Loading...
1/*
2 * net/sunrpc/cache.c
3 *
4 * Generic code for various authentication-related caches
5 * used by sunrpc clients and servers.
6 *
7 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
8 *
9 * Released under terms in GPL version 2. See COPYING.
10 *
11 */
12
13#include <linux/types.h>
14#include <linux/fs.h>
15#include <linux/file.h>
16#include <linux/slab.h>
17#include <linux/signal.h>
18#include <linux/sched.h>
19#include <linux/kmod.h>
20#include <linux/list.h>
21#include <linux/module.h>
22#include <linux/ctype.h>
23#include <asm/uaccess.h>
24#include <linux/poll.h>
25#include <linux/seq_file.h>
26#include <linux/proc_fs.h>
27#include <linux/net.h>
28#include <linux/workqueue.h>
29#include <linux/mutex.h>
30#include <linux/pagemap.h>
31#include <asm/ioctls.h>
32#include <linux/sunrpc/types.h>
33#include <linux/sunrpc/cache.h>
34#include <linux/sunrpc/stats.h>
35#include <linux/sunrpc/rpc_pipe_fs.h>
36#include "netns.h"
37
38#define RPCDBG_FACILITY RPCDBG_CACHE
39
40static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
41static void cache_revisit_request(struct cache_head *item);
42
43static void cache_init(struct cache_head *h)
44{
45 time_t now = seconds_since_boot();
46 h->next = NULL;
47 h->flags = 0;
48 kref_init(&h->ref);
49 h->expiry_time = now + CACHE_NEW_EXPIRY;
50 h->last_refresh = now;
51}
52
53static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
54{
55 return (h->expiry_time < seconds_since_boot()) ||
56 (detail->flush_time > h->last_refresh);
57}
58
59struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
60 struct cache_head *key, int hash)
61{
62 struct cache_head **head, **hp;
63 struct cache_head *new = NULL, *freeme = NULL;
64
65 head = &detail->hash_table[hash];
66
67 read_lock(&detail->hash_lock);
68
69 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
70 struct cache_head *tmp = *hp;
71 if (detail->match(tmp, key)) {
72 if (cache_is_expired(detail, tmp))
73 /* This entry is expired, we will discard it. */
74 break;
75 cache_get(tmp);
76 read_unlock(&detail->hash_lock);
77 return tmp;
78 }
79 }
80 read_unlock(&detail->hash_lock);
81 /* Didn't find anything, insert an empty entry */
82
83 new = detail->alloc();
84 if (!new)
85 return NULL;
86 /* must fully initialise 'new', else
87 * we might get lose if we need to
88 * cache_put it soon.
89 */
90 cache_init(new);
91 detail->init(new, key);
92
93 write_lock(&detail->hash_lock);
94
95 /* check if entry appeared while we slept */
96 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
97 struct cache_head *tmp = *hp;
98 if (detail->match(tmp, key)) {
99 if (cache_is_expired(detail, tmp)) {
100 *hp = tmp->next;
101 tmp->next = NULL;
102 detail->entries --;
103 freeme = tmp;
104 break;
105 }
106 cache_get(tmp);
107 write_unlock(&detail->hash_lock);
108 cache_put(new, detail);
109 return tmp;
110 }
111 }
112 new->next = *head;
113 *head = new;
114 detail->entries++;
115 cache_get(new);
116 write_unlock(&detail->hash_lock);
117
118 if (freeme)
119 cache_put(freeme, detail);
120 return new;
121}
122EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
123
124
125static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
126
127static void cache_fresh_locked(struct cache_head *head, time_t expiry)
128{
129 head->expiry_time = expiry;
130 head->last_refresh = seconds_since_boot();
131 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
132 set_bit(CACHE_VALID, &head->flags);
133}
134
135static void cache_fresh_unlocked(struct cache_head *head,
136 struct cache_detail *detail)
137{
138 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
139 cache_revisit_request(head);
140 cache_dequeue(detail, head);
141 }
142}
143
144struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
145 struct cache_head *new, struct cache_head *old, int hash)
146{
147 /* The 'old' entry is to be replaced by 'new'.
148 * If 'old' is not VALID, we update it directly,
149 * otherwise we need to replace it
150 */
151 struct cache_head **head;
152 struct cache_head *tmp;
153
154 if (!test_bit(CACHE_VALID, &old->flags)) {
155 write_lock(&detail->hash_lock);
156 if (!test_bit(CACHE_VALID, &old->flags)) {
157 if (test_bit(CACHE_NEGATIVE, &new->flags))
158 set_bit(CACHE_NEGATIVE, &old->flags);
159 else
160 detail->update(old, new);
161 cache_fresh_locked(old, new->expiry_time);
162 write_unlock(&detail->hash_lock);
163 cache_fresh_unlocked(old, detail);
164 return old;
165 }
166 write_unlock(&detail->hash_lock);
167 }
168 /* We need to insert a new entry */
169 tmp = detail->alloc();
170 if (!tmp) {
171 cache_put(old, detail);
172 return NULL;
173 }
174 cache_init(tmp);
175 detail->init(tmp, old);
176 head = &detail->hash_table[hash];
177
178 write_lock(&detail->hash_lock);
179 if (test_bit(CACHE_NEGATIVE, &new->flags))
180 set_bit(CACHE_NEGATIVE, &tmp->flags);
181 else
182 detail->update(tmp, new);
183 tmp->next = *head;
184 *head = tmp;
185 detail->entries++;
186 cache_get(tmp);
187 cache_fresh_locked(tmp, new->expiry_time);
188 cache_fresh_locked(old, 0);
189 write_unlock(&detail->hash_lock);
190 cache_fresh_unlocked(tmp, detail);
191 cache_fresh_unlocked(old, detail);
192 cache_put(old, detail);
193 return tmp;
194}
195EXPORT_SYMBOL_GPL(sunrpc_cache_update);
196
197static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
198{
199 if (!cd->cache_upcall)
200 return -EINVAL;
201 return cd->cache_upcall(cd, h);
202}
203
204static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
205{
206 if (!test_bit(CACHE_VALID, &h->flags))
207 return -EAGAIN;
208 else {
209 /* entry is valid */
210 if (test_bit(CACHE_NEGATIVE, &h->flags))
211 return -ENOENT;
212 else {
213 /*
214 * In combination with write barrier in
215 * sunrpc_cache_update, ensures that anyone
216 * using the cache entry after this sees the
217 * updated contents:
218 */
219 smp_rmb();
220 return 0;
221 }
222 }
223}
224
225static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
226{
227 int rv;
228
229 write_lock(&detail->hash_lock);
230 rv = cache_is_valid(detail, h);
231 if (rv != -EAGAIN) {
232 write_unlock(&detail->hash_lock);
233 return rv;
234 }
235 set_bit(CACHE_NEGATIVE, &h->flags);
236 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
237 write_unlock(&detail->hash_lock);
238 cache_fresh_unlocked(h, detail);
239 return -ENOENT;
240}
241
242/*
243 * This is the generic cache management routine for all
244 * the authentication caches.
245 * It checks the currency of a cache item and will (later)
246 * initiate an upcall to fill it if needed.
247 *
248 *
249 * Returns 0 if the cache_head can be used, or cache_puts it and returns
250 * -EAGAIN if upcall is pending and request has been queued
251 * -ETIMEDOUT if upcall failed or request could not be queue or
252 * upcall completed but item is still invalid (implying that
253 * the cache item has been replaced with a newer one).
254 * -ENOENT if cache entry was negative
255 */
256int cache_check(struct cache_detail *detail,
257 struct cache_head *h, struct cache_req *rqstp)
258{
259 int rv;
260 long refresh_age, age;
261
262 /* First decide return status as best we can */
263 rv = cache_is_valid(detail, h);
264
265 /* now see if we want to start an upcall */
266 refresh_age = (h->expiry_time - h->last_refresh);
267 age = seconds_since_boot() - h->last_refresh;
268
269 if (rqstp == NULL) {
270 if (rv == -EAGAIN)
271 rv = -ENOENT;
272 } else if (rv == -EAGAIN || age > refresh_age/2) {
273 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
274 refresh_age, age);
275 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
276 switch (cache_make_upcall(detail, h)) {
277 case -EINVAL:
278 clear_bit(CACHE_PENDING, &h->flags);
279 cache_revisit_request(h);
280 rv = try_to_negate_entry(detail, h);
281 break;
282 case -EAGAIN:
283 clear_bit(CACHE_PENDING, &h->flags);
284 cache_revisit_request(h);
285 break;
286 }
287 }
288 }
289
290 if (rv == -EAGAIN) {
291 if (!cache_defer_req(rqstp, h)) {
292 /*
293 * Request was not deferred; handle it as best
294 * we can ourselves:
295 */
296 rv = cache_is_valid(detail, h);
297 if (rv == -EAGAIN)
298 rv = -ETIMEDOUT;
299 }
300 }
301 if (rv)
302 cache_put(h, detail);
303 return rv;
304}
305EXPORT_SYMBOL_GPL(cache_check);
306
307/*
308 * caches need to be periodically cleaned.
309 * For this we maintain a list of cache_detail and
310 * a current pointer into that list and into the table
311 * for that entry.
312 *
313 * Each time clean_cache is called it finds the next non-empty entry
314 * in the current table and walks the list in that entry
315 * looking for entries that can be removed.
316 *
317 * An entry gets removed if:
318 * - The expiry is before current time
319 * - The last_refresh time is before the flush_time for that cache
320 *
321 * later we might drop old entries with non-NEVER expiry if that table
322 * is getting 'full' for some definition of 'full'
323 *
324 * The question of "how often to scan a table" is an interesting one
325 * and is answered in part by the use of the "nextcheck" field in the
326 * cache_detail.
327 * When a scan of a table begins, the nextcheck field is set to a time
328 * that is well into the future.
329 * While scanning, if an expiry time is found that is earlier than the
330 * current nextcheck time, nextcheck is set to that expiry time.
331 * If the flush_time is ever set to a time earlier than the nextcheck
332 * time, the nextcheck time is then set to that flush_time.
333 *
334 * A table is then only scanned if the current time is at least
335 * the nextcheck time.
336 *
337 */
338
339static LIST_HEAD(cache_list);
340static DEFINE_SPINLOCK(cache_list_lock);
341static struct cache_detail *current_detail;
342static int current_index;
343
344static void do_cache_clean(struct work_struct *work);
345static struct delayed_work cache_cleaner;
346
347static void sunrpc_init_cache_detail(struct cache_detail *cd)
348{
349 rwlock_init(&cd->hash_lock);
350 INIT_LIST_HEAD(&cd->queue);
351 spin_lock(&cache_list_lock);
352 cd->nextcheck = 0;
353 cd->entries = 0;
354 atomic_set(&cd->readers, 0);
355 cd->last_close = 0;
356 cd->last_warn = -1;
357 list_add(&cd->others, &cache_list);
358 spin_unlock(&cache_list_lock);
359
360 /* start the cleaning process */
361 schedule_delayed_work(&cache_cleaner, 0);
362}
363
364static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
365{
366 cache_purge(cd);
367 spin_lock(&cache_list_lock);
368 write_lock(&cd->hash_lock);
369 if (cd->entries || atomic_read(&cd->inuse)) {
370 write_unlock(&cd->hash_lock);
371 spin_unlock(&cache_list_lock);
372 goto out;
373 }
374 if (current_detail == cd)
375 current_detail = NULL;
376 list_del_init(&cd->others);
377 write_unlock(&cd->hash_lock);
378 spin_unlock(&cache_list_lock);
379 if (list_empty(&cache_list)) {
380 /* module must be being unloaded so its safe to kill the worker */
381 cancel_delayed_work_sync(&cache_cleaner);
382 }
383 return;
384out:
385 printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
386}
387
388/* clean cache tries to find something to clean
389 * and cleans it.
390 * It returns 1 if it cleaned something,
391 * 0 if it didn't find anything this time
392 * -1 if it fell off the end of the list.
393 */
394static int cache_clean(void)
395{
396 int rv = 0;
397 struct list_head *next;
398
399 spin_lock(&cache_list_lock);
400
401 /* find a suitable table if we don't already have one */
402 while (current_detail == NULL ||
403 current_index >= current_detail->hash_size) {
404 if (current_detail)
405 next = current_detail->others.next;
406 else
407 next = cache_list.next;
408 if (next == &cache_list) {
409 current_detail = NULL;
410 spin_unlock(&cache_list_lock);
411 return -1;
412 }
413 current_detail = list_entry(next, struct cache_detail, others);
414 if (current_detail->nextcheck > seconds_since_boot())
415 current_index = current_detail->hash_size;
416 else {
417 current_index = 0;
418 current_detail->nextcheck = seconds_since_boot()+30*60;
419 }
420 }
421
422 /* find a non-empty bucket in the table */
423 while (current_detail &&
424 current_index < current_detail->hash_size &&
425 current_detail->hash_table[current_index] == NULL)
426 current_index++;
427
428 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
429
430 if (current_detail && current_index < current_detail->hash_size) {
431 struct cache_head *ch, **cp;
432 struct cache_detail *d;
433
434 write_lock(¤t_detail->hash_lock);
435
436 /* Ok, now to clean this strand */
437
438 cp = & current_detail->hash_table[current_index];
439 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
440 if (current_detail->nextcheck > ch->expiry_time)
441 current_detail->nextcheck = ch->expiry_time+1;
442 if (!cache_is_expired(current_detail, ch))
443 continue;
444
445 *cp = ch->next;
446 ch->next = NULL;
447 current_detail->entries--;
448 rv = 1;
449 break;
450 }
451
452 write_unlock(¤t_detail->hash_lock);
453 d = current_detail;
454 if (!ch)
455 current_index ++;
456 spin_unlock(&cache_list_lock);
457 if (ch) {
458 if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
459 cache_dequeue(current_detail, ch);
460 cache_revisit_request(ch);
461 cache_put(ch, d);
462 }
463 } else
464 spin_unlock(&cache_list_lock);
465
466 return rv;
467}
468
469/*
470 * We want to regularly clean the cache, so we need to schedule some work ...
471 */
472static void do_cache_clean(struct work_struct *work)
473{
474 int delay = 5;
475 if (cache_clean() == -1)
476 delay = round_jiffies_relative(30*HZ);
477
478 if (list_empty(&cache_list))
479 delay = 0;
480
481 if (delay)
482 schedule_delayed_work(&cache_cleaner, delay);
483}
484
485
486/*
487 * Clean all caches promptly. This just calls cache_clean
488 * repeatedly until we are sure that every cache has had a chance to
489 * be fully cleaned
490 */
491void cache_flush(void)
492{
493 while (cache_clean() != -1)
494 cond_resched();
495 while (cache_clean() != -1)
496 cond_resched();
497}
498EXPORT_SYMBOL_GPL(cache_flush);
499
500void cache_purge(struct cache_detail *detail)
501{
502 detail->flush_time = LONG_MAX;
503 detail->nextcheck = seconds_since_boot();
504 cache_flush();
505 detail->flush_time = 1;
506}
507EXPORT_SYMBOL_GPL(cache_purge);
508
509
510/*
511 * Deferral and Revisiting of Requests.
512 *
513 * If a cache lookup finds a pending entry, we
514 * need to defer the request and revisit it later.
515 * All deferred requests are stored in a hash table,
516 * indexed by "struct cache_head *".
517 * As it may be wasteful to store a whole request
518 * structure, we allow the request to provide a
519 * deferred form, which must contain a
520 * 'struct cache_deferred_req'
521 * This cache_deferred_req contains a method to allow
522 * it to be revisited when cache info is available
523 */
524
525#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
526#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
527
528#define DFR_MAX 300 /* ??? */
529
530static DEFINE_SPINLOCK(cache_defer_lock);
531static LIST_HEAD(cache_defer_list);
532static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
533static int cache_defer_cnt;
534
535static void __unhash_deferred_req(struct cache_deferred_req *dreq)
536{
537 hlist_del_init(&dreq->hash);
538 if (!list_empty(&dreq->recent)) {
539 list_del_init(&dreq->recent);
540 cache_defer_cnt--;
541 }
542}
543
544static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
545{
546 int hash = DFR_HASH(item);
547
548 INIT_LIST_HEAD(&dreq->recent);
549 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
550}
551
552static void setup_deferral(struct cache_deferred_req *dreq,
553 struct cache_head *item,
554 int count_me)
555{
556
557 dreq->item = item;
558
559 spin_lock(&cache_defer_lock);
560
561 __hash_deferred_req(dreq, item);
562
563 if (count_me) {
564 cache_defer_cnt++;
565 list_add(&dreq->recent, &cache_defer_list);
566 }
567
568 spin_unlock(&cache_defer_lock);
569
570}
571
572struct thread_deferred_req {
573 struct cache_deferred_req handle;
574 struct completion completion;
575};
576
577static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
578{
579 struct thread_deferred_req *dr =
580 container_of(dreq, struct thread_deferred_req, handle);
581 complete(&dr->completion);
582}
583
584static void cache_wait_req(struct cache_req *req, struct cache_head *item)
585{
586 struct thread_deferred_req sleeper;
587 struct cache_deferred_req *dreq = &sleeper.handle;
588
589 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
590 dreq->revisit = cache_restart_thread;
591
592 setup_deferral(dreq, item, 0);
593
594 if (!test_bit(CACHE_PENDING, &item->flags) ||
595 wait_for_completion_interruptible_timeout(
596 &sleeper.completion, req->thread_wait) <= 0) {
597 /* The completion wasn't completed, so we need
598 * to clean up
599 */
600 spin_lock(&cache_defer_lock);
601 if (!hlist_unhashed(&sleeper.handle.hash)) {
602 __unhash_deferred_req(&sleeper.handle);
603 spin_unlock(&cache_defer_lock);
604 } else {
605 /* cache_revisit_request already removed
606 * this from the hash table, but hasn't
607 * called ->revisit yet. It will very soon
608 * and we need to wait for it.
609 */
610 spin_unlock(&cache_defer_lock);
611 wait_for_completion(&sleeper.completion);
612 }
613 }
614}
615
616static void cache_limit_defers(void)
617{
618 /* Make sure we haven't exceed the limit of allowed deferred
619 * requests.
620 */
621 struct cache_deferred_req *discard = NULL;
622
623 if (cache_defer_cnt <= DFR_MAX)
624 return;
625
626 spin_lock(&cache_defer_lock);
627
628 /* Consider removing either the first or the last */
629 if (cache_defer_cnt > DFR_MAX) {
630 if (net_random() & 1)
631 discard = list_entry(cache_defer_list.next,
632 struct cache_deferred_req, recent);
633 else
634 discard = list_entry(cache_defer_list.prev,
635 struct cache_deferred_req, recent);
636 __unhash_deferred_req(discard);
637 }
638 spin_unlock(&cache_defer_lock);
639 if (discard)
640 discard->revisit(discard, 1);
641}
642
643/* Return true if and only if a deferred request is queued. */
644static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
645{
646 struct cache_deferred_req *dreq;
647
648 if (req->thread_wait) {
649 cache_wait_req(req, item);
650 if (!test_bit(CACHE_PENDING, &item->flags))
651 return false;
652 }
653 dreq = req->defer(req);
654 if (dreq == NULL)
655 return false;
656 setup_deferral(dreq, item, 1);
657 if (!test_bit(CACHE_PENDING, &item->flags))
658 /* Bit could have been cleared before we managed to
659 * set up the deferral, so need to revisit just in case
660 */
661 cache_revisit_request(item);
662
663 cache_limit_defers();
664 return true;
665}
666
667static void cache_revisit_request(struct cache_head *item)
668{
669 struct cache_deferred_req *dreq;
670 struct list_head pending;
671 struct hlist_node *lp, *tmp;
672 int hash = DFR_HASH(item);
673
674 INIT_LIST_HEAD(&pending);
675 spin_lock(&cache_defer_lock);
676
677 hlist_for_each_entry_safe(dreq, lp, tmp, &cache_defer_hash[hash], hash)
678 if (dreq->item == item) {
679 __unhash_deferred_req(dreq);
680 list_add(&dreq->recent, &pending);
681 }
682
683 spin_unlock(&cache_defer_lock);
684
685 while (!list_empty(&pending)) {
686 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
687 list_del_init(&dreq->recent);
688 dreq->revisit(dreq, 0);
689 }
690}
691
692void cache_clean_deferred(void *owner)
693{
694 struct cache_deferred_req *dreq, *tmp;
695 struct list_head pending;
696
697
698 INIT_LIST_HEAD(&pending);
699 spin_lock(&cache_defer_lock);
700
701 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
702 if (dreq->owner == owner) {
703 __unhash_deferred_req(dreq);
704 list_add(&dreq->recent, &pending);
705 }
706 }
707 spin_unlock(&cache_defer_lock);
708
709 while (!list_empty(&pending)) {
710 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
711 list_del_init(&dreq->recent);
712 dreq->revisit(dreq, 1);
713 }
714}
715
716/*
717 * communicate with user-space
718 *
719 * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
720 * On read, you get a full request, or block.
721 * On write, an update request is processed.
722 * Poll works if anything to read, and always allows write.
723 *
724 * Implemented by linked list of requests. Each open file has
725 * a ->private that also exists in this list. New requests are added
726 * to the end and may wakeup and preceding readers.
727 * New readers are added to the head. If, on read, an item is found with
728 * CACHE_UPCALLING clear, we free it from the list.
729 *
730 */
731
732static DEFINE_SPINLOCK(queue_lock);
733static DEFINE_MUTEX(queue_io_mutex);
734
735struct cache_queue {
736 struct list_head list;
737 int reader; /* if 0, then request */
738};
739struct cache_request {
740 struct cache_queue q;
741 struct cache_head *item;
742 char * buf;
743 int len;
744 int readers;
745};
746struct cache_reader {
747 struct cache_queue q;
748 int offset; /* if non-0, we have a refcnt on next request */
749};
750
751static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
752 loff_t *ppos, struct cache_detail *cd)
753{
754 struct cache_reader *rp = filp->private_data;
755 struct cache_request *rq;
756 struct inode *inode = filp->f_path.dentry->d_inode;
757 int err;
758
759 if (count == 0)
760 return 0;
761
762 mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
763 * readers on this file */
764 again:
765 spin_lock(&queue_lock);
766 /* need to find next request */
767 while (rp->q.list.next != &cd->queue &&
768 list_entry(rp->q.list.next, struct cache_queue, list)
769 ->reader) {
770 struct list_head *next = rp->q.list.next;
771 list_move(&rp->q.list, next);
772 }
773 if (rp->q.list.next == &cd->queue) {
774 spin_unlock(&queue_lock);
775 mutex_unlock(&inode->i_mutex);
776 BUG_ON(rp->offset);
777 return 0;
778 }
779 rq = container_of(rp->q.list.next, struct cache_request, q.list);
780 BUG_ON(rq->q.reader);
781 if (rp->offset == 0)
782 rq->readers++;
783 spin_unlock(&queue_lock);
784
785 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
786 err = -EAGAIN;
787 spin_lock(&queue_lock);
788 list_move(&rp->q.list, &rq->q.list);
789 spin_unlock(&queue_lock);
790 } else {
791 if (rp->offset + count > rq->len)
792 count = rq->len - rp->offset;
793 err = -EFAULT;
794 if (copy_to_user(buf, rq->buf + rp->offset, count))
795 goto out;
796 rp->offset += count;
797 if (rp->offset >= rq->len) {
798 rp->offset = 0;
799 spin_lock(&queue_lock);
800 list_move(&rp->q.list, &rq->q.list);
801 spin_unlock(&queue_lock);
802 }
803 err = 0;
804 }
805 out:
806 if (rp->offset == 0) {
807 /* need to release rq */
808 spin_lock(&queue_lock);
809 rq->readers--;
810 if (rq->readers == 0 &&
811 !test_bit(CACHE_PENDING, &rq->item->flags)) {
812 list_del(&rq->q.list);
813 spin_unlock(&queue_lock);
814 cache_put(rq->item, cd);
815 kfree(rq->buf);
816 kfree(rq);
817 } else
818 spin_unlock(&queue_lock);
819 }
820 if (err == -EAGAIN)
821 goto again;
822 mutex_unlock(&inode->i_mutex);
823 return err ? err : count;
824}
825
826static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
827 size_t count, struct cache_detail *cd)
828{
829 ssize_t ret;
830
831 if (copy_from_user(kaddr, buf, count))
832 return -EFAULT;
833 kaddr[count] = '\0';
834 ret = cd->cache_parse(cd, kaddr, count);
835 if (!ret)
836 ret = count;
837 return ret;
838}
839
840static ssize_t cache_slow_downcall(const char __user *buf,
841 size_t count, struct cache_detail *cd)
842{
843 static char write_buf[8192]; /* protected by queue_io_mutex */
844 ssize_t ret = -EINVAL;
845
846 if (count >= sizeof(write_buf))
847 goto out;
848 mutex_lock(&queue_io_mutex);
849 ret = cache_do_downcall(write_buf, buf, count, cd);
850 mutex_unlock(&queue_io_mutex);
851out:
852 return ret;
853}
854
855static ssize_t cache_downcall(struct address_space *mapping,
856 const char __user *buf,
857 size_t count, struct cache_detail *cd)
858{
859 struct page *page;
860 char *kaddr;
861 ssize_t ret = -ENOMEM;
862
863 if (count >= PAGE_CACHE_SIZE)
864 goto out_slow;
865
866 page = find_or_create_page(mapping, 0, GFP_KERNEL);
867 if (!page)
868 goto out_slow;
869
870 kaddr = kmap(page);
871 ret = cache_do_downcall(kaddr, buf, count, cd);
872 kunmap(page);
873 unlock_page(page);
874 page_cache_release(page);
875 return ret;
876out_slow:
877 return cache_slow_downcall(buf, count, cd);
878}
879
880static ssize_t cache_write(struct file *filp, const char __user *buf,
881 size_t count, loff_t *ppos,
882 struct cache_detail *cd)
883{
884 struct address_space *mapping = filp->f_mapping;
885 struct inode *inode = filp->f_path.dentry->d_inode;
886 ssize_t ret = -EINVAL;
887
888 if (!cd->cache_parse)
889 goto out;
890
891 mutex_lock(&inode->i_mutex);
892 ret = cache_downcall(mapping, buf, count, cd);
893 mutex_unlock(&inode->i_mutex);
894out:
895 return ret;
896}
897
898static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
899
900static unsigned int cache_poll(struct file *filp, poll_table *wait,
901 struct cache_detail *cd)
902{
903 unsigned int mask;
904 struct cache_reader *rp = filp->private_data;
905 struct cache_queue *cq;
906
907 poll_wait(filp, &queue_wait, wait);
908
909 /* alway allow write */
910 mask = POLL_OUT | POLLWRNORM;
911
912 if (!rp)
913 return mask;
914
915 spin_lock(&queue_lock);
916
917 for (cq= &rp->q; &cq->list != &cd->queue;
918 cq = list_entry(cq->list.next, struct cache_queue, list))
919 if (!cq->reader) {
920 mask |= POLLIN | POLLRDNORM;
921 break;
922 }
923 spin_unlock(&queue_lock);
924 return mask;
925}
926
927static int cache_ioctl(struct inode *ino, struct file *filp,
928 unsigned int cmd, unsigned long arg,
929 struct cache_detail *cd)
930{
931 int len = 0;
932 struct cache_reader *rp = filp->private_data;
933 struct cache_queue *cq;
934
935 if (cmd != FIONREAD || !rp)
936 return -EINVAL;
937
938 spin_lock(&queue_lock);
939
940 /* only find the length remaining in current request,
941 * or the length of the next request
942 */
943 for (cq= &rp->q; &cq->list != &cd->queue;
944 cq = list_entry(cq->list.next, struct cache_queue, list))
945 if (!cq->reader) {
946 struct cache_request *cr =
947 container_of(cq, struct cache_request, q);
948 len = cr->len - rp->offset;
949 break;
950 }
951 spin_unlock(&queue_lock);
952
953 return put_user(len, (int __user *)arg);
954}
955
956static int cache_open(struct inode *inode, struct file *filp,
957 struct cache_detail *cd)
958{
959 struct cache_reader *rp = NULL;
960
961 if (!cd || !try_module_get(cd->owner))
962 return -EACCES;
963 nonseekable_open(inode, filp);
964 if (filp->f_mode & FMODE_READ) {
965 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
966 if (!rp)
967 return -ENOMEM;
968 rp->offset = 0;
969 rp->q.reader = 1;
970 atomic_inc(&cd->readers);
971 spin_lock(&queue_lock);
972 list_add(&rp->q.list, &cd->queue);
973 spin_unlock(&queue_lock);
974 }
975 filp->private_data = rp;
976 return 0;
977}
978
979static int cache_release(struct inode *inode, struct file *filp,
980 struct cache_detail *cd)
981{
982 struct cache_reader *rp = filp->private_data;
983
984 if (rp) {
985 spin_lock(&queue_lock);
986 if (rp->offset) {
987 struct cache_queue *cq;
988 for (cq= &rp->q; &cq->list != &cd->queue;
989 cq = list_entry(cq->list.next, struct cache_queue, list))
990 if (!cq->reader) {
991 container_of(cq, struct cache_request, q)
992 ->readers--;
993 break;
994 }
995 rp->offset = 0;
996 }
997 list_del(&rp->q.list);
998 spin_unlock(&queue_lock);
999
1000 filp->private_data = NULL;
1001 kfree(rp);
1002
1003 cd->last_close = seconds_since_boot();
1004 atomic_dec(&cd->readers);
1005 }
1006 module_put(cd->owner);
1007 return 0;
1008}
1009
1010
1011
1012static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1013{
1014 struct cache_queue *cq;
1015 spin_lock(&queue_lock);
1016 list_for_each_entry(cq, &detail->queue, list)
1017 if (!cq->reader) {
1018 struct cache_request *cr = container_of(cq, struct cache_request, q);
1019 if (cr->item != ch)
1020 continue;
1021 if (cr->readers != 0)
1022 continue;
1023 list_del(&cr->q.list);
1024 spin_unlock(&queue_lock);
1025 cache_put(cr->item, detail);
1026 kfree(cr->buf);
1027 kfree(cr);
1028 return;
1029 }
1030 spin_unlock(&queue_lock);
1031}
1032
1033/*
1034 * Support routines for text-based upcalls.
1035 * Fields are separated by spaces.
1036 * Fields are either mangled to quote space tab newline slosh with slosh
1037 * or a hexified with a leading \x
1038 * Record is terminated with newline.
1039 *
1040 */
1041
1042void qword_add(char **bpp, int *lp, char *str)
1043{
1044 char *bp = *bpp;
1045 int len = *lp;
1046 char c;
1047
1048 if (len < 0) return;
1049
1050 while ((c=*str++) && len)
1051 switch(c) {
1052 case ' ':
1053 case '\t':
1054 case '\n':
1055 case '\\':
1056 if (len >= 4) {
1057 *bp++ = '\\';
1058 *bp++ = '0' + ((c & 0300)>>6);
1059 *bp++ = '0' + ((c & 0070)>>3);
1060 *bp++ = '0' + ((c & 0007)>>0);
1061 }
1062 len -= 4;
1063 break;
1064 default:
1065 *bp++ = c;
1066 len--;
1067 }
1068 if (c || len <1) len = -1;
1069 else {
1070 *bp++ = ' ';
1071 len--;
1072 }
1073 *bpp = bp;
1074 *lp = len;
1075}
1076EXPORT_SYMBOL_GPL(qword_add);
1077
1078void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1079{
1080 char *bp = *bpp;
1081 int len = *lp;
1082
1083 if (len < 0) return;
1084
1085 if (len > 2) {
1086 *bp++ = '\\';
1087 *bp++ = 'x';
1088 len -= 2;
1089 while (blen && len >= 2) {
1090 unsigned char c = *buf++;
1091 *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1092 *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1093 len -= 2;
1094 blen--;
1095 }
1096 }
1097 if (blen || len<1) len = -1;
1098 else {
1099 *bp++ = ' ';
1100 len--;
1101 }
1102 *bpp = bp;
1103 *lp = len;
1104}
1105EXPORT_SYMBOL_GPL(qword_addhex);
1106
1107static void warn_no_listener(struct cache_detail *detail)
1108{
1109 if (detail->last_warn != detail->last_close) {
1110 detail->last_warn = detail->last_close;
1111 if (detail->warn_no_listener)
1112 detail->warn_no_listener(detail, detail->last_close != 0);
1113 }
1114}
1115
1116static bool cache_listeners_exist(struct cache_detail *detail)
1117{
1118 if (atomic_read(&detail->readers))
1119 return true;
1120 if (detail->last_close == 0)
1121 /* This cache was never opened */
1122 return false;
1123 if (detail->last_close < seconds_since_boot() - 30)
1124 /*
1125 * We allow for the possibility that someone might
1126 * restart a userspace daemon without restarting the
1127 * server; but after 30 seconds, we give up.
1128 */
1129 return false;
1130 return true;
1131}
1132
1133/*
1134 * register an upcall request to user-space and queue it up for read() by the
1135 * upcall daemon.
1136 *
1137 * Each request is at most one page long.
1138 */
1139int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1140 void (*cache_request)(struct cache_detail *,
1141 struct cache_head *,
1142 char **,
1143 int *))
1144{
1145
1146 char *buf;
1147 struct cache_request *crq;
1148 char *bp;
1149 int len;
1150
1151 if (!cache_listeners_exist(detail)) {
1152 warn_no_listener(detail);
1153 return -EINVAL;
1154 }
1155
1156 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1157 if (!buf)
1158 return -EAGAIN;
1159
1160 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1161 if (!crq) {
1162 kfree(buf);
1163 return -EAGAIN;
1164 }
1165
1166 bp = buf; len = PAGE_SIZE;
1167
1168 cache_request(detail, h, &bp, &len);
1169
1170 if (len < 0) {
1171 kfree(buf);
1172 kfree(crq);
1173 return -EAGAIN;
1174 }
1175 crq->q.reader = 0;
1176 crq->item = cache_get(h);
1177 crq->buf = buf;
1178 crq->len = PAGE_SIZE - len;
1179 crq->readers = 0;
1180 spin_lock(&queue_lock);
1181 list_add_tail(&crq->q.list, &detail->queue);
1182 spin_unlock(&queue_lock);
1183 wake_up(&queue_wait);
1184 return 0;
1185}
1186EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1187
1188/*
1189 * parse a message from user-space and pass it
1190 * to an appropriate cache
1191 * Messages are, like requests, separated into fields by
1192 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1193 *
1194 * Message is
1195 * reply cachename expiry key ... content....
1196 *
1197 * key and content are both parsed by cache
1198 */
1199
1200#define isodigit(c) (isdigit(c) && c <= '7')
1201int qword_get(char **bpp, char *dest, int bufsize)
1202{
1203 /* return bytes copied, or -1 on error */
1204 char *bp = *bpp;
1205 int len = 0;
1206
1207 while (*bp == ' ') bp++;
1208
1209 if (bp[0] == '\\' && bp[1] == 'x') {
1210 /* HEX STRING */
1211 bp += 2;
1212 while (len < bufsize) {
1213 int h, l;
1214
1215 h = hex_to_bin(bp[0]);
1216 if (h < 0)
1217 break;
1218
1219 l = hex_to_bin(bp[1]);
1220 if (l < 0)
1221 break;
1222
1223 *dest++ = (h << 4) | l;
1224 bp += 2;
1225 len++;
1226 }
1227 } else {
1228 /* text with \nnn octal quoting */
1229 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1230 if (*bp == '\\' &&
1231 isodigit(bp[1]) && (bp[1] <= '3') &&
1232 isodigit(bp[2]) &&
1233 isodigit(bp[3])) {
1234 int byte = (*++bp -'0');
1235 bp++;
1236 byte = (byte << 3) | (*bp++ - '0');
1237 byte = (byte << 3) | (*bp++ - '0');
1238 *dest++ = byte;
1239 len++;
1240 } else {
1241 *dest++ = *bp++;
1242 len++;
1243 }
1244 }
1245 }
1246
1247 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1248 return -1;
1249 while (*bp == ' ') bp++;
1250 *bpp = bp;
1251 *dest = '\0';
1252 return len;
1253}
1254EXPORT_SYMBOL_GPL(qword_get);
1255
1256
1257/*
1258 * support /proc/sunrpc/cache/$CACHENAME/content
1259 * as a seqfile.
1260 * We call ->cache_show passing NULL for the item to
1261 * get a header, then pass each real item in the cache
1262 */
1263
1264struct handle {
1265 struct cache_detail *cd;
1266};
1267
1268static void *c_start(struct seq_file *m, loff_t *pos)
1269 __acquires(cd->hash_lock)
1270{
1271 loff_t n = *pos;
1272 unsigned hash, entry;
1273 struct cache_head *ch;
1274 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1275
1276
1277 read_lock(&cd->hash_lock);
1278 if (!n--)
1279 return SEQ_START_TOKEN;
1280 hash = n >> 32;
1281 entry = n & ((1LL<<32) - 1);
1282
1283 for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1284 if (!entry--)
1285 return ch;
1286 n &= ~((1LL<<32) - 1);
1287 do {
1288 hash++;
1289 n += 1LL<<32;
1290 } while(hash < cd->hash_size &&
1291 cd->hash_table[hash]==NULL);
1292 if (hash >= cd->hash_size)
1293 return NULL;
1294 *pos = n+1;
1295 return cd->hash_table[hash];
1296}
1297
1298static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1299{
1300 struct cache_head *ch = p;
1301 int hash = (*pos >> 32);
1302 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1303
1304 if (p == SEQ_START_TOKEN)
1305 hash = 0;
1306 else if (ch->next == NULL) {
1307 hash++;
1308 *pos += 1LL<<32;
1309 } else {
1310 ++*pos;
1311 return ch->next;
1312 }
1313 *pos &= ~((1LL<<32) - 1);
1314 while (hash < cd->hash_size &&
1315 cd->hash_table[hash] == NULL) {
1316 hash++;
1317 *pos += 1LL<<32;
1318 }
1319 if (hash >= cd->hash_size)
1320 return NULL;
1321 ++*pos;
1322 return cd->hash_table[hash];
1323}
1324
1325static void c_stop(struct seq_file *m, void *p)
1326 __releases(cd->hash_lock)
1327{
1328 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1329 read_unlock(&cd->hash_lock);
1330}
1331
1332static int c_show(struct seq_file *m, void *p)
1333{
1334 struct cache_head *cp = p;
1335 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1336
1337 if (p == SEQ_START_TOKEN)
1338 return cd->cache_show(m, cd, NULL);
1339
1340 ifdebug(CACHE)
1341 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1342 convert_to_wallclock(cp->expiry_time),
1343 atomic_read(&cp->ref.refcount), cp->flags);
1344 cache_get(cp);
1345 if (cache_check(cd, cp, NULL))
1346 /* cache_check does a cache_put on failure */
1347 seq_printf(m, "# ");
1348 else
1349 cache_put(cp, cd);
1350
1351 return cd->cache_show(m, cd, cp);
1352}
1353
1354static const struct seq_operations cache_content_op = {
1355 .start = c_start,
1356 .next = c_next,
1357 .stop = c_stop,
1358 .show = c_show,
1359};
1360
1361static int content_open(struct inode *inode, struct file *file,
1362 struct cache_detail *cd)
1363{
1364 struct handle *han;
1365
1366 if (!cd || !try_module_get(cd->owner))
1367 return -EACCES;
1368 han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1369 if (han == NULL) {
1370 module_put(cd->owner);
1371 return -ENOMEM;
1372 }
1373
1374 han->cd = cd;
1375 return 0;
1376}
1377
1378static int content_release(struct inode *inode, struct file *file,
1379 struct cache_detail *cd)
1380{
1381 int ret = seq_release_private(inode, file);
1382 module_put(cd->owner);
1383 return ret;
1384}
1385
1386static int open_flush(struct inode *inode, struct file *file,
1387 struct cache_detail *cd)
1388{
1389 if (!cd || !try_module_get(cd->owner))
1390 return -EACCES;
1391 return nonseekable_open(inode, file);
1392}
1393
1394static int release_flush(struct inode *inode, struct file *file,
1395 struct cache_detail *cd)
1396{
1397 module_put(cd->owner);
1398 return 0;
1399}
1400
1401static ssize_t read_flush(struct file *file, char __user *buf,
1402 size_t count, loff_t *ppos,
1403 struct cache_detail *cd)
1404{
1405 char tbuf[20];
1406 unsigned long p = *ppos;
1407 size_t len;
1408
1409 sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1410 len = strlen(tbuf);
1411 if (p >= len)
1412 return 0;
1413 len -= p;
1414 if (len > count)
1415 len = count;
1416 if (copy_to_user(buf, (void*)(tbuf+p), len))
1417 return -EFAULT;
1418 *ppos += len;
1419 return len;
1420}
1421
1422static ssize_t write_flush(struct file *file, const char __user *buf,
1423 size_t count, loff_t *ppos,
1424 struct cache_detail *cd)
1425{
1426 char tbuf[20];
1427 char *bp, *ep;
1428
1429 if (*ppos || count > sizeof(tbuf)-1)
1430 return -EINVAL;
1431 if (copy_from_user(tbuf, buf, count))
1432 return -EFAULT;
1433 tbuf[count] = 0;
1434 simple_strtoul(tbuf, &ep, 0);
1435 if (*ep && *ep != '\n')
1436 return -EINVAL;
1437
1438 bp = tbuf;
1439 cd->flush_time = get_expiry(&bp);
1440 cd->nextcheck = seconds_since_boot();
1441 cache_flush();
1442
1443 *ppos += count;
1444 return count;
1445}
1446
1447static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1448 size_t count, loff_t *ppos)
1449{
1450 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1451
1452 return cache_read(filp, buf, count, ppos, cd);
1453}
1454
1455static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1456 size_t count, loff_t *ppos)
1457{
1458 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1459
1460 return cache_write(filp, buf, count, ppos, cd);
1461}
1462
1463static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1464{
1465 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1466
1467 return cache_poll(filp, wait, cd);
1468}
1469
1470static long cache_ioctl_procfs(struct file *filp,
1471 unsigned int cmd, unsigned long arg)
1472{
1473 struct inode *inode = filp->f_path.dentry->d_inode;
1474 struct cache_detail *cd = PDE(inode)->data;
1475
1476 return cache_ioctl(inode, filp, cmd, arg, cd);
1477}
1478
1479static int cache_open_procfs(struct inode *inode, struct file *filp)
1480{
1481 struct cache_detail *cd = PDE(inode)->data;
1482
1483 return cache_open(inode, filp, cd);
1484}
1485
1486static int cache_release_procfs(struct inode *inode, struct file *filp)
1487{
1488 struct cache_detail *cd = PDE(inode)->data;
1489
1490 return cache_release(inode, filp, cd);
1491}
1492
1493static const struct file_operations cache_file_operations_procfs = {
1494 .owner = THIS_MODULE,
1495 .llseek = no_llseek,
1496 .read = cache_read_procfs,
1497 .write = cache_write_procfs,
1498 .poll = cache_poll_procfs,
1499 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1500 .open = cache_open_procfs,
1501 .release = cache_release_procfs,
1502};
1503
1504static int content_open_procfs(struct inode *inode, struct file *filp)
1505{
1506 struct cache_detail *cd = PDE(inode)->data;
1507
1508 return content_open(inode, filp, cd);
1509}
1510
1511static int content_release_procfs(struct inode *inode, struct file *filp)
1512{
1513 struct cache_detail *cd = PDE(inode)->data;
1514
1515 return content_release(inode, filp, cd);
1516}
1517
1518static const struct file_operations content_file_operations_procfs = {
1519 .open = content_open_procfs,
1520 .read = seq_read,
1521 .llseek = seq_lseek,
1522 .release = content_release_procfs,
1523};
1524
1525static int open_flush_procfs(struct inode *inode, struct file *filp)
1526{
1527 struct cache_detail *cd = PDE(inode)->data;
1528
1529 return open_flush(inode, filp, cd);
1530}
1531
1532static int release_flush_procfs(struct inode *inode, struct file *filp)
1533{
1534 struct cache_detail *cd = PDE(inode)->data;
1535
1536 return release_flush(inode, filp, cd);
1537}
1538
1539static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1540 size_t count, loff_t *ppos)
1541{
1542 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1543
1544 return read_flush(filp, buf, count, ppos, cd);
1545}
1546
1547static ssize_t write_flush_procfs(struct file *filp,
1548 const char __user *buf,
1549 size_t count, loff_t *ppos)
1550{
1551 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1552
1553 return write_flush(filp, buf, count, ppos, cd);
1554}
1555
1556static const struct file_operations cache_flush_operations_procfs = {
1557 .open = open_flush_procfs,
1558 .read = read_flush_procfs,
1559 .write = write_flush_procfs,
1560 .release = release_flush_procfs,
1561 .llseek = no_llseek,
1562};
1563
1564static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1565{
1566 struct sunrpc_net *sn;
1567
1568 if (cd->u.procfs.proc_ent == NULL)
1569 return;
1570 if (cd->u.procfs.flush_ent)
1571 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1572 if (cd->u.procfs.channel_ent)
1573 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1574 if (cd->u.procfs.content_ent)
1575 remove_proc_entry("content", cd->u.procfs.proc_ent);
1576 cd->u.procfs.proc_ent = NULL;
1577 sn = net_generic(net, sunrpc_net_id);
1578 remove_proc_entry(cd->name, sn->proc_net_rpc);
1579}
1580
1581#ifdef CONFIG_PROC_FS
1582static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1583{
1584 struct proc_dir_entry *p;
1585 struct sunrpc_net *sn;
1586
1587 sn = net_generic(net, sunrpc_net_id);
1588 cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1589 if (cd->u.procfs.proc_ent == NULL)
1590 goto out_nomem;
1591 cd->u.procfs.channel_ent = NULL;
1592 cd->u.procfs.content_ent = NULL;
1593
1594 p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1595 cd->u.procfs.proc_ent,
1596 &cache_flush_operations_procfs, cd);
1597 cd->u.procfs.flush_ent = p;
1598 if (p == NULL)
1599 goto out_nomem;
1600
1601 if (cd->cache_upcall || cd->cache_parse) {
1602 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1603 cd->u.procfs.proc_ent,
1604 &cache_file_operations_procfs, cd);
1605 cd->u.procfs.channel_ent = p;
1606 if (p == NULL)
1607 goto out_nomem;
1608 }
1609 if (cd->cache_show) {
1610 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1611 cd->u.procfs.proc_ent,
1612 &content_file_operations_procfs, cd);
1613 cd->u.procfs.content_ent = p;
1614 if (p == NULL)
1615 goto out_nomem;
1616 }
1617 return 0;
1618out_nomem:
1619 remove_cache_proc_entries(cd, net);
1620 return -ENOMEM;
1621}
1622#else /* CONFIG_PROC_FS */
1623static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1624{
1625 return 0;
1626}
1627#endif
1628
1629void __init cache_initialize(void)
1630{
1631 INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1632}
1633
1634int cache_register_net(struct cache_detail *cd, struct net *net)
1635{
1636 int ret;
1637
1638 sunrpc_init_cache_detail(cd);
1639 ret = create_cache_proc_entries(cd, net);
1640 if (ret)
1641 sunrpc_destroy_cache_detail(cd);
1642 return ret;
1643}
1644
1645int cache_register(struct cache_detail *cd)
1646{
1647 return cache_register_net(cd, &init_net);
1648}
1649EXPORT_SYMBOL_GPL(cache_register);
1650
1651void cache_unregister_net(struct cache_detail *cd, struct net *net)
1652{
1653 remove_cache_proc_entries(cd, net);
1654 sunrpc_destroy_cache_detail(cd);
1655}
1656
1657void cache_unregister(struct cache_detail *cd)
1658{
1659 cache_unregister_net(cd, &init_net);
1660}
1661EXPORT_SYMBOL_GPL(cache_unregister);
1662
1663static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1664 size_t count, loff_t *ppos)
1665{
1666 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1667
1668 return cache_read(filp, buf, count, ppos, cd);
1669}
1670
1671static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1672 size_t count, loff_t *ppos)
1673{
1674 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1675
1676 return cache_write(filp, buf, count, ppos, cd);
1677}
1678
1679static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1680{
1681 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1682
1683 return cache_poll(filp, wait, cd);
1684}
1685
1686static long cache_ioctl_pipefs(struct file *filp,
1687 unsigned int cmd, unsigned long arg)
1688{
1689 struct inode *inode = filp->f_dentry->d_inode;
1690 struct cache_detail *cd = RPC_I(inode)->private;
1691
1692 return cache_ioctl(inode, filp, cmd, arg, cd);
1693}
1694
1695static int cache_open_pipefs(struct inode *inode, struct file *filp)
1696{
1697 struct cache_detail *cd = RPC_I(inode)->private;
1698
1699 return cache_open(inode, filp, cd);
1700}
1701
1702static int cache_release_pipefs(struct inode *inode, struct file *filp)
1703{
1704 struct cache_detail *cd = RPC_I(inode)->private;
1705
1706 return cache_release(inode, filp, cd);
1707}
1708
1709const struct file_operations cache_file_operations_pipefs = {
1710 .owner = THIS_MODULE,
1711 .llseek = no_llseek,
1712 .read = cache_read_pipefs,
1713 .write = cache_write_pipefs,
1714 .poll = cache_poll_pipefs,
1715 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1716 .open = cache_open_pipefs,
1717 .release = cache_release_pipefs,
1718};
1719
1720static int content_open_pipefs(struct inode *inode, struct file *filp)
1721{
1722 struct cache_detail *cd = RPC_I(inode)->private;
1723
1724 return content_open(inode, filp, cd);
1725}
1726
1727static int content_release_pipefs(struct inode *inode, struct file *filp)
1728{
1729 struct cache_detail *cd = RPC_I(inode)->private;
1730
1731 return content_release(inode, filp, cd);
1732}
1733
1734const struct file_operations content_file_operations_pipefs = {
1735 .open = content_open_pipefs,
1736 .read = seq_read,
1737 .llseek = seq_lseek,
1738 .release = content_release_pipefs,
1739};
1740
1741static int open_flush_pipefs(struct inode *inode, struct file *filp)
1742{
1743 struct cache_detail *cd = RPC_I(inode)->private;
1744
1745 return open_flush(inode, filp, cd);
1746}
1747
1748static int release_flush_pipefs(struct inode *inode, struct file *filp)
1749{
1750 struct cache_detail *cd = RPC_I(inode)->private;
1751
1752 return release_flush(inode, filp, cd);
1753}
1754
1755static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1756 size_t count, loff_t *ppos)
1757{
1758 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1759
1760 return read_flush(filp, buf, count, ppos, cd);
1761}
1762
1763static ssize_t write_flush_pipefs(struct file *filp,
1764 const char __user *buf,
1765 size_t count, loff_t *ppos)
1766{
1767 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1768
1769 return write_flush(filp, buf, count, ppos, cd);
1770}
1771
1772const struct file_operations cache_flush_operations_pipefs = {
1773 .open = open_flush_pipefs,
1774 .read = read_flush_pipefs,
1775 .write = write_flush_pipefs,
1776 .release = release_flush_pipefs,
1777 .llseek = no_llseek,
1778};
1779
1780int sunrpc_cache_register_pipefs(struct dentry *parent,
1781 const char *name, mode_t umode,
1782 struct cache_detail *cd)
1783{
1784 struct qstr q;
1785 struct dentry *dir;
1786 int ret = 0;
1787
1788 sunrpc_init_cache_detail(cd);
1789 q.name = name;
1790 q.len = strlen(name);
1791 q.hash = full_name_hash(q.name, q.len);
1792 dir = rpc_create_cache_dir(parent, &q, umode, cd);
1793 if (!IS_ERR(dir))
1794 cd->u.pipefs.dir = dir;
1795 else {
1796 sunrpc_destroy_cache_detail(cd);
1797 ret = PTR_ERR(dir);
1798 }
1799 return ret;
1800}
1801EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1802
1803void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1804{
1805 rpc_remove_cache_dir(cd->u.pipefs.dir);
1806 cd->u.pipefs.dir = NULL;
1807 sunrpc_destroy_cache_detail(cd);
1808}
1809EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1810
1/*
2 * net/sunrpc/cache.c
3 *
4 * Generic code for various authentication-related caches
5 * used by sunrpc clients and servers.
6 *
7 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
8 *
9 * Released under terms in GPL version 2. See COPYING.
10 *
11 */
12
13#include <linux/types.h>
14#include <linux/fs.h>
15#include <linux/file.h>
16#include <linux/slab.h>
17#include <linux/signal.h>
18#include <linux/sched.h>
19#include <linux/kmod.h>
20#include <linux/list.h>
21#include <linux/module.h>
22#include <linux/ctype.h>
23#include <asm/uaccess.h>
24#include <linux/poll.h>
25#include <linux/seq_file.h>
26#include <linux/proc_fs.h>
27#include <linux/net.h>
28#include <linux/workqueue.h>
29#include <linux/mutex.h>
30#include <linux/pagemap.h>
31#include <asm/ioctls.h>
32#include <linux/sunrpc/types.h>
33#include <linux/sunrpc/cache.h>
34#include <linux/sunrpc/stats.h>
35#include <linux/sunrpc/rpc_pipe_fs.h>
36#include "netns.h"
37
38#define RPCDBG_FACILITY RPCDBG_CACHE
39
40static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
41static void cache_revisit_request(struct cache_head *item);
42
43static void cache_init(struct cache_head *h)
44{
45 time_t now = seconds_since_boot();
46 h->next = NULL;
47 h->flags = 0;
48 kref_init(&h->ref);
49 h->expiry_time = now + CACHE_NEW_EXPIRY;
50 h->last_refresh = now;
51}
52
53struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
54 struct cache_head *key, int hash)
55{
56 struct cache_head **head, **hp;
57 struct cache_head *new = NULL, *freeme = NULL;
58
59 head = &detail->hash_table[hash];
60
61 read_lock(&detail->hash_lock);
62
63 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
64 struct cache_head *tmp = *hp;
65 if (detail->match(tmp, key)) {
66 if (cache_is_expired(detail, tmp))
67 /* This entry is expired, we will discard it. */
68 break;
69 cache_get(tmp);
70 read_unlock(&detail->hash_lock);
71 return tmp;
72 }
73 }
74 read_unlock(&detail->hash_lock);
75 /* Didn't find anything, insert an empty entry */
76
77 new = detail->alloc();
78 if (!new)
79 return NULL;
80 /* must fully initialise 'new', else
81 * we might get lose if we need to
82 * cache_put it soon.
83 */
84 cache_init(new);
85 detail->init(new, key);
86
87 write_lock(&detail->hash_lock);
88
89 /* check if entry appeared while we slept */
90 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
91 struct cache_head *tmp = *hp;
92 if (detail->match(tmp, key)) {
93 if (cache_is_expired(detail, tmp)) {
94 *hp = tmp->next;
95 tmp->next = NULL;
96 detail->entries --;
97 freeme = tmp;
98 break;
99 }
100 cache_get(tmp);
101 write_unlock(&detail->hash_lock);
102 cache_put(new, detail);
103 return tmp;
104 }
105 }
106 new->next = *head;
107 *head = new;
108 detail->entries++;
109 cache_get(new);
110 write_unlock(&detail->hash_lock);
111
112 if (freeme)
113 cache_put(freeme, detail);
114 return new;
115}
116EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
117
118
119static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
120
121static void cache_fresh_locked(struct cache_head *head, time_t expiry)
122{
123 head->expiry_time = expiry;
124 head->last_refresh = seconds_since_boot();
125 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
126 set_bit(CACHE_VALID, &head->flags);
127}
128
129static void cache_fresh_unlocked(struct cache_head *head,
130 struct cache_detail *detail)
131{
132 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
133 cache_revisit_request(head);
134 cache_dequeue(detail, head);
135 }
136}
137
138struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
139 struct cache_head *new, struct cache_head *old, int hash)
140{
141 /* The 'old' entry is to be replaced by 'new'.
142 * If 'old' is not VALID, we update it directly,
143 * otherwise we need to replace it
144 */
145 struct cache_head **head;
146 struct cache_head *tmp;
147
148 if (!test_bit(CACHE_VALID, &old->flags)) {
149 write_lock(&detail->hash_lock);
150 if (!test_bit(CACHE_VALID, &old->flags)) {
151 if (test_bit(CACHE_NEGATIVE, &new->flags))
152 set_bit(CACHE_NEGATIVE, &old->flags);
153 else
154 detail->update(old, new);
155 cache_fresh_locked(old, new->expiry_time);
156 write_unlock(&detail->hash_lock);
157 cache_fresh_unlocked(old, detail);
158 return old;
159 }
160 write_unlock(&detail->hash_lock);
161 }
162 /* We need to insert a new entry */
163 tmp = detail->alloc();
164 if (!tmp) {
165 cache_put(old, detail);
166 return NULL;
167 }
168 cache_init(tmp);
169 detail->init(tmp, old);
170 head = &detail->hash_table[hash];
171
172 write_lock(&detail->hash_lock);
173 if (test_bit(CACHE_NEGATIVE, &new->flags))
174 set_bit(CACHE_NEGATIVE, &tmp->flags);
175 else
176 detail->update(tmp, new);
177 tmp->next = *head;
178 *head = tmp;
179 detail->entries++;
180 cache_get(tmp);
181 cache_fresh_locked(tmp, new->expiry_time);
182 cache_fresh_locked(old, 0);
183 write_unlock(&detail->hash_lock);
184 cache_fresh_unlocked(tmp, detail);
185 cache_fresh_unlocked(old, detail);
186 cache_put(old, detail);
187 return tmp;
188}
189EXPORT_SYMBOL_GPL(sunrpc_cache_update);
190
191static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
192{
193 if (cd->cache_upcall)
194 return cd->cache_upcall(cd, h);
195 return sunrpc_cache_pipe_upcall(cd, h);
196}
197
198static inline int cache_is_valid(struct cache_head *h)
199{
200 if (!test_bit(CACHE_VALID, &h->flags))
201 return -EAGAIN;
202 else {
203 /* entry is valid */
204 if (test_bit(CACHE_NEGATIVE, &h->flags))
205 return -ENOENT;
206 else {
207 /*
208 * In combination with write barrier in
209 * sunrpc_cache_update, ensures that anyone
210 * using the cache entry after this sees the
211 * updated contents:
212 */
213 smp_rmb();
214 return 0;
215 }
216 }
217}
218
219static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
220{
221 int rv;
222
223 write_lock(&detail->hash_lock);
224 rv = cache_is_valid(h);
225 if (rv == -EAGAIN) {
226 set_bit(CACHE_NEGATIVE, &h->flags);
227 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
228 rv = -ENOENT;
229 }
230 write_unlock(&detail->hash_lock);
231 cache_fresh_unlocked(h, detail);
232 return rv;
233}
234
235/*
236 * This is the generic cache management routine for all
237 * the authentication caches.
238 * It checks the currency of a cache item and will (later)
239 * initiate an upcall to fill it if needed.
240 *
241 *
242 * Returns 0 if the cache_head can be used, or cache_puts it and returns
243 * -EAGAIN if upcall is pending and request has been queued
244 * -ETIMEDOUT if upcall failed or request could not be queue or
245 * upcall completed but item is still invalid (implying that
246 * the cache item has been replaced with a newer one).
247 * -ENOENT if cache entry was negative
248 */
249int cache_check(struct cache_detail *detail,
250 struct cache_head *h, struct cache_req *rqstp)
251{
252 int rv;
253 long refresh_age, age;
254
255 /* First decide return status as best we can */
256 rv = cache_is_valid(h);
257
258 /* now see if we want to start an upcall */
259 refresh_age = (h->expiry_time - h->last_refresh);
260 age = seconds_since_boot() - h->last_refresh;
261
262 if (rqstp == NULL) {
263 if (rv == -EAGAIN)
264 rv = -ENOENT;
265 } else if (rv == -EAGAIN ||
266 (h->expiry_time != 0 && age > refresh_age/2)) {
267 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
268 refresh_age, age);
269 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
270 switch (cache_make_upcall(detail, h)) {
271 case -EINVAL:
272 rv = try_to_negate_entry(detail, h);
273 break;
274 case -EAGAIN:
275 cache_fresh_unlocked(h, detail);
276 break;
277 }
278 }
279 }
280
281 if (rv == -EAGAIN) {
282 if (!cache_defer_req(rqstp, h)) {
283 /*
284 * Request was not deferred; handle it as best
285 * we can ourselves:
286 */
287 rv = cache_is_valid(h);
288 if (rv == -EAGAIN)
289 rv = -ETIMEDOUT;
290 }
291 }
292 if (rv)
293 cache_put(h, detail);
294 return rv;
295}
296EXPORT_SYMBOL_GPL(cache_check);
297
298/*
299 * caches need to be periodically cleaned.
300 * For this we maintain a list of cache_detail and
301 * a current pointer into that list and into the table
302 * for that entry.
303 *
304 * Each time cache_clean is called it finds the next non-empty entry
305 * in the current table and walks the list in that entry
306 * looking for entries that can be removed.
307 *
308 * An entry gets removed if:
309 * - The expiry is before current time
310 * - The last_refresh time is before the flush_time for that cache
311 *
312 * later we might drop old entries with non-NEVER expiry if that table
313 * is getting 'full' for some definition of 'full'
314 *
315 * The question of "how often to scan a table" is an interesting one
316 * and is answered in part by the use of the "nextcheck" field in the
317 * cache_detail.
318 * When a scan of a table begins, the nextcheck field is set to a time
319 * that is well into the future.
320 * While scanning, if an expiry time is found that is earlier than the
321 * current nextcheck time, nextcheck is set to that expiry time.
322 * If the flush_time is ever set to a time earlier than the nextcheck
323 * time, the nextcheck time is then set to that flush_time.
324 *
325 * A table is then only scanned if the current time is at least
326 * the nextcheck time.
327 *
328 */
329
330static LIST_HEAD(cache_list);
331static DEFINE_SPINLOCK(cache_list_lock);
332static struct cache_detail *current_detail;
333static int current_index;
334
335static void do_cache_clean(struct work_struct *work);
336static struct delayed_work cache_cleaner;
337
338void sunrpc_init_cache_detail(struct cache_detail *cd)
339{
340 rwlock_init(&cd->hash_lock);
341 INIT_LIST_HEAD(&cd->queue);
342 spin_lock(&cache_list_lock);
343 cd->nextcheck = 0;
344 cd->entries = 0;
345 atomic_set(&cd->readers, 0);
346 cd->last_close = 0;
347 cd->last_warn = -1;
348 list_add(&cd->others, &cache_list);
349 spin_unlock(&cache_list_lock);
350
351 /* start the cleaning process */
352 schedule_delayed_work(&cache_cleaner, 0);
353}
354EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
355
356void sunrpc_destroy_cache_detail(struct cache_detail *cd)
357{
358 cache_purge(cd);
359 spin_lock(&cache_list_lock);
360 write_lock(&cd->hash_lock);
361 if (cd->entries || atomic_read(&cd->inuse)) {
362 write_unlock(&cd->hash_lock);
363 spin_unlock(&cache_list_lock);
364 goto out;
365 }
366 if (current_detail == cd)
367 current_detail = NULL;
368 list_del_init(&cd->others);
369 write_unlock(&cd->hash_lock);
370 spin_unlock(&cache_list_lock);
371 if (list_empty(&cache_list)) {
372 /* module must be being unloaded so its safe to kill the worker */
373 cancel_delayed_work_sync(&cache_cleaner);
374 }
375 return;
376out:
377 printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
378}
379EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
380
381/* clean cache tries to find something to clean
382 * and cleans it.
383 * It returns 1 if it cleaned something,
384 * 0 if it didn't find anything this time
385 * -1 if it fell off the end of the list.
386 */
387static int cache_clean(void)
388{
389 int rv = 0;
390 struct list_head *next;
391
392 spin_lock(&cache_list_lock);
393
394 /* find a suitable table if we don't already have one */
395 while (current_detail == NULL ||
396 current_index >= current_detail->hash_size) {
397 if (current_detail)
398 next = current_detail->others.next;
399 else
400 next = cache_list.next;
401 if (next == &cache_list) {
402 current_detail = NULL;
403 spin_unlock(&cache_list_lock);
404 return -1;
405 }
406 current_detail = list_entry(next, struct cache_detail, others);
407 if (current_detail->nextcheck > seconds_since_boot())
408 current_index = current_detail->hash_size;
409 else {
410 current_index = 0;
411 current_detail->nextcheck = seconds_since_boot()+30*60;
412 }
413 }
414
415 /* find a non-empty bucket in the table */
416 while (current_detail &&
417 current_index < current_detail->hash_size &&
418 current_detail->hash_table[current_index] == NULL)
419 current_index++;
420
421 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
422
423 if (current_detail && current_index < current_detail->hash_size) {
424 struct cache_head *ch, **cp;
425 struct cache_detail *d;
426
427 write_lock(¤t_detail->hash_lock);
428
429 /* Ok, now to clean this strand */
430
431 cp = & current_detail->hash_table[current_index];
432 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
433 if (current_detail->nextcheck > ch->expiry_time)
434 current_detail->nextcheck = ch->expiry_time+1;
435 if (!cache_is_expired(current_detail, ch))
436 continue;
437
438 *cp = ch->next;
439 ch->next = NULL;
440 current_detail->entries--;
441 rv = 1;
442 break;
443 }
444
445 write_unlock(¤t_detail->hash_lock);
446 d = current_detail;
447 if (!ch)
448 current_index ++;
449 spin_unlock(&cache_list_lock);
450 if (ch) {
451 set_bit(CACHE_CLEANED, &ch->flags);
452 cache_fresh_unlocked(ch, d);
453 cache_put(ch, d);
454 }
455 } else
456 spin_unlock(&cache_list_lock);
457
458 return rv;
459}
460
461/*
462 * We want to regularly clean the cache, so we need to schedule some work ...
463 */
464static void do_cache_clean(struct work_struct *work)
465{
466 int delay = 5;
467 if (cache_clean() == -1)
468 delay = round_jiffies_relative(30*HZ);
469
470 if (list_empty(&cache_list))
471 delay = 0;
472
473 if (delay)
474 schedule_delayed_work(&cache_cleaner, delay);
475}
476
477
478/*
479 * Clean all caches promptly. This just calls cache_clean
480 * repeatedly until we are sure that every cache has had a chance to
481 * be fully cleaned
482 */
483void cache_flush(void)
484{
485 while (cache_clean() != -1)
486 cond_resched();
487 while (cache_clean() != -1)
488 cond_resched();
489}
490EXPORT_SYMBOL_GPL(cache_flush);
491
492void cache_purge(struct cache_detail *detail)
493{
494 detail->flush_time = LONG_MAX;
495 detail->nextcheck = seconds_since_boot();
496 cache_flush();
497 detail->flush_time = 1;
498}
499EXPORT_SYMBOL_GPL(cache_purge);
500
501
502/*
503 * Deferral and Revisiting of Requests.
504 *
505 * If a cache lookup finds a pending entry, we
506 * need to defer the request and revisit it later.
507 * All deferred requests are stored in a hash table,
508 * indexed by "struct cache_head *".
509 * As it may be wasteful to store a whole request
510 * structure, we allow the request to provide a
511 * deferred form, which must contain a
512 * 'struct cache_deferred_req'
513 * This cache_deferred_req contains a method to allow
514 * it to be revisited when cache info is available
515 */
516
517#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
518#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
519
520#define DFR_MAX 300 /* ??? */
521
522static DEFINE_SPINLOCK(cache_defer_lock);
523static LIST_HEAD(cache_defer_list);
524static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
525static int cache_defer_cnt;
526
527static void __unhash_deferred_req(struct cache_deferred_req *dreq)
528{
529 hlist_del_init(&dreq->hash);
530 if (!list_empty(&dreq->recent)) {
531 list_del_init(&dreq->recent);
532 cache_defer_cnt--;
533 }
534}
535
536static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
537{
538 int hash = DFR_HASH(item);
539
540 INIT_LIST_HEAD(&dreq->recent);
541 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
542}
543
544static void setup_deferral(struct cache_deferred_req *dreq,
545 struct cache_head *item,
546 int count_me)
547{
548
549 dreq->item = item;
550
551 spin_lock(&cache_defer_lock);
552
553 __hash_deferred_req(dreq, item);
554
555 if (count_me) {
556 cache_defer_cnt++;
557 list_add(&dreq->recent, &cache_defer_list);
558 }
559
560 spin_unlock(&cache_defer_lock);
561
562}
563
564struct thread_deferred_req {
565 struct cache_deferred_req handle;
566 struct completion completion;
567};
568
569static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
570{
571 struct thread_deferred_req *dr =
572 container_of(dreq, struct thread_deferred_req, handle);
573 complete(&dr->completion);
574}
575
576static void cache_wait_req(struct cache_req *req, struct cache_head *item)
577{
578 struct thread_deferred_req sleeper;
579 struct cache_deferred_req *dreq = &sleeper.handle;
580
581 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
582 dreq->revisit = cache_restart_thread;
583
584 setup_deferral(dreq, item, 0);
585
586 if (!test_bit(CACHE_PENDING, &item->flags) ||
587 wait_for_completion_interruptible_timeout(
588 &sleeper.completion, req->thread_wait) <= 0) {
589 /* The completion wasn't completed, so we need
590 * to clean up
591 */
592 spin_lock(&cache_defer_lock);
593 if (!hlist_unhashed(&sleeper.handle.hash)) {
594 __unhash_deferred_req(&sleeper.handle);
595 spin_unlock(&cache_defer_lock);
596 } else {
597 /* cache_revisit_request already removed
598 * this from the hash table, but hasn't
599 * called ->revisit yet. It will very soon
600 * and we need to wait for it.
601 */
602 spin_unlock(&cache_defer_lock);
603 wait_for_completion(&sleeper.completion);
604 }
605 }
606}
607
608static void cache_limit_defers(void)
609{
610 /* Make sure we haven't exceed the limit of allowed deferred
611 * requests.
612 */
613 struct cache_deferred_req *discard = NULL;
614
615 if (cache_defer_cnt <= DFR_MAX)
616 return;
617
618 spin_lock(&cache_defer_lock);
619
620 /* Consider removing either the first or the last */
621 if (cache_defer_cnt > DFR_MAX) {
622 if (prandom_u32() & 1)
623 discard = list_entry(cache_defer_list.next,
624 struct cache_deferred_req, recent);
625 else
626 discard = list_entry(cache_defer_list.prev,
627 struct cache_deferred_req, recent);
628 __unhash_deferred_req(discard);
629 }
630 spin_unlock(&cache_defer_lock);
631 if (discard)
632 discard->revisit(discard, 1);
633}
634
635/* Return true if and only if a deferred request is queued. */
636static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
637{
638 struct cache_deferred_req *dreq;
639
640 if (req->thread_wait) {
641 cache_wait_req(req, item);
642 if (!test_bit(CACHE_PENDING, &item->flags))
643 return false;
644 }
645 dreq = req->defer(req);
646 if (dreq == NULL)
647 return false;
648 setup_deferral(dreq, item, 1);
649 if (!test_bit(CACHE_PENDING, &item->flags))
650 /* Bit could have been cleared before we managed to
651 * set up the deferral, so need to revisit just in case
652 */
653 cache_revisit_request(item);
654
655 cache_limit_defers();
656 return true;
657}
658
659static void cache_revisit_request(struct cache_head *item)
660{
661 struct cache_deferred_req *dreq;
662 struct list_head pending;
663 struct hlist_node *tmp;
664 int hash = DFR_HASH(item);
665
666 INIT_LIST_HEAD(&pending);
667 spin_lock(&cache_defer_lock);
668
669 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
670 if (dreq->item == item) {
671 __unhash_deferred_req(dreq);
672 list_add(&dreq->recent, &pending);
673 }
674
675 spin_unlock(&cache_defer_lock);
676
677 while (!list_empty(&pending)) {
678 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
679 list_del_init(&dreq->recent);
680 dreq->revisit(dreq, 0);
681 }
682}
683
684void cache_clean_deferred(void *owner)
685{
686 struct cache_deferred_req *dreq, *tmp;
687 struct list_head pending;
688
689
690 INIT_LIST_HEAD(&pending);
691 spin_lock(&cache_defer_lock);
692
693 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
694 if (dreq->owner == owner) {
695 __unhash_deferred_req(dreq);
696 list_add(&dreq->recent, &pending);
697 }
698 }
699 spin_unlock(&cache_defer_lock);
700
701 while (!list_empty(&pending)) {
702 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
703 list_del_init(&dreq->recent);
704 dreq->revisit(dreq, 1);
705 }
706}
707
708/*
709 * communicate with user-space
710 *
711 * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
712 * On read, you get a full request, or block.
713 * On write, an update request is processed.
714 * Poll works if anything to read, and always allows write.
715 *
716 * Implemented by linked list of requests. Each open file has
717 * a ->private that also exists in this list. New requests are added
718 * to the end and may wakeup and preceding readers.
719 * New readers are added to the head. If, on read, an item is found with
720 * CACHE_UPCALLING clear, we free it from the list.
721 *
722 */
723
724static DEFINE_SPINLOCK(queue_lock);
725static DEFINE_MUTEX(queue_io_mutex);
726
727struct cache_queue {
728 struct list_head list;
729 int reader; /* if 0, then request */
730};
731struct cache_request {
732 struct cache_queue q;
733 struct cache_head *item;
734 char * buf;
735 int len;
736 int readers;
737};
738struct cache_reader {
739 struct cache_queue q;
740 int offset; /* if non-0, we have a refcnt on next request */
741};
742
743static int cache_request(struct cache_detail *detail,
744 struct cache_request *crq)
745{
746 char *bp = crq->buf;
747 int len = PAGE_SIZE;
748
749 detail->cache_request(detail, crq->item, &bp, &len);
750 if (len < 0)
751 return -EAGAIN;
752 return PAGE_SIZE - len;
753}
754
755static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
756 loff_t *ppos, struct cache_detail *cd)
757{
758 struct cache_reader *rp = filp->private_data;
759 struct cache_request *rq;
760 struct inode *inode = file_inode(filp);
761 int err;
762
763 if (count == 0)
764 return 0;
765
766 mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
767 * readers on this file */
768 again:
769 spin_lock(&queue_lock);
770 /* need to find next request */
771 while (rp->q.list.next != &cd->queue &&
772 list_entry(rp->q.list.next, struct cache_queue, list)
773 ->reader) {
774 struct list_head *next = rp->q.list.next;
775 list_move(&rp->q.list, next);
776 }
777 if (rp->q.list.next == &cd->queue) {
778 spin_unlock(&queue_lock);
779 mutex_unlock(&inode->i_mutex);
780 WARN_ON_ONCE(rp->offset);
781 return 0;
782 }
783 rq = container_of(rp->q.list.next, struct cache_request, q.list);
784 WARN_ON_ONCE(rq->q.reader);
785 if (rp->offset == 0)
786 rq->readers++;
787 spin_unlock(&queue_lock);
788
789 if (rq->len == 0) {
790 err = cache_request(cd, rq);
791 if (err < 0)
792 goto out;
793 rq->len = err;
794 }
795
796 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
797 err = -EAGAIN;
798 spin_lock(&queue_lock);
799 list_move(&rp->q.list, &rq->q.list);
800 spin_unlock(&queue_lock);
801 } else {
802 if (rp->offset + count > rq->len)
803 count = rq->len - rp->offset;
804 err = -EFAULT;
805 if (copy_to_user(buf, rq->buf + rp->offset, count))
806 goto out;
807 rp->offset += count;
808 if (rp->offset >= rq->len) {
809 rp->offset = 0;
810 spin_lock(&queue_lock);
811 list_move(&rp->q.list, &rq->q.list);
812 spin_unlock(&queue_lock);
813 }
814 err = 0;
815 }
816 out:
817 if (rp->offset == 0) {
818 /* need to release rq */
819 spin_lock(&queue_lock);
820 rq->readers--;
821 if (rq->readers == 0 &&
822 !test_bit(CACHE_PENDING, &rq->item->flags)) {
823 list_del(&rq->q.list);
824 spin_unlock(&queue_lock);
825 cache_put(rq->item, cd);
826 kfree(rq->buf);
827 kfree(rq);
828 } else
829 spin_unlock(&queue_lock);
830 }
831 if (err == -EAGAIN)
832 goto again;
833 mutex_unlock(&inode->i_mutex);
834 return err ? err : count;
835}
836
837static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
838 size_t count, struct cache_detail *cd)
839{
840 ssize_t ret;
841
842 if (count == 0)
843 return -EINVAL;
844 if (copy_from_user(kaddr, buf, count))
845 return -EFAULT;
846 kaddr[count] = '\0';
847 ret = cd->cache_parse(cd, kaddr, count);
848 if (!ret)
849 ret = count;
850 return ret;
851}
852
853static ssize_t cache_slow_downcall(const char __user *buf,
854 size_t count, struct cache_detail *cd)
855{
856 static char write_buf[8192]; /* protected by queue_io_mutex */
857 ssize_t ret = -EINVAL;
858
859 if (count >= sizeof(write_buf))
860 goto out;
861 mutex_lock(&queue_io_mutex);
862 ret = cache_do_downcall(write_buf, buf, count, cd);
863 mutex_unlock(&queue_io_mutex);
864out:
865 return ret;
866}
867
868static ssize_t cache_downcall(struct address_space *mapping,
869 const char __user *buf,
870 size_t count, struct cache_detail *cd)
871{
872 struct page *page;
873 char *kaddr;
874 ssize_t ret = -ENOMEM;
875
876 if (count >= PAGE_CACHE_SIZE)
877 goto out_slow;
878
879 page = find_or_create_page(mapping, 0, GFP_KERNEL);
880 if (!page)
881 goto out_slow;
882
883 kaddr = kmap(page);
884 ret = cache_do_downcall(kaddr, buf, count, cd);
885 kunmap(page);
886 unlock_page(page);
887 page_cache_release(page);
888 return ret;
889out_slow:
890 return cache_slow_downcall(buf, count, cd);
891}
892
893static ssize_t cache_write(struct file *filp, const char __user *buf,
894 size_t count, loff_t *ppos,
895 struct cache_detail *cd)
896{
897 struct address_space *mapping = filp->f_mapping;
898 struct inode *inode = file_inode(filp);
899 ssize_t ret = -EINVAL;
900
901 if (!cd->cache_parse)
902 goto out;
903
904 mutex_lock(&inode->i_mutex);
905 ret = cache_downcall(mapping, buf, count, cd);
906 mutex_unlock(&inode->i_mutex);
907out:
908 return ret;
909}
910
911static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
912
913static unsigned int cache_poll(struct file *filp, poll_table *wait,
914 struct cache_detail *cd)
915{
916 unsigned int mask;
917 struct cache_reader *rp = filp->private_data;
918 struct cache_queue *cq;
919
920 poll_wait(filp, &queue_wait, wait);
921
922 /* alway allow write */
923 mask = POLL_OUT | POLLWRNORM;
924
925 if (!rp)
926 return mask;
927
928 spin_lock(&queue_lock);
929
930 for (cq= &rp->q; &cq->list != &cd->queue;
931 cq = list_entry(cq->list.next, struct cache_queue, list))
932 if (!cq->reader) {
933 mask |= POLLIN | POLLRDNORM;
934 break;
935 }
936 spin_unlock(&queue_lock);
937 return mask;
938}
939
940static int cache_ioctl(struct inode *ino, struct file *filp,
941 unsigned int cmd, unsigned long arg,
942 struct cache_detail *cd)
943{
944 int len = 0;
945 struct cache_reader *rp = filp->private_data;
946 struct cache_queue *cq;
947
948 if (cmd != FIONREAD || !rp)
949 return -EINVAL;
950
951 spin_lock(&queue_lock);
952
953 /* only find the length remaining in current request,
954 * or the length of the next request
955 */
956 for (cq= &rp->q; &cq->list != &cd->queue;
957 cq = list_entry(cq->list.next, struct cache_queue, list))
958 if (!cq->reader) {
959 struct cache_request *cr =
960 container_of(cq, struct cache_request, q);
961 len = cr->len - rp->offset;
962 break;
963 }
964 spin_unlock(&queue_lock);
965
966 return put_user(len, (int __user *)arg);
967}
968
969static int cache_open(struct inode *inode, struct file *filp,
970 struct cache_detail *cd)
971{
972 struct cache_reader *rp = NULL;
973
974 if (!cd || !try_module_get(cd->owner))
975 return -EACCES;
976 nonseekable_open(inode, filp);
977 if (filp->f_mode & FMODE_READ) {
978 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
979 if (!rp) {
980 module_put(cd->owner);
981 return -ENOMEM;
982 }
983 rp->offset = 0;
984 rp->q.reader = 1;
985 atomic_inc(&cd->readers);
986 spin_lock(&queue_lock);
987 list_add(&rp->q.list, &cd->queue);
988 spin_unlock(&queue_lock);
989 }
990 filp->private_data = rp;
991 return 0;
992}
993
994static int cache_release(struct inode *inode, struct file *filp,
995 struct cache_detail *cd)
996{
997 struct cache_reader *rp = filp->private_data;
998
999 if (rp) {
1000 spin_lock(&queue_lock);
1001 if (rp->offset) {
1002 struct cache_queue *cq;
1003 for (cq= &rp->q; &cq->list != &cd->queue;
1004 cq = list_entry(cq->list.next, struct cache_queue, list))
1005 if (!cq->reader) {
1006 container_of(cq, struct cache_request, q)
1007 ->readers--;
1008 break;
1009 }
1010 rp->offset = 0;
1011 }
1012 list_del(&rp->q.list);
1013 spin_unlock(&queue_lock);
1014
1015 filp->private_data = NULL;
1016 kfree(rp);
1017
1018 cd->last_close = seconds_since_boot();
1019 atomic_dec(&cd->readers);
1020 }
1021 module_put(cd->owner);
1022 return 0;
1023}
1024
1025
1026
1027static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1028{
1029 struct cache_queue *cq, *tmp;
1030 struct cache_request *cr;
1031 struct list_head dequeued;
1032
1033 INIT_LIST_HEAD(&dequeued);
1034 spin_lock(&queue_lock);
1035 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1036 if (!cq->reader) {
1037 cr = container_of(cq, struct cache_request, q);
1038 if (cr->item != ch)
1039 continue;
1040 if (test_bit(CACHE_PENDING, &ch->flags))
1041 /* Lost a race and it is pending again */
1042 break;
1043 if (cr->readers != 0)
1044 continue;
1045 list_move(&cr->q.list, &dequeued);
1046 }
1047 spin_unlock(&queue_lock);
1048 while (!list_empty(&dequeued)) {
1049 cr = list_entry(dequeued.next, struct cache_request, q.list);
1050 list_del(&cr->q.list);
1051 cache_put(cr->item, detail);
1052 kfree(cr->buf);
1053 kfree(cr);
1054 }
1055}
1056
1057/*
1058 * Support routines for text-based upcalls.
1059 * Fields are separated by spaces.
1060 * Fields are either mangled to quote space tab newline slosh with slosh
1061 * or a hexified with a leading \x
1062 * Record is terminated with newline.
1063 *
1064 */
1065
1066void qword_add(char **bpp, int *lp, char *str)
1067{
1068 char *bp = *bpp;
1069 int len = *lp;
1070 char c;
1071
1072 if (len < 0) return;
1073
1074 while ((c=*str++) && len)
1075 switch(c) {
1076 case ' ':
1077 case '\t':
1078 case '\n':
1079 case '\\':
1080 if (len >= 4) {
1081 *bp++ = '\\';
1082 *bp++ = '0' + ((c & 0300)>>6);
1083 *bp++ = '0' + ((c & 0070)>>3);
1084 *bp++ = '0' + ((c & 0007)>>0);
1085 }
1086 len -= 4;
1087 break;
1088 default:
1089 *bp++ = c;
1090 len--;
1091 }
1092 if (c || len <1) len = -1;
1093 else {
1094 *bp++ = ' ';
1095 len--;
1096 }
1097 *bpp = bp;
1098 *lp = len;
1099}
1100EXPORT_SYMBOL_GPL(qword_add);
1101
1102void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1103{
1104 char *bp = *bpp;
1105 int len = *lp;
1106
1107 if (len < 0) return;
1108
1109 if (len > 2) {
1110 *bp++ = '\\';
1111 *bp++ = 'x';
1112 len -= 2;
1113 while (blen && len >= 2) {
1114 bp = hex_byte_pack(bp, *buf++);
1115 len -= 2;
1116 blen--;
1117 }
1118 }
1119 if (blen || len<1) len = -1;
1120 else {
1121 *bp++ = ' ';
1122 len--;
1123 }
1124 *bpp = bp;
1125 *lp = len;
1126}
1127EXPORT_SYMBOL_GPL(qword_addhex);
1128
1129static void warn_no_listener(struct cache_detail *detail)
1130{
1131 if (detail->last_warn != detail->last_close) {
1132 detail->last_warn = detail->last_close;
1133 if (detail->warn_no_listener)
1134 detail->warn_no_listener(detail, detail->last_close != 0);
1135 }
1136}
1137
1138static bool cache_listeners_exist(struct cache_detail *detail)
1139{
1140 if (atomic_read(&detail->readers))
1141 return true;
1142 if (detail->last_close == 0)
1143 /* This cache was never opened */
1144 return false;
1145 if (detail->last_close < seconds_since_boot() - 30)
1146 /*
1147 * We allow for the possibility that someone might
1148 * restart a userspace daemon without restarting the
1149 * server; but after 30 seconds, we give up.
1150 */
1151 return false;
1152 return true;
1153}
1154
1155/*
1156 * register an upcall request to user-space and queue it up for read() by the
1157 * upcall daemon.
1158 *
1159 * Each request is at most one page long.
1160 */
1161int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1162{
1163
1164 char *buf;
1165 struct cache_request *crq;
1166 int ret = 0;
1167
1168 if (!detail->cache_request)
1169 return -EINVAL;
1170
1171 if (!cache_listeners_exist(detail)) {
1172 warn_no_listener(detail);
1173 return -EINVAL;
1174 }
1175 if (test_bit(CACHE_CLEANED, &h->flags))
1176 /* Too late to make an upcall */
1177 return -EAGAIN;
1178
1179 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1180 if (!buf)
1181 return -EAGAIN;
1182
1183 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1184 if (!crq) {
1185 kfree(buf);
1186 return -EAGAIN;
1187 }
1188
1189 crq->q.reader = 0;
1190 crq->item = cache_get(h);
1191 crq->buf = buf;
1192 crq->len = 0;
1193 crq->readers = 0;
1194 spin_lock(&queue_lock);
1195 if (test_bit(CACHE_PENDING, &h->flags))
1196 list_add_tail(&crq->q.list, &detail->queue);
1197 else
1198 /* Lost a race, no longer PENDING, so don't enqueue */
1199 ret = -EAGAIN;
1200 spin_unlock(&queue_lock);
1201 wake_up(&queue_wait);
1202 if (ret == -EAGAIN) {
1203 kfree(buf);
1204 kfree(crq);
1205 }
1206 return ret;
1207}
1208EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1209
1210/*
1211 * parse a message from user-space and pass it
1212 * to an appropriate cache
1213 * Messages are, like requests, separated into fields by
1214 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1215 *
1216 * Message is
1217 * reply cachename expiry key ... content....
1218 *
1219 * key and content are both parsed by cache
1220 */
1221
1222int qword_get(char **bpp, char *dest, int bufsize)
1223{
1224 /* return bytes copied, or -1 on error */
1225 char *bp = *bpp;
1226 int len = 0;
1227
1228 while (*bp == ' ') bp++;
1229
1230 if (bp[0] == '\\' && bp[1] == 'x') {
1231 /* HEX STRING */
1232 bp += 2;
1233 while (len < bufsize) {
1234 int h, l;
1235
1236 h = hex_to_bin(bp[0]);
1237 if (h < 0)
1238 break;
1239
1240 l = hex_to_bin(bp[1]);
1241 if (l < 0)
1242 break;
1243
1244 *dest++ = (h << 4) | l;
1245 bp += 2;
1246 len++;
1247 }
1248 } else {
1249 /* text with \nnn octal quoting */
1250 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1251 if (*bp == '\\' &&
1252 isodigit(bp[1]) && (bp[1] <= '3') &&
1253 isodigit(bp[2]) &&
1254 isodigit(bp[3])) {
1255 int byte = (*++bp -'0');
1256 bp++;
1257 byte = (byte << 3) | (*bp++ - '0');
1258 byte = (byte << 3) | (*bp++ - '0');
1259 *dest++ = byte;
1260 len++;
1261 } else {
1262 *dest++ = *bp++;
1263 len++;
1264 }
1265 }
1266 }
1267
1268 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1269 return -1;
1270 while (*bp == ' ') bp++;
1271 *bpp = bp;
1272 *dest = '\0';
1273 return len;
1274}
1275EXPORT_SYMBOL_GPL(qword_get);
1276
1277
1278/*
1279 * support /proc/sunrpc/cache/$CACHENAME/content
1280 * as a seqfile.
1281 * We call ->cache_show passing NULL for the item to
1282 * get a header, then pass each real item in the cache
1283 */
1284
1285struct handle {
1286 struct cache_detail *cd;
1287};
1288
1289static void *c_start(struct seq_file *m, loff_t *pos)
1290 __acquires(cd->hash_lock)
1291{
1292 loff_t n = *pos;
1293 unsigned int hash, entry;
1294 struct cache_head *ch;
1295 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1296
1297
1298 read_lock(&cd->hash_lock);
1299 if (!n--)
1300 return SEQ_START_TOKEN;
1301 hash = n >> 32;
1302 entry = n & ((1LL<<32) - 1);
1303
1304 for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1305 if (!entry--)
1306 return ch;
1307 n &= ~((1LL<<32) - 1);
1308 do {
1309 hash++;
1310 n += 1LL<<32;
1311 } while(hash < cd->hash_size &&
1312 cd->hash_table[hash]==NULL);
1313 if (hash >= cd->hash_size)
1314 return NULL;
1315 *pos = n+1;
1316 return cd->hash_table[hash];
1317}
1318
1319static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1320{
1321 struct cache_head *ch = p;
1322 int hash = (*pos >> 32);
1323 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1324
1325 if (p == SEQ_START_TOKEN)
1326 hash = 0;
1327 else if (ch->next == NULL) {
1328 hash++;
1329 *pos += 1LL<<32;
1330 } else {
1331 ++*pos;
1332 return ch->next;
1333 }
1334 *pos &= ~((1LL<<32) - 1);
1335 while (hash < cd->hash_size &&
1336 cd->hash_table[hash] == NULL) {
1337 hash++;
1338 *pos += 1LL<<32;
1339 }
1340 if (hash >= cd->hash_size)
1341 return NULL;
1342 ++*pos;
1343 return cd->hash_table[hash];
1344}
1345
1346static void c_stop(struct seq_file *m, void *p)
1347 __releases(cd->hash_lock)
1348{
1349 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1350 read_unlock(&cd->hash_lock);
1351}
1352
1353static int c_show(struct seq_file *m, void *p)
1354{
1355 struct cache_head *cp = p;
1356 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1357
1358 if (p == SEQ_START_TOKEN)
1359 return cd->cache_show(m, cd, NULL);
1360
1361 ifdebug(CACHE)
1362 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1363 convert_to_wallclock(cp->expiry_time),
1364 atomic_read(&cp->ref.refcount), cp->flags);
1365 cache_get(cp);
1366 if (cache_check(cd, cp, NULL))
1367 /* cache_check does a cache_put on failure */
1368 seq_printf(m, "# ");
1369 else {
1370 if (cache_is_expired(cd, cp))
1371 seq_printf(m, "# ");
1372 cache_put(cp, cd);
1373 }
1374
1375 return cd->cache_show(m, cd, cp);
1376}
1377
1378static const struct seq_operations cache_content_op = {
1379 .start = c_start,
1380 .next = c_next,
1381 .stop = c_stop,
1382 .show = c_show,
1383};
1384
1385static int content_open(struct inode *inode, struct file *file,
1386 struct cache_detail *cd)
1387{
1388 struct handle *han;
1389
1390 if (!cd || !try_module_get(cd->owner))
1391 return -EACCES;
1392 han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1393 if (han == NULL) {
1394 module_put(cd->owner);
1395 return -ENOMEM;
1396 }
1397
1398 han->cd = cd;
1399 return 0;
1400}
1401
1402static int content_release(struct inode *inode, struct file *file,
1403 struct cache_detail *cd)
1404{
1405 int ret = seq_release_private(inode, file);
1406 module_put(cd->owner);
1407 return ret;
1408}
1409
1410static int open_flush(struct inode *inode, struct file *file,
1411 struct cache_detail *cd)
1412{
1413 if (!cd || !try_module_get(cd->owner))
1414 return -EACCES;
1415 return nonseekable_open(inode, file);
1416}
1417
1418static int release_flush(struct inode *inode, struct file *file,
1419 struct cache_detail *cd)
1420{
1421 module_put(cd->owner);
1422 return 0;
1423}
1424
1425static ssize_t read_flush(struct file *file, char __user *buf,
1426 size_t count, loff_t *ppos,
1427 struct cache_detail *cd)
1428{
1429 char tbuf[22];
1430 unsigned long p = *ppos;
1431 size_t len;
1432
1433 snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1434 len = strlen(tbuf);
1435 if (p >= len)
1436 return 0;
1437 len -= p;
1438 if (len > count)
1439 len = count;
1440 if (copy_to_user(buf, (void*)(tbuf+p), len))
1441 return -EFAULT;
1442 *ppos += len;
1443 return len;
1444}
1445
1446static ssize_t write_flush(struct file *file, const char __user *buf,
1447 size_t count, loff_t *ppos,
1448 struct cache_detail *cd)
1449{
1450 char tbuf[20];
1451 char *bp, *ep;
1452
1453 if (*ppos || count > sizeof(tbuf)-1)
1454 return -EINVAL;
1455 if (copy_from_user(tbuf, buf, count))
1456 return -EFAULT;
1457 tbuf[count] = 0;
1458 simple_strtoul(tbuf, &ep, 0);
1459 if (*ep && *ep != '\n')
1460 return -EINVAL;
1461
1462 bp = tbuf;
1463 cd->flush_time = get_expiry(&bp);
1464 cd->nextcheck = seconds_since_boot();
1465 cache_flush();
1466
1467 *ppos += count;
1468 return count;
1469}
1470
1471static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1472 size_t count, loff_t *ppos)
1473{
1474 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1475
1476 return cache_read(filp, buf, count, ppos, cd);
1477}
1478
1479static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1480 size_t count, loff_t *ppos)
1481{
1482 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1483
1484 return cache_write(filp, buf, count, ppos, cd);
1485}
1486
1487static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1488{
1489 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1490
1491 return cache_poll(filp, wait, cd);
1492}
1493
1494static long cache_ioctl_procfs(struct file *filp,
1495 unsigned int cmd, unsigned long arg)
1496{
1497 struct inode *inode = file_inode(filp);
1498 struct cache_detail *cd = PDE_DATA(inode);
1499
1500 return cache_ioctl(inode, filp, cmd, arg, cd);
1501}
1502
1503static int cache_open_procfs(struct inode *inode, struct file *filp)
1504{
1505 struct cache_detail *cd = PDE_DATA(inode);
1506
1507 return cache_open(inode, filp, cd);
1508}
1509
1510static int cache_release_procfs(struct inode *inode, struct file *filp)
1511{
1512 struct cache_detail *cd = PDE_DATA(inode);
1513
1514 return cache_release(inode, filp, cd);
1515}
1516
1517static const struct file_operations cache_file_operations_procfs = {
1518 .owner = THIS_MODULE,
1519 .llseek = no_llseek,
1520 .read = cache_read_procfs,
1521 .write = cache_write_procfs,
1522 .poll = cache_poll_procfs,
1523 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1524 .open = cache_open_procfs,
1525 .release = cache_release_procfs,
1526};
1527
1528static int content_open_procfs(struct inode *inode, struct file *filp)
1529{
1530 struct cache_detail *cd = PDE_DATA(inode);
1531
1532 return content_open(inode, filp, cd);
1533}
1534
1535static int content_release_procfs(struct inode *inode, struct file *filp)
1536{
1537 struct cache_detail *cd = PDE_DATA(inode);
1538
1539 return content_release(inode, filp, cd);
1540}
1541
1542static const struct file_operations content_file_operations_procfs = {
1543 .open = content_open_procfs,
1544 .read = seq_read,
1545 .llseek = seq_lseek,
1546 .release = content_release_procfs,
1547};
1548
1549static int open_flush_procfs(struct inode *inode, struct file *filp)
1550{
1551 struct cache_detail *cd = PDE_DATA(inode);
1552
1553 return open_flush(inode, filp, cd);
1554}
1555
1556static int release_flush_procfs(struct inode *inode, struct file *filp)
1557{
1558 struct cache_detail *cd = PDE_DATA(inode);
1559
1560 return release_flush(inode, filp, cd);
1561}
1562
1563static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1564 size_t count, loff_t *ppos)
1565{
1566 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1567
1568 return read_flush(filp, buf, count, ppos, cd);
1569}
1570
1571static ssize_t write_flush_procfs(struct file *filp,
1572 const char __user *buf,
1573 size_t count, loff_t *ppos)
1574{
1575 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1576
1577 return write_flush(filp, buf, count, ppos, cd);
1578}
1579
1580static const struct file_operations cache_flush_operations_procfs = {
1581 .open = open_flush_procfs,
1582 .read = read_flush_procfs,
1583 .write = write_flush_procfs,
1584 .release = release_flush_procfs,
1585 .llseek = no_llseek,
1586};
1587
1588static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1589{
1590 struct sunrpc_net *sn;
1591
1592 if (cd->u.procfs.proc_ent == NULL)
1593 return;
1594 if (cd->u.procfs.flush_ent)
1595 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1596 if (cd->u.procfs.channel_ent)
1597 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1598 if (cd->u.procfs.content_ent)
1599 remove_proc_entry("content", cd->u.procfs.proc_ent);
1600 cd->u.procfs.proc_ent = NULL;
1601 sn = net_generic(net, sunrpc_net_id);
1602 remove_proc_entry(cd->name, sn->proc_net_rpc);
1603}
1604
1605#ifdef CONFIG_PROC_FS
1606static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1607{
1608 struct proc_dir_entry *p;
1609 struct sunrpc_net *sn;
1610
1611 sn = net_generic(net, sunrpc_net_id);
1612 cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1613 if (cd->u.procfs.proc_ent == NULL)
1614 goto out_nomem;
1615 cd->u.procfs.channel_ent = NULL;
1616 cd->u.procfs.content_ent = NULL;
1617
1618 p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1619 cd->u.procfs.proc_ent,
1620 &cache_flush_operations_procfs, cd);
1621 cd->u.procfs.flush_ent = p;
1622 if (p == NULL)
1623 goto out_nomem;
1624
1625 if (cd->cache_request || cd->cache_parse) {
1626 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1627 cd->u.procfs.proc_ent,
1628 &cache_file_operations_procfs, cd);
1629 cd->u.procfs.channel_ent = p;
1630 if (p == NULL)
1631 goto out_nomem;
1632 }
1633 if (cd->cache_show) {
1634 p = proc_create_data("content", S_IFREG|S_IRUSR,
1635 cd->u.procfs.proc_ent,
1636 &content_file_operations_procfs, cd);
1637 cd->u.procfs.content_ent = p;
1638 if (p == NULL)
1639 goto out_nomem;
1640 }
1641 return 0;
1642out_nomem:
1643 remove_cache_proc_entries(cd, net);
1644 return -ENOMEM;
1645}
1646#else /* CONFIG_PROC_FS */
1647static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1648{
1649 return 0;
1650}
1651#endif
1652
1653void __init cache_initialize(void)
1654{
1655 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1656}
1657
1658int cache_register_net(struct cache_detail *cd, struct net *net)
1659{
1660 int ret;
1661
1662 sunrpc_init_cache_detail(cd);
1663 ret = create_cache_proc_entries(cd, net);
1664 if (ret)
1665 sunrpc_destroy_cache_detail(cd);
1666 return ret;
1667}
1668EXPORT_SYMBOL_GPL(cache_register_net);
1669
1670void cache_unregister_net(struct cache_detail *cd, struct net *net)
1671{
1672 remove_cache_proc_entries(cd, net);
1673 sunrpc_destroy_cache_detail(cd);
1674}
1675EXPORT_SYMBOL_GPL(cache_unregister_net);
1676
1677struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1678{
1679 struct cache_detail *cd;
1680
1681 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1682 if (cd == NULL)
1683 return ERR_PTR(-ENOMEM);
1684
1685 cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1686 GFP_KERNEL);
1687 if (cd->hash_table == NULL) {
1688 kfree(cd);
1689 return ERR_PTR(-ENOMEM);
1690 }
1691 cd->net = net;
1692 return cd;
1693}
1694EXPORT_SYMBOL_GPL(cache_create_net);
1695
1696void cache_destroy_net(struct cache_detail *cd, struct net *net)
1697{
1698 kfree(cd->hash_table);
1699 kfree(cd);
1700}
1701EXPORT_SYMBOL_GPL(cache_destroy_net);
1702
1703static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1704 size_t count, loff_t *ppos)
1705{
1706 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1707
1708 return cache_read(filp, buf, count, ppos, cd);
1709}
1710
1711static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1712 size_t count, loff_t *ppos)
1713{
1714 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1715
1716 return cache_write(filp, buf, count, ppos, cd);
1717}
1718
1719static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1720{
1721 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1722
1723 return cache_poll(filp, wait, cd);
1724}
1725
1726static long cache_ioctl_pipefs(struct file *filp,
1727 unsigned int cmd, unsigned long arg)
1728{
1729 struct inode *inode = file_inode(filp);
1730 struct cache_detail *cd = RPC_I(inode)->private;
1731
1732 return cache_ioctl(inode, filp, cmd, arg, cd);
1733}
1734
1735static int cache_open_pipefs(struct inode *inode, struct file *filp)
1736{
1737 struct cache_detail *cd = RPC_I(inode)->private;
1738
1739 return cache_open(inode, filp, cd);
1740}
1741
1742static int cache_release_pipefs(struct inode *inode, struct file *filp)
1743{
1744 struct cache_detail *cd = RPC_I(inode)->private;
1745
1746 return cache_release(inode, filp, cd);
1747}
1748
1749const struct file_operations cache_file_operations_pipefs = {
1750 .owner = THIS_MODULE,
1751 .llseek = no_llseek,
1752 .read = cache_read_pipefs,
1753 .write = cache_write_pipefs,
1754 .poll = cache_poll_pipefs,
1755 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1756 .open = cache_open_pipefs,
1757 .release = cache_release_pipefs,
1758};
1759
1760static int content_open_pipefs(struct inode *inode, struct file *filp)
1761{
1762 struct cache_detail *cd = RPC_I(inode)->private;
1763
1764 return content_open(inode, filp, cd);
1765}
1766
1767static int content_release_pipefs(struct inode *inode, struct file *filp)
1768{
1769 struct cache_detail *cd = RPC_I(inode)->private;
1770
1771 return content_release(inode, filp, cd);
1772}
1773
1774const struct file_operations content_file_operations_pipefs = {
1775 .open = content_open_pipefs,
1776 .read = seq_read,
1777 .llseek = seq_lseek,
1778 .release = content_release_pipefs,
1779};
1780
1781static int open_flush_pipefs(struct inode *inode, struct file *filp)
1782{
1783 struct cache_detail *cd = RPC_I(inode)->private;
1784
1785 return open_flush(inode, filp, cd);
1786}
1787
1788static int release_flush_pipefs(struct inode *inode, struct file *filp)
1789{
1790 struct cache_detail *cd = RPC_I(inode)->private;
1791
1792 return release_flush(inode, filp, cd);
1793}
1794
1795static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1796 size_t count, loff_t *ppos)
1797{
1798 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1799
1800 return read_flush(filp, buf, count, ppos, cd);
1801}
1802
1803static ssize_t write_flush_pipefs(struct file *filp,
1804 const char __user *buf,
1805 size_t count, loff_t *ppos)
1806{
1807 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1808
1809 return write_flush(filp, buf, count, ppos, cd);
1810}
1811
1812const struct file_operations cache_flush_operations_pipefs = {
1813 .open = open_flush_pipefs,
1814 .read = read_flush_pipefs,
1815 .write = write_flush_pipefs,
1816 .release = release_flush_pipefs,
1817 .llseek = no_llseek,
1818};
1819
1820int sunrpc_cache_register_pipefs(struct dentry *parent,
1821 const char *name, umode_t umode,
1822 struct cache_detail *cd)
1823{
1824 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1825 if (IS_ERR(dir))
1826 return PTR_ERR(dir);
1827 cd->u.pipefs.dir = dir;
1828 return 0;
1829}
1830EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1831
1832void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1833{
1834 rpc_remove_cache_dir(cd->u.pipefs.dir);
1835 cd->u.pipefs.dir = NULL;
1836}
1837EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1838