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