[libvirt] Re: [PATCH] libvirtd: new config-file option: unix_sock_dir [was Re: adding tests....
"Daniel P. Berrange" <berrange(a)redhat.com> wrote:
On Mon, Feb 09, 2009 at 02:01:19PM +0100, Jim Meyering wrote:
> However, first things first:
>
> Here's a patch that adds two blocks, neither pretty,
> but with less duplication than the 3rd alternative,
> which duplicates both the snprintf and the result comparison.
> (of course, I'll use only one of them)
>
> BTW, this also eliminates the uses of PATH_MAX that were
> vestiges of a messy rebase. Now we test against maxlen.
>
> Of these two, I prefer the latter (slightly less duplication).
> Do you care?
I don't particular like either option - too much line noise in there.
Two thoughts - printf is redundant in the first place, since the compiler
will happily concatenate 2 static strings, so we can just strdup(). In
the second case, we could asprintf instead, and so have something like
if (uid == SYSTEM_UID)
server->logDir = strdup(LOCAL_STATE_DIR "/log/libvirt")
else
virAsprintf(&server->logDir, "%s/.libvirt/log", dir_prefix))
if (!server->logDir)
... oom handling ...
Can't do that. The logDir member is declared like this:
qemud/qemud.h: char logDir[PATH_MAX];
which made me see why it was using PATH_MAX.
and we always want to check for buffer overrun.
so here's what I'll do:
diff --git a/qemud/qemud.c b/qemud/qemud.c
index f8c3c97..d9d7ce3 100644
--- a/qemud/qemud.c
+++ b/qemud/qemud.c
@@ -766,8 +766,10 @@ static int qemudInitPaths(struct qemud_server *server,
goto snprintf_error;
}
- if (snprintf(server->logDir, PATH_MAX, "%s/.libvirt/log",
- dir_prefix) >= PATH_MAX)
+ if (snprintf(server->logDir, sizeof (server->logDir),
+ (uid == SYSTEM_UID ? "%s/log/libvirt" :
"%s/.libvirt/log"),
+ (uid == SYSTEM_UID ? LOCAL_STATE_DIR : dir_prefix))
+ >= sizeof (server->logDir))
goto snprintf_error;
ret = 0;