1/* Cache page management and data I/O routines
2 *
3 * Copyright (C) 2004-2008 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 License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL PAGE
13#include <linux/module.h>
14#include <linux/fscache-cache.h>
15#include <linux/buffer_head.h>
16#include <linux/pagevec.h>
17#include <linux/slab.h>
18#include "internal.h"
19
20/*
21 * check to see if a page is being written to the cache
22 */
23bool __fscache_check_page_write(struct fscache_cookie *cookie, struct page *page)
24{
25	void *val;
26
27	rcu_read_lock();
28	val = radix_tree_lookup(&cookie->stores, page->index);
29	rcu_read_unlock();
30
31	return val != NULL;
32}
33EXPORT_SYMBOL(__fscache_check_page_write);
34
35/*
36 * wait for a page to finish being written to the cache
37 */
38void __fscache_wait_on_page_write(struct fscache_cookie *cookie, struct page *page)
39{
40	wait_queue_head_t *wq = bit_waitqueue(&cookie->flags, 0);
41
42	wait_event(*wq, !__fscache_check_page_write(cookie, page));
43}
44EXPORT_SYMBOL(__fscache_wait_on_page_write);
45
46/*
47 * wait for a page to finish being written to the cache. Put a timeout here
48 * since we might be called recursively via parent fs.
49 */
50static
51bool release_page_wait_timeout(struct fscache_cookie *cookie, struct page *page)
52{
53	wait_queue_head_t *wq = bit_waitqueue(&cookie->flags, 0);
54
55	return wait_event_timeout(*wq, !__fscache_check_page_write(cookie, page),
56				  HZ);
57}
58
59/*
60 * decide whether a page can be released, possibly by cancelling a store to it
61 * - we're allowed to sleep if __GFP_WAIT is flagged
62 */
63bool __fscache_maybe_release_page(struct fscache_cookie *cookie,
64				  struct page *page,
65				  gfp_t gfp)
66{
67	struct page *xpage;
68	void *val;
69
70	_enter("%p,%p,%x", cookie, page, gfp);
71
72try_again:
73	rcu_read_lock();
74	val = radix_tree_lookup(&cookie->stores, page->index);
75	if (!val) {
76		rcu_read_unlock();
77		fscache_stat(&fscache_n_store_vmscan_not_storing);
78		__fscache_uncache_page(cookie, page);
79		return true;
80	}
81
82	/* see if the page is actually undergoing storage - if so we can't get
83	 * rid of it till the cache has finished with it */
84	if (radix_tree_tag_get(&cookie->stores, page->index,
85			       FSCACHE_COOKIE_STORING_TAG)) {
86		rcu_read_unlock();
87		goto page_busy;
88	}
89
90	/* the page is pending storage, so we attempt to cancel the store and
91	 * discard the store request so that the page can be reclaimed */
92	spin_lock(&cookie->stores_lock);
93	rcu_read_unlock();
94
95	if (radix_tree_tag_get(&cookie->stores, page->index,
96			       FSCACHE_COOKIE_STORING_TAG)) {
97		/* the page started to undergo storage whilst we were looking,
98		 * so now we can only wait or return */
99		spin_unlock(&cookie->stores_lock);
100		goto page_busy;
101	}
102
103	xpage = radix_tree_delete(&cookie->stores, page->index);
104	spin_unlock(&cookie->stores_lock);
105
106	if (xpage) {
107		fscache_stat(&fscache_n_store_vmscan_cancelled);
108		fscache_stat(&fscache_n_store_radix_deletes);
109		ASSERTCMP(xpage, ==, page);
110	} else {
111		fscache_stat(&fscache_n_store_vmscan_gone);
112	}
113
114	wake_up_bit(&cookie->flags, 0);
115	if (xpage)
116		page_cache_release(xpage);
117	__fscache_uncache_page(cookie, page);
118	return true;
119
120page_busy:
121	/* We will wait here if we're allowed to, but that could deadlock the
122	 * allocator as the work threads writing to the cache may all end up
123	 * sleeping on memory allocation, so we may need to impose a timeout
124	 * too. */
125	if (!(gfp & __GFP_WAIT) || !(gfp & __GFP_FS)) {
126		fscache_stat(&fscache_n_store_vmscan_busy);
127		return false;
128	}
129
130	fscache_stat(&fscache_n_store_vmscan_wait);
131	if (!release_page_wait_timeout(cookie, page))
132		_debug("fscache writeout timeout page: %p{%lx}",
133			page, page->index);
134
135	gfp &= ~__GFP_WAIT;
136	goto try_again;
137}
138EXPORT_SYMBOL(__fscache_maybe_release_page);
139
140/*
141 * note that a page has finished being written to the cache
142 */
143static void fscache_end_page_write(struct fscache_object *object,
144				   struct page *page)
145{
146	struct fscache_cookie *cookie;
147	struct page *xpage = NULL;
148
149	spin_lock(&object->lock);
150	cookie = object->cookie;
151	if (cookie) {
152		/* delete the page from the tree if it is now no longer
153		 * pending */
154		spin_lock(&cookie->stores_lock);
155		radix_tree_tag_clear(&cookie->stores, page->index,
156				     FSCACHE_COOKIE_STORING_TAG);
157		if (!radix_tree_tag_get(&cookie->stores, page->index,
158					FSCACHE_COOKIE_PENDING_TAG)) {
159			fscache_stat(&fscache_n_store_radix_deletes);
160			xpage = radix_tree_delete(&cookie->stores, page->index);
161		}
162		spin_unlock(&cookie->stores_lock);
163		wake_up_bit(&cookie->flags, 0);
164	}
165	spin_unlock(&object->lock);
166	if (xpage)
167		page_cache_release(xpage);
168}
169
170/*
171 * actually apply the changed attributes to a cache object
172 */
173static void fscache_attr_changed_op(struct fscache_operation *op)
174{
175	struct fscache_object *object = op->object;
176	int ret;
177
178	_enter("{OBJ%x OP%x}", object->debug_id, op->debug_id);
179
180	fscache_stat(&fscache_n_attr_changed_calls);
181
182	if (fscache_object_is_active(object)) {
183		fscache_stat(&fscache_n_cop_attr_changed);
184		ret = object->cache->ops->attr_changed(object);
185		fscache_stat_d(&fscache_n_cop_attr_changed);
186		if (ret < 0)
187			fscache_abort_object(object);
188	}
189
190	fscache_op_complete(op, true);
191	_leave("");
192}
193
194/*
195 * notification that the attributes on an object have changed
196 */
197int __fscache_attr_changed(struct fscache_cookie *cookie)
198{
199	struct fscache_operation *op;
200	struct fscache_object *object;
201	bool wake_cookie = false;
202
203	_enter("%p", cookie);
204
205	ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
206
207	fscache_stat(&fscache_n_attr_changed);
208
209	op = kzalloc(sizeof(*op), GFP_KERNEL);
210	if (!op) {
211		fscache_stat(&fscache_n_attr_changed_nomem);
212		_leave(" = -ENOMEM");
213		return -ENOMEM;
214	}
215
216	fscache_operation_init(op, fscache_attr_changed_op, NULL);
217	op->flags = FSCACHE_OP_ASYNC |
218		(1 << FSCACHE_OP_EXCLUSIVE) |
219		(1 << FSCACHE_OP_UNUSE_COOKIE);
220
221	spin_lock(&cookie->lock);
222
223	if (!fscache_cookie_enabled(cookie) ||
224	    hlist_empty(&cookie->backing_objects))
225		goto nobufs;
226	object = hlist_entry(cookie->backing_objects.first,
227			     struct fscache_object, cookie_link);
228
229	__fscache_use_cookie(cookie);
230	if (fscache_submit_exclusive_op(object, op) < 0)
231		goto nobufs_dec;
232	spin_unlock(&cookie->lock);
233	fscache_stat(&fscache_n_attr_changed_ok);
234	fscache_put_operation(op);
235	_leave(" = 0");
236	return 0;
237
238nobufs_dec:
239	wake_cookie = __fscache_unuse_cookie(cookie);
240nobufs:
241	spin_unlock(&cookie->lock);
242	kfree(op);
243	if (wake_cookie)
244		__fscache_wake_unused_cookie(cookie);
245	fscache_stat(&fscache_n_attr_changed_nobufs);
246	_leave(" = %d", -ENOBUFS);
247	return -ENOBUFS;
248}
249EXPORT_SYMBOL(__fscache_attr_changed);
250
251/*
252 * release a retrieval op reference
253 */
254static void fscache_release_retrieval_op(struct fscache_operation *_op)
255{
256	struct fscache_retrieval *op =
257		container_of(_op, struct fscache_retrieval, op);
258
259	_enter("{OP%x}", op->op.debug_id);
260
261	ASSERTCMP(atomic_read(&op->n_pages), ==, 0);
262
263	fscache_hist(fscache_retrieval_histogram, op->start_time);
264	if (op->context)
265		fscache_put_context(op->op.object->cookie, op->context);
266
267	_leave("");
268}
269
270/*
271 * allocate a retrieval op
272 */
273static struct fscache_retrieval *fscache_alloc_retrieval(
274	struct fscache_cookie *cookie,
275	struct address_space *mapping,
276	fscache_rw_complete_t end_io_func,
277	void *context)
278{
279	struct fscache_retrieval *op;
280
281	/* allocate a retrieval operation and attempt to submit it */
282	op = kzalloc(sizeof(*op), GFP_NOIO);
283	if (!op) {
284		fscache_stat(&fscache_n_retrievals_nomem);
285		return NULL;
286	}
287
288	fscache_operation_init(&op->op, NULL, fscache_release_retrieval_op);
289	op->op.flags	= FSCACHE_OP_MYTHREAD |
290		(1UL << FSCACHE_OP_WAITING) |
291		(1UL << FSCACHE_OP_UNUSE_COOKIE);
292	op->mapping	= mapping;
293	op->end_io_func	= end_io_func;
294	op->context	= context;
295	op->start_time	= jiffies;
296	INIT_LIST_HEAD(&op->to_do);
297	return op;
298}
299
300/*
301 * wait for a deferred lookup to complete
302 */
303int fscache_wait_for_deferred_lookup(struct fscache_cookie *cookie)
304{
305	unsigned long jif;
306
307	_enter("");
308
309	if (!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags)) {
310		_leave(" = 0 [imm]");
311		return 0;
312	}
313
314	fscache_stat(&fscache_n_retrievals_wait);
315
316	jif = jiffies;
317	if (wait_on_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP,
318			TASK_INTERRUPTIBLE) != 0) {
319		fscache_stat(&fscache_n_retrievals_intr);
320		_leave(" = -ERESTARTSYS");
321		return -ERESTARTSYS;
322	}
323
324	ASSERT(!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags));
325
326	smp_rmb();
327	fscache_hist(fscache_retrieval_delay_histogram, jif);
328	_leave(" = 0 [dly]");
329	return 0;
330}
331
332/*
333 * Handle cancellation of a pending retrieval op
334 */
335static void fscache_do_cancel_retrieval(struct fscache_operation *_op)
336{
337	struct fscache_retrieval *op =
338		container_of(_op, struct fscache_retrieval, op);
339
340	atomic_set(&op->n_pages, 0);
341}
342
343/*
344 * wait for an object to become active (or dead)
345 */
346int fscache_wait_for_operation_activation(struct fscache_object *object,
347					  struct fscache_operation *op,
348					  atomic_t *stat_op_waits,
349					  atomic_t *stat_object_dead,
350					  void (*do_cancel)(struct fscache_operation *))
351{
352	int ret;
353
354	if (!test_bit(FSCACHE_OP_WAITING, &op->flags))
355		goto check_if_dead;
356
357	_debug(">>> WT");
358	if (stat_op_waits)
359		fscache_stat(stat_op_waits);
360	if (wait_on_bit(&op->flags, FSCACHE_OP_WAITING,
361			TASK_INTERRUPTIBLE) != 0) {
362		ret = fscache_cancel_op(op, do_cancel);
363		if (ret == 0)
364			return -ERESTARTSYS;
365
366		/* it's been removed from the pending queue by another party,
367		 * so we should get to run shortly */
368		wait_on_bit(&op->flags, FSCACHE_OP_WAITING,
369			    TASK_UNINTERRUPTIBLE);
370	}
371	_debug("<<< GO");
372
373check_if_dead:
374	if (op->state == FSCACHE_OP_ST_CANCELLED) {
375		if (stat_object_dead)
376			fscache_stat(stat_object_dead);
377		_leave(" = -ENOBUFS [cancelled]");
378		return -ENOBUFS;
379	}
380	if (unlikely(fscache_object_is_dead(object))) {
381		pr_err("%s() = -ENOBUFS [obj dead %d]\n", __func__, op->state);
382		fscache_cancel_op(op, do_cancel);
383		if (stat_object_dead)
384			fscache_stat(stat_object_dead);
385		return -ENOBUFS;
386	}
387	return 0;
388}
389
390/*
391 * read a page from the cache or allocate a block in which to store it
392 * - we return:
393 *   -ENOMEM	- out of memory, nothing done
394 *   -ERESTARTSYS - interrupted
395 *   -ENOBUFS	- no backing object available in which to cache the block
396 *   -ENODATA	- no data available in the backing object for this block
397 *   0		- dispatched a read - it'll call end_io_func() when finished
398 */
399int __fscache_read_or_alloc_page(struct fscache_cookie *cookie,
400				 struct page *page,
401				 fscache_rw_complete_t end_io_func,
402				 void *context,
403				 gfp_t gfp)
404{
405	struct fscache_retrieval *op;
406	struct fscache_object *object;
407	bool wake_cookie = false;
408	int ret;
409
410	_enter("%p,%p,,,", cookie, page);
411
412	fscache_stat(&fscache_n_retrievals);
413
414	if (hlist_empty(&cookie->backing_objects))
415		goto nobufs;
416
417	if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
418		_leave(" = -ENOBUFS [invalidating]");
419		return -ENOBUFS;
420	}
421
422	ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
423	ASSERTCMP(page, !=, NULL);
424
425	if (fscache_wait_for_deferred_lookup(cookie) < 0)
426		return -ERESTARTSYS;
427
428	op = fscache_alloc_retrieval(cookie, page->mapping,
429				     end_io_func, context);
430	if (!op) {
431		_leave(" = -ENOMEM");
432		return -ENOMEM;
433	}
434	atomic_set(&op->n_pages, 1);
435
436	spin_lock(&cookie->lock);
437
438	if (!fscache_cookie_enabled(cookie) ||
439	    hlist_empty(&cookie->backing_objects))
440		goto nobufs_unlock;
441	object = hlist_entry(cookie->backing_objects.first,
442			     struct fscache_object, cookie_link);
443
444	ASSERT(test_bit(FSCACHE_OBJECT_IS_LOOKED_UP, &object->flags));
445
446	__fscache_use_cookie(cookie);
447	atomic_inc(&object->n_reads);
448	__set_bit(FSCACHE_OP_DEC_READ_CNT, &op->op.flags);
449
450	if (fscache_submit_op(object, &op->op) < 0)
451		goto nobufs_unlock_dec;
452	spin_unlock(&cookie->lock);
453
454	fscache_stat(&fscache_n_retrieval_ops);
455
456	/* pin the netfs read context in case we need to do the actual netfs
457	 * read because we've encountered a cache read failure */
458	fscache_get_context(object->cookie, op->context);
459
460	/* we wait for the operation to become active, and then process it
461	 * *here*, in this thread, and not in the thread pool */
462	ret = fscache_wait_for_operation_activation(
463		object, &op->op,
464		__fscache_stat(&fscache_n_retrieval_op_waits),
465		__fscache_stat(&fscache_n_retrievals_object_dead),
466		fscache_do_cancel_retrieval);
467	if (ret < 0)
468		goto error;
469
470	/* ask the cache to honour the operation */
471	if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
472		fscache_stat(&fscache_n_cop_allocate_page);
473		ret = object->cache->ops->allocate_page(op, page, gfp);
474		fscache_stat_d(&fscache_n_cop_allocate_page);
475		if (ret == 0)
476			ret = -ENODATA;
477	} else {
478		fscache_stat(&fscache_n_cop_read_or_alloc_page);
479		ret = object->cache->ops->read_or_alloc_page(op, page, gfp);
480		fscache_stat_d(&fscache_n_cop_read_or_alloc_page);
481	}
482
483error:
484	if (ret == -ENOMEM)
485		fscache_stat(&fscache_n_retrievals_nomem);
486	else if (ret == -ERESTARTSYS)
487		fscache_stat(&fscache_n_retrievals_intr);
488	else if (ret == -ENODATA)
489		fscache_stat(&fscache_n_retrievals_nodata);
490	else if (ret < 0)
491		fscache_stat(&fscache_n_retrievals_nobufs);
492	else
493		fscache_stat(&fscache_n_retrievals_ok);
494
495	fscache_put_retrieval(op);
496	_leave(" = %d", ret);
497	return ret;
498
499nobufs_unlock_dec:
500	atomic_dec(&object->n_reads);
501	wake_cookie = __fscache_unuse_cookie(cookie);
502nobufs_unlock:
503	spin_unlock(&cookie->lock);
504	if (wake_cookie)
505		__fscache_wake_unused_cookie(cookie);
506	kfree(op);
507nobufs:
508	fscache_stat(&fscache_n_retrievals_nobufs);
509	_leave(" = -ENOBUFS");
510	return -ENOBUFS;
511}
512EXPORT_SYMBOL(__fscache_read_or_alloc_page);
513
514/*
515 * read a list of page from the cache or allocate a block in which to store
516 * them
517 * - we return:
518 *   -ENOMEM	- out of memory, some pages may be being read
519 *   -ERESTARTSYS - interrupted, some pages may be being read
520 *   -ENOBUFS	- no backing object or space available in which to cache any
521 *                pages not being read
522 *   -ENODATA	- no data available in the backing object for some or all of
523 *                the pages
524 *   0		- dispatched a read on all pages
525 *
526 * end_io_func() will be called for each page read from the cache as it is
527 * finishes being read
528 *
529 * any pages for which a read is dispatched will be removed from pages and
530 * nr_pages
531 */
532int __fscache_read_or_alloc_pages(struct fscache_cookie *cookie,
533				  struct address_space *mapping,
534				  struct list_head *pages,
535				  unsigned *nr_pages,
536				  fscache_rw_complete_t end_io_func,
537				  void *context,
538				  gfp_t gfp)
539{
540	struct fscache_retrieval *op;
541	struct fscache_object *object;
542	bool wake_cookie = false;
543	int ret;
544
545	_enter("%p,,%d,,,", cookie, *nr_pages);
546
547	fscache_stat(&fscache_n_retrievals);
548
549	if (hlist_empty(&cookie->backing_objects))
550		goto nobufs;
551
552	if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
553		_leave(" = -ENOBUFS [invalidating]");
554		return -ENOBUFS;
555	}
556
557	ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
558	ASSERTCMP(*nr_pages, >, 0);
559	ASSERT(!list_empty(pages));
560
561	if (fscache_wait_for_deferred_lookup(cookie) < 0)
562		return -ERESTARTSYS;
563
564	op = fscache_alloc_retrieval(cookie, mapping, end_io_func, context);
565	if (!op)
566		return -ENOMEM;
567	atomic_set(&op->n_pages, *nr_pages);
568
569	spin_lock(&cookie->lock);
570
571	if (!fscache_cookie_enabled(cookie) ||
572	    hlist_empty(&cookie->backing_objects))
573		goto nobufs_unlock;
574	object = hlist_entry(cookie->backing_objects.first,
575			     struct fscache_object, cookie_link);
576
577	__fscache_use_cookie(cookie);
578	atomic_inc(&object->n_reads);
579	__set_bit(FSCACHE_OP_DEC_READ_CNT, &op->op.flags);
580
581	if (fscache_submit_op(object, &op->op) < 0)
582		goto nobufs_unlock_dec;
583	spin_unlock(&cookie->lock);
584
585	fscache_stat(&fscache_n_retrieval_ops);
586
587	/* pin the netfs read context in case we need to do the actual netfs
588	 * read because we've encountered a cache read failure */
589	fscache_get_context(object->cookie, op->context);
590
591	/* we wait for the operation to become active, and then process it
592	 * *here*, in this thread, and not in the thread pool */
593	ret = fscache_wait_for_operation_activation(
594		object, &op->op,
595		__fscache_stat(&fscache_n_retrieval_op_waits),
596		__fscache_stat(&fscache_n_retrievals_object_dead),
597		fscache_do_cancel_retrieval);
598	if (ret < 0)
599		goto error;
600
601	/* ask the cache to honour the operation */
602	if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
603		fscache_stat(&fscache_n_cop_allocate_pages);
604		ret = object->cache->ops->allocate_pages(
605			op, pages, nr_pages, gfp);
606		fscache_stat_d(&fscache_n_cop_allocate_pages);
607	} else {
608		fscache_stat(&fscache_n_cop_read_or_alloc_pages);
609		ret = object->cache->ops->read_or_alloc_pages(
610			op, pages, nr_pages, gfp);
611		fscache_stat_d(&fscache_n_cop_read_or_alloc_pages);
612	}
613
614error:
615	if (ret == -ENOMEM)
616		fscache_stat(&fscache_n_retrievals_nomem);
617	else if (ret == -ERESTARTSYS)
618		fscache_stat(&fscache_n_retrievals_intr);
619	else if (ret == -ENODATA)
620		fscache_stat(&fscache_n_retrievals_nodata);
621	else if (ret < 0)
622		fscache_stat(&fscache_n_retrievals_nobufs);
623	else
624		fscache_stat(&fscache_n_retrievals_ok);
625
626	fscache_put_retrieval(op);
627	_leave(" = %d", ret);
628	return ret;
629
630nobufs_unlock_dec:
631	atomic_dec(&object->n_reads);
632	wake_cookie = __fscache_unuse_cookie(cookie);
633nobufs_unlock:
634	spin_unlock(&cookie->lock);
635	kfree(op);
636	if (wake_cookie)
637		__fscache_wake_unused_cookie(cookie);
638nobufs:
639	fscache_stat(&fscache_n_retrievals_nobufs);
640	_leave(" = -ENOBUFS");
641	return -ENOBUFS;
642}
643EXPORT_SYMBOL(__fscache_read_or_alloc_pages);
644
645/*
646 * allocate a block in the cache on which to store a page
647 * - we return:
648 *   -ENOMEM	- out of memory, nothing done
649 *   -ERESTARTSYS - interrupted
650 *   -ENOBUFS	- no backing object available in which to cache the block
651 *   0		- block allocated
652 */
653int __fscache_alloc_page(struct fscache_cookie *cookie,
654			 struct page *page,
655			 gfp_t gfp)
656{
657	struct fscache_retrieval *op;
658	struct fscache_object *object;
659	bool wake_cookie = false;
660	int ret;
661
662	_enter("%p,%p,,,", cookie, page);
663
664	fscache_stat(&fscache_n_allocs);
665
666	if (hlist_empty(&cookie->backing_objects))
667		goto nobufs;
668
669	ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
670	ASSERTCMP(page, !=, NULL);
671
672	if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
673		_leave(" = -ENOBUFS [invalidating]");
674		return -ENOBUFS;
675	}
676
677	if (fscache_wait_for_deferred_lookup(cookie) < 0)
678		return -ERESTARTSYS;
679
680	op = fscache_alloc_retrieval(cookie, page->mapping, NULL, NULL);
681	if (!op)
682		return -ENOMEM;
683	atomic_set(&op->n_pages, 1);
684
685	spin_lock(&cookie->lock);
686
687	if (!fscache_cookie_enabled(cookie) ||
688	    hlist_empty(&cookie->backing_objects))
689		goto nobufs_unlock;
690	object = hlist_entry(cookie->backing_objects.first,
691			     struct fscache_object, cookie_link);
692
693	__fscache_use_cookie(cookie);
694	if (fscache_submit_op(object, &op->op) < 0)
695		goto nobufs_unlock_dec;
696	spin_unlock(&cookie->lock);
697
698	fscache_stat(&fscache_n_alloc_ops);
699
700	ret = fscache_wait_for_operation_activation(
701		object, &op->op,
702		__fscache_stat(&fscache_n_alloc_op_waits),
703		__fscache_stat(&fscache_n_allocs_object_dead),
704		fscache_do_cancel_retrieval);
705	if (ret < 0)
706		goto error;
707
708	/* ask the cache to honour the operation */
709	fscache_stat(&fscache_n_cop_allocate_page);
710	ret = object->cache->ops->allocate_page(op, page, gfp);
711	fscache_stat_d(&fscache_n_cop_allocate_page);
712
713error:
714	if (ret == -ERESTARTSYS)
715		fscache_stat(&fscache_n_allocs_intr);
716	else if (ret < 0)
717		fscache_stat(&fscache_n_allocs_nobufs);
718	else
719		fscache_stat(&fscache_n_allocs_ok);
720
721	fscache_put_retrieval(op);
722	_leave(" = %d", ret);
723	return ret;
724
725nobufs_unlock_dec:
726	wake_cookie = __fscache_unuse_cookie(cookie);
727nobufs_unlock:
728	spin_unlock(&cookie->lock);
729	kfree(op);
730	if (wake_cookie)
731		__fscache_wake_unused_cookie(cookie);
732nobufs:
733	fscache_stat(&fscache_n_allocs_nobufs);
734	_leave(" = -ENOBUFS");
735	return -ENOBUFS;
736}
737EXPORT_SYMBOL(__fscache_alloc_page);
738
739/*
740 * Unmark pages allocate in the readahead code path (via:
741 * fscache_readpages_or_alloc) after delegating to the base filesystem
742 */
743void __fscache_readpages_cancel(struct fscache_cookie *cookie,
744				struct list_head *pages)
745{
746	struct page *page;
747
748	list_for_each_entry(page, pages, lru) {
749		if (PageFsCache(page))
750			__fscache_uncache_page(cookie, page);
751	}
752}
753EXPORT_SYMBOL(__fscache_readpages_cancel);
754
755/*
756 * release a write op reference
757 */
758static void fscache_release_write_op(struct fscache_operation *_op)
759{
760	_enter("{OP%x}", _op->debug_id);
761}
762
763/*
764 * perform the background storage of a page into the cache
765 */
766static void fscache_write_op(struct fscache_operation *_op)
767{
768	struct fscache_storage *op =
769		container_of(_op, struct fscache_storage, op);
770	struct fscache_object *object = op->op.object;
771	struct fscache_cookie *cookie;
772	struct page *page;
773	unsigned n;
774	void *results[1];
775	int ret;
776
777	_enter("{OP%x,%d}", op->op.debug_id, atomic_read(&op->op.usage));
778
779	spin_lock(&object->lock);
780	cookie = object->cookie;
781
782	if (!fscache_object_is_active(object)) {
783		/* If we get here, then the on-disk cache object likely longer
784		 * exists, so we should just cancel this write operation.
785		 */
786		spin_unlock(&object->lock);
787		fscache_op_complete(&op->op, false);
788		_leave(" [inactive]");
789		return;
790	}
791
792	if (!cookie) {
793		/* If we get here, then the cookie belonging to the object was
794		 * detached, probably by the cookie being withdrawn due to
795		 * memory pressure, which means that the pages we might write
796		 * to the cache from no longer exist - therefore, we can just
797		 * cancel this write operation.
798		 */
799		spin_unlock(&object->lock);
800		fscache_op_complete(&op->op, false);
801		_leave(" [cancel] op{f=%lx s=%u} obj{s=%s f=%lx}",
802		       _op->flags, _op->state, object->state->short_name,
803		       object->flags);
804		return;
805	}
806
807	spin_lock(&cookie->stores_lock);
808
809	fscache_stat(&fscache_n_store_calls);
810
811	/* find a page to store */
812	page = NULL;
813	n = radix_tree_gang_lookup_tag(&cookie->stores, results, 0, 1,
814				       FSCACHE_COOKIE_PENDING_TAG);
815	if (n != 1)
816		goto superseded;
817	page = results[0];
818	_debug("gang %d [%lx]", n, page->index);
819	if (page->index > op->store_limit) {
820		fscache_stat(&fscache_n_store_pages_over_limit);
821		goto superseded;
822	}
823
824	radix_tree_tag_set(&cookie->stores, page->index,
825			   FSCACHE_COOKIE_STORING_TAG);
826	radix_tree_tag_clear(&cookie->stores, page->index,
827			     FSCACHE_COOKIE_PENDING_TAG);
828
829	spin_unlock(&cookie->stores_lock);
830	spin_unlock(&object->lock);
831
832	fscache_stat(&fscache_n_store_pages);
833	fscache_stat(&fscache_n_cop_write_page);
834	ret = object->cache->ops->write_page(op, page);
835	fscache_stat_d(&fscache_n_cop_write_page);
836	fscache_end_page_write(object, page);
837	if (ret < 0) {
838		fscache_abort_object(object);
839		fscache_op_complete(&op->op, true);
840	} else {
841		fscache_enqueue_operation(&op->op);
842	}
843
844	_leave("");
845	return;
846
847superseded:
848	/* this writer is going away and there aren't any more things to
849	 * write */
850	_debug("cease");
851	spin_unlock(&cookie->stores_lock);
852	clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
853	spin_unlock(&object->lock);
854	fscache_op_complete(&op->op, true);
855	_leave("");
856}
857
858/*
859 * Clear the pages pending writing for invalidation
860 */
861void fscache_invalidate_writes(struct fscache_cookie *cookie)
862{
863	struct page *page;
864	void *results[16];
865	int n, i;
866
867	_enter("");
868
869	for (;;) {
870		spin_lock(&cookie->stores_lock);
871		n = radix_tree_gang_lookup_tag(&cookie->stores, results, 0,
872					       ARRAY_SIZE(results),
873					       FSCACHE_COOKIE_PENDING_TAG);
874		if (n == 0) {
875			spin_unlock(&cookie->stores_lock);
876			break;
877		}
878
879		for (i = n - 1; i >= 0; i--) {
880			page = results[i];
881			radix_tree_delete(&cookie->stores, page->index);
882		}
883
884		spin_unlock(&cookie->stores_lock);
885
886		for (i = n - 1; i >= 0; i--)
887			page_cache_release(results[i]);
888	}
889
890	_leave("");
891}
892
893/*
894 * request a page be stored in the cache
895 * - returns:
896 *   -ENOMEM	- out of memory, nothing done
897 *   -ENOBUFS	- no backing object available in which to cache the page
898 *   0		- dispatched a write - it'll call end_io_func() when finished
899 *
900 * if the cookie still has a backing object at this point, that object can be
901 * in one of a few states with respect to storage processing:
902 *
903 *  (1) negative lookup, object not yet created (FSCACHE_COOKIE_CREATING is
904 *      set)
905 *
906 *	(a) no writes yet
907 *
908 *	(b) writes deferred till post-creation (mark page for writing and
909 *	    return immediately)
910 *
911 *  (2) negative lookup, object created, initial fill being made from netfs
912 *
913 *	(a) fill point not yet reached this page (mark page for writing and
914 *          return)
915 *
916 *	(b) fill point passed this page (queue op to store this page)
917 *
918 *  (3) object extant (queue op to store this page)
919 *
920 * any other state is invalid
921 */
922int __fscache_write_page(struct fscache_cookie *cookie,
923			 struct page *page,
924			 gfp_t gfp)
925{
926	struct fscache_storage *op;
927	struct fscache_object *object;
928	bool wake_cookie = false;
929	int ret;
930
931	_enter("%p,%x,", cookie, (u32) page->flags);
932
933	ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
934	ASSERT(PageFsCache(page));
935
936	fscache_stat(&fscache_n_stores);
937
938	if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
939		_leave(" = -ENOBUFS [invalidating]");
940		return -ENOBUFS;
941	}
942
943	op = kzalloc(sizeof(*op), GFP_NOIO | __GFP_NOMEMALLOC | __GFP_NORETRY);
944	if (!op)
945		goto nomem;
946
947	fscache_operation_init(&op->op, fscache_write_op,
948			       fscache_release_write_op);
949	op->op.flags = FSCACHE_OP_ASYNC |
950		(1 << FSCACHE_OP_WAITING) |
951		(1 << FSCACHE_OP_UNUSE_COOKIE);
952
953	ret = radix_tree_maybe_preload(gfp & ~__GFP_HIGHMEM);
954	if (ret < 0)
955		goto nomem_free;
956
957	ret = -ENOBUFS;
958	spin_lock(&cookie->lock);
959
960	if (!fscache_cookie_enabled(cookie) ||
961	    hlist_empty(&cookie->backing_objects))
962		goto nobufs;
963	object = hlist_entry(cookie->backing_objects.first,
964			     struct fscache_object, cookie_link);
965	if (test_bit(FSCACHE_IOERROR, &object->cache->flags))
966		goto nobufs;
967
968	/* add the page to the pending-storage radix tree on the backing
969	 * object */
970	spin_lock(&object->lock);
971	spin_lock(&cookie->stores_lock);
972
973	_debug("store limit %llx", (unsigned long long) object->store_limit);
974
975	ret = radix_tree_insert(&cookie->stores, page->index, page);
976	if (ret < 0) {
977		if (ret == -EEXIST)
978			goto already_queued;
979		_debug("insert failed %d", ret);
980		goto nobufs_unlock_obj;
981	}
982
983	radix_tree_tag_set(&cookie->stores, page->index,
984			   FSCACHE_COOKIE_PENDING_TAG);
985	page_cache_get(page);
986
987	/* we only want one writer at a time, but we do need to queue new
988	 * writers after exclusive ops */
989	if (test_and_set_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags))
990		goto already_pending;
991
992	spin_unlock(&cookie->stores_lock);
993	spin_unlock(&object->lock);
994
995	op->op.debug_id	= atomic_inc_return(&fscache_op_debug_id);
996	op->store_limit = object->store_limit;
997
998	__fscache_use_cookie(cookie);
999	if (fscache_submit_op(object, &op->op) < 0)
1000		goto submit_failed;
1001
1002	spin_unlock(&cookie->lock);
1003	radix_tree_preload_end();
1004	fscache_stat(&fscache_n_store_ops);
1005	fscache_stat(&fscache_n_stores_ok);
1006
1007	/* the work queue now carries its own ref on the object */
1008	fscache_put_operation(&op->op);
1009	_leave(" = 0");
1010	return 0;
1011
1012already_queued:
1013	fscache_stat(&fscache_n_stores_again);
1014already_pending:
1015	spin_unlock(&cookie->stores_lock);
1016	spin_unlock(&object->lock);
1017	spin_unlock(&cookie->lock);
1018	radix_tree_preload_end();
1019	kfree(op);
1020	fscache_stat(&fscache_n_stores_ok);
1021	_leave(" = 0");
1022	return 0;
1023
1024submit_failed:
1025	spin_lock(&cookie->stores_lock);
1026	radix_tree_delete(&cookie->stores, page->index);
1027	spin_unlock(&cookie->stores_lock);
1028	wake_cookie = __fscache_unuse_cookie(cookie);
1029	page_cache_release(page);
1030	ret = -ENOBUFS;
1031	goto nobufs;
1032
1033nobufs_unlock_obj:
1034	spin_unlock(&cookie->stores_lock);
1035	spin_unlock(&object->lock);
1036nobufs:
1037	spin_unlock(&cookie->lock);
1038	radix_tree_preload_end();
1039	kfree(op);
1040	if (wake_cookie)
1041		__fscache_wake_unused_cookie(cookie);
1042	fscache_stat(&fscache_n_stores_nobufs);
1043	_leave(" = -ENOBUFS");
1044	return -ENOBUFS;
1045
1046nomem_free:
1047	kfree(op);
1048nomem:
1049	fscache_stat(&fscache_n_stores_oom);
1050	_leave(" = -ENOMEM");
1051	return -ENOMEM;
1052}
1053EXPORT_SYMBOL(__fscache_write_page);
1054
1055/*
1056 * remove a page from the cache
1057 */
1058void __fscache_uncache_page(struct fscache_cookie *cookie, struct page *page)
1059{
1060	struct fscache_object *object;
1061
1062	_enter(",%p", page);
1063
1064	ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
1065	ASSERTCMP(page, !=, NULL);
1066
1067	fscache_stat(&fscache_n_uncaches);
1068
1069	/* cache withdrawal may beat us to it */
1070	if (!PageFsCache(page))
1071		goto done;
1072
1073	/* get the object */
1074	spin_lock(&cookie->lock);
1075
1076	if (hlist_empty(&cookie->backing_objects)) {
1077		ClearPageFsCache(page);
1078		goto done_unlock;
1079	}
1080
1081	object = hlist_entry(cookie->backing_objects.first,
1082			     struct fscache_object, cookie_link);
1083
1084	/* there might now be stuff on disk we could read */
1085	clear_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
1086
1087	/* only invoke the cache backend if we managed to mark the page
1088	 * uncached here; this deals with synchronisation vs withdrawal */
1089	if (TestClearPageFsCache(page) &&
1090	    object->cache->ops->uncache_page) {
1091		/* the cache backend releases the cookie lock */
1092		fscache_stat(&fscache_n_cop_uncache_page);
1093		object->cache->ops->uncache_page(object, page);
1094		fscache_stat_d(&fscache_n_cop_uncache_page);
1095		goto done;
1096	}
1097
1098done_unlock:
1099	spin_unlock(&cookie->lock);
1100done:
1101	_leave("");
1102}
1103EXPORT_SYMBOL(__fscache_uncache_page);
1104
1105/**
1106 * fscache_mark_page_cached - Mark a page as being cached
1107 * @op: The retrieval op pages are being marked for
1108 * @page: The page to be marked
1109 *
1110 * Mark a netfs page as being cached.  After this is called, the netfs
1111 * must call fscache_uncache_page() to remove the mark.
1112 */
1113void fscache_mark_page_cached(struct fscache_retrieval *op, struct page *page)
1114{
1115	struct fscache_cookie *cookie = op->op.object->cookie;
1116
1117#ifdef CONFIG_FSCACHE_STATS
1118	atomic_inc(&fscache_n_marks);
1119#endif
1120
1121	_debug("- mark %p{%lx}", page, page->index);
1122	if (TestSetPageFsCache(page)) {
1123		static bool once_only;
1124		if (!once_only) {
1125			once_only = true;
1126			pr_warn("Cookie type %s marked page %lx multiple times\n",
1127				cookie->def->name, page->index);
1128		}
1129	}
1130
1131	if (cookie->def->mark_page_cached)
1132		cookie->def->mark_page_cached(cookie->netfs_data,
1133					      op->mapping, page);
1134}
1135EXPORT_SYMBOL(fscache_mark_page_cached);
1136
1137/**
1138 * fscache_mark_pages_cached - Mark pages as being cached
1139 * @op: The retrieval op pages are being marked for
1140 * @pagevec: The pages to be marked
1141 *
1142 * Mark a bunch of netfs pages as being cached.  After this is called,
1143 * the netfs must call fscache_uncache_page() to remove the mark.
1144 */
1145void fscache_mark_pages_cached(struct fscache_retrieval *op,
1146			       struct pagevec *pagevec)
1147{
1148	unsigned long loop;
1149
1150	for (loop = 0; loop < pagevec->nr; loop++)
1151		fscache_mark_page_cached(op, pagevec->pages[loop]);
1152
1153	pagevec_reinit(pagevec);
1154}
1155EXPORT_SYMBOL(fscache_mark_pages_cached);
1156
1157/*
1158 * Uncache all the pages in an inode that are marked PG_fscache, assuming them
1159 * to be associated with the given cookie.
1160 */
1161void __fscache_uncache_all_inode_pages(struct fscache_cookie *cookie,
1162				       struct inode *inode)
1163{
1164	struct address_space *mapping = inode->i_mapping;
1165	struct pagevec pvec;
1166	pgoff_t next;
1167	int i;
1168
1169	_enter("%p,%p", cookie, inode);
1170
1171	if (!mapping || mapping->nrpages == 0) {
1172		_leave(" [no pages]");
1173		return;
1174	}
1175
1176	pagevec_init(&pvec, 0);
1177	next = 0;
1178	do {
1179		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE))
1180			break;
1181		for (i = 0; i < pagevec_count(&pvec); i++) {
1182			struct page *page = pvec.pages[i];
1183			next = page->index;
1184			if (PageFsCache(page)) {
1185				__fscache_wait_on_page_write(cookie, page);
1186				__fscache_uncache_page(cookie, page);
1187			}
1188		}
1189		pagevec_release(&pvec);
1190		cond_resched();
1191	} while (++next);
1192
1193	_leave("");
1194}
1195EXPORT_SYMBOL(__fscache_uncache_all_inode_pages);
1196