Commit | Line | Data |
---|---|---|
9ae326a6 DH |
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> | |
5a0e3ad6 | 13 | #include <linux/slab.h> |
9ae326a6 DH |
14 | #include <linux/file.h> |
15 | #include "internal.h" | |
16 | ||
17 | /* | |
18 | * detect wake up events generated by the unlocking of pages in which we're | |
19 | * interested | |
20 | * - we use this to detect read completion of backing pages | |
21 | * - the caller holds the waitqueue lock | |
22 | */ | |
23 | static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode, | |
24 | int sync, void *_key) | |
25 | { | |
26 | struct cachefiles_one_read *monitor = | |
27 | container_of(wait, struct cachefiles_one_read, monitor); | |
28 | struct cachefiles_object *object; | |
29 | struct wait_bit_key *key = _key; | |
30 | struct page *page = wait->private; | |
31 | ||
32 | ASSERT(key); | |
33 | ||
34 | _enter("{%lu},%u,%d,{%p,%u}", | |
35 | monitor->netfs_page->index, mode, sync, | |
36 | key->flags, key->bit_nr); | |
37 | ||
38 | if (key->flags != &page->flags || | |
39 | key->bit_nr != PG_locked) | |
40 | return 0; | |
41 | ||
42 | _debug("--- monitor %p %lx ---", page, page->flags); | |
43 | ||
5e929b33 DH |
44 | if (!PageUptodate(page) && !PageError(page)) { |
45 | /* unlocked, not uptodate and not erronous? */ | |
46 | _debug("page probably truncated"); | |
47 | } | |
9ae326a6 DH |
48 | |
49 | /* remove from the waitqueue */ | |
50 | list_del(&wait->task_list); | |
51 | ||
52 | /* move onto the action list and queue for FS-Cache thread pool */ | |
53 | ASSERT(monitor->op); | |
54 | ||
55 | object = container_of(monitor->op->op.object, | |
56 | struct cachefiles_object, fscache); | |
57 | ||
58 | spin_lock(&object->work_lock); | |
59 | list_add_tail(&monitor->op_link, &monitor->op->to_do); | |
60 | spin_unlock(&object->work_lock); | |
61 | ||
62 | fscache_enqueue_retrieval(monitor->op); | |
63 | return 0; | |
64 | } | |
65 | ||
5e929b33 DH |
66 | /* |
67 | * handle a probably truncated page | |
68 | * - check to see if the page is still relevant and reissue the read if | |
69 | * possible | |
70 | * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we | |
71 | * must wait again and 0 if successful | |
72 | */ | |
73 | static int cachefiles_read_reissue(struct cachefiles_object *object, | |
74 | struct cachefiles_one_read *monitor) | |
75 | { | |
76 | struct address_space *bmapping = object->backer->d_inode->i_mapping; | |
77 | struct page *backpage = monitor->back_page, *backpage2; | |
78 | int ret; | |
79 | ||
37491a13 | 80 | _enter("{ino=%lx},{%lx,%lx}", |
5e929b33 DH |
81 | object->backer->d_inode->i_ino, |
82 | backpage->index, backpage->flags); | |
83 | ||
84 | /* skip if the page was truncated away completely */ | |
85 | if (backpage->mapping != bmapping) { | |
37491a13 | 86 | _leave(" = -ENODATA [mapping]"); |
5e929b33 DH |
87 | return -ENODATA; |
88 | } | |
89 | ||
90 | backpage2 = find_get_page(bmapping, backpage->index); | |
91 | if (!backpage2) { | |
37491a13 | 92 | _leave(" = -ENODATA [gone]"); |
5e929b33 DH |
93 | return -ENODATA; |
94 | } | |
95 | ||
96 | if (backpage != backpage2) { | |
97 | put_page(backpage2); | |
37491a13 | 98 | _leave(" = -ENODATA [different]"); |
5e929b33 DH |
99 | return -ENODATA; |
100 | } | |
101 | ||
102 | /* the page is still there and we already have a ref on it, so we don't | |
103 | * need a second */ | |
104 | put_page(backpage2); | |
105 | ||
106 | INIT_LIST_HEAD(&monitor->op_link); | |
107 | add_page_wait_queue(backpage, &monitor->monitor); | |
108 | ||
109 | if (trylock_page(backpage)) { | |
110 | ret = -EIO; | |
111 | if (PageError(backpage)) | |
112 | goto unlock_discard; | |
113 | ret = 0; | |
114 | if (PageUptodate(backpage)) | |
115 | goto unlock_discard; | |
116 | ||
37491a13 | 117 | _debug("reissue read"); |
5e929b33 DH |
118 | ret = bmapping->a_ops->readpage(NULL, backpage); |
119 | if (ret < 0) | |
120 | goto unlock_discard; | |
121 | } | |
122 | ||
123 | /* but the page may have been read before the monitor was installed, so | |
124 | * the monitor may miss the event - so we have to ensure that we do get | |
125 | * one in such a case */ | |
126 | if (trylock_page(backpage)) { | |
127 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
128 | unlock_page(backpage); | |
129 | } | |
130 | ||
131 | /* it'll reappear on the todo list */ | |
37491a13 | 132 | _leave(" = -EINPROGRESS"); |
5e929b33 DH |
133 | return -EINPROGRESS; |
134 | ||
135 | unlock_discard: | |
136 | unlock_page(backpage); | |
137 | spin_lock_irq(&object->work_lock); | |
138 | list_del(&monitor->op_link); | |
139 | spin_unlock_irq(&object->work_lock); | |
37491a13 | 140 | _leave(" = %d", ret); |
5e929b33 DH |
141 | return ret; |
142 | } | |
143 | ||
9ae326a6 DH |
144 | /* |
145 | * copy data from backing pages to netfs pages to complete a read operation | |
146 | * - driven by FS-Cache's thread pool | |
147 | */ | |
148 | static void cachefiles_read_copier(struct fscache_operation *_op) | |
149 | { | |
150 | struct cachefiles_one_read *monitor; | |
151 | struct cachefiles_object *object; | |
152 | struct fscache_retrieval *op; | |
153 | struct pagevec pagevec; | |
154 | int error, max; | |
155 | ||
156 | op = container_of(_op, struct fscache_retrieval, op); | |
157 | object = container_of(op->op.object, | |
158 | struct cachefiles_object, fscache); | |
159 | ||
160 | _enter("{ino=%lu}", object->backer->d_inode->i_ino); | |
161 | ||
162 | pagevec_init(&pagevec, 0); | |
163 | ||
164 | max = 8; | |
165 | spin_lock_irq(&object->work_lock); | |
166 | ||
167 | while (!list_empty(&op->to_do)) { | |
168 | monitor = list_entry(op->to_do.next, | |
169 | struct cachefiles_one_read, op_link); | |
170 | list_del(&monitor->op_link); | |
171 | ||
172 | spin_unlock_irq(&object->work_lock); | |
173 | ||
174 | _debug("- copy {%lu}", monitor->back_page->index); | |
175 | ||
5e929b33 | 176 | recheck: |
9dc8d9bf DH |
177 | if (test_bit(FSCACHE_COOKIE_INVALIDATING, |
178 | &object->fscache.cookie->flags)) { | |
179 | error = -ESTALE; | |
180 | } else if (PageUptodate(monitor->back_page)) { | |
9ae326a6 | 181 | copy_highpage(monitor->netfs_page, monitor->back_page); |
c4d6d8db DH |
182 | fscache_mark_page_cached(monitor->op, |
183 | monitor->netfs_page); | |
9ae326a6 | 184 | error = 0; |
5e929b33 DH |
185 | } else if (!PageError(monitor->back_page)) { |
186 | /* the page has probably been truncated */ | |
187 | error = cachefiles_read_reissue(object, monitor); | |
188 | if (error == -EINPROGRESS) | |
189 | goto next; | |
190 | goto recheck; | |
191 | } else { | |
9ae326a6 DH |
192 | cachefiles_io_error_obj( |
193 | object, | |
194 | "Readpage failed on backing file %lx", | |
195 | (unsigned long) monitor->back_page->flags); | |
5e929b33 DH |
196 | error = -EIO; |
197 | } | |
9ae326a6 DH |
198 | |
199 | page_cache_release(monitor->back_page); | |
200 | ||
201 | fscache_end_io(op, monitor->netfs_page, error); | |
202 | page_cache_release(monitor->netfs_page); | |
9f10523f | 203 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
204 | fscache_put_retrieval(op); |
205 | kfree(monitor); | |
206 | ||
5e929b33 | 207 | next: |
9ae326a6 DH |
208 | /* let the thread pool have some air occasionally */ |
209 | max--; | |
210 | if (max < 0 || need_resched()) { | |
211 | if (!list_empty(&op->to_do)) | |
212 | fscache_enqueue_retrieval(op); | |
213 | _leave(" [maxed out]"); | |
214 | return; | |
215 | } | |
216 | ||
217 | spin_lock_irq(&object->work_lock); | |
218 | } | |
219 | ||
220 | spin_unlock_irq(&object->work_lock); | |
221 | _leave(""); | |
222 | } | |
223 | ||
224 | /* | |
225 | * read the corresponding page to the given set from the backing file | |
226 | * - an uncertain page is simply discarded, to be tried again another time | |
227 | */ | |
228 | static int cachefiles_read_backing_file_one(struct cachefiles_object *object, | |
229 | struct fscache_retrieval *op, | |
230 | struct page *netpage, | |
231 | struct pagevec *pagevec) | |
232 | { | |
233 | struct cachefiles_one_read *monitor; | |
234 | struct address_space *bmapping; | |
235 | struct page *newpage, *backpage; | |
236 | int ret; | |
237 | ||
238 | _enter(""); | |
239 | ||
240 | pagevec_reinit(pagevec); | |
241 | ||
242 | _debug("read back %p{%lu,%d}", | |
243 | netpage, netpage->index, page_count(netpage)); | |
244 | ||
5f4f9f4a | 245 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
246 | if (!monitor) |
247 | goto nomem; | |
248 | ||
249 | monitor->netfs_page = netpage; | |
250 | monitor->op = fscache_get_retrieval(op); | |
251 | ||
252 | init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter); | |
253 | ||
254 | /* attempt to get hold of the backing page */ | |
255 | bmapping = object->backer->d_inode->i_mapping; | |
256 | newpage = NULL; | |
257 | ||
258 | for (;;) { | |
259 | backpage = find_get_page(bmapping, netpage->index); | |
260 | if (backpage) | |
261 | goto backing_page_already_present; | |
262 | ||
263 | if (!newpage) { | |
5f4f9f4a DH |
264 | newpage = __page_cache_alloc(cachefiles_gfp | |
265 | __GFP_COLD); | |
9ae326a6 DH |
266 | if (!newpage) |
267 | goto nomem_monitor; | |
268 | } | |
269 | ||
270 | ret = add_to_page_cache(newpage, bmapping, | |
5f4f9f4a | 271 | netpage->index, cachefiles_gfp); |
9ae326a6 DH |
272 | if (ret == 0) |
273 | goto installed_new_backing_page; | |
274 | if (ret != -EEXIST) | |
275 | goto nomem_page; | |
276 | } | |
277 | ||
278 | /* we've installed a new backing page, so now we need to add it | |
279 | * to the LRU list and start it reading */ | |
280 | installed_new_backing_page: | |
281 | _debug("- new %p", newpage); | |
282 | ||
283 | backpage = newpage; | |
284 | newpage = NULL; | |
285 | ||
286 | page_cache_get(backpage); | |
287 | pagevec_add(pagevec, backpage); | |
288 | __pagevec_lru_add_file(pagevec); | |
289 | ||
290 | read_backing_page: | |
291 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
292 | if (ret < 0) | |
293 | goto read_error; | |
294 | ||
295 | /* set the monitor to transfer the data across */ | |
296 | monitor_backing_page: | |
297 | _debug("- monitor add"); | |
298 | ||
299 | /* install the monitor */ | |
300 | page_cache_get(monitor->netfs_page); | |
301 | page_cache_get(backpage); | |
302 | monitor->back_page = backpage; | |
303 | monitor->monitor.private = backpage; | |
304 | add_page_wait_queue(backpage, &monitor->monitor); | |
305 | monitor = NULL; | |
306 | ||
307 | /* but the page may have been read before the monitor was installed, so | |
308 | * the monitor may miss the event - so we have to ensure that we do get | |
309 | * one in such a case */ | |
310 | if (trylock_page(backpage)) { | |
311 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
312 | unlock_page(backpage); | |
313 | } | |
314 | goto success; | |
315 | ||
316 | /* if the backing page is already present, it can be in one of | |
317 | * three states: read in progress, read failed or read okay */ | |
318 | backing_page_already_present: | |
319 | _debug("- present"); | |
320 | ||
321 | if (newpage) { | |
322 | page_cache_release(newpage); | |
323 | newpage = NULL; | |
324 | } | |
325 | ||
326 | if (PageError(backpage)) | |
327 | goto io_error; | |
328 | ||
329 | if (PageUptodate(backpage)) | |
330 | goto backing_page_already_uptodate; | |
331 | ||
332 | if (!trylock_page(backpage)) | |
333 | goto monitor_backing_page; | |
334 | _debug("read %p {%lx}", backpage, backpage->flags); | |
335 | goto read_backing_page; | |
336 | ||
337 | /* the backing page is already up to date, attach the netfs | |
338 | * page to the pagecache and LRU and copy the data across */ | |
339 | backing_page_already_uptodate: | |
340 | _debug("- uptodate"); | |
341 | ||
c4d6d8db | 342 | fscache_mark_page_cached(op, netpage); |
9ae326a6 DH |
343 | |
344 | copy_highpage(netpage, backpage); | |
345 | fscache_end_io(op, netpage, 0); | |
9f10523f | 346 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
347 | |
348 | success: | |
349 | _debug("success"); | |
350 | ret = 0; | |
351 | ||
352 | out: | |
353 | if (backpage) | |
354 | page_cache_release(backpage); | |
355 | if (monitor) { | |
356 | fscache_put_retrieval(monitor->op); | |
357 | kfree(monitor); | |
358 | } | |
359 | _leave(" = %d", ret); | |
360 | return ret; | |
361 | ||
362 | read_error: | |
363 | _debug("read error %d", ret); | |
364 | if (ret == -ENOMEM) | |
365 | goto out; | |
366 | io_error: | |
367 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
9f10523f | 368 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
369 | ret = -ENOBUFS; |
370 | goto out; | |
371 | ||
372 | nomem_page: | |
373 | page_cache_release(newpage); | |
374 | nomem_monitor: | |
375 | fscache_put_retrieval(monitor->op); | |
376 | kfree(monitor); | |
377 | nomem: | |
9f10523f | 378 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
379 | _leave(" = -ENOMEM"); |
380 | return -ENOMEM; | |
381 | } | |
382 | ||
383 | /* | |
384 | * read a page from the cache or allocate a block in which to store it | |
385 | * - cache withdrawal is prevented by the caller | |
386 | * - returns -EINTR if interrupted | |
387 | * - returns -ENOMEM if ran out of memory | |
388 | * - returns -ENOBUFS if no buffers can be made available | |
389 | * - returns -ENOBUFS if page is beyond EOF | |
390 | * - if the page is backed by a block in the cache: | |
391 | * - a read will be started which will call the callback on completion | |
392 | * - 0 will be returned | |
393 | * - else if the page is unbacked: | |
394 | * - the metadata will be retained | |
395 | * - -ENODATA will be returned | |
396 | */ | |
397 | int cachefiles_read_or_alloc_page(struct fscache_retrieval *op, | |
398 | struct page *page, | |
399 | gfp_t gfp) | |
400 | { | |
401 | struct cachefiles_object *object; | |
402 | struct cachefiles_cache *cache; | |
403 | struct pagevec pagevec; | |
404 | struct inode *inode; | |
405 | sector_t block0, block; | |
406 | unsigned shift; | |
407 | int ret; | |
408 | ||
409 | object = container_of(op->op.object, | |
410 | struct cachefiles_object, fscache); | |
411 | cache = container_of(object->fscache.cache, | |
412 | struct cachefiles_cache, cache); | |
413 | ||
414 | _enter("{%p},{%lx},,,", object, page->index); | |
415 | ||
416 | if (!object->backer) | |
9f10523f | 417 | goto enobufs; |
9ae326a6 DH |
418 | |
419 | inode = object->backer->d_inode; | |
420 | ASSERT(S_ISREG(inode->i_mode)); | |
421 | ASSERT(inode->i_mapping->a_ops->bmap); | |
422 | ASSERT(inode->i_mapping->a_ops->readpages); | |
423 | ||
424 | /* calculate the shift required to use bmap */ | |
425 | if (inode->i_sb->s_blocksize > PAGE_SIZE) | |
9f10523f | 426 | goto enobufs; |
9ae326a6 DH |
427 | |
428 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; | |
429 | ||
4fbf4291 | 430 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 431 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
432 | op->op.processor = cachefiles_read_copier; |
433 | ||
434 | pagevec_init(&pagevec, 0); | |
435 | ||
436 | /* we assume the absence or presence of the first block is a good | |
437 | * enough indication for the page as a whole | |
438 | * - TODO: don't use bmap() for this as it is _not_ actually good | |
439 | * enough for this as it doesn't indicate errors, but it's all we've | |
440 | * got for the moment | |
441 | */ | |
442 | block0 = page->index; | |
443 | block0 <<= shift; | |
444 | ||
445 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0); | |
446 | _debug("%llx -> %llx", | |
447 | (unsigned long long) block0, | |
448 | (unsigned long long) block); | |
449 | ||
450 | if (block) { | |
451 | /* submit the apparently valid page to the backing fs to be | |
452 | * read from disk */ | |
453 | ret = cachefiles_read_backing_file_one(object, op, page, | |
454 | &pagevec); | |
455 | } else if (cachefiles_has_space(cache, 0, 1) == 0) { | |
456 | /* there's space in the cache we can use */ | |
c4d6d8db | 457 | fscache_mark_page_cached(op, page); |
9f10523f | 458 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
459 | ret = -ENODATA; |
460 | } else { | |
9f10523f | 461 | goto enobufs; |
9ae326a6 DH |
462 | } |
463 | ||
464 | _leave(" = %d", ret); | |
465 | return ret; | |
9f10523f DH |
466 | |
467 | enobufs: | |
468 | fscache_retrieval_complete(op, 1); | |
469 | _leave(" = -ENOBUFS"); | |
470 | return -ENOBUFS; | |
9ae326a6 DH |
471 | } |
472 | ||
473 | /* | |
474 | * read the corresponding pages to the given set from the backing file | |
475 | * - any uncertain pages are simply discarded, to be tried again another time | |
476 | */ | |
477 | static int cachefiles_read_backing_file(struct cachefiles_object *object, | |
478 | struct fscache_retrieval *op, | |
c4d6d8db | 479 | struct list_head *list) |
9ae326a6 DH |
480 | { |
481 | struct cachefiles_one_read *monitor = NULL; | |
482 | struct address_space *bmapping = object->backer->d_inode->i_mapping; | |
483 | struct pagevec lru_pvec; | |
484 | struct page *newpage = NULL, *netpage, *_n, *backpage = NULL; | |
485 | int ret = 0; | |
486 | ||
487 | _enter(""); | |
488 | ||
489 | pagevec_init(&lru_pvec, 0); | |
490 | ||
491 | list_for_each_entry_safe(netpage, _n, list, lru) { | |
492 | list_del(&netpage->lru); | |
493 | ||
494 | _debug("read back %p{%lu,%d}", | |
495 | netpage, netpage->index, page_count(netpage)); | |
496 | ||
497 | if (!monitor) { | |
5f4f9f4a | 498 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
499 | if (!monitor) |
500 | goto nomem; | |
501 | ||
502 | monitor->op = fscache_get_retrieval(op); | |
503 | init_waitqueue_func_entry(&monitor->monitor, | |
504 | cachefiles_read_waiter); | |
505 | } | |
506 | ||
507 | for (;;) { | |
508 | backpage = find_get_page(bmapping, netpage->index); | |
509 | if (backpage) | |
510 | goto backing_page_already_present; | |
511 | ||
512 | if (!newpage) { | |
5f4f9f4a DH |
513 | newpage = __page_cache_alloc(cachefiles_gfp | |
514 | __GFP_COLD); | |
9ae326a6 DH |
515 | if (!newpage) |
516 | goto nomem; | |
517 | } | |
518 | ||
519 | ret = add_to_page_cache(newpage, bmapping, | |
5f4f9f4a | 520 | netpage->index, cachefiles_gfp); |
9ae326a6 DH |
521 | if (ret == 0) |
522 | goto installed_new_backing_page; | |
523 | if (ret != -EEXIST) | |
524 | goto nomem; | |
525 | } | |
526 | ||
527 | /* we've installed a new backing page, so now we need to add it | |
528 | * to the LRU list and start it reading */ | |
529 | installed_new_backing_page: | |
530 | _debug("- new %p", newpage); | |
531 | ||
532 | backpage = newpage; | |
533 | newpage = NULL; | |
534 | ||
535 | page_cache_get(backpage); | |
536 | if (!pagevec_add(&lru_pvec, backpage)) | |
537 | __pagevec_lru_add_file(&lru_pvec); | |
538 | ||
539 | reread_backing_page: | |
540 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
541 | if (ret < 0) | |
542 | goto read_error; | |
543 | ||
544 | /* add the netfs page to the pagecache and LRU, and set the | |
545 | * monitor to transfer the data across */ | |
546 | monitor_backing_page: | |
547 | _debug("- monitor add"); | |
548 | ||
549 | ret = add_to_page_cache(netpage, op->mapping, netpage->index, | |
5f4f9f4a | 550 | cachefiles_gfp); |
9ae326a6 DH |
551 | if (ret < 0) { |
552 | if (ret == -EEXIST) { | |
553 | page_cache_release(netpage); | |
554 | continue; | |
555 | } | |
556 | goto nomem; | |
557 | } | |
558 | ||
559 | page_cache_get(netpage); | |
560 | if (!pagevec_add(&lru_pvec, netpage)) | |
561 | __pagevec_lru_add_file(&lru_pvec); | |
562 | ||
563 | /* install a monitor */ | |
564 | page_cache_get(netpage); | |
565 | monitor->netfs_page = netpage; | |
566 | ||
567 | page_cache_get(backpage); | |
568 | monitor->back_page = backpage; | |
569 | monitor->monitor.private = backpage; | |
570 | add_page_wait_queue(backpage, &monitor->monitor); | |
571 | monitor = NULL; | |
572 | ||
573 | /* but the page may have been read before the monitor was | |
574 | * installed, so the monitor may miss the event - so we have to | |
575 | * ensure that we do get one in such a case */ | |
576 | if (trylock_page(backpage)) { | |
577 | _debug("2unlock %p {%lx}", backpage, backpage->flags); | |
578 | unlock_page(backpage); | |
579 | } | |
580 | ||
581 | page_cache_release(backpage); | |
582 | backpage = NULL; | |
583 | ||
584 | page_cache_release(netpage); | |
585 | netpage = NULL; | |
586 | continue; | |
587 | ||
588 | /* if the backing page is already present, it can be in one of | |
589 | * three states: read in progress, read failed or read okay */ | |
590 | backing_page_already_present: | |
591 | _debug("- present %p", backpage); | |
592 | ||
593 | if (PageError(backpage)) | |
594 | goto io_error; | |
595 | ||
596 | if (PageUptodate(backpage)) | |
597 | goto backing_page_already_uptodate; | |
598 | ||
599 | _debug("- not ready %p{%lx}", backpage, backpage->flags); | |
600 | ||
601 | if (!trylock_page(backpage)) | |
602 | goto monitor_backing_page; | |
603 | ||
604 | if (PageError(backpage)) { | |
605 | _debug("error %lx", backpage->flags); | |
606 | unlock_page(backpage); | |
607 | goto io_error; | |
608 | } | |
609 | ||
610 | if (PageUptodate(backpage)) | |
611 | goto backing_page_already_uptodate_unlock; | |
612 | ||
613 | /* we've locked a page that's neither up to date nor erroneous, | |
614 | * so we need to attempt to read it again */ | |
615 | goto reread_backing_page; | |
616 | ||
617 | /* the backing page is already up to date, attach the netfs | |
618 | * page to the pagecache and LRU and copy the data across */ | |
619 | backing_page_already_uptodate_unlock: | |
620 | _debug("uptodate %lx", backpage->flags); | |
621 | unlock_page(backpage); | |
622 | backing_page_already_uptodate: | |
623 | _debug("- uptodate"); | |
624 | ||
625 | ret = add_to_page_cache(netpage, op->mapping, netpage->index, | |
5f4f9f4a | 626 | cachefiles_gfp); |
9ae326a6 DH |
627 | if (ret < 0) { |
628 | if (ret == -EEXIST) { | |
629 | page_cache_release(netpage); | |
630 | continue; | |
631 | } | |
632 | goto nomem; | |
633 | } | |
634 | ||
635 | copy_highpage(netpage, backpage); | |
636 | ||
637 | page_cache_release(backpage); | |
638 | backpage = NULL; | |
639 | ||
c4d6d8db | 640 | fscache_mark_page_cached(op, netpage); |
9ae326a6 DH |
641 | |
642 | page_cache_get(netpage); | |
643 | if (!pagevec_add(&lru_pvec, netpage)) | |
644 | __pagevec_lru_add_file(&lru_pvec); | |
645 | ||
c4d6d8db | 646 | /* the netpage is unlocked and marked up to date here */ |
9ae326a6 | 647 | fscache_end_io(op, netpage, 0); |
9f10523f | 648 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
649 | page_cache_release(netpage); |
650 | netpage = NULL; | |
651 | continue; | |
652 | } | |
653 | ||
654 | netpage = NULL; | |
655 | ||
656 | _debug("out"); | |
657 | ||
658 | out: | |
659 | /* tidy up */ | |
660 | pagevec_lru_add_file(&lru_pvec); | |
661 | ||
662 | if (newpage) | |
663 | page_cache_release(newpage); | |
664 | if (netpage) | |
665 | page_cache_release(netpage); | |
666 | if (backpage) | |
667 | page_cache_release(backpage); | |
668 | if (monitor) { | |
669 | fscache_put_retrieval(op); | |
670 | kfree(monitor); | |
671 | } | |
672 | ||
673 | list_for_each_entry_safe(netpage, _n, list, lru) { | |
674 | list_del(&netpage->lru); | |
675 | page_cache_release(netpage); | |
9f10523f | 676 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
677 | } |
678 | ||
679 | _leave(" = %d", ret); | |
680 | return ret; | |
681 | ||
682 | nomem: | |
683 | _debug("nomem"); | |
684 | ret = -ENOMEM; | |
685 | goto out; | |
686 | ||
687 | read_error: | |
688 | _debug("read error %d", ret); | |
689 | if (ret == -ENOMEM) | |
690 | goto out; | |
691 | io_error: | |
692 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
693 | ret = -ENOBUFS; | |
694 | goto out; | |
695 | } | |
696 | ||
697 | /* | |
698 | * read a list of pages from the cache or allocate blocks in which to store | |
699 | * them | |
700 | */ | |
701 | int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op, | |
702 | struct list_head *pages, | |
703 | unsigned *nr_pages, | |
704 | gfp_t gfp) | |
705 | { | |
706 | struct cachefiles_object *object; | |
707 | struct cachefiles_cache *cache; | |
708 | struct list_head backpages; | |
709 | struct pagevec pagevec; | |
710 | struct inode *inode; | |
711 | struct page *page, *_n; | |
712 | unsigned shift, nrbackpages; | |
713 | int ret, ret2, space; | |
714 | ||
715 | object = container_of(op->op.object, | |
716 | struct cachefiles_object, fscache); | |
717 | cache = container_of(object->fscache.cache, | |
718 | struct cachefiles_cache, cache); | |
719 | ||
720 | _enter("{OBJ%x,%d},,%d,,", | |
721 | object->fscache.debug_id, atomic_read(&op->op.usage), | |
722 | *nr_pages); | |
723 | ||
724 | if (!object->backer) | |
9f10523f | 725 | goto all_enobufs; |
9ae326a6 DH |
726 | |
727 | space = 1; | |
728 | if (cachefiles_has_space(cache, 0, *nr_pages) < 0) | |
729 | space = 0; | |
730 | ||
731 | inode = object->backer->d_inode; | |
732 | ASSERT(S_ISREG(inode->i_mode)); | |
733 | ASSERT(inode->i_mapping->a_ops->bmap); | |
734 | ASSERT(inode->i_mapping->a_ops->readpages); | |
735 | ||
736 | /* calculate the shift required to use bmap */ | |
737 | if (inode->i_sb->s_blocksize > PAGE_SIZE) | |
9f10523f | 738 | goto all_enobufs; |
9ae326a6 DH |
739 | |
740 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; | |
741 | ||
742 | pagevec_init(&pagevec, 0); | |
743 | ||
4fbf4291 | 744 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 745 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
746 | op->op.processor = cachefiles_read_copier; |
747 | ||
748 | INIT_LIST_HEAD(&backpages); | |
749 | nrbackpages = 0; | |
750 | ||
751 | ret = space ? -ENODATA : -ENOBUFS; | |
752 | list_for_each_entry_safe(page, _n, pages, lru) { | |
753 | sector_t block0, block; | |
754 | ||
755 | /* we assume the absence or presence of the first block is a | |
756 | * good enough indication for the page as a whole | |
757 | * - TODO: don't use bmap() for this as it is _not_ actually | |
758 | * good enough for this as it doesn't indicate errors, but | |
759 | * it's all we've got for the moment | |
760 | */ | |
761 | block0 = page->index; | |
762 | block0 <<= shift; | |
763 | ||
764 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, | |
765 | block0); | |
766 | _debug("%llx -> %llx", | |
767 | (unsigned long long) block0, | |
768 | (unsigned long long) block); | |
769 | ||
770 | if (block) { | |
771 | /* we have data - add it to the list to give to the | |
772 | * backing fs */ | |
773 | list_move(&page->lru, &backpages); | |
774 | (*nr_pages)--; | |
775 | nrbackpages++; | |
776 | } else if (space && pagevec_add(&pagevec, page) == 0) { | |
777 | fscache_mark_pages_cached(op, &pagevec); | |
9f10523f | 778 | fscache_retrieval_complete(op, 1); |
9ae326a6 | 779 | ret = -ENODATA; |
9f10523f DH |
780 | } else { |
781 | fscache_retrieval_complete(op, 1); | |
9ae326a6 DH |
782 | } |
783 | } | |
784 | ||
785 | if (pagevec_count(&pagevec) > 0) | |
786 | fscache_mark_pages_cached(op, &pagevec); | |
787 | ||
788 | if (list_empty(pages)) | |
789 | ret = 0; | |
790 | ||
791 | /* submit the apparently valid pages to the backing fs to be read from | |
792 | * disk */ | |
793 | if (nrbackpages > 0) { | |
c4d6d8db | 794 | ret2 = cachefiles_read_backing_file(object, op, &backpages); |
9ae326a6 DH |
795 | if (ret2 == -ENOMEM || ret2 == -EINTR) |
796 | ret = ret2; | |
797 | } | |
798 | ||
9ae326a6 DH |
799 | _leave(" = %d [nr=%u%s]", |
800 | ret, *nr_pages, list_empty(pages) ? " empty" : ""); | |
801 | return ret; | |
9f10523f DH |
802 | |
803 | all_enobufs: | |
804 | fscache_retrieval_complete(op, *nr_pages); | |
805 | return -ENOBUFS; | |
9ae326a6 DH |
806 | } |
807 | ||
808 | /* | |
809 | * allocate a block in the cache in which to store a page | |
810 | * - cache withdrawal is prevented by the caller | |
811 | * - returns -EINTR if interrupted | |
812 | * - returns -ENOMEM if ran out of memory | |
813 | * - returns -ENOBUFS if no buffers can be made available | |
814 | * - returns -ENOBUFS if page is beyond EOF | |
815 | * - otherwise: | |
816 | * - the metadata will be retained | |
817 | * - 0 will be returned | |
818 | */ | |
819 | int cachefiles_allocate_page(struct fscache_retrieval *op, | |
820 | struct page *page, | |
821 | gfp_t gfp) | |
822 | { | |
823 | struct cachefiles_object *object; | |
824 | struct cachefiles_cache *cache; | |
9ae326a6 DH |
825 | int ret; |
826 | ||
827 | object = container_of(op->op.object, | |
828 | struct cachefiles_object, fscache); | |
829 | cache = container_of(object->fscache.cache, | |
830 | struct cachefiles_cache, cache); | |
831 | ||
832 | _enter("%p,{%lx},", object, page->index); | |
833 | ||
834 | ret = cachefiles_has_space(cache, 0, 1); | |
c4d6d8db DH |
835 | if (ret == 0) |
836 | fscache_mark_page_cached(op, page); | |
837 | else | |
9ae326a6 | 838 | ret = -ENOBUFS; |
9ae326a6 | 839 | |
9f10523f | 840 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
841 | _leave(" = %d", ret); |
842 | return ret; | |
843 | } | |
844 | ||
845 | /* | |
846 | * allocate blocks in the cache in which to store a set of pages | |
847 | * - cache withdrawal is prevented by the caller | |
848 | * - returns -EINTR if interrupted | |
849 | * - returns -ENOMEM if ran out of memory | |
850 | * - returns -ENOBUFS if some buffers couldn't be made available | |
851 | * - returns -ENOBUFS if some pages are beyond EOF | |
852 | * - otherwise: | |
853 | * - -ENODATA will be returned | |
854 | * - metadata will be retained for any page marked | |
855 | */ | |
856 | int cachefiles_allocate_pages(struct fscache_retrieval *op, | |
857 | struct list_head *pages, | |
858 | unsigned *nr_pages, | |
859 | gfp_t gfp) | |
860 | { | |
861 | struct cachefiles_object *object; | |
862 | struct cachefiles_cache *cache; | |
863 | struct pagevec pagevec; | |
864 | struct page *page; | |
865 | int ret; | |
866 | ||
867 | object = container_of(op->op.object, | |
868 | struct cachefiles_object, fscache); | |
869 | cache = container_of(object->fscache.cache, | |
870 | struct cachefiles_cache, cache); | |
871 | ||
872 | _enter("%p,,,%d,", object, *nr_pages); | |
873 | ||
874 | ret = cachefiles_has_space(cache, 0, *nr_pages); | |
875 | if (ret == 0) { | |
876 | pagevec_init(&pagevec, 0); | |
877 | ||
878 | list_for_each_entry(page, pages, lru) { | |
879 | if (pagevec_add(&pagevec, page) == 0) | |
880 | fscache_mark_pages_cached(op, &pagevec); | |
881 | } | |
882 | ||
883 | if (pagevec_count(&pagevec) > 0) | |
884 | fscache_mark_pages_cached(op, &pagevec); | |
885 | ret = -ENODATA; | |
886 | } else { | |
887 | ret = -ENOBUFS; | |
888 | } | |
889 | ||
9f10523f | 890 | fscache_retrieval_complete(op, *nr_pages); |
9ae326a6 DH |
891 | _leave(" = %d", ret); |
892 | return ret; | |
893 | } | |
894 | ||
895 | /* | |
896 | * request a page be stored in the cache | |
897 | * - cache withdrawal is prevented by the caller | |
898 | * - this request may be ignored if there's no cache block available, in which | |
899 | * case -ENOBUFS will be returned | |
900 | * - if the op is in progress, 0 will be returned | |
901 | */ | |
902 | int cachefiles_write_page(struct fscache_storage *op, struct page *page) | |
903 | { | |
904 | struct cachefiles_object *object; | |
905 | struct cachefiles_cache *cache; | |
906 | mm_segment_t old_fs; | |
907 | struct file *file; | |
765927b2 | 908 | struct path path; |
a17754fb DH |
909 | loff_t pos, eof; |
910 | size_t len; | |
9ae326a6 DH |
911 | void *data; |
912 | int ret; | |
913 | ||
914 | ASSERT(op != NULL); | |
915 | ASSERT(page != NULL); | |
916 | ||
917 | object = container_of(op->op.object, | |
918 | struct cachefiles_object, fscache); | |
919 | ||
920 | _enter("%p,%p{%lx},,,", object, page, page->index); | |
921 | ||
922 | if (!object->backer) { | |
923 | _leave(" = -ENOBUFS"); | |
924 | return -ENOBUFS; | |
925 | } | |
926 | ||
927 | ASSERT(S_ISREG(object->backer->d_inode->i_mode)); | |
928 | ||
929 | cache = container_of(object->fscache.cache, | |
930 | struct cachefiles_cache, cache); | |
931 | ||
932 | /* write the page to the backing filesystem and let it store it in its | |
933 | * own time */ | |
765927b2 AV |
934 | path.mnt = cache->mnt; |
935 | path.dentry = object->backer; | |
98c350cd | 936 | file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred); |
9ae326a6 DH |
937 | if (IS_ERR(file)) { |
938 | ret = PTR_ERR(file); | |
939 | } else { | |
940 | ret = -EIO; | |
941 | if (file->f_op->write) { | |
942 | pos = (loff_t) page->index << PAGE_SHIFT; | |
a17754fb DH |
943 | |
944 | /* we mustn't write more data than we have, so we have | |
945 | * to beware of a partial page at EOF */ | |
946 | eof = object->fscache.store_limit_l; | |
947 | len = PAGE_SIZE; | |
948 | if (eof & ~PAGE_MASK) { | |
949 | ASSERTCMP(pos, <, eof); | |
950 | if (eof - pos < PAGE_SIZE) { | |
951 | _debug("cut short %llx to %llx", | |
952 | pos, eof); | |
953 | len = eof - pos; | |
954 | ASSERTCMP(pos + len, ==, eof); | |
955 | } | |
956 | } | |
957 | ||
9ae326a6 DH |
958 | data = kmap(page); |
959 | old_fs = get_fs(); | |
960 | set_fs(KERNEL_DS); | |
961 | ret = file->f_op->write( | |
a17754fb | 962 | file, (const void __user *) data, len, &pos); |
9ae326a6 DH |
963 | set_fs(old_fs); |
964 | kunmap(page); | |
a17754fb | 965 | if (ret != len) |
9ae326a6 DH |
966 | ret = -EIO; |
967 | } | |
968 | fput(file); | |
969 | } | |
970 | ||
971 | if (ret < 0) { | |
972 | if (ret == -EIO) | |
973 | cachefiles_io_error_obj( | |
974 | object, "Write page to backing file failed"); | |
975 | ret = -ENOBUFS; | |
976 | } | |
977 | ||
978 | _leave(" = %d", ret); | |
979 | return ret; | |
980 | } | |
981 | ||
982 | /* | |
983 | * detach a backing block from a page | |
984 | * - cache withdrawal is prevented by the caller | |
985 | */ | |
986 | void cachefiles_uncache_page(struct fscache_object *_object, struct page *page) | |
987 | { | |
988 | struct cachefiles_object *object; | |
989 | struct cachefiles_cache *cache; | |
990 | ||
991 | object = container_of(_object, struct cachefiles_object, fscache); | |
992 | cache = container_of(object->fscache.cache, | |
993 | struct cachefiles_cache, cache); | |
994 | ||
995 | _enter("%p,{%lu}", object, page->index); | |
996 | ||
997 | spin_unlock(&object->fscache.cookie->lock); | |
998 | } |