On Mon, Apr 9, 2018 at 11:07 PM, Laura Abbott <labbott@xxxxxxxxxx> wrote:
There's an ongoing effort to remove VLAs[1] from the kernel to eventually
turn on -Wvla. The single VLA can either take a value of 2 or 4 so switch
to the upper bound.
[1] https://lkml.org/lkml/2018/3/7/621
Signed-off-by: Laura Abbott <labbott@xxxxxxxxxx>
---
drivers/misc/tifm_7xx1.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/misc/tifm_7xx1.c b/drivers/misc/tifm_7xx1.c
index e5f108713dd8..690eaaea5ce4 100644
--- a/drivers/misc/tifm_7xx1.c
+++ b/drivers/misc/tifm_7xx1.c
@@ -239,7 +239,8 @@ static int tifm_7xx1_resume(struct pci_dev *dev)
unsigned long timeout;
unsigned int good_sockets = 0, bad_sockets = 0;
unsigned long flags;
- unsigned char new_ids[fm->num_sockets];
+ /* Maximum number of entries is 4 */
+ unsigned char new_ids[4];
DECLARE_COMPLETION_ONSTACK(finish_resume);
I like the idea of removing all the VLAs, but this one appears to make
the code less robust rather than more: In case of an unexpected
fm->num_sockets value, we now not only overflow the kernel stack
area but also the local variable into the neighboring stack slots.
Maybe add an extra overflow check?
Arnd
diff --git a/drivers/misc/tifm_7xx1.c b/drivers/misc/tifm_7xx1.c
index e5f108713dd8..c702cd92d396 100644
--- a/drivers/misc/tifm_7xx1.c
+++ b/drivers/misc/tifm_7xx1.c
@@ -239,9 +239,12 @@ static int tifm_7xx1_resume(struct pci_dev *dev)
unsigned long timeout;
unsigned int good_sockets = 0, bad_sockets = 0;
unsigned long flags;
- unsigned char new_ids[fm->num_sockets];
+ /* Maximum number of entries is 4 */
+ unsigned char new_ids[4];
DECLARE_COMPLETION_ONSTACK(finish_resume);
+ if (WARN_ON(fm->num_sockets > ARRAY_SIZE(new_ids)))
+ return -ENXIO;
+
pci_set_power_state(dev, PCI_D0);
pci_restore_state(dev);
rc = pci_enable_device(dev);