Merge branch 'for-2.6.37' of git://linux-nfs.org/~bfields/linux
[~shefty/rdma-dev.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 void 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         set_bit(CACHE_VALID, &head->flags);
132 }
133
134 static void cache_fresh_unlocked(struct cache_head *head,
135                                  struct cache_detail *detail)
136 {
137         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
138                 cache_revisit_request(head);
139                 cache_dequeue(detail, head);
140         }
141 }
142
143 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
144                                        struct cache_head *new, struct cache_head *old, int hash)
145 {
146         /* The 'old' entry is to be replaced by 'new'.
147          * If 'old' is not VALID, we update it directly,
148          * otherwise we need to replace it
149          */
150         struct cache_head **head;
151         struct cache_head *tmp;
152
153         if (!test_bit(CACHE_VALID, &old->flags)) {
154                 write_lock(&detail->hash_lock);
155                 if (!test_bit(CACHE_VALID, &old->flags)) {
156                         if (test_bit(CACHE_NEGATIVE, &new->flags))
157                                 set_bit(CACHE_NEGATIVE, &old->flags);
158                         else
159                                 detail->update(old, new);
160                         cache_fresh_locked(old, new->expiry_time);
161                         write_unlock(&detail->hash_lock);
162                         cache_fresh_unlocked(old, detail);
163                         return old;
164                 }
165                 write_unlock(&detail->hash_lock);
166         }
167         /* We need to insert a new entry */
168         tmp = detail->alloc();
169         if (!tmp) {
170                 cache_put(old, detail);
171                 return NULL;
172         }
173         cache_init(tmp);
174         detail->init(tmp, old);
175         head = &detail->hash_table[hash];
176
177         write_lock(&detail->hash_lock);
178         if (test_bit(CACHE_NEGATIVE, &new->flags))
179                 set_bit(CACHE_NEGATIVE, &tmp->flags);
180         else
181                 detail->update(tmp, new);
182         tmp->next = *head;
183         *head = tmp;
184         detail->entries++;
185         cache_get(tmp);
186         cache_fresh_locked(tmp, new->expiry_time);
187         cache_fresh_locked(old, 0);
188         write_unlock(&detail->hash_lock);
189         cache_fresh_unlocked(tmp, detail);
190         cache_fresh_unlocked(old, detail);
191         cache_put(old, detail);
192         return tmp;
193 }
194 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
195
196 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
197 {
198         if (!cd->cache_upcall)
199                 return -EINVAL;
200         return cd->cache_upcall(cd, h);
201 }
202
203 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
204 {
205         if (!test_bit(CACHE_VALID, &h->flags))
206                 return -EAGAIN;
207         else {
208                 /* entry is valid */
209                 if (test_bit(CACHE_NEGATIVE, &h->flags))
210                         return -ENOENT;
211                 else
212                         return 0;
213         }
214 }
215
216 /*
217  * This is the generic cache management routine for all
218  * the authentication caches.
219  * It checks the currency of a cache item and will (later)
220  * initiate an upcall to fill it if needed.
221  *
222  *
223  * Returns 0 if the cache_head can be used, or cache_puts it and returns
224  * -EAGAIN if upcall is pending and request has been queued
225  * -ETIMEDOUT if upcall failed or request could not be queue or
226  *           upcall completed but item is still invalid (implying that
227  *           the cache item has been replaced with a newer one).
228  * -ENOENT if cache entry was negative
229  */
230 int cache_check(struct cache_detail *detail,
231                     struct cache_head *h, struct cache_req *rqstp)
232 {
233         int rv;
234         long refresh_age, age;
235
236         /* First decide return status as best we can */
237         rv = cache_is_valid(detail, h);
238
239         /* now see if we want to start an upcall */
240         refresh_age = (h->expiry_time - h->last_refresh);
241         age = seconds_since_boot() - h->last_refresh;
242
243         if (rqstp == NULL) {
244                 if (rv == -EAGAIN)
245                         rv = -ENOENT;
246         } else if (rv == -EAGAIN || age > refresh_age/2) {
247                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
248                                 refresh_age, age);
249                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
250                         switch (cache_make_upcall(detail, h)) {
251                         case -EINVAL:
252                                 clear_bit(CACHE_PENDING, &h->flags);
253                                 cache_revisit_request(h);
254                                 if (rv == -EAGAIN) {
255                                         set_bit(CACHE_NEGATIVE, &h->flags);
256                                         cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
257                                         cache_fresh_unlocked(h, detail);
258                                         rv = -ENOENT;
259                                 }
260                                 break;
261
262                         case -EAGAIN:
263                                 clear_bit(CACHE_PENDING, &h->flags);
264                                 cache_revisit_request(h);
265                                 break;
266                         }
267                 }
268         }
269
270         if (rv == -EAGAIN) {
271                 cache_defer_req(rqstp, h);
272                 if (!test_bit(CACHE_PENDING, &h->flags)) {
273                         /* Request is not deferred */
274                         rv = cache_is_valid(detail, h);
275                         if (rv == -EAGAIN)
276                                 rv = -ETIMEDOUT;
277                 }
278         }
279         if (rv)
280                 cache_put(h, detail);
281         return rv;
282 }
283 EXPORT_SYMBOL_GPL(cache_check);
284
285 /*
286  * caches need to be periodically cleaned.
287  * For this we maintain a list of cache_detail and
288  * a current pointer into that list and into the table
289  * for that entry.
290  *
291  * Each time clean_cache is called it finds the next non-empty entry
292  * in the current table and walks the list in that entry
293  * looking for entries that can be removed.
294  *
295  * An entry gets removed if:
296  * - The expiry is before current time
297  * - The last_refresh time is before the flush_time for that cache
298  *
299  * later we might drop old entries with non-NEVER expiry if that table
300  * is getting 'full' for some definition of 'full'
301  *
302  * The question of "how often to scan a table" is an interesting one
303  * and is answered in part by the use of the "nextcheck" field in the
304  * cache_detail.
305  * When a scan of a table begins, the nextcheck field is set to a time
306  * that is well into the future.
307  * While scanning, if an expiry time is found that is earlier than the
308  * current nextcheck time, nextcheck is set to that expiry time.
309  * If the flush_time is ever set to a time earlier than the nextcheck
310  * time, the nextcheck time is then set to that flush_time.
311  *
312  * A table is then only scanned if the current time is at least
313  * the nextcheck time.
314  *
315  */
316
317 static LIST_HEAD(cache_list);
318 static DEFINE_SPINLOCK(cache_list_lock);
319 static struct cache_detail *current_detail;
320 static int current_index;
321
322 static void do_cache_clean(struct work_struct *work);
323 static struct delayed_work cache_cleaner;
324
325 static void sunrpc_init_cache_detail(struct cache_detail *cd)
326 {
327         rwlock_init(&cd->hash_lock);
328         INIT_LIST_HEAD(&cd->queue);
329         spin_lock(&cache_list_lock);
330         cd->nextcheck = 0;
331         cd->entries = 0;
332         atomic_set(&cd->readers, 0);
333         cd->last_close = 0;
334         cd->last_warn = -1;
335         list_add(&cd->others, &cache_list);
336         spin_unlock(&cache_list_lock);
337
338         /* start the cleaning process */
339         schedule_delayed_work(&cache_cleaner, 0);
340 }
341
342 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
343 {
344         cache_purge(cd);
345         spin_lock(&cache_list_lock);
346         write_lock(&cd->hash_lock);
347         if (cd->entries || atomic_read(&cd->inuse)) {
348                 write_unlock(&cd->hash_lock);
349                 spin_unlock(&cache_list_lock);
350                 goto out;
351         }
352         if (current_detail == cd)
353                 current_detail = NULL;
354         list_del_init(&cd->others);
355         write_unlock(&cd->hash_lock);
356         spin_unlock(&cache_list_lock);
357         if (list_empty(&cache_list)) {
358                 /* module must be being unloaded so its safe to kill the worker */
359                 cancel_delayed_work_sync(&cache_cleaner);
360         }
361         return;
362 out:
363         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
364 }
365
366 /* clean cache tries to find something to clean
367  * and cleans it.
368  * It returns 1 if it cleaned something,
369  *            0 if it didn't find anything this time
370  *           -1 if it fell off the end of the list.
371  */
372 static int cache_clean(void)
373 {
374         int rv = 0;
375         struct list_head *next;
376
377         spin_lock(&cache_list_lock);
378
379         /* find a suitable table if we don't already have one */
380         while (current_detail == NULL ||
381             current_index >= current_detail->hash_size) {
382                 if (current_detail)
383                         next = current_detail->others.next;
384                 else
385                         next = cache_list.next;
386                 if (next == &cache_list) {
387                         current_detail = NULL;
388                         spin_unlock(&cache_list_lock);
389                         return -1;
390                 }
391                 current_detail = list_entry(next, struct cache_detail, others);
392                 if (current_detail->nextcheck > seconds_since_boot())
393                         current_index = current_detail->hash_size;
394                 else {
395                         current_index = 0;
396                         current_detail->nextcheck = seconds_since_boot()+30*60;
397                 }
398         }
399
400         /* find a non-empty bucket in the table */
401         while (current_detail &&
402                current_index < current_detail->hash_size &&
403                current_detail->hash_table[current_index] == NULL)
404                 current_index++;
405
406         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
407
408         if (current_detail && current_index < current_detail->hash_size) {
409                 struct cache_head *ch, **cp;
410                 struct cache_detail *d;
411
412                 write_lock(&current_detail->hash_lock);
413
414                 /* Ok, now to clean this strand */
415
416                 cp = & current_detail->hash_table[current_index];
417                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
418                         if (current_detail->nextcheck > ch->expiry_time)
419                                 current_detail->nextcheck = ch->expiry_time+1;
420                         if (!cache_is_expired(current_detail, ch))
421                                 continue;
422
423                         *cp = ch->next;
424                         ch->next = NULL;
425                         current_detail->entries--;
426                         rv = 1;
427                         break;
428                 }
429
430                 write_unlock(&current_detail->hash_lock);
431                 d = current_detail;
432                 if (!ch)
433                         current_index ++;
434                 spin_unlock(&cache_list_lock);
435                 if (ch) {
436                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
437                                 cache_dequeue(current_detail, ch);
438                         cache_revisit_request(ch);
439                         cache_put(ch, d);
440                 }
441         } else
442                 spin_unlock(&cache_list_lock);
443
444         return rv;
445 }
446
447 /*
448  * We want to regularly clean the cache, so we need to schedule some work ...
449  */
450 static void do_cache_clean(struct work_struct *work)
451 {
452         int delay = 5;
453         if (cache_clean() == -1)
454                 delay = round_jiffies_relative(30*HZ);
455
456         if (list_empty(&cache_list))
457                 delay = 0;
458
459         if (delay)
460                 schedule_delayed_work(&cache_cleaner, delay);
461 }
462
463
464 /*
465  * Clean all caches promptly.  This just calls cache_clean
466  * repeatedly until we are sure that every cache has had a chance to
467  * be fully cleaned
468  */
469 void cache_flush(void)
470 {
471         while (cache_clean() != -1)
472                 cond_resched();
473         while (cache_clean() != -1)
474                 cond_resched();
475 }
476 EXPORT_SYMBOL_GPL(cache_flush);
477
478 void cache_purge(struct cache_detail *detail)
479 {
480         detail->flush_time = LONG_MAX;
481         detail->nextcheck = seconds_since_boot();
482         cache_flush();
483         detail->flush_time = 1;
484 }
485 EXPORT_SYMBOL_GPL(cache_purge);
486
487
488 /*
489  * Deferral and Revisiting of Requests.
490  *
491  * If a cache lookup finds a pending entry, we
492  * need to defer the request and revisit it later.
493  * All deferred requests are stored in a hash table,
494  * indexed by "struct cache_head *".
495  * As it may be wasteful to store a whole request
496  * structure, we allow the request to provide a
497  * deferred form, which must contain a
498  * 'struct cache_deferred_req'
499  * This cache_deferred_req contains a method to allow
500  * it to be revisited when cache info is available
501  */
502
503 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
504 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
505
506 #define DFR_MAX 300     /* ??? */
507
508 static DEFINE_SPINLOCK(cache_defer_lock);
509 static LIST_HEAD(cache_defer_list);
510 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
511 static int cache_defer_cnt;
512
513 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
514 {
515         hlist_del_init(&dreq->hash);
516         if (!list_empty(&dreq->recent)) {
517                 list_del_init(&dreq->recent);
518                 cache_defer_cnt--;
519         }
520 }
521
522 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
523 {
524         int hash = DFR_HASH(item);
525
526         INIT_LIST_HEAD(&dreq->recent);
527         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
528 }
529
530 static void setup_deferral(struct cache_deferred_req *dreq,
531                            struct cache_head *item,
532                            int count_me)
533 {
534
535         dreq->item = item;
536
537         spin_lock(&cache_defer_lock);
538
539         __hash_deferred_req(dreq, item);
540
541         if (count_me) {
542                 cache_defer_cnt++;
543                 list_add(&dreq->recent, &cache_defer_list);
544         }
545
546         spin_unlock(&cache_defer_lock);
547
548 }
549
550 struct thread_deferred_req {
551         struct cache_deferred_req handle;
552         struct completion completion;
553 };
554
555 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
556 {
557         struct thread_deferred_req *dr =
558                 container_of(dreq, struct thread_deferred_req, handle);
559         complete(&dr->completion);
560 }
561
562 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
563 {
564         struct thread_deferred_req sleeper;
565         struct cache_deferred_req *dreq = &sleeper.handle;
566
567         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
568         dreq->revisit = cache_restart_thread;
569
570         setup_deferral(dreq, item, 0);
571
572         if (!test_bit(CACHE_PENDING, &item->flags) ||
573             wait_for_completion_interruptible_timeout(
574                     &sleeper.completion, req->thread_wait) <= 0) {
575                 /* The completion wasn't completed, so we need
576                  * to clean up
577                  */
578                 spin_lock(&cache_defer_lock);
579                 if (!hlist_unhashed(&sleeper.handle.hash)) {
580                         __unhash_deferred_req(&sleeper.handle);
581                         spin_unlock(&cache_defer_lock);
582                 } else {
583                         /* cache_revisit_request already removed
584                          * this from the hash table, but hasn't
585                          * called ->revisit yet.  It will very soon
586                          * and we need to wait for it.
587                          */
588                         spin_unlock(&cache_defer_lock);
589                         wait_for_completion(&sleeper.completion);
590                 }
591         }
592 }
593
594 static void cache_limit_defers(void)
595 {
596         /* Make sure we haven't exceed the limit of allowed deferred
597          * requests.
598          */
599         struct cache_deferred_req *discard = NULL;
600
601         if (cache_defer_cnt <= DFR_MAX)
602                 return;
603
604         spin_lock(&cache_defer_lock);
605
606         /* Consider removing either the first or the last */
607         if (cache_defer_cnt > DFR_MAX) {
608                 if (net_random() & 1)
609                         discard = list_entry(cache_defer_list.next,
610                                              struct cache_deferred_req, recent);
611                 else
612                         discard = list_entry(cache_defer_list.prev,
613                                              struct cache_deferred_req, recent);
614                 __unhash_deferred_req(discard);
615         }
616         spin_unlock(&cache_defer_lock);
617         if (discard)
618                 discard->revisit(discard, 1);
619 }
620
621 static void cache_defer_req(struct cache_req *req, struct cache_head *item)
622 {
623         struct cache_deferred_req *dreq;
624
625         if (req->thread_wait) {
626                 cache_wait_req(req, item);
627                 if (!test_bit(CACHE_PENDING, &item->flags))
628                         return;
629         }
630         dreq = req->defer(req);
631         if (dreq == NULL)
632                 return;
633         setup_deferral(dreq, item, 1);
634         if (!test_bit(CACHE_PENDING, &item->flags))
635                 /* Bit could have been cleared before we managed to
636                  * set up the deferral, so need to revisit just in case
637                  */
638                 cache_revisit_request(item);
639
640         cache_limit_defers();
641 }
642
643 static void cache_revisit_request(struct cache_head *item)
644 {
645         struct cache_deferred_req *dreq;
646         struct list_head pending;
647         struct hlist_node *lp, *tmp;
648         int hash = DFR_HASH(item);
649
650         INIT_LIST_HEAD(&pending);
651         spin_lock(&cache_defer_lock);
652
653         hlist_for_each_entry_safe(dreq, lp, tmp, &cache_defer_hash[hash], hash)
654                 if (dreq->item == item) {
655                         __unhash_deferred_req(dreq);
656                         list_add(&dreq->recent, &pending);
657                 }
658
659         spin_unlock(&cache_defer_lock);
660
661         while (!list_empty(&pending)) {
662                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
663                 list_del_init(&dreq->recent);
664                 dreq->revisit(dreq, 0);
665         }
666 }
667
668 void cache_clean_deferred(void *owner)
669 {
670         struct cache_deferred_req *dreq, *tmp;
671         struct list_head pending;
672
673
674         INIT_LIST_HEAD(&pending);
675         spin_lock(&cache_defer_lock);
676
677         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
678                 if (dreq->owner == owner) {
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, 1);
689         }
690 }
691
692 /*
693  * communicate with user-space
694  *
695  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
696  * On read, you get a full request, or block.
697  * On write, an update request is processed.
698  * Poll works if anything to read, and always allows write.
699  *
700  * Implemented by linked list of requests.  Each open file has
701  * a ->private that also exists in this list.  New requests are added
702  * to the end and may wakeup and preceding readers.
703  * New readers are added to the head.  If, on read, an item is found with
704  * CACHE_UPCALLING clear, we free it from the list.
705  *
706  */
707
708 static DEFINE_SPINLOCK(queue_lock);
709 static DEFINE_MUTEX(queue_io_mutex);
710
711 struct cache_queue {
712         struct list_head        list;
713         int                     reader; /* if 0, then request */
714 };
715 struct cache_request {
716         struct cache_queue      q;
717         struct cache_head       *item;
718         char                    * buf;
719         int                     len;
720         int                     readers;
721 };
722 struct cache_reader {
723         struct cache_queue      q;
724         int                     offset; /* if non-0, we have a refcnt on next request */
725 };
726
727 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
728                           loff_t *ppos, struct cache_detail *cd)
729 {
730         struct cache_reader *rp = filp->private_data;
731         struct cache_request *rq;
732         struct inode *inode = filp->f_path.dentry->d_inode;
733         int err;
734
735         if (count == 0)
736                 return 0;
737
738         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
739                               * readers on this file */
740  again:
741         spin_lock(&queue_lock);
742         /* need to find next request */
743         while (rp->q.list.next != &cd->queue &&
744                list_entry(rp->q.list.next, struct cache_queue, list)
745                ->reader) {
746                 struct list_head *next = rp->q.list.next;
747                 list_move(&rp->q.list, next);
748         }
749         if (rp->q.list.next == &cd->queue) {
750                 spin_unlock(&queue_lock);
751                 mutex_unlock(&inode->i_mutex);
752                 BUG_ON(rp->offset);
753                 return 0;
754         }
755         rq = container_of(rp->q.list.next, struct cache_request, q.list);
756         BUG_ON(rq->q.reader);
757         if (rp->offset == 0)
758                 rq->readers++;
759         spin_unlock(&queue_lock);
760
761         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
762                 err = -EAGAIN;
763                 spin_lock(&queue_lock);
764                 list_move(&rp->q.list, &rq->q.list);
765                 spin_unlock(&queue_lock);
766         } else {
767                 if (rp->offset + count > rq->len)
768                         count = rq->len - rp->offset;
769                 err = -EFAULT;
770                 if (copy_to_user(buf, rq->buf + rp->offset, count))
771                         goto out;
772                 rp->offset += count;
773                 if (rp->offset >= rq->len) {
774                         rp->offset = 0;
775                         spin_lock(&queue_lock);
776                         list_move(&rp->q.list, &rq->q.list);
777                         spin_unlock(&queue_lock);
778                 }
779                 err = 0;
780         }
781  out:
782         if (rp->offset == 0) {
783                 /* need to release rq */
784                 spin_lock(&queue_lock);
785                 rq->readers--;
786                 if (rq->readers == 0 &&
787                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
788                         list_del(&rq->q.list);
789                         spin_unlock(&queue_lock);
790                         cache_put(rq->item, cd);
791                         kfree(rq->buf);
792                         kfree(rq);
793                 } else
794                         spin_unlock(&queue_lock);
795         }
796         if (err == -EAGAIN)
797                 goto again;
798         mutex_unlock(&inode->i_mutex);
799         return err ? err :  count;
800 }
801
802 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
803                                  size_t count, struct cache_detail *cd)
804 {
805         ssize_t ret;
806
807         if (copy_from_user(kaddr, buf, count))
808                 return -EFAULT;
809         kaddr[count] = '\0';
810         ret = cd->cache_parse(cd, kaddr, count);
811         if (!ret)
812                 ret = count;
813         return ret;
814 }
815
816 static ssize_t cache_slow_downcall(const char __user *buf,
817                                    size_t count, struct cache_detail *cd)
818 {
819         static char write_buf[8192]; /* protected by queue_io_mutex */
820         ssize_t ret = -EINVAL;
821
822         if (count >= sizeof(write_buf))
823                 goto out;
824         mutex_lock(&queue_io_mutex);
825         ret = cache_do_downcall(write_buf, buf, count, cd);
826         mutex_unlock(&queue_io_mutex);
827 out:
828         return ret;
829 }
830
831 static ssize_t cache_downcall(struct address_space *mapping,
832                               const char __user *buf,
833                               size_t count, struct cache_detail *cd)
834 {
835         struct page *page;
836         char *kaddr;
837         ssize_t ret = -ENOMEM;
838
839         if (count >= PAGE_CACHE_SIZE)
840                 goto out_slow;
841
842         page = find_or_create_page(mapping, 0, GFP_KERNEL);
843         if (!page)
844                 goto out_slow;
845
846         kaddr = kmap(page);
847         ret = cache_do_downcall(kaddr, buf, count, cd);
848         kunmap(page);
849         unlock_page(page);
850         page_cache_release(page);
851         return ret;
852 out_slow:
853         return cache_slow_downcall(buf, count, cd);
854 }
855
856 static ssize_t cache_write(struct file *filp, const char __user *buf,
857                            size_t count, loff_t *ppos,
858                            struct cache_detail *cd)
859 {
860         struct address_space *mapping = filp->f_mapping;
861         struct inode *inode = filp->f_path.dentry->d_inode;
862         ssize_t ret = -EINVAL;
863
864         if (!cd->cache_parse)
865                 goto out;
866
867         mutex_lock(&inode->i_mutex);
868         ret = cache_downcall(mapping, buf, count, cd);
869         mutex_unlock(&inode->i_mutex);
870 out:
871         return ret;
872 }
873
874 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
875
876 static unsigned int cache_poll(struct file *filp, poll_table *wait,
877                                struct cache_detail *cd)
878 {
879         unsigned int mask;
880         struct cache_reader *rp = filp->private_data;
881         struct cache_queue *cq;
882
883         poll_wait(filp, &queue_wait, wait);
884
885         /* alway allow write */
886         mask = POLL_OUT | POLLWRNORM;
887
888         if (!rp)
889                 return mask;
890
891         spin_lock(&queue_lock);
892
893         for (cq= &rp->q; &cq->list != &cd->queue;
894              cq = list_entry(cq->list.next, struct cache_queue, list))
895                 if (!cq->reader) {
896                         mask |= POLLIN | POLLRDNORM;
897                         break;
898                 }
899         spin_unlock(&queue_lock);
900         return mask;
901 }
902
903 static int cache_ioctl(struct inode *ino, struct file *filp,
904                        unsigned int cmd, unsigned long arg,
905                        struct cache_detail *cd)
906 {
907         int len = 0;
908         struct cache_reader *rp = filp->private_data;
909         struct cache_queue *cq;
910
911         if (cmd != FIONREAD || !rp)
912                 return -EINVAL;
913
914         spin_lock(&queue_lock);
915
916         /* only find the length remaining in current request,
917          * or the length of the next request
918          */
919         for (cq= &rp->q; &cq->list != &cd->queue;
920              cq = list_entry(cq->list.next, struct cache_queue, list))
921                 if (!cq->reader) {
922                         struct cache_request *cr =
923                                 container_of(cq, struct cache_request, q);
924                         len = cr->len - rp->offset;
925                         break;
926                 }
927         spin_unlock(&queue_lock);
928
929         return put_user(len, (int __user *)arg);
930 }
931
932 static int cache_open(struct inode *inode, struct file *filp,
933                       struct cache_detail *cd)
934 {
935         struct cache_reader *rp = NULL;
936
937         if (!cd || !try_module_get(cd->owner))
938                 return -EACCES;
939         nonseekable_open(inode, filp);
940         if (filp->f_mode & FMODE_READ) {
941                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
942                 if (!rp)
943                         return -ENOMEM;
944                 rp->offset = 0;
945                 rp->q.reader = 1;
946                 atomic_inc(&cd->readers);
947                 spin_lock(&queue_lock);
948                 list_add(&rp->q.list, &cd->queue);
949                 spin_unlock(&queue_lock);
950         }
951         filp->private_data = rp;
952         return 0;
953 }
954
955 static int cache_release(struct inode *inode, struct file *filp,
956                          struct cache_detail *cd)
957 {
958         struct cache_reader *rp = filp->private_data;
959
960         if (rp) {
961                 spin_lock(&queue_lock);
962                 if (rp->offset) {
963                         struct cache_queue *cq;
964                         for (cq= &rp->q; &cq->list != &cd->queue;
965                              cq = list_entry(cq->list.next, struct cache_queue, list))
966                                 if (!cq->reader) {
967                                         container_of(cq, struct cache_request, q)
968                                                 ->readers--;
969                                         break;
970                                 }
971                         rp->offset = 0;
972                 }
973                 list_del(&rp->q.list);
974                 spin_unlock(&queue_lock);
975
976                 filp->private_data = NULL;
977                 kfree(rp);
978
979                 cd->last_close = seconds_since_boot();
980                 atomic_dec(&cd->readers);
981         }
982         module_put(cd->owner);
983         return 0;
984 }
985
986
987
988 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
989 {
990         struct cache_queue *cq;
991         spin_lock(&queue_lock);
992         list_for_each_entry(cq, &detail->queue, list)
993                 if (!cq->reader) {
994                         struct cache_request *cr = container_of(cq, struct cache_request, q);
995                         if (cr->item != ch)
996                                 continue;
997                         if (cr->readers != 0)
998                                 continue;
999                         list_del(&cr->q.list);
1000                         spin_unlock(&queue_lock);
1001                         cache_put(cr->item, detail);
1002                         kfree(cr->buf);
1003                         kfree(cr);
1004                         return;
1005                 }
1006         spin_unlock(&queue_lock);
1007 }
1008
1009 /*
1010  * Support routines for text-based upcalls.
1011  * Fields are separated by spaces.
1012  * Fields are either mangled to quote space tab newline slosh with slosh
1013  * or a hexified with a leading \x
1014  * Record is terminated with newline.
1015  *
1016  */
1017
1018 void qword_add(char **bpp, int *lp, char *str)
1019 {
1020         char *bp = *bpp;
1021         int len = *lp;
1022         char c;
1023
1024         if (len < 0) return;
1025
1026         while ((c=*str++) && len)
1027                 switch(c) {
1028                 case ' ':
1029                 case '\t':
1030                 case '\n':
1031                 case '\\':
1032                         if (len >= 4) {
1033                                 *bp++ = '\\';
1034                                 *bp++ = '0' + ((c & 0300)>>6);
1035                                 *bp++ = '0' + ((c & 0070)>>3);
1036                                 *bp++ = '0' + ((c & 0007)>>0);
1037                         }
1038                         len -= 4;
1039                         break;
1040                 default:
1041                         *bp++ = c;
1042                         len--;
1043                 }
1044         if (c || len <1) len = -1;
1045         else {
1046                 *bp++ = ' ';
1047                 len--;
1048         }
1049         *bpp = bp;
1050         *lp = len;
1051 }
1052 EXPORT_SYMBOL_GPL(qword_add);
1053
1054 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1055 {
1056         char *bp = *bpp;
1057         int len = *lp;
1058
1059         if (len < 0) return;
1060
1061         if (len > 2) {
1062                 *bp++ = '\\';
1063                 *bp++ = 'x';
1064                 len -= 2;
1065                 while (blen && len >= 2) {
1066                         unsigned char c = *buf++;
1067                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1068                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1069                         len -= 2;
1070                         blen--;
1071                 }
1072         }
1073         if (blen || len<1) len = -1;
1074         else {
1075                 *bp++ = ' ';
1076                 len--;
1077         }
1078         *bpp = bp;
1079         *lp = len;
1080 }
1081 EXPORT_SYMBOL_GPL(qword_addhex);
1082
1083 static void warn_no_listener(struct cache_detail *detail)
1084 {
1085         if (detail->last_warn != detail->last_close) {
1086                 detail->last_warn = detail->last_close;
1087                 if (detail->warn_no_listener)
1088                         detail->warn_no_listener(detail, detail->last_close != 0);
1089         }
1090 }
1091
1092 static bool cache_listeners_exist(struct cache_detail *detail)
1093 {
1094         if (atomic_read(&detail->readers))
1095                 return true;
1096         if (detail->last_close == 0)
1097                 /* This cache was never opened */
1098                 return false;
1099         if (detail->last_close < seconds_since_boot() - 30)
1100                 /*
1101                  * We allow for the possibility that someone might
1102                  * restart a userspace daemon without restarting the
1103                  * server; but after 30 seconds, we give up.
1104                  */
1105                  return false;
1106         return true;
1107 }
1108
1109 /*
1110  * register an upcall request to user-space and queue it up for read() by the
1111  * upcall daemon.
1112  *
1113  * Each request is at most one page long.
1114  */
1115 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1116                 void (*cache_request)(struct cache_detail *,
1117                                       struct cache_head *,
1118                                       char **,
1119                                       int *))
1120 {
1121
1122         char *buf;
1123         struct cache_request *crq;
1124         char *bp;
1125         int len;
1126
1127         if (!cache_listeners_exist(detail)) {
1128                 warn_no_listener(detail);
1129                 return -EINVAL;
1130         }
1131
1132         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1133         if (!buf)
1134                 return -EAGAIN;
1135
1136         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1137         if (!crq) {
1138                 kfree(buf);
1139                 return -EAGAIN;
1140         }
1141
1142         bp = buf; len = PAGE_SIZE;
1143
1144         cache_request(detail, h, &bp, &len);
1145
1146         if (len < 0) {
1147                 kfree(buf);
1148                 kfree(crq);
1149                 return -EAGAIN;
1150         }
1151         crq->q.reader = 0;
1152         crq->item = cache_get(h);
1153         crq->buf = buf;
1154         crq->len = PAGE_SIZE - len;
1155         crq->readers = 0;
1156         spin_lock(&queue_lock);
1157         list_add_tail(&crq->q.list, &detail->queue);
1158         spin_unlock(&queue_lock);
1159         wake_up(&queue_wait);
1160         return 0;
1161 }
1162 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1163
1164 /*
1165  * parse a message from user-space and pass it
1166  * to an appropriate cache
1167  * Messages are, like requests, separated into fields by
1168  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1169  *
1170  * Message is
1171  *   reply cachename expiry key ... content....
1172  *
1173  * key and content are both parsed by cache
1174  */
1175
1176 #define isodigit(c) (isdigit(c) && c <= '7')
1177 int qword_get(char **bpp, char *dest, int bufsize)
1178 {
1179         /* return bytes copied, or -1 on error */
1180         char *bp = *bpp;
1181         int len = 0;
1182
1183         while (*bp == ' ') bp++;
1184
1185         if (bp[0] == '\\' && bp[1] == 'x') {
1186                 /* HEX STRING */
1187                 bp += 2;
1188                 while (len < bufsize) {
1189                         int h, l;
1190
1191                         h = hex_to_bin(bp[0]);
1192                         if (h < 0)
1193                                 break;
1194
1195                         l = hex_to_bin(bp[1]);
1196                         if (l < 0)
1197                                 break;
1198
1199                         *dest++ = (h << 4) | l;
1200                         bp += 2;
1201                         len++;
1202                 }
1203         } else {
1204                 /* text with \nnn octal quoting */
1205                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1206                         if (*bp == '\\' &&
1207                             isodigit(bp[1]) && (bp[1] <= '3') &&
1208                             isodigit(bp[2]) &&
1209                             isodigit(bp[3])) {
1210                                 int byte = (*++bp -'0');
1211                                 bp++;
1212                                 byte = (byte << 3) | (*bp++ - '0');
1213                                 byte = (byte << 3) | (*bp++ - '0');
1214                                 *dest++ = byte;
1215                                 len++;
1216                         } else {
1217                                 *dest++ = *bp++;
1218                                 len++;
1219                         }
1220                 }
1221         }
1222
1223         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1224                 return -1;
1225         while (*bp == ' ') bp++;
1226         *bpp = bp;
1227         *dest = '\0';
1228         return len;
1229 }
1230 EXPORT_SYMBOL_GPL(qword_get);
1231
1232
1233 /*
1234  * support /proc/sunrpc/cache/$CACHENAME/content
1235  * as a seqfile.
1236  * We call ->cache_show passing NULL for the item to
1237  * get a header, then pass each real item in the cache
1238  */
1239
1240 struct handle {
1241         struct cache_detail *cd;
1242 };
1243
1244 static void *c_start(struct seq_file *m, loff_t *pos)
1245         __acquires(cd->hash_lock)
1246 {
1247         loff_t n = *pos;
1248         unsigned hash, entry;
1249         struct cache_head *ch;
1250         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1251
1252
1253         read_lock(&cd->hash_lock);
1254         if (!n--)
1255                 return SEQ_START_TOKEN;
1256         hash = n >> 32;
1257         entry = n & ((1LL<<32) - 1);
1258
1259         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1260                 if (!entry--)
1261                         return ch;
1262         n &= ~((1LL<<32) - 1);
1263         do {
1264                 hash++;
1265                 n += 1LL<<32;
1266         } while(hash < cd->hash_size &&
1267                 cd->hash_table[hash]==NULL);
1268         if (hash >= cd->hash_size)
1269                 return NULL;
1270         *pos = n+1;
1271         return cd->hash_table[hash];
1272 }
1273
1274 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1275 {
1276         struct cache_head *ch = p;
1277         int hash = (*pos >> 32);
1278         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1279
1280         if (p == SEQ_START_TOKEN)
1281                 hash = 0;
1282         else if (ch->next == NULL) {
1283                 hash++;
1284                 *pos += 1LL<<32;
1285         } else {
1286                 ++*pos;
1287                 return ch->next;
1288         }
1289         *pos &= ~((1LL<<32) - 1);
1290         while (hash < cd->hash_size &&
1291                cd->hash_table[hash] == NULL) {
1292                 hash++;
1293                 *pos += 1LL<<32;
1294         }
1295         if (hash >= cd->hash_size)
1296                 return NULL;
1297         ++*pos;
1298         return cd->hash_table[hash];
1299 }
1300
1301 static void c_stop(struct seq_file *m, void *p)
1302         __releases(cd->hash_lock)
1303 {
1304         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1305         read_unlock(&cd->hash_lock);
1306 }
1307
1308 static int c_show(struct seq_file *m, void *p)
1309 {
1310         struct cache_head *cp = p;
1311         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1312
1313         if (p == SEQ_START_TOKEN)
1314                 return cd->cache_show(m, cd, NULL);
1315
1316         ifdebug(CACHE)
1317                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1318                            convert_to_wallclock(cp->expiry_time),
1319                            atomic_read(&cp->ref.refcount), cp->flags);
1320         cache_get(cp);
1321         if (cache_check(cd, cp, NULL))
1322                 /* cache_check does a cache_put on failure */
1323                 seq_printf(m, "# ");
1324         else
1325                 cache_put(cp, cd);
1326
1327         return cd->cache_show(m, cd, cp);
1328 }
1329
1330 static const struct seq_operations cache_content_op = {
1331         .start  = c_start,
1332         .next   = c_next,
1333         .stop   = c_stop,
1334         .show   = c_show,
1335 };
1336
1337 static int content_open(struct inode *inode, struct file *file,
1338                         struct cache_detail *cd)
1339 {
1340         struct handle *han;
1341
1342         if (!cd || !try_module_get(cd->owner))
1343                 return -EACCES;
1344         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1345         if (han == NULL) {
1346                 module_put(cd->owner);
1347                 return -ENOMEM;
1348         }
1349
1350         han->cd = cd;
1351         return 0;
1352 }
1353
1354 static int content_release(struct inode *inode, struct file *file,
1355                 struct cache_detail *cd)
1356 {
1357         int ret = seq_release_private(inode, file);
1358         module_put(cd->owner);
1359         return ret;
1360 }
1361
1362 static int open_flush(struct inode *inode, struct file *file,
1363                         struct cache_detail *cd)
1364 {
1365         if (!cd || !try_module_get(cd->owner))
1366                 return -EACCES;
1367         return nonseekable_open(inode, file);
1368 }
1369
1370 static int release_flush(struct inode *inode, struct file *file,
1371                         struct cache_detail *cd)
1372 {
1373         module_put(cd->owner);
1374         return 0;
1375 }
1376
1377 static ssize_t read_flush(struct file *file, char __user *buf,
1378                           size_t count, loff_t *ppos,
1379                           struct cache_detail *cd)
1380 {
1381         char tbuf[20];
1382         unsigned long p = *ppos;
1383         size_t len;
1384
1385         sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1386         len = strlen(tbuf);
1387         if (p >= len)
1388                 return 0;
1389         len -= p;
1390         if (len > count)
1391                 len = count;
1392         if (copy_to_user(buf, (void*)(tbuf+p), len))
1393                 return -EFAULT;
1394         *ppos += len;
1395         return len;
1396 }
1397
1398 static ssize_t write_flush(struct file *file, const char __user *buf,
1399                            size_t count, loff_t *ppos,
1400                            struct cache_detail *cd)
1401 {
1402         char tbuf[20];
1403         char *bp, *ep;
1404
1405         if (*ppos || count > sizeof(tbuf)-1)
1406                 return -EINVAL;
1407         if (copy_from_user(tbuf, buf, count))
1408                 return -EFAULT;
1409         tbuf[count] = 0;
1410         simple_strtoul(tbuf, &ep, 0);
1411         if (*ep && *ep != '\n')
1412                 return -EINVAL;
1413
1414         bp = tbuf;
1415         cd->flush_time = get_expiry(&bp);
1416         cd->nextcheck = seconds_since_boot();
1417         cache_flush();
1418
1419         *ppos += count;
1420         return count;
1421 }
1422
1423 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1424                                  size_t count, loff_t *ppos)
1425 {
1426         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1427
1428         return cache_read(filp, buf, count, ppos, cd);
1429 }
1430
1431 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1432                                   size_t count, loff_t *ppos)
1433 {
1434         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1435
1436         return cache_write(filp, buf, count, ppos, cd);
1437 }
1438
1439 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1440 {
1441         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1442
1443         return cache_poll(filp, wait, cd);
1444 }
1445
1446 static long cache_ioctl_procfs(struct file *filp,
1447                                unsigned int cmd, unsigned long arg)
1448 {
1449         struct inode *inode = filp->f_path.dentry->d_inode;
1450         struct cache_detail *cd = PDE(inode)->data;
1451
1452         return cache_ioctl(inode, filp, cmd, arg, cd);
1453 }
1454
1455 static int cache_open_procfs(struct inode *inode, struct file *filp)
1456 {
1457         struct cache_detail *cd = PDE(inode)->data;
1458
1459         return cache_open(inode, filp, cd);
1460 }
1461
1462 static int cache_release_procfs(struct inode *inode, struct file *filp)
1463 {
1464         struct cache_detail *cd = PDE(inode)->data;
1465
1466         return cache_release(inode, filp, cd);
1467 }
1468
1469 static const struct file_operations cache_file_operations_procfs = {
1470         .owner          = THIS_MODULE,
1471         .llseek         = no_llseek,
1472         .read           = cache_read_procfs,
1473         .write          = cache_write_procfs,
1474         .poll           = cache_poll_procfs,
1475         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1476         .open           = cache_open_procfs,
1477         .release        = cache_release_procfs,
1478 };
1479
1480 static int content_open_procfs(struct inode *inode, struct file *filp)
1481 {
1482         struct cache_detail *cd = PDE(inode)->data;
1483
1484         return content_open(inode, filp, cd);
1485 }
1486
1487 static int content_release_procfs(struct inode *inode, struct file *filp)
1488 {
1489         struct cache_detail *cd = PDE(inode)->data;
1490
1491         return content_release(inode, filp, cd);
1492 }
1493
1494 static const struct file_operations content_file_operations_procfs = {
1495         .open           = content_open_procfs,
1496         .read           = seq_read,
1497         .llseek         = seq_lseek,
1498         .release        = content_release_procfs,
1499 };
1500
1501 static int open_flush_procfs(struct inode *inode, struct file *filp)
1502 {
1503         struct cache_detail *cd = PDE(inode)->data;
1504
1505         return open_flush(inode, filp, cd);
1506 }
1507
1508 static int release_flush_procfs(struct inode *inode, struct file *filp)
1509 {
1510         struct cache_detail *cd = PDE(inode)->data;
1511
1512         return release_flush(inode, filp, cd);
1513 }
1514
1515 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1516                             size_t count, loff_t *ppos)
1517 {
1518         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1519
1520         return read_flush(filp, buf, count, ppos, cd);
1521 }
1522
1523 static ssize_t write_flush_procfs(struct file *filp,
1524                                   const char __user *buf,
1525                                   size_t count, loff_t *ppos)
1526 {
1527         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1528
1529         return write_flush(filp, buf, count, ppos, cd);
1530 }
1531
1532 static const struct file_operations cache_flush_operations_procfs = {
1533         .open           = open_flush_procfs,
1534         .read           = read_flush_procfs,
1535         .write          = write_flush_procfs,
1536         .release        = release_flush_procfs,
1537         .llseek         = no_llseek,
1538 };
1539
1540 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1541 {
1542         struct sunrpc_net *sn;
1543
1544         if (cd->u.procfs.proc_ent == NULL)
1545                 return;
1546         if (cd->u.procfs.flush_ent)
1547                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1548         if (cd->u.procfs.channel_ent)
1549                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1550         if (cd->u.procfs.content_ent)
1551                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1552         cd->u.procfs.proc_ent = NULL;
1553         sn = net_generic(net, sunrpc_net_id);
1554         remove_proc_entry(cd->name, sn->proc_net_rpc);
1555 }
1556
1557 #ifdef CONFIG_PROC_FS
1558 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1559 {
1560         struct proc_dir_entry *p;
1561         struct sunrpc_net *sn;
1562
1563         sn = net_generic(net, sunrpc_net_id);
1564         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1565         if (cd->u.procfs.proc_ent == NULL)
1566                 goto out_nomem;
1567         cd->u.procfs.channel_ent = NULL;
1568         cd->u.procfs.content_ent = NULL;
1569
1570         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1571                              cd->u.procfs.proc_ent,
1572                              &cache_flush_operations_procfs, cd);
1573         cd->u.procfs.flush_ent = p;
1574         if (p == NULL)
1575                 goto out_nomem;
1576
1577         if (cd->cache_upcall || cd->cache_parse) {
1578                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1579                                      cd->u.procfs.proc_ent,
1580                                      &cache_file_operations_procfs, cd);
1581                 cd->u.procfs.channel_ent = p;
1582                 if (p == NULL)
1583                         goto out_nomem;
1584         }
1585         if (cd->cache_show) {
1586                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1587                                 cd->u.procfs.proc_ent,
1588                                 &content_file_operations_procfs, cd);
1589                 cd->u.procfs.content_ent = p;
1590                 if (p == NULL)
1591                         goto out_nomem;
1592         }
1593         return 0;
1594 out_nomem:
1595         remove_cache_proc_entries(cd, net);
1596         return -ENOMEM;
1597 }
1598 #else /* CONFIG_PROC_FS */
1599 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1600 {
1601         return 0;
1602 }
1603 #endif
1604
1605 void __init cache_initialize(void)
1606 {
1607         INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1608 }
1609
1610 int cache_register_net(struct cache_detail *cd, struct net *net)
1611 {
1612         int ret;
1613
1614         sunrpc_init_cache_detail(cd);
1615         ret = create_cache_proc_entries(cd, net);
1616         if (ret)
1617                 sunrpc_destroy_cache_detail(cd);
1618         return ret;
1619 }
1620
1621 int cache_register(struct cache_detail *cd)
1622 {
1623         return cache_register_net(cd, &init_net);
1624 }
1625 EXPORT_SYMBOL_GPL(cache_register);
1626
1627 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1628 {
1629         remove_cache_proc_entries(cd, net);
1630         sunrpc_destroy_cache_detail(cd);
1631 }
1632
1633 void cache_unregister(struct cache_detail *cd)
1634 {
1635         cache_unregister_net(cd, &init_net);
1636 }
1637 EXPORT_SYMBOL_GPL(cache_unregister);
1638
1639 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1640                                  size_t count, loff_t *ppos)
1641 {
1642         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1643
1644         return cache_read(filp, buf, count, ppos, cd);
1645 }
1646
1647 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1648                                   size_t count, loff_t *ppos)
1649 {
1650         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1651
1652         return cache_write(filp, buf, count, ppos, cd);
1653 }
1654
1655 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1656 {
1657         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1658
1659         return cache_poll(filp, wait, cd);
1660 }
1661
1662 static long cache_ioctl_pipefs(struct file *filp,
1663                               unsigned int cmd, unsigned long arg)
1664 {
1665         struct inode *inode = filp->f_dentry->d_inode;
1666         struct cache_detail *cd = RPC_I(inode)->private;
1667
1668         return cache_ioctl(inode, filp, cmd, arg, cd);
1669 }
1670
1671 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1672 {
1673         struct cache_detail *cd = RPC_I(inode)->private;
1674
1675         return cache_open(inode, filp, cd);
1676 }
1677
1678 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1679 {
1680         struct cache_detail *cd = RPC_I(inode)->private;
1681
1682         return cache_release(inode, filp, cd);
1683 }
1684
1685 const struct file_operations cache_file_operations_pipefs = {
1686         .owner          = THIS_MODULE,
1687         .llseek         = no_llseek,
1688         .read           = cache_read_pipefs,
1689         .write          = cache_write_pipefs,
1690         .poll           = cache_poll_pipefs,
1691         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1692         .open           = cache_open_pipefs,
1693         .release        = cache_release_pipefs,
1694 };
1695
1696 static int content_open_pipefs(struct inode *inode, struct file *filp)
1697 {
1698         struct cache_detail *cd = RPC_I(inode)->private;
1699
1700         return content_open(inode, filp, cd);
1701 }
1702
1703 static int content_release_pipefs(struct inode *inode, struct file *filp)
1704 {
1705         struct cache_detail *cd = RPC_I(inode)->private;
1706
1707         return content_release(inode, filp, cd);
1708 }
1709
1710 const struct file_operations content_file_operations_pipefs = {
1711         .open           = content_open_pipefs,
1712         .read           = seq_read,
1713         .llseek         = seq_lseek,
1714         .release        = content_release_pipefs,
1715 };
1716
1717 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1718 {
1719         struct cache_detail *cd = RPC_I(inode)->private;
1720
1721         return open_flush(inode, filp, cd);
1722 }
1723
1724 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1725 {
1726         struct cache_detail *cd = RPC_I(inode)->private;
1727
1728         return release_flush(inode, filp, cd);
1729 }
1730
1731 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1732                             size_t count, loff_t *ppos)
1733 {
1734         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1735
1736         return read_flush(filp, buf, count, ppos, cd);
1737 }
1738
1739 static ssize_t write_flush_pipefs(struct file *filp,
1740                                   const char __user *buf,
1741                                   size_t count, loff_t *ppos)
1742 {
1743         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1744
1745         return write_flush(filp, buf, count, ppos, cd);
1746 }
1747
1748 const struct file_operations cache_flush_operations_pipefs = {
1749         .open           = open_flush_pipefs,
1750         .read           = read_flush_pipefs,
1751         .write          = write_flush_pipefs,
1752         .release        = release_flush_pipefs,
1753         .llseek         = no_llseek,
1754 };
1755
1756 int sunrpc_cache_register_pipefs(struct dentry *parent,
1757                                  const char *name, mode_t umode,
1758                                  struct cache_detail *cd)
1759 {
1760         struct qstr q;
1761         struct dentry *dir;
1762         int ret = 0;
1763
1764         sunrpc_init_cache_detail(cd);
1765         q.name = name;
1766         q.len = strlen(name);
1767         q.hash = full_name_hash(q.name, q.len);
1768         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1769         if (!IS_ERR(dir))
1770                 cd->u.pipefs.dir = dir;
1771         else {
1772                 sunrpc_destroy_cache_detail(cd);
1773                 ret = PTR_ERR(dir);
1774         }
1775         return ret;
1776 }
1777 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1778
1779 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1780 {
1781         rpc_remove_cache_dir(cd->u.pipefs.dir);
1782         cd->u.pipefs.dir = NULL;
1783         sunrpc_destroy_cache_detail(cd);
1784 }
1785 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1786