Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux...
[safe/jmp/linux-2.6] / fs / lockd / svclock.c
index 49f9597..84055d3 100644 (file)
@@ -20,8 +20,8 @@
  * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
  */
 
-#include <linux/config.h>
 #include <linux/types.h>
+#include <linux/slab.h>
 #include <linux/errno.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
@@ -30,6 +30,7 @@
 #include <linux/sunrpc/svc.h>
 #include <linux/lockd/nlm.h>
 #include <linux/lockd/lockd.h>
+#include <linux/kthread.h>
 
 #define NLMDBG_FACILITY                NLMDBG_SVCLOCK
 
 #define nlm_deadlock   nlm_lck_denied
 #endif
 
+static void nlmsvc_release_block(struct nlm_block *block);
 static void    nlmsvc_insert_block(struct nlm_block *block, unsigned long);
-static int     nlmsvc_remove_block(struct nlm_block *block);
-static void    nlmsvc_grant_callback(struct rpc_task *task);
+static void    nlmsvc_remove_block(struct nlm_block *block);
+
+static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
+static void nlmsvc_freegrantargs(struct nlm_rqst *call);
+static const struct rpc_call_ops nlmsvc_grant_ops;
 
 /*
  * The list of blocked locks to retry
  */
-static struct nlm_block *      nlm_blocked;
+static LIST_HEAD(nlm_blocked);
 
 /*
  * Insert a blocked lock into the global list
@@ -54,74 +59,68 @@ static struct nlm_block *   nlm_blocked;
 static void
 nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
 {
-       struct nlm_block **bp, *b;
+       struct nlm_block *b;
+       struct list_head *pos;
 
        dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
-       if (block->b_queued)
-               nlmsvc_remove_block(block);
-       bp = &nlm_blocked;
+       if (list_empty(&block->b_list)) {
+               kref_get(&block->b_count);
+       } else {
+               list_del_init(&block->b_list);
+       }
+
+       pos = &nlm_blocked;
        if (when != NLM_NEVER) {
                if ((when += jiffies) == NLM_NEVER)
                        when ++;
-               while ((b = *bp) && time_before_eq(b->b_when,when) && b->b_when != NLM_NEVER)
-                       bp = &b->b_next;
-       } else
-               while ((b = *bp) != 0)
-                       bp = &b->b_next;
+               list_for_each(pos, &nlm_blocked) {
+                       b = list_entry(pos, struct nlm_block, b_list);
+                       if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
+                               break;
+               }
+               /* On normal exit from the loop, pos == &nlm_blocked,
+                * so we will be adding to the end of the list - good
+                */
+       }
 
-       block->b_queued = 1;
+       list_add_tail(&block->b_list, pos);
        block->b_when = when;
-       block->b_next = b;
-       *bp = block;
 }
 
 /*
  * Remove a block from the global list
  */
-static int
+static inline void
 nlmsvc_remove_block(struct nlm_block *block)
 {
-       struct nlm_block **bp, *b;
-
-       if (!block->b_queued)
-               return 1;
-       for (bp = &nlm_blocked; (b = *bp) != 0; bp = &b->b_next) {
-               if (b == block) {
-                       *bp = block->b_next;
-                       block->b_queued = 0;
-                       return 1;
-               }
+       if (!list_empty(&block->b_list)) {
+               list_del_init(&block->b_list);
+               nlmsvc_release_block(block);
        }
-
-       return 0;
 }
 
 /*
- * Find a block for a given lock and optionally remove it from
- * the list.
+ * Find a block for a given lock
  */
 static struct nlm_block *
-nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock, int remove)
+nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
 {
-       struct nlm_block        **head, *block;
+       struct nlm_block        *block;
        struct file_lock        *fl;
 
        dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
                                file, lock->fl.fl_pid,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end, lock->fl.fl_type);
-       for (head = &nlm_blocked; (block = *head) != 0; head = &block->b_next) {
-               fl = &block->b_call.a_args.lock.fl;
+       list_for_each_entry(block, &nlm_blocked, b_list) {
+               fl = &block->b_call->a_args.lock.fl;
                dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
                                block->b_file, fl->fl_pid,
                                (long long)fl->fl_start,
                                (long long)fl->fl_end, fl->fl_type,
-                               nlmdbg_cookie2a(&block->b_call.a_args.cookie));
+                               nlmdbg_cookie2a(&block->b_call->a_args.cookie));
                if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
-                       if (remove) {
-                               *head = block->b_next;
-                               block->b_queued = 0;
-                       }
+                       kref_get(&block->b_count);
                        return block;
                }
        }
@@ -131,9 +130,9 @@ nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock, int remove)
 
 static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
 {
-       if(a->len != b->len)
+       if (a->len != b->len)
                return 0;
-       if(memcmp(a->data,b->data,a->len))
+       if (memcmp(a->data, b->data, a->len))
                return 0;
        return 1;
 }
@@ -142,18 +141,20 @@ static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
  * Find a block with a given NLM cookie.
  */
 static inline struct nlm_block *
-nlmsvc_find_block(struct nlm_cookie *cookie,  struct sockaddr_in *sin)
+nlmsvc_find_block(struct nlm_cookie *cookie)
 {
        struct nlm_block *block;
 
-       for (block = nlm_blocked; block; block = block->b_next) {
-               dprintk("cookie: head of blocked queue %p, block %p\n", 
-                       nlm_blocked, block);
-               if (nlm_cookie_match(&block->b_call.a_args.cookie,cookie)
-                               && nlm_cmp_addr(sin, &block->b_host->h_addr))
-                       break;
+       list_for_each_entry(block, &nlm_blocked, b_list) {
+               if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
+                       goto found;
        }
 
+       return NULL;
+
+found:
+       dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
+       kref_get(&block->b_count);
        return block;
 }
 
@@ -166,34 +167,40 @@ nlmsvc_find_block(struct nlm_cookie *cookie,  struct sockaddr_in *sin)
  * request, but (as I found out later) that's because some implementations
  * do just this. Never mind the standards comittees, they support our
  * logging industries.
+ *
+ * 10 years later: I hope we can safely ignore these old and broken
+ * clients by now. Let's fix this so we can uniquely identify an incoming
+ * GRANTED_RES message by cookie, without having to rely on the client's IP
+ * address. --okir
  */
-static inline struct nlm_block *
-nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_file *file,
-                               struct nlm_lock *lock, struct nlm_cookie *cookie)
+static struct nlm_block *
+nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
+                   struct nlm_file *file, struct nlm_lock *lock,
+                   struct nlm_cookie *cookie)
 {
        struct nlm_block        *block;
-       struct nlm_host         *host;
-       struct nlm_rqst         *call;
+       struct nlm_rqst         *call = NULL;
 
-       /* Create host handle for callback */
-       host = nlmclnt_lookup_host(&rqstp->rq_addr,
-                               rqstp->rq_prot, rqstp->rq_vers);
-       if (host == NULL)
+       nlm_get_host(host);
+       call = nlm_alloc_call(host);
+       if (call == NULL)
                return NULL;
 
        /* Allocate memory for block, and initialize arguments */
-       if (!(block = (struct nlm_block *) kmalloc(sizeof(*block), GFP_KERNEL)))
+       block = kzalloc(sizeof(*block), GFP_KERNEL);
+       if (block == NULL)
                goto failed;
-       memset(block, 0, sizeof(*block));
-       locks_init_lock(&block->b_call.a_args.lock.fl);
-       locks_init_lock(&block->b_call.a_res.lock.fl);
+       kref_init(&block->b_count);
+       INIT_LIST_HEAD(&block->b_list);
+       INIT_LIST_HEAD(&block->b_flist);
 
-       if (!nlmclnt_setgrantargs(&block->b_call, lock))
+       if (!nlmsvc_setgrantargs(call, lock))
                goto failed_free;
 
        /* Set notifier function for VFS, and init args */
-       block->b_call.a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
-       block->b_call.a_args.cookie = *cookie;  /* see above */
+       call->a_args.lock.fl.fl_flags |= FL_SLEEP;
+       call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
+       nlmclnt_next_cookie(&call->a_args.cookie);
 
        dprintk("lockd: created block %p...\n", block);
 
@@ -201,208 +208,353 @@ nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_file *file,
        block->b_daemon = rqstp->rq_server;
        block->b_host   = host;
        block->b_file   = file;
+       block->b_fl = NULL;
+       file->f_count++;
 
        /* Add to file's list of blocks */
-       block->b_fnext  = file->f_blocks;
-       file->f_blocks  = block;
+       list_add(&block->b_flist, &file->f_blocks);
 
        /* Set up RPC arguments for callback */
-       call = &block->b_call;
-       call->a_host    = host;
+       block->b_call = call;
        call->a_flags   = RPC_TASK_ASYNC;
+       call->a_block = block;
 
        return block;
 
 failed_free:
        kfree(block);
 failed:
-       nlm_release_host(host);
+       nlm_release_call(call);
        return NULL;
 }
 
 /*
- * Delete a block. If the lock was cancelled or the grant callback
- * failed, unlock is set to 1.
+ * Delete a block.
  * It is the caller's responsibility to check whether the file
  * can be closed hereafter.
  */
-static void
-nlmsvc_delete_block(struct nlm_block *block, int unlock)
+static int nlmsvc_unlink_block(struct nlm_block *block)
 {
-       struct file_lock        *fl = &block->b_call.a_args.lock.fl;
-       struct nlm_file         *file = block->b_file;
-       struct nlm_block        **bp;
-
-       dprintk("lockd: deleting block %p...\n", block);
+       int status;
+       dprintk("lockd: unlinking block %p...\n", block);
 
        /* Remove block from list */
+       status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl);
        nlmsvc_remove_block(block);
-       if (fl->fl_next)
-               posix_unblock_lock(file->f_file, fl);
-       if (unlock) {
-               fl->fl_type = F_UNLCK;
-               posix_lock_file(file->f_file, fl);
-               block->b_granted = 0;
-       }
+       return status;
+}
 
-       /* If the block is in the middle of a GRANT callback,
-        * don't kill it yet. */
-       if (block->b_incall) {
-               nlmsvc_insert_block(block, NLM_NEVER);
-               block->b_done = 1;
-               return;
-       }
+static void nlmsvc_free_block(struct kref *kref)
+{
+       struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
+       struct nlm_file         *file = block->b_file;
+
+       dprintk("lockd: freeing block %p...\n", block);
 
        /* Remove block from file's list of blocks */
-       for (bp = &file->f_blocks; *bp; bp = &(*bp)->b_fnext) {
-               if (*bp == block) {
-                       *bp = block->b_fnext;
-                       break;
-               }
+       mutex_lock(&file->f_mutex);
+       list_del_init(&block->b_flist);
+       mutex_unlock(&file->f_mutex);
+
+       nlmsvc_freegrantargs(block->b_call);
+       nlm_release_call(block->b_call);
+       nlm_release_file(block->b_file);
+       kfree(block->b_fl);
+       kfree(block);
+}
+
+static void nlmsvc_release_block(struct nlm_block *block)
+{
+       if (block != NULL)
+               kref_put(&block->b_count, nlmsvc_free_block);
+}
+
+/*
+ * Loop over all blocks and delete blocks held by
+ * a matching host.
+ */
+void nlmsvc_traverse_blocks(struct nlm_host *host,
+                       struct nlm_file *file,
+                       nlm_host_match_fn_t match)
+{
+       struct nlm_block *block, *next;
+
+restart:
+       mutex_lock(&file->f_mutex);
+       list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
+               if (!match(block->b_host, host))
+                       continue;
+               /* Do not destroy blocks that are not on
+                * the global retry list - why? */
+               if (list_empty(&block->b_list))
+                       continue;
+               kref_get(&block->b_count);
+               mutex_unlock(&file->f_mutex);
+               nlmsvc_unlink_block(block);
+               nlmsvc_release_block(block);
+               goto restart;
        }
+       mutex_unlock(&file->f_mutex);
+}
 
-       if (block->b_host)
-               nlm_release_host(block->b_host);
-       nlmclnt_freegrantargs(&block->b_call);
-       kfree(block);
+/*
+ * Initialize arguments for GRANTED call. The nlm_rqst structure
+ * has been cleared already.
+ */
+static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
+{
+       locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
+       memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
+       call->a_args.lock.caller = utsname()->nodename;
+       call->a_args.lock.oh.len = lock->oh.len;
+
+       /* set default data area */
+       call->a_args.lock.oh.data = call->a_owner;
+       call->a_args.lock.svid = lock->fl.fl_pid;
+
+       if (lock->oh.len > NLMCLNT_OHSIZE) {
+               void *data = kmalloc(lock->oh.len, GFP_KERNEL);
+               if (!data)
+                       return 0;
+               call->a_args.lock.oh.data = (u8 *) data;
+       }
+
+       memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
+       return 1;
+}
+
+static void nlmsvc_freegrantargs(struct nlm_rqst *call)
+{
+       if (call->a_args.lock.oh.data != call->a_owner)
+               kfree(call->a_args.lock.oh.data);
+
+       locks_release_private(&call->a_args.lock.fl);
 }
 
 /*
- * Loop over all blocks and perform the action specified.
- * (NLM_ACT_CHECK handled by nlmsvc_inspect_file).
+ * Deferred lock request handling for non-blocking lock
  */
-int
-nlmsvc_traverse_blocks(struct nlm_host *host, struct nlm_file *file, int action)
+static __be32
+nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
 {
-       struct nlm_block        *block, *next;
-
-       down(&file->f_sema);
-       for (block = file->f_blocks; block; block = next) {
-               next = block->b_fnext;
-               if (action == NLM_ACT_MARK)
-                       block->b_host->h_inuse = 1;
-               else if (action == NLM_ACT_UNLOCK) {
-                       if (host == NULL || host == block->b_host)
-                               nlmsvc_delete_block(block, 1);
-               }
+       __be32 status = nlm_lck_denied_nolocks;
+
+       block->b_flags |= B_QUEUED;
+
+       nlmsvc_insert_block(block, NLM_TIMEOUT);
+
+       block->b_cache_req = &rqstp->rq_chandle;
+       if (rqstp->rq_chandle.defer) {
+               block->b_deferred_req =
+                       rqstp->rq_chandle.defer(block->b_cache_req);
+               if (block->b_deferred_req != NULL)
+                       status = nlm_drop_reply;
        }
-       up(&file->f_sema);
-       return 0;
+       dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
+               block, block->b_flags, ntohl(status));
+
+       return status;
 }
 
 /*
  * Attempt to establish a lock, and if it can't be granted, block it
  * if required.
  */
-u32
+__be32
 nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
-                       struct nlm_lock *lock, int wait, struct nlm_cookie *cookie)
+           struct nlm_host *host, struct nlm_lock *lock, int wait,
+           struct nlm_cookie *cookie, int reclaim)
 {
-       struct file_lock        *conflock;
-       struct nlm_block        *block;
+       struct nlm_block        *block = NULL;
        int                     error;
+       __be32                  ret;
 
        dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
-                               file->f_file->f_dentry->d_inode->i_sb->s_id,
-                               file->f_file->f_dentry->d_inode->i_ino,
+                               file->f_file->f_path.dentry->d_inode->i_sb->s_id,
+                               file->f_file->f_path.dentry->d_inode->i_ino,
                                lock->fl.fl_type, lock->fl.fl_pid,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end,
                                wait);
 
-
-       /* Get existing block (in case client is busy-waiting) */
-       block = nlmsvc_lookup_block(file, lock, 0);
-
-       lock->fl.fl_flags |= FL_LOCKD;
-
-again:
        /* Lock file against concurrent access */
-       down(&file->f_sema);
-
-       if (!(conflock = posix_test_lock(file->f_file, &lock->fl))) {
-               error = posix_lock_file(file->f_file, &lock->fl);
-
-               if (block)
-                       nlmsvc_delete_block(block, 0);
-               up(&file->f_sema);
-
-               dprintk("lockd: posix_lock_file returned %d\n", -error);
-               switch(-error) {
-               case 0:
-                       return nlm_granted;
-               case EDEADLK:
-                       return nlm_deadlock;
-               case EAGAIN:
-                       return nlm_lck_denied;
-               default:                        /* includes ENOLCK */
-                       return nlm_lck_denied_nolocks;
+       mutex_lock(&file->f_mutex);
+       /* Get existing block (in case client is busy-waiting)
+        * or create new block
+        */
+       block = nlmsvc_lookup_block(file, lock);
+       if (block == NULL) {
+               block = nlmsvc_create_block(rqstp, host, 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;
+
+       if (block->b_flags & B_QUEUED) {
+               dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
+                                                       block, block->b_flags);
+               if (block->b_granted) {
+                       nlmsvc_unlink_block(block);
+                       ret = nlm_granted;
+                       goto out;
+               }
+               if (block->b_flags & B_TIMED_OUT) {
+                       nlmsvc_unlink_block(block);
+                       ret = nlm_lck_denied;
+                       goto out;
                }
+               ret = nlm_drop_reply;
+               goto out;
        }
 
-       if (!wait) {
-               up(&file->f_sema);
-               return nlm_lck_denied;
+       if (locks_in_grace() && !reclaim) {
+               ret = nlm_lck_denied_grace_period;
+               goto out;
        }
-
-       if (posix_locks_deadlock(&lock->fl, conflock)) {
-               up(&file->f_sema);
-               return nlm_deadlock;
+       if (reclaim && !locks_in_grace()) {
+               ret = nlm_lck_denied_grace_period;
+               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_sema as nlmsvc_create_block may try to
-        * to claim it while doing host garbage collection */
-       if (block == NULL) {
-               up(&file->f_sema);
-               dprintk("lockd: blocking on this lock (allocating).\n");
-               if (!(block = nlmsvc_create_block(rqstp, file, lock, cookie)))
-                       return nlm_lck_denied_nolocks;
-               goto again;
+       if (!wait)
+               lock->fl.fl_flags &= ~FL_SLEEP;
+       error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
+       lock->fl.fl_flags &= ~FL_SLEEP;
+
+       dprintk("lockd: vfs_lock_file returned %d\n", error);
+       switch (error) {
+               case 0:
+                       ret = nlm_granted;
+                       goto out;
+               case -EAGAIN:
+                       /*
+                        * If this is a blocking request for an
+                        * already pending lock request then we need
+                        * to put it back on lockd's block list
+                        */
+                       if (wait)
+                               break;
+                       ret = nlm_lck_denied;
+                       goto out;
+               case FILE_LOCK_DEFERRED:
+                       if (wait)
+                               break;
+                       /* Filesystem lock operation is in progress
+                          Add it to the queue waiting for callback */
+                       ret = nlmsvc_defer_lock_rqst(rqstp, block);
+                       goto out;
+               case -EDEADLK:
+                       ret = nlm_deadlock;
+                       goto out;
+               default:                        /* includes ENOLCK */
+                       ret = nlm_lck_denied_nolocks;
+                       goto out;
        }
 
+       ret = nlm_lck_blocked;
+
        /* Append to list of blocked */
        nlmsvc_insert_block(block, NLM_NEVER);
-
-       if (list_empty(&block->b_call.a_args.lock.fl.fl_block)) {
-               /* Now add block to block list of the conflicting lock
-                  if we haven't done so. */
-               dprintk("lockd: blocking on this lock.\n");
-               posix_block_lock(conflock, &block->b_call.a_args.lock.fl);
-       }
-
-       up(&file->f_sema);
-       return nlm_lck_blocked;
+out:
+       mutex_unlock(&file->f_mutex);
+       nlmsvc_release_block(block);
+       dprintk("lockd: nlmsvc_lock returned %u\n", ret);
+       return ret;
 }
 
 /*
  * Test for presence of a conflicting lock.
  */
-u32
-nlmsvc_testlock(struct nlm_file *file, struct nlm_lock *lock,
-                                      struct nlm_lock *conflock)
+__be32
+nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
+               struct nlm_host *host, struct nlm_lock *lock,
+               struct nlm_lock *conflock, struct nlm_cookie *cookie)
 {
-       struct file_lock        *fl;
+       struct nlm_block        *block = NULL;
+       int                     error;
+       __be32                  ret;
 
        dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
-                               file->f_file->f_dentry->d_inode->i_sb->s_id,
-                               file->f_file->f_dentry->d_inode->i_ino,
+                               file->f_file->f_path.dentry->d_inode->i_sb->s_id,
+                               file->f_file->f_path.dentry->d_inode->i_ino,
                                lock->fl.fl_type,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end);
 
-       if ((fl = posix_test_lock(file->f_file, &lock->fl)) != NULL) {
-               dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
-                               fl->fl_type, (long long)fl->fl_start,
-                               (long long)fl->fl_end);
-               conflock->caller = "somehost";  /* FIXME */
-               conflock->oh.len = 0;           /* don't return OH info */
-               conflock->fl = *fl;
-               return nlm_lck_denied;
+       /* Get existing block (in case client is busy-waiting) */
+       block = nlmsvc_lookup_block(file, lock);
+
+       if (block == NULL) {
+               struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
+
+               if (conf == NULL)
+                       return nlm_granted;
+               block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
+               if (block == NULL) {
+                       kfree(conf);
+                       return nlm_granted;
+               }
+               block->b_fl = conf;
+       }
+       if (block->b_flags & B_QUEUED) {
+               dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n",
+                       block, block->b_flags, block->b_fl);
+               if (block->b_flags & B_TIMED_OUT) {
+                       nlmsvc_unlink_block(block);
+                       ret = nlm_lck_denied;
+                       goto out;
+               }
+               if (block->b_flags & B_GOT_CALLBACK) {
+                       nlmsvc_unlink_block(block);
+                       if (block->b_fl != NULL
+                                       && block->b_fl->fl_type != F_UNLCK) {
+                               lock->fl = *block->b_fl;
+                               goto conf_lock;
+                       } else {
+                               ret = nlm_granted;
+                               goto out;
+                       }
+               }
+               ret = nlm_drop_reply;
+               goto out;
+       }
+
+       if (locks_in_grace()) {
+               ret = nlm_lck_denied_grace_period;
+               goto out;
+       }
+       error = vfs_test_lock(file->f_file, &lock->fl);
+       if (error == FILE_LOCK_DEFERRED) {
+               ret = nlmsvc_defer_lock_rqst(rqstp, block);
+               goto out;
+       }
+       if (error) {
+               ret = nlm_lck_denied_nolocks;
+               goto out;
+       }
+       if (lock->fl.fl_type == F_UNLCK) {
+               ret = nlm_granted;
+               goto out;
        }
 
-       return nlm_granted;
+conf_lock:
+       dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
+               lock->fl.fl_type, (long long)lock->fl.fl_start,
+               (long long)lock->fl.fl_end);
+       conflock->caller = "somehost";  /* FIXME */
+       conflock->len = strlen(conflock->caller);
+       conflock->oh.len = 0;           /* don't return OH info */
+       conflock->svid = lock->fl.fl_pid;
+       conflock->fl.fl_type = lock->fl.fl_type;
+       conflock->fl.fl_start = lock->fl.fl_start;
+       conflock->fl.fl_end = lock->fl.fl_end;
+       ret = nlm_lck_denied;
+out:
+       if (block)
+               nlmsvc_release_block(block);
+       return ret;
 }
 
 /*
@@ -412,14 +564,14 @@ nlmsvc_testlock(struct nlm_file *file, struct nlm_lock *lock,
  * afterwards. In this case the block will still be there, and hence
  * must be removed.
  */
-u32
+__be32
 nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
 {
        int     error;
 
        dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
-                               file->f_file->f_dentry->d_inode->i_sb->s_id,
-                               file->f_file->f_dentry->d_inode->i_ino,
+                               file->f_file->f_path.dentry->d_inode->i_sb->s_id,
+                               file->f_file->f_path.dentry->d_inode->i_ino,
                                lock->fl.fl_pid,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end);
@@ -428,7 +580,7 @@ nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
        nlmsvc_cancel_blocked(file, lock);
 
        lock->fl.fl_type = F_UNLCK;
-       error = posix_lock_file(file->f_file, &lock->fl);
+       error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
 
        return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
 }
@@ -440,23 +592,89 @@ nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
  * be in progress.
  * The calling procedure must check whether the file can be closed.
  */
-u32
+__be32
 nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
 {
        struct nlm_block        *block;
+       int status = 0;
 
        dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
-                               file->f_file->f_dentry->d_inode->i_sb->s_id,
-                               file->f_file->f_dentry->d_inode->i_ino,
+                               file->f_file->f_path.dentry->d_inode->i_sb->s_id,
+                               file->f_file->f_path.dentry->d_inode->i_ino,
                                lock->fl.fl_pid,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end);
 
-       down(&file->f_sema);
-       if ((block = nlmsvc_lookup_block(file, lock, 1)) != NULL)
-               nlmsvc_delete_block(block, 1);
-       up(&file->f_sema);
-       return nlm_granted;
+       if (locks_in_grace())
+               return nlm_lck_denied_grace_period;
+
+       mutex_lock(&file->f_mutex);
+       block = nlmsvc_lookup_block(file, lock);
+       mutex_unlock(&file->f_mutex);
+       if (block != NULL) {
+               vfs_cancel_lock(block->b_file->f_file,
+                               &block->b_call->a_args.lock.fl);
+               status = nlmsvc_unlink_block(block);
+               nlmsvc_release_block(block);
+       }
+       return status ? nlm_lck_denied : nlm_granted;
+}
+
+/*
+ * This is a callback from the filesystem for VFS file lock requests.
+ * It will be used if fl_grant is defined and the filesystem can not
+ * respond to the request immediately.
+ * For GETLK request it will copy the reply to the nlm_block.
+ * For SETLK or SETLKW request it will get the local posix lock.
+ * In all cases it will move the block to the head of nlm_blocked q where
+ * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
+ * deferred rpc for GETLK and SETLK.
+ */
+static void
+nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf,
+                            int result)
+{
+       block->b_flags |= B_GOT_CALLBACK;
+       if (result == 0)
+               block->b_granted = 1;
+       else
+               block->b_flags |= B_TIMED_OUT;
+       if (conf) {
+               if (block->b_fl)
+                       __locks_copy_lock(block->b_fl, conf);
+       }
+}
+
+static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf,
+                                       int result)
+{
+       struct nlm_block *block;
+       int rc = -ENOENT;
+
+       lock_kernel();
+       list_for_each_entry(block, &nlm_blocked, b_list) {
+               if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
+                       dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
+                                                       block, block->b_flags);
+                       if (block->b_flags & B_QUEUED) {
+                               if (block->b_flags & B_TIMED_OUT) {
+                                       rc = -ENOLCK;
+                                       break;
+                               }
+                               nlmsvc_update_deferred_block(block, conf, result);
+                       } else if (result == 0)
+                               block->b_granted = 1;
+
+                       nlmsvc_insert_block(block, 0);
+                       svc_wake_up(block->b_daemon);
+                       rc = 0;
+                       break;
+               }
+       }
+       unlock_kernel();
+       if (rc == -ENOENT)
+               printk(KERN_WARNING "lockd: grant for unknown block\n");
+       return rc;
 }
 
 /*
@@ -469,11 +687,11 @@ nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
 static void
 nlmsvc_notify_blocked(struct file_lock *fl)
 {
-       struct nlm_block        **bp, *block;
+       struct nlm_block        *block;
 
        dprintk("lockd: VFS unblock notification for block %p\n", fl);
-       for (bp = &nlm_blocked; (block = *bp) != 0; bp = &block->b_next) {
-               if (nlm_compare_locks(&block->b_call.a_args.lock.fl, fl)) {
+       list_for_each_entry(block, &nlm_blocked, b_list) {
+               if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
                        nlmsvc_insert_block(block, 0);
                        svc_wake_up(block->b_daemon);
                        return;
@@ -488,9 +706,10 @@ static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
        return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
 }
 
-struct lock_manager_operations nlmsvc_lock_operations = {
+const struct lock_manager_operations nlmsvc_lock_operations = {
        .fl_compare_owner = nlmsvc_same_owner,
        .fl_notify = nlmsvc_notify_blocked,
+       .fl_grant = nlmsvc_grant_deferred,
 };
 
 /*
@@ -508,17 +727,15 @@ static void
 nlmsvc_grant_blocked(struct nlm_block *block)
 {
        struct nlm_file         *file = block->b_file;
-       struct nlm_lock         *lock = &block->b_call.a_args.lock;
-       struct file_lock        *conflock;
+       struct nlm_lock         *lock = &block->b_call->a_args.lock;
        int                     error;
 
        dprintk("lockd: grant blocked lock %p\n", block);
 
-       /* First thing is lock the file */
-       down(&file->f_sema);
+       kref_get(&block->b_count);
 
        /* Unlink block request from list */
-       nlmsvc_remove_block(block);
+       nlmsvc_unlink_block(block);
 
        /* If b_granted is true this means we've been here before.
         * Just retry the grant callback, possibly refreshing the RPC
@@ -529,24 +746,23 @@ nlmsvc_grant_blocked(struct nlm_block *block)
        }
 
        /* Try the lock operation again */
-       if ((conflock = posix_test_lock(file->f_file, &lock->fl)) != NULL) {
-               /* Bummer, we blocked again */
-               dprintk("lockd: lock still blocked\n");
+       lock->fl.fl_flags |= FL_SLEEP;
+       error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
+       lock->fl.fl_flags &= ~FL_SLEEP;
+
+       switch (error) {
+       case 0:
+               break;
+       case FILE_LOCK_DEFERRED:
+               dprintk("lockd: lock still blocked error %d\n", error);
                nlmsvc_insert_block(block, NLM_NEVER);
-               posix_block_lock(conflock, &lock->fl);
-               up(&file->f_sema);
+               nlmsvc_release_block(block);
                return;
-       }
-
-       /* Alright, no conflicting lock. Now lock it for real. If the
-        * following yields an error, this is most probably due to low
-        * memory. Retry the lock in a few seconds.
-        */
-       if ((error = posix_lock_file(file->f_file, &lock->fl)) < 0) {
+       default:
                printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
-                               -error, __FUNCTION__);
+                               -error, __func__);
                nlmsvc_insert_block(block, 10 * HZ);
-               up(&file->f_sema);
+               nlmsvc_release_block(block);
                return;
        }
 
@@ -554,17 +770,21 @@ callback:
        /* Lock was granted by VFS. */
        dprintk("lockd: GRANTing blocked lock.\n");
        block->b_granted = 1;
-       block->b_incall  = 1;
 
-       /* Schedule next grant callback in 30 seconds */
-       nlmsvc_insert_block(block, 30 * HZ);
+       /* keep block on the list, but don't reattempt until the RPC
+        * completes or the submission fails
+        */
+       nlmsvc_insert_block(block, NLM_NEVER);
 
-       /* Call the client */
-       nlm_get_host(block->b_call.a_host);
-       if (nlmsvc_async_call(&block->b_call, NLMPROC_GRANTED_MSG,
-                                               nlmsvc_grant_callback) < 0)
-               nlm_release_host(block->b_call.a_host);
-       up(&file->f_sema);
+       /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked
+        * will queue up a new one if this one times out
+        */
+       error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
+                               &nlmsvc_grant_ops);
+
+       /* RPC submission failed, wait a bit and retry */
+       if (error < 0)
+               nlmsvc_insert_block(block, 10 * HZ);
 }
 
 /*
@@ -575,24 +795,25 @@ callback:
  * chain once more in order to have it removed by lockd itself (which can
  * then sleep on the file semaphore without disrupting e.g. the nfs client).
  */
-static void
-nlmsvc_grant_callback(struct rpc_task *task)
+static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
 {
-       struct nlm_rqst         *call = (struct nlm_rqst *) task->tk_calldata;
-       struct nlm_block        *block;
+       struct nlm_rqst         *call = data;
+       struct nlm_block        *block = call->a_block;
        unsigned long           timeout;
-       struct sockaddr_in      *peer_addr = RPC_PEERADDR(task->tk_client);
 
        dprintk("lockd: GRANT_MSG RPC callback\n");
-       dprintk("callback: looking for cookie %s, host (%u.%u.%u.%u)\n",
-               nlmdbg_cookie2a(&call->a_args.cookie),
-               NIPQUAD(peer_addr->sin_addr.s_addr));
-       if (!(block = nlmsvc_find_block(&call->a_args.cookie, peer_addr))) {
-               dprintk("lockd: no block for cookie %s, host (%u.%u.%u.%u)\n",
-                       nlmdbg_cookie2a(&call->a_args.cookie),
-                       NIPQUAD(peer_addr->sin_addr.s_addr));
-               return;
-       }
+
+       lock_kernel();
+       /* if the block is not on a list at this point then it has
+        * been invalidated. Don't try to requeue it.
+        *
+        * FIXME: it's possible that the block is removed from the list
+        * after this check but before the nlmsvc_insert_block. In that
+        * case it will be added back. Perhaps we need better locking
+        * for nlm_blocked?
+        */
+       if (list_empty(&block->b_list))
+               goto out;
 
        /* Technically, we should down the file semaphore here. Since we
         * move the block towards the head of the queue only, no harm
@@ -600,57 +821,72 @@ nlmsvc_grant_callback(struct rpc_task *task)
        if (task->tk_status < 0) {
                /* RPC error: Re-insert for retransmission */
                timeout = 10 * HZ;
-       } else if (block->b_done) {
-               /* Block already removed, kill it for real */
-               timeout = 0;
        } else {
                /* Call was successful, now wait for client callback */
                timeout = 60 * HZ;
        }
        nlmsvc_insert_block(block, timeout);
        svc_wake_up(block->b_daemon);
-       block->b_incall = 0;
+out:
+       unlock_kernel();
+}
+
+static void nlmsvc_grant_release(void *data)
+{
+       struct nlm_rqst         *call = data;
 
-       nlm_release_host(call->a_host);
+       lock_kernel();
+       nlmsvc_release_block(call->a_block);
+       unlock_kernel();
 }
 
+static const struct rpc_call_ops nlmsvc_grant_ops = {
+       .rpc_call_done = nlmsvc_grant_callback,
+       .rpc_release = nlmsvc_grant_release,
+};
+
 /*
  * We received a GRANT_RES callback. Try to find the corresponding
  * block.
  */
 void
-nlmsvc_grant_reply(struct svc_rqst *rqstp, struct nlm_cookie *cookie, u32 status)
+nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
 {
        struct nlm_block        *block;
-       struct nlm_file         *file;
 
-       dprintk("grant_reply: looking for cookie %x, host (%08x), s=%d \n", 
-               *(unsigned int *)(cookie->data), 
-               ntohl(rqstp->rq_addr.sin_addr.s_addr), status);
-       if (!(block = nlmsvc_find_block(cookie, &rqstp->rq_addr)))
+       dprintk("grant_reply: looking for cookie %x, s=%d \n",
+               *(unsigned int *)(cookie->data), status);
+       if (!(block = nlmsvc_find_block(cookie)))
                return;
-       file = block->b_file;
 
-       file->f_count++;
-       down(&file->f_sema);
-       if ((block = nlmsvc_find_block(cookie,&rqstp->rq_addr)) != NULL) {
-               if (status == NLM_LCK_DENIED_GRACE_PERIOD) {
+       if (block) {
+               if (status == nlm_lck_denied_grace_period) {
                        /* Try again in a couple of seconds */
                        nlmsvc_insert_block(block, 10 * HZ);
-                       block = NULL;
                } else {
                        /* Lock is now held by client, or has been rejected.
                         * In both cases, the block should be removed. */
-                       up(&file->f_sema);
-                       if (status == NLM_LCK_GRANTED)
-                               nlmsvc_delete_block(block, 0);
-                       else
-                               nlmsvc_delete_block(block, 1);
+                       nlmsvc_unlink_block(block);
                }
        }
-       if (!block)
-               up(&file->f_sema);
-       nlm_release_file(file);
+       nlmsvc_release_block(block);
+}
+
+/* Helper function to handle retry of a deferred block.
+ * If it is a blocking lock, call grant_blocked.
+ * For a non-blocking lock or test lock, revisit the request.
+ */
+static void
+retry_deferred_block(struct nlm_block *block)
+{
+       if (!(block->b_flags & B_GOT_CALLBACK))
+               block->b_flags |= B_TIMED_OUT;
+       nlmsvc_insert_block(block, NLM_TIMEOUT);
+       dprintk("revisit block %p flags %d\n",  block, block->b_flags);
+       if (block->b_deferred_req) {
+               block->b_deferred_req->revisit(block->b_deferred_req, 0);
+               block->b_deferred_req = NULL;
+       }
 }
 
 /*
@@ -661,26 +897,28 @@ nlmsvc_grant_reply(struct svc_rqst *rqstp, struct nlm_cookie *cookie, u32 status
 unsigned long
 nlmsvc_retry_blocked(void)
 {
-       struct nlm_block        *block;
+       unsigned long   timeout = MAX_SCHEDULE_TIMEOUT;
+       struct nlm_block *block;
+
+       while (!list_empty(&nlm_blocked) && !kthread_should_stop()) {
+               block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
 
-       dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
-                       nlm_blocked,
-                       nlm_blocked? nlm_blocked->b_when : 0);
-       while ((block = nlm_blocked) != 0) {
                if (block->b_when == NLM_NEVER)
                        break;
-               if (time_after(block->b_when,jiffies))
+               if (time_after(block->b_when, jiffies)) {
+                       timeout = block->b_when - jiffies;
                        break;
-               dprintk("nlmsvc_retry_blocked(%p, when=%ld, done=%d)\n",
-                       block, block->b_when, block->b_done);
-               if (block->b_done)
-                       nlmsvc_delete_block(block, 0);
-               else
+               }
+
+               dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
+                       block, block->b_when);
+               if (block->b_flags & B_QUEUED) {
+                       dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
+                               block, block->b_granted, block->b_flags);
+                       retry_deferred_block(block);
+               } else
                        nlmsvc_grant_blocked(block);
        }
 
-       if ((block = nlm_blocked) && block->b_when != NLM_NEVER)
-               return (block->b_when - jiffies);
-
-       return MAX_SCHEDULE_TIMEOUT;
+       return timeout;
 }