[IRDA] sem2mutex: drivers/net/irda
[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 <asm/ioctls.h>
30 #include <linux/sunrpc/types.h>
31 #include <linux/sunrpc/cache.h>
32 #include <linux/sunrpc/stats.h>
33
34 #define  RPCDBG_FACILITY RPCDBG_CACHE
35
36 static void cache_defer_req(struct cache_req *req, struct cache_head *item);
37 static void cache_revisit_request(struct cache_head *item);
38
39 void cache_init(struct cache_head *h)
40 {
41         time_t now = get_seconds();
42         h->next = NULL;
43         h->flags = 0;
44         atomic_set(&h->refcnt, 1);
45         h->expiry_time = now + CACHE_NEW_EXPIRY;
46         h->last_refresh = now;
47 }
48
49
50 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
51 /*
52  * This is the generic cache management routine for all
53  * the authentication caches.
54  * It checks the currency of a cache item and will (later)
55  * initiate an upcall to fill it if needed.
56  *
57  *
58  * Returns 0 if the cache_head can be used, or cache_puts it and returns
59  * -EAGAIN if upcall is pending,
60  * -ENOENT if cache entry was negative
61  */
62 int cache_check(struct cache_detail *detail,
63                     struct cache_head *h, struct cache_req *rqstp)
64 {
65         int rv;
66         long refresh_age, age;
67
68         /* First decide return status as best we can */
69         if (!test_bit(CACHE_VALID, &h->flags) ||
70             h->expiry_time < get_seconds())
71                 rv = -EAGAIN;
72         else if (detail->flush_time > h->last_refresh)
73                 rv = -EAGAIN;
74         else {
75                 /* entry is valid */
76                 if (test_bit(CACHE_NEGATIVE, &h->flags))
77                         rv = -ENOENT;
78                 else rv = 0;
79         }
80
81         /* now see if we want to start an upcall */
82         refresh_age = (h->expiry_time - h->last_refresh);
83         age = get_seconds() - h->last_refresh;
84
85         if (rqstp == NULL) {
86                 if (rv == -EAGAIN)
87                         rv = -ENOENT;
88         } else if (rv == -EAGAIN || age > refresh_age/2) {
89                 dprintk("Want update, refage=%ld, age=%ld\n", refresh_age, age);
90                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
91                         switch (cache_make_upcall(detail, h)) {
92                         case -EINVAL:
93                                 clear_bit(CACHE_PENDING, &h->flags);
94                                 if (rv == -EAGAIN) {
95                                         set_bit(CACHE_NEGATIVE, &h->flags);
96                                         cache_fresh(detail, h, get_seconds()+CACHE_NEW_EXPIRY);
97                                         rv = -ENOENT;
98                                 }
99                                 break;
100
101                         case -EAGAIN:
102                                 clear_bit(CACHE_PENDING, &h->flags);
103                                 cache_revisit_request(h);
104                                 break;
105                         }
106                 }
107         }
108
109         if (rv == -EAGAIN)
110                 cache_defer_req(rqstp, h);
111
112         if (rv && h)
113                 detail->cache_put(h, detail);
114         return rv;
115 }
116
117 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
118
119 void cache_fresh(struct cache_detail *detail,
120                  struct cache_head *head, time_t expiry)
121 {
122
123         head->expiry_time = expiry;
124         head->last_refresh = get_seconds();
125         if (!test_and_set_bit(CACHE_VALID, &head->flags))
126                 cache_revisit_request(head);
127         if (test_and_clear_bit(CACHE_PENDING, &head->flags))
128                 queue_loose(detail, head);
129 }
130
131 /*
132  * caches need to be periodically cleaned.
133  * For this we maintain a list of cache_detail and
134  * a current pointer into that list and into the table
135  * for that entry.
136  *
137  * Each time clean_cache is called it finds the next non-empty entry
138  * in the current table and walks the list in that entry
139  * looking for entries that can be removed.
140  *
141  * An entry gets removed if:
142  * - The expiry is before current time
143  * - The last_refresh time is before the flush_time for that cache
144  *
145  * later we might drop old entries with non-NEVER expiry if that table
146  * is getting 'full' for some definition of 'full'
147  *
148  * The question of "how often to scan a table" is an interesting one
149  * and is answered in part by the use of the "nextcheck" field in the
150  * cache_detail.
151  * When a scan of a table begins, the nextcheck field is set to a time
152  * that is well into the future.
153  * While scanning, if an expiry time is found that is earlier than the
154  * current nextcheck time, nextcheck is set to that expiry time.
155  * If the flush_time is ever set to a time earlier than the nextcheck
156  * time, the nextcheck time is then set to that flush_time.
157  *
158  * A table is then only scanned if the current time is at least
159  * the nextcheck time.
160  * 
161  */
162
163 static LIST_HEAD(cache_list);
164 static DEFINE_SPINLOCK(cache_list_lock);
165 static struct cache_detail *current_detail;
166 static int current_index;
167
168 static struct file_operations cache_file_operations;
169 static struct file_operations content_file_operations;
170 static struct file_operations cache_flush_operations;
171
172 static void do_cache_clean(void *data);
173 static DECLARE_WORK(cache_cleaner, do_cache_clean, NULL);
174
175 void cache_register(struct cache_detail *cd)
176 {
177         cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
178         if (cd->proc_ent) {
179                 struct proc_dir_entry *p;
180                 cd->proc_ent->owner = cd->owner;
181                 cd->channel_ent = cd->content_ent = NULL;
182                 
183                 p = create_proc_entry("flush", S_IFREG|S_IRUSR|S_IWUSR,
184                                       cd->proc_ent);
185                 cd->flush_ent =  p;
186                 if (p) {
187                         p->proc_fops = &cache_flush_operations;
188                         p->owner = cd->owner;
189                         p->data = cd;
190                 }
191  
192                 if (cd->cache_request || cd->cache_parse) {
193                         p = create_proc_entry("channel", S_IFREG|S_IRUSR|S_IWUSR,
194                                               cd->proc_ent);
195                         cd->channel_ent = p;
196                         if (p) {
197                                 p->proc_fops = &cache_file_operations;
198                                 p->owner = cd->owner;
199                                 p->data = cd;
200                         }
201                 }
202                 if (cd->cache_show) {
203                         p = create_proc_entry("content", S_IFREG|S_IRUSR|S_IWUSR,
204                                               cd->proc_ent);
205                         cd->content_ent = p;
206                         if (p) {
207                                 p->proc_fops = &content_file_operations;
208                                 p->owner = cd->owner;
209                                 p->data = cd;
210                         }
211                 }
212         }
213         rwlock_init(&cd->hash_lock);
214         INIT_LIST_HEAD(&cd->queue);
215         spin_lock(&cache_list_lock);
216         cd->nextcheck = 0;
217         cd->entries = 0;
218         atomic_set(&cd->readers, 0);
219         cd->last_close = 0;
220         cd->last_warn = -1;
221         list_add(&cd->others, &cache_list);
222         spin_unlock(&cache_list_lock);
223
224         /* start the cleaning process */
225         schedule_work(&cache_cleaner);
226 }
227
228 int cache_unregister(struct cache_detail *cd)
229 {
230         cache_purge(cd);
231         spin_lock(&cache_list_lock);
232         write_lock(&cd->hash_lock);
233         if (cd->entries || atomic_read(&cd->inuse)) {
234                 write_unlock(&cd->hash_lock);
235                 spin_unlock(&cache_list_lock);
236                 return -EBUSY;
237         }
238         if (current_detail == cd)
239                 current_detail = NULL;
240         list_del_init(&cd->others);
241         write_unlock(&cd->hash_lock);
242         spin_unlock(&cache_list_lock);
243         if (cd->proc_ent) {
244                 if (cd->flush_ent)
245                         remove_proc_entry("flush", cd->proc_ent);
246                 if (cd->channel_ent)
247                         remove_proc_entry("channel", cd->proc_ent);
248                 if (cd->content_ent)
249                         remove_proc_entry("content", cd->proc_ent);
250
251                 cd->proc_ent = NULL;
252                 remove_proc_entry(cd->name, proc_net_rpc);
253         }
254         if (list_empty(&cache_list)) {
255                 /* module must be being unloaded so its safe to kill the worker */
256                 cancel_delayed_work(&cache_cleaner);
257                 flush_scheduled_work();
258         }
259         return 0;
260 }
261
262 /* clean cache tries to find something to clean
263  * and cleans it.
264  * It returns 1 if it cleaned something,
265  *            0 if it didn't find anything this time
266  *           -1 if it fell off the end of the list.
267  */
268 static int cache_clean(void)
269 {
270         int rv = 0;
271         struct list_head *next;
272
273         spin_lock(&cache_list_lock);
274
275         /* find a suitable table if we don't already have one */
276         while (current_detail == NULL ||
277             current_index >= current_detail->hash_size) {
278                 if (current_detail)
279                         next = current_detail->others.next;
280                 else
281                         next = cache_list.next;
282                 if (next == &cache_list) {
283                         current_detail = NULL;
284                         spin_unlock(&cache_list_lock);
285                         return -1;
286                 }
287                 current_detail = list_entry(next, struct cache_detail, others);
288                 if (current_detail->nextcheck > get_seconds())
289                         current_index = current_detail->hash_size;
290                 else {
291                         current_index = 0;
292                         current_detail->nextcheck = get_seconds()+30*60;
293                 }
294         }
295
296         /* find a non-empty bucket in the table */
297         while (current_detail &&
298                current_index < current_detail->hash_size &&
299                current_detail->hash_table[current_index] == NULL)
300                 current_index++;
301
302         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
303         
304         if (current_detail && current_index < current_detail->hash_size) {
305                 struct cache_head *ch, **cp;
306                 struct cache_detail *d;
307                 
308                 write_lock(&current_detail->hash_lock);
309
310                 /* Ok, now to clean this strand */
311                         
312                 cp = & current_detail->hash_table[current_index];
313                 ch = *cp;
314                 for (; ch; cp= & ch->next, ch= *cp) {
315                         if (current_detail->nextcheck > ch->expiry_time)
316                                 current_detail->nextcheck = ch->expiry_time+1;
317                         if (ch->expiry_time >= get_seconds()
318                             && ch->last_refresh >= current_detail->flush_time
319                                 )
320                                 continue;
321                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
322                                 queue_loose(current_detail, ch);
323
324                         if (atomic_read(&ch->refcnt) == 1)
325                                 break;
326                 }
327                 if (ch) {
328                         *cp = ch->next;
329                         ch->next = NULL;
330                         current_detail->entries--;
331                         rv = 1;
332                 }
333                 write_unlock(&current_detail->hash_lock);
334                 d = current_detail;
335                 if (!ch)
336                         current_index ++;
337                 spin_unlock(&cache_list_lock);
338                 if (ch)
339                         d->cache_put(ch, d);
340         } else
341                 spin_unlock(&cache_list_lock);
342
343         return rv;
344 }
345
346 /*
347  * We want to regularly clean the cache, so we need to schedule some work ...
348  */
349 static void do_cache_clean(void *data)
350 {
351         int delay = 5;
352         if (cache_clean() == -1)
353                 delay = 30*HZ;
354
355         if (list_empty(&cache_list))
356                 delay = 0;
357
358         if (delay)
359                 schedule_delayed_work(&cache_cleaner, delay);
360 }
361
362
363 /* 
364  * Clean all caches promptly.  This just calls cache_clean
365  * repeatedly until we are sure that every cache has had a chance to 
366  * be fully cleaned
367  */
368 void cache_flush(void)
369 {
370         while (cache_clean() != -1)
371                 cond_resched();
372         while (cache_clean() != -1)
373                 cond_resched();
374 }
375
376 void cache_purge(struct cache_detail *detail)
377 {
378         detail->flush_time = LONG_MAX;
379         detail->nextcheck = get_seconds();
380         cache_flush();
381         detail->flush_time = 1;
382 }
383
384
385
386 /*
387  * Deferral and Revisiting of Requests.
388  *
389  * If a cache lookup finds a pending entry, we
390  * need to defer the request and revisit it later.
391  * All deferred requests are stored in a hash table,
392  * indexed by "struct cache_head *".
393  * As it may be wasteful to store a whole request
394  * structure, we allow the request to provide a 
395  * deferred form, which must contain a
396  * 'struct cache_deferred_req'
397  * This cache_deferred_req contains a method to allow
398  * it to be revisited when cache info is available
399  */
400
401 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
402 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
403
404 #define DFR_MAX 300     /* ??? */
405
406 static DEFINE_SPINLOCK(cache_defer_lock);
407 static LIST_HEAD(cache_defer_list);
408 static struct list_head cache_defer_hash[DFR_HASHSIZE];
409 static int cache_defer_cnt;
410
411 static void cache_defer_req(struct cache_req *req, struct cache_head *item)
412 {
413         struct cache_deferred_req *dreq;
414         int hash = DFR_HASH(item);
415
416         dreq = req->defer(req);
417         if (dreq == NULL)
418                 return;
419
420         dreq->item = item;
421         dreq->recv_time = get_seconds();
422
423         spin_lock(&cache_defer_lock);
424
425         list_add(&dreq->recent, &cache_defer_list);
426
427         if (cache_defer_hash[hash].next == NULL)
428                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
429         list_add(&dreq->hash, &cache_defer_hash[hash]);
430
431         /* it is in, now maybe clean up */
432         dreq = NULL;
433         if (++cache_defer_cnt > DFR_MAX) {
434                 /* too much in the cache, randomly drop
435                  * first or last
436                  */
437                 if (net_random()&1) 
438                         dreq = list_entry(cache_defer_list.next,
439                                           struct cache_deferred_req,
440                                           recent);
441                 else
442                         dreq = list_entry(cache_defer_list.prev,
443                                           struct cache_deferred_req,
444                                           recent);
445                 list_del(&dreq->recent);
446                 list_del(&dreq->hash);
447                 cache_defer_cnt--;
448         }
449         spin_unlock(&cache_defer_lock);
450
451         if (dreq) {
452                 /* there was one too many */
453                 dreq->revisit(dreq, 1);
454         }
455         if (test_bit(CACHE_VALID, &item->flags)) {
456                 /* must have just been validated... */
457                 cache_revisit_request(item);
458         }
459 }
460
461 static void cache_revisit_request(struct cache_head *item)
462 {
463         struct cache_deferred_req *dreq;
464         struct list_head pending;
465
466         struct list_head *lp;
467         int hash = DFR_HASH(item);
468
469         INIT_LIST_HEAD(&pending);
470         spin_lock(&cache_defer_lock);
471         
472         lp = cache_defer_hash[hash].next;
473         if (lp) {
474                 while (lp != &cache_defer_hash[hash]) {
475                         dreq = list_entry(lp, struct cache_deferred_req, hash);
476                         lp = lp->next;
477                         if (dreq->item == item) {
478                                 list_del(&dreq->hash);
479                                 list_move(&dreq->recent, &pending);
480                                 cache_defer_cnt--;
481                         }
482                 }
483         }
484         spin_unlock(&cache_defer_lock);
485
486         while (!list_empty(&pending)) {
487                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
488                 list_del_init(&dreq->recent);
489                 dreq->revisit(dreq, 0);
490         }
491 }
492
493 void cache_clean_deferred(void *owner)
494 {
495         struct cache_deferred_req *dreq, *tmp;
496         struct list_head pending;
497
498
499         INIT_LIST_HEAD(&pending);
500         spin_lock(&cache_defer_lock);
501         
502         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
503                 if (dreq->owner == owner) {
504                         list_del(&dreq->hash);
505                         list_move(&dreq->recent, &pending);
506                         cache_defer_cnt--;
507                 }
508         }
509         spin_unlock(&cache_defer_lock);
510
511         while (!list_empty(&pending)) {
512                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
513                 list_del_init(&dreq->recent);
514                 dreq->revisit(dreq, 1);
515         }
516 }
517
518 /*
519  * communicate with user-space
520  *
521  * We have a magic /proc file - /proc/sunrpc/cache
522  * On read, you get a full request, or block
523  * On write, an update request is processed
524  * Poll works if anything to read, and always allows write
525  *
526  * Implemented by linked list of requests.  Each open file has 
527  * a ->private that also exists in this list.  New request are added
528  * to the end and may wakeup and preceding readers.
529  * New readers are added to the head.  If, on read, an item is found with
530  * CACHE_UPCALLING clear, we free it from the list.
531  *
532  */
533
534 static DEFINE_SPINLOCK(queue_lock);
535 static DECLARE_MUTEX(queue_io_sem);
536
537 struct cache_queue {
538         struct list_head        list;
539         int                     reader; /* if 0, then request */
540 };
541 struct cache_request {
542         struct cache_queue      q;
543         struct cache_head       *item;
544         char                    * buf;
545         int                     len;
546         int                     readers;
547 };
548 struct cache_reader {
549         struct cache_queue      q;
550         int                     offset; /* if non-0, we have a refcnt on next request */
551 };
552
553 static ssize_t
554 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
555 {
556         struct cache_reader *rp = filp->private_data;
557         struct cache_request *rq;
558         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
559         int err;
560
561         if (count == 0)
562                 return 0;
563
564         down(&queue_io_sem); /* protect against multiple concurrent
565                               * readers on this file */
566  again:
567         spin_lock(&queue_lock);
568         /* need to find next request */
569         while (rp->q.list.next != &cd->queue &&
570                list_entry(rp->q.list.next, struct cache_queue, list)
571                ->reader) {
572                 struct list_head *next = rp->q.list.next;
573                 list_move(&rp->q.list, next);
574         }
575         if (rp->q.list.next == &cd->queue) {
576                 spin_unlock(&queue_lock);
577                 up(&queue_io_sem);
578                 BUG_ON(rp->offset);
579                 return 0;
580         }
581         rq = container_of(rp->q.list.next, struct cache_request, q.list);
582         BUG_ON(rq->q.reader);
583         if (rp->offset == 0)
584                 rq->readers++;
585         spin_unlock(&queue_lock);
586
587         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
588                 err = -EAGAIN;
589                 spin_lock(&queue_lock);
590                 list_move(&rp->q.list, &rq->q.list);
591                 spin_unlock(&queue_lock);
592         } else {
593                 if (rp->offset + count > rq->len)
594                         count = rq->len - rp->offset;
595                 err = -EFAULT;
596                 if (copy_to_user(buf, rq->buf + rp->offset, count))
597                         goto out;
598                 rp->offset += count;
599                 if (rp->offset >= rq->len) {
600                         rp->offset = 0;
601                         spin_lock(&queue_lock);
602                         list_move(&rp->q.list, &rq->q.list);
603                         spin_unlock(&queue_lock);
604                 }
605                 err = 0;
606         }
607  out:
608         if (rp->offset == 0) {
609                 /* need to release rq */
610                 spin_lock(&queue_lock);
611                 rq->readers--;
612                 if (rq->readers == 0 &&
613                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
614                         list_del(&rq->q.list);
615                         spin_unlock(&queue_lock);
616                         cd->cache_put(rq->item, cd);
617                         kfree(rq->buf);
618                         kfree(rq);
619                 } else
620                         spin_unlock(&queue_lock);
621         }
622         if (err == -EAGAIN)
623                 goto again;
624         up(&queue_io_sem);
625         return err ? err :  count;
626 }
627
628 static char write_buf[8192]; /* protected by queue_io_sem */
629
630 static ssize_t
631 cache_write(struct file *filp, const char __user *buf, size_t count,
632             loff_t *ppos)
633 {
634         int err;
635         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
636
637         if (count == 0)
638                 return 0;
639         if (count >= sizeof(write_buf))
640                 return -EINVAL;
641
642         down(&queue_io_sem);
643
644         if (copy_from_user(write_buf, buf, count)) {
645                 up(&queue_io_sem);
646                 return -EFAULT;
647         }
648         write_buf[count] = '\0';
649         if (cd->cache_parse)
650                 err = cd->cache_parse(cd, write_buf, count);
651         else
652                 err = -EINVAL;
653
654         up(&queue_io_sem);
655         return err ? err : count;
656 }
657
658 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
659
660 static unsigned int
661 cache_poll(struct file *filp, poll_table *wait)
662 {
663         unsigned int mask;
664         struct cache_reader *rp = filp->private_data;
665         struct cache_queue *cq;
666         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
667
668         poll_wait(filp, &queue_wait, wait);
669
670         /* alway allow write */
671         mask = POLL_OUT | POLLWRNORM;
672
673         if (!rp)
674                 return mask;
675
676         spin_lock(&queue_lock);
677
678         for (cq= &rp->q; &cq->list != &cd->queue;
679              cq = list_entry(cq->list.next, struct cache_queue, list))
680                 if (!cq->reader) {
681                         mask |= POLLIN | POLLRDNORM;
682                         break;
683                 }
684         spin_unlock(&queue_lock);
685         return mask;
686 }
687
688 static int
689 cache_ioctl(struct inode *ino, struct file *filp,
690             unsigned int cmd, unsigned long arg)
691 {
692         int len = 0;
693         struct cache_reader *rp = filp->private_data;
694         struct cache_queue *cq;
695         struct cache_detail *cd = PDE(ino)->data;
696
697         if (cmd != FIONREAD || !rp)
698                 return -EINVAL;
699
700         spin_lock(&queue_lock);
701
702         /* only find the length remaining in current request,
703          * or the length of the next request
704          */
705         for (cq= &rp->q; &cq->list != &cd->queue;
706              cq = list_entry(cq->list.next, struct cache_queue, list))
707                 if (!cq->reader) {
708                         struct cache_request *cr =
709                                 container_of(cq, struct cache_request, q);
710                         len = cr->len - rp->offset;
711                         break;
712                 }
713         spin_unlock(&queue_lock);
714
715         return put_user(len, (int __user *)arg);
716 }
717
718 static int
719 cache_open(struct inode *inode, struct file *filp)
720 {
721         struct cache_reader *rp = NULL;
722
723         nonseekable_open(inode, filp);
724         if (filp->f_mode & FMODE_READ) {
725                 struct cache_detail *cd = PDE(inode)->data;
726
727                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
728                 if (!rp)
729                         return -ENOMEM;
730                 rp->offset = 0;
731                 rp->q.reader = 1;
732                 atomic_inc(&cd->readers);
733                 spin_lock(&queue_lock);
734                 list_add(&rp->q.list, &cd->queue);
735                 spin_unlock(&queue_lock);
736         }
737         filp->private_data = rp;
738         return 0;
739 }
740
741 static int
742 cache_release(struct inode *inode, struct file *filp)
743 {
744         struct cache_reader *rp = filp->private_data;
745         struct cache_detail *cd = PDE(inode)->data;
746
747         if (rp) {
748                 spin_lock(&queue_lock);
749                 if (rp->offset) {
750                         struct cache_queue *cq;
751                         for (cq= &rp->q; &cq->list != &cd->queue;
752                              cq = list_entry(cq->list.next, struct cache_queue, list))
753                                 if (!cq->reader) {
754                                         container_of(cq, struct cache_request, q)
755                                                 ->readers--;
756                                         break;
757                                 }
758                         rp->offset = 0;
759                 }
760                 list_del(&rp->q.list);
761                 spin_unlock(&queue_lock);
762
763                 filp->private_data = NULL;
764                 kfree(rp);
765
766                 cd->last_close = get_seconds();
767                 atomic_dec(&cd->readers);
768         }
769         return 0;
770 }
771
772
773
774 static struct file_operations cache_file_operations = {
775         .owner          = THIS_MODULE,
776         .llseek         = no_llseek,
777         .read           = cache_read,
778         .write          = cache_write,
779         .poll           = cache_poll,
780         .ioctl          = cache_ioctl, /* for FIONREAD */
781         .open           = cache_open,
782         .release        = cache_release,
783 };
784
785
786 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
787 {
788         struct cache_queue *cq;
789         spin_lock(&queue_lock);
790         list_for_each_entry(cq, &detail->queue, list)
791                 if (!cq->reader) {
792                         struct cache_request *cr = container_of(cq, struct cache_request, q);
793                         if (cr->item != ch)
794                                 continue;
795                         if (cr->readers != 0)
796                                 break;
797                         list_del(&cr->q.list);
798                         spin_unlock(&queue_lock);
799                         detail->cache_put(cr->item, detail);
800                         kfree(cr->buf);
801                         kfree(cr);
802                         return;
803                 }
804         spin_unlock(&queue_lock);
805 }
806
807 /*
808  * Support routines for text-based upcalls.
809  * Fields are separated by spaces.
810  * Fields are either mangled to quote space tab newline slosh with slosh
811  * or a hexified with a leading \x
812  * Record is terminated with newline.
813  *
814  */
815
816 void qword_add(char **bpp, int *lp, char *str)
817 {
818         char *bp = *bpp;
819         int len = *lp;
820         char c;
821
822         if (len < 0) return;
823
824         while ((c=*str++) && len)
825                 switch(c) {
826                 case ' ':
827                 case '\t':
828                 case '\n':
829                 case '\\':
830                         if (len >= 4) {
831                                 *bp++ = '\\';
832                                 *bp++ = '0' + ((c & 0300)>>6);
833                                 *bp++ = '0' + ((c & 0070)>>3);
834                                 *bp++ = '0' + ((c & 0007)>>0);
835                         }
836                         len -= 4;
837                         break;
838                 default:
839                         *bp++ = c;
840                         len--;
841                 }
842         if (c || len <1) len = -1;
843         else {
844                 *bp++ = ' ';
845                 len--;
846         }
847         *bpp = bp;
848         *lp = len;
849 }
850
851 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
852 {
853         char *bp = *bpp;
854         int len = *lp;
855
856         if (len < 0) return;
857
858         if (len > 2) {
859                 *bp++ = '\\';
860                 *bp++ = 'x';
861                 len -= 2;
862                 while (blen && len >= 2) {
863                         unsigned char c = *buf++;
864                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
865                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
866                         len -= 2;
867                         blen--;
868                 }
869         }
870         if (blen || len<1) len = -1;
871         else {
872                 *bp++ = ' ';
873                 len--;
874         }
875         *bpp = bp;
876         *lp = len;
877 }
878
879 static void warn_no_listener(struct cache_detail *detail)
880 {
881         if (detail->last_warn != detail->last_close) {
882                 detail->last_warn = detail->last_close;
883                 if (detail->warn_no_listener)
884                         detail->warn_no_listener(detail);
885         }
886 }
887
888 /*
889  * register an upcall request to user-space.
890  * Each request is at most one page long.
891  */
892 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
893 {
894
895         char *buf;
896         struct cache_request *crq;
897         char *bp;
898         int len;
899
900         if (detail->cache_request == NULL)
901                 return -EINVAL;
902
903         if (atomic_read(&detail->readers) == 0 &&
904             detail->last_close < get_seconds() - 30) {
905                         warn_no_listener(detail);
906                         return -EINVAL;
907         }
908
909         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
910         if (!buf)
911                 return -EAGAIN;
912
913         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
914         if (!crq) {
915                 kfree(buf);
916                 return -EAGAIN;
917         }
918
919         bp = buf; len = PAGE_SIZE;
920
921         detail->cache_request(detail, h, &bp, &len);
922
923         if (len < 0) {
924                 kfree(buf);
925                 kfree(crq);
926                 return -EAGAIN;
927         }
928         crq->q.reader = 0;
929         crq->item = cache_get(h);
930         crq->buf = buf;
931         crq->len = PAGE_SIZE - len;
932         crq->readers = 0;
933         spin_lock(&queue_lock);
934         list_add_tail(&crq->q.list, &detail->queue);
935         spin_unlock(&queue_lock);
936         wake_up(&queue_wait);
937         return 0;
938 }
939
940 /*
941  * parse a message from user-space and pass it
942  * to an appropriate cache
943  * Messages are, like requests, separated into fields by
944  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
945  *
946  * Message is 
947  *   reply cachename expiry key ... content....
948  *
949  * key and content are both parsed by cache 
950  */
951
952 #define isodigit(c) (isdigit(c) && c <= '7')
953 int qword_get(char **bpp, char *dest, int bufsize)
954 {
955         /* return bytes copied, or -1 on error */
956         char *bp = *bpp;
957         int len = 0;
958
959         while (*bp == ' ') bp++;
960
961         if (bp[0] == '\\' && bp[1] == 'x') {
962                 /* HEX STRING */
963                 bp += 2;
964                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
965                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
966                         bp++;
967                         byte <<= 4;
968                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
969                         *dest++ = byte;
970                         bp++;
971                         len++;
972                 }
973         } else {
974                 /* text with \nnn octal quoting */
975                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
976                         if (*bp == '\\' &&
977                             isodigit(bp[1]) && (bp[1] <= '3') &&
978                             isodigit(bp[2]) &&
979                             isodigit(bp[3])) {
980                                 int byte = (*++bp -'0');
981                                 bp++;
982                                 byte = (byte << 3) | (*bp++ - '0');
983                                 byte = (byte << 3) | (*bp++ - '0');
984                                 *dest++ = byte;
985                                 len++;
986                         } else {
987                                 *dest++ = *bp++;
988                                 len++;
989                         }
990                 }
991         }
992
993         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
994                 return -1;
995         while (*bp == ' ') bp++;
996         *bpp = bp;
997         *dest = '\0';
998         return len;
999 }
1000
1001
1002 /*
1003  * support /proc/sunrpc/cache/$CACHENAME/content
1004  * as a seqfile.
1005  * We call ->cache_show passing NULL for the item to
1006  * get a header, then pass each real item in the cache
1007  */
1008
1009 struct handle {
1010         struct cache_detail *cd;
1011 };
1012
1013 static void *c_start(struct seq_file *m, loff_t *pos)
1014 {
1015         loff_t n = *pos;
1016         unsigned hash, entry;
1017         struct cache_head *ch;
1018         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1019         
1020
1021         read_lock(&cd->hash_lock);
1022         if (!n--)
1023                 return SEQ_START_TOKEN;
1024         hash = n >> 32;
1025         entry = n & ((1LL<<32) - 1);
1026
1027         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1028                 if (!entry--)
1029                         return ch;
1030         n &= ~((1LL<<32) - 1);
1031         do {
1032                 hash++;
1033                 n += 1LL<<32;
1034         } while(hash < cd->hash_size && 
1035                 cd->hash_table[hash]==NULL);
1036         if (hash >= cd->hash_size)
1037                 return NULL;
1038         *pos = n+1;
1039         return cd->hash_table[hash];
1040 }
1041
1042 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1043 {
1044         struct cache_head *ch = p;
1045         int hash = (*pos >> 32);
1046         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1047
1048         if (p == SEQ_START_TOKEN)
1049                 hash = 0;
1050         else if (ch->next == NULL) {
1051                 hash++;
1052                 *pos += 1LL<<32;
1053         } else {
1054                 ++*pos;
1055                 return ch->next;
1056         }
1057         *pos &= ~((1LL<<32) - 1);
1058         while (hash < cd->hash_size &&
1059                cd->hash_table[hash] == NULL) {
1060                 hash++;
1061                 *pos += 1LL<<32;
1062         }
1063         if (hash >= cd->hash_size)
1064                 return NULL;
1065         ++*pos;
1066         return cd->hash_table[hash];
1067 }
1068
1069 static void c_stop(struct seq_file *m, void *p)
1070 {
1071         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1072         read_unlock(&cd->hash_lock);
1073 }
1074
1075 static int c_show(struct seq_file *m, void *p)
1076 {
1077         struct cache_head *cp = p;
1078         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1079
1080         if (p == SEQ_START_TOKEN)
1081                 return cd->cache_show(m, cd, NULL);
1082
1083         ifdebug(CACHE)
1084                 seq_printf(m, "# expiry=%ld refcnt=%d\n",
1085                            cp->expiry_time, atomic_read(&cp->refcnt));
1086         cache_get(cp);
1087         if (cache_check(cd, cp, NULL))
1088                 /* cache_check does a cache_put on failure */
1089                 seq_printf(m, "# ");
1090         else
1091                 cache_put(cp, cd);
1092
1093         return cd->cache_show(m, cd, cp);
1094 }
1095
1096 static struct seq_operations cache_content_op = {
1097         .start  = c_start,
1098         .next   = c_next,
1099         .stop   = c_stop,
1100         .show   = c_show,
1101 };
1102
1103 static int content_open(struct inode *inode, struct file *file)
1104 {
1105         int res;
1106         struct handle *han;
1107         struct cache_detail *cd = PDE(inode)->data;
1108
1109         han = kmalloc(sizeof(*han), GFP_KERNEL);
1110         if (han == NULL)
1111                 return -ENOMEM;
1112
1113         han->cd = cd;
1114
1115         res = seq_open(file, &cache_content_op);
1116         if (res)
1117                 kfree(han);
1118         else
1119                 ((struct seq_file *)file->private_data)->private = han;
1120
1121         return res;
1122 }
1123 static int content_release(struct inode *inode, struct file *file)
1124 {
1125         struct seq_file *m = (struct seq_file *)file->private_data;
1126         struct handle *han = m->private;
1127         kfree(han);
1128         m->private = NULL;
1129         return seq_release(inode, file);
1130 }
1131
1132 static struct file_operations content_file_operations = {
1133         .open           = content_open,
1134         .read           = seq_read,
1135         .llseek         = seq_lseek,
1136         .release        = content_release,
1137 };
1138
1139 static ssize_t read_flush(struct file *file, char __user *buf,
1140                             size_t count, loff_t *ppos)
1141 {
1142         struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1143         char tbuf[20];
1144         unsigned long p = *ppos;
1145         int len;
1146
1147         sprintf(tbuf, "%lu\n", cd->flush_time);
1148         len = strlen(tbuf);
1149         if (p >= len)
1150                 return 0;
1151         len -= p;
1152         if (len > count) len = count;
1153         if (copy_to_user(buf, (void*)(tbuf+p), len))
1154                 len = -EFAULT;
1155         else
1156                 *ppos += len;
1157         return len;
1158 }
1159
1160 static ssize_t write_flush(struct file * file, const char __user * buf,
1161                              size_t count, loff_t *ppos)
1162 {
1163         struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1164         char tbuf[20];
1165         char *ep;
1166         long flushtime;
1167         if (*ppos || count > sizeof(tbuf)-1)
1168                 return -EINVAL;
1169         if (copy_from_user(tbuf, buf, count))
1170                 return -EFAULT;
1171         tbuf[count] = 0;
1172         flushtime = simple_strtoul(tbuf, &ep, 0);
1173         if (*ep && *ep != '\n')
1174                 return -EINVAL;
1175
1176         cd->flush_time = flushtime;
1177         cd->nextcheck = get_seconds();
1178         cache_flush();
1179
1180         *ppos += count;
1181         return count;
1182 }
1183
1184 static struct file_operations cache_flush_operations = {
1185         .open           = nonseekable_open,
1186         .read           = read_flush,
1187         .write          = write_flush,
1188 };