824e8534e02210bc755b9a4b37b72067984caa4e
[linux-2.6.git] / net / sunrpc / cache.c
1 /*
2  * net/sunrpc/cache.c
3  *
4  * Generic code for various authentication-related caches
5  * used by sunrpc clients and servers.
6  *
7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
8  *
9  * Released under terms in GPL version 2.  See COPYING.
10  *
11  */
12
13 #include <linux/types.h>
14 #include <linux/fs.h>
15 #include <linux/file.h>
16 #include <linux/slab.h>
17 #include <linux/signal.h>
18 #include <linux/sched.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/module.h>
22 #include <linux/ctype.h>
23 #include <asm/uaccess.h>
24 #include <linux/poll.h>
25 #include <linux/seq_file.h>
26 #include <linux/proc_fs.h>
27 #include <linux/net.h>
28 #include <linux/workqueue.h>
29 #include <linux/mutex.h>
30 #include <asm/ioctls.h>
31 #include <linux/sunrpc/types.h>
32 #include <linux/sunrpc/cache.h>
33 #include <linux/sunrpc/stats.h>
34
35 #define  RPCDBG_FACILITY RPCDBG_CACHE
36
37 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
38 static void cache_revisit_request(struct cache_head *item);
39
40 static void cache_init(struct cache_head *h)
41 {
42         time_t now = get_seconds();
43         h->next = NULL;
44         h->flags = 0;
45         kref_init(&h->ref);
46         h->expiry_time = now + CACHE_NEW_EXPIRY;
47         h->last_refresh = now;
48 }
49
50 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
51                                        struct cache_head *key, int hash)
52 {
53         struct cache_head **head,  **hp;
54         struct cache_head *new = NULL;
55
56         head = &detail->hash_table[hash];
57
58         read_lock(&detail->hash_lock);
59
60         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
61                 struct cache_head *tmp = *hp;
62                 if (detail->match(tmp, key)) {
63                         cache_get(tmp);
64                         read_unlock(&detail->hash_lock);
65                         return tmp;
66                 }
67         }
68         read_unlock(&detail->hash_lock);
69         /* Didn't find anything, insert an empty entry */
70
71         new = detail->alloc();
72         if (!new)
73                 return NULL;
74         /* must fully initialise 'new', else
75          * we might get lose if we need to
76          * cache_put it soon.
77          */
78         cache_init(new);
79         detail->init(new, key);
80
81         write_lock(&detail->hash_lock);
82
83         /* check if entry appeared while we slept */
84         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
85                 struct cache_head *tmp = *hp;
86                 if (detail->match(tmp, key)) {
87                         cache_get(tmp);
88                         write_unlock(&detail->hash_lock);
89                         cache_put(new, detail);
90                         return tmp;
91                 }
92         }
93         new->next = *head;
94         *head = new;
95         detail->entries++;
96         cache_get(new);
97         write_unlock(&detail->hash_lock);
98
99         return new;
100 }
101 EXPORT_SYMBOL(sunrpc_cache_lookup);
102
103
104 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
105
106 static int cache_fresh_locked(struct cache_head *head, time_t expiry)
107 {
108         head->expiry_time = expiry;
109         head->last_refresh = get_seconds();
110         return !test_and_set_bit(CACHE_VALID, &head->flags);
111 }
112
113 static void cache_fresh_unlocked(struct cache_head *head,
114                         struct cache_detail *detail, int new)
115 {
116         if (new)
117                 cache_revisit_request(head);
118         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
119                 cache_revisit_request(head);
120                 queue_loose(detail, head);
121         }
122 }
123
124 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
125                                        struct cache_head *new, struct cache_head *old, int hash)
126 {
127         /* The 'old' entry is to be replaced by 'new'.
128          * If 'old' is not VALID, we update it directly,
129          * otherwise we need to replace it
130          */
131         struct cache_head **head;
132         struct cache_head *tmp;
133         int is_new;
134
135         if (!test_bit(CACHE_VALID, &old->flags)) {
136                 write_lock(&detail->hash_lock);
137                 if (!test_bit(CACHE_VALID, &old->flags)) {
138                         if (test_bit(CACHE_NEGATIVE, &new->flags))
139                                 set_bit(CACHE_NEGATIVE, &old->flags);
140                         else
141                                 detail->update(old, new);
142                         is_new = cache_fresh_locked(old, new->expiry_time);
143                         write_unlock(&detail->hash_lock);
144                         cache_fresh_unlocked(old, detail, is_new);
145                         return old;
146                 }
147                 write_unlock(&detail->hash_lock);
148         }
149         /* We need to insert a new entry */
150         tmp = detail->alloc();
151         if (!tmp) {
152                 cache_put(old, detail);
153                 return NULL;
154         }
155         cache_init(tmp);
156         detail->init(tmp, old);
157         head = &detail->hash_table[hash];
158
159         write_lock(&detail->hash_lock);
160         if (test_bit(CACHE_NEGATIVE, &new->flags))
161                 set_bit(CACHE_NEGATIVE, &tmp->flags);
162         else
163                 detail->update(tmp, new);
164         tmp->next = *head;
165         *head = tmp;
166         detail->entries++;
167         cache_get(tmp);
168         is_new = cache_fresh_locked(tmp, new->expiry_time);
169         cache_fresh_locked(old, 0);
170         write_unlock(&detail->hash_lock);
171         cache_fresh_unlocked(tmp, detail, is_new);
172         cache_fresh_unlocked(old, detail, 0);
173         cache_put(old, detail);
174         return tmp;
175 }
176 EXPORT_SYMBOL(sunrpc_cache_update);
177
178 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
179 /*
180  * This is the generic cache management routine for all
181  * the authentication caches.
182  * It checks the currency of a cache item and will (later)
183  * initiate an upcall to fill it if needed.
184  *
185  *
186  * Returns 0 if the cache_head can be used, or cache_puts it and returns
187  * -EAGAIN if upcall is pending,
188  * -ETIMEDOUT if upcall failed and should be retried,
189  * -ENOENT if cache entry was negative
190  */
191 int cache_check(struct cache_detail *detail,
192                     struct cache_head *h, struct cache_req *rqstp)
193 {
194         int rv;
195         long refresh_age, age;
196
197         /* First decide return status as best we can */
198         if (!test_bit(CACHE_VALID, &h->flags) ||
199             h->expiry_time < get_seconds())
200                 rv = -EAGAIN;
201         else if (detail->flush_time > h->last_refresh)
202                 rv = -EAGAIN;
203         else {
204                 /* entry is valid */
205                 if (test_bit(CACHE_NEGATIVE, &h->flags))
206                         rv = -ENOENT;
207                 else rv = 0;
208         }
209
210         /* now see if we want to start an upcall */
211         refresh_age = (h->expiry_time - h->last_refresh);
212         age = get_seconds() - h->last_refresh;
213
214         if (rqstp == NULL) {
215                 if (rv == -EAGAIN)
216                         rv = -ENOENT;
217         } else if (rv == -EAGAIN || age > refresh_age/2) {
218                 dprintk("Want update, refage=%ld, age=%ld\n", refresh_age, age);
219                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
220                         switch (cache_make_upcall(detail, h)) {
221                         case -EINVAL:
222                                 clear_bit(CACHE_PENDING, &h->flags);
223                                 if (rv == -EAGAIN) {
224                                         set_bit(CACHE_NEGATIVE, &h->flags);
225                                         cache_fresh_unlocked(h, detail,
226                                              cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY));
227                                         rv = -ENOENT;
228                                 }
229                                 break;
230
231                         case -EAGAIN:
232                                 clear_bit(CACHE_PENDING, &h->flags);
233                                 cache_revisit_request(h);
234                                 break;
235                         }
236                 }
237         }
238
239         if (rv == -EAGAIN)
240                 if (cache_defer_req(rqstp, h) != 0)
241                         rv = -ETIMEDOUT;
242
243         if (rv)
244                 cache_put(h, detail);
245         return rv;
246 }
247
248 /*
249  * caches need to be periodically cleaned.
250  * For this we maintain a list of cache_detail and
251  * a current pointer into that list and into the table
252  * for that entry.
253  *
254  * Each time clean_cache is called it finds the next non-empty entry
255  * in the current table and walks the list in that entry
256  * looking for entries that can be removed.
257  *
258  * An entry gets removed if:
259  * - The expiry is before current time
260  * - The last_refresh time is before the flush_time for that cache
261  *
262  * later we might drop old entries with non-NEVER expiry if that table
263  * is getting 'full' for some definition of 'full'
264  *
265  * The question of "how often to scan a table" is an interesting one
266  * and is answered in part by the use of the "nextcheck" field in the
267  * cache_detail.
268  * When a scan of a table begins, the nextcheck field is set to a time
269  * that is well into the future.
270  * While scanning, if an expiry time is found that is earlier than the
271  * current nextcheck time, nextcheck is set to that expiry time.
272  * If the flush_time is ever set to a time earlier than the nextcheck
273  * time, the nextcheck time is then set to that flush_time.
274  *
275  * A table is then only scanned if the current time is at least
276  * the nextcheck time.
277  * 
278  */
279
280 static LIST_HEAD(cache_list);
281 static DEFINE_SPINLOCK(cache_list_lock);
282 static struct cache_detail *current_detail;
283 static int current_index;
284
285 static struct file_operations cache_file_operations;
286 static struct file_operations content_file_operations;
287 static struct file_operations cache_flush_operations;
288
289 static void do_cache_clean(struct work_struct *work);
290 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
291
292 void cache_register(struct cache_detail *cd)
293 {
294         cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
295         if (cd->proc_ent) {
296                 struct proc_dir_entry *p;
297                 cd->proc_ent->owner = cd->owner;
298                 cd->channel_ent = cd->content_ent = NULL;
299                 
300                 p = create_proc_entry("flush", S_IFREG|S_IRUSR|S_IWUSR,
301                                       cd->proc_ent);
302                 cd->flush_ent =  p;
303                 if (p) {
304                         p->proc_fops = &cache_flush_operations;
305                         p->owner = cd->owner;
306                         p->data = cd;
307                 }
308  
309                 if (cd->cache_request || cd->cache_parse) {
310                         p = create_proc_entry("channel", S_IFREG|S_IRUSR|S_IWUSR,
311                                               cd->proc_ent);
312                         cd->channel_ent = p;
313                         if (p) {
314                                 p->proc_fops = &cache_file_operations;
315                                 p->owner = cd->owner;
316                                 p->data = cd;
317                         }
318                 }
319                 if (cd->cache_show) {
320                         p = create_proc_entry("content", S_IFREG|S_IRUSR|S_IWUSR,
321                                               cd->proc_ent);
322                         cd->content_ent = p;
323                         if (p) {
324                                 p->proc_fops = &content_file_operations;
325                                 p->owner = cd->owner;
326                                 p->data = cd;
327                         }
328                 }
329         }
330         rwlock_init(&cd->hash_lock);
331         INIT_LIST_HEAD(&cd->queue);
332         spin_lock(&cache_list_lock);
333         cd->nextcheck = 0;
334         cd->entries = 0;
335         atomic_set(&cd->readers, 0);
336         cd->last_close = 0;
337         cd->last_warn = -1;
338         list_add(&cd->others, &cache_list);
339         spin_unlock(&cache_list_lock);
340
341         /* start the cleaning process */
342         schedule_delayed_work(&cache_cleaner, 0);
343 }
344
345 int cache_unregister(struct cache_detail *cd)
346 {
347         cache_purge(cd);
348         spin_lock(&cache_list_lock);
349         write_lock(&cd->hash_lock);
350         if (cd->entries || atomic_read(&cd->inuse)) {
351                 write_unlock(&cd->hash_lock);
352                 spin_unlock(&cache_list_lock);
353                 return -EBUSY;
354         }
355         if (current_detail == cd)
356                 current_detail = NULL;
357         list_del_init(&cd->others);
358         write_unlock(&cd->hash_lock);
359         spin_unlock(&cache_list_lock);
360         if (cd->proc_ent) {
361                 if (cd->flush_ent)
362                         remove_proc_entry("flush", cd->proc_ent);
363                 if (cd->channel_ent)
364                         remove_proc_entry("channel", cd->proc_ent);
365                 if (cd->content_ent)
366                         remove_proc_entry("content", cd->proc_ent);
367
368                 cd->proc_ent = NULL;
369                 remove_proc_entry(cd->name, proc_net_rpc);
370         }
371         if (list_empty(&cache_list)) {
372                 /* module must be being unloaded so its safe to kill the worker */
373                 cancel_delayed_work(&cache_cleaner);
374                 flush_scheduled_work();
375         }
376         return 0;
377 }
378
379 /* clean cache tries to find something to clean
380  * and cleans it.
381  * It returns 1 if it cleaned something,
382  *            0 if it didn't find anything this time
383  *           -1 if it fell off the end of the list.
384  */
385 static int cache_clean(void)
386 {
387         int rv = 0;
388         struct list_head *next;
389
390         spin_lock(&cache_list_lock);
391
392         /* find a suitable table if we don't already have one */
393         while (current_detail == NULL ||
394             current_index >= current_detail->hash_size) {
395                 if (current_detail)
396                         next = current_detail->others.next;
397                 else
398                         next = cache_list.next;
399                 if (next == &cache_list) {
400                         current_detail = NULL;
401                         spin_unlock(&cache_list_lock);
402                         return -1;
403                 }
404                 current_detail = list_entry(next, struct cache_detail, others);
405                 if (current_detail->nextcheck > get_seconds())
406                         current_index = current_detail->hash_size;
407                 else {
408                         current_index = 0;
409                         current_detail->nextcheck = get_seconds()+30*60;
410                 }
411         }
412
413         /* find a non-empty bucket in the table */
414         while (current_detail &&
415                current_index < current_detail->hash_size &&
416                current_detail->hash_table[current_index] == NULL)
417                 current_index++;
418
419         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
420         
421         if (current_detail && current_index < current_detail->hash_size) {
422                 struct cache_head *ch, **cp;
423                 struct cache_detail *d;
424                 
425                 write_lock(&current_detail->hash_lock);
426
427                 /* Ok, now to clean this strand */
428                         
429                 cp = & current_detail->hash_table[current_index];
430                 ch = *cp;
431                 for (; ch; cp= & ch->next, ch= *cp) {
432                         if (current_detail->nextcheck > ch->expiry_time)
433                                 current_detail->nextcheck = ch->expiry_time+1;
434                         if (ch->expiry_time >= get_seconds()
435                             && ch->last_refresh >= current_detail->flush_time
436                                 )
437                                 continue;
438                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
439                                 queue_loose(current_detail, ch);
440
441                         if (atomic_read(&ch->ref.refcount) == 1)
442                                 break;
443                 }
444                 if (ch) {
445                         *cp = ch->next;
446                         ch->next = NULL;
447                         current_detail->entries--;
448                         rv = 1;
449                 }
450                 write_unlock(&current_detail->hash_lock);
451                 d = current_detail;
452                 if (!ch)
453                         current_index ++;
454                 spin_unlock(&cache_list_lock);
455                 if (ch)
456                         cache_put(ch, d);
457         } else
458                 spin_unlock(&cache_list_lock);
459
460         return rv;
461 }
462
463 /*
464  * We want to regularly clean the cache, so we need to schedule some work ...
465  */
466 static void do_cache_clean(struct work_struct *work)
467 {
468         int delay = 5;
469         if (cache_clean() == -1)
470                 delay = 30*HZ;
471
472         if (list_empty(&cache_list))
473                 delay = 0;
474
475         if (delay)
476                 schedule_delayed_work(&cache_cleaner, delay);
477 }
478
479
480 /* 
481  * Clean all caches promptly.  This just calls cache_clean
482  * repeatedly until we are sure that every cache has had a chance to 
483  * be fully cleaned
484  */
485 void cache_flush(void)
486 {
487         while (cache_clean() != -1)
488                 cond_resched();
489         while (cache_clean() != -1)
490                 cond_resched();
491 }
492
493 void cache_purge(struct cache_detail *detail)
494 {
495         detail->flush_time = LONG_MAX;
496         detail->nextcheck = get_seconds();
497         cache_flush();
498         detail->flush_time = 1;
499 }
500
501
502
503 /*
504  * Deferral and Revisiting of Requests.
505  *
506  * If a cache lookup finds a pending entry, we
507  * need to defer the request and revisit it later.
508  * All deferred requests are stored in a hash table,
509  * indexed by "struct cache_head *".
510  * As it may be wasteful to store a whole request
511  * structure, we allow the request to provide a 
512  * deferred form, which must contain a
513  * 'struct cache_deferred_req'
514  * This cache_deferred_req contains a method to allow
515  * it to be revisited when cache info is available
516  */
517
518 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
519 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
520
521 #define DFR_MAX 300     /* ??? */
522
523 static DEFINE_SPINLOCK(cache_defer_lock);
524 static LIST_HEAD(cache_defer_list);
525 static struct list_head cache_defer_hash[DFR_HASHSIZE];
526 static int cache_defer_cnt;
527
528 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
529 {
530         struct cache_deferred_req *dreq;
531         int hash = DFR_HASH(item);
532
533         dreq = req->defer(req);
534         if (dreq == NULL)
535                 return -ETIMEDOUT;
536
537         dreq->item = item;
538         dreq->recv_time = get_seconds();
539
540         spin_lock(&cache_defer_lock);
541
542         list_add(&dreq->recent, &cache_defer_list);
543
544         if (cache_defer_hash[hash].next == NULL)
545                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
546         list_add(&dreq->hash, &cache_defer_hash[hash]);
547
548         /* it is in, now maybe clean up */
549         dreq = NULL;
550         if (++cache_defer_cnt > DFR_MAX) {
551                 /* too much in the cache, randomly drop
552                  * first or last
553                  */
554                 if (net_random()&1) 
555                         dreq = list_entry(cache_defer_list.next,
556                                           struct cache_deferred_req,
557                                           recent);
558                 else
559                         dreq = list_entry(cache_defer_list.prev,
560                                           struct cache_deferred_req,
561                                           recent);
562                 list_del(&dreq->recent);
563                 list_del(&dreq->hash);
564                 cache_defer_cnt--;
565         }
566         spin_unlock(&cache_defer_lock);
567
568         if (dreq) {
569                 /* there was one too many */
570                 dreq->revisit(dreq, 1);
571         }
572         if (!test_bit(CACHE_PENDING, &item->flags)) {
573                 /* must have just been validated... */
574                 cache_revisit_request(item);
575         }
576         return 0;
577 }
578
579 static void cache_revisit_request(struct cache_head *item)
580 {
581         struct cache_deferred_req *dreq;
582         struct list_head pending;
583
584         struct list_head *lp;
585         int hash = DFR_HASH(item);
586
587         INIT_LIST_HEAD(&pending);
588         spin_lock(&cache_defer_lock);
589         
590         lp = cache_defer_hash[hash].next;
591         if (lp) {
592                 while (lp != &cache_defer_hash[hash]) {
593                         dreq = list_entry(lp, struct cache_deferred_req, hash);
594                         lp = lp->next;
595                         if (dreq->item == item) {
596                                 list_del(&dreq->hash);
597                                 list_move(&dreq->recent, &pending);
598                                 cache_defer_cnt--;
599                         }
600                 }
601         }
602         spin_unlock(&cache_defer_lock);
603
604         while (!list_empty(&pending)) {
605                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
606                 list_del_init(&dreq->recent);
607                 dreq->revisit(dreq, 0);
608         }
609 }
610
611 void cache_clean_deferred(void *owner)
612 {
613         struct cache_deferred_req *dreq, *tmp;
614         struct list_head pending;
615
616
617         INIT_LIST_HEAD(&pending);
618         spin_lock(&cache_defer_lock);
619         
620         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
621                 if (dreq->owner == owner) {
622                         list_del(&dreq->hash);
623                         list_move(&dreq->recent, &pending);
624                         cache_defer_cnt--;
625                 }
626         }
627         spin_unlock(&cache_defer_lock);
628
629         while (!list_empty(&pending)) {
630                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
631                 list_del_init(&dreq->recent);
632                 dreq->revisit(dreq, 1);
633         }
634 }
635
636 /*
637  * communicate with user-space
638  *
639  * We have a magic /proc file - /proc/sunrpc/cache
640  * On read, you get a full request, or block
641  * On write, an update request is processed
642  * Poll works if anything to read, and always allows write
643  *
644  * Implemented by linked list of requests.  Each open file has 
645  * a ->private that also exists in this list.  New request are added
646  * to the end and may wakeup and preceding readers.
647  * New readers are added to the head.  If, on read, an item is found with
648  * CACHE_UPCALLING clear, we free it from the list.
649  *
650  */
651
652 static DEFINE_SPINLOCK(queue_lock);
653 static DEFINE_MUTEX(queue_io_mutex);
654
655 struct cache_queue {
656         struct list_head        list;
657         int                     reader; /* if 0, then request */
658 };
659 struct cache_request {
660         struct cache_queue      q;
661         struct cache_head       *item;
662         char                    * buf;
663         int                     len;
664         int                     readers;
665 };
666 struct cache_reader {
667         struct cache_queue      q;
668         int                     offset; /* if non-0, we have a refcnt on next request */
669 };
670
671 static ssize_t
672 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
673 {
674         struct cache_reader *rp = filp->private_data;
675         struct cache_request *rq;
676         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
677         int err;
678
679         if (count == 0)
680                 return 0;
681
682         mutex_lock(&queue_io_mutex); /* protect against multiple concurrent
683                               * readers on this file */
684  again:
685         spin_lock(&queue_lock);
686         /* need to find next request */
687         while (rp->q.list.next != &cd->queue &&
688                list_entry(rp->q.list.next, struct cache_queue, list)
689                ->reader) {
690                 struct list_head *next = rp->q.list.next;
691                 list_move(&rp->q.list, next);
692         }
693         if (rp->q.list.next == &cd->queue) {
694                 spin_unlock(&queue_lock);
695                 mutex_unlock(&queue_io_mutex);
696                 BUG_ON(rp->offset);
697                 return 0;
698         }
699         rq = container_of(rp->q.list.next, struct cache_request, q.list);
700         BUG_ON(rq->q.reader);
701         if (rp->offset == 0)
702                 rq->readers++;
703         spin_unlock(&queue_lock);
704
705         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
706                 err = -EAGAIN;
707                 spin_lock(&queue_lock);
708                 list_move(&rp->q.list, &rq->q.list);
709                 spin_unlock(&queue_lock);
710         } else {
711                 if (rp->offset + count > rq->len)
712                         count = rq->len - rp->offset;
713                 err = -EFAULT;
714                 if (copy_to_user(buf, rq->buf + rp->offset, count))
715                         goto out;
716                 rp->offset += count;
717                 if (rp->offset >= rq->len) {
718                         rp->offset = 0;
719                         spin_lock(&queue_lock);
720                         list_move(&rp->q.list, &rq->q.list);
721                         spin_unlock(&queue_lock);
722                 }
723                 err = 0;
724         }
725  out:
726         if (rp->offset == 0) {
727                 /* need to release rq */
728                 spin_lock(&queue_lock);
729                 rq->readers--;
730                 if (rq->readers == 0 &&
731                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
732                         list_del(&rq->q.list);
733                         spin_unlock(&queue_lock);
734                         cache_put(rq->item, cd);
735                         kfree(rq->buf);
736                         kfree(rq);
737                 } else
738                         spin_unlock(&queue_lock);
739         }
740         if (err == -EAGAIN)
741                 goto again;
742         mutex_unlock(&queue_io_mutex);
743         return err ? err :  count;
744 }
745
746 static char write_buf[8192]; /* protected by queue_io_mutex */
747
748 static ssize_t
749 cache_write(struct file *filp, const char __user *buf, size_t count,
750             loff_t *ppos)
751 {
752         int err;
753         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
754
755         if (count == 0)
756                 return 0;
757         if (count >= sizeof(write_buf))
758                 return -EINVAL;
759
760         mutex_lock(&queue_io_mutex);
761
762         if (copy_from_user(write_buf, buf, count)) {
763                 mutex_unlock(&queue_io_mutex);
764                 return -EFAULT;
765         }
766         write_buf[count] = '\0';
767         if (cd->cache_parse)
768                 err = cd->cache_parse(cd, write_buf, count);
769         else
770                 err = -EINVAL;
771
772         mutex_unlock(&queue_io_mutex);
773         return err ? err : count;
774 }
775
776 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
777
778 static unsigned int
779 cache_poll(struct file *filp, poll_table *wait)
780 {
781         unsigned int mask;
782         struct cache_reader *rp = filp->private_data;
783         struct cache_queue *cq;
784         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
785
786         poll_wait(filp, &queue_wait, wait);
787
788         /* alway allow write */
789         mask = POLL_OUT | POLLWRNORM;
790
791         if (!rp)
792                 return mask;
793
794         spin_lock(&queue_lock);
795
796         for (cq= &rp->q; &cq->list != &cd->queue;
797              cq = list_entry(cq->list.next, struct cache_queue, list))
798                 if (!cq->reader) {
799                         mask |= POLLIN | POLLRDNORM;
800                         break;
801                 }
802         spin_unlock(&queue_lock);
803         return mask;
804 }
805
806 static int
807 cache_ioctl(struct inode *ino, struct file *filp,
808             unsigned int cmd, unsigned long arg)
809 {
810         int len = 0;
811         struct cache_reader *rp = filp->private_data;
812         struct cache_queue *cq;
813         struct cache_detail *cd = PDE(ino)->data;
814
815         if (cmd != FIONREAD || !rp)
816                 return -EINVAL;
817
818         spin_lock(&queue_lock);
819
820         /* only find the length remaining in current request,
821          * or the length of the next request
822          */
823         for (cq= &rp->q; &cq->list != &cd->queue;
824              cq = list_entry(cq->list.next, struct cache_queue, list))
825                 if (!cq->reader) {
826                         struct cache_request *cr =
827                                 container_of(cq, struct cache_request, q);
828                         len = cr->len - rp->offset;
829                         break;
830                 }
831         spin_unlock(&queue_lock);
832
833         return put_user(len, (int __user *)arg);
834 }
835
836 static int
837 cache_open(struct inode *inode, struct file *filp)
838 {
839         struct cache_reader *rp = NULL;
840
841         nonseekable_open(inode, filp);
842         if (filp->f_mode & FMODE_READ) {
843                 struct cache_detail *cd = PDE(inode)->data;
844
845                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
846                 if (!rp)
847                         return -ENOMEM;
848                 rp->offset = 0;
849                 rp->q.reader = 1;
850                 atomic_inc(&cd->readers);
851                 spin_lock(&queue_lock);
852                 list_add(&rp->q.list, &cd->queue);
853                 spin_unlock(&queue_lock);
854         }
855         filp->private_data = rp;
856         return 0;
857 }
858
859 static int
860 cache_release(struct inode *inode, struct file *filp)
861 {
862         struct cache_reader *rp = filp->private_data;
863         struct cache_detail *cd = PDE(inode)->data;
864
865         if (rp) {
866                 spin_lock(&queue_lock);
867                 if (rp->offset) {
868                         struct cache_queue *cq;
869                         for (cq= &rp->q; &cq->list != &cd->queue;
870                              cq = list_entry(cq->list.next, struct cache_queue, list))
871                                 if (!cq->reader) {
872                                         container_of(cq, struct cache_request, q)
873                                                 ->readers--;
874                                         break;
875                                 }
876                         rp->offset = 0;
877                 }
878                 list_del(&rp->q.list);
879                 spin_unlock(&queue_lock);
880
881                 filp->private_data = NULL;
882                 kfree(rp);
883
884                 cd->last_close = get_seconds();
885                 atomic_dec(&cd->readers);
886         }
887         return 0;
888 }
889
890
891
892 static struct file_operations cache_file_operations = {
893         .owner          = THIS_MODULE,
894         .llseek         = no_llseek,
895         .read           = cache_read,
896         .write          = cache_write,
897         .poll           = cache_poll,
898         .ioctl          = cache_ioctl, /* for FIONREAD */
899         .open           = cache_open,
900         .release        = cache_release,
901 };
902
903
904 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
905 {
906         struct cache_queue *cq;
907         spin_lock(&queue_lock);
908         list_for_each_entry(cq, &detail->queue, list)
909                 if (!cq->reader) {
910                         struct cache_request *cr = container_of(cq, struct cache_request, q);
911                         if (cr->item != ch)
912                                 continue;
913                         if (cr->readers != 0)
914                                 continue;
915                         list_del(&cr->q.list);
916                         spin_unlock(&queue_lock);
917                         cache_put(cr->item, detail);
918                         kfree(cr->buf);
919                         kfree(cr);
920                         return;
921                 }
922         spin_unlock(&queue_lock);
923 }
924
925 /*
926  * Support routines for text-based upcalls.
927  * Fields are separated by spaces.
928  * Fields are either mangled to quote space tab newline slosh with slosh
929  * or a hexified with a leading \x
930  * Record is terminated with newline.
931  *
932  */
933
934 void qword_add(char **bpp, int *lp, char *str)
935 {
936         char *bp = *bpp;
937         int len = *lp;
938         char c;
939
940         if (len < 0) return;
941
942         while ((c=*str++) && len)
943                 switch(c) {
944                 case ' ':
945                 case '\t':
946                 case '\n':
947                 case '\\':
948                         if (len >= 4) {
949                                 *bp++ = '\\';
950                                 *bp++ = '0' + ((c & 0300)>>6);
951                                 *bp++ = '0' + ((c & 0070)>>3);
952                                 *bp++ = '0' + ((c & 0007)>>0);
953                         }
954                         len -= 4;
955                         break;
956                 default:
957                         *bp++ = c;
958                         len--;
959                 }
960         if (c || len <1) len = -1;
961         else {
962                 *bp++ = ' ';
963                 len--;
964         }
965         *bpp = bp;
966         *lp = len;
967 }
968
969 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
970 {
971         char *bp = *bpp;
972         int len = *lp;
973
974         if (len < 0) return;
975
976         if (len > 2) {
977                 *bp++ = '\\';
978                 *bp++ = 'x';
979                 len -= 2;
980                 while (blen && len >= 2) {
981                         unsigned char c = *buf++;
982                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
983                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
984                         len -= 2;
985                         blen--;
986                 }
987         }
988         if (blen || len<1) len = -1;
989         else {
990                 *bp++ = ' ';
991                 len--;
992         }
993         *bpp = bp;
994         *lp = len;
995 }
996
997 static void warn_no_listener(struct cache_detail *detail)
998 {
999         if (detail->last_warn != detail->last_close) {
1000                 detail->last_warn = detail->last_close;
1001                 if (detail->warn_no_listener)
1002                         detail->warn_no_listener(detail);
1003         }
1004 }
1005
1006 /*
1007  * register an upcall request to user-space.
1008  * Each request is at most one page long.
1009  */
1010 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
1011 {
1012
1013         char *buf;
1014         struct cache_request *crq;
1015         char *bp;
1016         int len;
1017
1018         if (detail->cache_request == NULL)
1019                 return -EINVAL;
1020
1021         if (atomic_read(&detail->readers) == 0 &&
1022             detail->last_close < get_seconds() - 30) {
1023                         warn_no_listener(detail);
1024                         return -EINVAL;
1025         }
1026
1027         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1028         if (!buf)
1029                 return -EAGAIN;
1030
1031         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1032         if (!crq) {
1033                 kfree(buf);
1034                 return -EAGAIN;
1035         }
1036
1037         bp = buf; len = PAGE_SIZE;
1038
1039         detail->cache_request(detail, h, &bp, &len);
1040
1041         if (len < 0) {
1042                 kfree(buf);
1043                 kfree(crq);
1044                 return -EAGAIN;
1045         }
1046         crq->q.reader = 0;
1047         crq->item = cache_get(h);
1048         crq->buf = buf;
1049         crq->len = PAGE_SIZE - len;
1050         crq->readers = 0;
1051         spin_lock(&queue_lock);
1052         list_add_tail(&crq->q.list, &detail->queue);
1053         spin_unlock(&queue_lock);
1054         wake_up(&queue_wait);
1055         return 0;
1056 }
1057
1058 /*
1059  * parse a message from user-space and pass it
1060  * to an appropriate cache
1061  * Messages are, like requests, separated into fields by
1062  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1063  *
1064  * Message is 
1065  *   reply cachename expiry key ... content....
1066  *
1067  * key and content are both parsed by cache 
1068  */
1069
1070 #define isodigit(c) (isdigit(c) && c <= '7')
1071 int qword_get(char **bpp, char *dest, int bufsize)
1072 {
1073         /* return bytes copied, or -1 on error */
1074         char *bp = *bpp;
1075         int len = 0;
1076
1077         while (*bp == ' ') bp++;
1078
1079         if (bp[0] == '\\' && bp[1] == 'x') {
1080                 /* HEX STRING */
1081                 bp += 2;
1082                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1083                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1084                         bp++;
1085                         byte <<= 4;
1086                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1087                         *dest++ = byte;
1088                         bp++;
1089                         len++;
1090                 }
1091         } else {
1092                 /* text with \nnn octal quoting */
1093                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1094                         if (*bp == '\\' &&
1095                             isodigit(bp[1]) && (bp[1] <= '3') &&
1096                             isodigit(bp[2]) &&
1097                             isodigit(bp[3])) {
1098                                 int byte = (*++bp -'0');
1099                                 bp++;
1100                                 byte = (byte << 3) | (*bp++ - '0');
1101                                 byte = (byte << 3) | (*bp++ - '0');
1102                                 *dest++ = byte;
1103                                 len++;
1104                         } else {
1105                                 *dest++ = *bp++;
1106                                 len++;
1107                         }
1108                 }
1109         }
1110
1111         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1112                 return -1;
1113         while (*bp == ' ') bp++;
1114         *bpp = bp;
1115         *dest = '\0';
1116         return len;
1117 }
1118
1119
1120 /*
1121  * support /proc/sunrpc/cache/$CACHENAME/content
1122  * as a seqfile.
1123  * We call ->cache_show passing NULL for the item to
1124  * get a header, then pass each real item in the cache
1125  */
1126
1127 struct handle {
1128         struct cache_detail *cd;
1129 };
1130
1131 static void *c_start(struct seq_file *m, loff_t *pos)
1132 {
1133         loff_t n = *pos;
1134         unsigned hash, entry;
1135         struct cache_head *ch;
1136         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1137         
1138
1139         read_lock(&cd->hash_lock);
1140         if (!n--)
1141                 return SEQ_START_TOKEN;
1142         hash = n >> 32;
1143         entry = n & ((1LL<<32) - 1);
1144
1145         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1146                 if (!entry--)
1147                         return ch;
1148         n &= ~((1LL<<32) - 1);
1149         do {
1150                 hash++;
1151                 n += 1LL<<32;
1152         } while(hash < cd->hash_size && 
1153                 cd->hash_table[hash]==NULL);
1154         if (hash >= cd->hash_size)
1155                 return NULL;
1156         *pos = n+1;
1157         return cd->hash_table[hash];
1158 }
1159
1160 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1161 {
1162         struct cache_head *ch = p;
1163         int hash = (*pos >> 32);
1164         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1165
1166         if (p == SEQ_START_TOKEN)
1167                 hash = 0;
1168         else if (ch->next == NULL) {
1169                 hash++;
1170                 *pos += 1LL<<32;
1171         } else {
1172                 ++*pos;
1173                 return ch->next;
1174         }
1175         *pos &= ~((1LL<<32) - 1);
1176         while (hash < cd->hash_size &&
1177                cd->hash_table[hash] == NULL) {
1178                 hash++;
1179                 *pos += 1LL<<32;
1180         }
1181         if (hash >= cd->hash_size)
1182                 return NULL;
1183         ++*pos;
1184         return cd->hash_table[hash];
1185 }
1186
1187 static void c_stop(struct seq_file *m, void *p)
1188 {
1189         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1190         read_unlock(&cd->hash_lock);
1191 }
1192
1193 static int c_show(struct seq_file *m, void *p)
1194 {
1195         struct cache_head *cp = p;
1196         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1197
1198         if (p == SEQ_START_TOKEN)
1199                 return cd->cache_show(m, cd, NULL);
1200
1201         ifdebug(CACHE)
1202                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1203                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1204         cache_get(cp);
1205         if (cache_check(cd, cp, NULL))
1206                 /* cache_check does a cache_put on failure */
1207                 seq_printf(m, "# ");
1208         else
1209                 cache_put(cp, cd);
1210
1211         return cd->cache_show(m, cd, cp);
1212 }
1213
1214 static struct seq_operations cache_content_op = {
1215         .start  = c_start,
1216         .next   = c_next,
1217         .stop   = c_stop,
1218         .show   = c_show,
1219 };
1220
1221 static int content_open(struct inode *inode, struct file *file)
1222 {
1223         int res;
1224         struct handle *han;
1225         struct cache_detail *cd = PDE(inode)->data;
1226
1227         han = kmalloc(sizeof(*han), GFP_KERNEL);
1228         if (han == NULL)
1229                 return -ENOMEM;
1230
1231         han->cd = cd;
1232
1233         res = seq_open(file, &cache_content_op);
1234         if (res)
1235                 kfree(han);
1236         else
1237                 ((struct seq_file *)file->private_data)->private = han;
1238
1239         return res;
1240 }
1241 static int content_release(struct inode *inode, struct file *file)
1242 {
1243         struct seq_file *m = (struct seq_file *)file->private_data;
1244         struct handle *han = m->private;
1245         kfree(han);
1246         m->private = NULL;
1247         return seq_release(inode, file);
1248 }
1249
1250 static struct file_operations content_file_operations = {
1251         .open           = content_open,
1252         .read           = seq_read,
1253         .llseek         = seq_lseek,
1254         .release        = content_release,
1255 };
1256
1257 static ssize_t read_flush(struct file *file, char __user *buf,
1258                             size_t count, loff_t *ppos)
1259 {
1260         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1261         char tbuf[20];
1262         unsigned long p = *ppos;
1263         int len;
1264
1265         sprintf(tbuf, "%lu\n", cd->flush_time);
1266         len = strlen(tbuf);
1267         if (p >= len)
1268                 return 0;
1269         len -= p;
1270         if (len > count) len = count;
1271         if (copy_to_user(buf, (void*)(tbuf+p), len))
1272                 len = -EFAULT;
1273         else
1274                 *ppos += len;
1275         return len;
1276 }
1277
1278 static ssize_t write_flush(struct file * file, const char __user * buf,
1279                              size_t count, loff_t *ppos)
1280 {
1281         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1282         char tbuf[20];
1283         char *ep;
1284         long flushtime;
1285         if (*ppos || count > sizeof(tbuf)-1)
1286                 return -EINVAL;
1287         if (copy_from_user(tbuf, buf, count))
1288                 return -EFAULT;
1289         tbuf[count] = 0;
1290         flushtime = simple_strtoul(tbuf, &ep, 0);
1291         if (*ep && *ep != '\n')
1292                 return -EINVAL;
1293
1294         cd->flush_time = flushtime;
1295         cd->nextcheck = get_seconds();
1296         cache_flush();
1297
1298         *ppos += count;
1299         return count;
1300 }
1301
1302 static struct file_operations cache_flush_operations = {
1303         .open           = nonseekable_open,
1304         .read           = read_flush,
1305         .write          = write_flush,
1306 };