fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / net / sunrpc / cache.c
index d508a51..1427449 100644 (file)
@@ -26,6 +26,7 @@
 #include <linux/proc_fs.h>
 #include <linux/net.h>
 #include <linux/workqueue.h>
+#include <linux/mutex.h>
 #include <asm/ioctls.h>
 #include <linux/sunrpc/types.h>
 #include <linux/sunrpc/cache.h>
 
 #define         RPCDBG_FACILITY RPCDBG_CACHE
 
-void cache_init(struct cache_head *h)
+static int cache_defer_req(struct cache_req *req, struct cache_head *item);
+static void cache_revisit_request(struct cache_head *item);
+
+static void cache_init(struct cache_head *h)
 {
        time_t now = get_seconds();
        h->next = NULL;
        h->flags = 0;
-       atomic_set(&h->refcnt, 1);
+       kref_init(&h->ref);
        h->expiry_time = now + CACHE_NEW_EXPIRY;
        h->last_refresh = now;
 }
 
+struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
+                                      struct cache_head *key, int hash)
+{
+       struct cache_head **head,  **hp;
+       struct cache_head *new = NULL;
+
+       head = &detail->hash_table[hash];
+
+       read_lock(&detail->hash_lock);
+
+       for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
+               struct cache_head *tmp = *hp;
+               if (detail->match(tmp, key)) {
+                       cache_get(tmp);
+                       read_unlock(&detail->hash_lock);
+                       return tmp;
+               }
+       }
+       read_unlock(&detail->hash_lock);
+       /* Didn't find anything, insert an empty entry */
+
+       new = detail->alloc();
+       if (!new)
+               return NULL;
+       /* must fully initialise 'new', else
+        * we might get lose if we need to
+        * cache_put it soon.
+        */
+       cache_init(new);
+       detail->init(new, key);
+
+       write_lock(&detail->hash_lock);
+
+       /* check if entry appeared while we slept */
+       for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
+               struct cache_head *tmp = *hp;
+               if (detail->match(tmp, key)) {
+                       cache_get(tmp);
+                       write_unlock(&detail->hash_lock);
+                       cache_put(new, detail);
+                       return tmp;
+               }
+       }
+       new->next = *head;
+       *head = new;
+       detail->entries++;
+       cache_get(new);
+       write_unlock(&detail->hash_lock);
+
+       return new;
+}
+EXPORT_SYMBOL(sunrpc_cache_lookup);
+
+
+static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
+
+static int cache_fresh_locked(struct cache_head *head, time_t expiry)
+{
+       head->expiry_time = expiry;
+       head->last_refresh = get_seconds();
+       return !test_and_set_bit(CACHE_VALID, &head->flags);
+}
+
+static void cache_fresh_unlocked(struct cache_head *head,
+                       struct cache_detail *detail, int new)
+{
+       if (new)
+               cache_revisit_request(head);
+       if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
+               cache_revisit_request(head);
+               queue_loose(detail, head);
+       }
+}
+
+struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
+                                      struct cache_head *new, struct cache_head *old, int hash)
+{
+       /* The 'old' entry is to be replaced by 'new'.
+        * If 'old' is not VALID, we update it directly,
+        * otherwise we need to replace it
+        */
+       struct cache_head **head;
+       struct cache_head *tmp;
+       int is_new;
+
+       if (!test_bit(CACHE_VALID, &old->flags)) {
+               write_lock(&detail->hash_lock);
+               if (!test_bit(CACHE_VALID, &old->flags)) {
+                       if (test_bit(CACHE_NEGATIVE, &new->flags))
+                               set_bit(CACHE_NEGATIVE, &old->flags);
+                       else
+                               detail->update(old, new);
+                       is_new = cache_fresh_locked(old, new->expiry_time);
+                       write_unlock(&detail->hash_lock);
+                       cache_fresh_unlocked(old, detail, is_new);
+                       return old;
+               }
+               write_unlock(&detail->hash_lock);
+       }
+       /* We need to insert a new entry */
+       tmp = detail->alloc();
+       if (!tmp) {
+               cache_put(old, detail);
+               return NULL;
+       }
+       cache_init(tmp);
+       detail->init(tmp, old);
+       head = &detail->hash_table[hash];
+
+       write_lock(&detail->hash_lock);
+       if (test_bit(CACHE_NEGATIVE, &new->flags))
+               set_bit(CACHE_NEGATIVE, &tmp->flags);
+       else
+               detail->update(tmp, new);
+       tmp->next = *head;
+       *head = tmp;
+       detail->entries++;
+       cache_get(tmp);
+       is_new = cache_fresh_locked(tmp, new->expiry_time);
+       cache_fresh_locked(old, 0);
+       write_unlock(&detail->hash_lock);
+       cache_fresh_unlocked(tmp, detail, is_new);
+       cache_fresh_unlocked(old, detail, 0);
+       cache_put(old, detail);
+       return tmp;
+}
+EXPORT_SYMBOL(sunrpc_cache_update);
 
 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
 /*
@@ -54,6 +185,7 @@ static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
  *
  * Returns 0 if the cache_head can be used, or cache_puts it and returns
  * -EAGAIN if upcall is pending,
+ * -ETIMEDOUT if upcall failed and should be retried,
  * -ENOENT if cache entry was negative
  */
 int cache_check(struct cache_detail *detail,
@@ -90,7 +222,8 @@ int cache_check(struct cache_detail *detail,
                                clear_bit(CACHE_PENDING, &h->flags);
                                if (rv == -EAGAIN) {
                                        set_bit(CACHE_NEGATIVE, &h->flags);
-                                       cache_fresh(detail, h, get_seconds()+CACHE_NEW_EXPIRY);
+                                       cache_fresh_unlocked(h, detail,
+                                            cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY));
                                        rv = -ENOENT;
                                }
                                break;
@@ -104,27 +237,14 @@ int cache_check(struct cache_detail *detail,
        }
 
        if (rv == -EAGAIN)
-               cache_defer_req(rqstp, h);
+               if (cache_defer_req(rqstp, h) != 0)
+                       rv = -ETIMEDOUT;
 
-       if (rv && h)
-               detail->cache_put(h, detail);
+       if (rv)
+               cache_put(h, detail);
        return rv;
 }
 
-static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
-
-void cache_fresh(struct cache_detail *detail,
-                struct cache_head *head, time_t expiry)
-{
-
-       head->expiry_time = expiry;
-       head->last_refresh = get_seconds();
-       if (!test_and_set_bit(CACHE_VALID, &head->flags))
-               cache_revisit_request(head);
-       if (test_and_clear_bit(CACHE_PENDING, &head->flags))
-               queue_loose(detail, head);
-}
-
 /*
  * caches need to be periodically cleaned.
  * For this we maintain a list of cache_detail and
@@ -158,7 +278,7 @@ void cache_fresh(struct cache_detail *detail,
  */
 
 static LIST_HEAD(cache_list);
-static spinlock_t cache_list_lock = SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(cache_list_lock);
 static struct cache_detail *current_detail;
 static int current_index;
 
@@ -166,15 +286,15 @@ static struct file_operations cache_file_operations;
 static struct file_operations content_file_operations;
 static struct file_operations cache_flush_operations;
 
-static void do_cache_clean(void *data);
-static DECLARE_WORK(cache_cleaner, do_cache_clean, NULL);
+static void do_cache_clean(struct work_struct *work);
+static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
 
 void cache_register(struct cache_detail *cd)
 {
        cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
        if (cd->proc_ent) {
                struct proc_dir_entry *p;
-               cd->proc_ent->owner = THIS_MODULE;
+               cd->proc_ent->owner = cd->owner;
                cd->channel_ent = cd->content_ent = NULL;
                
                p = create_proc_entry("flush", S_IFREG|S_IRUSR|S_IWUSR,
@@ -182,7 +302,7 @@ void cache_register(struct cache_detail *cd)
                cd->flush_ent =  p;
                if (p) {
                        p->proc_fops = &cache_flush_operations;
-                       p->owner = THIS_MODULE;
+                       p->owner = cd->owner;
                        p->data = cd;
                }
  
@@ -192,7 +312,7 @@ void cache_register(struct cache_detail *cd)
                        cd->channel_ent = p;
                        if (p) {
                                p->proc_fops = &cache_file_operations;
-                               p->owner = THIS_MODULE;
+                               p->owner = cd->owner;
                                p->data = cd;
                        }
                }
@@ -202,7 +322,7 @@ void cache_register(struct cache_detail *cd)
                        cd->content_ent = p;
                        if (p) {
                                p->proc_fops = &content_file_operations;
-                               p->owner = THIS_MODULE;
+                               p->owner = cd->owner;
                                p->data = cd;
                        }
                }
@@ -219,7 +339,7 @@ void cache_register(struct cache_detail *cd)
        spin_unlock(&cache_list_lock);
 
        /* start the cleaning process */
-       schedule_work(&cache_cleaner);
+       schedule_delayed_work(&cache_cleaner, 0);
 }
 
 int cache_unregister(struct cache_detail *cd)
@@ -256,39 +376,13 @@ int cache_unregister(struct cache_detail *cd)
        return 0;
 }
 
-struct cache_detail *cache_find(char *name)
-{
-       struct list_head *l;
-
-       spin_lock(&cache_list_lock);
-       list_for_each(l, &cache_list) {
-               struct cache_detail *cd = list_entry(l, struct cache_detail, others);
-               
-               if (strcmp(cd->name, name)==0) {
-                       atomic_inc(&cd->inuse);
-                       spin_unlock(&cache_list_lock);
-                       return cd;
-               }
-       }
-       spin_unlock(&cache_list_lock);
-       return NULL;
-}
-
-/* cache_drop must be called on any cache returned by
- * cache_find, after it has been used
- */
-void cache_drop(struct cache_detail *detail)
-{
-       atomic_dec(&detail->inuse);
-}
-
 /* clean cache tries to find something to clean
  * and cleans it.
  * It returns 1 if it cleaned something,
  *            0 if it didn't find anything this time
  *           -1 if it fell off the end of the list.
  */
-int cache_clean(void)
+static int cache_clean(void)
 {
        int rv = 0;
        struct list_head *next;
@@ -344,12 +438,10 @@ int cache_clean(void)
                        if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
                                queue_loose(current_detail, ch);
 
-                       if (!atomic_read(&ch->refcnt))
+                       if (atomic_read(&ch->ref.refcount) == 1)
                                break;
                }
                if (ch) {
-                       cache_get(ch);
-                       clear_bit(CACHE_HASHED, &ch->flags);
                        *cp = ch->next;
                        ch->next = NULL;
                        current_detail->entries--;
@@ -361,7 +453,7 @@ int cache_clean(void)
                        current_index ++;
                spin_unlock(&cache_list_lock);
                if (ch)
-                       d->cache_put(ch, d);
+                       cache_put(ch, d);
        } else
                spin_unlock(&cache_list_lock);
 
@@ -371,7 +463,7 @@ int cache_clean(void)
 /*
  * We want to regularly clean the cache, so we need to schedule some work ...
  */
-static void do_cache_clean(void *data)
+static void do_cache_clean(struct work_struct *work)
 {
        int delay = 5;
        if (cache_clean() == -1)
@@ -428,19 +520,26 @@ void cache_purge(struct cache_detail *detail)
 
 #define        DFR_MAX 300     /* ??? */
 
-spinlock_t cache_defer_lock = SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(cache_defer_lock);
 static LIST_HEAD(cache_defer_list);
 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 static int cache_defer_cnt;
 
-void cache_defer_req(struct cache_req *req, struct cache_head *item)
+static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 {
        struct cache_deferred_req *dreq;
        int hash = DFR_HASH(item);
 
+       if (cache_defer_cnt >= DFR_MAX) {
+               /* too much in the cache, randomly drop this one,
+                * or continue and drop the oldest below
+                */
+               if (net_random()&1)
+                       return -ETIMEDOUT;
+       }
        dreq = req->defer(req);
        if (dreq == NULL)
-               return;
+               return -ETIMEDOUT;
 
        dreq->item = item;
        dreq->recv_time = get_seconds();
@@ -456,17 +555,8 @@ void cache_defer_req(struct cache_req *req, struct cache_head *item)
        /* it is in, now maybe clean up */
        dreq = NULL;
        if (++cache_defer_cnt > DFR_MAX) {
-               /* too much in the cache, randomly drop
-                * first or last
-                */
-               if (net_random()&1) 
-                       dreq = list_entry(cache_defer_list.next,
-                                         struct cache_deferred_req,
-                                         recent);
-               else
-                       dreq = list_entry(cache_defer_list.prev,
-                                         struct cache_deferred_req,
-                                         recent);
+               dreq = list_entry(cache_defer_list.prev,
+                                 struct cache_deferred_req, recent);
                list_del(&dreq->recent);
                list_del(&dreq->hash);
                cache_defer_cnt--;
@@ -477,13 +567,14 @@ void cache_defer_req(struct cache_req *req, struct cache_head *item)
                /* there was one too many */
                dreq->revisit(dreq, 1);
        }
-       if (test_bit(CACHE_VALID, &item->flags)) {
+       if (!test_bit(CACHE_PENDING, &item->flags)) {
                /* must have just been validated... */
                cache_revisit_request(item);
        }
+       return 0;
 }
 
-void cache_revisit_request(struct cache_head *item)
+static void cache_revisit_request(struct cache_head *item)
 {
        struct cache_deferred_req *dreq;
        struct list_head pending;
@@ -556,8 +647,8 @@ void cache_clean_deferred(void *owner)
  *
  */
 
-static spinlock_t queue_lock = SPIN_LOCK_UNLOCKED;
-static DECLARE_MUTEX(queue_io_sem);
+static DEFINE_SPINLOCK(queue_lock);
+static DEFINE_MUTEX(queue_io_mutex);
 
 struct cache_queue {
        struct list_head        list;
@@ -580,13 +671,13 @@ cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 {
        struct cache_reader *rp = filp->private_data;
        struct cache_request *rq;
-       struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
+       struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
        int err;
 
        if (count == 0)
                return 0;
 
-       down(&queue_io_sem); /* protect against multiple concurrent
+       mutex_lock(&queue_io_mutex); /* protect against multiple concurrent
                              * readers on this file */
  again:
        spin_lock(&queue_lock);
@@ -599,13 +690,12 @@ cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
        }
        if (rp->q.list.next == &cd->queue) {
                spin_unlock(&queue_lock);
-               up(&queue_io_sem);
-               if (rp->offset)
-                       BUG();
+               mutex_unlock(&queue_io_mutex);
+               BUG_ON(rp->offset);
                return 0;
        }
        rq = container_of(rp->q.list.next, struct cache_request, q.list);
-       if (rq->q.reader) BUG();
+       BUG_ON(rq->q.reader);
        if (rp->offset == 0)
                rq->readers++;
        spin_unlock(&queue_lock);
@@ -639,7 +729,7 @@ cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
                    !test_bit(CACHE_PENDING, &rq->item->flags)) {
                        list_del(&rq->q.list);
                        spin_unlock(&queue_lock);
-                       cd->cache_put(rq->item, cd);
+                       cache_put(rq->item, cd);
                        kfree(rq->buf);
                        kfree(rq);
                } else
@@ -647,28 +737,28 @@ cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
        }
        if (err == -EAGAIN)
                goto again;
-       up(&queue_io_sem);
+       mutex_unlock(&queue_io_mutex);
        return err ? err :  count;
 }
 
-static char write_buf[8192]; /* protected by queue_io_sem */
+static char write_buf[8192]; /* protected by queue_io_mutex */
 
 static ssize_t
 cache_write(struct file *filp, const char __user *buf, size_t count,
            loff_t *ppos)
 {
        int err;
-       struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
+       struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 
        if (count == 0)
                return 0;
        if (count >= sizeof(write_buf))
                return -EINVAL;
 
-       down(&queue_io_sem);
+       mutex_lock(&queue_io_mutex);
 
        if (copy_from_user(write_buf, buf, count)) {
-               up(&queue_io_sem);
+               mutex_unlock(&queue_io_mutex);
                return -EFAULT;
        }
        write_buf[count] = '\0';
@@ -677,7 +767,7 @@ cache_write(struct file *filp, const char __user *buf, size_t count,
        else
                err = -EINVAL;
 
-       up(&queue_io_sem);
+       mutex_unlock(&queue_io_mutex);
        return err ? err : count;
 }
 
@@ -689,7 +779,7 @@ cache_poll(struct file *filp, poll_table *wait)
        unsigned int mask;
        struct cache_reader *rp = filp->private_data;
        struct cache_queue *cq;
-       struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
+       struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 
        poll_wait(filp, &queue_wait, wait);
 
@@ -819,10 +909,10 @@ static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
                        if (cr->item != ch)
                                continue;
                        if (cr->readers != 0)
-                               break;
+                               continue;
                        list_del(&cr->q.list);
                        spin_unlock(&queue_lock);
-                       detail->cache_put(cr->item, detail);
+                       cache_put(cr->item, detail);
                        kfree(cr->buf);
                        kfree(cr);
                        return;
@@ -902,7 +992,7 @@ void qword_addhex(char **bpp, int *lp, char *buf, int blen)
        *lp = len;
 }
 
-void warn_no_listener(struct cache_detail *detail)
+static void warn_no_listener(struct cache_detail *detail)
 {
        if (detail->last_warn != detail->last_close) {
                detail->last_warn = detail->last_close;
@@ -1107,8 +1197,8 @@ static int c_show(struct seq_file *m, void *p)
                return cd->cache_show(m, cd, NULL);
 
        ifdebug(CACHE)
-               seq_printf(m, "# expiry=%ld refcnt=%d\n",
-                          cp->expiry_time, atomic_read(&cp->refcnt));
+               seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
+                          cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
        cache_get(cp);
        if (cache_check(cd, cp, NULL))
                /* cache_check does a cache_put on failure */
@@ -1119,7 +1209,7 @@ static int c_show(struct seq_file *m, void *p)
        return cd->cache_show(m, cd, cp);
 }
 
-struct seq_operations cache_content_op = {
+static struct seq_operations cache_content_op = {
        .start  = c_start,
        .next   = c_next,
        .stop   = c_stop,
@@ -1165,7 +1255,7 @@ static struct file_operations content_file_operations = {
 static ssize_t read_flush(struct file *file, char __user *buf,
                            size_t count, loff_t *ppos)
 {
-       struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
+       struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
        char tbuf[20];
        unsigned long p = *ppos;
        int len;
@@ -1186,7 +1276,7 @@ static ssize_t read_flush(struct file *file, char __user *buf,
 static ssize_t write_flush(struct file * file, const char __user * buf,
                             size_t count, loff_t *ppos)
 {
-       struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
+       struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
        char tbuf[20];
        char *ep;
        long flushtime;