Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/sunrpc/cache.c
4 *
5 * Generic code for various authentication-related caches
6 * used by sunrpc clients and servers.
7 *
8 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
9 */
10
11#include <linux/types.h>
12#include <linux/fs.h>
13#include <linux/file.h>
14#include <linux/slab.h>
15#include <linux/signal.h>
16#include <linux/sched.h>
17#include <linux/kmod.h>
18#include <linux/list.h>
19#include <linux/module.h>
20#include <linux/ctype.h>
21#include <linux/string_helpers.h>
22#include <linux/uaccess.h>
23#include <linux/poll.h>
24#include <linux/seq_file.h>
25#include <linux/proc_fs.h>
26#include <linux/net.h>
27#include <linux/workqueue.h>
28#include <linux/mutex.h>
29#include <linux/pagemap.h>
30#include <asm/ioctls.h>
31#include <linux/sunrpc/types.h>
32#include <linux/sunrpc/cache.h>
33#include <linux/sunrpc/stats.h>
34#include <linux/sunrpc/rpc_pipe_fs.h>
35#include "netns.h"
36
37#define RPCDBG_FACILITY RPCDBG_CACHE
38
39static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
40static void cache_revisit_request(struct cache_head *item);
41static bool cache_listeners_exist(struct cache_detail *detail);
42
43static void cache_init(struct cache_head *h, struct cache_detail *detail)
44{
45 time_t now = seconds_since_boot();
46 INIT_HLIST_NODE(&h->cache_list);
47 h->flags = 0;
48 kref_init(&h->ref);
49 h->expiry_time = now + CACHE_NEW_EXPIRY;
50 if (now <= detail->flush_time)
51 /* ensure it isn't already expired */
52 now = detail->flush_time + 1;
53 h->last_refresh = now;
54}
55
56static inline int cache_is_valid(struct cache_head *h);
57static void cache_fresh_locked(struct cache_head *head, time_t expiry,
58 struct cache_detail *detail);
59static void cache_fresh_unlocked(struct cache_head *head,
60 struct cache_detail *detail);
61
62static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
63 struct cache_head *key,
64 int hash)
65{
66 struct hlist_head *head = &detail->hash_table[hash];
67 struct cache_head *tmp;
68
69 rcu_read_lock();
70 hlist_for_each_entry_rcu(tmp, head, cache_list) {
71 if (detail->match(tmp, key)) {
72 if (cache_is_expired(detail, tmp))
73 continue;
74 tmp = cache_get_rcu(tmp);
75 rcu_read_unlock();
76 return tmp;
77 }
78 }
79 rcu_read_unlock();
80 return NULL;
81}
82
83static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
84 struct cache_head *key,
85 int hash)
86{
87 struct cache_head *new, *tmp, *freeme = NULL;
88 struct hlist_head *head = &detail->hash_table[hash];
89
90 new = detail->alloc();
91 if (!new)
92 return NULL;
93 /* must fully initialise 'new', else
94 * we might get lose if we need to
95 * cache_put it soon.
96 */
97 cache_init(new, detail);
98 detail->init(new, key);
99
100 spin_lock(&detail->hash_lock);
101
102 /* check if entry appeared while we slept */
103 hlist_for_each_entry_rcu(tmp, head, cache_list) {
104 if (detail->match(tmp, key)) {
105 if (cache_is_expired(detail, tmp)) {
106 hlist_del_init_rcu(&tmp->cache_list);
107 detail->entries --;
108 if (cache_is_valid(tmp) == -EAGAIN)
109 set_bit(CACHE_NEGATIVE, &tmp->flags);
110 cache_fresh_locked(tmp, 0, detail);
111 freeme = tmp;
112 break;
113 }
114 cache_get(tmp);
115 spin_unlock(&detail->hash_lock);
116 cache_put(new, detail);
117 return tmp;
118 }
119 }
120
121 hlist_add_head_rcu(&new->cache_list, head);
122 detail->entries++;
123 cache_get(new);
124 spin_unlock(&detail->hash_lock);
125
126 if (freeme) {
127 cache_fresh_unlocked(freeme, detail);
128 cache_put(freeme, detail);
129 }
130 return new;
131}
132
133struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
134 struct cache_head *key, int hash)
135{
136 struct cache_head *ret;
137
138 ret = sunrpc_cache_find_rcu(detail, key, hash);
139 if (ret)
140 return ret;
141 /* Didn't find anything, insert an empty entry */
142 return sunrpc_cache_add_entry(detail, key, hash);
143}
144EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
145
146static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
147
148static void cache_fresh_locked(struct cache_head *head, time_t expiry,
149 struct cache_detail *detail)
150{
151 time_t now = seconds_since_boot();
152 if (now <= detail->flush_time)
153 /* ensure it isn't immediately treated as expired */
154 now = detail->flush_time + 1;
155 head->expiry_time = expiry;
156 head->last_refresh = now;
157 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
158 set_bit(CACHE_VALID, &head->flags);
159}
160
161static void cache_fresh_unlocked(struct cache_head *head,
162 struct cache_detail *detail)
163{
164 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
165 cache_revisit_request(head);
166 cache_dequeue(detail, head);
167 }
168}
169
170struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
171 struct cache_head *new, struct cache_head *old, int hash)
172{
173 /* The 'old' entry is to be replaced by 'new'.
174 * If 'old' is not VALID, we update it directly,
175 * otherwise we need to replace it
176 */
177 struct cache_head *tmp;
178
179 if (!test_bit(CACHE_VALID, &old->flags)) {
180 spin_lock(&detail->hash_lock);
181 if (!test_bit(CACHE_VALID, &old->flags)) {
182 if (test_bit(CACHE_NEGATIVE, &new->flags))
183 set_bit(CACHE_NEGATIVE, &old->flags);
184 else
185 detail->update(old, new);
186 cache_fresh_locked(old, new->expiry_time, detail);
187 spin_unlock(&detail->hash_lock);
188 cache_fresh_unlocked(old, detail);
189 return old;
190 }
191 spin_unlock(&detail->hash_lock);
192 }
193 /* We need to insert a new entry */
194 tmp = detail->alloc();
195 if (!tmp) {
196 cache_put(old, detail);
197 return NULL;
198 }
199 cache_init(tmp, detail);
200 detail->init(tmp, old);
201
202 spin_lock(&detail->hash_lock);
203 if (test_bit(CACHE_NEGATIVE, &new->flags))
204 set_bit(CACHE_NEGATIVE, &tmp->flags);
205 else
206 detail->update(tmp, new);
207 hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
208 detail->entries++;
209 cache_get(tmp);
210 cache_fresh_locked(tmp, new->expiry_time, detail);
211 cache_fresh_locked(old, 0, detail);
212 spin_unlock(&detail->hash_lock);
213 cache_fresh_unlocked(tmp, detail);
214 cache_fresh_unlocked(old, detail);
215 cache_put(old, detail);
216 return tmp;
217}
218EXPORT_SYMBOL_GPL(sunrpc_cache_update);
219
220static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
221{
222 if (cd->cache_upcall)
223 return cd->cache_upcall(cd, h);
224 return sunrpc_cache_pipe_upcall(cd, h);
225}
226
227static inline int cache_is_valid(struct cache_head *h)
228{
229 if (!test_bit(CACHE_VALID, &h->flags))
230 return -EAGAIN;
231 else {
232 /* entry is valid */
233 if (test_bit(CACHE_NEGATIVE, &h->flags))
234 return -ENOENT;
235 else {
236 /*
237 * In combination with write barrier in
238 * sunrpc_cache_update, ensures that anyone
239 * using the cache entry after this sees the
240 * updated contents:
241 */
242 smp_rmb();
243 return 0;
244 }
245 }
246}
247
248static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
249{
250 int rv;
251
252 spin_lock(&detail->hash_lock);
253 rv = cache_is_valid(h);
254 if (rv == -EAGAIN) {
255 set_bit(CACHE_NEGATIVE, &h->flags);
256 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
257 detail);
258 rv = -ENOENT;
259 }
260 spin_unlock(&detail->hash_lock);
261 cache_fresh_unlocked(h, detail);
262 return rv;
263}
264
265/*
266 * This is the generic cache management routine for all
267 * the authentication caches.
268 * It checks the currency of a cache item and will (later)
269 * initiate an upcall to fill it if needed.
270 *
271 *
272 * Returns 0 if the cache_head can be used, or cache_puts it and returns
273 * -EAGAIN if upcall is pending and request has been queued
274 * -ETIMEDOUT if upcall failed or request could not be queue or
275 * upcall completed but item is still invalid (implying that
276 * the cache item has been replaced with a newer one).
277 * -ENOENT if cache entry was negative
278 */
279int cache_check(struct cache_detail *detail,
280 struct cache_head *h, struct cache_req *rqstp)
281{
282 int rv;
283 long refresh_age, age;
284
285 /* First decide return status as best we can */
286 rv = cache_is_valid(h);
287
288 /* now see if we want to start an upcall */
289 refresh_age = (h->expiry_time - h->last_refresh);
290 age = seconds_since_boot() - h->last_refresh;
291
292 if (rqstp == NULL) {
293 if (rv == -EAGAIN)
294 rv = -ENOENT;
295 } else if (rv == -EAGAIN ||
296 (h->expiry_time != 0 && age > refresh_age/2)) {
297 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
298 refresh_age, age);
299 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
300 switch (cache_make_upcall(detail, h)) {
301 case -EINVAL:
302 rv = try_to_negate_entry(detail, h);
303 break;
304 case -EAGAIN:
305 cache_fresh_unlocked(h, detail);
306 break;
307 }
308 } else if (!cache_listeners_exist(detail))
309 rv = try_to_negate_entry(detail, h);
310 }
311
312 if (rv == -EAGAIN) {
313 if (!cache_defer_req(rqstp, h)) {
314 /*
315 * Request was not deferred; handle it as best
316 * we can ourselves:
317 */
318 rv = cache_is_valid(h);
319 if (rv == -EAGAIN)
320 rv = -ETIMEDOUT;
321 }
322 }
323 if (rv)
324 cache_put(h, detail);
325 return rv;
326}
327EXPORT_SYMBOL_GPL(cache_check);
328
329/*
330 * caches need to be periodically cleaned.
331 * For this we maintain a list of cache_detail and
332 * a current pointer into that list and into the table
333 * for that entry.
334 *
335 * Each time cache_clean is called it finds the next non-empty entry
336 * in the current table and walks the list in that entry
337 * looking for entries that can be removed.
338 *
339 * An entry gets removed if:
340 * - The expiry is before current time
341 * - The last_refresh time is before the flush_time for that cache
342 *
343 * later we might drop old entries with non-NEVER expiry if that table
344 * is getting 'full' for some definition of 'full'
345 *
346 * The question of "how often to scan a table" is an interesting one
347 * and is answered in part by the use of the "nextcheck" field in the
348 * cache_detail.
349 * When a scan of a table begins, the nextcheck field is set to a time
350 * that is well into the future.
351 * While scanning, if an expiry time is found that is earlier than the
352 * current nextcheck time, nextcheck is set to that expiry time.
353 * If the flush_time is ever set to a time earlier than the nextcheck
354 * time, the nextcheck time is then set to that flush_time.
355 *
356 * A table is then only scanned if the current time is at least
357 * the nextcheck time.
358 *
359 */
360
361static LIST_HEAD(cache_list);
362static DEFINE_SPINLOCK(cache_list_lock);
363static struct cache_detail *current_detail;
364static int current_index;
365
366static void do_cache_clean(struct work_struct *work);
367static struct delayed_work cache_cleaner;
368
369void sunrpc_init_cache_detail(struct cache_detail *cd)
370{
371 spin_lock_init(&cd->hash_lock);
372 INIT_LIST_HEAD(&cd->queue);
373 spin_lock(&cache_list_lock);
374 cd->nextcheck = 0;
375 cd->entries = 0;
376 atomic_set(&cd->writers, 0);
377 cd->last_close = 0;
378 cd->last_warn = -1;
379 list_add(&cd->others, &cache_list);
380 spin_unlock(&cache_list_lock);
381
382 /* start the cleaning process */
383 queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
384}
385EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
386
387void sunrpc_destroy_cache_detail(struct cache_detail *cd)
388{
389 cache_purge(cd);
390 spin_lock(&cache_list_lock);
391 spin_lock(&cd->hash_lock);
392 if (current_detail == cd)
393 current_detail = NULL;
394 list_del_init(&cd->others);
395 spin_unlock(&cd->hash_lock);
396 spin_unlock(&cache_list_lock);
397 if (list_empty(&cache_list)) {
398 /* module must be being unloaded so its safe to kill the worker */
399 cancel_delayed_work_sync(&cache_cleaner);
400 }
401}
402EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
403
404/* clean cache tries to find something to clean
405 * and cleans it.
406 * It returns 1 if it cleaned something,
407 * 0 if it didn't find anything this time
408 * -1 if it fell off the end of the list.
409 */
410static int cache_clean(void)
411{
412 int rv = 0;
413 struct list_head *next;
414
415 spin_lock(&cache_list_lock);
416
417 /* find a suitable table if we don't already have one */
418 while (current_detail == NULL ||
419 current_index >= current_detail->hash_size) {
420 if (current_detail)
421 next = current_detail->others.next;
422 else
423 next = cache_list.next;
424 if (next == &cache_list) {
425 current_detail = NULL;
426 spin_unlock(&cache_list_lock);
427 return -1;
428 }
429 current_detail = list_entry(next, struct cache_detail, others);
430 if (current_detail->nextcheck > seconds_since_boot())
431 current_index = current_detail->hash_size;
432 else {
433 current_index = 0;
434 current_detail->nextcheck = seconds_since_boot()+30*60;
435 }
436 }
437
438 /* find a non-empty bucket in the table */
439 while (current_detail &&
440 current_index < current_detail->hash_size &&
441 hlist_empty(¤t_detail->hash_table[current_index]))
442 current_index++;
443
444 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
445
446 if (current_detail && current_index < current_detail->hash_size) {
447 struct cache_head *ch = NULL;
448 struct cache_detail *d;
449 struct hlist_head *head;
450 struct hlist_node *tmp;
451
452 spin_lock(¤t_detail->hash_lock);
453
454 /* Ok, now to clean this strand */
455
456 head = ¤t_detail->hash_table[current_index];
457 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
458 if (current_detail->nextcheck > ch->expiry_time)
459 current_detail->nextcheck = ch->expiry_time+1;
460 if (!cache_is_expired(current_detail, ch))
461 continue;
462
463 hlist_del_init_rcu(&ch->cache_list);
464 current_detail->entries--;
465 rv = 1;
466 break;
467 }
468
469 spin_unlock(¤t_detail->hash_lock);
470 d = current_detail;
471 if (!ch)
472 current_index ++;
473 spin_unlock(&cache_list_lock);
474 if (ch) {
475 set_bit(CACHE_CLEANED, &ch->flags);
476 cache_fresh_unlocked(ch, d);
477 cache_put(ch, d);
478 }
479 } else
480 spin_unlock(&cache_list_lock);
481
482 return rv;
483}
484
485/*
486 * We want to regularly clean the cache, so we need to schedule some work ...
487 */
488static void do_cache_clean(struct work_struct *work)
489{
490 int delay = 5;
491 if (cache_clean() == -1)
492 delay = round_jiffies_relative(30*HZ);
493
494 if (list_empty(&cache_list))
495 delay = 0;
496
497 if (delay)
498 queue_delayed_work(system_power_efficient_wq,
499 &cache_cleaner, delay);
500}
501
502
503/*
504 * Clean all caches promptly. This just calls cache_clean
505 * repeatedly until we are sure that every cache has had a chance to
506 * be fully cleaned
507 */
508void cache_flush(void)
509{
510 while (cache_clean() != -1)
511 cond_resched();
512 while (cache_clean() != -1)
513 cond_resched();
514}
515EXPORT_SYMBOL_GPL(cache_flush);
516
517void cache_purge(struct cache_detail *detail)
518{
519 struct cache_head *ch = NULL;
520 struct hlist_head *head = NULL;
521 struct hlist_node *tmp = NULL;
522 int i = 0;
523
524 spin_lock(&detail->hash_lock);
525 if (!detail->entries) {
526 spin_unlock(&detail->hash_lock);
527 return;
528 }
529
530 dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
531 for (i = 0; i < detail->hash_size; i++) {
532 head = &detail->hash_table[i];
533 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
534 hlist_del_init_rcu(&ch->cache_list);
535 detail->entries--;
536
537 set_bit(CACHE_CLEANED, &ch->flags);
538 spin_unlock(&detail->hash_lock);
539 cache_fresh_unlocked(ch, detail);
540 cache_put(ch, detail);
541 spin_lock(&detail->hash_lock);
542 }
543 }
544 spin_unlock(&detail->hash_lock);
545}
546EXPORT_SYMBOL_GPL(cache_purge);
547
548
549/*
550 * Deferral and Revisiting of Requests.
551 *
552 * If a cache lookup finds a pending entry, we
553 * need to defer the request and revisit it later.
554 * All deferred requests are stored in a hash table,
555 * indexed by "struct cache_head *".
556 * As it may be wasteful to store a whole request
557 * structure, we allow the request to provide a
558 * deferred form, which must contain a
559 * 'struct cache_deferred_req'
560 * This cache_deferred_req contains a method to allow
561 * it to be revisited when cache info is available
562 */
563
564#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
565#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
566
567#define DFR_MAX 300 /* ??? */
568
569static DEFINE_SPINLOCK(cache_defer_lock);
570static LIST_HEAD(cache_defer_list);
571static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
572static int cache_defer_cnt;
573
574static void __unhash_deferred_req(struct cache_deferred_req *dreq)
575{
576 hlist_del_init(&dreq->hash);
577 if (!list_empty(&dreq->recent)) {
578 list_del_init(&dreq->recent);
579 cache_defer_cnt--;
580 }
581}
582
583static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
584{
585 int hash = DFR_HASH(item);
586
587 INIT_LIST_HEAD(&dreq->recent);
588 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
589}
590
591static void setup_deferral(struct cache_deferred_req *dreq,
592 struct cache_head *item,
593 int count_me)
594{
595
596 dreq->item = item;
597
598 spin_lock(&cache_defer_lock);
599
600 __hash_deferred_req(dreq, item);
601
602 if (count_me) {
603 cache_defer_cnt++;
604 list_add(&dreq->recent, &cache_defer_list);
605 }
606
607 spin_unlock(&cache_defer_lock);
608
609}
610
611struct thread_deferred_req {
612 struct cache_deferred_req handle;
613 struct completion completion;
614};
615
616static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
617{
618 struct thread_deferred_req *dr =
619 container_of(dreq, struct thread_deferred_req, handle);
620 complete(&dr->completion);
621}
622
623static void cache_wait_req(struct cache_req *req, struct cache_head *item)
624{
625 struct thread_deferred_req sleeper;
626 struct cache_deferred_req *dreq = &sleeper.handle;
627
628 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
629 dreq->revisit = cache_restart_thread;
630
631 setup_deferral(dreq, item, 0);
632
633 if (!test_bit(CACHE_PENDING, &item->flags) ||
634 wait_for_completion_interruptible_timeout(
635 &sleeper.completion, req->thread_wait) <= 0) {
636 /* The completion wasn't completed, so we need
637 * to clean up
638 */
639 spin_lock(&cache_defer_lock);
640 if (!hlist_unhashed(&sleeper.handle.hash)) {
641 __unhash_deferred_req(&sleeper.handle);
642 spin_unlock(&cache_defer_lock);
643 } else {
644 /* cache_revisit_request already removed
645 * this from the hash table, but hasn't
646 * called ->revisit yet. It will very soon
647 * and we need to wait for it.
648 */
649 spin_unlock(&cache_defer_lock);
650 wait_for_completion(&sleeper.completion);
651 }
652 }
653}
654
655static void cache_limit_defers(void)
656{
657 /* Make sure we haven't exceed the limit of allowed deferred
658 * requests.
659 */
660 struct cache_deferred_req *discard = NULL;
661
662 if (cache_defer_cnt <= DFR_MAX)
663 return;
664
665 spin_lock(&cache_defer_lock);
666
667 /* Consider removing either the first or the last */
668 if (cache_defer_cnt > DFR_MAX) {
669 if (prandom_u32() & 1)
670 discard = list_entry(cache_defer_list.next,
671 struct cache_deferred_req, recent);
672 else
673 discard = list_entry(cache_defer_list.prev,
674 struct cache_deferred_req, recent);
675 __unhash_deferred_req(discard);
676 }
677 spin_unlock(&cache_defer_lock);
678 if (discard)
679 discard->revisit(discard, 1);
680}
681
682/* Return true if and only if a deferred request is queued. */
683static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
684{
685 struct cache_deferred_req *dreq;
686
687 if (req->thread_wait) {
688 cache_wait_req(req, item);
689 if (!test_bit(CACHE_PENDING, &item->flags))
690 return false;
691 }
692 dreq = req->defer(req);
693 if (dreq == NULL)
694 return false;
695 setup_deferral(dreq, item, 1);
696 if (!test_bit(CACHE_PENDING, &item->flags))
697 /* Bit could have been cleared before we managed to
698 * set up the deferral, so need to revisit just in case
699 */
700 cache_revisit_request(item);
701
702 cache_limit_defers();
703 return true;
704}
705
706static void cache_revisit_request(struct cache_head *item)
707{
708 struct cache_deferred_req *dreq;
709 struct list_head pending;
710 struct hlist_node *tmp;
711 int hash = DFR_HASH(item);
712
713 INIT_LIST_HEAD(&pending);
714 spin_lock(&cache_defer_lock);
715
716 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
717 if (dreq->item == item) {
718 __unhash_deferred_req(dreq);
719 list_add(&dreq->recent, &pending);
720 }
721
722 spin_unlock(&cache_defer_lock);
723
724 while (!list_empty(&pending)) {
725 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
726 list_del_init(&dreq->recent);
727 dreq->revisit(dreq, 0);
728 }
729}
730
731void cache_clean_deferred(void *owner)
732{
733 struct cache_deferred_req *dreq, *tmp;
734 struct list_head pending;
735
736
737 INIT_LIST_HEAD(&pending);
738 spin_lock(&cache_defer_lock);
739
740 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
741 if (dreq->owner == owner) {
742 __unhash_deferred_req(dreq);
743 list_add(&dreq->recent, &pending);
744 }
745 }
746 spin_unlock(&cache_defer_lock);
747
748 while (!list_empty(&pending)) {
749 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
750 list_del_init(&dreq->recent);
751 dreq->revisit(dreq, 1);
752 }
753}
754
755/*
756 * communicate with user-space
757 *
758 * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
759 * On read, you get a full request, or block.
760 * On write, an update request is processed.
761 * Poll works if anything to read, and always allows write.
762 *
763 * Implemented by linked list of requests. Each open file has
764 * a ->private that also exists in this list. New requests are added
765 * to the end and may wakeup and preceding readers.
766 * New readers are added to the head. If, on read, an item is found with
767 * CACHE_UPCALLING clear, we free it from the list.
768 *
769 */
770
771static DEFINE_SPINLOCK(queue_lock);
772static DEFINE_MUTEX(queue_io_mutex);
773
774struct cache_queue {
775 struct list_head list;
776 int reader; /* if 0, then request */
777};
778struct cache_request {
779 struct cache_queue q;
780 struct cache_head *item;
781 char * buf;
782 int len;
783 int readers;
784};
785struct cache_reader {
786 struct cache_queue q;
787 int offset; /* if non-0, we have a refcnt on next request */
788};
789
790static int cache_request(struct cache_detail *detail,
791 struct cache_request *crq)
792{
793 char *bp = crq->buf;
794 int len = PAGE_SIZE;
795
796 detail->cache_request(detail, crq->item, &bp, &len);
797 if (len < 0)
798 return -EAGAIN;
799 return PAGE_SIZE - len;
800}
801
802static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
803 loff_t *ppos, struct cache_detail *cd)
804{
805 struct cache_reader *rp = filp->private_data;
806 struct cache_request *rq;
807 struct inode *inode = file_inode(filp);
808 int err;
809
810 if (count == 0)
811 return 0;
812
813 inode_lock(inode); /* protect against multiple concurrent
814 * readers on this file */
815 again:
816 spin_lock(&queue_lock);
817 /* need to find next request */
818 while (rp->q.list.next != &cd->queue &&
819 list_entry(rp->q.list.next, struct cache_queue, list)
820 ->reader) {
821 struct list_head *next = rp->q.list.next;
822 list_move(&rp->q.list, next);
823 }
824 if (rp->q.list.next == &cd->queue) {
825 spin_unlock(&queue_lock);
826 inode_unlock(inode);
827 WARN_ON_ONCE(rp->offset);
828 return 0;
829 }
830 rq = container_of(rp->q.list.next, struct cache_request, q.list);
831 WARN_ON_ONCE(rq->q.reader);
832 if (rp->offset == 0)
833 rq->readers++;
834 spin_unlock(&queue_lock);
835
836 if (rq->len == 0) {
837 err = cache_request(cd, rq);
838 if (err < 0)
839 goto out;
840 rq->len = err;
841 }
842
843 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
844 err = -EAGAIN;
845 spin_lock(&queue_lock);
846 list_move(&rp->q.list, &rq->q.list);
847 spin_unlock(&queue_lock);
848 } else {
849 if (rp->offset + count > rq->len)
850 count = rq->len - rp->offset;
851 err = -EFAULT;
852 if (copy_to_user(buf, rq->buf + rp->offset, count))
853 goto out;
854 rp->offset += count;
855 if (rp->offset >= rq->len) {
856 rp->offset = 0;
857 spin_lock(&queue_lock);
858 list_move(&rp->q.list, &rq->q.list);
859 spin_unlock(&queue_lock);
860 }
861 err = 0;
862 }
863 out:
864 if (rp->offset == 0) {
865 /* need to release rq */
866 spin_lock(&queue_lock);
867 rq->readers--;
868 if (rq->readers == 0 &&
869 !test_bit(CACHE_PENDING, &rq->item->flags)) {
870 list_del(&rq->q.list);
871 spin_unlock(&queue_lock);
872 cache_put(rq->item, cd);
873 kfree(rq->buf);
874 kfree(rq);
875 } else
876 spin_unlock(&queue_lock);
877 }
878 if (err == -EAGAIN)
879 goto again;
880 inode_unlock(inode);
881 return err ? err : count;
882}
883
884static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
885 size_t count, struct cache_detail *cd)
886{
887 ssize_t ret;
888
889 if (count == 0)
890 return -EINVAL;
891 if (copy_from_user(kaddr, buf, count))
892 return -EFAULT;
893 kaddr[count] = '\0';
894 ret = cd->cache_parse(cd, kaddr, count);
895 if (!ret)
896 ret = count;
897 return ret;
898}
899
900static ssize_t cache_slow_downcall(const char __user *buf,
901 size_t count, struct cache_detail *cd)
902{
903 static char write_buf[8192]; /* protected by queue_io_mutex */
904 ssize_t ret = -EINVAL;
905
906 if (count >= sizeof(write_buf))
907 goto out;
908 mutex_lock(&queue_io_mutex);
909 ret = cache_do_downcall(write_buf, buf, count, cd);
910 mutex_unlock(&queue_io_mutex);
911out:
912 return ret;
913}
914
915static ssize_t cache_downcall(struct address_space *mapping,
916 const char __user *buf,
917 size_t count, struct cache_detail *cd)
918{
919 struct page *page;
920 char *kaddr;
921 ssize_t ret = -ENOMEM;
922
923 if (count >= PAGE_SIZE)
924 goto out_slow;
925
926 page = find_or_create_page(mapping, 0, GFP_KERNEL);
927 if (!page)
928 goto out_slow;
929
930 kaddr = kmap(page);
931 ret = cache_do_downcall(kaddr, buf, count, cd);
932 kunmap(page);
933 unlock_page(page);
934 put_page(page);
935 return ret;
936out_slow:
937 return cache_slow_downcall(buf, count, cd);
938}
939
940static ssize_t cache_write(struct file *filp, const char __user *buf,
941 size_t count, loff_t *ppos,
942 struct cache_detail *cd)
943{
944 struct address_space *mapping = filp->f_mapping;
945 struct inode *inode = file_inode(filp);
946 ssize_t ret = -EINVAL;
947
948 if (!cd->cache_parse)
949 goto out;
950
951 inode_lock(inode);
952 ret = cache_downcall(mapping, buf, count, cd);
953 inode_unlock(inode);
954out:
955 return ret;
956}
957
958static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
959
960static __poll_t cache_poll(struct file *filp, poll_table *wait,
961 struct cache_detail *cd)
962{
963 __poll_t mask;
964 struct cache_reader *rp = filp->private_data;
965 struct cache_queue *cq;
966
967 poll_wait(filp, &queue_wait, wait);
968
969 /* alway allow write */
970 mask = EPOLLOUT | EPOLLWRNORM;
971
972 if (!rp)
973 return mask;
974
975 spin_lock(&queue_lock);
976
977 for (cq= &rp->q; &cq->list != &cd->queue;
978 cq = list_entry(cq->list.next, struct cache_queue, list))
979 if (!cq->reader) {
980 mask |= EPOLLIN | EPOLLRDNORM;
981 break;
982 }
983 spin_unlock(&queue_lock);
984 return mask;
985}
986
987static int cache_ioctl(struct inode *ino, struct file *filp,
988 unsigned int cmd, unsigned long arg,
989 struct cache_detail *cd)
990{
991 int len = 0;
992 struct cache_reader *rp = filp->private_data;
993 struct cache_queue *cq;
994
995 if (cmd != FIONREAD || !rp)
996 return -EINVAL;
997
998 spin_lock(&queue_lock);
999
1000 /* only find the length remaining in current request,
1001 * or the length of the next request
1002 */
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 struct cache_request *cr =
1007 container_of(cq, struct cache_request, q);
1008 len = cr->len - rp->offset;
1009 break;
1010 }
1011 spin_unlock(&queue_lock);
1012
1013 return put_user(len, (int __user *)arg);
1014}
1015
1016static int cache_open(struct inode *inode, struct file *filp,
1017 struct cache_detail *cd)
1018{
1019 struct cache_reader *rp = NULL;
1020
1021 if (!cd || !try_module_get(cd->owner))
1022 return -EACCES;
1023 nonseekable_open(inode, filp);
1024 if (filp->f_mode & FMODE_READ) {
1025 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1026 if (!rp) {
1027 module_put(cd->owner);
1028 return -ENOMEM;
1029 }
1030 rp->offset = 0;
1031 rp->q.reader = 1;
1032
1033 spin_lock(&queue_lock);
1034 list_add(&rp->q.list, &cd->queue);
1035 spin_unlock(&queue_lock);
1036 }
1037 if (filp->f_mode & FMODE_WRITE)
1038 atomic_inc(&cd->writers);
1039 filp->private_data = rp;
1040 return 0;
1041}
1042
1043static int cache_release(struct inode *inode, struct file *filp,
1044 struct cache_detail *cd)
1045{
1046 struct cache_reader *rp = filp->private_data;
1047
1048 if (rp) {
1049 spin_lock(&queue_lock);
1050 if (rp->offset) {
1051 struct cache_queue *cq;
1052 for (cq= &rp->q; &cq->list != &cd->queue;
1053 cq = list_entry(cq->list.next, struct cache_queue, list))
1054 if (!cq->reader) {
1055 container_of(cq, struct cache_request, q)
1056 ->readers--;
1057 break;
1058 }
1059 rp->offset = 0;
1060 }
1061 list_del(&rp->q.list);
1062 spin_unlock(&queue_lock);
1063
1064 filp->private_data = NULL;
1065 kfree(rp);
1066
1067 }
1068 if (filp->f_mode & FMODE_WRITE) {
1069 atomic_dec(&cd->writers);
1070 cd->last_close = seconds_since_boot();
1071 }
1072 module_put(cd->owner);
1073 return 0;
1074}
1075
1076
1077
1078static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1079{
1080 struct cache_queue *cq, *tmp;
1081 struct cache_request *cr;
1082 struct list_head dequeued;
1083
1084 INIT_LIST_HEAD(&dequeued);
1085 spin_lock(&queue_lock);
1086 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1087 if (!cq->reader) {
1088 cr = container_of(cq, struct cache_request, q);
1089 if (cr->item != ch)
1090 continue;
1091 if (test_bit(CACHE_PENDING, &ch->flags))
1092 /* Lost a race and it is pending again */
1093 break;
1094 if (cr->readers != 0)
1095 continue;
1096 list_move(&cr->q.list, &dequeued);
1097 }
1098 spin_unlock(&queue_lock);
1099 while (!list_empty(&dequeued)) {
1100 cr = list_entry(dequeued.next, struct cache_request, q.list);
1101 list_del(&cr->q.list);
1102 cache_put(cr->item, detail);
1103 kfree(cr->buf);
1104 kfree(cr);
1105 }
1106}
1107
1108/*
1109 * Support routines for text-based upcalls.
1110 * Fields are separated by spaces.
1111 * Fields are either mangled to quote space tab newline slosh with slosh
1112 * or a hexified with a leading \x
1113 * Record is terminated with newline.
1114 *
1115 */
1116
1117void qword_add(char **bpp, int *lp, char *str)
1118{
1119 char *bp = *bpp;
1120 int len = *lp;
1121 int ret;
1122
1123 if (len < 0) return;
1124
1125 ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1126 if (ret >= len) {
1127 bp += len;
1128 len = -1;
1129 } else {
1130 bp += ret;
1131 len -= ret;
1132 *bp++ = ' ';
1133 len--;
1134 }
1135 *bpp = bp;
1136 *lp = len;
1137}
1138EXPORT_SYMBOL_GPL(qword_add);
1139
1140void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1141{
1142 char *bp = *bpp;
1143 int len = *lp;
1144
1145 if (len < 0) return;
1146
1147 if (len > 2) {
1148 *bp++ = '\\';
1149 *bp++ = 'x';
1150 len -= 2;
1151 while (blen && len >= 2) {
1152 bp = hex_byte_pack(bp, *buf++);
1153 len -= 2;
1154 blen--;
1155 }
1156 }
1157 if (blen || len<1) len = -1;
1158 else {
1159 *bp++ = ' ';
1160 len--;
1161 }
1162 *bpp = bp;
1163 *lp = len;
1164}
1165EXPORT_SYMBOL_GPL(qword_addhex);
1166
1167static void warn_no_listener(struct cache_detail *detail)
1168{
1169 if (detail->last_warn != detail->last_close) {
1170 detail->last_warn = detail->last_close;
1171 if (detail->warn_no_listener)
1172 detail->warn_no_listener(detail, detail->last_close != 0);
1173 }
1174}
1175
1176static bool cache_listeners_exist(struct cache_detail *detail)
1177{
1178 if (atomic_read(&detail->writers))
1179 return true;
1180 if (detail->last_close == 0)
1181 /* This cache was never opened */
1182 return false;
1183 if (detail->last_close < seconds_since_boot() - 30)
1184 /*
1185 * We allow for the possibility that someone might
1186 * restart a userspace daemon without restarting the
1187 * server; but after 30 seconds, we give up.
1188 */
1189 return false;
1190 return true;
1191}
1192
1193/*
1194 * register an upcall request to user-space and queue it up for read() by the
1195 * upcall daemon.
1196 *
1197 * Each request is at most one page long.
1198 */
1199int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1200{
1201
1202 char *buf;
1203 struct cache_request *crq;
1204 int ret = 0;
1205
1206 if (!detail->cache_request)
1207 return -EINVAL;
1208
1209 if (!cache_listeners_exist(detail)) {
1210 warn_no_listener(detail);
1211 return -EINVAL;
1212 }
1213 if (test_bit(CACHE_CLEANED, &h->flags))
1214 /* Too late to make an upcall */
1215 return -EAGAIN;
1216
1217 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1218 if (!buf)
1219 return -EAGAIN;
1220
1221 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1222 if (!crq) {
1223 kfree(buf);
1224 return -EAGAIN;
1225 }
1226
1227 crq->q.reader = 0;
1228 crq->buf = buf;
1229 crq->len = 0;
1230 crq->readers = 0;
1231 spin_lock(&queue_lock);
1232 if (test_bit(CACHE_PENDING, &h->flags)) {
1233 crq->item = cache_get(h);
1234 list_add_tail(&crq->q.list, &detail->queue);
1235 } else
1236 /* Lost a race, no longer PENDING, so don't enqueue */
1237 ret = -EAGAIN;
1238 spin_unlock(&queue_lock);
1239 wake_up(&queue_wait);
1240 if (ret == -EAGAIN) {
1241 kfree(buf);
1242 kfree(crq);
1243 }
1244 return ret;
1245}
1246EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1247
1248/*
1249 * parse a message from user-space and pass it
1250 * to an appropriate cache
1251 * Messages are, like requests, separated into fields by
1252 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1253 *
1254 * Message is
1255 * reply cachename expiry key ... content....
1256 *
1257 * key and content are both parsed by cache
1258 */
1259
1260int qword_get(char **bpp, char *dest, int bufsize)
1261{
1262 /* return bytes copied, or -1 on error */
1263 char *bp = *bpp;
1264 int len = 0;
1265
1266 while (*bp == ' ') bp++;
1267
1268 if (bp[0] == '\\' && bp[1] == 'x') {
1269 /* HEX STRING */
1270 bp += 2;
1271 while (len < bufsize - 1) {
1272 int h, l;
1273
1274 h = hex_to_bin(bp[0]);
1275 if (h < 0)
1276 break;
1277
1278 l = hex_to_bin(bp[1]);
1279 if (l < 0)
1280 break;
1281
1282 *dest++ = (h << 4) | l;
1283 bp += 2;
1284 len++;
1285 }
1286 } else {
1287 /* text with \nnn octal quoting */
1288 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1289 if (*bp == '\\' &&
1290 isodigit(bp[1]) && (bp[1] <= '3') &&
1291 isodigit(bp[2]) &&
1292 isodigit(bp[3])) {
1293 int byte = (*++bp -'0');
1294 bp++;
1295 byte = (byte << 3) | (*bp++ - '0');
1296 byte = (byte << 3) | (*bp++ - '0');
1297 *dest++ = byte;
1298 len++;
1299 } else {
1300 *dest++ = *bp++;
1301 len++;
1302 }
1303 }
1304 }
1305
1306 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1307 return -1;
1308 while (*bp == ' ') bp++;
1309 *bpp = bp;
1310 *dest = '\0';
1311 return len;
1312}
1313EXPORT_SYMBOL_GPL(qword_get);
1314
1315
1316/*
1317 * support /proc/net/rpc/$CACHENAME/content
1318 * as a seqfile.
1319 * We call ->cache_show passing NULL for the item to
1320 * get a header, then pass each real item in the cache
1321 */
1322
1323static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1324{
1325 loff_t n = *pos;
1326 unsigned int hash, entry;
1327 struct cache_head *ch;
1328 struct cache_detail *cd = m->private;
1329
1330 if (!n--)
1331 return SEQ_START_TOKEN;
1332 hash = n >> 32;
1333 entry = n & ((1LL<<32) - 1);
1334
1335 hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1336 if (!entry--)
1337 return ch;
1338 n &= ~((1LL<<32) - 1);
1339 do {
1340 hash++;
1341 n += 1LL<<32;
1342 } while(hash < cd->hash_size &&
1343 hlist_empty(&cd->hash_table[hash]));
1344 if (hash >= cd->hash_size)
1345 return NULL;
1346 *pos = n+1;
1347 return hlist_entry_safe(rcu_dereference_raw(
1348 hlist_first_rcu(&cd->hash_table[hash])),
1349 struct cache_head, cache_list);
1350}
1351
1352static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1353{
1354 struct cache_head *ch = p;
1355 int hash = (*pos >> 32);
1356 struct cache_detail *cd = m->private;
1357
1358 if (p == SEQ_START_TOKEN)
1359 hash = 0;
1360 else if (ch->cache_list.next == NULL) {
1361 hash++;
1362 *pos += 1LL<<32;
1363 } else {
1364 ++*pos;
1365 return hlist_entry_safe(rcu_dereference_raw(
1366 hlist_next_rcu(&ch->cache_list)),
1367 struct cache_head, cache_list);
1368 }
1369 *pos &= ~((1LL<<32) - 1);
1370 while (hash < cd->hash_size &&
1371 hlist_empty(&cd->hash_table[hash])) {
1372 hash++;
1373 *pos += 1LL<<32;
1374 }
1375 if (hash >= cd->hash_size)
1376 return NULL;
1377 ++*pos;
1378 return hlist_entry_safe(rcu_dereference_raw(
1379 hlist_first_rcu(&cd->hash_table[hash])),
1380 struct cache_head, cache_list);
1381}
1382
1383void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1384 __acquires(RCU)
1385{
1386 rcu_read_lock();
1387 return __cache_seq_start(m, pos);
1388}
1389EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1390
1391void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1392{
1393 return cache_seq_next(file, p, pos);
1394}
1395EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1396
1397void cache_seq_stop_rcu(struct seq_file *m, void *p)
1398 __releases(RCU)
1399{
1400 rcu_read_unlock();
1401}
1402EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1403
1404static int c_show(struct seq_file *m, void *p)
1405{
1406 struct cache_head *cp = p;
1407 struct cache_detail *cd = m->private;
1408
1409 if (p == SEQ_START_TOKEN)
1410 return cd->cache_show(m, cd, NULL);
1411
1412 ifdebug(CACHE)
1413 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1414 convert_to_wallclock(cp->expiry_time),
1415 kref_read(&cp->ref), cp->flags);
1416 cache_get(cp);
1417 if (cache_check(cd, cp, NULL))
1418 /* cache_check does a cache_put on failure */
1419 seq_printf(m, "# ");
1420 else {
1421 if (cache_is_expired(cd, cp))
1422 seq_printf(m, "# ");
1423 cache_put(cp, cd);
1424 }
1425
1426 return cd->cache_show(m, cd, cp);
1427}
1428
1429static const struct seq_operations cache_content_op = {
1430 .start = cache_seq_start_rcu,
1431 .next = cache_seq_next_rcu,
1432 .stop = cache_seq_stop_rcu,
1433 .show = c_show,
1434};
1435
1436static int content_open(struct inode *inode, struct file *file,
1437 struct cache_detail *cd)
1438{
1439 struct seq_file *seq;
1440 int err;
1441
1442 if (!cd || !try_module_get(cd->owner))
1443 return -EACCES;
1444
1445 err = seq_open(file, &cache_content_op);
1446 if (err) {
1447 module_put(cd->owner);
1448 return err;
1449 }
1450
1451 seq = file->private_data;
1452 seq->private = cd;
1453 return 0;
1454}
1455
1456static int content_release(struct inode *inode, struct file *file,
1457 struct cache_detail *cd)
1458{
1459 int ret = seq_release(inode, file);
1460 module_put(cd->owner);
1461 return ret;
1462}
1463
1464static int open_flush(struct inode *inode, struct file *file,
1465 struct cache_detail *cd)
1466{
1467 if (!cd || !try_module_get(cd->owner))
1468 return -EACCES;
1469 return nonseekable_open(inode, file);
1470}
1471
1472static int release_flush(struct inode *inode, struct file *file,
1473 struct cache_detail *cd)
1474{
1475 module_put(cd->owner);
1476 return 0;
1477}
1478
1479static ssize_t read_flush(struct file *file, char __user *buf,
1480 size_t count, loff_t *ppos,
1481 struct cache_detail *cd)
1482{
1483 char tbuf[22];
1484 size_t len;
1485
1486 len = snprintf(tbuf, sizeof(tbuf), "%lu\n",
1487 convert_to_wallclock(cd->flush_time));
1488 return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1489}
1490
1491static ssize_t write_flush(struct file *file, const char __user *buf,
1492 size_t count, loff_t *ppos,
1493 struct cache_detail *cd)
1494{
1495 char tbuf[20];
1496 char *ep;
1497 time_t now;
1498
1499 if (*ppos || count > sizeof(tbuf)-1)
1500 return -EINVAL;
1501 if (copy_from_user(tbuf, buf, count))
1502 return -EFAULT;
1503 tbuf[count] = 0;
1504 simple_strtoul(tbuf, &ep, 0);
1505 if (*ep && *ep != '\n')
1506 return -EINVAL;
1507 /* Note that while we check that 'buf' holds a valid number,
1508 * we always ignore the value and just flush everything.
1509 * Making use of the number leads to races.
1510 */
1511
1512 now = seconds_since_boot();
1513 /* Always flush everything, so behave like cache_purge()
1514 * Do this by advancing flush_time to the current time,
1515 * or by one second if it has already reached the current time.
1516 * Newly added cache entries will always have ->last_refresh greater
1517 * that ->flush_time, so they don't get flushed prematurely.
1518 */
1519
1520 if (cd->flush_time >= now)
1521 now = cd->flush_time + 1;
1522
1523 cd->flush_time = now;
1524 cd->nextcheck = now;
1525 cache_flush();
1526
1527 if (cd->flush)
1528 cd->flush();
1529
1530 *ppos += count;
1531 return count;
1532}
1533
1534static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1535 size_t count, loff_t *ppos)
1536{
1537 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1538
1539 return cache_read(filp, buf, count, ppos, cd);
1540}
1541
1542static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1543 size_t count, loff_t *ppos)
1544{
1545 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1546
1547 return cache_write(filp, buf, count, ppos, cd);
1548}
1549
1550static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1551{
1552 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1553
1554 return cache_poll(filp, wait, cd);
1555}
1556
1557static long cache_ioctl_procfs(struct file *filp,
1558 unsigned int cmd, unsigned long arg)
1559{
1560 struct inode *inode = file_inode(filp);
1561 struct cache_detail *cd = PDE_DATA(inode);
1562
1563 return cache_ioctl(inode, filp, cmd, arg, cd);
1564}
1565
1566static int cache_open_procfs(struct inode *inode, struct file *filp)
1567{
1568 struct cache_detail *cd = PDE_DATA(inode);
1569
1570 return cache_open(inode, filp, cd);
1571}
1572
1573static int cache_release_procfs(struct inode *inode, struct file *filp)
1574{
1575 struct cache_detail *cd = PDE_DATA(inode);
1576
1577 return cache_release(inode, filp, cd);
1578}
1579
1580static const struct file_operations cache_file_operations_procfs = {
1581 .owner = THIS_MODULE,
1582 .llseek = no_llseek,
1583 .read = cache_read_procfs,
1584 .write = cache_write_procfs,
1585 .poll = cache_poll_procfs,
1586 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1587 .open = cache_open_procfs,
1588 .release = cache_release_procfs,
1589};
1590
1591static int content_open_procfs(struct inode *inode, struct file *filp)
1592{
1593 struct cache_detail *cd = PDE_DATA(inode);
1594
1595 return content_open(inode, filp, cd);
1596}
1597
1598static int content_release_procfs(struct inode *inode, struct file *filp)
1599{
1600 struct cache_detail *cd = PDE_DATA(inode);
1601
1602 return content_release(inode, filp, cd);
1603}
1604
1605static const struct file_operations content_file_operations_procfs = {
1606 .open = content_open_procfs,
1607 .read = seq_read,
1608 .llseek = seq_lseek,
1609 .release = content_release_procfs,
1610};
1611
1612static int open_flush_procfs(struct inode *inode, struct file *filp)
1613{
1614 struct cache_detail *cd = PDE_DATA(inode);
1615
1616 return open_flush(inode, filp, cd);
1617}
1618
1619static int release_flush_procfs(struct inode *inode, struct file *filp)
1620{
1621 struct cache_detail *cd = PDE_DATA(inode);
1622
1623 return release_flush(inode, filp, cd);
1624}
1625
1626static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1627 size_t count, loff_t *ppos)
1628{
1629 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1630
1631 return read_flush(filp, buf, count, ppos, cd);
1632}
1633
1634static ssize_t write_flush_procfs(struct file *filp,
1635 const char __user *buf,
1636 size_t count, loff_t *ppos)
1637{
1638 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1639
1640 return write_flush(filp, buf, count, ppos, cd);
1641}
1642
1643static const struct file_operations cache_flush_operations_procfs = {
1644 .open = open_flush_procfs,
1645 .read = read_flush_procfs,
1646 .write = write_flush_procfs,
1647 .release = release_flush_procfs,
1648 .llseek = no_llseek,
1649};
1650
1651static void remove_cache_proc_entries(struct cache_detail *cd)
1652{
1653 if (cd->procfs) {
1654 proc_remove(cd->procfs);
1655 cd->procfs = NULL;
1656 }
1657}
1658
1659#ifdef CONFIG_PROC_FS
1660static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1661{
1662 struct proc_dir_entry *p;
1663 struct sunrpc_net *sn;
1664
1665 sn = net_generic(net, sunrpc_net_id);
1666 cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1667 if (cd->procfs == NULL)
1668 goto out_nomem;
1669
1670 p = proc_create_data("flush", S_IFREG | 0600,
1671 cd->procfs, &cache_flush_operations_procfs, cd);
1672 if (p == NULL)
1673 goto out_nomem;
1674
1675 if (cd->cache_request || cd->cache_parse) {
1676 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1677 &cache_file_operations_procfs, cd);
1678 if (p == NULL)
1679 goto out_nomem;
1680 }
1681 if (cd->cache_show) {
1682 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1683 &content_file_operations_procfs, cd);
1684 if (p == NULL)
1685 goto out_nomem;
1686 }
1687 return 0;
1688out_nomem:
1689 remove_cache_proc_entries(cd);
1690 return -ENOMEM;
1691}
1692#else /* CONFIG_PROC_FS */
1693static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1694{
1695 return 0;
1696}
1697#endif
1698
1699void __init cache_initialize(void)
1700{
1701 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1702}
1703
1704int cache_register_net(struct cache_detail *cd, struct net *net)
1705{
1706 int ret;
1707
1708 sunrpc_init_cache_detail(cd);
1709 ret = create_cache_proc_entries(cd, net);
1710 if (ret)
1711 sunrpc_destroy_cache_detail(cd);
1712 return ret;
1713}
1714EXPORT_SYMBOL_GPL(cache_register_net);
1715
1716void cache_unregister_net(struct cache_detail *cd, struct net *net)
1717{
1718 remove_cache_proc_entries(cd);
1719 sunrpc_destroy_cache_detail(cd);
1720}
1721EXPORT_SYMBOL_GPL(cache_unregister_net);
1722
1723struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1724{
1725 struct cache_detail *cd;
1726 int i;
1727
1728 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1729 if (cd == NULL)
1730 return ERR_PTR(-ENOMEM);
1731
1732 cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1733 GFP_KERNEL);
1734 if (cd->hash_table == NULL) {
1735 kfree(cd);
1736 return ERR_PTR(-ENOMEM);
1737 }
1738
1739 for (i = 0; i < cd->hash_size; i++)
1740 INIT_HLIST_HEAD(&cd->hash_table[i]);
1741 cd->net = net;
1742 return cd;
1743}
1744EXPORT_SYMBOL_GPL(cache_create_net);
1745
1746void cache_destroy_net(struct cache_detail *cd, struct net *net)
1747{
1748 kfree(cd->hash_table);
1749 kfree(cd);
1750}
1751EXPORT_SYMBOL_GPL(cache_destroy_net);
1752
1753static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1754 size_t count, loff_t *ppos)
1755{
1756 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1757
1758 return cache_read(filp, buf, count, ppos, cd);
1759}
1760
1761static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1762 size_t count, loff_t *ppos)
1763{
1764 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1765
1766 return cache_write(filp, buf, count, ppos, cd);
1767}
1768
1769static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1770{
1771 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1772
1773 return cache_poll(filp, wait, cd);
1774}
1775
1776static long cache_ioctl_pipefs(struct file *filp,
1777 unsigned int cmd, unsigned long arg)
1778{
1779 struct inode *inode = file_inode(filp);
1780 struct cache_detail *cd = RPC_I(inode)->private;
1781
1782 return cache_ioctl(inode, filp, cmd, arg, cd);
1783}
1784
1785static int cache_open_pipefs(struct inode *inode, struct file *filp)
1786{
1787 struct cache_detail *cd = RPC_I(inode)->private;
1788
1789 return cache_open(inode, filp, cd);
1790}
1791
1792static int cache_release_pipefs(struct inode *inode, struct file *filp)
1793{
1794 struct cache_detail *cd = RPC_I(inode)->private;
1795
1796 return cache_release(inode, filp, cd);
1797}
1798
1799const struct file_operations cache_file_operations_pipefs = {
1800 .owner = THIS_MODULE,
1801 .llseek = no_llseek,
1802 .read = cache_read_pipefs,
1803 .write = cache_write_pipefs,
1804 .poll = cache_poll_pipefs,
1805 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1806 .open = cache_open_pipefs,
1807 .release = cache_release_pipefs,
1808};
1809
1810static int content_open_pipefs(struct inode *inode, struct file *filp)
1811{
1812 struct cache_detail *cd = RPC_I(inode)->private;
1813
1814 return content_open(inode, filp, cd);
1815}
1816
1817static int content_release_pipefs(struct inode *inode, struct file *filp)
1818{
1819 struct cache_detail *cd = RPC_I(inode)->private;
1820
1821 return content_release(inode, filp, cd);
1822}
1823
1824const struct file_operations content_file_operations_pipefs = {
1825 .open = content_open_pipefs,
1826 .read = seq_read,
1827 .llseek = seq_lseek,
1828 .release = content_release_pipefs,
1829};
1830
1831static int open_flush_pipefs(struct inode *inode, struct file *filp)
1832{
1833 struct cache_detail *cd = RPC_I(inode)->private;
1834
1835 return open_flush(inode, filp, cd);
1836}
1837
1838static int release_flush_pipefs(struct inode *inode, struct file *filp)
1839{
1840 struct cache_detail *cd = RPC_I(inode)->private;
1841
1842 return release_flush(inode, filp, cd);
1843}
1844
1845static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1846 size_t count, loff_t *ppos)
1847{
1848 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1849
1850 return read_flush(filp, buf, count, ppos, cd);
1851}
1852
1853static ssize_t write_flush_pipefs(struct file *filp,
1854 const char __user *buf,
1855 size_t count, loff_t *ppos)
1856{
1857 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1858
1859 return write_flush(filp, buf, count, ppos, cd);
1860}
1861
1862const struct file_operations cache_flush_operations_pipefs = {
1863 .open = open_flush_pipefs,
1864 .read = read_flush_pipefs,
1865 .write = write_flush_pipefs,
1866 .release = release_flush_pipefs,
1867 .llseek = no_llseek,
1868};
1869
1870int sunrpc_cache_register_pipefs(struct dentry *parent,
1871 const char *name, umode_t umode,
1872 struct cache_detail *cd)
1873{
1874 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1875 if (IS_ERR(dir))
1876 return PTR_ERR(dir);
1877 cd->pipefs = dir;
1878 return 0;
1879}
1880EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1881
1882void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1883{
1884 if (cd->pipefs) {
1885 rpc_remove_cache_dir(cd->pipefs);
1886 cd->pipefs = NULL;
1887 }
1888}
1889EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1890
1891void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1892{
1893 spin_lock(&cd->hash_lock);
1894 if (!hlist_unhashed(&h->cache_list)){
1895 hlist_del_init_rcu(&h->cache_list);
1896 cd->entries--;
1897 spin_unlock(&cd->hash_lock);
1898 cache_put(h, cd);
1899 } else
1900 spin_unlock(&cd->hash_lock);
1901}
1902EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);
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 <linux/string_helpers.h>
24#include <linux/uaccess.h>
25#include <linux/poll.h>
26#include <linux/seq_file.h>
27#include <linux/proc_fs.h>
28#include <linux/net.h>
29#include <linux/workqueue.h>
30#include <linux/mutex.h>
31#include <linux/pagemap.h>
32#include <asm/ioctls.h>
33#include <linux/sunrpc/types.h>
34#include <linux/sunrpc/cache.h>
35#include <linux/sunrpc/stats.h>
36#include <linux/sunrpc/rpc_pipe_fs.h>
37#include "netns.h"
38
39#define RPCDBG_FACILITY RPCDBG_CACHE
40
41static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
42static void cache_revisit_request(struct cache_head *item);
43
44static void cache_init(struct cache_head *h, struct cache_detail *detail)
45{
46 time_t now = seconds_since_boot();
47 INIT_HLIST_NODE(&h->cache_list);
48 h->flags = 0;
49 kref_init(&h->ref);
50 h->expiry_time = now + CACHE_NEW_EXPIRY;
51 if (now <= detail->flush_time)
52 /* ensure it isn't already expired */
53 now = detail->flush_time + 1;
54 h->last_refresh = now;
55}
56
57struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
58 struct cache_head *key, int hash)
59{
60 struct cache_head *new = NULL, *freeme = NULL, *tmp = NULL;
61 struct hlist_head *head;
62
63 head = &detail->hash_table[hash];
64
65 read_lock(&detail->hash_lock);
66
67 hlist_for_each_entry(tmp, head, cache_list) {
68 if (detail->match(tmp, key)) {
69 if (cache_is_expired(detail, tmp))
70 /* This entry is expired, we will discard it. */
71 break;
72 cache_get(tmp);
73 read_unlock(&detail->hash_lock);
74 return tmp;
75 }
76 }
77 read_unlock(&detail->hash_lock);
78 /* Didn't find anything, insert an empty entry */
79
80 new = detail->alloc();
81 if (!new)
82 return NULL;
83 /* must fully initialise 'new', else
84 * we might get lose if we need to
85 * cache_put it soon.
86 */
87 cache_init(new, detail);
88 detail->init(new, key);
89
90 write_lock(&detail->hash_lock);
91
92 /* check if entry appeared while we slept */
93 hlist_for_each_entry(tmp, head, cache_list) {
94 if (detail->match(tmp, key)) {
95 if (cache_is_expired(detail, tmp)) {
96 hlist_del_init(&tmp->cache_list);
97 detail->entries --;
98 freeme = tmp;
99 break;
100 }
101 cache_get(tmp);
102 write_unlock(&detail->hash_lock);
103 cache_put(new, detail);
104 return tmp;
105 }
106 }
107
108 hlist_add_head(&new->cache_list, head);
109 detail->entries++;
110 cache_get(new);
111 write_unlock(&detail->hash_lock);
112
113 if (freeme)
114 cache_put(freeme, detail);
115 return new;
116}
117EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
118
119
120static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
121
122static void cache_fresh_locked(struct cache_head *head, time_t expiry,
123 struct cache_detail *detail)
124{
125 time_t now = seconds_since_boot();
126 if (now <= detail->flush_time)
127 /* ensure it isn't immediately treated as expired */
128 now = detail->flush_time + 1;
129 head->expiry_time = expiry;
130 head->last_refresh = now;
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 *tmp;
152
153 if (!test_bit(CACHE_VALID, &old->flags)) {
154 write_lock(&detail->hash_lock);
155 if (!test_bit(CACHE_VALID, &old->flags)) {
156 if (test_bit(CACHE_NEGATIVE, &new->flags))
157 set_bit(CACHE_NEGATIVE, &old->flags);
158 else
159 detail->update(old, new);
160 cache_fresh_locked(old, new->expiry_time, detail);
161 write_unlock(&detail->hash_lock);
162 cache_fresh_unlocked(old, detail);
163 return old;
164 }
165 write_unlock(&detail->hash_lock);
166 }
167 /* We need to insert a new entry */
168 tmp = detail->alloc();
169 if (!tmp) {
170 cache_put(old, detail);
171 return NULL;
172 }
173 cache_init(tmp, detail);
174 detail->init(tmp, old);
175
176 write_lock(&detail->hash_lock);
177 if (test_bit(CACHE_NEGATIVE, &new->flags))
178 set_bit(CACHE_NEGATIVE, &tmp->flags);
179 else
180 detail->update(tmp, new);
181 hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
182 detail->entries++;
183 cache_get(tmp);
184 cache_fresh_locked(tmp, new->expiry_time, detail);
185 cache_fresh_locked(old, 0, detail);
186 write_unlock(&detail->hash_lock);
187 cache_fresh_unlocked(tmp, detail);
188 cache_fresh_unlocked(old, detail);
189 cache_put(old, detail);
190 return tmp;
191}
192EXPORT_SYMBOL_GPL(sunrpc_cache_update);
193
194static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
195{
196 if (cd->cache_upcall)
197 return cd->cache_upcall(cd, h);
198 return sunrpc_cache_pipe_upcall(cd, h);
199}
200
201static inline int cache_is_valid(struct cache_head *h)
202{
203 if (!test_bit(CACHE_VALID, &h->flags))
204 return -EAGAIN;
205 else {
206 /* entry is valid */
207 if (test_bit(CACHE_NEGATIVE, &h->flags))
208 return -ENOENT;
209 else {
210 /*
211 * In combination with write barrier in
212 * sunrpc_cache_update, ensures that anyone
213 * using the cache entry after this sees the
214 * updated contents:
215 */
216 smp_rmb();
217 return 0;
218 }
219 }
220}
221
222static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
223{
224 int rv;
225
226 write_lock(&detail->hash_lock);
227 rv = cache_is_valid(h);
228 if (rv == -EAGAIN) {
229 set_bit(CACHE_NEGATIVE, &h->flags);
230 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
231 detail);
232 rv = -ENOENT;
233 }
234 write_unlock(&detail->hash_lock);
235 cache_fresh_unlocked(h, detail);
236 return rv;
237}
238
239/*
240 * This is the generic cache management routine for all
241 * the authentication caches.
242 * It checks the currency of a cache item and will (later)
243 * initiate an upcall to fill it if needed.
244 *
245 *
246 * Returns 0 if the cache_head can be used, or cache_puts it and returns
247 * -EAGAIN if upcall is pending and request has been queued
248 * -ETIMEDOUT if upcall failed or request could not be queue or
249 * upcall completed but item is still invalid (implying that
250 * the cache item has been replaced with a newer one).
251 * -ENOENT if cache entry was negative
252 */
253int cache_check(struct cache_detail *detail,
254 struct cache_head *h, struct cache_req *rqstp)
255{
256 int rv;
257 long refresh_age, age;
258
259 /* First decide return status as best we can */
260 rv = cache_is_valid(h);
261
262 /* now see if we want to start an upcall */
263 refresh_age = (h->expiry_time - h->last_refresh);
264 age = seconds_since_boot() - h->last_refresh;
265
266 if (rqstp == NULL) {
267 if (rv == -EAGAIN)
268 rv = -ENOENT;
269 } else if (rv == -EAGAIN ||
270 (h->expiry_time != 0 && age > refresh_age/2)) {
271 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
272 refresh_age, age);
273 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
274 switch (cache_make_upcall(detail, h)) {
275 case -EINVAL:
276 rv = try_to_negate_entry(detail, h);
277 break;
278 case -EAGAIN:
279 cache_fresh_unlocked(h, detail);
280 break;
281 }
282 }
283 }
284
285 if (rv == -EAGAIN) {
286 if (!cache_defer_req(rqstp, h)) {
287 /*
288 * Request was not deferred; handle it as best
289 * we can ourselves:
290 */
291 rv = cache_is_valid(h);
292 if (rv == -EAGAIN)
293 rv = -ETIMEDOUT;
294 }
295 }
296 if (rv)
297 cache_put(h, detail);
298 return rv;
299}
300EXPORT_SYMBOL_GPL(cache_check);
301
302/*
303 * caches need to be periodically cleaned.
304 * For this we maintain a list of cache_detail and
305 * a current pointer into that list and into the table
306 * for that entry.
307 *
308 * Each time cache_clean is called it finds the next non-empty entry
309 * in the current table and walks the list in that entry
310 * looking for entries that can be removed.
311 *
312 * An entry gets removed if:
313 * - The expiry is before current time
314 * - The last_refresh time is before the flush_time for that cache
315 *
316 * later we might drop old entries with non-NEVER expiry if that table
317 * is getting 'full' for some definition of 'full'
318 *
319 * The question of "how often to scan a table" is an interesting one
320 * and is answered in part by the use of the "nextcheck" field in the
321 * cache_detail.
322 * When a scan of a table begins, the nextcheck field is set to a time
323 * that is well into the future.
324 * While scanning, if an expiry time is found that is earlier than the
325 * current nextcheck time, nextcheck is set to that expiry time.
326 * If the flush_time is ever set to a time earlier than the nextcheck
327 * time, the nextcheck time is then set to that flush_time.
328 *
329 * A table is then only scanned if the current time is at least
330 * the nextcheck time.
331 *
332 */
333
334static LIST_HEAD(cache_list);
335static DEFINE_SPINLOCK(cache_list_lock);
336static struct cache_detail *current_detail;
337static int current_index;
338
339static void do_cache_clean(struct work_struct *work);
340static struct delayed_work cache_cleaner;
341
342void sunrpc_init_cache_detail(struct cache_detail *cd)
343{
344 rwlock_init(&cd->hash_lock);
345 INIT_LIST_HEAD(&cd->queue);
346 spin_lock(&cache_list_lock);
347 cd->nextcheck = 0;
348 cd->entries = 0;
349 atomic_set(&cd->readers, 0);
350 cd->last_close = 0;
351 cd->last_warn = -1;
352 list_add(&cd->others, &cache_list);
353 spin_unlock(&cache_list_lock);
354
355 /* start the cleaning process */
356 queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
357}
358EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
359
360void sunrpc_destroy_cache_detail(struct cache_detail *cd)
361{
362 cache_purge(cd);
363 spin_lock(&cache_list_lock);
364 write_lock(&cd->hash_lock);
365 if (current_detail == cd)
366 current_detail = NULL;
367 list_del_init(&cd->others);
368 write_unlock(&cd->hash_lock);
369 spin_unlock(&cache_list_lock);
370 if (list_empty(&cache_list)) {
371 /* module must be being unloaded so its safe to kill the worker */
372 cancel_delayed_work_sync(&cache_cleaner);
373 }
374}
375EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
376
377/* clean cache tries to find something to clean
378 * and cleans it.
379 * It returns 1 if it cleaned something,
380 * 0 if it didn't find anything this time
381 * -1 if it fell off the end of the list.
382 */
383static int cache_clean(void)
384{
385 int rv = 0;
386 struct list_head *next;
387
388 spin_lock(&cache_list_lock);
389
390 /* find a suitable table if we don't already have one */
391 while (current_detail == NULL ||
392 current_index >= current_detail->hash_size) {
393 if (current_detail)
394 next = current_detail->others.next;
395 else
396 next = cache_list.next;
397 if (next == &cache_list) {
398 current_detail = NULL;
399 spin_unlock(&cache_list_lock);
400 return -1;
401 }
402 current_detail = list_entry(next, struct cache_detail, others);
403 if (current_detail->nextcheck > seconds_since_boot())
404 current_index = current_detail->hash_size;
405 else {
406 current_index = 0;
407 current_detail->nextcheck = seconds_since_boot()+30*60;
408 }
409 }
410
411 /* find a non-empty bucket in the table */
412 while (current_detail &&
413 current_index < current_detail->hash_size &&
414 hlist_empty(¤t_detail->hash_table[current_index]))
415 current_index++;
416
417 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
418
419 if (current_detail && current_index < current_detail->hash_size) {
420 struct cache_head *ch = NULL;
421 struct cache_detail *d;
422 struct hlist_head *head;
423 struct hlist_node *tmp;
424
425 write_lock(¤t_detail->hash_lock);
426
427 /* Ok, now to clean this strand */
428
429 head = ¤t_detail->hash_table[current_index];
430 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
431 if (current_detail->nextcheck > ch->expiry_time)
432 current_detail->nextcheck = ch->expiry_time+1;
433 if (!cache_is_expired(current_detail, ch))
434 continue;
435
436 hlist_del_init(&ch->cache_list);
437 current_detail->entries--;
438 rv = 1;
439 break;
440 }
441
442 write_unlock(¤t_detail->hash_lock);
443 d = current_detail;
444 if (!ch)
445 current_index ++;
446 spin_unlock(&cache_list_lock);
447 if (ch) {
448 set_bit(CACHE_CLEANED, &ch->flags);
449 cache_fresh_unlocked(ch, d);
450 cache_put(ch, d);
451 }
452 } else
453 spin_unlock(&cache_list_lock);
454
455 return rv;
456}
457
458/*
459 * We want to regularly clean the cache, so we need to schedule some work ...
460 */
461static void do_cache_clean(struct work_struct *work)
462{
463 int delay = 5;
464 if (cache_clean() == -1)
465 delay = round_jiffies_relative(30*HZ);
466
467 if (list_empty(&cache_list))
468 delay = 0;
469
470 if (delay)
471 queue_delayed_work(system_power_efficient_wq,
472 &cache_cleaner, delay);
473}
474
475
476/*
477 * Clean all caches promptly. This just calls cache_clean
478 * repeatedly until we are sure that every cache has had a chance to
479 * be fully cleaned
480 */
481void cache_flush(void)
482{
483 while (cache_clean() != -1)
484 cond_resched();
485 while (cache_clean() != -1)
486 cond_resched();
487}
488EXPORT_SYMBOL_GPL(cache_flush);
489
490void cache_purge(struct cache_detail *detail)
491{
492 struct cache_head *ch = NULL;
493 struct hlist_head *head = NULL;
494 struct hlist_node *tmp = NULL;
495 int i = 0;
496
497 write_lock(&detail->hash_lock);
498 if (!detail->entries) {
499 write_unlock(&detail->hash_lock);
500 return;
501 }
502
503 dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
504 for (i = 0; i < detail->hash_size; i++) {
505 head = &detail->hash_table[i];
506 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
507 hlist_del_init(&ch->cache_list);
508 detail->entries--;
509
510 set_bit(CACHE_CLEANED, &ch->flags);
511 write_unlock(&detail->hash_lock);
512 cache_fresh_unlocked(ch, detail);
513 cache_put(ch, detail);
514 write_lock(&detail->hash_lock);
515 }
516 }
517 write_unlock(&detail->hash_lock);
518}
519EXPORT_SYMBOL_GPL(cache_purge);
520
521
522/*
523 * Deferral and Revisiting of Requests.
524 *
525 * If a cache lookup finds a pending entry, we
526 * need to defer the request and revisit it later.
527 * All deferred requests are stored in a hash table,
528 * indexed by "struct cache_head *".
529 * As it may be wasteful to store a whole request
530 * structure, we allow the request to provide a
531 * deferred form, which must contain a
532 * 'struct cache_deferred_req'
533 * This cache_deferred_req contains a method to allow
534 * it to be revisited when cache info is available
535 */
536
537#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
538#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
539
540#define DFR_MAX 300 /* ??? */
541
542static DEFINE_SPINLOCK(cache_defer_lock);
543static LIST_HEAD(cache_defer_list);
544static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
545static int cache_defer_cnt;
546
547static void __unhash_deferred_req(struct cache_deferred_req *dreq)
548{
549 hlist_del_init(&dreq->hash);
550 if (!list_empty(&dreq->recent)) {
551 list_del_init(&dreq->recent);
552 cache_defer_cnt--;
553 }
554}
555
556static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
557{
558 int hash = DFR_HASH(item);
559
560 INIT_LIST_HEAD(&dreq->recent);
561 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
562}
563
564static void setup_deferral(struct cache_deferred_req *dreq,
565 struct cache_head *item,
566 int count_me)
567{
568
569 dreq->item = item;
570
571 spin_lock(&cache_defer_lock);
572
573 __hash_deferred_req(dreq, item);
574
575 if (count_me) {
576 cache_defer_cnt++;
577 list_add(&dreq->recent, &cache_defer_list);
578 }
579
580 spin_unlock(&cache_defer_lock);
581
582}
583
584struct thread_deferred_req {
585 struct cache_deferred_req handle;
586 struct completion completion;
587};
588
589static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
590{
591 struct thread_deferred_req *dr =
592 container_of(dreq, struct thread_deferred_req, handle);
593 complete(&dr->completion);
594}
595
596static void cache_wait_req(struct cache_req *req, struct cache_head *item)
597{
598 struct thread_deferred_req sleeper;
599 struct cache_deferred_req *dreq = &sleeper.handle;
600
601 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
602 dreq->revisit = cache_restart_thread;
603
604 setup_deferral(dreq, item, 0);
605
606 if (!test_bit(CACHE_PENDING, &item->flags) ||
607 wait_for_completion_interruptible_timeout(
608 &sleeper.completion, req->thread_wait) <= 0) {
609 /* The completion wasn't completed, so we need
610 * to clean up
611 */
612 spin_lock(&cache_defer_lock);
613 if (!hlist_unhashed(&sleeper.handle.hash)) {
614 __unhash_deferred_req(&sleeper.handle);
615 spin_unlock(&cache_defer_lock);
616 } else {
617 /* cache_revisit_request already removed
618 * this from the hash table, but hasn't
619 * called ->revisit yet. It will very soon
620 * and we need to wait for it.
621 */
622 spin_unlock(&cache_defer_lock);
623 wait_for_completion(&sleeper.completion);
624 }
625 }
626}
627
628static void cache_limit_defers(void)
629{
630 /* Make sure we haven't exceed the limit of allowed deferred
631 * requests.
632 */
633 struct cache_deferred_req *discard = NULL;
634
635 if (cache_defer_cnt <= DFR_MAX)
636 return;
637
638 spin_lock(&cache_defer_lock);
639
640 /* Consider removing either the first or the last */
641 if (cache_defer_cnt > DFR_MAX) {
642 if (prandom_u32() & 1)
643 discard = list_entry(cache_defer_list.next,
644 struct cache_deferred_req, recent);
645 else
646 discard = list_entry(cache_defer_list.prev,
647 struct cache_deferred_req, recent);
648 __unhash_deferred_req(discard);
649 }
650 spin_unlock(&cache_defer_lock);
651 if (discard)
652 discard->revisit(discard, 1);
653}
654
655/* Return true if and only if a deferred request is queued. */
656static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
657{
658 struct cache_deferred_req *dreq;
659
660 if (req->thread_wait) {
661 cache_wait_req(req, item);
662 if (!test_bit(CACHE_PENDING, &item->flags))
663 return false;
664 }
665 dreq = req->defer(req);
666 if (dreq == NULL)
667 return false;
668 setup_deferral(dreq, item, 1);
669 if (!test_bit(CACHE_PENDING, &item->flags))
670 /* Bit could have been cleared before we managed to
671 * set up the deferral, so need to revisit just in case
672 */
673 cache_revisit_request(item);
674
675 cache_limit_defers();
676 return true;
677}
678
679static void cache_revisit_request(struct cache_head *item)
680{
681 struct cache_deferred_req *dreq;
682 struct list_head pending;
683 struct hlist_node *tmp;
684 int hash = DFR_HASH(item);
685
686 INIT_LIST_HEAD(&pending);
687 spin_lock(&cache_defer_lock);
688
689 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
690 if (dreq->item == item) {
691 __unhash_deferred_req(dreq);
692 list_add(&dreq->recent, &pending);
693 }
694
695 spin_unlock(&cache_defer_lock);
696
697 while (!list_empty(&pending)) {
698 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
699 list_del_init(&dreq->recent);
700 dreq->revisit(dreq, 0);
701 }
702}
703
704void cache_clean_deferred(void *owner)
705{
706 struct cache_deferred_req *dreq, *tmp;
707 struct list_head pending;
708
709
710 INIT_LIST_HEAD(&pending);
711 spin_lock(&cache_defer_lock);
712
713 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
714 if (dreq->owner == owner) {
715 __unhash_deferred_req(dreq);
716 list_add(&dreq->recent, &pending);
717 }
718 }
719 spin_unlock(&cache_defer_lock);
720
721 while (!list_empty(&pending)) {
722 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
723 list_del_init(&dreq->recent);
724 dreq->revisit(dreq, 1);
725 }
726}
727
728/*
729 * communicate with user-space
730 *
731 * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
732 * On read, you get a full request, or block.
733 * On write, an update request is processed.
734 * Poll works if anything to read, and always allows write.
735 *
736 * Implemented by linked list of requests. Each open file has
737 * a ->private that also exists in this list. New requests are added
738 * to the end and may wakeup and preceding readers.
739 * New readers are added to the head. If, on read, an item is found with
740 * CACHE_UPCALLING clear, we free it from the list.
741 *
742 */
743
744static DEFINE_SPINLOCK(queue_lock);
745static DEFINE_MUTEX(queue_io_mutex);
746
747struct cache_queue {
748 struct list_head list;
749 int reader; /* if 0, then request */
750};
751struct cache_request {
752 struct cache_queue q;
753 struct cache_head *item;
754 char * buf;
755 int len;
756 int readers;
757};
758struct cache_reader {
759 struct cache_queue q;
760 int offset; /* if non-0, we have a refcnt on next request */
761};
762
763static int cache_request(struct cache_detail *detail,
764 struct cache_request *crq)
765{
766 char *bp = crq->buf;
767 int len = PAGE_SIZE;
768
769 detail->cache_request(detail, crq->item, &bp, &len);
770 if (len < 0)
771 return -EAGAIN;
772 return PAGE_SIZE - len;
773}
774
775static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
776 loff_t *ppos, struct cache_detail *cd)
777{
778 struct cache_reader *rp = filp->private_data;
779 struct cache_request *rq;
780 struct inode *inode = file_inode(filp);
781 int err;
782
783 if (count == 0)
784 return 0;
785
786 inode_lock(inode); /* protect against multiple concurrent
787 * readers on this file */
788 again:
789 spin_lock(&queue_lock);
790 /* need to find next request */
791 while (rp->q.list.next != &cd->queue &&
792 list_entry(rp->q.list.next, struct cache_queue, list)
793 ->reader) {
794 struct list_head *next = rp->q.list.next;
795 list_move(&rp->q.list, next);
796 }
797 if (rp->q.list.next == &cd->queue) {
798 spin_unlock(&queue_lock);
799 inode_unlock(inode);
800 WARN_ON_ONCE(rp->offset);
801 return 0;
802 }
803 rq = container_of(rp->q.list.next, struct cache_request, q.list);
804 WARN_ON_ONCE(rq->q.reader);
805 if (rp->offset == 0)
806 rq->readers++;
807 spin_unlock(&queue_lock);
808
809 if (rq->len == 0) {
810 err = cache_request(cd, rq);
811 if (err < 0)
812 goto out;
813 rq->len = err;
814 }
815
816 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
817 err = -EAGAIN;
818 spin_lock(&queue_lock);
819 list_move(&rp->q.list, &rq->q.list);
820 spin_unlock(&queue_lock);
821 } else {
822 if (rp->offset + count > rq->len)
823 count = rq->len - rp->offset;
824 err = -EFAULT;
825 if (copy_to_user(buf, rq->buf + rp->offset, count))
826 goto out;
827 rp->offset += count;
828 if (rp->offset >= rq->len) {
829 rp->offset = 0;
830 spin_lock(&queue_lock);
831 list_move(&rp->q.list, &rq->q.list);
832 spin_unlock(&queue_lock);
833 }
834 err = 0;
835 }
836 out:
837 if (rp->offset == 0) {
838 /* need to release rq */
839 spin_lock(&queue_lock);
840 rq->readers--;
841 if (rq->readers == 0 &&
842 !test_bit(CACHE_PENDING, &rq->item->flags)) {
843 list_del(&rq->q.list);
844 spin_unlock(&queue_lock);
845 cache_put(rq->item, cd);
846 kfree(rq->buf);
847 kfree(rq);
848 } else
849 spin_unlock(&queue_lock);
850 }
851 if (err == -EAGAIN)
852 goto again;
853 inode_unlock(inode);
854 return err ? err : count;
855}
856
857static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
858 size_t count, struct cache_detail *cd)
859{
860 ssize_t ret;
861
862 if (count == 0)
863 return -EINVAL;
864 if (copy_from_user(kaddr, buf, count))
865 return -EFAULT;
866 kaddr[count] = '\0';
867 ret = cd->cache_parse(cd, kaddr, count);
868 if (!ret)
869 ret = count;
870 return ret;
871}
872
873static ssize_t cache_slow_downcall(const char __user *buf,
874 size_t count, struct cache_detail *cd)
875{
876 static char write_buf[8192]; /* protected by queue_io_mutex */
877 ssize_t ret = -EINVAL;
878
879 if (count >= sizeof(write_buf))
880 goto out;
881 mutex_lock(&queue_io_mutex);
882 ret = cache_do_downcall(write_buf, buf, count, cd);
883 mutex_unlock(&queue_io_mutex);
884out:
885 return ret;
886}
887
888static ssize_t cache_downcall(struct address_space *mapping,
889 const char __user *buf,
890 size_t count, struct cache_detail *cd)
891{
892 struct page *page;
893 char *kaddr;
894 ssize_t ret = -ENOMEM;
895
896 if (count >= PAGE_SIZE)
897 goto out_slow;
898
899 page = find_or_create_page(mapping, 0, GFP_KERNEL);
900 if (!page)
901 goto out_slow;
902
903 kaddr = kmap(page);
904 ret = cache_do_downcall(kaddr, buf, count, cd);
905 kunmap(page);
906 unlock_page(page);
907 put_page(page);
908 return ret;
909out_slow:
910 return cache_slow_downcall(buf, count, cd);
911}
912
913static ssize_t cache_write(struct file *filp, const char __user *buf,
914 size_t count, loff_t *ppos,
915 struct cache_detail *cd)
916{
917 struct address_space *mapping = filp->f_mapping;
918 struct inode *inode = file_inode(filp);
919 ssize_t ret = -EINVAL;
920
921 if (!cd->cache_parse)
922 goto out;
923
924 inode_lock(inode);
925 ret = cache_downcall(mapping, buf, count, cd);
926 inode_unlock(inode);
927out:
928 return ret;
929}
930
931static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
932
933static __poll_t cache_poll(struct file *filp, poll_table *wait,
934 struct cache_detail *cd)
935{
936 __poll_t mask;
937 struct cache_reader *rp = filp->private_data;
938 struct cache_queue *cq;
939
940 poll_wait(filp, &queue_wait, wait);
941
942 /* alway allow write */
943 mask = EPOLLOUT | EPOLLWRNORM;
944
945 if (!rp)
946 return mask;
947
948 spin_lock(&queue_lock);
949
950 for (cq= &rp->q; &cq->list != &cd->queue;
951 cq = list_entry(cq->list.next, struct cache_queue, list))
952 if (!cq->reader) {
953 mask |= EPOLLIN | EPOLLRDNORM;
954 break;
955 }
956 spin_unlock(&queue_lock);
957 return mask;
958}
959
960static int cache_ioctl(struct inode *ino, struct file *filp,
961 unsigned int cmd, unsigned long arg,
962 struct cache_detail *cd)
963{
964 int len = 0;
965 struct cache_reader *rp = filp->private_data;
966 struct cache_queue *cq;
967
968 if (cmd != FIONREAD || !rp)
969 return -EINVAL;
970
971 spin_lock(&queue_lock);
972
973 /* only find the length remaining in current request,
974 * or the length of the next request
975 */
976 for (cq= &rp->q; &cq->list != &cd->queue;
977 cq = list_entry(cq->list.next, struct cache_queue, list))
978 if (!cq->reader) {
979 struct cache_request *cr =
980 container_of(cq, struct cache_request, q);
981 len = cr->len - rp->offset;
982 break;
983 }
984 spin_unlock(&queue_lock);
985
986 return put_user(len, (int __user *)arg);
987}
988
989static int cache_open(struct inode *inode, struct file *filp,
990 struct cache_detail *cd)
991{
992 struct cache_reader *rp = NULL;
993
994 if (!cd || !try_module_get(cd->owner))
995 return -EACCES;
996 nonseekable_open(inode, filp);
997 if (filp->f_mode & FMODE_READ) {
998 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
999 if (!rp) {
1000 module_put(cd->owner);
1001 return -ENOMEM;
1002 }
1003 rp->offset = 0;
1004 rp->q.reader = 1;
1005 atomic_inc(&cd->readers);
1006 spin_lock(&queue_lock);
1007 list_add(&rp->q.list, &cd->queue);
1008 spin_unlock(&queue_lock);
1009 }
1010 filp->private_data = rp;
1011 return 0;
1012}
1013
1014static int cache_release(struct inode *inode, struct file *filp,
1015 struct cache_detail *cd)
1016{
1017 struct cache_reader *rp = filp->private_data;
1018
1019 if (rp) {
1020 spin_lock(&queue_lock);
1021 if (rp->offset) {
1022 struct cache_queue *cq;
1023 for (cq= &rp->q; &cq->list != &cd->queue;
1024 cq = list_entry(cq->list.next, struct cache_queue, list))
1025 if (!cq->reader) {
1026 container_of(cq, struct cache_request, q)
1027 ->readers--;
1028 break;
1029 }
1030 rp->offset = 0;
1031 }
1032 list_del(&rp->q.list);
1033 spin_unlock(&queue_lock);
1034
1035 filp->private_data = NULL;
1036 kfree(rp);
1037
1038 cd->last_close = seconds_since_boot();
1039 atomic_dec(&cd->readers);
1040 }
1041 module_put(cd->owner);
1042 return 0;
1043}
1044
1045
1046
1047static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1048{
1049 struct cache_queue *cq, *tmp;
1050 struct cache_request *cr;
1051 struct list_head dequeued;
1052
1053 INIT_LIST_HEAD(&dequeued);
1054 spin_lock(&queue_lock);
1055 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1056 if (!cq->reader) {
1057 cr = container_of(cq, struct cache_request, q);
1058 if (cr->item != ch)
1059 continue;
1060 if (test_bit(CACHE_PENDING, &ch->flags))
1061 /* Lost a race and it is pending again */
1062 break;
1063 if (cr->readers != 0)
1064 continue;
1065 list_move(&cr->q.list, &dequeued);
1066 }
1067 spin_unlock(&queue_lock);
1068 while (!list_empty(&dequeued)) {
1069 cr = list_entry(dequeued.next, struct cache_request, q.list);
1070 list_del(&cr->q.list);
1071 cache_put(cr->item, detail);
1072 kfree(cr->buf);
1073 kfree(cr);
1074 }
1075}
1076
1077/*
1078 * Support routines for text-based upcalls.
1079 * Fields are separated by spaces.
1080 * Fields are either mangled to quote space tab newline slosh with slosh
1081 * or a hexified with a leading \x
1082 * Record is terminated with newline.
1083 *
1084 */
1085
1086void qword_add(char **bpp, int *lp, char *str)
1087{
1088 char *bp = *bpp;
1089 int len = *lp;
1090 int ret;
1091
1092 if (len < 0) return;
1093
1094 ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1095 if (ret >= len) {
1096 bp += len;
1097 len = -1;
1098 } else {
1099 bp += ret;
1100 len -= ret;
1101 *bp++ = ' ';
1102 len--;
1103 }
1104 *bpp = bp;
1105 *lp = len;
1106}
1107EXPORT_SYMBOL_GPL(qword_add);
1108
1109void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1110{
1111 char *bp = *bpp;
1112 int len = *lp;
1113
1114 if (len < 0) return;
1115
1116 if (len > 2) {
1117 *bp++ = '\\';
1118 *bp++ = 'x';
1119 len -= 2;
1120 while (blen && len >= 2) {
1121 bp = hex_byte_pack(bp, *buf++);
1122 len -= 2;
1123 blen--;
1124 }
1125 }
1126 if (blen || len<1) len = -1;
1127 else {
1128 *bp++ = ' ';
1129 len--;
1130 }
1131 *bpp = bp;
1132 *lp = len;
1133}
1134EXPORT_SYMBOL_GPL(qword_addhex);
1135
1136static void warn_no_listener(struct cache_detail *detail)
1137{
1138 if (detail->last_warn != detail->last_close) {
1139 detail->last_warn = detail->last_close;
1140 if (detail->warn_no_listener)
1141 detail->warn_no_listener(detail, detail->last_close != 0);
1142 }
1143}
1144
1145static bool cache_listeners_exist(struct cache_detail *detail)
1146{
1147 if (atomic_read(&detail->readers))
1148 return true;
1149 if (detail->last_close == 0)
1150 /* This cache was never opened */
1151 return false;
1152 if (detail->last_close < seconds_since_boot() - 30)
1153 /*
1154 * We allow for the possibility that someone might
1155 * restart a userspace daemon without restarting the
1156 * server; but after 30 seconds, we give up.
1157 */
1158 return false;
1159 return true;
1160}
1161
1162/*
1163 * register an upcall request to user-space and queue it up for read() by the
1164 * upcall daemon.
1165 *
1166 * Each request is at most one page long.
1167 */
1168int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1169{
1170
1171 char *buf;
1172 struct cache_request *crq;
1173 int ret = 0;
1174
1175 if (!detail->cache_request)
1176 return -EINVAL;
1177
1178 if (!cache_listeners_exist(detail)) {
1179 warn_no_listener(detail);
1180 return -EINVAL;
1181 }
1182 if (test_bit(CACHE_CLEANED, &h->flags))
1183 /* Too late to make an upcall */
1184 return -EAGAIN;
1185
1186 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1187 if (!buf)
1188 return -EAGAIN;
1189
1190 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1191 if (!crq) {
1192 kfree(buf);
1193 return -EAGAIN;
1194 }
1195
1196 crq->q.reader = 0;
1197 crq->buf = buf;
1198 crq->len = 0;
1199 crq->readers = 0;
1200 spin_lock(&queue_lock);
1201 if (test_bit(CACHE_PENDING, &h->flags)) {
1202 crq->item = cache_get(h);
1203 list_add_tail(&crq->q.list, &detail->queue);
1204 } else
1205 /* Lost a race, no longer PENDING, so don't enqueue */
1206 ret = -EAGAIN;
1207 spin_unlock(&queue_lock);
1208 wake_up(&queue_wait);
1209 if (ret == -EAGAIN) {
1210 kfree(buf);
1211 kfree(crq);
1212 }
1213 return ret;
1214}
1215EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1216
1217/*
1218 * parse a message from user-space and pass it
1219 * to an appropriate cache
1220 * Messages are, like requests, separated into fields by
1221 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1222 *
1223 * Message is
1224 * reply cachename expiry key ... content....
1225 *
1226 * key and content are both parsed by cache
1227 */
1228
1229int qword_get(char **bpp, char *dest, int bufsize)
1230{
1231 /* return bytes copied, or -1 on error */
1232 char *bp = *bpp;
1233 int len = 0;
1234
1235 while (*bp == ' ') bp++;
1236
1237 if (bp[0] == '\\' && bp[1] == 'x') {
1238 /* HEX STRING */
1239 bp += 2;
1240 while (len < bufsize - 1) {
1241 int h, l;
1242
1243 h = hex_to_bin(bp[0]);
1244 if (h < 0)
1245 break;
1246
1247 l = hex_to_bin(bp[1]);
1248 if (l < 0)
1249 break;
1250
1251 *dest++ = (h << 4) | l;
1252 bp += 2;
1253 len++;
1254 }
1255 } else {
1256 /* text with \nnn octal quoting */
1257 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1258 if (*bp == '\\' &&
1259 isodigit(bp[1]) && (bp[1] <= '3') &&
1260 isodigit(bp[2]) &&
1261 isodigit(bp[3])) {
1262 int byte = (*++bp -'0');
1263 bp++;
1264 byte = (byte << 3) | (*bp++ - '0');
1265 byte = (byte << 3) | (*bp++ - '0');
1266 *dest++ = byte;
1267 len++;
1268 } else {
1269 *dest++ = *bp++;
1270 len++;
1271 }
1272 }
1273 }
1274
1275 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1276 return -1;
1277 while (*bp == ' ') bp++;
1278 *bpp = bp;
1279 *dest = '\0';
1280 return len;
1281}
1282EXPORT_SYMBOL_GPL(qword_get);
1283
1284
1285/*
1286 * support /proc/net/rpc/$CACHENAME/content
1287 * as a seqfile.
1288 * We call ->cache_show passing NULL for the item to
1289 * get a header, then pass each real item in the cache
1290 */
1291
1292void *cache_seq_start(struct seq_file *m, loff_t *pos)
1293 __acquires(cd->hash_lock)
1294{
1295 loff_t n = *pos;
1296 unsigned int hash, entry;
1297 struct cache_head *ch;
1298 struct cache_detail *cd = m->private;
1299
1300 read_lock(&cd->hash_lock);
1301 if (!n--)
1302 return SEQ_START_TOKEN;
1303 hash = n >> 32;
1304 entry = n & ((1LL<<32) - 1);
1305
1306 hlist_for_each_entry(ch, &cd->hash_table[hash], cache_list)
1307 if (!entry--)
1308 return ch;
1309 n &= ~((1LL<<32) - 1);
1310 do {
1311 hash++;
1312 n += 1LL<<32;
1313 } while(hash < cd->hash_size &&
1314 hlist_empty(&cd->hash_table[hash]));
1315 if (hash >= cd->hash_size)
1316 return NULL;
1317 *pos = n+1;
1318 return hlist_entry_safe(cd->hash_table[hash].first,
1319 struct cache_head, cache_list);
1320}
1321EXPORT_SYMBOL_GPL(cache_seq_start);
1322
1323void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1324{
1325 struct cache_head *ch = p;
1326 int hash = (*pos >> 32);
1327 struct cache_detail *cd = m->private;
1328
1329 if (p == SEQ_START_TOKEN)
1330 hash = 0;
1331 else if (ch->cache_list.next == NULL) {
1332 hash++;
1333 *pos += 1LL<<32;
1334 } else {
1335 ++*pos;
1336 return hlist_entry_safe(ch->cache_list.next,
1337 struct cache_head, cache_list);
1338 }
1339 *pos &= ~((1LL<<32) - 1);
1340 while (hash < cd->hash_size &&
1341 hlist_empty(&cd->hash_table[hash])) {
1342 hash++;
1343 *pos += 1LL<<32;
1344 }
1345 if (hash >= cd->hash_size)
1346 return NULL;
1347 ++*pos;
1348 return hlist_entry_safe(cd->hash_table[hash].first,
1349 struct cache_head, cache_list);
1350}
1351EXPORT_SYMBOL_GPL(cache_seq_next);
1352
1353void cache_seq_stop(struct seq_file *m, void *p)
1354 __releases(cd->hash_lock)
1355{
1356 struct cache_detail *cd = m->private;
1357 read_unlock(&cd->hash_lock);
1358}
1359EXPORT_SYMBOL_GPL(cache_seq_stop);
1360
1361static int c_show(struct seq_file *m, void *p)
1362{
1363 struct cache_head *cp = p;
1364 struct cache_detail *cd = m->private;
1365
1366 if (p == SEQ_START_TOKEN)
1367 return cd->cache_show(m, cd, NULL);
1368
1369 ifdebug(CACHE)
1370 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1371 convert_to_wallclock(cp->expiry_time),
1372 kref_read(&cp->ref), cp->flags);
1373 cache_get(cp);
1374 if (cache_check(cd, cp, NULL))
1375 /* cache_check does a cache_put on failure */
1376 seq_printf(m, "# ");
1377 else {
1378 if (cache_is_expired(cd, cp))
1379 seq_printf(m, "# ");
1380 cache_put(cp, cd);
1381 }
1382
1383 return cd->cache_show(m, cd, cp);
1384}
1385
1386static const struct seq_operations cache_content_op = {
1387 .start = cache_seq_start,
1388 .next = cache_seq_next,
1389 .stop = cache_seq_stop,
1390 .show = c_show,
1391};
1392
1393static int content_open(struct inode *inode, struct file *file,
1394 struct cache_detail *cd)
1395{
1396 struct seq_file *seq;
1397 int err;
1398
1399 if (!cd || !try_module_get(cd->owner))
1400 return -EACCES;
1401
1402 err = seq_open(file, &cache_content_op);
1403 if (err) {
1404 module_put(cd->owner);
1405 return err;
1406 }
1407
1408 seq = file->private_data;
1409 seq->private = cd;
1410 return 0;
1411}
1412
1413static int content_release(struct inode *inode, struct file *file,
1414 struct cache_detail *cd)
1415{
1416 int ret = seq_release(inode, file);
1417 module_put(cd->owner);
1418 return ret;
1419}
1420
1421static int open_flush(struct inode *inode, struct file *file,
1422 struct cache_detail *cd)
1423{
1424 if (!cd || !try_module_get(cd->owner))
1425 return -EACCES;
1426 return nonseekable_open(inode, file);
1427}
1428
1429static int release_flush(struct inode *inode, struct file *file,
1430 struct cache_detail *cd)
1431{
1432 module_put(cd->owner);
1433 return 0;
1434}
1435
1436static ssize_t read_flush(struct file *file, char __user *buf,
1437 size_t count, loff_t *ppos,
1438 struct cache_detail *cd)
1439{
1440 char tbuf[22];
1441 size_t len;
1442
1443 len = snprintf(tbuf, sizeof(tbuf), "%lu\n",
1444 convert_to_wallclock(cd->flush_time));
1445 return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1446}
1447
1448static ssize_t write_flush(struct file *file, const char __user *buf,
1449 size_t count, loff_t *ppos,
1450 struct cache_detail *cd)
1451{
1452 char tbuf[20];
1453 char *ep;
1454 time_t now;
1455
1456 if (*ppos || count > sizeof(tbuf)-1)
1457 return -EINVAL;
1458 if (copy_from_user(tbuf, buf, count))
1459 return -EFAULT;
1460 tbuf[count] = 0;
1461 simple_strtoul(tbuf, &ep, 0);
1462 if (*ep && *ep != '\n')
1463 return -EINVAL;
1464 /* Note that while we check that 'buf' holds a valid number,
1465 * we always ignore the value and just flush everything.
1466 * Making use of the number leads to races.
1467 */
1468
1469 now = seconds_since_boot();
1470 /* Always flush everything, so behave like cache_purge()
1471 * Do this by advancing flush_time to the current time,
1472 * or by one second if it has already reached the current time.
1473 * Newly added cache entries will always have ->last_refresh greater
1474 * that ->flush_time, so they don't get flushed prematurely.
1475 */
1476
1477 if (cd->flush_time >= now)
1478 now = cd->flush_time + 1;
1479
1480 cd->flush_time = now;
1481 cd->nextcheck = now;
1482 cache_flush();
1483
1484 *ppos += count;
1485 return count;
1486}
1487
1488static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1489 size_t count, loff_t *ppos)
1490{
1491 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1492
1493 return cache_read(filp, buf, count, ppos, cd);
1494}
1495
1496static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1497 size_t count, loff_t *ppos)
1498{
1499 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1500
1501 return cache_write(filp, buf, count, ppos, cd);
1502}
1503
1504static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1505{
1506 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1507
1508 return cache_poll(filp, wait, cd);
1509}
1510
1511static long cache_ioctl_procfs(struct file *filp,
1512 unsigned int cmd, unsigned long arg)
1513{
1514 struct inode *inode = file_inode(filp);
1515 struct cache_detail *cd = PDE_DATA(inode);
1516
1517 return cache_ioctl(inode, filp, cmd, arg, cd);
1518}
1519
1520static int cache_open_procfs(struct inode *inode, struct file *filp)
1521{
1522 struct cache_detail *cd = PDE_DATA(inode);
1523
1524 return cache_open(inode, filp, cd);
1525}
1526
1527static int cache_release_procfs(struct inode *inode, struct file *filp)
1528{
1529 struct cache_detail *cd = PDE_DATA(inode);
1530
1531 return cache_release(inode, filp, cd);
1532}
1533
1534static const struct file_operations cache_file_operations_procfs = {
1535 .owner = THIS_MODULE,
1536 .llseek = no_llseek,
1537 .read = cache_read_procfs,
1538 .write = cache_write_procfs,
1539 .poll = cache_poll_procfs,
1540 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1541 .open = cache_open_procfs,
1542 .release = cache_release_procfs,
1543};
1544
1545static int content_open_procfs(struct inode *inode, struct file *filp)
1546{
1547 struct cache_detail *cd = PDE_DATA(inode);
1548
1549 return content_open(inode, filp, cd);
1550}
1551
1552static int content_release_procfs(struct inode *inode, struct file *filp)
1553{
1554 struct cache_detail *cd = PDE_DATA(inode);
1555
1556 return content_release(inode, filp, cd);
1557}
1558
1559static const struct file_operations content_file_operations_procfs = {
1560 .open = content_open_procfs,
1561 .read = seq_read,
1562 .llseek = seq_lseek,
1563 .release = content_release_procfs,
1564};
1565
1566static int open_flush_procfs(struct inode *inode, struct file *filp)
1567{
1568 struct cache_detail *cd = PDE_DATA(inode);
1569
1570 return open_flush(inode, filp, cd);
1571}
1572
1573static int release_flush_procfs(struct inode *inode, struct file *filp)
1574{
1575 struct cache_detail *cd = PDE_DATA(inode);
1576
1577 return release_flush(inode, filp, cd);
1578}
1579
1580static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1581 size_t count, loff_t *ppos)
1582{
1583 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1584
1585 return read_flush(filp, buf, count, ppos, cd);
1586}
1587
1588static ssize_t write_flush_procfs(struct file *filp,
1589 const char __user *buf,
1590 size_t count, loff_t *ppos)
1591{
1592 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1593
1594 return write_flush(filp, buf, count, ppos, cd);
1595}
1596
1597static const struct file_operations cache_flush_operations_procfs = {
1598 .open = open_flush_procfs,
1599 .read = read_flush_procfs,
1600 .write = write_flush_procfs,
1601 .release = release_flush_procfs,
1602 .llseek = no_llseek,
1603};
1604
1605static void remove_cache_proc_entries(struct cache_detail *cd)
1606{
1607 if (cd->procfs) {
1608 proc_remove(cd->procfs);
1609 cd->procfs = NULL;
1610 }
1611}
1612
1613#ifdef CONFIG_PROC_FS
1614static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1615{
1616 struct proc_dir_entry *p;
1617 struct sunrpc_net *sn;
1618
1619 sn = net_generic(net, sunrpc_net_id);
1620 cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1621 if (cd->procfs == NULL)
1622 goto out_nomem;
1623
1624 p = proc_create_data("flush", S_IFREG | 0600,
1625 cd->procfs, &cache_flush_operations_procfs, cd);
1626 if (p == NULL)
1627 goto out_nomem;
1628
1629 if (cd->cache_request || cd->cache_parse) {
1630 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1631 &cache_file_operations_procfs, cd);
1632 if (p == NULL)
1633 goto out_nomem;
1634 }
1635 if (cd->cache_show) {
1636 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1637 &content_file_operations_procfs, cd);
1638 if (p == NULL)
1639 goto out_nomem;
1640 }
1641 return 0;
1642out_nomem:
1643 remove_cache_proc_entries(cd);
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);
1673 sunrpc_destroy_cache_detail(cd);
1674}
1675EXPORT_SYMBOL_GPL(cache_unregister_net);
1676
1677struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1678{
1679 struct cache_detail *cd;
1680 int i;
1681
1682 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1683 if (cd == NULL)
1684 return ERR_PTR(-ENOMEM);
1685
1686 cd->hash_table = kzalloc(cd->hash_size * sizeof(struct hlist_head),
1687 GFP_KERNEL);
1688 if (cd->hash_table == NULL) {
1689 kfree(cd);
1690 return ERR_PTR(-ENOMEM);
1691 }
1692
1693 for (i = 0; i < cd->hash_size; i++)
1694 INIT_HLIST_HEAD(&cd->hash_table[i]);
1695 cd->net = net;
1696 return cd;
1697}
1698EXPORT_SYMBOL_GPL(cache_create_net);
1699
1700void cache_destroy_net(struct cache_detail *cd, struct net *net)
1701{
1702 kfree(cd->hash_table);
1703 kfree(cd);
1704}
1705EXPORT_SYMBOL_GPL(cache_destroy_net);
1706
1707static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1708 size_t count, loff_t *ppos)
1709{
1710 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1711
1712 return cache_read(filp, buf, count, ppos, cd);
1713}
1714
1715static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1716 size_t count, loff_t *ppos)
1717{
1718 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1719
1720 return cache_write(filp, buf, count, ppos, cd);
1721}
1722
1723static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1724{
1725 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1726
1727 return cache_poll(filp, wait, cd);
1728}
1729
1730static long cache_ioctl_pipefs(struct file *filp,
1731 unsigned int cmd, unsigned long arg)
1732{
1733 struct inode *inode = file_inode(filp);
1734 struct cache_detail *cd = RPC_I(inode)->private;
1735
1736 return cache_ioctl(inode, filp, cmd, arg, cd);
1737}
1738
1739static int cache_open_pipefs(struct inode *inode, struct file *filp)
1740{
1741 struct cache_detail *cd = RPC_I(inode)->private;
1742
1743 return cache_open(inode, filp, cd);
1744}
1745
1746static int cache_release_pipefs(struct inode *inode, struct file *filp)
1747{
1748 struct cache_detail *cd = RPC_I(inode)->private;
1749
1750 return cache_release(inode, filp, cd);
1751}
1752
1753const struct file_operations cache_file_operations_pipefs = {
1754 .owner = THIS_MODULE,
1755 .llseek = no_llseek,
1756 .read = cache_read_pipefs,
1757 .write = cache_write_pipefs,
1758 .poll = cache_poll_pipefs,
1759 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1760 .open = cache_open_pipefs,
1761 .release = cache_release_pipefs,
1762};
1763
1764static int content_open_pipefs(struct inode *inode, struct file *filp)
1765{
1766 struct cache_detail *cd = RPC_I(inode)->private;
1767
1768 return content_open(inode, filp, cd);
1769}
1770
1771static int content_release_pipefs(struct inode *inode, struct file *filp)
1772{
1773 struct cache_detail *cd = RPC_I(inode)->private;
1774
1775 return content_release(inode, filp, cd);
1776}
1777
1778const struct file_operations content_file_operations_pipefs = {
1779 .open = content_open_pipefs,
1780 .read = seq_read,
1781 .llseek = seq_lseek,
1782 .release = content_release_pipefs,
1783};
1784
1785static int open_flush_pipefs(struct inode *inode, struct file *filp)
1786{
1787 struct cache_detail *cd = RPC_I(inode)->private;
1788
1789 return open_flush(inode, filp, cd);
1790}
1791
1792static int release_flush_pipefs(struct inode *inode, struct file *filp)
1793{
1794 struct cache_detail *cd = RPC_I(inode)->private;
1795
1796 return release_flush(inode, filp, cd);
1797}
1798
1799static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1800 size_t count, loff_t *ppos)
1801{
1802 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1803
1804 return read_flush(filp, buf, count, ppos, cd);
1805}
1806
1807static ssize_t write_flush_pipefs(struct file *filp,
1808 const char __user *buf,
1809 size_t count, loff_t *ppos)
1810{
1811 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1812
1813 return write_flush(filp, buf, count, ppos, cd);
1814}
1815
1816const struct file_operations cache_flush_operations_pipefs = {
1817 .open = open_flush_pipefs,
1818 .read = read_flush_pipefs,
1819 .write = write_flush_pipefs,
1820 .release = release_flush_pipefs,
1821 .llseek = no_llseek,
1822};
1823
1824int sunrpc_cache_register_pipefs(struct dentry *parent,
1825 const char *name, umode_t umode,
1826 struct cache_detail *cd)
1827{
1828 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1829 if (IS_ERR(dir))
1830 return PTR_ERR(dir);
1831 cd->pipefs = dir;
1832 return 0;
1833}
1834EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1835
1836void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1837{
1838 if (cd->pipefs) {
1839 rpc_remove_cache_dir(cd->pipefs);
1840 cd->pipefs = NULL;
1841 }
1842}
1843EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1844
1845void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1846{
1847 write_lock(&cd->hash_lock);
1848 if (!hlist_unhashed(&h->cache_list)){
1849 hlist_del_init(&h->cache_list);
1850 cd->entries--;
1851 write_unlock(&cd->hash_lock);
1852 cache_put(h, cd);
1853 } else
1854 write_unlock(&cd->hash_lock);
1855}
1856EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);