Following up, I built openafs-stable-1_4_x from CVS today on RHEL5, which
includes this patch, and I re-enabled afs_linux_flock (which is disabled
in current CVS). It seemed to work correctly and didn't crash. RHEL5 is
2.6.18 based thus it uses the new API.
I haven't tried 2.4 yet, maybe I will try RHEL3 next week.
-chris
On Thu, 8 Feb 2007, Christopher Allen Wing wrote:
The following fixes the crash and seems ok on RHEL4 (2.6.9 kernel). I
haven't tried it on newer 2.6 kernels. I might have an opportunity to try it
out on RHEL5 tomorrow, which is 2.6.18 based and thus has the 'new' API for
flock_lock_file*().
I believe it is safe to just clear FL_SLEEP in the structure handed to us,
assuming that we want to retain the current semantics (tell the kernel never
to wait when we ask it to perform a locking operation):
--- openafs-1.4.x/src/afs/LINUX/osi_vnodeops.c.orig 2007-02-08
14:22:27.000000000 -0500
+++ openafs-1.4.x/src/afs/LINUX/osi_vnodeops.c 2007-02-08 14:28:26.000000000
-0500
@@ -475,10 +475,8 @@
if ((code == 0 || flp->fl_type == F_UNLCK) &&
(cmd == F_SETLK || cmd == F_SETLKW)) {
#ifdef AFS_LINUX26_ENV
- struct file_lock flp2;
- flp2 = *flp;
- flp2.fl_flags &=~ FL_SLEEP;
- code = posix_lock_file(fp, &flp2);
+ flp->fl_flags &=~ FL_SLEEP;
+ code = posix_lock_file(fp, flp);
#else
code = posix_lock_file(fp, flp, 0);
#endif @@ -534,10 +532,8 @@
if ((code == 0 || flp->fl_type == F_UNLCK) &&
(cmd == F_SETLK || cmd == F_SETLKW)) {
- struct file_lock flp2;
- flp2 = *flp;
- flp2.fl_flags &=~ FL_SLEEP;
- code = flock_lock_file_wait(fp, &flp2);
+ flp->fl_flags &=~ FL_SLEEP;
+ code = flock_lock_file_wait(fp, flp);
if (code && flp->fl_type != F_UNLCK) {
struct AFS_FLOCK flock2;
flock2 = flock;
Anyone interested in trying this on linux-2.4 or linux-2.6.17+?
-Chris
[EMAIL PROTECTED]
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel
