lockd: always preallocate block in nlmsvc_lock()
authorMarc Eshel <eshel@almaden.ibm.com>
Wed, 6 Dec 2006 04:48:10 +0000 (23:48 -0500)
committerJ. Bruce Fields <bfields@citi.umich.edu>
Mon, 7 May 2007 00:38:50 +0000 (20:38 -0400)
Normally we could skip ever having to allocate a block in the case where
the client asks for a non-blocking lock, or asks for a blocking lock that
succeeds immediately.

However we're going to want to always look up a block first in order to
check whether we're revisiting a deferred lock call, and to be prepared to
handle the case where the filesystem returns -EINPROGRESS--in that case we
want to make sure the lock we've given the filesystem is the one embedded
in the block that we'll use to track the deferred request.

Signed-off-by: Marc Eshel <eshel@almaden.ibm.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
fs/lockd/svclock.c

index b7a8174fd1dc1b50d669606b5256e645f1572356..0d7398e3804f663e9ca2eba8bca8e90ba28a150f 100644 (file)
@@ -365,7 +365,7 @@ __be32
 nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
                        struct nlm_lock *lock, int wait, struct nlm_cookie *cookie)
 {
-       struct nlm_block        *block, *newblock = NULL;
+       struct nlm_block        *block = NULL;
        int                     error;
        __be32                  ret;
 
@@ -378,17 +378,20 @@ nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
                                wait);
 
 
-       lock->fl.fl_flags &= ~FL_SLEEP;
-again:
        /* Lock file against concurrent access */
        mutex_lock(&file->f_mutex);
-       /* Get existing block (in case client is busy-waiting) */
+       /* Get existing block (in case client is busy-waiting)
+        * or create new block
+        */
        block = nlmsvc_lookup_block(file, lock);
        if (block == NULL) {
-               if (newblock != NULL)
-                       lock = &newblock->b_call->a_args.lock;
-       } else
+               block = nlmsvc_create_block(rqstp, file, lock, cookie);
+               ret = nlm_lck_denied_nolocks;
+               if (block == NULL)
+                       goto out;
                lock = &block->b_call->a_args.lock;
+       } else
+               lock->fl.fl_flags &= ~FL_SLEEP;
 
        error = posix_lock_file(file->f_file, &lock->fl, NULL);
        lock->fl.fl_flags &= ~FL_SLEEP;
@@ -414,26 +417,11 @@ again:
                goto out;
 
        ret = nlm_lck_blocked;
-       if (block != NULL)
-               goto out;
-
-       /* If we don't have a block, create and initialize it. Then
-        * retry because we may have slept in kmalloc. */
-       /* We have to release f_mutex as nlmsvc_create_block may try to
-        * to claim it while doing host garbage collection */
-       if (newblock == NULL) {
-               mutex_unlock(&file->f_mutex);
-               dprintk("lockd: blocking on this lock (allocating).\n");
-               if (!(newblock = nlmsvc_create_block(rqstp, file, lock, cookie)))
-                       return nlm_lck_denied_nolocks;
-               goto again;
-       }
 
        /* Append to list of blocked */
-       nlmsvc_insert_block(newblock, NLM_NEVER);
+       nlmsvc_insert_block(block, NLM_NEVER);
 out:
        mutex_unlock(&file->f_mutex);
-       nlmsvc_release_block(newblock);
        nlmsvc_release_block(block);
        dprintk("lockd: nlmsvc_lock returned %u\n", ret);
        return ret;