Merge branch 'for-3.3' of git://linux-nfs.org/~bfields/linux
[linux-2.6.git] / net / sunrpc / cache.c
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
40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
41 static void cache_revisit_request(struct cache_head *item);
42
43 static 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
53 static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
54 {
55         return  (h->expiry_time < seconds_since_boot()) ||
56                 (detail->flush_time > h->last_refresh);
57 }
58
59 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
60                                        struct cache_head *key, int hash)
61 {
62         struct cache_head **head,  **hp;
63         struct cache_head *new = NULL, *freeme = NULL;
64
65         head = &detail->hash_table[hash];
66
67         read_lock(&detail->hash_lock);
68
69         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
70                 struct cache_head *tmp = *hp;
71                 if (detail->match(tmp, key)) {
72                         if (cache_is_expired(detail, tmp))
73                                 /* This entry is expired, we will discard it. */
74                                 break;
75                         cache_get(tmp);
76                         read_unlock(&detail->hash_lock);
77                         return tmp;
78                 }
79         }
80         read_unlock(&detail->hash_lock);
81         /* Didn't find anything, insert an empty entry */
82
83         new = detail->alloc();
84         if (!new)
85                 return NULL;
86         /* must fully initialise 'new', else
87          * we might get lose if we need to
88          * cache_put it soon.
89          */
90         cache_init(new);
91         detail->init(new, key);
92
93         write_lock(&detail->hash_lock);
94
95         /* check if entry appeared while we slept */
96         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
97                 struct cache_head *tmp = *hp;
98                 if (detail->match(tmp, key)) {
99                         if (cache_is_expired(detail, tmp)) {
100                                 *hp = tmp->next;
101                                 tmp->next = NULL;
102                                 detail->entries --;
103                                 freeme = tmp;
104                                 break;
105                         }
106                         cache_get(tmp);
107                         write_unlock(&detail->hash_lock);
108                         cache_put(new, detail);
109                         return tmp;
110                 }
111         }
112         new->next = *head;
113         *head = new;
114         detail->entries++;
115         cache_get(new);
116         write_unlock(&detail->hash_lock);
117
118         if (freeme)
119                 cache_put(freeme, detail);
120         return new;
121 }
122 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
123
124
125 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
126
127 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
128 {
129         head->expiry_time = expiry;
130         head->last_refresh = seconds_since_boot();
131         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
132         set_bit(CACHE_VALID, &head->flags);
133 }
134
135 static 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
144 struct 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 **head;
152         struct cache_head *tmp;
153
154         if (!test_bit(CACHE_VALID, &old->flags)) {
155                 write_lock(&detail->hash_lock);
156                 if (!test_bit(CACHE_VALID, &old->flags)) {
157                         if (test_bit(CACHE_NEGATIVE, &new->flags))
158                                 set_bit(CACHE_NEGATIVE, &old->flags);
159                         else
160                                 detail->update(old, new);
161                         cache_fresh_locked(old, new->expiry_time);
162                         write_unlock(&detail->hash_lock);
163                         cache_fresh_unlocked(old, detail);
164                         return old;
165                 }
166                 write_unlock(&detail->hash_lock);
167         }
168         /* We need to insert a new entry */
169         tmp = detail->alloc();
170         if (!tmp) {
171                 cache_put(old, detail);
172                 return NULL;
173         }
174         cache_init(tmp);
175         detail->init(tmp, old);
176         head = &detail->hash_table[hash];
177
178         write_lock(&detail->hash_lock);
179         if (test_bit(CACHE_NEGATIVE, &new->flags))
180                 set_bit(CACHE_NEGATIVE, &tmp->flags);
181         else
182                 detail->update(tmp, new);
183         tmp->next = *head;
184         *head = tmp;
185         detail->entries++;
186         cache_get(tmp);
187         cache_fresh_locked(tmp, new->expiry_time);
188         cache_fresh_locked(old, 0);
189         write_unlock(&detail->hash_lock);
190         cache_fresh_unlocked(tmp, detail);
191         cache_fresh_unlocked(old, detail);
192         cache_put(old, detail);
193         return tmp;
194 }
195 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
196
197 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
198 {
199         if (!cd->cache_upcall)
200                 return -EINVAL;
201         return cd->cache_upcall(cd, h);
202 }
203
204 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
205 {
206         if (!test_bit(CACHE_VALID, &h->flags))
207                 return -EAGAIN;
208         else {
209                 /* entry is valid */
210                 if (test_bit(CACHE_NEGATIVE, &h->flags))
211                         return -ENOENT;
212                 else {
213                         /*
214                          * In combination with write barrier in
215                          * sunrpc_cache_update, ensures that anyone
216                          * using the cache entry after this sees the
217                          * updated contents:
218                          */
219                         smp_rmb();
220                         return 0;
221                 }
222         }
223 }
224
225 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
226 {
227         int rv;
228
229         write_lock(&detail->hash_lock);
230         rv = cache_is_valid(detail, h);
231         if (rv != -EAGAIN) {
232                 write_unlock(&detail->hash_lock);
233                 return rv;
234         }
235         set_bit(CACHE_NEGATIVE, &h->flags);
236         cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
237         write_unlock(&detail->hash_lock);
238         cache_fresh_unlocked(h, detail);
239         return -ENOENT;
240 }
241
242 /*
243  * This is the generic cache management routine for all
244  * the authentication caches.
245  * It checks the currency of a cache item and will (later)
246  * initiate an upcall to fill it if needed.
247  *
248  *
249  * Returns 0 if the cache_head can be used, or cache_puts it and returns
250  * -EAGAIN if upcall is pending and request has been queued
251  * -ETIMEDOUT if upcall failed or request could not be queue or
252  *           upcall completed but item is still invalid (implying that
253  *           the cache item has been replaced with a newer one).
254  * -ENOENT if cache entry was negative
255  */
256 int cache_check(struct cache_detail *detail,
257                     struct cache_head *h, struct cache_req *rqstp)
258 {
259         int rv;
260         long refresh_age, age;
261
262         /* First decide return status as best we can */
263         rv = cache_is_valid(detail, h);
264
265         /* now see if we want to start an upcall */
266         refresh_age = (h->expiry_time - h->last_refresh);
267         age = seconds_since_boot() - h->last_refresh;
268
269         if (rqstp == NULL) {
270                 if (rv == -EAGAIN)
271                         rv = -ENOENT;
272         } else if (rv == -EAGAIN || age > refresh_age/2) {
273                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
274                                 refresh_age, age);
275                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
276                         switch (cache_make_upcall(detail, h)) {
277                         case -EINVAL:
278                                 clear_bit(CACHE_PENDING, &h->flags);
279                                 cache_revisit_request(h);
280                                 rv = try_to_negate_entry(detail, h);
281                                 break;
282                         case -EAGAIN:
283                                 clear_bit(CACHE_PENDING, &h->flags);
284                                 cache_revisit_request(h);
285                                 break;
286                         }
287                 }
288         }
289
290         if (rv == -EAGAIN) {
291                 if (!cache_defer_req(rqstp, h)) {
292                         /*
293                          * Request was not deferred; handle it as best
294                          * we can ourselves:
295                          */
296                         rv = cache_is_valid(detail, h);
297                         if (rv == -EAGAIN)
298                                 rv = -ETIMEDOUT;
299                 }
300         }
301         if (rv)
302                 cache_put(h, detail);
303         return rv;
304 }
305 EXPORT_SYMBOL_GPL(cache_check);
306
307 /*
308  * caches need to be periodically cleaned.
309  * For this we maintain a list of cache_detail and
310  * a current pointer into that list and into the table
311  * for that entry.
312  *
313  * Each time clean_cache is called it finds the next non-empty entry
314  * in the current table and walks the list in that entry
315  * looking for entries that can be removed.
316  *
317  * An entry gets removed if:
318  * - The expiry is before current time
319  * - The last_refresh time is before the flush_time for that cache
320  *
321  * later we might drop old entries with non-NEVER expiry if that table
322  * is getting 'full' for some definition of 'full'
323  *
324  * The question of "how often to scan a table" is an interesting one
325  * and is answered in part by the use of the "nextcheck" field in the
326  * cache_detail.
327  * When a scan of a table begins, the nextcheck field is set to a time
328  * that is well into the future.
329  * While scanning, if an expiry time is found that is earlier than the
330  * current nextcheck time, nextcheck is set to that expiry time.
331  * If the flush_time is ever set to a time earlier than the nextcheck
332  * time, the nextcheck time is then set to that flush_time.
333  *
334  * A table is then only scanned if the current time is at least
335  * the nextcheck time.
336  *
337  */
338
339 static LIST_HEAD(cache_list);
340 static DEFINE_SPINLOCK(cache_list_lock);
341 static struct cache_detail *current_detail;
342 static int current_index;
343
344 static void do_cache_clean(struct work_struct *work);
345 static struct delayed_work cache_cleaner;
346
347 static void sunrpc_init_cache_detail(struct cache_detail *cd)
348 {
349         rwlock_init(&cd->hash_lock);
350         INIT_LIST_HEAD(&cd->queue);
351         spin_lock(&cache_list_lock);
352         cd->nextcheck = 0;
353         cd->entries = 0;
354         atomic_set(&cd->readers, 0);
355         cd->last_close = 0;
356         cd->last_warn = -1;
357         list_add(&cd->others, &cache_list);
358         spin_unlock(&cache_list_lock);
359
360         /* start the cleaning process */
361         schedule_delayed_work(&cache_cleaner, 0);
362 }
363
364 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
365 {
366         cache_purge(cd);
367         spin_lock(&cache_list_lock);
368         write_lock(&cd->hash_lock);
369         if (cd->entries || atomic_read(&cd->inuse)) {
370                 write_unlock(&cd->hash_lock);
371                 spin_unlock(&cache_list_lock);
372                 goto out;
373         }
374         if (current_detail == cd)
375                 current_detail = NULL;
376         list_del_init(&cd->others);
377         write_unlock(&cd->hash_lock);
378         spin_unlock(&cache_list_lock);
379         if (list_empty(&cache_list)) {
380                 /* module must be being unloaded so its safe to kill the worker */
381                 cancel_delayed_work_sync(&cache_cleaner);
382         }
383         return;
384 out:
385         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
386 }
387
388 /* clean cache tries to find something to clean
389  * and cleans it.
390  * It returns 1 if it cleaned something,
391  *            0 if it didn't find anything this time
392  *           -1 if it fell off the end of the list.
393  */
394 static int cache_clean(void)
395 {
396         int rv = 0;
397         struct list_head *next;
398
399         spin_lock(&cache_list_lock);
400
401         /* find a suitable table if we don't already have one */
402         while (current_detail == NULL ||
403             current_index >= current_detail->hash_size) {
404                 if (current_detail)
405                         next = current_detail->others.next;
406                 else
407                         next = cache_list.next;
408                 if (next == &cache_list) {
409                         current_detail = NULL;
410                         spin_unlock(&cache_list_lock);
411                         return -1;
412                 }
413                 current_detail = list_entry(next, struct cache_detail, others);
414                 if (current_detail->nextcheck > seconds_since_boot())
415                         current_index = current_detail->hash_size;
416                 else {
417                         current_index = 0;
418                         current_detail->nextcheck = seconds_since_boot()+30*60;
419                 }
420         }
421
422         /* find a non-empty bucket in the table */
423         while (current_detail &&
424                current_index < current_detail->hash_size &&
425                current_detail->hash_table[current_index] == NULL)
426                 current_index++;
427
428         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
429
430         if (current_detail && current_index < current_detail->hash_size) {
431                 struct cache_head *ch, **cp;
432                 struct cache_detail *d;
433
434                 write_lock(&current_detail->hash_lock);
435
436                 /* Ok, now to clean this strand */
437
438                 cp = & current_detail->hash_table[current_index];
439                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
440                         if (current_detail->nextcheck > ch->expiry_time)
441                                 current_detail->nextcheck = ch->expiry_time+1;
442                         if (!cache_is_expired(current_detail, ch))
443                                 continue;
444
445                         *cp = ch->next;
446                         ch->next = NULL;
447                         current_detail->entries--;
448                         rv = 1;
449                         break;
450                 }
451
452                 write_unlock(&current_detail->hash_lock);
453                 d = current_detail;
454                 if (!ch)
455                         current_index ++;
456                 spin_unlock(&cache_list_lock);
457                 if (ch) {
458                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
459                                 cache_dequeue(current_detail, ch);
460                         cache_revisit_request(ch);
461                         cache_put(ch, d);
462                 }
463         } else
464                 spin_unlock(&cache_list_lock);
465
466         return rv;
467 }
468
469 /*
470  * We want to regularly clean the cache, so we need to schedule some work ...
471  */
472 static void do_cache_clean(struct work_struct *work)
473 {
474         int delay = 5;
475         if (cache_clean() == -1)
476                 delay = round_jiffies_relative(30*HZ);
477
478         if (list_empty(&cache_list))
479                 delay = 0;
480
481         if (delay)
482                 schedule_delayed_work(&cache_cleaner, delay);
483 }
484
485
486 /*
487  * Clean all caches promptly.  This just calls cache_clean
488  * repeatedly until we are sure that every cache has had a chance to
489  * be fully cleaned
490  */
491 void cache_flush(void)
492 {
493         while (cache_clean() != -1)
494                 cond_resched();
495         while (cache_clean() != -1)
496                 cond_resched();
497 }
498 EXPORT_SYMBOL_GPL(cache_flush);
499
500 void cache_purge(struct cache_detail *detail)
501 {
502         detail->flush_time = LONG_MAX;
503         detail->nextcheck = seconds_since_boot();
504         cache_flush();
505         detail->flush_time = 1;
506 }
507 EXPORT_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
530 static DEFINE_SPINLOCK(cache_defer_lock);
531 static LIST_HEAD(cache_defer_list);
532 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
533 static int cache_defer_cnt;
534
535 static 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
544 static 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
552 static 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
572 struct thread_deferred_req {
573         struct cache_deferred_req handle;
574         struct completion completion;
575 };
576
577 static 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
584 static 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
616 static 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 (net_random() & 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. */
644 static 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
667 static void cache_revisit_request(struct cache_head *item)
668 {
669         struct cache_deferred_req *dreq;
670         struct list_head pending;
671         struct hlist_node *lp, *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, lp, 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
692 void 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
732 static DEFINE_SPINLOCK(queue_lock);
733 static DEFINE_MUTEX(queue_io_mutex);
734
735 struct cache_queue {
736         struct list_head        list;
737         int                     reader; /* if 0, then request */
738 };
739 struct cache_request {
740         struct cache_queue      q;
741         struct cache_head       *item;
742         char                    * buf;
743         int                     len;
744         int                     readers;
745 };
746 struct cache_reader {
747         struct cache_queue      q;
748         int                     offset; /* if non-0, we have a refcnt on next request */
749 };
750
751 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
752                           loff_t *ppos, struct cache_detail *cd)
753 {
754         struct cache_reader *rp = filp->private_data;
755         struct cache_request *rq;
756         struct inode *inode = filp->f_path.dentry->d_inode;
757         int err;
758
759         if (count == 0)
760                 return 0;
761
762         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
763                               * readers on this file */
764  again:
765         spin_lock(&queue_lock);
766         /* need to find next request */
767         while (rp->q.list.next != &cd->queue &&
768                list_entry(rp->q.list.next, struct cache_queue, list)
769                ->reader) {
770                 struct list_head *next = rp->q.list.next;
771                 list_move(&rp->q.list, next);
772         }
773         if (rp->q.list.next == &cd->queue) {
774                 spin_unlock(&queue_lock);
775                 mutex_unlock(&inode->i_mutex);
776                 BUG_ON(rp->offset);
777                 return 0;
778         }
779         rq = container_of(rp->q.list.next, struct cache_request, q.list);
780         BUG_ON(rq->q.reader);
781         if (rp->offset == 0)
782                 rq->readers++;
783         spin_unlock(&queue_lock);
784
785         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
786                 err = -EAGAIN;
787                 spin_lock(&queue_lock);
788                 list_move(&rp->q.list, &rq->q.list);
789                 spin_unlock(&queue_lock);
790         } else {
791                 if (rp->offset + count > rq->len)
792                         count = rq->len - rp->offset;
793                 err = -EFAULT;
794                 if (copy_to_user(buf, rq->buf + rp->offset, count))
795                         goto out;
796                 rp->offset += count;
797                 if (rp->offset >= rq->len) {
798                         rp->offset = 0;
799                         spin_lock(&queue_lock);
800                         list_move(&rp->q.list, &rq->q.list);
801                         spin_unlock(&queue_lock);
802                 }
803                 err = 0;
804         }
805  out:
806         if (rp->offset == 0) {
807                 /* need to release rq */
808                 spin_lock(&queue_lock);
809                 rq->readers--;
810                 if (rq->readers == 0 &&
811                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
812                         list_del(&rq->q.list);
813                         spin_unlock(&queue_lock);
814                         cache_put(rq->item, cd);
815                         kfree(rq->buf);
816                         kfree(rq);
817                 } else
818                         spin_unlock(&queue_lock);
819         }
820         if (err == -EAGAIN)
821                 goto again;
822         mutex_unlock(&inode->i_mutex);
823         return err ? err :  count;
824 }
825
826 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
827                                  size_t count, struct cache_detail *cd)
828 {
829         ssize_t ret;
830
831         if (copy_from_user(kaddr, buf, count))
832                 return -EFAULT;
833         kaddr[count] = '\0';
834         ret = cd->cache_parse(cd, kaddr, count);
835         if (!ret)
836                 ret = count;
837         return ret;
838 }
839
840 static ssize_t cache_slow_downcall(const char __user *buf,
841                                    size_t count, struct cache_detail *cd)
842 {
843         static char write_buf[8192]; /* protected by queue_io_mutex */
844         ssize_t ret = -EINVAL;
845
846         if (count >= sizeof(write_buf))
847                 goto out;
848         mutex_lock(&queue_io_mutex);
849         ret = cache_do_downcall(write_buf, buf, count, cd);
850         mutex_unlock(&queue_io_mutex);
851 out:
852         return ret;
853 }
854
855 static ssize_t cache_downcall(struct address_space *mapping,
856                               const char __user *buf,
857                               size_t count, struct cache_detail *cd)
858 {
859         struct page *page;
860         char *kaddr;
861         ssize_t ret = -ENOMEM;
862
863         if (count >= PAGE_CACHE_SIZE)
864                 goto out_slow;
865
866         page = find_or_create_page(mapping, 0, GFP_KERNEL);
867         if (!page)
868                 goto out_slow;
869
870         kaddr = kmap(page);
871         ret = cache_do_downcall(kaddr, buf, count, cd);
872         kunmap(page);
873         unlock_page(page);
874         page_cache_release(page);
875         return ret;
876 out_slow:
877         return cache_slow_downcall(buf, count, cd);
878 }
879
880 static ssize_t cache_write(struct file *filp, const char __user *buf,
881                            size_t count, loff_t *ppos,
882                            struct cache_detail *cd)
883 {
884         struct address_space *mapping = filp->f_mapping;
885         struct inode *inode = filp->f_path.dentry->d_inode;
886         ssize_t ret = -EINVAL;
887
888         if (!cd->cache_parse)
889                 goto out;
890
891         mutex_lock(&inode->i_mutex);
892         ret = cache_downcall(mapping, buf, count, cd);
893         mutex_unlock(&inode->i_mutex);
894 out:
895         return ret;
896 }
897
898 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
899
900 static unsigned int cache_poll(struct file *filp, poll_table *wait,
901                                struct cache_detail *cd)
902 {
903         unsigned int mask;
904         struct cache_reader *rp = filp->private_data;
905         struct cache_queue *cq;
906
907         poll_wait(filp, &queue_wait, wait);
908
909         /* alway allow write */
910         mask = POLL_OUT | POLLWRNORM;
911
912         if (!rp)
913                 return mask;
914
915         spin_lock(&queue_lock);
916
917         for (cq= &rp->q; &cq->list != &cd->queue;
918              cq = list_entry(cq->list.next, struct cache_queue, list))
919                 if (!cq->reader) {
920                         mask |= POLLIN | POLLRDNORM;
921                         break;
922                 }
923         spin_unlock(&queue_lock);
924         return mask;
925 }
926
927 static int cache_ioctl(struct inode *ino, struct file *filp,
928                        unsigned int cmd, unsigned long arg,
929                        struct cache_detail *cd)
930 {
931         int len = 0;
932         struct cache_reader *rp = filp->private_data;
933         struct cache_queue *cq;
934
935         if (cmd != FIONREAD || !rp)
936                 return -EINVAL;
937
938         spin_lock(&queue_lock);
939
940         /* only find the length remaining in current request,
941          * or the length of the next request
942          */
943         for (cq= &rp->q; &cq->list != &cd->queue;
944              cq = list_entry(cq->list.next, struct cache_queue, list))
945                 if (!cq->reader) {
946                         struct cache_request *cr =
947                                 container_of(cq, struct cache_request, q);
948                         len = cr->len - rp->offset;
949                         break;
950                 }
951         spin_unlock(&queue_lock);
952
953         return put_user(len, (int __user *)arg);
954 }
955
956 static int cache_open(struct inode *inode, struct file *filp,
957                       struct cache_detail *cd)
958 {
959         struct cache_reader *rp = NULL;
960
961         if (!cd || !try_module_get(cd->owner))
962                 return -EACCES;
963         nonseekable_open(inode, filp);
964         if (filp->f_mode & FMODE_READ) {
965                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
966                 if (!rp)
967                         return -ENOMEM;
968                 rp->offset = 0;
969                 rp->q.reader = 1;
970                 atomic_inc(&cd->readers);
971                 spin_lock(&queue_lock);
972                 list_add(&rp->q.list, &cd->queue);
973                 spin_unlock(&queue_lock);
974         }
975         filp->private_data = rp;
976         return 0;
977 }
978
979 static int cache_release(struct inode *inode, struct file *filp,
980                          struct cache_detail *cd)
981 {
982         struct cache_reader *rp = filp->private_data;
983
984         if (rp) {
985                 spin_lock(&queue_lock);
986                 if (rp->offset) {
987                         struct cache_queue *cq;
988                         for (cq= &rp->q; &cq->list != &cd->queue;
989                              cq = list_entry(cq->list.next, struct cache_queue, list))
990                                 if (!cq->reader) {
991                                         container_of(cq, struct cache_request, q)
992                                                 ->readers--;
993                                         break;
994                                 }
995                         rp->offset = 0;
996                 }
997                 list_del(&rp->q.list);
998                 spin_unlock(&queue_lock);
999
1000                 filp->private_data = NULL;
1001                 kfree(rp);
1002
1003                 cd->last_close = seconds_since_boot();
1004                 atomic_dec(&cd->readers);
1005         }
1006         module_put(cd->owner);
1007         return 0;
1008 }
1009
1010
1011
1012 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1013 {
1014         struct cache_queue *cq;
1015         spin_lock(&queue_lock);
1016         list_for_each_entry(cq, &detail->queue, list)
1017                 if (!cq->reader) {
1018                         struct cache_request *cr = container_of(cq, struct cache_request, q);
1019                         if (cr->item != ch)
1020                                 continue;
1021                         if (cr->readers != 0)
1022                                 continue;
1023                         list_del(&cr->q.list);
1024                         spin_unlock(&queue_lock);
1025                         cache_put(cr->item, detail);
1026                         kfree(cr->buf);
1027                         kfree(cr);
1028                         return;
1029                 }
1030         spin_unlock(&queue_lock);
1031 }
1032
1033 /*
1034  * Support routines for text-based upcalls.
1035  * Fields are separated by spaces.
1036  * Fields are either mangled to quote space tab newline slosh with slosh
1037  * or a hexified with a leading \x
1038  * Record is terminated with newline.
1039  *
1040  */
1041
1042 void qword_add(char **bpp, int *lp, char *str)
1043 {
1044         char *bp = *bpp;
1045         int len = *lp;
1046         char c;
1047
1048         if (len < 0) return;
1049
1050         while ((c=*str++) && len)
1051                 switch(c) {
1052                 case ' ':
1053                 case '\t':
1054                 case '\n':
1055                 case '\\':
1056                         if (len >= 4) {
1057                                 *bp++ = '\\';
1058                                 *bp++ = '0' + ((c & 0300)>>6);
1059                                 *bp++ = '0' + ((c & 0070)>>3);
1060                                 *bp++ = '0' + ((c & 0007)>>0);
1061                         }
1062                         len -= 4;
1063                         break;
1064                 default:
1065                         *bp++ = c;
1066                         len--;
1067                 }
1068         if (c || len <1) len = -1;
1069         else {
1070                 *bp++ = ' ';
1071                 len--;
1072         }
1073         *bpp = bp;
1074         *lp = len;
1075 }
1076 EXPORT_SYMBOL_GPL(qword_add);
1077
1078 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1079 {
1080         char *bp = *bpp;
1081         int len = *lp;
1082
1083         if (len < 0) return;
1084
1085         if (len > 2) {
1086                 *bp++ = '\\';
1087                 *bp++ = 'x';
1088                 len -= 2;
1089                 while (blen && len >= 2) {
1090                         unsigned char c = *buf++;
1091                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1092                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1093                         len -= 2;
1094                         blen--;
1095                 }
1096         }
1097         if (blen || len<1) len = -1;
1098         else {
1099                 *bp++ = ' ';
1100                 len--;
1101         }
1102         *bpp = bp;
1103         *lp = len;
1104 }
1105 EXPORT_SYMBOL_GPL(qword_addhex);
1106
1107 static void warn_no_listener(struct cache_detail *detail)
1108 {
1109         if (detail->last_warn != detail->last_close) {
1110                 detail->last_warn = detail->last_close;
1111                 if (detail->warn_no_listener)
1112                         detail->warn_no_listener(detail, detail->last_close != 0);
1113         }
1114 }
1115
1116 static bool cache_listeners_exist(struct cache_detail *detail)
1117 {
1118         if (atomic_read(&detail->readers))
1119                 return true;
1120         if (detail->last_close == 0)
1121                 /* This cache was never opened */
1122                 return false;
1123         if (detail->last_close < seconds_since_boot() - 30)
1124                 /*
1125                  * We allow for the possibility that someone might
1126                  * restart a userspace daemon without restarting the
1127                  * server; but after 30 seconds, we give up.
1128                  */
1129                  return false;
1130         return true;
1131 }
1132
1133 /*
1134  * register an upcall request to user-space and queue it up for read() by the
1135  * upcall daemon.
1136  *
1137  * Each request is at most one page long.
1138  */
1139 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1140                 void (*cache_request)(struct cache_detail *,
1141                                       struct cache_head *,
1142                                       char **,
1143                                       int *))
1144 {
1145
1146         char *buf;
1147         struct cache_request *crq;
1148         char *bp;
1149         int len;
1150
1151         if (!cache_listeners_exist(detail)) {
1152                 warn_no_listener(detail);
1153                 return -EINVAL;
1154         }
1155
1156         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1157         if (!buf)
1158                 return -EAGAIN;
1159
1160         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1161         if (!crq) {
1162                 kfree(buf);
1163                 return -EAGAIN;
1164         }
1165
1166         bp = buf; len = PAGE_SIZE;
1167
1168         cache_request(detail, h, &bp, &len);
1169
1170         if (len < 0) {
1171                 kfree(buf);
1172                 kfree(crq);
1173                 return -EAGAIN;
1174         }
1175         crq->q.reader = 0;
1176         crq->item = cache_get(h);
1177         crq->buf = buf;
1178         crq->len = PAGE_SIZE - len;
1179         crq->readers = 0;
1180         spin_lock(&queue_lock);
1181         list_add_tail(&crq->q.list, &detail->queue);
1182         spin_unlock(&queue_lock);
1183         wake_up(&queue_wait);
1184         return 0;
1185 }
1186 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1187
1188 /*
1189  * parse a message from user-space and pass it
1190  * to an appropriate cache
1191  * Messages are, like requests, separated into fields by
1192  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1193  *
1194  * Message is
1195  *   reply cachename expiry key ... content....
1196  *
1197  * key and content are both parsed by cache
1198  */
1199
1200 #define isodigit(c) (isdigit(c) && c <= '7')
1201 int qword_get(char **bpp, char *dest, int bufsize)
1202 {
1203         /* return bytes copied, or -1 on error */
1204         char *bp = *bpp;
1205         int len = 0;
1206
1207         while (*bp == ' ') bp++;
1208
1209         if (bp[0] == '\\' && bp[1] == 'x') {
1210                 /* HEX STRING */
1211                 bp += 2;
1212                 while (len < bufsize) {
1213                         int h, l;
1214
1215                         h = hex_to_bin(bp[0]);
1216                         if (h < 0)
1217                                 break;
1218
1219                         l = hex_to_bin(bp[1]);
1220                         if (l < 0)
1221                                 break;
1222
1223                         *dest++ = (h << 4) | l;
1224                         bp += 2;
1225                         len++;
1226                 }
1227         } else {
1228                 /* text with \nnn octal quoting */
1229                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1230                         if (*bp == '\\' &&
1231                             isodigit(bp[1]) && (bp[1] <= '3') &&
1232                             isodigit(bp[2]) &&
1233                             isodigit(bp[3])) {
1234                                 int byte = (*++bp -'0');
1235                                 bp++;
1236                                 byte = (byte << 3) | (*bp++ - '0');
1237                                 byte = (byte << 3) | (*bp++ - '0');
1238                                 *dest++ = byte;
1239                                 len++;
1240                         } else {
1241                                 *dest++ = *bp++;
1242                                 len++;
1243                         }
1244                 }
1245         }
1246
1247         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1248                 return -1;
1249         while (*bp == ' ') bp++;
1250         *bpp = bp;
1251         *dest = '\0';
1252         return len;
1253 }
1254 EXPORT_SYMBOL_GPL(qword_get);
1255
1256
1257 /*
1258  * support /proc/sunrpc/cache/$CACHENAME/content
1259  * as a seqfile.
1260  * We call ->cache_show passing NULL for the item to
1261  * get a header, then pass each real item in the cache
1262  */
1263
1264 struct handle {
1265         struct cache_detail *cd;
1266 };
1267
1268 static void *c_start(struct seq_file *m, loff_t *pos)
1269         __acquires(cd->hash_lock)
1270 {
1271         loff_t n = *pos;
1272         unsigned hash, entry;
1273         struct cache_head *ch;
1274         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1275
1276
1277         read_lock(&cd->hash_lock);
1278         if (!n--)
1279                 return SEQ_START_TOKEN;
1280         hash = n >> 32;
1281         entry = n & ((1LL<<32) - 1);
1282
1283         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1284                 if (!entry--)
1285                         return ch;
1286         n &= ~((1LL<<32) - 1);
1287         do {
1288                 hash++;
1289                 n += 1LL<<32;
1290         } while(hash < cd->hash_size &&
1291                 cd->hash_table[hash]==NULL);
1292         if (hash >= cd->hash_size)
1293                 return NULL;
1294         *pos = n+1;
1295         return cd->hash_table[hash];
1296 }
1297
1298 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1299 {
1300         struct cache_head *ch = p;
1301         int hash = (*pos >> 32);
1302         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1303
1304         if (p == SEQ_START_TOKEN)
1305                 hash = 0;
1306         else if (ch->next == NULL) {
1307                 hash++;
1308                 *pos += 1LL<<32;
1309         } else {
1310                 ++*pos;
1311                 return ch->next;
1312         }
1313         *pos &= ~((1LL<<32) - 1);
1314         while (hash < cd->hash_size &&
1315                cd->hash_table[hash] == NULL) {
1316                 hash++;
1317                 *pos += 1LL<<32;
1318         }
1319         if (hash >= cd->hash_size)
1320                 return NULL;
1321         ++*pos;
1322         return cd->hash_table[hash];
1323 }
1324
1325 static void c_stop(struct seq_file *m, void *p)
1326         __releases(cd->hash_lock)
1327 {
1328         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1329         read_unlock(&cd->hash_lock);
1330 }
1331
1332 static int c_show(struct seq_file *m, void *p)
1333 {
1334         struct cache_head *cp = p;
1335         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1336
1337         if (p == SEQ_START_TOKEN)
1338                 return cd->cache_show(m, cd, NULL);
1339
1340         ifdebug(CACHE)
1341                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1342                            convert_to_wallclock(cp->expiry_time),
1343                            atomic_read(&cp->ref.refcount), cp->flags);
1344         cache_get(cp);
1345         if (cache_check(cd, cp, NULL))
1346                 /* cache_check does a cache_put on failure */
1347                 seq_printf(m, "# ");
1348         else
1349                 cache_put(cp, cd);
1350
1351         return cd->cache_show(m, cd, cp);
1352 }
1353
1354 static const struct seq_operations cache_content_op = {
1355         .start  = c_start,
1356         .next   = c_next,
1357         .stop   = c_stop,
1358         .show   = c_show,
1359 };
1360
1361 static int content_open(struct inode *inode, struct file *file,
1362                         struct cache_detail *cd)
1363 {
1364         struct handle *han;
1365
1366         if (!cd || !try_module_get(cd->owner))
1367                 return -EACCES;
1368         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1369         if (han == NULL) {
1370                 module_put(cd->owner);
1371                 return -ENOMEM;
1372         }
1373
1374         han->cd = cd;
1375         return 0;
1376 }
1377
1378 static int content_release(struct inode *inode, struct file *file,
1379                 struct cache_detail *cd)
1380 {
1381         int ret = seq_release_private(inode, file);
1382         module_put(cd->owner);
1383         return ret;
1384 }
1385
1386 static int open_flush(struct inode *inode, struct file *file,
1387                         struct cache_detail *cd)
1388 {
1389         if (!cd || !try_module_get(cd->owner))
1390                 return -EACCES;
1391         return nonseekable_open(inode, file);
1392 }
1393
1394 static int release_flush(struct inode *inode, struct file *file,
1395                         struct cache_detail *cd)
1396 {
1397         module_put(cd->owner);
1398         return 0;
1399 }
1400
1401 static ssize_t read_flush(struct file *file, char __user *buf,
1402                           size_t count, loff_t *ppos,
1403                           struct cache_detail *cd)
1404 {
1405         char tbuf[20];
1406         unsigned long p = *ppos;
1407         size_t len;
1408
1409         sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1410         len = strlen(tbuf);
1411         if (p >= len)
1412                 return 0;
1413         len -= p;
1414         if (len > count)
1415                 len = count;
1416         if (copy_to_user(buf, (void*)(tbuf+p), len))
1417                 return -EFAULT;
1418         *ppos += len;
1419         return len;
1420 }
1421
1422 static ssize_t write_flush(struct file *file, const char __user *buf,
1423                            size_t count, loff_t *ppos,
1424                            struct cache_detail *cd)
1425 {
1426         char tbuf[20];
1427         char *bp, *ep;
1428
1429         if (*ppos || count > sizeof(tbuf)-1)
1430                 return -EINVAL;
1431         if (copy_from_user(tbuf, buf, count))
1432                 return -EFAULT;
1433         tbuf[count] = 0;
1434         simple_strtoul(tbuf, &ep, 0);
1435         if (*ep && *ep != '\n')
1436                 return -EINVAL;
1437
1438         bp = tbuf;
1439         cd->flush_time = get_expiry(&bp);
1440         cd->nextcheck = seconds_since_boot();
1441         cache_flush();
1442
1443         *ppos += count;
1444         return count;
1445 }
1446
1447 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1448                                  size_t count, loff_t *ppos)
1449 {
1450         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1451
1452         return cache_read(filp, buf, count, ppos, cd);
1453 }
1454
1455 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1456                                   size_t count, loff_t *ppos)
1457 {
1458         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1459
1460         return cache_write(filp, buf, count, ppos, cd);
1461 }
1462
1463 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1464 {
1465         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1466
1467         return cache_poll(filp, wait, cd);
1468 }
1469
1470 static long cache_ioctl_procfs(struct file *filp,
1471                                unsigned int cmd, unsigned long arg)
1472 {
1473         struct inode *inode = filp->f_path.dentry->d_inode;
1474         struct cache_detail *cd = PDE(inode)->data;
1475
1476         return cache_ioctl(inode, filp, cmd, arg, cd);
1477 }
1478
1479 static int cache_open_procfs(struct inode *inode, struct file *filp)
1480 {
1481         struct cache_detail *cd = PDE(inode)->data;
1482
1483         return cache_open(inode, filp, cd);
1484 }
1485
1486 static int cache_release_procfs(struct inode *inode, struct file *filp)
1487 {
1488         struct cache_detail *cd = PDE(inode)->data;
1489
1490         return cache_release(inode, filp, cd);
1491 }
1492
1493 static const struct file_operations cache_file_operations_procfs = {
1494         .owner          = THIS_MODULE,
1495         .llseek         = no_llseek,
1496         .read           = cache_read_procfs,
1497         .write          = cache_write_procfs,
1498         .poll           = cache_poll_procfs,
1499         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1500         .open           = cache_open_procfs,
1501         .release        = cache_release_procfs,
1502 };
1503
1504 static int content_open_procfs(struct inode *inode, struct file *filp)
1505 {
1506         struct cache_detail *cd = PDE(inode)->data;
1507
1508         return content_open(inode, filp, cd);
1509 }
1510
1511 static int content_release_procfs(struct inode *inode, struct file *filp)
1512 {
1513         struct cache_detail *cd = PDE(inode)->data;
1514
1515         return content_release(inode, filp, cd);
1516 }
1517
1518 static const struct file_operations content_file_operations_procfs = {
1519         .open           = content_open_procfs,
1520         .read           = seq_read,
1521         .llseek         = seq_lseek,
1522         .release        = content_release_procfs,
1523 };
1524
1525 static int open_flush_procfs(struct inode *inode, struct file *filp)
1526 {
1527         struct cache_detail *cd = PDE(inode)->data;
1528
1529         return open_flush(inode, filp, cd);
1530 }
1531
1532 static int release_flush_procfs(struct inode *inode, struct file *filp)
1533 {
1534         struct cache_detail *cd = PDE(inode)->data;
1535
1536         return release_flush(inode, filp, cd);
1537 }
1538
1539 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1540                             size_t count, loff_t *ppos)
1541 {
1542         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1543
1544         return read_flush(filp, buf, count, ppos, cd);
1545 }
1546
1547 static ssize_t write_flush_procfs(struct file *filp,
1548                                   const char __user *buf,
1549                                   size_t count, loff_t *ppos)
1550 {
1551         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1552
1553         return write_flush(filp, buf, count, ppos, cd);
1554 }
1555
1556 static const struct file_operations cache_flush_operations_procfs = {
1557         .open           = open_flush_procfs,
1558         .read           = read_flush_procfs,
1559         .write          = write_flush_procfs,
1560         .release        = release_flush_procfs,
1561         .llseek         = no_llseek,
1562 };
1563
1564 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1565 {
1566         struct sunrpc_net *sn;
1567
1568         if (cd->u.procfs.proc_ent == NULL)
1569                 return;
1570         if (cd->u.procfs.flush_ent)
1571                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1572         if (cd->u.procfs.channel_ent)
1573                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1574         if (cd->u.procfs.content_ent)
1575                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1576         cd->u.procfs.proc_ent = NULL;
1577         sn = net_generic(net, sunrpc_net_id);
1578         remove_proc_entry(cd->name, sn->proc_net_rpc);
1579 }
1580
1581 #ifdef CONFIG_PROC_FS
1582 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1583 {
1584         struct proc_dir_entry *p;
1585         struct sunrpc_net *sn;
1586
1587         sn = net_generic(net, sunrpc_net_id);
1588         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1589         if (cd->u.procfs.proc_ent == NULL)
1590                 goto out_nomem;
1591         cd->u.procfs.channel_ent = NULL;
1592         cd->u.procfs.content_ent = NULL;
1593
1594         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1595                              cd->u.procfs.proc_ent,
1596                              &cache_flush_operations_procfs, cd);
1597         cd->u.procfs.flush_ent = p;
1598         if (p == NULL)
1599                 goto out_nomem;
1600
1601         if (cd->cache_upcall || cd->cache_parse) {
1602                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1603                                      cd->u.procfs.proc_ent,
1604                                      &cache_file_operations_procfs, cd);
1605                 cd->u.procfs.channel_ent = p;
1606                 if (p == NULL)
1607                         goto out_nomem;
1608         }
1609         if (cd->cache_show) {
1610                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1611                                 cd->u.procfs.proc_ent,
1612                                 &content_file_operations_procfs, cd);
1613                 cd->u.procfs.content_ent = p;
1614                 if (p == NULL)
1615                         goto out_nomem;
1616         }
1617         return 0;
1618 out_nomem:
1619         remove_cache_proc_entries(cd, net);
1620         return -ENOMEM;
1621 }
1622 #else /* CONFIG_PROC_FS */
1623 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1624 {
1625         return 0;
1626 }
1627 #endif
1628
1629 void __init cache_initialize(void)
1630 {
1631         INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1632 }
1633
1634 int cache_register_net(struct cache_detail *cd, struct net *net)
1635 {
1636         int ret;
1637
1638         sunrpc_init_cache_detail(cd);
1639         ret = create_cache_proc_entries(cd, net);
1640         if (ret)
1641                 sunrpc_destroy_cache_detail(cd);
1642         return ret;
1643 }
1644 EXPORT_SYMBOL_GPL(cache_register_net);
1645
1646 int cache_register(struct cache_detail *cd)
1647 {
1648         return cache_register_net(cd, &init_net);
1649 }
1650 EXPORT_SYMBOL_GPL(cache_register);
1651
1652 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1653 {
1654         remove_cache_proc_entries(cd, net);
1655         sunrpc_destroy_cache_detail(cd);
1656 }
1657 EXPORT_SYMBOL_GPL(cache_unregister_net);
1658
1659 void cache_unregister(struct cache_detail *cd)
1660 {
1661         cache_unregister_net(cd, &init_net);
1662 }
1663 EXPORT_SYMBOL_GPL(cache_unregister);
1664
1665 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1666                                  size_t count, loff_t *ppos)
1667 {
1668         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1669
1670         return cache_read(filp, buf, count, ppos, cd);
1671 }
1672
1673 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1674                                   size_t count, loff_t *ppos)
1675 {
1676         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1677
1678         return cache_write(filp, buf, count, ppos, cd);
1679 }
1680
1681 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1682 {
1683         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1684
1685         return cache_poll(filp, wait, cd);
1686 }
1687
1688 static long cache_ioctl_pipefs(struct file *filp,
1689                               unsigned int cmd, unsigned long arg)
1690 {
1691         struct inode *inode = filp->f_dentry->d_inode;
1692         struct cache_detail *cd = RPC_I(inode)->private;
1693
1694         return cache_ioctl(inode, filp, cmd, arg, cd);
1695 }
1696
1697 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1698 {
1699         struct cache_detail *cd = RPC_I(inode)->private;
1700
1701         return cache_open(inode, filp, cd);
1702 }
1703
1704 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1705 {
1706         struct cache_detail *cd = RPC_I(inode)->private;
1707
1708         return cache_release(inode, filp, cd);
1709 }
1710
1711 const struct file_operations cache_file_operations_pipefs = {
1712         .owner          = THIS_MODULE,
1713         .llseek         = no_llseek,
1714         .read           = cache_read_pipefs,
1715         .write          = cache_write_pipefs,
1716         .poll           = cache_poll_pipefs,
1717         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1718         .open           = cache_open_pipefs,
1719         .release        = cache_release_pipefs,
1720 };
1721
1722 static int content_open_pipefs(struct inode *inode, struct file *filp)
1723 {
1724         struct cache_detail *cd = RPC_I(inode)->private;
1725
1726         return content_open(inode, filp, cd);
1727 }
1728
1729 static int content_release_pipefs(struct inode *inode, struct file *filp)
1730 {
1731         struct cache_detail *cd = RPC_I(inode)->private;
1732
1733         return content_release(inode, filp, cd);
1734 }
1735
1736 const struct file_operations content_file_operations_pipefs = {
1737         .open           = content_open_pipefs,
1738         .read           = seq_read,
1739         .llseek         = seq_lseek,
1740         .release        = content_release_pipefs,
1741 };
1742
1743 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1744 {
1745         struct cache_detail *cd = RPC_I(inode)->private;
1746
1747         return open_flush(inode, filp, cd);
1748 }
1749
1750 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1751 {
1752         struct cache_detail *cd = RPC_I(inode)->private;
1753
1754         return release_flush(inode, filp, cd);
1755 }
1756
1757 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1758                             size_t count, loff_t *ppos)
1759 {
1760         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1761
1762         return read_flush(filp, buf, count, ppos, cd);
1763 }
1764
1765 static ssize_t write_flush_pipefs(struct file *filp,
1766                                   const char __user *buf,
1767                                   size_t count, loff_t *ppos)
1768 {
1769         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1770
1771         return write_flush(filp, buf, count, ppos, cd);
1772 }
1773
1774 const struct file_operations cache_flush_operations_pipefs = {
1775         .open           = open_flush_pipefs,
1776         .read           = read_flush_pipefs,
1777         .write          = write_flush_pipefs,
1778         .release        = release_flush_pipefs,
1779         .llseek         = no_llseek,
1780 };
1781
1782 int sunrpc_cache_register_pipefs(struct dentry *parent,
1783                                  const char *name, umode_t umode,
1784                                  struct cache_detail *cd)
1785 {
1786         struct qstr q;
1787         struct dentry *dir;
1788         int ret = 0;
1789
1790         sunrpc_init_cache_detail(cd);
1791         q.name = name;
1792         q.len = strlen(name);
1793         q.hash = full_name_hash(q.name, q.len);
1794         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1795         if (!IS_ERR(dir))
1796                 cd->u.pipefs.dir = dir;
1797         else {
1798                 sunrpc_destroy_cache_detail(cd);
1799                 ret = PTR_ERR(dir);
1800         }
1801         return ret;
1802 }
1803 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1804
1805 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1806 {
1807         rpc_remove_cache_dir(cd->u.pipefs.dir);
1808         cd->u.pipefs.dir = NULL;
1809         sunrpc_destroy_cache_detail(cd);
1810 }
1811 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1812