MIME-Version: 1.0
References: <CAGOe9RiRUK9K8gUbsMfg8nWDsM2Fd9py-2oe4VG1Uaggu8fQGA@mail.gmail.com>
 <13e3100fc7c7d14919c37943dcfd76b263cecce2.camel@cybertec.at>
 <609925.1752502040@sss.pgh.pa.us> <20250714182016.23ypyxgtq6vbgl4c@hjp.at> <708241.1752517845@sss.pgh.pa.us>
In-Reply-To: <708241.1752517845@sss.pgh.pa.us>
From: Benjamin Wang <benjamin.ahrtr@gmail.com>
Date: Mon, 14 Jul 2025 20:02:27 +0100
Message-ID: <CADVOZZ5D7bR_2fcqonVc5CE6wkwDb1smzZf+V7MVspw+1NKhRQ@mail.gmail.com>
Subject: Re: Bypassing Directory Ownership Check in PostgreSQL 16.6 with
 Secure z/OS NFS (AT-TLS)
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: "Peter J. Holzer" <hjp-pgsql@hjp.at>, pgsql-general@lists.postgresql.org
Content-Type: multipart/alternative; boundary="000000000000b279170639e84dab"
Archived-At: <https://www.postgresql.org/message-id/CADVOZZ5D7bR_2fcqonVc5CE6wkwDb1smzZf%2BV7MVspw%2B1NKhRQ%40mail.gmail.com>
Precedence: bulk

--000000000000b279170639e84dab
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

I am not sure whether PostgreSQL depends on system call `fsyncdata` to
 sync data to disk. If yes, then I don't think it's safe to use NFS.
When `fsyncdata` returns success, it doesn't mean the data has
really been synced to disk. But if PostgreSQL crashes right after
it returns success to clients. Eventually it breaks the D (Durability) of
ACID.

Benjamin

On Mon, Jul 14, 2025 at 7:31=E2=80=AFPM Tom Lane <tgl@sss.pgh.pa.us> wrote:

> "Peter J. Holzer" <hjp-pgsql@hjp.at> writes:
> > On 2025-07-14 10:07:20 -0400, Tom Lane wrote:
> >> That is primarily for safety reasons: if for some reason the
> >> filesystem gets dismounted, or hasn't come on-line yet during
> >> a reboot, you do not want Postgres to be able to write on the
> >> underlying mount-point directory.
>
> > Be careful: There are two different directorys involved in a mount
> > point. The one in the parent filesystem and the one in the mounted file
> > system.
>
> True, and the safety requirement really is only that the parent
> filesystem's mount-point directory not be writable by us.
> But normal practice is that both directories are root-owned,
> or at least owned by highly privileged users.
>
> (I have a vague idea that there are system-level security hazards,
> not specific to Postgres, if mount-point directories are publicly
> writable.  Don't feel like researching that though.)
>
>                         regards, tom lane
>
>
>

--000000000000b279170639e84dab
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">I am not sure whether PostgreSQL depends on system call `f=
syncdata` to<div>=C2=A0sync data to disk. If yes, then I don&#39;t think it=
&#39;s safe to use NFS.=C2=A0</div><div>When `fsyncdata` returns success, i=
t doesn&#39;t mean the data has</div><div>really been synced to disk. But i=
f PostgreSQL crashes right after</div><div>it returns success to clients. E=
ventually it breaks the D (Durability) of</div><div>ACID.</div><div><br></d=
iv><div>Benjamin=C2=A0</div></div><br><div class=3D"gmail_quote gmail_quote=
_container"><div dir=3D"ltr" class=3D"gmail_attr">On Mon, Jul 14, 2025 at 7=
:31=E2=80=AFPM Tom Lane &lt;<a href=3D"mailto:tgl@sss.pgh.pa.us">tgl@sss.pg=
h.pa.us</a>&gt; wrote:<br></div><blockquote class=3D"gmail_quote" style=3D"=
margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-lef=
t:1ex">&quot;Peter J. Holzer&quot; &lt;<a href=3D"mailto:hjp-pgsql@hjp.at" =
target=3D"_blank">hjp-pgsql@hjp.at</a>&gt; writes:<br>
&gt; On 2025-07-14 10:07:20 -0400, Tom Lane wrote:<br>
&gt;&gt; That is primarily for safety reasons: if for some reason the<br>
&gt;&gt; filesystem gets dismounted, or hasn&#39;t come on-line yet during<=
br>
&gt;&gt; a reboot, you do not want Postgres to be able to write on the<br>
&gt;&gt; underlying mount-point directory.<br>
<br>
&gt; Be careful: There are two different directorys involved in a mount<br>
&gt; point. The one in the parent filesystem and the one in the mounted fil=
e<br>
&gt; system.<br>
<br>
True, and the safety requirement really is only that the parent<br>
filesystem&#39;s mount-point directory not be writable by us.<br>
But normal practice is that both directories are root-owned,<br>
or at least owned by highly privileged users.<br>
<br>
(I have a vague idea that there are system-level security hazards,<br>
not specific to Postgres, if mount-point directories are publicly<br>
writable.=C2=A0 Don&#39;t feel like researching that though.)<br>
<br>
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 regards, tom lane<br>
<br>
<br>
</blockquote></div>

--000000000000b279170639e84dab--