[PATCH] An option to make fcntl & flock locks fair

Matthew Wilcox (willy@debian.org)
Thu, 22 Aug 2002 15:18:48 +0100


Shlomi Fish asked about including first-come, first-served style locking
for posix and flock locks. After some back-and-forth, we came up with
the following patch which seems unintrusive enough to bother including.
Personally, I doubt the utility of this, but someone might have an
application for it, and the code's already written.

diff -urpNX dontdiff linux-2.5.31/fs/locks.c linux-2.5.31-willy/fs/locks.c
--- linux-2.5.31/fs/locks.c 2002-08-01 14:16:39.000000000 -0700
+++ linux-2.5.31-willy/fs/locks.c 2002-08-21 08:29:29.000000000 -0700
@@ -126,6 +126,14 @@
#include <asm/semaphore.h>
#include <asm/uaccess.h>

+/* Defining FAIR_LOCKS prevents readers from starving writers. I'm not
+ * sure anyone has a usage pattern where this is actually a problem, but
+ * it's a trivial change to allow you to enable this, so I added it.
+ * Note that we record the owner and check it first, so you can't deadlock
+ * on an attempt to modify your own lock.
+ */
+#undef FAIR_LOCKS
+
#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
@@ -503,24 +508,22 @@ static void locks_delete_lock(struct fil
locks_free_lock(fl);
}

-/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
- * checks for shared/exclusive status of overlapping locks.
+/*
+ * Determine if lock sys_fl blocks lock caller_fl. Write locks require
+ * exclusive access, read locks can share. If you define FAIR_LOCKS,
+ * readers can only share until a writer comes along and blocks.
*/
static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
{
- switch (caller_fl->fl_type) {
- case F_RDLCK:
- return (sys_fl->fl_type == F_WRLCK);
-
- case F_WRLCK:
- return (1);
-
- default:
- printk(KERN_ERR "locks_conflict(): impossible lock type - %d\n",
- caller_fl->fl_type);
- break;
- }
- return (0); /* This should never happen */
+ if (caller_fl->fl_type == F_WRLCK)
+ return 1;
+ if (sys_fl->fl_type == F_WRLCK)
+ return 1;
+#ifdef FAIR_LOCKS
+ if (!list_empty(&sys_fl->fl_block))
+ return 1;
+#endif
+ return 0;
}

/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific

-- 
Revolutions do not require corporate support.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/