Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/sunrpc/cache.c
4 *
5 * Generic code for various authentication-related caches
6 * used by sunrpc clients and servers.
7 *
8 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
9 */
10
11#include <linux/types.h>
12#include <linux/fs.h>
13#include <linux/file.h>
14#include <linux/slab.h>
15#include <linux/signal.h>
16#include <linux/sched.h>
17#include <linux/kmod.h>
18#include <linux/list.h>
19#include <linux/module.h>
20#include <linux/ctype.h>
21#include <linux/string_helpers.h>
22#include <linux/uaccess.h>
23#include <linux/poll.h>
24#include <linux/seq_file.h>
25#include <linux/proc_fs.h>
26#include <linux/net.h>
27#include <linux/workqueue.h>
28#include <linux/mutex.h>
29#include <linux/pagemap.h>
30#include <asm/ioctls.h>
31#include <linux/sunrpc/types.h>
32#include <linux/sunrpc/cache.h>
33#include <linux/sunrpc/stats.h>
34#include <linux/sunrpc/rpc_pipe_fs.h>
35#include <trace/events/sunrpc.h>
36#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, struct cache_detail *detail)
44{
45 time64_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 void cache_fresh_unlocked(struct cache_head *head,
57 struct cache_detail *detail);
58
59static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
60 struct cache_head *key,
61 int hash)
62{
63 struct hlist_head *head = &detail->hash_table[hash];
64 struct cache_head *tmp;
65
66 rcu_read_lock();
67 hlist_for_each_entry_rcu(tmp, head, cache_list) {
68 if (!detail->match(tmp, key))
69 continue;
70 if (test_bit(CACHE_VALID, &tmp->flags) &&
71 cache_is_expired(detail, tmp))
72 continue;
73 tmp = cache_get_rcu(tmp);
74 rcu_read_unlock();
75 return tmp;
76 }
77 rcu_read_unlock();
78 return NULL;
79}
80
81static void sunrpc_begin_cache_remove_entry(struct cache_head *ch,
82 struct cache_detail *cd)
83{
84 /* Must be called under cd->hash_lock */
85 hlist_del_init_rcu(&ch->cache_list);
86 set_bit(CACHE_CLEANED, &ch->flags);
87 cd->entries --;
88}
89
90static void sunrpc_end_cache_remove_entry(struct cache_head *ch,
91 struct cache_detail *cd)
92{
93 cache_fresh_unlocked(ch, cd);
94 cache_put(ch, cd);
95}
96
97static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
98 struct cache_head *key,
99 int hash)
100{
101 struct cache_head *new, *tmp, *freeme = NULL;
102 struct hlist_head *head = &detail->hash_table[hash];
103
104 new = detail->alloc();
105 if (!new)
106 return NULL;
107 /* must fully initialise 'new', else
108 * we might get lose if we need to
109 * cache_put it soon.
110 */
111 cache_init(new, detail);
112 detail->init(new, key);
113
114 spin_lock(&detail->hash_lock);
115
116 /* check if entry appeared while we slept */
117 hlist_for_each_entry_rcu(tmp, head, cache_list,
118 lockdep_is_held(&detail->hash_lock)) {
119 if (!detail->match(tmp, key))
120 continue;
121 if (test_bit(CACHE_VALID, &tmp->flags) &&
122 cache_is_expired(detail, tmp)) {
123 sunrpc_begin_cache_remove_entry(tmp, detail);
124 trace_cache_entry_expired(detail, tmp);
125 freeme = tmp;
126 break;
127 }
128 cache_get(tmp);
129 spin_unlock(&detail->hash_lock);
130 cache_put(new, detail);
131 return tmp;
132 }
133
134 hlist_add_head_rcu(&new->cache_list, head);
135 detail->entries++;
136 cache_get(new);
137 spin_unlock(&detail->hash_lock);
138
139 if (freeme)
140 sunrpc_end_cache_remove_entry(freeme, detail);
141 return new;
142}
143
144struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
145 struct cache_head *key, int hash)
146{
147 struct cache_head *ret;
148
149 ret = sunrpc_cache_find_rcu(detail, key, hash);
150 if (ret)
151 return ret;
152 /* Didn't find anything, insert an empty entry */
153 return sunrpc_cache_add_entry(detail, key, hash);
154}
155EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
156
157static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
158
159static void cache_fresh_locked(struct cache_head *head, time64_t expiry,
160 struct cache_detail *detail)
161{
162 time64_t now = seconds_since_boot();
163 if (now <= detail->flush_time)
164 /* ensure it isn't immediately treated as expired */
165 now = detail->flush_time + 1;
166 head->expiry_time = expiry;
167 head->last_refresh = now;
168 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
169 set_bit(CACHE_VALID, &head->flags);
170}
171
172static void cache_fresh_unlocked(struct cache_head *head,
173 struct cache_detail *detail)
174{
175 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
176 cache_revisit_request(head);
177 cache_dequeue(detail, head);
178 }
179}
180
181static void cache_make_negative(struct cache_detail *detail,
182 struct cache_head *h)
183{
184 set_bit(CACHE_NEGATIVE, &h->flags);
185 trace_cache_entry_make_negative(detail, h);
186}
187
188static void cache_entry_update(struct cache_detail *detail,
189 struct cache_head *h,
190 struct cache_head *new)
191{
192 if (!test_bit(CACHE_NEGATIVE, &new->flags)) {
193 detail->update(h, new);
194 trace_cache_entry_update(detail, h);
195 } else {
196 cache_make_negative(detail, h);
197 }
198}
199
200struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
201 struct cache_head *new, struct cache_head *old, int hash)
202{
203 /* The 'old' entry is to be replaced by 'new'.
204 * If 'old' is not VALID, we update it directly,
205 * otherwise we need to replace it
206 */
207 struct cache_head *tmp;
208
209 if (!test_bit(CACHE_VALID, &old->flags)) {
210 spin_lock(&detail->hash_lock);
211 if (!test_bit(CACHE_VALID, &old->flags)) {
212 cache_entry_update(detail, old, new);
213 cache_fresh_locked(old, new->expiry_time, detail);
214 spin_unlock(&detail->hash_lock);
215 cache_fresh_unlocked(old, detail);
216 return old;
217 }
218 spin_unlock(&detail->hash_lock);
219 }
220 /* We need to insert a new entry */
221 tmp = detail->alloc();
222 if (!tmp) {
223 cache_put(old, detail);
224 return NULL;
225 }
226 cache_init(tmp, detail);
227 detail->init(tmp, old);
228
229 spin_lock(&detail->hash_lock);
230 cache_entry_update(detail, tmp, new);
231 hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
232 detail->entries++;
233 cache_get(tmp);
234 cache_fresh_locked(tmp, new->expiry_time, detail);
235 cache_fresh_locked(old, 0, detail);
236 spin_unlock(&detail->hash_lock);
237 cache_fresh_unlocked(tmp, detail);
238 cache_fresh_unlocked(old, detail);
239 cache_put(old, detail);
240 return tmp;
241}
242EXPORT_SYMBOL_GPL(sunrpc_cache_update);
243
244static inline int cache_is_valid(struct cache_head *h)
245{
246 if (!test_bit(CACHE_VALID, &h->flags))
247 return -EAGAIN;
248 else {
249 /* entry is valid */
250 if (test_bit(CACHE_NEGATIVE, &h->flags))
251 return -ENOENT;
252 else {
253 /*
254 * In combination with write barrier in
255 * sunrpc_cache_update, ensures that anyone
256 * using the cache entry after this sees the
257 * updated contents:
258 */
259 smp_rmb();
260 return 0;
261 }
262 }
263}
264
265static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
266{
267 int rv;
268
269 spin_lock(&detail->hash_lock);
270 rv = cache_is_valid(h);
271 if (rv == -EAGAIN) {
272 cache_make_negative(detail, h);
273 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
274 detail);
275 rv = -ENOENT;
276 }
277 spin_unlock(&detail->hash_lock);
278 cache_fresh_unlocked(h, detail);
279 return rv;
280}
281
282/*
283 * This is the generic cache management routine for all
284 * the authentication caches.
285 * It checks the currency of a cache item and will (later)
286 * initiate an upcall to fill it if needed.
287 *
288 *
289 * Returns 0 if the cache_head can be used, or cache_puts it and returns
290 * -EAGAIN if upcall is pending and request has been queued
291 * -ETIMEDOUT if upcall failed or request could not be queue or
292 * upcall completed but item is still invalid (implying that
293 * the cache item has been replaced with a newer one).
294 * -ENOENT if cache entry was negative
295 */
296int cache_check(struct cache_detail *detail,
297 struct cache_head *h, struct cache_req *rqstp)
298{
299 int rv;
300 time64_t refresh_age, age;
301
302 /* First decide return status as best we can */
303 rv = cache_is_valid(h);
304
305 /* now see if we want to start an upcall */
306 refresh_age = (h->expiry_time - h->last_refresh);
307 age = seconds_since_boot() - h->last_refresh;
308
309 if (rqstp == NULL) {
310 if (rv == -EAGAIN)
311 rv = -ENOENT;
312 } else if (rv == -EAGAIN ||
313 (h->expiry_time != 0 && age > refresh_age/2)) {
314 dprintk("RPC: Want update, refage=%lld, age=%lld\n",
315 refresh_age, age);
316 switch (detail->cache_upcall(detail, h)) {
317 case -EINVAL:
318 rv = try_to_negate_entry(detail, h);
319 break;
320 case -EAGAIN:
321 cache_fresh_unlocked(h, detail);
322 break;
323 }
324 }
325
326 if (rv == -EAGAIN) {
327 if (!cache_defer_req(rqstp, h)) {
328 /*
329 * Request was not deferred; handle it as best
330 * we can ourselves:
331 */
332 rv = cache_is_valid(h);
333 if (rv == -EAGAIN)
334 rv = -ETIMEDOUT;
335 }
336 }
337 if (rv)
338 cache_put(h, detail);
339 return rv;
340}
341EXPORT_SYMBOL_GPL(cache_check);
342
343/*
344 * caches need to be periodically cleaned.
345 * For this we maintain a list of cache_detail and
346 * a current pointer into that list and into the table
347 * for that entry.
348 *
349 * Each time cache_clean is called it finds the next non-empty entry
350 * in the current table and walks the list in that entry
351 * looking for entries that can be removed.
352 *
353 * An entry gets removed if:
354 * - The expiry is before current time
355 * - The last_refresh time is before the flush_time for that cache
356 *
357 * later we might drop old entries with non-NEVER expiry if that table
358 * is getting 'full' for some definition of 'full'
359 *
360 * The question of "how often to scan a table" is an interesting one
361 * and is answered in part by the use of the "nextcheck" field in the
362 * cache_detail.
363 * When a scan of a table begins, the nextcheck field is set to a time
364 * that is well into the future.
365 * While scanning, if an expiry time is found that is earlier than the
366 * current nextcheck time, nextcheck is set to that expiry time.
367 * If the flush_time is ever set to a time earlier than the nextcheck
368 * time, the nextcheck time is then set to that flush_time.
369 *
370 * A table is then only scanned if the current time is at least
371 * the nextcheck time.
372 *
373 */
374
375static LIST_HEAD(cache_list);
376static DEFINE_SPINLOCK(cache_list_lock);
377static struct cache_detail *current_detail;
378static int current_index;
379
380static void do_cache_clean(struct work_struct *work);
381static struct delayed_work cache_cleaner;
382
383void sunrpc_init_cache_detail(struct cache_detail *cd)
384{
385 spin_lock_init(&cd->hash_lock);
386 INIT_LIST_HEAD(&cd->queue);
387 spin_lock(&cache_list_lock);
388 cd->nextcheck = 0;
389 cd->entries = 0;
390 atomic_set(&cd->writers, 0);
391 cd->last_close = 0;
392 cd->last_warn = -1;
393 list_add(&cd->others, &cache_list);
394 spin_unlock(&cache_list_lock);
395
396 /* start the cleaning process */
397 queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
398}
399EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
400
401void sunrpc_destroy_cache_detail(struct cache_detail *cd)
402{
403 cache_purge(cd);
404 spin_lock(&cache_list_lock);
405 spin_lock(&cd->hash_lock);
406 if (current_detail == cd)
407 current_detail = NULL;
408 list_del_init(&cd->others);
409 spin_unlock(&cd->hash_lock);
410 spin_unlock(&cache_list_lock);
411 if (list_empty(&cache_list)) {
412 /* module must be being unloaded so its safe to kill the worker */
413 cancel_delayed_work_sync(&cache_cleaner);
414 }
415}
416EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
417
418/* clean cache tries to find something to clean
419 * and cleans it.
420 * It returns 1 if it cleaned something,
421 * 0 if it didn't find anything this time
422 * -1 if it fell off the end of the list.
423 */
424static int cache_clean(void)
425{
426 int rv = 0;
427 struct list_head *next;
428
429 spin_lock(&cache_list_lock);
430
431 /* find a suitable table if we don't already have one */
432 while (current_detail == NULL ||
433 current_index >= current_detail->hash_size) {
434 if (current_detail)
435 next = current_detail->others.next;
436 else
437 next = cache_list.next;
438 if (next == &cache_list) {
439 current_detail = NULL;
440 spin_unlock(&cache_list_lock);
441 return -1;
442 }
443 current_detail = list_entry(next, struct cache_detail, others);
444 if (current_detail->nextcheck > seconds_since_boot())
445 current_index = current_detail->hash_size;
446 else {
447 current_index = 0;
448 current_detail->nextcheck = seconds_since_boot()+30*60;
449 }
450 }
451
452 /* find a non-empty bucket in the table */
453 while (current_detail &&
454 current_index < current_detail->hash_size &&
455 hlist_empty(¤t_detail->hash_table[current_index]))
456 current_index++;
457
458 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
459
460 if (current_detail && current_index < current_detail->hash_size) {
461 struct cache_head *ch = NULL;
462 struct cache_detail *d;
463 struct hlist_head *head;
464 struct hlist_node *tmp;
465
466 spin_lock(¤t_detail->hash_lock);
467
468 /* Ok, now to clean this strand */
469
470 head = ¤t_detail->hash_table[current_index];
471 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
472 if (current_detail->nextcheck > ch->expiry_time)
473 current_detail->nextcheck = ch->expiry_time+1;
474 if (!cache_is_expired(current_detail, ch))
475 continue;
476
477 sunrpc_begin_cache_remove_entry(ch, current_detail);
478 trace_cache_entry_expired(current_detail, ch);
479 rv = 1;
480 break;
481 }
482
483 spin_unlock(¤t_detail->hash_lock);
484 d = current_detail;
485 if (!ch)
486 current_index ++;
487 spin_unlock(&cache_list_lock);
488 if (ch)
489 sunrpc_end_cache_remove_entry(ch, d);
490 } else
491 spin_unlock(&cache_list_lock);
492
493 return rv;
494}
495
496/*
497 * We want to regularly clean the cache, so we need to schedule some work ...
498 */
499static void do_cache_clean(struct work_struct *work)
500{
501 int delay = 5;
502 if (cache_clean() == -1)
503 delay = round_jiffies_relative(30*HZ);
504
505 if (list_empty(&cache_list))
506 delay = 0;
507
508 if (delay)
509 queue_delayed_work(system_power_efficient_wq,
510 &cache_cleaner, delay);
511}
512
513
514/*
515 * Clean all caches promptly. This just calls cache_clean
516 * repeatedly until we are sure that every cache has had a chance to
517 * be fully cleaned
518 */
519void cache_flush(void)
520{
521 while (cache_clean() != -1)
522 cond_resched();
523 while (cache_clean() != -1)
524 cond_resched();
525}
526EXPORT_SYMBOL_GPL(cache_flush);
527
528void cache_purge(struct cache_detail *detail)
529{
530 struct cache_head *ch = NULL;
531 struct hlist_head *head = NULL;
532 int i = 0;
533
534 spin_lock(&detail->hash_lock);
535 if (!detail->entries) {
536 spin_unlock(&detail->hash_lock);
537 return;
538 }
539
540 dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
541 for (i = 0; i < detail->hash_size; i++) {
542 head = &detail->hash_table[i];
543 while (!hlist_empty(head)) {
544 ch = hlist_entry(head->first, struct cache_head,
545 cache_list);
546 sunrpc_begin_cache_remove_entry(ch, detail);
547 spin_unlock(&detail->hash_lock);
548 sunrpc_end_cache_remove_entry(ch, detail);
549 spin_lock(&detail->hash_lock);
550 }
551 }
552 spin_unlock(&detail->hash_lock);
553}
554EXPORT_SYMBOL_GPL(cache_purge);
555
556
557/*
558 * Deferral and Revisiting of Requests.
559 *
560 * If a cache lookup finds a pending entry, we
561 * need to defer the request and revisit it later.
562 * All deferred requests are stored in a hash table,
563 * indexed by "struct cache_head *".
564 * As it may be wasteful to store a whole request
565 * structure, we allow the request to provide a
566 * deferred form, which must contain a
567 * 'struct cache_deferred_req'
568 * This cache_deferred_req contains a method to allow
569 * it to be revisited when cache info is available
570 */
571
572#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
573#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
574
575#define DFR_MAX 300 /* ??? */
576
577static DEFINE_SPINLOCK(cache_defer_lock);
578static LIST_HEAD(cache_defer_list);
579static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
580static int cache_defer_cnt;
581
582static void __unhash_deferred_req(struct cache_deferred_req *dreq)
583{
584 hlist_del_init(&dreq->hash);
585 if (!list_empty(&dreq->recent)) {
586 list_del_init(&dreq->recent);
587 cache_defer_cnt--;
588 }
589}
590
591static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
592{
593 int hash = DFR_HASH(item);
594
595 INIT_LIST_HEAD(&dreq->recent);
596 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
597}
598
599static void setup_deferral(struct cache_deferred_req *dreq,
600 struct cache_head *item,
601 int count_me)
602{
603
604 dreq->item = item;
605
606 spin_lock(&cache_defer_lock);
607
608 __hash_deferred_req(dreq, item);
609
610 if (count_me) {
611 cache_defer_cnt++;
612 list_add(&dreq->recent, &cache_defer_list);
613 }
614
615 spin_unlock(&cache_defer_lock);
616
617}
618
619struct thread_deferred_req {
620 struct cache_deferred_req handle;
621 struct completion completion;
622};
623
624static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
625{
626 struct thread_deferred_req *dr =
627 container_of(dreq, struct thread_deferred_req, handle);
628 complete(&dr->completion);
629}
630
631static void cache_wait_req(struct cache_req *req, struct cache_head *item)
632{
633 struct thread_deferred_req sleeper;
634 struct cache_deferred_req *dreq = &sleeper.handle;
635
636 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
637 dreq->revisit = cache_restart_thread;
638
639 setup_deferral(dreq, item, 0);
640
641 if (!test_bit(CACHE_PENDING, &item->flags) ||
642 wait_for_completion_interruptible_timeout(
643 &sleeper.completion, req->thread_wait) <= 0) {
644 /* The completion wasn't completed, so we need
645 * to clean up
646 */
647 spin_lock(&cache_defer_lock);
648 if (!hlist_unhashed(&sleeper.handle.hash)) {
649 __unhash_deferred_req(&sleeper.handle);
650 spin_unlock(&cache_defer_lock);
651 } else {
652 /* cache_revisit_request already removed
653 * this from the hash table, but hasn't
654 * called ->revisit yet. It will very soon
655 * and we need to wait for it.
656 */
657 spin_unlock(&cache_defer_lock);
658 wait_for_completion(&sleeper.completion);
659 }
660 }
661}
662
663static void cache_limit_defers(void)
664{
665 /* Make sure we haven't exceed the limit of allowed deferred
666 * requests.
667 */
668 struct cache_deferred_req *discard = NULL;
669
670 if (cache_defer_cnt <= DFR_MAX)
671 return;
672
673 spin_lock(&cache_defer_lock);
674
675 /* Consider removing either the first or the last */
676 if (cache_defer_cnt > DFR_MAX) {
677 if (prandom_u32() & 1)
678 discard = list_entry(cache_defer_list.next,
679 struct cache_deferred_req, recent);
680 else
681 discard = list_entry(cache_defer_list.prev,
682 struct cache_deferred_req, recent);
683 __unhash_deferred_req(discard);
684 }
685 spin_unlock(&cache_defer_lock);
686 if (discard)
687 discard->revisit(discard, 1);
688}
689
690/* Return true if and only if a deferred request is queued. */
691static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
692{
693 struct cache_deferred_req *dreq;
694
695 if (req->thread_wait) {
696 cache_wait_req(req, item);
697 if (!test_bit(CACHE_PENDING, &item->flags))
698 return false;
699 }
700 dreq = req->defer(req);
701 if (dreq == NULL)
702 return false;
703 setup_deferral(dreq, item, 1);
704 if (!test_bit(CACHE_PENDING, &item->flags))
705 /* Bit could have been cleared before we managed to
706 * set up the deferral, so need to revisit just in case
707 */
708 cache_revisit_request(item);
709
710 cache_limit_defers();
711 return true;
712}
713
714static void cache_revisit_request(struct cache_head *item)
715{
716 struct cache_deferred_req *dreq;
717 struct list_head pending;
718 struct hlist_node *tmp;
719 int hash = DFR_HASH(item);
720
721 INIT_LIST_HEAD(&pending);
722 spin_lock(&cache_defer_lock);
723
724 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
725 if (dreq->item == item) {
726 __unhash_deferred_req(dreq);
727 list_add(&dreq->recent, &pending);
728 }
729
730 spin_unlock(&cache_defer_lock);
731
732 while (!list_empty(&pending)) {
733 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
734 list_del_init(&dreq->recent);
735 dreq->revisit(dreq, 0);
736 }
737}
738
739void cache_clean_deferred(void *owner)
740{
741 struct cache_deferred_req *dreq, *tmp;
742 struct list_head pending;
743
744
745 INIT_LIST_HEAD(&pending);
746 spin_lock(&cache_defer_lock);
747
748 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
749 if (dreq->owner == owner) {
750 __unhash_deferred_req(dreq);
751 list_add(&dreq->recent, &pending);
752 }
753 }
754 spin_unlock(&cache_defer_lock);
755
756 while (!list_empty(&pending)) {
757 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
758 list_del_init(&dreq->recent);
759 dreq->revisit(dreq, 1);
760 }
761}
762
763/*
764 * communicate with user-space
765 *
766 * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
767 * On read, you get a full request, or block.
768 * On write, an update request is processed.
769 * Poll works if anything to read, and always allows write.
770 *
771 * Implemented by linked list of requests. Each open file has
772 * a ->private that also exists in this list. New requests are added
773 * to the end and may wakeup and preceding readers.
774 * New readers are added to the head. If, on read, an item is found with
775 * CACHE_UPCALLING clear, we free it from the list.
776 *
777 */
778
779static DEFINE_SPINLOCK(queue_lock);
780static DEFINE_MUTEX(queue_io_mutex);
781
782struct cache_queue {
783 struct list_head list;
784 int reader; /* if 0, then request */
785};
786struct cache_request {
787 struct cache_queue q;
788 struct cache_head *item;
789 char * buf;
790 int len;
791 int readers;
792};
793struct cache_reader {
794 struct cache_queue q;
795 int offset; /* if non-0, we have a refcnt on next request */
796};
797
798static int cache_request(struct cache_detail *detail,
799 struct cache_request *crq)
800{
801 char *bp = crq->buf;
802 int len = PAGE_SIZE;
803
804 detail->cache_request(detail, crq->item, &bp, &len);
805 if (len < 0)
806 return -EAGAIN;
807 return PAGE_SIZE - len;
808}
809
810static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
811 loff_t *ppos, struct cache_detail *cd)
812{
813 struct cache_reader *rp = filp->private_data;
814 struct cache_request *rq;
815 struct inode *inode = file_inode(filp);
816 int err;
817
818 if (count == 0)
819 return 0;
820
821 inode_lock(inode); /* protect against multiple concurrent
822 * readers on this file */
823 again:
824 spin_lock(&queue_lock);
825 /* need to find next request */
826 while (rp->q.list.next != &cd->queue &&
827 list_entry(rp->q.list.next, struct cache_queue, list)
828 ->reader) {
829 struct list_head *next = rp->q.list.next;
830 list_move(&rp->q.list, next);
831 }
832 if (rp->q.list.next == &cd->queue) {
833 spin_unlock(&queue_lock);
834 inode_unlock(inode);
835 WARN_ON_ONCE(rp->offset);
836 return 0;
837 }
838 rq = container_of(rp->q.list.next, struct cache_request, q.list);
839 WARN_ON_ONCE(rq->q.reader);
840 if (rp->offset == 0)
841 rq->readers++;
842 spin_unlock(&queue_lock);
843
844 if (rq->len == 0) {
845 err = cache_request(cd, rq);
846 if (err < 0)
847 goto out;
848 rq->len = err;
849 }
850
851 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
852 err = -EAGAIN;
853 spin_lock(&queue_lock);
854 list_move(&rp->q.list, &rq->q.list);
855 spin_unlock(&queue_lock);
856 } else {
857 if (rp->offset + count > rq->len)
858 count = rq->len - rp->offset;
859 err = -EFAULT;
860 if (copy_to_user(buf, rq->buf + rp->offset, count))
861 goto out;
862 rp->offset += count;
863 if (rp->offset >= rq->len) {
864 rp->offset = 0;
865 spin_lock(&queue_lock);
866 list_move(&rp->q.list, &rq->q.list);
867 spin_unlock(&queue_lock);
868 }
869 err = 0;
870 }
871 out:
872 if (rp->offset == 0) {
873 /* need to release rq */
874 spin_lock(&queue_lock);
875 rq->readers--;
876 if (rq->readers == 0 &&
877 !test_bit(CACHE_PENDING, &rq->item->flags)) {
878 list_del(&rq->q.list);
879 spin_unlock(&queue_lock);
880 cache_put(rq->item, cd);
881 kfree(rq->buf);
882 kfree(rq);
883 } else
884 spin_unlock(&queue_lock);
885 }
886 if (err == -EAGAIN)
887 goto again;
888 inode_unlock(inode);
889 return err ? err : count;
890}
891
892static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
893 size_t count, struct cache_detail *cd)
894{
895 ssize_t ret;
896
897 if (count == 0)
898 return -EINVAL;
899 if (copy_from_user(kaddr, buf, count))
900 return -EFAULT;
901 kaddr[count] = '\0';
902 ret = cd->cache_parse(cd, kaddr, count);
903 if (!ret)
904 ret = count;
905 return ret;
906}
907
908static ssize_t cache_slow_downcall(const char __user *buf,
909 size_t count, struct cache_detail *cd)
910{
911 static char write_buf[8192]; /* protected by queue_io_mutex */
912 ssize_t ret = -EINVAL;
913
914 if (count >= sizeof(write_buf))
915 goto out;
916 mutex_lock(&queue_io_mutex);
917 ret = cache_do_downcall(write_buf, buf, count, cd);
918 mutex_unlock(&queue_io_mutex);
919out:
920 return ret;
921}
922
923static ssize_t cache_downcall(struct address_space *mapping,
924 const char __user *buf,
925 size_t count, struct cache_detail *cd)
926{
927 struct page *page;
928 char *kaddr;
929 ssize_t ret = -ENOMEM;
930
931 if (count >= PAGE_SIZE)
932 goto out_slow;
933
934 page = find_or_create_page(mapping, 0, GFP_KERNEL);
935 if (!page)
936 goto out_slow;
937
938 kaddr = kmap(page);
939 ret = cache_do_downcall(kaddr, buf, count, cd);
940 kunmap(page);
941 unlock_page(page);
942 put_page(page);
943 return ret;
944out_slow:
945 return cache_slow_downcall(buf, count, cd);
946}
947
948static ssize_t cache_write(struct file *filp, const char __user *buf,
949 size_t count, loff_t *ppos,
950 struct cache_detail *cd)
951{
952 struct address_space *mapping = filp->f_mapping;
953 struct inode *inode = file_inode(filp);
954 ssize_t ret = -EINVAL;
955
956 if (!cd->cache_parse)
957 goto out;
958
959 inode_lock(inode);
960 ret = cache_downcall(mapping, buf, count, cd);
961 inode_unlock(inode);
962out:
963 return ret;
964}
965
966static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
967
968static __poll_t cache_poll(struct file *filp, poll_table *wait,
969 struct cache_detail *cd)
970{
971 __poll_t mask;
972 struct cache_reader *rp = filp->private_data;
973 struct cache_queue *cq;
974
975 poll_wait(filp, &queue_wait, wait);
976
977 /* alway allow write */
978 mask = EPOLLOUT | EPOLLWRNORM;
979
980 if (!rp)
981 return mask;
982
983 spin_lock(&queue_lock);
984
985 for (cq= &rp->q; &cq->list != &cd->queue;
986 cq = list_entry(cq->list.next, struct cache_queue, list))
987 if (!cq->reader) {
988 mask |= EPOLLIN | EPOLLRDNORM;
989 break;
990 }
991 spin_unlock(&queue_lock);
992 return mask;
993}
994
995static int cache_ioctl(struct inode *ino, struct file *filp,
996 unsigned int cmd, unsigned long arg,
997 struct cache_detail *cd)
998{
999 int len = 0;
1000 struct cache_reader *rp = filp->private_data;
1001 struct cache_queue *cq;
1002
1003 if (cmd != FIONREAD || !rp)
1004 return -EINVAL;
1005
1006 spin_lock(&queue_lock);
1007
1008 /* only find the length remaining in current request,
1009 * or the length of the next request
1010 */
1011 for (cq= &rp->q; &cq->list != &cd->queue;
1012 cq = list_entry(cq->list.next, struct cache_queue, list))
1013 if (!cq->reader) {
1014 struct cache_request *cr =
1015 container_of(cq, struct cache_request, q);
1016 len = cr->len - rp->offset;
1017 break;
1018 }
1019 spin_unlock(&queue_lock);
1020
1021 return put_user(len, (int __user *)arg);
1022}
1023
1024static int cache_open(struct inode *inode, struct file *filp,
1025 struct cache_detail *cd)
1026{
1027 struct cache_reader *rp = NULL;
1028
1029 if (!cd || !try_module_get(cd->owner))
1030 return -EACCES;
1031 nonseekable_open(inode, filp);
1032 if (filp->f_mode & FMODE_READ) {
1033 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1034 if (!rp) {
1035 module_put(cd->owner);
1036 return -ENOMEM;
1037 }
1038 rp->offset = 0;
1039 rp->q.reader = 1;
1040
1041 spin_lock(&queue_lock);
1042 list_add(&rp->q.list, &cd->queue);
1043 spin_unlock(&queue_lock);
1044 }
1045 if (filp->f_mode & FMODE_WRITE)
1046 atomic_inc(&cd->writers);
1047 filp->private_data = rp;
1048 return 0;
1049}
1050
1051static int cache_release(struct inode *inode, struct file *filp,
1052 struct cache_detail *cd)
1053{
1054 struct cache_reader *rp = filp->private_data;
1055
1056 if (rp) {
1057 spin_lock(&queue_lock);
1058 if (rp->offset) {
1059 struct cache_queue *cq;
1060 for (cq= &rp->q; &cq->list != &cd->queue;
1061 cq = list_entry(cq->list.next, struct cache_queue, list))
1062 if (!cq->reader) {
1063 container_of(cq, struct cache_request, q)
1064 ->readers--;
1065 break;
1066 }
1067 rp->offset = 0;
1068 }
1069 list_del(&rp->q.list);
1070 spin_unlock(&queue_lock);
1071
1072 filp->private_data = NULL;
1073 kfree(rp);
1074
1075 }
1076 if (filp->f_mode & FMODE_WRITE) {
1077 atomic_dec(&cd->writers);
1078 cd->last_close = seconds_since_boot();
1079 }
1080 module_put(cd->owner);
1081 return 0;
1082}
1083
1084
1085
1086static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1087{
1088 struct cache_queue *cq, *tmp;
1089 struct cache_request *cr;
1090 struct list_head dequeued;
1091
1092 INIT_LIST_HEAD(&dequeued);
1093 spin_lock(&queue_lock);
1094 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1095 if (!cq->reader) {
1096 cr = container_of(cq, struct cache_request, q);
1097 if (cr->item != ch)
1098 continue;
1099 if (test_bit(CACHE_PENDING, &ch->flags))
1100 /* Lost a race and it is pending again */
1101 break;
1102 if (cr->readers != 0)
1103 continue;
1104 list_move(&cr->q.list, &dequeued);
1105 }
1106 spin_unlock(&queue_lock);
1107 while (!list_empty(&dequeued)) {
1108 cr = list_entry(dequeued.next, struct cache_request, q.list);
1109 list_del(&cr->q.list);
1110 cache_put(cr->item, detail);
1111 kfree(cr->buf);
1112 kfree(cr);
1113 }
1114}
1115
1116/*
1117 * Support routines for text-based upcalls.
1118 * Fields are separated by spaces.
1119 * Fields are either mangled to quote space tab newline slosh with slosh
1120 * or a hexified with a leading \x
1121 * Record is terminated with newline.
1122 *
1123 */
1124
1125void qword_add(char **bpp, int *lp, char *str)
1126{
1127 char *bp = *bpp;
1128 int len = *lp;
1129 int ret;
1130
1131 if (len < 0) return;
1132
1133 ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1134 if (ret >= len) {
1135 bp += len;
1136 len = -1;
1137 } else {
1138 bp += ret;
1139 len -= ret;
1140 *bp++ = ' ';
1141 len--;
1142 }
1143 *bpp = bp;
1144 *lp = len;
1145}
1146EXPORT_SYMBOL_GPL(qword_add);
1147
1148void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1149{
1150 char *bp = *bpp;
1151 int len = *lp;
1152
1153 if (len < 0) return;
1154
1155 if (len > 2) {
1156 *bp++ = '\\';
1157 *bp++ = 'x';
1158 len -= 2;
1159 while (blen && len >= 2) {
1160 bp = hex_byte_pack(bp, *buf++);
1161 len -= 2;
1162 blen--;
1163 }
1164 }
1165 if (blen || len<1) len = -1;
1166 else {
1167 *bp++ = ' ';
1168 len--;
1169 }
1170 *bpp = bp;
1171 *lp = len;
1172}
1173EXPORT_SYMBOL_GPL(qword_addhex);
1174
1175static void warn_no_listener(struct cache_detail *detail)
1176{
1177 if (detail->last_warn != detail->last_close) {
1178 detail->last_warn = detail->last_close;
1179 if (detail->warn_no_listener)
1180 detail->warn_no_listener(detail, detail->last_close != 0);
1181 }
1182}
1183
1184static bool cache_listeners_exist(struct cache_detail *detail)
1185{
1186 if (atomic_read(&detail->writers))
1187 return true;
1188 if (detail->last_close == 0)
1189 /* This cache was never opened */
1190 return false;
1191 if (detail->last_close < seconds_since_boot() - 30)
1192 /*
1193 * We allow for the possibility that someone might
1194 * restart a userspace daemon without restarting the
1195 * server; but after 30 seconds, we give up.
1196 */
1197 return false;
1198 return true;
1199}
1200
1201/*
1202 * register an upcall request to user-space and queue it up for read() by the
1203 * upcall daemon.
1204 *
1205 * Each request is at most one page long.
1206 */
1207static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1208{
1209 char *buf;
1210 struct cache_request *crq;
1211 int ret = 0;
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 trace_cache_entry_upcall(detail, h);
1236 } else
1237 /* Lost a race, no longer PENDING, so don't enqueue */
1238 ret = -EAGAIN;
1239 spin_unlock(&queue_lock);
1240 wake_up(&queue_wait);
1241 if (ret == -EAGAIN) {
1242 kfree(buf);
1243 kfree(crq);
1244 }
1245 return ret;
1246}
1247
1248int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1249{
1250 if (test_and_set_bit(CACHE_PENDING, &h->flags))
1251 return 0;
1252 return cache_pipe_upcall(detail, h);
1253}
1254EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1255
1256int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1257 struct cache_head *h)
1258{
1259 if (!cache_listeners_exist(detail)) {
1260 warn_no_listener(detail);
1261 trace_cache_entry_no_listener(detail, h);
1262 return -EINVAL;
1263 }
1264 return sunrpc_cache_pipe_upcall(detail, h);
1265}
1266EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1267
1268/*
1269 * parse a message from user-space and pass it
1270 * to an appropriate cache
1271 * Messages are, like requests, separated into fields by
1272 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1273 *
1274 * Message is
1275 * reply cachename expiry key ... content....
1276 *
1277 * key and content are both parsed by cache
1278 */
1279
1280int qword_get(char **bpp, char *dest, int bufsize)
1281{
1282 /* return bytes copied, or -1 on error */
1283 char *bp = *bpp;
1284 int len = 0;
1285
1286 while (*bp == ' ') bp++;
1287
1288 if (bp[0] == '\\' && bp[1] == 'x') {
1289 /* HEX STRING */
1290 bp += 2;
1291 while (len < bufsize - 1) {
1292 int h, l;
1293
1294 h = hex_to_bin(bp[0]);
1295 if (h < 0)
1296 break;
1297
1298 l = hex_to_bin(bp[1]);
1299 if (l < 0)
1300 break;
1301
1302 *dest++ = (h << 4) | l;
1303 bp += 2;
1304 len++;
1305 }
1306 } else {
1307 /* text with \nnn octal quoting */
1308 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1309 if (*bp == '\\' &&
1310 isodigit(bp[1]) && (bp[1] <= '3') &&
1311 isodigit(bp[2]) &&
1312 isodigit(bp[3])) {
1313 int byte = (*++bp -'0');
1314 bp++;
1315 byte = (byte << 3) | (*bp++ - '0');
1316 byte = (byte << 3) | (*bp++ - '0');
1317 *dest++ = byte;
1318 len++;
1319 } else {
1320 *dest++ = *bp++;
1321 len++;
1322 }
1323 }
1324 }
1325
1326 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1327 return -1;
1328 while (*bp == ' ') bp++;
1329 *bpp = bp;
1330 *dest = '\0';
1331 return len;
1332}
1333EXPORT_SYMBOL_GPL(qword_get);
1334
1335
1336/*
1337 * support /proc/net/rpc/$CACHENAME/content
1338 * as a seqfile.
1339 * We call ->cache_show passing NULL for the item to
1340 * get a header, then pass each real item in the cache
1341 */
1342
1343static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1344{
1345 loff_t n = *pos;
1346 unsigned int hash, entry;
1347 struct cache_head *ch;
1348 struct cache_detail *cd = m->private;
1349
1350 if (!n--)
1351 return SEQ_START_TOKEN;
1352 hash = n >> 32;
1353 entry = n & ((1LL<<32) - 1);
1354
1355 hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1356 if (!entry--)
1357 return ch;
1358 n &= ~((1LL<<32) - 1);
1359 do {
1360 hash++;
1361 n += 1LL<<32;
1362 } while(hash < cd->hash_size &&
1363 hlist_empty(&cd->hash_table[hash]));
1364 if (hash >= cd->hash_size)
1365 return NULL;
1366 *pos = n+1;
1367 return hlist_entry_safe(rcu_dereference_raw(
1368 hlist_first_rcu(&cd->hash_table[hash])),
1369 struct cache_head, cache_list);
1370}
1371
1372static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1373{
1374 struct cache_head *ch = p;
1375 int hash = (*pos >> 32);
1376 struct cache_detail *cd = m->private;
1377
1378 if (p == SEQ_START_TOKEN)
1379 hash = 0;
1380 else if (ch->cache_list.next == NULL) {
1381 hash++;
1382 *pos += 1LL<<32;
1383 } else {
1384 ++*pos;
1385 return hlist_entry_safe(rcu_dereference_raw(
1386 hlist_next_rcu(&ch->cache_list)),
1387 struct cache_head, cache_list);
1388 }
1389 *pos &= ~((1LL<<32) - 1);
1390 while (hash < cd->hash_size &&
1391 hlist_empty(&cd->hash_table[hash])) {
1392 hash++;
1393 *pos += 1LL<<32;
1394 }
1395 if (hash >= cd->hash_size)
1396 return NULL;
1397 ++*pos;
1398 return hlist_entry_safe(rcu_dereference_raw(
1399 hlist_first_rcu(&cd->hash_table[hash])),
1400 struct cache_head, cache_list);
1401}
1402
1403void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1404 __acquires(RCU)
1405{
1406 rcu_read_lock();
1407 return __cache_seq_start(m, pos);
1408}
1409EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1410
1411void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1412{
1413 return cache_seq_next(file, p, pos);
1414}
1415EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1416
1417void cache_seq_stop_rcu(struct seq_file *m, void *p)
1418 __releases(RCU)
1419{
1420 rcu_read_unlock();
1421}
1422EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1423
1424static int c_show(struct seq_file *m, void *p)
1425{
1426 struct cache_head *cp = p;
1427 struct cache_detail *cd = m->private;
1428
1429 if (p == SEQ_START_TOKEN)
1430 return cd->cache_show(m, cd, NULL);
1431
1432 ifdebug(CACHE)
1433 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1434 convert_to_wallclock(cp->expiry_time),
1435 kref_read(&cp->ref), cp->flags);
1436 cache_get(cp);
1437 if (cache_check(cd, cp, NULL))
1438 /* cache_check does a cache_put on failure */
1439 seq_printf(m, "# ");
1440 else {
1441 if (cache_is_expired(cd, cp))
1442 seq_printf(m, "# ");
1443 cache_put(cp, cd);
1444 }
1445
1446 return cd->cache_show(m, cd, cp);
1447}
1448
1449static const struct seq_operations cache_content_op = {
1450 .start = cache_seq_start_rcu,
1451 .next = cache_seq_next_rcu,
1452 .stop = cache_seq_stop_rcu,
1453 .show = c_show,
1454};
1455
1456static int content_open(struct inode *inode, struct file *file,
1457 struct cache_detail *cd)
1458{
1459 struct seq_file *seq;
1460 int err;
1461
1462 if (!cd || !try_module_get(cd->owner))
1463 return -EACCES;
1464
1465 err = seq_open(file, &cache_content_op);
1466 if (err) {
1467 module_put(cd->owner);
1468 return err;
1469 }
1470
1471 seq = file->private_data;
1472 seq->private = cd;
1473 return 0;
1474}
1475
1476static int content_release(struct inode *inode, struct file *file,
1477 struct cache_detail *cd)
1478{
1479 int ret = seq_release(inode, file);
1480 module_put(cd->owner);
1481 return ret;
1482}
1483
1484static int open_flush(struct inode *inode, struct file *file,
1485 struct cache_detail *cd)
1486{
1487 if (!cd || !try_module_get(cd->owner))
1488 return -EACCES;
1489 return nonseekable_open(inode, file);
1490}
1491
1492static int release_flush(struct inode *inode, struct file *file,
1493 struct cache_detail *cd)
1494{
1495 module_put(cd->owner);
1496 return 0;
1497}
1498
1499static ssize_t read_flush(struct file *file, char __user *buf,
1500 size_t count, loff_t *ppos,
1501 struct cache_detail *cd)
1502{
1503 char tbuf[22];
1504 size_t len;
1505
1506 len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1507 convert_to_wallclock(cd->flush_time));
1508 return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1509}
1510
1511static ssize_t write_flush(struct file *file, const char __user *buf,
1512 size_t count, loff_t *ppos,
1513 struct cache_detail *cd)
1514{
1515 char tbuf[20];
1516 char *ep;
1517 time64_t now;
1518
1519 if (*ppos || count > sizeof(tbuf)-1)
1520 return -EINVAL;
1521 if (copy_from_user(tbuf, buf, count))
1522 return -EFAULT;
1523 tbuf[count] = 0;
1524 simple_strtoul(tbuf, &ep, 0);
1525 if (*ep && *ep != '\n')
1526 return -EINVAL;
1527 /* Note that while we check that 'buf' holds a valid number,
1528 * we always ignore the value and just flush everything.
1529 * Making use of the number leads to races.
1530 */
1531
1532 now = seconds_since_boot();
1533 /* Always flush everything, so behave like cache_purge()
1534 * Do this by advancing flush_time to the current time,
1535 * or by one second if it has already reached the current time.
1536 * Newly added cache entries will always have ->last_refresh greater
1537 * that ->flush_time, so they don't get flushed prematurely.
1538 */
1539
1540 if (cd->flush_time >= now)
1541 now = cd->flush_time + 1;
1542
1543 cd->flush_time = now;
1544 cd->nextcheck = now;
1545 cache_flush();
1546
1547 if (cd->flush)
1548 cd->flush();
1549
1550 *ppos += count;
1551 return count;
1552}
1553
1554static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1555 size_t count, loff_t *ppos)
1556{
1557 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1558
1559 return cache_read(filp, buf, count, ppos, cd);
1560}
1561
1562static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1563 size_t count, loff_t *ppos)
1564{
1565 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1566
1567 return cache_write(filp, buf, count, ppos, cd);
1568}
1569
1570static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1571{
1572 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1573
1574 return cache_poll(filp, wait, cd);
1575}
1576
1577static long cache_ioctl_procfs(struct file *filp,
1578 unsigned int cmd, unsigned long arg)
1579{
1580 struct inode *inode = file_inode(filp);
1581 struct cache_detail *cd = PDE_DATA(inode);
1582
1583 return cache_ioctl(inode, filp, cmd, arg, cd);
1584}
1585
1586static int cache_open_procfs(struct inode *inode, struct file *filp)
1587{
1588 struct cache_detail *cd = PDE_DATA(inode);
1589
1590 return cache_open(inode, filp, cd);
1591}
1592
1593static int cache_release_procfs(struct inode *inode, struct file *filp)
1594{
1595 struct cache_detail *cd = PDE_DATA(inode);
1596
1597 return cache_release(inode, filp, cd);
1598}
1599
1600static const struct proc_ops cache_channel_proc_ops = {
1601 .proc_lseek = no_llseek,
1602 .proc_read = cache_read_procfs,
1603 .proc_write = cache_write_procfs,
1604 .proc_poll = cache_poll_procfs,
1605 .proc_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1606 .proc_open = cache_open_procfs,
1607 .proc_release = cache_release_procfs,
1608};
1609
1610static int content_open_procfs(struct inode *inode, struct file *filp)
1611{
1612 struct cache_detail *cd = PDE_DATA(inode);
1613
1614 return content_open(inode, filp, cd);
1615}
1616
1617static int content_release_procfs(struct inode *inode, struct file *filp)
1618{
1619 struct cache_detail *cd = PDE_DATA(inode);
1620
1621 return content_release(inode, filp, cd);
1622}
1623
1624static const struct proc_ops content_proc_ops = {
1625 .proc_open = content_open_procfs,
1626 .proc_read = seq_read,
1627 .proc_lseek = seq_lseek,
1628 .proc_release = content_release_procfs,
1629};
1630
1631static int open_flush_procfs(struct inode *inode, struct file *filp)
1632{
1633 struct cache_detail *cd = PDE_DATA(inode);
1634
1635 return open_flush(inode, filp, cd);
1636}
1637
1638static int release_flush_procfs(struct inode *inode, struct file *filp)
1639{
1640 struct cache_detail *cd = PDE_DATA(inode);
1641
1642 return release_flush(inode, filp, cd);
1643}
1644
1645static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1646 size_t count, loff_t *ppos)
1647{
1648 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1649
1650 return read_flush(filp, buf, count, ppos, cd);
1651}
1652
1653static ssize_t write_flush_procfs(struct file *filp,
1654 const char __user *buf,
1655 size_t count, loff_t *ppos)
1656{
1657 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1658
1659 return write_flush(filp, buf, count, ppos, cd);
1660}
1661
1662static const struct proc_ops cache_flush_proc_ops = {
1663 .proc_open = open_flush_procfs,
1664 .proc_read = read_flush_procfs,
1665 .proc_write = write_flush_procfs,
1666 .proc_release = release_flush_procfs,
1667 .proc_lseek = no_llseek,
1668};
1669
1670static void remove_cache_proc_entries(struct cache_detail *cd)
1671{
1672 if (cd->procfs) {
1673 proc_remove(cd->procfs);
1674 cd->procfs = NULL;
1675 }
1676}
1677
1678#ifdef CONFIG_PROC_FS
1679static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1680{
1681 struct proc_dir_entry *p;
1682 struct sunrpc_net *sn;
1683
1684 sn = net_generic(net, sunrpc_net_id);
1685 cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1686 if (cd->procfs == NULL)
1687 goto out_nomem;
1688
1689 p = proc_create_data("flush", S_IFREG | 0600,
1690 cd->procfs, &cache_flush_proc_ops, cd);
1691 if (p == NULL)
1692 goto out_nomem;
1693
1694 if (cd->cache_request || cd->cache_parse) {
1695 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1696 &cache_channel_proc_ops, cd);
1697 if (p == NULL)
1698 goto out_nomem;
1699 }
1700 if (cd->cache_show) {
1701 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1702 &content_proc_ops, cd);
1703 if (p == NULL)
1704 goto out_nomem;
1705 }
1706 return 0;
1707out_nomem:
1708 remove_cache_proc_entries(cd);
1709 return -ENOMEM;
1710}
1711#else /* CONFIG_PROC_FS */
1712static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1713{
1714 return 0;
1715}
1716#endif
1717
1718void __init cache_initialize(void)
1719{
1720 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1721}
1722
1723int cache_register_net(struct cache_detail *cd, struct net *net)
1724{
1725 int ret;
1726
1727 sunrpc_init_cache_detail(cd);
1728 ret = create_cache_proc_entries(cd, net);
1729 if (ret)
1730 sunrpc_destroy_cache_detail(cd);
1731 return ret;
1732}
1733EXPORT_SYMBOL_GPL(cache_register_net);
1734
1735void cache_unregister_net(struct cache_detail *cd, struct net *net)
1736{
1737 remove_cache_proc_entries(cd);
1738 sunrpc_destroy_cache_detail(cd);
1739}
1740EXPORT_SYMBOL_GPL(cache_unregister_net);
1741
1742struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1743{
1744 struct cache_detail *cd;
1745 int i;
1746
1747 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1748 if (cd == NULL)
1749 return ERR_PTR(-ENOMEM);
1750
1751 cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1752 GFP_KERNEL);
1753 if (cd->hash_table == NULL) {
1754 kfree(cd);
1755 return ERR_PTR(-ENOMEM);
1756 }
1757
1758 for (i = 0; i < cd->hash_size; i++)
1759 INIT_HLIST_HEAD(&cd->hash_table[i]);
1760 cd->net = net;
1761 return cd;
1762}
1763EXPORT_SYMBOL_GPL(cache_create_net);
1764
1765void cache_destroy_net(struct cache_detail *cd, struct net *net)
1766{
1767 kfree(cd->hash_table);
1768 kfree(cd);
1769}
1770EXPORT_SYMBOL_GPL(cache_destroy_net);
1771
1772static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1773 size_t count, loff_t *ppos)
1774{
1775 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1776
1777 return cache_read(filp, buf, count, ppos, cd);
1778}
1779
1780static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1781 size_t count, loff_t *ppos)
1782{
1783 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1784
1785 return cache_write(filp, buf, count, ppos, cd);
1786}
1787
1788static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1789{
1790 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1791
1792 return cache_poll(filp, wait, cd);
1793}
1794
1795static long cache_ioctl_pipefs(struct file *filp,
1796 unsigned int cmd, unsigned long arg)
1797{
1798 struct inode *inode = file_inode(filp);
1799 struct cache_detail *cd = RPC_I(inode)->private;
1800
1801 return cache_ioctl(inode, filp, cmd, arg, cd);
1802}
1803
1804static int cache_open_pipefs(struct inode *inode, struct file *filp)
1805{
1806 struct cache_detail *cd = RPC_I(inode)->private;
1807
1808 return cache_open(inode, filp, cd);
1809}
1810
1811static int cache_release_pipefs(struct inode *inode, struct file *filp)
1812{
1813 struct cache_detail *cd = RPC_I(inode)->private;
1814
1815 return cache_release(inode, filp, cd);
1816}
1817
1818const struct file_operations cache_file_operations_pipefs = {
1819 .owner = THIS_MODULE,
1820 .llseek = no_llseek,
1821 .read = cache_read_pipefs,
1822 .write = cache_write_pipefs,
1823 .poll = cache_poll_pipefs,
1824 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1825 .open = cache_open_pipefs,
1826 .release = cache_release_pipefs,
1827};
1828
1829static int content_open_pipefs(struct inode *inode, struct file *filp)
1830{
1831 struct cache_detail *cd = RPC_I(inode)->private;
1832
1833 return content_open(inode, filp, cd);
1834}
1835
1836static int content_release_pipefs(struct inode *inode, struct file *filp)
1837{
1838 struct cache_detail *cd = RPC_I(inode)->private;
1839
1840 return content_release(inode, filp, cd);
1841}
1842
1843const struct file_operations content_file_operations_pipefs = {
1844 .open = content_open_pipefs,
1845 .read = seq_read,
1846 .llseek = seq_lseek,
1847 .release = content_release_pipefs,
1848};
1849
1850static int open_flush_pipefs(struct inode *inode, struct file *filp)
1851{
1852 struct cache_detail *cd = RPC_I(inode)->private;
1853
1854 return open_flush(inode, filp, cd);
1855}
1856
1857static int release_flush_pipefs(struct inode *inode, struct file *filp)
1858{
1859 struct cache_detail *cd = RPC_I(inode)->private;
1860
1861 return release_flush(inode, filp, cd);
1862}
1863
1864static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1865 size_t count, loff_t *ppos)
1866{
1867 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1868
1869 return read_flush(filp, buf, count, ppos, cd);
1870}
1871
1872static ssize_t write_flush_pipefs(struct file *filp,
1873 const char __user *buf,
1874 size_t count, loff_t *ppos)
1875{
1876 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1877
1878 return write_flush(filp, buf, count, ppos, cd);
1879}
1880
1881const struct file_operations cache_flush_operations_pipefs = {
1882 .open = open_flush_pipefs,
1883 .read = read_flush_pipefs,
1884 .write = write_flush_pipefs,
1885 .release = release_flush_pipefs,
1886 .llseek = no_llseek,
1887};
1888
1889int sunrpc_cache_register_pipefs(struct dentry *parent,
1890 const char *name, umode_t umode,
1891 struct cache_detail *cd)
1892{
1893 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1894 if (IS_ERR(dir))
1895 return PTR_ERR(dir);
1896 cd->pipefs = dir;
1897 return 0;
1898}
1899EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1900
1901void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1902{
1903 if (cd->pipefs) {
1904 rpc_remove_cache_dir(cd->pipefs);
1905 cd->pipefs = NULL;
1906 }
1907}
1908EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1909
1910void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1911{
1912 spin_lock(&cd->hash_lock);
1913 if (!hlist_unhashed(&h->cache_list)){
1914 sunrpc_begin_cache_remove_entry(h, cd);
1915 spin_unlock(&cd->hash_lock);
1916 sunrpc_end_cache_remove_entry(h, cd);
1917 } else
1918 spin_unlock(&cd->hash_lock);
1919}
1920EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);
1/*
2 * net/sunrpc/cache.c
3 *
4 * Generic code for various authentication-related caches
5 * used by sunrpc clients and servers.
6 *
7 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
8 *
9 * Released under terms in GPL version 2. See COPYING.
10 *
11 */
12
13#include <linux/types.h>
14#include <linux/fs.h>
15#include <linux/file.h>
16#include <linux/slab.h>
17#include <linux/signal.h>
18#include <linux/sched.h>
19#include <linux/kmod.h>
20#include <linux/list.h>
21#include <linux/module.h>
22#include <linux/ctype.h>
23#include <linux/string_helpers.h>
24#include <linux/uaccess.h>
25#include <linux/poll.h>
26#include <linux/seq_file.h>
27#include <linux/proc_fs.h>
28#include <linux/net.h>
29#include <linux/workqueue.h>
30#include <linux/mutex.h>
31#include <linux/pagemap.h>
32#include <asm/ioctls.h>
33#include <linux/sunrpc/types.h>
34#include <linux/sunrpc/cache.h>
35#include <linux/sunrpc/stats.h>
36#include <linux/sunrpc/rpc_pipe_fs.h>
37#include "netns.h"
38
39#define RPCDBG_FACILITY RPCDBG_CACHE
40
41static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
42static void cache_revisit_request(struct cache_head *item);
43
44static void cache_init(struct cache_head *h, struct cache_detail *detail)
45{
46 time_t now = seconds_since_boot();
47 INIT_HLIST_NODE(&h->cache_list);
48 h->flags = 0;
49 kref_init(&h->ref);
50 h->expiry_time = now + CACHE_NEW_EXPIRY;
51 if (now <= detail->flush_time)
52 /* ensure it isn't already expired */
53 now = detail->flush_time + 1;
54 h->last_refresh = now;
55}
56
57struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
58 struct cache_head *key, int hash)
59{
60 struct cache_head *new = NULL, *freeme = NULL, *tmp = NULL;
61 struct hlist_head *head;
62
63 head = &detail->hash_table[hash];
64
65 read_lock(&detail->hash_lock);
66
67 hlist_for_each_entry(tmp, head, cache_list) {
68 if (detail->match(tmp, key)) {
69 if (cache_is_expired(detail, tmp))
70 /* This entry is expired, we will discard it. */
71 break;
72 cache_get(tmp);
73 read_unlock(&detail->hash_lock);
74 return tmp;
75 }
76 }
77 read_unlock(&detail->hash_lock);
78 /* Didn't find anything, insert an empty entry */
79
80 new = detail->alloc();
81 if (!new)
82 return NULL;
83 /* must fully initialise 'new', else
84 * we might get lose if we need to
85 * cache_put it soon.
86 */
87 cache_init(new, detail);
88 detail->init(new, key);
89
90 write_lock(&detail->hash_lock);
91
92 /* check if entry appeared while we slept */
93 hlist_for_each_entry(tmp, head, cache_list) {
94 if (detail->match(tmp, key)) {
95 if (cache_is_expired(detail, tmp)) {
96 hlist_del_init(&tmp->cache_list);
97 detail->entries --;
98 freeme = tmp;
99 break;
100 }
101 cache_get(tmp);
102 write_unlock(&detail->hash_lock);
103 cache_put(new, detail);
104 return tmp;
105 }
106 }
107
108 hlist_add_head(&new->cache_list, head);
109 detail->entries++;
110 cache_get(new);
111 write_unlock(&detail->hash_lock);
112
113 if (freeme)
114 cache_put(freeme, detail);
115 return new;
116}
117EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
118
119
120static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
121
122static void cache_fresh_locked(struct cache_head *head, time_t expiry,
123 struct cache_detail *detail)
124{
125 time_t now = seconds_since_boot();
126 if (now <= detail->flush_time)
127 /* ensure it isn't immediately treated as expired */
128 now = detail->flush_time + 1;
129 head->expiry_time = expiry;
130 head->last_refresh = now;
131 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
132 set_bit(CACHE_VALID, &head->flags);
133}
134
135static void cache_fresh_unlocked(struct cache_head *head,
136 struct cache_detail *detail)
137{
138 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
139 cache_revisit_request(head);
140 cache_dequeue(detail, head);
141 }
142}
143
144struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
145 struct cache_head *new, struct cache_head *old, int hash)
146{
147 /* The 'old' entry is to be replaced by 'new'.
148 * If 'old' is not VALID, we update it directly,
149 * otherwise we need to replace it
150 */
151 struct cache_head *tmp;
152
153 if (!test_bit(CACHE_VALID, &old->flags)) {
154 write_lock(&detail->hash_lock);
155 if (!test_bit(CACHE_VALID, &old->flags)) {
156 if (test_bit(CACHE_NEGATIVE, &new->flags))
157 set_bit(CACHE_NEGATIVE, &old->flags);
158 else
159 detail->update(old, new);
160 cache_fresh_locked(old, new->expiry_time, detail);
161 write_unlock(&detail->hash_lock);
162 cache_fresh_unlocked(old, detail);
163 return old;
164 }
165 write_unlock(&detail->hash_lock);
166 }
167 /* We need to insert a new entry */
168 tmp = detail->alloc();
169 if (!tmp) {
170 cache_put(old, detail);
171 return NULL;
172 }
173 cache_init(tmp, detail);
174 detail->init(tmp, old);
175
176 write_lock(&detail->hash_lock);
177 if (test_bit(CACHE_NEGATIVE, &new->flags))
178 set_bit(CACHE_NEGATIVE, &tmp->flags);
179 else
180 detail->update(tmp, new);
181 hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
182 detail->entries++;
183 cache_get(tmp);
184 cache_fresh_locked(tmp, new->expiry_time, detail);
185 cache_fresh_locked(old, 0, detail);
186 write_unlock(&detail->hash_lock);
187 cache_fresh_unlocked(tmp, detail);
188 cache_fresh_unlocked(old, detail);
189 cache_put(old, detail);
190 return tmp;
191}
192EXPORT_SYMBOL_GPL(sunrpc_cache_update);
193
194static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
195{
196 if (cd->cache_upcall)
197 return cd->cache_upcall(cd, h);
198 return sunrpc_cache_pipe_upcall(cd, h);
199}
200
201static inline int cache_is_valid(struct cache_head *h)
202{
203 if (!test_bit(CACHE_VALID, &h->flags))
204 return -EAGAIN;
205 else {
206 /* entry is valid */
207 if (test_bit(CACHE_NEGATIVE, &h->flags))
208 return -ENOENT;
209 else {
210 /*
211 * In combination with write barrier in
212 * sunrpc_cache_update, ensures that anyone
213 * using the cache entry after this sees the
214 * updated contents:
215 */
216 smp_rmb();
217 return 0;
218 }
219 }
220}
221
222static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
223{
224 int rv;
225
226 write_lock(&detail->hash_lock);
227 rv = cache_is_valid(h);
228 if (rv == -EAGAIN) {
229 set_bit(CACHE_NEGATIVE, &h->flags);
230 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
231 detail);
232 rv = -ENOENT;
233 }
234 write_unlock(&detail->hash_lock);
235 cache_fresh_unlocked(h, detail);
236 return rv;
237}
238
239/*
240 * This is the generic cache management routine for all
241 * the authentication caches.
242 * It checks the currency of a cache item and will (later)
243 * initiate an upcall to fill it if needed.
244 *
245 *
246 * Returns 0 if the cache_head can be used, or cache_puts it and returns
247 * -EAGAIN if upcall is pending and request has been queued
248 * -ETIMEDOUT if upcall failed or request could not be queue or
249 * upcall completed but item is still invalid (implying that
250 * the cache item has been replaced with a newer one).
251 * -ENOENT if cache entry was negative
252 */
253int cache_check(struct cache_detail *detail,
254 struct cache_head *h, struct cache_req *rqstp)
255{
256 int rv;
257 long refresh_age, age;
258
259 /* First decide return status as best we can */
260 rv = cache_is_valid(h);
261
262 /* now see if we want to start an upcall */
263 refresh_age = (h->expiry_time - h->last_refresh);
264 age = seconds_since_boot() - h->last_refresh;
265
266 if (rqstp == NULL) {
267 if (rv == -EAGAIN)
268 rv = -ENOENT;
269 } else if (rv == -EAGAIN ||
270 (h->expiry_time != 0 && age > refresh_age/2)) {
271 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
272 refresh_age, age);
273 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
274 switch (cache_make_upcall(detail, h)) {
275 case -EINVAL:
276 rv = try_to_negate_entry(detail, h);
277 break;
278 case -EAGAIN:
279 cache_fresh_unlocked(h, detail);
280 break;
281 }
282 }
283 }
284
285 if (rv == -EAGAIN) {
286 if (!cache_defer_req(rqstp, h)) {
287 /*
288 * Request was not deferred; handle it as best
289 * we can ourselves:
290 */
291 rv = cache_is_valid(h);
292 if (rv == -EAGAIN)
293 rv = -ETIMEDOUT;
294 }
295 }
296 if (rv)
297 cache_put(h, detail);
298 return rv;
299}
300EXPORT_SYMBOL_GPL(cache_check);
301
302/*
303 * caches need to be periodically cleaned.
304 * For this we maintain a list of cache_detail and
305 * a current pointer into that list and into the table
306 * for that entry.
307 *
308 * Each time cache_clean is called it finds the next non-empty entry
309 * in the current table and walks the list in that entry
310 * looking for entries that can be removed.
311 *
312 * An entry gets removed if:
313 * - The expiry is before current time
314 * - The last_refresh time is before the flush_time for that cache
315 *
316 * later we might drop old entries with non-NEVER expiry if that table
317 * is getting 'full' for some definition of 'full'
318 *
319 * The question of "how often to scan a table" is an interesting one
320 * and is answered in part by the use of the "nextcheck" field in the
321 * cache_detail.
322 * When a scan of a table begins, the nextcheck field is set to a time
323 * that is well into the future.
324 * While scanning, if an expiry time is found that is earlier than the
325 * current nextcheck time, nextcheck is set to that expiry time.
326 * If the flush_time is ever set to a time earlier than the nextcheck
327 * time, the nextcheck time is then set to that flush_time.
328 *
329 * A table is then only scanned if the current time is at least
330 * the nextcheck time.
331 *
332 */
333
334static LIST_HEAD(cache_list);
335static DEFINE_SPINLOCK(cache_list_lock);
336static struct cache_detail *current_detail;
337static int current_index;
338
339static void do_cache_clean(struct work_struct *work);
340static struct delayed_work cache_cleaner;
341
342void sunrpc_init_cache_detail(struct cache_detail *cd)
343{
344 rwlock_init(&cd->hash_lock);
345 INIT_LIST_HEAD(&cd->queue);
346 spin_lock(&cache_list_lock);
347 cd->nextcheck = 0;
348 cd->entries = 0;
349 atomic_set(&cd->readers, 0);
350 cd->last_close = 0;
351 cd->last_warn = -1;
352 list_add(&cd->others, &cache_list);
353 spin_unlock(&cache_list_lock);
354
355 /* start the cleaning process */
356 queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
357}
358EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
359
360void sunrpc_destroy_cache_detail(struct cache_detail *cd)
361{
362 cache_purge(cd);
363 spin_lock(&cache_list_lock);
364 write_lock(&cd->hash_lock);
365 if (current_detail == cd)
366 current_detail = NULL;
367 list_del_init(&cd->others);
368 write_unlock(&cd->hash_lock);
369 spin_unlock(&cache_list_lock);
370 if (list_empty(&cache_list)) {
371 /* module must be being unloaded so its safe to kill the worker */
372 cancel_delayed_work_sync(&cache_cleaner);
373 }
374}
375EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
376
377/* clean cache tries to find something to clean
378 * and cleans it.
379 * It returns 1 if it cleaned something,
380 * 0 if it didn't find anything this time
381 * -1 if it fell off the end of the list.
382 */
383static int cache_clean(void)
384{
385 int rv = 0;
386 struct list_head *next;
387
388 spin_lock(&cache_list_lock);
389
390 /* find a suitable table if we don't already have one */
391 while (current_detail == NULL ||
392 current_index >= current_detail->hash_size) {
393 if (current_detail)
394 next = current_detail->others.next;
395 else
396 next = cache_list.next;
397 if (next == &cache_list) {
398 current_detail = NULL;
399 spin_unlock(&cache_list_lock);
400 return -1;
401 }
402 current_detail = list_entry(next, struct cache_detail, others);
403 if (current_detail->nextcheck > seconds_since_boot())
404 current_index = current_detail->hash_size;
405 else {
406 current_index = 0;
407 current_detail->nextcheck = seconds_since_boot()+30*60;
408 }
409 }
410
411 /* find a non-empty bucket in the table */
412 while (current_detail &&
413 current_index < current_detail->hash_size &&
414 hlist_empty(¤t_detail->hash_table[current_index]))
415 current_index++;
416
417 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
418
419 if (current_detail && current_index < current_detail->hash_size) {
420 struct cache_head *ch = NULL;
421 struct cache_detail *d;
422 struct hlist_head *head;
423 struct hlist_node *tmp;
424
425 write_lock(¤t_detail->hash_lock);
426
427 /* Ok, now to clean this strand */
428
429 head = ¤t_detail->hash_table[current_index];
430 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
431 if (current_detail->nextcheck > ch->expiry_time)
432 current_detail->nextcheck = ch->expiry_time+1;
433 if (!cache_is_expired(current_detail, ch))
434 continue;
435
436 hlist_del_init(&ch->cache_list);
437 current_detail->entries--;
438 rv = 1;
439 break;
440 }
441
442 write_unlock(¤t_detail->hash_lock);
443 d = current_detail;
444 if (!ch)
445 current_index ++;
446 spin_unlock(&cache_list_lock);
447 if (ch) {
448 set_bit(CACHE_CLEANED, &ch->flags);
449 cache_fresh_unlocked(ch, d);
450 cache_put(ch, d);
451 }
452 } else
453 spin_unlock(&cache_list_lock);
454
455 return rv;
456}
457
458/*
459 * We want to regularly clean the cache, so we need to schedule some work ...
460 */
461static void do_cache_clean(struct work_struct *work)
462{
463 int delay = 5;
464 if (cache_clean() == -1)
465 delay = round_jiffies_relative(30*HZ);
466
467 if (list_empty(&cache_list))
468 delay = 0;
469
470 if (delay)
471 queue_delayed_work(system_power_efficient_wq,
472 &cache_cleaner, delay);
473}
474
475
476/*
477 * Clean all caches promptly. This just calls cache_clean
478 * repeatedly until we are sure that every cache has had a chance to
479 * be fully cleaned
480 */
481void cache_flush(void)
482{
483 while (cache_clean() != -1)
484 cond_resched();
485 while (cache_clean() != -1)
486 cond_resched();
487}
488EXPORT_SYMBOL_GPL(cache_flush);
489
490void cache_purge(struct cache_detail *detail)
491{
492 struct cache_head *ch = NULL;
493 struct hlist_head *head = NULL;
494 struct hlist_node *tmp = NULL;
495 int i = 0;
496
497 write_lock(&detail->hash_lock);
498 if (!detail->entries) {
499 write_unlock(&detail->hash_lock);
500 return;
501 }
502
503 dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
504 for (i = 0; i < detail->hash_size; i++) {
505 head = &detail->hash_table[i];
506 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
507 hlist_del_init(&ch->cache_list);
508 detail->entries--;
509
510 set_bit(CACHE_CLEANED, &ch->flags);
511 write_unlock(&detail->hash_lock);
512 cache_fresh_unlocked(ch, detail);
513 cache_put(ch, detail);
514 write_lock(&detail->hash_lock);
515 }
516 }
517 write_unlock(&detail->hash_lock);
518}
519EXPORT_SYMBOL_GPL(cache_purge);
520
521
522/*
523 * Deferral and Revisiting of Requests.
524 *
525 * If a cache lookup finds a pending entry, we
526 * need to defer the request and revisit it later.
527 * All deferred requests are stored in a hash table,
528 * indexed by "struct cache_head *".
529 * As it may be wasteful to store a whole request
530 * structure, we allow the request to provide a
531 * deferred form, which must contain a
532 * 'struct cache_deferred_req'
533 * This cache_deferred_req contains a method to allow
534 * it to be revisited when cache info is available
535 */
536
537#define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
538#define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
539
540#define DFR_MAX 300 /* ??? */
541
542static DEFINE_SPINLOCK(cache_defer_lock);
543static LIST_HEAD(cache_defer_list);
544static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
545static int cache_defer_cnt;
546
547static void __unhash_deferred_req(struct cache_deferred_req *dreq)
548{
549 hlist_del_init(&dreq->hash);
550 if (!list_empty(&dreq->recent)) {
551 list_del_init(&dreq->recent);
552 cache_defer_cnt--;
553 }
554}
555
556static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
557{
558 int hash = DFR_HASH(item);
559
560 INIT_LIST_HEAD(&dreq->recent);
561 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
562}
563
564static void setup_deferral(struct cache_deferred_req *dreq,
565 struct cache_head *item,
566 int count_me)
567{
568
569 dreq->item = item;
570
571 spin_lock(&cache_defer_lock);
572
573 __hash_deferred_req(dreq, item);
574
575 if (count_me) {
576 cache_defer_cnt++;
577 list_add(&dreq->recent, &cache_defer_list);
578 }
579
580 spin_unlock(&cache_defer_lock);
581
582}
583
584struct thread_deferred_req {
585 struct cache_deferred_req handle;
586 struct completion completion;
587};
588
589static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
590{
591 struct thread_deferred_req *dr =
592 container_of(dreq, struct thread_deferred_req, handle);
593 complete(&dr->completion);
594}
595
596static void cache_wait_req(struct cache_req *req, struct cache_head *item)
597{
598 struct thread_deferred_req sleeper;
599 struct cache_deferred_req *dreq = &sleeper.handle;
600
601 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
602 dreq->revisit = cache_restart_thread;
603
604 setup_deferral(dreq, item, 0);
605
606 if (!test_bit(CACHE_PENDING, &item->flags) ||
607 wait_for_completion_interruptible_timeout(
608 &sleeper.completion, req->thread_wait) <= 0) {
609 /* The completion wasn't completed, so we need
610 * to clean up
611 */
612 spin_lock(&cache_defer_lock);
613 if (!hlist_unhashed(&sleeper.handle.hash)) {
614 __unhash_deferred_req(&sleeper.handle);
615 spin_unlock(&cache_defer_lock);
616 } else {
617 /* cache_revisit_request already removed
618 * this from the hash table, but hasn't
619 * called ->revisit yet. It will very soon
620 * and we need to wait for it.
621 */
622 spin_unlock(&cache_defer_lock);
623 wait_for_completion(&sleeper.completion);
624 }
625 }
626}
627
628static void cache_limit_defers(void)
629{
630 /* Make sure we haven't exceed the limit of allowed deferred
631 * requests.
632 */
633 struct cache_deferred_req *discard = NULL;
634
635 if (cache_defer_cnt <= DFR_MAX)
636 return;
637
638 spin_lock(&cache_defer_lock);
639
640 /* Consider removing either the first or the last */
641 if (cache_defer_cnt > DFR_MAX) {
642 if (prandom_u32() & 1)
643 discard = list_entry(cache_defer_list.next,
644 struct cache_deferred_req, recent);
645 else
646 discard = list_entry(cache_defer_list.prev,
647 struct cache_deferred_req, recent);
648 __unhash_deferred_req(discard);
649 }
650 spin_unlock(&cache_defer_lock);
651 if (discard)
652 discard->revisit(discard, 1);
653}
654
655/* Return true if and only if a deferred request is queued. */
656static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
657{
658 struct cache_deferred_req *dreq;
659
660 if (req->thread_wait) {
661 cache_wait_req(req, item);
662 if (!test_bit(CACHE_PENDING, &item->flags))
663 return false;
664 }
665 dreq = req->defer(req);
666 if (dreq == NULL)
667 return false;
668 setup_deferral(dreq, item, 1);
669 if (!test_bit(CACHE_PENDING, &item->flags))
670 /* Bit could have been cleared before we managed to
671 * set up the deferral, so need to revisit just in case
672 */
673 cache_revisit_request(item);
674
675 cache_limit_defers();
676 return true;
677}
678
679static void cache_revisit_request(struct cache_head *item)
680{
681 struct cache_deferred_req *dreq;
682 struct list_head pending;
683 struct hlist_node *tmp;
684 int hash = DFR_HASH(item);
685
686 INIT_LIST_HEAD(&pending);
687 spin_lock(&cache_defer_lock);
688
689 hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
690 if (dreq->item == item) {
691 __unhash_deferred_req(dreq);
692 list_add(&dreq->recent, &pending);
693 }
694
695 spin_unlock(&cache_defer_lock);
696
697 while (!list_empty(&pending)) {
698 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
699 list_del_init(&dreq->recent);
700 dreq->revisit(dreq, 0);
701 }
702}
703
704void cache_clean_deferred(void *owner)
705{
706 struct cache_deferred_req *dreq, *tmp;
707 struct list_head pending;
708
709
710 INIT_LIST_HEAD(&pending);
711 spin_lock(&cache_defer_lock);
712
713 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
714 if (dreq->owner == owner) {
715 __unhash_deferred_req(dreq);
716 list_add(&dreq->recent, &pending);
717 }
718 }
719 spin_unlock(&cache_defer_lock);
720
721 while (!list_empty(&pending)) {
722 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
723 list_del_init(&dreq->recent);
724 dreq->revisit(dreq, 1);
725 }
726}
727
728/*
729 * communicate with user-space
730 *
731 * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
732 * On read, you get a full request, or block.
733 * On write, an update request is processed.
734 * Poll works if anything to read, and always allows write.
735 *
736 * Implemented by linked list of requests. Each open file has
737 * a ->private that also exists in this list. New requests are added
738 * to the end and may wakeup and preceding readers.
739 * New readers are added to the head. If, on read, an item is found with
740 * CACHE_UPCALLING clear, we free it from the list.
741 *
742 */
743
744static DEFINE_SPINLOCK(queue_lock);
745static DEFINE_MUTEX(queue_io_mutex);
746
747struct cache_queue {
748 struct list_head list;
749 int reader; /* if 0, then request */
750};
751struct cache_request {
752 struct cache_queue q;
753 struct cache_head *item;
754 char * buf;
755 int len;
756 int readers;
757};
758struct cache_reader {
759 struct cache_queue q;
760 int offset; /* if non-0, we have a refcnt on next request */
761};
762
763static int cache_request(struct cache_detail *detail,
764 struct cache_request *crq)
765{
766 char *bp = crq->buf;
767 int len = PAGE_SIZE;
768
769 detail->cache_request(detail, crq->item, &bp, &len);
770 if (len < 0)
771 return -EAGAIN;
772 return PAGE_SIZE - len;
773}
774
775static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
776 loff_t *ppos, struct cache_detail *cd)
777{
778 struct cache_reader *rp = filp->private_data;
779 struct cache_request *rq;
780 struct inode *inode = file_inode(filp);
781 int err;
782
783 if (count == 0)
784 return 0;
785
786 inode_lock(inode); /* protect against multiple concurrent
787 * readers on this file */
788 again:
789 spin_lock(&queue_lock);
790 /* need to find next request */
791 while (rp->q.list.next != &cd->queue &&
792 list_entry(rp->q.list.next, struct cache_queue, list)
793 ->reader) {
794 struct list_head *next = rp->q.list.next;
795 list_move(&rp->q.list, next);
796 }
797 if (rp->q.list.next == &cd->queue) {
798 spin_unlock(&queue_lock);
799 inode_unlock(inode);
800 WARN_ON_ONCE(rp->offset);
801 return 0;
802 }
803 rq = container_of(rp->q.list.next, struct cache_request, q.list);
804 WARN_ON_ONCE(rq->q.reader);
805 if (rp->offset == 0)
806 rq->readers++;
807 spin_unlock(&queue_lock);
808
809 if (rq->len == 0) {
810 err = cache_request(cd, rq);
811 if (err < 0)
812 goto out;
813 rq->len = err;
814 }
815
816 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
817 err = -EAGAIN;
818 spin_lock(&queue_lock);
819 list_move(&rp->q.list, &rq->q.list);
820 spin_unlock(&queue_lock);
821 } else {
822 if (rp->offset + count > rq->len)
823 count = rq->len - rp->offset;
824 err = -EFAULT;
825 if (copy_to_user(buf, rq->buf + rp->offset, count))
826 goto out;
827 rp->offset += count;
828 if (rp->offset >= rq->len) {
829 rp->offset = 0;
830 spin_lock(&queue_lock);
831 list_move(&rp->q.list, &rq->q.list);
832 spin_unlock(&queue_lock);
833 }
834 err = 0;
835 }
836 out:
837 if (rp->offset == 0) {
838 /* need to release rq */
839 spin_lock(&queue_lock);
840 rq->readers--;
841 if (rq->readers == 0 &&
842 !test_bit(CACHE_PENDING, &rq->item->flags)) {
843 list_del(&rq->q.list);
844 spin_unlock(&queue_lock);
845 cache_put(rq->item, cd);
846 kfree(rq->buf);
847 kfree(rq);
848 } else
849 spin_unlock(&queue_lock);
850 }
851 if (err == -EAGAIN)
852 goto again;
853 inode_unlock(inode);
854 return err ? err : count;
855}
856
857static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
858 size_t count, struct cache_detail *cd)
859{
860 ssize_t ret;
861
862 if (count == 0)
863 return -EINVAL;
864 if (copy_from_user(kaddr, buf, count))
865 return -EFAULT;
866 kaddr[count] = '\0';
867 ret = cd->cache_parse(cd, kaddr, count);
868 if (!ret)
869 ret = count;
870 return ret;
871}
872
873static ssize_t cache_slow_downcall(const char __user *buf,
874 size_t count, struct cache_detail *cd)
875{
876 static char write_buf[8192]; /* protected by queue_io_mutex */
877 ssize_t ret = -EINVAL;
878
879 if (count >= sizeof(write_buf))
880 goto out;
881 mutex_lock(&queue_io_mutex);
882 ret = cache_do_downcall(write_buf, buf, count, cd);
883 mutex_unlock(&queue_io_mutex);
884out:
885 return ret;
886}
887
888static ssize_t cache_downcall(struct address_space *mapping,
889 const char __user *buf,
890 size_t count, struct cache_detail *cd)
891{
892 struct page *page;
893 char *kaddr;
894 ssize_t ret = -ENOMEM;
895
896 if (count >= PAGE_SIZE)
897 goto out_slow;
898
899 page = find_or_create_page(mapping, 0, GFP_KERNEL);
900 if (!page)
901 goto out_slow;
902
903 kaddr = kmap(page);
904 ret = cache_do_downcall(kaddr, buf, count, cd);
905 kunmap(page);
906 unlock_page(page);
907 put_page(page);
908 return ret;
909out_slow:
910 return cache_slow_downcall(buf, count, cd);
911}
912
913static ssize_t cache_write(struct file *filp, const char __user *buf,
914 size_t count, loff_t *ppos,
915 struct cache_detail *cd)
916{
917 struct address_space *mapping = filp->f_mapping;
918 struct inode *inode = file_inode(filp);
919 ssize_t ret = -EINVAL;
920
921 if (!cd->cache_parse)
922 goto out;
923
924 inode_lock(inode);
925 ret = cache_downcall(mapping, buf, count, cd);
926 inode_unlock(inode);
927out:
928 return ret;
929}
930
931static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
932
933static __poll_t cache_poll(struct file *filp, poll_table *wait,
934 struct cache_detail *cd)
935{
936 __poll_t mask;
937 struct cache_reader *rp = filp->private_data;
938 struct cache_queue *cq;
939
940 poll_wait(filp, &queue_wait, wait);
941
942 /* alway allow write */
943 mask = EPOLLOUT | EPOLLWRNORM;
944
945 if (!rp)
946 return mask;
947
948 spin_lock(&queue_lock);
949
950 for (cq= &rp->q; &cq->list != &cd->queue;
951 cq = list_entry(cq->list.next, struct cache_queue, list))
952 if (!cq->reader) {
953 mask |= EPOLLIN | EPOLLRDNORM;
954 break;
955 }
956 spin_unlock(&queue_lock);
957 return mask;
958}
959
960static int cache_ioctl(struct inode *ino, struct file *filp,
961 unsigned int cmd, unsigned long arg,
962 struct cache_detail *cd)
963{
964 int len = 0;
965 struct cache_reader *rp = filp->private_data;
966 struct cache_queue *cq;
967
968 if (cmd != FIONREAD || !rp)
969 return -EINVAL;
970
971 spin_lock(&queue_lock);
972
973 /* only find the length remaining in current request,
974 * or the length of the next request
975 */
976 for (cq= &rp->q; &cq->list != &cd->queue;
977 cq = list_entry(cq->list.next, struct cache_queue, list))
978 if (!cq->reader) {
979 struct cache_request *cr =
980 container_of(cq, struct cache_request, q);
981 len = cr->len - rp->offset;
982 break;
983 }
984 spin_unlock(&queue_lock);
985
986 return put_user(len, (int __user *)arg);
987}
988
989static int cache_open(struct inode *inode, struct file *filp,
990 struct cache_detail *cd)
991{
992 struct cache_reader *rp = NULL;
993
994 if (!cd || !try_module_get(cd->owner))
995 return -EACCES;
996 nonseekable_open(inode, filp);
997 if (filp->f_mode & FMODE_READ) {
998 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
999 if (!rp) {
1000 module_put(cd->owner);
1001 return -ENOMEM;
1002 }
1003 rp->offset = 0;
1004 rp->q.reader = 1;
1005 atomic_inc(&cd->readers);
1006 spin_lock(&queue_lock);
1007 list_add(&rp->q.list, &cd->queue);
1008 spin_unlock(&queue_lock);
1009 }
1010 filp->private_data = rp;
1011 return 0;
1012}
1013
1014static int cache_release(struct inode *inode, struct file *filp,
1015 struct cache_detail *cd)
1016{
1017 struct cache_reader *rp = filp->private_data;
1018
1019 if (rp) {
1020 spin_lock(&queue_lock);
1021 if (rp->offset) {
1022 struct cache_queue *cq;
1023 for (cq= &rp->q; &cq->list != &cd->queue;
1024 cq = list_entry(cq->list.next, struct cache_queue, list))
1025 if (!cq->reader) {
1026 container_of(cq, struct cache_request, q)
1027 ->readers--;
1028 break;
1029 }
1030 rp->offset = 0;
1031 }
1032 list_del(&rp->q.list);
1033 spin_unlock(&queue_lock);
1034
1035 filp->private_data = NULL;
1036 kfree(rp);
1037
1038 cd->last_close = seconds_since_boot();
1039 atomic_dec(&cd->readers);
1040 }
1041 module_put(cd->owner);
1042 return 0;
1043}
1044
1045
1046
1047static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1048{
1049 struct cache_queue *cq, *tmp;
1050 struct cache_request *cr;
1051 struct list_head dequeued;
1052
1053 INIT_LIST_HEAD(&dequeued);
1054 spin_lock(&queue_lock);
1055 list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1056 if (!cq->reader) {
1057 cr = container_of(cq, struct cache_request, q);
1058 if (cr->item != ch)
1059 continue;
1060 if (test_bit(CACHE_PENDING, &ch->flags))
1061 /* Lost a race and it is pending again */
1062 break;
1063 if (cr->readers != 0)
1064 continue;
1065 list_move(&cr->q.list, &dequeued);
1066 }
1067 spin_unlock(&queue_lock);
1068 while (!list_empty(&dequeued)) {
1069 cr = list_entry(dequeued.next, struct cache_request, q.list);
1070 list_del(&cr->q.list);
1071 cache_put(cr->item, detail);
1072 kfree(cr->buf);
1073 kfree(cr);
1074 }
1075}
1076
1077/*
1078 * Support routines for text-based upcalls.
1079 * Fields are separated by spaces.
1080 * Fields are either mangled to quote space tab newline slosh with slosh
1081 * or a hexified with a leading \x
1082 * Record is terminated with newline.
1083 *
1084 */
1085
1086void qword_add(char **bpp, int *lp, char *str)
1087{
1088 char *bp = *bpp;
1089 int len = *lp;
1090 int ret;
1091
1092 if (len < 0) return;
1093
1094 ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1095 if (ret >= len) {
1096 bp += len;
1097 len = -1;
1098 } else {
1099 bp += ret;
1100 len -= ret;
1101 *bp++ = ' ';
1102 len--;
1103 }
1104 *bpp = bp;
1105 *lp = len;
1106}
1107EXPORT_SYMBOL_GPL(qword_add);
1108
1109void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1110{
1111 char *bp = *bpp;
1112 int len = *lp;
1113
1114 if (len < 0) return;
1115
1116 if (len > 2) {
1117 *bp++ = '\\';
1118 *bp++ = 'x';
1119 len -= 2;
1120 while (blen && len >= 2) {
1121 bp = hex_byte_pack(bp, *buf++);
1122 len -= 2;
1123 blen--;
1124 }
1125 }
1126 if (blen || len<1) len = -1;
1127 else {
1128 *bp++ = ' ';
1129 len--;
1130 }
1131 *bpp = bp;
1132 *lp = len;
1133}
1134EXPORT_SYMBOL_GPL(qword_addhex);
1135
1136static void warn_no_listener(struct cache_detail *detail)
1137{
1138 if (detail->last_warn != detail->last_close) {
1139 detail->last_warn = detail->last_close;
1140 if (detail->warn_no_listener)
1141 detail->warn_no_listener(detail, detail->last_close != 0);
1142 }
1143}
1144
1145static bool cache_listeners_exist(struct cache_detail *detail)
1146{
1147 if (atomic_read(&detail->readers))
1148 return true;
1149 if (detail->last_close == 0)
1150 /* This cache was never opened */
1151 return false;
1152 if (detail->last_close < seconds_since_boot() - 30)
1153 /*
1154 * We allow for the possibility that someone might
1155 * restart a userspace daemon without restarting the
1156 * server; but after 30 seconds, we give up.
1157 */
1158 return false;
1159 return true;
1160}
1161
1162/*
1163 * register an upcall request to user-space and queue it up for read() by the
1164 * upcall daemon.
1165 *
1166 * Each request is at most one page long.
1167 */
1168int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1169{
1170
1171 char *buf;
1172 struct cache_request *crq;
1173 int ret = 0;
1174
1175 if (!detail->cache_request)
1176 return -EINVAL;
1177
1178 if (!cache_listeners_exist(detail)) {
1179 warn_no_listener(detail);
1180 return -EINVAL;
1181 }
1182 if (test_bit(CACHE_CLEANED, &h->flags))
1183 /* Too late to make an upcall */
1184 return -EAGAIN;
1185
1186 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1187 if (!buf)
1188 return -EAGAIN;
1189
1190 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1191 if (!crq) {
1192 kfree(buf);
1193 return -EAGAIN;
1194 }
1195
1196 crq->q.reader = 0;
1197 crq->buf = buf;
1198 crq->len = 0;
1199 crq->readers = 0;
1200 spin_lock(&queue_lock);
1201 if (test_bit(CACHE_PENDING, &h->flags)) {
1202 crq->item = cache_get(h);
1203 list_add_tail(&crq->q.list, &detail->queue);
1204 } else
1205 /* Lost a race, no longer PENDING, so don't enqueue */
1206 ret = -EAGAIN;
1207 spin_unlock(&queue_lock);
1208 wake_up(&queue_wait);
1209 if (ret == -EAGAIN) {
1210 kfree(buf);
1211 kfree(crq);
1212 }
1213 return ret;
1214}
1215EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1216
1217/*
1218 * parse a message from user-space and pass it
1219 * to an appropriate cache
1220 * Messages are, like requests, separated into fields by
1221 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1222 *
1223 * Message is
1224 * reply cachename expiry key ... content....
1225 *
1226 * key and content are both parsed by cache
1227 */
1228
1229int qword_get(char **bpp, char *dest, int bufsize)
1230{
1231 /* return bytes copied, or -1 on error */
1232 char *bp = *bpp;
1233 int len = 0;
1234
1235 while (*bp == ' ') bp++;
1236
1237 if (bp[0] == '\\' && bp[1] == 'x') {
1238 /* HEX STRING */
1239 bp += 2;
1240 while (len < bufsize - 1) {
1241 int h, l;
1242
1243 h = hex_to_bin(bp[0]);
1244 if (h < 0)
1245 break;
1246
1247 l = hex_to_bin(bp[1]);
1248 if (l < 0)
1249 break;
1250
1251 *dest++ = (h << 4) | l;
1252 bp += 2;
1253 len++;
1254 }
1255 } else {
1256 /* text with \nnn octal quoting */
1257 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1258 if (*bp == '\\' &&
1259 isodigit(bp[1]) && (bp[1] <= '3') &&
1260 isodigit(bp[2]) &&
1261 isodigit(bp[3])) {
1262 int byte = (*++bp -'0');
1263 bp++;
1264 byte = (byte << 3) | (*bp++ - '0');
1265 byte = (byte << 3) | (*bp++ - '0');
1266 *dest++ = byte;
1267 len++;
1268 } else {
1269 *dest++ = *bp++;
1270 len++;
1271 }
1272 }
1273 }
1274
1275 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1276 return -1;
1277 while (*bp == ' ') bp++;
1278 *bpp = bp;
1279 *dest = '\0';
1280 return len;
1281}
1282EXPORT_SYMBOL_GPL(qword_get);
1283
1284
1285/*
1286 * support /proc/net/rpc/$CACHENAME/content
1287 * as a seqfile.
1288 * We call ->cache_show passing NULL for the item to
1289 * get a header, then pass each real item in the cache
1290 */
1291
1292void *cache_seq_start(struct seq_file *m, loff_t *pos)
1293 __acquires(cd->hash_lock)
1294{
1295 loff_t n = *pos;
1296 unsigned int hash, entry;
1297 struct cache_head *ch;
1298 struct cache_detail *cd = m->private;
1299
1300 read_lock(&cd->hash_lock);
1301 if (!n--)
1302 return SEQ_START_TOKEN;
1303 hash = n >> 32;
1304 entry = n & ((1LL<<32) - 1);
1305
1306 hlist_for_each_entry(ch, &cd->hash_table[hash], cache_list)
1307 if (!entry--)
1308 return ch;
1309 n &= ~((1LL<<32) - 1);
1310 do {
1311 hash++;
1312 n += 1LL<<32;
1313 } while(hash < cd->hash_size &&
1314 hlist_empty(&cd->hash_table[hash]));
1315 if (hash >= cd->hash_size)
1316 return NULL;
1317 *pos = n+1;
1318 return hlist_entry_safe(cd->hash_table[hash].first,
1319 struct cache_head, cache_list);
1320}
1321EXPORT_SYMBOL_GPL(cache_seq_start);
1322
1323void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1324{
1325 struct cache_head *ch = p;
1326 int hash = (*pos >> 32);
1327 struct cache_detail *cd = m->private;
1328
1329 if (p == SEQ_START_TOKEN)
1330 hash = 0;
1331 else if (ch->cache_list.next == NULL) {
1332 hash++;
1333 *pos += 1LL<<32;
1334 } else {
1335 ++*pos;
1336 return hlist_entry_safe(ch->cache_list.next,
1337 struct cache_head, cache_list);
1338 }
1339 *pos &= ~((1LL<<32) - 1);
1340 while (hash < cd->hash_size &&
1341 hlist_empty(&cd->hash_table[hash])) {
1342 hash++;
1343 *pos += 1LL<<32;
1344 }
1345 if (hash >= cd->hash_size)
1346 return NULL;
1347 ++*pos;
1348 return hlist_entry_safe(cd->hash_table[hash].first,
1349 struct cache_head, cache_list);
1350}
1351EXPORT_SYMBOL_GPL(cache_seq_next);
1352
1353void cache_seq_stop(struct seq_file *m, void *p)
1354 __releases(cd->hash_lock)
1355{
1356 struct cache_detail *cd = m->private;
1357 read_unlock(&cd->hash_lock);
1358}
1359EXPORT_SYMBOL_GPL(cache_seq_stop);
1360
1361static int c_show(struct seq_file *m, void *p)
1362{
1363 struct cache_head *cp = p;
1364 struct cache_detail *cd = m->private;
1365
1366 if (p == SEQ_START_TOKEN)
1367 return cd->cache_show(m, cd, NULL);
1368
1369 ifdebug(CACHE)
1370 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1371 convert_to_wallclock(cp->expiry_time),
1372 kref_read(&cp->ref), cp->flags);
1373 cache_get(cp);
1374 if (cache_check(cd, cp, NULL))
1375 /* cache_check does a cache_put on failure */
1376 seq_printf(m, "# ");
1377 else {
1378 if (cache_is_expired(cd, cp))
1379 seq_printf(m, "# ");
1380 cache_put(cp, cd);
1381 }
1382
1383 return cd->cache_show(m, cd, cp);
1384}
1385
1386static const struct seq_operations cache_content_op = {
1387 .start = cache_seq_start,
1388 .next = cache_seq_next,
1389 .stop = cache_seq_stop,
1390 .show = c_show,
1391};
1392
1393static int content_open(struct inode *inode, struct file *file,
1394 struct cache_detail *cd)
1395{
1396 struct seq_file *seq;
1397 int err;
1398
1399 if (!cd || !try_module_get(cd->owner))
1400 return -EACCES;
1401
1402 err = seq_open(file, &cache_content_op);
1403 if (err) {
1404 module_put(cd->owner);
1405 return err;
1406 }
1407
1408 seq = file->private_data;
1409 seq->private = cd;
1410 return 0;
1411}
1412
1413static int content_release(struct inode *inode, struct file *file,
1414 struct cache_detail *cd)
1415{
1416 int ret = seq_release(inode, file);
1417 module_put(cd->owner);
1418 return ret;
1419}
1420
1421static int open_flush(struct inode *inode, struct file *file,
1422 struct cache_detail *cd)
1423{
1424 if (!cd || !try_module_get(cd->owner))
1425 return -EACCES;
1426 return nonseekable_open(inode, file);
1427}
1428
1429static int release_flush(struct inode *inode, struct file *file,
1430 struct cache_detail *cd)
1431{
1432 module_put(cd->owner);
1433 return 0;
1434}
1435
1436static ssize_t read_flush(struct file *file, char __user *buf,
1437 size_t count, loff_t *ppos,
1438 struct cache_detail *cd)
1439{
1440 char tbuf[22];
1441 size_t len;
1442
1443 len = snprintf(tbuf, sizeof(tbuf), "%lu\n",
1444 convert_to_wallclock(cd->flush_time));
1445 return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1446}
1447
1448static ssize_t write_flush(struct file *file, const char __user *buf,
1449 size_t count, loff_t *ppos,
1450 struct cache_detail *cd)
1451{
1452 char tbuf[20];
1453 char *ep;
1454 time_t now;
1455
1456 if (*ppos || count > sizeof(tbuf)-1)
1457 return -EINVAL;
1458 if (copy_from_user(tbuf, buf, count))
1459 return -EFAULT;
1460 tbuf[count] = 0;
1461 simple_strtoul(tbuf, &ep, 0);
1462 if (*ep && *ep != '\n')
1463 return -EINVAL;
1464 /* Note that while we check that 'buf' holds a valid number,
1465 * we always ignore the value and just flush everything.
1466 * Making use of the number leads to races.
1467 */
1468
1469 now = seconds_since_boot();
1470 /* Always flush everything, so behave like cache_purge()
1471 * Do this by advancing flush_time to the current time,
1472 * or by one second if it has already reached the current time.
1473 * Newly added cache entries will always have ->last_refresh greater
1474 * that ->flush_time, so they don't get flushed prematurely.
1475 */
1476
1477 if (cd->flush_time >= now)
1478 now = cd->flush_time + 1;
1479
1480 cd->flush_time = now;
1481 cd->nextcheck = now;
1482 cache_flush();
1483
1484 *ppos += count;
1485 return count;
1486}
1487
1488static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1489 size_t count, loff_t *ppos)
1490{
1491 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1492
1493 return cache_read(filp, buf, count, ppos, cd);
1494}
1495
1496static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1497 size_t count, loff_t *ppos)
1498{
1499 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1500
1501 return cache_write(filp, buf, count, ppos, cd);
1502}
1503
1504static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1505{
1506 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1507
1508 return cache_poll(filp, wait, cd);
1509}
1510
1511static long cache_ioctl_procfs(struct file *filp,
1512 unsigned int cmd, unsigned long arg)
1513{
1514 struct inode *inode = file_inode(filp);
1515 struct cache_detail *cd = PDE_DATA(inode);
1516
1517 return cache_ioctl(inode, filp, cmd, arg, cd);
1518}
1519
1520static int cache_open_procfs(struct inode *inode, struct file *filp)
1521{
1522 struct cache_detail *cd = PDE_DATA(inode);
1523
1524 return cache_open(inode, filp, cd);
1525}
1526
1527static int cache_release_procfs(struct inode *inode, struct file *filp)
1528{
1529 struct cache_detail *cd = PDE_DATA(inode);
1530
1531 return cache_release(inode, filp, cd);
1532}
1533
1534static const struct file_operations cache_file_operations_procfs = {
1535 .owner = THIS_MODULE,
1536 .llseek = no_llseek,
1537 .read = cache_read_procfs,
1538 .write = cache_write_procfs,
1539 .poll = cache_poll_procfs,
1540 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1541 .open = cache_open_procfs,
1542 .release = cache_release_procfs,
1543};
1544
1545static int content_open_procfs(struct inode *inode, struct file *filp)
1546{
1547 struct cache_detail *cd = PDE_DATA(inode);
1548
1549 return content_open(inode, filp, cd);
1550}
1551
1552static int content_release_procfs(struct inode *inode, struct file *filp)
1553{
1554 struct cache_detail *cd = PDE_DATA(inode);
1555
1556 return content_release(inode, filp, cd);
1557}
1558
1559static const struct file_operations content_file_operations_procfs = {
1560 .open = content_open_procfs,
1561 .read = seq_read,
1562 .llseek = seq_lseek,
1563 .release = content_release_procfs,
1564};
1565
1566static int open_flush_procfs(struct inode *inode, struct file *filp)
1567{
1568 struct cache_detail *cd = PDE_DATA(inode);
1569
1570 return open_flush(inode, filp, cd);
1571}
1572
1573static int release_flush_procfs(struct inode *inode, struct file *filp)
1574{
1575 struct cache_detail *cd = PDE_DATA(inode);
1576
1577 return release_flush(inode, filp, cd);
1578}
1579
1580static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1581 size_t count, loff_t *ppos)
1582{
1583 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1584
1585 return read_flush(filp, buf, count, ppos, cd);
1586}
1587
1588static ssize_t write_flush_procfs(struct file *filp,
1589 const char __user *buf,
1590 size_t count, loff_t *ppos)
1591{
1592 struct cache_detail *cd = PDE_DATA(file_inode(filp));
1593
1594 return write_flush(filp, buf, count, ppos, cd);
1595}
1596
1597static const struct file_operations cache_flush_operations_procfs = {
1598 .open = open_flush_procfs,
1599 .read = read_flush_procfs,
1600 .write = write_flush_procfs,
1601 .release = release_flush_procfs,
1602 .llseek = no_llseek,
1603};
1604
1605static void remove_cache_proc_entries(struct cache_detail *cd)
1606{
1607 if (cd->procfs) {
1608 proc_remove(cd->procfs);
1609 cd->procfs = NULL;
1610 }
1611}
1612
1613#ifdef CONFIG_PROC_FS
1614static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1615{
1616 struct proc_dir_entry *p;
1617 struct sunrpc_net *sn;
1618
1619 sn = net_generic(net, sunrpc_net_id);
1620 cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1621 if (cd->procfs == NULL)
1622 goto out_nomem;
1623
1624 p = proc_create_data("flush", S_IFREG | 0600,
1625 cd->procfs, &cache_flush_operations_procfs, cd);
1626 if (p == NULL)
1627 goto out_nomem;
1628
1629 if (cd->cache_request || cd->cache_parse) {
1630 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1631 &cache_file_operations_procfs, cd);
1632 if (p == NULL)
1633 goto out_nomem;
1634 }
1635 if (cd->cache_show) {
1636 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1637 &content_file_operations_procfs, cd);
1638 if (p == NULL)
1639 goto out_nomem;
1640 }
1641 return 0;
1642out_nomem:
1643 remove_cache_proc_entries(cd);
1644 return -ENOMEM;
1645}
1646#else /* CONFIG_PROC_FS */
1647static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1648{
1649 return 0;
1650}
1651#endif
1652
1653void __init cache_initialize(void)
1654{
1655 INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1656}
1657
1658int cache_register_net(struct cache_detail *cd, struct net *net)
1659{
1660 int ret;
1661
1662 sunrpc_init_cache_detail(cd);
1663 ret = create_cache_proc_entries(cd, net);
1664 if (ret)
1665 sunrpc_destroy_cache_detail(cd);
1666 return ret;
1667}
1668EXPORT_SYMBOL_GPL(cache_register_net);
1669
1670void cache_unregister_net(struct cache_detail *cd, struct net *net)
1671{
1672 remove_cache_proc_entries(cd);
1673 sunrpc_destroy_cache_detail(cd);
1674}
1675EXPORT_SYMBOL_GPL(cache_unregister_net);
1676
1677struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1678{
1679 struct cache_detail *cd;
1680 int i;
1681
1682 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1683 if (cd == NULL)
1684 return ERR_PTR(-ENOMEM);
1685
1686 cd->hash_table = kzalloc(cd->hash_size * sizeof(struct hlist_head),
1687 GFP_KERNEL);
1688 if (cd->hash_table == NULL) {
1689 kfree(cd);
1690 return ERR_PTR(-ENOMEM);
1691 }
1692
1693 for (i = 0; i < cd->hash_size; i++)
1694 INIT_HLIST_HEAD(&cd->hash_table[i]);
1695 cd->net = net;
1696 return cd;
1697}
1698EXPORT_SYMBOL_GPL(cache_create_net);
1699
1700void cache_destroy_net(struct cache_detail *cd, struct net *net)
1701{
1702 kfree(cd->hash_table);
1703 kfree(cd);
1704}
1705EXPORT_SYMBOL_GPL(cache_destroy_net);
1706
1707static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1708 size_t count, loff_t *ppos)
1709{
1710 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1711
1712 return cache_read(filp, buf, count, ppos, cd);
1713}
1714
1715static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1716 size_t count, loff_t *ppos)
1717{
1718 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1719
1720 return cache_write(filp, buf, count, ppos, cd);
1721}
1722
1723static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1724{
1725 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1726
1727 return cache_poll(filp, wait, cd);
1728}
1729
1730static long cache_ioctl_pipefs(struct file *filp,
1731 unsigned int cmd, unsigned long arg)
1732{
1733 struct inode *inode = file_inode(filp);
1734 struct cache_detail *cd = RPC_I(inode)->private;
1735
1736 return cache_ioctl(inode, filp, cmd, arg, cd);
1737}
1738
1739static int cache_open_pipefs(struct inode *inode, struct file *filp)
1740{
1741 struct cache_detail *cd = RPC_I(inode)->private;
1742
1743 return cache_open(inode, filp, cd);
1744}
1745
1746static int cache_release_pipefs(struct inode *inode, struct file *filp)
1747{
1748 struct cache_detail *cd = RPC_I(inode)->private;
1749
1750 return cache_release(inode, filp, cd);
1751}
1752
1753const struct file_operations cache_file_operations_pipefs = {
1754 .owner = THIS_MODULE,
1755 .llseek = no_llseek,
1756 .read = cache_read_pipefs,
1757 .write = cache_write_pipefs,
1758 .poll = cache_poll_pipefs,
1759 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1760 .open = cache_open_pipefs,
1761 .release = cache_release_pipefs,
1762};
1763
1764static int content_open_pipefs(struct inode *inode, struct file *filp)
1765{
1766 struct cache_detail *cd = RPC_I(inode)->private;
1767
1768 return content_open(inode, filp, cd);
1769}
1770
1771static int content_release_pipefs(struct inode *inode, struct file *filp)
1772{
1773 struct cache_detail *cd = RPC_I(inode)->private;
1774
1775 return content_release(inode, filp, cd);
1776}
1777
1778const struct file_operations content_file_operations_pipefs = {
1779 .open = content_open_pipefs,
1780 .read = seq_read,
1781 .llseek = seq_lseek,
1782 .release = content_release_pipefs,
1783};
1784
1785static int open_flush_pipefs(struct inode *inode, struct file *filp)
1786{
1787 struct cache_detail *cd = RPC_I(inode)->private;
1788
1789 return open_flush(inode, filp, cd);
1790}
1791
1792static int release_flush_pipefs(struct inode *inode, struct file *filp)
1793{
1794 struct cache_detail *cd = RPC_I(inode)->private;
1795
1796 return release_flush(inode, filp, cd);
1797}
1798
1799static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1800 size_t count, loff_t *ppos)
1801{
1802 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1803
1804 return read_flush(filp, buf, count, ppos, cd);
1805}
1806
1807static ssize_t write_flush_pipefs(struct file *filp,
1808 const char __user *buf,
1809 size_t count, loff_t *ppos)
1810{
1811 struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1812
1813 return write_flush(filp, buf, count, ppos, cd);
1814}
1815
1816const struct file_operations cache_flush_operations_pipefs = {
1817 .open = open_flush_pipefs,
1818 .read = read_flush_pipefs,
1819 .write = write_flush_pipefs,
1820 .release = release_flush_pipefs,
1821 .llseek = no_llseek,
1822};
1823
1824int sunrpc_cache_register_pipefs(struct dentry *parent,
1825 const char *name, umode_t umode,
1826 struct cache_detail *cd)
1827{
1828 struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1829 if (IS_ERR(dir))
1830 return PTR_ERR(dir);
1831 cd->pipefs = dir;
1832 return 0;
1833}
1834EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1835
1836void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1837{
1838 if (cd->pipefs) {
1839 rpc_remove_cache_dir(cd->pipefs);
1840 cd->pipefs = NULL;
1841 }
1842}
1843EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1844
1845void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1846{
1847 write_lock(&cd->hash_lock);
1848 if (!hlist_unhashed(&h->cache_list)){
1849 hlist_del_init(&h->cache_list);
1850 cd->entries--;
1851 write_unlock(&cd->hash_lock);
1852 cache_put(h, cd);
1853 } else
1854 write_unlock(&cd->hash_lock);
1855}
1856EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);