I just tried to upgrade my tor exit node to 0.2.4.16-rc, and got this:
Aug 22 09:55:09.000 [warn] Something tried to close an or_connection_t without going through channels at src/or/connection.c:3185 Aug 22 09:55:10.000 [warn] Something tried to close an or_connection_t without going through channels at src/or/connection.c:3185 Aug 22 09:55:12.000 [warn] Something tried to close an or_connection_t without going through channels at src/or/connection.c:3185 Aug 22 09:55:12.000 [warn] Something tried to close an or_connection_t without going through channels at src/or/connection.c:3185 Aug 22 09:55:12.000 [warn] Something tried to close an or_connection_t without going through channels at src/or/connection.c:3185 Aug 22 09:55:16.000 [warn] Something tried to close an or_connection_t without going through channels at src/or/connection.c:3185 Aug 22 09:55:17.000 [warn] Something tried to close an or_connection_t without going through channels at src/or/connection.c:3185 Segmentation fault
The configure line was:
./configure --with-openssl-dir=../openssl-1.0.1e --enable-bufferevents
Running under gdb, I see:
Program received signal SIGSEGV, Segmentation fault. buf_datalen (buf=0x0) at src/or/buffers.c:523 523 return buf->datalen;
(gdb) bt #0 buf_datalen (buf=0x0) at src/or/buffers.c:523 #1 0x0000555555611b89 in connection_write_to_buf_impl_ ( string=0x7fffffffdd20 "\026\003\001", len=168, conn=0x55555682d070, zlib=0) at src/or/connection.c:3598 #2 0x0000555555587779 in connection_write_to_buf (conn=0x55555682d070, len=<optimized out>, string=<optimized out>) at src/or/connection.h:136 #3 connection_edge_process_relay_cell (cell=0x7fffffffdd10, circ=0x555557282610, conn=0x55555682d070, layer_hint=<optimized out>) at src/or/relay.c:1340 #4 0x0000555555589391 in circuit_receive_relay_cell (cell=0x7fffffffdd10, circ=0x555557282610, cell_direction=CELL_DIRECTION_OUT) at src/or/relay.c:212 #5 0x00005555555fa57c in command_process_relay_cell (chan=0x5555575b9f40, cell=0x7fffffffdd10) at src/or/command.c:462 #6 command_process_cell (chan=0x5555575b9f40, cell=0x7fffffffdd10) at src/or/command.c:148 #7 0x00005555555db873 in channel_tls_handle_cell (cell=0x7fffffffdd10, conn=0x555557ccc140) at src/or/channeltls.c:923 #8 0x000055555561c3f9 in connection_or_process_cells_from_inbuf ( conn=0x555557ccc140) at src/or/connection_or.c:1972 #9 0x000055555560f965 in connection_handle_read_cb (bufev=<optimized out>, arg=0x555557ccc140) at src/or/connection.c:3114 #10 0x00007ffff749457e in ?? () from /usr/lib/libevent-2.0.so.5 ---Type <return> to continue, or q <return> to quit--- #11 0x00007ffff748ae47 in event_base_loop () from /usr/lib/libevent-2.0.so.5 #12 0x000055555556b1fe in do_main_loop () at src/or/main.c:1987 #13 0x000055555556c9cf in tor_main (argc=<optimized out>, argv=0x7fffffffe3c8) at src/or/main.c:2703 #14 0x00007ffff689c76d in __libc_start_main () from /lib/x86_64-linux-gnu/libc.so.6 #15 0x0000555555566cdd in _start ()
I rebuilt without bufferevents, and it hasn't crashed yet. (I also don't see the "Something tried to close an or_connection_t" warnings.)
- Ian
On Thu, Aug 22, 2013 at 10:17 AM, Ian Goldberg iang@cs.uwaterloo.ca wrote:
I just tried to upgrade my tor exit node to 0.2.4.16-rc, and got this:
I rebuilt without bufferevents, and it hasn't crashed yet. (I also don't see the "Something tried to close an or_connection_t" warnings.)
Seems likely. This is a dup of bug https://trac.torproject.org/projects/tor/ticket/7734 .
When you started Tor, did you get this warning?
log_warn(LD_GENERAL, "Tor was compiled with the --enable-bufferevents " "option. This is still experimental, and might cause strange " "bugs. If you want a more stable Tor, be sure to build without " "--enable-bufferevents.");
(The bufferevent backend is not preferred in 0.2.3 or 0.2.4. Maybe 0.2.5 is where somebody hammers all the bugs out of it.)
On Thu, Aug 22, 2013 at 10:25:00AM -0400, Nick Mathewson wrote:
On Thu, Aug 22, 2013 at 10:17 AM, Ian Goldberg iang@cs.uwaterloo.ca wrote:
I just tried to upgrade my tor exit node to 0.2.4.16-rc, and got this:
I rebuilt without bufferevents, and it hasn't crashed yet. (I also don't see the "Something tried to close an or_connection_t" warnings.)
Seems likely. This is a dup of bug https://trac.torproject.org/projects/tor/ticket/7734 .
Indeed. (Though that bug doesn't report a segfault.)
When you started Tor, did you get this warning?
log_warn(LD_GENERAL, "Tor was compiled with the --enable-bufferevents " "option. This is still experimental, and might cause strange " "bugs. If you want a more stable Tor, be sure to build without " "--enable-bufferevents.");
I didn't notice it, but things scrolled by quickly.
(The bufferevent backend is not preferred in 0.2.3 or 0.2.4. Maybe 0.2.5 is where somebody hammers all the bugs out of it.)
Ah, I had been under the mistaken impression it was good in 0.2.4. No worries.
Thanks,
- Ian