4 * Generic code for various authentication-related caches
5 * used by sunrpc clients and servers.
7 * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
9 * Released under terms in GPL version 2. See COPYING.
13 #include <linux/types.h>
15 #include <linux/file.h>
16 #include <linux/slab.h>
17 #include <linux/signal.h>
18 #include <linux/sched.h>
19 #include <linux/kmod.h>
20 #include <linux/list.h>
21 #include <linux/module.h>
22 #include <linux/ctype.h>
23 #include <asm/uaccess.h>
24 #include <linux/poll.h>
25 #include <linux/seq_file.h>
26 #include <linux/proc_fs.h>
27 #include <linux/net.h>
28 #include <linux/workqueue.h>
29 #include <linux/mutex.h>
30 #include <linux/pagemap.h>
31 #include <asm/ioctls.h>
32 #include <linux/sunrpc/types.h>
33 #include <linux/sunrpc/cache.h>
34 #include <linux/sunrpc/stats.h>
35 #include <linux/sunrpc/rpc_pipe_fs.h>
38 #define RPCDBG_FACILITY RPCDBG_CACHE
40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
41 static void cache_revisit_request(struct cache_head *item);
43 static void cache_init(struct cache_head *h)
45 time_t now = seconds_since_boot();
49 h->expiry_time = now + CACHE_NEW_EXPIRY;
50 h->last_refresh = now;
53 static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
55 return (h->expiry_time < seconds_since_boot()) ||
56 (detail->flush_time > h->last_refresh);
59 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
60 struct cache_head *key, int hash)
62 struct cache_head **head, **hp;
63 struct cache_head *new = NULL, *freeme = NULL;
65 head = &detail->hash_table[hash];
67 read_lock(&detail->hash_lock);
69 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
70 struct cache_head *tmp = *hp;
71 if (detail->match(tmp, key)) {
72 if (cache_is_expired(detail, tmp))
73 /* This entry is expired, we will discard it. */
76 read_unlock(&detail->hash_lock);
80 read_unlock(&detail->hash_lock);
81 /* Didn't find anything, insert an empty entry */
83 new = detail->alloc();
86 /* must fully initialise 'new', else
87 * we might get lose if we need to
91 detail->init(new, key);
93 write_lock(&detail->hash_lock);
95 /* check if entry appeared while we slept */
96 for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
97 struct cache_head *tmp = *hp;
98 if (detail->match(tmp, key)) {
99 if (cache_is_expired(detail, tmp)) {
107 write_unlock(&detail->hash_lock);
108 cache_put(new, detail);
116 write_unlock(&detail->hash_lock);
119 cache_put(freeme, detail);
122 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
125 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
127 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
129 head->expiry_time = expiry;
130 head->last_refresh = seconds_since_boot();
131 smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
132 set_bit(CACHE_VALID, &head->flags);
135 static void cache_fresh_unlocked(struct cache_head *head,
136 struct cache_detail *detail)
138 if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
139 cache_revisit_request(head);
140 cache_dequeue(detail, head);
144 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
145 struct cache_head *new, struct cache_head *old, int hash)
147 /* The 'old' entry is to be replaced by 'new'.
148 * If 'old' is not VALID, we update it directly,
149 * otherwise we need to replace it
151 struct cache_head **head;
152 struct cache_head *tmp;
154 if (!test_bit(CACHE_VALID, &old->flags)) {
155 write_lock(&detail->hash_lock);
156 if (!test_bit(CACHE_VALID, &old->flags)) {
157 if (test_bit(CACHE_NEGATIVE, &new->flags))
158 set_bit(CACHE_NEGATIVE, &old->flags);
160 detail->update(old, new);
161 cache_fresh_locked(old, new->expiry_time);
162 write_unlock(&detail->hash_lock);
163 cache_fresh_unlocked(old, detail);
166 write_unlock(&detail->hash_lock);
168 /* We need to insert a new entry */
169 tmp = detail->alloc();
171 cache_put(old, detail);
175 detail->init(tmp, old);
176 head = &detail->hash_table[hash];
178 write_lock(&detail->hash_lock);
179 if (test_bit(CACHE_NEGATIVE, &new->flags))
180 set_bit(CACHE_NEGATIVE, &tmp->flags);
182 detail->update(tmp, new);
187 cache_fresh_locked(tmp, new->expiry_time);
188 cache_fresh_locked(old, 0);
189 write_unlock(&detail->hash_lock);
190 cache_fresh_unlocked(tmp, detail);
191 cache_fresh_unlocked(old, detail);
192 cache_put(old, detail);
195 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
197 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
199 if (!cd->cache_upcall)
201 return cd->cache_upcall(cd, h);
204 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
206 if (!test_bit(CACHE_VALID, &h->flags))
210 if (test_bit(CACHE_NEGATIVE, &h->flags))
214 * In combination with write barrier in
215 * sunrpc_cache_update, ensures that anyone
216 * using the cache entry after this sees the
225 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
229 write_lock(&detail->hash_lock);
230 rv = cache_is_valid(detail, h);
232 write_unlock(&detail->hash_lock);
235 set_bit(CACHE_NEGATIVE, &h->flags);
236 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
237 write_unlock(&detail->hash_lock);
238 cache_fresh_unlocked(h, detail);
243 * This is the generic cache management routine for all
244 * the authentication caches.
245 * It checks the currency of a cache item and will (later)
246 * initiate an upcall to fill it if needed.
249 * Returns 0 if the cache_head can be used, or cache_puts it and returns
250 * -EAGAIN if upcall is pending and request has been queued
251 * -ETIMEDOUT if upcall failed or request could not be queue or
252 * upcall completed but item is still invalid (implying that
253 * the cache item has been replaced with a newer one).
254 * -ENOENT if cache entry was negative
256 int cache_check(struct cache_detail *detail,
257 struct cache_head *h, struct cache_req *rqstp)
260 long refresh_age, age;
262 /* First decide return status as best we can */
263 rv = cache_is_valid(detail, h);
265 /* now see if we want to start an upcall */
266 refresh_age = (h->expiry_time - h->last_refresh);
267 age = seconds_since_boot() - h->last_refresh;
272 } else if (rv == -EAGAIN || age > refresh_age/2) {
273 dprintk("RPC: Want update, refage=%ld, age=%ld\n",
275 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
276 switch (cache_make_upcall(detail, h)) {
278 clear_bit(CACHE_PENDING, &h->flags);
279 cache_revisit_request(h);
280 rv = try_to_negate_entry(detail, h);
283 clear_bit(CACHE_PENDING, &h->flags);
284 cache_revisit_request(h);
291 if (!cache_defer_req(rqstp, h)) {
293 * Request was not deferred; handle it as best
296 rv = cache_is_valid(detail, h);
302 cache_put(h, detail);
305 EXPORT_SYMBOL_GPL(cache_check);
308 * caches need to be periodically cleaned.
309 * For this we maintain a list of cache_detail and
310 * a current pointer into that list and into the table
313 * Each time clean_cache is called it finds the next non-empty entry
314 * in the current table and walks the list in that entry
315 * looking for entries that can be removed.
317 * An entry gets removed if:
318 * - The expiry is before current time
319 * - The last_refresh time is before the flush_time for that cache
321 * later we might drop old entries with non-NEVER expiry if that table
322 * is getting 'full' for some definition of 'full'
324 * The question of "how often to scan a table" is an interesting one
325 * and is answered in part by the use of the "nextcheck" field in the
327 * When a scan of a table begins, the nextcheck field is set to a time
328 * that is well into the future.
329 * While scanning, if an expiry time is found that is earlier than the
330 * current nextcheck time, nextcheck is set to that expiry time.
331 * If the flush_time is ever set to a time earlier than the nextcheck
332 * time, the nextcheck time is then set to that flush_time.
334 * A table is then only scanned if the current time is at least
335 * the nextcheck time.
339 static LIST_HEAD(cache_list);
340 static DEFINE_SPINLOCK(cache_list_lock);
341 static struct cache_detail *current_detail;
342 static int current_index;
344 static void do_cache_clean(struct work_struct *work);
345 static struct delayed_work cache_cleaner;
347 void sunrpc_init_cache_detail(struct cache_detail *cd)
349 rwlock_init(&cd->hash_lock);
350 INIT_LIST_HEAD(&cd->queue);
351 spin_lock(&cache_list_lock);
354 atomic_set(&cd->readers, 0);
357 list_add(&cd->others, &cache_list);
358 spin_unlock(&cache_list_lock);
360 /* start the cleaning process */
361 schedule_delayed_work(&cache_cleaner, 0);
363 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
365 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
368 spin_lock(&cache_list_lock);
369 write_lock(&cd->hash_lock);
370 if (cd->entries || atomic_read(&cd->inuse)) {
371 write_unlock(&cd->hash_lock);
372 spin_unlock(&cache_list_lock);
375 if (current_detail == cd)
376 current_detail = NULL;
377 list_del_init(&cd->others);
378 write_unlock(&cd->hash_lock);
379 spin_unlock(&cache_list_lock);
380 if (list_empty(&cache_list)) {
381 /* module must be being unloaded so its safe to kill the worker */
382 cancel_delayed_work_sync(&cache_cleaner);
386 printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
388 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
390 /* clean cache tries to find something to clean
392 * It returns 1 if it cleaned something,
393 * 0 if it didn't find anything this time
394 * -1 if it fell off the end of the list.
396 static int cache_clean(void)
399 struct list_head *next;
401 spin_lock(&cache_list_lock);
403 /* find a suitable table if we don't already have one */
404 while (current_detail == NULL ||
405 current_index >= current_detail->hash_size) {
407 next = current_detail->others.next;
409 next = cache_list.next;
410 if (next == &cache_list) {
411 current_detail = NULL;
412 spin_unlock(&cache_list_lock);
415 current_detail = list_entry(next, struct cache_detail, others);
416 if (current_detail->nextcheck > seconds_since_boot())
417 current_index = current_detail->hash_size;
420 current_detail->nextcheck = seconds_since_boot()+30*60;
424 /* find a non-empty bucket in the table */
425 while (current_detail &&
426 current_index < current_detail->hash_size &&
427 current_detail->hash_table[current_index] == NULL)
430 /* find a cleanable entry in the bucket and clean it, or set to next bucket */
432 if (current_detail && current_index < current_detail->hash_size) {
433 struct cache_head *ch, **cp;
434 struct cache_detail *d;
436 write_lock(¤t_detail->hash_lock);
438 /* Ok, now to clean this strand */
440 cp = & current_detail->hash_table[current_index];
441 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
442 if (current_detail->nextcheck > ch->expiry_time)
443 current_detail->nextcheck = ch->expiry_time+1;
444 if (!cache_is_expired(current_detail, ch))
449 current_detail->entries--;
454 write_unlock(¤t_detail->hash_lock);
458 spin_unlock(&cache_list_lock);
460 if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
461 cache_dequeue(current_detail, ch);
462 cache_revisit_request(ch);
466 spin_unlock(&cache_list_lock);
472 * We want to regularly clean the cache, so we need to schedule some work ...
474 static void do_cache_clean(struct work_struct *work)
477 if (cache_clean() == -1)
478 delay = round_jiffies_relative(30*HZ);
480 if (list_empty(&cache_list))
484 schedule_delayed_work(&cache_cleaner, delay);
489 * Clean all caches promptly. This just calls cache_clean
490 * repeatedly until we are sure that every cache has had a chance to
493 void cache_flush(void)
495 while (cache_clean() != -1)
497 while (cache_clean() != -1)
500 EXPORT_SYMBOL_GPL(cache_flush);
502 void cache_purge(struct cache_detail *detail)
504 detail->flush_time = LONG_MAX;
505 detail->nextcheck = seconds_since_boot();
507 detail->flush_time = 1;
509 EXPORT_SYMBOL_GPL(cache_purge);
513 * Deferral and Revisiting of Requests.
515 * If a cache lookup finds a pending entry, we
516 * need to defer the request and revisit it later.
517 * All deferred requests are stored in a hash table,
518 * indexed by "struct cache_head *".
519 * As it may be wasteful to store a whole request
520 * structure, we allow the request to provide a
521 * deferred form, which must contain a
522 * 'struct cache_deferred_req'
523 * This cache_deferred_req contains a method to allow
524 * it to be revisited when cache info is available
527 #define DFR_HASHSIZE (PAGE_SIZE/sizeof(struct list_head))
528 #define DFR_HASH(item) ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
530 #define DFR_MAX 300 /* ??? */
532 static DEFINE_SPINLOCK(cache_defer_lock);
533 static LIST_HEAD(cache_defer_list);
534 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
535 static int cache_defer_cnt;
537 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
539 hlist_del_init(&dreq->hash);
540 if (!list_empty(&dreq->recent)) {
541 list_del_init(&dreq->recent);
546 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
548 int hash = DFR_HASH(item);
550 INIT_LIST_HEAD(&dreq->recent);
551 hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
554 static void setup_deferral(struct cache_deferred_req *dreq,
555 struct cache_head *item,
561 spin_lock(&cache_defer_lock);
563 __hash_deferred_req(dreq, item);
567 list_add(&dreq->recent, &cache_defer_list);
570 spin_unlock(&cache_defer_lock);
574 struct thread_deferred_req {
575 struct cache_deferred_req handle;
576 struct completion completion;
579 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
581 struct thread_deferred_req *dr =
582 container_of(dreq, struct thread_deferred_req, handle);
583 complete(&dr->completion);
586 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
588 struct thread_deferred_req sleeper;
589 struct cache_deferred_req *dreq = &sleeper.handle;
591 sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
592 dreq->revisit = cache_restart_thread;
594 setup_deferral(dreq, item, 0);
596 if (!test_bit(CACHE_PENDING, &item->flags) ||
597 wait_for_completion_interruptible_timeout(
598 &sleeper.completion, req->thread_wait) <= 0) {
599 /* The completion wasn't completed, so we need
602 spin_lock(&cache_defer_lock);
603 if (!hlist_unhashed(&sleeper.handle.hash)) {
604 __unhash_deferred_req(&sleeper.handle);
605 spin_unlock(&cache_defer_lock);
607 /* cache_revisit_request already removed
608 * this from the hash table, but hasn't
609 * called ->revisit yet. It will very soon
610 * and we need to wait for it.
612 spin_unlock(&cache_defer_lock);
613 wait_for_completion(&sleeper.completion);
618 static void cache_limit_defers(void)
620 /* Make sure we haven't exceed the limit of allowed deferred
623 struct cache_deferred_req *discard = NULL;
625 if (cache_defer_cnt <= DFR_MAX)
628 spin_lock(&cache_defer_lock);
630 /* Consider removing either the first or the last */
631 if (cache_defer_cnt > DFR_MAX) {
632 if (net_random() & 1)
633 discard = list_entry(cache_defer_list.next,
634 struct cache_deferred_req, recent);
636 discard = list_entry(cache_defer_list.prev,
637 struct cache_deferred_req, recent);
638 __unhash_deferred_req(discard);
640 spin_unlock(&cache_defer_lock);
642 discard->revisit(discard, 1);
645 /* Return true if and only if a deferred request is queued. */
646 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
648 struct cache_deferred_req *dreq;
650 if (req->thread_wait) {
651 cache_wait_req(req, item);
652 if (!test_bit(CACHE_PENDING, &item->flags))
655 dreq = req->defer(req);
658 setup_deferral(dreq, item, 1);
659 if (!test_bit(CACHE_PENDING, &item->flags))
660 /* Bit could have been cleared before we managed to
661 * set up the deferral, so need to revisit just in case
663 cache_revisit_request(item);
665 cache_limit_defers();
669 static void cache_revisit_request(struct cache_head *item)
671 struct cache_deferred_req *dreq;
672 struct list_head pending;
673 struct hlist_node *lp, *tmp;
674 int hash = DFR_HASH(item);
676 INIT_LIST_HEAD(&pending);
677 spin_lock(&cache_defer_lock);
679 hlist_for_each_entry_safe(dreq, lp, tmp, &cache_defer_hash[hash], hash)
680 if (dreq->item == item) {
681 __unhash_deferred_req(dreq);
682 list_add(&dreq->recent, &pending);
685 spin_unlock(&cache_defer_lock);
687 while (!list_empty(&pending)) {
688 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
689 list_del_init(&dreq->recent);
690 dreq->revisit(dreq, 0);
694 void cache_clean_deferred(void *owner)
696 struct cache_deferred_req *dreq, *tmp;
697 struct list_head pending;
700 INIT_LIST_HEAD(&pending);
701 spin_lock(&cache_defer_lock);
703 list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
704 if (dreq->owner == owner) {
705 __unhash_deferred_req(dreq);
706 list_add(&dreq->recent, &pending);
709 spin_unlock(&cache_defer_lock);
711 while (!list_empty(&pending)) {
712 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
713 list_del_init(&dreq->recent);
714 dreq->revisit(dreq, 1);
719 * communicate with user-space
721 * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
722 * On read, you get a full request, or block.
723 * On write, an update request is processed.
724 * Poll works if anything to read, and always allows write.
726 * Implemented by linked list of requests. Each open file has
727 * a ->private that also exists in this list. New requests are added
728 * to the end and may wakeup and preceding readers.
729 * New readers are added to the head. If, on read, an item is found with
730 * CACHE_UPCALLING clear, we free it from the list.
734 static DEFINE_SPINLOCK(queue_lock);
735 static DEFINE_MUTEX(queue_io_mutex);
738 struct list_head list;
739 int reader; /* if 0, then request */
741 struct cache_request {
742 struct cache_queue q;
743 struct cache_head *item;
748 struct cache_reader {
749 struct cache_queue q;
750 int offset; /* if non-0, we have a refcnt on next request */
753 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
754 loff_t *ppos, struct cache_detail *cd)
756 struct cache_reader *rp = filp->private_data;
757 struct cache_request *rq;
758 struct inode *inode = filp->f_path.dentry->d_inode;
764 mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
765 * readers on this file */
767 spin_lock(&queue_lock);
768 /* need to find next request */
769 while (rp->q.list.next != &cd->queue &&
770 list_entry(rp->q.list.next, struct cache_queue, list)
772 struct list_head *next = rp->q.list.next;
773 list_move(&rp->q.list, next);
775 if (rp->q.list.next == &cd->queue) {
776 spin_unlock(&queue_lock);
777 mutex_unlock(&inode->i_mutex);
781 rq = container_of(rp->q.list.next, struct cache_request, q.list);
782 BUG_ON(rq->q.reader);
785 spin_unlock(&queue_lock);
787 if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
789 spin_lock(&queue_lock);
790 list_move(&rp->q.list, &rq->q.list);
791 spin_unlock(&queue_lock);
793 if (rp->offset + count > rq->len)
794 count = rq->len - rp->offset;
796 if (copy_to_user(buf, rq->buf + rp->offset, count))
799 if (rp->offset >= rq->len) {
801 spin_lock(&queue_lock);
802 list_move(&rp->q.list, &rq->q.list);
803 spin_unlock(&queue_lock);
808 if (rp->offset == 0) {
809 /* need to release rq */
810 spin_lock(&queue_lock);
812 if (rq->readers == 0 &&
813 !test_bit(CACHE_PENDING, &rq->item->flags)) {
814 list_del(&rq->q.list);
815 spin_unlock(&queue_lock);
816 cache_put(rq->item, cd);
820 spin_unlock(&queue_lock);
824 mutex_unlock(&inode->i_mutex);
825 return err ? err : count;
828 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
829 size_t count, struct cache_detail *cd)
833 if (copy_from_user(kaddr, buf, count))
836 ret = cd->cache_parse(cd, kaddr, count);
842 static ssize_t cache_slow_downcall(const char __user *buf,
843 size_t count, struct cache_detail *cd)
845 static char write_buf[8192]; /* protected by queue_io_mutex */
846 ssize_t ret = -EINVAL;
848 if (count >= sizeof(write_buf))
850 mutex_lock(&queue_io_mutex);
851 ret = cache_do_downcall(write_buf, buf, count, cd);
852 mutex_unlock(&queue_io_mutex);
857 static ssize_t cache_downcall(struct address_space *mapping,
858 const char __user *buf,
859 size_t count, struct cache_detail *cd)
863 ssize_t ret = -ENOMEM;
865 if (count >= PAGE_CACHE_SIZE)
868 page = find_or_create_page(mapping, 0, GFP_KERNEL);
873 ret = cache_do_downcall(kaddr, buf, count, cd);
876 page_cache_release(page);
879 return cache_slow_downcall(buf, count, cd);
882 static ssize_t cache_write(struct file *filp, const char __user *buf,
883 size_t count, loff_t *ppos,
884 struct cache_detail *cd)
886 struct address_space *mapping = filp->f_mapping;
887 struct inode *inode = filp->f_path.dentry->d_inode;
888 ssize_t ret = -EINVAL;
890 if (!cd->cache_parse)
893 mutex_lock(&inode->i_mutex);
894 ret = cache_downcall(mapping, buf, count, cd);
895 mutex_unlock(&inode->i_mutex);
900 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
902 static unsigned int cache_poll(struct file *filp, poll_table *wait,
903 struct cache_detail *cd)
906 struct cache_reader *rp = filp->private_data;
907 struct cache_queue *cq;
909 poll_wait(filp, &queue_wait, wait);
911 /* alway allow write */
912 mask = POLL_OUT | POLLWRNORM;
917 spin_lock(&queue_lock);
919 for (cq= &rp->q; &cq->list != &cd->queue;
920 cq = list_entry(cq->list.next, struct cache_queue, list))
922 mask |= POLLIN | POLLRDNORM;
925 spin_unlock(&queue_lock);
929 static int cache_ioctl(struct inode *ino, struct file *filp,
930 unsigned int cmd, unsigned long arg,
931 struct cache_detail *cd)
934 struct cache_reader *rp = filp->private_data;
935 struct cache_queue *cq;
937 if (cmd != FIONREAD || !rp)
940 spin_lock(&queue_lock);
942 /* only find the length remaining in current request,
943 * or the length of the next request
945 for (cq= &rp->q; &cq->list != &cd->queue;
946 cq = list_entry(cq->list.next, struct cache_queue, list))
948 struct cache_request *cr =
949 container_of(cq, struct cache_request, q);
950 len = cr->len - rp->offset;
953 spin_unlock(&queue_lock);
955 return put_user(len, (int __user *)arg);
958 static int cache_open(struct inode *inode, struct file *filp,
959 struct cache_detail *cd)
961 struct cache_reader *rp = NULL;
963 if (!cd || !try_module_get(cd->owner))
965 nonseekable_open(inode, filp);
966 if (filp->f_mode & FMODE_READ) {
967 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
972 atomic_inc(&cd->readers);
973 spin_lock(&queue_lock);
974 list_add(&rp->q.list, &cd->queue);
975 spin_unlock(&queue_lock);
977 filp->private_data = rp;
981 static int cache_release(struct inode *inode, struct file *filp,
982 struct cache_detail *cd)
984 struct cache_reader *rp = filp->private_data;
987 spin_lock(&queue_lock);
989 struct cache_queue *cq;
990 for (cq= &rp->q; &cq->list != &cd->queue;
991 cq = list_entry(cq->list.next, struct cache_queue, list))
993 container_of(cq, struct cache_request, q)
999 list_del(&rp->q.list);
1000 spin_unlock(&queue_lock);
1002 filp->private_data = NULL;
1005 cd->last_close = seconds_since_boot();
1006 atomic_dec(&cd->readers);
1008 module_put(cd->owner);
1014 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1016 struct cache_queue *cq;
1017 spin_lock(&queue_lock);
1018 list_for_each_entry(cq, &detail->queue, list)
1020 struct cache_request *cr = container_of(cq, struct cache_request, q);
1023 if (cr->readers != 0)
1025 list_del(&cr->q.list);
1026 spin_unlock(&queue_lock);
1027 cache_put(cr->item, detail);
1032 spin_unlock(&queue_lock);
1036 * Support routines for text-based upcalls.
1037 * Fields are separated by spaces.
1038 * Fields are either mangled to quote space tab newline slosh with slosh
1039 * or a hexified with a leading \x
1040 * Record is terminated with newline.
1044 void qword_add(char **bpp, int *lp, char *str)
1050 if (len < 0) return;
1052 while ((c=*str++) && len)
1060 *bp++ = '0' + ((c & 0300)>>6);
1061 *bp++ = '0' + ((c & 0070)>>3);
1062 *bp++ = '0' + ((c & 0007)>>0);
1070 if (c || len <1) len = -1;
1078 EXPORT_SYMBOL_GPL(qword_add);
1080 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1085 if (len < 0) return;
1091 while (blen && len >= 2) {
1092 unsigned char c = *buf++;
1093 *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1094 *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1099 if (blen || len<1) len = -1;
1107 EXPORT_SYMBOL_GPL(qword_addhex);
1109 static void warn_no_listener(struct cache_detail *detail)
1111 if (detail->last_warn != detail->last_close) {
1112 detail->last_warn = detail->last_close;
1113 if (detail->warn_no_listener)
1114 detail->warn_no_listener(detail, detail->last_close != 0);
1118 static bool cache_listeners_exist(struct cache_detail *detail)
1120 if (atomic_read(&detail->readers))
1122 if (detail->last_close == 0)
1123 /* This cache was never opened */
1125 if (detail->last_close < seconds_since_boot() - 30)
1127 * We allow for the possibility that someone might
1128 * restart a userspace daemon without restarting the
1129 * server; but after 30 seconds, we give up.
1136 * register an upcall request to user-space and queue it up for read() by the
1139 * Each request is at most one page long.
1141 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1142 void (*cache_request)(struct cache_detail *,
1143 struct cache_head *,
1149 struct cache_request *crq;
1153 if (!cache_listeners_exist(detail)) {
1154 warn_no_listener(detail);
1158 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1162 crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1168 bp = buf; len = PAGE_SIZE;
1170 cache_request(detail, h, &bp, &len);
1178 crq->item = cache_get(h);
1180 crq->len = PAGE_SIZE - len;
1182 spin_lock(&queue_lock);
1183 list_add_tail(&crq->q.list, &detail->queue);
1184 spin_unlock(&queue_lock);
1185 wake_up(&queue_wait);
1188 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1191 * parse a message from user-space and pass it
1192 * to an appropriate cache
1193 * Messages are, like requests, separated into fields by
1194 * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1197 * reply cachename expiry key ... content....
1199 * key and content are both parsed by cache
1202 #define isodigit(c) (isdigit(c) && c <= '7')
1203 int qword_get(char **bpp, char *dest, int bufsize)
1205 /* return bytes copied, or -1 on error */
1209 while (*bp == ' ') bp++;
1211 if (bp[0] == '\\' && bp[1] == 'x') {
1214 while (len < bufsize) {
1217 h = hex_to_bin(bp[0]);
1221 l = hex_to_bin(bp[1]);
1225 *dest++ = (h << 4) | l;
1230 /* text with \nnn octal quoting */
1231 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1233 isodigit(bp[1]) && (bp[1] <= '3') &&
1236 int byte = (*++bp -'0');
1238 byte = (byte << 3) | (*bp++ - '0');
1239 byte = (byte << 3) | (*bp++ - '0');
1249 if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1251 while (*bp == ' ') bp++;
1256 EXPORT_SYMBOL_GPL(qword_get);
1260 * support /proc/sunrpc/cache/$CACHENAME/content
1262 * We call ->cache_show passing NULL for the item to
1263 * get a header, then pass each real item in the cache
1267 struct cache_detail *cd;
1270 static void *c_start(struct seq_file *m, loff_t *pos)
1271 __acquires(cd->hash_lock)
1274 unsigned hash, entry;
1275 struct cache_head *ch;
1276 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1279 read_lock(&cd->hash_lock);
1281 return SEQ_START_TOKEN;
1283 entry = n & ((1LL<<32) - 1);
1285 for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1288 n &= ~((1LL<<32) - 1);
1292 } while(hash < cd->hash_size &&
1293 cd->hash_table[hash]==NULL);
1294 if (hash >= cd->hash_size)
1297 return cd->hash_table[hash];
1300 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1302 struct cache_head *ch = p;
1303 int hash = (*pos >> 32);
1304 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1306 if (p == SEQ_START_TOKEN)
1308 else if (ch->next == NULL) {
1315 *pos &= ~((1LL<<32) - 1);
1316 while (hash < cd->hash_size &&
1317 cd->hash_table[hash] == NULL) {
1321 if (hash >= cd->hash_size)
1324 return cd->hash_table[hash];
1327 static void c_stop(struct seq_file *m, void *p)
1328 __releases(cd->hash_lock)
1330 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1331 read_unlock(&cd->hash_lock);
1334 static int c_show(struct seq_file *m, void *p)
1336 struct cache_head *cp = p;
1337 struct cache_detail *cd = ((struct handle*)m->private)->cd;
1339 if (p == SEQ_START_TOKEN)
1340 return cd->cache_show(m, cd, NULL);
1343 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1344 convert_to_wallclock(cp->expiry_time),
1345 atomic_read(&cp->ref.refcount), cp->flags);
1347 if (cache_check(cd, cp, NULL))
1348 /* cache_check does a cache_put on failure */
1349 seq_printf(m, "# ");
1353 return cd->cache_show(m, cd, cp);
1356 static const struct seq_operations cache_content_op = {
1363 static int content_open(struct inode *inode, struct file *file,
1364 struct cache_detail *cd)
1368 if (!cd || !try_module_get(cd->owner))
1370 han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1372 module_put(cd->owner);
1380 static int content_release(struct inode *inode, struct file *file,
1381 struct cache_detail *cd)
1383 int ret = seq_release_private(inode, file);
1384 module_put(cd->owner);
1388 static int open_flush(struct inode *inode, struct file *file,
1389 struct cache_detail *cd)
1391 if (!cd || !try_module_get(cd->owner))
1393 return nonseekable_open(inode, file);
1396 static int release_flush(struct inode *inode, struct file *file,
1397 struct cache_detail *cd)
1399 module_put(cd->owner);
1403 static ssize_t read_flush(struct file *file, char __user *buf,
1404 size_t count, loff_t *ppos,
1405 struct cache_detail *cd)
1408 unsigned long p = *ppos;
1411 sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1418 if (copy_to_user(buf, (void*)(tbuf+p), len))
1424 static ssize_t write_flush(struct file *file, const char __user *buf,
1425 size_t count, loff_t *ppos,
1426 struct cache_detail *cd)
1431 if (*ppos || count > sizeof(tbuf)-1)
1433 if (copy_from_user(tbuf, buf, count))
1436 simple_strtoul(tbuf, &ep, 0);
1437 if (*ep && *ep != '\n')
1441 cd->flush_time = get_expiry(&bp);
1442 cd->nextcheck = seconds_since_boot();
1449 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1450 size_t count, loff_t *ppos)
1452 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1454 return cache_read(filp, buf, count, ppos, cd);
1457 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1458 size_t count, loff_t *ppos)
1460 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1462 return cache_write(filp, buf, count, ppos, cd);
1465 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1467 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1469 return cache_poll(filp, wait, cd);
1472 static long cache_ioctl_procfs(struct file *filp,
1473 unsigned int cmd, unsigned long arg)
1475 struct inode *inode = filp->f_path.dentry->d_inode;
1476 struct cache_detail *cd = PDE(inode)->data;
1478 return cache_ioctl(inode, filp, cmd, arg, cd);
1481 static int cache_open_procfs(struct inode *inode, struct file *filp)
1483 struct cache_detail *cd = PDE(inode)->data;
1485 return cache_open(inode, filp, cd);
1488 static int cache_release_procfs(struct inode *inode, struct file *filp)
1490 struct cache_detail *cd = PDE(inode)->data;
1492 return cache_release(inode, filp, cd);
1495 static const struct file_operations cache_file_operations_procfs = {
1496 .owner = THIS_MODULE,
1497 .llseek = no_llseek,
1498 .read = cache_read_procfs,
1499 .write = cache_write_procfs,
1500 .poll = cache_poll_procfs,
1501 .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1502 .open = cache_open_procfs,
1503 .release = cache_release_procfs,
1506 static int content_open_procfs(struct inode *inode, struct file *filp)
1508 struct cache_detail *cd = PDE(inode)->data;
1510 return content_open(inode, filp, cd);
1513 static int content_release_procfs(struct inode *inode, struct file *filp)
1515 struct cache_detail *cd = PDE(inode)->data;
1517 return content_release(inode, filp, cd);
1520 static const struct file_operations content_file_operations_procfs = {
1521 .open = content_open_procfs,
1523 .llseek = seq_lseek,
1524 .release = content_release_procfs,
1527 static int open_flush_procfs(struct inode *inode, struct file *filp)
1529 struct cache_detail *cd = PDE(inode)->data;
1531 return open_flush(inode, filp, cd);
1534 static int release_flush_procfs(struct inode *inode, struct file *filp)
1536 struct cache_detail *cd = PDE(inode)->data;
1538 return release_flush(inode, filp, cd);
1541 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1542 size_t count, loff_t *ppos)
1544 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1546 return read_flush(filp, buf, count, ppos, cd);
1549 static ssize_t write_flush_procfs(struct file *filp,
1550 const char __user *buf,
1551 size_t count, loff_t *ppos)
1553 struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1555 return write_flush(filp, buf, count, ppos, cd);
1558 static const struct file_operations cache_flush_operations_procfs = {
1559 .open = open_flush_procfs,
1560 .read = read_flush_procfs,
1561 .write = write_flush_procfs,
1562 .release = release_flush_procfs,
1563 .llseek = no_llseek,
1566 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1568 struct sunrpc_net *sn;
1570 if (cd->u.procfs.proc_ent == NULL)
1572 if (cd->u.procfs.flush_ent)
1573 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1574 if (cd->u.procfs.channel_ent)
1575 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1576 if (cd->u.procfs.content_ent)
1577 remove_proc_entry("content", cd->u.procfs.proc_ent);
1578 cd->u.procfs.proc_ent = NULL;
1579 sn = net_generic(net, sunrpc_net_id);
1580 remove_proc_entry(cd->name, sn->proc_net_rpc);
1583 #ifdef CONFIG_PROC_FS
1584 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1586 struct proc_dir_entry *p;
1587 struct sunrpc_net *sn;
1589 sn = net_generic(net, sunrpc_net_id);
1590 cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1591 if (cd->u.procfs.proc_ent == NULL)
1593 cd->u.procfs.channel_ent = NULL;
1594 cd->u.procfs.content_ent = NULL;
1596 p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1597 cd->u.procfs.proc_ent,
1598 &cache_flush_operations_procfs, cd);
1599 cd->u.procfs.flush_ent = p;
1603 if (cd->cache_upcall || cd->cache_parse) {
1604 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1605 cd->u.procfs.proc_ent,
1606 &cache_file_operations_procfs, cd);
1607 cd->u.procfs.channel_ent = p;
1611 if (cd->cache_show) {
1612 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1613 cd->u.procfs.proc_ent,
1614 &content_file_operations_procfs, cd);
1615 cd->u.procfs.content_ent = p;
1621 remove_cache_proc_entries(cd, net);
1624 #else /* CONFIG_PROC_FS */
1625 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1631 void __init cache_initialize(void)
1633 INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1636 int cache_register_net(struct cache_detail *cd, struct net *net)
1640 sunrpc_init_cache_detail(cd);
1641 ret = create_cache_proc_entries(cd, net);
1643 sunrpc_destroy_cache_detail(cd);
1646 EXPORT_SYMBOL_GPL(cache_register_net);
1648 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1650 remove_cache_proc_entries(cd, net);
1651 sunrpc_destroy_cache_detail(cd);
1653 EXPORT_SYMBOL_GPL(cache_unregister_net);
1655 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1657 struct cache_detail *cd;
1659 cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1661 return ERR_PTR(-ENOMEM);
1663 cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1665 if (cd->hash_table == NULL) {
1667 return ERR_PTR(-ENOMEM);
1672 EXPORT_SYMBOL_GPL(cache_create_net);
1674 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1676 kfree(cd->hash_table);
1679 EXPORT_SYMBOL_GPL(cache_destroy_net);
1681 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1682 size_t count, loff_t *ppos)
1684 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1686 return cache_read(filp, buf, count, ppos, cd);
1689 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1690 size_t count, loff_t *ppos)
1692 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1694 return cache_write(filp, buf, count, ppos, cd);
1697 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1699 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1701 return cache_poll(filp, wait, cd);
1704 static long cache_ioctl_pipefs(struct file *filp,
1705 unsigned int cmd, unsigned long arg)
1707 struct inode *inode = filp->f_dentry->d_inode;
1708 struct cache_detail *cd = RPC_I(inode)->private;
1710 return cache_ioctl(inode, filp, cmd, arg, cd);
1713 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1715 struct cache_detail *cd = RPC_I(inode)->private;
1717 return cache_open(inode, filp, cd);
1720 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1722 struct cache_detail *cd = RPC_I(inode)->private;
1724 return cache_release(inode, filp, cd);
1727 const struct file_operations cache_file_operations_pipefs = {
1728 .owner = THIS_MODULE,
1729 .llseek = no_llseek,
1730 .read = cache_read_pipefs,
1731 .write = cache_write_pipefs,
1732 .poll = cache_poll_pipefs,
1733 .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1734 .open = cache_open_pipefs,
1735 .release = cache_release_pipefs,
1738 static int content_open_pipefs(struct inode *inode, struct file *filp)
1740 struct cache_detail *cd = RPC_I(inode)->private;
1742 return content_open(inode, filp, cd);
1745 static int content_release_pipefs(struct inode *inode, struct file *filp)
1747 struct cache_detail *cd = RPC_I(inode)->private;
1749 return content_release(inode, filp, cd);
1752 const struct file_operations content_file_operations_pipefs = {
1753 .open = content_open_pipefs,
1755 .llseek = seq_lseek,
1756 .release = content_release_pipefs,
1759 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1761 struct cache_detail *cd = RPC_I(inode)->private;
1763 return open_flush(inode, filp, cd);
1766 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1768 struct cache_detail *cd = RPC_I(inode)->private;
1770 return release_flush(inode, filp, cd);
1773 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1774 size_t count, loff_t *ppos)
1776 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1778 return read_flush(filp, buf, count, ppos, cd);
1781 static ssize_t write_flush_pipefs(struct file *filp,
1782 const char __user *buf,
1783 size_t count, loff_t *ppos)
1785 struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1787 return write_flush(filp, buf, count, ppos, cd);
1790 const struct file_operations cache_flush_operations_pipefs = {
1791 .open = open_flush_pipefs,
1792 .read = read_flush_pipefs,
1793 .write = write_flush_pipefs,
1794 .release = release_flush_pipefs,
1795 .llseek = no_llseek,
1798 int sunrpc_cache_register_pipefs(struct dentry *parent,
1799 const char *name, umode_t umode,
1800 struct cache_detail *cd)
1807 q.len = strlen(name);
1808 q.hash = full_name_hash(q.name, q.len);
1809 dir = rpc_create_cache_dir(parent, &q, umode, cd);
1811 cd->u.pipefs.dir = dir;
1816 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1818 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1820 rpc_remove_cache_dir(cd->u.pipefs.dir);
1821 cd->u.pipefs.dir = NULL;
1823 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);