SUNRPC: Move procfs-specific stuff out of the generic sunrpc cache code
[safe/jmp/linux-2.6] / 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
36 #define  RPCDBG_FACILITY RPCDBG_CACHE
37
38 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
39 static void cache_revisit_request(struct cache_head *item);
40
41 static void cache_init(struct cache_head *h)
42 {
43         time_t now = get_seconds();
44         h->next = NULL;
45         h->flags = 0;
46         kref_init(&h->ref);
47         h->expiry_time = now + CACHE_NEW_EXPIRY;
48         h->last_refresh = now;
49 }
50
51 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
52                                        struct cache_head *key, int hash)
53 {
54         struct cache_head **head,  **hp;
55         struct cache_head *new = NULL;
56
57         head = &detail->hash_table[hash];
58
59         read_lock(&detail->hash_lock);
60
61         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
62                 struct cache_head *tmp = *hp;
63                 if (detail->match(tmp, key)) {
64                         cache_get(tmp);
65                         read_unlock(&detail->hash_lock);
66                         return tmp;
67                 }
68         }
69         read_unlock(&detail->hash_lock);
70         /* Didn't find anything, insert an empty entry */
71
72         new = detail->alloc();
73         if (!new)
74                 return NULL;
75         /* must fully initialise 'new', else
76          * we might get lose if we need to
77          * cache_put it soon.
78          */
79         cache_init(new);
80         detail->init(new, key);
81
82         write_lock(&detail->hash_lock);
83
84         /* check if entry appeared while we slept */
85         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
86                 struct cache_head *tmp = *hp;
87                 if (detail->match(tmp, key)) {
88                         cache_get(tmp);
89                         write_unlock(&detail->hash_lock);
90                         cache_put(new, detail);
91                         return tmp;
92                 }
93         }
94         new->next = *head;
95         *head = new;
96         detail->entries++;
97         cache_get(new);
98         write_unlock(&detail->hash_lock);
99
100         return new;
101 }
102 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
103
104
105 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
106
107 static int cache_fresh_locked(struct cache_head *head, time_t expiry)
108 {
109         head->expiry_time = expiry;
110         head->last_refresh = get_seconds();
111         return !test_and_set_bit(CACHE_VALID, &head->flags);
112 }
113
114 static void cache_fresh_unlocked(struct cache_head *head,
115                         struct cache_detail *detail, int new)
116 {
117         if (new)
118                 cache_revisit_request(head);
119         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
120                 cache_revisit_request(head);
121                 queue_loose(detail, head);
122         }
123 }
124
125 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
126                                        struct cache_head *new, struct cache_head *old, int hash)
127 {
128         /* The 'old' entry is to be replaced by 'new'.
129          * If 'old' is not VALID, we update it directly,
130          * otherwise we need to replace it
131          */
132         struct cache_head **head;
133         struct cache_head *tmp;
134         int is_new;
135
136         if (!test_bit(CACHE_VALID, &old->flags)) {
137                 write_lock(&detail->hash_lock);
138                 if (!test_bit(CACHE_VALID, &old->flags)) {
139                         if (test_bit(CACHE_NEGATIVE, &new->flags))
140                                 set_bit(CACHE_NEGATIVE, &old->flags);
141                         else
142                                 detail->update(old, new);
143                         is_new = cache_fresh_locked(old, new->expiry_time);
144                         write_unlock(&detail->hash_lock);
145                         cache_fresh_unlocked(old, detail, is_new);
146                         return old;
147                 }
148                 write_unlock(&detail->hash_lock);
149         }
150         /* We need to insert a new entry */
151         tmp = detail->alloc();
152         if (!tmp) {
153                 cache_put(old, detail);
154                 return NULL;
155         }
156         cache_init(tmp);
157         detail->init(tmp, old);
158         head = &detail->hash_table[hash];
159
160         write_lock(&detail->hash_lock);
161         if (test_bit(CACHE_NEGATIVE, &new->flags))
162                 set_bit(CACHE_NEGATIVE, &tmp->flags);
163         else
164                 detail->update(tmp, new);
165         tmp->next = *head;
166         *head = tmp;
167         detail->entries++;
168         cache_get(tmp);
169         is_new = cache_fresh_locked(tmp, new->expiry_time);
170         cache_fresh_locked(old, 0);
171         write_unlock(&detail->hash_lock);
172         cache_fresh_unlocked(tmp, detail, is_new);
173         cache_fresh_unlocked(old, detail, 0);
174         cache_put(old, detail);
175         return tmp;
176 }
177 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
178
179 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
180 {
181         if (!cd->cache_upcall)
182                 return -EINVAL;
183         return cd->cache_upcall(cd, h);
184 }
185
186 /*
187  * This is the generic cache management routine for all
188  * the authentication caches.
189  * It checks the currency of a cache item and will (later)
190  * initiate an upcall to fill it if needed.
191  *
192  *
193  * Returns 0 if the cache_head can be used, or cache_puts it and returns
194  * -EAGAIN if upcall is pending,
195  * -ETIMEDOUT if upcall failed and should be retried,
196  * -ENOENT if cache entry was negative
197  */
198 int cache_check(struct cache_detail *detail,
199                     struct cache_head *h, struct cache_req *rqstp)
200 {
201         int rv;
202         long refresh_age, age;
203
204         /* First decide return status as best we can */
205         if (!test_bit(CACHE_VALID, &h->flags) ||
206             h->expiry_time < get_seconds())
207                 rv = -EAGAIN;
208         else if (detail->flush_time > h->last_refresh)
209                 rv = -EAGAIN;
210         else {
211                 /* entry is valid */
212                 if (test_bit(CACHE_NEGATIVE, &h->flags))
213                         rv = -ENOENT;
214                 else rv = 0;
215         }
216
217         /* now see if we want to start an upcall */
218         refresh_age = (h->expiry_time - h->last_refresh);
219         age = get_seconds() - h->last_refresh;
220
221         if (rqstp == NULL) {
222                 if (rv == -EAGAIN)
223                         rv = -ENOENT;
224         } else if (rv == -EAGAIN || age > refresh_age/2) {
225                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
226                                 refresh_age, age);
227                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
228                         switch (cache_make_upcall(detail, h)) {
229                         case -EINVAL:
230                                 clear_bit(CACHE_PENDING, &h->flags);
231                                 if (rv == -EAGAIN) {
232                                         set_bit(CACHE_NEGATIVE, &h->flags);
233                                         cache_fresh_unlocked(h, detail,
234                                              cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY));
235                                         rv = -ENOENT;
236                                 }
237                                 break;
238
239                         case -EAGAIN:
240                                 clear_bit(CACHE_PENDING, &h->flags);
241                                 cache_revisit_request(h);
242                                 break;
243                         }
244                 }
245         }
246
247         if (rv == -EAGAIN)
248                 if (cache_defer_req(rqstp, h) != 0)
249                         rv = -ETIMEDOUT;
250
251         if (rv)
252                 cache_put(h, detail);
253         return rv;
254 }
255 EXPORT_SYMBOL_GPL(cache_check);
256
257 /*
258  * caches need to be periodically cleaned.
259  * For this we maintain a list of cache_detail and
260  * a current pointer into that list and into the table
261  * for that entry.
262  *
263  * Each time clean_cache is called it finds the next non-empty entry
264  * in the current table and walks the list in that entry
265  * looking for entries that can be removed.
266  *
267  * An entry gets removed if:
268  * - The expiry is before current time
269  * - The last_refresh time is before the flush_time for that cache
270  *
271  * later we might drop old entries with non-NEVER expiry if that table
272  * is getting 'full' for some definition of 'full'
273  *
274  * The question of "how often to scan a table" is an interesting one
275  * and is answered in part by the use of the "nextcheck" field in the
276  * cache_detail.
277  * When a scan of a table begins, the nextcheck field is set to a time
278  * that is well into the future.
279  * While scanning, if an expiry time is found that is earlier than the
280  * current nextcheck time, nextcheck is set to that expiry time.
281  * If the flush_time is ever set to a time earlier than the nextcheck
282  * time, the nextcheck time is then set to that flush_time.
283  *
284  * A table is then only scanned if the current time is at least
285  * the nextcheck time.
286  *
287  */
288
289 static LIST_HEAD(cache_list);
290 static DEFINE_SPINLOCK(cache_list_lock);
291 static struct cache_detail *current_detail;
292 static int current_index;
293
294 static void do_cache_clean(struct work_struct *work);
295 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
296
297 static void sunrpc_init_cache_detail(struct cache_detail *cd)
298 {
299         rwlock_init(&cd->hash_lock);
300         INIT_LIST_HEAD(&cd->queue);
301         spin_lock(&cache_list_lock);
302         cd->nextcheck = 0;
303         cd->entries = 0;
304         atomic_set(&cd->readers, 0);
305         cd->last_close = 0;
306         cd->last_warn = -1;
307         list_add(&cd->others, &cache_list);
308         spin_unlock(&cache_list_lock);
309
310         /* start the cleaning process */
311         schedule_delayed_work(&cache_cleaner, 0);
312 }
313
314 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
315 {
316         cache_purge(cd);
317         spin_lock(&cache_list_lock);
318         write_lock(&cd->hash_lock);
319         if (cd->entries || atomic_read(&cd->inuse)) {
320                 write_unlock(&cd->hash_lock);
321                 spin_unlock(&cache_list_lock);
322                 goto out;
323         }
324         if (current_detail == cd)
325                 current_detail = NULL;
326         list_del_init(&cd->others);
327         write_unlock(&cd->hash_lock);
328         spin_unlock(&cache_list_lock);
329         if (list_empty(&cache_list)) {
330                 /* module must be being unloaded so its safe to kill the worker */
331                 cancel_delayed_work_sync(&cache_cleaner);
332         }
333         return;
334 out:
335         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
336 }
337
338 /* clean cache tries to find something to clean
339  * and cleans it.
340  * It returns 1 if it cleaned something,
341  *            0 if it didn't find anything this time
342  *           -1 if it fell off the end of the list.
343  */
344 static int cache_clean(void)
345 {
346         int rv = 0;
347         struct list_head *next;
348
349         spin_lock(&cache_list_lock);
350
351         /* find a suitable table if we don't already have one */
352         while (current_detail == NULL ||
353             current_index >= current_detail->hash_size) {
354                 if (current_detail)
355                         next = current_detail->others.next;
356                 else
357                         next = cache_list.next;
358                 if (next == &cache_list) {
359                         current_detail = NULL;
360                         spin_unlock(&cache_list_lock);
361                         return -1;
362                 }
363                 current_detail = list_entry(next, struct cache_detail, others);
364                 if (current_detail->nextcheck > get_seconds())
365                         current_index = current_detail->hash_size;
366                 else {
367                         current_index = 0;
368                         current_detail->nextcheck = get_seconds()+30*60;
369                 }
370         }
371
372         /* find a non-empty bucket in the table */
373         while (current_detail &&
374                current_index < current_detail->hash_size &&
375                current_detail->hash_table[current_index] == NULL)
376                 current_index++;
377
378         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
379
380         if (current_detail && current_index < current_detail->hash_size) {
381                 struct cache_head *ch, **cp;
382                 struct cache_detail *d;
383
384                 write_lock(&current_detail->hash_lock);
385
386                 /* Ok, now to clean this strand */
387
388                 cp = & current_detail->hash_table[current_index];
389                 ch = *cp;
390                 for (; ch; cp= & ch->next, ch= *cp) {
391                         if (current_detail->nextcheck > ch->expiry_time)
392                                 current_detail->nextcheck = ch->expiry_time+1;
393                         if (ch->expiry_time >= get_seconds()
394                             && ch->last_refresh >= current_detail->flush_time
395                                 )
396                                 continue;
397                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
398                                 queue_loose(current_detail, ch);
399
400                         if (atomic_read(&ch->ref.refcount) == 1)
401                                 break;
402                 }
403                 if (ch) {
404                         *cp = ch->next;
405                         ch->next = NULL;
406                         current_detail->entries--;
407                         rv = 1;
408                 }
409                 write_unlock(&current_detail->hash_lock);
410                 d = current_detail;
411                 if (!ch)
412                         current_index ++;
413                 spin_unlock(&cache_list_lock);
414                 if (ch)
415                         cache_put(ch, d);
416         } else
417                 spin_unlock(&cache_list_lock);
418
419         return rv;
420 }
421
422 /*
423  * We want to regularly clean the cache, so we need to schedule some work ...
424  */
425 static void do_cache_clean(struct work_struct *work)
426 {
427         int delay = 5;
428         if (cache_clean() == -1)
429                 delay = round_jiffies_relative(30*HZ);
430
431         if (list_empty(&cache_list))
432                 delay = 0;
433
434         if (delay)
435                 schedule_delayed_work(&cache_cleaner, delay);
436 }
437
438
439 /*
440  * Clean all caches promptly.  This just calls cache_clean
441  * repeatedly until we are sure that every cache has had a chance to
442  * be fully cleaned
443  */
444 void cache_flush(void)
445 {
446         while (cache_clean() != -1)
447                 cond_resched();
448         while (cache_clean() != -1)
449                 cond_resched();
450 }
451 EXPORT_SYMBOL_GPL(cache_flush);
452
453 void cache_purge(struct cache_detail *detail)
454 {
455         detail->flush_time = LONG_MAX;
456         detail->nextcheck = get_seconds();
457         cache_flush();
458         detail->flush_time = 1;
459 }
460 EXPORT_SYMBOL_GPL(cache_purge);
461
462
463 /*
464  * Deferral and Revisiting of Requests.
465  *
466  * If a cache lookup finds a pending entry, we
467  * need to defer the request and revisit it later.
468  * All deferred requests are stored in a hash table,
469  * indexed by "struct cache_head *".
470  * As it may be wasteful to store a whole request
471  * structure, we allow the request to provide a
472  * deferred form, which must contain a
473  * 'struct cache_deferred_req'
474  * This cache_deferred_req contains a method to allow
475  * it to be revisited when cache info is available
476  */
477
478 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
479 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
480
481 #define DFR_MAX 300     /* ??? */
482
483 static DEFINE_SPINLOCK(cache_defer_lock);
484 static LIST_HEAD(cache_defer_list);
485 static struct list_head cache_defer_hash[DFR_HASHSIZE];
486 static int cache_defer_cnt;
487
488 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
489 {
490         struct cache_deferred_req *dreq;
491         int hash = DFR_HASH(item);
492
493         if (cache_defer_cnt >= DFR_MAX) {
494                 /* too much in the cache, randomly drop this one,
495                  * or continue and drop the oldest below
496                  */
497                 if (net_random()&1)
498                         return -ETIMEDOUT;
499         }
500         dreq = req->defer(req);
501         if (dreq == NULL)
502                 return -ETIMEDOUT;
503
504         dreq->item = item;
505
506         spin_lock(&cache_defer_lock);
507
508         list_add(&dreq->recent, &cache_defer_list);
509
510         if (cache_defer_hash[hash].next == NULL)
511                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
512         list_add(&dreq->hash, &cache_defer_hash[hash]);
513
514         /* it is in, now maybe clean up */
515         dreq = NULL;
516         if (++cache_defer_cnt > DFR_MAX) {
517                 dreq = list_entry(cache_defer_list.prev,
518                                   struct cache_deferred_req, recent);
519                 list_del(&dreq->recent);
520                 list_del(&dreq->hash);
521                 cache_defer_cnt--;
522         }
523         spin_unlock(&cache_defer_lock);
524
525         if (dreq) {
526                 /* there was one too many */
527                 dreq->revisit(dreq, 1);
528         }
529         if (!test_bit(CACHE_PENDING, &item->flags)) {
530                 /* must have just been validated... */
531                 cache_revisit_request(item);
532         }
533         return 0;
534 }
535
536 static void cache_revisit_request(struct cache_head *item)
537 {
538         struct cache_deferred_req *dreq;
539         struct list_head pending;
540
541         struct list_head *lp;
542         int hash = DFR_HASH(item);
543
544         INIT_LIST_HEAD(&pending);
545         spin_lock(&cache_defer_lock);
546
547         lp = cache_defer_hash[hash].next;
548         if (lp) {
549                 while (lp != &cache_defer_hash[hash]) {
550                         dreq = list_entry(lp, struct cache_deferred_req, hash);
551                         lp = lp->next;
552                         if (dreq->item == item) {
553                                 list_del(&dreq->hash);
554                                 list_move(&dreq->recent, &pending);
555                                 cache_defer_cnt--;
556                         }
557                 }
558         }
559         spin_unlock(&cache_defer_lock);
560
561         while (!list_empty(&pending)) {
562                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
563                 list_del_init(&dreq->recent);
564                 dreq->revisit(dreq, 0);
565         }
566 }
567
568 void cache_clean_deferred(void *owner)
569 {
570         struct cache_deferred_req *dreq, *tmp;
571         struct list_head pending;
572
573
574         INIT_LIST_HEAD(&pending);
575         spin_lock(&cache_defer_lock);
576
577         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
578                 if (dreq->owner == owner) {
579                         list_del(&dreq->hash);
580                         list_move(&dreq->recent, &pending);
581                         cache_defer_cnt--;
582                 }
583         }
584         spin_unlock(&cache_defer_lock);
585
586         while (!list_empty(&pending)) {
587                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
588                 list_del_init(&dreq->recent);
589                 dreq->revisit(dreq, 1);
590         }
591 }
592
593 /*
594  * communicate with user-space
595  *
596  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
597  * On read, you get a full request, or block.
598  * On write, an update request is processed.
599  * Poll works if anything to read, and always allows write.
600  *
601  * Implemented by linked list of requests.  Each open file has
602  * a ->private that also exists in this list.  New requests are added
603  * to the end and may wakeup and preceding readers.
604  * New readers are added to the head.  If, on read, an item is found with
605  * CACHE_UPCALLING clear, we free it from the list.
606  *
607  */
608
609 static DEFINE_SPINLOCK(queue_lock);
610 static DEFINE_MUTEX(queue_io_mutex);
611
612 struct cache_queue {
613         struct list_head        list;
614         int                     reader; /* if 0, then request */
615 };
616 struct cache_request {
617         struct cache_queue      q;
618         struct cache_head       *item;
619         char                    * buf;
620         int                     len;
621         int                     readers;
622 };
623 struct cache_reader {
624         struct cache_queue      q;
625         int                     offset; /* if non-0, we have a refcnt on next request */
626 };
627
628 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
629                           loff_t *ppos, struct cache_detail *cd)
630 {
631         struct cache_reader *rp = filp->private_data;
632         struct cache_request *rq;
633         struct inode *inode = filp->f_path.dentry->d_inode;
634         int err;
635
636         if (count == 0)
637                 return 0;
638
639         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
640                               * readers on this file */
641  again:
642         spin_lock(&queue_lock);
643         /* need to find next request */
644         while (rp->q.list.next != &cd->queue &&
645                list_entry(rp->q.list.next, struct cache_queue, list)
646                ->reader) {
647                 struct list_head *next = rp->q.list.next;
648                 list_move(&rp->q.list, next);
649         }
650         if (rp->q.list.next == &cd->queue) {
651                 spin_unlock(&queue_lock);
652                 mutex_unlock(&inode->i_mutex);
653                 BUG_ON(rp->offset);
654                 return 0;
655         }
656         rq = container_of(rp->q.list.next, struct cache_request, q.list);
657         BUG_ON(rq->q.reader);
658         if (rp->offset == 0)
659                 rq->readers++;
660         spin_unlock(&queue_lock);
661
662         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
663                 err = -EAGAIN;
664                 spin_lock(&queue_lock);
665                 list_move(&rp->q.list, &rq->q.list);
666                 spin_unlock(&queue_lock);
667         } else {
668                 if (rp->offset + count > rq->len)
669                         count = rq->len - rp->offset;
670                 err = -EFAULT;
671                 if (copy_to_user(buf, rq->buf + rp->offset, count))
672                         goto out;
673                 rp->offset += count;
674                 if (rp->offset >= rq->len) {
675                         rp->offset = 0;
676                         spin_lock(&queue_lock);
677                         list_move(&rp->q.list, &rq->q.list);
678                         spin_unlock(&queue_lock);
679                 }
680                 err = 0;
681         }
682  out:
683         if (rp->offset == 0) {
684                 /* need to release rq */
685                 spin_lock(&queue_lock);
686                 rq->readers--;
687                 if (rq->readers == 0 &&
688                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
689                         list_del(&rq->q.list);
690                         spin_unlock(&queue_lock);
691                         cache_put(rq->item, cd);
692                         kfree(rq->buf);
693                         kfree(rq);
694                 } else
695                         spin_unlock(&queue_lock);
696         }
697         if (err == -EAGAIN)
698                 goto again;
699         mutex_unlock(&inode->i_mutex);
700         return err ? err :  count;
701 }
702
703 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
704                                  size_t count, struct cache_detail *cd)
705 {
706         ssize_t ret;
707
708         if (copy_from_user(kaddr, buf, count))
709                 return -EFAULT;
710         kaddr[count] = '\0';
711         ret = cd->cache_parse(cd, kaddr, count);
712         if (!ret)
713                 ret = count;
714         return ret;
715 }
716
717 static ssize_t cache_slow_downcall(const char __user *buf,
718                                    size_t count, struct cache_detail *cd)
719 {
720         static char write_buf[8192]; /* protected by queue_io_mutex */
721         ssize_t ret = -EINVAL;
722
723         if (count >= sizeof(write_buf))
724                 goto out;
725         mutex_lock(&queue_io_mutex);
726         ret = cache_do_downcall(write_buf, buf, count, cd);
727         mutex_unlock(&queue_io_mutex);
728 out:
729         return ret;
730 }
731
732 static ssize_t cache_downcall(struct address_space *mapping,
733                               const char __user *buf,
734                               size_t count, struct cache_detail *cd)
735 {
736         struct page *page;
737         char *kaddr;
738         ssize_t ret = -ENOMEM;
739
740         if (count >= PAGE_CACHE_SIZE)
741                 goto out_slow;
742
743         page = find_or_create_page(mapping, 0, GFP_KERNEL);
744         if (!page)
745                 goto out_slow;
746
747         kaddr = kmap(page);
748         ret = cache_do_downcall(kaddr, buf, count, cd);
749         kunmap(page);
750         unlock_page(page);
751         page_cache_release(page);
752         return ret;
753 out_slow:
754         return cache_slow_downcall(buf, count, cd);
755 }
756
757 static ssize_t cache_write(struct file *filp, const char __user *buf,
758                            size_t count, loff_t *ppos,
759                            struct cache_detail *cd)
760 {
761         struct address_space *mapping = filp->f_mapping;
762         struct inode *inode = filp->f_path.dentry->d_inode;
763         ssize_t ret = -EINVAL;
764
765         if (!cd->cache_parse)
766                 goto out;
767
768         mutex_lock(&inode->i_mutex);
769         ret = cache_downcall(mapping, buf, count, cd);
770         mutex_unlock(&inode->i_mutex);
771 out:
772         return ret;
773 }
774
775 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
776
777 static unsigned int cache_poll(struct file *filp, poll_table *wait,
778                                struct cache_detail *cd)
779 {
780         unsigned int mask;
781         struct cache_reader *rp = filp->private_data;
782         struct cache_queue *cq;
783
784         poll_wait(filp, &queue_wait, wait);
785
786         /* alway allow write */
787         mask = POLL_OUT | POLLWRNORM;
788
789         if (!rp)
790                 return mask;
791
792         spin_lock(&queue_lock);
793
794         for (cq= &rp->q; &cq->list != &cd->queue;
795              cq = list_entry(cq->list.next, struct cache_queue, list))
796                 if (!cq->reader) {
797                         mask |= POLLIN | POLLRDNORM;
798                         break;
799                 }
800         spin_unlock(&queue_lock);
801         return mask;
802 }
803
804 static int cache_ioctl(struct inode *ino, struct file *filp,
805                        unsigned int cmd, unsigned long arg,
806                        struct cache_detail *cd)
807 {
808         int len = 0;
809         struct cache_reader *rp = filp->private_data;
810         struct cache_queue *cq;
811
812         if (cmd != FIONREAD || !rp)
813                 return -EINVAL;
814
815         spin_lock(&queue_lock);
816
817         /* only find the length remaining in current request,
818          * or the length of the next request
819          */
820         for (cq= &rp->q; &cq->list != &cd->queue;
821              cq = list_entry(cq->list.next, struct cache_queue, list))
822                 if (!cq->reader) {
823                         struct cache_request *cr =
824                                 container_of(cq, struct cache_request, q);
825                         len = cr->len - rp->offset;
826                         break;
827                 }
828         spin_unlock(&queue_lock);
829
830         return put_user(len, (int __user *)arg);
831 }
832
833 static int cache_open(struct inode *inode, struct file *filp,
834                       struct cache_detail *cd)
835 {
836         struct cache_reader *rp = NULL;
837
838         nonseekable_open(inode, filp);
839         if (filp->f_mode & FMODE_READ) {
840                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
841                 if (!rp)
842                         return -ENOMEM;
843                 rp->offset = 0;
844                 rp->q.reader = 1;
845                 atomic_inc(&cd->readers);
846                 spin_lock(&queue_lock);
847                 list_add(&rp->q.list, &cd->queue);
848                 spin_unlock(&queue_lock);
849         }
850         filp->private_data = rp;
851         return 0;
852 }
853
854 static int cache_release(struct inode *inode, struct file *filp,
855                          struct cache_detail *cd)
856 {
857         struct cache_reader *rp = filp->private_data;
858
859         if (rp) {
860                 spin_lock(&queue_lock);
861                 if (rp->offset) {
862                         struct cache_queue *cq;
863                         for (cq= &rp->q; &cq->list != &cd->queue;
864                              cq = list_entry(cq->list.next, struct cache_queue, list))
865                                 if (!cq->reader) {
866                                         container_of(cq, struct cache_request, q)
867                                                 ->readers--;
868                                         break;
869                                 }
870                         rp->offset = 0;
871                 }
872                 list_del(&rp->q.list);
873                 spin_unlock(&queue_lock);
874
875                 filp->private_data = NULL;
876                 kfree(rp);
877
878                 cd->last_close = get_seconds();
879                 atomic_dec(&cd->readers);
880         }
881         return 0;
882 }
883
884
885
886 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
887 {
888         struct cache_queue *cq;
889         spin_lock(&queue_lock);
890         list_for_each_entry(cq, &detail->queue, list)
891                 if (!cq->reader) {
892                         struct cache_request *cr = container_of(cq, struct cache_request, q);
893                         if (cr->item != ch)
894                                 continue;
895                         if (cr->readers != 0)
896                                 continue;
897                         list_del(&cr->q.list);
898                         spin_unlock(&queue_lock);
899                         cache_put(cr->item, detail);
900                         kfree(cr->buf);
901                         kfree(cr);
902                         return;
903                 }
904         spin_unlock(&queue_lock);
905 }
906
907 /*
908  * Support routines for text-based upcalls.
909  * Fields are separated by spaces.
910  * Fields are either mangled to quote space tab newline slosh with slosh
911  * or a hexified with a leading \x
912  * Record is terminated with newline.
913  *
914  */
915
916 void qword_add(char **bpp, int *lp, char *str)
917 {
918         char *bp = *bpp;
919         int len = *lp;
920         char c;
921
922         if (len < 0) return;
923
924         while ((c=*str++) && len)
925                 switch(c) {
926                 case ' ':
927                 case '\t':
928                 case '\n':
929                 case '\\':
930                         if (len >= 4) {
931                                 *bp++ = '\\';
932                                 *bp++ = '0' + ((c & 0300)>>6);
933                                 *bp++ = '0' + ((c & 0070)>>3);
934                                 *bp++ = '0' + ((c & 0007)>>0);
935                         }
936                         len -= 4;
937                         break;
938                 default:
939                         *bp++ = c;
940                         len--;
941                 }
942         if (c || len <1) len = -1;
943         else {
944                 *bp++ = ' ';
945                 len--;
946         }
947         *bpp = bp;
948         *lp = len;
949 }
950 EXPORT_SYMBOL_GPL(qword_add);
951
952 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
953 {
954         char *bp = *bpp;
955         int len = *lp;
956
957         if (len < 0) return;
958
959         if (len > 2) {
960                 *bp++ = '\\';
961                 *bp++ = 'x';
962                 len -= 2;
963                 while (blen && len >= 2) {
964                         unsigned char c = *buf++;
965                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
966                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
967                         len -= 2;
968                         blen--;
969                 }
970         }
971         if (blen || len<1) len = -1;
972         else {
973                 *bp++ = ' ';
974                 len--;
975         }
976         *bpp = bp;
977         *lp = len;
978 }
979 EXPORT_SYMBOL_GPL(qword_addhex);
980
981 static void warn_no_listener(struct cache_detail *detail)
982 {
983         if (detail->last_warn != detail->last_close) {
984                 detail->last_warn = detail->last_close;
985                 if (detail->warn_no_listener)
986                         detail->warn_no_listener(detail, detail->last_close != 0);
987         }
988 }
989
990 /*
991  * register an upcall request to user-space and queue it up for read() by the
992  * upcall daemon.
993  *
994  * Each request is at most one page long.
995  */
996 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
997                 void (*cache_request)(struct cache_detail *,
998                                       struct cache_head *,
999                                       char **,
1000                                       int *))
1001 {
1002
1003         char *buf;
1004         struct cache_request *crq;
1005         char *bp;
1006         int len;
1007
1008         if (atomic_read(&detail->readers) == 0 &&
1009             detail->last_close < get_seconds() - 30) {
1010                         warn_no_listener(detail);
1011                         return -EINVAL;
1012         }
1013
1014         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1015         if (!buf)
1016                 return -EAGAIN;
1017
1018         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1019         if (!crq) {
1020                 kfree(buf);
1021                 return -EAGAIN;
1022         }
1023
1024         bp = buf; len = PAGE_SIZE;
1025
1026         cache_request(detail, h, &bp, &len);
1027
1028         if (len < 0) {
1029                 kfree(buf);
1030                 kfree(crq);
1031                 return -EAGAIN;
1032         }
1033         crq->q.reader = 0;
1034         crq->item = cache_get(h);
1035         crq->buf = buf;
1036         crq->len = PAGE_SIZE - len;
1037         crq->readers = 0;
1038         spin_lock(&queue_lock);
1039         list_add_tail(&crq->q.list, &detail->queue);
1040         spin_unlock(&queue_lock);
1041         wake_up(&queue_wait);
1042         return 0;
1043 }
1044 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1045
1046 /*
1047  * parse a message from user-space and pass it
1048  * to an appropriate cache
1049  * Messages are, like requests, separated into fields by
1050  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1051  *
1052  * Message is
1053  *   reply cachename expiry key ... content....
1054  *
1055  * key and content are both parsed by cache
1056  */
1057
1058 #define isodigit(c) (isdigit(c) && c <= '7')
1059 int qword_get(char **bpp, char *dest, int bufsize)
1060 {
1061         /* return bytes copied, or -1 on error */
1062         char *bp = *bpp;
1063         int len = 0;
1064
1065         while (*bp == ' ') bp++;
1066
1067         if (bp[0] == '\\' && bp[1] == 'x') {
1068                 /* HEX STRING */
1069                 bp += 2;
1070                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1071                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1072                         bp++;
1073                         byte <<= 4;
1074                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1075                         *dest++ = byte;
1076                         bp++;
1077                         len++;
1078                 }
1079         } else {
1080                 /* text with \nnn octal quoting */
1081                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1082                         if (*bp == '\\' &&
1083                             isodigit(bp[1]) && (bp[1] <= '3') &&
1084                             isodigit(bp[2]) &&
1085                             isodigit(bp[3])) {
1086                                 int byte = (*++bp -'0');
1087                                 bp++;
1088                                 byte = (byte << 3) | (*bp++ - '0');
1089                                 byte = (byte << 3) | (*bp++ - '0');
1090                                 *dest++ = byte;
1091                                 len++;
1092                         } else {
1093                                 *dest++ = *bp++;
1094                                 len++;
1095                         }
1096                 }
1097         }
1098
1099         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1100                 return -1;
1101         while (*bp == ' ') bp++;
1102         *bpp = bp;
1103         *dest = '\0';
1104         return len;
1105 }
1106 EXPORT_SYMBOL_GPL(qword_get);
1107
1108
1109 /*
1110  * support /proc/sunrpc/cache/$CACHENAME/content
1111  * as a seqfile.
1112  * We call ->cache_show passing NULL for the item to
1113  * get a header, then pass each real item in the cache
1114  */
1115
1116 struct handle {
1117         struct cache_detail *cd;
1118 };
1119
1120 static void *c_start(struct seq_file *m, loff_t *pos)
1121         __acquires(cd->hash_lock)
1122 {
1123         loff_t n = *pos;
1124         unsigned hash, entry;
1125         struct cache_head *ch;
1126         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1127
1128
1129         read_lock(&cd->hash_lock);
1130         if (!n--)
1131                 return SEQ_START_TOKEN;
1132         hash = n >> 32;
1133         entry = n & ((1LL<<32) - 1);
1134
1135         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1136                 if (!entry--)
1137                         return ch;
1138         n &= ~((1LL<<32) - 1);
1139         do {
1140                 hash++;
1141                 n += 1LL<<32;
1142         } while(hash < cd->hash_size &&
1143                 cd->hash_table[hash]==NULL);
1144         if (hash >= cd->hash_size)
1145                 return NULL;
1146         *pos = n+1;
1147         return cd->hash_table[hash];
1148 }
1149
1150 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1151 {
1152         struct cache_head *ch = p;
1153         int hash = (*pos >> 32);
1154         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1155
1156         if (p == SEQ_START_TOKEN)
1157                 hash = 0;
1158         else if (ch->next == NULL) {
1159                 hash++;
1160                 *pos += 1LL<<32;
1161         } else {
1162                 ++*pos;
1163                 return ch->next;
1164         }
1165         *pos &= ~((1LL<<32) - 1);
1166         while (hash < cd->hash_size &&
1167                cd->hash_table[hash] == NULL) {
1168                 hash++;
1169                 *pos += 1LL<<32;
1170         }
1171         if (hash >= cd->hash_size)
1172                 return NULL;
1173         ++*pos;
1174         return cd->hash_table[hash];
1175 }
1176
1177 static void c_stop(struct seq_file *m, void *p)
1178         __releases(cd->hash_lock)
1179 {
1180         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1181         read_unlock(&cd->hash_lock);
1182 }
1183
1184 static int c_show(struct seq_file *m, void *p)
1185 {
1186         struct cache_head *cp = p;
1187         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1188
1189         if (p == SEQ_START_TOKEN)
1190                 return cd->cache_show(m, cd, NULL);
1191
1192         ifdebug(CACHE)
1193                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1194                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1195         cache_get(cp);
1196         if (cache_check(cd, cp, NULL))
1197                 /* cache_check does a cache_put on failure */
1198                 seq_printf(m, "# ");
1199         else
1200                 cache_put(cp, cd);
1201
1202         return cd->cache_show(m, cd, cp);
1203 }
1204
1205 static const struct seq_operations cache_content_op = {
1206         .start  = c_start,
1207         .next   = c_next,
1208         .stop   = c_stop,
1209         .show   = c_show,
1210 };
1211
1212 static int content_open(struct inode *inode, struct file *file,
1213                         struct cache_detail *cd)
1214 {
1215         struct handle *han;
1216
1217         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1218         if (han == NULL)
1219                 return -ENOMEM;
1220
1221         han->cd = cd;
1222         return 0;
1223 }
1224
1225 static ssize_t read_flush(struct file *file, char __user *buf,
1226                           size_t count, loff_t *ppos,
1227                           struct cache_detail *cd)
1228 {
1229         char tbuf[20];
1230         unsigned long p = *ppos;
1231         size_t len;
1232
1233         sprintf(tbuf, "%lu\n", cd->flush_time);
1234         len = strlen(tbuf);
1235         if (p >= len)
1236                 return 0;
1237         len -= p;
1238         if (len > count)
1239                 len = count;
1240         if (copy_to_user(buf, (void*)(tbuf+p), len))
1241                 return -EFAULT;
1242         *ppos += len;
1243         return len;
1244 }
1245
1246 static ssize_t write_flush(struct file *file, const char __user *buf,
1247                            size_t count, loff_t *ppos,
1248                            struct cache_detail *cd)
1249 {
1250         char tbuf[20];
1251         char *ep;
1252         long flushtime;
1253         if (*ppos || count > sizeof(tbuf)-1)
1254                 return -EINVAL;
1255         if (copy_from_user(tbuf, buf, count))
1256                 return -EFAULT;
1257         tbuf[count] = 0;
1258         flushtime = simple_strtoul(tbuf, &ep, 0);
1259         if (*ep && *ep != '\n')
1260                 return -EINVAL;
1261
1262         cd->flush_time = flushtime;
1263         cd->nextcheck = get_seconds();
1264         cache_flush();
1265
1266         *ppos += count;
1267         return count;
1268 }
1269
1270 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1271                                  size_t count, loff_t *ppos)
1272 {
1273         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1274
1275         return cache_read(filp, buf, count, ppos, cd);
1276 }
1277
1278 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1279                                   size_t count, loff_t *ppos)
1280 {
1281         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1282
1283         return cache_write(filp, buf, count, ppos, cd);
1284 }
1285
1286 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1287 {
1288         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1289
1290         return cache_poll(filp, wait, cd);
1291 }
1292
1293 static int cache_ioctl_procfs(struct inode *inode, struct file *filp,
1294                               unsigned int cmd, unsigned long arg)
1295 {
1296         struct cache_detail *cd = PDE(inode)->data;
1297
1298         return cache_ioctl(inode, filp, cmd, arg, cd);
1299 }
1300
1301 static int cache_open_procfs(struct inode *inode, struct file *filp)
1302 {
1303         struct cache_detail *cd = PDE(inode)->data;
1304
1305         return cache_open(inode, filp, cd);
1306 }
1307
1308 static int cache_release_procfs(struct inode *inode, struct file *filp)
1309 {
1310         struct cache_detail *cd = PDE(inode)->data;
1311
1312         return cache_release(inode, filp, cd);
1313 }
1314
1315 static const struct file_operations cache_file_operations_procfs = {
1316         .owner          = THIS_MODULE,
1317         .llseek         = no_llseek,
1318         .read           = cache_read_procfs,
1319         .write          = cache_write_procfs,
1320         .poll           = cache_poll_procfs,
1321         .ioctl          = cache_ioctl_procfs, /* for FIONREAD */
1322         .open           = cache_open_procfs,
1323         .release        = cache_release_procfs,
1324 };
1325
1326 static int content_open_procfs(struct inode *inode, struct file *filp)
1327 {
1328         struct cache_detail *cd = PDE(inode)->data;
1329
1330         return content_open(inode, filp, cd);
1331 }
1332
1333 static const struct file_operations content_file_operations_procfs = {
1334         .open           = content_open_procfs,
1335         .read           = seq_read,
1336         .llseek         = seq_lseek,
1337         .release        = seq_release_private,
1338 };
1339
1340 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1341                             size_t count, loff_t *ppos)
1342 {
1343         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1344
1345         return read_flush(filp, buf, count, ppos, cd);
1346 }
1347
1348 static ssize_t write_flush_procfs(struct file *filp,
1349                                   const char __user *buf,
1350                                   size_t count, loff_t *ppos)
1351 {
1352         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1353
1354         return write_flush(filp, buf, count, ppos, cd);
1355 }
1356
1357 static const struct file_operations cache_flush_operations_procfs = {
1358         .open           = nonseekable_open,
1359         .read           = read_flush_procfs,
1360         .write          = write_flush_procfs,
1361 };
1362
1363 static void remove_cache_proc_entries(struct cache_detail *cd)
1364 {
1365         if (cd->u.procfs.proc_ent == NULL)
1366                 return;
1367         if (cd->u.procfs.flush_ent)
1368                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1369         if (cd->u.procfs.channel_ent)
1370                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1371         if (cd->u.procfs.content_ent)
1372                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1373         cd->u.procfs.proc_ent = NULL;
1374         remove_proc_entry(cd->name, proc_net_rpc);
1375 }
1376
1377 #ifdef CONFIG_PROC_FS
1378 static int create_cache_proc_entries(struct cache_detail *cd)
1379 {
1380         struct proc_dir_entry *p;
1381
1382         cd->u.procfs.proc_ent = proc_mkdir(cd->name, proc_net_rpc);
1383         if (cd->u.procfs.proc_ent == NULL)
1384                 goto out_nomem;
1385         cd->u.procfs.channel_ent = NULL;
1386         cd->u.procfs.content_ent = NULL;
1387
1388         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1389                              cd->u.procfs.proc_ent,
1390                              &cache_flush_operations_procfs, cd);
1391         cd->u.procfs.flush_ent = p;
1392         if (p == NULL)
1393                 goto out_nomem;
1394
1395         if (cd->cache_upcall || cd->cache_parse) {
1396                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1397                                      cd->u.procfs.proc_ent,
1398                                      &cache_file_operations_procfs, cd);
1399                 cd->u.procfs.channel_ent = p;
1400                 if (p == NULL)
1401                         goto out_nomem;
1402         }
1403         if (cd->cache_show) {
1404                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1405                                 cd->u.procfs.proc_ent,
1406                                 &content_file_operations_procfs, cd);
1407                 cd->u.procfs.content_ent = p;
1408                 if (p == NULL)
1409                         goto out_nomem;
1410         }
1411         return 0;
1412 out_nomem:
1413         remove_cache_proc_entries(cd);
1414         return -ENOMEM;
1415 }
1416 #else /* CONFIG_PROC_FS */
1417 static int create_cache_proc_entries(struct cache_detail *cd)
1418 {
1419         return 0;
1420 }
1421 #endif
1422
1423 int cache_register(struct cache_detail *cd)
1424 {
1425         int ret;
1426
1427         sunrpc_init_cache_detail(cd);
1428         ret = create_cache_proc_entries(cd);
1429         if (ret)
1430                 sunrpc_destroy_cache_detail(cd);
1431         return ret;
1432 }
1433 EXPORT_SYMBOL_GPL(cache_register);
1434
1435 void cache_unregister(struct cache_detail *cd)
1436 {
1437         remove_cache_proc_entries(cd);
1438         sunrpc_destroy_cache_detail(cd);
1439 }
1440 EXPORT_SYMBOL_GPL(cache_unregister);