qemu locks up on typing 41 characters at once into serial console

Bug #1154328 reported by Andreas Gustafsson
8
This bug affects 1 person
Affects Status Importance Assigned to Milestone
QEMU
Fix Released
Undecided
Unassigned

Bug Description

I am running daily automated tests that involve booting a NetBSD 6.0.1
guest in qemu freshly built from git. The tests are scripted using
pexpect, which interacts with the NetBSD guest over the emulated
serial console. Recently, the tests stopped working; the guest boots
and pexpect is able to log in, but when it sends a long shell command
(more than 40 characters) to the guest, the command is neither echoed
nor executed, and no further output is seen from the guest.

The problem can be reproduced manually (without pexpect) as follows.

Run the following commands in a terminal window on a host of
your choice (Linux will do fine):

  wget http://www.gson.org/bugs/qemu/NetBSD-6.0.1-i386-live-wd0root-com0.img.gz
  gunzip NetBSD-6.0.1-i386-live-wd0root-com0.img.gz
  qemu-system-i386 -m 32 -nographic -snapshot -hda NetBSD-6.0.1-i386-live-wd0root-com0.img

This will download a disk image (some 144 MB compressed, 2 GB
uncompressed) containing a NetBSD system configured to use a serial
console, and boot it in qemu. Make sure the qemu-system-i386
in your PATH is one recently built from git, or adjust the command
as needed.

Once the VM has booted, log in as root (there is no password). You
will now be in a functional NetBSD root shell.

Now cut-and-paste a string containing at least 41 characters into the
terminal window. I used a string containing 41 copies of the letter
"X". You can use other strings, but beware of pasting strings
containing valid shell commands, as they may end up being executed on
the host (see below).

If your copy of qemu is suffering from the bug, it will lock up. Not
only will the virtual machine no longer respond to keystrokes, but
qemu itself will no longer respond to commands such as "control-a c".
You will have to kill it from a different terminal window. When the
qemu process is killed, any pasted characters after the first 40 will
be read and executed by the host shell, suggesting that they were never
even read by the qemu process. As I had typed a return after pasting
the 41 X:es, the host shell executed the command "X", thereby
accidentally attempting (unsuccessfully) to start an X server.

"git bisect" implicates the following commit:

  commit a29753f8aa79a34a324afebe340182a51a5aef11
  Author: Anthony Liguori <email address hidden>
  Date: Tue Mar 5 23:21:19 2013 +0530

      qemu-char: convert fd_chr to use a GIOChannel

         This uses the newly introduced IOWatchPoll source.

      Signed-off-by: Anthony Liguori <email address hidden>
      Signed-off-by: Amit Shah <email address hidden>
      Message-id: 0cb5d14510ee835a0ebc23676d10a2cce9280da5<email address hidden>
      Signed-off-by: Anthony Liguori <email address hidden>

Revision history for this message
Andreas Gustafsson (gson) wrote :

My automated test is still hanging, but since commit 893986fe94eb229f2317f50fac0e35e068eb66ba,
it no longer hangs silently, but instead outputs a seemingly endless stream of assertion failure messages:

(process:5928): GLib-CRITICAL **: g_source_get_context: assertion `!SOURCE_DESTROYED (source)' failed

(process:5928): GLib-CRITICAL **: g_source_get_context: assertion `!SOURCE_DESTROYED (source)' failed

(process:5928): GLib-CRITICAL **: g_source_get_context: assertion `!SOURCE_DESTROYED (source)' failed

(process:5928): GLib-CRITICAL **: g_source_attach: assertion `source->context == NULL' failed

(process:5928): GLib-CRITICAL **: g_source_get_context: assertion `!SOURCE_DESTROYED (source)' failed

(process:5928): GLib-CRITICAL **: g_source_attach: assertion `source->context == NULL' failed

(process:5928): GLib-CRITICAL **: g_source_get_context: assertion `!SOURCE_DESTROYED (source)' failed

(process:5928): GLib-CRITICAL **: g_source_attach: assertion `source->context == NULL' failed

(process:5928): GLib-CRITICAL **: g_source_get_context: assertion `!SOURCE_DESTROYED (source)' failed

Revision history for this message
Andreas Gustafsson (gson) wrote :

Testing git revision 47b5264eb3e1cd2825e48d28fd0d1b239ed53974, the assertion failure messages are gone. Presumably they were fixed by commit 1e885b25275fb6763eb947b1e53b2d6911b967a8.

Changed in qemu:
status: New → Fix Committed
Aurelien Jarno (aurel32)
Changed in qemu:
status: Fix Committed → Fix Released
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.