If the device minor cannot be allocated or the cdrom fails to be
registered then the mutex should be destroyed.
Signed-off-by: Simon Arlott <[email protected]>
Fixes: 51a858817dcd ("scsi: sr: get rid of sr global mutex")
Cc: [email protected]
---
On 30/05/2020 17:41, James Bottomley wrote:
> On Sat, 2020-05-30 at 09:24 -0700, Bart Van Assche wrote:
>> Please add Fixes: and Cc: stable tags.
I've added a Fixes: tag and Cc:'d stable.
> This isn't really a bug, is it? mutex_destroy is a nop unless lock
> debugging is enabled in which case it checks the lock is unlocked and
> marks it as unusable to detect a use after destroy. Since the
> structure containing the mutex is kfree'd in the next statement, kasan
> would also detect any use after free. That's not to say we shouldn't
> do this to be fully correct ... just that it has no potential ever to
> have user visible impact so there doesn't seem to be much point
> cluttering up the stable process with it.
If the current lock debugging implementation in stable will be ok with
it then I'd agree there's no reason to put it in stable kernels, except
that the commit this fixes was added to stable with this bug and one in
sr_block_release (72655c0ebd1d).
drivers/scsi/sr.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/scsi/sr.c b/drivers/scsi/sr.c
index d2fe3fa470f9..8d062d4f3ce0 100644
--- a/drivers/scsi/sr.c
+++ b/drivers/scsi/sr.c
@@ -817,6 +817,7 @@ static int sr_probe(struct device *dev)
fail_put:
put_disk(disk);
+ mutex_destroy(&cd->lock);
fail_free:
kfree(cd);
fail:
--
2.17.1
--
Simon Arlott
If the cdrom fails to be registered then the device minor should be
deallocated.
Signed-off-by: Simon Arlott <[email protected]>
Cc: [email protected]
---
On 30/05/2020 17:24, Bart Van Assche wrote:
> On 2020-05-30 02:33, Simon Arlott wrote:
>> If the cdrom fails to be registered then the device minor should be
>> deallocated.
>
> Also for this patch, please add Fixes: and Cc: stable tags.
I've Cc:'d stable.
There is no specific previous commit that this fixes. I was just
checking the rest of sr_probe when making a patch for the first issue.
drivers/scsi/sr.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/drivers/scsi/sr.c b/drivers/scsi/sr.c
index 8d062d4f3ce0..1e13c6a0f0ca 100644
--- a/drivers/scsi/sr.c
+++ b/drivers/scsi/sr.c
@@ -797,7 +797,7 @@ static int sr_probe(struct device *dev)
cd->cdi.disk = disk;
if (register_cdrom(&cd->cdi))
- goto fail_put;
+ goto fail_minor;
/*
* Initialize block layer runtime PM stuffs before the
@@ -815,6 +815,10 @@ static int sr_probe(struct device *dev)
return 0;
+fail_minor:
+ spin_lock(&sr_index_lock);
+ clear_bit(minor, sr_index_bits);
+ spin_unlock(&sr_index_lock);
fail_put:
put_disk(disk);
mutex_destroy(&cd->lock);
--
2.17.1
--
Simon Arlott
On Sat, 30 May 2020 18:58:25 +0100, Simon Arlott wrote:
> If the device minor cannot be allocated or the cdrom fails to be
> registered then the mutex should be destroyed.
Applied to 5.8/scsi-queue, thanks!
[1/2] scsi: sr: Fix sr_probe() missing mutex_destroy
https://git.kernel.org/mkp/scsi/c/a247e07f8dad
[2/2] scsi: sr: Fix sr_probe() missing deallocate of device minor
https://git.kernel.org/mkp/scsi/c/6555781b3fde
--
Martin K. Petersen Oracle Linux Engineering