lockd: replace semaphore, sleep_on

Use wait_event, completion instead of the obsolete sleep_on functions and the abused semaphore

---

 2.6.9-bk-041020-thomas/fs/lockd/clntlock.c |   12 +++++++-----
 2.6.9-bk-041020-thomas/fs/lockd/svc.c      |   23
+++++++++++++----------
 2 files changed, 20 insertions(+), 15 deletions(-)

diff -puN fs/lockd/svc.c~lockd fs/lockd/svc.c
--- 2.6.9-bk-041020/fs/lockd/svc.c~lockd 2004-10-20 15:56:52.000000000
+0200
+++ 2.6.9-bk-041020-thomas/fs/lockd/svc.c 2004-10-20 15:56:52.000000000
+0200
@@ -46,7 +46,7 @@ static pid_t nlmsvc_pid;
int nlmsvc_grace_period;
unsigned long nlmsvc_timeout;

-static DECLARE_MUTEX_LOCKED(lockd_start);
+static DECLARE_WAIT_QUEUE_HEAD(lockd_start);
static DECLARE_WAIT_QUEUE_HEAD(lockd_exit);

/*
@@ -109,7 +109,7 @@ lockd(struct svc_rqst *rqstp)
* Let our maker know we're running.
*/
nlmsvc_pid = current->pid;
- up(&lockd_start);
+ wake_up(&lockd_start);

daemonize("lockd");

@@ -258,8 +258,15 @@ lockd_up(void)
"lockd_up: create thread failed, error=%d\n", error);
goto destroy_and_out;
}
- down(&lockd_start);
-
+ /*
+ * Wait for the lockd process to start, but since we're holding
+ * the lockd semaphore, we can't wait around forever ...
+ */
+ if (wait_event_interruptible_timeout(lockd_start,
+ nlmsvc_pid != 0, HZ) <= 0) {
+ printk(KERN_WARNING
+ "lockd_down: lockd failed to start\n");
+ }
/*
* Note: svc_serv structures have an initial use count of 1,
* so we exit through here on both success and failure.
@@ -298,16 +305,12 @@ lockd_down(void)
* Wait for the lockd process to exit, but since we're holding
* the lockd semaphore, we can't wait around forever ... */
- clear_thread_flag(TIF_SIGPENDING);
- interruptible_sleep_on_timeout(&lockd_exit, HZ);
- if (nlmsvc_pid) {
+ if (wait_event_interruptible_timeout(lockd_exit,
+ nlmsvc_pid == 0, HZ) <= 0) {
printk(KERN_WARNING
"lockd_down: lockd failed to exit, clearing pid\n");
nlmsvc_pid = 0;
}
- spin_lock_irq(&current->sighand->siglock);
- recalc_sigpending();
- spin_unlock_irq(&current->sighand->siglock);
out:
up(&nlmsvc_sema);
}
diff -puN fs/lockd/clntlock.c~lockd fs/lockd/clntlock.c
--- 2.6.9-bk-041020/fs/lockd/clntlock.c~lockd 2004-10-20
15:56:52.000000000 +0200
+++ 2.6.9-bk-041020-thomas/fs/lockd/clntlock.c 2004-10-20
15:56:52.000000000 +0200
@@ -6,6 +6,7 @@
* Copyright (C) 1996, Olaf Kirch <[email protected]>
*/

+#include <linux/completion.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/time.h>
@@ -32,7 +33,7 @@ static int reclaimer(void *ptr);
*/
struct nlm_wait {
struct nlm_wait * b_next; /* linked list */
- wait_queue_head_t b_wait; /* where to wait on */
+ struct completion b_wait; /* where to wait on */
struct nlm_host * b_host;
struct file_lock * b_lock; /* local file lock */
unsigned short b_reclaim; /* got to reclaim lock */
@@ -53,7 +54,7 @@ nlmclnt_block(struct nlm_host *host, str

block.b_host = host;
block.b_lock = fl;
- init_waitqueue_head(&block.b_wait);
+ init_completion(&block.b_wait);
block.b_status = NLM_LCK_BLOCKED;
block.b_next = nlm_blocked;
nlm_blocked = &block;
@@ -69,7 +70,8 @@ nlmclnt_block(struct nlm_host *host, str
* a 1 minute timeout would do. See the comment before
* nlmclnt_lock for an explanation.
*/
- sleep_on_timeout(&block.b_wait, 30*HZ);
+
+ wait_for_completion_timeout(&block.b_wait, 30*HZ);

for (head = &nlm_blocked; *head; head = &(*head)->b_next) {
if (*head == &block) {
@@ -118,7 +120,7 @@ nlmclnt_grant(struct nlm_lock *lock)
* wake up the caller.
*/
block->b_status = NLM_LCK_GRANTED;
- wake_up(&block->b_wait);
+ complete(&block->b_wait);

return nlm_granted;
}
@@ -233,7 +235,7 @@ restart:
for (block = nlm_blocked; block; block = block->b_next) {
if (block->b_host == host) {
block->b_status = NLM_LCK_DENIED_GRACE_PERIOD;
- wake_up(&block->b_wait);
+ complete(&block->b_wait);
}
}

_

Not really my area, you want to ask Trond about this.

"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

What is the point of all this timeout business when starting/stopping lockd? We sure as hell don't want lockd to survive us if we are the process that is responsible for killing it, and neither do we want to start locking if there is no lockd to handle callbacks. Please see the example of rpciod in net/sunrpc/sched.c for how to do this correctly.

As for the client changes to the GRANTED mechanism, where is wait_for_completion_timeout() defined? It is not part of Linus' kernel.

Cheers,
Trond

Trond Myklebust <[email protected]>