Subject: Re: pgsql: Validate page level checksums in base backups
To: Magnus Hagander <magnus@hagander.net>,
 Michael Banck <michael.banck@credativ.de>
Cc: Alvaro Herrera <alvherre@alvh.no-ip.org>, Tom Lane <tgl@sss.pgh.pa.us>,
 Andres Freund <andres@anarazel.de>, David Steele <david@pgmasters.net>,
 PostgreSQL-development <pgsql-hackers@postgresql.org>
References: <20180404121930.GE20852@nighthawk.caipicrew.dd-dns.de>
 <20180404134732.tsxx2osvooolxy24@alvherre.pgsql>
 <CABUevEwrCF7TzAyD=4WPfTyzH+5xaFYf+-=ATW58XLi7b4dURA@mail.gmail.com>
 <20180404182247.GG20852@nighthawk.caipicrew.dd-dns.de>
 <CABUevExp8BOo3ZjEXM5xkz5P0zrjZT4_eM7QM4DLyPorDVUQUQ@mail.gmail.com>
 <20180405124113.GH20852@nighthawk.caipicrew.dd-dns.de>
 <CABUevEzZ55qgLVtMxW=hXhzjHV3oU-gChpeKtufyf_siL-rBWg@mail.gmail.com>
From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Message-ID: <f23e92ec-118d-e6ea-0c81-876ad62a588a@2ndquadrant.com>
Date: Tue, 10 Apr 2018 23:24:35 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.4.0
MIME-Version: 1.0
In-Reply-To: 
 <CABUevEzZ55qgLVtMxW=hXhzjHV3oU-gChpeKtufyf_siL-rBWg@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi,

I think there's a bug in sendFile(). We do check checksums on all pages
that pass this LSN check:

    /*
     * Only check pages which have not been modified since the
     * start of the base backup. Otherwise, they might have been
     * written only halfway and the checksum would not be valid.
     * However, replaying WAL would reinstate the correct page in
     * this case.
     */
    if (PageGetLSN(page) < startptr)
    {
        ...
    }

Now, imagine the page is new, i.e. all-zeroes. That means the LSN is 0/0
too, and we'll try to verify the checksum - but we actually do not set
checksums on empty pages.

So I think it should be something like this:

    if ((!PageIsNew(page)) && (PageGetLSN(page) < startptr))
    {
        ...
    }

It might be worth verifying that the page is actually all-zeroes (and
not just with corrupted pd_upper value. Not sure it's worth it.

I've found this by fairly trivial stress testing - running pgbench and
pg_basebackup in a loop. It was failing pretty reliably (~75% of runs).
With the proposed change I see no further failures.

regards

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services