nfsd41: sessionid hashing
authorMarc Eshel <eshel@almaden.ibm.com>
Fri, 3 Apr 2009 05:27:52 +0000 (08:27 +0300)
committerJ. Bruce Fields <bfields@citi.umich.edu>
Sat, 4 Apr 2009 00:41:14 +0000 (17:41 -0700)
Simple sessionid hashing using its monotonically increasing sequence number.

Locking considerations:
sessionid_hashtbl access is controlled by the sessionid_lock spin lock.
It must be taken for insert, delete, and lookup.
nfsd4_sequence looks up the session id and if the session is found,
it calls nfsd4_get_session (still under the sessionid_lock).
nfsd4_destroy_session calls nfsd4_put_session after unhashing
it, so when the session's kref reaches zero it's going to get freed.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[we don't use a prime for sessionid hash table size]
[use sessionid_lock spin lock]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
fs/nfsd/nfs4state.c
include/linux/nfsd/state.h

index f23385b..fc20e1f 100644 (file)
@@ -382,11 +382,62 @@ static void release_openowner(struct nfs4_stateowner *sop)
        nfs4_put_stateowner(sop);
 }
 
+static DEFINE_SPINLOCK(sessionid_lock);
+#define SESSION_HASH_SIZE      512
+static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
+
+static inline int
+hash_sessionid(struct nfs4_sessionid *sessionid)
+{
+       struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
+
+       return sid->sequence % SESSION_HASH_SIZE;
+}
+
+static inline void
+dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
+{
+       u32 *ptr = (u32 *)(&sessionid->data[0]);
+       dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
+}
+
+/* caller must hold sessionid_lock */
+static struct nfsd4_session *
+find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
+{
+       struct nfsd4_session *elem;
+       int idx;
+
+       dump_sessionid(__func__, sessionid);
+       idx = hash_sessionid(sessionid);
+       dprintk("%s: idx is %d\n", __func__, idx);
+       /* Search in the appropriate list */
+       list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
+               dump_sessionid("list traversal", &elem->se_sessionid);
+               if (!memcmp(elem->se_sessionid.data, sessionid->data,
+                           NFS4_MAX_SESSIONID_LEN)) {
+                       return elem;
+               }
+       }
+
+       dprintk("%s: session not found\n", __func__);
+       return NULL;
+}
+
+/* caller must hold sessionid_lock */
 static void
-release_session(struct nfsd4_session *ses)
+unhash_session(struct nfsd4_session *ses)
 {
        list_del(&ses->se_hash);
        list_del(&ses->se_perclnt);
+}
+
+static void
+release_session(struct nfsd4_session *ses)
+{
+       spin_lock(&sessionid_lock);
+       unhash_session(ses);
+       spin_unlock(&sessionid_lock);
        nfsd4_put_session(ses);
 }
 
@@ -3205,6 +3256,8 @@ nfs4_state_init(void)
                INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
                INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
        }
+       for (i = 0; i < SESSION_HASH_SIZE; i++)
+               INIT_LIST_HEAD(&sessionid_hashtbl[i]);
        for (i = 0; i < FILE_HASH_SIZE; i++) {
                INIT_LIST_HEAD(&file_hashtbl[i]);
        }
index 7faefc7..5b3a666 100644 (file)
@@ -132,6 +132,13 @@ nfsd4_get_session(struct nfsd4_session *ses)
        kref_get(&ses->se_ref);
 }
 
+/* formatted contents of nfs4_sessionid */
+struct nfsd4_sessionid {
+       clientid_t      clientid;
+       u32             sequence;
+       u32             reserved;
+};
+
 #define HEXDIR_LEN     33 /* hex version of 16 byte md5 of cl_name plus '\0' */
 
 /*