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 	fscache_enqueue_retrieval(op);
68 	spin_unlock(&object->work_lock);
69 
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 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 discard:
146 	spin_lock_irq(&object->work_lock);
147 	list_del(&monitor->op_link);
148 	spin_unlock_irq(&object->work_lock);
149 	_leave(" = %d", ret);
150 	return ret;
151 }
152 
153 /*
154  * copy data from backing pages to netfs pages to complete a read operation
155  * - driven by FS-Cache's thread pool
156  */
cachefiles_read_copier(struct fscache_operation * _op)157 static void cachefiles_read_copier(struct fscache_operation *_op)
158 {
159 	struct cachefiles_one_read *monitor;
160 	struct cachefiles_object *object;
161 	struct fscache_retrieval *op;
162 	int error, max;
163 
164 	op = container_of(_op, struct fscache_retrieval, op);
165 	object = container_of(op->op.object,
166 			      struct cachefiles_object, fscache);
167 
168 	_enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
169 
170 	max = 8;
171 	spin_lock_irq(&object->work_lock);
172 
173 	while (!list_empty(&op->to_do)) {
174 		monitor = list_entry(op->to_do.next,
175 				     struct cachefiles_one_read, op_link);
176 		list_del(&monitor->op_link);
177 
178 		spin_unlock_irq(&object->work_lock);
179 
180 		_debug("- copy {%lu}", monitor->back_page->index);
181 
182 	recheck:
183 		if (test_bit(FSCACHE_COOKIE_INVALIDATING,
184 			     &object->fscache.cookie->flags)) {
185 			error = -ESTALE;
186 		} else if (PageUptodate(monitor->back_page)) {
187 			copy_highpage(monitor->netfs_page, monitor->back_page);
188 			fscache_mark_page_cached(monitor->op,
189 						 monitor->netfs_page);
190 			error = 0;
191 		} else if (!PageError(monitor->back_page)) {
192 			/* the page has probably been truncated */
193 			error = cachefiles_read_reissue(object, monitor);
194 			if (error == -EINPROGRESS)
195 				goto next;
196 			goto recheck;
197 		} else {
198 			cachefiles_io_error_obj(
199 				object,
200 				"Readpage failed on backing file %lx",
201 				(unsigned long) monitor->back_page->flags);
202 			error = -EIO;
203 		}
204 
205 		put_page(monitor->back_page);
206 
207 		fscache_end_io(op, monitor->netfs_page, error);
208 		put_page(monitor->netfs_page);
209 		fscache_retrieval_complete(op, 1);
210 		fscache_put_retrieval(op);
211 		kfree(monitor);
212 
213 	next:
214 		/* let the thread pool have some air occasionally */
215 		max--;
216 		if (max < 0 || need_resched()) {
217 			if (!list_empty(&op->to_do))
218 				fscache_enqueue_retrieval(op);
219 			_leave(" [maxed out]");
220 			return;
221 		}
222 
223 		spin_lock_irq(&object->work_lock);
224 	}
225 
226 	spin_unlock_irq(&object->work_lock);
227 	_leave("");
228 }
229 
230 /*
231  * read the corresponding page to the given set from the backing file
232  * - an uncertain page is simply discarded, to be tried again another time
233  */
cachefiles_read_backing_file_one(struct cachefiles_object * object,struct fscache_retrieval * op,struct page * netpage)234 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
235 					    struct fscache_retrieval *op,
236 					    struct page *netpage)
237 {
238 	struct cachefiles_one_read *monitor;
239 	struct address_space *bmapping;
240 	struct page *newpage, *backpage;
241 	int ret;
242 
243 	_enter("");
244 
245 	_debug("read back %p{%lu,%d}",
246 	       netpage, netpage->index, page_count(netpage));
247 
248 	monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
249 	if (!monitor)
250 		goto nomem;
251 
252 	monitor->netfs_page = netpage;
253 	monitor->op = fscache_get_retrieval(op);
254 
255 	init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
256 
257 	/* attempt to get hold of the backing page */
258 	bmapping = d_backing_inode(object->backer)->i_mapping;
259 	newpage = NULL;
260 
261 	for (;;) {
262 		backpage = find_get_page(bmapping, netpage->index);
263 		if (backpage)
264 			goto backing_page_already_present;
265 
266 		if (!newpage) {
267 			newpage = __page_cache_alloc(cachefiles_gfp);
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 				if (!newpage)
505 					goto nomem;
506 			}
507 
508 			ret = add_to_page_cache_lru(newpage, bmapping,
509 						    netpage->index,
510 						    cachefiles_gfp);
511 			if (ret == 0)
512 				goto installed_new_backing_page;
513 			if (ret != -EEXIST)
514 				goto nomem;
515 		}
516 
517 		/* we've installed a new backing page, so now we need
518 		 * to start it reading */
519 	installed_new_backing_page:
520 		_debug("- new %p", newpage);
521 
522 		backpage = newpage;
523 		newpage = NULL;
524 
525 	reread_backing_page:
526 		ret = bmapping->a_ops->readpage(NULL, backpage);
527 		if (ret < 0)
528 			goto read_error;
529 
530 		/* add the netfs page to the pagecache and LRU, and set the
531 		 * monitor to transfer the data across */
532 	monitor_backing_page:
533 		_debug("- monitor add");
534 
535 		ret = add_to_page_cache_lru(netpage, op->mapping,
536 					    netpage->index, cachefiles_gfp);
537 		if (ret < 0) {
538 			if (ret == -EEXIST) {
539 				put_page(backpage);
540 				backpage = NULL;
541 				put_page(netpage);
542 				netpage = NULL;
543 				fscache_retrieval_complete(op, 1);
544 				continue;
545 			}
546 			goto nomem;
547 		}
548 
549 		/* install a monitor */
550 		get_page(netpage);
551 		monitor->netfs_page = netpage;
552 
553 		get_page(backpage);
554 		monitor->back_page = backpage;
555 		monitor->monitor.private = backpage;
556 		add_page_wait_queue(backpage, &monitor->monitor);
557 		monitor = NULL;
558 
559 		/* but the page may have been read before the monitor was
560 		 * installed, so the monitor may miss the event - so we have to
561 		 * ensure that we do get one in such a case */
562 		if (trylock_page(backpage)) {
563 			_debug("2unlock %p {%lx}", backpage, backpage->flags);
564 			unlock_page(backpage);
565 		}
566 
567 		put_page(backpage);
568 		backpage = NULL;
569 
570 		put_page(netpage);
571 		netpage = NULL;
572 		continue;
573 
574 		/* if the backing page is already present, it can be in one of
575 		 * three states: read in progress, read failed or read okay */
576 	backing_page_already_present:
577 		_debug("- present %p", backpage);
578 
579 		if (PageError(backpage))
580 			goto io_error;
581 
582 		if (PageUptodate(backpage))
583 			goto backing_page_already_uptodate;
584 
585 		_debug("- not ready %p{%lx}", backpage, backpage->flags);
586 
587 		if (!trylock_page(backpage))
588 			goto monitor_backing_page;
589 
590 		if (PageError(backpage)) {
591 			_debug("error %lx", backpage->flags);
592 			unlock_page(backpage);
593 			goto io_error;
594 		}
595 
596 		if (PageUptodate(backpage))
597 			goto backing_page_already_uptodate_unlock;
598 
599 		/* we've locked a page that's neither up to date nor erroneous,
600 		 * so we need to attempt to read it again */
601 		goto reread_backing_page;
602 
603 		/* the backing page is already up to date, attach the netfs
604 		 * page to the pagecache and LRU and copy the data across */
605 	backing_page_already_uptodate_unlock:
606 		_debug("uptodate %lx", backpage->flags);
607 		unlock_page(backpage);
608 	backing_page_already_uptodate:
609 		_debug("- uptodate");
610 
611 		ret = add_to_page_cache_lru(netpage, op->mapping,
612 					    netpage->index, cachefiles_gfp);
613 		if (ret < 0) {
614 			if (ret == -EEXIST) {
615 				put_page(backpage);
616 				backpage = NULL;
617 				put_page(netpage);
618 				netpage = NULL;
619 				fscache_retrieval_complete(op, 1);
620 				continue;
621 			}
622 			goto nomem;
623 		}
624 
625 		copy_highpage(netpage, backpage);
626 
627 		put_page(backpage);
628 		backpage = NULL;
629 
630 		fscache_mark_page_cached(op, netpage);
631 
632 		/* the netpage is unlocked and marked up to date here */
633 		fscache_end_io(op, netpage, 0);
634 		put_page(netpage);
635 		netpage = NULL;
636 		fscache_retrieval_complete(op, 1);
637 		continue;
638 	}
639 
640 	netpage = NULL;
641 
642 	_debug("out");
643 
644 out:
645 	/* tidy up */
646 	if (newpage)
647 		put_page(newpage);
648 	if (netpage)
649 		put_page(netpage);
650 	if (backpage)
651 		put_page(backpage);
652 	if (monitor) {
653 		fscache_put_retrieval(op);
654 		kfree(monitor);
655 	}
656 
657 	list_for_each_entry_safe(netpage, _n, list, lru) {
658 		list_del(&netpage->lru);
659 		put_page(netpage);
660 		fscache_retrieval_complete(op, 1);
661 	}
662 
663 	_leave(" = %d", ret);
664 	return ret;
665 
666 nomem:
667 	_debug("nomem");
668 	ret = -ENOMEM;
669 	goto record_page_complete;
670 
671 read_error:
672 	_debug("read error %d", ret);
673 	if (ret == -ENOMEM)
674 		goto record_page_complete;
675 io_error:
676 	cachefiles_io_error_obj(object, "Page read error on backing file");
677 	ret = -ENOBUFS;
678 record_page_complete:
679 	fscache_retrieval_complete(op, 1);
680 	goto out;
681 }
682 
683 /*
684  * read a list of pages from the cache or allocate blocks in which to store
685  * them
686  */
cachefiles_read_or_alloc_pages(struct fscache_retrieval * op,struct list_head * pages,unsigned * nr_pages,gfp_t gfp)687 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
688 				   struct list_head *pages,
689 				   unsigned *nr_pages,
690 				   gfp_t gfp)
691 {
692 	struct cachefiles_object *object;
693 	struct cachefiles_cache *cache;
694 	struct list_head backpages;
695 	struct pagevec pagevec;
696 	struct inode *inode;
697 	struct page *page, *_n;
698 	unsigned shift, nrbackpages;
699 	int ret, ret2, space;
700 
701 	object = container_of(op->op.object,
702 			      struct cachefiles_object, fscache);
703 	cache = container_of(object->fscache.cache,
704 			     struct cachefiles_cache, cache);
705 
706 	_enter("{OBJ%x,%d},,%d,,",
707 	       object->fscache.debug_id, atomic_read(&op->op.usage),
708 	       *nr_pages);
709 
710 	if (!object->backer)
711 		goto all_enobufs;
712 
713 	space = 1;
714 	if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
715 		space = 0;
716 
717 	inode = d_backing_inode(object->backer);
718 	ASSERT(S_ISREG(inode->i_mode));
719 	ASSERT(inode->i_mapping->a_ops->bmap);
720 	ASSERT(inode->i_mapping->a_ops->readpages);
721 
722 	/* calculate the shift required to use bmap */
723 	shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
724 
725 	pagevec_init(&pagevec);
726 
727 	op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
728 	op->op.flags |= FSCACHE_OP_ASYNC;
729 	op->op.processor = cachefiles_read_copier;
730 
731 	INIT_LIST_HEAD(&backpages);
732 	nrbackpages = 0;
733 
734 	ret = space ? -ENODATA : -ENOBUFS;
735 	list_for_each_entry_safe(page, _n, pages, lru) {
736 		sector_t block0, block;
737 
738 		/* we assume the absence or presence of the first block is a
739 		 * good enough indication for the page as a whole
740 		 * - TODO: don't use bmap() for this as it is _not_ actually
741 		 *   good enough for this as it doesn't indicate errors, but
742 		 *   it's all we've got for the moment
743 		 */
744 		block0 = page->index;
745 		block0 <<= shift;
746 
747 		block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
748 						      block0);
749 		_debug("%llx -> %llx",
750 		       (unsigned long long) block0,
751 		       (unsigned long long) block);
752 
753 		if (block) {
754 			/* we have data - add it to the list to give to the
755 			 * backing fs */
756 			list_move(&page->lru, &backpages);
757 			(*nr_pages)--;
758 			nrbackpages++;
759 		} else if (space && pagevec_add(&pagevec, page) == 0) {
760 			fscache_mark_pages_cached(op, &pagevec);
761 			fscache_retrieval_complete(op, 1);
762 			ret = -ENODATA;
763 		} else {
764 			fscache_retrieval_complete(op, 1);
765 		}
766 	}
767 
768 	if (pagevec_count(&pagevec) > 0)
769 		fscache_mark_pages_cached(op, &pagevec);
770 
771 	if (list_empty(pages))
772 		ret = 0;
773 
774 	/* submit the apparently valid pages to the backing fs to be read from
775 	 * disk */
776 	if (nrbackpages > 0) {
777 		ret2 = cachefiles_read_backing_file(object, op, &backpages);
778 		if (ret2 == -ENOMEM || ret2 == -EINTR)
779 			ret = ret2;
780 	}
781 
782 	_leave(" = %d [nr=%u%s]",
783 	       ret, *nr_pages, list_empty(pages) ? " empty" : "");
784 	return ret;
785 
786 all_enobufs:
787 	fscache_retrieval_complete(op, *nr_pages);
788 	return -ENOBUFS;
789 }
790 
791 /*
792  * allocate a block in the cache in which to store a page
793  * - cache withdrawal is prevented by the caller
794  * - returns -EINTR if interrupted
795  * - returns -ENOMEM if ran out of memory
796  * - returns -ENOBUFS if no buffers can be made available
797  * - returns -ENOBUFS if page is beyond EOF
798  * - otherwise:
799  *   - the metadata will be retained
800  *   - 0 will be returned
801  */
cachefiles_allocate_page(struct fscache_retrieval * op,struct page * page,gfp_t gfp)802 int cachefiles_allocate_page(struct fscache_retrieval *op,
803 			     struct page *page,
804 			     gfp_t gfp)
805 {
806 	struct cachefiles_object *object;
807 	struct cachefiles_cache *cache;
808 	int ret;
809 
810 	object = container_of(op->op.object,
811 			      struct cachefiles_object, fscache);
812 	cache = container_of(object->fscache.cache,
813 			     struct cachefiles_cache, cache);
814 
815 	_enter("%p,{%lx},", object, page->index);
816 
817 	ret = cachefiles_has_space(cache, 0, 1);
818 	if (ret == 0)
819 		fscache_mark_page_cached(op, page);
820 	else
821 		ret = -ENOBUFS;
822 
823 	fscache_retrieval_complete(op, 1);
824 	_leave(" = %d", ret);
825 	return ret;
826 }
827 
828 /*
829  * allocate blocks in the cache in which to store a set of pages
830  * - cache withdrawal is prevented by the caller
831  * - returns -EINTR if interrupted
832  * - returns -ENOMEM if ran out of memory
833  * - returns -ENOBUFS if some buffers couldn't be made available
834  * - returns -ENOBUFS if some pages are beyond EOF
835  * - otherwise:
836  *   - -ENODATA will be returned
837  * - metadata will be retained for any page marked
838  */
cachefiles_allocate_pages(struct fscache_retrieval * op,struct list_head * pages,unsigned * nr_pages,gfp_t gfp)839 int cachefiles_allocate_pages(struct fscache_retrieval *op,
840 			      struct list_head *pages,
841 			      unsigned *nr_pages,
842 			      gfp_t gfp)
843 {
844 	struct cachefiles_object *object;
845 	struct cachefiles_cache *cache;
846 	struct pagevec pagevec;
847 	struct page *page;
848 	int ret;
849 
850 	object = container_of(op->op.object,
851 			      struct cachefiles_object, fscache);
852 	cache = container_of(object->fscache.cache,
853 			     struct cachefiles_cache, cache);
854 
855 	_enter("%p,,,%d,", object, *nr_pages);
856 
857 	ret = cachefiles_has_space(cache, 0, *nr_pages);
858 	if (ret == 0) {
859 		pagevec_init(&pagevec);
860 
861 		list_for_each_entry(page, pages, lru) {
862 			if (pagevec_add(&pagevec, page) == 0)
863 				fscache_mark_pages_cached(op, &pagevec);
864 		}
865 
866 		if (pagevec_count(&pagevec) > 0)
867 			fscache_mark_pages_cached(op, &pagevec);
868 		ret = -ENODATA;
869 	} else {
870 		ret = -ENOBUFS;
871 	}
872 
873 	fscache_retrieval_complete(op, *nr_pages);
874 	_leave(" = %d", ret);
875 	return ret;
876 }
877 
878 /*
879  * request a page be stored in the cache
880  * - cache withdrawal is prevented by the caller
881  * - this request may be ignored if there's no cache block available, in which
882  *   case -ENOBUFS will be returned
883  * - if the op is in progress, 0 will be returned
884  */
cachefiles_write_page(struct fscache_storage * op,struct page * page)885 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
886 {
887 	struct cachefiles_object *object;
888 	struct cachefiles_cache *cache;
889 	struct file *file;
890 	struct path path;
891 	loff_t pos, eof;
892 	size_t len;
893 	void *data;
894 	int ret = -ENOBUFS;
895 
896 	ASSERT(op != NULL);
897 	ASSERT(page != NULL);
898 
899 	object = container_of(op->op.object,
900 			      struct cachefiles_object, fscache);
901 
902 	_enter("%p,%p{%lx},,,", object, page, page->index);
903 
904 	if (!object->backer) {
905 		_leave(" = -ENOBUFS");
906 		return -ENOBUFS;
907 	}
908 
909 	ASSERT(d_is_reg(object->backer));
910 
911 	cache = container_of(object->fscache.cache,
912 			     struct cachefiles_cache, cache);
913 
914 	pos = (loff_t)page->index << PAGE_SHIFT;
915 
916 	/* We mustn't write more data than we have, so we have to beware of a
917 	 * partial page at EOF.
918 	 */
919 	eof = object->fscache.store_limit_l;
920 	if (pos >= eof)
921 		goto error;
922 
923 	/* write the page to the backing filesystem and let it store it in its
924 	 * own time */
925 	path.mnt = cache->mnt;
926 	path.dentry = object->backer;
927 	file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
928 	if (IS_ERR(file)) {
929 		ret = PTR_ERR(file);
930 		goto error_2;
931 	}
932 
933 	len = PAGE_SIZE;
934 	if (eof & ~PAGE_MASK) {
935 		if (eof - pos < PAGE_SIZE) {
936 			_debug("cut short %llx to %llx",
937 			       pos, eof);
938 			len = eof - pos;
939 			ASSERTCMP(pos + len, ==, eof);
940 		}
941 	}
942 
943 	data = kmap(page);
944 	ret = __kernel_write(file, data, len, &pos);
945 	kunmap(page);
946 	fput(file);
947 	if (ret != len)
948 		goto error_eio;
949 
950 	_leave(" = 0");
951 	return 0;
952 
953 error_eio:
954 	ret = -EIO;
955 error_2:
956 	if (ret == -EIO)
957 		cachefiles_io_error_obj(object,
958 					"Write page to backing file failed");
959 error:
960 	_leave(" = -ENOBUFS [%d]", ret);
961 	return -ENOBUFS;
962 }
963 
964 /*
965  * detach a backing block from a page
966  * - cache withdrawal is prevented by the caller
967  */
cachefiles_uncache_page(struct fscache_object * _object,struct page * page)968 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
969 	__releases(&object->fscache.cookie->lock)
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