| Summary: | prefix_pton is inefficient | ||
|---|---|---|---|
| Product: | py-radix | Reporter: | Göran Weinholt <weinholt> |
| Component: | Default | Assignee: | Assigned to nobody <unassigned-bugs> |
| Status: | CLOSED MOVED | ||
| Severity: | enhancement | CC: | djm |
| Priority: | P2 | ||
| Version: | -current | ||
| Hardware: | ix86 | ||
| OS: | Linux | ||
| Attachments: | change prefix_pton to use inet_pton() | ||
py-radix is not longer in this bugtracker closing bugs resolved before the openssh-8.9 release |
Created attachment 1710 [details] change prefix_pton to use inet_pton() py-radix 0.5 uses getaddrinfo() to parse IPv4/IPv6 addresses. On GNU/Linux this procedure creates a socket and performs a few syscalls on it. It would be better to parse addresses without performing syscalls. I noticed this when writing a program that generates DHCP configuration for ~15k hosts. When running the program on the server running the DHCP daemon it ran out of some sort of kernel memory (NFS stopped working, neighbor table filled, new TCP connections failed). For some reason the development machine didn't have the same problem. I traced it down to the getaddrinfo() call. To solve this problem I patched py-radix to use inet_pton instead. Please review the attached patch. Also, the testbed runs in 3.1 seconds when it previously took 4.3s. The patch also fixes the error messages. Previously invalid prefix lengths would result in a garbage error message, and a failed call to getaddrinfo() would return a pointer to inside the function's stack.