If the bf_get() call in lpfc_mbx_cmpl_rdp_page_a2() does succeeds, execution
continues normally and mp gets kfree()d.
If the subsequent call to lpfc_sli_issue_mbox() fails execution jumps to the
error label where lpfc_mbuf_free() is called with mp->virt and mp->phys as
function arguments. This is the use after free. Following the use after free mp
gets kfree()d again which is a double free.
Signed-off-by: Johannes Thumshirn <[email protected]>
---
drivers/scsi/lpfc/lpfc_mbox.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/lpfc/lpfc_mbox.c b/drivers/scsi/lpfc/lpfc_mbox.c
index eb62772..4abb93a 100644
--- a/drivers/scsi/lpfc/lpfc_mbox.c
+++ b/drivers/scsi/lpfc/lpfc_mbox.c
@@ -2284,7 +2284,7 @@ lpfc_mbx_cmpl_rdp_page_a2(struct lpfc_hba *phba, LPFC_MBOXQ_t *mbox)
(struct lpfc_rdp_context *)(mbox->context2);
if (bf_get(lpfc_mqe_status, &mbox->u.mqe))
- goto error;
+ goto error_mbuf_free;
lpfc_sli_bemem_bcopy(mp->virt, &rdp_context->page_a2,
DMP_SFF_PAGE_A2_SIZE);
@@ -2299,13 +2299,14 @@ lpfc_mbx_cmpl_rdp_page_a2(struct lpfc_hba *phba, LPFC_MBOXQ_t *mbox)
mbox->mbox_cmpl = lpfc_mbx_cmpl_rdp_link_stat;
mbox->context2 = (struct lpfc_rdp_context *) rdp_context;
if (lpfc_sli_issue_mbox(phba, mbox, MBX_NOWAIT) == MBX_NOT_FINISHED)
- goto error;
+ goto error_cmd_free;
return;
-error:
+error_mbuf_free:
lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
+error_cmd_free:
lpfc_sli4_mbox_cmd_free(phba, mbox);
rdp_context->cmpl(phba, rdp_context, FAILURE);
}
--
2.5.0
Johannes Thumshirn wrote:
> If the bf_get() call in lpfc_mbx_cmpl_rdp_page_a2() does succeeds, execution
> continues normally and mp gets kfree()d.
>
> If the subsequent call to lpfc_sli_issue_mbox() fails execution jumps to the
> error label where lpfc_mbuf_free() is called with mp->virt and mp->phys as
> function arguments. This is the use after free. Following the use after free mp
> gets kfree()d again which is a double free.
A similar patch was posted by Colin Ian King on 2015-07-31 [1].
[1] http://marc.info/?l=linux-scsi&m=143835937206204&w=2
Sebastian
Sebastian Herbszt <[email protected]> writes:
> Johannes Thumshirn wrote:
>> If the bf_get() call in lpfc_mbx_cmpl_rdp_page_a2() does succeeds, execution
>> continues normally and mp gets kfree()d.
>>
>> If the subsequent call to lpfc_sli_issue_mbox() fails execution jumps to the
>> error label where lpfc_mbuf_free() is called with mp->virt and mp->phys as
>> function arguments. This is the use after free. Following the use after free mp
>> gets kfree()d again which is a double free.
>
> A similar patch was posted by Colin Ian King on 2015-07-31 [1].
>
> [1] http://marc.info/?l=linux-scsi&m=143835937206204&w=2
OK,
Is it already in James' tree (haven't checked)? The problematic code was
merged for 4.2-rc1 so if the fix (Collin's or mine I don't care) could go
in while we're still in the rc phase, we could avoid all that stable
circus.
Thanks for digging this out.
Byte,
Johannes
Johannes Thumshirn wrote:
> Sebastian Herbszt <[email protected]> writes:
>
> > Johannes Thumshirn wrote:
> >> If the bf_get() call in lpfc_mbx_cmpl_rdp_page_a2() does succeeds, execution
> >> continues normally and mp gets kfree()d.
> >>
> >> If the subsequent call to lpfc_sli_issue_mbox() fails execution jumps to the
> >> error label where lpfc_mbuf_free() is called with mp->virt and mp->phys as
> >> function arguments. This is the use after free. Following the use after free mp
> >> gets kfree()d again which is a double free.
> >
> > A similar patch was posted by Colin Ian King on 2015-07-31 [1].
> >
> > [1] http://marc.info/?l=linux-scsi&m=143835937206204&w=2
>
> OK,
>
> Is it already in James' tree (haven't checked)? The problematic code was
> merged for 4.2-rc1 so if the fix (Collin's or mine I don't care) could go
> in while we're still in the rc phase, we could avoid all that stable
> circus.
>
> Thanks for digging this out.
>
> Byte,
> Johannes
It is not yet in scsi.git.
James S., Dick, which patch do you prefer?
Sebastian
On 8/18/2015 6:27 PM, Sebastian Herbszt wrote:
> Johannes Thumshirn wrote:
>> Sebastian Herbszt <[email protected]> writes:
>>
>>> Johannes Thumshirn wrote:
>>>> If the bf_get() call in lpfc_mbx_cmpl_rdp_page_a2() does succeeds, execution
>>>> continues normally and mp gets kfree()d.
>>>>
>>>> If the subsequent call to lpfc_sli_issue_mbox() fails execution jumps to the
>>>> error label where lpfc_mbuf_free() is called with mp->virt and mp->phys as
>>>> function arguments. This is the use after free. Following the use after free mp
>>>> gets kfree()d again which is a double free.
>>> A similar patch was posted by Colin Ian King on 2015-07-31 [1].
>>>
>>> [1] http://marc.info/?l=linux-scsi&m=143835937206204&w=2
>> OK,
>>
>> Is it already in James' tree (haven't checked)? The problematic code was
>> merged for 4.2-rc1 so if the fix (Collin's or mine I don't care) could go
>> in while we're still in the rc phase, we could avoid all that stable
>> circus.
>>
>> Thanks for digging this out.
>>
>> Byte,
>> Johannes
> It is not yet in scsi.git.
>
> James S., Dick, which patch do you prefer?
>
> Sebastian
I looked at both and liked Johannes patch better.
-- james s