X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_lxc.py;h=a08b0ce693dd7ab8434aa7136d3e15ab3d585746;hb=729dc5bb424cf4f8355e80c7a80f6751ff3c0370;hp=432f42644586970cc866775df426e50060569240;hpb=d88a6b97acdd56a303918cac7ed16ffad269d648;p=nodemanager.git diff --git a/sliver_lxc.py b/sliver_lxc.py index 432f426..a08b0ce 100644 --- a/sliver_lxc.py +++ b/sliver_lxc.py @@ -10,11 +10,19 @@ import grp 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 + 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): @@ -48,7 +56,7 @@ 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 @@ -82,8 +90,8 @@ class Sliver_LXC(Sliver_Libvirt, Initscript): 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) @@ -103,33 +111,6 @@ class Sliver_LXC(Sliver_Libvirt, Initscript): 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 '_' @@ -189,12 +170,20 @@ unset pathmunge 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("%s:x:%d:%d::/home/%s:/bin/bash\n" % (name, uid, uid, name)) + 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 etc/passwd") + logger.log_exc("exception while updating %s"%etcpasswd) + logger.log("adding group slices with gid %(slices_gid)d to %(etcgroup)s"%locals()) + try: + file(etcgroup,'a').write("slices:x:%(slices_gid)d\n"%locals()) + except: + logger.log_exc("exception while updating %s"%etcgroup) sudoers = os.path.join(containerDir, 'etc/sudoers') if os.path.exists(sudoers): try: @@ -202,6 +191,56 @@ unset pathmunge 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) @@ -220,7 +259,7 @@ unset pathmunge 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 @@ -231,7 +270,7 @@ unset pathmunge 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 @@ -243,41 +282,51 @@ unset pathmunge containerDir = Sliver_LXC.CON_BASE_DIR + '/%s'%(name) + # 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. + removeSliverFromVsys (name) + try: # 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 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) + if not os.path.exists(containerDir): + logger.log('sliver_lxc.destroy: %s cleanly destroyed.'%name) + else: + # oh no, it's still here... + logger.log("sliver_lxc.destroy: 1st warning: could not delete %s" % containerDir) + # this is for debugging but does not seem to be of much use + logger.log_call (['lsof']) + # what I can see on running nodes is that a second subvolume delete seems to do the trick here + # so let's check if that could be a workaround + logger.log("sliver_lxc.destroy: 2nd attempt at btrfs subvolume delete %s" % containerDir) + command = ['btrfs', 'subvolume', 'delete', containerDir] + logger.log_call(command, timeout=60) + if not os.path.exists(containerDir): + logger.log("sliver_lxc.destroy: WARNING: failed to delete %s after 2 attempts"%containerDir) - logger.verbose('sliver_libvirt: %s destroyed.'%name)