[PATCH] knfsd: nfsd4: allow multiple lockowners
authorNeilBrown <neilb@cse.unsw.edu.au>
Fri, 24 Jun 2005 05:04:20 +0000 (22:04 -0700)
committerLinus Torvalds <torvalds@ppc970.osdl.org>
Fri, 24 Jun 2005 07:06:35 +0000 (00:06 -0700)
>From the language of rfc3530 section 8.1.3 (e.g., the suggestion that a
"process id" might be a reasonable lockowner value) it's conceivable that a
client might want to use the same lockowner string on multiple files, so we may
as well allow that.  We expect each use of open_to_lockowner to create a
distinct seqid stream, though.

For now we're also allowing multiple uses of open_to_lockowner with the same
open, though it seems unlikely clients would actually do that.

Also add a comment reminding myself of some very non-scalable data structures.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
fs/nfsd/nfs4state.c

index 22e76e3f06a57fe40d1d06c9216320caf3431074..26d00465c28aa630e120d74c7798df31efef1925 100644 (file)
@@ -2583,22 +2583,6 @@ nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
                deny->ld_type = NFS4_WRITE_LT;
 }
 
-static struct nfs4_stateowner *
-find_lockstateowner(struct xdr_netobj *owner, clientid_t *clid)
-{
-       struct nfs4_stateowner *local = NULL;
-       int i;
-
-       for (i = 0; i < LOCK_HASH_SIZE; i++) {
-               list_for_each_entry(local, &lock_ownerid_hashtbl[i], so_idhash) {
-                       if (!cmp_owner_str(local, owner, clid))
-                               continue;
-                       return local;
-               }
-       }
-       return NULL;
-}
-
 static struct nfs4_stateowner *
 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
                struct xdr_netobj *owner)
@@ -2697,7 +2681,7 @@ check_lock_length(u64 offset, u64 length)
 int
 nfsd4_lock(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_lock *lock)
 {
-       struct nfs4_stateowner *lock_sop = NULL, *open_sop = NULL;
+       struct nfs4_stateowner *open_sop = NULL;
        struct nfs4_stateid *lock_stp;
        struct file *filp;
        struct file_lock file_lock;
@@ -2756,16 +2740,9 @@ nfsd4_lock(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_lock
                strhashval = lock_ownerstr_hashval(fp->fi_inode, 
                                open_sop->so_client->cl_clientid.cl_id, 
                                &lock->v.new.owner);
-               /* 
-                * If we already have this lock owner, the client is in 
-                * error (or our bookeeping is wrong!) 
-                * for asking for a 'new lock'.
-                */
-               status = nfserr_bad_stateid;
-               lock_sop = find_lockstateowner(&lock->v.new.owner,
-                                               &lock->v.new.clientid);
-               if (lock_sop)
-                       goto out;
+               /* XXX: Do we need to check for duplicate stateowners on
+                * the same file, or should they just be allowed (and
+                * create new stateids)? */
                status = nfserr_resource;
                if (!(lock->lk_stateowner = alloc_init_lock_stateowner(strhashval, open_sop->so_client, open_stp, lock)))
                        goto out;
@@ -3056,8 +3033,11 @@ int
 nfsd4_release_lockowner(struct svc_rqst *rqstp, struct nfsd4_release_lockowner *rlockowner)
 {
        clientid_t *clid = &rlockowner->rl_clientid;
-       struct nfs4_stateowner *local = NULL;
+       struct nfs4_stateowner *sop;
+       struct nfs4_stateid *stp;
        struct xdr_netobj *owner = &rlockowner->rl_owner;
+       struct list_head matches;
+       int i;
        int status;
 
        dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
@@ -3073,22 +3053,32 @@ nfsd4_release_lockowner(struct svc_rqst *rqstp, struct nfsd4_release_lockowner *
 
        nfs4_lock_state();
 
-       status = nfs_ok;
-       local = find_lockstateowner(owner, clid);
-       if (local) {
-               struct nfs4_stateid *stp;
-
-               /* check for any locks held by any stateid
-                * associated with the (lock) stateowner */
-               status = nfserr_locks_held;
-               list_for_each_entry(stp, &local->so_stateids,
-                               st_perstateowner) {
-                       if (check_for_locks(stp->st_vfs_file, local))
-                               goto out;
+       status = nfserr_locks_held;
+       /* XXX: we're doing a linear search through all the lockowners.
+        * Yipes!  For now we'll just hope clients aren't really using
+        * release_lockowner much, but eventually we have to fix these
+        * data structures. */
+       INIT_LIST_HEAD(&matches);
+       for (i = 0; i < LOCK_HASH_SIZE; i++) {
+               list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
+                       if (!cmp_owner_str(sop, owner, clid))
+                               continue;
+                       list_for_each_entry(stp, &sop->so_stateids,
+                                       st_perstateowner) {
+                               if (check_for_locks(stp->st_vfs_file, sop))
+                                       goto out;
+                               /* Note: so_perclient unused for lockowners,
+                                * so it's OK to fool with here. */
+                               list_add(&sop->so_perclient, &matches);
+                       }
                }
-               /* no locks held by (lock) stateowner */
-               status = nfs_ok;
-               release_stateowner(local);
+       }
+       /* Clients probably won't expect us to return with some (but not all)
+        * of the lockowner state released; so don't release any until all
+        * have been checked. */
+       status = nfs_ok;
+       list_for_each_entry(sop, &matches, so_perclient) {
+               release_stateowner(sop);
        }
 out:
        nfs4_unlock_state();