• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Storage object read/write
2  *
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public Licence
8  * as published by the Free Software Foundation; either version
9  * 2 of the Licence, or (at your option) any later version.
10  */
11 
12 #include <linux/mount.h>
13 #include <linux/slab.h>
14 #include <linux/file.h>
15 #include <linux/swap.h>
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  */
cachefiles_read_waiter(wait_queue_entry_t * wait,unsigned mode,int sync,void * _key)24 static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode,
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 fscache_retrieval *op = monitor->op;
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 
46 	if (!PageUptodate(page) && !PageError(page)) {
47 		/* unlocked, not uptodate and not erronous? */
48 		_debug("page probably truncated");
49 	}
50 
51 	/* remove from the waitqueue */
52 	list_del(&wait->entry);
53 
54 	/* move onto the action list and queue for FS-Cache thread pool */
55 	ASSERT(op);
56 
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);
63 
64 	object = container_of(op->op.object, struct cachefiles_object, fscache);
65 	spin_lock(&object->work_lock);
66 	list_add_tail(&monitor->op_link, &op->to_do);
67 	spin_unlock(&object->work_lock);
68 
69 	fscache_enqueue_retrieval(op);
70 	fscache_put_retrieval(op);
71 	return 0;
72 }
73 
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  */
cachefiles_read_reissue(struct cachefiles_object * object,struct cachefiles_one_read * monitor)81 static int cachefiles_read_reissue(struct cachefiles_object *object,
82 				   struct cachefiles_one_read *monitor)
83 {
84 	struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
85 	struct page *backpage = monitor->back_page, *backpage2;
86 	int ret;
87 
88 	_enter("{ino=%lx},{%lx,%lx}",
89 	       d_backing_inode(object->backer)->i_ino,
90 	       backpage->index, backpage->flags);
91 
92 	/* skip if the page was truncated away completely */
93 	if (backpage->mapping != bmapping) {
94 		_leave(" = -ENODATA [mapping]");
95 		return -ENODATA;
96 	}
97 
98 	backpage2 = find_get_page(bmapping, backpage->index);
99 	if (!backpage2) {
100 		_leave(" = -ENODATA [gone]");
101 		return -ENODATA;
102 	}
103 
104 	if (backpage != backpage2) {
105 		put_page(backpage2);
106 		_leave(" = -ENODATA [different]");
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 
125 		_debug("reissue read");
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 */
140 	_leave(" = -EINPROGRESS");
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);
148 	_leave(" = %d", ret);
149 	return ret;
150 }
151 
152 /*
153  * copy data from backing pages to netfs pages to complete a read operation
154  * - driven by FS-Cache's thread pool
155  */
cachefiles_read_copier(struct fscache_operation * _op)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;
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 
167 	_enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
168 
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 
181 	recheck:
182 		if (test_bit(FSCACHE_COOKIE_INVALIDATING,
183 			     &object->fscache.cookie->flags)) {
184 			error = -ESTALE;
185 		} else if (PageUptodate(monitor->back_page)) {
186 			copy_highpage(monitor->netfs_page, monitor->back_page);
187 			fscache_mark_page_cached(monitor->op,
188 						 monitor->netfs_page);
189 			error = 0;
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 {
197 			cachefiles_io_error_obj(
198 				object,
199 				"Readpage failed on backing file %lx",
200 				(unsigned long) monitor->back_page->flags);
201 			error = -EIO;
202 		}
203 
204 		put_page(monitor->back_page);
205 
206 		fscache_end_io(op, monitor->netfs_page, error);
207 		put_page(monitor->netfs_page);
208 		fscache_retrieval_complete(op, 1);
209 		fscache_put_retrieval(op);
210 		kfree(monitor);
211 
212 	next:
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  */
cachefiles_read_backing_file_one(struct cachefiles_object * object,struct fscache_retrieval * op,struct page * netpage)233 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
234 					    struct fscache_retrieval *op,
235 					    struct page *netpage)
236 {
237 	struct cachefiles_one_read *monitor;
238 	struct address_space *bmapping;
239 	struct page *newpage, *backpage;
240 	int ret;
241 
242 	_enter("");
243 
244 	_debug("read back %p{%lu,%d}",
245 	       netpage, netpage->index, page_count(netpage));
246 
247 	monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
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 */
257 	bmapping = d_backing_inode(object->backer)->i_mapping;
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) {
266 			newpage = __page_cache_alloc(cachefiles_gfp |
267 						     __GFP_COLD);
268 			if (!newpage)
269 				goto nomem_monitor;
270 		}
271 
272 		ret = add_to_page_cache_lru(newpage, bmapping,
273 					    netpage->index, cachefiles_gfp);
274 		if (ret == 0)
275 			goto installed_new_backing_page;
276 		if (ret != -EEXIST)
277 			goto nomem_page;
278 	}
279 
280 	/* we've installed a new backing page, so now we need to start
281 	 * it reading */
282 installed_new_backing_page:
283 	_debug("- new %p", newpage);
284 
285 	backpage = newpage;
286 	newpage = NULL;
287 
288 read_backing_page:
289 	ret = bmapping->a_ops->readpage(NULL, backpage);
290 	if (ret < 0)
291 		goto read_error;
292 
293 	/* set the monitor to transfer the data across */
294 monitor_backing_page:
295 	_debug("- monitor add");
296 
297 	/* install the monitor */
298 	get_page(monitor->netfs_page);
299 	get_page(backpage);
300 	monitor->back_page = backpage;
301 	monitor->monitor.private = backpage;
302 	add_page_wait_queue(backpage, &monitor->monitor);
303 	monitor = NULL;
304 
305 	/* but the page may have been read before the monitor was installed, so
306 	 * the monitor may miss the event - so we have to ensure that we do get
307 	 * one in such a case */
308 	if (trylock_page(backpage)) {
309 		_debug("jumpstart %p {%lx}", backpage, backpage->flags);
310 		unlock_page(backpage);
311 	}
312 	goto success;
313 
314 	/* if the backing page is already present, it can be in one of
315 	 * three states: read in progress, read failed or read okay */
316 backing_page_already_present:
317 	_debug("- present");
318 
319 	if (newpage) {
320 		put_page(newpage);
321 		newpage = NULL;
322 	}
323 
324 	if (PageError(backpage))
325 		goto io_error;
326 
327 	if (PageUptodate(backpage))
328 		goto backing_page_already_uptodate;
329 
330 	if (!trylock_page(backpage))
331 		goto monitor_backing_page;
332 	_debug("read %p {%lx}", backpage, backpage->flags);
333 	goto read_backing_page;
334 
335 	/* the backing page is already up to date, attach the netfs
336 	 * page to the pagecache and LRU and copy the data across */
337 backing_page_already_uptodate:
338 	_debug("- uptodate");
339 
340 	fscache_mark_page_cached(op, netpage);
341 
342 	copy_highpage(netpage, backpage);
343 	fscache_end_io(op, netpage, 0);
344 	fscache_retrieval_complete(op, 1);
345 
346 success:
347 	_debug("success");
348 	ret = 0;
349 
350 out:
351 	if (backpage)
352 		put_page(backpage);
353 	if (monitor) {
354 		fscache_put_retrieval(monitor->op);
355 		kfree(monitor);
356 	}
357 	_leave(" = %d", ret);
358 	return ret;
359 
360 read_error:
361 	_debug("read error %d", ret);
362 	if (ret == -ENOMEM) {
363 		fscache_retrieval_complete(op, 1);
364 		goto out;
365 	}
366 io_error:
367 	cachefiles_io_error_obj(object, "Page read error on backing file");
368 	fscache_retrieval_complete(op, 1);
369 	ret = -ENOBUFS;
370 	goto out;
371 
372 nomem_page:
373 	put_page(newpage);
374 nomem_monitor:
375 	fscache_put_retrieval(monitor->op);
376 	kfree(monitor);
377 nomem:
378 	fscache_retrieval_complete(op, 1);
379 	_leave(" = -ENOMEM");
380 	return -ENOMEM;
381 }
382 
383 /*
384  * read a page from the cache or allocate a block in which to store it
385  * - cache withdrawal is prevented by the caller
386  * - returns -EINTR if interrupted
387  * - returns -ENOMEM if ran out of memory
388  * - returns -ENOBUFS if no buffers can be made available
389  * - returns -ENOBUFS if page is beyond EOF
390  * - if the page is backed by a block in the cache:
391  *   - a read will be started which will call the callback on completion
392  *   - 0 will be returned
393  * - else if the page is unbacked:
394  *   - the metadata will be retained
395  *   - -ENODATA will be returned
396  */
cachefiles_read_or_alloc_page(struct fscache_retrieval * op,struct page * page,gfp_t gfp)397 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
398 				  struct page *page,
399 				  gfp_t gfp)
400 {
401 	struct cachefiles_object *object;
402 	struct cachefiles_cache *cache;
403 	struct inode *inode;
404 	sector_t block0, block;
405 	unsigned shift;
406 	int ret;
407 
408 	object = container_of(op->op.object,
409 			      struct cachefiles_object, fscache);
410 	cache = container_of(object->fscache.cache,
411 			     struct cachefiles_cache, cache);
412 
413 	_enter("{%p},{%lx},,,", object, page->index);
414 
415 	if (!object->backer)
416 		goto enobufs;
417 
418 	inode = d_backing_inode(object->backer);
419 	ASSERT(S_ISREG(inode->i_mode));
420 	ASSERT(inode->i_mapping->a_ops->bmap);
421 	ASSERT(inode->i_mapping->a_ops->readpages);
422 
423 	/* calculate the shift required to use bmap */
424 	shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
425 
426 	op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
427 	op->op.flags |= FSCACHE_OP_ASYNC;
428 	op->op.processor = cachefiles_read_copier;
429 
430 	/* we assume the absence or presence of the first block is a good
431 	 * enough indication for the page as a whole
432 	 * - TODO: don't use bmap() for this as it is _not_ actually good
433 	 *   enough for this as it doesn't indicate errors, but it's all we've
434 	 *   got for the moment
435 	 */
436 	block0 = page->index;
437 	block0 <<= shift;
438 
439 	block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
440 	_debug("%llx -> %llx",
441 	       (unsigned long long) block0,
442 	       (unsigned long long) block);
443 
444 	if (block) {
445 		/* submit the apparently valid page to the backing fs to be
446 		 * read from disk */
447 		ret = cachefiles_read_backing_file_one(object, op, page);
448 	} else if (cachefiles_has_space(cache, 0, 1) == 0) {
449 		/* there's space in the cache we can use */
450 		fscache_mark_page_cached(op, page);
451 		fscache_retrieval_complete(op, 1);
452 		ret = -ENODATA;
453 	} else {
454 		goto enobufs;
455 	}
456 
457 	_leave(" = %d", ret);
458 	return ret;
459 
460 enobufs:
461 	fscache_retrieval_complete(op, 1);
462 	_leave(" = -ENOBUFS");
463 	return -ENOBUFS;
464 }
465 
466 /*
467  * read the corresponding pages to the given set from the backing file
468  * - any uncertain pages are simply discarded, to be tried again another time
469  */
cachefiles_read_backing_file(struct cachefiles_object * object,struct fscache_retrieval * op,struct list_head * list)470 static int cachefiles_read_backing_file(struct cachefiles_object *object,
471 					struct fscache_retrieval *op,
472 					struct list_head *list)
473 {
474 	struct cachefiles_one_read *monitor = NULL;
475 	struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
476 	struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
477 	int ret = 0;
478 
479 	_enter("");
480 
481 	list_for_each_entry_safe(netpage, _n, list, lru) {
482 		list_del(&netpage->lru);
483 
484 		_debug("read back %p{%lu,%d}",
485 		       netpage, netpage->index, page_count(netpage));
486 
487 		if (!monitor) {
488 			monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
489 			if (!monitor)
490 				goto nomem;
491 
492 			monitor->op = fscache_get_retrieval(op);
493 			init_waitqueue_func_entry(&monitor->monitor,
494 						  cachefiles_read_waiter);
495 		}
496 
497 		for (;;) {
498 			backpage = find_get_page(bmapping, netpage->index);
499 			if (backpage)
500 				goto backing_page_already_present;
501 
502 			if (!newpage) {
503 				newpage = __page_cache_alloc(cachefiles_gfp |
504 							     __GFP_COLD);
505 				if (!newpage)
506 					goto nomem;
507 			}
508 
509 			ret = add_to_page_cache_lru(newpage, bmapping,
510 						    netpage->index,
511 						    cachefiles_gfp);
512 			if (ret == 0)
513 				goto installed_new_backing_page;
514 			if (ret != -EEXIST)
515 				goto nomem;
516 		}
517 
518 		/* we've installed a new backing page, so now we need
519 		 * to start it reading */
520 	installed_new_backing_page:
521 		_debug("- new %p", newpage);
522 
523 		backpage = newpage;
524 		newpage = NULL;
525 
526 	reread_backing_page:
527 		ret = bmapping->a_ops->readpage(NULL, backpage);
528 		if (ret < 0)
529 			goto read_error;
530 
531 		/* add the netfs page to the pagecache and LRU, and set the
532 		 * monitor to transfer the data across */
533 	monitor_backing_page:
534 		_debug("- monitor add");
535 
536 		ret = add_to_page_cache_lru(netpage, op->mapping,
537 					    netpage->index, cachefiles_gfp);
538 		if (ret < 0) {
539 			if (ret == -EEXIST) {
540 				put_page(backpage);
541 				backpage = NULL;
542 				put_page(netpage);
543 				netpage = NULL;
544 				fscache_retrieval_complete(op, 1);
545 				continue;
546 			}
547 			goto nomem;
548 		}
549 
550 		/* install a monitor */
551 		get_page(netpage);
552 		monitor->netfs_page = netpage;
553 
554 		get_page(backpage);
555 		monitor->back_page = backpage;
556 		monitor->monitor.private = backpage;
557 		add_page_wait_queue(backpage, &monitor->monitor);
558 		monitor = NULL;
559 
560 		/* but the page may have been read before the monitor was
561 		 * installed, so the monitor may miss the event - so we have to
562 		 * ensure that we do get one in such a case */
563 		if (trylock_page(backpage)) {
564 			_debug("2unlock %p {%lx}", backpage, backpage->flags);
565 			unlock_page(backpage);
566 		}
567 
568 		put_page(backpage);
569 		backpage = NULL;
570 
571 		put_page(netpage);
572 		netpage = NULL;
573 		continue;
574 
575 		/* if the backing page is already present, it can be in one of
576 		 * three states: read in progress, read failed or read okay */
577 	backing_page_already_present:
578 		_debug("- present %p", backpage);
579 
580 		if (PageError(backpage))
581 			goto io_error;
582 
583 		if (PageUptodate(backpage))
584 			goto backing_page_already_uptodate;
585 
586 		_debug("- not ready %p{%lx}", backpage, backpage->flags);
587 
588 		if (!trylock_page(backpage))
589 			goto monitor_backing_page;
590 
591 		if (PageError(backpage)) {
592 			_debug("error %lx", backpage->flags);
593 			unlock_page(backpage);
594 			goto io_error;
595 		}
596 
597 		if (PageUptodate(backpage))
598 			goto backing_page_already_uptodate_unlock;
599 
600 		/* we've locked a page that's neither up to date nor erroneous,
601 		 * so we need to attempt to read it again */
602 		goto reread_backing_page;
603 
604 		/* the backing page is already up to date, attach the netfs
605 		 * page to the pagecache and LRU and copy the data across */
606 	backing_page_already_uptodate_unlock:
607 		_debug("uptodate %lx", backpage->flags);
608 		unlock_page(backpage);
609 	backing_page_already_uptodate:
610 		_debug("- uptodate");
611 
612 		ret = add_to_page_cache_lru(netpage, op->mapping,
613 					    netpage->index, cachefiles_gfp);
614 		if (ret < 0) {
615 			if (ret == -EEXIST) {
616 				put_page(backpage);
617 				backpage = NULL;
618 				put_page(netpage);
619 				netpage = NULL;
620 				fscache_retrieval_complete(op, 1);
621 				continue;
622 			}
623 			goto nomem;
624 		}
625 
626 		copy_highpage(netpage, backpage);
627 
628 		put_page(backpage);
629 		backpage = NULL;
630 
631 		fscache_mark_page_cached(op, netpage);
632 
633 		/* the netpage is unlocked and marked up to date here */
634 		fscache_end_io(op, netpage, 0);
635 		put_page(netpage);
636 		netpage = NULL;
637 		fscache_retrieval_complete(op, 1);
638 		continue;
639 	}
640 
641 	netpage = NULL;
642 
643 	_debug("out");
644 
645 out:
646 	/* tidy up */
647 	if (newpage)
648 		put_page(newpage);
649 	if (netpage)
650 		put_page(netpage);
651 	if (backpage)
652 		put_page(backpage);
653 	if (monitor) {
654 		fscache_put_retrieval(op);
655 		kfree(monitor);
656 	}
657 
658 	list_for_each_entry_safe(netpage, _n, list, lru) {
659 		list_del(&netpage->lru);
660 		put_page(netpage);
661 		fscache_retrieval_complete(op, 1);
662 	}
663 
664 	_leave(" = %d", ret);
665 	return ret;
666 
667 nomem:
668 	_debug("nomem");
669 	ret = -ENOMEM;
670 	goto record_page_complete;
671 
672 read_error:
673 	_debug("read error %d", ret);
674 	if (ret == -ENOMEM)
675 		goto record_page_complete;
676 io_error:
677 	cachefiles_io_error_obj(object, "Page read error on backing file");
678 	ret = -ENOBUFS;
679 record_page_complete:
680 	fscache_retrieval_complete(op, 1);
681 	goto out;
682 }
683 
684 /*
685  * read a list of pages from the cache or allocate blocks in which to store
686  * them
687  */
cachefiles_read_or_alloc_pages(struct fscache_retrieval * op,struct list_head * pages,unsigned * nr_pages,gfp_t gfp)688 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
689 				   struct list_head *pages,
690 				   unsigned *nr_pages,
691 				   gfp_t gfp)
692 {
693 	struct cachefiles_object *object;
694 	struct cachefiles_cache *cache;
695 	struct list_head backpages;
696 	struct pagevec pagevec;
697 	struct inode *inode;
698 	struct page *page, *_n;
699 	unsigned shift, nrbackpages;
700 	int ret, ret2, space;
701 
702 	object = container_of(op->op.object,
703 			      struct cachefiles_object, fscache);
704 	cache = container_of(object->fscache.cache,
705 			     struct cachefiles_cache, cache);
706 
707 	_enter("{OBJ%x,%d},,%d,,",
708 	       object->fscache.debug_id, atomic_read(&op->op.usage),
709 	       *nr_pages);
710 
711 	if (!object->backer)
712 		goto all_enobufs;
713 
714 	space = 1;
715 	if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
716 		space = 0;
717 
718 	inode = d_backing_inode(object->backer);
719 	ASSERT(S_ISREG(inode->i_mode));
720 	ASSERT(inode->i_mapping->a_ops->bmap);
721 	ASSERT(inode->i_mapping->a_ops->readpages);
722 
723 	/* calculate the shift required to use bmap */
724 	shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
725 
726 	pagevec_init(&pagevec, 0);
727 
728 	op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
729 	op->op.flags |= FSCACHE_OP_ASYNC;
730 	op->op.processor = cachefiles_read_copier;
731 
732 	INIT_LIST_HEAD(&backpages);
733 	nrbackpages = 0;
734 
735 	ret = space ? -ENODATA : -ENOBUFS;
736 	list_for_each_entry_safe(page, _n, pages, lru) {
737 		sector_t block0, block;
738 
739 		/* we assume the absence or presence of the first block is a
740 		 * good enough indication for the page as a whole
741 		 * - TODO: don't use bmap() for this as it is _not_ actually
742 		 *   good enough for this as it doesn't indicate errors, but
743 		 *   it's all we've got for the moment
744 		 */
745 		block0 = page->index;
746 		block0 <<= shift;
747 
748 		block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
749 						      block0);
750 		_debug("%llx -> %llx",
751 		       (unsigned long long) block0,
752 		       (unsigned long long) block);
753 
754 		if (block) {
755 			/* we have data - add it to the list to give to the
756 			 * backing fs */
757 			list_move(&page->lru, &backpages);
758 			(*nr_pages)--;
759 			nrbackpages++;
760 		} else if (space && pagevec_add(&pagevec, page) == 0) {
761 			fscache_mark_pages_cached(op, &pagevec);
762 			fscache_retrieval_complete(op, 1);
763 			ret = -ENODATA;
764 		} else {
765 			fscache_retrieval_complete(op, 1);
766 		}
767 	}
768 
769 	if (pagevec_count(&pagevec) > 0)
770 		fscache_mark_pages_cached(op, &pagevec);
771 
772 	if (list_empty(pages))
773 		ret = 0;
774 
775 	/* submit the apparently valid pages to the backing fs to be read from
776 	 * disk */
777 	if (nrbackpages > 0) {
778 		ret2 = cachefiles_read_backing_file(object, op, &backpages);
779 		if (ret2 == -ENOMEM || ret2 == -EINTR)
780 			ret = ret2;
781 	}
782 
783 	_leave(" = %d [nr=%u%s]",
784 	       ret, *nr_pages, list_empty(pages) ? " empty" : "");
785 	return ret;
786 
787 all_enobufs:
788 	fscache_retrieval_complete(op, *nr_pages);
789 	return -ENOBUFS;
790 }
791 
792 /*
793  * allocate a block in the cache in which to store a page
794  * - cache withdrawal is prevented by the caller
795  * - returns -EINTR if interrupted
796  * - returns -ENOMEM if ran out of memory
797  * - returns -ENOBUFS if no buffers can be made available
798  * - returns -ENOBUFS if page is beyond EOF
799  * - otherwise:
800  *   - the metadata will be retained
801  *   - 0 will be returned
802  */
cachefiles_allocate_page(struct fscache_retrieval * op,struct page * page,gfp_t gfp)803 int cachefiles_allocate_page(struct fscache_retrieval *op,
804 			     struct page *page,
805 			     gfp_t gfp)
806 {
807 	struct cachefiles_object *object;
808 	struct cachefiles_cache *cache;
809 	int ret;
810 
811 	object = container_of(op->op.object,
812 			      struct cachefiles_object, fscache);
813 	cache = container_of(object->fscache.cache,
814 			     struct cachefiles_cache, cache);
815 
816 	_enter("%p,{%lx},", object, page->index);
817 
818 	ret = cachefiles_has_space(cache, 0, 1);
819 	if (ret == 0)
820 		fscache_mark_page_cached(op, page);
821 	else
822 		ret = -ENOBUFS;
823 
824 	fscache_retrieval_complete(op, 1);
825 	_leave(" = %d", ret);
826 	return ret;
827 }
828 
829 /*
830  * allocate blocks in the cache in which to store a set of pages
831  * - cache withdrawal is prevented by the caller
832  * - returns -EINTR if interrupted
833  * - returns -ENOMEM if ran out of memory
834  * - returns -ENOBUFS if some buffers couldn't be made available
835  * - returns -ENOBUFS if some pages are beyond EOF
836  * - otherwise:
837  *   - -ENODATA will be returned
838  * - metadata will be retained for any page marked
839  */
cachefiles_allocate_pages(struct fscache_retrieval * op,struct list_head * pages,unsigned * nr_pages,gfp_t gfp)840 int cachefiles_allocate_pages(struct fscache_retrieval *op,
841 			      struct list_head *pages,
842 			      unsigned *nr_pages,
843 			      gfp_t gfp)
844 {
845 	struct cachefiles_object *object;
846 	struct cachefiles_cache *cache;
847 	struct pagevec pagevec;
848 	struct page *page;
849 	int ret;
850 
851 	object = container_of(op->op.object,
852 			      struct cachefiles_object, fscache);
853 	cache = container_of(object->fscache.cache,
854 			     struct cachefiles_cache, cache);
855 
856 	_enter("%p,,,%d,", object, *nr_pages);
857 
858 	ret = cachefiles_has_space(cache, 0, *nr_pages);
859 	if (ret == 0) {
860 		pagevec_init(&pagevec, 0);
861 
862 		list_for_each_entry(page, pages, lru) {
863 			if (pagevec_add(&pagevec, page) == 0)
864 				fscache_mark_pages_cached(op, &pagevec);
865 		}
866 
867 		if (pagevec_count(&pagevec) > 0)
868 			fscache_mark_pages_cached(op, &pagevec);
869 		ret = -ENODATA;
870 	} else {
871 		ret = -ENOBUFS;
872 	}
873 
874 	fscache_retrieval_complete(op, *nr_pages);
875 	_leave(" = %d", ret);
876 	return ret;
877 }
878 
879 /*
880  * request a page be stored in the cache
881  * - cache withdrawal is prevented by the caller
882  * - this request may be ignored if there's no cache block available, in which
883  *   case -ENOBUFS will be returned
884  * - if the op is in progress, 0 will be returned
885  */
cachefiles_write_page(struct fscache_storage * op,struct page * page)886 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
887 {
888 	struct cachefiles_object *object;
889 	struct cachefiles_cache *cache;
890 	struct file *file;
891 	struct path path;
892 	loff_t pos, eof;
893 	size_t len;
894 	void *data;
895 	int ret = -ENOBUFS;
896 
897 	ASSERT(op != NULL);
898 	ASSERT(page != NULL);
899 
900 	object = container_of(op->op.object,
901 			      struct cachefiles_object, fscache);
902 
903 	_enter("%p,%p{%lx},,,", object, page, page->index);
904 
905 	if (!object->backer) {
906 		_leave(" = -ENOBUFS");
907 		return -ENOBUFS;
908 	}
909 
910 	ASSERT(d_is_reg(object->backer));
911 
912 	cache = container_of(object->fscache.cache,
913 			     struct cachefiles_cache, cache);
914 
915 	pos = (loff_t)page->index << PAGE_SHIFT;
916 
917 	/* We mustn't write more data than we have, so we have to beware of a
918 	 * partial page at EOF.
919 	 */
920 	eof = object->fscache.store_limit_l;
921 	if (pos >= eof)
922 		goto error;
923 
924 	/* write the page to the backing filesystem and let it store it in its
925 	 * own time */
926 	path.mnt = cache->mnt;
927 	path.dentry = object->backer;
928 	file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
929 	if (IS_ERR(file)) {
930 		ret = PTR_ERR(file);
931 		goto error_2;
932 	}
933 
934 	len = PAGE_SIZE;
935 	if (eof & ~PAGE_MASK) {
936 		if (eof - pos < PAGE_SIZE) {
937 			_debug("cut short %llx to %llx",
938 			       pos, eof);
939 			len = eof - pos;
940 			ASSERTCMP(pos + len, ==, eof);
941 		}
942 	}
943 
944 	data = kmap(page);
945 	ret = __kernel_write(file, data, len, &pos);
946 	kunmap(page);
947 	fput(file);
948 	if (ret != len)
949 		goto error_eio;
950 
951 	_leave(" = 0");
952 	return 0;
953 
954 error_eio:
955 	ret = -EIO;
956 error_2:
957 	if (ret == -EIO)
958 		cachefiles_io_error_obj(object,
959 					"Write page to backing file failed");
960 error:
961 	_leave(" = -ENOBUFS [%d]", ret);
962 	return -ENOBUFS;
963 }
964 
965 /*
966  * detach a backing block from a page
967  * - cache withdrawal is prevented by the caller
968  */
cachefiles_uncache_page(struct fscache_object * _object,struct page * page)969 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
970 {
971 	struct cachefiles_object *object;
972 
973 	object = container_of(_object, struct cachefiles_object, fscache);
974 
975 	_enter("%p,{%lu}", object, page->index);
976 
977 	spin_unlock(&object->fscache.cookie->lock);
978 }
979