From: Tom Lane <tgl@sss.pgh.pa.us>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
cc: Justin Pryzby <pryzby@telsasoft.com>, pgsql-hackers@postgresql.org,
        Kunshchikov Vladimir <Vladimir.Kunshchikov@infotecs.ru>
Subject: Re: pg_dump, gzwrite, and errno
In-reply-to: <20200618212027.GA17103@alvherre.pgsql>
References: <20200618212027.GA17103@alvherre.pgsql>
Comments: In-reply-to Alvaro Herrera <alvherre@2ndquadrant.com>
	message dated "Thu, 18 Jun 2020 17:20:27 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1597084.1592515818.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Thu, 18 Jun 2020 17:30:18 -0400
Message-ID: <1597085.1592515818@sss.pgh.pa.us>
Precedence: bulk

Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2020-Jun-11, Justin Pryzby wrote:
>> --- a/src/bin/pg_dump/pg_backup_directory.c
>> +++ b/src/bin/pg_dump/pg_backup_directory.c
>> @@ -347,8 +347,12 @@ _WriteData(ArchiveHandle *AH, const void *data, si=
ze_t dLen)
>>         lclContext *ctx =3D (lclContext *) AH->formatData;
>>  =

>>         if (dLen > 0 && cfwrite(data, dLen, ctx->dataFH) !=3D dLen)
>> +       {
>> +               if (errno =3D=3D 0)
>> +                       errno =3D ENOSPC;
>>                 fatal("could not write to output file: %s",
>>                           get_cfp_error(ctx->dataFH));
>> +       }
>>  }

> This seems correct to me.

Surely it's insufficient as-is, because there is no reason to suppose
that errno is zero at entry.  You'd need to set errno =3D 0 first.

Also it's fairly customary in our sources to include a comment about
this machination; so the full ritual is usually more like

    errno =3D 0;
    if (pg_pwrite(fd, data, len, xlrec->offset) !=3D len)
    {
        /* if write didn't set errno, assume problem is no disk space */
        if (errno =3D=3D 0)
            errno =3D ENOSPC;
        ereport ...

> (I spent a long time looking at zlib sources
> to convince myself that it does work with compressed files too)

Yeah, it's not obvious that gzwrite has the same behavior w.r.t. errno
as a plain write.  But there's not much we can do to improve matters
if it does not, so we might as well assume it does.

			regards, tom lane