xref: /openbmc/linux/net/sunrpc/cache.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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 <asm/ioctls.h>
30 #include <linux/sunrpc/types.h>
31 #include <linux/sunrpc/cache.h>
32 #include <linux/sunrpc/stats.h>
33 
34 #define	 RPCDBG_FACILITY RPCDBG_CACHE
35 
36 static void cache_defer_req(struct cache_req *req, struct cache_head *item);
37 static void cache_revisit_request(struct cache_head *item);
38 
39 void cache_init(struct cache_head *h)
40 {
41 	time_t now = get_seconds();
42 	h->next = NULL;
43 	h->flags = 0;
44 	atomic_set(&h->refcnt, 1);
45 	h->expiry_time = now + CACHE_NEW_EXPIRY;
46 	h->last_refresh = now;
47 }
48 
49 
50 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
51 /*
52  * This is the generic cache management routine for all
53  * the authentication caches.
54  * It checks the currency of a cache item and will (later)
55  * initiate an upcall to fill it if needed.
56  *
57  *
58  * Returns 0 if the cache_head can be used, or cache_puts it and returns
59  * -EAGAIN if upcall is pending,
60  * -ENOENT if cache entry was negative
61  */
62 int cache_check(struct cache_detail *detail,
63 		    struct cache_head *h, struct cache_req *rqstp)
64 {
65 	int rv;
66 	long refresh_age, age;
67 
68 	/* First decide return status as best we can */
69 	if (!test_bit(CACHE_VALID, &h->flags) ||
70 	    h->expiry_time < get_seconds())
71 		rv = -EAGAIN;
72 	else if (detail->flush_time > h->last_refresh)
73 		rv = -EAGAIN;
74 	else {
75 		/* entry is valid */
76 		if (test_bit(CACHE_NEGATIVE, &h->flags))
77 			rv = -ENOENT;
78 		else rv = 0;
79 	}
80 
81 	/* now see if we want to start an upcall */
82 	refresh_age = (h->expiry_time - h->last_refresh);
83 	age = get_seconds() - h->last_refresh;
84 
85 	if (rqstp == NULL) {
86 		if (rv == -EAGAIN)
87 			rv = -ENOENT;
88 	} else if (rv == -EAGAIN || age > refresh_age/2) {
89 		dprintk("Want update, refage=%ld, age=%ld\n", refresh_age, age);
90 		if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
91 			switch (cache_make_upcall(detail, h)) {
92 			case -EINVAL:
93 				clear_bit(CACHE_PENDING, &h->flags);
94 				if (rv == -EAGAIN) {
95 					set_bit(CACHE_NEGATIVE, &h->flags);
96 					cache_fresh(detail, h, get_seconds()+CACHE_NEW_EXPIRY);
97 					rv = -ENOENT;
98 				}
99 				break;
100 
101 			case -EAGAIN:
102 				clear_bit(CACHE_PENDING, &h->flags);
103 				cache_revisit_request(h);
104 				break;
105 			}
106 		}
107 	}
108 
109 	if (rv == -EAGAIN)
110 		cache_defer_req(rqstp, h);
111 
112 	if (rv && h)
113 		detail->cache_put(h, detail);
114 	return rv;
115 }
116 
117 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
118 
119 void cache_fresh(struct cache_detail *detail,
120 		 struct cache_head *head, time_t expiry)
121 {
122 
123 	head->expiry_time = expiry;
124 	head->last_refresh = get_seconds();
125 	if (!test_and_set_bit(CACHE_VALID, &head->flags))
126 		cache_revisit_request(head);
127 	if (test_and_clear_bit(CACHE_PENDING, &head->flags))
128 		queue_loose(detail, head);
129 }
130 
131 /*
132  * caches need to be periodically cleaned.
133  * For this we maintain a list of cache_detail and
134  * a current pointer into that list and into the table
135  * for that entry.
136  *
137  * Each time clean_cache is called it finds the next non-empty entry
138  * in the current table and walks the list in that entry
139  * looking for entries that can be removed.
140  *
141  * An entry gets removed if:
142  * - The expiry is before current time
143  * - The last_refresh time is before the flush_time for that cache
144  *
145  * later we might drop old entries with non-NEVER expiry if that table
146  * is getting 'full' for some definition of 'full'
147  *
148  * The question of "how often to scan a table" is an interesting one
149  * and is answered in part by the use of the "nextcheck" field in the
150  * cache_detail.
151  * When a scan of a table begins, the nextcheck field is set to a time
152  * that is well into the future.
153  * While scanning, if an expiry time is found that is earlier than the
154  * current nextcheck time, nextcheck is set to that expiry time.
155  * If the flush_time is ever set to a time earlier than the nextcheck
156  * time, the nextcheck time is then set to that flush_time.
157  *
158  * A table is then only scanned if the current time is at least
159  * the nextcheck time.
160  *
161  */
162 
163 static LIST_HEAD(cache_list);
164 static DEFINE_SPINLOCK(cache_list_lock);
165 static struct cache_detail *current_detail;
166 static int current_index;
167 
168 static struct file_operations cache_file_operations;
169 static struct file_operations content_file_operations;
170 static struct file_operations cache_flush_operations;
171 
172 static void do_cache_clean(void *data);
173 static DECLARE_WORK(cache_cleaner, do_cache_clean, NULL);
174 
175 void cache_register(struct cache_detail *cd)
176 {
177 	cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
178 	if (cd->proc_ent) {
179 		struct proc_dir_entry *p;
180 		cd->proc_ent->owner = cd->owner;
181 		cd->channel_ent = cd->content_ent = NULL;
182 
183  		p = create_proc_entry("flush", S_IFREG|S_IRUSR|S_IWUSR,
184  				      cd->proc_ent);
185 		cd->flush_ent =  p;
186  		if (p) {
187  			p->proc_fops = &cache_flush_operations;
188  			p->owner = cd->owner;
189  			p->data = cd;
190  		}
191 
192 		if (cd->cache_request || cd->cache_parse) {
193 			p = create_proc_entry("channel", S_IFREG|S_IRUSR|S_IWUSR,
194 					      cd->proc_ent);
195 			cd->channel_ent = p;
196 			if (p) {
197 				p->proc_fops = &cache_file_operations;
198 				p->owner = cd->owner;
199 				p->data = cd;
200 			}
201 		}
202  		if (cd->cache_show) {
203  			p = create_proc_entry("content", S_IFREG|S_IRUSR|S_IWUSR,
204  					      cd->proc_ent);
205 			cd->content_ent = p;
206  			if (p) {
207  				p->proc_fops = &content_file_operations;
208  				p->owner = cd->owner;
209  				p->data = cd;
210  			}
211  		}
212 	}
213 	rwlock_init(&cd->hash_lock);
214 	INIT_LIST_HEAD(&cd->queue);
215 	spin_lock(&cache_list_lock);
216 	cd->nextcheck = 0;
217 	cd->entries = 0;
218 	atomic_set(&cd->readers, 0);
219 	cd->last_close = 0;
220 	cd->last_warn = -1;
221 	list_add(&cd->others, &cache_list);
222 	spin_unlock(&cache_list_lock);
223 
224 	/* start the cleaning process */
225 	schedule_work(&cache_cleaner);
226 }
227 
228 int cache_unregister(struct cache_detail *cd)
229 {
230 	cache_purge(cd);
231 	spin_lock(&cache_list_lock);
232 	write_lock(&cd->hash_lock);
233 	if (cd->entries || atomic_read(&cd->inuse)) {
234 		write_unlock(&cd->hash_lock);
235 		spin_unlock(&cache_list_lock);
236 		return -EBUSY;
237 	}
238 	if (current_detail == cd)
239 		current_detail = NULL;
240 	list_del_init(&cd->others);
241 	write_unlock(&cd->hash_lock);
242 	spin_unlock(&cache_list_lock);
243 	if (cd->proc_ent) {
244 		if (cd->flush_ent)
245 			remove_proc_entry("flush", cd->proc_ent);
246 		if (cd->channel_ent)
247 			remove_proc_entry("channel", cd->proc_ent);
248 		if (cd->content_ent)
249 			remove_proc_entry("content", cd->proc_ent);
250 
251 		cd->proc_ent = NULL;
252 		remove_proc_entry(cd->name, proc_net_rpc);
253 	}
254 	if (list_empty(&cache_list)) {
255 		/* module must be being unloaded so its safe to kill the worker */
256 		cancel_delayed_work(&cache_cleaner);
257 		flush_scheduled_work();
258 	}
259 	return 0;
260 }
261 
262 /* clean cache tries to find something to clean
263  * and cleans it.
264  * It returns 1 if it cleaned something,
265  *            0 if it didn't find anything this time
266  *           -1 if it fell off the end of the list.
267  */
268 static int cache_clean(void)
269 {
270 	int rv = 0;
271 	struct list_head *next;
272 
273 	spin_lock(&cache_list_lock);
274 
275 	/* find a suitable table if we don't already have one */
276 	while (current_detail == NULL ||
277 	    current_index >= current_detail->hash_size) {
278 		if (current_detail)
279 			next = current_detail->others.next;
280 		else
281 			next = cache_list.next;
282 		if (next == &cache_list) {
283 			current_detail = NULL;
284 			spin_unlock(&cache_list_lock);
285 			return -1;
286 		}
287 		current_detail = list_entry(next, struct cache_detail, others);
288 		if (current_detail->nextcheck > get_seconds())
289 			current_index = current_detail->hash_size;
290 		else {
291 			current_index = 0;
292 			current_detail->nextcheck = get_seconds()+30*60;
293 		}
294 	}
295 
296 	/* find a non-empty bucket in the table */
297 	while (current_detail &&
298 	       current_index < current_detail->hash_size &&
299 	       current_detail->hash_table[current_index] == NULL)
300 		current_index++;
301 
302 	/* find a cleanable entry in the bucket and clean it, or set to next bucket */
303 
304 	if (current_detail && current_index < current_detail->hash_size) {
305 		struct cache_head *ch, **cp;
306 		struct cache_detail *d;
307 
308 		write_lock(&current_detail->hash_lock);
309 
310 		/* Ok, now to clean this strand */
311 
312 		cp = & current_detail->hash_table[current_index];
313 		ch = *cp;
314 		for (; ch; cp= & ch->next, ch= *cp) {
315 			if (current_detail->nextcheck > ch->expiry_time)
316 				current_detail->nextcheck = ch->expiry_time+1;
317 			if (ch->expiry_time >= get_seconds()
318 			    && ch->last_refresh >= current_detail->flush_time
319 				)
320 				continue;
321 			if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
322 				queue_loose(current_detail, ch);
323 
324 			if (atomic_read(&ch->refcnt) == 1)
325 				break;
326 		}
327 		if (ch) {
328 			*cp = ch->next;
329 			ch->next = NULL;
330 			current_detail->entries--;
331 			rv = 1;
332 		}
333 		write_unlock(&current_detail->hash_lock);
334 		d = current_detail;
335 		if (!ch)
336 			current_index ++;
337 		spin_unlock(&cache_list_lock);
338 		if (ch)
339 			d->cache_put(ch, d);
340 	} else
341 		spin_unlock(&cache_list_lock);
342 
343 	return rv;
344 }
345 
346 /*
347  * We want to regularly clean the cache, so we need to schedule some work ...
348  */
349 static void do_cache_clean(void *data)
350 {
351 	int delay = 5;
352 	if (cache_clean() == -1)
353 		delay = 30*HZ;
354 
355 	if (list_empty(&cache_list))
356 		delay = 0;
357 
358 	if (delay)
359 		schedule_delayed_work(&cache_cleaner, delay);
360 }
361 
362 
363 /*
364  * Clean all caches promptly.  This just calls cache_clean
365  * repeatedly until we are sure that every cache has had a chance to
366  * be fully cleaned
367  */
368 void cache_flush(void)
369 {
370 	while (cache_clean() != -1)
371 		cond_resched();
372 	while (cache_clean() != -1)
373 		cond_resched();
374 }
375 
376 void cache_purge(struct cache_detail *detail)
377 {
378 	detail->flush_time = LONG_MAX;
379 	detail->nextcheck = get_seconds();
380 	cache_flush();
381 	detail->flush_time = 1;
382 }
383 
384 
385 
386 /*
387  * Deferral and Revisiting of Requests.
388  *
389  * If a cache lookup finds a pending entry, we
390  * need to defer the request and revisit it later.
391  * All deferred requests are stored in a hash table,
392  * indexed by "struct cache_head *".
393  * As it may be wasteful to store a whole request
394  * structure, we allow the request to provide a
395  * deferred form, which must contain a
396  * 'struct cache_deferred_req'
397  * This cache_deferred_req contains a method to allow
398  * it to be revisited when cache info is available
399  */
400 
401 #define	DFR_HASHSIZE	(PAGE_SIZE/sizeof(struct list_head))
402 #define	DFR_HASH(item)	((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
403 
404 #define	DFR_MAX	300	/* ??? */
405 
406 static DEFINE_SPINLOCK(cache_defer_lock);
407 static LIST_HEAD(cache_defer_list);
408 static struct list_head cache_defer_hash[DFR_HASHSIZE];
409 static int cache_defer_cnt;
410 
411 static void cache_defer_req(struct cache_req *req, struct cache_head *item)
412 {
413 	struct cache_deferred_req *dreq;
414 	int hash = DFR_HASH(item);
415 
416 	dreq = req->defer(req);
417 	if (dreq == NULL)
418 		return;
419 
420 	dreq->item = item;
421 	dreq->recv_time = get_seconds();
422 
423 	spin_lock(&cache_defer_lock);
424 
425 	list_add(&dreq->recent, &cache_defer_list);
426 
427 	if (cache_defer_hash[hash].next == NULL)
428 		INIT_LIST_HEAD(&cache_defer_hash[hash]);
429 	list_add(&dreq->hash, &cache_defer_hash[hash]);
430 
431 	/* it is in, now maybe clean up */
432 	dreq = NULL;
433 	if (++cache_defer_cnt > DFR_MAX) {
434 		/* too much in the cache, randomly drop
435 		 * first or last
436 		 */
437 		if (net_random()&1)
438 			dreq = list_entry(cache_defer_list.next,
439 					  struct cache_deferred_req,
440 					  recent);
441 		else
442 			dreq = list_entry(cache_defer_list.prev,
443 					  struct cache_deferred_req,
444 					  recent);
445 		list_del(&dreq->recent);
446 		list_del(&dreq->hash);
447 		cache_defer_cnt--;
448 	}
449 	spin_unlock(&cache_defer_lock);
450 
451 	if (dreq) {
452 		/* there was one too many */
453 		dreq->revisit(dreq, 1);
454 	}
455 	if (test_bit(CACHE_VALID, &item->flags)) {
456 		/* must have just been validated... */
457 		cache_revisit_request(item);
458 	}
459 }
460 
461 static void cache_revisit_request(struct cache_head *item)
462 {
463 	struct cache_deferred_req *dreq;
464 	struct list_head pending;
465 
466 	struct list_head *lp;
467 	int hash = DFR_HASH(item);
468 
469 	INIT_LIST_HEAD(&pending);
470 	spin_lock(&cache_defer_lock);
471 
472 	lp = cache_defer_hash[hash].next;
473 	if (lp) {
474 		while (lp != &cache_defer_hash[hash]) {
475 			dreq = list_entry(lp, struct cache_deferred_req, hash);
476 			lp = lp->next;
477 			if (dreq->item == item) {
478 				list_del(&dreq->hash);
479 				list_move(&dreq->recent, &pending);
480 				cache_defer_cnt--;
481 			}
482 		}
483 	}
484 	spin_unlock(&cache_defer_lock);
485 
486 	while (!list_empty(&pending)) {
487 		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
488 		list_del_init(&dreq->recent);
489 		dreq->revisit(dreq, 0);
490 	}
491 }
492 
493 void cache_clean_deferred(void *owner)
494 {
495 	struct cache_deferred_req *dreq, *tmp;
496 	struct list_head pending;
497 
498 
499 	INIT_LIST_HEAD(&pending);
500 	spin_lock(&cache_defer_lock);
501 
502 	list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
503 		if (dreq->owner == owner) {
504 			list_del(&dreq->hash);
505 			list_move(&dreq->recent, &pending);
506 			cache_defer_cnt--;
507 		}
508 	}
509 	spin_unlock(&cache_defer_lock);
510 
511 	while (!list_empty(&pending)) {
512 		dreq = list_entry(pending.next, struct cache_deferred_req, recent);
513 		list_del_init(&dreq->recent);
514 		dreq->revisit(dreq, 1);
515 	}
516 }
517 
518 /*
519  * communicate with user-space
520  *
521  * We have a magic /proc file - /proc/sunrpc/cache
522  * On read, you get a full request, or block
523  * On write, an update request is processed
524  * Poll works if anything to read, and always allows write
525  *
526  * Implemented by linked list of requests.  Each open file has
527  * a ->private that also exists in this list.  New request are added
528  * to the end and may wakeup and preceding readers.
529  * New readers are added to the head.  If, on read, an item is found with
530  * CACHE_UPCALLING clear, we free it from the list.
531  *
532  */
533 
534 static DEFINE_SPINLOCK(queue_lock);
535 static DECLARE_MUTEX(queue_io_sem);
536 
537 struct cache_queue {
538 	struct list_head	list;
539 	int			reader;	/* if 0, then request */
540 };
541 struct cache_request {
542 	struct cache_queue	q;
543 	struct cache_head	*item;
544 	char			* buf;
545 	int			len;
546 	int			readers;
547 };
548 struct cache_reader {
549 	struct cache_queue	q;
550 	int			offset;	/* if non-0, we have a refcnt on next request */
551 };
552 
553 static ssize_t
554 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
555 {
556 	struct cache_reader *rp = filp->private_data;
557 	struct cache_request *rq;
558 	struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
559 	int err;
560 
561 	if (count == 0)
562 		return 0;
563 
564 	down(&queue_io_sem); /* protect against multiple concurrent
565 			      * readers on this file */
566  again:
567 	spin_lock(&queue_lock);
568 	/* need to find next request */
569 	while (rp->q.list.next != &cd->queue &&
570 	       list_entry(rp->q.list.next, struct cache_queue, list)
571 	       ->reader) {
572 		struct list_head *next = rp->q.list.next;
573 		list_move(&rp->q.list, next);
574 	}
575 	if (rp->q.list.next == &cd->queue) {
576 		spin_unlock(&queue_lock);
577 		up(&queue_io_sem);
578 		if (rp->offset)
579 			BUG();
580 		return 0;
581 	}
582 	rq = container_of(rp->q.list.next, struct cache_request, q.list);
583 	if (rq->q.reader) BUG();
584 	if (rp->offset == 0)
585 		rq->readers++;
586 	spin_unlock(&queue_lock);
587 
588 	if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
589 		err = -EAGAIN;
590 		spin_lock(&queue_lock);
591 		list_move(&rp->q.list, &rq->q.list);
592 		spin_unlock(&queue_lock);
593 	} else {
594 		if (rp->offset + count > rq->len)
595 			count = rq->len - rp->offset;
596 		err = -EFAULT;
597 		if (copy_to_user(buf, rq->buf + rp->offset, count))
598 			goto out;
599 		rp->offset += count;
600 		if (rp->offset >= rq->len) {
601 			rp->offset = 0;
602 			spin_lock(&queue_lock);
603 			list_move(&rp->q.list, &rq->q.list);
604 			spin_unlock(&queue_lock);
605 		}
606 		err = 0;
607 	}
608  out:
609 	if (rp->offset == 0) {
610 		/* need to release rq */
611 		spin_lock(&queue_lock);
612 		rq->readers--;
613 		if (rq->readers == 0 &&
614 		    !test_bit(CACHE_PENDING, &rq->item->flags)) {
615 			list_del(&rq->q.list);
616 			spin_unlock(&queue_lock);
617 			cd->cache_put(rq->item, cd);
618 			kfree(rq->buf);
619 			kfree(rq);
620 		} else
621 			spin_unlock(&queue_lock);
622 	}
623 	if (err == -EAGAIN)
624 		goto again;
625 	up(&queue_io_sem);
626 	return err ? err :  count;
627 }
628 
629 static char write_buf[8192]; /* protected by queue_io_sem */
630 
631 static ssize_t
632 cache_write(struct file *filp, const char __user *buf, size_t count,
633 	    loff_t *ppos)
634 {
635 	int err;
636 	struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
637 
638 	if (count == 0)
639 		return 0;
640 	if (count >= sizeof(write_buf))
641 		return -EINVAL;
642 
643 	down(&queue_io_sem);
644 
645 	if (copy_from_user(write_buf, buf, count)) {
646 		up(&queue_io_sem);
647 		return -EFAULT;
648 	}
649 	write_buf[count] = '\0';
650 	if (cd->cache_parse)
651 		err = cd->cache_parse(cd, write_buf, count);
652 	else
653 		err = -EINVAL;
654 
655 	up(&queue_io_sem);
656 	return err ? err : count;
657 }
658 
659 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
660 
661 static unsigned int
662 cache_poll(struct file *filp, poll_table *wait)
663 {
664 	unsigned int mask;
665 	struct cache_reader *rp = filp->private_data;
666 	struct cache_queue *cq;
667 	struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
668 
669 	poll_wait(filp, &queue_wait, wait);
670 
671 	/* alway allow write */
672 	mask = POLL_OUT | POLLWRNORM;
673 
674 	if (!rp)
675 		return mask;
676 
677 	spin_lock(&queue_lock);
678 
679 	for (cq= &rp->q; &cq->list != &cd->queue;
680 	     cq = list_entry(cq->list.next, struct cache_queue, list))
681 		if (!cq->reader) {
682 			mask |= POLLIN | POLLRDNORM;
683 			break;
684 		}
685 	spin_unlock(&queue_lock);
686 	return mask;
687 }
688 
689 static int
690 cache_ioctl(struct inode *ino, struct file *filp,
691 	    unsigned int cmd, unsigned long arg)
692 {
693 	int len = 0;
694 	struct cache_reader *rp = filp->private_data;
695 	struct cache_queue *cq;
696 	struct cache_detail *cd = PDE(ino)->data;
697 
698 	if (cmd != FIONREAD || !rp)
699 		return -EINVAL;
700 
701 	spin_lock(&queue_lock);
702 
703 	/* only find the length remaining in current request,
704 	 * or the length of the next request
705 	 */
706 	for (cq= &rp->q; &cq->list != &cd->queue;
707 	     cq = list_entry(cq->list.next, struct cache_queue, list))
708 		if (!cq->reader) {
709 			struct cache_request *cr =
710 				container_of(cq, struct cache_request, q);
711 			len = cr->len - rp->offset;
712 			break;
713 		}
714 	spin_unlock(&queue_lock);
715 
716 	return put_user(len, (int __user *)arg);
717 }
718 
719 static int
720 cache_open(struct inode *inode, struct file *filp)
721 {
722 	struct cache_reader *rp = NULL;
723 
724 	nonseekable_open(inode, filp);
725 	if (filp->f_mode & FMODE_READ) {
726 		struct cache_detail *cd = PDE(inode)->data;
727 
728 		rp = kmalloc(sizeof(*rp), GFP_KERNEL);
729 		if (!rp)
730 			return -ENOMEM;
731 		rp->offset = 0;
732 		rp->q.reader = 1;
733 		atomic_inc(&cd->readers);
734 		spin_lock(&queue_lock);
735 		list_add(&rp->q.list, &cd->queue);
736 		spin_unlock(&queue_lock);
737 	}
738 	filp->private_data = rp;
739 	return 0;
740 }
741 
742 static int
743 cache_release(struct inode *inode, struct file *filp)
744 {
745 	struct cache_reader *rp = filp->private_data;
746 	struct cache_detail *cd = PDE(inode)->data;
747 
748 	if (rp) {
749 		spin_lock(&queue_lock);
750 		if (rp->offset) {
751 			struct cache_queue *cq;
752 			for (cq= &rp->q; &cq->list != &cd->queue;
753 			     cq = list_entry(cq->list.next, struct cache_queue, list))
754 				if (!cq->reader) {
755 					container_of(cq, struct cache_request, q)
756 						->readers--;
757 					break;
758 				}
759 			rp->offset = 0;
760 		}
761 		list_del(&rp->q.list);
762 		spin_unlock(&queue_lock);
763 
764 		filp->private_data = NULL;
765 		kfree(rp);
766 
767 		cd->last_close = get_seconds();
768 		atomic_dec(&cd->readers);
769 	}
770 	return 0;
771 }
772 
773 
774 
775 static struct file_operations cache_file_operations = {
776 	.owner		= THIS_MODULE,
777 	.llseek		= no_llseek,
778 	.read		= cache_read,
779 	.write		= cache_write,
780 	.poll		= cache_poll,
781 	.ioctl		= cache_ioctl, /* for FIONREAD */
782 	.open		= cache_open,
783 	.release	= cache_release,
784 };
785 
786 
787 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
788 {
789 	struct cache_queue *cq;
790 	spin_lock(&queue_lock);
791 	list_for_each_entry(cq, &detail->queue, list)
792 		if (!cq->reader) {
793 			struct cache_request *cr = container_of(cq, struct cache_request, q);
794 			if (cr->item != ch)
795 				continue;
796 			if (cr->readers != 0)
797 				break;
798 			list_del(&cr->q.list);
799 			spin_unlock(&queue_lock);
800 			detail->cache_put(cr->item, detail);
801 			kfree(cr->buf);
802 			kfree(cr);
803 			return;
804 		}
805 	spin_unlock(&queue_lock);
806 }
807 
808 /*
809  * Support routines for text-based upcalls.
810  * Fields are separated by spaces.
811  * Fields are either mangled to quote space tab newline slosh with slosh
812  * or a hexified with a leading \x
813  * Record is terminated with newline.
814  *
815  */
816 
817 void qword_add(char **bpp, int *lp, char *str)
818 {
819 	char *bp = *bpp;
820 	int len = *lp;
821 	char c;
822 
823 	if (len < 0) return;
824 
825 	while ((c=*str++) && len)
826 		switch(c) {
827 		case ' ':
828 		case '\t':
829 		case '\n':
830 		case '\\':
831 			if (len >= 4) {
832 				*bp++ = '\\';
833 				*bp++ = '0' + ((c & 0300)>>6);
834 				*bp++ = '0' + ((c & 0070)>>3);
835 				*bp++ = '0' + ((c & 0007)>>0);
836 			}
837 			len -= 4;
838 			break;
839 		default:
840 			*bp++ = c;
841 			len--;
842 		}
843 	if (c || len <1) len = -1;
844 	else {
845 		*bp++ = ' ';
846 		len--;
847 	}
848 	*bpp = bp;
849 	*lp = len;
850 }
851 
852 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
853 {
854 	char *bp = *bpp;
855 	int len = *lp;
856 
857 	if (len < 0) return;
858 
859 	if (len > 2) {
860 		*bp++ = '\\';
861 		*bp++ = 'x';
862 		len -= 2;
863 		while (blen && len >= 2) {
864 			unsigned char c = *buf++;
865 			*bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
866 			*bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
867 			len -= 2;
868 			blen--;
869 		}
870 	}
871 	if (blen || len<1) len = -1;
872 	else {
873 		*bp++ = ' ';
874 		len--;
875 	}
876 	*bpp = bp;
877 	*lp = len;
878 }
879 
880 static void warn_no_listener(struct cache_detail *detail)
881 {
882 	if (detail->last_warn != detail->last_close) {
883 		detail->last_warn = detail->last_close;
884 		if (detail->warn_no_listener)
885 			detail->warn_no_listener(detail);
886 	}
887 }
888 
889 /*
890  * register an upcall request to user-space.
891  * Each request is at most one page long.
892  */
893 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
894 {
895 
896 	char *buf;
897 	struct cache_request *crq;
898 	char *bp;
899 	int len;
900 
901 	if (detail->cache_request == NULL)
902 		return -EINVAL;
903 
904 	if (atomic_read(&detail->readers) == 0 &&
905 	    detail->last_close < get_seconds() - 30) {
906 			warn_no_listener(detail);
907 			return -EINVAL;
908 	}
909 
910 	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
911 	if (!buf)
912 		return -EAGAIN;
913 
914 	crq = kmalloc(sizeof (*crq), GFP_KERNEL);
915 	if (!crq) {
916 		kfree(buf);
917 		return -EAGAIN;
918 	}
919 
920 	bp = buf; len = PAGE_SIZE;
921 
922 	detail->cache_request(detail, h, &bp, &len);
923 
924 	if (len < 0) {
925 		kfree(buf);
926 		kfree(crq);
927 		return -EAGAIN;
928 	}
929 	crq->q.reader = 0;
930 	crq->item = cache_get(h);
931 	crq->buf = buf;
932 	crq->len = PAGE_SIZE - len;
933 	crq->readers = 0;
934 	spin_lock(&queue_lock);
935 	list_add_tail(&crq->q.list, &detail->queue);
936 	spin_unlock(&queue_lock);
937 	wake_up(&queue_wait);
938 	return 0;
939 }
940 
941 /*
942  * parse a message from user-space and pass it
943  * to an appropriate cache
944  * Messages are, like requests, separated into fields by
945  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
946  *
947  * Message is
948  *   reply cachename expiry key ... content....
949  *
950  * key and content are both parsed by cache
951  */
952 
953 #define isodigit(c) (isdigit(c) && c <= '7')
954 int qword_get(char **bpp, char *dest, int bufsize)
955 {
956 	/* return bytes copied, or -1 on error */
957 	char *bp = *bpp;
958 	int len = 0;
959 
960 	while (*bp == ' ') bp++;
961 
962 	if (bp[0] == '\\' && bp[1] == 'x') {
963 		/* HEX STRING */
964 		bp += 2;
965 		while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
966 			int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
967 			bp++;
968 			byte <<= 4;
969 			byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
970 			*dest++ = byte;
971 			bp++;
972 			len++;
973 		}
974 	} else {
975 		/* text with \nnn octal quoting */
976 		while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
977 			if (*bp == '\\' &&
978 			    isodigit(bp[1]) && (bp[1] <= '3') &&
979 			    isodigit(bp[2]) &&
980 			    isodigit(bp[3])) {
981 				int byte = (*++bp -'0');
982 				bp++;
983 				byte = (byte << 3) | (*bp++ - '0');
984 				byte = (byte << 3) | (*bp++ - '0');
985 				*dest++ = byte;
986 				len++;
987 			} else {
988 				*dest++ = *bp++;
989 				len++;
990 			}
991 		}
992 	}
993 
994 	if (*bp != ' ' && *bp != '\n' && *bp != '\0')
995 		return -1;
996 	while (*bp == ' ') bp++;
997 	*bpp = bp;
998 	*dest = '\0';
999 	return len;
1000 }
1001 
1002 
1003 /*
1004  * support /proc/sunrpc/cache/$CACHENAME/content
1005  * as a seqfile.
1006  * We call ->cache_show passing NULL for the item to
1007  * get a header, then pass each real item in the cache
1008  */
1009 
1010 struct handle {
1011 	struct cache_detail *cd;
1012 };
1013 
1014 static void *c_start(struct seq_file *m, loff_t *pos)
1015 {
1016 	loff_t n = *pos;
1017 	unsigned hash, entry;
1018 	struct cache_head *ch;
1019 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1020 
1021 
1022 	read_lock(&cd->hash_lock);
1023 	if (!n--)
1024 		return SEQ_START_TOKEN;
1025 	hash = n >> 32;
1026 	entry = n & ((1LL<<32) - 1);
1027 
1028 	for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1029 		if (!entry--)
1030 			return ch;
1031 	n &= ~((1LL<<32) - 1);
1032 	do {
1033 		hash++;
1034 		n += 1LL<<32;
1035 	} while(hash < cd->hash_size &&
1036 		cd->hash_table[hash]==NULL);
1037 	if (hash >= cd->hash_size)
1038 		return NULL;
1039 	*pos = n+1;
1040 	return cd->hash_table[hash];
1041 }
1042 
1043 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1044 {
1045 	struct cache_head *ch = p;
1046 	int hash = (*pos >> 32);
1047 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1048 
1049 	if (p == SEQ_START_TOKEN)
1050 		hash = 0;
1051 	else if (ch->next == NULL) {
1052 		hash++;
1053 		*pos += 1LL<<32;
1054 	} else {
1055 		++*pos;
1056 		return ch->next;
1057 	}
1058 	*pos &= ~((1LL<<32) - 1);
1059 	while (hash < cd->hash_size &&
1060 	       cd->hash_table[hash] == NULL) {
1061 		hash++;
1062 		*pos += 1LL<<32;
1063 	}
1064 	if (hash >= cd->hash_size)
1065 		return NULL;
1066 	++*pos;
1067 	return cd->hash_table[hash];
1068 }
1069 
1070 static void c_stop(struct seq_file *m, void *p)
1071 {
1072 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1073 	read_unlock(&cd->hash_lock);
1074 }
1075 
1076 static int c_show(struct seq_file *m, void *p)
1077 {
1078 	struct cache_head *cp = p;
1079 	struct cache_detail *cd = ((struct handle*)m->private)->cd;
1080 
1081 	if (p == SEQ_START_TOKEN)
1082 		return cd->cache_show(m, cd, NULL);
1083 
1084 	ifdebug(CACHE)
1085 		seq_printf(m, "# expiry=%ld refcnt=%d\n",
1086 			   cp->expiry_time, atomic_read(&cp->refcnt));
1087 	cache_get(cp);
1088 	if (cache_check(cd, cp, NULL))
1089 		/* cache_check does a cache_put on failure */
1090 		seq_printf(m, "# ");
1091 	else
1092 		cache_put(cp, cd);
1093 
1094 	return cd->cache_show(m, cd, cp);
1095 }
1096 
1097 static struct seq_operations cache_content_op = {
1098 	.start	= c_start,
1099 	.next	= c_next,
1100 	.stop	= c_stop,
1101 	.show	= c_show,
1102 };
1103 
1104 static int content_open(struct inode *inode, struct file *file)
1105 {
1106 	int res;
1107 	struct handle *han;
1108 	struct cache_detail *cd = PDE(inode)->data;
1109 
1110 	han = kmalloc(sizeof(*han), GFP_KERNEL);
1111 	if (han == NULL)
1112 		return -ENOMEM;
1113 
1114 	han->cd = cd;
1115 
1116 	res = seq_open(file, &cache_content_op);
1117 	if (res)
1118 		kfree(han);
1119 	else
1120 		((struct seq_file *)file->private_data)->private = han;
1121 
1122 	return res;
1123 }
1124 static int content_release(struct inode *inode, struct file *file)
1125 {
1126 	struct seq_file *m = (struct seq_file *)file->private_data;
1127 	struct handle *han = m->private;
1128 	kfree(han);
1129 	m->private = NULL;
1130 	return seq_release(inode, file);
1131 }
1132 
1133 static struct file_operations content_file_operations = {
1134 	.open		= content_open,
1135 	.read		= seq_read,
1136 	.llseek		= seq_lseek,
1137 	.release	= content_release,
1138 };
1139 
1140 static ssize_t read_flush(struct file *file, char __user *buf,
1141 			    size_t count, loff_t *ppos)
1142 {
1143 	struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1144 	char tbuf[20];
1145 	unsigned long p = *ppos;
1146 	int len;
1147 
1148 	sprintf(tbuf, "%lu\n", cd->flush_time);
1149 	len = strlen(tbuf);
1150 	if (p >= len)
1151 		return 0;
1152 	len -= p;
1153 	if (len > count) len = count;
1154 	if (copy_to_user(buf, (void*)(tbuf+p), len))
1155 		len = -EFAULT;
1156 	else
1157 		*ppos += len;
1158 	return len;
1159 }
1160 
1161 static ssize_t write_flush(struct file * file, const char __user * buf,
1162 			     size_t count, loff_t *ppos)
1163 {
1164 	struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1165 	char tbuf[20];
1166 	char *ep;
1167 	long flushtime;
1168 	if (*ppos || count > sizeof(tbuf)-1)
1169 		return -EINVAL;
1170 	if (copy_from_user(tbuf, buf, count))
1171 		return -EFAULT;
1172 	tbuf[count] = 0;
1173 	flushtime = simple_strtoul(tbuf, &ep, 0);
1174 	if (*ep && *ep != '\n')
1175 		return -EINVAL;
1176 
1177 	cd->flush_time = flushtime;
1178 	cd->nextcheck = get_seconds();
1179 	cache_flush();
1180 
1181 	*ppos += count;
1182 	return count;
1183 }
1184 
1185 static struct file_operations cache_flush_operations = {
1186 	.open		= nonseekable_open,
1187 	.read		= read_flush,
1188 	.write		= write_flush,
1189 };
1190