lxcsu

   1 #!/usr/bin/python
   2
   3 import sys
   4 import os
   5 import setns
   6
   7 from argparse import ArgumentParser
   8
   9 drop_capabilities='cap_sys_admin,cap_sys_boot,cap_sys_module'
  10
  11 # can set to True here, but also use the -d option
  12 debug = False
  13
  14 def getarch(f):
  15     output = os.popen('readelf -h %s 2>&1'%f).readlines()
  16     classlines = [x for x in output if ('Class' in x.split(':')[0])]
  17     line = classlines[0]
  18     c = line.split(':')[1]
  19     if ('ELF64' in c):
  20         return 'x86_64'
  21     elif ('ELF32' in c):
  22         return 'i686'
  23     else:
  24         raise Exception('Could not determine architecture')
  25
  26 def umount(fs_dir, opts=''):
  27     output = os.popen('/bin/umount %s %s 2>&1'%(opts, fs_dir)).read()
  28     return ('device is busy' not in output)
  29
  30 def main ():
  31     parser = ArgumentParser()
  32     parser.add_argument("-n", "--nonet",
  33                                         action="store_true", dest="netns", default=False,
  34                                         help="Don't enter network namespace")
  35     parser.add_argument("-m", "--nomnt",
  36                                         action="store_true", dest="mntns", default=False,
  37                                         help="Don't enter mount namespace")
  38     parser.add_argument("-p", "--nopid",
  39                                         action="store_true", dest="pidns", default=False,
  40                                         help="Don't enter pid namespace")
  41     parser.add_argument("-r", "--root",
  42                                         action="store_true", dest="root", default=False,
  43                                         help="Enter as root: be careful")
  44     parser.add_argument("-i","--internal",
  45                                         action="store_true", dest="internal", default=False,
  46                                         help="does *not* prepend '-- -c' to arguments - or invoke lxcsu-internal")
  47     parser.add_argument("-d","--debug",
  48                                         action='store_true', dest='debug', default=False,
  49                                         help="debug option")
  50     parser.add_argument("-s","--nosliceuid",
  51                                         action='store_true', dest="nosliceuid", default=False,
  52                                         help="do not change to slice uid inside of slice")
  53     parser.add_argument("-o","--noslicehome",
  54                                         action='store_true', dest="noslicehome", default=False,
  55                                         help="do not change to slice home directory inside of slice")
  56
  57     if os.path.exists("/etc/lxcsu_default"):
  58         defaults = parser.parse_args(file("/etc/lxcsu_default","r").read().split())
  59         parser.set_defaults(**defaults.__dict__)
  60
  61         parser.add_argument ("slice_name")
  62         parser.add_argument ("command_to_run",nargs="*")
  63
  64         args = parser.parse_args()
  65         slice_name=args.slice_name
  66
  67         # unless we run the symlink 'lxcsu-internal', or we specify the -i option, prepend '--' '-c'
  68         if sys.argv[0].find('internal')>=0: args.internal=True
  69
  70         if len(args.command_to_run)>0 and (args.command_to_run[0] == "/sbin/service"):
  71             # A quick hack to support nodemanager interfaces.py when restarting
  72             # networking in a slice.
  73             args.nosliceuid = True
  74
  75         # plain lxcsu
  76         if not args.internal:
  77             # no command given: enter interactive shell
  78             if not args.command_to_run: args.command_to_run=['/bin/sh']
  79             args.command_to_run = [ '-c' ] + [" ".join(args.command_to_run)]
  80
  81         # support for either setting debug at the top of this file, or on the command-line
  82         if args.debug:
  83             global debug
  84             debug=True
  85
  86         try:
  87             cmd = 'grep %s /proc/*/cgroup | grep freezer'%slice_name
  88             output = os.popen(cmd).readlines()
  89         except:
  90             print "Error finding slice %s"%slice_name
  91             exit(1)
  92
  93         slice_spec = None
  94
  95         # provide a default as this is not always properly computed
  96         arch = None
  97
  98         for e in output:
  99             try:
 100                 l = e.rstrip()
 101                 path = l.split(':')[0]
 102                 comp = l.rsplit(':')[-1]
 103                 slice_name_check = comp.rsplit('/')[-1]
 104                 if debug: print "dealing with >%s<"%slice_name_check
 105
 106                 if (slice_name_check == slice_name):
 107                     if debug: print "found %s"%slice_name
 108                     slice_path = path
 109                     pid = slice_path.split('/')[2]
 110                     cmdline = open('/proc/%s/cmdline'%pid).read().rstrip('\n\x00')
 111                     if (cmdline == '/sbin/init') or (cmdline.startswith("init [")):
 112                         slice_spec = slice_path
 113                         arch = getarch('/proc/%s/exe'%pid)
 114                         break
 115             except Exception,e:
 116                 if debug:
 117                     import traceback
 118                     print "BEG lxcsu - ignoring exception"
 119                     traceback.print_exc()
 120                     print "END lxcsu - ignoring exception"
 121                     pass
 122
 123         if (not slice_spec or not pid):
 124             print "lxcsu: domain not started: %s"%slice_name
 125             exit(1)
 126
 127         if arch is None:
 128             arch = 'x86_64'
 129
 130         # Set sysctls specific to slice
 131         sysctls = []
 132         sysctl_dir = '/etc/planetlab/vsys-attributes/%s'%slice_name
 133         if (os.access(sysctl_dir,0)):
 134             entries = os.listdir(sysctl_dir)
 135             for e in entries:
 136                 prefix = 'vsys_sysctl.'
 137                 if (e.startswith(prefix)):
 138                     sysctl_file = '/'.join([sysctl_dir,e])
 139                     sysctl_name = e[len(prefix):]
 140                     sysctl_val = open(sysctl_file).read()
 141                     sysctls.append((sysctl_file, sysctl_name, sysctl_val))
 142
 143         # Enter cgroups
 144         try:
 145             for subsystem in ['cpuset','memory','blkio']:
 146                 open('/sys/fs/cgroup/%s/libvirt/lxc/%s/tasks'%(subsystem,slice_name),'w').write(str(os.getpid()))
 147
 148         except:
 149             print "Error assigning resources: %s"%slice_name
 150             exit(1)
 151
 152         try:
 153             open('/sys/fs/cgroup/cpuacct/system/libvirtd.service/libvirt/lxc/%s/tasks'%slice_name,'w').write(str(os.getpid()))
 154         except:
 155             print "Error assigning cpuacct: %s" % slice_name
 156             exit(1)
 157
 158         # If the slice is frozen, then we'll get an EBUSY when trying to write to the task
 159         # list for the freezer cgroup. Since the user couldn't do anything anyway, it's best
 160         # in this case to error out the shell. (an alternative would be to un-freeze it,
 161         # add the task, and re-freeze it)
 162         try:
 163             f=open('/sys/fs/cgroup/freezer/libvirt/lxc/%s/tasks'%(slice_name),'w')
 164             f.write(str(os.getpid()))
 165             # note: we need to call f.close() explicitly, or we'll get an exception in
 166             # the object destructor, which will not be caught
 167             f.close()
 168         except:
 169             print "Error adding task to freezer cgroup. Slice is probably frozen: %s" % slice_name
 170             exit(1)
 171
 172         setns.chcontext('/proc/%s/ns/uts'%pid)
 173         setns.chcontext('/proc/%s/ns/ipc'%pid)
 174
 175         if (not args.pidns):
 176             setns.chcontext('/proc/%s/ns/pid'%pid)
 177
 178         if (not args.netns):
 179             setns.chcontext('/proc/%s/ns/net'%pid)
 180
 181         if (not args.mntns):
 182             setns.chcontext('/proc/%s/ns/mnt'%pid)
 183
 184         proc_mounted = False
 185         if (not os.access('/proc/self',0)):
 186             proc_mounted = True
 187             setns.proc_mount()
 188
 189         for (sysctl_file, sysctl_name, sysctl_val) in sysctls:
 190             for fn in ["/sbin/sysctl", "/usr/sbin/sysctl", "/bin/sysctl", "/usr/bin/sysctl"]:
 191                 if os.path.exists(fn):
 192                     os.system('%s -w %s=%s  >/dev/null 2>&1'%(fn, sysctl_name,sysctl_val))
 193                     break
 194                 else:
 195                     print "Error: image does not have a sysctl binary"
 196
 197         # cgroups is not yet LXC-safe, so we need to use the coarse grained access control
 198         # strategy of unmounting the filesystem
 199
 200         umount_result = True
 201         for subsystem in ['cpuset','cpu,cpuacct','memory','devices','freezer','net_cls','blkio','perf_event','systemd']:
 202             fs_path = '/sys/fs/cgroup/%s'%subsystem
 203             if (not umount(fs_path,'-l')):
 204                 pass
 205             # Leaving these comments for historical reference
 206                         #print "Error disabling cgroup access"
 207                         #exit(1) - Don't need this because failure here implies failure in the call to umount /sys/fs/cgroup
 208
 209         if (not umount('/sys/fs/cgroup')):
 210             print "Error disabling cgroup access"
 211             exit(1)
 212
 213         pid = os.fork()
 214
 215         # capsh has a --user option starting with f14
 216         # so if only for f12 we need to fake this one
 217         #
 218         # capsh.c does essentially the following when invoked with --user:
 219         #               pwd = getpwnam(user); ...
 220         #               ngroups = MAX_GROUPS;
 221         #               status = getgrouplist(user, pwd->pw_gid, groups, &ngroups); ...
 222         #               status = setgroups(ngroups, groups); ...
 223         #               status = setgid(pwd->pw_gid); ...
 224         #               status = setuid(pwd->pw_uid); ...
 225         # however we cannot simulate that ourselves because if we did in this process then
 226         # capsh could not be allowed to mess with caps any more
 227
 228         def getuid (slicename):
 229             import pwd
 230             try:
 231                 return pwd.getpwnam(slicename).pw_uid
 232             except:
 233                 return
 234
 235         if (pid == 0):
 236             cap_arg = '--drop='+drop_capabilities
 237
 238             if (not args.root):
 239                 if (args.nosliceuid):
 240                     # we still want to drop capabilities, but don't want to switch UIDs
 241                     exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--','--login',]+args.command_to_run
 242                 else:
 243                     uid = getuid (slice_name)
 244                     if not uid:
 245                         print "lxcsu could not spot %s in /etc/passwd - exiting"%slice_name
 246                         exit(1)
 247                         exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--uid=%s'%uid,'--','--login',]+args.command_to_run
 248 # once we can drop f12, it would be nicer to instead go for
 249 #                                exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--user=%s'%slice_name,'--','--login',]+args.command_to_run
 250                     else:
 251                         exec_args = [arch,'/usr/sbin/capsh','--','--login']+args.command_to_run
 252
 253                 os.environ['SHELL'] = '/bin/sh'
 254                 if os.path.exists('/etc/planetlab/lib/bind_public.so'):
 255                     os.environ['LD_PRELOAD'] = '/etc/planetlab/lib/bind_public.so'
 256                 if not args.noslicehome:
 257                     os.environ['HOME'] = '/home/%s'%slice_name
 258                     os.chdir("/home/%s"%(slice_name))
 259                 if debug: print 'lxcsu:execv:','/usr/bin/setarch',exec_args
 260                 os.execv('/usr/bin/setarch',exec_args)
 261         else:
 262             setns.proc_umount()
 263             _,status = os.waitpid(pid,0)
 264             exit(os.WEXITSTATUS(status))
 265
 266 if __name__ == '__main__':
 267     main()