Thanks Richard!
The patches are similar to what we did locally, and applying them fixed the issue with the actual use-case (not just the sanitized repro).

On Tue, Feb 14, 2023 at 8:10 PM Richard W.M. Jones <rjones@redhat.com> wrote:
On Tue, Feb 14, 2023 at 08:02:59PM +0200, Yonatan Shtarkman wrote:
> The attached code snippet reproduces the issue when running:
> python3 libugestfs_segfault_repro.py
>
> With:
> guestfs version: 1.48.6
> guestfs-python version: 1.48.6

Can confirm with {python3-,}libguestfs-1.51.1-1.fc39.x86_64

The stack trace is roughly the same as yours.

Rich.

> Stacktrace:
> #0  0x00007f6dd17b4701 in _Py_INCREF (op=<optimized out>) at /usr/include/
> python3.9/object.h:408
> 408    op->ob_refcnt++;
> (gdb) bt
> #0  0x00007f6dd17b4701 in _Py_INCREF (op=<optimized out>) at /usr/include/
> python3.9/object.h:408
> #1  guestfs_int_py_event_callback_wrapper
>     (g=<optimized out>, flags=<optimized out>, array_len=0, array=0x0, buf_len=
> 43, buf=0xd1dc90 "commandrvf: udevadm --debug settle -E /\347\232\204\303
> available\r\ned7
> #2  guestfs_int_py_event_callback_wrapper
>     (g=<optimized out>, callback=0x7f6dd19ec040, event=16, event_handle=0,
> flags=<optimized out>, buf=0xd1dc90 "commandrvf: udevadm --debug settle -E /\
> 347\232\204\303 4
> #3  0x00007f6dd166f9aa in guestfs_int_call_callbacks_message
>     (g=0xae06f0, event=16, buf=0xd1dc90 "commandrvf: udevadm --debug settle -E
> /\347\232\204\303 available\r\ned to rvf:e run  in, ignor data av", buf_len=43)
>     at events.c:117
> #4  0x00007f6dd168560e in guestfs_int_log_message_callback
>     (g=g@entry=0xae06f0, buf=0xd1dc90 "commandrvf: udevadm --debug settle -E /\
> 347\232\204\303 available\r\ned to rvf:e run  in, ignor data av", len=len@entry
> =43)
>     at proto.c:145
> #5  0x00007f6dd166ac29 in handle_log_message (g=g@entry=0xae06f0, conn=
> conn@entry=0xd19570) at conn-socket.c:395
> #6  0x00007f6dd166b233 in read_data (len=4, bufv=<optimized out>, connv=
> <optimized out>, g=<optimized out>) at conn-socket.c:179
> #7  read_data (g=0xae06f0, connv=0xd19570, bufv=<optimized out>, len=4) at
> conn-socket.c:142
> #8  0x00007f6dd1685c2a in recv_from_daemon (buf_rtn=0x7fffa95cc548, size_rtn=
> 0x7fffa95cc544, g=0xae06f0) at proto.c:545
> #9  guestfs_int_recv_from_daemon (g=g@entry=0xae06f0, size_rtn=size_rtn@entry=
> 0x7fffa95cc544, buf_rtn=buf_rtn@entry=0x7fffa95cc548) at proto.c:623
> #10 0x00007f6dd168603a in guestfs_int_recv
>     (g=g@entry=0xae06f0, fn=fn@entry=0x7f6dd16aa308 "download", hdr=hdr@entry=
> 0x7fffa95cc610, err=err@entry=0x7fffa95cc5e0, xdrp=xdrp@entry=0x0, ret=
> ret@entry=0x0)
>     at proto.c:668
> #11 0x00007f6dd163825e in guestfs_download (g=g@entry=0xae06f0, remotefilename=
> <optimized out>, filename=0x7f6dd17edea0 "/tmp/1") at actions-5.c:2984
> #12 0x00007f6dd17ab01a in guestfs_int_py_download (self=<optimized out>, args=
> <optimized out>) at actions-5.c:794
>  
>
>
> On Mon, Feb 13, 2023 at 10:16 PM Richard W.M. Jones <rjones@redhat.com> wrote:
>
>     On Mon, Feb 13, 2023 at 09:38:50PM +0200, Yonatan Shtarkman wrote:
>     > Also, as a workaround, I avoided calling the event callback if null is
>     returned
>     > by Py_BuildValue (still print the error and release the thread).
>     > This seems to work for our usage (we only use the event callbacks for
>     logging),
>     > any potential issues I'm missing?
>
>     Oh I see so the issue is actually when logging and not in the download
>     call?  That's strange that it should be happening intermittently.  I'd
>     expect a UTF-8 encoding error would happen every time not sometimes.
>
>     Rich.
>
>     --
>     Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/
>     ~rjones
>     Read my programming and virtualization blog: http://rwmj.wordpress.com
>     virt-top is 'top' for virtual machines.  Tiny program with many
>     powerful monitoring features, net stats, disk stats, logging, etc.
>     http://people.redhat.com/~rjones/virt-top
>
>

> import guestfs
> from guestfs import GuestFS
>
> g = GuestFS(python_return_dict=True)
> def callback(event, _event_handle, buf: str, array):
>       pass
>       
> g.set_event_callback(callback, guestfs.EVENT_ALL)
> g.set_verbose(True)
> path = '/tmp/tmp_disk'
> g.disk_create(path, format='raw', size=32*1024*1024)
> g.add_drive(path)
> g.launch()
> g.mkfs('ext4', '/dev/sda')
> g.mount('/dev/sda', '/')
> for _ in range(20000):
>     try:
>        g.download ('/的ó', '/tmp/1')
>     except:
>        pass
>     


--
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html