[libvirt] [PATCH] avoid a probable EINVAL from lseek
Jim Meyering
jim at meyering.net
Tue Feb 2 09:36:08 UTC 2010
Daniel Veillard wrote:
> On Mon, Feb 01, 2010 at 10:18:27PM +0100, Jim Meyering wrote:
>>
>> In src/qemu/qemu_driver.c, coverity reports this:
>>
>> Event negative_return_fn: Called negative-returning function "lseek(logfile, 0L, 2)"
>> Event var_assign: NEGATIVE return value of "lseek" assigned to signed variable "pos"
>> At conditional (1): "(pos = lseek(logfile, 0L, 2)) < 0" taking true path
>> 2877 if ((pos = lseek(logfile, 0, SEEK_END)) < 0)
>> 2878 VIR_WARN(_("Unable to seek to end of logfile: %s"),
>> 2879 virStrerror(errno, ebuf, sizeof ebuf));
>>
>> since later in that same function, a negative "pos" may
>> be used like this:
>>
>> Event negative_returns: Tracked variable "pos" was passed to a negative sink. [details]
>> 2930 if (qemudWaitForMonitor(conn, driver, vm, pos) < 0)
>> 2931 goto abort;
>> 2932
>>
>> which is a legitimate problem, since
>> qemudWaitForMonitor calls qemudLogReadFD, which calls lseek
>> with that same "pos" value:
>>
>> Event neg_sink_parm_call: Parameter "pos" passed to negative sink "lseek"
>> 560 if (lseek(fd, pos, SEEK_SET) < 0) {
>> 561 virReportSystemError(conn, errno,
>> 562 _("Unable to seek to %lld in %s"),
>> 563 (long long) pos, logfile);
>> 564 close(fd);
>> 565 }
>>
>>
>> One approach is to detect the negative offset in that final bit
>> of code and skip the lseek:
>>
>> >From 0ef617935462c314ed0b44bcaa3dd5bf58ccbc1b Mon Sep 17 00:00:00 2001
>> From: Jim Meyering <meyering at redhat.com>
>> Date: Mon, 1 Feb 2010 22:17:44 +0100
>> Subject: [PATCH] avoid a probable EINVAL from lseek
>>
>> * src/qemu/qemu_driver.c (qemudLogReadFD): Don't pass a negative
>> offset (from a preceding failed attempt to seek to EOF) to this use
>> of lseek.
>> ---
>> src/qemu/qemu_driver.c | 4 ++--
>> 1 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c
>> index 22593bf..676a27b 100644
>> --- a/src/qemu/qemu_driver.c
>> +++ b/src/qemu/qemu_driver.c
>> @@ -558,8 +558,8 @@ qemudLogReadFD(virConnectPtr conn, const char* logDir, const char* name, off_t p
>> close(fd);
>> return -1;
>> }
>> - if (lseek(fd, pos, SEEK_SET) < 0) {
>> - virReportSystemError(conn, errno,
>> + if (pos < 0 || lseek(fd, pos, SEEK_SET) < 0) {
>> + virReportSystemError(conn, pos < 0 ? 0 : errno,
>> _("Unable to seek to %lld in %s"),
>> (long long) pos, logfile);
>> close(fd);
>
> I was wondering if it wasn't simpler to abort earlier on when
> pos < 0 was returned from lseek, but after rereading the code I
> agree with your patch,
Returning early (failing) after the initial log-lseek failure
did not seem justified, considering other log-related failures
merely get a warning there.
Pushed.
More information about the libvir-list
mailing list