]>
Commit | Line | Data |
---|---|---|
b4d0d230 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
9ae326a6 DH |
2 | /* Storage object read/write |
3 | * | |
4 | * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. | |
5 | * Written by David Howells (dhowells@redhat.com) | |
9ae326a6 DH |
6 | */ |
7 | ||
8 | #include <linux/mount.h> | |
5a0e3ad6 | 9 | #include <linux/slab.h> |
9ae326a6 | 10 | #include <linux/file.h> |
a0b8cab3 | 11 | #include <linux/swap.h> |
9ae326a6 DH |
12 | #include "internal.h" |
13 | ||
14 | /* | |
15 | * detect wake up events generated by the unlocking of pages in which we're | |
16 | * interested | |
17 | * - we use this to detect read completion of backing pages | |
18 | * - the caller holds the waitqueue lock | |
19 | */ | |
ac6424b9 | 20 | static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode, |
9ae326a6 DH |
21 | int sync, void *_key) |
22 | { | |
23 | struct cachefiles_one_read *monitor = | |
24 | container_of(wait, struct cachefiles_one_read, monitor); | |
25 | struct cachefiles_object *object; | |
934140ab | 26 | struct fscache_retrieval *op = monitor->op; |
9ae326a6 DH |
27 | struct wait_bit_key *key = _key; |
28 | struct page *page = wait->private; | |
29 | ||
30 | ASSERT(key); | |
31 | ||
32 | _enter("{%lu},%u,%d,{%p,%u}", | |
33 | monitor->netfs_page->index, mode, sync, | |
34 | key->flags, key->bit_nr); | |
35 | ||
36 | if (key->flags != &page->flags || | |
37 | key->bit_nr != PG_locked) | |
38 | return 0; | |
39 | ||
40 | _debug("--- monitor %p %lx ---", page, page->flags); | |
41 | ||
5e929b33 DH |
42 | if (!PageUptodate(page) && !PageError(page)) { |
43 | /* unlocked, not uptodate and not erronous? */ | |
44 | _debug("page probably truncated"); | |
45 | } | |
9ae326a6 DH |
46 | |
47 | /* remove from the waitqueue */ | |
2055da97 | 48 | list_del(&wait->entry); |
9ae326a6 DH |
49 | |
50 | /* move onto the action list and queue for FS-Cache thread pool */ | |
934140ab | 51 | ASSERT(op); |
9ae326a6 | 52 | |
934140ab KKM |
53 | /* We need to temporarily bump the usage count as we don't own a ref |
54 | * here otherwise cachefiles_read_copier() may free the op between the | |
55 | * monitor being enqueued on the op->to_do list and the op getting | |
56 | * enqueued on the work queue. | |
57 | */ | |
58 | fscache_get_retrieval(op); | |
9ae326a6 | 59 | |
934140ab | 60 | object = container_of(op->op.object, struct cachefiles_object, fscache); |
9ae326a6 | 61 | spin_lock(&object->work_lock); |
934140ab | 62 | list_add_tail(&monitor->op_link, &op->to_do); |
7bb0c533 | 63 | fscache_enqueue_retrieval(op); |
9ae326a6 DH |
64 | spin_unlock(&object->work_lock); |
65 | ||
934140ab | 66 | fscache_put_retrieval(op); |
9ae326a6 DH |
67 | return 0; |
68 | } | |
69 | ||
5e929b33 DH |
70 | /* |
71 | * handle a probably truncated page | |
72 | * - check to see if the page is still relevant and reissue the read if | |
73 | * possible | |
74 | * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we | |
75 | * must wait again and 0 if successful | |
76 | */ | |
77 | static int cachefiles_read_reissue(struct cachefiles_object *object, | |
78 | struct cachefiles_one_read *monitor) | |
79 | { | |
466b77bc | 80 | struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping; |
5e929b33 DH |
81 | struct page *backpage = monitor->back_page, *backpage2; |
82 | int ret; | |
83 | ||
37491a13 | 84 | _enter("{ino=%lx},{%lx,%lx}", |
466b77bc | 85 | d_backing_inode(object->backer)->i_ino, |
5e929b33 DH |
86 | backpage->index, backpage->flags); |
87 | ||
88 | /* skip if the page was truncated away completely */ | |
89 | if (backpage->mapping != bmapping) { | |
37491a13 | 90 | _leave(" = -ENODATA [mapping]"); |
5e929b33 DH |
91 | return -ENODATA; |
92 | } | |
93 | ||
94 | backpage2 = find_get_page(bmapping, backpage->index); | |
95 | if (!backpage2) { | |
37491a13 | 96 | _leave(" = -ENODATA [gone]"); |
5e929b33 DH |
97 | return -ENODATA; |
98 | } | |
99 | ||
100 | if (backpage != backpage2) { | |
101 | put_page(backpage2); | |
37491a13 | 102 | _leave(" = -ENODATA [different]"); |
5e929b33 DH |
103 | return -ENODATA; |
104 | } | |
105 | ||
106 | /* the page is still there and we already have a ref on it, so we don't | |
107 | * need a second */ | |
108 | put_page(backpage2); | |
109 | ||
110 | INIT_LIST_HEAD(&monitor->op_link); | |
111 | add_page_wait_queue(backpage, &monitor->monitor); | |
112 | ||
113 | if (trylock_page(backpage)) { | |
114 | ret = -EIO; | |
115 | if (PageError(backpage)) | |
116 | goto unlock_discard; | |
117 | ret = 0; | |
118 | if (PageUptodate(backpage)) | |
119 | goto unlock_discard; | |
120 | ||
37491a13 | 121 | _debug("reissue read"); |
5e929b33 DH |
122 | ret = bmapping->a_ops->readpage(NULL, backpage); |
123 | if (ret < 0) | |
124 | goto unlock_discard; | |
125 | } | |
126 | ||
127 | /* but the page may have been read before the monitor was installed, so | |
128 | * the monitor may miss the event - so we have to ensure that we do get | |
129 | * one in such a case */ | |
130 | if (trylock_page(backpage)) { | |
131 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
132 | unlock_page(backpage); | |
133 | } | |
134 | ||
135 | /* it'll reappear on the todo list */ | |
37491a13 | 136 | _leave(" = -EINPROGRESS"); |
5e929b33 DH |
137 | return -EINPROGRESS; |
138 | ||
139 | unlock_discard: | |
140 | unlock_page(backpage); | |
141 | spin_lock_irq(&object->work_lock); | |
142 | list_del(&monitor->op_link); | |
143 | spin_unlock_irq(&object->work_lock); | |
37491a13 | 144 | _leave(" = %d", ret); |
5e929b33 DH |
145 | return ret; |
146 | } | |
147 | ||
9ae326a6 DH |
148 | /* |
149 | * copy data from backing pages to netfs pages to complete a read operation | |
150 | * - driven by FS-Cache's thread pool | |
151 | */ | |
152 | static void cachefiles_read_copier(struct fscache_operation *_op) | |
153 | { | |
154 | struct cachefiles_one_read *monitor; | |
155 | struct cachefiles_object *object; | |
156 | struct fscache_retrieval *op; | |
9ae326a6 DH |
157 | int error, max; |
158 | ||
159 | op = container_of(_op, struct fscache_retrieval, op); | |
160 | object = container_of(op->op.object, | |
161 | struct cachefiles_object, fscache); | |
162 | ||
466b77bc | 163 | _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino); |
9ae326a6 | 164 | |
9ae326a6 DH |
165 | max = 8; |
166 | spin_lock_irq(&object->work_lock); | |
167 | ||
168 | while (!list_empty(&op->to_do)) { | |
169 | monitor = list_entry(op->to_do.next, | |
170 | struct cachefiles_one_read, op_link); | |
171 | list_del(&monitor->op_link); | |
172 | ||
173 | spin_unlock_irq(&object->work_lock); | |
174 | ||
175 | _debug("- copy {%lu}", monitor->back_page->index); | |
176 | ||
5e929b33 | 177 | recheck: |
9dc8d9bf DH |
178 | if (test_bit(FSCACHE_COOKIE_INVALIDATING, |
179 | &object->fscache.cookie->flags)) { | |
180 | error = -ESTALE; | |
181 | } else if (PageUptodate(monitor->back_page)) { | |
9ae326a6 | 182 | copy_highpage(monitor->netfs_page, monitor->back_page); |
c4d6d8db DH |
183 | fscache_mark_page_cached(monitor->op, |
184 | monitor->netfs_page); | |
9ae326a6 | 185 | error = 0; |
5e929b33 DH |
186 | } else if (!PageError(monitor->back_page)) { |
187 | /* the page has probably been truncated */ | |
188 | error = cachefiles_read_reissue(object, monitor); | |
189 | if (error == -EINPROGRESS) | |
190 | goto next; | |
191 | goto recheck; | |
192 | } else { | |
9ae326a6 DH |
193 | cachefiles_io_error_obj( |
194 | object, | |
195 | "Readpage failed on backing file %lx", | |
196 | (unsigned long) monitor->back_page->flags); | |
5e929b33 DH |
197 | error = -EIO; |
198 | } | |
9ae326a6 | 199 | |
09cbfeaf | 200 | put_page(monitor->back_page); |
9ae326a6 DH |
201 | |
202 | fscache_end_io(op, monitor->netfs_page, error); | |
09cbfeaf | 203 | put_page(monitor->netfs_page); |
9f10523f | 204 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
205 | fscache_put_retrieval(op); |
206 | kfree(monitor); | |
207 | ||
5e929b33 | 208 | next: |
9ae326a6 DH |
209 | /* let the thread pool have some air occasionally */ |
210 | max--; | |
211 | if (max < 0 || need_resched()) { | |
212 | if (!list_empty(&op->to_do)) | |
213 | fscache_enqueue_retrieval(op); | |
214 | _leave(" [maxed out]"); | |
215 | return; | |
216 | } | |
217 | ||
218 | spin_lock_irq(&object->work_lock); | |
219 | } | |
220 | ||
221 | spin_unlock_irq(&object->work_lock); | |
222 | _leave(""); | |
223 | } | |
224 | ||
225 | /* | |
226 | * read the corresponding page to the given set from the backing file | |
227 | * - an uncertain page is simply discarded, to be tried again another time | |
228 | */ | |
229 | static int cachefiles_read_backing_file_one(struct cachefiles_object *object, | |
230 | struct fscache_retrieval *op, | |
a0b8cab3 | 231 | struct page *netpage) |
9ae326a6 DH |
232 | { |
233 | struct cachefiles_one_read *monitor; | |
234 | struct address_space *bmapping; | |
235 | struct page *newpage, *backpage; | |
236 | int ret; | |
237 | ||
238 | _enter(""); | |
239 | ||
9ae326a6 DH |
240 | _debug("read back %p{%lu,%d}", |
241 | netpage, netpage->index, page_count(netpage)); | |
242 | ||
5f4f9f4a | 243 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
244 | if (!monitor) |
245 | goto nomem; | |
246 | ||
247 | monitor->netfs_page = netpage; | |
248 | monitor->op = fscache_get_retrieval(op); | |
249 | ||
250 | init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter); | |
251 | ||
252 | /* attempt to get hold of the backing page */ | |
466b77bc | 253 | bmapping = d_backing_inode(object->backer)->i_mapping; |
9ae326a6 DH |
254 | newpage = NULL; |
255 | ||
256 | for (;;) { | |
257 | backpage = find_get_page(bmapping, netpage->index); | |
258 | if (backpage) | |
259 | goto backing_page_already_present; | |
260 | ||
261 | if (!newpage) { | |
453f85d4 | 262 | newpage = __page_cache_alloc(cachefiles_gfp); |
9ae326a6 DH |
263 | if (!newpage) |
264 | goto nomem_monitor; | |
265 | } | |
266 | ||
55881bc7 JW |
267 | ret = add_to_page_cache_lru(newpage, bmapping, |
268 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
269 | if (ret == 0) |
270 | goto installed_new_backing_page; | |
271 | if (ret != -EEXIST) | |
272 | goto nomem_page; | |
273 | } | |
274 | ||
55881bc7 JW |
275 | /* we've installed a new backing page, so now we need to start |
276 | * it reading */ | |
9ae326a6 DH |
277 | installed_new_backing_page: |
278 | _debug("- new %p", newpage); | |
279 | ||
280 | backpage = newpage; | |
281 | newpage = NULL; | |
282 | ||
9ae326a6 DH |
283 | read_backing_page: |
284 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
285 | if (ret < 0) | |
286 | goto read_error; | |
287 | ||
288 | /* set the monitor to transfer the data across */ | |
289 | monitor_backing_page: | |
290 | _debug("- monitor add"); | |
291 | ||
292 | /* install the monitor */ | |
09cbfeaf KS |
293 | get_page(monitor->netfs_page); |
294 | get_page(backpage); | |
9ae326a6 DH |
295 | monitor->back_page = backpage; |
296 | monitor->monitor.private = backpage; | |
297 | add_page_wait_queue(backpage, &monitor->monitor); | |
298 | monitor = NULL; | |
299 | ||
300 | /* but the page may have been read before the monitor was installed, so | |
301 | * the monitor may miss the event - so we have to ensure that we do get | |
302 | * one in such a case */ | |
303 | if (trylock_page(backpage)) { | |
304 | _debug("jumpstart %p {%lx}", backpage, backpage->flags); | |
305 | unlock_page(backpage); | |
306 | } | |
307 | goto success; | |
308 | ||
309 | /* if the backing page is already present, it can be in one of | |
310 | * three states: read in progress, read failed or read okay */ | |
311 | backing_page_already_present: | |
312 | _debug("- present"); | |
313 | ||
314 | if (newpage) { | |
09cbfeaf | 315 | put_page(newpage); |
9ae326a6 DH |
316 | newpage = NULL; |
317 | } | |
318 | ||
319 | if (PageError(backpage)) | |
320 | goto io_error; | |
321 | ||
322 | if (PageUptodate(backpage)) | |
323 | goto backing_page_already_uptodate; | |
324 | ||
325 | if (!trylock_page(backpage)) | |
326 | goto monitor_backing_page; | |
327 | _debug("read %p {%lx}", backpage, backpage->flags); | |
328 | goto read_backing_page; | |
329 | ||
330 | /* the backing page is already up to date, attach the netfs | |
331 | * page to the pagecache and LRU and copy the data across */ | |
332 | backing_page_already_uptodate: | |
333 | _debug("- uptodate"); | |
334 | ||
c4d6d8db | 335 | fscache_mark_page_cached(op, netpage); |
9ae326a6 DH |
336 | |
337 | copy_highpage(netpage, backpage); | |
338 | fscache_end_io(op, netpage, 0); | |
9f10523f | 339 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
340 | |
341 | success: | |
342 | _debug("success"); | |
343 | ret = 0; | |
344 | ||
345 | out: | |
346 | if (backpage) | |
09cbfeaf | 347 | put_page(backpage); |
9ae326a6 DH |
348 | if (monitor) { |
349 | fscache_put_retrieval(monitor->op); | |
350 | kfree(monitor); | |
351 | } | |
352 | _leave(" = %d", ret); | |
353 | return ret; | |
354 | ||
355 | read_error: | |
356 | _debug("read error %d", ret); | |
b4cf1e08 DH |
357 | if (ret == -ENOMEM) { |
358 | fscache_retrieval_complete(op, 1); | |
9ae326a6 | 359 | goto out; |
b4cf1e08 | 360 | } |
9ae326a6 DH |
361 | io_error: |
362 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
9f10523f | 363 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
364 | ret = -ENOBUFS; |
365 | goto out; | |
366 | ||
367 | nomem_page: | |
09cbfeaf | 368 | put_page(newpage); |
9ae326a6 DH |
369 | nomem_monitor: |
370 | fscache_put_retrieval(monitor->op); | |
371 | kfree(monitor); | |
372 | nomem: | |
9f10523f | 373 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
374 | _leave(" = -ENOMEM"); |
375 | return -ENOMEM; | |
376 | } | |
377 | ||
378 | /* | |
379 | * read a page from the cache or allocate a block in which to store it | |
380 | * - cache withdrawal is prevented by the caller | |
381 | * - returns -EINTR if interrupted | |
382 | * - returns -ENOMEM if ran out of memory | |
383 | * - returns -ENOBUFS if no buffers can be made available | |
384 | * - returns -ENOBUFS if page is beyond EOF | |
385 | * - if the page is backed by a block in the cache: | |
386 | * - a read will be started which will call the callback on completion | |
387 | * - 0 will be returned | |
388 | * - else if the page is unbacked: | |
389 | * - the metadata will be retained | |
390 | * - -ENODATA will be returned | |
391 | */ | |
392 | int cachefiles_read_or_alloc_page(struct fscache_retrieval *op, | |
393 | struct page *page, | |
394 | gfp_t gfp) | |
395 | { | |
396 | struct cachefiles_object *object; | |
397 | struct cachefiles_cache *cache; | |
9ae326a6 | 398 | struct inode *inode; |
10d83e11 | 399 | sector_t block; |
9ae326a6 | 400 | unsigned shift; |
c5f9d9db | 401 | int ret, ret2; |
9ae326a6 DH |
402 | |
403 | object = container_of(op->op.object, | |
404 | struct cachefiles_object, fscache); | |
405 | cache = container_of(object->fscache.cache, | |
406 | struct cachefiles_cache, cache); | |
407 | ||
408 | _enter("{%p},{%lx},,,", object, page->index); | |
409 | ||
410 | if (!object->backer) | |
9f10523f | 411 | goto enobufs; |
9ae326a6 | 412 | |
466b77bc | 413 | inode = d_backing_inode(object->backer); |
9ae326a6 | 414 | ASSERT(S_ISREG(inode->i_mode)); |
9ae326a6 DH |
415 | ASSERT(inode->i_mapping->a_ops->readpages); |
416 | ||
417 | /* calculate the shift required to use bmap */ | |
9ae326a6 DH |
418 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
419 | ||
4fbf4291 | 420 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 421 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
422 | op->op.processor = cachefiles_read_copier; |
423 | ||
9ae326a6 DH |
424 | /* we assume the absence or presence of the first block is a good |
425 | * enough indication for the page as a whole | |
426 | * - TODO: don't use bmap() for this as it is _not_ actually good | |
427 | * enough for this as it doesn't indicate errors, but it's all we've | |
428 | * got for the moment | |
429 | */ | |
10d83e11 CM |
430 | block = page->index; |
431 | block <<= shift; | |
432 | ||
c5f9d9db DH |
433 | ret2 = bmap(inode, &block); |
434 | ASSERT(ret2 == 0); | |
9ae326a6 | 435 | |
9ae326a6 | 436 | _debug("%llx -> %llx", |
10d83e11 | 437 | (unsigned long long) (page->index << shift), |
9ae326a6 DH |
438 | (unsigned long long) block); |
439 | ||
440 | if (block) { | |
441 | /* submit the apparently valid page to the backing fs to be | |
442 | * read from disk */ | |
a0b8cab3 | 443 | ret = cachefiles_read_backing_file_one(object, op, page); |
9ae326a6 DH |
444 | } else if (cachefiles_has_space(cache, 0, 1) == 0) { |
445 | /* there's space in the cache we can use */ | |
c4d6d8db | 446 | fscache_mark_page_cached(op, page); |
9f10523f | 447 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
448 | ret = -ENODATA; |
449 | } else { | |
9f10523f | 450 | goto enobufs; |
9ae326a6 DH |
451 | } |
452 | ||
453 | _leave(" = %d", ret); | |
454 | return ret; | |
9f10523f DH |
455 | |
456 | enobufs: | |
457 | fscache_retrieval_complete(op, 1); | |
458 | _leave(" = -ENOBUFS"); | |
459 | return -ENOBUFS; | |
9ae326a6 DH |
460 | } |
461 | ||
462 | /* | |
463 | * read the corresponding pages to the given set from the backing file | |
464 | * - any uncertain pages are simply discarded, to be tried again another time | |
465 | */ | |
466 | static int cachefiles_read_backing_file(struct cachefiles_object *object, | |
467 | struct fscache_retrieval *op, | |
c4d6d8db | 468 | struct list_head *list) |
9ae326a6 DH |
469 | { |
470 | struct cachefiles_one_read *monitor = NULL; | |
466b77bc | 471 | struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping; |
9ae326a6 DH |
472 | struct page *newpage = NULL, *netpage, *_n, *backpage = NULL; |
473 | int ret = 0; | |
474 | ||
475 | _enter(""); | |
476 | ||
9ae326a6 DH |
477 | list_for_each_entry_safe(netpage, _n, list, lru) { |
478 | list_del(&netpage->lru); | |
479 | ||
480 | _debug("read back %p{%lu,%d}", | |
481 | netpage, netpage->index, page_count(netpage)); | |
482 | ||
483 | if (!monitor) { | |
5f4f9f4a | 484 | monitor = kzalloc(sizeof(*monitor), cachefiles_gfp); |
9ae326a6 DH |
485 | if (!monitor) |
486 | goto nomem; | |
487 | ||
488 | monitor->op = fscache_get_retrieval(op); | |
489 | init_waitqueue_func_entry(&monitor->monitor, | |
490 | cachefiles_read_waiter); | |
491 | } | |
492 | ||
493 | for (;;) { | |
494 | backpage = find_get_page(bmapping, netpage->index); | |
495 | if (backpage) | |
496 | goto backing_page_already_present; | |
497 | ||
498 | if (!newpage) { | |
453f85d4 | 499 | newpage = __page_cache_alloc(cachefiles_gfp); |
9ae326a6 DH |
500 | if (!newpage) |
501 | goto nomem; | |
502 | } | |
503 | ||
55881bc7 JW |
504 | ret = add_to_page_cache_lru(newpage, bmapping, |
505 | netpage->index, | |
506 | cachefiles_gfp); | |
9ae326a6 DH |
507 | if (ret == 0) |
508 | goto installed_new_backing_page; | |
509 | if (ret != -EEXIST) | |
510 | goto nomem; | |
511 | } | |
512 | ||
55881bc7 JW |
513 | /* we've installed a new backing page, so now we need |
514 | * to start it reading */ | |
9ae326a6 DH |
515 | installed_new_backing_page: |
516 | _debug("- new %p", newpage); | |
517 | ||
518 | backpage = newpage; | |
519 | newpage = NULL; | |
520 | ||
9ae326a6 DH |
521 | reread_backing_page: |
522 | ret = bmapping->a_ops->readpage(NULL, backpage); | |
523 | if (ret < 0) | |
524 | goto read_error; | |
525 | ||
526 | /* add the netfs page to the pagecache and LRU, and set the | |
527 | * monitor to transfer the data across */ | |
528 | monitor_backing_page: | |
529 | _debug("- monitor add"); | |
530 | ||
55881bc7 JW |
531 | ret = add_to_page_cache_lru(netpage, op->mapping, |
532 | netpage->index, cachefiles_gfp); | |
9ae326a6 DH |
533 | if (ret < 0) { |
534 | if (ret == -EEXIST) { | |
9a24ce5b KKM |
535 | put_page(backpage); |
536 | backpage = NULL; | |
09cbfeaf | 537 | put_page(netpage); |
9a24ce5b | 538 | netpage = NULL; |
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) { | |
9a24ce5b KKM |
611 | put_page(backpage); |
612 | backpage = NULL; | |
09cbfeaf | 613 | put_page(netpage); |
9a24ce5b | 614 | netpage = NULL; |
b4cf1e08 | 615 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
616 | continue; |
617 | } | |
618 | goto nomem; | |
619 | } | |
620 | ||
621 | copy_highpage(netpage, backpage); | |
622 | ||
09cbfeaf | 623 | put_page(backpage); |
9ae326a6 DH |
624 | backpage = NULL; |
625 | ||
c4d6d8db | 626 | fscache_mark_page_cached(op, netpage); |
9ae326a6 | 627 | |
c4d6d8db | 628 | /* the netpage is unlocked and marked up to date here */ |
9ae326a6 | 629 | fscache_end_io(op, netpage, 0); |
09cbfeaf | 630 | put_page(netpage); |
9ae326a6 | 631 | netpage = NULL; |
b4cf1e08 | 632 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
633 | continue; |
634 | } | |
635 | ||
636 | netpage = NULL; | |
637 | ||
638 | _debug("out"); | |
639 | ||
640 | out: | |
641 | /* tidy up */ | |
9ae326a6 | 642 | if (newpage) |
09cbfeaf | 643 | put_page(newpage); |
9ae326a6 | 644 | if (netpage) |
09cbfeaf | 645 | put_page(netpage); |
9ae326a6 | 646 | if (backpage) |
09cbfeaf | 647 | put_page(backpage); |
9ae326a6 DH |
648 | if (monitor) { |
649 | fscache_put_retrieval(op); | |
650 | kfree(monitor); | |
651 | } | |
652 | ||
653 | list_for_each_entry_safe(netpage, _n, list, lru) { | |
654 | list_del(&netpage->lru); | |
09cbfeaf | 655 | put_page(netpage); |
9f10523f | 656 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
657 | } |
658 | ||
659 | _leave(" = %d", ret); | |
660 | return ret; | |
661 | ||
662 | nomem: | |
663 | _debug("nomem"); | |
664 | ret = -ENOMEM; | |
b4cf1e08 | 665 | goto record_page_complete; |
9ae326a6 DH |
666 | |
667 | read_error: | |
668 | _debug("read error %d", ret); | |
669 | if (ret == -ENOMEM) | |
b4cf1e08 | 670 | goto record_page_complete; |
9ae326a6 DH |
671 | io_error: |
672 | cachefiles_io_error_obj(object, "Page read error on backing file"); | |
673 | ret = -ENOBUFS; | |
b4cf1e08 DH |
674 | record_page_complete: |
675 | fscache_retrieval_complete(op, 1); | |
9ae326a6 DH |
676 | goto out; |
677 | } | |
678 | ||
679 | /* | |
680 | * read a list of pages from the cache or allocate blocks in which to store | |
681 | * them | |
682 | */ | |
683 | int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op, | |
684 | struct list_head *pages, | |
685 | unsigned *nr_pages, | |
686 | gfp_t gfp) | |
687 | { | |
688 | struct cachefiles_object *object; | |
689 | struct cachefiles_cache *cache; | |
690 | struct list_head backpages; | |
691 | struct pagevec pagevec; | |
692 | struct inode *inode; | |
693 | struct page *page, *_n; | |
694 | unsigned shift, nrbackpages; | |
695 | int ret, ret2, space; | |
696 | ||
697 | object = container_of(op->op.object, | |
698 | struct cachefiles_object, fscache); | |
699 | cache = container_of(object->fscache.cache, | |
700 | struct cachefiles_cache, cache); | |
701 | ||
702 | _enter("{OBJ%x,%d},,%d,,", | |
703 | object->fscache.debug_id, atomic_read(&op->op.usage), | |
704 | *nr_pages); | |
705 | ||
706 | if (!object->backer) | |
9f10523f | 707 | goto all_enobufs; |
9ae326a6 DH |
708 | |
709 | space = 1; | |
710 | if (cachefiles_has_space(cache, 0, *nr_pages) < 0) | |
711 | space = 0; | |
712 | ||
466b77bc | 713 | inode = d_backing_inode(object->backer); |
9ae326a6 | 714 | ASSERT(S_ISREG(inode->i_mode)); |
9ae326a6 DH |
715 | ASSERT(inode->i_mapping->a_ops->readpages); |
716 | ||
717 | /* calculate the shift required to use bmap */ | |
9ae326a6 DH |
718 | shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits; |
719 | ||
86679820 | 720 | pagevec_init(&pagevec); |
9ae326a6 | 721 | |
4fbf4291 | 722 | op->op.flags &= FSCACHE_OP_KEEP_FLAGS; |
8af7c124 | 723 | op->op.flags |= FSCACHE_OP_ASYNC; |
9ae326a6 DH |
724 | op->op.processor = cachefiles_read_copier; |
725 | ||
726 | INIT_LIST_HEAD(&backpages); | |
727 | nrbackpages = 0; | |
728 | ||
729 | ret = space ? -ENODATA : -ENOBUFS; | |
730 | list_for_each_entry_safe(page, _n, pages, lru) { | |
10d83e11 | 731 | sector_t block; |
9ae326a6 DH |
732 | |
733 | /* we assume the absence or presence of the first block is a | |
734 | * good enough indication for the page as a whole | |
735 | * - TODO: don't use bmap() for this as it is _not_ actually | |
736 | * good enough for this as it doesn't indicate errors, but | |
737 | * it's all we've got for the moment | |
738 | */ | |
10d83e11 CM |
739 | block = page->index; |
740 | block <<= shift; | |
741 | ||
c5f9d9db DH |
742 | ret2 = bmap(inode, &block); |
743 | ASSERT(ret2 == 0); | |
9ae326a6 | 744 | |
9ae326a6 | 745 | _debug("%llx -> %llx", |
10d83e11 | 746 | (unsigned long long) (page->index << shift), |
9ae326a6 DH |
747 | (unsigned long long) block); |
748 | ||
749 | if (block) { | |
750 | /* we have data - add it to the list to give to the | |
751 | * backing fs */ | |
752 | list_move(&page->lru, &backpages); | |
753 | (*nr_pages)--; | |
754 | nrbackpages++; | |
755 | } else if (space && pagevec_add(&pagevec, page) == 0) { | |
756 | fscache_mark_pages_cached(op, &pagevec); | |
9f10523f | 757 | fscache_retrieval_complete(op, 1); |
9ae326a6 | 758 | ret = -ENODATA; |
9f10523f DH |
759 | } else { |
760 | fscache_retrieval_complete(op, 1); | |
9ae326a6 DH |
761 | } |
762 | } | |
763 | ||
764 | if (pagevec_count(&pagevec) > 0) | |
765 | fscache_mark_pages_cached(op, &pagevec); | |
766 | ||
767 | if (list_empty(pages)) | |
768 | ret = 0; | |
769 | ||
770 | /* submit the apparently valid pages to the backing fs to be read from | |
771 | * disk */ | |
772 | if (nrbackpages > 0) { | |
c4d6d8db | 773 | ret2 = cachefiles_read_backing_file(object, op, &backpages); |
9ae326a6 DH |
774 | if (ret2 == -ENOMEM || ret2 == -EINTR) |
775 | ret = ret2; | |
776 | } | |
777 | ||
9ae326a6 DH |
778 | _leave(" = %d [nr=%u%s]", |
779 | ret, *nr_pages, list_empty(pages) ? " empty" : ""); | |
780 | return ret; | |
9f10523f DH |
781 | |
782 | all_enobufs: | |
783 | fscache_retrieval_complete(op, *nr_pages); | |
784 | return -ENOBUFS; | |
9ae326a6 DH |
785 | } |
786 | ||
787 | /* | |
788 | * allocate a block in the cache in which to store a page | |
789 | * - cache withdrawal is prevented by the caller | |
790 | * - returns -EINTR if interrupted | |
791 | * - returns -ENOMEM if ran out of memory | |
792 | * - returns -ENOBUFS if no buffers can be made available | |
793 | * - returns -ENOBUFS if page is beyond EOF | |
794 | * - otherwise: | |
795 | * - the metadata will be retained | |
796 | * - 0 will be returned | |
797 | */ | |
798 | int cachefiles_allocate_page(struct fscache_retrieval *op, | |
799 | struct page *page, | |
800 | gfp_t gfp) | |
801 | { | |
802 | struct cachefiles_object *object; | |
803 | struct cachefiles_cache *cache; | |
9ae326a6 DH |
804 | int ret; |
805 | ||
806 | object = container_of(op->op.object, | |
807 | struct cachefiles_object, fscache); | |
808 | cache = container_of(object->fscache.cache, | |
809 | struct cachefiles_cache, cache); | |
810 | ||
811 | _enter("%p,{%lx},", object, page->index); | |
812 | ||
813 | ret = cachefiles_has_space(cache, 0, 1); | |
c4d6d8db DH |
814 | if (ret == 0) |
815 | fscache_mark_page_cached(op, page); | |
816 | else | |
9ae326a6 | 817 | ret = -ENOBUFS; |
9ae326a6 | 818 | |
9f10523f | 819 | fscache_retrieval_complete(op, 1); |
9ae326a6 DH |
820 | _leave(" = %d", ret); |
821 | return ret; | |
822 | } | |
823 | ||
824 | /* | |
825 | * allocate blocks in the cache in which to store a set of pages | |
826 | * - cache withdrawal is prevented by the caller | |
827 | * - returns -EINTR if interrupted | |
828 | * - returns -ENOMEM if ran out of memory | |
829 | * - returns -ENOBUFS if some buffers couldn't be made available | |
830 | * - returns -ENOBUFS if some pages are beyond EOF | |
831 | * - otherwise: | |
832 | * - -ENODATA will be returned | |
833 | * - metadata will be retained for any page marked | |
834 | */ | |
835 | int cachefiles_allocate_pages(struct fscache_retrieval *op, | |
836 | struct list_head *pages, | |
837 | unsigned *nr_pages, | |
838 | gfp_t gfp) | |
839 | { | |
840 | struct cachefiles_object *object; | |
841 | struct cachefiles_cache *cache; | |
842 | struct pagevec pagevec; | |
843 | struct page *page; | |
844 | int ret; | |
845 | ||
846 | object = container_of(op->op.object, | |
847 | struct cachefiles_object, fscache); | |
848 | cache = container_of(object->fscache.cache, | |
849 | struct cachefiles_cache, cache); | |
850 | ||
851 | _enter("%p,,,%d,", object, *nr_pages); | |
852 | ||
853 | ret = cachefiles_has_space(cache, 0, *nr_pages); | |
854 | if (ret == 0) { | |
86679820 | 855 | pagevec_init(&pagevec); |
9ae326a6 DH |
856 | |
857 | list_for_each_entry(page, pages, lru) { | |
858 | if (pagevec_add(&pagevec, page) == 0) | |
859 | fscache_mark_pages_cached(op, &pagevec); | |
860 | } | |
861 | ||
862 | if (pagevec_count(&pagevec) > 0) | |
863 | fscache_mark_pages_cached(op, &pagevec); | |
864 | ret = -ENODATA; | |
865 | } else { | |
866 | ret = -ENOBUFS; | |
867 | } | |
868 | ||
9f10523f | 869 | fscache_retrieval_complete(op, *nr_pages); |
9ae326a6 DH |
870 | _leave(" = %d", ret); |
871 | return ret; | |
872 | } | |
873 | ||
874 | /* | |
875 | * request a page be stored in the cache | |
876 | * - cache withdrawal is prevented by the caller | |
877 | * - this request may be ignored if there's no cache block available, in which | |
878 | * case -ENOBUFS will be returned | |
879 | * - if the op is in progress, 0 will be returned | |
880 | */ | |
881 | int cachefiles_write_page(struct fscache_storage *op, struct page *page) | |
882 | { | |
883 | struct cachefiles_object *object; | |
884 | struct cachefiles_cache *cache; | |
9ae326a6 | 885 | struct file *file; |
765927b2 | 886 | struct path path; |
a17754fb DH |
887 | loff_t pos, eof; |
888 | size_t len; | |
9ae326a6 | 889 | void *data; |
cf897526 | 890 | int ret = -ENOBUFS; |
9ae326a6 DH |
891 | |
892 | ASSERT(op != NULL); | |
893 | ASSERT(page != NULL); | |
894 | ||
895 | object = container_of(op->op.object, | |
896 | struct cachefiles_object, fscache); | |
897 | ||
898 | _enter("%p,%p{%lx},,,", object, page, page->index); | |
899 | ||
900 | if (!object->backer) { | |
901 | _leave(" = -ENOBUFS"); | |
902 | return -ENOBUFS; | |
903 | } | |
904 | ||
ce40fa78 | 905 | ASSERT(d_is_reg(object->backer)); |
9ae326a6 DH |
906 | |
907 | cache = container_of(object->fscache.cache, | |
908 | struct cachefiles_cache, cache); | |
909 | ||
102f4d90 DH |
910 | pos = (loff_t)page->index << PAGE_SHIFT; |
911 | ||
912 | /* We mustn't write more data than we have, so we have to beware of a | |
913 | * partial page at EOF. | |
914 | */ | |
915 | eof = object->fscache.store_limit_l; | |
916 | if (pos >= eof) | |
917 | goto error; | |
918 | ||
9ae326a6 DH |
919 | /* write the page to the backing filesystem and let it store it in its |
920 | * own time */ | |
765927b2 AV |
921 | path.mnt = cache->mnt; |
922 | path.dentry = object->backer; | |
98c350cd | 923 | file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred); |
9ae326a6 DH |
924 | if (IS_ERR(file)) { |
925 | ret = PTR_ERR(file); | |
102f4d90 | 926 | goto error_2; |
9ae326a6 DH |
927 | } |
928 | ||
102f4d90 DH |
929 | len = PAGE_SIZE; |
930 | if (eof & ~PAGE_MASK) { | |
931 | if (eof - pos < PAGE_SIZE) { | |
932 | _debug("cut short %llx to %llx", | |
933 | pos, eof); | |
934 | len = eof - pos; | |
935 | ASSERTCMP(pos + len, ==, eof); | |
936 | } | |
9ae326a6 DH |
937 | } |
938 | ||
102f4d90 DH |
939 | data = kmap(page); |
940 | ret = __kernel_write(file, data, len, &pos); | |
941 | kunmap(page); | |
942 | fput(file); | |
943 | if (ret != len) | |
944 | goto error_eio; | |
945 | ||
946 | _leave(" = 0"); | |
947 | return 0; | |
948 | ||
949 | error_eio: | |
950 | ret = -EIO; | |
951 | error_2: | |
952 | if (ret == -EIO) | |
953 | cachefiles_io_error_obj(object, | |
954 | "Write page to backing file failed"); | |
955 | error: | |
956 | _leave(" = -ENOBUFS [%d]", ret); | |
957 | return -ENOBUFS; | |
9ae326a6 DH |
958 | } |
959 | ||
960 | /* | |
961 | * detach a backing block from a page | |
962 | * - cache withdrawal is prevented by the caller | |
963 | */ | |
964 | void cachefiles_uncache_page(struct fscache_object *_object, struct page *page) | |
bfa3837e | 965 | __releases(&object->fscache.cookie->lock) |
9ae326a6 DH |
966 | { |
967 | struct cachefiles_object *object; | |
9ae326a6 DH |
968 | |
969 | object = container_of(_object, struct cachefiles_object, fscache); | |
9ae326a6 DH |
970 | |
971 | _enter("%p,{%lu}", object, page->index); | |
972 | ||
973 | spin_unlock(&object->fscache.cookie->lock); | |
974 | } |