Hi,
We notice that skb is already freed by dev_kfree_skb in pn533_fill_fragment_skbs, but follow error handler branch #line 2288 and #line 2356, skb is freed again, seems like a double free issue. Would you like to have a look at them? We will provide patch for them after confirmation.
https://github.com/torvalds/linux/blob/master/drivers/nfc/pn533/pn533.c#L2288
Thanks so much,
Chengfeng
On Fri, Nov 05, 2021 at 09:22:12AM +0000, YE Chengfeng wrote:
> Hi,
>
> We notice that skb is already freed by dev_kfree_skb in
> pn533_fill_fragment_skbs, but follow error handler branch #line 2288
> and #line 2356, skb is freed again, seems like a double free issue.
> Would you like to have a look at them?
>
> https://github.com/torvalds/linux/blob/master/drivers/nfc/pn533/pn533.c#L2288
>
The code is buggy, yes, but it's a bit tricky to fix.
pn533_fill_fragment_skbs() never returns error codes, it returns zero
on error. Specifically it clears out the &dev->fragment_skb list and
then returns the length of the list "skb_queue_len(&dev->fragment_skb)"
which is now zero.
Returning success on transmit failure is fine because the network stack
thinks it was lost somewhere in the network and resends it. But
probably it should return -ENOMEM? But changing the return would make
the other caller into a double free now.
So probably the correct fix is to
1) Make pn533_fill_fragment_skbs() return -ENOMEM on error
2) Don't call dev_kfree_skb(skb); on error in pn533_fill_fragment_skbs().
Only call it on the success path.
3) Change the callers to check for negatives instead of <= 0
> We will provide patch for them after confirmation.
Sounds great. You can fix it however you want. My ideas are a
suggestion only.
regards,
dan carpenter
Thanks for your advice. I follow it and send a v2 patch, is it correct?
Best regards,
Chengfeng
-----?ʼ?ԭ??-----
??????: Dan Carpenter <[email protected]>
????ʱ??: 2021??11??5?? 20:17
?ռ???: YE Chengfeng <[email protected]>
????: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
????: Re: nfc: pn533: suspected double free when pn533_fill_fragment_skbs() return value <= 0
On Fri, Nov 05, 2021 at 09:22:12AM +0000, YE Chengfeng wrote:
> Hi,
>
> We notice that skb is already freed by dev_kfree_skb in
> pn533_fill_fragment_skbs, but follow error handler branch #line 2288
> and #line 2356, skb is freed again, seems like a double free issue.
> Would you like to have a look at them?
>
> https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgith
> ub.com%2Ftorvalds%2Flinux%2Fblob%2Fmaster%2Fdrivers%2Fnfc%2Fpn533%2Fpn
> 533.c%23L2288&data=04%7C01%7Ccyeaa%40connect.ust.hk%7C29df72d1f8a3
> 4473a15008d9a05642cf%7C6c1d415239d044ca88d9b8d6ddca0708%7C1%7C0%7C6377
> 17114624016520%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2l
> uMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=9nssVI5NtyEUvnH6k
> NP60dqSmtgfWILoKXimcJWIiKc%3D&reserved=0
>
The code is buggy, yes, but it's a bit tricky to fix.
pn533_fill_fragment_skbs() never returns error codes, it returns zero on error. Specifically it clears out the &dev->fragment_skb list and then returns the length of the list "skb_queue_len(&dev->fragment_skb)"
which is now zero.
Returning success on transmit failure is fine because the network stack thinks it was lost somewhere in the network and resends it. But probably it should return -ENOMEM? But changing the return would make the other caller into a double free now.
So probably the correct fix is to
1) Make pn533_fill_fragment_skbs() return -ENOMEM on error
2) Don't call dev_kfree_skb(skb); on error in pn533_fill_fragment_skbs().
Only call it on the success path.
3) Change the callers to check for negatives instead of <= 0
> We will provide patch for them after confirmation.
Sounds great. You can fix it however you want. My ideas are a suggestion only.
regards,
dan carpenter