You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@trafficserver.apache.org by "Peter Chou (JIRA)" <ji...@apache.org> on 2016/03/30 23:59:25 UTC
[jira] [Updated] (TS-4252) Some plugins are causing seg-faults when
using getopt_long with optind = 1.
[ https://issues.apache.org/jira/browse/TS-4252?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Peter Chou updated TS-4252:
---------------------------
Affects Version/s: 6.1.1
Environment: Linux Intel 64-bit Ubuntu 14.04 and RHEL 7.
Description:
There are several global plugins which experience segmentation fault related to parsing (argc, argv) arguments using getopt_long(). Often, the plugins display debug output showing corrupted arguments, e.g., arguments belonging to previous entries in plugins.config. This has been confirmed to happen with background_fetch.so and regex_revalidate.so. The other plugins remap_stats and stale_while_revalidate may also be affected based on code review.
This issue is corrected if the plugins are modified to use optind = 0 instead of optind = 1 before calling getopt_long(). Note that the majority of plugins are using optind = 0 already. Per the Linux man page, you should only need to set optind = 1 between scanning different argument vectors, but you must set optind = 0 to cause some re-initialization to occur if you make use of GNU extensions in the opstring argument of getopt_long(). I am not sure if this applies to prior plugin using GNU extensions or current one (or going between one or the other), but it would seem safer to use optind = 0 always.
was:
In "plugin.config" if we just do background_fetch.so with no arguments we get a segmentation fault with messages saying invalid option with argument text from previous lines in the configuration file. If I just add a garbage argument like "background_fetch.so bleah" there is no fault.
I noticed that this plugin initialized optind to 1 before calling getopt_long while others such as tcpinfo set it to 0. Setting it to 0 also prevents the fault. Is this the correct fix?
Summary: Some plugins are causing seg-faults when using getopt_long with optind = 1. (was: background_fetch.so segfaults with no arguments as a global plugin.)
Updated description to indicate multiple plugins affected.
> Some plugins are causing seg-faults when using getopt_long with optind = 1.
> ---------------------------------------------------------------------------
>
> Key: TS-4252
> URL: https://issues.apache.org/jira/browse/TS-4252
> Project: Traffic Server
> Issue Type: Bug
> Components: Plugins
> Affects Versions: 6.1.1
> Environment: Linux Intel 64-bit Ubuntu 14.04 and RHEL 7.
> Reporter: Peter Chou
> Assignee: Leif Hedstrom
> Fix For: 6.2.0
>
>
> There are several global plugins which experience segmentation fault related to parsing (argc, argv) arguments using getopt_long(). Often, the plugins display debug output showing corrupted arguments, e.g., arguments belonging to previous entries in plugins.config. This has been confirmed to happen with background_fetch.so and regex_revalidate.so. The other plugins remap_stats and stale_while_revalidate may also be affected based on code review.
> This issue is corrected if the plugins are modified to use optind = 0 instead of optind = 1 before calling getopt_long(). Note that the majority of plugins are using optind = 0 already. Per the Linux man page, you should only need to set optind = 1 between scanning different argument vectors, but you must set optind = 0 to cause some re-initialization to occur if you make use of GNU extensions in the opstring argument of getopt_long(). I am not sure if this applies to prior plugin using GNU extensions or current one (or going between one or the other), but it would seem safer to use optind = 0 always.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)