]>
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; | |
7bf1efb0 | 30 | struct fscache_retrieval *op = monitor->op; |
9ae326a6 DH |
31 | struct wait_bit_key *key = _key; |
32 | struct page *page = wait->private; | |
33 | ||
34 | ASSERT(key); | |
35 | ||
36 | _enter("{%lu},%u,%d,{%p,%u}", | |
37 | monitor->netfs_page->index, mode, sync, | |
38 | key->flags, key->bit_nr); | |
39 | ||
40 | if (key->flags != &page->flags || | |
41 | key->bit_nr != PG_locked) | |
42 | return 0; | |
43 | ||
44 | _debug("--- monitor %p %lx ---", page, page->flags); | |
45 | ||
5e929b33 DH |
46 | if (!PageUptodate(page) && !PageError(page)) { |
47 | /* unlocked, not uptodate and not erronous? */ | |
48 | _debug("page probably truncated"); | |
49 | } | |
9ae326a6 DH |
50 | |
51 | /* remove from the waitqueue */ | |
2055da97 | 52 | list_del(&wait->entry); |
9ae326a6 DH |
53 | |
54 | /* move onto the action list and queue for FS-Cache thread pool */ | |
7bf1efb0 | 55 | ASSERT(op); |
9ae326a6 | 56 | |
7bf1efb0 KKM |
57 | /* We need to temporarily bump the usage count as we don't own a ref |
58 | * here otherwise cachefiles_read_copier() may free the op between the | |
59 | * monitor being enqueued on the op->to_do list and the op getting | |
60 | * enqueued on the work queue. | |
61 | */ | |
62 | fscache_get_retrieval(op); | |
9ae326a6 | 63 | |
7bf1efb0 | 64 | object = container_of(op->op.object, struct cachefiles_object, fscache); |
9ae326a6 | 65 | spin_lock(&object->work_lock); |
7bf1efb0 | 66 | list_add_tail(&monitor->op_link, &op->to_do); |
9ae326a6 DH |
67 | spin_unlock(&object->work_lock); |
68 | ||
7bf1efb0 KKM |
69 | fscache_enqueue_retrieval(op); |
70 | fscache_put_retrieval(op); | |
9ae326a6 DH |
71 | return 0; |
72 | } | |
73 | ||
5e929b33 DH |
74 | /* |
75 | * handle a probably truncated page | |
76 | * - check to see if the page is still relevant and reissue the read if | |
77 | * possible | |
78 | * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we | |
79 | * must wait again and 0 if successful | |
80 | */ | |
81 | static int cachefiles_read_reissue(struct cachefiles_object *object, | |
82 | struct cachefiles_one_read *monitor) | |
83 | { | |
466b77bc | 84 | struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping; |
5e929b33 DH |
85 | struct page *backpage = monitor->back_page, *backpage2; |
86 | int ret; | |
87 | ||
37491a13 | 88 | _enter("{ino=%lx},{%lx,%lx}", |
466b77bc | 89 | d_backing_inode(object->backer)->i_ino, |
5e929b33 DH |
90 | backpage->index, backpage->flags); |
91 | ||
92 | /* skip if the page was truncated away completely */ | |
93 | if (backpage->mapping != bmapping) { | |
37491a13 | 94 | _leave(" = -ENODATA [mapping]"); |
5e929b33 DH |
95 | return -ENODATA; |
96 | } | |
97 | ||
98 | backpage2 = find_get_page(bmapping, backpage->index); | |
99 | if (!backpage2) { | |
37491a13 | 100 | _leave(" = -ENODATA [gone]"); |
5e929b33 DH |
101 | return -ENODATA; |
102 | } | |
103 | ||
104 | if (backpage != backpage2) { | |
105 | put_page(backpage2); | |
37491a13 | 106 | _leave(" = -ENODATA [different]"); |
5e929b33 DH |
107 | return -ENODATA; |
108 | } | |
109 | ||
110 | /* the page is still there and we already have a ref on it, so we don't | |
111 | * need a second */ | |
112 | put_page(backpage2); | |
113 | ||
114 | INIT_LIST_HEAD(&monitor->op_link); | |
115 | add_page_wait_queue(backpage, &monitor->monitor); | |
116 | ||
117 | if (trylock_page(backpage)) { | |
118 | ret = -EIO; | |
119 | if (PageError(backpage)) | |
120 | goto unlock_discard; | |
121 | ret = 0; | |
122 | if (PageUptodate(backpage)) | |
123 | goto unlock_discard; | |
124 | ||
37491a13 | 125 | _debug("reissue read"); |
5e929b33 DH |
126 | ret = bmapping->a_ops->readpage(NULL, backpage); |
127 | if (ret < 0) | |
128 | goto unlock_discard; | |
129 | } | |
130 | ||
131 | /* but the page may have been read before the monitor was installed, so | |
132 | * the monitor may miss the event - so we have to ensure that we do get | |
133 | * one in such a case */ | |
134 | if (trylock_page(backpage)) { | |
135 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
136 | unlock_page(backpage); | |
137 | } | |
138 | ||
139 | /* it'll reappear on the todo list */ | |
37491a13 | 140 | _leave(" = -EINPROGRESS"); |
5e929b33 DH |
141 | return -EINPROGRESS; |
142 | ||
143 | unlock_discard: | |
144 | unlock_page(backpage); | |
145 | spin_lock_irq(&object->work_lock); | |
146 | list_del(&monitor->op_link); | |
147 | spin_unlock_irq(&object->work_lock); | |
37491a13 | 148 | _leave(" = %d", ret); |
5e929b33 DH |
149 | return ret; |
150 | } | |
151 | ||
9ae326a6 DH |
152 | /* |
153 | * copy data from backing pages to netfs pages to complete a read operation | |
154 | * - driven by FS-Cache's thread pool | |
155 | */ | |
156 | static void cachefiles_read_copier(struct fscache_operation *_op) | |
157 | { | |
158 | struct cachefiles_one_read *monitor; | |
159 | struct cachefiles_object *object; | |
160 | struct fscache_retrieval *op; | |
9ae326a6 DH |
161 | int error, max; |
162 | ||
163 | op = container_of(_op, struct fscache_retrieval, op); | |
164 | object = container_of(op->op.object, | |
165 | struct cachefiles_object, fscache); | |
166 | ||
466b77bc | 167 | _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino); |
9ae326a6 | 168 | |
9ae326a6 DH |
169 | max = 8; |
170 | spin_lock_irq(&object->work_lock); | |
171 | ||
172 | while (!list_empty(&op->to_do)) { | |
173 | monitor = list_entry(op->to_do.next, | |
174 | struct cachefiles_one_read, op_link); | |
175 | list_del(&monitor->op_link); | |
176 | ||
177 | spin_unlock_irq(&object->work_lock); | |
178 | ||
179 | _debug("- copy {%lu}", monitor->back_page->index); | |
180 | ||
5e929b33 | 181 | recheck: |
9dc8d9bf DH |
182 | if (test_bit(FSCACHE_COOKIE_INVALIDATING, |
183 | &object->fscache.cookie->flags)) { | |
184 | error = -ESTALE; | |
185 | } else if (PageUptodate(monitor->back_page)) { | |
9ae326a6 | 186 | copy_highpage(monitor->netfs_page, monitor->back_page); |
c4d6d8db DH |
187 | fscache_mark_page_cached(monitor->op, |
188 | monitor->netfs_page); | |
9ae326a6 | 189 | error = 0; |
5e929b33 DH |
190 | } else if (!PageError(monitor->back_page)) { |
191 | /* the page has probably been truncated */ | |
192 | error = cachefiles_read_reissue(object, monitor); | |
193 | if (error == -EINPROGRESS) | |
194 | goto next; | |
195 | goto recheck; | |
196 | } else { | |
9ae326a6 DH |
197 | cachefiles_io_error_obj( |
198 | object, | |
199 | "Readpage failed on backing file %lx", | |
200 | (unsigned long) monitor->back_page->flags); | |
5e929b33 DH |
201 | error = -EIO; |
202 | } | |
9ae326a6 | 203 | |
09cbfeaf | 204 | put_page(monitor->back_page); |
9ae326a6 DH |
205 | |
206 | fscache_end_io(op, monitor->netfs_page, error); | |
09cbfeaf | 207 | put_page(monitor->netfs_page); |
9f10523f | 208 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
209 | fscache_put_retrieval(op); |
210 | kfree(monitor); | |
211 | ||
5e929b33 | 212 | next: |
9ae326a6 DH |
213 | /* let the thread pool have some air occasionally */ |
214 | max--; | |
215 | if (max < 0 || need_resched()) { | |
216 | if (!list_empty(&op->to_do)) | |
217 | fscache_enqueue_retrieval(op); | |
218 | _leave(" [maxed out]"); | |
219 | return; | |
220 | } | |
221 | ||
222 | spin_lock_irq(&object->work_lock); | |
223 | } | |
224 | ||
225 | spin_unlock_irq(&object->work_lock); | |
226 | _leave(""); | |
227 | } | |
228 | ||
229 | /* | |
230 | * read the corresponding page to the given set from the backing file | |
231 | * - an uncertain page is simply discarded, to be tried again another time | |
232 | */ | |
233 | static int cachefiles_read_backing_file_one(struct cachefiles_object *object, | |
234 | struct fscache_retrieval *op, | |
a0b8cab3 | 235 | struct page *netpage) |
9ae326a6 DH |
236 | { |
237 | struct cachefiles_one_read *monitor; | |
238 | struct address_space *bmapping; | |
239 | struct page *newpage, *backpage; | |
240 | int ret; | |
241 | ||
242 | _enter(""); | |
243 | ||
9ae326a6 DH |
244 | _debug("read back %p{%lu,%d}", |
245 | netpage, netpage->index, page_count(netpage)); | |
246 | ||
5f4f9f4a | 247 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
248 | if (!monitor) |
249 | goto nomem; | |
250 | ||
251 | monitor->netfs_page = netpage; | |
252 | monitor->op = fscache_get_retrieval(op); | |
253 | ||
254 | init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter); | |
255 | ||
256 | /* attempt to get hold of the backing page */ | |
466b77bc | 257 | bmapping = d_backing_inode(object->backer)->i_mapping; |
9ae326a6 DH |
258 | newpage = NULL; |
259 | ||
260 | for (;;) { | |
261 | backpage = find_get_page(bmapping, netpage->index); | |
262 | if (backpage) | |
263 | goto backing_page_already_present; | |
264 | ||
265 | if (!newpage) { | |
453f85d4 | 266 | newpage = __page_cache_alloc(cachefiles_gfp); |
9ae326a6 DH |
267 | if (!newpage) |
268 | goto nomem_monitor; | |
269 | } | |
270 | ||
55881bc7 JW |
271 | ret = add_to_page_cache_lru(newpage, bmapping, |
272 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
273 | if (ret == 0) |
274 | goto installed_new_backing_page; | |
275 | if (ret != -EEXIST) | |
276 | goto nomem_page; | |
277 | } | |
278 | ||
55881bc7 JW |
279 | /* we've installed a new backing page, so now we need to start |
280 | * it reading */ | |
9ae326a6 DH |
281 | installed_new_backing_page: |
282 | _debug("- new %p", newpage); | |
283 | ||
284 | backpage = newpage; | |
285 | newpage = NULL; | |
286 | ||
9ae326a6 DH |
287 | read_backing_page: |
288 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
289 | if (ret < 0) | |
290 | goto read_error; | |
291 | ||
292 | /* set the monitor to transfer the data across */ | |
293 | monitor_backing_page: | |
294 | _debug("- monitor add"); | |
295 | ||
296 | /* install the monitor */ | |
09cbfeaf KS |
297 | get_page(monitor->netfs_page); |
298 | get_page(backpage); | |
9ae326a6 DH |
299 | monitor->back_page = backpage; |
300 | monitor->monitor.private = backpage; | |
301 | add_page_wait_queue(backpage, &monitor->monitor); | |
302 | monitor = NULL; | |
303 | ||
304 | /* but the page may have been read before the monitor was installed, so | |
305 | * the monitor may miss the event - so we have to ensure that we do get | |
306 | * one in such a case */ | |
307 | if (trylock_page(backpage)) { | |
308 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
309 | unlock_page(backpage); | |
310 | } | |
311 | goto success; | |
312 | ||
313 | /* if the backing page is already present, it can be in one of | |
314 | * three states: read in progress, read failed or read okay */ | |
315 | backing_page_already_present: | |
316 | _debug("- present"); | |
317 | ||
318 | if (newpage) { | |
09cbfeaf | 319 | put_page(newpage); |
9ae326a6 DH |
320 | newpage = NULL; |
321 | } | |
322 | ||
323 | if (PageError(backpage)) | |
324 | goto io_error; | |
325 | ||
326 | if (PageUptodate(backpage)) | |
327 | goto backing_page_already_uptodate; | |
328 | ||
329 | if (!trylock_page(backpage)) | |
330 | goto monitor_backing_page; | |
331 | _debug("read %p {%lx}", backpage, backpage->flags); | |
332 | goto read_backing_page; | |
333 | ||
334 | /* the backing page is already up to date, attach the netfs | |
335 | * page to the pagecache and LRU and copy the data across */ | |
336 | backing_page_already_uptodate: | |
337 | _debug("- uptodate"); | |
338 | ||
c4d6d8db | 339 | fscache_mark_page_cached(op, netpage); |
9ae326a6 DH |
340 | |
341 | copy_highpage(netpage, backpage); | |
342 | fscache_end_io(op, netpage, 0); | |
9f10523f | 343 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
344 | |
345 | success: | |
346 | _debug("success"); | |
347 | ret = 0; | |
348 | ||
349 | out: | |
350 | if (backpage) | |
09cbfeaf | 351 | put_page(backpage); |
9ae326a6 DH |
352 | if (monitor) { |
353 | fscache_put_retrieval(monitor->op); | |
354 | kfree(monitor); | |
355 | } | |
356 | _leave(" = %d", ret); | |
357 | return ret; | |
358 | ||
359 | read_error: | |
360 | _debug("read error %d", ret); | |
b4cf1e08 DH |
361 | if (ret == -ENOMEM) { |
362 | fscache_retrieval_complete(op, 1); | |
9ae326a6 | 363 | goto out; |
b4cf1e08 | 364 | } |
9ae326a6 DH |
365 | io_error: |
366 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
9f10523f | 367 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
368 | ret = -ENOBUFS; |
369 | goto out; | |
370 | ||
371 | nomem_page: | |
09cbfeaf | 372 | put_page(newpage); |
9ae326a6 DH |
373 | nomem_monitor: |
374 | fscache_put_retrieval(monitor->op); | |
375 | kfree(monitor); | |
376 | nomem: | |
9f10523f | 377 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
378 | _leave(" = -ENOMEM"); |
379 | return -ENOMEM; | |
380 | } | |
381 | ||
382 | /* | |
383 | * read a page from the cache or allocate a block in which to store it | |
384 | * - cache withdrawal is prevented by the caller | |
385 | * - returns -EINTR if interrupted | |
386 | * - returns -ENOMEM if ran out of memory | |
387 | * - returns -ENOBUFS if no buffers can be made available | |
388 | * - returns -ENOBUFS if page is beyond EOF | |
389 | * - if the page is backed by a block in the cache: | |
390 | * - a read will be started which will call the callback on completion | |
391 | * - 0 will be returned | |
392 | * - else if the page is unbacked: | |
393 | * - the metadata will be retained | |
394 | * - -ENODATA will be returned | |
395 | */ | |
396 | int cachefiles_read_or_alloc_page(struct fscache_retrieval *op, | |
397 | struct page *page, | |
398 | gfp_t gfp) | |
399 | { | |
400 | struct cachefiles_object *object; | |
401 | struct cachefiles_cache *cache; | |
9ae326a6 DH |
402 | struct inode *inode; |
403 | sector_t block0, block; | |
404 | unsigned shift; | |
405 | int ret; | |
406 | ||
407 | object = container_of(op->op.object, | |
408 | struct cachefiles_object, fscache); | |
409 | cache = container_of(object->fscache.cache, | |
410 | struct cachefiles_cache, cache); | |
411 | ||
412 | _enter("{%p},{%lx},,,", object, page->index); | |
413 | ||
414 | if (!object->backer) | |
9f10523f | 415 | goto enobufs; |
9ae326a6 | 416 | |
466b77bc | 417 | inode = d_backing_inode(object->backer); |
9ae326a6 DH |
418 | ASSERT(S_ISREG(inode->i_mode)); |
419 | ASSERT(inode->i_mapping->a_ops->bmap); | |
420 | ASSERT(inode->i_mapping->a_ops->readpages); | |
421 | ||
422 | /* calculate the shift required to use bmap */ | |
9ae326a6 DH |
423 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
424 | ||
4fbf4291 | 425 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 426 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
427 | op->op.processor = cachefiles_read_copier; |
428 | ||
9ae326a6 DH |
429 | /* we assume the absence or presence of the first block is a good |
430 | * enough indication for the page as a whole | |
431 | * - TODO: don't use bmap() for this as it is _not_ actually good | |
432 | * enough for this as it doesn't indicate errors, but it's all we've | |
433 | * got for the moment | |
434 | */ | |
435 | block0 = page->index; | |
436 | block0 <<= shift; | |
437 | ||
438 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0); | |
439 | _debug("%llx -> %llx", | |
440 | (unsigned long long) block0, | |
441 | (unsigned long long) block); | |
442 | ||
443 | if (block) { | |
444 | /* submit the apparently valid page to the backing fs to be | |
445 | * read from disk */ | |
a0b8cab3 | 446 | ret = cachefiles_read_backing_file_one(object, op, page); |
9ae326a6 DH |
447 | } else if (cachefiles_has_space(cache, 0, 1) == 0) { |
448 | /* there's space in the cache we can use */ | |
c4d6d8db | 449 | fscache_mark_page_cached(op, page); |
9f10523f | 450 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
451 | ret = -ENODATA; |
452 | } else { | |
9f10523f | 453 | goto enobufs; |
9ae326a6 DH |
454 | } |
455 | ||
456 | _leave(" = %d", ret); | |
457 | return ret; | |
9f10523f DH |
458 | |
459 | enobufs: | |
460 | fscache_retrieval_complete(op, 1); | |
461 | _leave(" = -ENOBUFS"); | |
462 | return -ENOBUFS; | |
9ae326a6 DH |
463 | } |
464 | ||
465 | /* | |
466 | * read the corresponding pages to the given set from the backing file | |
467 | * - any uncertain pages are simply discarded, to be tried again another time | |
468 | */ | |
469 | static int cachefiles_read_backing_file(struct cachefiles_object *object, | |
470 | struct fscache_retrieval *op, | |
c4d6d8db | 471 | struct list_head *list) |
9ae326a6 DH |
472 | { |
473 | struct cachefiles_one_read *monitor = NULL; | |
466b77bc | 474 | struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping; |
9ae326a6 DH |
475 | struct page *newpage = NULL, *netpage, *_n, *backpage = NULL; |
476 | int ret = 0; | |
477 | ||
478 | _enter(""); | |
479 | ||
9ae326a6 DH |
480 | list_for_each_entry_safe(netpage, _n, list, lru) { |
481 | list_del(&netpage->lru); | |
482 | ||
483 | _debug("read back %p{%lu,%d}", | |
484 | netpage, netpage->index, page_count(netpage)); | |
485 | ||
486 | if (!monitor) { | |
5f4f9f4a | 487 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
488 | if (!monitor) |
489 | goto nomem; | |
490 | ||
491 | monitor->op = fscache_get_retrieval(op); | |
492 | init_waitqueue_func_entry(&monitor->monitor, | |
493 | cachefiles_read_waiter); | |
494 | } | |
495 | ||
496 | for (;;) { | |
497 | backpage = find_get_page(bmapping, netpage->index); | |
498 | if (backpage) | |
499 | goto backing_page_already_present; | |
500 | ||
501 | if (!newpage) { | |
453f85d4 | 502 | newpage = __page_cache_alloc(cachefiles_gfp); |
9ae326a6 DH |
503 | if (!newpage) |
504 | goto nomem; | |
505 | } | |
506 | ||
55881bc7 JW |
507 | ret = add_to_page_cache_lru(newpage, bmapping, |
508 | netpage->index, | |
509 | cachefiles_gfp); | |
9ae326a6 DH |
510 | if (ret == 0) |
511 | goto installed_new_backing_page; | |
512 | if (ret != -EEXIST) | |
513 | goto nomem; | |
514 | } | |
515 | ||
55881bc7 JW |
516 | /* we've installed a new backing page, so now we need |
517 | * to start it reading */ | |
9ae326a6 DH |
518 | installed_new_backing_page: |
519 | _debug("- new %p", newpage); | |
520 | ||
521 | backpage = newpage; | |
522 | newpage = NULL; | |
523 | ||
9ae326a6 DH |
524 | reread_backing_page: |
525 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
526 | if (ret < 0) | |
527 | goto read_error; | |
528 | ||
529 | /* add the netfs page to the pagecache and LRU, and set the | |
530 | * monitor to transfer the data across */ | |
531 | monitor_backing_page: | |
532 | _debug("- monitor add"); | |
533 | ||
55881bc7 JW |
534 | ret = add_to_page_cache_lru(netpage, op->mapping, |
535 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
536 | if (ret < 0) { |
537 | if (ret == -EEXIST) { | |
09cbfeaf | 538 | put_page(netpage); |
b4cf1e08 | 539 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
540 | continue; |
541 | } | |
542 | goto nomem; | |
543 | } | |
544 | ||
9ae326a6 | 545 | /* install a monitor */ |
09cbfeaf | 546 | get_page(netpage); |
9ae326a6 DH |
547 | monitor->netfs_page = netpage; |
548 | ||
09cbfeaf | 549 | get_page(backpage); |
9ae326a6 DH |
550 | monitor->back_page = backpage; |
551 | monitor->monitor.private = backpage; | |
552 | add_page_wait_queue(backpage, &monitor->monitor); | |
553 | monitor = NULL; | |
554 | ||
555 | /* but the page may have been read before the monitor was | |
556 | * installed, so the monitor may miss the event - so we have to | |
557 | * ensure that we do get one in such a case */ | |
558 | if (trylock_page(backpage)) { | |
559 | _debug("2unlock %p {%lx}", backpage, backpage->flags); | |
560 | unlock_page(backpage); | |
561 | } | |
562 | ||
09cbfeaf | 563 | put_page(backpage); |
9ae326a6 DH |
564 | backpage = NULL; |
565 | ||
09cbfeaf | 566 | put_page(netpage); |
9ae326a6 DH |
567 | netpage = NULL; |
568 | continue; | |
569 | ||
570 | /* if the backing page is already present, it can be in one of | |
571 | * three states: read in progress, read failed or read okay */ | |
572 | backing_page_already_present: | |
573 | _debug("- present %p", backpage); | |
574 | ||
575 | if (PageError(backpage)) | |
576 | goto io_error; | |
577 | ||
578 | if (PageUptodate(backpage)) | |
579 | goto backing_page_already_uptodate; | |
580 | ||
581 | _debug("- not ready %p{%lx}", backpage, backpage->flags); | |
582 | ||
583 | if (!trylock_page(backpage)) | |
584 | goto monitor_backing_page; | |
585 | ||
586 | if (PageError(backpage)) { | |
587 | _debug("error %lx", backpage->flags); | |
588 | unlock_page(backpage); | |
589 | goto io_error; | |
590 | } | |
591 | ||
592 | if (PageUptodate(backpage)) | |
593 | goto backing_page_already_uptodate_unlock; | |
594 | ||
595 | /* we've locked a page that's neither up to date nor erroneous, | |
596 | * so we need to attempt to read it again */ | |
597 | goto reread_backing_page; | |
598 | ||
599 | /* the backing page is already up to date, attach the netfs | |
600 | * page to the pagecache and LRU and copy the data across */ | |
601 | backing_page_already_uptodate_unlock: | |
602 | _debug("uptodate %lx", backpage->flags); | |
603 | unlock_page(backpage); | |
604 | backing_page_already_uptodate: | |
605 | _debug("- uptodate"); | |
606 | ||
55881bc7 JW |
607 | ret = add_to_page_cache_lru(netpage, op->mapping, |
608 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
609 | if (ret < 0) { |
610 | if (ret == -EEXIST) { | |
09cbfeaf | 611 | put_page(netpage); |
b4cf1e08 | 612 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
613 | continue; |
614 | } | |
615 | goto nomem; | |
616 | } | |
617 | ||
618 | copy_highpage(netpage, backpage); | |
619 | ||
09cbfeaf | 620 | put_page(backpage); |
9ae326a6 DH |
621 | backpage = NULL; |
622 | ||
c4d6d8db | 623 | fscache_mark_page_cached(op, netpage); |
9ae326a6 | 624 | |
c4d6d8db | 625 | /* the netpage is unlocked and marked up to date here */ |
9ae326a6 | 626 | fscache_end_io(op, netpage, 0); |
09cbfeaf | 627 | put_page(netpage); |
9ae326a6 | 628 | netpage = NULL; |
b4cf1e08 | 629 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
630 | continue; |
631 | } | |
632 | ||
633 | netpage = NULL; | |
634 | ||
635 | _debug("out"); | |
636 | ||
637 | out: | |
638 | /* tidy up */ | |
9ae326a6 | 639 | if (newpage) |
09cbfeaf | 640 | put_page(newpage); |
9ae326a6 | 641 | if (netpage) |
09cbfeaf | 642 | put_page(netpage); |
9ae326a6 | 643 | if (backpage) |
09cbfeaf | 644 | put_page(backpage); |
9ae326a6 DH |
645 | if (monitor) { |
646 | fscache_put_retrieval(op); | |
647 | kfree(monitor); | |
648 | } | |
649 | ||
650 | list_for_each_entry_safe(netpage, _n, list, lru) { | |
651 | list_del(&netpage->lru); | |
09cbfeaf | 652 | put_page(netpage); |
9f10523f | 653 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
654 | } |
655 | ||
656 | _leave(" = %d", ret); | |
657 | return ret; | |
658 | ||
659 | nomem: | |
660 | _debug("nomem"); | |
661 | ret = -ENOMEM; | |
b4cf1e08 | 662 | goto record_page_complete; |
9ae326a6 DH |
663 | |
664 | read_error: | |
665 | _debug("read error %d", ret); | |
666 | if (ret == -ENOMEM) | |
b4cf1e08 | 667 | goto record_page_complete; |
9ae326a6 DH |
668 | io_error: |
669 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
670 | ret = -ENOBUFS; | |
b4cf1e08 DH |
671 | record_page_complete: |
672 | fscache_retrieval_complete(op, 1); | |
9ae326a6 DH |
673 | goto out; |
674 | } | |
675 | ||
676 | /* | |
677 | * read a list of pages from the cache or allocate blocks in which to store | |
678 | * them | |
679 | */ | |
680 | int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op, | |
681 | struct list_head *pages, | |
682 | unsigned *nr_pages, | |
683 | gfp_t gfp) | |
684 | { | |
685 | struct cachefiles_object *object; | |
686 | struct cachefiles_cache *cache; | |
687 | struct list_head backpages; | |
688 | struct pagevec pagevec; | |
689 | struct inode *inode; | |
690 | struct page *page, *_n; | |
691 | unsigned shift, nrbackpages; | |
692 | int ret, ret2, space; | |
693 | ||
694 | object = container_of(op->op.object, | |
695 | struct cachefiles_object, fscache); | |
696 | cache = container_of(object->fscache.cache, | |
697 | struct cachefiles_cache, cache); | |
698 | ||
699 | _enter("{OBJ%x,%d},,%d,,", | |
700 | object->fscache.debug_id, atomic_read(&op->op.usage), | |
701 | *nr_pages); | |
702 | ||
703 | if (!object->backer) | |
9f10523f | 704 | goto all_enobufs; |
9ae326a6 DH |
705 | |
706 | space = 1; | |
707 | if (cachefiles_has_space(cache, 0, *nr_pages) < 0) | |
708 | space = 0; | |
709 | ||
466b77bc | 710 | inode = d_backing_inode(object->backer); |
9ae326a6 DH |
711 | ASSERT(S_ISREG(inode->i_mode)); |
712 | ASSERT(inode->i_mapping->a_ops->bmap); | |
713 | ASSERT(inode->i_mapping->a_ops->readpages); | |
714 | ||
715 | /* calculate the shift required to use bmap */ | |
9ae326a6 DH |
716 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
717 | ||
86679820 | 718 | pagevec_init(&pagevec); |
9ae326a6 | 719 | |
4fbf4291 | 720 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 721 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
722 | op->op.processor = cachefiles_read_copier; |
723 | ||
724 | INIT_LIST_HEAD(&backpages); | |
725 | nrbackpages = 0; | |
726 | ||
727 | ret = space ? -ENODATA : -ENOBUFS; | |
728 | list_for_each_entry_safe(page, _n, pages, lru) { | |
729 | sector_t block0, block; | |
730 | ||
731 | /* we assume the absence or presence of the first block is a | |
732 | * good enough indication for the page as a whole | |
733 | * - TODO: don't use bmap() for this as it is _not_ actually | |
734 | * good enough for this as it doesn't indicate errors, but | |
735 | * it's all we've got for the moment | |
736 | */ | |
737 | block0 = page->index; | |
738 | block0 <<= shift; | |
739 | ||
740 | block = inode->i_mapping->a_ops->bmap(inode->i_mapping, | |
741 | block0); | |
742 | _debug("%llx -> %llx", | |
743 | (unsigned long long) block0, | |
744 | (unsigned long long) block); | |
745 | ||
746 | if (block) { | |
747 | /* we have data - add it to the list to give to the | |
748 | * backing fs */ | |
749 | list_move(&page->lru, &backpages); | |
750 | (*nr_pages)--; | |
751 | nrbackpages++; | |
752 | } else if (space && pagevec_add(&pagevec, page) == 0) { | |
753 | fscache_mark_pages_cached(op, &pagevec); | |
9f10523f | 754 | fscache_retrieval_complete(op, 1); |
9ae326a6 | 755 | ret = -ENODATA; |
9f10523f DH |
756 | } else { |
757 | fscache_retrieval_complete(op, 1); | |
9ae326a6 DH |
758 | } |
759 | } | |
760 | ||
761 | if (pagevec_count(&pagevec) > 0) | |
762 | fscache_mark_pages_cached(op, &pagevec); | |
763 | ||
764 | if (list_empty(pages)) | |
765 | ret = 0; | |
766 | ||
767 | /* submit the apparently valid pages to the backing fs to be read from | |
768 | * disk */ | |
769 | if (nrbackpages > 0) { | |
c4d6d8db | 770 | ret2 = cachefiles_read_backing_file(object, op, &backpages); |
9ae326a6 DH |
771 | if (ret2 == -ENOMEM || ret2 == -EINTR) |
772 | ret = ret2; | |
773 | } | |
774 | ||
9ae326a6 DH |
775 | _leave(" = %d [nr=%u%s]", |
776 | ret, *nr_pages, list_empty(pages) ? " empty" : ""); | |
777 | return ret; | |
9f10523f DH |
778 | |
779 | all_enobufs: | |
780 | fscache_retrieval_complete(op, *nr_pages); | |
781 | return -ENOBUFS; | |
9ae326a6 DH |
782 | } |
783 | ||
784 | /* | |
785 | * allocate a block in the cache in which to store a page | |
786 | * - cache withdrawal is prevented by the caller | |
787 | * - returns -EINTR if interrupted | |
788 | * - returns -ENOMEM if ran out of memory | |
789 | * - returns -ENOBUFS if no buffers can be made available | |
790 | * - returns -ENOBUFS if page is beyond EOF | |
791 | * - otherwise: | |
792 | * - the metadata will be retained | |
793 | * - 0 will be returned | |
794 | */ | |
795 | int cachefiles_allocate_page(struct fscache_retrieval *op, | |
796 | struct page *page, | |
797 | gfp_t gfp) | |
798 | { | |
799 | struct cachefiles_object *object; | |
800 | struct cachefiles_cache *cache; | |
9ae326a6 DH |
801 | int ret; |
802 | ||
803 | object = container_of(op->op.object, | |
804 | struct cachefiles_object, fscache); | |
805 | cache = container_of(object->fscache.cache, | |
806 | struct cachefiles_cache, cache); | |
807 | ||
808 | _enter("%p,{%lx},", object, page->index); | |
809 | ||
810 | ret = cachefiles_has_space(cache, 0, 1); | |
c4d6d8db DH |
811 | if (ret == 0) |
812 | fscache_mark_page_cached(op, page); | |
813 | else | |
9ae326a6 | 814 | ret = -ENOBUFS; |
9ae326a6 | 815 | |
9f10523f | 816 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
817 | _leave(" = %d", ret); |
818 | return ret; | |
819 | } | |
820 | ||
821 | /* | |
822 | * allocate blocks in the cache in which to store a set of pages | |
823 | * - cache withdrawal is prevented by the caller | |
824 | * - returns -EINTR if interrupted | |
825 | * - returns -ENOMEM if ran out of memory | |
826 | * - returns -ENOBUFS if some buffers couldn't be made available | |
827 | * - returns -ENOBUFS if some pages are beyond EOF | |
828 | * - otherwise: | |
829 | * - -ENODATA will be returned | |
830 | * - metadata will be retained for any page marked | |
831 | */ | |
832 | int cachefiles_allocate_pages(struct fscache_retrieval *op, | |
833 | struct list_head *pages, | |
834 | unsigned *nr_pages, | |
835 | gfp_t gfp) | |
836 | { | |
837 | struct cachefiles_object *object; | |
838 | struct cachefiles_cache *cache; | |
839 | struct pagevec pagevec; | |
840 | struct page *page; | |
841 | int ret; | |
842 | ||
843 | object = container_of(op->op.object, | |
844 | struct cachefiles_object, fscache); | |
845 | cache = container_of(object->fscache.cache, | |
846 | struct cachefiles_cache, cache); | |
847 | ||
848 | _enter("%p,,,%d,", object, *nr_pages); | |
849 | ||
850 | ret = cachefiles_has_space(cache, 0, *nr_pages); | |
851 | if (ret == 0) { | |
86679820 | 852 | pagevec_init(&pagevec); |
9ae326a6 DH |
853 | |
854 | list_for_each_entry(page, pages, lru) { | |
855 | if (pagevec_add(&pagevec, page) == 0) | |
856 | fscache_mark_pages_cached(op, &pagevec); | |
857 | } | |
858 | ||
859 | if (pagevec_count(&pagevec) > 0) | |
860 | fscache_mark_pages_cached(op, &pagevec); | |
861 | ret = -ENODATA; | |
862 | } else { | |
863 | ret = -ENOBUFS; | |
864 | } | |
865 | ||
9f10523f | 866 | fscache_retrieval_complete(op, *nr_pages); |
9ae326a6 DH |
867 | _leave(" = %d", ret); |
868 | return ret; | |
869 | } | |
870 | ||
871 | /* | |
872 | * request a page be stored in the cache | |
873 | * - cache withdrawal is prevented by the caller | |
874 | * - this request may be ignored if there's no cache block available, in which | |
875 | * case -ENOBUFS will be returned | |
876 | * - if the op is in progress, 0 will be returned | |
877 | */ | |
878 | int cachefiles_write_page(struct fscache_storage *op, struct page *page) | |
879 | { | |
880 | struct cachefiles_object *object; | |
881 | struct cachefiles_cache *cache; | |
9ae326a6 | 882 | struct file *file; |
765927b2 | 883 | struct path path; |
a17754fb DH |
884 | loff_t pos, eof; |
885 | size_t len; | |
9ae326a6 | 886 | void *data; |
cf897526 | 887 | int ret = -ENOBUFS; |
9ae326a6 DH |
888 | |
889 | ASSERT(op != NULL); | |
890 | ASSERT(page != NULL); | |
891 | ||
892 | object = container_of(op->op.object, | |
893 | struct cachefiles_object, fscache); | |
894 | ||
895 | _enter("%p,%p{%lx},,,", object, page, page->index); | |
896 | ||
897 | if (!object->backer) { | |
898 | _leave(" = -ENOBUFS"); | |
899 | return -ENOBUFS; | |
900 | } | |
901 | ||
ce40fa78 | 902 | ASSERT(d_is_reg(object->backer)); |
9ae326a6 DH |
903 | |
904 | cache = container_of(object->fscache.cache, | |
905 | struct cachefiles_cache, cache); | |
906 | ||
102f4d90 DH |
907 | pos = (loff_t)page->index << PAGE_SHIFT; |
908 | ||
909 | /* We mustn't write more data than we have, so we have to beware of a | |
910 | * partial page at EOF. | |
911 | */ | |
912 | eof = object->fscache.store_limit_l; | |
913 | if (pos >= eof) | |
914 | goto error; | |
915 | ||
9ae326a6 DH |
916 | /* write the page to the backing filesystem and let it store it in its |
917 | * own time */ | |
765927b2 AV |
918 | path.mnt = cache->mnt; |
919 | path.dentry = object->backer; | |
98c350cd | 920 | file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred); |
9ae326a6 DH |
921 | if (IS_ERR(file)) { |
922 | ret = PTR_ERR(file); | |
102f4d90 | 923 | goto error_2; |
9ae326a6 DH |
924 | } |
925 | ||
102f4d90 DH |
926 | len = PAGE_SIZE; |
927 | if (eof & ~PAGE_MASK) { | |
928 | if (eof - pos < PAGE_SIZE) { | |
929 | _debug("cut short %llx to %llx", | |
930 | pos, eof); | |
931 | len = eof - pos; | |
932 | ASSERTCMP(pos + len, ==, eof); | |
933 | } | |
9ae326a6 DH |
934 | } |
935 | ||
102f4d90 DH |
936 | data = kmap(page); |
937 | ret = __kernel_write(file, data, len, &pos); | |
938 | kunmap(page); | |
939 | fput(file); | |
940 | if (ret != len) | |
941 | goto error_eio; | |
942 | ||
943 | _leave(" = 0"); | |
944 | return 0; | |
945 | ||
946 | error_eio: | |
947 | ret = -EIO; | |
948 | error_2: | |
949 | if (ret == -EIO) | |
950 | cachefiles_io_error_obj(object, | |
951 | "Write page to backing file failed"); | |
952 | error: | |
953 | _leave(" = -ENOBUFS [%d]", ret); | |
954 | return -ENOBUFS; | |
9ae326a6 DH |
955 | } |
956 | ||
957 | /* | |
958 | * detach a backing block from a page | |
959 | * - cache withdrawal is prevented by the caller | |
960 | */ | |
961 | void cachefiles_uncache_page(struct fscache_object *_object, struct page *page) | |
962 | { | |
963 | struct cachefiles_object *object; | |
964 | struct cachefiles_cache *cache; | |
965 | ||
966 | object = container_of(_object, struct cachefiles_object, fscache); | |
967 | cache = container_of(object->fscache.cache, | |
968 | struct cachefiles_cache, cache); | |
969 | ||
970 | _enter("%p,{%lu}", object, page->index); | |
971 | ||
972 | spin_unlock(&object->fscache.cookie->lock); | |
973 | } |