There are potential double free bugs in nfcmrvl usb driver among
fw_dnld_rx_work(), fw_dnld_timeout() and nfcmrvl_nci_unregister_dev().
All these three functions will call fw_dnld_over(). The fw_dnld_rx_work()
is a work item, the fw_dnld_timeout() is a timer handler and the
nfcmrvl_nci_unregister_dev() is called when nfcmrvl_nci device is
detaching. So these three functions could execute concurrently.
The race between fw_dnld_rx_work() and nfcmrvl_nci_unregister_dev()
can be shown as below:
(Thread 1) | (Thread 2)
| fw_dnld_rx_work
| fw_dnld_over
| release_firmware
| kfree(fw); //(1)
nfcmrvl_disconnect |
nfcmrvl_nci_unregister_dev |
nfcmrvl_fw_dnld_abort |
fw_dnld_over | ...
if (priv->fw_dnld.fw) |
release_firmware |
kfree(fw); //(2) |
... | priv->fw_dnld.fw = NULL;
When the fw_dnld_rx_work work item is executing, we detach the device.
The release_firmware() will deallocate firmware in position (1),
but firmware will be deallocated again in position (2), which
leads to double free.
The race between fw_dnld_rx_work() and fw_dnld_timeout()
can be shown as below:
(Thread 1) | (Thread 2)
| fw_dnld_rx_work
| fw_dnld_over
| release_firmware
| kfree(fw); //(1)
fw_dnld_timeout |
fw_dnld_over | ...
if (priv->fw_dnld.fw) |
release_firmware |
kfree(fw); //(2) |
... | priv->fw_dnld.fw = NULL;
The release_firmware() will deallocate firmware in position (1),
but firmware will be deallocated again in position (2), which
leads to double free.
The race between fw_dnld_timeout() and nfcmrvl_nci_unregister_dev()
can be shown as below.
(Thread 1) | (Thread 2)
| nfcmrvl_disconnect
| nfcmrvl_nci_unregister_dev
| nfcmrvl_fw_dnld_abort
| fw_dnld_over
| release_firmware
| kfree(fw); //(1)
fw_dnld_timeout |
fw_dnld_over | ...
if (priv->fw_dnld.fw) |
release_firmware |
kfree(fw); //(2) |
... | priv->fw_dnld.fw = NULL;
The release_firmware() will deallocate firmware in position (1),
but firmware will be deallocated again in position (2), which
leads to double free.
This patch adds spin_lock_irq in fw_dnld_over() in order to synchronize
among different threads that call fw_dnld_over(). The priv->fw_dnld.fw will
be set to NULL after work item is finished and fw_dnld_over() called by
other threads will check whether priv->fw_dnld.fw is NULL. So the double
free bug could be prevented.
Fixes: 3194c6870158e3 ("NFC: nfcmrvl: add firmware download support")
Signed-off-by: Duoming Zhou <[email protected]>
Reviewed-by: Lin Ma <[email protected]>
---
Changes in V3:
- Make commit message more clearer.
- Use spin_lock_irq to synchronize.
drivers/nfc/nfcmrvl/fw_dnld.c | 3 +++
drivers/nfc/nfcmrvl/fw_dnld.h | 2 ++
2 files changed, 5 insertions(+)
diff --git a/drivers/nfc/nfcmrvl/fw_dnld.c b/drivers/nfc/nfcmrvl/fw_dnld.c
index e83f65596a8..c22a4556db5 100644
--- a/drivers/nfc/nfcmrvl/fw_dnld.c
+++ b/drivers/nfc/nfcmrvl/fw_dnld.c
@@ -92,12 +92,14 @@ static struct sk_buff *alloc_lc_skb(struct nfcmrvl_private *priv, uint8_t plen)
static void fw_dnld_over(struct nfcmrvl_private *priv, u32 error)
{
+ spin_lock_irq(&priv->fw_dnld.lock);
if (priv->fw_dnld.fw) {
release_firmware(priv->fw_dnld.fw);
priv->fw_dnld.fw = NULL;
priv->fw_dnld.header = NULL;
priv->fw_dnld.binary_config = NULL;
}
+ spin_unlock_irq(&priv->fw_dnld.lock);
atomic_set(&priv->ndev->cmd_cnt, 0);
@@ -451,6 +453,7 @@ int nfcmrvl_fw_dnld_init(struct nfcmrvl_private *priv)
if (!priv->fw_dnld.rx_wq)
return -ENOMEM;
skb_queue_head_init(&priv->fw_dnld.rx_q);
+ spin_lock_init(&priv->fw_dnld.lock);
return 0;
}
diff --git a/drivers/nfc/nfcmrvl/fw_dnld.h b/drivers/nfc/nfcmrvl/fw_dnld.h
index 7c4d91b0191..6974c307947 100644
--- a/drivers/nfc/nfcmrvl/fw_dnld.h
+++ b/drivers/nfc/nfcmrvl/fw_dnld.h
@@ -75,6 +75,8 @@ struct nfcmrvl_fw_dnld {
struct sk_buff_head rx_q;
struct timer_list timer;
+ /* To synchronize among different threads that call firmware.*/
+ spinlock_t lock;
};
int nfcmrvl_fw_dnld_init(struct nfcmrvl_private *priv);
--
2.17.1
Hello there,
Sorry here but actually I didn't "review" this patch but only offer some suggestions.
It seems that the current version of patch mainly focus on solving the data race. However, I'd prefer to make sure this function
> static void fw_dnld_over(struct nfcmrvl_private *priv, u32 error)
never be able to be called more than once. Maybe add some additional flag with lock is more appropriate.
Regards
Lin