Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) International Business Machines Corp., 2000-2003 | |
3 | * Portions Copyright (C) Christoph Hellwig, 2001-2002 | |
4 | * | |
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. | |
9 | * | |
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. | |
14 | * | |
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 | |
18 | */ | |
19 | ||
20 | #include <linux/fs.h> | |
21 | #include <linux/init.h> | |
22 | #include <linux/buffer_head.h> | |
23 | #include <linux/mempool.h> | |
24 | #include <linux/delay.h> | |
25 | #include "jfs_incore.h" | |
26 | #include "jfs_superblock.h" | |
27 | #include "jfs_filsys.h" | |
28 | #include "jfs_metapage.h" | |
29 | #include "jfs_txnmgr.h" | |
30 | #include "jfs_debug.h" | |
31 | ||
32 | static DEFINE_SPINLOCK(meta_lock); | |
33 | ||
34 | #ifdef CONFIG_JFS_STATISTICS | |
35 | static struct { | |
36 | uint pagealloc; /* # of page allocations */ | |
37 | uint pagefree; /* # of page frees */ | |
38 | uint lockwait; /* # of sleeping lock_metapage() calls */ | |
39 | } mpStat; | |
40 | #endif | |
41 | ||
42 | ||
43 | #define HASH_BITS 10 /* This makes hash_table 1 4K page */ | |
44 | #define HASH_SIZE (1 << HASH_BITS) | |
45 | static struct metapage **hash_table = NULL; | |
46 | static unsigned long hash_order; | |
47 | ||
48 | ||
49 | static inline int metapage_locked(struct metapage *mp) | |
50 | { | |
51 | return test_bit(META_locked, &mp->flag); | |
52 | } | |
53 | ||
54 | static inline int trylock_metapage(struct metapage *mp) | |
55 | { | |
56 | return test_and_set_bit(META_locked, &mp->flag); | |
57 | } | |
58 | ||
59 | static inline void unlock_metapage(struct metapage *mp) | |
60 | { | |
61 | clear_bit(META_locked, &mp->flag); | |
62 | wake_up(&mp->wait); | |
63 | } | |
64 | ||
65 | static void __lock_metapage(struct metapage *mp) | |
66 | { | |
67 | DECLARE_WAITQUEUE(wait, current); | |
68 | ||
69 | INCREMENT(mpStat.lockwait); | |
70 | ||
71 | add_wait_queue_exclusive(&mp->wait, &wait); | |
72 | do { | |
73 | set_current_state(TASK_UNINTERRUPTIBLE); | |
74 | if (metapage_locked(mp)) { | |
75 | spin_unlock(&meta_lock); | |
76 | schedule(); | |
77 | spin_lock(&meta_lock); | |
78 | } | |
79 | } while (trylock_metapage(mp)); | |
80 | __set_current_state(TASK_RUNNING); | |
81 | remove_wait_queue(&mp->wait, &wait); | |
82 | } | |
83 | ||
84 | /* needs meta_lock */ | |
85 | static inline void lock_metapage(struct metapage *mp) | |
86 | { | |
87 | if (trylock_metapage(mp)) | |
88 | __lock_metapage(mp); | |
89 | } | |
90 | ||
91 | #define METAPOOL_MIN_PAGES 32 | |
92 | static kmem_cache_t *metapage_cache; | |
93 | static mempool_t *metapage_mempool; | |
94 | ||
95 | static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags) | |
96 | { | |
97 | struct metapage *mp = (struct metapage *)foo; | |
98 | ||
99 | if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == | |
100 | SLAB_CTOR_CONSTRUCTOR) { | |
101 | mp->lid = 0; | |
102 | mp->lsn = 0; | |
103 | mp->flag = 0; | |
104 | mp->data = NULL; | |
105 | mp->clsn = 0; | |
106 | mp->log = NULL; | |
107 | set_bit(META_free, &mp->flag); | |
108 | init_waitqueue_head(&mp->wait); | |
109 | } | |
110 | } | |
111 | ||
112 | static inline struct metapage *alloc_metapage(int gfp_mask) | |
113 | { | |
114 | return mempool_alloc(metapage_mempool, gfp_mask); | |
115 | } | |
116 | ||
117 | static inline void free_metapage(struct metapage *mp) | |
118 | { | |
119 | mp->flag = 0; | |
120 | set_bit(META_free, &mp->flag); | |
121 | ||
122 | mempool_free(mp, metapage_mempool); | |
123 | } | |
124 | ||
125 | int __init metapage_init(void) | |
126 | { | |
127 | /* | |
128 | * Allocate the metapage structures | |
129 | */ | |
130 | metapage_cache = kmem_cache_create("jfs_mp", sizeof(struct metapage), | |
131 | 0, 0, init_once, NULL); | |
132 | if (metapage_cache == NULL) | |
133 | return -ENOMEM; | |
134 | ||
135 | metapage_mempool = mempool_create(METAPOOL_MIN_PAGES, mempool_alloc_slab, | |
136 | mempool_free_slab, metapage_cache); | |
137 | ||
138 | if (metapage_mempool == NULL) { | |
139 | kmem_cache_destroy(metapage_cache); | |
140 | return -ENOMEM; | |
141 | } | |
142 | /* | |
143 | * Now the hash list | |
144 | */ | |
145 | for (hash_order = 0; | |
146 | ((PAGE_SIZE << hash_order) / sizeof(void *)) < HASH_SIZE; | |
147 | hash_order++); | |
148 | hash_table = | |
149 | (struct metapage **) __get_free_pages(GFP_KERNEL, hash_order); | |
150 | assert(hash_table); | |
151 | memset(hash_table, 0, PAGE_SIZE << hash_order); | |
152 | ||
153 | return 0; | |
154 | } | |
155 | ||
156 | void metapage_exit(void) | |
157 | { | |
158 | mempool_destroy(metapage_mempool); | |
159 | kmem_cache_destroy(metapage_cache); | |
160 | } | |
161 | ||
162 | /* | |
163 | * Basically same hash as in pagemap.h, but using our hash table | |
164 | */ | |
165 | static struct metapage **meta_hash(struct address_space *mapping, | |
166 | unsigned long index) | |
167 | { | |
168 | #define i (((unsigned long)mapping)/ \ | |
169 | (sizeof(struct inode) & ~(sizeof(struct inode) -1 ))) | |
170 | #define s(x) ((x) + ((x) >> HASH_BITS)) | |
171 | return hash_table + (s(i + index) & (HASH_SIZE - 1)); | |
172 | #undef i | |
173 | #undef s | |
174 | } | |
175 | ||
176 | static struct metapage *search_hash(struct metapage ** hash_ptr, | |
177 | struct address_space *mapping, | |
178 | unsigned long index) | |
179 | { | |
180 | struct metapage *ptr; | |
181 | ||
182 | for (ptr = *hash_ptr; ptr; ptr = ptr->hash_next) { | |
183 | if ((ptr->mapping == mapping) && (ptr->index == index)) | |
184 | return ptr; | |
185 | } | |
186 | ||
187 | return NULL; | |
188 | } | |
189 | ||
190 | static void add_to_hash(struct metapage * mp, struct metapage ** hash_ptr) | |
191 | { | |
192 | if (*hash_ptr) | |
193 | (*hash_ptr)->hash_prev = mp; | |
194 | ||
195 | mp->hash_prev = NULL; | |
196 | mp->hash_next = *hash_ptr; | |
197 | *hash_ptr = mp; | |
198 | } | |
199 | ||
200 | static void remove_from_hash(struct metapage * mp, struct metapage ** hash_ptr) | |
201 | { | |
202 | if (mp->hash_prev) | |
203 | mp->hash_prev->hash_next = mp->hash_next; | |
204 | else { | |
205 | assert(*hash_ptr == mp); | |
206 | *hash_ptr = mp->hash_next; | |
207 | } | |
208 | ||
209 | if (mp->hash_next) | |
210 | mp->hash_next->hash_prev = mp->hash_prev; | |
211 | } | |
212 | ||
213 | struct metapage *__get_metapage(struct inode *inode, unsigned long lblock, | |
214 | unsigned int size, int absolute, | |
215 | unsigned long new) | |
216 | { | |
217 | struct metapage **hash_ptr; | |
218 | int l2BlocksPerPage; | |
219 | int l2bsize; | |
220 | struct address_space *mapping; | |
221 | struct metapage *mp; | |
222 | unsigned long page_index; | |
223 | unsigned long page_offset; | |
224 | ||
225 | jfs_info("__get_metapage: inode = 0x%p, lblock = 0x%lx", inode, lblock); | |
226 | ||
227 | if (absolute) | |
228 | mapping = inode->i_sb->s_bdev->bd_inode->i_mapping; | |
229 | else { | |
230 | /* | |
231 | * If an nfs client tries to read an inode that is larger | |
232 | * than any existing inodes, we may try to read past the | |
233 | * end of the inode map | |
234 | */ | |
235 | if ((lblock << inode->i_blkbits) >= inode->i_size) | |
236 | return NULL; | |
237 | mapping = inode->i_mapping; | |
238 | } | |
239 | ||
240 | hash_ptr = meta_hash(mapping, lblock); | |
241 | again: | |
242 | spin_lock(&meta_lock); | |
243 | mp = search_hash(hash_ptr, mapping, lblock); | |
244 | if (mp) { | |
245 | page_found: | |
246 | if (test_bit(META_stale, &mp->flag)) { | |
247 | spin_unlock(&meta_lock); | |
248 | msleep(1); | |
249 | goto again; | |
250 | } | |
251 | mp->count++; | |
252 | lock_metapage(mp); | |
253 | spin_unlock(&meta_lock); | |
254 | if (test_bit(META_discard, &mp->flag)) { | |
255 | if (!new) { | |
256 | jfs_error(inode->i_sb, | |
257 | "__get_metapage: using a " | |
258 | "discarded metapage"); | |
259 | release_metapage(mp); | |
260 | return NULL; | |
261 | } | |
262 | clear_bit(META_discard, &mp->flag); | |
263 | } | |
264 | jfs_info("__get_metapage: found 0x%p, in hash", mp); | |
265 | if (mp->logical_size != size) { | |
266 | jfs_error(inode->i_sb, | |
267 | "__get_metapage: mp->logical_size != size"); | |
268 | release_metapage(mp); | |
269 | return NULL; | |
270 | } | |
271 | } else { | |
272 | l2bsize = inode->i_blkbits; | |
273 | l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize; | |
274 | page_index = lblock >> l2BlocksPerPage; | |
275 | page_offset = (lblock - (page_index << l2BlocksPerPage)) << | |
276 | l2bsize; | |
277 | if ((page_offset + size) > PAGE_CACHE_SIZE) { | |
278 | spin_unlock(&meta_lock); | |
279 | jfs_err("MetaData crosses page boundary!!"); | |
280 | return NULL; | |
281 | } | |
282 | ||
283 | /* | |
284 | * Locks held on aggregate inode pages are usually | |
285 | * not held long, and they are taken in critical code | |
286 | * paths (committing dirty inodes, txCommit thread) | |
287 | * | |
288 | * Attempt to get metapage without blocking, tapping into | |
289 | * reserves if necessary. | |
290 | */ | |
291 | mp = NULL; | |
292 | if (JFS_IP(inode)->fileset == AGGREGATE_I) { | |
293 | mp = alloc_metapage(GFP_ATOMIC); | |
294 | if (!mp) { | |
295 | /* | |
296 | * mempool is supposed to protect us from | |
297 | * failing here. We will try a blocking | |
298 | * call, but a deadlock is possible here | |
299 | */ | |
300 | printk(KERN_WARNING | |
301 | "__get_metapage: atomic call to mempool_alloc failed.\n"); | |
302 | printk(KERN_WARNING | |
303 | "Will attempt blocking call\n"); | |
304 | } | |
305 | } | |
306 | if (!mp) { | |
307 | struct metapage *mp2; | |
308 | ||
309 | spin_unlock(&meta_lock); | |
310 | mp = alloc_metapage(GFP_NOFS); | |
311 | spin_lock(&meta_lock); | |
312 | ||
313 | /* we dropped the meta_lock, we need to search the | |
314 | * hash again. | |
315 | */ | |
316 | mp2 = search_hash(hash_ptr, mapping, lblock); | |
317 | if (mp2) { | |
318 | free_metapage(mp); | |
319 | mp = mp2; | |
320 | goto page_found; | |
321 | } | |
322 | } | |
323 | mp->flag = 0; | |
324 | lock_metapage(mp); | |
325 | if (absolute) | |
326 | set_bit(META_absolute, &mp->flag); | |
327 | mp->xflag = COMMIT_PAGE; | |
328 | mp->count = 1; | |
329 | atomic_set(&mp->nohomeok,0); | |
330 | mp->mapping = mapping; | |
331 | mp->index = lblock; | |
332 | mp->page = NULL; | |
333 | mp->logical_size = size; | |
334 | add_to_hash(mp, hash_ptr); | |
335 | spin_unlock(&meta_lock); | |
336 | ||
337 | if (new) { | |
338 | jfs_info("__get_metapage: Calling grab_cache_page"); | |
339 | mp->page = grab_cache_page(mapping, page_index); | |
340 | if (!mp->page) { | |
341 | jfs_err("grab_cache_page failed!"); | |
342 | goto freeit; | |
343 | } else { | |
344 | INCREMENT(mpStat.pagealloc); | |
345 | unlock_page(mp->page); | |
346 | } | |
347 | } else { | |
348 | jfs_info("__get_metapage: Calling read_cache_page"); | |
349 | mp->page = read_cache_page(mapping, lblock, | |
350 | (filler_t *)mapping->a_ops->readpage, NULL); | |
351 | if (IS_ERR(mp->page)) { | |
352 | jfs_err("read_cache_page failed!"); | |
353 | goto freeit; | |
354 | } else | |
355 | INCREMENT(mpStat.pagealloc); | |
356 | } | |
357 | mp->data = kmap(mp->page) + page_offset; | |
358 | } | |
359 | ||
360 | if (new) | |
361 | memset(mp->data, 0, PSIZE); | |
362 | ||
363 | jfs_info("__get_metapage: returning = 0x%p", mp); | |
364 | return mp; | |
365 | ||
366 | freeit: | |
367 | spin_lock(&meta_lock); | |
368 | remove_from_hash(mp, hash_ptr); | |
369 | free_metapage(mp); | |
370 | spin_unlock(&meta_lock); | |
371 | return NULL; | |
372 | } | |
373 | ||
374 | void hold_metapage(struct metapage * mp, int force) | |
375 | { | |
376 | spin_lock(&meta_lock); | |
377 | ||
378 | mp->count++; | |
379 | ||
380 | if (force) { | |
381 | ASSERT (!(test_bit(META_forced, &mp->flag))); | |
382 | if (trylock_metapage(mp)) | |
383 | set_bit(META_forced, &mp->flag); | |
384 | } else | |
385 | lock_metapage(mp); | |
386 | ||
387 | spin_unlock(&meta_lock); | |
388 | } | |
389 | ||
390 | static void __write_metapage(struct metapage * mp) | |
391 | { | |
392 | int l2bsize = mp->mapping->host->i_blkbits; | |
393 | int l2BlocksPerPage = PAGE_CACHE_SHIFT - l2bsize; | |
394 | unsigned long page_index; | |
395 | unsigned long page_offset; | |
396 | int rc; | |
397 | ||
398 | jfs_info("__write_metapage: mp = 0x%p", mp); | |
399 | ||
400 | page_index = mp->page->index; | |
401 | page_offset = | |
402 | (mp->index - (page_index << l2BlocksPerPage)) << l2bsize; | |
403 | ||
404 | lock_page(mp->page); | |
405 | rc = mp->mapping->a_ops->prepare_write(NULL, mp->page, page_offset, | |
406 | page_offset + | |
407 | mp->logical_size); | |
408 | if (rc) { | |
409 | jfs_err("prepare_write return %d!", rc); | |
410 | ClearPageUptodate(mp->page); | |
411 | unlock_page(mp->page); | |
412 | clear_bit(META_dirty, &mp->flag); | |
413 | return; | |
414 | } | |
415 | rc = mp->mapping->a_ops->commit_write(NULL, mp->page, page_offset, | |
416 | page_offset + | |
417 | mp->logical_size); | |
418 | if (rc) { | |
419 | jfs_err("commit_write returned %d", rc); | |
420 | } | |
421 | ||
422 | unlock_page(mp->page); | |
423 | clear_bit(META_dirty, &mp->flag); | |
424 | ||
425 | jfs_info("__write_metapage done"); | |
426 | } | |
427 | ||
428 | static inline void sync_metapage(struct metapage *mp) | |
429 | { | |
430 | struct page *page = mp->page; | |
431 | ||
432 | page_cache_get(page); | |
433 | lock_page(page); | |
434 | ||
435 | /* we're done with this page - no need to check for errors */ | |
436 | if (page_has_buffers(page)) | |
437 | write_one_page(page, 1); | |
438 | else | |
439 | unlock_page(page); | |
440 | page_cache_release(page); | |
441 | } | |
442 | ||
443 | void release_metapage(struct metapage * mp) | |
444 | { | |
445 | struct jfs_log *log; | |
446 | ||
447 | jfs_info("release_metapage: mp = 0x%p, flag = 0x%lx", mp, mp->flag); | |
448 | ||
449 | spin_lock(&meta_lock); | |
450 | if (test_bit(META_forced, &mp->flag)) { | |
451 | clear_bit(META_forced, &mp->flag); | |
452 | mp->count--; | |
453 | spin_unlock(&meta_lock); | |
454 | return; | |
455 | } | |
456 | ||
457 | assert(mp->count); | |
458 | if (--mp->count || atomic_read(&mp->nohomeok)) { | |
459 | unlock_metapage(mp); | |
460 | spin_unlock(&meta_lock); | |
461 | return; | |
462 | } | |
463 | ||
464 | if (mp->page) { | |
465 | set_bit(META_stale, &mp->flag); | |
466 | spin_unlock(&meta_lock); | |
467 | kunmap(mp->page); | |
468 | mp->data = NULL; | |
469 | if (test_bit(META_dirty, &mp->flag)) | |
470 | __write_metapage(mp); | |
471 | if (test_bit(META_sync, &mp->flag)) { | |
472 | sync_metapage(mp); | |
473 | clear_bit(META_sync, &mp->flag); | |
474 | } | |
475 | ||
476 | if (test_bit(META_discard, &mp->flag)) { | |
477 | lock_page(mp->page); | |
478 | block_invalidatepage(mp->page, 0); | |
479 | unlock_page(mp->page); | |
480 | } | |
481 | ||
482 | page_cache_release(mp->page); | |
483 | mp->page = NULL; | |
484 | INCREMENT(mpStat.pagefree); | |
485 | spin_lock(&meta_lock); | |
486 | } | |
487 | ||
488 | if (mp->lsn) { | |
489 | /* | |
490 | * Remove metapage from logsynclist. | |
491 | */ | |
492 | log = mp->log; | |
493 | LOGSYNC_LOCK(log); | |
494 | mp->log = NULL; | |
495 | mp->lsn = 0; | |
496 | mp->clsn = 0; | |
497 | log->count--; | |
498 | list_del(&mp->synclist); | |
499 | LOGSYNC_UNLOCK(log); | |
500 | } | |
501 | remove_from_hash(mp, meta_hash(mp->mapping, mp->index)); | |
502 | spin_unlock(&meta_lock); | |
503 | ||
504 | free_metapage(mp); | |
505 | } | |
506 | ||
507 | void __invalidate_metapages(struct inode *ip, s64 addr, int len) | |
508 | { | |
509 | struct metapage **hash_ptr; | |
510 | unsigned long lblock; | |
511 | int l2BlocksPerPage = PAGE_CACHE_SHIFT - ip->i_blkbits; | |
512 | /* All callers are interested in block device's mapping */ | |
513 | struct address_space *mapping = ip->i_sb->s_bdev->bd_inode->i_mapping; | |
514 | struct metapage *mp; | |
515 | struct page *page; | |
516 | ||
517 | /* | |
518 | * First, mark metapages to discard. They will eventually be | |
519 | * released, but should not be written. | |
520 | */ | |
521 | for (lblock = addr; lblock < addr + len; | |
522 | lblock += 1 << l2BlocksPerPage) { | |
523 | hash_ptr = meta_hash(mapping, lblock); | |
524 | again: | |
525 | spin_lock(&meta_lock); | |
526 | mp = search_hash(hash_ptr, mapping, lblock); | |
527 | if (mp) { | |
528 | if (test_bit(META_stale, &mp->flag)) { | |
529 | spin_unlock(&meta_lock); | |
530 | msleep(1); | |
531 | goto again; | |
532 | } | |
533 | ||
534 | clear_bit(META_dirty, &mp->flag); | |
535 | set_bit(META_discard, &mp->flag); | |
536 | spin_unlock(&meta_lock); | |
537 | } else { | |
538 | spin_unlock(&meta_lock); | |
539 | page = find_lock_page(mapping, lblock>>l2BlocksPerPage); | |
540 | if (page) { | |
541 | block_invalidatepage(page, 0); | |
542 | unlock_page(page); | |
543 | page_cache_release(page); | |
544 | } | |
545 | } | |
546 | } | |
547 | } | |
548 | ||
549 | #ifdef CONFIG_JFS_STATISTICS | |
550 | int jfs_mpstat_read(char *buffer, char **start, off_t offset, int length, | |
551 | int *eof, void *data) | |
552 | { | |
553 | int len = 0; | |
554 | off_t begin; | |
555 | ||
556 | len += sprintf(buffer, | |
557 | "JFS Metapage statistics\n" | |
558 | "=======================\n" | |
559 | "page allocations = %d\n" | |
560 | "page frees = %d\n" | |
561 | "lock waits = %d\n", | |
562 | mpStat.pagealloc, | |
563 | mpStat.pagefree, | |
564 | mpStat.lockwait); | |
565 | ||
566 | begin = offset; | |
567 | *start = buffer + begin; | |
568 | len -= begin; | |
569 | ||
570 | if (len > length) | |
571 | len = length; | |
572 | else | |
573 | *eof = 1; | |
574 | ||
575 | if (len < 0) | |
576 | len = 0; | |
577 | ||
578 | return len; | |
579 | } | |
580 | #endif |