cache.c revision e7fe23363bab0488c7ce09626900e7d621ea2292
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 <asm/ioctls.h>
31#include <linux/sunrpc/types.h>
32#include <linux/sunrpc/cache.h>
33#include <linux/sunrpc/stats.h>
34
35#define	 RPCDBG_FACILITY RPCDBG_CACHE
36
37static int cache_defer_req(struct cache_req *req, struct cache_head *item);
38static void cache_revisit_request(struct cache_head *item);
39
40static void cache_init(struct cache_head *h)
41{
42	time_t now = get_seconds();
43	h->next = NULL;
44	h->flags = 0;
45	kref_init(&h->ref);
46	h->expiry_time = now + CACHE_NEW_EXPIRY;
47	h->last_refresh = now;
48}
49
50struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
51				       struct cache_head *key, int hash)
52{
53	struct cache_head **head,  **hp;
54	struct cache_head *new = NULL;
55
56	head = &detail->hash_table[hash];
57
58	read_lock(&detail->hash_lock);
59
60	for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
61		struct cache_head *tmp = *hp;
62		if (detail->match(tmp, key)) {
63			cache_get(tmp);
64			read_unlock(&detail->hash_lock);
65			return tmp;
66		}
67	}
68	read_unlock(&detail->hash_lock);
69	/* Didn't find anything, insert an empty entry */
70
71	new = detail->alloc();
72	if (!new)
73		return NULL;
74	/* must fully initialise 'new', else
75	 * we might get lose if we need to
76	 * cache_put it soon.
77	 */
78	cache_init(new);
79	detail->init(new, key);
80
81	write_lock(&detail->hash_lock);
82
83	/* check if entry appeared while we slept */
84	for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
85		struct cache_head *tmp = *hp;
86		if (detail->match(tmp, key)) {
87			cache_get(tmp);
88			write_unlock(&detail->hash_lock);
89			cache_put(new, detail);
90			return tmp;
91		}
92	}
93	new->next = *head;
94	*head = new;
95	detail->entries++;
96	cache_get(new);
97	write_unlock(&detail->hash_lock);
98
99	return new;
100}
101EXPORT_SYMBOL(sunrpc_cache_lookup);
102
103
104static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
105
106static int cache_fresh_locked(struct cache_head *head, time_t expiry)
107{
108	head->expiry_time = expiry;
109	head->last_refresh = get_seconds();
110	return !test_and_set_bit(CACHE_VALID, &head->flags);
111}
112
113static void cache_fresh_unlocked(struct cache_head *head,
114			struct cache_detail *detail, int new)
115{
116	if (new)
117		cache_revisit_request(head);
118	if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
119		cache_revisit_request(head);
120		queue_loose(detail, head);
121	}
122}
123
124struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
125				       struct cache_head *new, struct cache_head *old, int hash)
126{
127	/* The 'old' entry is to be replaced by 'new'.
128	 * If 'old' is not VALID, we update it directly,
129	 * otherwise we need to replace it
130	 */
131	struct cache_head **head;
132	struct cache_head *tmp;
133	int is_new;
134
135	if (!test_bit(CACHE_VALID, &old->flags)) {
136		write_lock(&detail->hash_lock);
137		if (!test_bit(CACHE_VALID, &old->flags)) {
138			if (test_bit(CACHE_NEGATIVE, &new->flags))
139				set_bit(CACHE_NEGATIVE, &old->flags);
140			else
141				detail->update(old, new);
142			is_new = cache_fresh_locked(old, new->expiry_time);
143			write_unlock(&detail->hash_lock);
144			cache_fresh_unlocked(old, detail, is_new);
145			return old;
146		}
147		write_unlock(&detail->hash_lock);
148	}
149	/* We need to insert a new entry */
150	tmp = detail->alloc();
151	if (!tmp) {
152		cache_put(old, detail);
153		return NULL;
154	}
155	cache_init(tmp);
156	detail->init(tmp, old);
157	head = &detail->hash_table[hash];
158
159	write_lock(&detail->hash_lock);
160	if (test_bit(CACHE_NEGATIVE, &new->flags))
161		set_bit(CACHE_NEGATIVE, &tmp->flags);
162	else
163		detail->update(tmp, new);
164	tmp->next = *head;
165	*head = tmp;
166	detail->entries++;
167	cache_get(tmp);
168	is_new = cache_fresh_locked(tmp, new->expiry_time);
169	cache_fresh_locked(old, 0);
170	write_unlock(&detail->hash_lock);
171	cache_fresh_unlocked(tmp, detail, is_new);
172	cache_fresh_unlocked(old, detail, 0);
173	cache_put(old, detail);
174	return tmp;
175}
176EXPORT_SYMBOL(sunrpc_cache_update);
177
178static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
179/*
180 * This is the generic cache management routine for all
181 * the authentication caches.
182 * It checks the currency of a cache item and will (later)
183 * initiate an upcall to fill it if needed.
184 *
185 *
186 * Returns 0 if the cache_head can be used, or cache_puts it and returns
187 * -EAGAIN if upcall is pending,
188 * -ETIMEDOUT if upcall failed and should be retried,
189 * -ENOENT if cache entry was negative
190 */
191int cache_check(struct cache_detail *detail,
192		    struct cache_head *h, struct cache_req *rqstp)
193{
194	int rv;
195	long refresh_age, age;
196
197	/* First decide return status as best we can */
198	if (!test_bit(CACHE_VALID, &h->flags) ||
199	    h->expiry_time < get_seconds())
200		rv = -EAGAIN;
201	else if (detail->flush_time > h->last_refresh)
202		rv = -EAGAIN;
203	else {
204		/* entry is valid */
205		if (test_bit(CACHE_NEGATIVE, &h->flags))
206			rv = -ENOENT;
207		else rv = 0;
208	}
209
210	/* now see if we want to start an upcall */
211	refresh_age = (h->expiry_time - h->last_refresh);
212	age = get_seconds() - h->last_refresh;
213
214	if (rqstp == NULL) {
215		if (rv == -EAGAIN)
216			rv = -ENOENT;
217	} else if (rv == -EAGAIN || age > refresh_age/2) {
218		dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
219				refresh_age, age);
220		if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
221			switch (cache_make_upcall(detail, h)) {
222			case -EINVAL:
223				clear_bit(CACHE_PENDING, &h->flags);
224				if (rv == -EAGAIN) {
225					set_bit(CACHE_NEGATIVE, &h->flags);
226					cache_fresh_unlocked(h, detail,
227					     cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY));
228					rv = -ENOENT;
229				}
230				break;
231
232			case -EAGAIN:
233				clear_bit(CACHE_PENDING, &h->flags);
234				cache_revisit_request(h);
235				break;
236			}
237		}
238	}
239
240	if (rv == -EAGAIN)
241		if (cache_defer_req(rqstp, h) != 0)
242			rv = -ETIMEDOUT;
243
244	if (rv)
245		cache_put(h, detail);
246	return rv;
247}
248EXPORT_SYMBOL(cache_check);
249
250/*
251 * caches need to be periodically cleaned.
252 * For this we maintain a list of cache_detail and
253 * a current pointer into that list and into the table
254 * for that entry.
255 *
256 * Each time clean_cache is called it finds the next non-empty entry
257 * in the current table and walks the list in that entry
258 * looking for entries that can be removed.
259 *
260 * An entry gets removed if:
261 * - The expiry is before current time
262 * - The last_refresh time is before the flush_time for that cache
263 *
264 * later we might drop old entries with non-NEVER expiry if that table
265 * is getting 'full' for some definition of 'full'
266 *
267 * The question of "how often to scan a table" is an interesting one
268 * and is answered in part by the use of the "nextcheck" field in the
269 * cache_detail.
270 * When a scan of a table begins, the nextcheck field is set to a time
271 * that is well into the future.
272 * While scanning, if an expiry time is found that is earlier than the
273 * current nextcheck time, nextcheck is set to that expiry time.
274 * If the flush_time is ever set to a time earlier than the nextcheck
275 * time, the nextcheck time is then set to that flush_time.
276 *
277 * A table is then only scanned if the current time is at least
278 * the nextcheck time.
279 *
280 */
281
282static LIST_HEAD(cache_list);
283static DEFINE_SPINLOCK(cache_list_lock);
284static struct cache_detail *current_detail;
285static int current_index;
286
287static const struct file_operations cache_file_operations;
288static const struct file_operations content_file_operations;
289static const struct file_operations cache_flush_operations;
290
291static void do_cache_clean(struct work_struct *work);
292static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
293
294static void remove_cache_proc_entries(struct cache_detail *cd)
295{
296	if (cd->proc_ent == NULL)
297		return;
298	if (cd->flush_ent)
299		remove_proc_entry("flush", cd->proc_ent);
300	if (cd->channel_ent)
301		remove_proc_entry("channel", cd->proc_ent);
302	if (cd->content_ent)
303		remove_proc_entry("content", cd->proc_ent);
304	cd->proc_ent = NULL;
305	remove_proc_entry(cd->name, proc_net_rpc);
306}
307
308#ifdef CONFIG_PROC_FS
309static int create_cache_proc_entries(struct cache_detail *cd)
310{
311	struct proc_dir_entry *p;
312
313	cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
314	if (cd->proc_ent == NULL)
315		goto out_nomem;
316	cd->proc_ent->owner = cd->owner;
317	cd->channel_ent = cd->content_ent = NULL;
318
319	p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
320			     cd->proc_ent, &cache_flush_operations, cd);
321	cd->flush_ent = p;
322	if (p == NULL)
323		goto out_nomem;
324	p->owner = cd->owner;
325
326	if (cd->cache_request || cd->cache_parse) {
327		p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
328				     cd->proc_ent, &cache_file_operations, cd);
329		cd->channel_ent = p;
330		if (p == NULL)
331			goto out_nomem;
332		p->owner = cd->owner;
333	}
334	if (cd->cache_show) {
335		p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
336				cd->proc_ent, &content_file_operations, cd);
337		cd->content_ent = p;
338		if (p == NULL)
339			goto out_nomem;
340		p->owner = cd->owner;
341	}
342	return 0;
343out_nomem:
344	remove_cache_proc_entries(cd);
345	return -ENOMEM;
346}
347#else /* CONFIG_PROC_FS */
348static int create_cache_proc_entries(struct cache_detail *cd)
349{
350	return 0;
351}
352#endif
353
354int cache_register(struct cache_detail *cd)
355{
356	int ret;
357
358	ret = create_cache_proc_entries(cd);
359	if (ret)
360		return ret;
361	rwlock_init(&cd->hash_lock);
362	INIT_LIST_HEAD(&cd->queue);
363	spin_lock(&cache_list_lock);
364	cd->nextcheck = 0;
365	cd->entries = 0;
366	atomic_set(&cd->readers, 0);
367	cd->last_close = 0;
368	cd->last_warn = -1;
369	list_add(&cd->others, &cache_list);
370	spin_unlock(&cache_list_lock);
371
372	/* start the cleaning process */
373	schedule_delayed_work(&cache_cleaner, 0);
374	return 0;
375}
376EXPORT_SYMBOL(cache_register);
377
378void cache_unregister(struct cache_detail *cd)
379{
380	cache_purge(cd);
381	spin_lock(&cache_list_lock);
382	write_lock(&cd->hash_lock);
383	if (cd->entries || atomic_read(&cd->inuse)) {
384		write_unlock(&cd->hash_lock);
385		spin_unlock(&cache_list_lock);
386		goto out;
387	}
388	if (current_detail == cd)
389		current_detail = NULL;
390	list_del_init(&cd->others);
391	write_unlock(&cd->hash_lock);
392	spin_unlock(&cache_list_lock);
393	remove_cache_proc_entries(cd);
394	if (list_empty(&cache_list)) {
395		/* module must be being unloaded so its safe to kill the worker */
396		cancel_delayed_work_sync(&cache_cleaner);
397	}
398	return;
399out:
400	printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
401}
402EXPORT_SYMBOL(cache_unregister);
403
404/* clean cache tries to find something to clean
405 * and cleans it.
406 * It returns 1 if it cleaned something,
407 *            0 if it didn't find anything this time
408 *           -1 if it fell off the end of the list.
409 */
410static int cache_clean(void)
411{
412	int rv = 0;
413	struct list_head *next;
414
415	spin_lock(&cache_list_lock);
416
417	/* find a suitable table if we don't already have one */
418	while (current_detail == NULL ||
419	    current_index >= current_detail->hash_size) {
420		if (current_detail)
421			next = current_detail->others.next;
422		else
423			next = cache_list.next;
424		if (next == &cache_list) {
425			current_detail = NULL;
426			spin_unlock(&cache_list_lock);
427			return -1;
428		}
429		current_detail = list_entry(next, struct cache_detail, others);
430		if (current_detail->nextcheck > get_seconds())
431			current_index = current_detail->hash_size;
432		else {
433			current_index = 0;
434			current_detail->nextcheck = get_seconds()+30*60;
435		}
436	}
437
438	/* find a non-empty bucket in the table */
439	while (current_detail &&
440	       current_index < current_detail->hash_size &&
441	       current_detail->hash_table[current_index] == NULL)
442		current_index++;
443
444	/* find a cleanable entry in the bucket and clean it, or set to next bucket */
445
446	if (current_detail && current_index < current_detail->hash_size) {
447		struct cache_head *ch, **cp;
448		struct cache_detail *d;
449
450		write_lock(&current_detail->hash_lock);
451
452		/* Ok, now to clean this strand */
453
454		cp = & current_detail->hash_table[current_index];
455		ch = *cp;
456		for (; ch; cp= & ch->next, ch= *cp) {
457			if (current_detail->nextcheck > ch->expiry_time)
458				current_detail->nextcheck = ch->expiry_time+1;
459			if (ch->expiry_time >= get_seconds()
460			    && ch->last_refresh >= current_detail->flush_time
461				)
462				continue;
463			if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
464				queue_loose(current_detail, ch);
465
466			if (atomic_read(&ch->ref.refcount) == 1)
467				break;
468		}
469		if (ch) {
470			*cp = ch->next;
471			ch->next = NULL;
472			current_detail->entries--;
473			rv = 1;
474		}
475		write_unlock(&current_detail->hash_lock);
476		d = current_detail;
477		if (!ch)
478			current_index ++;
479		spin_unlock(&cache_list_lock);
480		if (ch)
481			cache_put(ch, d);
482	} else
483		spin_unlock(&cache_list_lock);
484
485	return rv;
486}
487
488/*
489 * We want to regularly clean the cache, so we need to schedule some work ...
490 */
491static void do_cache_clean(struct work_struct *work)
492{
493	int delay = 5;
494	if (cache_clean() == -1)
495		delay = 30*HZ;
496
497	if (list_empty(&cache_list))
498		delay = 0;
499
500	if (delay)
501		schedule_delayed_work(&cache_cleaner, delay);
502}
503
504
505/*
506 * Clean all caches promptly.  This just calls cache_clean
507 * repeatedly until we are sure that every cache has had a chance to
508 * be fully cleaned
509 */
510void cache_flush(void)
511{
512	while (cache_clean() != -1)
513		cond_resched();
514	while (cache_clean() != -1)
515		cond_resched();
516}
517EXPORT_SYMBOL(cache_flush);
518
519void cache_purge(struct cache_detail *detail)
520{
521	detail->flush_time = LONG_MAX;
522	detail->nextcheck = get_seconds();
523	cache_flush();
524	detail->flush_time = 1;
525}
526EXPORT_SYMBOL(cache_purge);
527
528
529/*
530 * Deferral and Revisiting of Requests.
531 *
532 * If a cache lookup finds a pending entry, we
533 * need to defer the request and revisit it later.
534 * All deferred requests are stored in a hash table,
535 * indexed by "struct cache_head *".
536 * As it may be wasteful to store a whole request
537 * structure, we allow the request to provide a
538 * deferred form, which must contain a
539 * 'struct cache_deferred_req'
540 * This cache_deferred_req contains a method to allow
541 * it to be revisited when cache info is available
542 */
543
544#define	DFR_HASHSIZE	(PAGE_SIZE/sizeof(struct list_head))
545#define	DFR_HASH(item)	((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
546
547#define	DFR_MAX	300	/* ??? */
548
549static DEFINE_SPINLOCK(cache_defer_lock);
550static LIST_HEAD(cache_defer_list);
551static struct list_head cache_defer_hash[DFR_HASHSIZE];
552static int cache_defer_cnt;
553
554static int cache_defer_req(struct cache_req *req, struct cache_head *item)
555{
556	struct cache_deferred_req *dreq;
557	int hash = DFR_HASH(item);
558
559	if (cache_defer_cnt >= DFR_MAX) {
560		/* too much in the cache, randomly drop this one,
561		 * or continue and drop the oldest below
562		 */
563		if (net_random()&1)
564			return -ETIMEDOUT;
565	}
566	dreq = req->defer(req);
567	if (dreq == NULL)
568		return -ETIMEDOUT;
569
570	dreq->item = item;
571
572	spin_lock(&cache_defer_lock);
573
574	list_add(&dreq->recent, &cache_defer_list);
575
576	if (cache_defer_hash[hash].next == NULL)
577		INIT_LIST_HEAD(&cache_defer_hash[hash]);
578	list_add(&dreq->hash, &cache_defer_hash[hash]);
579
580	/* it is in, now maybe clean up */
581	dreq = NULL;
582	if (++cache_defer_cnt > DFR_MAX) {
583		dreq = list_entry(cache_defer_list.prev,
584				  struct cache_deferred_req, recent);
585		list_del(&dreq->recent);
586		list_del(&dreq->hash);
587		cache_defer_cnt--;
588	}
589	spin_unlock(&cache_defer_lock);
590
591	if (dreq) {
592		/* there was one too many */
593		dreq->revisit(dreq, 1);
594	}
595	if (!test_bit(CACHE_PENDING, &item->flags)) {
596		/* must have just been validated... */
597		cache_revisit_request(item);
598	}
599	return 0;
600}
601
602static void cache_revisit_request(struct cache_head *item)
603{
604	struct cache_deferred_req *dreq;
605	struct list_head pending;
606
607	struct list_head *lp;
608	int hash = DFR_HASH(item);
609
610	INIT_LIST_HEAD(&pending);
611	spin_lock(&cache_defer_lock);
612
613	lp = cache_defer_hash[hash].next;
614	if (lp) {
615		while (lp != &cache_defer_hash[hash]) {
616			dreq = list_entry(lp, struct cache_deferred_req, hash);
617			lp = lp->next;
618			if (dreq->item == item) {
619				list_del(&dreq->hash);
620				list_move(&dreq->recent, &pending);
621				cache_defer_cnt--;
622			}
623		}
624	}
625	spin_unlock(&cache_defer_lock);
626
627	while (!list_empty(&pending)) {
628		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
629		list_del_init(&dreq->recent);
630		dreq->revisit(dreq, 0);
631	}
632}
633
634void cache_clean_deferred(void *owner)
635{
636	struct cache_deferred_req *dreq, *tmp;
637	struct list_head pending;
638
639
640	INIT_LIST_HEAD(&pending);
641	spin_lock(&cache_defer_lock);
642
643	list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
644		if (dreq->owner == owner) {
645			list_del(&dreq->hash);
646			list_move(&dreq->recent, &pending);
647			cache_defer_cnt--;
648		}
649	}
650	spin_unlock(&cache_defer_lock);
651
652	while (!list_empty(&pending)) {
653		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
654		list_del_init(&dreq->recent);
655		dreq->revisit(dreq, 1);
656	}
657}
658
659/*
660 * communicate with user-space
661 *
662 * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
663 * On read, you get a full request, or block.
664 * On write, an update request is processed.
665 * Poll works if anything to read, and always allows write.
666 *
667 * Implemented by linked list of requests.  Each open file has
668 * a ->private that also exists in this list.  New requests are added
669 * to the end and may wakeup and preceding readers.
670 * New readers are added to the head.  If, on read, an item is found with
671 * CACHE_UPCALLING clear, we free it from the list.
672 *
673 */
674
675static DEFINE_SPINLOCK(queue_lock);
676static DEFINE_MUTEX(queue_io_mutex);
677
678struct cache_queue {
679	struct list_head	list;
680	int			reader;	/* if 0, then request */
681};
682struct cache_request {
683	struct cache_queue	q;
684	struct cache_head	*item;
685	char			* buf;
686	int			len;
687	int			readers;
688};
689struct cache_reader {
690	struct cache_queue	q;
691	int			offset;	/* if non-0, we have a refcnt on next request */
692};
693
694static ssize_t
695cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
696{
697	struct cache_reader *rp = filp->private_data;
698	struct cache_request *rq;
699	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
700	int err;
701
702	if (count == 0)
703		return 0;
704
705	mutex_lock(&queue_io_mutex); /* protect against multiple concurrent
706			      * readers on this file */
707 again:
708	spin_lock(&queue_lock);
709	/* need to find next request */
710	while (rp->q.list.next != &cd->queue &&
711	       list_entry(rp->q.list.next, struct cache_queue, list)
712	       ->reader) {
713		struct list_head *next = rp->q.list.next;
714		list_move(&rp->q.list, next);
715	}
716	if (rp->q.list.next == &cd->queue) {
717		spin_unlock(&queue_lock);
718		mutex_unlock(&queue_io_mutex);
719		BUG_ON(rp->offset);
720		return 0;
721	}
722	rq = container_of(rp->q.list.next, struct cache_request, q.list);
723	BUG_ON(rq->q.reader);
724	if (rp->offset == 0)
725		rq->readers++;
726	spin_unlock(&queue_lock);
727
728	if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
729		err = -EAGAIN;
730		spin_lock(&queue_lock);
731		list_move(&rp->q.list, &rq->q.list);
732		spin_unlock(&queue_lock);
733	} else {
734		if (rp->offset + count > rq->len)
735			count = rq->len - rp->offset;
736		err = -EFAULT;
737		if (copy_to_user(buf, rq->buf + rp->offset, count))
738			goto out;
739		rp->offset += count;
740		if (rp->offset >= rq->len) {
741			rp->offset = 0;
742			spin_lock(&queue_lock);
743			list_move(&rp->q.list, &rq->q.list);
744			spin_unlock(&queue_lock);
745		}
746		err = 0;
747	}
748 out:
749	if (rp->offset == 0) {
750		/* need to release rq */
751		spin_lock(&queue_lock);
752		rq->readers--;
753		if (rq->readers == 0 &&
754		    !test_bit(CACHE_PENDING, &rq->item->flags)) {
755			list_del(&rq->q.list);
756			spin_unlock(&queue_lock);
757			cache_put(rq->item, cd);
758			kfree(rq->buf);
759			kfree(rq);
760		} else
761			spin_unlock(&queue_lock);
762	}
763	if (err == -EAGAIN)
764		goto again;
765	mutex_unlock(&queue_io_mutex);
766	return err ? err :  count;
767}
768
769static char write_buf[8192]; /* protected by queue_io_mutex */
770
771static ssize_t
772cache_write(struct file *filp, const char __user *buf, size_t count,
773	    loff_t *ppos)
774{
775	int err;
776	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
777
778	if (count == 0)
779		return 0;
780	if (count >= sizeof(write_buf))
781		return -EINVAL;
782
783	mutex_lock(&queue_io_mutex);
784
785	if (copy_from_user(write_buf, buf, count)) {
786		mutex_unlock(&queue_io_mutex);
787		return -EFAULT;
788	}
789	write_buf[count] = '\0';
790	if (cd->cache_parse)
791		err = cd->cache_parse(cd, write_buf, count);
792	else
793		err = -EINVAL;
794
795	mutex_unlock(&queue_io_mutex);
796	return err ? err : count;
797}
798
799static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
800
801static unsigned int
802cache_poll(struct file *filp, poll_table *wait)
803{
804	unsigned int mask;
805	struct cache_reader *rp = filp->private_data;
806	struct cache_queue *cq;
807	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
808
809	poll_wait(filp, &queue_wait, wait);
810
811	/* alway allow write */
812	mask = POLL_OUT | POLLWRNORM;
813
814	if (!rp)
815		return mask;
816
817	spin_lock(&queue_lock);
818
819	for (cq= &rp->q; &cq->list != &cd->queue;
820	     cq = list_entry(cq->list.next, struct cache_queue, list))
821		if (!cq->reader) {
822			mask |= POLLIN | POLLRDNORM;
823			break;
824		}
825	spin_unlock(&queue_lock);
826	return mask;
827}
828
829static int
830cache_ioctl(struct inode *ino, struct file *filp,
831	    unsigned int cmd, unsigned long arg)
832{
833	int len = 0;
834	struct cache_reader *rp = filp->private_data;
835	struct cache_queue *cq;
836	struct cache_detail *cd = PDE(ino)->data;
837
838	if (cmd != FIONREAD || !rp)
839		return -EINVAL;
840
841	spin_lock(&queue_lock);
842
843	/* only find the length remaining in current request,
844	 * or the length of the next request
845	 */
846	for (cq= &rp->q; &cq->list != &cd->queue;
847	     cq = list_entry(cq->list.next, struct cache_queue, list))
848		if (!cq->reader) {
849			struct cache_request *cr =
850				container_of(cq, struct cache_request, q);
851			len = cr->len - rp->offset;
852			break;
853		}
854	spin_unlock(&queue_lock);
855
856	return put_user(len, (int __user *)arg);
857}
858
859static int
860cache_open(struct inode *inode, struct file *filp)
861{
862	struct cache_reader *rp = NULL;
863
864	nonseekable_open(inode, filp);
865	if (filp->f_mode & FMODE_READ) {
866		struct cache_detail *cd = PDE(inode)->data;
867
868		rp = kmalloc(sizeof(*rp), GFP_KERNEL);
869		if (!rp)
870			return -ENOMEM;
871		rp->offset = 0;
872		rp->q.reader = 1;
873		atomic_inc(&cd->readers);
874		spin_lock(&queue_lock);
875		list_add(&rp->q.list, &cd->queue);
876		spin_unlock(&queue_lock);
877	}
878	filp->private_data = rp;
879	return 0;
880}
881
882static int
883cache_release(struct inode *inode, struct file *filp)
884{
885	struct cache_reader *rp = filp->private_data;
886	struct cache_detail *cd = PDE(inode)->data;
887
888	if (rp) {
889		spin_lock(&queue_lock);
890		if (rp->offset) {
891			struct cache_queue *cq;
892			for (cq= &rp->q; &cq->list != &cd->queue;
893			     cq = list_entry(cq->list.next, struct cache_queue, list))
894				if (!cq->reader) {
895					container_of(cq, struct cache_request, q)
896						->readers--;
897					break;
898				}
899			rp->offset = 0;
900		}
901		list_del(&rp->q.list);
902		spin_unlock(&queue_lock);
903
904		filp->private_data = NULL;
905		kfree(rp);
906
907		cd->last_close = get_seconds();
908		atomic_dec(&cd->readers);
909	}
910	return 0;
911}
912
913
914
915static const struct file_operations cache_file_operations = {
916	.owner		= THIS_MODULE,
917	.llseek		= no_llseek,
918	.read		= cache_read,
919	.write		= cache_write,
920	.poll		= cache_poll,
921	.ioctl		= cache_ioctl, /* for FIONREAD */
922	.open		= cache_open,
923	.release	= cache_release,
924};
925
926
927static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
928{
929	struct cache_queue *cq;
930	spin_lock(&queue_lock);
931	list_for_each_entry(cq, &detail->queue, list)
932		if (!cq->reader) {
933			struct cache_request *cr = container_of(cq, struct cache_request, q);
934			if (cr->item != ch)
935				continue;
936			if (cr->readers != 0)
937				continue;
938			list_del(&cr->q.list);
939			spin_unlock(&queue_lock);
940			cache_put(cr->item, detail);
941			kfree(cr->buf);
942			kfree(cr);
943			return;
944		}
945	spin_unlock(&queue_lock);
946}
947
948/*
949 * Support routines for text-based upcalls.
950 * Fields are separated by spaces.
951 * Fields are either mangled to quote space tab newline slosh with slosh
952 * or a hexified with a leading \x
953 * Record is terminated with newline.
954 *
955 */
956
957void qword_add(char **bpp, int *lp, char *str)
958{
959	char *bp = *bpp;
960	int len = *lp;
961	char c;
962
963	if (len < 0) return;
964
965	while ((c=*str++) && len)
966		switch(c) {
967		case ' ':
968		case '\t':
969		case '\n':
970		case '\\':
971			if (len >= 4) {
972				*bp++ = '\\';
973				*bp++ = '0' + ((c & 0300)>>6);
974				*bp++ = '0' + ((c & 0070)>>3);
975				*bp++ = '0' + ((c & 0007)>>0);
976			}
977			len -= 4;
978			break;
979		default:
980			*bp++ = c;
981			len--;
982		}
983	if (c || len <1) len = -1;
984	else {
985		*bp++ = ' ';
986		len--;
987	}
988	*bpp = bp;
989	*lp = len;
990}
991EXPORT_SYMBOL(qword_add);
992
993void qword_addhex(char **bpp, int *lp, char *buf, int blen)
994{
995	char *bp = *bpp;
996	int len = *lp;
997
998	if (len < 0) return;
999
1000	if (len > 2) {
1001		*bp++ = '\\';
1002		*bp++ = 'x';
1003		len -= 2;
1004		while (blen && len >= 2) {
1005			unsigned char c = *buf++;
1006			*bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1007			*bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1008			len -= 2;
1009			blen--;
1010		}
1011	}
1012	if (blen || len<1) len = -1;
1013	else {
1014		*bp++ = ' ';
1015		len--;
1016	}
1017	*bpp = bp;
1018	*lp = len;
1019}
1020EXPORT_SYMBOL(qword_addhex);
1021
1022static void warn_no_listener(struct cache_detail *detail)
1023{
1024	if (detail->last_warn != detail->last_close) {
1025		detail->last_warn = detail->last_close;
1026		if (detail->warn_no_listener)
1027			detail->warn_no_listener(detail);
1028	}
1029}
1030
1031/*
1032 * register an upcall request to user-space.
1033 * Each request is at most one page long.
1034 */
1035static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
1036{
1037
1038	char *buf;
1039	struct cache_request *crq;
1040	char *bp;
1041	int len;
1042
1043	if (detail->cache_request == NULL)
1044		return -EINVAL;
1045
1046	if (atomic_read(&detail->readers) == 0 &&
1047	    detail->last_close < get_seconds() - 30) {
1048			warn_no_listener(detail);
1049			return -EINVAL;
1050	}
1051
1052	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1053	if (!buf)
1054		return -EAGAIN;
1055
1056	crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1057	if (!crq) {
1058		kfree(buf);
1059		return -EAGAIN;
1060	}
1061
1062	bp = buf; len = PAGE_SIZE;
1063
1064	detail->cache_request(detail, h, &bp, &len);
1065
1066	if (len < 0) {
1067		kfree(buf);
1068		kfree(crq);
1069		return -EAGAIN;
1070	}
1071	crq->q.reader = 0;
1072	crq->item = cache_get(h);
1073	crq->buf = buf;
1074	crq->len = PAGE_SIZE - len;
1075	crq->readers = 0;
1076	spin_lock(&queue_lock);
1077	list_add_tail(&crq->q.list, &detail->queue);
1078	spin_unlock(&queue_lock);
1079	wake_up(&queue_wait);
1080	return 0;
1081}
1082
1083/*
1084 * parse a message from user-space and pass it
1085 * to an appropriate cache
1086 * Messages are, like requests, separated into fields by
1087 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1088 *
1089 * Message is
1090 *   reply cachename expiry key ... content....
1091 *
1092 * key and content are both parsed by cache
1093 */
1094
1095#define isodigit(c) (isdigit(c) && c <= '7')
1096int qword_get(char **bpp, char *dest, int bufsize)
1097{
1098	/* return bytes copied, or -1 on error */
1099	char *bp = *bpp;
1100	int len = 0;
1101
1102	while (*bp == ' ') bp++;
1103
1104	if (bp[0] == '\\' && bp[1] == 'x') {
1105		/* HEX STRING */
1106		bp += 2;
1107		while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1108			int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1109			bp++;
1110			byte <<= 4;
1111			byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1112			*dest++ = byte;
1113			bp++;
1114			len++;
1115		}
1116	} else {
1117		/* text with \nnn octal quoting */
1118		while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1119			if (*bp == '\\' &&
1120			    isodigit(bp[1]) && (bp[1] <= '3') &&
1121			    isodigit(bp[2]) &&
1122			    isodigit(bp[3])) {
1123				int byte = (*++bp -'0');
1124				bp++;
1125				byte = (byte << 3) | (*bp++ - '0');
1126				byte = (byte << 3) | (*bp++ - '0');
1127				*dest++ = byte;
1128				len++;
1129			} else {
1130				*dest++ = *bp++;
1131				len++;
1132			}
1133		}
1134	}
1135
1136	if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1137		return -1;
1138	while (*bp == ' ') bp++;
1139	*bpp = bp;
1140	*dest = '\0';
1141	return len;
1142}
1143EXPORT_SYMBOL(qword_get);
1144
1145
1146/*
1147 * support /proc/sunrpc/cache/$CACHENAME/content
1148 * as a seqfile.
1149 * We call ->cache_show passing NULL for the item to
1150 * get a header, then pass each real item in the cache
1151 */
1152
1153struct handle {
1154	struct cache_detail *cd;
1155};
1156
1157static void *c_start(struct seq_file *m, loff_t *pos)
1158	__acquires(cd->hash_lock)
1159{
1160	loff_t n = *pos;
1161	unsigned hash, entry;
1162	struct cache_head *ch;
1163	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1164
1165
1166	read_lock(&cd->hash_lock);
1167	if (!n--)
1168		return SEQ_START_TOKEN;
1169	hash = n >> 32;
1170	entry = n & ((1LL<<32) - 1);
1171
1172	for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1173		if (!entry--)
1174			return ch;
1175	n &= ~((1LL<<32) - 1);
1176	do {
1177		hash++;
1178		n += 1LL<<32;
1179	} while(hash < cd->hash_size &&
1180		cd->hash_table[hash]==NULL);
1181	if (hash >= cd->hash_size)
1182		return NULL;
1183	*pos = n+1;
1184	return cd->hash_table[hash];
1185}
1186
1187static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1188{
1189	struct cache_head *ch = p;
1190	int hash = (*pos >> 32);
1191	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1192
1193	if (p == SEQ_START_TOKEN)
1194		hash = 0;
1195	else if (ch->next == NULL) {
1196		hash++;
1197		*pos += 1LL<<32;
1198	} else {
1199		++*pos;
1200		return ch->next;
1201	}
1202	*pos &= ~((1LL<<32) - 1);
1203	while (hash < cd->hash_size &&
1204	       cd->hash_table[hash] == NULL) {
1205		hash++;
1206		*pos += 1LL<<32;
1207	}
1208	if (hash >= cd->hash_size)
1209		return NULL;
1210	++*pos;
1211	return cd->hash_table[hash];
1212}
1213
1214static void c_stop(struct seq_file *m, void *p)
1215	__releases(cd->hash_lock)
1216{
1217	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1218	read_unlock(&cd->hash_lock);
1219}
1220
1221static int c_show(struct seq_file *m, void *p)
1222{
1223	struct cache_head *cp = p;
1224	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1225
1226	if (p == SEQ_START_TOKEN)
1227		return cd->cache_show(m, cd, NULL);
1228
1229	ifdebug(CACHE)
1230		seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1231			   cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1232	cache_get(cp);
1233	if (cache_check(cd, cp, NULL))
1234		/* cache_check does a cache_put on failure */
1235		seq_printf(m, "# ");
1236	else
1237		cache_put(cp, cd);
1238
1239	return cd->cache_show(m, cd, cp);
1240}
1241
1242static const struct seq_operations cache_content_op = {
1243	.start	= c_start,
1244	.next	= c_next,
1245	.stop	= c_stop,
1246	.show	= c_show,
1247};
1248
1249static int content_open(struct inode *inode, struct file *file)
1250{
1251	struct handle *han;
1252	struct cache_detail *cd = PDE(inode)->data;
1253
1254	han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1255	if (han == NULL)
1256		return -ENOMEM;
1257
1258	han->cd = cd;
1259	return 0;
1260}
1261
1262static const struct file_operations content_file_operations = {
1263	.open		= content_open,
1264	.read		= seq_read,
1265	.llseek		= seq_lseek,
1266	.release	= seq_release_private,
1267};
1268
1269static ssize_t read_flush(struct file *file, char __user *buf,
1270			    size_t count, loff_t *ppos)
1271{
1272	struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1273	char tbuf[20];
1274	unsigned long p = *ppos;
1275	size_t len;
1276
1277	sprintf(tbuf, "%lu\n", cd->flush_time);
1278	len = strlen(tbuf);
1279	if (p >= len)
1280		return 0;
1281	len -= p;
1282	if (len > count)
1283		len = count;
1284	if (copy_to_user(buf, (void*)(tbuf+p), len))
1285		return -EFAULT;
1286	*ppos += len;
1287	return len;
1288}
1289
1290static ssize_t write_flush(struct file * file, const char __user * buf,
1291			     size_t count, loff_t *ppos)
1292{
1293	struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1294	char tbuf[20];
1295	char *ep;
1296	long flushtime;
1297	if (*ppos || count > sizeof(tbuf)-1)
1298		return -EINVAL;
1299	if (copy_from_user(tbuf, buf, count))
1300		return -EFAULT;
1301	tbuf[count] = 0;
1302	flushtime = simple_strtoul(tbuf, &ep, 0);
1303	if (*ep && *ep != '\n')
1304		return -EINVAL;
1305
1306	cd->flush_time = flushtime;
1307	cd->nextcheck = get_seconds();
1308	cache_flush();
1309
1310	*ppos += count;
1311	return count;
1312}
1313
1314static const struct file_operations cache_flush_operations = {
1315	.open		= nonseekable_open,
1316	.read		= read_flush,
1317	.write		= write_flush,
1318};
1319