diff options
Diffstat (limited to 'fs/cachefiles/rdwr.c')
-rw-r--r-- | fs/cachefiles/rdwr.c | 879 |
1 files changed, 879 insertions, 0 deletions
diff --git a/fs/cachefiles/rdwr.c b/fs/cachefiles/rdwr.c new file mode 100644 index 000000000000..a69787e7dd96 --- /dev/null +++ b/fs/cachefiles/rdwr.c | |||
@@ -0,0 +1,879 @@ | |||
1 | /* Storage object read/write | ||
2 | * | ||
3 | * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. | ||
4 | * Written by David Howells (dhowells@redhat.com) | ||
5 | * | ||
6 | * This program is free software; you can redistribute it and/or | ||
7 | * modify it under the terms of the GNU General Public Licence | ||
8 | * as published by the Free Software Foundation; either version | ||
9 | * 2 of the Licence, or (at your option) any later version. | ||
10 | */ | ||
11 | |||
12 | #include <linux/mount.h> | ||
13 | #include <linux/file.h> | ||
14 | #include "internal.h" | ||
15 | |||
16 | /* | ||
17 | * detect wake up events generated by the unlocking of pages in which we're | ||
18 | * interested | ||
19 | * - we use this to detect read completion of backing pages | ||
20 | * - the caller holds the waitqueue lock | ||
21 | */ | ||
22 | static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode, | ||
23 | int sync, void *_key) | ||
24 | { | ||
25 | struct cachefiles_one_read *monitor = | ||
26 | container_of(wait, struct cachefiles_one_read, monitor); | ||
27 | struct cachefiles_object *object; | ||
28 | struct wait_bit_key *key = _key; | ||
29 | struct page *page = wait->private; | ||
30 | |||
31 | ASSERT(key); | ||
32 | |||
33 | _enter("{%lu},%u,%d,{%p,%u}", | ||
34 | monitor->netfs_page->index, mode, sync, | ||
35 | key->flags, key->bit_nr); | ||
36 | |||
37 | if (key->flags != &page->flags || | ||
38 | key->bit_nr != PG_locked) | ||
39 | return 0; | ||
40 | |||
41 | _debug("--- monitor %p %lx ---", page, page->flags); | ||
42 | |||
43 | if (!PageUptodate(page) && !PageError(page)) | ||
44 | dump_stack(); | ||
45 | |||
46 | /* remove from the waitqueue */ | ||
47 | list_del(&wait->task_list); | ||
48 | |||
49 | /* move onto the action list and queue for FS-Cache thread pool */ | ||
50 | ASSERT(monitor->op); | ||
51 | |||
52 | object = container_of(monitor->op->op.object, | ||
53 | struct cachefiles_object, fscache); | ||
54 | |||
55 | spin_lock(&object->work_lock); | ||
56 | list_add_tail(&monitor->op_link, &monitor->op->to_do); | ||
57 | spin_unlock(&object->work_lock); | ||
58 | |||
59 | fscache_enqueue_retrieval(monitor->op); | ||
60 | return 0; | ||
61 | } | ||
62 | |||
63 | /* | ||
64 | * copy data from backing pages to netfs pages to complete a read operation | ||
65 | * - driven by FS-Cache's thread pool | ||
66 | */ | ||
67 | static void cachefiles_read_copier(struct fscache_operation *_op) | ||
68 | { | ||
69 | struct cachefiles_one_read *monitor; | ||
70 | struct cachefiles_object *object; | ||
71 | struct fscache_retrieval *op; | ||
72 | struct pagevec pagevec; | ||
73 | int error, max; | ||
74 | |||
75 | op = container_of(_op, struct fscache_retrieval, op); | ||
76 | object = container_of(op->op.object, | ||
77 | struct cachefiles_object, fscache); | ||
78 | |||
79 | _enter("{ino=%lu}", object->backer->d_inode->i_ino); | ||
80 | |||
81 | pagevec_init(&pagevec, 0); | ||
82 | |||
83 | max = 8; | ||
84 | spin_lock_irq(&object->work_lock); | ||
85 | |||
86 | while (!list_empty(&op->to_do)) { | ||
87 | monitor = list_entry(op->to_do.next, | ||
88 | struct cachefiles_one_read, op_link); | ||
89 | list_del(&monitor->op_link); | ||
90 | |||
91 | spin_unlock_irq(&object->work_lock); | ||
92 | |||
93 | _debug("- copy {%lu}", monitor->back_page->index); | ||
94 | |||
95 | error = -EIO; | ||
96 | if (PageUptodate(monitor->back_page)) { | ||
97 | copy_highpage(monitor->netfs_page, monitor->back_page); | ||
98 | |||
99 | pagevec_add(&pagevec, monitor->netfs_page); | ||
100 | fscache_mark_pages_cached(monitor->op, &pagevec); | ||
101 | error = 0; | ||
102 | } | ||
103 | |||
104 | if (error) | ||
105 | cachefiles_io_error_obj( | ||
106 | object, | ||
107 | "Readpage failed on backing file %lx", | ||
108 | (unsigned long) monitor->back_page->flags); | ||
109 | |||
110 | page_cache_release(monitor->back_page); | ||
111 | |||
112 | fscache_end_io(op, monitor->netfs_page, error); | ||
113 | page_cache_release(monitor->netfs_page); | ||
114 | fscache_put_retrieval(op); | ||
115 | kfree(monitor); | ||
116 | |||
117 | /* let the thread pool have some air occasionally */ | ||
118 | max--; | ||
119 | if (max < 0 || need_resched()) { | ||
120 | if (!list_empty(&op->to_do)) | ||
121 | fscache_enqueue_retrieval(op); | ||
122 | _leave(" [maxed out]"); | ||
123 | return; | ||
124 | } | ||
125 | |||
126 | spin_lock_irq(&object->work_lock); | ||
127 | } | ||
128 | |||
129 | spin_unlock_irq(&object->work_lock); | ||
130 | _leave(""); | ||
131 | } | ||
132 | |||
133 | /* | ||
134 | * read the corresponding page to the given set from the backing file | ||
135 | * - an uncertain page is simply discarded, to be tried again another time | ||
136 | */ | ||
137 | static int cachefiles_read_backing_file_one(struct cachefiles_object *object, | ||
138 | struct fscache_retrieval *op, | ||
139 | struct page *netpage, | ||
140 | struct pagevec *pagevec) | ||
141 | { | ||
142 | struct cachefiles_one_read *monitor; | ||
143 | struct address_space *bmapping; | ||
144 | struct page *newpage, *backpage; | ||
145 | int ret; | ||
146 | |||
147 | _enter(""); | ||
148 | |||
149 | pagevec_reinit(pagevec); | ||
150 | |||
151 | _debug("read back %p{%lu,%d}", | ||
152 | netpage, netpage->index, page_count(netpage)); | ||
153 | |||
154 | monitor = kzalloc(sizeof(*monitor), GFP_KERNEL); | ||
155 | if (!monitor) | ||
156 | goto nomem; | ||
157 | |||
158 | monitor->netfs_page = netpage; | ||
159 | monitor->op = fscache_get_retrieval(op); | ||
160 | |||
161 | init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter); | ||
162 | |||
163 | /* attempt to get hold of the backing page */ | ||
164 | bmapping = object->backer->d_inode->i_mapping; | ||
165 | newpage = NULL; | ||
166 | |||
167 | for (;;) { | ||
168 | backpage = find_get_page(bmapping, netpage->index); | ||
169 | if (backpage) | ||
170 | goto backing_page_already_present; | ||
171 | |||
172 | if (!newpage) { | ||
173 | newpage = page_cache_alloc_cold(bmapping); | ||
174 | if (!newpage) | ||
175 | goto nomem_monitor; | ||
176 | } | ||
177 | |||
178 | ret = add_to_page_cache(newpage, bmapping, | ||
179 | netpage->index, GFP_KERNEL); | ||
180 | if (ret == 0) | ||
181 | goto installed_new_backing_page; | ||
182 | if (ret != -EEXIST) | ||
183 | goto nomem_page; | ||
184 | } | ||
185 | |||
186 | /* we've installed a new backing page, so now we need to add it | ||
187 | * to the LRU list and start it reading */ | ||
188 | installed_new_backing_page: | ||
189 | _debug("- new %p", newpage); | ||
190 | |||
191 | backpage = newpage; | ||
192 | newpage = NULL; | ||
193 | |||
194 | page_cache_get(backpage); | ||
195 | pagevec_add(pagevec, backpage); | ||
196 | __pagevec_lru_add_file(pagevec); | ||
197 | |||
198 | read_backing_page: | ||
199 | ret = bmapping->a_ops->readpage(NULL, backpage); | ||
200 | if (ret < 0) | ||
201 | goto read_error; | ||
202 | |||
203 | /* set the monitor to transfer the data across */ | ||
204 | monitor_backing_page: | ||
205 | _debug("- monitor add"); | ||
206 | |||
207 | /* install the monitor */ | ||
208 | page_cache_get(monitor->netfs_page); | ||
209 | page_cache_get(backpage); | ||
210 | monitor->back_page = backpage; | ||
211 | monitor->monitor.private = backpage; | ||
212 | add_page_wait_queue(backpage, &monitor->monitor); | ||
213 | monitor = NULL; | ||
214 | |||
215 | /* but the page may have been read before the monitor was installed, so | ||
216 | * the monitor may miss the event - so we have to ensure that we do get | ||
217 | * one in such a case */ | ||
218 | if (trylock_page(backpage)) { | ||
219 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | ||
220 | unlock_page(backpage); | ||
221 | } | ||
222 | goto success; | ||
223 | |||
224 | /* if the backing page is already present, it can be in one of | ||
225 | * three states: read in progress, read failed or read okay */ | ||
226 | backing_page_already_present: | ||
227 | _debug("- present"); | ||
228 | |||
229 | if (newpage) { | ||
230 | page_cache_release(newpage); | ||
231 | newpage = NULL; | ||
232 | } | ||
233 | |||
234 | if (PageError(backpage)) | ||
235 | goto io_error; | ||
236 | |||
237 | if (PageUptodate(backpage)) | ||
238 | goto backing_page_already_uptodate; | ||
239 | |||
240 | if (!trylock_page(backpage)) | ||
241 | goto monitor_backing_page; | ||
242 | _debug("read %p {%lx}", backpage, backpage->flags); | ||
243 | goto read_backing_page; | ||
244 | |||
245 | /* the backing page is already up to date, attach the netfs | ||
246 | * page to the pagecache and LRU and copy the data across */ | ||
247 | backing_page_already_uptodate: | ||
248 | _debug("- uptodate"); | ||
249 | |||
250 | pagevec_add(pagevec, netpage); | ||
251 | fscache_mark_pages_cached(op, pagevec); | ||
252 | |||
253 | copy_highpage(netpage, backpage); | ||
254 | fscache_end_io(op, netpage, 0); | ||
255 | |||
256 | success: | ||
257 | _debug("success"); | ||
258 | ret = 0; | ||
259 | |||
260 | out: | ||
261 | if (backpage) | ||
262 | page_cache_release(backpage); | ||
263 | if (monitor) { | ||
264 | fscache_put_retrieval(monitor->op); | ||
265 | kfree(monitor); | ||
266 | } | ||
267 | _leave(" = %d", ret); | ||
268 | return ret; | ||
269 | |||
270 | read_error: | ||
271 | _debug("read error %d", ret); | ||
272 | if (ret == -ENOMEM) | ||
273 | goto out; | ||
274 | io_error: | ||
275 | cachefiles_io_error_obj(object, "Page read error on backing file"); | ||
276 | ret = -ENOBUFS; | ||
277 | goto out; | ||
278 | |||
279 | nomem_page: | ||
280 | page_cache_release(newpage); | ||
281 | nomem_monitor: | ||
282 | fscache_put_retrieval(monitor->op); | ||
283 | kfree(monitor); | ||
284 | nomem: | ||
285 | _leave(" = -ENOMEM"); | ||
286 | return -ENOMEM; | ||
287 | } | ||
288 | |||
289 | /* | ||
290 | * read a page from the cache or allocate a block in which to store it | ||
291 | * - cache withdrawal is prevented by the caller | ||
292 | * - returns -EINTR if interrupted | ||
293 | * - returns -ENOMEM if ran out of memory | ||
294 | * - returns -ENOBUFS if no buffers can be made available | ||
295 | * - returns -ENOBUFS if page is beyond EOF | ||
296 | * - if the page is backed by a block in the cache: | ||
297 | * - a read will be started which will call the callback on completion | ||
298 | * - 0 will be returned | ||
299 | * - else if the page is unbacked: | ||
300 | * - the metadata will be retained | ||
301 | * - -ENODATA will be returned | ||
302 | */ | ||
303 | int cachefiles_read_or_alloc_page(struct fscache_retrieval *op, | ||
304 | struct page *page, | ||
305 | gfp_t gfp) | ||
306 | { | ||
307 | struct cachefiles_object *object; | ||
308 | struct cachefiles_cache *cache; | ||
309 | struct pagevec pagevec; | ||
310 | struct inode *inode; | ||
311 | sector_t block0, block; | ||
312 | unsigned shift; | ||
313 | int ret; | ||
314 | |||
315 | object = container_of(op->op.object, | ||
316 | struct cachefiles_object, fscache); | ||
317 | cache = container_of(object->fscache.cache, | ||
318 | struct cachefiles_cache, cache); | ||
319 | |||
320 | _enter("{%p},{%lx},,,", object, page->index); | ||
321 | |||
322 | if (!object->backer) | ||
323 | return -ENOBUFS; | ||
324 | |||
325 | inode = object->backer->d_inode; | ||
326 | ASSERT(S_ISREG(inode->i_mode)); | ||
327 | ASSERT(inode->i_mapping->a_ops->bmap); | ||
328 | ASSERT(inode->i_mapping->a_ops->readpages); | ||
329 | |||
330 | /* calculate the shift required to use bmap */ | ||
331 | if (inode->i_sb->s_blocksize > PAGE_SIZE) | ||
332 | return -ENOBUFS; | ||
333 | |||
334 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; | ||
335 | |||
336 | op->op.flags = FSCACHE_OP_FAST; | ||
337 | op->op.processor = cachefiles_read_copier; | ||
338 | |||
339 | pagevec_init(&pagevec, 0); | ||
340 | |||
341 | /* we assume the absence or presence of the first block is a good | ||
342 | * enough indication for the page as a whole | ||
343 | * - TODO: don't use bmap() for this as it is _not_ actually good | ||
344 | * enough for this as it doesn't indicate errors, but it's all we've | ||
345 | * got for the moment | ||
346 | */ | ||
347 | block0 = page->index; | ||
348 | block0 <<= shift; | ||
349 | |||
350 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0); | ||
351 | _debug("%llx -> %llx", | ||
352 | (unsigned long long) block0, | ||
353 | (unsigned long long) block); | ||
354 | |||
355 | if (block) { | ||
356 | /* submit the apparently valid page to the backing fs to be | ||
357 | * read from disk */ | ||
358 | ret = cachefiles_read_backing_file_one(object, op, page, | ||
359 | &pagevec); | ||
360 | } else if (cachefiles_has_space(cache, 0, 1) == 0) { | ||
361 | /* there's space in the cache we can use */ | ||
362 | pagevec_add(&pagevec, page); | ||
363 | fscache_mark_pages_cached(op, &pagevec); | ||
364 | ret = -ENODATA; | ||
365 | } else { | ||
366 | ret = -ENOBUFS; | ||
367 | } | ||
368 | |||
369 | _leave(" = %d", ret); | ||
370 | return ret; | ||
371 | } | ||
372 | |||
373 | /* | ||
374 | * read the corresponding pages to the given set from the backing file | ||
375 | * - any uncertain pages are simply discarded, to be tried again another time | ||
376 | */ | ||
377 | static int cachefiles_read_backing_file(struct cachefiles_object *object, | ||
378 | struct fscache_retrieval *op, | ||
379 | struct list_head *list, | ||
380 | struct pagevec *mark_pvec) | ||
381 | { | ||
382 | struct cachefiles_one_read *monitor = NULL; | ||
383 | struct address_space *bmapping = object->backer->d_inode->i_mapping; | ||
384 | struct pagevec lru_pvec; | ||
385 | struct page *newpage = NULL, *netpage, *_n, *backpage = NULL; | ||
386 | int ret = 0; | ||
387 | |||
388 | _enter(""); | ||
389 | |||
390 | pagevec_init(&lru_pvec, 0); | ||
391 | |||
392 | list_for_each_entry_safe(netpage, _n, list, lru) { | ||
393 | list_del(&netpage->lru); | ||
394 | |||
395 | _debug("read back %p{%lu,%d}", | ||
396 | netpage, netpage->index, page_count(netpage)); | ||
397 | |||
398 | if (!monitor) { | ||
399 | monitor = kzalloc(sizeof(*monitor), GFP_KERNEL); | ||
400 | if (!monitor) | ||
401 | goto nomem; | ||
402 | |||
403 | monitor->op = fscache_get_retrieval(op); | ||
404 | init_waitqueue_func_entry(&monitor->monitor, | ||
405 | cachefiles_read_waiter); | ||
406 | } | ||
407 | |||
408 | for (;;) { | ||
409 | backpage = find_get_page(bmapping, netpage->index); | ||
410 | if (backpage) | ||
411 | goto backing_page_already_present; | ||
412 | |||
413 | if (!newpage) { | ||
414 | newpage = page_cache_alloc_cold(bmapping); | ||
415 | if (!newpage) | ||
416 | goto nomem; | ||
417 | } | ||
418 | |||
419 | ret = add_to_page_cache(newpage, bmapping, | ||
420 | netpage->index, GFP_KERNEL); | ||
421 | if (ret == 0) | ||
422 | goto installed_new_backing_page; | ||
423 | if (ret != -EEXIST) | ||
424 | goto nomem; | ||
425 | } | ||
426 | |||
427 | /* we've installed a new backing page, so now we need to add it | ||
428 | * to the LRU list and start it reading */ | ||
429 | installed_new_backing_page: | ||
430 | _debug("- new %p", newpage); | ||
431 | |||
432 | backpage = newpage; | ||
433 | newpage = NULL; | ||
434 | |||
435 | page_cache_get(backpage); | ||
436 | if (!pagevec_add(&lru_pvec, backpage)) | ||
437 | __pagevec_lru_add_file(&lru_pvec); | ||
438 | |||
439 | reread_backing_page: | ||
440 | ret = bmapping->a_ops->readpage(NULL, backpage); | ||
441 | if (ret < 0) | ||
442 | goto read_error; | ||
443 | |||
444 | /* add the netfs page to the pagecache and LRU, and set the | ||
445 | * monitor to transfer the data across */ | ||
446 | monitor_backing_page: | ||
447 | _debug("- monitor add"); | ||
448 | |||
449 | ret = add_to_page_cache(netpage, op->mapping, netpage->index, | ||
450 | GFP_KERNEL); | ||
451 | if (ret < 0) { | ||
452 | if (ret == -EEXIST) { | ||
453 | page_cache_release(netpage); | ||
454 | continue; | ||
455 | } | ||
456 | goto nomem; | ||
457 | } | ||
458 | |||
459 | page_cache_get(netpage); | ||
460 | if (!pagevec_add(&lru_pvec, netpage)) | ||
461 | __pagevec_lru_add_file(&lru_pvec); | ||
462 | |||
463 | /* install a monitor */ | ||
464 | page_cache_get(netpage); | ||
465 | monitor->netfs_page = netpage; | ||
466 | |||
467 | page_cache_get(backpage); | ||
468 | monitor->back_page = backpage; | ||
469 | monitor->monitor.private = backpage; | ||
470 | add_page_wait_queue(backpage, &monitor->monitor); | ||
471 | monitor = NULL; | ||
472 | |||
473 | /* but the page may have been read before the monitor was | ||
474 | * installed, so the monitor may miss the event - so we have to | ||
475 | * ensure that we do get one in such a case */ | ||
476 | if (trylock_page(backpage)) { | ||
477 | _debug("2unlock %p {%lx}", backpage, backpage->flags); | ||
478 | unlock_page(backpage); | ||
479 | } | ||
480 | |||
481 | page_cache_release(backpage); | ||
482 | backpage = NULL; | ||
483 | |||
484 | page_cache_release(netpage); | ||
485 | netpage = NULL; | ||
486 | continue; | ||
487 | |||
488 | /* if the backing page is already present, it can be in one of | ||
489 | * three states: read in progress, read failed or read okay */ | ||
490 | backing_page_already_present: | ||
491 | _debug("- present %p", backpage); | ||
492 | |||
493 | if (PageError(backpage)) | ||
494 | goto io_error; | ||
495 | |||
496 | if (PageUptodate(backpage)) | ||
497 | goto backing_page_already_uptodate; | ||
498 | |||
499 | _debug("- not ready %p{%lx}", backpage, backpage->flags); | ||
500 | |||
501 | if (!trylock_page(backpage)) | ||
502 | goto monitor_backing_page; | ||
503 | |||
504 | if (PageError(backpage)) { | ||
505 | _debug("error %lx", backpage->flags); | ||
506 | unlock_page(backpage); | ||
507 | goto io_error; | ||
508 | } | ||
509 | |||
510 | if (PageUptodate(backpage)) | ||
511 | goto backing_page_already_uptodate_unlock; | ||
512 | |||
513 | /* we've locked a page that's neither up to date nor erroneous, | ||
514 | * so we need to attempt to read it again */ | ||
515 | goto reread_backing_page; | ||
516 | |||
517 | /* the backing page is already up to date, attach the netfs | ||
518 | * page to the pagecache and LRU and copy the data across */ | ||
519 | backing_page_already_uptodate_unlock: | ||
520 | _debug("uptodate %lx", backpage->flags); | ||
521 | unlock_page(backpage); | ||
522 | backing_page_already_uptodate: | ||
523 | _debug("- uptodate"); | ||
524 | |||
525 | ret = add_to_page_cache(netpage, op->mapping, netpage->index, | ||
526 | GFP_KERNEL); | ||
527 | if (ret < 0) { | ||
528 | if (ret == -EEXIST) { | ||
529 | page_cache_release(netpage); | ||
530 | continue; | ||
531 | } | ||
532 | goto nomem; | ||
533 | } | ||
534 | |||
535 | copy_highpage(netpage, backpage); | ||
536 | |||
537 | page_cache_release(backpage); | ||
538 | backpage = NULL; | ||
539 | |||
540 | if (!pagevec_add(mark_pvec, netpage)) | ||
541 | fscache_mark_pages_cached(op, mark_pvec); | ||
542 | |||
543 | page_cache_get(netpage); | ||
544 | if (!pagevec_add(&lru_pvec, netpage)) | ||
545 | __pagevec_lru_add_file(&lru_pvec); | ||
546 | |||
547 | fscache_end_io(op, netpage, 0); | ||
548 | page_cache_release(netpage); | ||
549 | netpage = NULL; | ||
550 | continue; | ||
551 | } | ||
552 | |||
553 | netpage = NULL; | ||
554 | |||
555 | _debug("out"); | ||
556 | |||
557 | out: | ||
558 | /* tidy up */ | ||
559 | pagevec_lru_add_file(&lru_pvec); | ||
560 | |||
561 | if (newpage) | ||
562 | page_cache_release(newpage); | ||
563 | if (netpage) | ||
564 | page_cache_release(netpage); | ||
565 | if (backpage) | ||
566 | page_cache_release(backpage); | ||
567 | if (monitor) { | ||
568 | fscache_put_retrieval(op); | ||
569 | kfree(monitor); | ||
570 | } | ||
571 | |||
572 | list_for_each_entry_safe(netpage, _n, list, lru) { | ||
573 | list_del(&netpage->lru); | ||
574 | page_cache_release(netpage); | ||
575 | } | ||
576 | |||
577 | _leave(" = %d", ret); | ||
578 | return ret; | ||
579 | |||
580 | nomem: | ||
581 | _debug("nomem"); | ||
582 | ret = -ENOMEM; | ||
583 | goto out; | ||
584 | |||
585 | read_error: | ||
586 | _debug("read error %d", ret); | ||
587 | if (ret == -ENOMEM) | ||
588 | goto out; | ||
589 | io_error: | ||
590 | cachefiles_io_error_obj(object, "Page read error on backing file"); | ||
591 | ret = -ENOBUFS; | ||
592 | goto out; | ||
593 | } | ||
594 | |||
595 | /* | ||
596 | * read a list of pages from the cache or allocate blocks in which to store | ||
597 | * them | ||
598 | */ | ||
599 | int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op, | ||
600 | struct list_head *pages, | ||
601 | unsigned *nr_pages, | ||
602 | gfp_t gfp) | ||
603 | { | ||
604 | struct cachefiles_object *object; | ||
605 | struct cachefiles_cache *cache; | ||
606 | struct list_head backpages; | ||
607 | struct pagevec pagevec; | ||
608 | struct inode *inode; | ||
609 | struct page *page, *_n; | ||
610 | unsigned shift, nrbackpages; | ||
611 | int ret, ret2, space; | ||
612 | |||
613 | object = container_of(op->op.object, | ||
614 | struct cachefiles_object, fscache); | ||
615 | cache = container_of(object->fscache.cache, | ||
616 | struct cachefiles_cache, cache); | ||
617 | |||
618 | _enter("{OBJ%x,%d},,%d,,", | ||
619 | object->fscache.debug_id, atomic_read(&op->op.usage), | ||
620 | *nr_pages); | ||
621 | |||
622 | if (!object->backer) | ||
623 | return -ENOBUFS; | ||
624 | |||
625 | space = 1; | ||
626 | if (cachefiles_has_space(cache, 0, *nr_pages) < 0) | ||
627 | space = 0; | ||
628 | |||
629 | inode = object->backer->d_inode; | ||
630 | ASSERT(S_ISREG(inode->i_mode)); | ||
631 | ASSERT(inode->i_mapping->a_ops->bmap); | ||
632 | ASSERT(inode->i_mapping->a_ops->readpages); | ||
633 | |||
634 | /* calculate the shift required to use bmap */ | ||
635 | if (inode->i_sb->s_blocksize > PAGE_SIZE) | ||
636 | return -ENOBUFS; | ||
637 | |||
638 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; | ||
639 | |||
640 | pagevec_init(&pagevec, 0); | ||
641 | |||
642 | op->op.flags = FSCACHE_OP_FAST; | ||
643 | op->op.processor = cachefiles_read_copier; | ||
644 | |||
645 | INIT_LIST_HEAD(&backpages); | ||
646 | nrbackpages = 0; | ||
647 | |||
648 | ret = space ? -ENODATA : -ENOBUFS; | ||
649 | list_for_each_entry_safe(page, _n, pages, lru) { | ||
650 | sector_t block0, block; | ||
651 | |||
652 | /* we assume the absence or presence of the first block is a | ||
653 | * good enough indication for the page as a whole | ||
654 | * - TODO: don't use bmap() for this as it is _not_ actually | ||
655 | * good enough for this as it doesn't indicate errors, but | ||
656 | * it's all we've got for the moment | ||
657 | */ | ||
658 | block0 = page->index; | ||
659 | block0 <<= shift; | ||
660 | |||
661 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, | ||
662 | block0); | ||
663 | _debug("%llx -> %llx", | ||
664 | (unsigned long long) block0, | ||
665 | (unsigned long long) block); | ||
666 | |||
667 | if (block) { | ||
668 | /* we have data - add it to the list to give to the | ||
669 | * backing fs */ | ||
670 | list_move(&page->lru, &backpages); | ||
671 | (*nr_pages)--; | ||
672 | nrbackpages++; | ||
673 | } else if (space && pagevec_add(&pagevec, page) == 0) { | ||
674 | fscache_mark_pages_cached(op, &pagevec); | ||
675 | ret = -ENODATA; | ||
676 | } | ||
677 | } | ||
678 | |||
679 | if (pagevec_count(&pagevec) > 0) | ||
680 | fscache_mark_pages_cached(op, &pagevec); | ||
681 | |||
682 | if (list_empty(pages)) | ||
683 | ret = 0; | ||
684 | |||
685 | /* submit the apparently valid pages to the backing fs to be read from | ||
686 | * disk */ | ||
687 | if (nrbackpages > 0) { | ||
688 | ret2 = cachefiles_read_backing_file(object, op, &backpages, | ||
689 | &pagevec); | ||
690 | if (ret2 == -ENOMEM || ret2 == -EINTR) | ||
691 | ret = ret2; | ||
692 | } | ||
693 | |||
694 | if (pagevec_count(&pagevec) > 0) | ||
695 | fscache_mark_pages_cached(op, &pagevec); | ||
696 | |||
697 | _leave(" = %d [nr=%u%s]", | ||
698 | ret, *nr_pages, list_empty(pages) ? " empty" : ""); | ||
699 | return ret; | ||
700 | } | ||
701 | |||
702 | /* | ||
703 | * allocate a block in the cache in which to store a page | ||
704 | * - cache withdrawal is prevented by the caller | ||
705 | * - returns -EINTR if interrupted | ||
706 | * - returns -ENOMEM if ran out of memory | ||
707 | * - returns -ENOBUFS if no buffers can be made available | ||
708 | * - returns -ENOBUFS if page is beyond EOF | ||
709 | * - otherwise: | ||
710 | * - the metadata will be retained | ||
711 | * - 0 will be returned | ||
712 | */ | ||
713 | int cachefiles_allocate_page(struct fscache_retrieval *op, | ||
714 | struct page *page, | ||
715 | gfp_t gfp) | ||
716 | { | ||
717 | struct cachefiles_object *object; | ||
718 | struct cachefiles_cache *cache; | ||
719 | struct pagevec pagevec; | ||
720 | int ret; | ||
721 | |||
722 | object = container_of(op->op.object, | ||
723 | struct cachefiles_object, fscache); | ||
724 | cache = container_of(object->fscache.cache, | ||
725 | struct cachefiles_cache, cache); | ||
726 | |||
727 | _enter("%p,{%lx},", object, page->index); | ||
728 | |||
729 | ret = cachefiles_has_space(cache, 0, 1); | ||
730 | if (ret == 0) { | ||
731 | pagevec_init(&pagevec, 0); | ||
732 | pagevec_add(&pagevec, page); | ||
733 | fscache_mark_pages_cached(op, &pagevec); | ||
734 | } else { | ||
735 | ret = -ENOBUFS; | ||
736 | } | ||
737 | |||
738 | _leave(" = %d", ret); | ||
739 | return ret; | ||
740 | } | ||
741 | |||
742 | /* | ||
743 | * allocate blocks in the cache in which to store a set of pages | ||
744 | * - cache withdrawal is prevented by the caller | ||
745 | * - returns -EINTR if interrupted | ||
746 | * - returns -ENOMEM if ran out of memory | ||
747 | * - returns -ENOBUFS if some buffers couldn't be made available | ||
748 | * - returns -ENOBUFS if some pages are beyond EOF | ||
749 | * - otherwise: | ||
750 | * - -ENODATA will be returned | ||
751 | * - metadata will be retained for any page marked | ||
752 | */ | ||
753 | int cachefiles_allocate_pages(struct fscache_retrieval *op, | ||
754 | struct list_head *pages, | ||
755 | unsigned *nr_pages, | ||
756 | gfp_t gfp) | ||
757 | { | ||
758 | struct cachefiles_object *object; | ||
759 | struct cachefiles_cache *cache; | ||
760 | struct pagevec pagevec; | ||
761 | struct page *page; | ||
762 | int ret; | ||
763 | |||
764 | object = container_of(op->op.object, | ||
765 | struct cachefiles_object, fscache); | ||
766 | cache = container_of(object->fscache.cache, | ||
767 | struct cachefiles_cache, cache); | ||
768 | |||
769 | _enter("%p,,,%d,", object, *nr_pages); | ||
770 | |||
771 | ret = cachefiles_has_space(cache, 0, *nr_pages); | ||
772 | if (ret == 0) { | ||
773 | pagevec_init(&pagevec, 0); | ||
774 | |||
775 | list_for_each_entry(page, pages, lru) { | ||
776 | if (pagevec_add(&pagevec, page) == 0) | ||
777 | fscache_mark_pages_cached(op, &pagevec); | ||
778 | } | ||
779 | |||
780 | if (pagevec_count(&pagevec) > 0) | ||
781 | fscache_mark_pages_cached(op, &pagevec); | ||
782 | ret = -ENODATA; | ||
783 | } else { | ||
784 | ret = -ENOBUFS; | ||
785 | } | ||
786 | |||
787 | _leave(" = %d", ret); | ||
788 | return ret; | ||
789 | } | ||
790 | |||
791 | /* | ||
792 | * request a page be stored in the cache | ||
793 | * - cache withdrawal is prevented by the caller | ||
794 | * - this request may be ignored if there's no cache block available, in which | ||
795 | * case -ENOBUFS will be returned | ||
796 | * - if the op is in progress, 0 will be returned | ||
797 | */ | ||
798 | int cachefiles_write_page(struct fscache_storage *op, struct page *page) | ||
799 | { | ||
800 | struct cachefiles_object *object; | ||
801 | struct cachefiles_cache *cache; | ||
802 | mm_segment_t old_fs; | ||
803 | struct file *file; | ||
804 | loff_t pos; | ||
805 | void *data; | ||
806 | int ret; | ||
807 | |||
808 | ASSERT(op != NULL); | ||
809 | ASSERT(page != NULL); | ||
810 | |||
811 | object = container_of(op->op.object, | ||
812 | struct cachefiles_object, fscache); | ||
813 | |||
814 | _enter("%p,%p{%lx},,,", object, page, page->index); | ||
815 | |||
816 | if (!object->backer) { | ||
817 | _leave(" = -ENOBUFS"); | ||
818 | return -ENOBUFS; | ||
819 | } | ||
820 | |||
821 | ASSERT(S_ISREG(object->backer->d_inode->i_mode)); | ||
822 | |||
823 | cache = container_of(object->fscache.cache, | ||
824 | struct cachefiles_cache, cache); | ||
825 | |||
826 | /* write the page to the backing filesystem and let it store it in its | ||
827 | * own time */ | ||
828 | dget(object->backer); | ||
829 | mntget(cache->mnt); | ||
830 | file = dentry_open(object->backer, cache->mnt, O_RDWR, | ||
831 | cache->cache_cred); | ||
832 | if (IS_ERR(file)) { | ||
833 | ret = PTR_ERR(file); | ||
834 | } else { | ||
835 | ret = -EIO; | ||
836 | if (file->f_op->write) { | ||
837 | pos = (loff_t) page->index << PAGE_SHIFT; | ||
838 | data = kmap(page); | ||
839 | old_fs = get_fs(); | ||
840 | set_fs(KERNEL_DS); | ||
841 | ret = file->f_op->write( | ||
842 | file, (const void __user *) data, PAGE_SIZE, | ||
843 | &pos); | ||
844 | set_fs(old_fs); | ||
845 | kunmap(page); | ||
846 | if (ret != PAGE_SIZE) | ||
847 | ret = -EIO; | ||
848 | } | ||
849 | fput(file); | ||
850 | } | ||
851 | |||
852 | if (ret < 0) { | ||
853 | if (ret == -EIO) | ||
854 | cachefiles_io_error_obj( | ||
855 | object, "Write page to backing file failed"); | ||
856 | ret = -ENOBUFS; | ||
857 | } | ||
858 | |||
859 | _leave(" = %d", ret); | ||
860 | return ret; | ||
861 | } | ||
862 | |||
863 | /* | ||
864 | * detach a backing block from a page | ||
865 | * - cache withdrawal is prevented by the caller | ||
866 | */ | ||
867 | void cachefiles_uncache_page(struct fscache_object *_object, struct page *page) | ||
868 | { | ||
869 | struct cachefiles_object *object; | ||
870 | struct cachefiles_cache *cache; | ||
871 | |||
872 | object = container_of(_object, struct cachefiles_object, fscache); | ||
873 | cache = container_of(object->fscache.cache, | ||
874 | struct cachefiles_cache, cache); | ||
875 | |||
876 | _enter("%p,{%lu}", object, page->index); | ||
877 | |||
878 | spin_unlock(&object->fscache.cookie->lock); | ||
879 | } | ||