gnrc_netif: fix packet leak with gnrc_netif_pktq & netdev_new_api#20983
gnrc_netif: fix packet leak with gnrc_netif_pktq & netdev_new_api#20983maribu merged 3 commits intoRIOT-OS:masterfrom
Conversation
|
Hmm not so fast, looks like we are getting a double free now 😕 |
|
So |
But also test the other combinations so you do not introduce regressions there ;-). |
|
I think I know what was going on here: |
Since all the new addions are now the I'm not saying there are no leaks when combining !netdev_new with netdev_new, but those would be unrelated to the fixes for the netdev_new-only case - I'll give that a try tomorrow. |
miri64
left a comment
There was a problem hiding this comment.
Not sure this works. I am not sure if just skipping pkt (without releasing it) now generates a leak.
f238c84 to
8318a48
Compare
|
A border router ( |
8318a48 to
1629a6a
Compare
Contribution description
With
netdev_new_apipackets are not released on send anymore.gnrc_netif_pktqadds a hold on a packet to not release it on send.Even though 1 + 1 - 1 - 1 should be 0, we are leaking packets on send.
Testing procedure
Run a driver that makes use of both
netdev_new_apiandgnrc_netif_pktq(e.g.at86rf215):master
pktbuf stats
this PR
pktbuf is empty
Issues/PRs references
#11263 (comment)