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