Merge tag 'jfs-3.10' of git://github.com/kleikamp/linux-shaggy
[linux-3.10.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 cd->cache_upcall(cd, h);
201         return sunrpc_cache_pipe_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 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 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
364
365 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
366 {
367         cache_purge(cd);
368         spin_lock(&cache_list_lock);
369         write_lock(&cd->hash_lock);
370         if (cd->entries || atomic_read(&cd->inuse)) {
371                 write_unlock(&cd->hash_lock);
372                 spin_unlock(&cache_list_lock);
373                 goto out;
374         }
375         if (current_detail == cd)
376                 current_detail = NULL;
377         list_del_init(&cd->others);
378         write_unlock(&cd->hash_lock);
379         spin_unlock(&cache_list_lock);
380         if (list_empty(&cache_list)) {
381                 /* module must be being unloaded so its safe to kill the worker */
382                 cancel_delayed_work_sync(&cache_cleaner);
383         }
384         return;
385 out:
386         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
387 }
388 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
389
390 /* clean cache tries to find something to clean
391  * and cleans it.
392  * It returns 1 if it cleaned something,
393  *            0 if it didn't find anything this time
394  *           -1 if it fell off the end of the list.
395  */
396 static int cache_clean(void)
397 {
398         int rv = 0;
399         struct list_head *next;
400
401         spin_lock(&cache_list_lock);
402
403         /* find a suitable table if we don't already have one */
404         while (current_detail == NULL ||
405             current_index >= current_detail->hash_size) {
406                 if (current_detail)
407                         next = current_detail->others.next;
408                 else
409                         next = cache_list.next;
410                 if (next == &cache_list) {
411                         current_detail = NULL;
412                         spin_unlock(&cache_list_lock);
413                         return -1;
414                 }
415                 current_detail = list_entry(next, struct cache_detail, others);
416                 if (current_detail->nextcheck > seconds_since_boot())
417                         current_index = current_detail->hash_size;
418                 else {
419                         current_index = 0;
420                         current_detail->nextcheck = seconds_since_boot()+30*60;
421                 }
422         }
423
424         /* find a non-empty bucket in the table */
425         while (current_detail &&
426                current_index < current_detail->hash_size &&
427                current_detail->hash_table[current_index] == NULL)
428                 current_index++;
429
430         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
431
432         if (current_detail && current_index < current_detail->hash_size) {
433                 struct cache_head *ch, **cp;
434                 struct cache_detail *d;
435
436                 write_lock(&current_detail->hash_lock);
437
438                 /* Ok, now to clean this strand */
439
440                 cp = & current_detail->hash_table[current_index];
441                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
442                         if (current_detail->nextcheck > ch->expiry_time)
443                                 current_detail->nextcheck = ch->expiry_time+1;
444                         if (!cache_is_expired(current_detail, ch))
445                                 continue;
446
447                         *cp = ch->next;
448                         ch->next = NULL;
449                         current_detail->entries--;
450                         rv = 1;
451                         break;
452                 }
453
454                 write_unlock(&current_detail->hash_lock);
455                 d = current_detail;
456                 if (!ch)
457                         current_index ++;
458                 spin_unlock(&cache_list_lock);
459                 if (ch) {
460                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
461                                 cache_dequeue(current_detail, ch);
462                         cache_revisit_request(ch);
463                         cache_put(ch, d);
464                 }
465         } else
466                 spin_unlock(&cache_list_lock);
467
468         return rv;
469 }
470
471 /*
472  * We want to regularly clean the cache, so we need to schedule some work ...
473  */
474 static void do_cache_clean(struct work_struct *work)
475 {
476         int delay = 5;
477         if (cache_clean() == -1)
478                 delay = round_jiffies_relative(30*HZ);
479
480         if (list_empty(&cache_list))
481                 delay = 0;
482
483         if (delay)
484                 schedule_delayed_work(&cache_cleaner, delay);
485 }
486
487
488 /*
489  * Clean all caches promptly.  This just calls cache_clean
490  * repeatedly until we are sure that every cache has had a chance to
491  * be fully cleaned
492  */
493 void cache_flush(void)
494 {
495         while (cache_clean() != -1)
496                 cond_resched();
497         while (cache_clean() != -1)
498                 cond_resched();
499 }
500 EXPORT_SYMBOL_GPL(cache_flush);
501
502 void cache_purge(struct cache_detail *detail)
503 {
504         detail->flush_time = LONG_MAX;
505         detail->nextcheck = seconds_since_boot();
506         cache_flush();
507         detail->flush_time = 1;
508 }
509 EXPORT_SYMBOL_GPL(cache_purge);
510
511
512 /*
513  * Deferral and Revisiting of Requests.
514  *
515  * If a cache lookup finds a pending entry, we
516  * need to defer the request and revisit it later.
517  * All deferred requests are stored in a hash table,
518  * indexed by "struct cache_head *".
519  * As it may be wasteful to store a whole request
520  * structure, we allow the request to provide a
521  * deferred form, which must contain a
522  * 'struct cache_deferred_req'
523  * This cache_deferred_req contains a method to allow
524  * it to be revisited when cache info is available
525  */
526
527 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
528 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
529
530 #define DFR_MAX 300     /* ??? */
531
532 static DEFINE_SPINLOCK(cache_defer_lock);
533 static LIST_HEAD(cache_defer_list);
534 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
535 static int cache_defer_cnt;
536
537 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
538 {
539         hlist_del_init(&dreq->hash);
540         if (!list_empty(&dreq->recent)) {
541                 list_del_init(&dreq->recent);
542                 cache_defer_cnt--;
543         }
544 }
545
546 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
547 {
548         int hash = DFR_HASH(item);
549
550         INIT_LIST_HEAD(&dreq->recent);
551         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
552 }
553
554 static void setup_deferral(struct cache_deferred_req *dreq,
555                            struct cache_head *item,
556                            int count_me)
557 {
558
559         dreq->item = item;
560
561         spin_lock(&cache_defer_lock);
562
563         __hash_deferred_req(dreq, item);
564
565         if (count_me) {
566                 cache_defer_cnt++;
567                 list_add(&dreq->recent, &cache_defer_list);
568         }
569
570         spin_unlock(&cache_defer_lock);
571
572 }
573
574 struct thread_deferred_req {
575         struct cache_deferred_req handle;
576         struct completion completion;
577 };
578
579 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
580 {
581         struct thread_deferred_req *dr =
582                 container_of(dreq, struct thread_deferred_req, handle);
583         complete(&dr->completion);
584 }
585
586 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
587 {
588         struct thread_deferred_req sleeper;
589         struct cache_deferred_req *dreq = &sleeper.handle;
590
591         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
592         dreq->revisit = cache_restart_thread;
593
594         setup_deferral(dreq, item, 0);
595
596         if (!test_bit(CACHE_PENDING, &item->flags) ||
597             wait_for_completion_interruptible_timeout(
598                     &sleeper.completion, req->thread_wait) <= 0) {
599                 /* The completion wasn't completed, so we need
600                  * to clean up
601                  */
602                 spin_lock(&cache_defer_lock);
603                 if (!hlist_unhashed(&sleeper.handle.hash)) {
604                         __unhash_deferred_req(&sleeper.handle);
605                         spin_unlock(&cache_defer_lock);
606                 } else {
607                         /* cache_revisit_request already removed
608                          * this from the hash table, but hasn't
609                          * called ->revisit yet.  It will very soon
610                          * and we need to wait for it.
611                          */
612                         spin_unlock(&cache_defer_lock);
613                         wait_for_completion(&sleeper.completion);
614                 }
615         }
616 }
617
618 static void cache_limit_defers(void)
619 {
620         /* Make sure we haven't exceed the limit of allowed deferred
621          * requests.
622          */
623         struct cache_deferred_req *discard = NULL;
624
625         if (cache_defer_cnt <= DFR_MAX)
626                 return;
627
628         spin_lock(&cache_defer_lock);
629
630         /* Consider removing either the first or the last */
631         if (cache_defer_cnt > DFR_MAX) {
632                 if (net_random() & 1)
633                         discard = list_entry(cache_defer_list.next,
634                                              struct cache_deferred_req, recent);
635                 else
636                         discard = list_entry(cache_defer_list.prev,
637                                              struct cache_deferred_req, recent);
638                 __unhash_deferred_req(discard);
639         }
640         spin_unlock(&cache_defer_lock);
641         if (discard)
642                 discard->revisit(discard, 1);
643 }
644
645 /* Return true if and only if a deferred request is queued. */
646 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
647 {
648         struct cache_deferred_req *dreq;
649
650         if (req->thread_wait) {
651                 cache_wait_req(req, item);
652                 if (!test_bit(CACHE_PENDING, &item->flags))
653                         return false;
654         }
655         dreq = req->defer(req);
656         if (dreq == NULL)
657                 return false;
658         setup_deferral(dreq, item, 1);
659         if (!test_bit(CACHE_PENDING, &item->flags))
660                 /* Bit could have been cleared before we managed to
661                  * set up the deferral, so need to revisit just in case
662                  */
663                 cache_revisit_request(item);
664
665         cache_limit_defers();
666         return true;
667 }
668
669 static void cache_revisit_request(struct cache_head *item)
670 {
671         struct cache_deferred_req *dreq;
672         struct list_head pending;
673         struct hlist_node *tmp;
674         int hash = DFR_HASH(item);
675
676         INIT_LIST_HEAD(&pending);
677         spin_lock(&cache_defer_lock);
678
679         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
680                 if (dreq->item == item) {
681                         __unhash_deferred_req(dreq);
682                         list_add(&dreq->recent, &pending);
683                 }
684
685         spin_unlock(&cache_defer_lock);
686
687         while (!list_empty(&pending)) {
688                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
689                 list_del_init(&dreq->recent);
690                 dreq->revisit(dreq, 0);
691         }
692 }
693
694 void cache_clean_deferred(void *owner)
695 {
696         struct cache_deferred_req *dreq, *tmp;
697         struct list_head pending;
698
699
700         INIT_LIST_HEAD(&pending);
701         spin_lock(&cache_defer_lock);
702
703         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
704                 if (dreq->owner == owner) {
705                         __unhash_deferred_req(dreq);
706                         list_add(&dreq->recent, &pending);
707                 }
708         }
709         spin_unlock(&cache_defer_lock);
710
711         while (!list_empty(&pending)) {
712                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
713                 list_del_init(&dreq->recent);
714                 dreq->revisit(dreq, 1);
715         }
716 }
717
718 /*
719  * communicate with user-space
720  *
721  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
722  * On read, you get a full request, or block.
723  * On write, an update request is processed.
724  * Poll works if anything to read, and always allows write.
725  *
726  * Implemented by linked list of requests.  Each open file has
727  * a ->private that also exists in this list.  New requests are added
728  * to the end and may wakeup and preceding readers.
729  * New readers are added to the head.  If, on read, an item is found with
730  * CACHE_UPCALLING clear, we free it from the list.
731  *
732  */
733
734 static DEFINE_SPINLOCK(queue_lock);
735 static DEFINE_MUTEX(queue_io_mutex);
736
737 struct cache_queue {
738         struct list_head        list;
739         int                     reader; /* if 0, then request */
740 };
741 struct cache_request {
742         struct cache_queue      q;
743         struct cache_head       *item;
744         char                    * buf;
745         int                     len;
746         int                     readers;
747 };
748 struct cache_reader {
749         struct cache_queue      q;
750         int                     offset; /* if non-0, we have a refcnt on next request */
751 };
752
753 static int cache_request(struct cache_detail *detail,
754                                struct cache_request *crq)
755 {
756         char *bp = crq->buf;
757         int len = PAGE_SIZE;
758
759         detail->cache_request(detail, crq->item, &bp, &len);
760         if (len < 0)
761                 return -EAGAIN;
762         return PAGE_SIZE - len;
763 }
764
765 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
766                           loff_t *ppos, struct cache_detail *cd)
767 {
768         struct cache_reader *rp = filp->private_data;
769         struct cache_request *rq;
770         struct inode *inode = file_inode(filp);
771         int err;
772
773         if (count == 0)
774                 return 0;
775
776         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
777                               * readers on this file */
778  again:
779         spin_lock(&queue_lock);
780         /* need to find next request */
781         while (rp->q.list.next != &cd->queue &&
782                list_entry(rp->q.list.next, struct cache_queue, list)
783                ->reader) {
784                 struct list_head *next = rp->q.list.next;
785                 list_move(&rp->q.list, next);
786         }
787         if (rp->q.list.next == &cd->queue) {
788                 spin_unlock(&queue_lock);
789                 mutex_unlock(&inode->i_mutex);
790                 WARN_ON_ONCE(rp->offset);
791                 return 0;
792         }
793         rq = container_of(rp->q.list.next, struct cache_request, q.list);
794         WARN_ON_ONCE(rq->q.reader);
795         if (rp->offset == 0)
796                 rq->readers++;
797         spin_unlock(&queue_lock);
798
799         if (rq->len == 0) {
800                 err = cache_request(cd, rq);
801                 if (err < 0)
802                         goto out;
803                 rq->len = err;
804         }
805
806         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
807                 err = -EAGAIN;
808                 spin_lock(&queue_lock);
809                 list_move(&rp->q.list, &rq->q.list);
810                 spin_unlock(&queue_lock);
811         } else {
812                 if (rp->offset + count > rq->len)
813                         count = rq->len - rp->offset;
814                 err = -EFAULT;
815                 if (copy_to_user(buf, rq->buf + rp->offset, count))
816                         goto out;
817                 rp->offset += count;
818                 if (rp->offset >= rq->len) {
819                         rp->offset = 0;
820                         spin_lock(&queue_lock);
821                         list_move(&rp->q.list, &rq->q.list);
822                         spin_unlock(&queue_lock);
823                 }
824                 err = 0;
825         }
826  out:
827         if (rp->offset == 0) {
828                 /* need to release rq */
829                 spin_lock(&queue_lock);
830                 rq->readers--;
831                 if (rq->readers == 0 &&
832                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
833                         list_del(&rq->q.list);
834                         spin_unlock(&queue_lock);
835                         cache_put(rq->item, cd);
836                         kfree(rq->buf);
837                         kfree(rq);
838                 } else
839                         spin_unlock(&queue_lock);
840         }
841         if (err == -EAGAIN)
842                 goto again;
843         mutex_unlock(&inode->i_mutex);
844         return err ? err :  count;
845 }
846
847 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
848                                  size_t count, struct cache_detail *cd)
849 {
850         ssize_t ret;
851
852         if (count == 0)
853                 return -EINVAL;
854         if (copy_from_user(kaddr, buf, count))
855                 return -EFAULT;
856         kaddr[count] = '\0';
857         ret = cd->cache_parse(cd, kaddr, count);
858         if (!ret)
859                 ret = count;
860         return ret;
861 }
862
863 static ssize_t cache_slow_downcall(const char __user *buf,
864                                    size_t count, struct cache_detail *cd)
865 {
866         static char write_buf[8192]; /* protected by queue_io_mutex */
867         ssize_t ret = -EINVAL;
868
869         if (count >= sizeof(write_buf))
870                 goto out;
871         mutex_lock(&queue_io_mutex);
872         ret = cache_do_downcall(write_buf, buf, count, cd);
873         mutex_unlock(&queue_io_mutex);
874 out:
875         return ret;
876 }
877
878 static ssize_t cache_downcall(struct address_space *mapping,
879                               const char __user *buf,
880                               size_t count, struct cache_detail *cd)
881 {
882         struct page *page;
883         char *kaddr;
884         ssize_t ret = -ENOMEM;
885
886         if (count >= PAGE_CACHE_SIZE)
887                 goto out_slow;
888
889         page = find_or_create_page(mapping, 0, GFP_KERNEL);
890         if (!page)
891                 goto out_slow;
892
893         kaddr = kmap(page);
894         ret = cache_do_downcall(kaddr, buf, count, cd);
895         kunmap(page);
896         unlock_page(page);
897         page_cache_release(page);
898         return ret;
899 out_slow:
900         return cache_slow_downcall(buf, count, cd);
901 }
902
903 static ssize_t cache_write(struct file *filp, const char __user *buf,
904                            size_t count, loff_t *ppos,
905                            struct cache_detail *cd)
906 {
907         struct address_space *mapping = filp->f_mapping;
908         struct inode *inode = file_inode(filp);
909         ssize_t ret = -EINVAL;
910
911         if (!cd->cache_parse)
912                 goto out;
913
914         mutex_lock(&inode->i_mutex);
915         ret = cache_downcall(mapping, buf, count, cd);
916         mutex_unlock(&inode->i_mutex);
917 out:
918         return ret;
919 }
920
921 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
922
923 static unsigned int cache_poll(struct file *filp, poll_table *wait,
924                                struct cache_detail *cd)
925 {
926         unsigned int mask;
927         struct cache_reader *rp = filp->private_data;
928         struct cache_queue *cq;
929
930         poll_wait(filp, &queue_wait, wait);
931
932         /* alway allow write */
933         mask = POLL_OUT | POLLWRNORM;
934
935         if (!rp)
936                 return mask;
937
938         spin_lock(&queue_lock);
939
940         for (cq= &rp->q; &cq->list != &cd->queue;
941              cq = list_entry(cq->list.next, struct cache_queue, list))
942                 if (!cq->reader) {
943                         mask |= POLLIN | POLLRDNORM;
944                         break;
945                 }
946         spin_unlock(&queue_lock);
947         return mask;
948 }
949
950 static int cache_ioctl(struct inode *ino, struct file *filp,
951                        unsigned int cmd, unsigned long arg,
952                        struct cache_detail *cd)
953 {
954         int len = 0;
955         struct cache_reader *rp = filp->private_data;
956         struct cache_queue *cq;
957
958         if (cmd != FIONREAD || !rp)
959                 return -EINVAL;
960
961         spin_lock(&queue_lock);
962
963         /* only find the length remaining in current request,
964          * or the length of the next request
965          */
966         for (cq= &rp->q; &cq->list != &cd->queue;
967              cq = list_entry(cq->list.next, struct cache_queue, list))
968                 if (!cq->reader) {
969                         struct cache_request *cr =
970                                 container_of(cq, struct cache_request, q);
971                         len = cr->len - rp->offset;
972                         break;
973                 }
974         spin_unlock(&queue_lock);
975
976         return put_user(len, (int __user *)arg);
977 }
978
979 static int cache_open(struct inode *inode, struct file *filp,
980                       struct cache_detail *cd)
981 {
982         struct cache_reader *rp = NULL;
983
984         if (!cd || !try_module_get(cd->owner))
985                 return -EACCES;
986         nonseekable_open(inode, filp);
987         if (filp->f_mode & FMODE_READ) {
988                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
989                 if (!rp)
990                         return -ENOMEM;
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
1002 static 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
1035 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1036 {
1037         struct cache_queue *cq;
1038         spin_lock(&queue_lock);
1039         list_for_each_entry(cq, &detail->queue, list)
1040                 if (!cq->reader) {
1041                         struct cache_request *cr = container_of(cq, struct cache_request, q);
1042                         if (cr->item != ch)
1043                                 continue;
1044                         if (cr->readers != 0)
1045                                 continue;
1046                         list_del(&cr->q.list);
1047                         spin_unlock(&queue_lock);
1048                         cache_put(cr->item, detail);
1049                         kfree(cr->buf);
1050                         kfree(cr);
1051                         return;
1052                 }
1053         spin_unlock(&queue_lock);
1054 }
1055
1056 /*
1057  * Support routines for text-based upcalls.
1058  * Fields are separated by spaces.
1059  * Fields are either mangled to quote space tab newline slosh with slosh
1060  * or a hexified with a leading \x
1061  * Record is terminated with newline.
1062  *
1063  */
1064
1065 void qword_add(char **bpp, int *lp, char *str)
1066 {
1067         char *bp = *bpp;
1068         int len = *lp;
1069         char c;
1070
1071         if (len < 0) return;
1072
1073         while ((c=*str++) && len)
1074                 switch(c) {
1075                 case ' ':
1076                 case '\t':
1077                 case '\n':
1078                 case '\\':
1079                         if (len >= 4) {
1080                                 *bp++ = '\\';
1081                                 *bp++ = '0' + ((c & 0300)>>6);
1082                                 *bp++ = '0' + ((c & 0070)>>3);
1083                                 *bp++ = '0' + ((c & 0007)>>0);
1084                         }
1085                         len -= 4;
1086                         break;
1087                 default:
1088                         *bp++ = c;
1089                         len--;
1090                 }
1091         if (c || len <1) len = -1;
1092         else {
1093                 *bp++ = ' ';
1094                 len--;
1095         }
1096         *bpp = bp;
1097         *lp = len;
1098 }
1099 EXPORT_SYMBOL_GPL(qword_add);
1100
1101 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1102 {
1103         char *bp = *bpp;
1104         int len = *lp;
1105
1106         if (len < 0) return;
1107
1108         if (len > 2) {
1109                 *bp++ = '\\';
1110                 *bp++ = 'x';
1111                 len -= 2;
1112                 while (blen && len >= 2) {
1113                         unsigned char c = *buf++;
1114                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1115                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1116                         len -= 2;
1117                         blen--;
1118                 }
1119         }
1120         if (blen || len<1) len = -1;
1121         else {
1122                 *bp++ = ' ';
1123                 len--;
1124         }
1125         *bpp = bp;
1126         *lp = len;
1127 }
1128 EXPORT_SYMBOL_GPL(qword_addhex);
1129
1130 static void warn_no_listener(struct cache_detail *detail)
1131 {
1132         if (detail->last_warn != detail->last_close) {
1133                 detail->last_warn = detail->last_close;
1134                 if (detail->warn_no_listener)
1135                         detail->warn_no_listener(detail, detail->last_close != 0);
1136         }
1137 }
1138
1139 static bool cache_listeners_exist(struct cache_detail *detail)
1140 {
1141         if (atomic_read(&detail->readers))
1142                 return true;
1143         if (detail->last_close == 0)
1144                 /* This cache was never opened */
1145                 return false;
1146         if (detail->last_close < seconds_since_boot() - 30)
1147                 /*
1148                  * We allow for the possibility that someone might
1149                  * restart a userspace daemon without restarting the
1150                  * server; but after 30 seconds, we give up.
1151                  */
1152                  return false;
1153         return true;
1154 }
1155
1156 /*
1157  * register an upcall request to user-space and queue it up for read() by the
1158  * upcall daemon.
1159  *
1160  * Each request is at most one page long.
1161  */
1162 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1163 {
1164
1165         char *buf;
1166         struct cache_request *crq;
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
1176         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1177         if (!buf)
1178                 return -EAGAIN;
1179
1180         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1181         if (!crq) {
1182                 kfree(buf);
1183                 return -EAGAIN;
1184         }
1185
1186         crq->q.reader = 0;
1187         crq->item = cache_get(h);
1188         crq->buf = buf;
1189         crq->len = 0;
1190         crq->readers = 0;
1191         spin_lock(&queue_lock);
1192         list_add_tail(&crq->q.list, &detail->queue);
1193         spin_unlock(&queue_lock);
1194         wake_up(&queue_wait);
1195         return 0;
1196 }
1197 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1198
1199 /*
1200  * parse a message from user-space and pass it
1201  * to an appropriate cache
1202  * Messages are, like requests, separated into fields by
1203  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1204  *
1205  * Message is
1206  *   reply cachename expiry key ... content....
1207  *
1208  * key and content are both parsed by cache
1209  */
1210
1211 int qword_get(char **bpp, char *dest, int bufsize)
1212 {
1213         /* return bytes copied, or -1 on error */
1214         char *bp = *bpp;
1215         int len = 0;
1216
1217         while (*bp == ' ') bp++;
1218
1219         if (bp[0] == '\\' && bp[1] == 'x') {
1220                 /* HEX STRING */
1221                 bp += 2;
1222                 while (len < bufsize) {
1223                         int h, l;
1224
1225                         h = hex_to_bin(bp[0]);
1226                         if (h < 0)
1227                                 break;
1228
1229                         l = hex_to_bin(bp[1]);
1230                         if (l < 0)
1231                                 break;
1232
1233                         *dest++ = (h << 4) | l;
1234                         bp += 2;
1235                         len++;
1236                 }
1237         } else {
1238                 /* text with \nnn octal quoting */
1239                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1240                         if (*bp == '\\' &&
1241                             isodigit(bp[1]) && (bp[1] <= '3') &&
1242                             isodigit(bp[2]) &&
1243                             isodigit(bp[3])) {
1244                                 int byte = (*++bp -'0');
1245                                 bp++;
1246                                 byte = (byte << 3) | (*bp++ - '0');
1247                                 byte = (byte << 3) | (*bp++ - '0');
1248                                 *dest++ = byte;
1249                                 len++;
1250                         } else {
1251                                 *dest++ = *bp++;
1252                                 len++;
1253                         }
1254                 }
1255         }
1256
1257         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1258                 return -1;
1259         while (*bp == ' ') bp++;
1260         *bpp = bp;
1261         *dest = '\0';
1262         return len;
1263 }
1264 EXPORT_SYMBOL_GPL(qword_get);
1265
1266
1267 /*
1268  * support /proc/sunrpc/cache/$CACHENAME/content
1269  * as a seqfile.
1270  * We call ->cache_show passing NULL for the item to
1271  * get a header, then pass each real item in the cache
1272  */
1273
1274 struct handle {
1275         struct cache_detail *cd;
1276 };
1277
1278 static void *c_start(struct seq_file *m, loff_t *pos)
1279         __acquires(cd->hash_lock)
1280 {
1281         loff_t n = *pos;
1282         unsigned int hash, entry;
1283         struct cache_head *ch;
1284         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1285
1286
1287         read_lock(&cd->hash_lock);
1288         if (!n--)
1289                 return SEQ_START_TOKEN;
1290         hash = n >> 32;
1291         entry = n & ((1LL<<32) - 1);
1292
1293         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1294                 if (!entry--)
1295                         return ch;
1296         n &= ~((1LL<<32) - 1);
1297         do {
1298                 hash++;
1299                 n += 1LL<<32;
1300         } while(hash < cd->hash_size &&
1301                 cd->hash_table[hash]==NULL);
1302         if (hash >= cd->hash_size)
1303                 return NULL;
1304         *pos = n+1;
1305         return cd->hash_table[hash];
1306 }
1307
1308 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1309 {
1310         struct cache_head *ch = p;
1311         int hash = (*pos >> 32);
1312         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1313
1314         if (p == SEQ_START_TOKEN)
1315                 hash = 0;
1316         else if (ch->next == NULL) {
1317                 hash++;
1318                 *pos += 1LL<<32;
1319         } else {
1320                 ++*pos;
1321                 return ch->next;
1322         }
1323         *pos &= ~((1LL<<32) - 1);
1324         while (hash < cd->hash_size &&
1325                cd->hash_table[hash] == NULL) {
1326                 hash++;
1327                 *pos += 1LL<<32;
1328         }
1329         if (hash >= cd->hash_size)
1330                 return NULL;
1331         ++*pos;
1332         return cd->hash_table[hash];
1333 }
1334
1335 static void c_stop(struct seq_file *m, void *p)
1336         __releases(cd->hash_lock)
1337 {
1338         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1339         read_unlock(&cd->hash_lock);
1340 }
1341
1342 static int c_show(struct seq_file *m, void *p)
1343 {
1344         struct cache_head *cp = p;
1345         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1346
1347         if (p == SEQ_START_TOKEN)
1348                 return cd->cache_show(m, cd, NULL);
1349
1350         ifdebug(CACHE)
1351                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1352                            convert_to_wallclock(cp->expiry_time),
1353                            atomic_read(&cp->ref.refcount), cp->flags);
1354         cache_get(cp);
1355         if (cache_check(cd, cp, NULL))
1356                 /* cache_check does a cache_put on failure */
1357                 seq_printf(m, "# ");
1358         else {
1359                 if (cache_is_expired(cd, cp))
1360                         seq_printf(m, "# ");
1361                 cache_put(cp, cd);
1362         }
1363
1364         return cd->cache_show(m, cd, cp);
1365 }
1366
1367 static const struct seq_operations cache_content_op = {
1368         .start  = c_start,
1369         .next   = c_next,
1370         .stop   = c_stop,
1371         .show   = c_show,
1372 };
1373
1374 static int content_open(struct inode *inode, struct file *file,
1375                         struct cache_detail *cd)
1376 {
1377         struct handle *han;
1378
1379         if (!cd || !try_module_get(cd->owner))
1380                 return -EACCES;
1381         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1382         if (han == NULL) {
1383                 module_put(cd->owner);
1384                 return -ENOMEM;
1385         }
1386
1387         han->cd = cd;
1388         return 0;
1389 }
1390
1391 static int content_release(struct inode *inode, struct file *file,
1392                 struct cache_detail *cd)
1393 {
1394         int ret = seq_release_private(inode, file);
1395         module_put(cd->owner);
1396         return ret;
1397 }
1398
1399 static int open_flush(struct inode *inode, struct file *file,
1400                         struct cache_detail *cd)
1401 {
1402         if (!cd || !try_module_get(cd->owner))
1403                 return -EACCES;
1404         return nonseekable_open(inode, file);
1405 }
1406
1407 static int release_flush(struct inode *inode, struct file *file,
1408                         struct cache_detail *cd)
1409 {
1410         module_put(cd->owner);
1411         return 0;
1412 }
1413
1414 static ssize_t read_flush(struct file *file, char __user *buf,
1415                           size_t count, loff_t *ppos,
1416                           struct cache_detail *cd)
1417 {
1418         char tbuf[22];
1419         unsigned long p = *ppos;
1420         size_t len;
1421
1422         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1423         len = strlen(tbuf);
1424         if (p >= len)
1425                 return 0;
1426         len -= p;
1427         if (len > count)
1428                 len = count;
1429         if (copy_to_user(buf, (void*)(tbuf+p), len))
1430                 return -EFAULT;
1431         *ppos += len;
1432         return len;
1433 }
1434
1435 static ssize_t write_flush(struct file *file, const char __user *buf,
1436                            size_t count, loff_t *ppos,
1437                            struct cache_detail *cd)
1438 {
1439         char tbuf[20];
1440         char *bp, *ep;
1441
1442         if (*ppos || count > sizeof(tbuf)-1)
1443                 return -EINVAL;
1444         if (copy_from_user(tbuf, buf, count))
1445                 return -EFAULT;
1446         tbuf[count] = 0;
1447         simple_strtoul(tbuf, &ep, 0);
1448         if (*ep && *ep != '\n')
1449                 return -EINVAL;
1450
1451         bp = tbuf;
1452         cd->flush_time = get_expiry(&bp);
1453         cd->nextcheck = seconds_since_boot();
1454         cache_flush();
1455
1456         *ppos += count;
1457         return count;
1458 }
1459
1460 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1461                                  size_t count, loff_t *ppos)
1462 {
1463         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1464
1465         return cache_read(filp, buf, count, ppos, cd);
1466 }
1467
1468 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1469                                   size_t count, loff_t *ppos)
1470 {
1471         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1472
1473         return cache_write(filp, buf, count, ppos, cd);
1474 }
1475
1476 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1477 {
1478         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1479
1480         return cache_poll(filp, wait, cd);
1481 }
1482
1483 static long cache_ioctl_procfs(struct file *filp,
1484                                unsigned int cmd, unsigned long arg)
1485 {
1486         struct inode *inode = file_inode(filp);
1487         struct cache_detail *cd = PDE_DATA(inode);
1488
1489         return cache_ioctl(inode, filp, cmd, arg, cd);
1490 }
1491
1492 static int cache_open_procfs(struct inode *inode, struct file *filp)
1493 {
1494         struct cache_detail *cd = PDE_DATA(inode);
1495
1496         return cache_open(inode, filp, cd);
1497 }
1498
1499 static int cache_release_procfs(struct inode *inode, struct file *filp)
1500 {
1501         struct cache_detail *cd = PDE_DATA(inode);
1502
1503         return cache_release(inode, filp, cd);
1504 }
1505
1506 static const struct file_operations cache_file_operations_procfs = {
1507         .owner          = THIS_MODULE,
1508         .llseek         = no_llseek,
1509         .read           = cache_read_procfs,
1510         .write          = cache_write_procfs,
1511         .poll           = cache_poll_procfs,
1512         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1513         .open           = cache_open_procfs,
1514         .release        = cache_release_procfs,
1515 };
1516
1517 static int content_open_procfs(struct inode *inode, struct file *filp)
1518 {
1519         struct cache_detail *cd = PDE_DATA(inode);
1520
1521         return content_open(inode, filp, cd);
1522 }
1523
1524 static int content_release_procfs(struct inode *inode, struct file *filp)
1525 {
1526         struct cache_detail *cd = PDE_DATA(inode);
1527
1528         return content_release(inode, filp, cd);
1529 }
1530
1531 static const struct file_operations content_file_operations_procfs = {
1532         .open           = content_open_procfs,
1533         .read           = seq_read,
1534         .llseek         = seq_lseek,
1535         .release        = content_release_procfs,
1536 };
1537
1538 static int open_flush_procfs(struct inode *inode, struct file *filp)
1539 {
1540         struct cache_detail *cd = PDE_DATA(inode);
1541
1542         return open_flush(inode, filp, cd);
1543 }
1544
1545 static int release_flush_procfs(struct inode *inode, struct file *filp)
1546 {
1547         struct cache_detail *cd = PDE_DATA(inode);
1548
1549         return release_flush(inode, filp, cd);
1550 }
1551
1552 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1553                             size_t count, loff_t *ppos)
1554 {
1555         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1556
1557         return read_flush(filp, buf, count, ppos, cd);
1558 }
1559
1560 static ssize_t write_flush_procfs(struct file *filp,
1561                                   const char __user *buf,
1562                                   size_t count, loff_t *ppos)
1563 {
1564         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1565
1566         return write_flush(filp, buf, count, ppos, cd);
1567 }
1568
1569 static const struct file_operations cache_flush_operations_procfs = {
1570         .open           = open_flush_procfs,
1571         .read           = read_flush_procfs,
1572         .write          = write_flush_procfs,
1573         .release        = release_flush_procfs,
1574         .llseek         = no_llseek,
1575 };
1576
1577 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1578 {
1579         struct sunrpc_net *sn;
1580
1581         if (cd->u.procfs.proc_ent == NULL)
1582                 return;
1583         if (cd->u.procfs.flush_ent)
1584                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1585         if (cd->u.procfs.channel_ent)
1586                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1587         if (cd->u.procfs.content_ent)
1588                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1589         cd->u.procfs.proc_ent = NULL;
1590         sn = net_generic(net, sunrpc_net_id);
1591         remove_proc_entry(cd->name, sn->proc_net_rpc);
1592 }
1593
1594 #ifdef CONFIG_PROC_FS
1595 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1596 {
1597         struct proc_dir_entry *p;
1598         struct sunrpc_net *sn;
1599
1600         sn = net_generic(net, sunrpc_net_id);
1601         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1602         if (cd->u.procfs.proc_ent == NULL)
1603                 goto out_nomem;
1604         cd->u.procfs.channel_ent = NULL;
1605         cd->u.procfs.content_ent = NULL;
1606
1607         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1608                              cd->u.procfs.proc_ent,
1609                              &cache_flush_operations_procfs, cd);
1610         cd->u.procfs.flush_ent = p;
1611         if (p == NULL)
1612                 goto out_nomem;
1613
1614         if (cd->cache_request || cd->cache_parse) {
1615                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1616                                      cd->u.procfs.proc_ent,
1617                                      &cache_file_operations_procfs, cd);
1618                 cd->u.procfs.channel_ent = p;
1619                 if (p == NULL)
1620                         goto out_nomem;
1621         }
1622         if (cd->cache_show) {
1623                 p = proc_create_data("content", S_IFREG|S_IRUSR,
1624                                 cd->u.procfs.proc_ent,
1625                                 &content_file_operations_procfs, cd);
1626                 cd->u.procfs.content_ent = p;
1627                 if (p == NULL)
1628                         goto out_nomem;
1629         }
1630         return 0;
1631 out_nomem:
1632         remove_cache_proc_entries(cd, net);
1633         return -ENOMEM;
1634 }
1635 #else /* CONFIG_PROC_FS */
1636 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1637 {
1638         return 0;
1639 }
1640 #endif
1641
1642 void __init cache_initialize(void)
1643 {
1644         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1645 }
1646
1647 int cache_register_net(struct cache_detail *cd, struct net *net)
1648 {
1649         int ret;
1650
1651         sunrpc_init_cache_detail(cd);
1652         ret = create_cache_proc_entries(cd, net);
1653         if (ret)
1654                 sunrpc_destroy_cache_detail(cd);
1655         return ret;
1656 }
1657 EXPORT_SYMBOL_GPL(cache_register_net);
1658
1659 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1660 {
1661         remove_cache_proc_entries(cd, net);
1662         sunrpc_destroy_cache_detail(cd);
1663 }
1664 EXPORT_SYMBOL_GPL(cache_unregister_net);
1665
1666 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1667 {
1668         struct cache_detail *cd;
1669
1670         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1671         if (cd == NULL)
1672                 return ERR_PTR(-ENOMEM);
1673
1674         cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1675                                  GFP_KERNEL);
1676         if (cd->hash_table == NULL) {
1677                 kfree(cd);
1678                 return ERR_PTR(-ENOMEM);
1679         }
1680         cd->net = net;
1681         return cd;
1682 }
1683 EXPORT_SYMBOL_GPL(cache_create_net);
1684
1685 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1686 {
1687         kfree(cd->hash_table);
1688         kfree(cd);
1689 }
1690 EXPORT_SYMBOL_GPL(cache_destroy_net);
1691
1692 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1693                                  size_t count, loff_t *ppos)
1694 {
1695         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1696
1697         return cache_read(filp, buf, count, ppos, cd);
1698 }
1699
1700 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1701                                   size_t count, loff_t *ppos)
1702 {
1703         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1704
1705         return cache_write(filp, buf, count, ppos, cd);
1706 }
1707
1708 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1709 {
1710         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1711
1712         return cache_poll(filp, wait, cd);
1713 }
1714
1715 static long cache_ioctl_pipefs(struct file *filp,
1716                               unsigned int cmd, unsigned long arg)
1717 {
1718         struct inode *inode = file_inode(filp);
1719         struct cache_detail *cd = RPC_I(inode)->private;
1720
1721         return cache_ioctl(inode, filp, cmd, arg, cd);
1722 }
1723
1724 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1725 {
1726         struct cache_detail *cd = RPC_I(inode)->private;
1727
1728         return cache_open(inode, filp, cd);
1729 }
1730
1731 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1732 {
1733         struct cache_detail *cd = RPC_I(inode)->private;
1734
1735         return cache_release(inode, filp, cd);
1736 }
1737
1738 const struct file_operations cache_file_operations_pipefs = {
1739         .owner          = THIS_MODULE,
1740         .llseek         = no_llseek,
1741         .read           = cache_read_pipefs,
1742         .write          = cache_write_pipefs,
1743         .poll           = cache_poll_pipefs,
1744         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1745         .open           = cache_open_pipefs,
1746         .release        = cache_release_pipefs,
1747 };
1748
1749 static int content_open_pipefs(struct inode *inode, struct file *filp)
1750 {
1751         struct cache_detail *cd = RPC_I(inode)->private;
1752
1753         return content_open(inode, filp, cd);
1754 }
1755
1756 static int content_release_pipefs(struct inode *inode, struct file *filp)
1757 {
1758         struct cache_detail *cd = RPC_I(inode)->private;
1759
1760         return content_release(inode, filp, cd);
1761 }
1762
1763 const struct file_operations content_file_operations_pipefs = {
1764         .open           = content_open_pipefs,
1765         .read           = seq_read,
1766         .llseek         = seq_lseek,
1767         .release        = content_release_pipefs,
1768 };
1769
1770 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1771 {
1772         struct cache_detail *cd = RPC_I(inode)->private;
1773
1774         return open_flush(inode, filp, cd);
1775 }
1776
1777 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1778 {
1779         struct cache_detail *cd = RPC_I(inode)->private;
1780
1781         return release_flush(inode, filp, cd);
1782 }
1783
1784 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1785                             size_t count, loff_t *ppos)
1786 {
1787         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1788
1789         return read_flush(filp, buf, count, ppos, cd);
1790 }
1791
1792 static ssize_t write_flush_pipefs(struct file *filp,
1793                                   const char __user *buf,
1794                                   size_t count, loff_t *ppos)
1795 {
1796         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1797
1798         return write_flush(filp, buf, count, ppos, cd);
1799 }
1800
1801 const struct file_operations cache_flush_operations_pipefs = {
1802         .open           = open_flush_pipefs,
1803         .read           = read_flush_pipefs,
1804         .write          = write_flush_pipefs,
1805         .release        = release_flush_pipefs,
1806         .llseek         = no_llseek,
1807 };
1808
1809 int sunrpc_cache_register_pipefs(struct dentry *parent,
1810                                  const char *name, umode_t umode,
1811                                  struct cache_detail *cd)
1812 {
1813         struct qstr q;
1814         struct dentry *dir;
1815         int ret = 0;
1816
1817         q.name = name;
1818         q.len = strlen(name);
1819         q.hash = full_name_hash(q.name, q.len);
1820         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1821         if (!IS_ERR(dir))
1822                 cd->u.pipefs.dir = dir;
1823         else
1824                 ret = PTR_ERR(dir);
1825         return ret;
1826 }
1827 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1828
1829 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1830 {
1831         rpc_remove_cache_dir(cd->u.pipefs.dir);
1832         cd->u.pipefs.dir = NULL;
1833 }
1834 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1835