ParseTzFile doesn't FreeFile on error

Started by Kyotaro Horiguchialmost 4 years ago5 messageshackers
Jump to latest
#1Kyotaro Horiguchi
horikyota.ntt@gmail.com

While working on some patch, I saw the following error message when a
transaction ended successfully after a failed call to
parse_and_validate_value().

The cause is ParseTzFile() returns leaving an open file descriptor
unfreed in some error cases.

This happens only in a special case when the errors are ignored, but
in principle the file descriptor should be released before exiting the
function.

I'm not sure it's worth fixing but the attached fixes that.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center

Attachments:

0001-Fix-ParseTzFile-to-call-FreeFile-on-error.patchtext/x-patch; charset=us-asciiDownload+12-8
#2Tom Lane
tgl@sss.pgh.pa.us
In reply to: Kyotaro Horiguchi (#1)
Re: ParseTzFile doesn't FreeFile on error

Kyotaro Horiguchi <horikyota.ntt@gmail.com> writes:

The cause is ParseTzFile() returns leaving an open file descriptor
unfreed in some error cases.
This happens only in a special case when the errors are ignored, but
in principle the file descriptor should be released before exiting the
function.
I'm not sure it's worth fixing but the attached fixes that.

I agree this is worth fixing, but adding all these gotos seems a bit
inelegant. What do you think of the attached version?

BTW, my first thought about it was "what if one of the callees throws
elog(ERROR), eg palloc out-of-memory"? But I think that's all right
since then we'll reach transaction abort cleanup, which won't whine
about open files. The problem is limited to the case where no error
gets thrown.

regards, tom lane

Attachments:

0002-Fix-ParseTzFile-to-call-FreeFile-on-error.patchtext/x-diff; charset=us-ascii; name=0002-Fix-ParseTzFile-to-call-FreeFile-on-error.patchDownload+16-7
#3Kyotaro Horiguchi
horikyota.ntt@gmail.com
In reply to: Tom Lane (#2)
Re: ParseTzFile doesn't FreeFile on error

At Mon, 30 May 2022 13:11:04 -0400, Tom Lane <tgl@sss.pgh.pa.us> wrote in

Kyotaro Horiguchi <horikyota.ntt@gmail.com> writes:

The cause is ParseTzFile() returns leaving an open file descriptor
unfreed in some error cases.
This happens only in a special case when the errors are ignored, but
in principle the file descriptor should be released before exiting the
function.
I'm not sure it's worth fixing but the attached fixes that.

I agree this is worth fixing, but adding all these gotos seems a bit
inelegant. What do you think of the attached version?

It is what came up to me first. It is natural. So I'm fine with
it. The point of the "goto"s was that repeated "n = -1;break;" looked
somewhat noisy to me in the loop.

BTW, my first thought about it was "what if one of the callees throws
elog(ERROR), eg palloc out-of-memory"? But I think that's all right
since then we'll reach transaction abort cleanup, which won't whine
about open files. The problem is limited to the case where no error
gets thrown.

Right. This "issue" is not a problem unless the caller continues
without throwing an exception after the function errors out, which is
not done by the current code.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center

#4Tom Lane
tgl@sss.pgh.pa.us
In reply to: Kyotaro Horiguchi (#3)
Re: ParseTzFile doesn't FreeFile on error

Kyotaro Horiguchi <horikyota.ntt@gmail.com> writes:

At Mon, 30 May 2022 13:11:04 -0400, Tom Lane <tgl@sss.pgh.pa.us> wrote in

BTW, my first thought about it was "what if one of the callees throws
elog(ERROR), eg palloc out-of-memory"? But I think that's all right
since then we'll reach transaction abort cleanup, which won't whine
about open files. The problem is limited to the case where no error
gets thrown.

Right. This "issue" is not a problem unless the caller continues
without throwing an exception after the function errors out, which is
not done by the current code.

Actually the problem *is* reachable, if you intentionally break the
already-active timezone abbreviation file: newly started sessions
produce file-leak warnings after failing to apply the setting.
I concede that's not a likely scenario, but that's why I think it's
worth fixing.

regards, tom lane

#5Kyotaro Horiguchi
horikyota.ntt@gmail.com
In reply to: Tom Lane (#4)
Re: ParseTzFile doesn't FreeFile on error

At Tue, 31 May 2022 14:21:28 -0400, Tom Lane <tgl@sss.pgh.pa.us> wrote in

Actually the problem *is* reachable, if you intentionally break the
already-active timezone abbreviation file: newly started sessions
produce file-leak warnings after failing to apply the setting.
I concede that's not a likely scenario, but that's why I think it's
worth fixing.

Ah, I see. Thanks!

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center