Hello,
There appears to be a bug in the code used by DBD::Pg's 'lo_read'
function. I presume the code with the bug is a part of libpq, and
thus the postgresql-libs RPM binary distribution.
The nature of the bug is that an 'lo_read' operation performed
with DBD::Pg caused a segfault with postgresql-libs-7.3.2 and
"hangs" on files >= 32768 bytes with postgresql-libs-7.3.4. The
hang is actually a read() loop on the socket generating EAGAIN
error on each read().
I've attached a short perl script that demonstrates the bug.
The database server does not seem to matter; I get the same results
using 7.2 as I do with 7.3.x servers. I also get the same results
when I vary the perl-DBD-Pg and perl-DBI module versions. The bug
seems to be following libpg.
All of my testing has been on the i686 platform using linux 2.4.20
kernels.
Test case #1:
- client = postgresql-libs-7.2.3-5 (redhat-7.3)
perl-DBI-1.21-1 (redhat-7.3)
perl-DBD-Pg-1.01-8 (redhat-7.3)
perl-5.6.1
- server = postgresql-server-7.3.2-3 (redhat-9)
- transport = tcpip + ssl
$ perl test.pl
- reading testfile '/bin/ls'
- inserting testfile contents
- oid = '16265435'
- reading large object
success.. removing large object
Test case #2:
- client = postgresql-libs-7.3.2-3 (redhat-9)
perl-DBI-1.32-5 (redhat-9)
perl-DBD-Pg-1.21-2 (redhat-9)
-server = postgresql-server-7.3.2-3 (redhat-9)
-transport = tcpip + ssl
$ perl test.pl
- reading testfile '/bin/ls'
- inserting testfile contents
- oid = '16265436'
- reading large object
Segmentation fault
... after ~32768 bytes, process loops reading the socket
read(3, 0x81cc938, 5) = -1 EAGAIN (Resource temporarily unavailable)
and eventually segfaults
Test case #3:
- same as #2 except
- transport = tcpip (localhost)
$ perl test.pl
- reading testfile '/bin/ls'
- inserting testfile contents
- oid = '16265436'
- reading large object
Segmentation fault
Test case #4:
- same as #2 and #3 except
- postgresql-libs-7.3.4 PGDG used
$ perl test.pl
- reading testfile '/bin/ls'
- inserting testfile contents
- oid = '16265437'
- reading large object
... after ~32768 bytes, process loops reading the socket
read(3, 0x81cc938, 5) = -1 EAGAIN (Resource temporarily unavailable)
but doesn't appear to segfault (unless I am not patient enough)
This problem is a bastard as it makes getting data out of an
otherwise perfectly good database rather difficult. In other
words, its a show-stopper when migrating from 7.2 -> 7.3.
Regards,
Kevin
