {
140 kfree(a);
141 set_page_private(page, 0);
142 ClearPagePrivate(page);
143 kunmap(page);
144 }
145}
146
147static inline void inc_io(struct page *page)
148{
149 atomic_inc(&mp_anchor(page)->io_count);
150}
151
152static inline void dec_io(struct page *page, void (*handler) (struct page *))
153{
154 if (atomic_dec_and_test(&mp_anchor(page)->io_count))
155 handler(page);
156}
157
158#else
159static inline struct metapage *page_to_mp(struct page *page, int offset)
160{
161 return PagePrivate(page) ? (struct metapage *)page_private(page) : NULL;
162}
163
164static inline int insert_metapage(struct page *page, struct metapage *mp)
165{
166 if (mp) {
167 set_page_private(page, (unsigned long)mp);
168 SetPagePrivate(page);
169 kmap(page);
170 }
171 return 0;
172}
173
174static inline void remove_metapage(struct page *page, struct metapage *mp)
175{
176 set_page_private(page, 0);
177 ClearPagePrivate(page);
178 kunmap(page);
179}
180
181#define inc_io(page) do {} while(0)
182#define dec_io(page, handler) handler(page)
183
184#endif
185
186static void init_once(void *foo)
187{
188 struct metapage *mp = (struct metapage *)foo;
189
190 mp->lid = 0;
191 mp->lsn = 0;
192 mp->flag = 0;
193 mp->data = NULL;
194 mp->clsn = 0;
195 mp->log = NULL;
196 set_bit(META_free, &mp->flag);
197 init_waitqueue_head(&mp->wait);
198}
199
200static inline struct metapage *alloc_metapage(gfp_t gfp_mask)
201{
202 return mempool_alloc(metapage_mempool, gfp_mask);
203}
204
205static inline void free_metapage(struct metapage *mp)
206{
207 mp->flag = 0;
208 set_bit(META_free, &mp->flag);
209
210 mempool_free(mp, metapage_mempool);
211}
212
213int __init metapage_init(void)
214{
215
216
217
218 metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage),
219 0, 0, init_once);
220 if (metapage_cache == NULL)
221 return -ENOMEM;
222
223 metapage_mempool = mempool_create_slab_pool(METAPOOL_MIN_PAGES,
224 metapage_cache);
225
226 if (metapage_mempool == NULL) {
227 kmem_cache_destroy(metapage_cache);
228 return -ENOMEM;
229 }
230
231 return 0;
232}
233
234void metapage_exit(void)
235{
236 mempool_destroy(metapage_mempool);
237 kmem_cache_destroy(metapage_cache);
238}
239
240static inline void drop_metapage(struct page *page, struct metapage *mp)
241{
242 if (mp->count || mp->nohomeok || test_bit(META_dirty, &mp->flag) ||
243 test_bit(META_io, &mp->flag))
244 return;
245 remove_metapage(page, mp);
246 INCREMENT(mpStat.pagefree);
247 free_metapage(mp);
248}
249
250
251
252
253
254static sector_t metapage_get_blocks(struct inode *inode, sector_t lblock,
255 int *len)
256{
257 int rc = 0;
258 int xflag;
259 s64 xaddr;
260 sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
261 inode->i_blkbits;
262
263 if (lblock >= file_blocks)
264 return 0;
265 if (lblock + *len > file_blocks)
266 *len = file_blocks - lblock;
267
268 if (inode->i_ino) {
269 rc = xtLookup(inode, (s64)lblock, *len, &xflag, &xaddr, len, 0);
270 if ((rc == 0) && *len)
271 lblock = (sector_t)xaddr;
272 else
273 lblock = 0;
274 }
275
276 return lblock;
277}
278
279static void last_read_complete(struct page *page)
280{
281 if (!PageError(page))
282 SetPageUptodate(page);
283 unlock_page(page);
284}
285
286static void metapage_read_end_io(struct bio *bio, int err)
287{
288 struct page *page = bio->bi_private;
289
290 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
291 printk(KERN_ERR "metapage_read_end_io: I/O error\n");
292 SetPageError(page);
293 }
294
295 dec_io(page, last_read_complete);
296 bio_put(bio);
297}
298
299static void remove_from_logsync(struct metapage *mp)
300{
301 struct jfs_log *log = mp->log;
302 unsigned long flags;
303
304
305
306
307 if (!log)
308 return;
309
310 LOGSYNC_LOCK(log, flags);
311 if (mp->lsn) {
312 mp->log = NULL;
313 mp->lsn = 0;
314 mp->clsn = 0;
315 log->count--;
316 list_del(&mp->synclist);
317 }
318 LOGSYNC_UNLOCK(log, flags);
319}
320
321static void last_write_complete(struct page *page)
322{
323 struct metapage *mp;
324 unsigned int offset;
325
326 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
327 mp = page_to_mp(page, offset);
328 if (mp && test_bit(META_io, &mp->flag)) {
329 if (mp->lsn)
330 remove_from_logsync(mp);
331 clear_bit(META_io, &mp->flag);
332 }
333
334
335
336
337 }
338 end_page_writeback(page);
339}
340
341static void metapage_write_end_io(struct bio *bio, int err)
342{
343 struct page *page = bio->bi_private;
344
345 BUG_ON(!PagePrivate(page));
346
347 if (! test_bit(BIO_UPTODATE, &bio->bi_flags)) {
348 printk(KERN_ERR "metapage_write_end_io: I/O error\n");
349 SetPageError(page);
350 }
351 dec_io(page, last_write_complete);
352 bio_put(bio);
353}
354
355static int metapage_writepage(struct page *page, struct writeback_control *wbc)
356{
357 struct bio *bio = NULL;
358 int block_offset;
359 struct inode *inode = page->mapping->host;
360 int blocks_per_mp = JFS_SBI(inode->i_sb)->nbperpage;
361 int len;
362 int xlen;
363 struct metapage *mp;
364 int redirty = 0;
365 sector_t lblock;
366 int nr_underway = 0;
367 sector_t pblock;
368 sector_t next_block = 0;
369 sector_t page_start;
370 unsigned long bio_bytes = 0;
371 unsigned long bio_offset = 0;
372 int offset;
373 int bad_blocks = 0;
374
375 page_start = (sector_t)page->index <<
376 (PAGE_CACHE_SHIFT - inode->i_blkbits);
377 BUG_ON(!PageLocked(page));
378 BUG_ON(PageWriteback(page));
379 set_page_writeback(page);
380
381 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
382 mp = page_to_mp(page, offset);
383
384 if (!mp || !test_bit(META_dirty, &mp->flag))
385 continue;
386
387 if (mp->nohomeok && !test_bit(META_forcewrite, &mp->flag)) {
388 redirty = 1;
389
390
391
392
393 if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
394 jfs_flush_journal(mp->log, 0);
395 continue;
396 }
397
398 clear_bit(META_dirty, &mp->flag);
399 set_bit(META_io, &mp->flag);
400 block_offset = offset >> inode->i_blkbits;
401 lblock = page_start + block_offset;
402 if (bio) {
403 if (xlen && lblock == next_block) {
404
405 len = min(xlen, blocks_per_mp);
406 xlen -= len;
407 bio_bytes += len << inode->i_blkbits;
408 continue;
409 }
410
411 if (bio_add_page(bio, page, bio_bytes, bio_offset) <
412 bio_bytes)
413 goto add_failed;
414
415
416
417
418 inc_io(page);
419 if (!bio->bi_iter.bi_size)
420 goto dump_bio;
421 submit_bio(WRITE, bio);
422 nr_underway++;
423 bio = NULL;
424 } else
425 inc_io(page);
426 xlen = (PAGE_CACHE_SIZE - offset) >> inode->i_blkbits;
427 pblock = metapage_get_blocks(inode, lblock, &xlen);
428 if (!pblock) {
429 printk(KERN_ERR "JFS: metapage_get_blocks failed\n");
430
431
432
433
434 bad_blocks++;
435 continue;
436 }
437 len = min(xlen, (int)JFS_SBI(inode->i_sb)->nbperpage);
438
439 bio = bio_alloc(GFP_NOFS, 1);
440 bio->bi_bdev = inode->i_sb->s_bdev;
441 bio->bi_iter.bi_sector = pblock << (inode->i_blkbits - 9);
442 bio->bi_end_io = metapage_write_end_io;
443 bio->bi_private = page;
444
445
446 bio_offset = offset;
447 bio_bytes = len << inode->i_blkbits;
448
449 xlen -= len;
450 next_block = lblock + len;
451 }
452 if (bio) {
453 if (bio_add_page(bio, page, bio_bytes, bio_offset) < bio_bytes)
454 goto add_failed;
455 if (!bio->bi_iter.bi_size)
456 goto dump_bio;
457
458 submit_bio(WRITE, bio);
459 nr_underway++;
460 }
461 if (redirty)
462 redirty_page_for_writepage(wbc, page);
463
464 unlock_page(page);
465
466 if (bad_blocks)
467 goto err_out;
468
469 if (nr_underway == 0)
470 end_page_writeback(page);
471
472 return 0;
473add_failed:
474
475 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
476 goto skip;
477dump_bio:
478 print_hex_dump(KERN_ERR, "JFS: dump of bio: ", DUMP_PREFIX_ADDRESS, 16,
479 4, bio, sizeof(*bio), 0);
480skip:
481 bio_put(bio);
482 unlock_page(page);
483 dec_io(page, last_write_complete);
484err_out:
485 while (bad_blocks--)
486 dec_io(page, last_write_complete);
487 return -EIO;
488}
489
490static int metapage_readpage(struct file *fp, struct page *page)
491{
492 struct inode *inode = page->mapping->host;
493 struct bio *bio = NULL;
494 int block_offset;
495 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
496 sector_t page_start;
497 sector_t pblock;
498 int xlen;
499 unsigned int len;
500 int offset;
501
502 BUG_ON(!PageLocked(page));
503 page_start = (sector_t)page->index <<
504 (PAGE_CACHE_SHIFT - inode->i_blkbits);
505
506 block_offset = 0;
507 while (block_offset < blocks_per_page) {
508 xlen = blocks_per_page - block_offset;
509 pblock = metapage_get_blocks(inode, page_start + block_offset,
510 &xlen);
511 if (pblock) {
512 if (!PagePrivate(page))
513 insert_metapage(page, NULL);
514 inc_io(page);
515 if (bio)
516 submit_bio(READ, bio);
517
518 bio = bio_alloc(GFP_NOFS, 1);
519 bio->bi_bdev = inode->i_sb->s_bdev;
520 bio->bi_iter.bi_sector =
521 pblock << (inode->i_blkbits - 9);
522 bio->bi_end_io = metapage_read_end_io;
523 bio->bi_private = page;
524 len = xlen << inode->i_blkbits;
525 offset = block_offset << inode->i_blkbits;
526 if (bio_add_page(bio, page, len, offset) < len)
527 goto add_failed;
528 block_offset += xlen;
529 } else
530 block_offset++;
531 }
532 if (bio)
533 submit_bio(READ, bio);
534 else
535 unlock_page(page);
536
537 return 0;
538
539add_failed:
540 printk(KERN_ERR "JFS: bio_add_page failed unexpectedly\n");
541 bio_put(bio);
542 dec_io(page, last_read_complete);
543 return -EIO;
544}
545
546static int metapage_releasepage(struct page *page, gfp_t gfp_mask)
547{
548 struct metapage *mp;
549 int ret = 1;
550 int offset;
551
552 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
553 mp = page_to_mp(page, offset);
554
555 if (!mp)
556 continue;
557
558 jfs_info("metapage_releasepage: mp = 0x%p", mp);
559 if (mp->count || mp->nohomeok ||
560 test_bit(META_dirty, &mp->flag)) {
561 jfs_info("count = %ld, nohomeok = %d", mp->count,
562 mp->nohomeok);
563 ret = 0;
564 continue;
565 }
566 if (mp->lsn)
567 remove_from_logsync(mp);
568 remove_metapage(page, mp);
569 INCREMENT(mpStat.pagefree);
570 free_metapage(mp);
571 }
572 return ret;
573}
574
575static void metapage_invalidatepage(struct page *page, unsigned int offset,
576 unsigned int length)
577{
578 BUG_ON(offset || length < PAGE_CACHE_SIZE);
579
580 BUG_ON(PageWriteback(page));
581
582 metapage_releasepage(page, 0);
583}
584
585const struct address_space_operations jfs_metapage_aops = {
586 .readpage = metapage_readpage,
587 .writepage = metapage_writepage,
588 .releasepage = metapage_releasepage,
589 .invalidatepage = metapage_invalidatepage,
590 .set_page_dirty = __set_page_dirty_nobuffers,
591};
592
593struct metapage *__get_metapage(struct inode *inode, unsigned long lblock,
594 unsigned int size, int absolute,
595 unsigned long new)
596{
597 int l2BlocksPerPage;
598 int l2bsize;
599 struct address_space *mapping;
600 struct metapage *mp = NULL;
601 struct page *page;
602 unsigned long page_index;
603 unsigned long page_offset;
604
605 jfs_info("__get_metapage: ino = %ld, lblock = 0x%lx, abs=%d",
606 inode->i_ino, lblock, absolute);
607
608 l2bsize = inode->i_blkbits;
609 l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize;
610 page_index = lblock >> l2BlocksPerPage;
611 page_offset = (lblock - (page_index << l2BlocksPerPage)) << l2bsize;
612 if ((page_offset + size) > PAGE_CACHE_SIZE) {
613 jfs_err("MetaData crosses page boundary!!");
614 jfs_err("lblock = %lx, size = %d", lblock, size);
615 dump_stack();
616 return NULL;
617 }
618 if (absolute)
619 mapping = JFS_SBI(inode->i_sb)->direct_inode->i_mapping;
620 else {
621
622
623
624
625
626 if ((lblock << inode->i_blkbits) >= inode->i_size)
627 return NULL;
628 mapping = inode->i_mapping;
629 }
630
631 if (new && (PSIZE == PAGE_CACHE_SIZE)) {
632 page = grab_cache_page(mapping, page_index);
633 if (!page) {
634 jfs_err("grab_cache_page failed!");
635 return NULL;
636 }
637 SetPageUptodate(page);
638 } else {
639 page = read_mapping_page(mapping, page_index, NULL);
640 if (IS_ERR(page) || !PageUptodate(page)) {
641 jfs_err("read_mapping_page failed!");
642 return NULL;
643 }
644 lock_page(page);
645 }
646
647 mp = page_to_mp(page, page_offset);
648 if (mp) {
649 if (mp->logical_size != size) {
650 jfs_error(inode->i_sb,
651 "get_mp->logical_size != size\n");
652 jfs_err("logical_size = %d, size = %d",
653 mp->logical_size, size);
654 dump_stack();
655 goto unlock;
656 }
657 mp->count++;
658 lock_metapage(mp);
659 if (test_bit(META_discard, &mp->flag)) {
660 if (!new) {
661 jfs_error(inode->i_sb,
662 "using a discarded metapage\n");
663 discard_metapage(mp);
664 goto unlock;
665 }
666 clear_bit(META_discard, &mp->flag);
667 }
668 } else {
669 INCREMENT(mpStat.pagealloc);
670 mp = alloc_metapage(GFP_NOFS);
671 mp->page = page;
672 mp->flag = 0;
673 mp->xflag = COMMIT_PAGE;
674 mp->count = 1;
675 mp->nohomeok = 0;
676 mp->logical_size = size;
677 mp->data = page_address(page) + page_offset;
678 mp->index = lblock;
679 if (unlikely(insert_metapage(page, mp))) {
680 free_metapage(mp);
681 goto unlock;
682 }
683 lock_metapage(mp);
684 }
685
686 if (new) {
687 jfs_info("zeroing mp = 0x%p", mp);
688 memset(mp->data, 0, PSIZE);
689 }
690
691 unlock_page(page);
692 jfs_info("__get_metapage: returning = 0x%p data = 0x%p", mp, mp->data);
693 return mp;
694
695unlock:
696 unlock_page(page);
697 return NULL;
698}
699
700void grab_metapage(struct metapage * mp)
701{
702 jfs_info("grab_metapage: mp = 0x%p", mp);
703 page_cache_get(mp->page);
704 lock_page(mp->page);
705 mp->count++;
706 lock_metapage(mp);
707 unlock_page(mp->page);
708}
709
710void force_metapage(struct metapage *mp)
711{
712 struct page *page = mp->page;
713 jfs_info("force_metapage: mp = 0x%p", mp);
714 set_bit(META_forcewrite, &mp->flag);
715 clear_bit(META_sync, &mp->flag);
716 page_cache_get(page);
717 lock_page(page);
718 set_page_dirty(page);
719 write_one_page(page, 1);
720 clear_bit(META_forcewrite, &mp->flag);
721 page_cache_release(page);
722}
723
724void hold_metapage(struct metapage *mp)
725{
726 lock_page(mp->page);
727}
728
729void put_metapage(struct metapage *mp)
730{
731 if (mp->count || mp->nohomeok) {
732
733 unlock_page(mp->page);
734 return;
735 }
736 page_cache_get(mp->page);
737 mp->count++;
738 lock_metapage(mp);
739 unlock_page(mp->page);
740 release_metapage(mp);
741}
742
743void release_metapage(struct metapage * mp)
744{
745 struct page *page = mp->page;
746 jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag);
747
748 BUG_ON(!page);
749
750 lock_page(page);
751 unlock_metapage(mp);
752
753 assert(mp->count);
754 if (--mp->count || mp->nohomeok) {
755 unlock_page(page);
756 page_cache_release(page);
757 return;
758 }
759
760 if (test_bit(META_dirty, &mp->flag)) {
761 set_page_dirty(page);
762 if (test_bit(META_sync, &mp->flag)) {
763 clear_bit(META_sync, &mp->flag);
764 write_one_page(page, 1);
765 lock_page(page);
766 }
767 } else if (mp->lsn)
768 remove_from_logsync(mp);
769
770
771 drop_metapage(page, mp);
772
773 unlock_page(page);
774 page_cache_release(page);
775}
776
777void __invalidate_metapages(struct inode *ip, s64 addr, int len)
778{
779 sector_t lblock;
780 int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits;
781 int BlocksPerPage = 1 << l2BlocksPerPage;
782
783 struct address_space *mapping =
784 JFS_SBI(ip->i_sb)->direct_inode->i_mapping;
785 struct metapage *mp;
786 struct page *page;
787 unsigned int offset;
788
789
790
791
792
793 for (lblock = addr & ~(BlocksPerPage - 1); lblock < addr + len;
794 lblock += BlocksPerPage) {
795 page = find_lock_page(mapping, lblock >> l2BlocksPerPage);
796 if (!page)
797 continue;
798 for (offset = 0; offset < PAGE_CACHE_SIZE; offset += PSIZE) {
799 mp = page_to_mp(page, offset);
800 if (!mp)
801 continue;
802 if (mp->index < addr)
803 continue;
804 if (mp->index >= addr + len)
805 break;
806
807 clear_bit(META_dirty, &mp->flag);
808 set_bit(META_discard, &mp->flag);
809 if (mp->lsn)
810 remove_from_logsync(mp);
811 }
812 unlock_page(page);
813 page_cache_release(page);
814 }
815}
816
817#ifdef CONFIG_JFS_STATISTICS
818static int jfs_mpstat_proc_show(struct seq_file *m, void *v)
819{
820 seq_printf(m,
821 "JFS Metapage statistics\n"
822 "=======================\n"
823 "page allocations = %d\n"
824 "page frees = %d\n"
825 "lock waits = %d\n",
826 mpStat.pagealloc,
827 mpStat.pagefree,
828 mpStat.lockwait);
829 return 0;
830}
831
832static int jfs_mpstat_proc_open(struct inode *inode, struct file *file)
833{
834 return single_open(file, jfs_mpstat_proc_show, NULL);
835}
836
837const struct file_operations jfs_mpstat_proc_fops = {
838 .owner = THIS_MODULE,
839 .open = jfs_mpstat_proc_open,
840 .read = seq_read,
841 .llseek = seq_lseek,
842 .release = single_release,
843};
844#endif
845
