Device model needs to do xm destroy before it exits on
guest shutdown/poweroff/halt -p.
Signed-off-by: Edwin Zhai <edwin.zhai@intel.com>
Signed-off-by: Arun Sharma <arun.sharma@intel.com>
Without this, xend will not spawn a new device model process after the reboot.
Signed-off-by: Ke Yu <ke.yu@intel.com>
Signed-off-by: Arun Sharma <arun.sharma@intel.com>
This patch remembers the device model pid across xend restarts and
avoids creating duplicate device model processes.
Also, device models don't inherit socket fds from xend (which are closed
on exec now), which used to prevent xend restarting.
Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Xiaohui Xin <xiaohui.xin@intel.com>
Signed-off-by: Arun Sharma <arun.sharma@intel.com>
If we define "vif" in the config file, image.py will raise error "vmx:
missing vbd configuration". The reason is "vif" is dealt with as a "vbd"
device.
This patch fixes this issue by dealing with "vbd" and "vif " separately,
removing "macaddr" arg and parsing mac address from "vif" instead.
Also, the vbd doesn't have to be a file anymore, but can be a physical
disk partition.
Signed-off-by: Yunfeng Zhao <yunfeng.zhao@intel.com>
Signed-off-by: Arun Sharma <arun.sharma@intel.com>
recently stopped working if security was enabled.
xm list fails if security is enabled because of a single
additional "%" character in a print.
Signed-off-by: Reiner Sailer <sailer@watson.ibm.com>
1) make xenconsoled exit gracefully if xcs dies
2) daemonize xenstored before binding to xenstored socket
3) wait to close stdio in xenstored until we're ready to accept
connections (so that PID=`xenstored --output-pid` doesn't return until
it's ready to accept connections)
4) updates tools/misc/xend for these changes
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Rusty Russell <rusty@rustycorp.com.au>
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Mike Wray made this change because xend's callbacks were failing
with EROFS, because a transaction was going on. However, the
solution is to either block callbacks during transactions, or use
a different connection to the store.
We fail ex-transaction activity because it's not clear what the right
behaviour is: it allows deadlock if two people do it, and it implied
your transaction isn't rooted in the right place.
This hack caused the testsuite to fail (obviously), so as a result the
testsuite wasn't run very often. This caused it to bitrot.
Signed-off-by: Rusty Russel <rusty@rustcorp.com.au>
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
It does a number of things:
* It implements the standardized options proposed via my RFC for xm
* It restructures the calling of these functions to be function pointers out
of a dictionary, instead of subclasses found through inspection
* It consolidates all help info, making it easier to see that the help
screen looks consistent across commands. (Per command help isn't yet in
this patch just global help)
* It moves object importing into the functions that require them only,
allowing commands like "xm help" to be runable as non root
* It adds command aliasing to provide backwards compatibility with many
commands (i.e. xm balloon)
* It provides more useful error messages is most fail cases
* It implements a short help by default (which takes < 24 screen lines) as
requested by Mark W at OLS
* It attempts to trap more exceptions than before and provide useful errors
based on them (this requires more testing)
This patch makes no changes to any other elements of xen/xm, and works with
all other objects and libraries as is.
Signed-off-by: Sean Dague <sean@dague.net>
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
These bits are daemon-only, never come though to userspace. We already
changed them once and broke compilation: better they had never been
referenced.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
xs_watch_stress, reduce cycles in "make check" random test.
2) xs_crashme: corrupt random packets going to xenstored, watch it
crash.
3) Handle second input from before we finished output on first one.
4) Fix bug where one-arg operations are given zero args.
5) Fix bug where SET_PERMS fails after blocking on transaction.
6) Fix memory leak when DIRECTORY op given no argument.
7) Fail on first memory leak, for better testing.
8) Fix missing waiting_for_ack initialization for new connections.
9) Ensure all input and output is handled for domains so we don't stall.
10) Fix overrun bug in xs_count_strings on non-nul-terminated strings.
11) New test for clients which write without waiting for response.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
over to using xenstore for triggers. Dropping the control messages also
fixes the issue with not being able to use xm vcpu-hotplug on dom0.
Signed-off-by: Ryan Harper <ryanh@us.ibm.com>
of codes for signaling a shutdown request.
Also, we fix the return code for the shutdown notifier, so other
notifiers will get to run ;)
Signed-off-by: Dan Smith <danms@us.ibm.com>
targets into "make -s check" (the GNU standard) which is designed to run
in several seconds or less, with only "<dirname>:" and a seried of "."
for output, and "make fullcheck" which is more verbose and complete, and
runs in minutes.
I plan to submit a patch to plumb these tests in at the top level soon,
with the idea of having a "make -s check" which you can reasonable run
every checkin (time make check << time make), and a "make fullcheck"
which you can run after significant changes.
This patch speeds up testing by making tests one single "xs_test"
script, enhancing xs_test with a simple "expect" command. It also only
runs the "high value" tests under "make check", and leaves the rest for
"make fullcheck".
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
still takes one (and just doesn't use it). Noone uses it anyway, so
nothing should break by taking it away.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
xen-unstable.hg but it doesn't add itself to any of the Makefiles but
just as a subdirectory of consoled. There's a README in the directory
that should be read otherwise you won't be able to do anything useful
with it :-)
Right now, we can get pretty far before seeing data corruption.
Previously, we'd see it almost immediately (after ~600 bytes, which is
the max the control channel can hold before filling up).
It's actual general enough that it could be also used to test network
traffic corruption but there are certainly better tools for that.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
does nothing (see ctrl_interface.c:ctrl_chan_write_request()).
The following patch throttles the rate in which consoled writes data to
xcs. With the current values, you get a responsive console but avoid
data corruption in most scenarios.
I'm able to get pretty far in my regression test with this patch. With
higher throttle values I'm able to get even further (but the console
becomes painfully slow).
I implemented proper control channel queuing in xenctld in VM-Tools and
it's pretty nasty stuff. This should prevent corruption for most users
until we can get rid of xcs.
Regards,
Anthony Liguori
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com
1. Converts the shutdown driver and xend to use the store instead of
control messages,
2. Includes Anthony's xenstore notification code, and
3. Changes xend so that sysrq's are no longer sent as "special case"
shutdown messages. Store keys are cheap, so making the sysrq
delivery less obscure is good.
I think I have made all of the appropriate modifications to Xend, but
it is complex, so I may have missed something. Comments are welcome.
Signed-off-by: Dan Smith <danms@us.ibm.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
three patches to fix a problem with Xend starting consoled. consoled
depends on xenstored to be running and xenstored is started on demand in
Xend. The patches change xenstored to manage its own pidfile, and have
xend start actually start up xenstored.
