]>
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 | 14 | #include <linux/file.h> |
a0b8cab3 | 15 | #include <linux/swap.h> |
9ae326a6 DH |
16 | #include "internal.h" |
17 | ||
18 | /* | |
19 | * detect wake up events generated by the unlocking of pages in which we're | |
20 | * interested | |
21 | * - we use this to detect read completion of backing pages | |
22 | * - the caller holds the waitqueue lock | |
23 | */ | |
ac6424b9 | 24 | static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode, |
9ae326a6 DH |
25 | int sync, void *_key) |
26 | { | |
27 | struct cachefiles_one_read *monitor = | |
28 | container_of(wait, struct cachefiles_one_read, monitor); | |
29 | struct cachefiles_object *object; | |
30 | struct wait_bit_key *key = _key; | |
31 | struct page *page = wait->private; | |
32 | ||
33 | ASSERT(key); | |
34 | ||
35 | _enter("{%lu},%u,%d,{%p,%u}", | |
36 | monitor->netfs_page->index, mode, sync, | |
37 | key->flags, key->bit_nr); | |
38 | ||
39 | if (key->flags != &page->flags || | |
40 | key->bit_nr != PG_locked) | |
41 | return 0; | |
42 | ||
43 | _debug("--- monitor %p %lx ---", page, page->flags); | |
44 | ||
5e929b33 DH |
45 | if (!PageUptodate(page) && !PageError(page)) { |
46 | /* unlocked, not uptodate and not erronous? */ | |
47 | _debug("page probably truncated"); | |
48 | } | |
9ae326a6 DH |
49 | |
50 | /* remove from the waitqueue */ | |
2055da97 | 51 | list_del(&wait->entry); |
9ae326a6 DH |
52 | |
53 | /* move onto the action list and queue for FS-Cache thread pool */ | |
54 | ASSERT(monitor->op); | |
55 | ||
56 | object = container_of(monitor->op->op.object, | |
57 | struct cachefiles_object, fscache); | |
58 | ||
59 | spin_lock(&object->work_lock); | |
60 | list_add_tail(&monitor->op_link, &monitor->op->to_do); | |
61 | spin_unlock(&object->work_lock); | |
62 | ||
63 | fscache_enqueue_retrieval(monitor->op); | |
64 | return 0; | |
65 | } | |
66 | ||
5e929b33 DH |
67 | /* |
68 | * handle a probably truncated page | |
69 | * - check to see if the page is still relevant and reissue the read if | |
70 | * possible | |
71 | * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we | |
72 | * must wait again and 0 if successful | |
73 | */ | |
74 | static int cachefiles_read_reissue(struct cachefiles_object *object, | |
75 | struct cachefiles_one_read *monitor) | |
76 | { | |
466b77bc | 77 | struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping; |
5e929b33 DH |
78 | struct page *backpage = monitor->back_page, *backpage2; |
79 | int ret; | |
80 | ||
37491a13 | 81 | _enter("{ino=%lx},{%lx,%lx}", |
466b77bc | 82 | d_backing_inode(object->backer)->i_ino, |
5e929b33 DH |
83 | backpage->index, backpage->flags); |
84 | ||
85 | /* skip if the page was truncated away completely */ | |
86 | if (backpage->mapping != bmapping) { | |
37491a13 | 87 | _leave(" = -ENODATA [mapping]"); |
5e929b33 DH |
88 | return -ENODATA; |
89 | } | |
90 | ||
91 | backpage2 = find_get_page(bmapping, backpage->index); | |
92 | if (!backpage2) { | |
37491a13 | 93 | _leave(" = -ENODATA [gone]"); |
5e929b33 DH |
94 | return -ENODATA; |
95 | } | |
96 | ||
97 | if (backpage != backpage2) { | |
98 | put_page(backpage2); | |
37491a13 | 99 | _leave(" = -ENODATA [different]"); |
5e929b33 DH |
100 | return -ENODATA; |
101 | } | |
102 | ||
103 | /* the page is still there and we already have a ref on it, so we don't | |
104 | * need a second */ | |
105 | put_page(backpage2); | |
106 | ||
107 | INIT_LIST_HEAD(&monitor->op_link); | |
108 | add_page_wait_queue(backpage, &monitor->monitor); | |
109 | ||
110 | if (trylock_page(backpage)) { | |
111 | ret = -EIO; | |
112 | if (PageError(backpage)) | |
113 | goto unlock_discard; | |
114 | ret = 0; | |
115 | if (PageUptodate(backpage)) | |
116 | goto unlock_discard; | |
117 | ||
37491a13 | 118 | _debug("reissue read"); |
5e929b33 DH |
119 | ret = bmapping->a_ops->readpage(NULL, backpage); |
120 | if (ret < 0) | |
121 | goto unlock_discard; | |
122 | } | |
123 | ||
124 | /* but the page may have been read before the monitor was installed, so | |
125 | * the monitor may miss the event - so we have to ensure that we do get | |
126 | * one in such a case */ | |
127 | if (trylock_page(backpage)) { | |
128 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
129 | unlock_page(backpage); | |
130 | } | |
131 | ||
132 | /* it'll reappear on the todo list */ | |
37491a13 | 133 | _leave(" = -EINPROGRESS"); |
5e929b33 DH |
134 | return -EINPROGRESS; |
135 | ||
136 | unlock_discard: | |
137 | unlock_page(backpage); | |
138 | spin_lock_irq(&object->work_lock); | |
139 | list_del(&monitor->op_link); | |
140 | spin_unlock_irq(&object->work_lock); | |
37491a13 | 141 | _leave(" = %d", ret); |
5e929b33 DH |
142 | return ret; |
143 | } | |
144 | ||
9ae326a6 DH |
145 | /* |
146 | * copy data from backing pages to netfs pages to complete a read operation | |
147 | * - driven by FS-Cache's thread pool | |
148 | */ | |
149 | static void cachefiles_read_copier(struct fscache_operation *_op) | |
150 | { | |
151 | struct cachefiles_one_read *monitor; | |
152 | struct cachefiles_object *object; | |
153 | struct fscache_retrieval *op; | |
9ae326a6 DH |
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 | ||
466b77bc | 160 | _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino); |
9ae326a6 | 161 | |
9ae326a6 DH |
162 | max = 8; |
163 | spin_lock_irq(&object->work_lock); | |
164 | ||
165 | while (!list_empty(&op->to_do)) { | |
166 | monitor = list_entry(op->to_do.next, | |
167 | struct cachefiles_one_read, op_link); | |
168 | list_del(&monitor->op_link); | |
169 | ||
170 | spin_unlock_irq(&object->work_lock); | |
171 | ||
172 | _debug("- copy {%lu}", monitor->back_page->index); | |
173 | ||
5e929b33 | 174 | recheck: |
9dc8d9bf DH |
175 | if (test_bit(FSCACHE_COOKIE_INVALIDATING, |
176 | &object->fscache.cookie->flags)) { | |
177 | error = -ESTALE; | |
178 | } else if (PageUptodate(monitor->back_page)) { | |
9ae326a6 | 179 | copy_highpage(monitor->netfs_page, monitor->back_page); |
c4d6d8db DH |
180 | fscache_mark_page_cached(monitor->op, |
181 | monitor->netfs_page); | |
9ae326a6 | 182 | error = 0; |
5e929b33 DH |
183 | } else if (!PageError(monitor->back_page)) { |
184 | /* the page has probably been truncated */ | |
185 | error = cachefiles_read_reissue(object, monitor); | |
186 | if (error == -EINPROGRESS) | |
187 | goto next; | |
188 | goto recheck; | |
189 | } else { | |
9ae326a6 DH |
190 | cachefiles_io_error_obj( |
191 | object, | |
192 | "Readpage failed on backing file %lx", | |
193 | (unsigned long) monitor->back_page->flags); | |
5e929b33 DH |
194 | error = -EIO; |
195 | } | |
9ae326a6 | 196 | |
09cbfeaf | 197 | put_page(monitor->back_page); |
9ae326a6 DH |
198 | |
199 | fscache_end_io(op, monitor->netfs_page, error); | |
09cbfeaf | 200 | put_page(monitor->netfs_page); |
9f10523f | 201 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
202 | fscache_put_retrieval(op); |
203 | kfree(monitor); | |
204 | ||
5e929b33 | 205 | next: |
9ae326a6 DH |
206 | /* let the thread pool have some air occasionally */ |
207 | max--; | |
208 | if (max < 0 || need_resched()) { | |
209 | if (!list_empty(&op->to_do)) | |
210 | fscache_enqueue_retrieval(op); | |
211 | _leave(" [maxed out]"); | |
212 | return; | |
213 | } | |
214 | ||
215 | spin_lock_irq(&object->work_lock); | |
216 | } | |
217 | ||
218 | spin_unlock_irq(&object->work_lock); | |
219 | _leave(""); | |
220 | } | |
221 | ||
222 | /* | |
223 | * read the corresponding page to the given set from the backing file | |
224 | * - an uncertain page is simply discarded, to be tried again another time | |
225 | */ | |
226 | static int cachefiles_read_backing_file_one(struct cachefiles_object *object, | |
227 | struct fscache_retrieval *op, | |
a0b8cab3 | 228 | struct page *netpage) |
9ae326a6 DH |
229 | { |
230 | struct cachefiles_one_read *monitor; | |
231 | struct address_space *bmapping; | |
232 | struct page *newpage, *backpage; | |
233 | int ret; | |
234 | ||
235 | _enter(""); | |
236 | ||
9ae326a6 DH |
237 | _debug("read back %p{%lu,%d}", |
238 | netpage, netpage->index, page_count(netpage)); | |
239 | ||
5f4f9f4a | 240 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
241 | if (!monitor) |
242 | goto nomem; | |
243 | ||
244 | monitor->netfs_page = netpage; | |
245 | monitor->op = fscache_get_retrieval(op); | |
246 | ||
247 | init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter); | |
248 | ||
249 | /* attempt to get hold of the backing page */ | |
466b77bc | 250 | bmapping = d_backing_inode(object->backer)->i_mapping; |
9ae326a6 DH |
251 | newpage = NULL; |
252 | ||
253 | for (;;) { | |
254 | backpage = find_get_page(bmapping, netpage->index); | |
255 | if (backpage) | |
256 | goto backing_page_already_present; | |
257 | ||
258 | if (!newpage) { | |
453f85d4 | 259 | newpage = __page_cache_alloc(cachefiles_gfp); |
9ae326a6 DH |
260 | if (!newpage) |
261 | goto nomem_monitor; | |
262 | } | |
263 | ||
55881bc7 JW |
264 | ret = add_to_page_cache_lru(newpage, bmapping, |
265 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
266 | if (ret == 0) |
267 | goto installed_new_backing_page; | |
268 | if (ret != -EEXIST) | |
269 | goto nomem_page; | |
270 | } | |
271 | ||
55881bc7 JW |
272 | /* we've installed a new backing page, so now we need to start |
273 | * it reading */ | |
9ae326a6 DH |
274 | installed_new_backing_page: |
275 | _debug("- new %p", newpage); | |
276 | ||
277 | backpage = newpage; | |
278 | newpage = NULL; | |
279 | ||
9ae326a6 DH |
280 | read_backing_page: |
281 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
282 | if (ret < 0) | |
283 | goto read_error; | |
284 | ||
285 | /* set the monitor to transfer the data across */ | |
286 | monitor_backing_page: | |
287 | _debug("- monitor add"); | |
288 | ||
289 | /* install the monitor */ | |
09cbfeaf KS |
290 | get_page(monitor->netfs_page); |
291 | get_page(backpage); | |
9ae326a6 DH |
292 | monitor->back_page = backpage; |
293 | monitor->monitor.private = backpage; | |
294 | add_page_wait_queue(backpage, &monitor->monitor); | |
295 | monitor = NULL; | |
296 | ||
297 | /* but the page may have been read before the monitor was installed, so | |
298 | * the monitor may miss the event - so we have to ensure that we do get | |
299 | * one in such a case */ | |
300 | if (trylock_page(backpage)) { | |
301 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
302 | unlock_page(backpage); | |
303 | } | |
304 | goto success; | |
305 | ||
306 | /* if the backing page is already present, it can be in one of | |
307 | * three states: read in progress, read failed or read okay */ | |
308 | backing_page_already_present: | |
309 | _debug("- present"); | |
310 | ||
311 | if (newpage) { | |
09cbfeaf | 312 | put_page(newpage); |
9ae326a6 DH |
313 | newpage = NULL; |
314 | } | |
315 | ||
316 | if (PageError(backpage)) | |
317 | goto io_error; | |
318 | ||
319 | if (PageUptodate(backpage)) | |
320 | goto backing_page_already_uptodate; | |
321 | ||
322 | if (!trylock_page(backpage)) | |
323 | goto monitor_backing_page; | |
324 | _debug("read %p {%lx}", backpage, backpage->flags); | |
325 | goto read_backing_page; | |
326 | ||
327 | /* the backing page is already up to date, attach the netfs | |
328 | * page to the pagecache and LRU and copy the data across */ | |
329 | backing_page_already_uptodate: | |
330 | _debug("- uptodate"); | |
331 | ||
c4d6d8db | 332 | fscache_mark_page_cached(op, netpage); |
9ae326a6 DH |
333 | |
334 | copy_highpage(netpage, backpage); | |
335 | fscache_end_io(op, netpage, 0); | |
9f10523f | 336 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
337 | |
338 | success: | |
339 | _debug("success"); | |
340 | ret = 0; | |
341 | ||
342 | out: | |
343 | if (backpage) | |
09cbfeaf | 344 | put_page(backpage); |
9ae326a6 DH |
345 | if (monitor) { |
346 | fscache_put_retrieval(monitor->op); | |
347 | kfree(monitor); | |
348 | } | |
349 | _leave(" = %d", ret); | |
350 | return ret; | |
351 | ||
352 | read_error: | |
353 | _debug("read error %d", ret); | |
b4cf1e08 DH |
354 | if (ret == -ENOMEM) { |
355 | fscache_retrieval_complete(op, 1); | |
9ae326a6 | 356 | goto out; |
b4cf1e08 | 357 | } |
9ae326a6 DH |
358 | io_error: |
359 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
9f10523f | 360 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
361 | ret = -ENOBUFS; |
362 | goto out; | |
363 | ||
364 | nomem_page: | |
09cbfeaf | 365 | put_page(newpage); |
9ae326a6 DH |
366 | nomem_monitor: |
367 | fscache_put_retrieval(monitor->op); | |
368 | kfree(monitor); | |
369 | nomem: | |
9f10523f | 370 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
371 | _leave(" = -ENOMEM"); |
372 | return -ENOMEM; | |
373 | } | |
374 | ||
375 | /* | |
376 | * read a page from the cache or allocate a block in which to store it | |
377 | * - cache withdrawal is prevented by the caller | |
378 | * - returns -EINTR if interrupted | |
379 | * - returns -ENOMEM if ran out of memory | |
380 | * - returns -ENOBUFS if no buffers can be made available | |
381 | * - returns -ENOBUFS if page is beyond EOF | |
382 | * - if the page is backed by a block in the cache: | |
383 | * - a read will be started which will call the callback on completion | |
384 | * - 0 will be returned | |
385 | * - else if the page is unbacked: | |
386 | * - the metadata will be retained | |
387 | * - -ENODATA will be returned | |
388 | */ | |
389 | int cachefiles_read_or_alloc_page(struct fscache_retrieval *op, | |
390 | struct page *page, | |
391 | gfp_t gfp) | |
392 | { | |
393 | struct cachefiles_object *object; | |
394 | struct cachefiles_cache *cache; | |
9ae326a6 DH |
395 | struct inode *inode; |
396 | sector_t block0, block; | |
397 | unsigned shift; | |
398 | int ret; | |
399 | ||
400 | object = container_of(op->op.object, | |
401 | struct cachefiles_object, fscache); | |
402 | cache = container_of(object->fscache.cache, | |
403 | struct cachefiles_cache, cache); | |
404 | ||
405 | _enter("{%p},{%lx},,,", object, page->index); | |
406 | ||
407 | if (!object->backer) | |
9f10523f | 408 | goto enobufs; |
9ae326a6 | 409 | |
466b77bc | 410 | inode = d_backing_inode(object->backer); |
9ae326a6 DH |
411 | ASSERT(S_ISREG(inode->i_mode)); |
412 | ASSERT(inode->i_mapping->a_ops->bmap); | |
413 | ASSERT(inode->i_mapping->a_ops->readpages); | |
414 | ||
415 | /* calculate the shift required to use bmap */ | |
9ae326a6 DH |
416 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
417 | ||
4fbf4291 | 418 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 419 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
420 | op->op.processor = cachefiles_read_copier; |
421 | ||
9ae326a6 DH |
422 | /* we assume the absence or presence of the first block is a good |
423 | * enough indication for the page as a whole | |
424 | * - TODO: don't use bmap() for this as it is _not_ actually good | |
425 | * enough for this as it doesn't indicate errors, but it's all we've | |
426 | * got for the moment | |
427 | */ | |
428 | block0 = page->index; | |
429 | block0 <<= shift; | |
430 | ||
431 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0); | |
432 | _debug("%llx -> %llx", | |
433 | (unsigned long long) block0, | |
434 | (unsigned long long) block); | |
435 | ||
436 | if (block) { | |
437 | /* submit the apparently valid page to the backing fs to be | |
438 | * read from disk */ | |
a0b8cab3 | 439 | ret = cachefiles_read_backing_file_one(object, op, page); |
9ae326a6 DH |
440 | } else if (cachefiles_has_space(cache, 0, 1) == 0) { |
441 | /* there's space in the cache we can use */ | |
c4d6d8db | 442 | fscache_mark_page_cached(op, page); |
9f10523f | 443 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
444 | ret = -ENODATA; |
445 | } else { | |
9f10523f | 446 | goto enobufs; |
9ae326a6 DH |
447 | } |
448 | ||
449 | _leave(" = %d", ret); | |
450 | return ret; | |
9f10523f DH |
451 | |
452 | enobufs: | |
453 | fscache_retrieval_complete(op, 1); | |
454 | _leave(" = -ENOBUFS"); | |
455 | return -ENOBUFS; | |
9ae326a6 DH |
456 | } |
457 | ||
458 | /* | |
459 | * read the corresponding pages to the given set from the backing file | |
460 | * - any uncertain pages are simply discarded, to be tried again another time | |
461 | */ | |
462 | static int cachefiles_read_backing_file(struct cachefiles_object *object, | |
463 | struct fscache_retrieval *op, | |
c4d6d8db | 464 | struct list_head *list) |
9ae326a6 DH |
465 | { |
466 | struct cachefiles_one_read *monitor = NULL; | |
466b77bc | 467 | struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping; |
9ae326a6 DH |
468 | struct page *newpage = NULL, *netpage, *_n, *backpage = NULL; |
469 | int ret = 0; | |
470 | ||
471 | _enter(""); | |
472 | ||
9ae326a6 DH |
473 | list_for_each_entry_safe(netpage, _n, list, lru) { |
474 | list_del(&netpage->lru); | |
475 | ||
476 | _debug("read back %p{%lu,%d}", | |
477 | netpage, netpage->index, page_count(netpage)); | |
478 | ||
479 | if (!monitor) { | |
5f4f9f4a | 480 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
481 | if (!monitor) |
482 | goto nomem; | |
483 | ||
484 | monitor->op = fscache_get_retrieval(op); | |
485 | init_waitqueue_func_entry(&monitor->monitor, | |
486 | cachefiles_read_waiter); | |
487 | } | |
488 | ||
489 | for (;;) { | |
490 | backpage = find_get_page(bmapping, netpage->index); | |
491 | if (backpage) | |
492 | goto backing_page_already_present; | |
493 | ||
494 | if (!newpage) { | |
453f85d4 | 495 | newpage = __page_cache_alloc(cachefiles_gfp); |
9ae326a6 DH |
496 | if (!newpage) |
497 | goto nomem; | |
498 | } | |
499 | ||
55881bc7 JW |
500 | ret = add_to_page_cache_lru(newpage, bmapping, |
501 | netpage->index, | |
502 | cachefiles_gfp); | |
9ae326a6 DH |
503 | if (ret == 0) |
504 | goto installed_new_backing_page; | |
505 | if (ret != -EEXIST) | |
506 | goto nomem; | |
507 | } | |
508 | ||
55881bc7 JW |
509 | /* we've installed a new backing page, so now we need |
510 | * to start it reading */ | |
9ae326a6 DH |
511 | installed_new_backing_page: |
512 | _debug("- new %p", newpage); | |
513 | ||
514 | backpage = newpage; | |
515 | newpage = NULL; | |
516 | ||
9ae326a6 DH |
517 | reread_backing_page: |
518 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
519 | if (ret < 0) | |
520 | goto read_error; | |
521 | ||
522 | /* add the netfs page to the pagecache and LRU, and set the | |
523 | * monitor to transfer the data across */ | |
524 | monitor_backing_page: | |
525 | _debug("- monitor add"); | |
526 | ||
55881bc7 JW |
527 | ret = add_to_page_cache_lru(netpage, op->mapping, |
528 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
529 | if (ret < 0) { |
530 | if (ret == -EEXIST) { | |
09cbfeaf | 531 | put_page(netpage); |
b4cf1e08 | 532 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
533 | continue; |
534 | } | |
535 | goto nomem; | |
536 | } | |
537 | ||
9ae326a6 | 538 | /* install a monitor */ |
09cbfeaf | 539 | get_page(netpage); |
9ae326a6 DH |
540 | monitor->netfs_page = netpage; |
541 | ||
09cbfeaf | 542 | get_page(backpage); |
9ae326a6 DH |
543 | monitor->back_page = backpage; |
544 | monitor->monitor.private = backpage; | |
545 | add_page_wait_queue(backpage, &monitor->monitor); | |
546 | monitor = NULL; | |
547 | ||
548 | /* but the page may have been read before the monitor was | |
549 | * installed, so the monitor may miss the event - so we have to | |
550 | * ensure that we do get one in such a case */ | |
551 | if (trylock_page(backpage)) { | |
552 | _debug("2unlock %p {%lx}", backpage, backpage->flags); | |
553 | unlock_page(backpage); | |
554 | } | |
555 | ||
09cbfeaf | 556 | put_page(backpage); |
9ae326a6 DH |
557 | backpage = NULL; |
558 | ||
09cbfeaf | 559 | put_page(netpage); |
9ae326a6 DH |
560 | netpage = NULL; |
561 | continue; | |
562 | ||
563 | /* if the backing page is already present, it can be in one of | |
564 | * three states: read in progress, read failed or read okay */ | |
565 | backing_page_already_present: | |
566 | _debug("- present %p", backpage); | |
567 | ||
568 | if (PageError(backpage)) | |
569 | goto io_error; | |
570 | ||
571 | if (PageUptodate(backpage)) | |
572 | goto backing_page_already_uptodate; | |
573 | ||
574 | _debug("- not ready %p{%lx}", backpage, backpage->flags); | |
575 | ||
576 | if (!trylock_page(backpage)) | |
577 | goto monitor_backing_page; | |
578 | ||
579 | if (PageError(backpage)) { | |
580 | _debug("error %lx", backpage->flags); | |
581 | unlock_page(backpage); | |
582 | goto io_error; | |
583 | } | |
584 | ||
585 | if (PageUptodate(backpage)) | |
586 | goto backing_page_already_uptodate_unlock; | |
587 | ||
588 | /* we've locked a page that's neither up to date nor erroneous, | |
589 | * so we need to attempt to read it again */ | |
590 | goto reread_backing_page; | |
591 | ||
592 | /* the backing page is already up to date, attach the netfs | |
593 | * page to the pagecache and LRU and copy the data across */ | |
594 | backing_page_already_uptodate_unlock: | |
595 | _debug("uptodate %lx", backpage->flags); | |
596 | unlock_page(backpage); | |
597 | backing_page_already_uptodate: | |
598 | _debug("- uptodate"); | |
599 | ||
55881bc7 JW |
600 | ret = add_to_page_cache_lru(netpage, op->mapping, |
601 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
602 | if (ret < 0) { |
603 | if (ret == -EEXIST) { | |
09cbfeaf | 604 | put_page(netpage); |
b4cf1e08 | 605 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
606 | continue; |
607 | } | |
608 | goto nomem; | |
609 | } | |
610 | ||
611 | copy_highpage(netpage, backpage); | |
612 | ||
09cbfeaf | 613 | put_page(backpage); |
9ae326a6 DH |
614 | backpage = NULL; |
615 | ||
c4d6d8db | 616 | fscache_mark_page_cached(op, netpage); |
9ae326a6 | 617 | |
c4d6d8db | 618 | /* the netpage is unlocked and marked up to date here */ |
9ae326a6 | 619 | fscache_end_io(op, netpage, 0); |
09cbfeaf | 620 | put_page(netpage); |
9ae326a6 | 621 | netpage = NULL; |
b4cf1e08 | 622 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
623 | continue; |
624 | } | |
625 | ||
626 | netpage = NULL; | |
627 | ||
628 | _debug("out"); | |
629 | ||
630 | out: | |
631 | /* tidy up */ | |
9ae326a6 | 632 | if (newpage) |
09cbfeaf | 633 | put_page(newpage); |
9ae326a6 | 634 | if (netpage) |
09cbfeaf | 635 | put_page(netpage); |
9ae326a6 | 636 | if (backpage) |
09cbfeaf | 637 | put_page(backpage); |
9ae326a6 DH |
638 | if (monitor) { |
639 | fscache_put_retrieval(op); | |
640 | kfree(monitor); | |
641 | } | |
642 | ||
643 | list_for_each_entry_safe(netpage, _n, list, lru) { | |
644 | list_del(&netpage->lru); | |
09cbfeaf | 645 | put_page(netpage); |
9f10523f | 646 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
647 | } |
648 | ||
649 | _leave(" = %d", ret); | |
650 | return ret; | |
651 | ||
652 | nomem: | |
653 | _debug("nomem"); | |
654 | ret = -ENOMEM; | |
b4cf1e08 | 655 | goto record_page_complete; |
9ae326a6 DH |
656 | |
657 | read_error: | |
658 | _debug("read error %d", ret); | |
659 | if (ret == -ENOMEM) | |
b4cf1e08 | 660 | goto record_page_complete; |
9ae326a6 DH |
661 | io_error: |
662 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
663 | ret = -ENOBUFS; | |
b4cf1e08 DH |
664 | record_page_complete: |
665 | fscache_retrieval_complete(op, 1); | |
9ae326a6 DH |
666 | goto out; |
667 | } | |
668 | ||
669 | /* | |
670 | * read a list of pages from the cache or allocate blocks in which to store | |
671 | * them | |
672 | */ | |
673 | int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op, | |
674 | struct list_head *pages, | |
675 | unsigned *nr_pages, | |
676 | gfp_t gfp) | |
677 | { | |
678 | struct cachefiles_object *object; | |
679 | struct cachefiles_cache *cache; | |
680 | struct list_head backpages; | |
681 | struct pagevec pagevec; | |
682 | struct inode *inode; | |
683 | struct page *page, *_n; | |
684 | unsigned shift, nrbackpages; | |
685 | int ret, ret2, space; | |
686 | ||
687 | object = container_of(op->op.object, | |
688 | struct cachefiles_object, fscache); | |
689 | cache = container_of(object->fscache.cache, | |
690 | struct cachefiles_cache, cache); | |
691 | ||
692 | _enter("{OBJ%x,%d},,%d,,", | |
693 | object->fscache.debug_id, atomic_read(&op->op.usage), | |
694 | *nr_pages); | |
695 | ||
696 | if (!object->backer) | |
9f10523f | 697 | goto all_enobufs; |
9ae326a6 DH |
698 | |
699 | space = 1; | |
700 | if (cachefiles_has_space(cache, 0, *nr_pages) < 0) | |
701 | space = 0; | |
702 | ||
466b77bc | 703 | inode = d_backing_inode(object->backer); |
9ae326a6 DH |
704 | ASSERT(S_ISREG(inode->i_mode)); |
705 | ASSERT(inode->i_mapping->a_ops->bmap); | |
706 | ASSERT(inode->i_mapping->a_ops->readpages); | |
707 | ||
708 | /* calculate the shift required to use bmap */ | |
9ae326a6 DH |
709 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
710 | ||
86679820 | 711 | pagevec_init(&pagevec); |
9ae326a6 | 712 | |
4fbf4291 | 713 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 714 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
715 | op->op.processor = cachefiles_read_copier; |
716 | ||
717 | INIT_LIST_HEAD(&backpages); | |
718 | nrbackpages = 0; | |
719 | ||
720 | ret = space ? -ENODATA : -ENOBUFS; | |
721 | list_for_each_entry_safe(page, _n, pages, lru) { | |
722 | sector_t block0, block; | |
723 | ||
724 | /* we assume the absence or presence of the first block is a | |
725 | * good enough indication for the page as a whole | |
726 | * - TODO: don't use bmap() for this as it is _not_ actually | |
727 | * good enough for this as it doesn't indicate errors, but | |
728 | * it's all we've got for the moment | |
729 | */ | |
730 | block0 = page->index; | |
731 | block0 <<= shift; | |
732 | ||
733 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, | |
734 | block0); | |
735 | _debug("%llx -> %llx", | |
736 | (unsigned long long) block0, | |
737 | (unsigned long long) block); | |
738 | ||
739 | if (block) { | |
740 | /* we have data - add it to the list to give to the | |
741 | * backing fs */ | |
742 | list_move(&page->lru, &backpages); | |
743 | (*nr_pages)--; | |
744 | nrbackpages++; | |
745 | } else if (space && pagevec_add(&pagevec, page) == 0) { | |
746 | fscache_mark_pages_cached(op, &pagevec); | |
9f10523f | 747 | fscache_retrieval_complete(op, 1); |
9ae326a6 | 748 | ret = -ENODATA; |
9f10523f DH |
749 | } else { |
750 | fscache_retrieval_complete(op, 1); | |
9ae326a6 DH |
751 | } |
752 | } | |
753 | ||
754 | if (pagevec_count(&pagevec) > 0) | |
755 | fscache_mark_pages_cached(op, &pagevec); | |
756 | ||
757 | if (list_empty(pages)) | |
758 | ret = 0; | |
759 | ||
760 | /* submit the apparently valid pages to the backing fs to be read from | |
761 | * disk */ | |
762 | if (nrbackpages > 0) { | |
c4d6d8db | 763 | ret2 = cachefiles_read_backing_file(object, op, &backpages); |
9ae326a6 DH |
764 | if (ret2 == -ENOMEM || ret2 == -EINTR) |
765 | ret = ret2; | |
766 | } | |
767 | ||
9ae326a6 DH |
768 | _leave(" = %d [nr=%u%s]", |
769 | ret, *nr_pages, list_empty(pages) ? " empty" : ""); | |
770 | return ret; | |
9f10523f DH |
771 | |
772 | all_enobufs: | |
773 | fscache_retrieval_complete(op, *nr_pages); | |
774 | return -ENOBUFS; | |
9ae326a6 DH |
775 | } |
776 | ||
777 | /* | |
778 | * allocate a block in the cache in which to store a page | |
779 | * - cache withdrawal is prevented by the caller | |
780 | * - returns -EINTR if interrupted | |
781 | * - returns -ENOMEM if ran out of memory | |
782 | * - returns -ENOBUFS if no buffers can be made available | |
783 | * - returns -ENOBUFS if page is beyond EOF | |
784 | * - otherwise: | |
785 | * - the metadata will be retained | |
786 | * - 0 will be returned | |
787 | */ | |
788 | int cachefiles_allocate_page(struct fscache_retrieval *op, | |
789 | struct page *page, | |
790 | gfp_t gfp) | |
791 | { | |
792 | struct cachefiles_object *object; | |
793 | struct cachefiles_cache *cache; | |
9ae326a6 DH |
794 | int ret; |
795 | ||
796 | object = container_of(op->op.object, | |
797 | struct cachefiles_object, fscache); | |
798 | cache = container_of(object->fscache.cache, | |
799 | struct cachefiles_cache, cache); | |
800 | ||
801 | _enter("%p,{%lx},", object, page->index); | |
802 | ||
803 | ret = cachefiles_has_space(cache, 0, 1); | |
c4d6d8db DH |
804 | if (ret == 0) |
805 | fscache_mark_page_cached(op, page); | |
806 | else | |
9ae326a6 | 807 | ret = -ENOBUFS; |
9ae326a6 | 808 | |
9f10523f | 809 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
810 | _leave(" = %d", ret); |
811 | return ret; | |
812 | } | |
813 | ||
814 | /* | |
815 | * allocate blocks in the cache in which to store a set of pages | |
816 | * - cache withdrawal is prevented by the caller | |
817 | * - returns -EINTR if interrupted | |
818 | * - returns -ENOMEM if ran out of memory | |
819 | * - returns -ENOBUFS if some buffers couldn't be made available | |
820 | * - returns -ENOBUFS if some pages are beyond EOF | |
821 | * - otherwise: | |
822 | * - -ENODATA will be returned | |
823 | * - metadata will be retained for any page marked | |
824 | */ | |
825 | int cachefiles_allocate_pages(struct fscache_retrieval *op, | |
826 | struct list_head *pages, | |
827 | unsigned *nr_pages, | |
828 | gfp_t gfp) | |
829 | { | |
830 | struct cachefiles_object *object; | |
831 | struct cachefiles_cache *cache; | |
832 | struct pagevec pagevec; | |
833 | struct page *page; | |
834 | int ret; | |
835 | ||
836 | object = container_of(op->op.object, | |
837 | struct cachefiles_object, fscache); | |
838 | cache = container_of(object->fscache.cache, | |
839 | struct cachefiles_cache, cache); | |
840 | ||
841 | _enter("%p,,,%d,", object, *nr_pages); | |
842 | ||
843 | ret = cachefiles_has_space(cache, 0, *nr_pages); | |
844 | if (ret == 0) { | |
86679820 | 845 | pagevec_init(&pagevec); |
9ae326a6 DH |
846 | |
847 | list_for_each_entry(page, pages, lru) { | |
848 | if (pagevec_add(&pagevec, page) == 0) | |
849 | fscache_mark_pages_cached(op, &pagevec); | |
850 | } | |
851 | ||
852 | if (pagevec_count(&pagevec) > 0) | |
853 | fscache_mark_pages_cached(op, &pagevec); | |
854 | ret = -ENODATA; | |
855 | } else { | |
856 | ret = -ENOBUFS; | |
857 | } | |
858 | ||
9f10523f | 859 | fscache_retrieval_complete(op, *nr_pages); |
9ae326a6 DH |
860 | _leave(" = %d", ret); |
861 | return ret; | |
862 | } | |
863 | ||
864 | /* | |
865 | * request a page be stored in the cache | |
866 | * - cache withdrawal is prevented by the caller | |
867 | * - this request may be ignored if there's no cache block available, in which | |
868 | * case -ENOBUFS will be returned | |
869 | * - if the op is in progress, 0 will be returned | |
870 | */ | |
871 | int cachefiles_write_page(struct fscache_storage *op, struct page *page) | |
872 | { | |
873 | struct cachefiles_object *object; | |
874 | struct cachefiles_cache *cache; | |
9ae326a6 | 875 | struct file *file; |
765927b2 | 876 | struct path path; |
a17754fb DH |
877 | loff_t pos, eof; |
878 | size_t len; | |
9ae326a6 | 879 | void *data; |
cf897526 | 880 | int ret = -ENOBUFS; |
9ae326a6 DH |
881 | |
882 | ASSERT(op != NULL); | |
883 | ASSERT(page != NULL); | |
884 | ||
885 | object = container_of(op->op.object, | |
886 | struct cachefiles_object, fscache); | |
887 | ||
888 | _enter("%p,%p{%lx},,,", object, page, page->index); | |
889 | ||
890 | if (!object->backer) { | |
891 | _leave(" = -ENOBUFS"); | |
892 | return -ENOBUFS; | |
893 | } | |
894 | ||
ce40fa78 | 895 | ASSERT(d_is_reg(object->backer)); |
9ae326a6 DH |
896 | |
897 | cache = container_of(object->fscache.cache, | |
898 | struct cachefiles_cache, cache); | |
899 | ||
102f4d90 DH |
900 | pos = (loff_t)page->index << PAGE_SHIFT; |
901 | ||
902 | /* We mustn't write more data than we have, so we have to beware of a | |
903 | * partial page at EOF. | |
904 | */ | |
905 | eof = object->fscache.store_limit_l; | |
906 | if (pos >= eof) | |
907 | goto error; | |
908 | ||
9ae326a6 DH |
909 | /* write the page to the backing filesystem and let it store it in its |
910 | * own time */ | |
765927b2 AV |
911 | path.mnt = cache->mnt; |
912 | path.dentry = object->backer; | |
98c350cd | 913 | file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred); |
9ae326a6 DH |
914 | if (IS_ERR(file)) { |
915 | ret = PTR_ERR(file); | |
102f4d90 | 916 | goto error_2; |
9ae326a6 DH |
917 | } |
918 | ||
102f4d90 DH |
919 | len = PAGE_SIZE; |
920 | if (eof & ~PAGE_MASK) { | |
921 | if (eof - pos < PAGE_SIZE) { | |
922 | _debug("cut short %llx to %llx", | |
923 | pos, eof); | |
924 | len = eof - pos; | |
925 | ASSERTCMP(pos + len, ==, eof); | |
926 | } | |
9ae326a6 DH |
927 | } |
928 | ||
102f4d90 DH |
929 | data = kmap(page); |
930 | ret = __kernel_write(file, data, len, &pos); | |
931 | kunmap(page); | |
932 | fput(file); | |
933 | if (ret != len) | |
934 | goto error_eio; | |
935 | ||
936 | _leave(" = 0"); | |
937 | return 0; | |
938 | ||
939 | error_eio: | |
940 | ret = -EIO; | |
941 | error_2: | |
942 | if (ret == -EIO) | |
943 | cachefiles_io_error_obj(object, | |
944 | "Write page to backing file failed"); | |
945 | error: | |
946 | _leave(" = -ENOBUFS [%d]", ret); | |
947 | return -ENOBUFS; | |
9ae326a6 DH |
948 | } |
949 | ||
950 | /* | |
951 | * detach a backing block from a page | |
952 | * - cache withdrawal is prevented by the caller | |
953 | */ | |
954 | void cachefiles_uncache_page(struct fscache_object *_object, struct page *page) | |
955 | { | |
956 | struct cachefiles_object *object; | |
957 | struct cachefiles_cache *cache; | |
958 | ||
959 | object = container_of(_object, struct cachefiles_object, fscache); | |
960 | cache = container_of(object->fscache.cache, | |
961 | struct cachefiles_cache, cache); | |
962 | ||
963 | _enter("%p,{%lu}", object, page->index); | |
964 | ||
965 | spin_unlock(&object->fscache.cookie->lock); | |
966 | } |