xref: /linux/net/sunrpc/cache.c (revision 88f50c80748bf5238c88e70ee26c68ac48b94e68)
1 /*
2  * net/sunrpc/cache.c
3  *
4  * Generic code for various authentication-related caches
5  * used by sunrpc clients and servers.
6  *
7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
8  *
9  * Released under terms in GPL version 2.  See COPYING.
10  *
11  */
12 
13 #include <linux/types.h>
14 #include <linux/fs.h>
15 #include <linux/file.h>
16 #include <linux/slab.h>
17 #include <linux/signal.h>
18 #include <linux/sched.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/module.h>
22 #include <linux/ctype.h>
23 #include <asm/uaccess.h>
24 #include <linux/poll.h>
25 #include <linux/seq_file.h>
26 #include <linux/proc_fs.h>
27 #include <linux/net.h>
28 #include <linux/workqueue.h>
29 #include <linux/mutex.h>
30 #include <linux/pagemap.h>
31 #include <asm/ioctls.h>
32 #include <linux/sunrpc/types.h>
33 #include <linux/sunrpc/cache.h>
34 #include <linux/sunrpc/stats.h>
35 #include <linux/sunrpc/rpc_pipe_fs.h>
36 #include "netns.h"
37 
38 #define	 RPCDBG_FACILITY RPCDBG_CACHE
39 
40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
41 static void cache_revisit_request(struct cache_head *item);
42 
43 static void cache_init(struct cache_head *h)
44 {
45 	time_t now = seconds_since_boot();
46 	h->next = NULL;
47 	h->flags = 0;
48 	kref_init(&h->ref);
49 	h->expiry_time = now + CACHE_NEW_EXPIRY;
50 	h->last_refresh = now;
51 }
52 
53 static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
54 {
55 	return  (h->expiry_time < seconds_since_boot()) ||
56 		(detail->flush_time > h->last_refresh);
57 }
58 
59 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
60 				       struct cache_head *key, int hash)
61 {
62 	struct cache_head **head,  **hp;
63 	struct cache_head *new = NULL, *freeme = NULL;
64 
65 	head = &detail->hash_table[hash];
66 
67 	read_lock(&detail->hash_lock);
68 
69 	for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
70 		struct cache_head *tmp = *hp;
71 		if (detail->match(tmp, key)) {
72 			if (cache_is_expired(detail, tmp))
73 				/* This entry is expired, we will discard it. */
74 				break;
75 			cache_get(tmp);
76 			read_unlock(&detail->hash_lock);
77 			return tmp;
78 		}
79 	}
80 	read_unlock(&detail->hash_lock);
81 	/* Didn't find anything, insert an empty entry */
82 
83 	new = detail->alloc();
84 	if (!new)
85 		return NULL;
86 	/* must fully initialise 'new', else
87 	 * we might get lose if we need to
88 	 * cache_put it soon.
89 	 */
90 	cache_init(new);
91 	detail->init(new, key);
92 
93 	write_lock(&detail->hash_lock);
94 
95 	/* check if entry appeared while we slept */
96 	for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
97 		struct cache_head *tmp = *hp;
98 		if (detail->match(tmp, key)) {
99 			if (cache_is_expired(detail, tmp)) {
100 				*hp = tmp->next;
101 				tmp->next = NULL;
102 				detail->entries --;
103 				freeme = tmp;
104 				break;
105 			}
106 			cache_get(tmp);
107 			write_unlock(&detail->hash_lock);
108 			cache_put(new, detail);
109 			return tmp;
110 		}
111 	}
112 	new->next = *head;
113 	*head = new;
114 	detail->entries++;
115 	cache_get(new);
116 	write_unlock(&detail->hash_lock);
117 
118 	if (freeme)
119 		cache_put(freeme, detail);
120 	return new;
121 }
122 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
123 
124 
125 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
126 
127 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
128 {
129 	head->expiry_time = expiry;
130 	head->last_refresh = seconds_since_boot();
131 	smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
132 	set_bit(CACHE_VALID, &head->flags);
133 }
134 
135 static void cache_fresh_unlocked(struct cache_head *head,
136 				 struct cache_detail *detail)
137 {
138 	if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
139 		cache_revisit_request(head);
140 		cache_dequeue(detail, head);
141 	}
142 }
143 
144 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
145 				       struct cache_head *new, struct cache_head *old, int hash)
146 {
147 	/* The 'old' entry is to be replaced by 'new'.
148 	 * If 'old' is not VALID, we update it directly,
149 	 * otherwise we need to replace it
150 	 */
151 	struct cache_head **head;
152 	struct cache_head *tmp;
153 
154 	if (!test_bit(CACHE_VALID, &old->flags)) {
155 		write_lock(&detail->hash_lock);
156 		if (!test_bit(CACHE_VALID, &old->flags)) {
157 			if (test_bit(CACHE_NEGATIVE, &new->flags))
158 				set_bit(CACHE_NEGATIVE, &old->flags);
159 			else
160 				detail->update(old, new);
161 			cache_fresh_locked(old, new->expiry_time);
162 			write_unlock(&detail->hash_lock);
163 			cache_fresh_unlocked(old, detail);
164 			return old;
165 		}
166 		write_unlock(&detail->hash_lock);
167 	}
168 	/* We need to insert a new entry */
169 	tmp = detail->alloc();
170 	if (!tmp) {
171 		cache_put(old, detail);
172 		return NULL;
173 	}
174 	cache_init(tmp);
175 	detail->init(tmp, old);
176 	head = &detail->hash_table[hash];
177 
178 	write_lock(&detail->hash_lock);
179 	if (test_bit(CACHE_NEGATIVE, &new->flags))
180 		set_bit(CACHE_NEGATIVE, &tmp->flags);
181 	else
182 		detail->update(tmp, new);
183 	tmp->next = *head;
184 	*head = tmp;
185 	detail->entries++;
186 	cache_get(tmp);
187 	cache_fresh_locked(tmp, new->expiry_time);
188 	cache_fresh_locked(old, 0);
189 	write_unlock(&detail->hash_lock);
190 	cache_fresh_unlocked(tmp, detail);
191 	cache_fresh_unlocked(old, detail);
192 	cache_put(old, detail);
193 	return tmp;
194 }
195 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
196 
197 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
198 {
199 	if (!cd->cache_upcall)
200 		return -EINVAL;
201 	return cd->cache_upcall(cd, h);
202 }
203 
204 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
205 {
206 	if (!test_bit(CACHE_VALID, &h->flags))
207 		return -EAGAIN;
208 	else {
209 		/* entry is valid */
210 		if (test_bit(CACHE_NEGATIVE, &h->flags))
211 			return -ENOENT;
212 		else {
213 			/*
214 			 * In combination with write barrier in
215 			 * sunrpc_cache_update, ensures that anyone
216 			 * using the cache entry after this sees the
217 			 * updated contents:
218 			 */
219 			smp_rmb();
220 			return 0;
221 		}
222 	}
223 }
224 
225 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
226 {
227 	int rv;
228 
229 	write_lock(&detail->hash_lock);
230 	rv = cache_is_valid(detail, h);
231 	if (rv != -EAGAIN) {
232 		write_unlock(&detail->hash_lock);
233 		return rv;
234 	}
235 	set_bit(CACHE_NEGATIVE, &h->flags);
236 	cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
237 	write_unlock(&detail->hash_lock);
238 	cache_fresh_unlocked(h, detail);
239 	return -ENOENT;
240 }
241 
242 /*
243  * This is the generic cache management routine for all
244  * the authentication caches.
245  * It checks the currency of a cache item and will (later)
246  * initiate an upcall to fill it if needed.
247  *
248  *
249  * Returns 0 if the cache_head can be used, or cache_puts it and returns
250  * -EAGAIN if upcall is pending and request has been queued
251  * -ETIMEDOUT if upcall failed or request could not be queue or
252  *           upcall completed but item is still invalid (implying that
253  *           the cache item has been replaced with a newer one).
254  * -ENOENT if cache entry was negative
255  */
256 int cache_check(struct cache_detail *detail,
257 		    struct cache_head *h, struct cache_req *rqstp)
258 {
259 	int rv;
260 	long refresh_age, age;
261 
262 	/* First decide return status as best we can */
263 	rv = cache_is_valid(detail, h);
264 
265 	/* now see if we want to start an upcall */
266 	refresh_age = (h->expiry_time - h->last_refresh);
267 	age = seconds_since_boot() - h->last_refresh;
268 
269 	if (rqstp == NULL) {
270 		if (rv == -EAGAIN)
271 			rv = -ENOENT;
272 	} else if (rv == -EAGAIN || age > refresh_age/2) {
273 		dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
274 				refresh_age, age);
275 		if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
276 			switch (cache_make_upcall(detail, h)) {
277 			case -EINVAL:
278 				clear_bit(CACHE_PENDING, &h->flags);
279 				cache_revisit_request(h);
280 				rv = try_to_negate_entry(detail, h);
281 				break;
282 			case -EAGAIN:
283 				clear_bit(CACHE_PENDING, &h->flags);
284 				cache_revisit_request(h);
285 				break;
286 			}
287 		}
288 	}
289 
290 	if (rv == -EAGAIN) {
291 		if (!cache_defer_req(rqstp, h)) {
292 			/*
293 			 * Request was not deferred; handle it as best
294 			 * we can ourselves:
295 			 */
296 			rv = cache_is_valid(detail, h);
297 			if (rv == -EAGAIN)
298 				rv = -ETIMEDOUT;
299 		}
300 	}
301 	if (rv)
302 		cache_put(h, detail);
303 	return rv;
304 }
305 EXPORT_SYMBOL_GPL(cache_check);
306 
307 /*
308  * caches need to be periodically cleaned.
309  * For this we maintain a list of cache_detail and
310  * a current pointer into that list and into the table
311  * for that entry.
312  *
313  * Each time clean_cache is called it finds the next non-empty entry
314  * in the current table and walks the list in that entry
315  * looking for entries that can be removed.
316  *
317  * An entry gets removed if:
318  * - The expiry is before current time
319  * - The last_refresh time is before the flush_time for that cache
320  *
321  * later we might drop old entries with non-NEVER expiry if that table
322  * is getting 'full' for some definition of 'full'
323  *
324  * The question of "how often to scan a table" is an interesting one
325  * and is answered in part by the use of the "nextcheck" field in the
326  * cache_detail.
327  * When a scan of a table begins, the nextcheck field is set to a time
328  * that is well into the future.
329  * While scanning, if an expiry time is found that is earlier than the
330  * current nextcheck time, nextcheck is set to that expiry time.
331  * If the flush_time is ever set to a time earlier than the nextcheck
332  * time, the nextcheck time is then set to that flush_time.
333  *
334  * A table is then only scanned if the current time is at least
335  * the nextcheck time.
336  *
337  */
338 
339 static LIST_HEAD(cache_list);
340 static DEFINE_SPINLOCK(cache_list_lock);
341 static struct cache_detail *current_detail;
342 static int current_index;
343 
344 static void do_cache_clean(struct work_struct *work);
345 static struct delayed_work cache_cleaner;
346 
347 void sunrpc_init_cache_detail(struct cache_detail *cd)
348 {
349 	rwlock_init(&cd->hash_lock);
350 	INIT_LIST_HEAD(&cd->queue);
351 	spin_lock(&cache_list_lock);
352 	cd->nextcheck = 0;
353 	cd->entries = 0;
354 	atomic_set(&cd->readers, 0);
355 	cd->last_close = 0;
356 	cd->last_warn = -1;
357 	list_add(&cd->others, &cache_list);
358 	spin_unlock(&cache_list_lock);
359 
360 	/* start the cleaning process */
361 	schedule_delayed_work(&cache_cleaner, 0);
362 }
363 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
364 
365 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
366 {
367 	cache_purge(cd);
368 	spin_lock(&cache_list_lock);
369 	write_lock(&cd->hash_lock);
370 	if (cd->entries || atomic_read(&cd->inuse)) {
371 		write_unlock(&cd->hash_lock);
372 		spin_unlock(&cache_list_lock);
373 		goto out;
374 	}
375 	if (current_detail == cd)
376 		current_detail = NULL;
377 	list_del_init(&cd->others);
378 	write_unlock(&cd->hash_lock);
379 	spin_unlock(&cache_list_lock);
380 	if (list_empty(&cache_list)) {
381 		/* module must be being unloaded so its safe to kill the worker */
382 		cancel_delayed_work_sync(&cache_cleaner);
383 	}
384 	return;
385 out:
386 	printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
387 }
388 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
389 
390 /* clean cache tries to find something to clean
391  * and cleans it.
392  * It returns 1 if it cleaned something,
393  *            0 if it didn't find anything this time
394  *           -1 if it fell off the end of the list.
395  */
396 static int cache_clean(void)
397 {
398 	int rv = 0;
399 	struct list_head *next;
400 
401 	spin_lock(&cache_list_lock);
402 
403 	/* find a suitable table if we don't already have one */
404 	while (current_detail == NULL ||
405 	    current_index >= current_detail->hash_size) {
406 		if (current_detail)
407 			next = current_detail->others.next;
408 		else
409 			next = cache_list.next;
410 		if (next == &cache_list) {
411 			current_detail = NULL;
412 			spin_unlock(&cache_list_lock);
413 			return -1;
414 		}
415 		current_detail = list_entry(next, struct cache_detail, others);
416 		if (current_detail->nextcheck > seconds_since_boot())
417 			current_index = current_detail->hash_size;
418 		else {
419 			current_index = 0;
420 			current_detail->nextcheck = seconds_since_boot()+30*60;
421 		}
422 	}
423 
424 	/* find a non-empty bucket in the table */
425 	while (current_detail &&
426 	       current_index < current_detail->hash_size &&
427 	       current_detail->hash_table[current_index] == NULL)
428 		current_index++;
429 
430 	/* find a cleanable entry in the bucket and clean it, or set to next bucket */
431 
432 	if (current_detail && current_index < current_detail->hash_size) {
433 		struct cache_head *ch, **cp;
434 		struct cache_detail *d;
435 
436 		write_lock(&current_detail->hash_lock);
437 
438 		/* Ok, now to clean this strand */
439 
440 		cp = & current_detail->hash_table[current_index];
441 		for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
442 			if (current_detail->nextcheck > ch->expiry_time)
443 				current_detail->nextcheck = ch->expiry_time+1;
444 			if (!cache_is_expired(current_detail, ch))
445 				continue;
446 
447 			*cp = ch->next;
448 			ch->next = NULL;
449 			current_detail->entries--;
450 			rv = 1;
451 			break;
452 		}
453 
454 		write_unlock(&current_detail->hash_lock);
455 		d = current_detail;
456 		if (!ch)
457 			current_index ++;
458 		spin_unlock(&cache_list_lock);
459 		if (ch) {
460 			if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
461 				cache_dequeue(current_detail, ch);
462 			cache_revisit_request(ch);
463 			cache_put(ch, d);
464 		}
465 	} else
466 		spin_unlock(&cache_list_lock);
467 
468 	return rv;
469 }
470 
471 /*
472  * We want to regularly clean the cache, so we need to schedule some work ...
473  */
474 static void do_cache_clean(struct work_struct *work)
475 {
476 	int delay = 5;
477 	if (cache_clean() == -1)
478 		delay = round_jiffies_relative(30*HZ);
479 
480 	if (list_empty(&cache_list))
481 		delay = 0;
482 
483 	if (delay)
484 		schedule_delayed_work(&cache_cleaner, delay);
485 }
486 
487 
488 /*
489  * Clean all caches promptly.  This just calls cache_clean
490  * repeatedly until we are sure that every cache has had a chance to
491  * be fully cleaned
492  */
493 void cache_flush(void)
494 {
495 	while (cache_clean() != -1)
496 		cond_resched();
497 	while (cache_clean() != -1)
498 		cond_resched();
499 }
500 EXPORT_SYMBOL_GPL(cache_flush);
501 
502 void cache_purge(struct cache_detail *detail)
503 {
504 	detail->flush_time = LONG_MAX;
505 	detail->nextcheck = seconds_since_boot();
506 	cache_flush();
507 	detail->flush_time = 1;
508 }
509 EXPORT_SYMBOL_GPL(cache_purge);
510 
511 
512 /*
513  * Deferral and Revisiting of Requests.
514  *
515  * If a cache lookup finds a pending entry, we
516  * need to defer the request and revisit it later.
517  * All deferred requests are stored in a hash table,
518  * indexed by "struct cache_head *".
519  * As it may be wasteful to store a whole request
520  * structure, we allow the request to provide a
521  * deferred form, which must contain a
522  * 'struct cache_deferred_req'
523  * This cache_deferred_req contains a method to allow
524  * it to be revisited when cache info is available
525  */
526 
527 #define	DFR_HASHSIZE	(PAGE_SIZE/sizeof(struct list_head))
528 #define	DFR_HASH(item)	((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
529 
530 #define	DFR_MAX	300	/* ??? */
531 
532 static DEFINE_SPINLOCK(cache_defer_lock);
533 static LIST_HEAD(cache_defer_list);
534 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
535 static int cache_defer_cnt;
536 
537 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
538 {
539 	hlist_del_init(&dreq->hash);
540 	if (!list_empty(&dreq->recent)) {
541 		list_del_init(&dreq->recent);
542 		cache_defer_cnt--;
543 	}
544 }
545 
546 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
547 {
548 	int hash = DFR_HASH(item);
549 
550 	INIT_LIST_HEAD(&dreq->recent);
551 	hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
552 }
553 
554 static void setup_deferral(struct cache_deferred_req *dreq,
555 			   struct cache_head *item,
556 			   int count_me)
557 {
558 
559 	dreq->item = item;
560 
561 	spin_lock(&cache_defer_lock);
562 
563 	__hash_deferred_req(dreq, item);
564 
565 	if (count_me) {
566 		cache_defer_cnt++;
567 		list_add(&dreq->recent, &cache_defer_list);
568 	}
569 
570 	spin_unlock(&cache_defer_lock);
571 
572 }
573 
574 struct thread_deferred_req {
575 	struct cache_deferred_req handle;
576 	struct completion completion;
577 };
578 
579 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
580 {
581 	struct thread_deferred_req *dr =
582 		container_of(dreq, struct thread_deferred_req, handle);
583 	complete(&dr->completion);
584 }
585 
586 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
587 {
588 	struct thread_deferred_req sleeper;
589 	struct cache_deferred_req *dreq = &sleeper.handle;
590 
591 	sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
592 	dreq->revisit = cache_restart_thread;
593 
594 	setup_deferral(dreq, item, 0);
595 
596 	if (!test_bit(CACHE_PENDING, &item->flags) ||
597 	    wait_for_completion_interruptible_timeout(
598 		    &sleeper.completion, req->thread_wait) <= 0) {
599 		/* The completion wasn't completed, so we need
600 		 * to clean up
601 		 */
602 		spin_lock(&cache_defer_lock);
603 		if (!hlist_unhashed(&sleeper.handle.hash)) {
604 			__unhash_deferred_req(&sleeper.handle);
605 			spin_unlock(&cache_defer_lock);
606 		} else {
607 			/* cache_revisit_request already removed
608 			 * this from the hash table, but hasn't
609 			 * called ->revisit yet.  It will very soon
610 			 * and we need to wait for it.
611 			 */
612 			spin_unlock(&cache_defer_lock);
613 			wait_for_completion(&sleeper.completion);
614 		}
615 	}
616 }
617 
618 static void cache_limit_defers(void)
619 {
620 	/* Make sure we haven't exceed the limit of allowed deferred
621 	 * requests.
622 	 */
623 	struct cache_deferred_req *discard = NULL;
624 
625 	if (cache_defer_cnt <= DFR_MAX)
626 		return;
627 
628 	spin_lock(&cache_defer_lock);
629 
630 	/* Consider removing either the first or the last */
631 	if (cache_defer_cnt > DFR_MAX) {
632 		if (net_random() & 1)
633 			discard = list_entry(cache_defer_list.next,
634 					     struct cache_deferred_req, recent);
635 		else
636 			discard = list_entry(cache_defer_list.prev,
637 					     struct cache_deferred_req, recent);
638 		__unhash_deferred_req(discard);
639 	}
640 	spin_unlock(&cache_defer_lock);
641 	if (discard)
642 		discard->revisit(discard, 1);
643 }
644 
645 /* Return true if and only if a deferred request is queued. */
646 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
647 {
648 	struct cache_deferred_req *dreq;
649 
650 	if (req->thread_wait) {
651 		cache_wait_req(req, item);
652 		if (!test_bit(CACHE_PENDING, &item->flags))
653 			return false;
654 	}
655 	dreq = req->defer(req);
656 	if (dreq == NULL)
657 		return false;
658 	setup_deferral(dreq, item, 1);
659 	if (!test_bit(CACHE_PENDING, &item->flags))
660 		/* Bit could have been cleared before we managed to
661 		 * set up the deferral, so need to revisit just in case
662 		 */
663 		cache_revisit_request(item);
664 
665 	cache_limit_defers();
666 	return true;
667 }
668 
669 static void cache_revisit_request(struct cache_head *item)
670 {
671 	struct cache_deferred_req *dreq;
672 	struct list_head pending;
673 	struct hlist_node *lp, *tmp;
674 	int hash = DFR_HASH(item);
675 
676 	INIT_LIST_HEAD(&pending);
677 	spin_lock(&cache_defer_lock);
678 
679 	hlist_for_each_entry_safe(dreq, lp, tmp, &cache_defer_hash[hash], hash)
680 		if (dreq->item == item) {
681 			__unhash_deferred_req(dreq);
682 			list_add(&dreq->recent, &pending);
683 		}
684 
685 	spin_unlock(&cache_defer_lock);
686 
687 	while (!list_empty(&pending)) {
688 		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
689 		list_del_init(&dreq->recent);
690 		dreq->revisit(dreq, 0);
691 	}
692 }
693 
694 void cache_clean_deferred(void *owner)
695 {
696 	struct cache_deferred_req *dreq, *tmp;
697 	struct list_head pending;
698 
699 
700 	INIT_LIST_HEAD(&pending);
701 	spin_lock(&cache_defer_lock);
702 
703 	list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
704 		if (dreq->owner == owner) {
705 			__unhash_deferred_req(dreq);
706 			list_add(&dreq->recent, &pending);
707 		}
708 	}
709 	spin_unlock(&cache_defer_lock);
710 
711 	while (!list_empty(&pending)) {
712 		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
713 		list_del_init(&dreq->recent);
714 		dreq->revisit(dreq, 1);
715 	}
716 }
717 
718 /*
719  * communicate with user-space
720  *
721  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
722  * On read, you get a full request, or block.
723  * On write, an update request is processed.
724  * Poll works if anything to read, and always allows write.
725  *
726  * Implemented by linked list of requests.  Each open file has
727  * a ->private that also exists in this list.  New requests are added
728  * to the end and may wakeup and preceding readers.
729  * New readers are added to the head.  If, on read, an item is found with
730  * CACHE_UPCALLING clear, we free it from the list.
731  *
732  */
733 
734 static DEFINE_SPINLOCK(queue_lock);
735 static DEFINE_MUTEX(queue_io_mutex);
736 
737 struct cache_queue {
738 	struct list_head	list;
739 	int			reader;	/* if 0, then request */
740 };
741 struct cache_request {
742 	struct cache_queue	q;
743 	struct cache_head	*item;
744 	char			* buf;
745 	int			len;
746 	int			readers;
747 };
748 struct cache_reader {
749 	struct cache_queue	q;
750 	int			offset;	/* if non-0, we have a refcnt on next request */
751 };
752 
753 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
754 			  loff_t *ppos, struct cache_detail *cd)
755 {
756 	struct cache_reader *rp = filp->private_data;
757 	struct cache_request *rq;
758 	struct inode *inode = filp->f_path.dentry->d_inode;
759 	int err;
760 
761 	if (count == 0)
762 		return 0;
763 
764 	mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
765 			      * readers on this file */
766  again:
767 	spin_lock(&queue_lock);
768 	/* need to find next request */
769 	while (rp->q.list.next != &cd->queue &&
770 	       list_entry(rp->q.list.next, struct cache_queue, list)
771 	       ->reader) {
772 		struct list_head *next = rp->q.list.next;
773 		list_move(&rp->q.list, next);
774 	}
775 	if (rp->q.list.next == &cd->queue) {
776 		spin_unlock(&queue_lock);
777 		mutex_unlock(&inode->i_mutex);
778 		BUG_ON(rp->offset);
779 		return 0;
780 	}
781 	rq = container_of(rp->q.list.next, struct cache_request, q.list);
782 	BUG_ON(rq->q.reader);
783 	if (rp->offset == 0)
784 		rq->readers++;
785 	spin_unlock(&queue_lock);
786 
787 	if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
788 		err = -EAGAIN;
789 		spin_lock(&queue_lock);
790 		list_move(&rp->q.list, &rq->q.list);
791 		spin_unlock(&queue_lock);
792 	} else {
793 		if (rp->offset + count > rq->len)
794 			count = rq->len - rp->offset;
795 		err = -EFAULT;
796 		if (copy_to_user(buf, rq->buf + rp->offset, count))
797 			goto out;
798 		rp->offset += count;
799 		if (rp->offset >= rq->len) {
800 			rp->offset = 0;
801 			spin_lock(&queue_lock);
802 			list_move(&rp->q.list, &rq->q.list);
803 			spin_unlock(&queue_lock);
804 		}
805 		err = 0;
806 	}
807  out:
808 	if (rp->offset == 0) {
809 		/* need to release rq */
810 		spin_lock(&queue_lock);
811 		rq->readers--;
812 		if (rq->readers == 0 &&
813 		    !test_bit(CACHE_PENDING, &rq->item->flags)) {
814 			list_del(&rq->q.list);
815 			spin_unlock(&queue_lock);
816 			cache_put(rq->item, cd);
817 			kfree(rq->buf);
818 			kfree(rq);
819 		} else
820 			spin_unlock(&queue_lock);
821 	}
822 	if (err == -EAGAIN)
823 		goto again;
824 	mutex_unlock(&inode->i_mutex);
825 	return err ? err :  count;
826 }
827 
828 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
829 				 size_t count, struct cache_detail *cd)
830 {
831 	ssize_t ret;
832 
833 	if (copy_from_user(kaddr, buf, count))
834 		return -EFAULT;
835 	kaddr[count] = '\0';
836 	ret = cd->cache_parse(cd, kaddr, count);
837 	if (!ret)
838 		ret = count;
839 	return ret;
840 }
841 
842 static ssize_t cache_slow_downcall(const char __user *buf,
843 				   size_t count, struct cache_detail *cd)
844 {
845 	static char write_buf[8192]; /* protected by queue_io_mutex */
846 	ssize_t ret = -EINVAL;
847 
848 	if (count >= sizeof(write_buf))
849 		goto out;
850 	mutex_lock(&queue_io_mutex);
851 	ret = cache_do_downcall(write_buf, buf, count, cd);
852 	mutex_unlock(&queue_io_mutex);
853 out:
854 	return ret;
855 }
856 
857 static ssize_t cache_downcall(struct address_space *mapping,
858 			      const char __user *buf,
859 			      size_t count, struct cache_detail *cd)
860 {
861 	struct page *page;
862 	char *kaddr;
863 	ssize_t ret = -ENOMEM;
864 
865 	if (count >= PAGE_CACHE_SIZE)
866 		goto out_slow;
867 
868 	page = find_or_create_page(mapping, 0, GFP_KERNEL);
869 	if (!page)
870 		goto out_slow;
871 
872 	kaddr = kmap(page);
873 	ret = cache_do_downcall(kaddr, buf, count, cd);
874 	kunmap(page);
875 	unlock_page(page);
876 	page_cache_release(page);
877 	return ret;
878 out_slow:
879 	return cache_slow_downcall(buf, count, cd);
880 }
881 
882 static ssize_t cache_write(struct file *filp, const char __user *buf,
883 			   size_t count, loff_t *ppos,
884 			   struct cache_detail *cd)
885 {
886 	struct address_space *mapping = filp->f_mapping;
887 	struct inode *inode = filp->f_path.dentry->d_inode;
888 	ssize_t ret = -EINVAL;
889 
890 	if (!cd->cache_parse)
891 		goto out;
892 
893 	mutex_lock(&inode->i_mutex);
894 	ret = cache_downcall(mapping, buf, count, cd);
895 	mutex_unlock(&inode->i_mutex);
896 out:
897 	return ret;
898 }
899 
900 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
901 
902 static unsigned int cache_poll(struct file *filp, poll_table *wait,
903 			       struct cache_detail *cd)
904 {
905 	unsigned int mask;
906 	struct cache_reader *rp = filp->private_data;
907 	struct cache_queue *cq;
908 
909 	poll_wait(filp, &queue_wait, wait);
910 
911 	/* alway allow write */
912 	mask = POLL_OUT | POLLWRNORM;
913 
914 	if (!rp)
915 		return mask;
916 
917 	spin_lock(&queue_lock);
918 
919 	for (cq= &rp->q; &cq->list != &cd->queue;
920 	     cq = list_entry(cq->list.next, struct cache_queue, list))
921 		if (!cq->reader) {
922 			mask |= POLLIN | POLLRDNORM;
923 			break;
924 		}
925 	spin_unlock(&queue_lock);
926 	return mask;
927 }
928 
929 static int cache_ioctl(struct inode *ino, struct file *filp,
930 		       unsigned int cmd, unsigned long arg,
931 		       struct cache_detail *cd)
932 {
933 	int len = 0;
934 	struct cache_reader *rp = filp->private_data;
935 	struct cache_queue *cq;
936 
937 	if (cmd != FIONREAD || !rp)
938 		return -EINVAL;
939 
940 	spin_lock(&queue_lock);
941 
942 	/* only find the length remaining in current request,
943 	 * or the length of the next request
944 	 */
945 	for (cq= &rp->q; &cq->list != &cd->queue;
946 	     cq = list_entry(cq->list.next, struct cache_queue, list))
947 		if (!cq->reader) {
948 			struct cache_request *cr =
949 				container_of(cq, struct cache_request, q);
950 			len = cr->len - rp->offset;
951 			break;
952 		}
953 	spin_unlock(&queue_lock);
954 
955 	return put_user(len, (int __user *)arg);
956 }
957 
958 static int cache_open(struct inode *inode, struct file *filp,
959 		      struct cache_detail *cd)
960 {
961 	struct cache_reader *rp = NULL;
962 
963 	if (!cd || !try_module_get(cd->owner))
964 		return -EACCES;
965 	nonseekable_open(inode, filp);
966 	if (filp->f_mode & FMODE_READ) {
967 		rp = kmalloc(sizeof(*rp), GFP_KERNEL);
968 		if (!rp)
969 			return -ENOMEM;
970 		rp->offset = 0;
971 		rp->q.reader = 1;
972 		atomic_inc(&cd->readers);
973 		spin_lock(&queue_lock);
974 		list_add(&rp->q.list, &cd->queue);
975 		spin_unlock(&queue_lock);
976 	}
977 	filp->private_data = rp;
978 	return 0;
979 }
980 
981 static int cache_release(struct inode *inode, struct file *filp,
982 			 struct cache_detail *cd)
983 {
984 	struct cache_reader *rp = filp->private_data;
985 
986 	if (rp) {
987 		spin_lock(&queue_lock);
988 		if (rp->offset) {
989 			struct cache_queue *cq;
990 			for (cq= &rp->q; &cq->list != &cd->queue;
991 			     cq = list_entry(cq->list.next, struct cache_queue, list))
992 				if (!cq->reader) {
993 					container_of(cq, struct cache_request, q)
994 						->readers--;
995 					break;
996 				}
997 			rp->offset = 0;
998 		}
999 		list_del(&rp->q.list);
1000 		spin_unlock(&queue_lock);
1001 
1002 		filp->private_data = NULL;
1003 		kfree(rp);
1004 
1005 		cd->last_close = seconds_since_boot();
1006 		atomic_dec(&cd->readers);
1007 	}
1008 	module_put(cd->owner);
1009 	return 0;
1010 }
1011 
1012 
1013 
1014 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1015 {
1016 	struct cache_queue *cq;
1017 	spin_lock(&queue_lock);
1018 	list_for_each_entry(cq, &detail->queue, list)
1019 		if (!cq->reader) {
1020 			struct cache_request *cr = container_of(cq, struct cache_request, q);
1021 			if (cr->item != ch)
1022 				continue;
1023 			if (cr->readers != 0)
1024 				continue;
1025 			list_del(&cr->q.list);
1026 			spin_unlock(&queue_lock);
1027 			cache_put(cr->item, detail);
1028 			kfree(cr->buf);
1029 			kfree(cr);
1030 			return;
1031 		}
1032 	spin_unlock(&queue_lock);
1033 }
1034 
1035 /*
1036  * Support routines for text-based upcalls.
1037  * Fields are separated by spaces.
1038  * Fields are either mangled to quote space tab newline slosh with slosh
1039  * or a hexified with a leading \x
1040  * Record is terminated with newline.
1041  *
1042  */
1043 
1044 void qword_add(char **bpp, int *lp, char *str)
1045 {
1046 	char *bp = *bpp;
1047 	int len = *lp;
1048 	char c;
1049 
1050 	if (len < 0) return;
1051 
1052 	while ((c=*str++) && len)
1053 		switch(c) {
1054 		case ' ':
1055 		case '\t':
1056 		case '\n':
1057 		case '\\':
1058 			if (len >= 4) {
1059 				*bp++ = '\\';
1060 				*bp++ = '0' + ((c & 0300)>>6);
1061 				*bp++ = '0' + ((c & 0070)>>3);
1062 				*bp++ = '0' + ((c & 0007)>>0);
1063 			}
1064 			len -= 4;
1065 			break;
1066 		default:
1067 			*bp++ = c;
1068 			len--;
1069 		}
1070 	if (c || len <1) len = -1;
1071 	else {
1072 		*bp++ = ' ';
1073 		len--;
1074 	}
1075 	*bpp = bp;
1076 	*lp = len;
1077 }
1078 EXPORT_SYMBOL_GPL(qword_add);
1079 
1080 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1081 {
1082 	char *bp = *bpp;
1083 	int len = *lp;
1084 
1085 	if (len < 0) return;
1086 
1087 	if (len > 2) {
1088 		*bp++ = '\\';
1089 		*bp++ = 'x';
1090 		len -= 2;
1091 		while (blen && len >= 2) {
1092 			unsigned char c = *buf++;
1093 			*bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1094 			*bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1095 			len -= 2;
1096 			blen--;
1097 		}
1098 	}
1099 	if (blen || len<1) len = -1;
1100 	else {
1101 		*bp++ = ' ';
1102 		len--;
1103 	}
1104 	*bpp = bp;
1105 	*lp = len;
1106 }
1107 EXPORT_SYMBOL_GPL(qword_addhex);
1108 
1109 static void warn_no_listener(struct cache_detail *detail)
1110 {
1111 	if (detail->last_warn != detail->last_close) {
1112 		detail->last_warn = detail->last_close;
1113 		if (detail->warn_no_listener)
1114 			detail->warn_no_listener(detail, detail->last_close != 0);
1115 	}
1116 }
1117 
1118 static bool cache_listeners_exist(struct cache_detail *detail)
1119 {
1120 	if (atomic_read(&detail->readers))
1121 		return true;
1122 	if (detail->last_close == 0)
1123 		/* This cache was never opened */
1124 		return false;
1125 	if (detail->last_close < seconds_since_boot() - 30)
1126 		/*
1127 		 * We allow for the possibility that someone might
1128 		 * restart a userspace daemon without restarting the
1129 		 * server; but after 30 seconds, we give up.
1130 		 */
1131 		 return false;
1132 	return true;
1133 }
1134 
1135 /*
1136  * register an upcall request to user-space and queue it up for read() by the
1137  * upcall daemon.
1138  *
1139  * Each request is at most one page long.
1140  */
1141 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1142 		void (*cache_request)(struct cache_detail *,
1143 				      struct cache_head *,
1144 				      char **,
1145 				      int *))
1146 {
1147 
1148 	char *buf;
1149 	struct cache_request *crq;
1150 	char *bp;
1151 	int len;
1152 
1153 	if (!cache_listeners_exist(detail)) {
1154 		warn_no_listener(detail);
1155 		return -EINVAL;
1156 	}
1157 
1158 	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1159 	if (!buf)
1160 		return -EAGAIN;
1161 
1162 	crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1163 	if (!crq) {
1164 		kfree(buf);
1165 		return -EAGAIN;
1166 	}
1167 
1168 	bp = buf; len = PAGE_SIZE;
1169 
1170 	cache_request(detail, h, &bp, &len);
1171 
1172 	if (len < 0) {
1173 		kfree(buf);
1174 		kfree(crq);
1175 		return -EAGAIN;
1176 	}
1177 	crq->q.reader = 0;
1178 	crq->item = cache_get(h);
1179 	crq->buf = buf;
1180 	crq->len = PAGE_SIZE - len;
1181 	crq->readers = 0;
1182 	spin_lock(&queue_lock);
1183 	list_add_tail(&crq->q.list, &detail->queue);
1184 	spin_unlock(&queue_lock);
1185 	wake_up(&queue_wait);
1186 	return 0;
1187 }
1188 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1189 
1190 /*
1191  * parse a message from user-space and pass it
1192  * to an appropriate cache
1193  * Messages are, like requests, separated into fields by
1194  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1195  *
1196  * Message is
1197  *   reply cachename expiry key ... content....
1198  *
1199  * key and content are both parsed by cache
1200  */
1201 
1202 #define isodigit(c) (isdigit(c) && c <= '7')
1203 int qword_get(char **bpp, char *dest, int bufsize)
1204 {
1205 	/* return bytes copied, or -1 on error */
1206 	char *bp = *bpp;
1207 	int len = 0;
1208 
1209 	while (*bp == ' ') bp++;
1210 
1211 	if (bp[0] == '\\' && bp[1] == 'x') {
1212 		/* HEX STRING */
1213 		bp += 2;
1214 		while (len < bufsize) {
1215 			int h, l;
1216 
1217 			h = hex_to_bin(bp[0]);
1218 			if (h < 0)
1219 				break;
1220 
1221 			l = hex_to_bin(bp[1]);
1222 			if (l < 0)
1223 				break;
1224 
1225 			*dest++ = (h << 4) | l;
1226 			bp += 2;
1227 			len++;
1228 		}
1229 	} else {
1230 		/* text with \nnn octal quoting */
1231 		while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1232 			if (*bp == '\\' &&
1233 			    isodigit(bp[1]) && (bp[1] <= '3') &&
1234 			    isodigit(bp[2]) &&
1235 			    isodigit(bp[3])) {
1236 				int byte = (*++bp -'0');
1237 				bp++;
1238 				byte = (byte << 3) | (*bp++ - '0');
1239 				byte = (byte << 3) | (*bp++ - '0');
1240 				*dest++ = byte;
1241 				len++;
1242 			} else {
1243 				*dest++ = *bp++;
1244 				len++;
1245 			}
1246 		}
1247 	}
1248 
1249 	if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1250 		return -1;
1251 	while (*bp == ' ') bp++;
1252 	*bpp = bp;
1253 	*dest = '\0';
1254 	return len;
1255 }
1256 EXPORT_SYMBOL_GPL(qword_get);
1257 
1258 
1259 /*
1260  * support /proc/sunrpc/cache/$CACHENAME/content
1261  * as a seqfile.
1262  * We call ->cache_show passing NULL for the item to
1263  * get a header, then pass each real item in the cache
1264  */
1265 
1266 struct handle {
1267 	struct cache_detail *cd;
1268 };
1269 
1270 static void *c_start(struct seq_file *m, loff_t *pos)
1271 	__acquires(cd->hash_lock)
1272 {
1273 	loff_t n = *pos;
1274 	unsigned hash, entry;
1275 	struct cache_head *ch;
1276 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1277 
1278 
1279 	read_lock(&cd->hash_lock);
1280 	if (!n--)
1281 		return SEQ_START_TOKEN;
1282 	hash = n >> 32;
1283 	entry = n & ((1LL<<32) - 1);
1284 
1285 	for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1286 		if (!entry--)
1287 			return ch;
1288 	n &= ~((1LL<<32) - 1);
1289 	do {
1290 		hash++;
1291 		n += 1LL<<32;
1292 	} while(hash < cd->hash_size &&
1293 		cd->hash_table[hash]==NULL);
1294 	if (hash >= cd->hash_size)
1295 		return NULL;
1296 	*pos = n+1;
1297 	return cd->hash_table[hash];
1298 }
1299 
1300 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1301 {
1302 	struct cache_head *ch = p;
1303 	int hash = (*pos >> 32);
1304 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1305 
1306 	if (p == SEQ_START_TOKEN)
1307 		hash = 0;
1308 	else if (ch->next == NULL) {
1309 		hash++;
1310 		*pos += 1LL<<32;
1311 	} else {
1312 		++*pos;
1313 		return ch->next;
1314 	}
1315 	*pos &= ~((1LL<<32) - 1);
1316 	while (hash < cd->hash_size &&
1317 	       cd->hash_table[hash] == NULL) {
1318 		hash++;
1319 		*pos += 1LL<<32;
1320 	}
1321 	if (hash >= cd->hash_size)
1322 		return NULL;
1323 	++*pos;
1324 	return cd->hash_table[hash];
1325 }
1326 
1327 static void c_stop(struct seq_file *m, void *p)
1328 	__releases(cd->hash_lock)
1329 {
1330 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1331 	read_unlock(&cd->hash_lock);
1332 }
1333 
1334 static int c_show(struct seq_file *m, void *p)
1335 {
1336 	struct cache_head *cp = p;
1337 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1338 
1339 	if (p == SEQ_START_TOKEN)
1340 		return cd->cache_show(m, cd, NULL);
1341 
1342 	ifdebug(CACHE)
1343 		seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1344 			   convert_to_wallclock(cp->expiry_time),
1345 			   atomic_read(&cp->ref.refcount), cp->flags);
1346 	cache_get(cp);
1347 	if (cache_check(cd, cp, NULL))
1348 		/* cache_check does a cache_put on failure */
1349 		seq_printf(m, "# ");
1350 	else
1351 		cache_put(cp, cd);
1352 
1353 	return cd->cache_show(m, cd, cp);
1354 }
1355 
1356 static const struct seq_operations cache_content_op = {
1357 	.start	= c_start,
1358 	.next	= c_next,
1359 	.stop	= c_stop,
1360 	.show	= c_show,
1361 };
1362 
1363 static int content_open(struct inode *inode, struct file *file,
1364 			struct cache_detail *cd)
1365 {
1366 	struct handle *han;
1367 
1368 	if (!cd || !try_module_get(cd->owner))
1369 		return -EACCES;
1370 	han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1371 	if (han == NULL) {
1372 		module_put(cd->owner);
1373 		return -ENOMEM;
1374 	}
1375 
1376 	han->cd = cd;
1377 	return 0;
1378 }
1379 
1380 static int content_release(struct inode *inode, struct file *file,
1381 		struct cache_detail *cd)
1382 {
1383 	int ret = seq_release_private(inode, file);
1384 	module_put(cd->owner);
1385 	return ret;
1386 }
1387 
1388 static int open_flush(struct inode *inode, struct file *file,
1389 			struct cache_detail *cd)
1390 {
1391 	if (!cd || !try_module_get(cd->owner))
1392 		return -EACCES;
1393 	return nonseekable_open(inode, file);
1394 }
1395 
1396 static int release_flush(struct inode *inode, struct file *file,
1397 			struct cache_detail *cd)
1398 {
1399 	module_put(cd->owner);
1400 	return 0;
1401 }
1402 
1403 static ssize_t read_flush(struct file *file, char __user *buf,
1404 			  size_t count, loff_t *ppos,
1405 			  struct cache_detail *cd)
1406 {
1407 	char tbuf[20];
1408 	unsigned long p = *ppos;
1409 	size_t len;
1410 
1411 	sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1412 	len = strlen(tbuf);
1413 	if (p >= len)
1414 		return 0;
1415 	len -= p;
1416 	if (len > count)
1417 		len = count;
1418 	if (copy_to_user(buf, (void*)(tbuf+p), len))
1419 		return -EFAULT;
1420 	*ppos += len;
1421 	return len;
1422 }
1423 
1424 static ssize_t write_flush(struct file *file, const char __user *buf,
1425 			   size_t count, loff_t *ppos,
1426 			   struct cache_detail *cd)
1427 {
1428 	char tbuf[20];
1429 	char *bp, *ep;
1430 
1431 	if (*ppos || count > sizeof(tbuf)-1)
1432 		return -EINVAL;
1433 	if (copy_from_user(tbuf, buf, count))
1434 		return -EFAULT;
1435 	tbuf[count] = 0;
1436 	simple_strtoul(tbuf, &ep, 0);
1437 	if (*ep && *ep != '\n')
1438 		return -EINVAL;
1439 
1440 	bp = tbuf;
1441 	cd->flush_time = get_expiry(&bp);
1442 	cd->nextcheck = seconds_since_boot();
1443 	cache_flush();
1444 
1445 	*ppos += count;
1446 	return count;
1447 }
1448 
1449 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1450 				 size_t count, loff_t *ppos)
1451 {
1452 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1453 
1454 	return cache_read(filp, buf, count, ppos, cd);
1455 }
1456 
1457 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1458 				  size_t count, loff_t *ppos)
1459 {
1460 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1461 
1462 	return cache_write(filp, buf, count, ppos, cd);
1463 }
1464 
1465 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1466 {
1467 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1468 
1469 	return cache_poll(filp, wait, cd);
1470 }
1471 
1472 static long cache_ioctl_procfs(struct file *filp,
1473 			       unsigned int cmd, unsigned long arg)
1474 {
1475 	struct inode *inode = filp->f_path.dentry->d_inode;
1476 	struct cache_detail *cd = PDE(inode)->data;
1477 
1478 	return cache_ioctl(inode, filp, cmd, arg, cd);
1479 }
1480 
1481 static int cache_open_procfs(struct inode *inode, struct file *filp)
1482 {
1483 	struct cache_detail *cd = PDE(inode)->data;
1484 
1485 	return cache_open(inode, filp, cd);
1486 }
1487 
1488 static int cache_release_procfs(struct inode *inode, struct file *filp)
1489 {
1490 	struct cache_detail *cd = PDE(inode)->data;
1491 
1492 	return cache_release(inode, filp, cd);
1493 }
1494 
1495 static const struct file_operations cache_file_operations_procfs = {
1496 	.owner		= THIS_MODULE,
1497 	.llseek		= no_llseek,
1498 	.read		= cache_read_procfs,
1499 	.write		= cache_write_procfs,
1500 	.poll		= cache_poll_procfs,
1501 	.unlocked_ioctl	= cache_ioctl_procfs, /* for FIONREAD */
1502 	.open		= cache_open_procfs,
1503 	.release	= cache_release_procfs,
1504 };
1505 
1506 static int content_open_procfs(struct inode *inode, struct file *filp)
1507 {
1508 	struct cache_detail *cd = PDE(inode)->data;
1509 
1510 	return content_open(inode, filp, cd);
1511 }
1512 
1513 static int content_release_procfs(struct inode *inode, struct file *filp)
1514 {
1515 	struct cache_detail *cd = PDE(inode)->data;
1516 
1517 	return content_release(inode, filp, cd);
1518 }
1519 
1520 static const struct file_operations content_file_operations_procfs = {
1521 	.open		= content_open_procfs,
1522 	.read		= seq_read,
1523 	.llseek		= seq_lseek,
1524 	.release	= content_release_procfs,
1525 };
1526 
1527 static int open_flush_procfs(struct inode *inode, struct file *filp)
1528 {
1529 	struct cache_detail *cd = PDE(inode)->data;
1530 
1531 	return open_flush(inode, filp, cd);
1532 }
1533 
1534 static int release_flush_procfs(struct inode *inode, struct file *filp)
1535 {
1536 	struct cache_detail *cd = PDE(inode)->data;
1537 
1538 	return release_flush(inode, filp, cd);
1539 }
1540 
1541 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1542 			    size_t count, loff_t *ppos)
1543 {
1544 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1545 
1546 	return read_flush(filp, buf, count, ppos, cd);
1547 }
1548 
1549 static ssize_t write_flush_procfs(struct file *filp,
1550 				  const char __user *buf,
1551 				  size_t count, loff_t *ppos)
1552 {
1553 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1554 
1555 	return write_flush(filp, buf, count, ppos, cd);
1556 }
1557 
1558 static const struct file_operations cache_flush_operations_procfs = {
1559 	.open		= open_flush_procfs,
1560 	.read		= read_flush_procfs,
1561 	.write		= write_flush_procfs,
1562 	.release	= release_flush_procfs,
1563 	.llseek		= no_llseek,
1564 };
1565 
1566 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1567 {
1568 	struct sunrpc_net *sn;
1569 
1570 	if (cd->u.procfs.proc_ent == NULL)
1571 		return;
1572 	if (cd->u.procfs.flush_ent)
1573 		remove_proc_entry("flush", cd->u.procfs.proc_ent);
1574 	if (cd->u.procfs.channel_ent)
1575 		remove_proc_entry("channel", cd->u.procfs.proc_ent);
1576 	if (cd->u.procfs.content_ent)
1577 		remove_proc_entry("content", cd->u.procfs.proc_ent);
1578 	cd->u.procfs.proc_ent = NULL;
1579 	sn = net_generic(net, sunrpc_net_id);
1580 	remove_proc_entry(cd->name, sn->proc_net_rpc);
1581 }
1582 
1583 #ifdef CONFIG_PROC_FS
1584 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1585 {
1586 	struct proc_dir_entry *p;
1587 	struct sunrpc_net *sn;
1588 
1589 	sn = net_generic(net, sunrpc_net_id);
1590 	cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1591 	if (cd->u.procfs.proc_ent == NULL)
1592 		goto out_nomem;
1593 	cd->u.procfs.channel_ent = NULL;
1594 	cd->u.procfs.content_ent = NULL;
1595 
1596 	p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1597 			     cd->u.procfs.proc_ent,
1598 			     &cache_flush_operations_procfs, cd);
1599 	cd->u.procfs.flush_ent = p;
1600 	if (p == NULL)
1601 		goto out_nomem;
1602 
1603 	if (cd->cache_upcall || cd->cache_parse) {
1604 		p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1605 				     cd->u.procfs.proc_ent,
1606 				     &cache_file_operations_procfs, cd);
1607 		cd->u.procfs.channel_ent = p;
1608 		if (p == NULL)
1609 			goto out_nomem;
1610 	}
1611 	if (cd->cache_show) {
1612 		p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1613 				cd->u.procfs.proc_ent,
1614 				&content_file_operations_procfs, cd);
1615 		cd->u.procfs.content_ent = p;
1616 		if (p == NULL)
1617 			goto out_nomem;
1618 	}
1619 	return 0;
1620 out_nomem:
1621 	remove_cache_proc_entries(cd, net);
1622 	return -ENOMEM;
1623 }
1624 #else /* CONFIG_PROC_FS */
1625 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1626 {
1627 	return 0;
1628 }
1629 #endif
1630 
1631 void __init cache_initialize(void)
1632 {
1633 	INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1634 }
1635 
1636 int cache_register_net(struct cache_detail *cd, struct net *net)
1637 {
1638 	int ret;
1639 
1640 	sunrpc_init_cache_detail(cd);
1641 	ret = create_cache_proc_entries(cd, net);
1642 	if (ret)
1643 		sunrpc_destroy_cache_detail(cd);
1644 	return ret;
1645 }
1646 EXPORT_SYMBOL_GPL(cache_register_net);
1647 
1648 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1649 {
1650 	remove_cache_proc_entries(cd, net);
1651 	sunrpc_destroy_cache_detail(cd);
1652 }
1653 EXPORT_SYMBOL_GPL(cache_unregister_net);
1654 
1655 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1656 {
1657 	struct cache_detail *cd;
1658 
1659 	cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1660 	if (cd == NULL)
1661 		return ERR_PTR(-ENOMEM);
1662 
1663 	cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1664 				 GFP_KERNEL);
1665 	if (cd->hash_table == NULL) {
1666 		kfree(cd);
1667 		return ERR_PTR(-ENOMEM);
1668 	}
1669 	cd->net = net;
1670 	return cd;
1671 }
1672 EXPORT_SYMBOL_GPL(cache_create_net);
1673 
1674 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1675 {
1676 	kfree(cd->hash_table);
1677 	kfree(cd);
1678 }
1679 EXPORT_SYMBOL_GPL(cache_destroy_net);
1680 
1681 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1682 				 size_t count, loff_t *ppos)
1683 {
1684 	struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1685 
1686 	return cache_read(filp, buf, count, ppos, cd);
1687 }
1688 
1689 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1690 				  size_t count, loff_t *ppos)
1691 {
1692 	struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1693 
1694 	return cache_write(filp, buf, count, ppos, cd);
1695 }
1696 
1697 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1698 {
1699 	struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1700 
1701 	return cache_poll(filp, wait, cd);
1702 }
1703 
1704 static long cache_ioctl_pipefs(struct file *filp,
1705 			      unsigned int cmd, unsigned long arg)
1706 {
1707 	struct inode *inode = filp->f_dentry->d_inode;
1708 	struct cache_detail *cd = RPC_I(inode)->private;
1709 
1710 	return cache_ioctl(inode, filp, cmd, arg, cd);
1711 }
1712 
1713 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1714 {
1715 	struct cache_detail *cd = RPC_I(inode)->private;
1716 
1717 	return cache_open(inode, filp, cd);
1718 }
1719 
1720 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1721 {
1722 	struct cache_detail *cd = RPC_I(inode)->private;
1723 
1724 	return cache_release(inode, filp, cd);
1725 }
1726 
1727 const struct file_operations cache_file_operations_pipefs = {
1728 	.owner		= THIS_MODULE,
1729 	.llseek		= no_llseek,
1730 	.read		= cache_read_pipefs,
1731 	.write		= cache_write_pipefs,
1732 	.poll		= cache_poll_pipefs,
1733 	.unlocked_ioctl	= cache_ioctl_pipefs, /* for FIONREAD */
1734 	.open		= cache_open_pipefs,
1735 	.release	= cache_release_pipefs,
1736 };
1737 
1738 static int content_open_pipefs(struct inode *inode, struct file *filp)
1739 {
1740 	struct cache_detail *cd = RPC_I(inode)->private;
1741 
1742 	return content_open(inode, filp, cd);
1743 }
1744 
1745 static int content_release_pipefs(struct inode *inode, struct file *filp)
1746 {
1747 	struct cache_detail *cd = RPC_I(inode)->private;
1748 
1749 	return content_release(inode, filp, cd);
1750 }
1751 
1752 const struct file_operations content_file_operations_pipefs = {
1753 	.open		= content_open_pipefs,
1754 	.read		= seq_read,
1755 	.llseek		= seq_lseek,
1756 	.release	= content_release_pipefs,
1757 };
1758 
1759 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1760 {
1761 	struct cache_detail *cd = RPC_I(inode)->private;
1762 
1763 	return open_flush(inode, filp, cd);
1764 }
1765 
1766 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1767 {
1768 	struct cache_detail *cd = RPC_I(inode)->private;
1769 
1770 	return release_flush(inode, filp, cd);
1771 }
1772 
1773 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1774 			    size_t count, loff_t *ppos)
1775 {
1776 	struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1777 
1778 	return read_flush(filp, buf, count, ppos, cd);
1779 }
1780 
1781 static ssize_t write_flush_pipefs(struct file *filp,
1782 				  const char __user *buf,
1783 				  size_t count, loff_t *ppos)
1784 {
1785 	struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1786 
1787 	return write_flush(filp, buf, count, ppos, cd);
1788 }
1789 
1790 const struct file_operations cache_flush_operations_pipefs = {
1791 	.open		= open_flush_pipefs,
1792 	.read		= read_flush_pipefs,
1793 	.write		= write_flush_pipefs,
1794 	.release	= release_flush_pipefs,
1795 	.llseek		= no_llseek,
1796 };
1797 
1798 int sunrpc_cache_register_pipefs(struct dentry *parent,
1799 				 const char *name, umode_t umode,
1800 				 struct cache_detail *cd)
1801 {
1802 	struct qstr q;
1803 	struct dentry *dir;
1804 	int ret = 0;
1805 
1806 	q.name = name;
1807 	q.len = strlen(name);
1808 	q.hash = full_name_hash(q.name, q.len);
1809 	dir = rpc_create_cache_dir(parent, &q, umode, cd);
1810 	if (!IS_ERR(dir))
1811 		cd->u.pipefs.dir = dir;
1812 	else
1813 		ret = PTR_ERR(dir);
1814 	return ret;
1815 }
1816 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1817 
1818 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1819 {
1820 	rpc_remove_cache_dir(cd->u.pipefs.dir);
1821 	cd->u.pipefs.dir = NULL;
1822 }
1823 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1824 
1825