Re: [PATCH 2/8] esp: do not unref managed frag pages in esp_ssg_unref()
maher azz <[email protected]>
| Newsgroups | gmane.linux.network |
|---|---|
| Message-ID | <CAFQ-Uc8ek5+KFQ+3BXnNope+NoR-XMMz-u2xL-pL2aHdww8-Vg@mail.gmail.com> |
Hi Steffen, Thanks, the review was correct on all three points. I've sent a v2 that addresses them: - It now calls skb_zcopy_downgrade_managed() before ESP mutates the frag array (in esp_output_head/esp6_output_head), instead of the early return in esp_ssg_unref(). As the review noted, that early return also suppressed release of the ESP-owned trailer and destination page references; downgrading fixes both the underflow and that leak, the same way __ip_append_data() does. - Corrected the Fixes: tag to 753f1ca4e1e5 (managed frags infrastructure), not the 2017 skb_cow_data commits. One open question: for the stable floor I used 753f1ca4e1e5. Would you prefer the commit that first made managed-frag skbs reachable from esp_output(), or an explicit v6.1+ range? Please take a look. Thanks, Maher