ptmx fix duplicate idr_remove
RFC PATCH 1/1 ptmx fix duplicate idr_remove
Remove duplicate call to idr_remove() in ptmx_open.
Error during open can result in call to release_dev() followed by call to idr_remove(). release_dev already calls idr_remove so the second call can cause a stack dump in idr_remove()->sub_remove() flagging an attempt to release an already released entry.
I reproduces this on a machine with a misconfigured X server (attempting to restart multiple times rapidly) getting the same error as the 1st link below.
This also seems to be related to:
http://marc.theaimsgroup.com/?l=selinux&m=110536513426735&w=2
http://marc.theaimsgroup.com/?l=selinux&m=110596994916785&w=2
The stack dump can occur on close (as well as open) as shown in the 1st instance above, possible from something like:
process A - open (index=0), open fail to out1,
release_dev calls idr_remove (index 0), down(sem) sleeps
process B - open (index=0), open OK (idr allocated) process A - wake and call idr_remove on index 0
...
process B - close, release_dev, stack dump on idr_remove (index=0)
because entry already removed
Comments?
drivers/char/tty_io.c +1 -0
@@ -2188,6 +2188,7 @@ static int ptmx_open(struct inode * inodreturn 0 ; out1:release_dev (filp); +return retval; out:down (&allocated_ptys_lock);idr_remove (&allocated_ptys, index);
Quoted 32 lines · Paul Fulghum
Remove duplicate call to
idr_remove()inptmx_open.Error during open can result in call to
release_dev()followed by call toidr_remove().release_devalready callsidr_removeso the second call can cause a stack dump in idr_remove()->sub_remove() flagging an attempt to release an already released entry.I reproduces this on a machine with a misconfigured X server (attempting to restart multiple times rapidly) getting the same error as the 1st link below.
This also seems to be related to:
http://marc.theaimsgroup.com/?l=selinux&m=110536513426735&w=2
http://marc.theaimsgroup.com/?l=selinux&m=110596994916785&w=2The stack dump can occur on close (as well as open) as shown in the 1st instance above, possible from something like:
process A - open (index=0), open fail to out1,
release_devcallsidr_remove(index 0), down(sem) sleeps
process B - open (index=0), open OK (idr allocated) process A - wake and callidr_removeon index 0
...
process B - close,release_dev, stack dump onidr_remove(index=0)
because entry already removedComments?
drivers/char/tty_io.c static int ptmx_open(struct inode * inod · line 2194
Quoted 7 lines
drivers/char/tty_io.c +1
-0@@ -2188,6 +2188,7 @@ static int ptmx_open(struct inode * inodreturn 0 ; out1:release_dev (filp); +return retval; out:down (&allocated_ptys_lock);idr_remove (&allocated_ptys, index);
Look solid.
Except for this, in release_dev():
/* check whether both sides are closing ... */ if (!tty_closing || (o_tty && !o_tty_closing)) return;
if that's taken, we won't have run idr_remove() at all?
Quoted 7 lines · Andrew Morton
Except for this, in
release_dev():/* check whether both sides are closing ... */ if (!tty_closing || (o_tty && !o_tty_closing)) return;if that's taken, we won't have run
idr_remove()at all?
ptmx_open creates the ptm/pts pair.
If ptmx_open fails, the conditions
for releasing the ptm (and associated pts)
have been met: no active references.
Signature · 2 lines
Paul Fulghum Microgate Systems, Ltd.