| Summary: | <sys-libs/glibc-2.20-r2: getaddrinfo() writes DNS queries to random file descriptors under high load (CVE-2013-7423) | ||
|---|---|---|---|
| Product: | Gentoo Security | Reporter: | Kristian Fiskerstrand (RETIRED) <k_f> |
| Component: | Vulnerabilities | Assignee: | Gentoo Security <security> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | cyberbat83, hanno, toolchain |
| Priority: | Normal | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| URL: | https://sourceware.org/bugzilla/show_bug.cgi?id=15946 | ||
| Whiteboard: | A3 [glsa cleanup] | ||
| Package list: | Runtime testing required: | --- | |
| Bug Depends on: | 516884, 544034 | ||
| Bug Blocks: | |||
This issue seems to be confirmed fixed in version 2.20. There was some confusion whether the issue still persists after this version, however that seems to have been limited to a bad testcase. Versions prior to 2.20 are vulnerable to this issue. https://sourceware.org/ml/glibc-bugs/2015-01/msg00226.html CVE-2013-7423 (http://nvd.nist.gov/nvd.cfm?cvename=CVE-2013-7423): The send_dg function in resolv/res_send.c in GNU C Library (aka glibc or libc6) before 2.20 does not properly reuse file descriptors, which allows remote attackers to send DNS queries to unintended locations via a large number of request that trigger a call to the getaddrinfo function. This issue was resolved and addressed in GLSA 201602-02 at https://security.gentoo.org/glsa/201602-02 by GLSA coordinator Tobias Heinlein (keytoaster). |
From ${URL}: Under high load, getaddrinfo() starts sending DNS queries to random file descriptors, e.g. some unrelated socket connected to a remote service. The attached code reproduces the bug on at least the following configurations: Archlinux libc6 2.18 Debian libc6 2.6.11 Debian libc6 2.13-38 Debian libc6 2.17-92 Ubuntu libc6 2.17-0ubuntu5 What the code does is to fill the file descriptor space, closing and creating many file descriptors, to maximize the chances of reproducing the bug: - a thread listens to a local unix socket - a thread connects to the unix socket, never writes to it, dups the connection as much as possible (fills the fd space), closes the dups, and starts dup()ing again - lots of threads call getaddrinfo() Under less than a minute, the listener starts reading garbage. The garbage received by the listener seems to always be a full, well-formed, DNS query. It seems to always be an AAAA query, even when hints.ai_family is AF_INET. All queries are similar, only the id changes. - -- CVE assigned today at http://seclists.org/oss-sec/2015/q1/316