[PATCH] pdns_server crashes on SIGSTOP+SIGCONT, or under gdb

[anon]

(This was reported to Fedora EPEL as  RHBZ#652841.)

With pdns-2.9.22-3.el5 on CentOS 5 x86_64, sending SIGSTOP and SIGCONT to a running pdns_server (for example, by launching it from the shell, pressing Control-Z, and typing fg) reliably causes it to immediately segfault:

# yum -y install pdns
# echo 'launch=bind' >> /etc/pdns/pdns.conf
# pdns_server
Nov 12 22:23:12 Reading random entropy from '/dev/urandom'
Nov 12 22:23:12 This is a standalone pdns
Nov 12 22:23:12 Listening on controlsocket in '/var/run/pdns.controlsocket'
Nov 12 22:23:12 It is advised to bind to explicit addresses with the --local-address option
Nov 12 22:23:12 UDP server bound to 0.0.0.0:53
Nov 12 22:23:12 TCP server bound to 0.0.0.0:53
Nov 12 22:23:12 PowerDNS 2.9.22 (C) 2001-2009 PowerDNS.COM BV (Jan 26 2009, 12:01:27, gcc 4.1.2 20080704 (Red Hat 4.1.2-44)) starting up
Nov 12 22:23:12 PowerDNS comes with ABSOLUTELY NO WARRANTY. This is free software, and you are welcome to redistribute it according to the terms of the GPL version 2.
Nov 12 22:23:12 Creating backend connection for TCP
Nov 12 22:23:12 About to create 3 backend threads for UDP
Nov 12 22:23:12 Done launching threads, ready to distribute questions
^Z
[1]+  Stopped                 pdns_server
# fg
pdns_server
Nov 12 22:23:13 Got a signal 11, attempting to print trace: 
Nov 12 22:23:13 pdns_server [0x485ce6]
Nov 12 22:23:13 /lib64/libc.so.6 [0x2b0ec242e2d0]
Nov 12 22:23:13 pdns_server(_ZN11DistributorI9DNSPacketS0_13PacketHandlerE10makeThreadEPv+0x12c) [0x4c788c]
Nov 12 22:23:13 /lib64/libpthread.so.0 [0x2b0ec21e873d]
Nov 12 22:23:13 /lib64/libc.so.6(clone+0x6d) [0x2b0ec24d1f6d]
Aborted

Attempting to run pdns_server under gdb also causes it to immediately segfault (probably because SIGSTOP is involved in the implementation of ptrace):

(gdb) run
Starting program: /usr/sbin/pdns_server 
[Thread debugging using libthread_db enabled]
Nov 12 22:43:14 Reading random entropy from '/dev/urandom'
Nov 12 22:43:14 This is a standalone pdns
Nov 12 22:43:14 Listening on controlsocket in '/var/run/pdns.controlsocket'
Nov 12 22:43:14 It is advised to bind to explicit addresses with the --local-address option
Nov 12 22:43:14 UDP server bound to 0.0.0.0:53
Nov 12 22:43:14 TCP server bound to 0.0.0.0:53
Nov 12 22:43:14 PowerDNS 2.9.22 (C) 2001-2009 PowerDNS.COM BV (Jan 26 2009, 12:01:27, gcc 4.1.2 20080704 (Red Hat 4.1.2-44)) starting up
Nov 12 22:43:14 PowerDNS comes with ABSOLUTELY NO WARRANTY. This is free software, and you are welcome to redistribute it according to the terms of the GPL version 2.
[New Thread 0x40a00940 (LWP 2335)]
Nov 12 22:43:14 Creating backend connection for TCP
[New Thread 0x41401940 (LWP 2336)]
Nov 12 22:43:14 About to create 3 backend threads for UDP
[New Thread 0x41e02940 (LWP 2337)]
[New Thread 0x42803940 (LWP 2338)]

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x41e02940 (LWP 2337)]
0x00002aaaab197672 in __gnu_cxx::__exchange_and_add (__mem=0xfffffffffffffff8, 
    __val=-1) at atomicity.cc:41
41				  : "0" (__val), "m" (*__mem));
(gdb) bt full
#0  0x00002aaaab197672 in __gnu_cxx::__exchange_and_add (
    __mem=0xfffffffffffffff8, __val=-1) at atomicity.cc:41
        __result = -1
#1  0x00000000004c7c1e in Distributor<DNSPacket, DNSPacket, PacketHandler>::makeThread (p=0x791830) at /usr/include/c++/4.1.2/bits/basic_string.h:232
        b = 0x792150
        qcount = 0
        queuetimeout = 1500
#2  0x00002aaaab87873d in start_thread (arg=<value optimized out>)
    at pthread_create.c:301
        __res = <value optimized out>
        pd = 0x41e02940
        unwind_buf = {cancel_jmp_buf = {{jmp_buf = {1105209664, 
                197107770085693007, 10485760, 3, 1105211392, 4096, 
                197107769030834047, 197154333293990220}, mask_was_saved = 0}}, 
          priv = {pad = {0x0, 0x0, 0x0, 0x0}, data = {prev = 0x0, 
              cleanup = 0x0, canceltype = 0}}}
        not_first_call = 0
        robust = <value optimized out>
#3  0x00002aaaabb61f6d in clone () from /lib64/libc.so.6
No symbol table info available.

[anon]

Handle EINTR from sem_wait()

[anon]

The problem seems to be that Semaphore::wait() does not handle the case of sem_wait() returning EINTR. I’ve attached a patch to fix this.

The callers of Semaphore::wait() should probably also be updated to check the return value and handle other unexpected errors, instead of blindly assuming that the semaphore was successfully locked.

(Er, by the way, since this Trac shows everyone as ‘anon’: I am Anders Kaseorg <andersk@…>.)

[ahu]

Fixed in commit 1747