vzctl umount: fix exit code if CT is runningFixes the following issue:
# vzctl umount 10101
Container is running -- stop it first
# echo $?
0
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzoversell: handle unlimited RAM CTsSome containers might be configured as unlimited
(i.e. infinite physpages and privvmpages limit).
Also, a container might appear as unlimited while
it is being restored from dump (as beancounter limits
are initially set to unlimited and only set to proper
values at the end of the restoration process).
In case such unlimited containers are present,
overcommitment figures shown doesn't make sen...
store_devnodes: fix NULL derefAs pointed out by Linas Žilinskas in bug #3228,
we should check name for being non-NULL here.
https://bugzilla.openvz.org/3228
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vps-create.sh: use stat -f instead of dfOn a node with a stale NFS mount (not related to /vz), vzctl create
fails after a long timeout. It happens because vps-create calls
df $VE_ROOT, but for some odd reason df wants to stat() all the
mount points and it hangs for a while and then fails on that stale mount.
The solution is to use stat -f rather than df.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl.spec: require attr packageIt never occured to me that setfattr can not be available, but it happens:
# vzctl create ...
...
Performing postcreate actions
/etc/vz/dists/scripts/postcreate.sh: line 119: setfattr: command not found
...
So, let's require it from spec file.
This is an addition to commit f537353.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl.spec: drop the Conflicts: vzkernelThe change this patch is reverting was introduced to fix bug #3094.
Unfortunately, it breaks the update for RHEL5-based OpenVZ kernels,
which is the subject of bug #3219.
The only solution I found is to list all the old RHEL6-based kernel
explicitly. It would be good enough, but looks ugly.
As the newest RHEL6-based kernel conflicting with this vzctl is at least
2 years old, let's just drop t...
create_hardlink_dir(): fix wrong owner/perms caseIn case CPT_HARDLINK_DIR exists, is a directory, but its
ownership and/or permissions are wrong, the best course
of action is not to remove it, but just fix what's wrong.
This is what this patch does.
In addition:
* mark create_hardlink_dir() as static (as it should be)
* report and return a potential error from stat() syscall,
(as it should not happen and if it is, it's probably fatal)
Th...
vzmigrate: check CPU caps for suspended CTThis commit adds a check for CPU capabilities when migrating a
suspended container. Such a check is needed because if CPU caps
differ, a container might not survive restore after migrating.
To implement this check, a support from the kernel is required
(CPT_TEST_CAPS ioctl for /dev/cpt). Such support was errorneously
removed from the kernel 042stab101.5, but then restored in 042stab107.1,
so a...
vzmigrate: fix for vzfsync if VE_PRIVATE differsWhen referring to DDXML, we assumed VE_PRIVATE == VE_PRIVATE_REMOTE
which is not always the case. Fix it.
https://bugzilla.openvz.org/3170
Reported-by: toumin <983510@gmail.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
etc/vz.conf: improve NEIGHBOUR_DEVS descriptionThis is an addition to commit 2869ab3
https://bugzilla.openvz.org/1289
https://bugzilla.openvz.org/3192
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ct_enter(): enter mnt namespace lastmnt needs to be set last since it changes the filesystem.
Runnig vzctl start on upstream kernel fails for me on CentOS 6.5.
I used to run vzkernel but reverted to stock to be able to run
docker / lxc at the same time.
With strace I noticed it's failing to find /proc/<pid>/ns/pid and
I traced that to the loop in hooks_ct.c. After it does setns
on the mnt file the pid one no longer can be opene...
vz.conf: allow list of interfaces in NEIGHBOUR_DEVSHistorically, allowed values for NEIGHBOUR_DEVS were either "detect"
or any other value, in the last case "all devices" are assumed.
It's time to change it. From now on, NEIGHBOUR_DEVS can be one of:
* all (or empty string, or unset) -- old behavior (all devices are used
* detect -- same as before, devices to use for ARP are detected
* list:* -- devices are explicitly listed (as in "list:et...
vzctl set --devices: made cumulative, fixThis is the same as commits 90e8dd6 and b8708e2 for vzctl set
--devnodes. From now on:
* the option is now cumulative, so any new devices set by
"vzctl set --devices" command are added to the existing devices
* vzctl set --devices t:major:minor:none --save" removes the device
from the configuration.
Note if last device is deleted from the list, we have to add DEVICES=""
to the configurat...
vzctl set --devices: fix SIGSEGVThis is caused by the commit c2a48a7e. Note add_dev_param() is called
from parse_devices_str() in which name is not set, so we shouldn't
blindly use strdup().
https://bugzilla.openvz.org/3189
Reported-by: Devon <devon.b@virtualcomplete.com>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
init.d: exit code if vzctl or vz.conf not foundIn case vzctl binary of vz.conf global configuration file is not found,
we should not exit with 0. Instead, use codes 5 "program is not
installed" and 6 "program is not configured", as per LSB
http://refspecs.linuxbase.org/LSB_3.1.1/LSB-Core-generic/LSB-Core-generic/iniscrptact.html
https://bugzilla.openvz.org/show_bug.cgi?id=3195
Reported-by: nethubonline@gmail.com
Signed-off-by: Kir Kolysh...
init.d/vz-redhat: fix exit code according to LSBIn case an invalid command is specified, we should return exit code of
3 -- <<unimplemented feature (for example, "reload")>>
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzeventd(8): document new behaviorAs per previous commit, vzeventd now processes all known events.
Document it.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzevent: try to run a script for all known eventsPreviously, we only tried to execute vzevent-stop and vzevent-reboot
scripts. This change makes vzevent process all known events as they
come from the kernel. Unknown (not recognized) events are still logged
as errors (maybe we need to change that).
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzeventd: ignore non-existent event scriptsAdd a check that the executable event script file is present.
Ignore ENOENT but report other errors.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzlist -j: output valid JSON for no CTsIn case there are no containers found, still produce valid JSON
output.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
make_dir_mode(): ignore EEXIST from mkdir()It seems two instances of vzctl can create a directory
at the same time, so due to a time period between access()
and mkdir() a directory can be created, so we need to ignore
EEXIST error.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vps_start_custom(): fixupIn case VPS_PRESTART script failed, we just return an error,
no need to rollback anything yet.
This is a fix to commit 0807ef4.
Found by Coverity, CID 1271650, 1271651, 1271652.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Revert "dir_empty(): skip lost+found"As described in commit 07a6ee9, it was a bad idea to try allowing
VE_PRIVATE being a mount point. As we don't do it anymore,
this patch is no longer required -- so revert it
This reverts commit 320bc278ac7d2e3f712f15f802550bca45c2ca7b.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vps_create: minor fixes to cleanup logicfs_create() is supposed to remove VE_PRIVATE in case of error,
so we don't need to clean it up if it return an error. This fixes
the following error (lines marked with *):
# vzctl create 666 --ostemplate centos-6-x86-minimal --private /dev/shm
Can't create: private dir (/dev/shm) is a mount point. Suggestion: use --private /dev/shm/private
* Destroying container private area: /dev/shm
...
fs_create: lock privateWe are supposed to have VE_PRIVATE created for the duration of
fs_create() run. It's like some sort of locking.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl create: disallow VE_PRIVATE be a mount pointAs it is pointed out in bug #3166, when VE_PRIVATE is a mount point,
container creation leads to a bunch of weird errors:
# vzctl create $CTID --private '/data/$VEID'
Creating container private area (debian-7.0-x86_64-minimal)
Can't rename /data/10001.tmp to /data/10001: Device or resource busy
Destroying container private area: /data/10001
Warning: directory /data/10001 is not on the sam...
vzctl start/resume: load kernel modules needed for CTThis patch introduces vps-prestart script which is executed upon
container start or restore. Currently, this script checks CT
config and loads some kernel modules if reqired.
This should help NFS, NFSD, PPP, and Docker to work inside CT
(if configured).
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl chkpnt: workaround for ENOSPCIn order to be able to checkpoint/restore processes having opened
but deleted files, we have a kernel helper that hardlinks such files
to a special CPT_HARDLINK_DIR (see commit 16ee8dd).
The problem is if there is no disk space (or disk inodes) available,
checkpointing fails like this:
# vzctl chkpnt 74933 --dump --dumpfile /dev/shm/dump.74933
Setting up checkpoint...
join context..
dump...
[style] call clean_hardlink_dir() from _env_create()We need to call clean_hardlink_dir() when starting a container.
Calling it from fill_container_param() is not obvious and confusing.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl set --devnodes: remove devices from CTStrictly speaking it is not required to remove /dev/XXX
nodes from CT when permissions are revoked (as in
"vzctl set --devnodes XXX:none") since XXX is becoming
inaccessible from CT anyway. Having said that, removal
of revoked devices makes things look more clear, so let's
do it.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl set --devnodes: fix removing last deviceApparently removing a last device from the list is a special case,
as merge_conf() doesn't detect it unless we provide some string.
So, let's print DEVNODES="" in this case.
This fixes the following bug:
# grep DEVNODES /etc/vz/conf/101.conf
DEVNODES="ttyS3:rw "
# vzctl set 101 --devnodes ttyS3:none --save
No changes in CT configuration, not saving
# grep DEVNODES /etc/vz/conf/101.conf
...
vzctl set --devnodes: made cumulative, fixThis patch fixes the handling of --devnodes option.
First, the option is now cumulative, so any new devices set by
"vzctl set --devnodes" command are added to the existing devices.
Second, "vzctl set --devnodes DEV:none --save" removes DEV from the
configuration.
https://bugzilla.openvz.org/3186
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
add_dev_param(): don't modify second argumentCommit ea64ee6 has a bad side effect: after calling merge_dev()
the device names in vps_p list (which is supposed to have configuration
as read from ve.conf) disappear. This is caused by the hunk modifying
add_dev_param().
This patch fixes that side effect, also marking the second argument
const so this won't happen in the future.
This is not a bug fix per se (as no one was using dev in vps_p...
parse_dev(): rm freefree(dev.name) is useless here: parse_dev() is parsing --devices
option which doesn't ever set name.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
vzctl snapshot-delete: ignore ploop 'no guid found'If a snapshot is created by means of vzctl, but later removed
via ploop tool directly, it results in failure of vzctl snapshot-delete.
This patch solves it by ignoring 'no guid found" error from
ploop_delete_snapshot(), so we can proceed and remove other parts
of the snapshot (ve.conf, dump file) and a record in Snapshots.xml.
Based on patch by Igor Sukhih <igor@openvz.org>
As this requires ...
vz-postinstall: add way to disable distro kernelsThis is a frequent problem with e.g. CentOS then a new distro kernel
is installed and becomes bootable by default, so OpenVZ features are
lost.
A very easy way to prevent this is to set 'exclude=kernel' in
/etc/yum.conf. This doesn't affect OpenVZ kernels as they are
named vzkernel.
Run this on install, and clean up our changes on uninstall so
if vzctl is removed stock distro kernels will con...
dists/scripts/{funtoo,gentoo}*: remove env var docAn up-to-date documentation on environment variables passed
on to the scripts is available from etc/dists/distribution.conf-template
file, so let's remove it from these scripts to not create confusion.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
debian-add_ip.sh: silent an errorIf /etc/network/interfaces doesn't exist it's OK, no need to
print an error from grep.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Introduce funtoo-set_hostnameFuntoo needs its own script for setting hostname, different than
gentoo's. Please use the attached script. The current approach sets only
the local hostname "foo" and not FQDN "foo.bar.com".
https://bugs.funtoo.org/browse/FL-1418
https://bugzilla.openvz.org/show_bug.cgi?id=3097
is_mount_point(): fix for non-existing dirRecent commit 2b1db5b introduces a regression -- in case VE_PRIVATE
doesn't exist, is_mount_point() returns -1 and CT creation fails.
Fix is_mount_point() so if a directory doesn't exist, 0 is returned
(i.e. non-existing directory is definitely not a mount point).
https://bugzilla.openvz.org/3174
Reported-by: Pavel Odintsov <pavel.odintsov@gmail.com>
Signed-off-by: Kir Kolyshkin <kir@openvz....
vzctl delete: do rm config/dumpIn case we were not able to remove VE_PRIVATE, do remove container
configuration file(s) and its dumps anyway.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Kir Kolyshkin
Kirill Kolyshkin
Zach Pomerantz
