2006-09-04 13:31:25

by Alan

Subject: [PATCH] ide: Fix crash on repeated reset

[email protected] (Michał Mirosław) [1] reported a
problem (bugzilla #7023) where a user initiated reset while the IDE
layer was already resetting the channel caused a crash, and provided a
rough fix.

This is a slightly cleaner version of the fix which tracks the reset
state and blocks further reset requests while a reset is in progress.

Note this is not a security issue - random end users can't access the
ioctl in question anyway.


[1] This is what bugzilla.kernel.org calls him anyway


Signed-off-by: Alan Cox <[email protected]>

diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.18-rc5-mm1/include/linux/ide.h linux-2.6.18-rc5-mm1/include/linux/ide.h
--- linux.vanilla-2.6.18-rc5-mm1/include/linux/ide.h 2006-09-01 13:39:19.000000000 +0100
+++ linux-2.6.18-rc5-mm1/include/linux/ide.h 2006-09-01 13:55:32.000000000 +0100
@@ -825,6 +825,9 @@
unsigned int sleeping : 1;
/* BOOL: polling active & poll_timeout field valid */
unsigned int polling : 1;
+ /* BOOL: in a polling reset situation. Must not trigger another reset yet */
+ unsigned resetting : 1;
+
/* current drive */
ide_drive_t *drive;
/* ptr to current hwif in linked-list */
diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.18-rc5-mm1/drivers/ide/ide.c linux-2.6.18-rc5-mm1/drivers/ide/ide.c
--- linux.vanilla-2.6.18-rc5-mm1/drivers/ide/ide.c 2006-09-01 13:39:05.000000000 +0100
+++ linux-2.6.18-rc5-mm1/drivers/ide/ide.c 2006-09-01 13:53:03.000000000 +0100
@@ -1370,6 +1370,11 @@
*/

spin_lock_irqsave(&ide_lock, flags);
+
+ if (HWGROUP(drive)->resetting) {
+ spin_unlock_irqrestore(&ide_lock, flags);
+ return -EBUSY;
+ }

ide_abort(drive, "drive reset");

diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.18-rc5-mm1/drivers/ide/ide-iops.c linux-2.6.18-rc5-mm1/drivers/ide/ide-iops.c
--- linux.vanilla-2.6.18-rc5-mm1/drivers/ide/ide-iops.c 2006-08-30 18:31:46.000000000 +0100
+++ linux-2.6.18-rc5-mm1/drivers/ide/ide-iops.c 2006-09-01 13:53:03.000000000 +0100
@@ -998,6 +998,7 @@
}
/* done polling */
hwgroup->polling = 0;
+ hwgroup->resetting = 0;
return ide_stopped;
}

@@ -1057,6 +1058,7 @@
}
}
hwgroup->polling = 0; /* done polling */
+ hwgroup->resetting = 0; /* done reset attempt */
return ide_stopped;
}

@@ -1143,6 +1145,7 @@

/* For an ATAPI device, first try an ATAPI SRST. */
if (drive->media != ide_disk && !do_not_try_atapi) {
+ hwgroup->resetting = 1;
pre_reset(drive);
SELECT_DRIVE(drive);
udelay (20);
@@ -1168,6 +1171,7 @@
return ide_stopped;
}

+ hwgroup->resetting = 1;
/*
* Note that we also set nIEN while resetting the device,
* to mask unwanted interrupts from the interface during the reset.

2006-09-04 19:38:11

by Andreas Mohr

Subject: Re: [PATCH] ide: Fix crash on repeated reset

Hi,

On Mon, Sep 04, 2006 at 02:54:01PM +0100, Alan Cox wrote:
> unsigned int sleeping : 1;
> /* BOOL: polling active & poll_timeout field valid */
> unsigned int polling : 1;
> + /* BOOL: in a polling reset situation. Must not trigger another reset yet */
> + unsigned resetting : 1;
> +

Inconsistent variable type declarations/formatting?

Andreas Mohr

2006-09-05 13:18:52

by Alan

Subject: Re: [PATCH] ide: Fix crash on repeated reset

Ar Llu, 2006-09-04 am 21:38 +0200, ysgrifennodd Andreas Mohr:
> Hi,
>
> On Mon, Sep 04, 2006 at 02:54:01PM +0100, Alan Cox wrote:
> > unsigned int sleeping : 1;
> > /* BOOL: polling active & poll_timeout field valid */
> > unsigned int polling : 1;
> > + /* BOOL: in a polling reset situation. Must not trigger another reset yet */
> > + unsigned resetting : 1;
> > +
>
> Inconsistent variable type declarations/formatting?

Harmless but true. Updated in my tree.

Alan