xref: /openbmc/linux/net/sunrpc/cache.c (revision 643d1f7f)
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 
37 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
38 static void cache_revisit_request(struct cache_head *item);
39 
40 static 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 
50 struct 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 }
101 EXPORT_SYMBOL(sunrpc_cache_lookup);
102 
103 
104 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
105 
106 static 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 
113 static 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 
124 struct 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 }
176 EXPORT_SYMBOL(sunrpc_cache_update);
177 
178 static 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  */
191 int 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 }
248 EXPORT_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 
282 static LIST_HEAD(cache_list);
283 static DEFINE_SPINLOCK(cache_list_lock);
284 static struct cache_detail *current_detail;
285 static int current_index;
286 
287 static const struct file_operations cache_file_operations;
288 static const struct file_operations content_file_operations;
289 static const struct file_operations cache_flush_operations;
290 
291 static void do_cache_clean(struct work_struct *work);
292 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
293 
294 static 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
309 static 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 = create_proc_entry("flush", S_IFREG|S_IRUSR|S_IWUSR, cd->proc_ent);
320 	cd->flush_ent = p;
321 	if (p == NULL)
322 		goto out_nomem;
323 	p->proc_fops = &cache_flush_operations;
324 	p->owner = cd->owner;
325 	p->data = cd;
326 
327 	if (cd->cache_request || cd->cache_parse) {
328 		p = create_proc_entry("channel", S_IFREG|S_IRUSR|S_IWUSR,
329 				      cd->proc_ent);
330 		cd->channel_ent = p;
331 		if (p == NULL)
332 			goto out_nomem;
333 		p->proc_fops = &cache_file_operations;
334 		p->owner = cd->owner;
335 		p->data = cd;
336 	}
337 	if (cd->cache_show) {
338 		p = create_proc_entry("content", S_IFREG|S_IRUSR|S_IWUSR,
339 				      cd->proc_ent);
340 		cd->content_ent = p;
341 		if (p == NULL)
342 			goto out_nomem;
343 		p->proc_fops = &content_file_operations;
344 		p->owner = cd->owner;
345 		p->data = cd;
346 	}
347 	return 0;
348 out_nomem:
349 	remove_cache_proc_entries(cd);
350 	return -ENOMEM;
351 }
352 #else /* CONFIG_PROC_FS */
353 static int create_cache_proc_entries(struct cache_detail *cd)
354 {
355 	return 0;
356 }
357 #endif
358 
359 int cache_register(struct cache_detail *cd)
360 {
361 	int ret;
362 
363 	ret = create_cache_proc_entries(cd);
364 	if (ret)
365 		return ret;
366 	rwlock_init(&cd->hash_lock);
367 	INIT_LIST_HEAD(&cd->queue);
368 	spin_lock(&cache_list_lock);
369 	cd->nextcheck = 0;
370 	cd->entries = 0;
371 	atomic_set(&cd->readers, 0);
372 	cd->last_close = 0;
373 	cd->last_warn = -1;
374 	list_add(&cd->others, &cache_list);
375 	spin_unlock(&cache_list_lock);
376 
377 	/* start the cleaning process */
378 	schedule_delayed_work(&cache_cleaner, 0);
379 	return 0;
380 }
381 EXPORT_SYMBOL(cache_register);
382 
383 void cache_unregister(struct cache_detail *cd)
384 {
385 	cache_purge(cd);
386 	spin_lock(&cache_list_lock);
387 	write_lock(&cd->hash_lock);
388 	if (cd->entries || atomic_read(&cd->inuse)) {
389 		write_unlock(&cd->hash_lock);
390 		spin_unlock(&cache_list_lock);
391 		goto out;
392 	}
393 	if (current_detail == cd)
394 		current_detail = NULL;
395 	list_del_init(&cd->others);
396 	write_unlock(&cd->hash_lock);
397 	spin_unlock(&cache_list_lock);
398 	remove_cache_proc_entries(cd);
399 	if (list_empty(&cache_list)) {
400 		/* module must be being unloaded so its safe to kill the worker */
401 		cancel_delayed_work_sync(&cache_cleaner);
402 	}
403 	return;
404 out:
405 	printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
406 }
407 EXPORT_SYMBOL(cache_unregister);
408 
409 /* clean cache tries to find something to clean
410  * and cleans it.
411  * It returns 1 if it cleaned something,
412  *            0 if it didn't find anything this time
413  *           -1 if it fell off the end of the list.
414  */
415 static int cache_clean(void)
416 {
417 	int rv = 0;
418 	struct list_head *next;
419 
420 	spin_lock(&cache_list_lock);
421 
422 	/* find a suitable table if we don't already have one */
423 	while (current_detail == NULL ||
424 	    current_index >= current_detail->hash_size) {
425 		if (current_detail)
426 			next = current_detail->others.next;
427 		else
428 			next = cache_list.next;
429 		if (next == &cache_list) {
430 			current_detail = NULL;
431 			spin_unlock(&cache_list_lock);
432 			return -1;
433 		}
434 		current_detail = list_entry(next, struct cache_detail, others);
435 		if (current_detail->nextcheck > get_seconds())
436 			current_index = current_detail->hash_size;
437 		else {
438 			current_index = 0;
439 			current_detail->nextcheck = get_seconds()+30*60;
440 		}
441 	}
442 
443 	/* find a non-empty bucket in the table */
444 	while (current_detail &&
445 	       current_index < current_detail->hash_size &&
446 	       current_detail->hash_table[current_index] == NULL)
447 		current_index++;
448 
449 	/* find a cleanable entry in the bucket and clean it, or set to next bucket */
450 
451 	if (current_detail && current_index < current_detail->hash_size) {
452 		struct cache_head *ch, **cp;
453 		struct cache_detail *d;
454 
455 		write_lock(&current_detail->hash_lock);
456 
457 		/* Ok, now to clean this strand */
458 
459 		cp = & current_detail->hash_table[current_index];
460 		ch = *cp;
461 		for (; ch; cp= & ch->next, ch= *cp) {
462 			if (current_detail->nextcheck > ch->expiry_time)
463 				current_detail->nextcheck = ch->expiry_time+1;
464 			if (ch->expiry_time >= get_seconds()
465 			    && ch->last_refresh >= current_detail->flush_time
466 				)
467 				continue;
468 			if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
469 				queue_loose(current_detail, ch);
470 
471 			if (atomic_read(&ch->ref.refcount) == 1)
472 				break;
473 		}
474 		if (ch) {
475 			*cp = ch->next;
476 			ch->next = NULL;
477 			current_detail->entries--;
478 			rv = 1;
479 		}
480 		write_unlock(&current_detail->hash_lock);
481 		d = current_detail;
482 		if (!ch)
483 			current_index ++;
484 		spin_unlock(&cache_list_lock);
485 		if (ch)
486 			cache_put(ch, d);
487 	} else
488 		spin_unlock(&cache_list_lock);
489 
490 	return rv;
491 }
492 
493 /*
494  * We want to regularly clean the cache, so we need to schedule some work ...
495  */
496 static void do_cache_clean(struct work_struct *work)
497 {
498 	int delay = 5;
499 	if (cache_clean() == -1)
500 		delay = 30*HZ;
501 
502 	if (list_empty(&cache_list))
503 		delay = 0;
504 
505 	if (delay)
506 		schedule_delayed_work(&cache_cleaner, delay);
507 }
508 
509 
510 /*
511  * Clean all caches promptly.  This just calls cache_clean
512  * repeatedly until we are sure that every cache has had a chance to
513  * be fully cleaned
514  */
515 void cache_flush(void)
516 {
517 	while (cache_clean() != -1)
518 		cond_resched();
519 	while (cache_clean() != -1)
520 		cond_resched();
521 }
522 EXPORT_SYMBOL(cache_flush);
523 
524 void cache_purge(struct cache_detail *detail)
525 {
526 	detail->flush_time = LONG_MAX;
527 	detail->nextcheck = get_seconds();
528 	cache_flush();
529 	detail->flush_time = 1;
530 }
531 EXPORT_SYMBOL(cache_purge);
532 
533 
534 /*
535  * Deferral and Revisiting of Requests.
536  *
537  * If a cache lookup finds a pending entry, we
538  * need to defer the request and revisit it later.
539  * All deferred requests are stored in a hash table,
540  * indexed by "struct cache_head *".
541  * As it may be wasteful to store a whole request
542  * structure, we allow the request to provide a
543  * deferred form, which must contain a
544  * 'struct cache_deferred_req'
545  * This cache_deferred_req contains a method to allow
546  * it to be revisited when cache info is available
547  */
548 
549 #define	DFR_HASHSIZE	(PAGE_SIZE/sizeof(struct list_head))
550 #define	DFR_HASH(item)	((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
551 
552 #define	DFR_MAX	300	/* ??? */
553 
554 static DEFINE_SPINLOCK(cache_defer_lock);
555 static LIST_HEAD(cache_defer_list);
556 static struct list_head cache_defer_hash[DFR_HASHSIZE];
557 static int cache_defer_cnt;
558 
559 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
560 {
561 	struct cache_deferred_req *dreq;
562 	int hash = DFR_HASH(item);
563 
564 	if (cache_defer_cnt >= DFR_MAX) {
565 		/* too much in the cache, randomly drop this one,
566 		 * or continue and drop the oldest below
567 		 */
568 		if (net_random()&1)
569 			return -ETIMEDOUT;
570 	}
571 	dreq = req->defer(req);
572 	if (dreq == NULL)
573 		return -ETIMEDOUT;
574 
575 	dreq->item = item;
576 	dreq->recv_time = get_seconds();
577 
578 	spin_lock(&cache_defer_lock);
579 
580 	list_add(&dreq->recent, &cache_defer_list);
581 
582 	if (cache_defer_hash[hash].next == NULL)
583 		INIT_LIST_HEAD(&cache_defer_hash[hash]);
584 	list_add(&dreq->hash, &cache_defer_hash[hash]);
585 
586 	/* it is in, now maybe clean up */
587 	dreq = NULL;
588 	if (++cache_defer_cnt > DFR_MAX) {
589 		dreq = list_entry(cache_defer_list.prev,
590 				  struct cache_deferred_req, recent);
591 		list_del(&dreq->recent);
592 		list_del(&dreq->hash);
593 		cache_defer_cnt--;
594 	}
595 	spin_unlock(&cache_defer_lock);
596 
597 	if (dreq) {
598 		/* there was one too many */
599 		dreq->revisit(dreq, 1);
600 	}
601 	if (!test_bit(CACHE_PENDING, &item->flags)) {
602 		/* must have just been validated... */
603 		cache_revisit_request(item);
604 	}
605 	return 0;
606 }
607 
608 static void cache_revisit_request(struct cache_head *item)
609 {
610 	struct cache_deferred_req *dreq;
611 	struct list_head pending;
612 
613 	struct list_head *lp;
614 	int hash = DFR_HASH(item);
615 
616 	INIT_LIST_HEAD(&pending);
617 	spin_lock(&cache_defer_lock);
618 
619 	lp = cache_defer_hash[hash].next;
620 	if (lp) {
621 		while (lp != &cache_defer_hash[hash]) {
622 			dreq = list_entry(lp, struct cache_deferred_req, hash);
623 			lp = lp->next;
624 			if (dreq->item == item) {
625 				list_del(&dreq->hash);
626 				list_move(&dreq->recent, &pending);
627 				cache_defer_cnt--;
628 			}
629 		}
630 	}
631 	spin_unlock(&cache_defer_lock);
632 
633 	while (!list_empty(&pending)) {
634 		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
635 		list_del_init(&dreq->recent);
636 		dreq->revisit(dreq, 0);
637 	}
638 }
639 
640 void cache_clean_deferred(void *owner)
641 {
642 	struct cache_deferred_req *dreq, *tmp;
643 	struct list_head pending;
644 
645 
646 	INIT_LIST_HEAD(&pending);
647 	spin_lock(&cache_defer_lock);
648 
649 	list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
650 		if (dreq->owner == owner) {
651 			list_del(&dreq->hash);
652 			list_move(&dreq->recent, &pending);
653 			cache_defer_cnt--;
654 		}
655 	}
656 	spin_unlock(&cache_defer_lock);
657 
658 	while (!list_empty(&pending)) {
659 		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
660 		list_del_init(&dreq->recent);
661 		dreq->revisit(dreq, 1);
662 	}
663 }
664 
665 /*
666  * communicate with user-space
667  *
668  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
669  * On read, you get a full request, or block.
670  * On write, an update request is processed.
671  * Poll works if anything to read, and always allows write.
672  *
673  * Implemented by linked list of requests.  Each open file has
674  * a ->private that also exists in this list.  New requests are added
675  * to the end and may wakeup and preceding readers.
676  * New readers are added to the head.  If, on read, an item is found with
677  * CACHE_UPCALLING clear, we free it from the list.
678  *
679  */
680 
681 static DEFINE_SPINLOCK(queue_lock);
682 static DEFINE_MUTEX(queue_io_mutex);
683 
684 struct cache_queue {
685 	struct list_head	list;
686 	int			reader;	/* if 0, then request */
687 };
688 struct cache_request {
689 	struct cache_queue	q;
690 	struct cache_head	*item;
691 	char			* buf;
692 	int			len;
693 	int			readers;
694 };
695 struct cache_reader {
696 	struct cache_queue	q;
697 	int			offset;	/* if non-0, we have a refcnt on next request */
698 };
699 
700 static ssize_t
701 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
702 {
703 	struct cache_reader *rp = filp->private_data;
704 	struct cache_request *rq;
705 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
706 	int err;
707 
708 	if (count == 0)
709 		return 0;
710 
711 	mutex_lock(&queue_io_mutex); /* protect against multiple concurrent
712 			      * readers on this file */
713  again:
714 	spin_lock(&queue_lock);
715 	/* need to find next request */
716 	while (rp->q.list.next != &cd->queue &&
717 	       list_entry(rp->q.list.next, struct cache_queue, list)
718 	       ->reader) {
719 		struct list_head *next = rp->q.list.next;
720 		list_move(&rp->q.list, next);
721 	}
722 	if (rp->q.list.next == &cd->queue) {
723 		spin_unlock(&queue_lock);
724 		mutex_unlock(&queue_io_mutex);
725 		BUG_ON(rp->offset);
726 		return 0;
727 	}
728 	rq = container_of(rp->q.list.next, struct cache_request, q.list);
729 	BUG_ON(rq->q.reader);
730 	if (rp->offset == 0)
731 		rq->readers++;
732 	spin_unlock(&queue_lock);
733 
734 	if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
735 		err = -EAGAIN;
736 		spin_lock(&queue_lock);
737 		list_move(&rp->q.list, &rq->q.list);
738 		spin_unlock(&queue_lock);
739 	} else {
740 		if (rp->offset + count > rq->len)
741 			count = rq->len - rp->offset;
742 		err = -EFAULT;
743 		if (copy_to_user(buf, rq->buf + rp->offset, count))
744 			goto out;
745 		rp->offset += count;
746 		if (rp->offset >= rq->len) {
747 			rp->offset = 0;
748 			spin_lock(&queue_lock);
749 			list_move(&rp->q.list, &rq->q.list);
750 			spin_unlock(&queue_lock);
751 		}
752 		err = 0;
753 	}
754  out:
755 	if (rp->offset == 0) {
756 		/* need to release rq */
757 		spin_lock(&queue_lock);
758 		rq->readers--;
759 		if (rq->readers == 0 &&
760 		    !test_bit(CACHE_PENDING, &rq->item->flags)) {
761 			list_del(&rq->q.list);
762 			spin_unlock(&queue_lock);
763 			cache_put(rq->item, cd);
764 			kfree(rq->buf);
765 			kfree(rq);
766 		} else
767 			spin_unlock(&queue_lock);
768 	}
769 	if (err == -EAGAIN)
770 		goto again;
771 	mutex_unlock(&queue_io_mutex);
772 	return err ? err :  count;
773 }
774 
775 static char write_buf[8192]; /* protected by queue_io_mutex */
776 
777 static ssize_t
778 cache_write(struct file *filp, const char __user *buf, size_t count,
779 	    loff_t *ppos)
780 {
781 	int err;
782 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
783 
784 	if (count == 0)
785 		return 0;
786 	if (count >= sizeof(write_buf))
787 		return -EINVAL;
788 
789 	mutex_lock(&queue_io_mutex);
790 
791 	if (copy_from_user(write_buf, buf, count)) {
792 		mutex_unlock(&queue_io_mutex);
793 		return -EFAULT;
794 	}
795 	write_buf[count] = '\0';
796 	if (cd->cache_parse)
797 		err = cd->cache_parse(cd, write_buf, count);
798 	else
799 		err = -EINVAL;
800 
801 	mutex_unlock(&queue_io_mutex);
802 	return err ? err : count;
803 }
804 
805 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
806 
807 static unsigned int
808 cache_poll(struct file *filp, poll_table *wait)
809 {
810 	unsigned int mask;
811 	struct cache_reader *rp = filp->private_data;
812 	struct cache_queue *cq;
813 	struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
814 
815 	poll_wait(filp, &queue_wait, wait);
816 
817 	/* alway allow write */
818 	mask = POLL_OUT | POLLWRNORM;
819 
820 	if (!rp)
821 		return mask;
822 
823 	spin_lock(&queue_lock);
824 
825 	for (cq= &rp->q; &cq->list != &cd->queue;
826 	     cq = list_entry(cq->list.next, struct cache_queue, list))
827 		if (!cq->reader) {
828 			mask |= POLLIN | POLLRDNORM;
829 			break;
830 		}
831 	spin_unlock(&queue_lock);
832 	return mask;
833 }
834 
835 static int
836 cache_ioctl(struct inode *ino, struct file *filp,
837 	    unsigned int cmd, unsigned long arg)
838 {
839 	int len = 0;
840 	struct cache_reader *rp = filp->private_data;
841 	struct cache_queue *cq;
842 	struct cache_detail *cd = PDE(ino)->data;
843 
844 	if (cmd != FIONREAD || !rp)
845 		return -EINVAL;
846 
847 	spin_lock(&queue_lock);
848 
849 	/* only find the length remaining in current request,
850 	 * or the length of the next request
851 	 */
852 	for (cq= &rp->q; &cq->list != &cd->queue;
853 	     cq = list_entry(cq->list.next, struct cache_queue, list))
854 		if (!cq->reader) {
855 			struct cache_request *cr =
856 				container_of(cq, struct cache_request, q);
857 			len = cr->len - rp->offset;
858 			break;
859 		}
860 	spin_unlock(&queue_lock);
861 
862 	return put_user(len, (int __user *)arg);
863 }
864 
865 static int
866 cache_open(struct inode *inode, struct file *filp)
867 {
868 	struct cache_reader *rp = NULL;
869 
870 	nonseekable_open(inode, filp);
871 	if (filp->f_mode & FMODE_READ) {
872 		struct cache_detail *cd = PDE(inode)->data;
873 
874 		rp = kmalloc(sizeof(*rp), GFP_KERNEL);
875 		if (!rp)
876 			return -ENOMEM;
877 		rp->offset = 0;
878 		rp->q.reader = 1;
879 		atomic_inc(&cd->readers);
880 		spin_lock(&queue_lock);
881 		list_add(&rp->q.list, &cd->queue);
882 		spin_unlock(&queue_lock);
883 	}
884 	filp->private_data = rp;
885 	return 0;
886 }
887 
888 static int
889 cache_release(struct inode *inode, struct file *filp)
890 {
891 	struct cache_reader *rp = filp->private_data;
892 	struct cache_detail *cd = PDE(inode)->data;
893 
894 	if (rp) {
895 		spin_lock(&queue_lock);
896 		if (rp->offset) {
897 			struct cache_queue *cq;
898 			for (cq= &rp->q; &cq->list != &cd->queue;
899 			     cq = list_entry(cq->list.next, struct cache_queue, list))
900 				if (!cq->reader) {
901 					container_of(cq, struct cache_request, q)
902 						->readers--;
903 					break;
904 				}
905 			rp->offset = 0;
906 		}
907 		list_del(&rp->q.list);
908 		spin_unlock(&queue_lock);
909 
910 		filp->private_data = NULL;
911 		kfree(rp);
912 
913 		cd->last_close = get_seconds();
914 		atomic_dec(&cd->readers);
915 	}
916 	return 0;
917 }
918 
919 
920 
921 static const struct file_operations cache_file_operations = {
922 	.owner		= THIS_MODULE,
923 	.llseek		= no_llseek,
924 	.read		= cache_read,
925 	.write		= cache_write,
926 	.poll		= cache_poll,
927 	.ioctl		= cache_ioctl, /* for FIONREAD */
928 	.open		= cache_open,
929 	.release	= cache_release,
930 };
931 
932 
933 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
934 {
935 	struct cache_queue *cq;
936 	spin_lock(&queue_lock);
937 	list_for_each_entry(cq, &detail->queue, list)
938 		if (!cq->reader) {
939 			struct cache_request *cr = container_of(cq, struct cache_request, q);
940 			if (cr->item != ch)
941 				continue;
942 			if (cr->readers != 0)
943 				continue;
944 			list_del(&cr->q.list);
945 			spin_unlock(&queue_lock);
946 			cache_put(cr->item, detail);
947 			kfree(cr->buf);
948 			kfree(cr);
949 			return;
950 		}
951 	spin_unlock(&queue_lock);
952 }
953 
954 /*
955  * Support routines for text-based upcalls.
956  * Fields are separated by spaces.
957  * Fields are either mangled to quote space tab newline slosh with slosh
958  * or a hexified with a leading \x
959  * Record is terminated with newline.
960  *
961  */
962 
963 void qword_add(char **bpp, int *lp, char *str)
964 {
965 	char *bp = *bpp;
966 	int len = *lp;
967 	char c;
968 
969 	if (len < 0) return;
970 
971 	while ((c=*str++) && len)
972 		switch(c) {
973 		case ' ':
974 		case '\t':
975 		case '\n':
976 		case '\\':
977 			if (len >= 4) {
978 				*bp++ = '\\';
979 				*bp++ = '0' + ((c & 0300)>>6);
980 				*bp++ = '0' + ((c & 0070)>>3);
981 				*bp++ = '0' + ((c & 0007)>>0);
982 			}
983 			len -= 4;
984 			break;
985 		default:
986 			*bp++ = c;
987 			len--;
988 		}
989 	if (c || len <1) len = -1;
990 	else {
991 		*bp++ = ' ';
992 		len--;
993 	}
994 	*bpp = bp;
995 	*lp = len;
996 }
997 EXPORT_SYMBOL(qword_add);
998 
999 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1000 {
1001 	char *bp = *bpp;
1002 	int len = *lp;
1003 
1004 	if (len < 0) return;
1005 
1006 	if (len > 2) {
1007 		*bp++ = '\\';
1008 		*bp++ = 'x';
1009 		len -= 2;
1010 		while (blen && len >= 2) {
1011 			unsigned char c = *buf++;
1012 			*bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1013 			*bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1014 			len -= 2;
1015 			blen--;
1016 		}
1017 	}
1018 	if (blen || len<1) len = -1;
1019 	else {
1020 		*bp++ = ' ';
1021 		len--;
1022 	}
1023 	*bpp = bp;
1024 	*lp = len;
1025 }
1026 EXPORT_SYMBOL(qword_addhex);
1027 
1028 static void warn_no_listener(struct cache_detail *detail)
1029 {
1030 	if (detail->last_warn != detail->last_close) {
1031 		detail->last_warn = detail->last_close;
1032 		if (detail->warn_no_listener)
1033 			detail->warn_no_listener(detail);
1034 	}
1035 }
1036 
1037 /*
1038  * register an upcall request to user-space.
1039  * Each request is at most one page long.
1040  */
1041 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
1042 {
1043 
1044 	char *buf;
1045 	struct cache_request *crq;
1046 	char *bp;
1047 	int len;
1048 
1049 	if (detail->cache_request == NULL)
1050 		return -EINVAL;
1051 
1052 	if (atomic_read(&detail->readers) == 0 &&
1053 	    detail->last_close < get_seconds() - 30) {
1054 			warn_no_listener(detail);
1055 			return -EINVAL;
1056 	}
1057 
1058 	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1059 	if (!buf)
1060 		return -EAGAIN;
1061 
1062 	crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1063 	if (!crq) {
1064 		kfree(buf);
1065 		return -EAGAIN;
1066 	}
1067 
1068 	bp = buf; len = PAGE_SIZE;
1069 
1070 	detail->cache_request(detail, h, &bp, &len);
1071 
1072 	if (len < 0) {
1073 		kfree(buf);
1074 		kfree(crq);
1075 		return -EAGAIN;
1076 	}
1077 	crq->q.reader = 0;
1078 	crq->item = cache_get(h);
1079 	crq->buf = buf;
1080 	crq->len = PAGE_SIZE - len;
1081 	crq->readers = 0;
1082 	spin_lock(&queue_lock);
1083 	list_add_tail(&crq->q.list, &detail->queue);
1084 	spin_unlock(&queue_lock);
1085 	wake_up(&queue_wait);
1086 	return 0;
1087 }
1088 
1089 /*
1090  * parse a message from user-space and pass it
1091  * to an appropriate cache
1092  * Messages are, like requests, separated into fields by
1093  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1094  *
1095  * Message is
1096  *   reply cachename expiry key ... content....
1097  *
1098  * key and content are both parsed by cache
1099  */
1100 
1101 #define isodigit(c) (isdigit(c) && c <= '7')
1102 int qword_get(char **bpp, char *dest, int bufsize)
1103 {
1104 	/* return bytes copied, or -1 on error */
1105 	char *bp = *bpp;
1106 	int len = 0;
1107 
1108 	while (*bp == ' ') bp++;
1109 
1110 	if (bp[0] == '\\' && bp[1] == 'x') {
1111 		/* HEX STRING */
1112 		bp += 2;
1113 		while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1114 			int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1115 			bp++;
1116 			byte <<= 4;
1117 			byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1118 			*dest++ = byte;
1119 			bp++;
1120 			len++;
1121 		}
1122 	} else {
1123 		/* text with \nnn octal quoting */
1124 		while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1125 			if (*bp == '\\' &&
1126 			    isodigit(bp[1]) && (bp[1] <= '3') &&
1127 			    isodigit(bp[2]) &&
1128 			    isodigit(bp[3])) {
1129 				int byte = (*++bp -'0');
1130 				bp++;
1131 				byte = (byte << 3) | (*bp++ - '0');
1132 				byte = (byte << 3) | (*bp++ - '0');
1133 				*dest++ = byte;
1134 				len++;
1135 			} else {
1136 				*dest++ = *bp++;
1137 				len++;
1138 			}
1139 		}
1140 	}
1141 
1142 	if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1143 		return -1;
1144 	while (*bp == ' ') bp++;
1145 	*bpp = bp;
1146 	*dest = '\0';
1147 	return len;
1148 }
1149 EXPORT_SYMBOL(qword_get);
1150 
1151 
1152 /*
1153  * support /proc/sunrpc/cache/$CACHENAME/content
1154  * as a seqfile.
1155  * We call ->cache_show passing NULL for the item to
1156  * get a header, then pass each real item in the cache
1157  */
1158 
1159 struct handle {
1160 	struct cache_detail *cd;
1161 };
1162 
1163 static void *c_start(struct seq_file *m, loff_t *pos)
1164 	__acquires(cd->hash_lock)
1165 {
1166 	loff_t n = *pos;
1167 	unsigned hash, entry;
1168 	struct cache_head *ch;
1169 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1170 
1171 
1172 	read_lock(&cd->hash_lock);
1173 	if (!n--)
1174 		return SEQ_START_TOKEN;
1175 	hash = n >> 32;
1176 	entry = n & ((1LL<<32) - 1);
1177 
1178 	for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1179 		if (!entry--)
1180 			return ch;
1181 	n &= ~((1LL<<32) - 1);
1182 	do {
1183 		hash++;
1184 		n += 1LL<<32;
1185 	} while(hash < cd->hash_size &&
1186 		cd->hash_table[hash]==NULL);
1187 	if (hash >= cd->hash_size)
1188 		return NULL;
1189 	*pos = n+1;
1190 	return cd->hash_table[hash];
1191 }
1192 
1193 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1194 {
1195 	struct cache_head *ch = p;
1196 	int hash = (*pos >> 32);
1197 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1198 
1199 	if (p == SEQ_START_TOKEN)
1200 		hash = 0;
1201 	else if (ch->next == NULL) {
1202 		hash++;
1203 		*pos += 1LL<<32;
1204 	} else {
1205 		++*pos;
1206 		return ch->next;
1207 	}
1208 	*pos &= ~((1LL<<32) - 1);
1209 	while (hash < cd->hash_size &&
1210 	       cd->hash_table[hash] == NULL) {
1211 		hash++;
1212 		*pos += 1LL<<32;
1213 	}
1214 	if (hash >= cd->hash_size)
1215 		return NULL;
1216 	++*pos;
1217 	return cd->hash_table[hash];
1218 }
1219 
1220 static void c_stop(struct seq_file *m, void *p)
1221 	__releases(cd->hash_lock)
1222 {
1223 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1224 	read_unlock(&cd->hash_lock);
1225 }
1226 
1227 static int c_show(struct seq_file *m, void *p)
1228 {
1229 	struct cache_head *cp = p;
1230 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1231 
1232 	if (p == SEQ_START_TOKEN)
1233 		return cd->cache_show(m, cd, NULL);
1234 
1235 	ifdebug(CACHE)
1236 		seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1237 			   cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1238 	cache_get(cp);
1239 	if (cache_check(cd, cp, NULL))
1240 		/* cache_check does a cache_put on failure */
1241 		seq_printf(m, "# ");
1242 	else
1243 		cache_put(cp, cd);
1244 
1245 	return cd->cache_show(m, cd, cp);
1246 }
1247 
1248 static const struct seq_operations cache_content_op = {
1249 	.start	= c_start,
1250 	.next	= c_next,
1251 	.stop	= c_stop,
1252 	.show	= c_show,
1253 };
1254 
1255 static int content_open(struct inode *inode, struct file *file)
1256 {
1257 	struct handle *han;
1258 	struct cache_detail *cd = PDE(inode)->data;
1259 
1260 	han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1261 	if (han == NULL)
1262 		return -ENOMEM;
1263 
1264 	han->cd = cd;
1265 	return 0;
1266 }
1267 
1268 static const struct file_operations content_file_operations = {
1269 	.open		= content_open,
1270 	.read		= seq_read,
1271 	.llseek		= seq_lseek,
1272 	.release	= seq_release_private,
1273 };
1274 
1275 static ssize_t read_flush(struct file *file, char __user *buf,
1276 			    size_t count, loff_t *ppos)
1277 {
1278 	struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1279 	char tbuf[20];
1280 	unsigned long p = *ppos;
1281 	size_t len;
1282 
1283 	sprintf(tbuf, "%lu\n", cd->flush_time);
1284 	len = strlen(tbuf);
1285 	if (p >= len)
1286 		return 0;
1287 	len -= p;
1288 	if (len > count)
1289 		len = count;
1290 	if (copy_to_user(buf, (void*)(tbuf+p), len))
1291 		return -EFAULT;
1292 	*ppos += len;
1293 	return len;
1294 }
1295 
1296 static ssize_t write_flush(struct file * file, const char __user * buf,
1297 			     size_t count, loff_t *ppos)
1298 {
1299 	struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1300 	char tbuf[20];
1301 	char *ep;
1302 	long flushtime;
1303 	if (*ppos || count > sizeof(tbuf)-1)
1304 		return -EINVAL;
1305 	if (copy_from_user(tbuf, buf, count))
1306 		return -EFAULT;
1307 	tbuf[count] = 0;
1308 	flushtime = simple_strtoul(tbuf, &ep, 0);
1309 	if (*ep && *ep != '\n')
1310 		return -EINVAL;
1311 
1312 	cd->flush_time = flushtime;
1313 	cd->nextcheck = get_seconds();
1314 	cache_flush();
1315 
1316 	*ppos += count;
1317 	return count;
1318 }
1319 
1320 static const struct file_operations cache_flush_operations = {
1321 	.open		= nonseekable_open,
1322 	.read		= read_flush,
1323 	.write		= write_flush,
1324 };
1325