segfault when trying to issue a command to closed Unix domain socket

[mirkperl]

When using the redisConnectUnix() function to connect to redis via a Unix domain socket, if redis-server exits and you try to issue a command using the context returned from redisConnectUnix() you get a SIGPIPE which is resulting in a segfault, crashing the entire application that uses hiredis.

Here's a backtrace from a simple program that makes a connection to hiredis, sleeps for 5 seconds during which I kill redis in another terminal, then continues to try to issue a DEL command.

(gdb) run
sleeping
done sleeping
Program received signal SIGPIPE, Broken pipe.
0xffffe424 in __kernel_vsyscall ()
(gdb) bt
#0 0xffffe424 in __kernel_vsyscall ()
#1 0xb7682a23 in write () from /lib/libc.so.6
#2 0xb7707ea6 in redisBufferWrite () from /usr/lib/libhiredis.so.0.10
#3 0xb7709a1f in redisGetReply () from /usr/lib/libhiredis.so.0.10
#4 0xb7709b0f in redisvCommand () from /usr/lib/libhiredis.so.0.10
#5 0xb7709b4b in redisCommand () from /usr/lib/libhiredis.so.0.10
#6 0x0804863e in main (argc=2, argv=0xbfab6a54) at t.c:25

[pietern]

It's up to the application using hiredis to register a handler for, or ignore, SIGPIPE.

This line shows how to ignore SIGPIPE:

hiredis/example-libev.c

Line 35 in 0625e7c

signal(SIGPIPE, SIG_IGN);