Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/sunrpc/cache.c
4 *
5 * Generic code for various authentication-related caches
6 * used by sunrpc clients and servers.
7 *
8 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
9 */
10
11#include <linux/types.h>
12#include <linux/fs.h>
13#include <linux/file.h>
14#include <linux/slab.h>
15#include <linux/signal.h>
16#include <linux/sched.h>
17#include <linux/kmod.h>
18#include <linux/list.h>
19#include <linux/module.h>
20#include <linux/ctype.h>
21#include <linux/string_helpers.h>
22#include <linux/uaccess.h>
23#include <linux/poll.h>
24#include <linux/seq_file.h>
25#include <linux/proc_fs.h>
26#include <linux/net.h>
27#include <linux/workqueue.h>
28#include <linux/mutex.h>
29#include <linux/pagemap.h>
30#include <asm/ioctls.h>
31#include <linux/sunrpc/types.h>
32#include <linux/sunrpc/cache.h>
33#include <linux/sunrpc/stats.h>
34#include <linux/sunrpc/rpc_pipe_fs.h>
35#include "netns.h"
36
37#define RPCDBG_FACILITY RPCDBG_CACHE
38
39static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
40static void cache_revisit_request(struct cache_head *item);
41static bool cache_listeners_exist(struct cache_detail *detail);
42
43static void cache_init(struct cache_head *h, struct cache_detail *detail)
44{
45 time_t now = seconds_since_boot();
46 INIT_HLIST_NODE(&h->cache_list);
47 h->flags = 0;
48 kref_init(&h->ref);
49 h->expiry_time = now + CACHE_NEW_EXPIRY;
50 if (now <= detail->flush_time)
51 /* ensure it isn't already expired */
52 now = detail->flush_time + 1;
53 h->last_refresh = now;
54}
55
56static inline int cache_is_valid(struct cache_head *h);
57static void cache_fresh_locked(struct cache_head *head, time_t expiry,
58 struct cache_detail *detail);
59static void cache_fresh_unlocked(struct cache_head *head,
60 struct cache_detail *detail);
61
62static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
63 struct cache_head *key,
64 int hash)
65{
66 struct hlist_head *head = &detail->hash_table[hash];
67 struct cache_head *tmp;
68
69 rcu_read_lock();
70 hlist_for_each_entry_rcu(tmp, head, cache_list) {
71 if (detail->match(tmp, key)) {
72 if (cache_is_expired(detail, tmp))
73 continue;
74 tmp = cache_get_rcu(tmp);
75 rcu_read_unlock();
76 return tmp;
77 }
78 }
79 rcu_read_unlock();
80 return NULL;
81}
82
83static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
84 struct cache_head *key,
85 int hash)
86{
87 struct cache_head *new, *tmp, *freeme = NULL;
88 struct hlist_head *head = &detail->hash_table[hash];
89
90 new = detail->alloc();
91 if (!new)
92 return NULL;
93 /* must fully initialise 'new', else
94 * we might get lose if we need to
95 * cache_put it soon.
96 */
97 cache_init(new, detail);
98 detail->init(new, key);
99
100 spin_lock(&detail->hash_lock);
101
102 /* check if entry appeared while we slept */
103 hlist_for_each_entry_rcu(tmp, head, cache_list) {
104 if (detail->match(tmp, key)) {
105 if (cache_is_expired(detail, tmp)) {
106 hlist_del_init_rcu(&tmp->cache_list);
107 detail->entries --;
108 if (cache_is_valid(tmp) == -EAGAIN)
109 set_bit(CACHE_NEGATIVE, &tmp->flags);
110 cache_fresh_locked(tmp, 0, detail);
111 freeme = tmp;
112 break;
113 }
114 cache_get(tmp);
115 spin_unlock(&detail->hash_lock);
116 cache_put(new, detail);
117 return tmp;
118 }
119 }
120
121 hlist_add_head_rcu(&new->cache_list, head);
122 detail->entries++;
123 cache_get(new);
124 spin_unlock(&detail->hash_lock);
125
126 if (freeme) {
127 cache_fresh_unlocked(freeme, detail);
128 cache_put(freeme, detail);
129 }
130 return new;
131}
132
133struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
134 struct cache_head *key, int hash)
135{
136 struct cache_head *ret;
137
138 ret = sunrpc_cache_find_rcu(detail, key, hash);
139 if (ret)
140 return ret;
141 /* Didn't find anything, insert an empty entry */
142 return sunrpc_cache_add_entry(detail, key, hash);
143}
144EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
145
146static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
147
148static void cache_fresh_locked(struct cache_head *head, time_t expiry,
149 struct cache_detail *detail)
150{
151 time_t now = seconds_since_boot();
152 if (now <= detail->flush_time)
153 /* ensure it isn't immediately treated as expired */
154 now = detail->flush_time + 1;
155 head->expiry_time = expiry;
156 head->last_refresh = now;
157 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
158 set_bit(CACHE_VALID, &head->flags);
159}
160
161static void cache_fresh_unlocked(struct cache_head *head,
162 struct cache_detail *detail)
163{
164 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
165 cache_revisit_request(head);
166 cache_dequeue(detail, head);
167 }
168}
169
170struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
171 struct cache_head *new, struct cache_head *old, int hash)
172{
173 /* The 'old' entry is to be replaced by 'new'.
174 * If 'old' is not VALID, we update it directly,
175 * otherwise we need to replace it
176 */
177 struct cache_head *tmp;
178
179 if (!test_bit(CACHE_VALID, &old->flags)) {
180 spin_lock(&detail->hash_lock);
181 if (!test_bit(CACHE_VALID, &old->flags)) {
182 if (test_bit(CACHE_NEGATIVE, &new->flags))
183 set_bit(CACHE_NEGATIVE, &old->flags);
184 else
185 detail->update(old, new);
186 cache_fresh_locked(old, new->expiry_time, detail);
187 spin_unlock(&detail->hash_lock);
188 cache_fresh_unlocked(old, detail);
189 return old;
190 }
191 spin_unlock(&detail->hash_lock);
192 }
193 /* We need to insert a new entry */
194 tmp = detail->alloc();
195 if (!tmp) {
196 cache_put(old, detail);
197 return NULL;
198 }
199 cache_init(tmp, detail);
200 detail->init(tmp, old);
201
202 spin_lock(&detail->hash_lock);
203 if (test_bit(CACHE_NEGATIVE, &new->flags))
204 set_bit(CACHE_NEGATIVE, &tmp->flags);
205 else
206 detail->update(tmp, new);
207 hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
208 detail->entries++;
209 cache_get(tmp);
210 cache_fresh_locked(tmp, new->expiry_time, detail);
211 cache_fresh_locked(old, 0, detail);
212 spin_unlock(&detail->hash_lock);
213 cache_fresh_unlocked(tmp, detail);
214 cache_fresh_unlocked(old, detail);
215 cache_put(old, detail);
216 return tmp;
217}
218EXPORT_SYMBOL_GPL(sunrpc_cache_update);
219
220static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
221{
222 if (cd->cache_upcall)
223 return cd->cache_upcall(cd, h);
224 return sunrpc_cache_pipe_upcall(cd, h);
225}
226
227static inline int cache_is_valid(struct cache_head *h)
228{
229 if (!test_bit(CACHE_VALID, &h->flags))
230 return -EAGAIN;
231 else {
232 /* entry is valid */
233 if (test_bit(CACHE_NEGATIVE, &h->flags))
234 return -ENOENT;
235 else {
236 /*
237 * In combination with write barrier in
238 * sunrpc_cache_update, ensures that anyone
239 * using the cache entry after this sees the
240 * updated contents:
241 */
242 smp_rmb();
243 return 0;
244 }
245 }
246}
247
248static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
249{
250 int rv;
251
252 spin_lock(&detail->hash_lock);
253 rv = cache_is_valid(h);
254 if (rv == -EAGAIN) {
255 set_bit(CACHE_NEGATIVE, &h->flags);
256 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
257 detail);
258 rv = -ENOENT;
259 }
260 spin_unlock(&detail->hash_lock);
261 cache_fresh_unlocked(h, detail);
262 return rv;
263}
264
265/*
266 * This is the generic cache management routine for all
267 * the authentication caches.
268 * It checks the currency of a cache item and will (later)
269 * initiate an upcall to fill it if needed.
270 *
271 *
272 * Returns 0 if the cache_head can be used, or cache_puts it and returns
273 * -EAGAIN if upcall is pending and request has been queued
274 * -ETIMEDOUT if upcall failed or request could not be queue or
275 * upcall completed but item is still invalid (implying that
276 * the cache item has been replaced with a newer one).
277 * -ENOENT if cache entry was negative
278 */
279int cache_check(struct cache_detail *detail,
280 struct cache_head *h, struct cache_req *rqstp)
281{
282 int rv;
283 long refresh_age, age;
284
285 /* First decide return status as best we can */
286 rv = cache_is_valid(h);
287
288 /* now see if we want to start an upcall */
289 refresh_age = (h->expiry_time - h->last_refresh);
290 age = seconds_since_boot() - h->last_refresh;
291
292 if (rqstp == NULL) {
293 if (rv == -EAGAIN)
294 rv = -ENOENT;
295 } else if (rv == -EAGAIN ||
296 (h->expiry_time != 0 && age > refresh_age/2)) {
297 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
298 refresh_age, age);
299 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
300 switch (cache_make_upcall(detail, h)) {
301 case -EINVAL:
302 rv = try_to_negate_entry(detail, h);
303 break;
304 case -EAGAIN:
305 cache_fresh_unlocked(h, detail);
306 break;
307 }
308 } else if (!cache_listeners_exist(detail))
309 rv = try_to_negate_entry(detail, h);
310 }
311
312 if (rv == -EAGAIN) {
313 if (!cache_defer_req(rqstp, h)) {
314 /*
315 * Request was not deferred; handle it as best
316 * we can ourselves:
317 */
318 rv = cache_is_valid(h);
319 if (rv == -EAGAIN)
320 rv = -ETIMEDOUT;
321 }
322 }
323 if (rv)
324 cache_put(h, detail);
325 return rv;
326}
327EXPORT_SYMBOL_GPL(cache_check);
328
329/*
330 * caches need to be periodically cleaned.
331 * For this we maintain a list of cache_detail and
332 * a current pointer into that list and into the table
333 * for that entry.
334 *
335 * Each time cache_clean is called it finds the next non-empty entry
336 * in the current table and walks the list in that entry
337 * looking for entries that can be removed.
338 *
339 * An entry gets removed if:
340 * - The expiry is before current time
341 * - The last_refresh time is before the flush_time for that cache
342 *
343 * later we might drop old entries with non-NEVER expiry if that table
344 * is getting 'full' for some definition of 'full'
345 *
346 * The question of "how often to scan a table" is an interesting one
347 * and is answered in part by the use of the "nextcheck" field in the
348 * cache_detail.
349 * When a scan of a table begins, the nextcheck field is set to a time
350 * that is well into the future.
351 * While scanning, if an expiry time is found that is earlier than the
352 * current nextcheck time, nextcheck is set to that expiry time.
353 * If the flush_time is ever set to a time earlier than the nextcheck
354 * time, the nextcheck time is then set to that flush_time.
355 *
356 * A table is then only scanned if the current time is at least
357 * the nextcheck time.
358 *
359 */
360
361static LIST_HEAD(cache_list);
362static DEFINE_SPINLOCK(cache_list_lock);
363static struct cache_detail *current_detail;
364static int current_index;
365
366static void do_cache_clean(struct work_struct *work);
367static struct delayed_work cache_cleaner;
368
369void sunrpc_init_cache_detail(struct cache_detail *cd)
370{
371 spin_lock_init(&cd->hash_lock);
372 INIT_LIST_HEAD(&cd->queue);
373 spin_lock(&cache_list_lock);
374 cd->nextcheck = 0;
375 cd->entries = 0;
376 atomic_set(&cd->writers, 0);
377 cd->last_close = 0;
378 cd->last_warn = -1;
379 list_add(&cd->others, &cache_list);
380 spin_unlock(&cache_list_lock);
381
382 /* start the cleaning process */
383 queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
384}
385EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
386
387void sunrpc_destroy_cache_detail(struct cache_detail *cd)
388{
389 cache_purge(cd);
390 spin_lock(&cache_list_lock);
391 spin_lock(&cd->hash_lock);
392 if (current_detail == cd)
393 current_detail = NULL;
394 list_del_init(&cd->others);
395 spin_unlock(&cd->hash_lock);
396 spin_unlock(&cache_list_lock);
397 if (list_empty(&cache_list)) {
398 /* module must be being unloaded so its safe to kill the worker */
399 cancel_delayed_work_sync(&cache_cleaner);
400 }
401}
402EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
403
404/* clean cache tries to find something to clean
405 * and cleans it.
406 * It returns 1 if it cleaned something,
407 * 0 if it didn't find anything this time
408 * -1 if it fell off the end of the list.
409 */
410static int cache_clean(void)
411{
412 int rv = 0;
413 struct list_head *next;
414
415 spin_lock(&cache_list_lock);
416
417 /* find a suitable table if we don't already have one */
418 while (current_detail == NULL ||
419 current_index >= current_detail->hash_size) {
420 if (current_detail)
421 next = current_detail->others.next;
422 else
423 next = cache_list.next;
424 if (next == &cache_list) {
425 current_detail = NULL;
426 spin_unlock(&cache_list_lock);
427 return -1;
428 }
429 current_detail = list_entry(next, struct cache_detail, others);
430 if (current_detail->nextcheck > seconds_since_boot())
431 current_index = current_detail->hash_size;
432 else {
433 current_index = 0;
434 current_detail->nextcheck = seconds_since_boot()+30*60;
435 }
436 }
437
438 /* find a non-empty bucket in the table */
439 while (current_detail &&
440 current_index < current_detail->hash_size &&
441 hlist_empty(¤t_detail->hash_table[current_index]))
442 current_index++;
443
444 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
445
446 if (current_detail && current_index < current_detail->hash_size) {
447 struct cache_head *ch = NULL;
448 struct cache_detail *d;
449 struct hlist_head *head;
450 struct hlist_node *tmp;
451
452 spin_lock(¤t_detail->hash_lock);
453
454 /* Ok, now to clean this strand */
455
456 head = ¤t_detail->hash_table[current_index];
457 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
458 if (current_detail->nextcheck > ch->expiry_time)
459 current_detail->nextcheck = ch->expiry_time+1;
460 if (!cache_is_expired(current_detail, ch))
461 continue;
462
463 hlist_del_init_rcu(&ch->cache_list);
464 current_detail->entries--;
465 rv = 1;
466 break;
467 }
468
469 spin_unlock(¤t_detail->hash_lock);
470 d = current_detail;
471 if (!ch)
472 current_index ++;
473 spin_unlock(&cache_list_lock);
474 if (ch) {
475 set_bit(CACHE_CLEANED, &ch->flags);
476 cache_fresh_unlocked(ch, d);
477 cache_put(ch, d);
478 }
479 } else
480 spin_unlock(&cache_list_lock);
481
482 return rv;
483}
484
485/*
486 * We want to regularly clean the cache, so we need to schedule some work ...
487 */
488static void do_cache_clean(struct work_struct *work)
489{
490 int delay = 5;
491 if (cache_clean() == -1)
492 delay = round_jiffies_relative(30*HZ);
493
494 if (list_empty(&cache_list))
495 delay = 0;
496
497 if (delay)
498 queue_delayed_work(system_power_efficient_wq,
499 &cache_cleaner, delay);
500}
501
502
503/*
504 * Clean all caches promptly. This just calls cache_clean
505 * repeatedly until we are sure that every cache has had a chance to
506 * be fully cleaned
507 */
508void cache_flush(void)
509{
510 while (cache_clean() != -1)
511 cond_resched();
512 while (cache_clean() != -1)
513 cond_resched();
514}
515EXPORT_SYMBOL_GPL(cache_flush);
516
517void cache_purge(struct cache_detail *detail)
518{
519 struct cache_head *ch = NULL;
520 struct hlist_head *head = NULL;
521 struct hlist_node *tmp = NULL;
522 int i = 0;
523
524 spin_lock(&detail->hash_lock);
525 if (!detail->entries) {
526 spin_unlock(&detail->hash_lock);
527 return;
528 }
529
530 dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
531 for (i = 0; i < detail->hash_size; i++) {
532 head = &detail->hash_table[i];
533 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
534 hlist_del_init_rcu(&ch->cache_list);
535 detail->entries--;
536
537 set_bit(CACHE_CLEANED, &ch->flags);
538 spin_unlock(&detail->hash_lock);
539 cache_fresh_unlocked(ch, detail);
540 cache_put(ch, detail);
541 spin_lock(&detail->hash_lock);
542 }
543 }
544 spin_unlock(&detail->hash_lock);
545}
546EXPORT_SYMBOL_GPL(cache_purge);
547
548
549/*
550 * Deferral and Revisiting of Requests.
551 *
552 * If a cache lookup finds a pending entry, we
553 * need to defer the request and revisit it later.
554 * All deferred requests are stored in a hash table,
555 * indexed by "struct cache_head *".
556 * As it may be wasteful to store a whole request
557 * structure, we allow the request to provide a
558 * deferred form, which must contain a
559 * 'struct cache_deferred_req'
560 * This cache_deferred_req contains a method to allow
561 * it to be revisited when cache info is available
562 */
563
564#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
565#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
566
567#define DFR_MAX 300 /* ??? */
568
569static DEFINE_SPINLOCK(cache_defer_lock);
570static LIST_HEAD(cache_defer_list);
571static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
572static int cache_defer_cnt;
573
574static void __unhash_deferred_req(struct cache_deferred_req *dreq)
575{
576 hlist_del_init(&dreq->hash);
577 if (!list_empty(&dreq->recent)) {
578 list_del_init(&dreq->recent);
579 cache_defer_cnt--;
580 }
581}
582
583static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
584{
585 int hash = DFR_HASH(item);
586
587 INIT_LIST_HEAD(&dreq->recent);
588 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
589}
590
591static void setup_deferral(struct cache_deferred_req *dreq,
592 struct cache_head *item,
593 int count_me)
594{
595
596 dreq->item = item;
597
598 spin_lock(&cache_defer_lock);
599
600 __hash_deferred_req(dreq, item);
601
602 if (count_me) {
603 cache_defer_cnt++;
604 list_add(&dreq->recent, &cache_defer_list);
605 }
606
607 spin_unlock(&cache_defer_lock);
608
609}
610
611struct thread_deferred_req {
612 struct cache_deferred_req handle;
613 struct completion completion;
614};
615
616static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
617{
618 struct thread_deferred_req *dr =
619 container_of(dreq, struct thread_deferred_req, handle);
620 complete(&dr->completion);
621}
622
623static void cache_wait_req(struct cache_req *req, struct cache_head *item)
624{
625 struct thread_deferred_req sleeper;
626 struct cache_deferred_req *dreq = &sleeper.handle;
627
628 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
629 dreq->revisit = cache_restart_thread;
630
631 setup_deferral(dreq, item, 0);
632
633 if (!test_bit(CACHE_PENDING, &item->flags) ||
634 wait_for_completion_interruptible_timeout(
635 &sleeper.completion, req->thread_wait) <= 0) {
636 /* The completion wasn't completed, so we need
637 * to clean up
638 */
639 spin_lock(&cache_defer_lock);
640 if (!hlist_unhashed(&sleeper.handle.hash)) {
641 __unhash_deferred_req(&sleeper.handle);
642 spin_unlock(&cache_defer_lock);
643 } else {
644 /* cache_revisit_request already removed
645 * this from the hash table, but hasn't
646 * called ->revisit yet. It will very soon
647 * and we need to wait for it.
648 */
649 spin_unlock(&cache_defer_lock);
650 wait_for_completion(&sleeper.completion);
651 }
652 }
653}
654
655static void cache_limit_defers(void)
656{
657 /* Make sure we haven't exceed the limit of allowed deferred
658 * requests.
659 */
660 struct cache_deferred_req *discard = NULL;
661
662 if (cache_defer_cnt <= DFR_MAX)
663 return;
664
665 spin_lock(&cache_defer_lock);
666
667 /* Consider removing either the first or the last */
668 if (cache_defer_cnt > DFR_MAX) {
669 if (prandom_u32() & 1)
670 discard = list_entry(cache_defer_list.next,
671 struct cache_deferred_req, recent);
672 else
673 discard = list_entry(cache_defer_list.prev,
674 struct cache_deferred_req, recent);
675 __unhash_deferred_req(discard);
676 }
677 spin_unlock(&cache_defer_lock);
678 if (discard)
679 discard->revisit(discard, 1);
680}
681
682/* Return true if and only if a deferred request is queued. */
683static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
684{
685 struct cache_deferred_req *dreq;
686
687 if (req->thread_wait) {
688 cache_wait_req(req, item);
689 if (!test_bit(CACHE_PENDING, &item->flags))
690 return false;
691 }
692 dreq = req->defer(req);
693 if (dreq == NULL)
694 return false;
695 setup_deferral(dreq, item, 1);
696 if (!test_bit(CACHE_PENDING, &item->flags))
697 /* Bit could have been cleared before we managed to
698 * set up the deferral, so need to revisit just in case
699 */
700 cache_revisit_request(item);
701
702 cache_limit_defers();
703 return true;
704}
705
706static void cache_revisit_request(struct cache_head *item)
707{
708 struct cache_deferred_req *dreq;
709 struct list_head pending;
710 struct hlist_node *tmp;
711 int hash = DFR_HASH(item);
712
713 INIT_LIST_HEAD(&pending);
714 spin_lock(&cache_defer_lock);
715
716 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
717 if (dreq->item == item) {
718 __unhash_deferred_req(dreq);
719 list_add(&dreq->recent, &pending);
720 }
721
722 spin_unlock(&cache_defer_lock);
723
724 while (!list_empty(&pending)) {
725 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
726 list_del_init(&dreq->recent);
727 dreq->revisit(dreq, 0);
728 }
729}
730
731void cache_clean_deferred(void *owner)
732{
733 struct cache_deferred_req *dreq, *tmp;
734 struct list_head pending;
735
736
737 INIT_LIST_HEAD(&pending);
738 spin_lock(&cache_defer_lock);
739
740 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
741 if (dreq->owner == owner) {
742 __unhash_deferred_req(dreq);
743 list_add(&dreq->recent, &pending);
744 }
745 }
746 spin_unlock(&cache_defer_lock);
747
748 while (!list_empty(&pending)) {
749 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
750 list_del_init(&dreq->recent);
751 dreq->revisit(dreq, 1);
752 }
753}
754
755/*
756 * communicate with user-space
757 *
758 * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
759 * On read, you get a full request, or block.
760 * On write, an update request is processed.
761 * Poll works if anything to read, and always allows write.
762 *
763 * Implemented by linked list of requests. Each open file has
764 * a ->private that also exists in this list. New requests are added
765 * to the end and may wakeup and preceding readers.
766 * New readers are added to the head. If, on read, an item is found with
767 * CACHE_UPCALLING clear, we free it from the list.
768 *
769 */
770
771static DEFINE_SPINLOCK(queue_lock);
772static DEFINE_MUTEX(queue_io_mutex);
773
774struct cache_queue {
775 struct list_head list;
776 int reader; /* if 0, then request */
777};
778struct cache_request {
779 struct cache_queue q;
780 struct cache_head *item;
781 char * buf;
782 int len;
783 int readers;
784};
785struct cache_reader {
786 struct cache_queue q;
787 int offset; /* if non-0, we have a refcnt on next request */
788};
789
790static int cache_request(struct cache_detail *detail,
791 struct cache_request *crq)
792{
793 char *bp = crq->buf;
794 int len = PAGE_SIZE;
795
796 detail->cache_request(detail, crq->item, &bp, &len);
797 if (len < 0)
798 return -EAGAIN;
799 return PAGE_SIZE - len;
800}
801
802static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
803 loff_t *ppos, struct cache_detail *cd)
804{
805 struct cache_reader *rp = filp->private_data;
806 struct cache_request *rq;
807 struct inode *inode = file_inode(filp);
808 int err;
809
810 if (count == 0)
811 return 0;
812
813 inode_lock(inode); /* protect against multiple concurrent
814 * readers on this file */
815 again:
816 spin_lock(&queue_lock);
817 /* need to find next request */
818 while (rp->q.list.next != &cd->queue &&
819 list_entry(rp->q.list.next, struct cache_queue, list)
820 ->reader) {
821 struct list_head *next = rp->q.list.next;
822 list_move(&rp->q.list, next);
823 }
824 if (rp->q.list.next == &cd->queue) {
825 spin_unlock(&queue_lock);
826 inode_unlock(inode);
827 WARN_ON_ONCE(rp->offset);
828 return 0;
829 }
830 rq = container_of(rp->q.list.next, struct cache_request, q.list);
831 WARN_ON_ONCE(rq->q.reader);
832 if (rp->offset == 0)
833 rq->readers++;
834 spin_unlock(&queue_lock);
835
836 if (rq->len == 0) {
837 err = cache_request(cd, rq);
838 if (err < 0)
839 goto out;
840 rq->len = err;
841 }
842
843 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
844 err = -EAGAIN;
845 spin_lock(&queue_lock);
846 list_move(&rp->q.list, &rq->q.list);
847 spin_unlock(&queue_lock);
848 } else {
849 if (rp->offset + count > rq->len)
850 count = rq->len - rp->offset;
851 err = -EFAULT;
852 if (copy_to_user(buf, rq->buf + rp->offset, count))
853 goto out;
854 rp->offset += count;
855 if (rp->offset >= rq->len) {
856 rp->offset = 0;
857 spin_lock(&queue_lock);
858 list_move(&rp->q.list, &rq->q.list);
859 spin_unlock(&queue_lock);
860 }
861 err = 0;
862 }
863 out:
864 if (rp->offset == 0) {
865 /* need to release rq */
866 spin_lock(&queue_lock);
867 rq->readers--;
868 if (rq->readers == 0 &&
869 !test_bit(CACHE_PENDING, &rq->item->flags)) {
870 list_del(&rq->q.list);
871 spin_unlock(&queue_lock);
872 cache_put(rq->item, cd);
873 kfree(rq->buf);
874 kfree(rq);
875 } else
876 spin_unlock(&queue_lock);
877 }
878 if (err == -EAGAIN)
879 goto again;
880 inode_unlock(inode);
881 return err ? err : count;
882}
883
884static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
885 size_t count, struct cache_detail *cd)
886{
887 ssize_t ret;
888
889 if (count == 0)
890 return -EINVAL;
891 if (copy_from_user(kaddr, buf, count))
892 return -EFAULT;
893 kaddr[count] = '\0';
894 ret = cd->cache_parse(cd, kaddr, count);
895 if (!ret)
896 ret = count;
897 return ret;
898}
899
900static ssize_t cache_slow_downcall(const char __user *buf,
901 size_t count, struct cache_detail *cd)
902{
903 static char write_buf[8192]; /* protected by queue_io_mutex */
904 ssize_t ret = -EINVAL;
905
906 if (count >= sizeof(write_buf))
907 goto out;
908 mutex_lock(&queue_io_mutex);
909 ret = cache_do_downcall(write_buf, buf, count, cd);
910 mutex_unlock(&queue_io_mutex);
911out:
912 return ret;
913}
914
915static ssize_t cache_downcall(struct address_space *mapping,
916 const char __user *buf,
917 size_t count, struct cache_detail *cd)
918{
919 struct page *page;
920 char *kaddr;
921 ssize_t ret = -ENOMEM;
922
923 if (count >= PAGE_SIZE)
924 goto out_slow;
925
926 page = find_or_create_page(mapping, 0, GFP_KERNEL);
927 if (!page)
928 goto out_slow;
929
930 kaddr = kmap(page);
931 ret = cache_do_downcall(kaddr, buf, count, cd);
932 kunmap(page);
933 unlock_page(page);
934 put_page(page);
935 return ret;
936out_slow:
937 return cache_slow_downcall(buf, count, cd);
938}
939
940static ssize_t cache_write(struct file *filp, const char __user *buf,
941 size_t count, loff_t *ppos,
942 struct cache_detail *cd)
943{
944 struct address_space *mapping = filp->f_mapping;
945 struct inode *inode = file_inode(filp);
946 ssize_t ret = -EINVAL;
947
948 if (!cd->cache_parse)
949 goto out;
950
951 inode_lock(inode);
952 ret = cache_downcall(mapping, buf, count, cd);
953 inode_unlock(inode);
954out:
955 return ret;
956}
957
958static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
959
960static __poll_t cache_poll(struct file *filp, poll_table *wait,
961 struct cache_detail *cd)
962{
963 __poll_t mask;
964 struct cache_reader *rp = filp->private_data;
965 struct cache_queue *cq;
966
967 poll_wait(filp, &queue_wait, wait);
968
969 /* alway allow write */
970 mask = EPOLLOUT | EPOLLWRNORM;
971
972 if (!rp)
973 return mask;
974
975 spin_lock(&queue_lock);
976
977 for (cq= &rp->q; &cq->list != &cd->queue;
978 cq = list_entry(cq->list.next, struct cache_queue, list))
979 if (!cq->reader) {
980 mask |= EPOLLIN | EPOLLRDNORM;
981 break;
982 }
983 spin_unlock(&queue_lock);
984 return mask;
985}
986
987static int cache_ioctl(struct inode *ino, struct file *filp,
988 unsigned int cmd, unsigned long arg,
989 struct cache_detail *cd)
990{
991 int len = 0;
992 struct cache_reader *rp = filp->private_data;
993 struct cache_queue *cq;
994
995 if (cmd != FIONREAD || !rp)
996 return -EINVAL;
997
998 spin_lock(&queue_lock);
999
1000 /* only find the length remaining in current request,
1001 * or the length of the next request
1002 */
1003 for (cq= &rp->q; &cq->list != &cd->queue;
1004 cq = list_entry(cq->list.next, struct cache_queue, list))
1005 if (!cq->reader) {
1006 struct cache_request *cr =
1007 container_of(cq, struct cache_request, q);
1008 len = cr->len - rp->offset;
1009 break;
1010 }
1011 spin_unlock(&queue_lock);
1012
1013 return put_user(len, (int __user *)arg);
1014}
1015
1016static int cache_open(struct inode *inode, struct file *filp,
1017 struct cache_detail *cd)
1018{
1019 struct cache_reader *rp = NULL;
1020
1021 if (!cd || !try_module_get(cd->owner))
1022 return -EACCES;
1023 nonseekable_open(inode, filp);
1024 if (filp->f_mode & FMODE_READ) {
1025 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1026 if (!rp) {
1027 module_put(cd->owner);
1028 return -ENOMEM;
1029 }
1030 rp->offset = 0;
1031 rp->q.reader = 1;
1032
1033 spin_lock(&queue_lock);
1034 list_add(&rp->q.list, &cd->queue);
1035 spin_unlock(&queue_lock);
1036 }
1037 if (filp->f_mode & FMODE_WRITE)
1038 atomic_inc(&cd->writers);
1039 filp->private_data = rp;
1040 return 0;
1041}
1042
1043static int cache_release(struct inode *inode, struct file *filp,
1044 struct cache_detail *cd)
1045{
1046 struct cache_reader *rp = filp->private_data;
1047
1048 if (rp) {
1049 spin_lock(&queue_lock);
1050 if (rp->offset) {
1051 struct cache_queue *cq;
1052 for (cq= &rp->q; &cq->list != &cd->queue;
1053 cq = list_entry(cq->list.next, struct cache_queue, list))
1054 if (!cq->reader) {
1055 container_of(cq, struct cache_request, q)
1056 ->readers--;
1057 break;
1058 }
1059 rp->offset = 0;
1060 }
1061 list_del(&rp->q.list);
1062 spin_unlock(&queue_lock);
1063
1064 filp->private_data = NULL;
1065 kfree(rp);
1066
1067 }
1068 if (filp->f_mode & FMODE_WRITE) {
1069 atomic_dec(&cd->writers);
1070 cd->last_close = seconds_since_boot();
1071 }
1072 module_put(cd->owner);
1073 return 0;
1074}
1075
1076
1077
1078static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1079{
1080 struct cache_queue *cq, *tmp;
1081 struct cache_request *cr;
1082 struct list_head dequeued;
1083
1084 INIT_LIST_HEAD(&dequeued);
1085 spin_lock(&queue_lock);
1086 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1087 if (!cq->reader) {
1088 cr = container_of(cq, struct cache_request, q);
1089 if (cr->item != ch)
1090 continue;
1091 if (test_bit(CACHE_PENDING, &ch->flags))
1092 /* Lost a race and it is pending again */
1093 break;
1094 if (cr->readers != 0)
1095 continue;
1096 list_move(&cr->q.list, &dequeued);
1097 }
1098 spin_unlock(&queue_lock);
1099 while (!list_empty(&dequeued)) {
1100 cr = list_entry(dequeued.next, struct cache_request, q.list);
1101 list_del(&cr->q.list);
1102 cache_put(cr->item, detail);
1103 kfree(cr->buf);
1104 kfree(cr);
1105 }
1106}
1107
1108/*
1109 * Support routines for text-based upcalls.
1110 * Fields are separated by spaces.
1111 * Fields are either mangled to quote space tab newline slosh with slosh
1112 * or a hexified with a leading \x
1113 * Record is terminated with newline.
1114 *
1115 */
1116
1117void qword_add(char **bpp, int *lp, char *str)
1118{
1119 char *bp = *bpp;
1120 int len = *lp;
1121 int ret;
1122
1123 if (len < 0) return;
1124
1125 ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1126 if (ret >= len) {
1127 bp += len;
1128 len = -1;
1129 } else {
1130 bp += ret;
1131 len -= ret;
1132 *bp++ = ' ';
1133 len--;
1134 }
1135 *bpp = bp;
1136 *lp = len;
1137}
1138EXPORT_SYMBOL_GPL(qword_add);
1139
1140void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1141{
1142 char *bp = *bpp;
1143 int len = *lp;
1144
1145 if (len < 0) return;
1146
1147 if (len > 2) {
1148 *bp++ = '\\';
1149 *bp++ = 'x';
1150 len -= 2;
1151 while (blen && len >= 2) {
1152 bp = hex_byte_pack(bp, *buf++);
1153 len -= 2;
1154 blen--;
1155 }
1156 }
1157 if (blen || len<1) len = -1;
1158 else {
1159 *bp++ = ' ';
1160 len--;
1161 }
1162 *bpp = bp;
1163 *lp = len;
1164}
1165EXPORT_SYMBOL_GPL(qword_addhex);
1166
1167static void warn_no_listener(struct cache_detail *detail)
1168{
1169 if (detail->last_warn != detail->last_close) {
1170 detail->last_warn = detail->last_close;
1171 if (detail->warn_no_listener)
1172 detail->warn_no_listener(detail, detail->last_close != 0);
1173 }
1174}
1175
1176static bool cache_listeners_exist(struct cache_detail *detail)
1177{
1178 if (atomic_read(&detail->writers))
1179 return true;
1180 if (detail->last_close == 0)
1181 /* This cache was never opened */
1182 return false;
1183 if (detail->last_close < seconds_since_boot() - 30)
1184 /*
1185 * We allow for the possibility that someone might
1186 * restart a userspace daemon without restarting the
1187 * server; but after 30 seconds, we give up.
1188 */
1189 return false;
1190 return true;
1191}
1192
1193/*
1194 * register an upcall request to user-space and queue it up for read() by the
1195 * upcall daemon.
1196 *
1197 * Each request is at most one page long.
1198 */
1199int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1200{
1201
1202 char *buf;
1203 struct cache_request *crq;
1204 int ret = 0;
1205
1206 if (!detail->cache_request)
1207 return -EINVAL;
1208
1209 if (!cache_listeners_exist(detail)) {
1210 warn_no_listener(detail);
1211 return -EINVAL;
1212 }
1213 if (test_bit(CACHE_CLEANED, &h->flags))
1214 /* Too late to make an upcall */
1215 return -EAGAIN;
1216
1217 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1218 if (!buf)
1219 return -EAGAIN;
1220
1221 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1222 if (!crq) {
1223 kfree(buf);
1224 return -EAGAIN;
1225 }
1226
1227 crq->q.reader = 0;
1228 crq->buf = buf;
1229 crq->len = 0;
1230 crq->readers = 0;
1231 spin_lock(&queue_lock);
1232 if (test_bit(CACHE_PENDING, &h->flags)) {
1233 crq->item = cache_get(h);
1234 list_add_tail(&crq->q.list, &detail->queue);
1235 } else
1236 /* Lost a race, no longer PENDING, so don't enqueue */
1237 ret = -EAGAIN;
1238 spin_unlock(&queue_lock);
1239 wake_up(&queue_wait);
1240 if (ret == -EAGAIN) {
1241 kfree(buf);
1242 kfree(crq);
1243 }
1244 return ret;
1245}
1246EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1247
1248/*
1249 * parse a message from user-space and pass it
1250 * to an appropriate cache
1251 * Messages are, like requests, separated into fields by
1252 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1253 *
1254 * Message is
1255 * reply cachename expiry key ... content....
1256 *
1257 * key and content are both parsed by cache
1258 */
1259
1260int qword_get(char **bpp, char *dest, int bufsize)
1261{
1262 /* return bytes copied, or -1 on error */
1263 char *bp = *bpp;
1264 int len = 0;
1265
1266 while (*bp == ' ') bp++;
1267
1268 if (bp[0] == '\\' && bp[1] == 'x') {
1269 /* HEX STRING */
1270 bp += 2;
1271 while (len < bufsize - 1) {
1272 int h, l;
1273
1274 h = hex_to_bin(bp[0]);
1275 if (h < 0)
1276 break;
1277
1278 l = hex_to_bin(bp[1]);
1279 if (l < 0)
1280 break;
1281
1282 *dest++ = (h << 4) | l;
1283 bp += 2;
1284 len++;
1285 }
1286 } else {
1287 /* text with \nnn octal quoting */
1288 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1289 if (*bp == '\\' &&
1290 isodigit(bp[1]) && (bp[1] <= '3') &&
1291 isodigit(bp[2]) &&
1292 isodigit(bp[3])) {
1293 int byte = (*++bp -'0');
1294 bp++;
1295 byte = (byte << 3) | (*bp++ - '0');
1296 byte = (byte << 3) | (*bp++ - '0');
1297 *dest++ = byte;
1298 len++;
1299 } else {
1300 *dest++ = *bp++;
1301 len++;
1302 }
1303 }
1304 }
1305
1306 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1307 return -1;
1308 while (*bp == ' ') bp++;
1309 *bpp = bp;
1310 *dest = '\0';
1311 return len;
1312}
1313EXPORT_SYMBOL_GPL(qword_get);
1314
1315
1316/*
1317 * support /proc/net/rpc/$CACHENAME/content
1318 * as a seqfile.
1319 * We call ->cache_show passing NULL for the item to
1320 * get a header, then pass each real item in the cache
1321 */
1322
1323static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1324{
1325 loff_t n = *pos;
1326 unsigned int hash, entry;
1327 struct cache_head *ch;
1328 struct cache_detail *cd = m->private;
1329
1330 if (!n--)
1331 return SEQ_START_TOKEN;
1332 hash = n >> 32;
1333 entry = n & ((1LL<<32) - 1);
1334
1335 hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1336 if (!entry--)
1337 return ch;
1338 n &= ~((1LL<<32) - 1);
1339 do {
1340 hash++;
1341 n += 1LL<<32;
1342 } while(hash < cd->hash_size &&
1343 hlist_empty(&cd->hash_table[hash]));
1344 if (hash >= cd->hash_size)
1345 return NULL;
1346 *pos = n+1;
1347 return hlist_entry_safe(rcu_dereference_raw(
1348 hlist_first_rcu(&cd->hash_table[hash])),
1349 struct cache_head, cache_list);
1350}
1351
1352static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1353{
1354 struct cache_head *ch = p;
1355 int hash = (*pos >> 32);
1356 struct cache_detail *cd = m->private;
1357
1358 if (p == SEQ_START_TOKEN)
1359 hash = 0;
1360 else if (ch->cache_list.next == NULL) {
1361 hash++;
1362 *pos += 1LL<<32;
1363 } else {
1364 ++*pos;
1365 return hlist_entry_safe(rcu_dereference_raw(
1366 hlist_next_rcu(&ch->cache_list)),
1367 struct cache_head, cache_list);
1368 }
1369 *pos &= ~((1LL<<32) - 1);
1370 while (hash < cd->hash_size &&
1371 hlist_empty(&cd->hash_table[hash])) {
1372 hash++;
1373 *pos += 1LL<<32;
1374 }
1375 if (hash >= cd->hash_size)
1376 return NULL;
1377 ++*pos;
1378 return hlist_entry_safe(rcu_dereference_raw(
1379 hlist_first_rcu(&cd->hash_table[hash])),
1380 struct cache_head, cache_list);
1381}
1382
1383void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1384 __acquires(RCU)
1385{
1386 rcu_read_lock();
1387 return __cache_seq_start(m, pos);
1388}
1389EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1390
1391void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1392{
1393 return cache_seq_next(file, p, pos);
1394}
1395EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1396
1397void cache_seq_stop_rcu(struct seq_file *m, void *p)
1398 __releases(RCU)
1399{
1400 rcu_read_unlock();
1401}
1402EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1403
1404static int c_show(struct seq_file *m, void *p)
1405{
1406 struct cache_head *cp = p;
1407 struct cache_detail *cd = m->private;
1408
1409 if (p == SEQ_START_TOKEN)
1410 return cd->cache_show(m, cd, NULL);
1411
1412 ifdebug(CACHE)
1413 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1414 convert_to_wallclock(cp->expiry_time),
1415 kref_read(&cp->ref), cp->flags);
1416 cache_get(cp);
1417 if (cache_check(cd, cp, NULL))
1418 /* cache_check does a cache_put on failure */
1419 seq_printf(m, "# ");
1420 else {
1421 if (cache_is_expired(cd, cp))
1422 seq_printf(m, "# ");
1423 cache_put(cp, cd);
1424 }
1425
1426 return cd->cache_show(m, cd, cp);
1427}
1428
1429static const struct seq_operations cache_content_op = {
1430 .start = cache_seq_start_rcu,
1431 .next = cache_seq_next_rcu,
1432 .stop = cache_seq_stop_rcu,
1433 .show = c_show,
1434};
1435
1436static int content_open(struct inode *inode, struct file *file,
1437 struct cache_detail *cd)
1438{
1439 struct seq_file *seq;
1440 int err;
1441
1442 if (!cd || !try_module_get(cd->owner))
1443 return -EACCES;
1444
1445 err = seq_open(file, &cache_content_op);
1446 if (err) {
1447 module_put(cd->owner);
1448 return err;
1449 }
1450
1451 seq = file->private_data;
1452 seq->private = cd;
1453 return 0;
1454}
1455
1456static int content_release(struct inode *inode, struct file *file,
1457 struct cache_detail *cd)
1458{
1459 int ret = seq_release(inode, file);
1460 module_put(cd->owner);
1461 return ret;
1462}
1463
1464static int open_flush(struct inode *inode, struct file *file,
1465 struct cache_detail *cd)
1466{
1467 if (!cd || !try_module_get(cd->owner))
1468 return -EACCES;
1469 return nonseekable_open(inode, file);
1470}
1471
1472static int release_flush(struct inode *inode, struct file *file,
1473 struct cache_detail *cd)
1474{
1475 module_put(cd->owner);
1476 return 0;
1477}
1478
1479static ssize_t read_flush(struct file *file, char __user *buf,
1480 size_t count, loff_t *ppos,
1481 struct cache_detail *cd)
1482{
1483 char tbuf[22];
1484 size_t len;
1485
1486 len = snprintf(tbuf, sizeof(tbuf), "%lu\n",
1487 convert_to_wallclock(cd->flush_time));
1488 return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1489}
1490
1491static ssize_t write_flush(struct file *file, const char __user *buf,
1492 size_t count, loff_t *ppos,
1493 struct cache_detail *cd)
1494{
1495 char tbuf[20];
1496 char *ep;
1497 time_t now;
1498
1499 if (*ppos || count > sizeof(tbuf)-1)
1500 return -EINVAL;
1501 if (copy_from_user(tbuf, buf, count))
1502 return -EFAULT;
1503 tbuf[count] = 0;
1504 simple_strtoul(tbuf, &ep, 0);
1505 if (*ep && *ep != '\n')
1506 return -EINVAL;
1507 /* Note that while we check that 'buf' holds a valid number,
1508 * we always ignore the value and just flush everything.
1509 * Making use of the number leads to races.
1510 */
1511
1512 now = seconds_since_boot();
1513 /* Always flush everything, so behave like cache_purge()
1514 * Do this by advancing flush_time to the current time,
1515 * or by one second if it has already reached the current time.
1516 * Newly added cache entries will always have ->last_refresh greater
1517 * that ->flush_time, so they don't get flushed prematurely.
1518 */
1519
1520 if (cd->flush_time >= now)
1521 now = cd->flush_time + 1;
1522
1523 cd->flush_time = now;
1524 cd->nextcheck = now;
1525 cache_flush();
1526
1527 if (cd->flush)
1528 cd->flush();
1529
1530 *ppos += count;
1531 return count;
1532}
1533
1534static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1535 size_t count, loff_t *ppos)
1536{
1537 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1538
1539 return cache_read(filp, buf, count, ppos, cd);
1540}
1541
1542static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1543 size_t count, loff_t *ppos)
1544{
1545 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1546
1547 return cache_write(filp, buf, count, ppos, cd);
1548}
1549
1550static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1551{
1552 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1553
1554 return cache_poll(filp, wait, cd);
1555}
1556
1557static long cache_ioctl_procfs(struct file *filp,
1558 unsigned int cmd, unsigned long arg)
1559{
1560 struct inode *inode = file_inode(filp);
1561 struct cache_detail *cd = PDE_DATA(inode);
1562
1563 return cache_ioctl(inode, filp, cmd, arg, cd);
1564}
1565
1566static int cache_open_procfs(struct inode *inode, struct file *filp)
1567{
1568 struct cache_detail *cd = PDE_DATA(inode);
1569
1570 return cache_open(inode, filp, cd);
1571}
1572
1573static int cache_release_procfs(struct inode *inode, struct file *filp)
1574{
1575 struct cache_detail *cd = PDE_DATA(inode);
1576
1577 return cache_release(inode, filp, cd);
1578}
1579
1580static const struct file_operations cache_file_operations_procfs = {
1581 .owner = THIS_MODULE,
1582 .llseek = no_llseek,
1583 .read = cache_read_procfs,
1584 .write = cache_write_procfs,
1585 .poll = cache_poll_procfs,
1586 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1587 .open = cache_open_procfs,
1588 .release = cache_release_procfs,
1589};
1590
1591static int content_open_procfs(struct inode *inode, struct file *filp)
1592{
1593 struct cache_detail *cd = PDE_DATA(inode);
1594
1595 return content_open(inode, filp, cd);
1596}
1597
1598static int content_release_procfs(struct inode *inode, struct file *filp)
1599{
1600 struct cache_detail *cd = PDE_DATA(inode);
1601
1602 return content_release(inode, filp, cd);
1603}
1604
1605static const struct file_operations content_file_operations_procfs = {
1606 .open = content_open_procfs,
1607 .read = seq_read,
1608 .llseek = seq_lseek,
1609 .release = content_release_procfs,
1610};
1611
1612static int open_flush_procfs(struct inode *inode, struct file *filp)
1613{
1614 struct cache_detail *cd = PDE_DATA(inode);
1615
1616 return open_flush(inode, filp, cd);
1617}
1618
1619static int release_flush_procfs(struct inode *inode, struct file *filp)
1620{
1621 struct cache_detail *cd = PDE_DATA(inode);
1622
1623 return release_flush(inode, filp, cd);
1624}
1625
1626static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1627 size_t count, loff_t *ppos)
1628{
1629 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1630
1631 return read_flush(filp, buf, count, ppos, cd);
1632}
1633
1634static ssize_t write_flush_procfs(struct file *filp,
1635 const char __user *buf,
1636 size_t count, loff_t *ppos)
1637{
1638 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1639
1640 return write_flush(filp, buf, count, ppos, cd);
1641}
1642
1643static const struct file_operations cache_flush_operations_procfs = {
1644 .open = open_flush_procfs,
1645 .read = read_flush_procfs,
1646 .write = write_flush_procfs,
1647 .release = release_flush_procfs,
1648 .llseek = no_llseek,
1649};
1650
1651static void remove_cache_proc_entries(struct cache_detail *cd)
1652{
1653 if (cd->procfs) {
1654 proc_remove(cd->procfs);
1655 cd->procfs = NULL;
1656 }
1657}
1658
1659#ifdef CONFIG_PROC_FS
1660static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1661{
1662 struct proc_dir_entry *p;
1663 struct sunrpc_net *sn;
1664
1665 sn = net_generic(net, sunrpc_net_id);
1666 cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1667 if (cd->procfs == NULL)
1668 goto out_nomem;
1669
1670 p = proc_create_data("flush", S_IFREG | 0600,
1671 cd->procfs, &cache_flush_operations_procfs, cd);
1672 if (p == NULL)
1673 goto out_nomem;
1674
1675 if (cd->cache_request || cd->cache_parse) {
1676 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1677 &cache_file_operations_procfs, cd);
1678 if (p == NULL)
1679 goto out_nomem;
1680 }
1681 if (cd->cache_show) {
1682 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1683 &content_file_operations_procfs, cd);
1684 if (p == NULL)
1685 goto out_nomem;
1686 }
1687 return 0;
1688out_nomem:
1689 remove_cache_proc_entries(cd);
1690 return -ENOMEM;
1691}
1692#else /* CONFIG_PROC_FS */
1693static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1694{
1695 return 0;
1696}
1697#endif
1698
1699void __init cache_initialize(void)
1700{
1701 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1702}
1703
1704int cache_register_net(struct cache_detail *cd, struct net *net)
1705{
1706 int ret;
1707
1708 sunrpc_init_cache_detail(cd);
1709 ret = create_cache_proc_entries(cd, net);
1710 if (ret)
1711 sunrpc_destroy_cache_detail(cd);
1712 return ret;
1713}
1714EXPORT_SYMBOL_GPL(cache_register_net);
1715
1716void cache_unregister_net(struct cache_detail *cd, struct net *net)
1717{
1718 remove_cache_proc_entries(cd);
1719 sunrpc_destroy_cache_detail(cd);
1720}
1721EXPORT_SYMBOL_GPL(cache_unregister_net);
1722
1723struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1724{
1725 struct cache_detail *cd;
1726 int i;
1727
1728 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1729 if (cd == NULL)
1730 return ERR_PTR(-ENOMEM);
1731
1732 cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1733 GFP_KERNEL);
1734 if (cd->hash_table == NULL) {
1735 kfree(cd);
1736 return ERR_PTR(-ENOMEM);
1737 }
1738
1739 for (i = 0; i < cd->hash_size; i++)
1740 INIT_HLIST_HEAD(&cd->hash_table[i]);
1741 cd->net = net;
1742 return cd;
1743}
1744EXPORT_SYMBOL_GPL(cache_create_net);
1745
1746void cache_destroy_net(struct cache_detail *cd, struct net *net)
1747{
1748 kfree(cd->hash_table);
1749 kfree(cd);
1750}
1751EXPORT_SYMBOL_GPL(cache_destroy_net);
1752
1753static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1754 size_t count, loff_t *ppos)
1755{
1756 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1757
1758 return cache_read(filp, buf, count, ppos, cd);
1759}
1760
1761static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1762 size_t count, loff_t *ppos)
1763{
1764 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1765
1766 return cache_write(filp, buf, count, ppos, cd);
1767}
1768
1769static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1770{
1771 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1772
1773 return cache_poll(filp, wait, cd);
1774}
1775
1776static long cache_ioctl_pipefs(struct file *filp,
1777 unsigned int cmd, unsigned long arg)
1778{
1779 struct inode *inode = file_inode(filp);
1780 struct cache_detail *cd = RPC_I(inode)->private;
1781
1782 return cache_ioctl(inode, filp, cmd, arg, cd);
1783}
1784
1785static int cache_open_pipefs(struct inode *inode, struct file *filp)
1786{
1787 struct cache_detail *cd = RPC_I(inode)->private;
1788
1789 return cache_open(inode, filp, cd);
1790}
1791
1792static int cache_release_pipefs(struct inode *inode, struct file *filp)
1793{
1794 struct cache_detail *cd = RPC_I(inode)->private;
1795
1796 return cache_release(inode, filp, cd);
1797}
1798
1799const struct file_operations cache_file_operations_pipefs = {
1800 .owner = THIS_MODULE,
1801 .llseek = no_llseek,
1802 .read = cache_read_pipefs,
1803 .write = cache_write_pipefs,
1804 .poll = cache_poll_pipefs,
1805 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1806 .open = cache_open_pipefs,
1807 .release = cache_release_pipefs,
1808};
1809
1810static int content_open_pipefs(struct inode *inode, struct file *filp)
1811{
1812 struct cache_detail *cd = RPC_I(inode)->private;
1813
1814 return content_open(inode, filp, cd);
1815}
1816
1817static int content_release_pipefs(struct inode *inode, struct file *filp)
1818{
1819 struct cache_detail *cd = RPC_I(inode)->private;
1820
1821 return content_release(inode, filp, cd);
1822}
1823
1824const struct file_operations content_file_operations_pipefs = {
1825 .open = content_open_pipefs,
1826 .read = seq_read,
1827 .llseek = seq_lseek,
1828 .release = content_release_pipefs,
1829};
1830
1831static int open_flush_pipefs(struct inode *inode, struct file *filp)
1832{
1833 struct cache_detail *cd = RPC_I(inode)->private;
1834
1835 return open_flush(inode, filp, cd);
1836}
1837
1838static int release_flush_pipefs(struct inode *inode, struct file *filp)
1839{
1840 struct cache_detail *cd = RPC_I(inode)->private;
1841
1842 return release_flush(inode, filp, cd);
1843}
1844
1845static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1846 size_t count, loff_t *ppos)
1847{
1848 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1849
1850 return read_flush(filp, buf, count, ppos, cd);
1851}
1852
1853static ssize_t write_flush_pipefs(struct file *filp,
1854 const char __user *buf,
1855 size_t count, loff_t *ppos)
1856{
1857 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1858
1859 return write_flush(filp, buf, count, ppos, cd);
1860}
1861
1862const struct file_operations cache_flush_operations_pipefs = {
1863 .open = open_flush_pipefs,
1864 .read = read_flush_pipefs,
1865 .write = write_flush_pipefs,
1866 .release = release_flush_pipefs,
1867 .llseek = no_llseek,
1868};
1869
1870int sunrpc_cache_register_pipefs(struct dentry *parent,
1871 const char *name, umode_t umode,
1872 struct cache_detail *cd)
1873{
1874 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1875 if (IS_ERR(dir))
1876 return PTR_ERR(dir);
1877 cd->pipefs = dir;
1878 return 0;
1879}
1880EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1881
1882void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1883{
1884 if (cd->pipefs) {
1885 rpc_remove_cache_dir(cd->pipefs);
1886 cd->pipefs = NULL;
1887 }
1888}
1889EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1890
1891void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1892{
1893 spin_lock(&cd->hash_lock);
1894 if (!hlist_unhashed(&h->cache_list)){
1895 hlist_del_init_rcu(&h->cache_list);
1896 cd->entries--;
1897 spin_unlock(&cd->hash_lock);
1898 cache_put(h, cd);
1899 } else
1900 spin_unlock(&cd->hash_lock);
1901}
1902EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);
1/*
2 * net/sunrpc/cache.c
3 *
4 * Generic code for various authentication-related caches
5 * used by sunrpc clients and servers.
6 *
7 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
8 *
9 * Released under terms in GPL version 2. See COPYING.
10 *
11 */
12
13#include <linux/types.h>
14#include <linux/fs.h>
15#include <linux/file.h>
16#include <linux/slab.h>
17#include <linux/signal.h>
18#include <linux/sched.h>
19#include <linux/kmod.h>
20#include <linux/list.h>
21#include <linux/module.h>
22#include <linux/ctype.h>
23#include <asm/uaccess.h>
24#include <linux/poll.h>
25#include <linux/seq_file.h>
26#include <linux/proc_fs.h>
27#include <linux/net.h>
28#include <linux/workqueue.h>
29#include <linux/mutex.h>
30#include <linux/pagemap.h>
31#include <asm/ioctls.h>
32#include <linux/sunrpc/types.h>
33#include <linux/sunrpc/cache.h>
34#include <linux/sunrpc/stats.h>
35#include <linux/sunrpc/rpc_pipe_fs.h>
36#include "netns.h"
37
38#define RPCDBG_FACILITY RPCDBG_CACHE
39
40static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
41static void cache_revisit_request(struct cache_head *item);
42
43static void cache_init(struct cache_head *h)
44{
45 time_t now = seconds_since_boot();
46 h->next = NULL;
47 h->flags = 0;
48 kref_init(&h->ref);
49 h->expiry_time = now + CACHE_NEW_EXPIRY;
50 h->last_refresh = now;
51}
52
53struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
54 struct cache_head *key, int hash)
55{
56 struct cache_head **head, **hp;
57 struct cache_head *new = NULL, *freeme = NULL;
58
59 head = &detail->hash_table[hash];
60
61 read_lock(&detail->hash_lock);
62
63 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
64 struct cache_head *tmp = *hp;
65 if (detail->match(tmp, key)) {
66 if (cache_is_expired(detail, tmp))
67 /* This entry is expired, we will discard it. */
68 break;
69 cache_get(tmp);
70 read_unlock(&detail->hash_lock);
71 return tmp;
72 }
73 }
74 read_unlock(&detail->hash_lock);
75 /* Didn't find anything, insert an empty entry */
76
77 new = detail->alloc();
78 if (!new)
79 return NULL;
80 /* must fully initialise 'new', else
81 * we might get lose if we need to
82 * cache_put it soon.
83 */
84 cache_init(new);
85 detail->init(new, key);
86
87 write_lock(&detail->hash_lock);
88
89 /* check if entry appeared while we slept */
90 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
91 struct cache_head *tmp = *hp;
92 if (detail->match(tmp, key)) {
93 if (cache_is_expired(detail, tmp)) {
94 *hp = tmp->next;
95 tmp->next = NULL;
96 detail->entries --;
97 freeme = tmp;
98 break;
99 }
100 cache_get(tmp);
101 write_unlock(&detail->hash_lock);
102 cache_put(new, detail);
103 return tmp;
104 }
105 }
106 new->next = *head;
107 *head = new;
108 detail->entries++;
109 cache_get(new);
110 write_unlock(&detail->hash_lock);
111
112 if (freeme)
113 cache_put(freeme, detail);
114 return new;
115}
116EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
117
118
119static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
120
121static void cache_fresh_locked(struct cache_head *head, time_t expiry)
122{
123 head->expiry_time = expiry;
124 head->last_refresh = seconds_since_boot();
125 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
126 set_bit(CACHE_VALID, &head->flags);
127}
128
129static void cache_fresh_unlocked(struct cache_head *head,
130 struct cache_detail *detail)
131{
132 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
133 cache_revisit_request(head);
134 cache_dequeue(detail, head);
135 }
136}
137
138struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
139 struct cache_head *new, struct cache_head *old, int hash)
140{
141 /* The 'old' entry is to be replaced by 'new'.
142 * If 'old' is not VALID, we update it directly,
143 * otherwise we need to replace it
144 */
145 struct cache_head **head;
146 struct cache_head *tmp;
147
148 if (!test_bit(CACHE_VALID, &old->flags)) {
149 write_lock(&detail->hash_lock);
150 if (!test_bit(CACHE_VALID, &old->flags)) {
151 if (test_bit(CACHE_NEGATIVE, &new->flags))
152 set_bit(CACHE_NEGATIVE, &old->flags);
153 else
154 detail->update(old, new);
155 cache_fresh_locked(old, new->expiry_time);
156 write_unlock(&detail->hash_lock);
157 cache_fresh_unlocked(old, detail);
158 return old;
159 }
160 write_unlock(&detail->hash_lock);
161 }
162 /* We need to insert a new entry */
163 tmp = detail->alloc();
164 if (!tmp) {
165 cache_put(old, detail);
166 return NULL;
167 }
168 cache_init(tmp);
169 detail->init(tmp, old);
170 head = &detail->hash_table[hash];
171
172 write_lock(&detail->hash_lock);
173 if (test_bit(CACHE_NEGATIVE, &new->flags))
174 set_bit(CACHE_NEGATIVE, &tmp->flags);
175 else
176 detail->update(tmp, new);
177 tmp->next = *head;
178 *head = tmp;
179 detail->entries++;
180 cache_get(tmp);
181 cache_fresh_locked(tmp, new->expiry_time);
182 cache_fresh_locked(old, 0);
183 write_unlock(&detail->hash_lock);
184 cache_fresh_unlocked(tmp, detail);
185 cache_fresh_unlocked(old, detail);
186 cache_put(old, detail);
187 return tmp;
188}
189EXPORT_SYMBOL_GPL(sunrpc_cache_update);
190
191static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
192{
193 if (cd->cache_upcall)
194 return cd->cache_upcall(cd, h);
195 return sunrpc_cache_pipe_upcall(cd, h);
196}
197
198static inline int cache_is_valid(struct cache_head *h)
199{
200 if (!test_bit(CACHE_VALID, &h->flags))
201 return -EAGAIN;
202 else {
203 /* entry is valid */
204 if (test_bit(CACHE_NEGATIVE, &h->flags))
205 return -ENOENT;
206 else {
207 /*
208 * In combination with write barrier in
209 * sunrpc_cache_update, ensures that anyone
210 * using the cache entry after this sees the
211 * updated contents:
212 */
213 smp_rmb();
214 return 0;
215 }
216 }
217}
218
219static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
220{
221 int rv;
222
223 write_lock(&detail->hash_lock);
224 rv = cache_is_valid(h);
225 if (rv == -EAGAIN) {
226 set_bit(CACHE_NEGATIVE, &h->flags);
227 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
228 rv = -ENOENT;
229 }
230 write_unlock(&detail->hash_lock);
231 cache_fresh_unlocked(h, detail);
232 return rv;
233}
234
235/*
236 * This is the generic cache management routine for all
237 * the authentication caches.
238 * It checks the currency of a cache item and will (later)
239 * initiate an upcall to fill it if needed.
240 *
241 *
242 * Returns 0 if the cache_head can be used, or cache_puts it and returns
243 * -EAGAIN if upcall is pending and request has been queued
244 * -ETIMEDOUT if upcall failed or request could not be queue or
245 * upcall completed but item is still invalid (implying that
246 * the cache item has been replaced with a newer one).
247 * -ENOENT if cache entry was negative
248 */
249int cache_check(struct cache_detail *detail,
250 struct cache_head *h, struct cache_req *rqstp)
251{
252 int rv;
253 long refresh_age, age;
254
255 /* First decide return status as best we can */
256 rv = cache_is_valid(h);
257
258 /* now see if we want to start an upcall */
259 refresh_age = (h->expiry_time - h->last_refresh);
260 age = seconds_since_boot() - h->last_refresh;
261
262 if (rqstp == NULL) {
263 if (rv == -EAGAIN)
264 rv = -ENOENT;
265 } else if (rv == -EAGAIN ||
266 (h->expiry_time != 0 && age > refresh_age/2)) {
267 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
268 refresh_age, age);
269 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
270 switch (cache_make_upcall(detail, h)) {
271 case -EINVAL:
272 rv = try_to_negate_entry(detail, h);
273 break;
274 case -EAGAIN:
275 cache_fresh_unlocked(h, detail);
276 break;
277 }
278 }
279 }
280
281 if (rv == -EAGAIN) {
282 if (!cache_defer_req(rqstp, h)) {
283 /*
284 * Request was not deferred; handle it as best
285 * we can ourselves:
286 */
287 rv = cache_is_valid(h);
288 if (rv == -EAGAIN)
289 rv = -ETIMEDOUT;
290 }
291 }
292 if (rv)
293 cache_put(h, detail);
294 return rv;
295}
296EXPORT_SYMBOL_GPL(cache_check);
297
298/*
299 * caches need to be periodically cleaned.
300 * For this we maintain a list of cache_detail and
301 * a current pointer into that list and into the table
302 * for that entry.
303 *
304 * Each time cache_clean is called it finds the next non-empty entry
305 * in the current table and walks the list in that entry
306 * looking for entries that can be removed.
307 *
308 * An entry gets removed if:
309 * - The expiry is before current time
310 * - The last_refresh time is before the flush_time for that cache
311 *
312 * later we might drop old entries with non-NEVER expiry if that table
313 * is getting 'full' for some definition of 'full'
314 *
315 * The question of "how often to scan a table" is an interesting one
316 * and is answered in part by the use of the "nextcheck" field in the
317 * cache_detail.
318 * When a scan of a table begins, the nextcheck field is set to a time
319 * that is well into the future.
320 * While scanning, if an expiry time is found that is earlier than the
321 * current nextcheck time, nextcheck is set to that expiry time.
322 * If the flush_time is ever set to a time earlier than the nextcheck
323 * time, the nextcheck time is then set to that flush_time.
324 *
325 * A table is then only scanned if the current time is at least
326 * the nextcheck time.
327 *
328 */
329
330static LIST_HEAD(cache_list);
331static DEFINE_SPINLOCK(cache_list_lock);
332static struct cache_detail *current_detail;
333static int current_index;
334
335static void do_cache_clean(struct work_struct *work);
336static struct delayed_work cache_cleaner;
337
338void sunrpc_init_cache_detail(struct cache_detail *cd)
339{
340 rwlock_init(&cd->hash_lock);
341 INIT_LIST_HEAD(&cd->queue);
342 spin_lock(&cache_list_lock);
343 cd->nextcheck = 0;
344 cd->entries = 0;
345 atomic_set(&cd->readers, 0);
346 cd->last_close = 0;
347 cd->last_warn = -1;
348 list_add(&cd->others, &cache_list);
349 spin_unlock(&cache_list_lock);
350
351 /* start the cleaning process */
352 schedule_delayed_work(&cache_cleaner, 0);
353}
354EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
355
356void sunrpc_destroy_cache_detail(struct cache_detail *cd)
357{
358 cache_purge(cd);
359 spin_lock(&cache_list_lock);
360 write_lock(&cd->hash_lock);
361 if (cd->entries || atomic_read(&cd->inuse)) {
362 write_unlock(&cd->hash_lock);
363 spin_unlock(&cache_list_lock);
364 goto out;
365 }
366 if (current_detail == cd)
367 current_detail = NULL;
368 list_del_init(&cd->others);
369 write_unlock(&cd->hash_lock);
370 spin_unlock(&cache_list_lock);
371 if (list_empty(&cache_list)) {
372 /* module must be being unloaded so its safe to kill the worker */
373 cancel_delayed_work_sync(&cache_cleaner);
374 }
375 return;
376out:
377 printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
378}
379EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
380
381/* clean cache tries to find something to clean
382 * and cleans it.
383 * It returns 1 if it cleaned something,
384 * 0 if it didn't find anything this time
385 * -1 if it fell off the end of the list.
386 */
387static int cache_clean(void)
388{
389 int rv = 0;
390 struct list_head *next;
391
392 spin_lock(&cache_list_lock);
393
394 /* find a suitable table if we don't already have one */
395 while (current_detail == NULL ||
396 current_index >= current_detail->hash_size) {
397 if (current_detail)
398 next = current_detail->others.next;
399 else
400 next = cache_list.next;
401 if (next == &cache_list) {
402 current_detail = NULL;
403 spin_unlock(&cache_list_lock);
404 return -1;
405 }
406 current_detail = list_entry(next, struct cache_detail, others);
407 if (current_detail->nextcheck > seconds_since_boot())
408 current_index = current_detail->hash_size;
409 else {
410 current_index = 0;
411 current_detail->nextcheck = seconds_since_boot()+30*60;
412 }
413 }
414
415 /* find a non-empty bucket in the table */
416 while (current_detail &&
417 current_index < current_detail->hash_size &&
418 current_detail->hash_table[current_index] == NULL)
419 current_index++;
420
421 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
422
423 if (current_detail && current_index < current_detail->hash_size) {
424 struct cache_head *ch, **cp;
425 struct cache_detail *d;
426
427 write_lock(¤t_detail->hash_lock);
428
429 /* Ok, now to clean this strand */
430
431 cp = & current_detail->hash_table[current_index];
432 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
433 if (current_detail->nextcheck > ch->expiry_time)
434 current_detail->nextcheck = ch->expiry_time+1;
435 if (!cache_is_expired(current_detail, ch))
436 continue;
437
438 *cp = ch->next;
439 ch->next = NULL;
440 current_detail->entries--;
441 rv = 1;
442 break;
443 }
444
445 write_unlock(¤t_detail->hash_lock);
446 d = current_detail;
447 if (!ch)
448 current_index ++;
449 spin_unlock(&cache_list_lock);
450 if (ch) {
451 set_bit(CACHE_CLEANED, &ch->flags);
452 cache_fresh_unlocked(ch, d);
453 cache_put(ch, d);
454 }
455 } else
456 spin_unlock(&cache_list_lock);
457
458 return rv;
459}
460
461/*
462 * We want to regularly clean the cache, so we need to schedule some work ...
463 */
464static void do_cache_clean(struct work_struct *work)
465{
466 int delay = 5;
467 if (cache_clean() == -1)
468 delay = round_jiffies_relative(30*HZ);
469
470 if (list_empty(&cache_list))
471 delay = 0;
472
473 if (delay)
474 schedule_delayed_work(&cache_cleaner, delay);
475}
476
477
478/*
479 * Clean all caches promptly. This just calls cache_clean
480 * repeatedly until we are sure that every cache has had a chance to
481 * be fully cleaned
482 */
483void cache_flush(void)
484{
485 while (cache_clean() != -1)
486 cond_resched();
487 while (cache_clean() != -1)
488 cond_resched();
489}
490EXPORT_SYMBOL_GPL(cache_flush);
491
492void cache_purge(struct cache_detail *detail)
493{
494 detail->flush_time = LONG_MAX;
495 detail->nextcheck = seconds_since_boot();
496 cache_flush();
497 detail->flush_time = 1;
498}
499EXPORT_SYMBOL_GPL(cache_purge);
500
501
502/*
503 * Deferral and Revisiting of Requests.
504 *
505 * If a cache lookup finds a pending entry, we
506 * need to defer the request and revisit it later.
507 * All deferred requests are stored in a hash table,
508 * indexed by "struct cache_head *".
509 * As it may be wasteful to store a whole request
510 * structure, we allow the request to provide a
511 * deferred form, which must contain a
512 * 'struct cache_deferred_req'
513 * This cache_deferred_req contains a method to allow
514 * it to be revisited when cache info is available
515 */
516
517#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
518#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
519
520#define DFR_MAX 300 /* ??? */
521
522static DEFINE_SPINLOCK(cache_defer_lock);
523static LIST_HEAD(cache_defer_list);
524static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
525static int cache_defer_cnt;
526
527static void __unhash_deferred_req(struct cache_deferred_req *dreq)
528{
529 hlist_del_init(&dreq->hash);
530 if (!list_empty(&dreq->recent)) {
531 list_del_init(&dreq->recent);
532 cache_defer_cnt--;
533 }
534}
535
536static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
537{
538 int hash = DFR_HASH(item);
539
540 INIT_LIST_HEAD(&dreq->recent);
541 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
542}
543
544static void setup_deferral(struct cache_deferred_req *dreq,
545 struct cache_head *item,
546 int count_me)
547{
548
549 dreq->item = item;
550
551 spin_lock(&cache_defer_lock);
552
553 __hash_deferred_req(dreq, item);
554
555 if (count_me) {
556 cache_defer_cnt++;
557 list_add(&dreq->recent, &cache_defer_list);
558 }
559
560 spin_unlock(&cache_defer_lock);
561
562}
563
564struct thread_deferred_req {
565 struct cache_deferred_req handle;
566 struct completion completion;
567};
568
569static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
570{
571 struct thread_deferred_req *dr =
572 container_of(dreq, struct thread_deferred_req, handle);
573 complete(&dr->completion);
574}
575
576static void cache_wait_req(struct cache_req *req, struct cache_head *item)
577{
578 struct thread_deferred_req sleeper;
579 struct cache_deferred_req *dreq = &sleeper.handle;
580
581 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
582 dreq->revisit = cache_restart_thread;
583
584 setup_deferral(dreq, item, 0);
585
586 if (!test_bit(CACHE_PENDING, &item->flags) ||
587 wait_for_completion_interruptible_timeout(
588 &sleeper.completion, req->thread_wait) <= 0) {
589 /* The completion wasn't completed, so we need
590 * to clean up
591 */
592 spin_lock(&cache_defer_lock);
593 if (!hlist_unhashed(&sleeper.handle.hash)) {
594 __unhash_deferred_req(&sleeper.handle);
595 spin_unlock(&cache_defer_lock);
596 } else {
597 /* cache_revisit_request already removed
598 * this from the hash table, but hasn't
599 * called ->revisit yet. It will very soon
600 * and we need to wait for it.
601 */
602 spin_unlock(&cache_defer_lock);
603 wait_for_completion(&sleeper.completion);
604 }
605 }
606}
607
608static void cache_limit_defers(void)
609{
610 /* Make sure we haven't exceed the limit of allowed deferred
611 * requests.
612 */
613 struct cache_deferred_req *discard = NULL;
614
615 if (cache_defer_cnt <= DFR_MAX)
616 return;
617
618 spin_lock(&cache_defer_lock);
619
620 /* Consider removing either the first or the last */
621 if (cache_defer_cnt > DFR_MAX) {
622 if (prandom_u32() & 1)
623 discard = list_entry(cache_defer_list.next,
624 struct cache_deferred_req, recent);
625 else
626 discard = list_entry(cache_defer_list.prev,
627 struct cache_deferred_req, recent);
628 __unhash_deferred_req(discard);
629 }
630 spin_unlock(&cache_defer_lock);
631 if (discard)
632 discard->revisit(discard, 1);
633}
634
635/* Return true if and only if a deferred request is queued. */
636static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
637{
638 struct cache_deferred_req *dreq;
639
640 if (req->thread_wait) {
641 cache_wait_req(req, item);
642 if (!test_bit(CACHE_PENDING, &item->flags))
643 return false;
644 }
645 dreq = req->defer(req);
646 if (dreq == NULL)
647 return false;
648 setup_deferral(dreq, item, 1);
649 if (!test_bit(CACHE_PENDING, &item->flags))
650 /* Bit could have been cleared before we managed to
651 * set up the deferral, so need to revisit just in case
652 */
653 cache_revisit_request(item);
654
655 cache_limit_defers();
656 return true;
657}
658
659static void cache_revisit_request(struct cache_head *item)
660{
661 struct cache_deferred_req *dreq;
662 struct list_head pending;
663 struct hlist_node *tmp;
664 int hash = DFR_HASH(item);
665
666 INIT_LIST_HEAD(&pending);
667 spin_lock(&cache_defer_lock);
668
669 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
670 if (dreq->item == item) {
671 __unhash_deferred_req(dreq);
672 list_add(&dreq->recent, &pending);
673 }
674
675 spin_unlock(&cache_defer_lock);
676
677 while (!list_empty(&pending)) {
678 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
679 list_del_init(&dreq->recent);
680 dreq->revisit(dreq, 0);
681 }
682}
683
684void cache_clean_deferred(void *owner)
685{
686 struct cache_deferred_req *dreq, *tmp;
687 struct list_head pending;
688
689
690 INIT_LIST_HEAD(&pending);
691 spin_lock(&cache_defer_lock);
692
693 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
694 if (dreq->owner == owner) {
695 __unhash_deferred_req(dreq);
696 list_add(&dreq->recent, &pending);
697 }
698 }
699 spin_unlock(&cache_defer_lock);
700
701 while (!list_empty(&pending)) {
702 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
703 list_del_init(&dreq->recent);
704 dreq->revisit(dreq, 1);
705 }
706}
707
708/*
709 * communicate with user-space
710 *
711 * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
712 * On read, you get a full request, or block.
713 * On write, an update request is processed.
714 * Poll works if anything to read, and always allows write.
715 *
716 * Implemented by linked list of requests. Each open file has
717 * a ->private that also exists in this list. New requests are added
718 * to the end and may wakeup and preceding readers.
719 * New readers are added to the head. If, on read, an item is found with
720 * CACHE_UPCALLING clear, we free it from the list.
721 *
722 */
723
724static DEFINE_SPINLOCK(queue_lock);
725static DEFINE_MUTEX(queue_io_mutex);
726
727struct cache_queue {
728 struct list_head list;
729 int reader; /* if 0, then request */
730};
731struct cache_request {
732 struct cache_queue q;
733 struct cache_head *item;
734 char * buf;
735 int len;
736 int readers;
737};
738struct cache_reader {
739 struct cache_queue q;
740 int offset; /* if non-0, we have a refcnt on next request */
741};
742
743static int cache_request(struct cache_detail *detail,
744 struct cache_request *crq)
745{
746 char *bp = crq->buf;
747 int len = PAGE_SIZE;
748
749 detail->cache_request(detail, crq->item, &bp, &len);
750 if (len < 0)
751 return -EAGAIN;
752 return PAGE_SIZE - len;
753}
754
755static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
756 loff_t *ppos, struct cache_detail *cd)
757{
758 struct cache_reader *rp = filp->private_data;
759 struct cache_request *rq;
760 struct inode *inode = file_inode(filp);
761 int err;
762
763 if (count == 0)
764 return 0;
765
766 mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
767 * readers on this file */
768 again:
769 spin_lock(&queue_lock);
770 /* need to find next request */
771 while (rp->q.list.next != &cd->queue &&
772 list_entry(rp->q.list.next, struct cache_queue, list)
773 ->reader) {
774 struct list_head *next = rp->q.list.next;
775 list_move(&rp->q.list, next);
776 }
777 if (rp->q.list.next == &cd->queue) {
778 spin_unlock(&queue_lock);
779 mutex_unlock(&inode->i_mutex);
780 WARN_ON_ONCE(rp->offset);
781 return 0;
782 }
783 rq = container_of(rp->q.list.next, struct cache_request, q.list);
784 WARN_ON_ONCE(rq->q.reader);
785 if (rp->offset == 0)
786 rq->readers++;
787 spin_unlock(&queue_lock);
788
789 if (rq->len == 0) {
790 err = cache_request(cd, rq);
791 if (err < 0)
792 goto out;
793 rq->len = err;
794 }
795
796 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
797 err = -EAGAIN;
798 spin_lock(&queue_lock);
799 list_move(&rp->q.list, &rq->q.list);
800 spin_unlock(&queue_lock);
801 } else {
802 if (rp->offset + count > rq->len)
803 count = rq->len - rp->offset;
804 err = -EFAULT;
805 if (copy_to_user(buf, rq->buf + rp->offset, count))
806 goto out;
807 rp->offset += count;
808 if (rp->offset >= rq->len) {
809 rp->offset = 0;
810 spin_lock(&queue_lock);
811 list_move(&rp->q.list, &rq->q.list);
812 spin_unlock(&queue_lock);
813 }
814 err = 0;
815 }
816 out:
817 if (rp->offset == 0) {
818 /* need to release rq */
819 spin_lock(&queue_lock);
820 rq->readers--;
821 if (rq->readers == 0 &&
822 !test_bit(CACHE_PENDING, &rq->item->flags)) {
823 list_del(&rq->q.list);
824 spin_unlock(&queue_lock);
825 cache_put(rq->item, cd);
826 kfree(rq->buf);
827 kfree(rq);
828 } else
829 spin_unlock(&queue_lock);
830 }
831 if (err == -EAGAIN)
832 goto again;
833 mutex_unlock(&inode->i_mutex);
834 return err ? err : count;
835}
836
837static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
838 size_t count, struct cache_detail *cd)
839{
840 ssize_t ret;
841
842 if (count == 0)
843 return -EINVAL;
844 if (copy_from_user(kaddr, buf, count))
845 return -EFAULT;
846 kaddr[count] = '\0';
847 ret = cd->cache_parse(cd, kaddr, count);
848 if (!ret)
849 ret = count;
850 return ret;
851}
852
853static ssize_t cache_slow_downcall(const char __user *buf,
854 size_t count, struct cache_detail *cd)
855{
856 static char write_buf[8192]; /* protected by queue_io_mutex */
857 ssize_t ret = -EINVAL;
858
859 if (count >= sizeof(write_buf))
860 goto out;
861 mutex_lock(&queue_io_mutex);
862 ret = cache_do_downcall(write_buf, buf, count, cd);
863 mutex_unlock(&queue_io_mutex);
864out:
865 return ret;
866}
867
868static ssize_t cache_downcall(struct address_space *mapping,
869 const char __user *buf,
870 size_t count, struct cache_detail *cd)
871{
872 struct page *page;
873 char *kaddr;
874 ssize_t ret = -ENOMEM;
875
876 if (count >= PAGE_CACHE_SIZE)
877 goto out_slow;
878
879 page = find_or_create_page(mapping, 0, GFP_KERNEL);
880 if (!page)
881 goto out_slow;
882
883 kaddr = kmap(page);
884 ret = cache_do_downcall(kaddr, buf, count, cd);
885 kunmap(page);
886 unlock_page(page);
887 page_cache_release(page);
888 return ret;
889out_slow:
890 return cache_slow_downcall(buf, count, cd);
891}
892
893static ssize_t cache_write(struct file *filp, const char __user *buf,
894 size_t count, loff_t *ppos,
895 struct cache_detail *cd)
896{
897 struct address_space *mapping = filp->f_mapping;
898 struct inode *inode = file_inode(filp);
899 ssize_t ret = -EINVAL;
900
901 if (!cd->cache_parse)
902 goto out;
903
904 mutex_lock(&inode->i_mutex);
905 ret = cache_downcall(mapping, buf, count, cd);
906 mutex_unlock(&inode->i_mutex);
907out:
908 return ret;
909}
910
911static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
912
913static unsigned int cache_poll(struct file *filp, poll_table *wait,
914 struct cache_detail *cd)
915{
916 unsigned int mask;
917 struct cache_reader *rp = filp->private_data;
918 struct cache_queue *cq;
919
920 poll_wait(filp, &queue_wait, wait);
921
922 /* alway allow write */
923 mask = POLL_OUT | POLLWRNORM;
924
925 if (!rp)
926 return mask;
927
928 spin_lock(&queue_lock);
929
930 for (cq= &rp->q; &cq->list != &cd->queue;
931 cq = list_entry(cq->list.next, struct cache_queue, list))
932 if (!cq->reader) {
933 mask |= POLLIN | POLLRDNORM;
934 break;
935 }
936 spin_unlock(&queue_lock);
937 return mask;
938}
939
940static int cache_ioctl(struct inode *ino, struct file *filp,
941 unsigned int cmd, unsigned long arg,
942 struct cache_detail *cd)
943{
944 int len = 0;
945 struct cache_reader *rp = filp->private_data;
946 struct cache_queue *cq;
947
948 if (cmd != FIONREAD || !rp)
949 return -EINVAL;
950
951 spin_lock(&queue_lock);
952
953 /* only find the length remaining in current request,
954 * or the length of the next request
955 */
956 for (cq= &rp->q; &cq->list != &cd->queue;
957 cq = list_entry(cq->list.next, struct cache_queue, list))
958 if (!cq->reader) {
959 struct cache_request *cr =
960 container_of(cq, struct cache_request, q);
961 len = cr->len - rp->offset;
962 break;
963 }
964 spin_unlock(&queue_lock);
965
966 return put_user(len, (int __user *)arg);
967}
968
969static int cache_open(struct inode *inode, struct file *filp,
970 struct cache_detail *cd)
971{
972 struct cache_reader *rp = NULL;
973
974 if (!cd || !try_module_get(cd->owner))
975 return -EACCES;
976 nonseekable_open(inode, filp);
977 if (filp->f_mode & FMODE_READ) {
978 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
979 if (!rp) {
980 module_put(cd->owner);
981 return -ENOMEM;
982 }
983 rp->offset = 0;
984 rp->q.reader = 1;
985 atomic_inc(&cd->readers);
986 spin_lock(&queue_lock);
987 list_add(&rp->q.list, &cd->queue);
988 spin_unlock(&queue_lock);
989 }
990 filp->private_data = rp;
991 return 0;
992}
993
994static int cache_release(struct inode *inode, struct file *filp,
995 struct cache_detail *cd)
996{
997 struct cache_reader *rp = filp->private_data;
998
999 if (rp) {
1000 spin_lock(&queue_lock);
1001 if (rp->offset) {
1002 struct cache_queue *cq;
1003 for (cq= &rp->q; &cq->list != &cd->queue;
1004 cq = list_entry(cq->list.next, struct cache_queue, list))
1005 if (!cq->reader) {
1006 container_of(cq, struct cache_request, q)
1007 ->readers--;
1008 break;
1009 }
1010 rp->offset = 0;
1011 }
1012 list_del(&rp->q.list);
1013 spin_unlock(&queue_lock);
1014
1015 filp->private_data = NULL;
1016 kfree(rp);
1017
1018 cd->last_close = seconds_since_boot();
1019 atomic_dec(&cd->readers);
1020 }
1021 module_put(cd->owner);
1022 return 0;
1023}
1024
1025
1026
1027static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1028{
1029 struct cache_queue *cq, *tmp;
1030 struct cache_request *cr;
1031 struct list_head dequeued;
1032
1033 INIT_LIST_HEAD(&dequeued);
1034 spin_lock(&queue_lock);
1035 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1036 if (!cq->reader) {
1037 cr = container_of(cq, struct cache_request, q);
1038 if (cr->item != ch)
1039 continue;
1040 if (test_bit(CACHE_PENDING, &ch->flags))
1041 /* Lost a race and it is pending again */
1042 break;
1043 if (cr->readers != 0)
1044 continue;
1045 list_move(&cr->q.list, &dequeued);
1046 }
1047 spin_unlock(&queue_lock);
1048 while (!list_empty(&dequeued)) {
1049 cr = list_entry(dequeued.next, struct cache_request, q.list);
1050 list_del(&cr->q.list);
1051 cache_put(cr->item, detail);
1052 kfree(cr->buf);
1053 kfree(cr);
1054 }
1055}
1056
1057/*
1058 * Support routines for text-based upcalls.
1059 * Fields are separated by spaces.
1060 * Fields are either mangled to quote space tab newline slosh with slosh
1061 * or a hexified with a leading \x
1062 * Record is terminated with newline.
1063 *
1064 */
1065
1066void qword_add(char **bpp, int *lp, char *str)
1067{
1068 char *bp = *bpp;
1069 int len = *lp;
1070 char c;
1071
1072 if (len < 0) return;
1073
1074 while ((c=*str++) && len)
1075 switch(c) {
1076 case ' ':
1077 case '\t':
1078 case '\n':
1079 case '\\':
1080 if (len >= 4) {
1081 *bp++ = '\\';
1082 *bp++ = '0' + ((c & 0300)>>6);
1083 *bp++ = '0' + ((c & 0070)>>3);
1084 *bp++ = '0' + ((c & 0007)>>0);
1085 }
1086 len -= 4;
1087 break;
1088 default:
1089 *bp++ = c;
1090 len--;
1091 }
1092 if (c || len <1) len = -1;
1093 else {
1094 *bp++ = ' ';
1095 len--;
1096 }
1097 *bpp = bp;
1098 *lp = len;
1099}
1100EXPORT_SYMBOL_GPL(qword_add);
1101
1102void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1103{
1104 char *bp = *bpp;
1105 int len = *lp;
1106
1107 if (len < 0) return;
1108
1109 if (len > 2) {
1110 *bp++ = '\\';
1111 *bp++ = 'x';
1112 len -= 2;
1113 while (blen && len >= 2) {
1114 bp = hex_byte_pack(bp, *buf++);
1115 len -= 2;
1116 blen--;
1117 }
1118 }
1119 if (blen || len<1) len = -1;
1120 else {
1121 *bp++ = ' ';
1122 len--;
1123 }
1124 *bpp = bp;
1125 *lp = len;
1126}
1127EXPORT_SYMBOL_GPL(qword_addhex);
1128
1129static void warn_no_listener(struct cache_detail *detail)
1130{
1131 if (detail->last_warn != detail->last_close) {
1132 detail->last_warn = detail->last_close;
1133 if (detail->warn_no_listener)
1134 detail->warn_no_listener(detail, detail->last_close != 0);
1135 }
1136}
1137
1138static bool cache_listeners_exist(struct cache_detail *detail)
1139{
1140 if (atomic_read(&detail->readers))
1141 return true;
1142 if (detail->last_close == 0)
1143 /* This cache was never opened */
1144 return false;
1145 if (detail->last_close < seconds_since_boot() - 30)
1146 /*
1147 * We allow for the possibility that someone might
1148 * restart a userspace daemon without restarting the
1149 * server; but after 30 seconds, we give up.
1150 */
1151 return false;
1152 return true;
1153}
1154
1155/*
1156 * register an upcall request to user-space and queue it up for read() by the
1157 * upcall daemon.
1158 *
1159 * Each request is at most one page long.
1160 */
1161int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1162{
1163
1164 char *buf;
1165 struct cache_request *crq;
1166 int ret = 0;
1167
1168 if (!detail->cache_request)
1169 return -EINVAL;
1170
1171 if (!cache_listeners_exist(detail)) {
1172 warn_no_listener(detail);
1173 return -EINVAL;
1174 }
1175 if (test_bit(CACHE_CLEANED, &h->flags))
1176 /* Too late to make an upcall */
1177 return -EAGAIN;
1178
1179 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1180 if (!buf)
1181 return -EAGAIN;
1182
1183 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1184 if (!crq) {
1185 kfree(buf);
1186 return -EAGAIN;
1187 }
1188
1189 crq->q.reader = 0;
1190 crq->item = cache_get(h);
1191 crq->buf = buf;
1192 crq->len = 0;
1193 crq->readers = 0;
1194 spin_lock(&queue_lock);
1195 if (test_bit(CACHE_PENDING, &h->flags))
1196 list_add_tail(&crq->q.list, &detail->queue);
1197 else
1198 /* Lost a race, no longer PENDING, so don't enqueue */
1199 ret = -EAGAIN;
1200 spin_unlock(&queue_lock);
1201 wake_up(&queue_wait);
1202 if (ret == -EAGAIN) {
1203 kfree(buf);
1204 kfree(crq);
1205 }
1206 return ret;
1207}
1208EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1209
1210/*
1211 * parse a message from user-space and pass it
1212 * to an appropriate cache
1213 * Messages are, like requests, separated into fields by
1214 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1215 *
1216 * Message is
1217 * reply cachename expiry key ... content....
1218 *
1219 * key and content are both parsed by cache
1220 */
1221
1222int qword_get(char **bpp, char *dest, int bufsize)
1223{
1224 /* return bytes copied, or -1 on error */
1225 char *bp = *bpp;
1226 int len = 0;
1227
1228 while (*bp == ' ') bp++;
1229
1230 if (bp[0] == '\\' && bp[1] == 'x') {
1231 /* HEX STRING */
1232 bp += 2;
1233 while (len < bufsize) {
1234 int h, l;
1235
1236 h = hex_to_bin(bp[0]);
1237 if (h < 0)
1238 break;
1239
1240 l = hex_to_bin(bp[1]);
1241 if (l < 0)
1242 break;
1243
1244 *dest++ = (h << 4) | l;
1245 bp += 2;
1246 len++;
1247 }
1248 } else {
1249 /* text with \nnn octal quoting */
1250 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1251 if (*bp == '\\' &&
1252 isodigit(bp[1]) && (bp[1] <= '3') &&
1253 isodigit(bp[2]) &&
1254 isodigit(bp[3])) {
1255 int byte = (*++bp -'0');
1256 bp++;
1257 byte = (byte << 3) | (*bp++ - '0');
1258 byte = (byte << 3) | (*bp++ - '0');
1259 *dest++ = byte;
1260 len++;
1261 } else {
1262 *dest++ = *bp++;
1263 len++;
1264 }
1265 }
1266 }
1267
1268 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1269 return -1;
1270 while (*bp == ' ') bp++;
1271 *bpp = bp;
1272 *dest = '\0';
1273 return len;
1274}
1275EXPORT_SYMBOL_GPL(qword_get);
1276
1277
1278/*
1279 * support /proc/sunrpc/cache/$CACHENAME/content
1280 * as a seqfile.
1281 * We call ->cache_show passing NULL for the item to
1282 * get a header, then pass each real item in the cache
1283 */
1284
1285struct handle {
1286 struct cache_detail *cd;
1287};
1288
1289static void *c_start(struct seq_file *m, loff_t *pos)
1290 __acquires(cd->hash_lock)
1291{
1292 loff_t n = *pos;
1293 unsigned int hash, entry;
1294 struct cache_head *ch;
1295 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1296
1297
1298 read_lock(&cd->hash_lock);
1299 if (!n--)
1300 return SEQ_START_TOKEN;
1301 hash = n >> 32;
1302 entry = n & ((1LL<<32) - 1);
1303
1304 for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1305 if (!entry--)
1306 return ch;
1307 n &= ~((1LL<<32) - 1);
1308 do {
1309 hash++;
1310 n += 1LL<<32;
1311 } while(hash < cd->hash_size &&
1312 cd->hash_table[hash]==NULL);
1313 if (hash >= cd->hash_size)
1314 return NULL;
1315 *pos = n+1;
1316 return cd->hash_table[hash];
1317}
1318
1319static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1320{
1321 struct cache_head *ch = p;
1322 int hash = (*pos >> 32);
1323 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1324
1325 if (p == SEQ_START_TOKEN)
1326 hash = 0;
1327 else if (ch->next == NULL) {
1328 hash++;
1329 *pos += 1LL<<32;
1330 } else {
1331 ++*pos;
1332 return ch->next;
1333 }
1334 *pos &= ~((1LL<<32) - 1);
1335 while (hash < cd->hash_size &&
1336 cd->hash_table[hash] == NULL) {
1337 hash++;
1338 *pos += 1LL<<32;
1339 }
1340 if (hash >= cd->hash_size)
1341 return NULL;
1342 ++*pos;
1343 return cd->hash_table[hash];
1344}
1345
1346static void c_stop(struct seq_file *m, void *p)
1347 __releases(cd->hash_lock)
1348{
1349 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1350 read_unlock(&cd->hash_lock);
1351}
1352
1353static int c_show(struct seq_file *m, void *p)
1354{
1355 struct cache_head *cp = p;
1356 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1357
1358 if (p == SEQ_START_TOKEN)
1359 return cd->cache_show(m, cd, NULL);
1360
1361 ifdebug(CACHE)
1362 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1363 convert_to_wallclock(cp->expiry_time),
1364 atomic_read(&cp->ref.refcount), cp->flags);
1365 cache_get(cp);
1366 if (cache_check(cd, cp, NULL))
1367 /* cache_check does a cache_put on failure */
1368 seq_printf(m, "# ");
1369 else {
1370 if (cache_is_expired(cd, cp))
1371 seq_printf(m, "# ");
1372 cache_put(cp, cd);
1373 }
1374
1375 return cd->cache_show(m, cd, cp);
1376}
1377
1378static const struct seq_operations cache_content_op = {
1379 .start = c_start,
1380 .next = c_next,
1381 .stop = c_stop,
1382 .show = c_show,
1383};
1384
1385static int content_open(struct inode *inode, struct file *file,
1386 struct cache_detail *cd)
1387{
1388 struct handle *han;
1389
1390 if (!cd || !try_module_get(cd->owner))
1391 return -EACCES;
1392 han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1393 if (han == NULL) {
1394 module_put(cd->owner);
1395 return -ENOMEM;
1396 }
1397
1398 han->cd = cd;
1399 return 0;
1400}
1401
1402static int content_release(struct inode *inode, struct file *file,
1403 struct cache_detail *cd)
1404{
1405 int ret = seq_release_private(inode, file);
1406 module_put(cd->owner);
1407 return ret;
1408}
1409
1410static int open_flush(struct inode *inode, struct file *file,
1411 struct cache_detail *cd)
1412{
1413 if (!cd || !try_module_get(cd->owner))
1414 return -EACCES;
1415 return nonseekable_open(inode, file);
1416}
1417
1418static int release_flush(struct inode *inode, struct file *file,
1419 struct cache_detail *cd)
1420{
1421 module_put(cd->owner);
1422 return 0;
1423}
1424
1425static ssize_t read_flush(struct file *file, char __user *buf,
1426 size_t count, loff_t *ppos,
1427 struct cache_detail *cd)
1428{
1429 char tbuf[22];
1430 unsigned long p = *ppos;
1431 size_t len;
1432
1433 snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1434 len = strlen(tbuf);
1435 if (p >= len)
1436 return 0;
1437 len -= p;
1438 if (len > count)
1439 len = count;
1440 if (copy_to_user(buf, (void*)(tbuf+p), len))
1441 return -EFAULT;
1442 *ppos += len;
1443 return len;
1444}
1445
1446static ssize_t write_flush(struct file *file, const char __user *buf,
1447 size_t count, loff_t *ppos,
1448 struct cache_detail *cd)
1449{
1450 char tbuf[20];
1451 char *bp, *ep;
1452
1453 if (*ppos || count > sizeof(tbuf)-1)
1454 return -EINVAL;
1455 if (copy_from_user(tbuf, buf, count))
1456 return -EFAULT;
1457 tbuf[count] = 0;
1458 simple_strtoul(tbuf, &ep, 0);
1459 if (*ep && *ep != '\n')
1460 return -EINVAL;
1461
1462 bp = tbuf;
1463 cd->flush_time = get_expiry(&bp);
1464 cd->nextcheck = seconds_since_boot();
1465 cache_flush();
1466
1467 *ppos += count;
1468 return count;
1469}
1470
1471static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1472 size_t count, loff_t *ppos)
1473{
1474 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1475
1476 return cache_read(filp, buf, count, ppos, cd);
1477}
1478
1479static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1480 size_t count, loff_t *ppos)
1481{
1482 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1483
1484 return cache_write(filp, buf, count, ppos, cd);
1485}
1486
1487static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1488{
1489 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1490
1491 return cache_poll(filp, wait, cd);
1492}
1493
1494static long cache_ioctl_procfs(struct file *filp,
1495 unsigned int cmd, unsigned long arg)
1496{
1497 struct inode *inode = file_inode(filp);
1498 struct cache_detail *cd = PDE_DATA(inode);
1499
1500 return cache_ioctl(inode, filp, cmd, arg, cd);
1501}
1502
1503static int cache_open_procfs(struct inode *inode, struct file *filp)
1504{
1505 struct cache_detail *cd = PDE_DATA(inode);
1506
1507 return cache_open(inode, filp, cd);
1508}
1509
1510static int cache_release_procfs(struct inode *inode, struct file *filp)
1511{
1512 struct cache_detail *cd = PDE_DATA(inode);
1513
1514 return cache_release(inode, filp, cd);
1515}
1516
1517static const struct file_operations cache_file_operations_procfs = {
1518 .owner = THIS_MODULE,
1519 .llseek = no_llseek,
1520 .read = cache_read_procfs,
1521 .write = cache_write_procfs,
1522 .poll = cache_poll_procfs,
1523 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1524 .open = cache_open_procfs,
1525 .release = cache_release_procfs,
1526};
1527
1528static int content_open_procfs(struct inode *inode, struct file *filp)
1529{
1530 struct cache_detail *cd = PDE_DATA(inode);
1531
1532 return content_open(inode, filp, cd);
1533}
1534
1535static int content_release_procfs(struct inode *inode, struct file *filp)
1536{
1537 struct cache_detail *cd = PDE_DATA(inode);
1538
1539 return content_release(inode, filp, cd);
1540}
1541
1542static const struct file_operations content_file_operations_procfs = {
1543 .open = content_open_procfs,
1544 .read = seq_read,
1545 .llseek = seq_lseek,
1546 .release = content_release_procfs,
1547};
1548
1549static int open_flush_procfs(struct inode *inode, struct file *filp)
1550{
1551 struct cache_detail *cd = PDE_DATA(inode);
1552
1553 return open_flush(inode, filp, cd);
1554}
1555
1556static int release_flush_procfs(struct inode *inode, struct file *filp)
1557{
1558 struct cache_detail *cd = PDE_DATA(inode);
1559
1560 return release_flush(inode, filp, cd);
1561}
1562
1563static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1564 size_t count, loff_t *ppos)
1565{
1566 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1567
1568 return read_flush(filp, buf, count, ppos, cd);
1569}
1570
1571static ssize_t write_flush_procfs(struct file *filp,
1572 const char __user *buf,
1573 size_t count, loff_t *ppos)
1574{
1575 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1576
1577 return write_flush(filp, buf, count, ppos, cd);
1578}
1579
1580static const struct file_operations cache_flush_operations_procfs = {
1581 .open = open_flush_procfs,
1582 .read = read_flush_procfs,
1583 .write = write_flush_procfs,
1584 .release = release_flush_procfs,
1585 .llseek = no_llseek,
1586};
1587
1588static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1589{
1590 struct sunrpc_net *sn;
1591
1592 if (cd->u.procfs.proc_ent == NULL)
1593 return;
1594 if (cd->u.procfs.flush_ent)
1595 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1596 if (cd->u.procfs.channel_ent)
1597 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1598 if (cd->u.procfs.content_ent)
1599 remove_proc_entry("content", cd->u.procfs.proc_ent);
1600 cd->u.procfs.proc_ent = NULL;
1601 sn = net_generic(net, sunrpc_net_id);
1602 remove_proc_entry(cd->name, sn->proc_net_rpc);
1603}
1604
1605#ifdef CONFIG_PROC_FS
1606static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1607{
1608 struct proc_dir_entry *p;
1609 struct sunrpc_net *sn;
1610
1611 sn = net_generic(net, sunrpc_net_id);
1612 cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1613 if (cd->u.procfs.proc_ent == NULL)
1614 goto out_nomem;
1615 cd->u.procfs.channel_ent = NULL;
1616 cd->u.procfs.content_ent = NULL;
1617
1618 p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1619 cd->u.procfs.proc_ent,
1620 &cache_flush_operations_procfs, cd);
1621 cd->u.procfs.flush_ent = p;
1622 if (p == NULL)
1623 goto out_nomem;
1624
1625 if (cd->cache_request || cd->cache_parse) {
1626 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1627 cd->u.procfs.proc_ent,
1628 &cache_file_operations_procfs, cd);
1629 cd->u.procfs.channel_ent = p;
1630 if (p == NULL)
1631 goto out_nomem;
1632 }
1633 if (cd->cache_show) {
1634 p = proc_create_data("content", S_IFREG|S_IRUSR,
1635 cd->u.procfs.proc_ent,
1636 &content_file_operations_procfs, cd);
1637 cd->u.procfs.content_ent = p;
1638 if (p == NULL)
1639 goto out_nomem;
1640 }
1641 return 0;
1642out_nomem:
1643 remove_cache_proc_entries(cd, net);
1644 return -ENOMEM;
1645}
1646#else /* CONFIG_PROC_FS */
1647static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1648{
1649 return 0;
1650}
1651#endif
1652
1653void __init cache_initialize(void)
1654{
1655 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1656}
1657
1658int cache_register_net(struct cache_detail *cd, struct net *net)
1659{
1660 int ret;
1661
1662 sunrpc_init_cache_detail(cd);
1663 ret = create_cache_proc_entries(cd, net);
1664 if (ret)
1665 sunrpc_destroy_cache_detail(cd);
1666 return ret;
1667}
1668EXPORT_SYMBOL_GPL(cache_register_net);
1669
1670void cache_unregister_net(struct cache_detail *cd, struct net *net)
1671{
1672 remove_cache_proc_entries(cd, net);
1673 sunrpc_destroy_cache_detail(cd);
1674}
1675EXPORT_SYMBOL_GPL(cache_unregister_net);
1676
1677struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1678{
1679 struct cache_detail *cd;
1680
1681 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1682 if (cd == NULL)
1683 return ERR_PTR(-ENOMEM);
1684
1685 cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1686 GFP_KERNEL);
1687 if (cd->hash_table == NULL) {
1688 kfree(cd);
1689 return ERR_PTR(-ENOMEM);
1690 }
1691 cd->net = net;
1692 return cd;
1693}
1694EXPORT_SYMBOL_GPL(cache_create_net);
1695
1696void cache_destroy_net(struct cache_detail *cd, struct net *net)
1697{
1698 kfree(cd->hash_table);
1699 kfree(cd);
1700}
1701EXPORT_SYMBOL_GPL(cache_destroy_net);
1702
1703static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1704 size_t count, loff_t *ppos)
1705{
1706 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1707
1708 return cache_read(filp, buf, count, ppos, cd);
1709}
1710
1711static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1712 size_t count, loff_t *ppos)
1713{
1714 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1715
1716 return cache_write(filp, buf, count, ppos, cd);
1717}
1718
1719static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1720{
1721 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1722
1723 return cache_poll(filp, wait, cd);
1724}
1725
1726static long cache_ioctl_pipefs(struct file *filp,
1727 unsigned int cmd, unsigned long arg)
1728{
1729 struct inode *inode = file_inode(filp);
1730 struct cache_detail *cd = RPC_I(inode)->private;
1731
1732 return cache_ioctl(inode, filp, cmd, arg, cd);
1733}
1734
1735static int cache_open_pipefs(struct inode *inode, struct file *filp)
1736{
1737 struct cache_detail *cd = RPC_I(inode)->private;
1738
1739 return cache_open(inode, filp, cd);
1740}
1741
1742static int cache_release_pipefs(struct inode *inode, struct file *filp)
1743{
1744 struct cache_detail *cd = RPC_I(inode)->private;
1745
1746 return cache_release(inode, filp, cd);
1747}
1748
1749const struct file_operations cache_file_operations_pipefs = {
1750 .owner = THIS_MODULE,
1751 .llseek = no_llseek,
1752 .read = cache_read_pipefs,
1753 .write = cache_write_pipefs,
1754 .poll = cache_poll_pipefs,
1755 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1756 .open = cache_open_pipefs,
1757 .release = cache_release_pipefs,
1758};
1759
1760static int content_open_pipefs(struct inode *inode, struct file *filp)
1761{
1762 struct cache_detail *cd = RPC_I(inode)->private;
1763
1764 return content_open(inode, filp, cd);
1765}
1766
1767static int content_release_pipefs(struct inode *inode, struct file *filp)
1768{
1769 struct cache_detail *cd = RPC_I(inode)->private;
1770
1771 return content_release(inode, filp, cd);
1772}
1773
1774const struct file_operations content_file_operations_pipefs = {
1775 .open = content_open_pipefs,
1776 .read = seq_read,
1777 .llseek = seq_lseek,
1778 .release = content_release_pipefs,
1779};
1780
1781static int open_flush_pipefs(struct inode *inode, struct file *filp)
1782{
1783 struct cache_detail *cd = RPC_I(inode)->private;
1784
1785 return open_flush(inode, filp, cd);
1786}
1787
1788static int release_flush_pipefs(struct inode *inode, struct file *filp)
1789{
1790 struct cache_detail *cd = RPC_I(inode)->private;
1791
1792 return release_flush(inode, filp, cd);
1793}
1794
1795static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1796 size_t count, loff_t *ppos)
1797{
1798 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1799
1800 return read_flush(filp, buf, count, ppos, cd);
1801}
1802
1803static ssize_t write_flush_pipefs(struct file *filp,
1804 const char __user *buf,
1805 size_t count, loff_t *ppos)
1806{
1807 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1808
1809 return write_flush(filp, buf, count, ppos, cd);
1810}
1811
1812const struct file_operations cache_flush_operations_pipefs = {
1813 .open = open_flush_pipefs,
1814 .read = read_flush_pipefs,
1815 .write = write_flush_pipefs,
1816 .release = release_flush_pipefs,
1817 .llseek = no_llseek,
1818};
1819
1820int sunrpc_cache_register_pipefs(struct dentry *parent,
1821 const char *name, umode_t umode,
1822 struct cache_detail *cd)
1823{
1824 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1825 if (IS_ERR(dir))
1826 return PTR_ERR(dir);
1827 cd->u.pipefs.dir = dir;
1828 return 0;
1829}
1830EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1831
1832void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1833{
1834 rpc_remove_cache_dir(cd->u.pipefs.dir);
1835 cd->u.pipefs.dir = NULL;
1836}
1837EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1838