Ulrich found another small detail wrt. POSIX requirements for threads -
this time it's the recursion features (read-held lock being write-locked
means an upgrade if the same 'process' is the owner, means a deadlock if a
different 'process').
this requirement even makes some sense - the group of threads who own a
lock really own all rights to the lock as well.
the attached patch against BK-curr fixes this, all testcases pass now.
(inter-process testcases as well, which are not affected by this patch.)
(SIGURG and SIGIO semantics should also continue to work - there's some
more stuff we can optimize with the new pidhash in this area, but that's
for later.)
Ingo
--- linux/fs/locks.c.orig Wed Sep 25 10:28:26 2002
+++ linux/fs/locks.c Wed Sep 25 10:28:41 2002
@@ -252,7 +252,7 @@
return -ENOMEM;
fl->fl_file = filp;
- fl->fl_pid = current->pid;
+ fl->fl_pid = current->tgid;
fl->fl_flags = (cmd & LOCK_NB) ? FL_FLOCK : FL_FLOCK | FL_SLEEP;
fl->fl_type = type;
fl->fl_end = OFFSET_MAX;
@@ -308,7 +308,7 @@
fl->fl_end = OFFSET_MAX;
fl->fl_owner = current->files;
- fl->fl_pid = current->pid;
+ fl->fl_pid = current->tgid;
fl->fl_file = filp;
fl->fl_flags = FL_POSIX;
fl->fl_notify = NULL;
@@ -348,7 +348,7 @@
fl->fl_end = OFFSET_MAX;
fl->fl_owner = current->files;
- fl->fl_pid = current->pid;
+ fl->fl_pid = current->tgid;
fl->fl_file = filp;
fl->fl_flags = FL_POSIX;
fl->fl_notify = NULL;
@@ -377,7 +377,7 @@
return -ENOMEM;
fl->fl_owner = current->files;
- fl->fl_pid = current->pid;
+ fl->fl_pid = current->tgid;
fl->fl_file = filp;
fl->fl_flags = FL_LEASE;
@@ -669,7 +669,7 @@
int error;
fl.fl_owner = current->files;
- fl.fl_pid = current->pid;
+ fl.fl_pid = current->tgid;
fl.fl_file = filp;
fl.fl_flags = FL_POSIX | FL_ACCESS | FL_SLEEP;
fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
@@ -1241,7 +1241,7 @@
*before = fl;
list_add(&fl->fl_link, &file_lock_list);
- error = f_setown(filp, current->pid, 1);
+ error = f_setown(filp, current->tgid, 1);
out_unlock:
unlock_kernel();
return error;
@@ -1632,7 +1632,7 @@
lock.fl_start = 0;
lock.fl_end = OFFSET_MAX;
lock.fl_owner = owner;
- lock.fl_pid = current->pid;
+ lock.fl_pid = current->tgid;
lock.fl_file = filp;
if (filp->f_op && filp->f_op->lock != NULL) {
Ingo Molnar wrote:
>
> Ulrich found another small detail wrt. POSIX requirements for threads -
> this time it's the recursion features (read-held lock being write-locked
> means an upgrade if the same 'process' is the owner, means a deadlock if a
> different 'process').
>
I had submitted same patch on May,31st and got the following
response from Matthew Wilcox. Removing the pid check from the
locks_same_owner() will fix this problem.
==================== Matthew Wilcox's response ====================
Saurabh Desai believes that locks created by threads should not conflict
with each other. I'm inclined to agree; I don't know why the test for
->fl_pid was added, but the comment suggests that whoever added it wasn't
sure either.
Frankly, I have no clue about the intended semantics for threads, and
SUS v3 does not offer any enlightenment. But it seems reasonable that
processes which share a files_struct should share locks. After all,
if one process closes the fd, they'll remove locks belonging to the
other process.
Here's a patch generated against 2.4; it also applies to 2.5.
Please apply.
===== fs/locks.c 1.9 vs edited =====
--- 1.9/fs/locks.c Mon Jun 3 18:49:43 2002
+++ edited/fs/locks.c Fri Jun 7 21:24:12 2002
@@ -380,15 +380,12 @@
}
/*
- * Check whether two locks have the same owner
- * N.B. Do we need the test on PID as well as owner?
- * (Clone tasks should be considered as one "owner".)
+ * Locks are deemed to have the same owner if the tasks share files_struct.
*/
static inline int
locks_same_owner(struct file_lock *fl1, struct file_lock *fl2)
{
- return (fl1->fl_owner == fl2->fl_owner) &&
- (fl1->fl_pid == fl2->fl_pid);
+ return (fl1->fl_owner == fl2->fl_owner);
}
/* Remove waiter from blocker's block list.