from pwd import getpwnam
from string import Template
+# vsys probably should not be a plugin
+# the thing is, the right way to handle stuff would be that
+# if slivers get created by doing a,b,c
+# then they sohuld be delted by doing c,b,a
+# the current ordering model for vsys plugins completely fails to capture that
+from plugins.vsys import removeSliverFromVsys, startService as vsysStartService
+
import libvirt
import logger
import plnode.bwlimit as bwlimit
from initscript import Initscript
+from account import Account
from sliver_libvirt import Sliver_Libvirt
class Sliver_LXC(Sliver_Libvirt, Initscript):
# the generic /etc/init.d/vinit script is permanently refreshed, and enabled
self.install_and_enable_vinit()
# expose .ssh for omf_friendly slivers
- if 'omf_control' in self.rspec['tags']:
+ if 'tags' in self.rspec and 'omf_control' in self.rspec['tags']:
Account.mount_ssh_dir(self.name)
Sliver_Libvirt.start (self, delay)
# if a change has occured in the slice initscript, reflect this in /etc/init.d/vinit.slice
vref = rec['vref']
if vref is None:
- logger.log('sliver_libvirt: %s: WARNING - no vref attached defaults to lxc-f14' % (name))
- vref = "lxc-f14-x86_64"
+ vref = "lxc-f18-x86_64"
+ logger.log("sliver_libvirt: %s: WARNING - no vref attached, using hard-wired default %s" % (name,vref))
refImgDir = os.path.join(Sliver_LXC.REF_IMG_BASE_DIR, vref)
containerDir = os.path.join(Sliver_LXC.CON_BASE_DIR, name)
logger.log('sliver_lxc: %s: ERROR Expected reference image in %s'%(name,refImgDir))
return
+ # in fedora20 we have some difficulty in properly cleaning up /vservers/<slicename>
+ # also note that running e.g. btrfs subvolume create /vservers/.lvref/image /vservers/foo
+ # behaves differently, whether /vservers/foo exists or not:
+ # if /vservers/foo does not exist, it creates /vservers/foo
+ # but if it does exist, then it creates /vservers/foo/image !!
+ # so we need to check the expected container rootfs does not exist yet
+ if not os.path.exists (containerDir):
+ pass
+ else:
+ # if it's empty then let's clean it up
+ if not os.listdir(containerDir):
+ # clean up rootfs as userdel will only take care of /home/<slicename>
+ logger.log("sliver_lxc: %s: WARNING cleaning up empty %s"%(name,containerDir))
+ command = ['btrfs', 'subvolume', 'delete', containerDir]
+ logger.log_call(command, timeout=60)
+ # re-check
+ if os.path.exists (containerDir):
+ logger.log('sliver_lxc: %s: ERROR Could not create sliver - could not clean up empty %s'%(name,containerDir))
+ return
+ else:
+ logger.log('sliver_lxc: %s: ERROR Could not create sliver - could not clean up pre-existing %s'%(name,containerDir))
+ return
+
# Snapshot the reference image fs (assume the reference image is in its own
# subvolume)
command = ['btrfs', 'subvolume', 'snapshot', refImgDir, containerDir]
command = ['chmod', '755', containerDir]
logger.log_call(command, timeout=15*60)
- # customize prompt for slice owner, + LD_PRELOAD for transparently wrap bind
- dot_profile=os.path.join(containerDir,"root/.profile")
- ld_preload_msg="""# by default, we define this setting so that calls to bind(2),
-# when invoked on 0.0.0.0, get transparently redirected to the public interface of this node
-# see https://svn.planet-lab.org/wiki/LxcPortForwarding"""
- usrmove_path_msg="""# VM's before Features/UsrMove need /bin and /sbin in their PATH"""
- usrmove_path_code="""
-pathmunge () {
- if ! echo $PATH | /bin/egrep -q "(^|:)$1($|:)" ; then
- if [ "$2" = "after" ] ; then
- PATH=$PATH:$1
- else
- PATH=$1:$PATH
- fi
- fi
-}
-pathmunge /bin after
-pathmunge /sbin after
-unset pathmunge
-"""
- with open(dot_profile,'w') as f:
- f.write("export PS1='%s@\H \$ '\n"%(name))
- f.write("%s\n"%ld_preload_msg)
- f.write("export LD_PRELOAD=/etc/planetlab/lib/bind_public.so\n")
- f.write("%s\n"%usrmove_path_msg)
- f.write("%s\n"%usrmove_path_code)
-
# TODO: set quotas...
# Set hostname. A valid hostname cannot have '_'
command = ['chown', name, '%s/home/%s' % (containerDir, name)]
logger.log_call(command, timeout=10)
etcpasswd = os.path.join(containerDir, 'etc/passwd')
+ etcgroup = os.path.join(containerDir, 'etc/group')
if os.path.exists(etcpasswd):
- logger.log("adding user %s id %d to %s" % (name, uid, etcpasswd))
+ # create all accounts with gid=1001 - i.e. 'slices' like it is in the root context
+ slices_gid=1001
+ logger.log("adding user %(name)s id %(uid)d gid %(slices_gid)d to %(etcpasswd)s" % (locals()))
+ try:
+ file(etcpasswd,'a').write("%(name)s:x:%(uid)d:%(slices_gid)d::/home/%(name)s:/bin/bash\n" % locals())
+ except:
+ logger.log_exc("exception while updating %s"%etcpasswd)
+ logger.log("adding group slices with gid %(slices_gid)d to %(etcgroup)s"%locals())
try:
- file(etcpasswd,'a').write("%s:x:%d:%d::/home/%s:/bin/bash\n" % (name, uid, uid, name))
+ file(etcgroup,'a').write("slices:x:%(slices_gid)d\n"%locals())
except:
- logger.log_exc("exception while updating etc/passwd")
+ logger.log_exc("exception while updating %s"%etcgroup)
sudoers = os.path.join(containerDir, 'etc/sudoers')
if os.path.exists(sudoers):
try:
except:
logger.log_exc("exception while updating /etc/sudoers")
+ # customizations for the user environment - root or slice uid
+ # we save the whole business in /etc/planetlab.profile
+ # and source this file for both root and the slice uid's .profile
+ # prompt for slice owner, + LD_PRELOAD for transparently wrap bind
+ pl_profile=os.path.join(containerDir,"etc/planetlab.profile")
+ ld_preload_text="""# by default, we define this setting so that calls to bind(2),
+# when invoked on 0.0.0.0, get transparently redirected to the public interface of this node
+# see https://svn.planet-lab.org/wiki/LxcPortForwarding"""
+ usrmove_path_text="""# VM's before Features/UsrMove need /bin and /sbin in their PATH"""
+ usrmove_path_code="""
+pathmunge () {
+ if ! echo $PATH | /bin/egrep -q "(^|:)$1($|:)" ; then
+ if [ "$2" = "after" ] ; then
+ PATH=$PATH:$1
+ else
+ PATH=$1:$PATH
+ fi
+ fi
+}
+pathmunge /bin after
+pathmunge /sbin after
+unset pathmunge
+"""
+ with open(pl_profile,'w') as f:
+ f.write("export PS1='%s@\H \$ '\n"%(name))
+ f.write("%s\n"%ld_preload_text)
+ f.write("export LD_PRELOAD=/etc/planetlab/lib/bind_public.so\n")
+ f.write("%s\n"%usrmove_path_text)
+ f.write("%s\n"%usrmove_path_code)
+
+ # make sure this file is sourced from both root's and slice's .profile
+ enforced_line = "[ -f /etc/planetlab.profile ] && source /etc/planetlab.profile\n"
+ for path in [ 'root/.profile', 'home/%s/.profile'%name ]:
+ from_root=os.path.join(containerDir,path)
+ # if dir is not yet existing let's forget it for now
+ if not os.path.isdir(os.path.dirname(from_root)): continue
+ found=False
+ try:
+ contents=file(from_root).readlines()
+ for content in contents:
+ if content==enforced_line: found=True
+ except IOError: pass
+ if not found:
+ with open(from_root,"a") as user_profile:
+ user_profile.write(enforced_line)
+ # in case we create the slice's .profile when writing
+ if from_root.find("/home")>=0:
+ command=['chown','%s:slices'%name,from_root]
+ logger.log_call(command,timeout=5)
+
# Lookup for xid and create template after the user is created so we
# can get the correct xid based on the name of the slice
xid = bwlimit.get_xid(name)
try:
with open(template_filename) as f:
template = Template(f.read())
- xml = template.substitute(name=name, interfaces=interfaces, arch=arch)
+ xml = template.substitute(name=name, xid=xid, interfaces=interfaces, arch=arch)
except IOError:
logger.log('Failed to parse or use XML template file %s'%template_filename)
return
dom = conn.lookupByName(name)
except:
dom = conn.defineXML(xml)
- logger.verbose('lxc_create: %s -> %s'%(name, Sliver_Libvirt.debuginfo(dom)))
+ logger.verbose('lxc_create: %s -> %s'%(name, Sliver_Libvirt.dom_details(dom)))
@staticmethod
# Destroy libvirt domain
dom = conn.lookupByName(name)
except:
- logger.verbose('sliver_lxc: Domain %s does not exist!' % name)
+ logger.verbose('sliver_lxc.destroy: Domain %s does not exist!' % name)
+ return
+
+ # Slivers with vsys running will fail the subvolume delete
+ # removeSliverFromVsys return True if it stops vsys, telling us to start it again later
+ vsys_stopped = removeSliverFromVsys (name)
try:
+ logger.log("sliver_lxc.destroy: destroying domain %s"%name)
dom.destroy()
except:
- logger.verbose('sliver_lxc: Domain %s not running... continuing.' % name)
+ logger.verbose('sliver_lxc.destroy: Domain %s not running... continuing.' % name)
try:
+ logger.log("sliver_lxc.destroy: undefining domain %s"%name)
dom.undefine()
except:
- logger.verbose('sliver_lxc: Domain %s is not defined... continuing.' % name)
+ logger.verbose('sliver_lxc.destroy: Domain %s is not defined... continuing.' % name)
# Remove user after destroy domain to force logout
command = ['/usr/sbin/userdel', '-f', '-r', name]
logger.log_call(command, timeout=15*60)
- if os.path.exists(os.path.join(containerDir,"vsys")):
- # Slivers with vsys running will fail the subvolume delete.
- # A more permanent solution may be to ensure that the vsys module
- # is called before the sliver is destroyed.
- logger.log("destroying vsys directory and restarting vsys")
- logger.log_call(["rm", "-fR", os.path.join(containerDir, "vsys")])
- logger.log_call(["/etc/init.d/vsys", "restart", ])
-
# Remove rootfs of destroyed domain
command = ['btrfs', 'subvolume', 'delete', containerDir]
- logger.log_call(command, timeout=60)
-
- if os.path.exists(containerDir):
- # oh no, it's still here...
- logger.log("WARNING: failed to destroy container %s" % containerDir)
-
- logger.verbose('sliver_libvirt: %s destroyed.'%name)
-
+ logger.log_call(command, timeout=10)
+
+ # For some reason I am seeing this :
+ #log_call: running command btrfs subvolume delete /vservers/inri_sl1
+ #log_call: ERROR: cannot delete '/vservers/inri_sl1' - Device or resource busy
+ #log_call: Delete subvolume '/vservers/inri_sl1'
+ #log_call:end command (btrfs subvolume delete /vservers/inri_sl1) returned with code 1
+ #
+ # something must have an open handle to a file in there, but I can't find out what it is
+ # the following code aims at gathering data on what is going on in the system at this point in time
+ # note that some time later (typically when the sliver gets re-created) the same
+ # attempt at deleting the subvolume does work
+ # also lsof never shows anything relevant; this is painful..
+
+ if not os.path.exists(containerDir):
+ logger.log('sliver_lxc.destroy: %s cleanly destroyed.'%name)
+ else:
+ logger.log("-TMP-cwd %s : %s"%(name,os.getcwd()))
+ logger.log("-TMP-lsof %s"%name)
+ command=['lsof']
+ logger.log_call(command)
+ logger.log("-TMP-ls-l %s"%name)
+ command = ['ls', '-l', containerDir]
+ logger.log_call(command)
+ if os.path.exists(containerDir):
+ logger.log('sliver_lxc.destroy: ERROR could not cleanly destroy %s - giving up'%name)
+
+ if vsys_stopped: vsysStartService()
git://git.onelab.eu / nodemanager.git / blobdiff