SegFault in "nn_usock_handler" at src/aio/usock_posix.inc : Line 601

[Astra580]

We are facing segfault issue in nanomsg library. We use PUB/SUB mechanism for communication.
Following nanomsg thread routine is causing the crash.

[Backtrace]
#0 0x00007efd56091a2b in nn_usock_handler (self=0x56094724f870, src=1, type=1, srcptr=) at /usr/src/debug/git/nanomsg/src/aio/usock_posix.inc:601
usock = 0x56094724f870
s =
rc =
sz = 9
srcptr =
type = 1
src = 1
self = 0x56094724f870
usock = 0x56094724f870
#1 0x00007efd5607709e in nn_worker_routine (arg=0x7efd562c06e0 <self+64>) at /usr/src/debug/git/tnanomsg/src/aio/worker_posix.inc:249
rc =
self = 0x7efd562c06e0 <self+64>
pevent = 1
phndl = 0x56094724f8f0
thndl = 0x560947506368
tasks = {head = 0x0, tail = 0x0}
item =
task =
fd = 0x56094724f8e0
timer =
#2 0x00007efd56078ccd in nn_thread_main_routine (arg=) at /usr/src/debug/git/nanomsg/src/utils/thread_posix.inc:35
self =
#3 0x00007efd55144477 in start_thread (arg=0x7efd4e9cf700) at /usr/src/debug/glibc/2.27-r0/git/nptl/pthread_create.c:463
pd = 0x7efd4e9cf700
now =
unwind_buf = {cancel_jmp_buf = {{jmp_buf = {139626410735360, -9064765947889294268, 139626582381214, 139626582381215, 8396800, 139626582381216, 9208144196267959364, 9208121042860401732}, mask_was_saved = 0}}, priv = {pad = {0x0, 0x0, 0x0, 0x0}, data = {prev = 0x0, cleanup = 0x0, canceltype = 0}}}
not_first_call =
#4 0x00007efd54affd9f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
No locals.

Can anyone suggest if any patch available for this issue?

[gdamore]

This doesn't look familiar to me. I don't see anything specific in this crash that helps me understand it either.

I'd recommend trying out NNG if you're able to.

If you're still working on this, can you provide details about what the git version you're using is? I see you are using Linux. Also was there anything else you have about activity? If you have a reproduction case that is small, that would help.

[Astra580]

Currently we are using very old nanomsg version with git nanomsg master commit : 327a29b

We dont want to update directly to latest version available. Below is the scenario of crash.

We are using pub-sub mechanism for communication between process.

At Receiving end we are performing below operation :

Initial socket creation for subscription

-> nn_socket(AF_SP, NN_SUB)
-> nn_setsockopt(NNSocket, NN_SUB, NN_SUB_SUBSCRIBE, "", 0)
-> nn_connect(NNSocket, "URL_STRING")

When data is available to read

-> #define POS_MAX_BUFF_SIZE 8192
-> void* buff = nn_allocmsg(POS_MAX_BUFF_SIZE, 0)
-> nn_recv (fd, (uint8_t *)buff, POS_MAX_BUFFER_SIZE, NN_DONTWAIT)

As per our understanding crash is happening when we call nn_recv().
Unfortunately we could not reproduce the issue again. Crash is observed only once until now.

[gdamore]

That is an extremely old version version of libnanomsg. Please update and let me know if you still have problems.

[gdamore]

Essentially, I'm not willing to spend effort trying to diagnose a problem because you don't want to update to something newer. This project is already in sustaining mode, and looking back at 8 year old versions of the code is just not something that is practical normally.

Having said all that, if you want to get support for that ancient version please contact info@staysail.tech and we can discuss an hourly rate.

[gdamore]

Closing this for now. If you test a newer version and find a problem, let me know. If you want to proceed with this old version please reach out for commercial support options.