afb5d8b2d
Denys Vlasenko
2020-09-29
Subject
ntpd: fix the case where two replies received at once and first one causes a step
Binary
No binary (build failed)
Architecture & Compiler Flags
—
Build Log
(log yok)
Commit Diff GitHub →
networking/ntpd.c+9 -0
| @@ -2025,6 +2025,15 @@ recv_and_process_peer_pkt(peer_t *p) | |
| offset = 0; | |
| + | /* The below can happen as follows: |
| + | * = we receive two peer rsponses at once. |
| + | * = recv_and_process_peer_pkt(PEER1) -> update_local_clock() |
| + | * -> step_time() and it closes all other fds, sets all ->fd to -1. |
| + | * = recv_and_process_peer_pkt(PEER2) sees PEER2->fd == -1 |
| + | */ |
| + | if (p->p_fd < 0) |
| + | return; |
| + | |
| /* We can recvfrom here and check from.IP, but some multihomed | |
| * ntp servers reply from their *other IP*. | |
| * TODO: maybe we should check at least what we can: from.port == 123? | |