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 <trace/events/sunrpc.h>
36
37#include "netns.h"
38#include "fail.h"
39
40#define RPCDBG_FACILITY RPCDBG_CACHE
41
42static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
43static void cache_revisit_request(struct cache_head *item);
44
45static void cache_init(struct cache_head *h, struct cache_detail *detail)
46{
47 time64_t now = seconds_since_boot();
48 INIT_HLIST_NODE(&h->cache_list);
49 h->flags = 0;
50 kref_init(&h->ref);
51 h->expiry_time = now + CACHE_NEW_EXPIRY;
52 if (now <= detail->flush_time)
53 /* ensure it isn't already expired */
54 now = detail->flush_time + 1;
55 h->last_refresh = now;
56}
57
58static void cache_fresh_unlocked(struct cache_head *head,
59 struct cache_detail *detail);
60
61static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
62 struct cache_head *key,
63 int hash)
64{
65 struct hlist_head *head = &detail->hash_table[hash];
66 struct cache_head *tmp;
67
68 rcu_read_lock();
69 hlist_for_each_entry_rcu(tmp, head, cache_list) {
70 if (!detail->match(tmp, key))
71 continue;
72 if (test_bit(CACHE_VALID, &tmp->flags) &&
73 cache_is_expired(detail, tmp))
74 continue;
75 tmp = cache_get_rcu(tmp);
76 rcu_read_unlock();
77 return tmp;
78 }
79 rcu_read_unlock();
80 return NULL;
81}
82
83static void sunrpc_begin_cache_remove_entry(struct cache_head *ch,
84 struct cache_detail *cd)
85{
86 /* Must be called under cd->hash_lock */
87 hlist_del_init_rcu(&ch->cache_list);
88 set_bit(CACHE_CLEANED, &ch->flags);
89 cd->entries --;
90}
91
92static void sunrpc_end_cache_remove_entry(struct cache_head *ch,
93 struct cache_detail *cd)
94{
95 cache_fresh_unlocked(ch, cd);
96 cache_put(ch, cd);
97}
98
99static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
100 struct cache_head *key,
101 int hash)
102{
103 struct cache_head *new, *tmp, *freeme = NULL;
104 struct hlist_head *head = &detail->hash_table[hash];
105
106 new = detail->alloc();
107 if (!new)
108 return NULL;
109 /* must fully initialise 'new', else
110 * we might get lose if we need to
111 * cache_put it soon.
112 */
113 cache_init(new, detail);
114 detail->init(new, key);
115
116 spin_lock(&detail->hash_lock);
117
118 /* check if entry appeared while we slept */
119 hlist_for_each_entry_rcu(tmp, head, cache_list,
120 lockdep_is_held(&detail->hash_lock)) {
121 if (!detail->match(tmp, key))
122 continue;
123 if (test_bit(CACHE_VALID, &tmp->flags) &&
124 cache_is_expired(detail, tmp)) {
125 sunrpc_begin_cache_remove_entry(tmp, detail);
126 trace_cache_entry_expired(detail, tmp);
127 freeme = tmp;
128 break;
129 }
130 cache_get(tmp);
131 spin_unlock(&detail->hash_lock);
132 cache_put(new, detail);
133 return tmp;
134 }
135
136 hlist_add_head_rcu(&new->cache_list, head);
137 detail->entries++;
138 cache_get(new);
139 spin_unlock(&detail->hash_lock);
140
141 if (freeme)
142 sunrpc_end_cache_remove_entry(freeme, detail);
143 return new;
144}
145
146struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
147 struct cache_head *key, int hash)
148{
149 struct cache_head *ret;
150
151 ret = sunrpc_cache_find_rcu(detail, key, hash);
152 if (ret)
153 return ret;
154 /* Didn't find anything, insert an empty entry */
155 return sunrpc_cache_add_entry(detail, key, hash);
156}
157EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
158
159static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
160
161static void cache_fresh_locked(struct cache_head *head, time64_t expiry,
162 struct cache_detail *detail)
163{
164 time64_t now = seconds_since_boot();
165 if (now <= detail->flush_time)
166 /* ensure it isn't immediately treated as expired */
167 now = detail->flush_time + 1;
168 head->expiry_time = expiry;
169 head->last_refresh = now;
170 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
171 set_bit(CACHE_VALID, &head->flags);
172}
173
174static void cache_fresh_unlocked(struct cache_head *head,
175 struct cache_detail *detail)
176{
177 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
178 cache_revisit_request(head);
179 cache_dequeue(detail, head);
180 }
181}
182
183static void cache_make_negative(struct cache_detail *detail,
184 struct cache_head *h)
185{
186 set_bit(CACHE_NEGATIVE, &h->flags);
187 trace_cache_entry_make_negative(detail, h);
188}
189
190static void cache_entry_update(struct cache_detail *detail,
191 struct cache_head *h,
192 struct cache_head *new)
193{
194 if (!test_bit(CACHE_NEGATIVE, &new->flags)) {
195 detail->update(h, new);
196 trace_cache_entry_update(detail, h);
197 } else {
198 cache_make_negative(detail, h);
199 }
200}
201
202struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
203 struct cache_head *new, struct cache_head *old, int hash)
204{
205 /* The 'old' entry is to be replaced by 'new'.
206 * If 'old' is not VALID, we update it directly,
207 * otherwise we need to replace it
208 */
209 struct cache_head *tmp;
210
211 if (!test_bit(CACHE_VALID, &old->flags)) {
212 spin_lock(&detail->hash_lock);
213 if (!test_bit(CACHE_VALID, &old->flags)) {
214 cache_entry_update(detail, old, new);
215 cache_fresh_locked(old, new->expiry_time, detail);
216 spin_unlock(&detail->hash_lock);
217 cache_fresh_unlocked(old, detail);
218 return old;
219 }
220 spin_unlock(&detail->hash_lock);
221 }
222 /* We need to insert a new entry */
223 tmp = detail->alloc();
224 if (!tmp) {
225 cache_put(old, detail);
226 return NULL;
227 }
228 cache_init(tmp, detail);
229 detail->init(tmp, old);
230
231 spin_lock(&detail->hash_lock);
232 cache_entry_update(detail, tmp, new);
233 hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
234 detail->entries++;
235 cache_get(tmp);
236 cache_fresh_locked(tmp, new->expiry_time, detail);
237 cache_fresh_locked(old, 0, detail);
238 spin_unlock(&detail->hash_lock);
239 cache_fresh_unlocked(tmp, detail);
240 cache_fresh_unlocked(old, detail);
241 cache_put(old, detail);
242 return tmp;
243}
244EXPORT_SYMBOL_GPL(sunrpc_cache_update);
245
246static inline int cache_is_valid(struct cache_head *h)
247{
248 if (!test_bit(CACHE_VALID, &h->flags))
249 return -EAGAIN;
250 else {
251 /* entry is valid */
252 if (test_bit(CACHE_NEGATIVE, &h->flags))
253 return -ENOENT;
254 else {
255 /*
256 * In combination with write barrier in
257 * sunrpc_cache_update, ensures that anyone
258 * using the cache entry after this sees the
259 * updated contents:
260 */
261 smp_rmb();
262 return 0;
263 }
264 }
265}
266
267static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
268{
269 int rv;
270
271 spin_lock(&detail->hash_lock);
272 rv = cache_is_valid(h);
273 if (rv == -EAGAIN) {
274 cache_make_negative(detail, h);
275 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
276 detail);
277 rv = -ENOENT;
278 }
279 spin_unlock(&detail->hash_lock);
280 cache_fresh_unlocked(h, detail);
281 return rv;
282}
283
284/*
285 * This is the generic cache management routine for all
286 * the authentication caches.
287 * It checks the currency of a cache item and will (later)
288 * initiate an upcall to fill it if needed.
289 *
290 *
291 * Returns 0 if the cache_head can be used, or cache_puts it and returns
292 * -EAGAIN if upcall is pending and request has been queued
293 * -ETIMEDOUT if upcall failed or request could not be queue or
294 * upcall completed but item is still invalid (implying that
295 * the cache item has been replaced with a newer one).
296 * -ENOENT if cache entry was negative
297 */
298int cache_check(struct cache_detail *detail,
299 struct cache_head *h, struct cache_req *rqstp)
300{
301 int rv;
302 time64_t refresh_age, age;
303
304 /* First decide return status as best we can */
305 rv = cache_is_valid(h);
306
307 /* now see if we want to start an upcall */
308 refresh_age = (h->expiry_time - h->last_refresh);
309 age = seconds_since_boot() - h->last_refresh;
310
311 if (rqstp == NULL) {
312 if (rv == -EAGAIN)
313 rv = -ENOENT;
314 } else if (rv == -EAGAIN ||
315 (h->expiry_time != 0 && age > refresh_age/2)) {
316 dprintk("RPC: Want update, refage=%lld, age=%lld\n",
317 refresh_age, age);
318 switch (detail->cache_upcall(detail, h)) {
319 case -EINVAL:
320 rv = try_to_negate_entry(detail, h);
321 break;
322 case -EAGAIN:
323 cache_fresh_unlocked(h, detail);
324 break;
325 }
326 }
327
328 if (rv == -EAGAIN) {
329 if (!cache_defer_req(rqstp, h)) {
330 /*
331 * Request was not deferred; handle it as best
332 * we can ourselves:
333 */
334 rv = cache_is_valid(h);
335 if (rv == -EAGAIN)
336 rv = -ETIMEDOUT;
337 }
338 }
339 if (rv)
340 cache_put(h, detail);
341 return rv;
342}
343EXPORT_SYMBOL_GPL(cache_check);
344
345/*
346 * caches need to be periodically cleaned.
347 * For this we maintain a list of cache_detail and
348 * a current pointer into that list and into the table
349 * for that entry.
350 *
351 * Each time cache_clean is called it finds the next non-empty entry
352 * in the current table and walks the list in that entry
353 * looking for entries that can be removed.
354 *
355 * An entry gets removed if:
356 * - The expiry is before current time
357 * - The last_refresh time is before the flush_time for that cache
358 *
359 * later we might drop old entries with non-NEVER expiry if that table
360 * is getting 'full' for some definition of 'full'
361 *
362 * The question of "how often to scan a table" is an interesting one
363 * and is answered in part by the use of the "nextcheck" field in the
364 * cache_detail.
365 * When a scan of a table begins, the nextcheck field is set to a time
366 * that is well into the future.
367 * While scanning, if an expiry time is found that is earlier than the
368 * current nextcheck time, nextcheck is set to that expiry time.
369 * If the flush_time is ever set to a time earlier than the nextcheck
370 * time, the nextcheck time is then set to that flush_time.
371 *
372 * A table is then only scanned if the current time is at least
373 * the nextcheck time.
374 *
375 */
376
377static LIST_HEAD(cache_list);
378static DEFINE_SPINLOCK(cache_list_lock);
379static struct cache_detail *current_detail;
380static int current_index;
381
382static void do_cache_clean(struct work_struct *work);
383static struct delayed_work cache_cleaner;
384
385void sunrpc_init_cache_detail(struct cache_detail *cd)
386{
387 spin_lock_init(&cd->hash_lock);
388 INIT_LIST_HEAD(&cd->queue);
389 spin_lock(&cache_list_lock);
390 cd->nextcheck = 0;
391 cd->entries = 0;
392 atomic_set(&cd->writers, 0);
393 cd->last_close = 0;
394 cd->last_warn = -1;
395 list_add(&cd->others, &cache_list);
396 spin_unlock(&cache_list_lock);
397
398 /* start the cleaning process */
399 queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
400}
401EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
402
403void sunrpc_destroy_cache_detail(struct cache_detail *cd)
404{
405 cache_purge(cd);
406 spin_lock(&cache_list_lock);
407 spin_lock(&cd->hash_lock);
408 if (current_detail == cd)
409 current_detail = NULL;
410 list_del_init(&cd->others);
411 spin_unlock(&cd->hash_lock);
412 spin_unlock(&cache_list_lock);
413 if (list_empty(&cache_list)) {
414 /* module must be being unloaded so its safe to kill the worker */
415 cancel_delayed_work_sync(&cache_cleaner);
416 }
417}
418EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
419
420/* clean cache tries to find something to clean
421 * and cleans it.
422 * It returns 1 if it cleaned something,
423 * 0 if it didn't find anything this time
424 * -1 if it fell off the end of the list.
425 */
426static int cache_clean(void)
427{
428 int rv = 0;
429 struct list_head *next;
430
431 spin_lock(&cache_list_lock);
432
433 /* find a suitable table if we don't already have one */
434 while (current_detail == NULL ||
435 current_index >= current_detail->hash_size) {
436 if (current_detail)
437 next = current_detail->others.next;
438 else
439 next = cache_list.next;
440 if (next == &cache_list) {
441 current_detail = NULL;
442 spin_unlock(&cache_list_lock);
443 return -1;
444 }
445 current_detail = list_entry(next, struct cache_detail, others);
446 if (current_detail->nextcheck > seconds_since_boot())
447 current_index = current_detail->hash_size;
448 else {
449 current_index = 0;
450 current_detail->nextcheck = seconds_since_boot()+30*60;
451 }
452 }
453
454 /* find a non-empty bucket in the table */
455 while (current_detail &&
456 current_index < current_detail->hash_size &&
457 hlist_empty(¤t_detail->hash_table[current_index]))
458 current_index++;
459
460 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
461
462 if (current_detail && current_index < current_detail->hash_size) {
463 struct cache_head *ch = NULL;
464 struct cache_detail *d;
465 struct hlist_head *head;
466 struct hlist_node *tmp;
467
468 spin_lock(¤t_detail->hash_lock);
469
470 /* Ok, now to clean this strand */
471
472 head = ¤t_detail->hash_table[current_index];
473 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
474 if (current_detail->nextcheck > ch->expiry_time)
475 current_detail->nextcheck = ch->expiry_time+1;
476 if (!cache_is_expired(current_detail, ch))
477 continue;
478
479 sunrpc_begin_cache_remove_entry(ch, current_detail);
480 trace_cache_entry_expired(current_detail, ch);
481 rv = 1;
482 break;
483 }
484
485 spin_unlock(¤t_detail->hash_lock);
486 d = current_detail;
487 if (!ch)
488 current_index ++;
489 spin_unlock(&cache_list_lock);
490 if (ch)
491 sunrpc_end_cache_remove_entry(ch, d);
492 } else
493 spin_unlock(&cache_list_lock);
494
495 return rv;
496}
497
498/*
499 * We want to regularly clean the cache, so we need to schedule some work ...
500 */
501static void do_cache_clean(struct work_struct *work)
502{
503 int delay;
504
505 if (list_empty(&cache_list))
506 return;
507
508 if (cache_clean() == -1)
509 delay = round_jiffies_relative(30*HZ);
510 else
511 delay = 5;
512
513 queue_delayed_work(system_power_efficient_wq, &cache_cleaner, delay);
514}
515
516
517/*
518 * Clean all caches promptly. This just calls cache_clean
519 * repeatedly until we are sure that every cache has had a chance to
520 * be fully cleaned
521 */
522void cache_flush(void)
523{
524 while (cache_clean() != -1)
525 cond_resched();
526 while (cache_clean() != -1)
527 cond_resched();
528}
529EXPORT_SYMBOL_GPL(cache_flush);
530
531void cache_purge(struct cache_detail *detail)
532{
533 struct cache_head *ch = NULL;
534 struct hlist_head *head = NULL;
535 int i = 0;
536
537 spin_lock(&detail->hash_lock);
538 if (!detail->entries) {
539 spin_unlock(&detail->hash_lock);
540 return;
541 }
542
543 dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
544 for (i = 0; i < detail->hash_size; i++) {
545 head = &detail->hash_table[i];
546 while (!hlist_empty(head)) {
547 ch = hlist_entry(head->first, struct cache_head,
548 cache_list);
549 sunrpc_begin_cache_remove_entry(ch, detail);
550 spin_unlock(&detail->hash_lock);
551 sunrpc_end_cache_remove_entry(ch, detail);
552 spin_lock(&detail->hash_lock);
553 }
554 }
555 spin_unlock(&detail->hash_lock);
556}
557EXPORT_SYMBOL_GPL(cache_purge);
558
559
560/*
561 * Deferral and Revisiting of Requests.
562 *
563 * If a cache lookup finds a pending entry, we
564 * need to defer the request and revisit it later.
565 * All deferred requests are stored in a hash table,
566 * indexed by "struct cache_head *".
567 * As it may be wasteful to store a whole request
568 * structure, we allow the request to provide a
569 * deferred form, which must contain a
570 * 'struct cache_deferred_req'
571 * This cache_deferred_req contains a method to allow
572 * it to be revisited when cache info is available
573 */
574
575#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
576#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
577
578#define DFR_MAX 300 /* ??? */
579
580static DEFINE_SPINLOCK(cache_defer_lock);
581static LIST_HEAD(cache_defer_list);
582static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
583static int cache_defer_cnt;
584
585static void __unhash_deferred_req(struct cache_deferred_req *dreq)
586{
587 hlist_del_init(&dreq->hash);
588 if (!list_empty(&dreq->recent)) {
589 list_del_init(&dreq->recent);
590 cache_defer_cnt--;
591 }
592}
593
594static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
595{
596 int hash = DFR_HASH(item);
597
598 INIT_LIST_HEAD(&dreq->recent);
599 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
600}
601
602static void setup_deferral(struct cache_deferred_req *dreq,
603 struct cache_head *item,
604 int count_me)
605{
606
607 dreq->item = item;
608
609 spin_lock(&cache_defer_lock);
610
611 __hash_deferred_req(dreq, item);
612
613 if (count_me) {
614 cache_defer_cnt++;
615 list_add(&dreq->recent, &cache_defer_list);
616 }
617
618 spin_unlock(&cache_defer_lock);
619
620}
621
622struct thread_deferred_req {
623 struct cache_deferred_req handle;
624 struct completion completion;
625};
626
627static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
628{
629 struct thread_deferred_req *dr =
630 container_of(dreq, struct thread_deferred_req, handle);
631 complete(&dr->completion);
632}
633
634static void cache_wait_req(struct cache_req *req, struct cache_head *item)
635{
636 struct thread_deferred_req sleeper;
637 struct cache_deferred_req *dreq = &sleeper.handle;
638
639 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
640 dreq->revisit = cache_restart_thread;
641
642 setup_deferral(dreq, item, 0);
643
644 if (!test_bit(CACHE_PENDING, &item->flags) ||
645 wait_for_completion_interruptible_timeout(
646 &sleeper.completion, req->thread_wait) <= 0) {
647 /* The completion wasn't completed, so we need
648 * to clean up
649 */
650 spin_lock(&cache_defer_lock);
651 if (!hlist_unhashed(&sleeper.handle.hash)) {
652 __unhash_deferred_req(&sleeper.handle);
653 spin_unlock(&cache_defer_lock);
654 } else {
655 /* cache_revisit_request already removed
656 * this from the hash table, but hasn't
657 * called ->revisit yet. It will very soon
658 * and we need to wait for it.
659 */
660 spin_unlock(&cache_defer_lock);
661 wait_for_completion(&sleeper.completion);
662 }
663 }
664}
665
666static void cache_limit_defers(void)
667{
668 /* Make sure we haven't exceed the limit of allowed deferred
669 * requests.
670 */
671 struct cache_deferred_req *discard = NULL;
672
673 if (cache_defer_cnt <= DFR_MAX)
674 return;
675
676 spin_lock(&cache_defer_lock);
677
678 /* Consider removing either the first or the last */
679 if (cache_defer_cnt > DFR_MAX) {
680 if (get_random_u32_below(2))
681 discard = list_entry(cache_defer_list.next,
682 struct cache_deferred_req, recent);
683 else
684 discard = list_entry(cache_defer_list.prev,
685 struct cache_deferred_req, recent);
686 __unhash_deferred_req(discard);
687 }
688 spin_unlock(&cache_defer_lock);
689 if (discard)
690 discard->revisit(discard, 1);
691}
692
693#if IS_ENABLED(CONFIG_FAIL_SUNRPC)
694static inline bool cache_defer_immediately(void)
695{
696 return !fail_sunrpc.ignore_cache_wait &&
697 should_fail(&fail_sunrpc.attr, 1);
698}
699#else
700static inline bool cache_defer_immediately(void)
701{
702 return false;
703}
704#endif
705
706/* Return true if and only if a deferred request is queued. */
707static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
708{
709 struct cache_deferred_req *dreq;
710
711 if (!cache_defer_immediately()) {
712 cache_wait_req(req, item);
713 if (!test_bit(CACHE_PENDING, &item->flags))
714 return false;
715 }
716
717 dreq = req->defer(req);
718 if (dreq == NULL)
719 return false;
720 setup_deferral(dreq, item, 1);
721 if (!test_bit(CACHE_PENDING, &item->flags))
722 /* Bit could have been cleared before we managed to
723 * set up the deferral, so need to revisit just in case
724 */
725 cache_revisit_request(item);
726
727 cache_limit_defers();
728 return true;
729}
730
731static void cache_revisit_request(struct cache_head *item)
732{
733 struct cache_deferred_req *dreq;
734 struct list_head pending;
735 struct hlist_node *tmp;
736 int hash = DFR_HASH(item);
737
738 INIT_LIST_HEAD(&pending);
739 spin_lock(&cache_defer_lock);
740
741 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
742 if (dreq->item == item) {
743 __unhash_deferred_req(dreq);
744 list_add(&dreq->recent, &pending);
745 }
746
747 spin_unlock(&cache_defer_lock);
748
749 while (!list_empty(&pending)) {
750 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
751 list_del_init(&dreq->recent);
752 dreq->revisit(dreq, 0);
753 }
754}
755
756void cache_clean_deferred(void *owner)
757{
758 struct cache_deferred_req *dreq, *tmp;
759 struct list_head pending;
760
761
762 INIT_LIST_HEAD(&pending);
763 spin_lock(&cache_defer_lock);
764
765 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
766 if (dreq->owner == owner) {
767 __unhash_deferred_req(dreq);
768 list_add(&dreq->recent, &pending);
769 }
770 }
771 spin_unlock(&cache_defer_lock);
772
773 while (!list_empty(&pending)) {
774 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
775 list_del_init(&dreq->recent);
776 dreq->revisit(dreq, 1);
777 }
778}
779
780/*
781 * communicate with user-space
782 *
783 * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
784 * On read, you get a full request, or block.
785 * On write, an update request is processed.
786 * Poll works if anything to read, and always allows write.
787 *
788 * Implemented by linked list of requests. Each open file has
789 * a ->private that also exists in this list. New requests are added
790 * to the end and may wakeup and preceding readers.
791 * New readers are added to the head. If, on read, an item is found with
792 * CACHE_UPCALLING clear, we free it from the list.
793 *
794 */
795
796static DEFINE_SPINLOCK(queue_lock);
797
798struct cache_queue {
799 struct list_head list;
800 int reader; /* if 0, then request */
801};
802struct cache_request {
803 struct cache_queue q;
804 struct cache_head *item;
805 char * buf;
806 int len;
807 int readers;
808};
809struct cache_reader {
810 struct cache_queue q;
811 int offset; /* if non-0, we have a refcnt on next request */
812};
813
814static int cache_request(struct cache_detail *detail,
815 struct cache_request *crq)
816{
817 char *bp = crq->buf;
818 int len = PAGE_SIZE;
819
820 detail->cache_request(detail, crq->item, &bp, &len);
821 if (len < 0)
822 return -E2BIG;
823 return PAGE_SIZE - len;
824}
825
826static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
827 loff_t *ppos, struct cache_detail *cd)
828{
829 struct cache_reader *rp = filp->private_data;
830 struct cache_request *rq;
831 struct inode *inode = file_inode(filp);
832 int err;
833
834 if (count == 0)
835 return 0;
836
837 inode_lock(inode); /* protect against multiple concurrent
838 * readers on this file */
839 again:
840 spin_lock(&queue_lock);
841 /* need to find next request */
842 while (rp->q.list.next != &cd->queue &&
843 list_entry(rp->q.list.next, struct cache_queue, list)
844 ->reader) {
845 struct list_head *next = rp->q.list.next;
846 list_move(&rp->q.list, next);
847 }
848 if (rp->q.list.next == &cd->queue) {
849 spin_unlock(&queue_lock);
850 inode_unlock(inode);
851 WARN_ON_ONCE(rp->offset);
852 return 0;
853 }
854 rq = container_of(rp->q.list.next, struct cache_request, q.list);
855 WARN_ON_ONCE(rq->q.reader);
856 if (rp->offset == 0)
857 rq->readers++;
858 spin_unlock(&queue_lock);
859
860 if (rq->len == 0) {
861 err = cache_request(cd, rq);
862 if (err < 0)
863 goto out;
864 rq->len = err;
865 }
866
867 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
868 err = -EAGAIN;
869 spin_lock(&queue_lock);
870 list_move(&rp->q.list, &rq->q.list);
871 spin_unlock(&queue_lock);
872 } else {
873 if (rp->offset + count > rq->len)
874 count = rq->len - rp->offset;
875 err = -EFAULT;
876 if (copy_to_user(buf, rq->buf + rp->offset, count))
877 goto out;
878 rp->offset += count;
879 if (rp->offset >= rq->len) {
880 rp->offset = 0;
881 spin_lock(&queue_lock);
882 list_move(&rp->q.list, &rq->q.list);
883 spin_unlock(&queue_lock);
884 }
885 err = 0;
886 }
887 out:
888 if (rp->offset == 0) {
889 /* need to release rq */
890 spin_lock(&queue_lock);
891 rq->readers--;
892 if (rq->readers == 0 &&
893 !test_bit(CACHE_PENDING, &rq->item->flags)) {
894 list_del(&rq->q.list);
895 spin_unlock(&queue_lock);
896 cache_put(rq->item, cd);
897 kfree(rq->buf);
898 kfree(rq);
899 } else
900 spin_unlock(&queue_lock);
901 }
902 if (err == -EAGAIN)
903 goto again;
904 inode_unlock(inode);
905 return err ? err : count;
906}
907
908static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
909 size_t count, struct cache_detail *cd)
910{
911 ssize_t ret;
912
913 if (count == 0)
914 return -EINVAL;
915 if (copy_from_user(kaddr, buf, count))
916 return -EFAULT;
917 kaddr[count] = '\0';
918 ret = cd->cache_parse(cd, kaddr, count);
919 if (!ret)
920 ret = count;
921 return ret;
922}
923
924static ssize_t cache_downcall(struct address_space *mapping,
925 const char __user *buf,
926 size_t count, struct cache_detail *cd)
927{
928 char *write_buf;
929 ssize_t ret = -ENOMEM;
930
931 if (count >= 32768) { /* 32k is max userland buffer, lets check anyway */
932 ret = -EINVAL;
933 goto out;
934 }
935
936 write_buf = kvmalloc(count + 1, GFP_KERNEL);
937 if (!write_buf)
938 goto out;
939
940 ret = cache_do_downcall(write_buf, buf, count, cd);
941 kvfree(write_buf);
942out:
943 return ret;
944}
945
946static ssize_t cache_write(struct file *filp, const char __user *buf,
947 size_t count, loff_t *ppos,
948 struct cache_detail *cd)
949{
950 struct address_space *mapping = filp->f_mapping;
951 struct inode *inode = file_inode(filp);
952 ssize_t ret = -EINVAL;
953
954 if (!cd->cache_parse)
955 goto out;
956
957 inode_lock(inode);
958 ret = cache_downcall(mapping, buf, count, cd);
959 inode_unlock(inode);
960out:
961 return ret;
962}
963
964static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
965
966static __poll_t cache_poll(struct file *filp, poll_table *wait,
967 struct cache_detail *cd)
968{
969 __poll_t mask;
970 struct cache_reader *rp = filp->private_data;
971 struct cache_queue *cq;
972
973 poll_wait(filp, &queue_wait, wait);
974
975 /* alway allow write */
976 mask = EPOLLOUT | EPOLLWRNORM;
977
978 if (!rp)
979 return mask;
980
981 spin_lock(&queue_lock);
982
983 for (cq= &rp->q; &cq->list != &cd->queue;
984 cq = list_entry(cq->list.next, struct cache_queue, list))
985 if (!cq->reader) {
986 mask |= EPOLLIN | EPOLLRDNORM;
987 break;
988 }
989 spin_unlock(&queue_lock);
990 return mask;
991}
992
993static int cache_ioctl(struct inode *ino, struct file *filp,
994 unsigned int cmd, unsigned long arg,
995 struct cache_detail *cd)
996{
997 int len = 0;
998 struct cache_reader *rp = filp->private_data;
999 struct cache_queue *cq;
1000
1001 if (cmd != FIONREAD || !rp)
1002 return -EINVAL;
1003
1004 spin_lock(&queue_lock);
1005
1006 /* only find the length remaining in current request,
1007 * or the length of the next request
1008 */
1009 for (cq= &rp->q; &cq->list != &cd->queue;
1010 cq = list_entry(cq->list.next, struct cache_queue, list))
1011 if (!cq->reader) {
1012 struct cache_request *cr =
1013 container_of(cq, struct cache_request, q);
1014 len = cr->len - rp->offset;
1015 break;
1016 }
1017 spin_unlock(&queue_lock);
1018
1019 return put_user(len, (int __user *)arg);
1020}
1021
1022static int cache_open(struct inode *inode, struct file *filp,
1023 struct cache_detail *cd)
1024{
1025 struct cache_reader *rp = NULL;
1026
1027 if (!cd || !try_module_get(cd->owner))
1028 return -EACCES;
1029 nonseekable_open(inode, filp);
1030 if (filp->f_mode & FMODE_READ) {
1031 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1032 if (!rp) {
1033 module_put(cd->owner);
1034 return -ENOMEM;
1035 }
1036 rp->offset = 0;
1037 rp->q.reader = 1;
1038
1039 spin_lock(&queue_lock);
1040 list_add(&rp->q.list, &cd->queue);
1041 spin_unlock(&queue_lock);
1042 }
1043 if (filp->f_mode & FMODE_WRITE)
1044 atomic_inc(&cd->writers);
1045 filp->private_data = rp;
1046 return 0;
1047}
1048
1049static int cache_release(struct inode *inode, struct file *filp,
1050 struct cache_detail *cd)
1051{
1052 struct cache_reader *rp = filp->private_data;
1053
1054 if (rp) {
1055 spin_lock(&queue_lock);
1056 if (rp->offset) {
1057 struct cache_queue *cq;
1058 for (cq= &rp->q; &cq->list != &cd->queue;
1059 cq = list_entry(cq->list.next, struct cache_queue, list))
1060 if (!cq->reader) {
1061 container_of(cq, struct cache_request, q)
1062 ->readers--;
1063 break;
1064 }
1065 rp->offset = 0;
1066 }
1067 list_del(&rp->q.list);
1068 spin_unlock(&queue_lock);
1069
1070 filp->private_data = NULL;
1071 kfree(rp);
1072
1073 }
1074 if (filp->f_mode & FMODE_WRITE) {
1075 atomic_dec(&cd->writers);
1076 cd->last_close = seconds_since_boot();
1077 }
1078 module_put(cd->owner);
1079 return 0;
1080}
1081
1082
1083
1084static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1085{
1086 struct cache_queue *cq, *tmp;
1087 struct cache_request *cr;
1088 struct list_head dequeued;
1089
1090 INIT_LIST_HEAD(&dequeued);
1091 spin_lock(&queue_lock);
1092 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1093 if (!cq->reader) {
1094 cr = container_of(cq, struct cache_request, q);
1095 if (cr->item != ch)
1096 continue;
1097 if (test_bit(CACHE_PENDING, &ch->flags))
1098 /* Lost a race and it is pending again */
1099 break;
1100 if (cr->readers != 0)
1101 continue;
1102 list_move(&cr->q.list, &dequeued);
1103 }
1104 spin_unlock(&queue_lock);
1105 while (!list_empty(&dequeued)) {
1106 cr = list_entry(dequeued.next, struct cache_request, q.list);
1107 list_del(&cr->q.list);
1108 cache_put(cr->item, detail);
1109 kfree(cr->buf);
1110 kfree(cr);
1111 }
1112}
1113
1114/*
1115 * Support routines for text-based upcalls.
1116 * Fields are separated by spaces.
1117 * Fields are either mangled to quote space tab newline slosh with slosh
1118 * or a hexified with a leading \x
1119 * Record is terminated with newline.
1120 *
1121 */
1122
1123void qword_add(char **bpp, int *lp, char *str)
1124{
1125 char *bp = *bpp;
1126 int len = *lp;
1127 int ret;
1128
1129 if (len < 0) return;
1130
1131 ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1132 if (ret >= len) {
1133 bp += len;
1134 len = -1;
1135 } else {
1136 bp += ret;
1137 len -= ret;
1138 *bp++ = ' ';
1139 len--;
1140 }
1141 *bpp = bp;
1142 *lp = len;
1143}
1144EXPORT_SYMBOL_GPL(qword_add);
1145
1146void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1147{
1148 char *bp = *bpp;
1149 int len = *lp;
1150
1151 if (len < 0) return;
1152
1153 if (len > 2) {
1154 *bp++ = '\\';
1155 *bp++ = 'x';
1156 len -= 2;
1157 while (blen && len >= 2) {
1158 bp = hex_byte_pack(bp, *buf++);
1159 len -= 2;
1160 blen--;
1161 }
1162 }
1163 if (blen || len<1) len = -1;
1164 else {
1165 *bp++ = ' ';
1166 len--;
1167 }
1168 *bpp = bp;
1169 *lp = len;
1170}
1171EXPORT_SYMBOL_GPL(qword_addhex);
1172
1173static void warn_no_listener(struct cache_detail *detail)
1174{
1175 if (detail->last_warn != detail->last_close) {
1176 detail->last_warn = detail->last_close;
1177 if (detail->warn_no_listener)
1178 detail->warn_no_listener(detail, detail->last_close != 0);
1179 }
1180}
1181
1182static bool cache_listeners_exist(struct cache_detail *detail)
1183{
1184 if (atomic_read(&detail->writers))
1185 return true;
1186 if (detail->last_close == 0)
1187 /* This cache was never opened */
1188 return false;
1189 if (detail->last_close < seconds_since_boot() - 30)
1190 /*
1191 * We allow for the possibility that someone might
1192 * restart a userspace daemon without restarting the
1193 * server; but after 30 seconds, we give up.
1194 */
1195 return false;
1196 return true;
1197}
1198
1199/*
1200 * register an upcall request to user-space and queue it up for read() by the
1201 * upcall daemon.
1202 *
1203 * Each request is at most one page long.
1204 */
1205static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1206{
1207 char *buf;
1208 struct cache_request *crq;
1209 int ret = 0;
1210
1211 if (test_bit(CACHE_CLEANED, &h->flags))
1212 /* Too late to make an upcall */
1213 return -EAGAIN;
1214
1215 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1216 if (!buf)
1217 return -EAGAIN;
1218
1219 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1220 if (!crq) {
1221 kfree(buf);
1222 return -EAGAIN;
1223 }
1224
1225 crq->q.reader = 0;
1226 crq->buf = buf;
1227 crq->len = 0;
1228 crq->readers = 0;
1229 spin_lock(&queue_lock);
1230 if (test_bit(CACHE_PENDING, &h->flags)) {
1231 crq->item = cache_get(h);
1232 list_add_tail(&crq->q.list, &detail->queue);
1233 trace_cache_entry_upcall(detail, h);
1234 } else
1235 /* Lost a race, no longer PENDING, so don't enqueue */
1236 ret = -EAGAIN;
1237 spin_unlock(&queue_lock);
1238 wake_up(&queue_wait);
1239 if (ret == -EAGAIN) {
1240 kfree(buf);
1241 kfree(crq);
1242 }
1243 return ret;
1244}
1245
1246int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1247{
1248 if (test_and_set_bit(CACHE_PENDING, &h->flags))
1249 return 0;
1250 return cache_pipe_upcall(detail, h);
1251}
1252EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1253
1254int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1255 struct cache_head *h)
1256{
1257 if (!cache_listeners_exist(detail)) {
1258 warn_no_listener(detail);
1259 trace_cache_entry_no_listener(detail, h);
1260 return -EINVAL;
1261 }
1262 return sunrpc_cache_pipe_upcall(detail, h);
1263}
1264EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1265
1266/*
1267 * parse a message from user-space and pass it
1268 * to an appropriate cache
1269 * Messages are, like requests, separated into fields by
1270 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1271 *
1272 * Message is
1273 * reply cachename expiry key ... content....
1274 *
1275 * key and content are both parsed by cache
1276 */
1277
1278int qword_get(char **bpp, char *dest, int bufsize)
1279{
1280 /* return bytes copied, or -1 on error */
1281 char *bp = *bpp;
1282 int len = 0;
1283
1284 while (*bp == ' ') bp++;
1285
1286 if (bp[0] == '\\' && bp[1] == 'x') {
1287 /* HEX STRING */
1288 bp += 2;
1289 while (len < bufsize - 1) {
1290 int h, l;
1291
1292 h = hex_to_bin(bp[0]);
1293 if (h < 0)
1294 break;
1295
1296 l = hex_to_bin(bp[1]);
1297 if (l < 0)
1298 break;
1299
1300 *dest++ = (h << 4) | l;
1301 bp += 2;
1302 len++;
1303 }
1304 } else {
1305 /* text with \nnn octal quoting */
1306 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1307 if (*bp == '\\' &&
1308 isodigit(bp[1]) && (bp[1] <= '3') &&
1309 isodigit(bp[2]) &&
1310 isodigit(bp[3])) {
1311 int byte = (*++bp -'0');
1312 bp++;
1313 byte = (byte << 3) | (*bp++ - '0');
1314 byte = (byte << 3) | (*bp++ - '0');
1315 *dest++ = byte;
1316 len++;
1317 } else {
1318 *dest++ = *bp++;
1319 len++;
1320 }
1321 }
1322 }
1323
1324 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1325 return -1;
1326 while (*bp == ' ') bp++;
1327 *bpp = bp;
1328 *dest = '\0';
1329 return len;
1330}
1331EXPORT_SYMBOL_GPL(qword_get);
1332
1333
1334/*
1335 * support /proc/net/rpc/$CACHENAME/content
1336 * as a seqfile.
1337 * We call ->cache_show passing NULL for the item to
1338 * get a header, then pass each real item in the cache
1339 */
1340
1341static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1342{
1343 loff_t n = *pos;
1344 unsigned int hash, entry;
1345 struct cache_head *ch;
1346 struct cache_detail *cd = m->private;
1347
1348 if (!n--)
1349 return SEQ_START_TOKEN;
1350 hash = n >> 32;
1351 entry = n & ((1LL<<32) - 1);
1352
1353 hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1354 if (!entry--)
1355 return ch;
1356 n &= ~((1LL<<32) - 1);
1357 do {
1358 hash++;
1359 n += 1LL<<32;
1360 } while(hash < cd->hash_size &&
1361 hlist_empty(&cd->hash_table[hash]));
1362 if (hash >= cd->hash_size)
1363 return NULL;
1364 *pos = n+1;
1365 return hlist_entry_safe(rcu_dereference_raw(
1366 hlist_first_rcu(&cd->hash_table[hash])),
1367 struct cache_head, cache_list);
1368}
1369
1370static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1371{
1372 struct cache_head *ch = p;
1373 int hash = (*pos >> 32);
1374 struct cache_detail *cd = m->private;
1375
1376 if (p == SEQ_START_TOKEN)
1377 hash = 0;
1378 else if (ch->cache_list.next == NULL) {
1379 hash++;
1380 *pos += 1LL<<32;
1381 } else {
1382 ++*pos;
1383 return hlist_entry_safe(rcu_dereference_raw(
1384 hlist_next_rcu(&ch->cache_list)),
1385 struct cache_head, cache_list);
1386 }
1387 *pos &= ~((1LL<<32) - 1);
1388 while (hash < cd->hash_size &&
1389 hlist_empty(&cd->hash_table[hash])) {
1390 hash++;
1391 *pos += 1LL<<32;
1392 }
1393 if (hash >= cd->hash_size)
1394 return NULL;
1395 ++*pos;
1396 return hlist_entry_safe(rcu_dereference_raw(
1397 hlist_first_rcu(&cd->hash_table[hash])),
1398 struct cache_head, cache_list);
1399}
1400
1401void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1402 __acquires(RCU)
1403{
1404 rcu_read_lock();
1405 return __cache_seq_start(m, pos);
1406}
1407EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1408
1409void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1410{
1411 return cache_seq_next(file, p, pos);
1412}
1413EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1414
1415void cache_seq_stop_rcu(struct seq_file *m, void *p)
1416 __releases(RCU)
1417{
1418 rcu_read_unlock();
1419}
1420EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1421
1422static int c_show(struct seq_file *m, void *p)
1423{
1424 struct cache_head *cp = p;
1425 struct cache_detail *cd = m->private;
1426
1427 if (p == SEQ_START_TOKEN)
1428 return cd->cache_show(m, cd, NULL);
1429
1430 ifdebug(CACHE)
1431 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1432 convert_to_wallclock(cp->expiry_time),
1433 kref_read(&cp->ref), cp->flags);
1434 cache_get(cp);
1435 if (cache_check(cd, cp, NULL))
1436 /* cache_check does a cache_put on failure */
1437 seq_puts(m, "# ");
1438 else {
1439 if (cache_is_expired(cd, cp))
1440 seq_puts(m, "# ");
1441 cache_put(cp, cd);
1442 }
1443
1444 return cd->cache_show(m, cd, cp);
1445}
1446
1447static const struct seq_operations cache_content_op = {
1448 .start = cache_seq_start_rcu,
1449 .next = cache_seq_next_rcu,
1450 .stop = cache_seq_stop_rcu,
1451 .show = c_show,
1452};
1453
1454static int content_open(struct inode *inode, struct file *file,
1455 struct cache_detail *cd)
1456{
1457 struct seq_file *seq;
1458 int err;
1459
1460 if (!cd || !try_module_get(cd->owner))
1461 return -EACCES;
1462
1463 err = seq_open(file, &cache_content_op);
1464 if (err) {
1465 module_put(cd->owner);
1466 return err;
1467 }
1468
1469 seq = file->private_data;
1470 seq->private = cd;
1471 return 0;
1472}
1473
1474static int content_release(struct inode *inode, struct file *file,
1475 struct cache_detail *cd)
1476{
1477 int ret = seq_release(inode, file);
1478 module_put(cd->owner);
1479 return ret;
1480}
1481
1482static int open_flush(struct inode *inode, struct file *file,
1483 struct cache_detail *cd)
1484{
1485 if (!cd || !try_module_get(cd->owner))
1486 return -EACCES;
1487 return nonseekable_open(inode, file);
1488}
1489
1490static int release_flush(struct inode *inode, struct file *file,
1491 struct cache_detail *cd)
1492{
1493 module_put(cd->owner);
1494 return 0;
1495}
1496
1497static ssize_t read_flush(struct file *file, char __user *buf,
1498 size_t count, loff_t *ppos,
1499 struct cache_detail *cd)
1500{
1501 char tbuf[22];
1502 size_t len;
1503
1504 len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1505 convert_to_wallclock(cd->flush_time));
1506 return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1507}
1508
1509static ssize_t write_flush(struct file *file, const char __user *buf,
1510 size_t count, loff_t *ppos,
1511 struct cache_detail *cd)
1512{
1513 char tbuf[20];
1514 char *ep;
1515 time64_t now;
1516
1517 if (*ppos || count > sizeof(tbuf)-1)
1518 return -EINVAL;
1519 if (copy_from_user(tbuf, buf, count))
1520 return -EFAULT;
1521 tbuf[count] = 0;
1522 simple_strtoul(tbuf, &ep, 0);
1523 if (*ep && *ep != '\n')
1524 return -EINVAL;
1525 /* Note that while we check that 'buf' holds a valid number,
1526 * we always ignore the value and just flush everything.
1527 * Making use of the number leads to races.
1528 */
1529
1530 now = seconds_since_boot();
1531 /* Always flush everything, so behave like cache_purge()
1532 * Do this by advancing flush_time to the current time,
1533 * or by one second if it has already reached the current time.
1534 * Newly added cache entries will always have ->last_refresh greater
1535 * that ->flush_time, so they don't get flushed prematurely.
1536 */
1537
1538 if (cd->flush_time >= now)
1539 now = cd->flush_time + 1;
1540
1541 cd->flush_time = now;
1542 cd->nextcheck = now;
1543 cache_flush();
1544
1545 if (cd->flush)
1546 cd->flush();
1547
1548 *ppos += count;
1549 return count;
1550}
1551
1552static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1553 size_t count, loff_t *ppos)
1554{
1555 struct cache_detail *cd = pde_data(file_inode(filp));
1556
1557 return cache_read(filp, buf, count, ppos, cd);
1558}
1559
1560static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1561 size_t count, loff_t *ppos)
1562{
1563 struct cache_detail *cd = pde_data(file_inode(filp));
1564
1565 return cache_write(filp, buf, count, ppos, cd);
1566}
1567
1568static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1569{
1570 struct cache_detail *cd = pde_data(file_inode(filp));
1571
1572 return cache_poll(filp, wait, cd);
1573}
1574
1575static long cache_ioctl_procfs(struct file *filp,
1576 unsigned int cmd, unsigned long arg)
1577{
1578 struct inode *inode = file_inode(filp);
1579 struct cache_detail *cd = pde_data(inode);
1580
1581 return cache_ioctl(inode, filp, cmd, arg, cd);
1582}
1583
1584static int cache_open_procfs(struct inode *inode, struct file *filp)
1585{
1586 struct cache_detail *cd = pde_data(inode);
1587
1588 return cache_open(inode, filp, cd);
1589}
1590
1591static int cache_release_procfs(struct inode *inode, struct file *filp)
1592{
1593 struct cache_detail *cd = pde_data(inode);
1594
1595 return cache_release(inode, filp, cd);
1596}
1597
1598static const struct proc_ops cache_channel_proc_ops = {
1599 .proc_lseek = no_llseek,
1600 .proc_read = cache_read_procfs,
1601 .proc_write = cache_write_procfs,
1602 .proc_poll = cache_poll_procfs,
1603 .proc_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1604 .proc_open = cache_open_procfs,
1605 .proc_release = cache_release_procfs,
1606};
1607
1608static int content_open_procfs(struct inode *inode, struct file *filp)
1609{
1610 struct cache_detail *cd = pde_data(inode);
1611
1612 return content_open(inode, filp, cd);
1613}
1614
1615static int content_release_procfs(struct inode *inode, struct file *filp)
1616{
1617 struct cache_detail *cd = pde_data(inode);
1618
1619 return content_release(inode, filp, cd);
1620}
1621
1622static const struct proc_ops content_proc_ops = {
1623 .proc_open = content_open_procfs,
1624 .proc_read = seq_read,
1625 .proc_lseek = seq_lseek,
1626 .proc_release = content_release_procfs,
1627};
1628
1629static int open_flush_procfs(struct inode *inode, struct file *filp)
1630{
1631 struct cache_detail *cd = pde_data(inode);
1632
1633 return open_flush(inode, filp, cd);
1634}
1635
1636static int release_flush_procfs(struct inode *inode, struct file *filp)
1637{
1638 struct cache_detail *cd = pde_data(inode);
1639
1640 return release_flush(inode, filp, cd);
1641}
1642
1643static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1644 size_t count, loff_t *ppos)
1645{
1646 struct cache_detail *cd = pde_data(file_inode(filp));
1647
1648 return read_flush(filp, buf, count, ppos, cd);
1649}
1650
1651static ssize_t write_flush_procfs(struct file *filp,
1652 const char __user *buf,
1653 size_t count, loff_t *ppos)
1654{
1655 struct cache_detail *cd = pde_data(file_inode(filp));
1656
1657 return write_flush(filp, buf, count, ppos, cd);
1658}
1659
1660static const struct proc_ops cache_flush_proc_ops = {
1661 .proc_open = open_flush_procfs,
1662 .proc_read = read_flush_procfs,
1663 .proc_write = write_flush_procfs,
1664 .proc_release = release_flush_procfs,
1665 .proc_lseek = no_llseek,
1666};
1667
1668static void remove_cache_proc_entries(struct cache_detail *cd)
1669{
1670 if (cd->procfs) {
1671 proc_remove(cd->procfs);
1672 cd->procfs = NULL;
1673 }
1674}
1675
1676#ifdef CONFIG_PROC_FS
1677static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1678{
1679 struct proc_dir_entry *p;
1680 struct sunrpc_net *sn;
1681
1682 sn = net_generic(net, sunrpc_net_id);
1683 cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1684 if (cd->procfs == NULL)
1685 goto out_nomem;
1686
1687 p = proc_create_data("flush", S_IFREG | 0600,
1688 cd->procfs, &cache_flush_proc_ops, cd);
1689 if (p == NULL)
1690 goto out_nomem;
1691
1692 if (cd->cache_request || cd->cache_parse) {
1693 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1694 &cache_channel_proc_ops, cd);
1695 if (p == NULL)
1696 goto out_nomem;
1697 }
1698 if (cd->cache_show) {
1699 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1700 &content_proc_ops, cd);
1701 if (p == NULL)
1702 goto out_nomem;
1703 }
1704 return 0;
1705out_nomem:
1706 remove_cache_proc_entries(cd);
1707 return -ENOMEM;
1708}
1709#else /* CONFIG_PROC_FS */
1710static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1711{
1712 return 0;
1713}
1714#endif
1715
1716void __init cache_initialize(void)
1717{
1718 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1719}
1720
1721int cache_register_net(struct cache_detail *cd, struct net *net)
1722{
1723 int ret;
1724
1725 sunrpc_init_cache_detail(cd);
1726 ret = create_cache_proc_entries(cd, net);
1727 if (ret)
1728 sunrpc_destroy_cache_detail(cd);
1729 return ret;
1730}
1731EXPORT_SYMBOL_GPL(cache_register_net);
1732
1733void cache_unregister_net(struct cache_detail *cd, struct net *net)
1734{
1735 remove_cache_proc_entries(cd);
1736 sunrpc_destroy_cache_detail(cd);
1737}
1738EXPORT_SYMBOL_GPL(cache_unregister_net);
1739
1740struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1741{
1742 struct cache_detail *cd;
1743 int i;
1744
1745 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1746 if (cd == NULL)
1747 return ERR_PTR(-ENOMEM);
1748
1749 cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1750 GFP_KERNEL);
1751 if (cd->hash_table == NULL) {
1752 kfree(cd);
1753 return ERR_PTR(-ENOMEM);
1754 }
1755
1756 for (i = 0; i < cd->hash_size; i++)
1757 INIT_HLIST_HEAD(&cd->hash_table[i]);
1758 cd->net = net;
1759 return cd;
1760}
1761EXPORT_SYMBOL_GPL(cache_create_net);
1762
1763void cache_destroy_net(struct cache_detail *cd, struct net *net)
1764{
1765 kfree(cd->hash_table);
1766 kfree(cd);
1767}
1768EXPORT_SYMBOL_GPL(cache_destroy_net);
1769
1770static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1771 size_t count, loff_t *ppos)
1772{
1773 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1774
1775 return cache_read(filp, buf, count, ppos, cd);
1776}
1777
1778static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1779 size_t count, loff_t *ppos)
1780{
1781 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1782
1783 return cache_write(filp, buf, count, ppos, cd);
1784}
1785
1786static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1787{
1788 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1789
1790 return cache_poll(filp, wait, cd);
1791}
1792
1793static long cache_ioctl_pipefs(struct file *filp,
1794 unsigned int cmd, unsigned long arg)
1795{
1796 struct inode *inode = file_inode(filp);
1797 struct cache_detail *cd = RPC_I(inode)->private;
1798
1799 return cache_ioctl(inode, filp, cmd, arg, cd);
1800}
1801
1802static int cache_open_pipefs(struct inode *inode, struct file *filp)
1803{
1804 struct cache_detail *cd = RPC_I(inode)->private;
1805
1806 return cache_open(inode, filp, cd);
1807}
1808
1809static int cache_release_pipefs(struct inode *inode, struct file *filp)
1810{
1811 struct cache_detail *cd = RPC_I(inode)->private;
1812
1813 return cache_release(inode, filp, cd);
1814}
1815
1816const struct file_operations cache_file_operations_pipefs = {
1817 .owner = THIS_MODULE,
1818 .llseek = no_llseek,
1819 .read = cache_read_pipefs,
1820 .write = cache_write_pipefs,
1821 .poll = cache_poll_pipefs,
1822 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1823 .open = cache_open_pipefs,
1824 .release = cache_release_pipefs,
1825};
1826
1827static int content_open_pipefs(struct inode *inode, struct file *filp)
1828{
1829 struct cache_detail *cd = RPC_I(inode)->private;
1830
1831 return content_open(inode, filp, cd);
1832}
1833
1834static int content_release_pipefs(struct inode *inode, struct file *filp)
1835{
1836 struct cache_detail *cd = RPC_I(inode)->private;
1837
1838 return content_release(inode, filp, cd);
1839}
1840
1841const struct file_operations content_file_operations_pipefs = {
1842 .open = content_open_pipefs,
1843 .read = seq_read,
1844 .llseek = seq_lseek,
1845 .release = content_release_pipefs,
1846};
1847
1848static int open_flush_pipefs(struct inode *inode, struct file *filp)
1849{
1850 struct cache_detail *cd = RPC_I(inode)->private;
1851
1852 return open_flush(inode, filp, cd);
1853}
1854
1855static int release_flush_pipefs(struct inode *inode, struct file *filp)
1856{
1857 struct cache_detail *cd = RPC_I(inode)->private;
1858
1859 return release_flush(inode, filp, cd);
1860}
1861
1862static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1863 size_t count, loff_t *ppos)
1864{
1865 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1866
1867 return read_flush(filp, buf, count, ppos, cd);
1868}
1869
1870static ssize_t write_flush_pipefs(struct file *filp,
1871 const char __user *buf,
1872 size_t count, loff_t *ppos)
1873{
1874 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1875
1876 return write_flush(filp, buf, count, ppos, cd);
1877}
1878
1879const struct file_operations cache_flush_operations_pipefs = {
1880 .open = open_flush_pipefs,
1881 .read = read_flush_pipefs,
1882 .write = write_flush_pipefs,
1883 .release = release_flush_pipefs,
1884 .llseek = no_llseek,
1885};
1886
1887int sunrpc_cache_register_pipefs(struct dentry *parent,
1888 const char *name, umode_t umode,
1889 struct cache_detail *cd)
1890{
1891 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1892 if (IS_ERR(dir))
1893 return PTR_ERR(dir);
1894 cd->pipefs = dir;
1895 return 0;
1896}
1897EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1898
1899void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1900{
1901 if (cd->pipefs) {
1902 rpc_remove_cache_dir(cd->pipefs);
1903 cd->pipefs = NULL;
1904 }
1905}
1906EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1907
1908void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1909{
1910 spin_lock(&cd->hash_lock);
1911 if (!hlist_unhashed(&h->cache_list)){
1912 sunrpc_begin_cache_remove_entry(h, cd);
1913 spin_unlock(&cd->hash_lock);
1914 sunrpc_end_cache_remove_entry(h, cd);
1915 } else
1916 spin_unlock(&cd->hash_lock);
1917}
1918EXPORT_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 (cd->entries) {
366 write_unlock(&cd->hash_lock);
367 spin_unlock(&cache_list_lock);
368 goto out;
369 }
370 if (current_detail == cd)
371 current_detail = NULL;
372 list_del_init(&cd->others);
373 write_unlock(&cd->hash_lock);
374 spin_unlock(&cache_list_lock);
375 if (list_empty(&cache_list)) {
376 /* module must be being unloaded so its safe to kill the worker */
377 cancel_delayed_work_sync(&cache_cleaner);
378 }
379 return;
380out:
381 printk(KERN_ERR "RPC: failed to unregister %s cache\n", cd->name);
382}
383EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
384
385/* clean cache tries to find something to clean
386 * and cleans it.
387 * It returns 1 if it cleaned something,
388 * 0 if it didn't find anything this time
389 * -1 if it fell off the end of the list.
390 */
391static int cache_clean(void)
392{
393 int rv = 0;
394 struct list_head *next;
395
396 spin_lock(&cache_list_lock);
397
398 /* find a suitable table if we don't already have one */
399 while (current_detail == NULL ||
400 current_index >= current_detail->hash_size) {
401 if (current_detail)
402 next = current_detail->others.next;
403 else
404 next = cache_list.next;
405 if (next == &cache_list) {
406 current_detail = NULL;
407 spin_unlock(&cache_list_lock);
408 return -1;
409 }
410 current_detail = list_entry(next, struct cache_detail, others);
411 if (current_detail->nextcheck > seconds_since_boot())
412 current_index = current_detail->hash_size;
413 else {
414 current_index = 0;
415 current_detail->nextcheck = seconds_since_boot()+30*60;
416 }
417 }
418
419 /* find a non-empty bucket in the table */
420 while (current_detail &&
421 current_index < current_detail->hash_size &&
422 hlist_empty(¤t_detail->hash_table[current_index]))
423 current_index++;
424
425 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
426
427 if (current_detail && current_index < current_detail->hash_size) {
428 struct cache_head *ch = NULL;
429 struct cache_detail *d;
430 struct hlist_head *head;
431 struct hlist_node *tmp;
432
433 write_lock(¤t_detail->hash_lock);
434
435 /* Ok, now to clean this strand */
436
437 head = ¤t_detail->hash_table[current_index];
438 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
439 if (current_detail->nextcheck > ch->expiry_time)
440 current_detail->nextcheck = ch->expiry_time+1;
441 if (!cache_is_expired(current_detail, ch))
442 continue;
443
444 hlist_del_init(&ch->cache_list);
445 current_detail->entries--;
446 rv = 1;
447 break;
448 }
449
450 write_unlock(¤t_detail->hash_lock);
451 d = current_detail;
452 if (!ch)
453 current_index ++;
454 spin_unlock(&cache_list_lock);
455 if (ch) {
456 set_bit(CACHE_CLEANED, &ch->flags);
457 cache_fresh_unlocked(ch, d);
458 cache_put(ch, d);
459 }
460 } else
461 spin_unlock(&cache_list_lock);
462
463 return rv;
464}
465
466/*
467 * We want to regularly clean the cache, so we need to schedule some work ...
468 */
469static void do_cache_clean(struct work_struct *work)
470{
471 int delay = 5;
472 if (cache_clean() == -1)
473 delay = round_jiffies_relative(30*HZ);
474
475 if (list_empty(&cache_list))
476 delay = 0;
477
478 if (delay)
479 queue_delayed_work(system_power_efficient_wq,
480 &cache_cleaner, delay);
481}
482
483
484/*
485 * Clean all caches promptly. This just calls cache_clean
486 * repeatedly until we are sure that every cache has had a chance to
487 * be fully cleaned
488 */
489void cache_flush(void)
490{
491 while (cache_clean() != -1)
492 cond_resched();
493 while (cache_clean() != -1)
494 cond_resched();
495}
496EXPORT_SYMBOL_GPL(cache_flush);
497
498void cache_purge(struct cache_detail *detail)
499{
500 time_t now = seconds_since_boot();
501 if (detail->flush_time >= now)
502 now = detail->flush_time + 1;
503 /* 'now' is the maximum value any 'last_refresh' can have */
504 detail->flush_time = now;
505 detail->nextcheck = seconds_since_boot();
506 cache_flush();
507}
508EXPORT_SYMBOL_GPL(cache_purge);
509
510
511/*
512 * Deferral and Revisiting of Requests.
513 *
514 * If a cache lookup finds a pending entry, we
515 * need to defer the request and revisit it later.
516 * All deferred requests are stored in a hash table,
517 * indexed by "struct cache_head *".
518 * As it may be wasteful to store a whole request
519 * structure, we allow the request to provide a
520 * deferred form, which must contain a
521 * 'struct cache_deferred_req'
522 * This cache_deferred_req contains a method to allow
523 * it to be revisited when cache info is available
524 */
525
526#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
527#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
528
529#define DFR_MAX 300 /* ??? */
530
531static DEFINE_SPINLOCK(cache_defer_lock);
532static LIST_HEAD(cache_defer_list);
533static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
534static int cache_defer_cnt;
535
536static void __unhash_deferred_req(struct cache_deferred_req *dreq)
537{
538 hlist_del_init(&dreq->hash);
539 if (!list_empty(&dreq->recent)) {
540 list_del_init(&dreq->recent);
541 cache_defer_cnt--;
542 }
543}
544
545static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
546{
547 int hash = DFR_HASH(item);
548
549 INIT_LIST_HEAD(&dreq->recent);
550 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
551}
552
553static void setup_deferral(struct cache_deferred_req *dreq,
554 struct cache_head *item,
555 int count_me)
556{
557
558 dreq->item = item;
559
560 spin_lock(&cache_defer_lock);
561
562 __hash_deferred_req(dreq, item);
563
564 if (count_me) {
565 cache_defer_cnt++;
566 list_add(&dreq->recent, &cache_defer_list);
567 }
568
569 spin_unlock(&cache_defer_lock);
570
571}
572
573struct thread_deferred_req {
574 struct cache_deferred_req handle;
575 struct completion completion;
576};
577
578static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
579{
580 struct thread_deferred_req *dr =
581 container_of(dreq, struct thread_deferred_req, handle);
582 complete(&dr->completion);
583}
584
585static void cache_wait_req(struct cache_req *req, struct cache_head *item)
586{
587 struct thread_deferred_req sleeper;
588 struct cache_deferred_req *dreq = &sleeper.handle;
589
590 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
591 dreq->revisit = cache_restart_thread;
592
593 setup_deferral(dreq, item, 0);
594
595 if (!test_bit(CACHE_PENDING, &item->flags) ||
596 wait_for_completion_interruptible_timeout(
597 &sleeper.completion, req->thread_wait) <= 0) {
598 /* The completion wasn't completed, so we need
599 * to clean up
600 */
601 spin_lock(&cache_defer_lock);
602 if (!hlist_unhashed(&sleeper.handle.hash)) {
603 __unhash_deferred_req(&sleeper.handle);
604 spin_unlock(&cache_defer_lock);
605 } else {
606 /* cache_revisit_request already removed
607 * this from the hash table, but hasn't
608 * called ->revisit yet. It will very soon
609 * and we need to wait for it.
610 */
611 spin_unlock(&cache_defer_lock);
612 wait_for_completion(&sleeper.completion);
613 }
614 }
615}
616
617static void cache_limit_defers(void)
618{
619 /* Make sure we haven't exceed the limit of allowed deferred
620 * requests.
621 */
622 struct cache_deferred_req *discard = NULL;
623
624 if (cache_defer_cnt <= DFR_MAX)
625 return;
626
627 spin_lock(&cache_defer_lock);
628
629 /* Consider removing either the first or the last */
630 if (cache_defer_cnt > DFR_MAX) {
631 if (prandom_u32() & 1)
632 discard = list_entry(cache_defer_list.next,
633 struct cache_deferred_req, recent);
634 else
635 discard = list_entry(cache_defer_list.prev,
636 struct cache_deferred_req, recent);
637 __unhash_deferred_req(discard);
638 }
639 spin_unlock(&cache_defer_lock);
640 if (discard)
641 discard->revisit(discard, 1);
642}
643
644/* Return true if and only if a deferred request is queued. */
645static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
646{
647 struct cache_deferred_req *dreq;
648
649 if (req->thread_wait) {
650 cache_wait_req(req, item);
651 if (!test_bit(CACHE_PENDING, &item->flags))
652 return false;
653 }
654 dreq = req->defer(req);
655 if (dreq == NULL)
656 return false;
657 setup_deferral(dreq, item, 1);
658 if (!test_bit(CACHE_PENDING, &item->flags))
659 /* Bit could have been cleared before we managed to
660 * set up the deferral, so need to revisit just in case
661 */
662 cache_revisit_request(item);
663
664 cache_limit_defers();
665 return true;
666}
667
668static void cache_revisit_request(struct cache_head *item)
669{
670 struct cache_deferred_req *dreq;
671 struct list_head pending;
672 struct hlist_node *tmp;
673 int hash = DFR_HASH(item);
674
675 INIT_LIST_HEAD(&pending);
676 spin_lock(&cache_defer_lock);
677
678 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
679 if (dreq->item == item) {
680 __unhash_deferred_req(dreq);
681 list_add(&dreq->recent, &pending);
682 }
683
684 spin_unlock(&cache_defer_lock);
685
686 while (!list_empty(&pending)) {
687 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
688 list_del_init(&dreq->recent);
689 dreq->revisit(dreq, 0);
690 }
691}
692
693void cache_clean_deferred(void *owner)
694{
695 struct cache_deferred_req *dreq, *tmp;
696 struct list_head pending;
697
698
699 INIT_LIST_HEAD(&pending);
700 spin_lock(&cache_defer_lock);
701
702 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
703 if (dreq->owner == owner) {
704 __unhash_deferred_req(dreq);
705 list_add(&dreq->recent, &pending);
706 }
707 }
708 spin_unlock(&cache_defer_lock);
709
710 while (!list_empty(&pending)) {
711 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
712 list_del_init(&dreq->recent);
713 dreq->revisit(dreq, 1);
714 }
715}
716
717/*
718 * communicate with user-space
719 *
720 * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
721 * On read, you get a full request, or block.
722 * On write, an update request is processed.
723 * Poll works if anything to read, and always allows write.
724 *
725 * Implemented by linked list of requests. Each open file has
726 * a ->private that also exists in this list. New requests are added
727 * to the end and may wakeup and preceding readers.
728 * New readers are added to the head. If, on read, an item is found with
729 * CACHE_UPCALLING clear, we free it from the list.
730 *
731 */
732
733static DEFINE_SPINLOCK(queue_lock);
734static DEFINE_MUTEX(queue_io_mutex);
735
736struct cache_queue {
737 struct list_head list;
738 int reader; /* if 0, then request */
739};
740struct cache_request {
741 struct cache_queue q;
742 struct cache_head *item;
743 char * buf;
744 int len;
745 int readers;
746};
747struct cache_reader {
748 struct cache_queue q;
749 int offset; /* if non-0, we have a refcnt on next request */
750};
751
752static int cache_request(struct cache_detail *detail,
753 struct cache_request *crq)
754{
755 char *bp = crq->buf;
756 int len = PAGE_SIZE;
757
758 detail->cache_request(detail, crq->item, &bp, &len);
759 if (len < 0)
760 return -EAGAIN;
761 return PAGE_SIZE - len;
762}
763
764static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
765 loff_t *ppos, struct cache_detail *cd)
766{
767 struct cache_reader *rp = filp->private_data;
768 struct cache_request *rq;
769 struct inode *inode = file_inode(filp);
770 int err;
771
772 if (count == 0)
773 return 0;
774
775 inode_lock(inode); /* protect against multiple concurrent
776 * readers on this file */
777 again:
778 spin_lock(&queue_lock);
779 /* need to find next request */
780 while (rp->q.list.next != &cd->queue &&
781 list_entry(rp->q.list.next, struct cache_queue, list)
782 ->reader) {
783 struct list_head *next = rp->q.list.next;
784 list_move(&rp->q.list, next);
785 }
786 if (rp->q.list.next == &cd->queue) {
787 spin_unlock(&queue_lock);
788 inode_unlock(inode);
789 WARN_ON_ONCE(rp->offset);
790 return 0;
791 }
792 rq = container_of(rp->q.list.next, struct cache_request, q.list);
793 WARN_ON_ONCE(rq->q.reader);
794 if (rp->offset == 0)
795 rq->readers++;
796 spin_unlock(&queue_lock);
797
798 if (rq->len == 0) {
799 err = cache_request(cd, rq);
800 if (err < 0)
801 goto out;
802 rq->len = err;
803 }
804
805 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
806 err = -EAGAIN;
807 spin_lock(&queue_lock);
808 list_move(&rp->q.list, &rq->q.list);
809 spin_unlock(&queue_lock);
810 } else {
811 if (rp->offset + count > rq->len)
812 count = rq->len - rp->offset;
813 err = -EFAULT;
814 if (copy_to_user(buf, rq->buf + rp->offset, count))
815 goto out;
816 rp->offset += count;
817 if (rp->offset >= rq->len) {
818 rp->offset = 0;
819 spin_lock(&queue_lock);
820 list_move(&rp->q.list, &rq->q.list);
821 spin_unlock(&queue_lock);
822 }
823 err = 0;
824 }
825 out:
826 if (rp->offset == 0) {
827 /* need to release rq */
828 spin_lock(&queue_lock);
829 rq->readers--;
830 if (rq->readers == 0 &&
831 !test_bit(CACHE_PENDING, &rq->item->flags)) {
832 list_del(&rq->q.list);
833 spin_unlock(&queue_lock);
834 cache_put(rq->item, cd);
835 kfree(rq->buf);
836 kfree(rq);
837 } else
838 spin_unlock(&queue_lock);
839 }
840 if (err == -EAGAIN)
841 goto again;
842 inode_unlock(inode);
843 return err ? err : count;
844}
845
846static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
847 size_t count, struct cache_detail *cd)
848{
849 ssize_t ret;
850
851 if (count == 0)
852 return -EINVAL;
853 if (copy_from_user(kaddr, buf, count))
854 return -EFAULT;
855 kaddr[count] = '\0';
856 ret = cd->cache_parse(cd, kaddr, count);
857 if (!ret)
858 ret = count;
859 return ret;
860}
861
862static ssize_t cache_slow_downcall(const char __user *buf,
863 size_t count, struct cache_detail *cd)
864{
865 static char write_buf[8192]; /* protected by queue_io_mutex */
866 ssize_t ret = -EINVAL;
867
868 if (count >= sizeof(write_buf))
869 goto out;
870 mutex_lock(&queue_io_mutex);
871 ret = cache_do_downcall(write_buf, buf, count, cd);
872 mutex_unlock(&queue_io_mutex);
873out:
874 return ret;
875}
876
877static ssize_t cache_downcall(struct address_space *mapping,
878 const char __user *buf,
879 size_t count, struct cache_detail *cd)
880{
881 struct page *page;
882 char *kaddr;
883 ssize_t ret = -ENOMEM;
884
885 if (count >= PAGE_SIZE)
886 goto out_slow;
887
888 page = find_or_create_page(mapping, 0, GFP_KERNEL);
889 if (!page)
890 goto out_slow;
891
892 kaddr = kmap(page);
893 ret = cache_do_downcall(kaddr, buf, count, cd);
894 kunmap(page);
895 unlock_page(page);
896 put_page(page);
897 return ret;
898out_slow:
899 return cache_slow_downcall(buf, count, cd);
900}
901
902static ssize_t cache_write(struct file *filp, const char __user *buf,
903 size_t count, loff_t *ppos,
904 struct cache_detail *cd)
905{
906 struct address_space *mapping = filp->f_mapping;
907 struct inode *inode = file_inode(filp);
908 ssize_t ret = -EINVAL;
909
910 if (!cd->cache_parse)
911 goto out;
912
913 inode_lock(inode);
914 ret = cache_downcall(mapping, buf, count, cd);
915 inode_unlock(inode);
916out:
917 return ret;
918}
919
920static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
921
922static unsigned int cache_poll(struct file *filp, poll_table *wait,
923 struct cache_detail *cd)
924{
925 unsigned int mask;
926 struct cache_reader *rp = filp->private_data;
927 struct cache_queue *cq;
928
929 poll_wait(filp, &queue_wait, wait);
930
931 /* alway allow write */
932 mask = POLLOUT | POLLWRNORM;
933
934 if (!rp)
935 return mask;
936
937 spin_lock(&queue_lock);
938
939 for (cq= &rp->q; &cq->list != &cd->queue;
940 cq = list_entry(cq->list.next, struct cache_queue, list))
941 if (!cq->reader) {
942 mask |= POLLIN | POLLRDNORM;
943 break;
944 }
945 spin_unlock(&queue_lock);
946 return mask;
947}
948
949static int cache_ioctl(struct inode *ino, struct file *filp,
950 unsigned int cmd, unsigned long arg,
951 struct cache_detail *cd)
952{
953 int len = 0;
954 struct cache_reader *rp = filp->private_data;
955 struct cache_queue *cq;
956
957 if (cmd != FIONREAD || !rp)
958 return -EINVAL;
959
960 spin_lock(&queue_lock);
961
962 /* only find the length remaining in current request,
963 * or the length of the next request
964 */
965 for (cq= &rp->q; &cq->list != &cd->queue;
966 cq = list_entry(cq->list.next, struct cache_queue, list))
967 if (!cq->reader) {
968 struct cache_request *cr =
969 container_of(cq, struct cache_request, q);
970 len = cr->len - rp->offset;
971 break;
972 }
973 spin_unlock(&queue_lock);
974
975 return put_user(len, (int __user *)arg);
976}
977
978static int cache_open(struct inode *inode, struct file *filp,
979 struct cache_detail *cd)
980{
981 struct cache_reader *rp = NULL;
982
983 if (!cd || !try_module_get(cd->owner))
984 return -EACCES;
985 nonseekable_open(inode, filp);
986 if (filp->f_mode & FMODE_READ) {
987 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
988 if (!rp) {
989 module_put(cd->owner);
990 return -ENOMEM;
991 }
992 rp->offset = 0;
993 rp->q.reader = 1;
994 atomic_inc(&cd->readers);
995 spin_lock(&queue_lock);
996 list_add(&rp->q.list, &cd->queue);
997 spin_unlock(&queue_lock);
998 }
999 filp->private_data = rp;
1000 return 0;
1001}
1002
1003static int cache_release(struct inode *inode, struct file *filp,
1004 struct cache_detail *cd)
1005{
1006 struct cache_reader *rp = filp->private_data;
1007
1008 if (rp) {
1009 spin_lock(&queue_lock);
1010 if (rp->offset) {
1011 struct cache_queue *cq;
1012 for (cq= &rp->q; &cq->list != &cd->queue;
1013 cq = list_entry(cq->list.next, struct cache_queue, list))
1014 if (!cq->reader) {
1015 container_of(cq, struct cache_request, q)
1016 ->readers--;
1017 break;
1018 }
1019 rp->offset = 0;
1020 }
1021 list_del(&rp->q.list);
1022 spin_unlock(&queue_lock);
1023
1024 filp->private_data = NULL;
1025 kfree(rp);
1026
1027 cd->last_close = seconds_since_boot();
1028 atomic_dec(&cd->readers);
1029 }
1030 module_put(cd->owner);
1031 return 0;
1032}
1033
1034
1035
1036static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1037{
1038 struct cache_queue *cq, *tmp;
1039 struct cache_request *cr;
1040 struct list_head dequeued;
1041
1042 INIT_LIST_HEAD(&dequeued);
1043 spin_lock(&queue_lock);
1044 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1045 if (!cq->reader) {
1046 cr = container_of(cq, struct cache_request, q);
1047 if (cr->item != ch)
1048 continue;
1049 if (test_bit(CACHE_PENDING, &ch->flags))
1050 /* Lost a race and it is pending again */
1051 break;
1052 if (cr->readers != 0)
1053 continue;
1054 list_move(&cr->q.list, &dequeued);
1055 }
1056 spin_unlock(&queue_lock);
1057 while (!list_empty(&dequeued)) {
1058 cr = list_entry(dequeued.next, struct cache_request, q.list);
1059 list_del(&cr->q.list);
1060 cache_put(cr->item, detail);
1061 kfree(cr->buf);
1062 kfree(cr);
1063 }
1064}
1065
1066/*
1067 * Support routines for text-based upcalls.
1068 * Fields are separated by spaces.
1069 * Fields are either mangled to quote space tab newline slosh with slosh
1070 * or a hexified with a leading \x
1071 * Record is terminated with newline.
1072 *
1073 */
1074
1075void qword_add(char **bpp, int *lp, char *str)
1076{
1077 char *bp = *bpp;
1078 int len = *lp;
1079 int ret;
1080
1081 if (len < 0) return;
1082
1083 ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1084 if (ret >= len) {
1085 bp += len;
1086 len = -1;
1087 } else {
1088 bp += ret;
1089 len -= ret;
1090 *bp++ = ' ';
1091 len--;
1092 }
1093 *bpp = bp;
1094 *lp = len;
1095}
1096EXPORT_SYMBOL_GPL(qword_add);
1097
1098void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1099{
1100 char *bp = *bpp;
1101 int len = *lp;
1102
1103 if (len < 0) return;
1104
1105 if (len > 2) {
1106 *bp++ = '\\';
1107 *bp++ = 'x';
1108 len -= 2;
1109 while (blen && len >= 2) {
1110 bp = hex_byte_pack(bp, *buf++);
1111 len -= 2;
1112 blen--;
1113 }
1114 }
1115 if (blen || len<1) len = -1;
1116 else {
1117 *bp++ = ' ';
1118 len--;
1119 }
1120 *bpp = bp;
1121 *lp = len;
1122}
1123EXPORT_SYMBOL_GPL(qword_addhex);
1124
1125static void warn_no_listener(struct cache_detail *detail)
1126{
1127 if (detail->last_warn != detail->last_close) {
1128 detail->last_warn = detail->last_close;
1129 if (detail->warn_no_listener)
1130 detail->warn_no_listener(detail, detail->last_close != 0);
1131 }
1132}
1133
1134static bool cache_listeners_exist(struct cache_detail *detail)
1135{
1136 if (atomic_read(&detail->readers))
1137 return true;
1138 if (detail->last_close == 0)
1139 /* This cache was never opened */
1140 return false;
1141 if (detail->last_close < seconds_since_boot() - 30)
1142 /*
1143 * We allow for the possibility that someone might
1144 * restart a userspace daemon without restarting the
1145 * server; but after 30 seconds, we give up.
1146 */
1147 return false;
1148 return true;
1149}
1150
1151/*
1152 * register an upcall request to user-space and queue it up for read() by the
1153 * upcall daemon.
1154 *
1155 * Each request is at most one page long.
1156 */
1157int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1158{
1159
1160 char *buf;
1161 struct cache_request *crq;
1162 int ret = 0;
1163
1164 if (!detail->cache_request)
1165 return -EINVAL;
1166
1167 if (!cache_listeners_exist(detail)) {
1168 warn_no_listener(detail);
1169 return -EINVAL;
1170 }
1171 if (test_bit(CACHE_CLEANED, &h->flags))
1172 /* Too late to make an upcall */
1173 return -EAGAIN;
1174
1175 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1176 if (!buf)
1177 return -EAGAIN;
1178
1179 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1180 if (!crq) {
1181 kfree(buf);
1182 return -EAGAIN;
1183 }
1184
1185 crq->q.reader = 0;
1186 crq->buf = buf;
1187 crq->len = 0;
1188 crq->readers = 0;
1189 spin_lock(&queue_lock);
1190 if (test_bit(CACHE_PENDING, &h->flags)) {
1191 crq->item = cache_get(h);
1192 list_add_tail(&crq->q.list, &detail->queue);
1193 } else
1194 /* Lost a race, no longer PENDING, so don't enqueue */
1195 ret = -EAGAIN;
1196 spin_unlock(&queue_lock);
1197 wake_up(&queue_wait);
1198 if (ret == -EAGAIN) {
1199 kfree(buf);
1200 kfree(crq);
1201 }
1202 return ret;
1203}
1204EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1205
1206/*
1207 * parse a message from user-space and pass it
1208 * to an appropriate cache
1209 * Messages are, like requests, separated into fields by
1210 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1211 *
1212 * Message is
1213 * reply cachename expiry key ... content....
1214 *
1215 * key and content are both parsed by cache
1216 */
1217
1218int qword_get(char **bpp, char *dest, int bufsize)
1219{
1220 /* return bytes copied, or -1 on error */
1221 char *bp = *bpp;
1222 int len = 0;
1223
1224 while (*bp == ' ') bp++;
1225
1226 if (bp[0] == '\\' && bp[1] == 'x') {
1227 /* HEX STRING */
1228 bp += 2;
1229 while (len < bufsize - 1) {
1230 int h, l;
1231
1232 h = hex_to_bin(bp[0]);
1233 if (h < 0)
1234 break;
1235
1236 l = hex_to_bin(bp[1]);
1237 if (l < 0)
1238 break;
1239
1240 *dest++ = (h << 4) | l;
1241 bp += 2;
1242 len++;
1243 }
1244 } else {
1245 /* text with \nnn octal quoting */
1246 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1247 if (*bp == '\\' &&
1248 isodigit(bp[1]) && (bp[1] <= '3') &&
1249 isodigit(bp[2]) &&
1250 isodigit(bp[3])) {
1251 int byte = (*++bp -'0');
1252 bp++;
1253 byte = (byte << 3) | (*bp++ - '0');
1254 byte = (byte << 3) | (*bp++ - '0');
1255 *dest++ = byte;
1256 len++;
1257 } else {
1258 *dest++ = *bp++;
1259 len++;
1260 }
1261 }
1262 }
1263
1264 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1265 return -1;
1266 while (*bp == ' ') bp++;
1267 *bpp = bp;
1268 *dest = '\0';
1269 return len;
1270}
1271EXPORT_SYMBOL_GPL(qword_get);
1272
1273
1274/*
1275 * support /proc/sunrpc/cache/$CACHENAME/content
1276 * as a seqfile.
1277 * We call ->cache_show passing NULL for the item to
1278 * get a header, then pass each real item in the cache
1279 */
1280
1281void *cache_seq_start(struct seq_file *m, loff_t *pos)
1282 __acquires(cd->hash_lock)
1283{
1284 loff_t n = *pos;
1285 unsigned int hash, entry;
1286 struct cache_head *ch;
1287 struct cache_detail *cd = m->private;
1288
1289 read_lock(&cd->hash_lock);
1290 if (!n--)
1291 return SEQ_START_TOKEN;
1292 hash = n >> 32;
1293 entry = n & ((1LL<<32) - 1);
1294
1295 hlist_for_each_entry(ch, &cd->hash_table[hash], cache_list)
1296 if (!entry--)
1297 return ch;
1298 n &= ~((1LL<<32) - 1);
1299 do {
1300 hash++;
1301 n += 1LL<<32;
1302 } while(hash < cd->hash_size &&
1303 hlist_empty(&cd->hash_table[hash]));
1304 if (hash >= cd->hash_size)
1305 return NULL;
1306 *pos = n+1;
1307 return hlist_entry_safe(cd->hash_table[hash].first,
1308 struct cache_head, cache_list);
1309}
1310EXPORT_SYMBOL_GPL(cache_seq_start);
1311
1312void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1313{
1314 struct cache_head *ch = p;
1315 int hash = (*pos >> 32);
1316 struct cache_detail *cd = m->private;
1317
1318 if (p == SEQ_START_TOKEN)
1319 hash = 0;
1320 else if (ch->cache_list.next == NULL) {
1321 hash++;
1322 *pos += 1LL<<32;
1323 } else {
1324 ++*pos;
1325 return hlist_entry_safe(ch->cache_list.next,
1326 struct cache_head, cache_list);
1327 }
1328 *pos &= ~((1LL<<32) - 1);
1329 while (hash < cd->hash_size &&
1330 hlist_empty(&cd->hash_table[hash])) {
1331 hash++;
1332 *pos += 1LL<<32;
1333 }
1334 if (hash >= cd->hash_size)
1335 return NULL;
1336 ++*pos;
1337 return hlist_entry_safe(cd->hash_table[hash].first,
1338 struct cache_head, cache_list);
1339}
1340EXPORT_SYMBOL_GPL(cache_seq_next);
1341
1342void cache_seq_stop(struct seq_file *m, void *p)
1343 __releases(cd->hash_lock)
1344{
1345 struct cache_detail *cd = m->private;
1346 read_unlock(&cd->hash_lock);
1347}
1348EXPORT_SYMBOL_GPL(cache_seq_stop);
1349
1350static int c_show(struct seq_file *m, void *p)
1351{
1352 struct cache_head *cp = p;
1353 struct cache_detail *cd = m->private;
1354
1355 if (p == SEQ_START_TOKEN)
1356 return cd->cache_show(m, cd, NULL);
1357
1358 ifdebug(CACHE)
1359 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1360 convert_to_wallclock(cp->expiry_time),
1361 atomic_read(&cp->ref.refcount), cp->flags);
1362 cache_get(cp);
1363 if (cache_check(cd, cp, NULL))
1364 /* cache_check does a cache_put on failure */
1365 seq_printf(m, "# ");
1366 else {
1367 if (cache_is_expired(cd, cp))
1368 seq_printf(m, "# ");
1369 cache_put(cp, cd);
1370 }
1371
1372 return cd->cache_show(m, cd, cp);
1373}
1374
1375static const struct seq_operations cache_content_op = {
1376 .start = cache_seq_start,
1377 .next = cache_seq_next,
1378 .stop = cache_seq_stop,
1379 .show = c_show,
1380};
1381
1382static int content_open(struct inode *inode, struct file *file,
1383 struct cache_detail *cd)
1384{
1385 struct seq_file *seq;
1386 int err;
1387
1388 if (!cd || !try_module_get(cd->owner))
1389 return -EACCES;
1390
1391 err = seq_open(file, &cache_content_op);
1392 if (err) {
1393 module_put(cd->owner);
1394 return err;
1395 }
1396
1397 seq = file->private_data;
1398 seq->private = cd;
1399 return 0;
1400}
1401
1402static int content_release(struct inode *inode, struct file *file,
1403 struct cache_detail *cd)
1404{
1405 int ret = seq_release(inode, file);
1406 module_put(cd->owner);
1407 return ret;
1408}
1409
1410static int open_flush(struct inode *inode, struct file *file,
1411 struct cache_detail *cd)
1412{
1413 if (!cd || !try_module_get(cd->owner))
1414 return -EACCES;
1415 return nonseekable_open(inode, file);
1416}
1417
1418static int release_flush(struct inode *inode, struct file *file,
1419 struct cache_detail *cd)
1420{
1421 module_put(cd->owner);
1422 return 0;
1423}
1424
1425static ssize_t read_flush(struct file *file, char __user *buf,
1426 size_t count, loff_t *ppos,
1427 struct cache_detail *cd)
1428{
1429 char tbuf[22];
1430 unsigned long p = *ppos;
1431 size_t len;
1432
1433 snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1434 len = strlen(tbuf);
1435 if (p >= len)
1436 return 0;
1437 len -= p;
1438 if (len > count)
1439 len = count;
1440 if (copy_to_user(buf, (void*)(tbuf+p), len))
1441 return -EFAULT;
1442 *ppos += len;
1443 return len;
1444}
1445
1446static ssize_t write_flush(struct file *file, const char __user *buf,
1447 size_t count, loff_t *ppos,
1448 struct cache_detail *cd)
1449{
1450 char tbuf[20];
1451 char *bp, *ep;
1452 time_t then, now;
1453
1454 if (*ppos || count > sizeof(tbuf)-1)
1455 return -EINVAL;
1456 if (copy_from_user(tbuf, buf, count))
1457 return -EFAULT;
1458 tbuf[count] = 0;
1459 simple_strtoul(tbuf, &ep, 0);
1460 if (*ep && *ep != '\n')
1461 return -EINVAL;
1462
1463 bp = tbuf;
1464 then = get_expiry(&bp);
1465 now = seconds_since_boot();
1466 cd->nextcheck = now;
1467 /* Can only set flush_time to 1 second beyond "now", or
1468 * possibly 1 second beyond flushtime. This is because
1469 * flush_time never goes backwards so it mustn't get too far
1470 * ahead of time.
1471 */
1472 if (then >= now) {
1473 /* Want to flush everything, so behave like cache_purge() */
1474 if (cd->flush_time >= now)
1475 now = cd->flush_time + 1;
1476 then = now;
1477 }
1478
1479 cd->flush_time = then;
1480 cache_flush();
1481
1482 *ppos += count;
1483 return count;
1484}
1485
1486static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1487 size_t count, loff_t *ppos)
1488{
1489 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1490
1491 return cache_read(filp, buf, count, ppos, cd);
1492}
1493
1494static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1495 size_t count, loff_t *ppos)
1496{
1497 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1498
1499 return cache_write(filp, buf, count, ppos, cd);
1500}
1501
1502static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1503{
1504 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1505
1506 return cache_poll(filp, wait, cd);
1507}
1508
1509static long cache_ioctl_procfs(struct file *filp,
1510 unsigned int cmd, unsigned long arg)
1511{
1512 struct inode *inode = file_inode(filp);
1513 struct cache_detail *cd = PDE_DATA(inode);
1514
1515 return cache_ioctl(inode, filp, cmd, arg, cd);
1516}
1517
1518static int cache_open_procfs(struct inode *inode, struct file *filp)
1519{
1520 struct cache_detail *cd = PDE_DATA(inode);
1521
1522 return cache_open(inode, filp, cd);
1523}
1524
1525static int cache_release_procfs(struct inode *inode, struct file *filp)
1526{
1527 struct cache_detail *cd = PDE_DATA(inode);
1528
1529 return cache_release(inode, filp, cd);
1530}
1531
1532static const struct file_operations cache_file_operations_procfs = {
1533 .owner = THIS_MODULE,
1534 .llseek = no_llseek,
1535 .read = cache_read_procfs,
1536 .write = cache_write_procfs,
1537 .poll = cache_poll_procfs,
1538 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1539 .open = cache_open_procfs,
1540 .release = cache_release_procfs,
1541};
1542
1543static int content_open_procfs(struct inode *inode, struct file *filp)
1544{
1545 struct cache_detail *cd = PDE_DATA(inode);
1546
1547 return content_open(inode, filp, cd);
1548}
1549
1550static int content_release_procfs(struct inode *inode, struct file *filp)
1551{
1552 struct cache_detail *cd = PDE_DATA(inode);
1553
1554 return content_release(inode, filp, cd);
1555}
1556
1557static const struct file_operations content_file_operations_procfs = {
1558 .open = content_open_procfs,
1559 .read = seq_read,
1560 .llseek = seq_lseek,
1561 .release = content_release_procfs,
1562};
1563
1564static int open_flush_procfs(struct inode *inode, struct file *filp)
1565{
1566 struct cache_detail *cd = PDE_DATA(inode);
1567
1568 return open_flush(inode, filp, cd);
1569}
1570
1571static int release_flush_procfs(struct inode *inode, struct file *filp)
1572{
1573 struct cache_detail *cd = PDE_DATA(inode);
1574
1575 return release_flush(inode, filp, cd);
1576}
1577
1578static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1579 size_t count, loff_t *ppos)
1580{
1581 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1582
1583 return read_flush(filp, buf, count, ppos, cd);
1584}
1585
1586static ssize_t write_flush_procfs(struct file *filp,
1587 const char __user *buf,
1588 size_t count, loff_t *ppos)
1589{
1590 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1591
1592 return write_flush(filp, buf, count, ppos, cd);
1593}
1594
1595static const struct file_operations cache_flush_operations_procfs = {
1596 .open = open_flush_procfs,
1597 .read = read_flush_procfs,
1598 .write = write_flush_procfs,
1599 .release = release_flush_procfs,
1600 .llseek = no_llseek,
1601};
1602
1603static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1604{
1605 struct sunrpc_net *sn;
1606
1607 if (cd->u.procfs.proc_ent == NULL)
1608 return;
1609 if (cd->u.procfs.flush_ent)
1610 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1611 if (cd->u.procfs.channel_ent)
1612 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1613 if (cd->u.procfs.content_ent)
1614 remove_proc_entry("content", cd->u.procfs.proc_ent);
1615 cd->u.procfs.proc_ent = NULL;
1616 sn = net_generic(net, sunrpc_net_id);
1617 remove_proc_entry(cd->name, sn->proc_net_rpc);
1618}
1619
1620#ifdef CONFIG_PROC_FS
1621static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1622{
1623 struct proc_dir_entry *p;
1624 struct sunrpc_net *sn;
1625
1626 sn = net_generic(net, sunrpc_net_id);
1627 cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1628 if (cd->u.procfs.proc_ent == NULL)
1629 goto out_nomem;
1630 cd->u.procfs.channel_ent = NULL;
1631 cd->u.procfs.content_ent = NULL;
1632
1633 p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1634 cd->u.procfs.proc_ent,
1635 &cache_flush_operations_procfs, cd);
1636 cd->u.procfs.flush_ent = p;
1637 if (p == NULL)
1638 goto out_nomem;
1639
1640 if (cd->cache_request || cd->cache_parse) {
1641 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1642 cd->u.procfs.proc_ent,
1643 &cache_file_operations_procfs, cd);
1644 cd->u.procfs.channel_ent = p;
1645 if (p == NULL)
1646 goto out_nomem;
1647 }
1648 if (cd->cache_show) {
1649 p = proc_create_data("content", S_IFREG|S_IRUSR,
1650 cd->u.procfs.proc_ent,
1651 &content_file_operations_procfs, cd);
1652 cd->u.procfs.content_ent = p;
1653 if (p == NULL)
1654 goto out_nomem;
1655 }
1656 return 0;
1657out_nomem:
1658 remove_cache_proc_entries(cd, net);
1659 return -ENOMEM;
1660}
1661#else /* CONFIG_PROC_FS */
1662static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1663{
1664 return 0;
1665}
1666#endif
1667
1668void __init cache_initialize(void)
1669{
1670 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1671}
1672
1673int cache_register_net(struct cache_detail *cd, struct net *net)
1674{
1675 int ret;
1676
1677 sunrpc_init_cache_detail(cd);
1678 ret = create_cache_proc_entries(cd, net);
1679 if (ret)
1680 sunrpc_destroy_cache_detail(cd);
1681 return ret;
1682}
1683EXPORT_SYMBOL_GPL(cache_register_net);
1684
1685void cache_unregister_net(struct cache_detail *cd, struct net *net)
1686{
1687 remove_cache_proc_entries(cd, net);
1688 sunrpc_destroy_cache_detail(cd);
1689}
1690EXPORT_SYMBOL_GPL(cache_unregister_net);
1691
1692struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1693{
1694 struct cache_detail *cd;
1695 int i;
1696
1697 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1698 if (cd == NULL)
1699 return ERR_PTR(-ENOMEM);
1700
1701 cd->hash_table = kzalloc(cd->hash_size * sizeof(struct hlist_head),
1702 GFP_KERNEL);
1703 if (cd->hash_table == NULL) {
1704 kfree(cd);
1705 return ERR_PTR(-ENOMEM);
1706 }
1707
1708 for (i = 0; i < cd->hash_size; i++)
1709 INIT_HLIST_HEAD(&cd->hash_table[i]);
1710 cd->net = net;
1711 return cd;
1712}
1713EXPORT_SYMBOL_GPL(cache_create_net);
1714
1715void cache_destroy_net(struct cache_detail *cd, struct net *net)
1716{
1717 kfree(cd->hash_table);
1718 kfree(cd);
1719}
1720EXPORT_SYMBOL_GPL(cache_destroy_net);
1721
1722static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1723 size_t count, loff_t *ppos)
1724{
1725 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1726
1727 return cache_read(filp, buf, count, ppos, cd);
1728}
1729
1730static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1731 size_t count, loff_t *ppos)
1732{
1733 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1734
1735 return cache_write(filp, buf, count, ppos, cd);
1736}
1737
1738static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1739{
1740 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1741
1742 return cache_poll(filp, wait, cd);
1743}
1744
1745static long cache_ioctl_pipefs(struct file *filp,
1746 unsigned int cmd, unsigned long arg)
1747{
1748 struct inode *inode = file_inode(filp);
1749 struct cache_detail *cd = RPC_I(inode)->private;
1750
1751 return cache_ioctl(inode, filp, cmd, arg, cd);
1752}
1753
1754static int cache_open_pipefs(struct inode *inode, struct file *filp)
1755{
1756 struct cache_detail *cd = RPC_I(inode)->private;
1757
1758 return cache_open(inode, filp, cd);
1759}
1760
1761static int cache_release_pipefs(struct inode *inode, struct file *filp)
1762{
1763 struct cache_detail *cd = RPC_I(inode)->private;
1764
1765 return cache_release(inode, filp, cd);
1766}
1767
1768const struct file_operations cache_file_operations_pipefs = {
1769 .owner = THIS_MODULE,
1770 .llseek = no_llseek,
1771 .read = cache_read_pipefs,
1772 .write = cache_write_pipefs,
1773 .poll = cache_poll_pipefs,
1774 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1775 .open = cache_open_pipefs,
1776 .release = cache_release_pipefs,
1777};
1778
1779static int content_open_pipefs(struct inode *inode, struct file *filp)
1780{
1781 struct cache_detail *cd = RPC_I(inode)->private;
1782
1783 return content_open(inode, filp, cd);
1784}
1785
1786static int content_release_pipefs(struct inode *inode, struct file *filp)
1787{
1788 struct cache_detail *cd = RPC_I(inode)->private;
1789
1790 return content_release(inode, filp, cd);
1791}
1792
1793const struct file_operations content_file_operations_pipefs = {
1794 .open = content_open_pipefs,
1795 .read = seq_read,
1796 .llseek = seq_lseek,
1797 .release = content_release_pipefs,
1798};
1799
1800static int open_flush_pipefs(struct inode *inode, struct file *filp)
1801{
1802 struct cache_detail *cd = RPC_I(inode)->private;
1803
1804 return open_flush(inode, filp, cd);
1805}
1806
1807static int release_flush_pipefs(struct inode *inode, struct file *filp)
1808{
1809 struct cache_detail *cd = RPC_I(inode)->private;
1810
1811 return release_flush(inode, filp, cd);
1812}
1813
1814static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1815 size_t count, loff_t *ppos)
1816{
1817 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1818
1819 return read_flush(filp, buf, count, ppos, cd);
1820}
1821
1822static ssize_t write_flush_pipefs(struct file *filp,
1823 const char __user *buf,
1824 size_t count, loff_t *ppos)
1825{
1826 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1827
1828 return write_flush(filp, buf, count, ppos, cd);
1829}
1830
1831const struct file_operations cache_flush_operations_pipefs = {
1832 .open = open_flush_pipefs,
1833 .read = read_flush_pipefs,
1834 .write = write_flush_pipefs,
1835 .release = release_flush_pipefs,
1836 .llseek = no_llseek,
1837};
1838
1839int sunrpc_cache_register_pipefs(struct dentry *parent,
1840 const char *name, umode_t umode,
1841 struct cache_detail *cd)
1842{
1843 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1844 if (IS_ERR(dir))
1845 return PTR_ERR(dir);
1846 cd->u.pipefs.dir = dir;
1847 return 0;
1848}
1849EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1850
1851void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1852{
1853 rpc_remove_cache_dir(cd->u.pipefs.dir);
1854 cd->u.pipefs.dir = NULL;
1855}
1856EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1857