7 from argparse import ArgumentParser
9 drop_capabilities='cap_sys_admin,cap_sys_boot,cap_sys_module'
11 # can set to True here, but also use the -d option
15 output = os.popen('readelf -h %s 2>&1'%f).readlines()
16 classlines = [x for x in output if ('Class' in x.split(':')[0])]
18 c = line.split(':')[1]
24 raise Exception('Could not determine architecture')
27 output = os.popen('/bin/umount %s 2>&1'%fs_dir).read()
28 return ('device is busy' not in output)
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,
50 parser.add_argument("-s","--nosliceuid", dest="nosliceuid", default=False,
51 help="do not change to slice uid inside of slice")
52 parser.add_argument ("slice_name")
53 parser.add_argument ("command_to_run",nargs="*")
55 args = parser.parse_args()
56 slice_name=args.slice_name
58 # unless we run the symlink 'lxcsu-internal', or we specify the -i option, prepend '--' '-c'
59 if sys.argv[0].find('internal')>=0: args.internal=True
61 if len(args.command_to_run)>0 and (args.command_to_run[0] == "/sbin/service"):
62 # A quick hack to support nodemanager interfaces.py when restarting
63 # networking in a slice.
64 args.nosliceuid = True
68 # no command given: enter interactive shell
69 if not args.command_to_run: args.command_to_run=['/bin/sh']
70 args.command_to_run = [ '-c' ] + [" ".join(args.command_to_run)]
72 # support for either setting debug at the top of this file, or on the command-line
78 cmd = 'grep %s /proc/*/cgroup | grep freezer'%slice_name
79 output = os.popen(cmd).readlines()
81 print "Error finding slice %s"%slice_name
86 # provide a default as this is not always properly computed
92 path = l.split(':')[0]
93 comp = l.rsplit(':')[-1]
94 slice_name_check = comp.rsplit('/')[-1]
95 if debug: print "dealing with >%s<"%slice_name_check
97 if (slice_name_check == slice_name):
98 if debug: print "found %s"%slice_name
100 pid = slice_path.split('/')[2]
101 cmdline = open('/proc/%s/cmdline'%pid).read().rstrip('\n\x00')
102 if (cmdline == '/sbin/init') or (cmdline.startswith("init [")):
103 slice_spec = slice_path
104 arch = getarch('/proc/%s/exe'%pid)
109 print "BEG lxcsu - ignoring exception"
110 traceback.print_exc()
111 print "END lxcsu - ignoring exception"
114 if (not slice_spec or not pid):
115 print "Not started: %s"%slice_name
121 # Set sysctls specific to slice
123 sysctl_dir = '/etc/planetlab/vsys-attributes/%s'%slice_name
124 if (os.access(sysctl_dir,0)):
125 entries = os.listdir(sysctl_dir)
127 prefix = 'vsys_sysctl.'
128 if (e.startswith(prefix)):
129 sysctl_file = '/'.join([sysctl_dir,e])
130 sysctl_name = e[len(prefix):]
131 sysctl_val = open(sysctl_file).read()
132 sysctls.append((sysctl_file, sysctl_name, sysctl_val))
136 for subsystem in ['cpuset','memory','blkio']:
137 open('/sys/fs/cgroup/%s/libvirt/lxc/%s/tasks'%(subsystem,slice_name),'w').write(str(os.getpid()))
140 print "Error assigning resources: %s"%slice_name
144 open('/sys/fs/cgroup/cpuacct/system/libvirtd.service/libvirt/lxc/%s/tasks'%slice_name,'w').write(str(os.getpid()))
146 print "Error assigning cpuacct: %s" % slice_name
149 # If the slice is frozen, then we'll get an EBUSY when trying to write to the task
150 # list for the freezer cgroup. Since the user couldn't do anything anyway, it's best
151 # in this case to error out the shell. (an alternative would be to un-freeze it,
152 # add the task, and re-freeze it)
154 f=open('/sys/fs/cgroup/freezer/libvirt/lxc/%s/tasks'%(slice_name),'w')
155 f.write(str(os.getpid()))
156 # note: we need to call f.close() explicitly, or we'll get an exception in
157 # the object destructor, which will not be caught
160 print "Error adding task to freezer cgroup. Slice is probably frozen: %s" % slice_name
163 setns.chcontext('/proc/%s/ns/uts'%pid)
164 setns.chcontext('/proc/%s/ns/ipc'%pid)
167 setns.chcontext('/proc/%s/ns/pid'%pid)
170 setns.chcontext('/proc/%s/ns/net'%pid)
173 setns.chcontext('/proc/%s/ns/mnt'%pid)
176 if (not os.access('/proc/self',0)):
180 for (sysctl_file, sysctl_name, sysctl_val) in sysctls:
181 for fn in ["/sbin/sysctl", "/usr/sbin/sysctl", "/bin/sysctl", "/usr/bin/sysctl"]:
182 if os.path.exists(fn):
183 os.system("mount -o remount,rw none /proc/sys")
184 os.system('%s -w %s=%s'%(fn, sysctl_name,sysctl_val))
185 os.system("mount -o remount,ro none /proc/sys")
188 print "Error: image does not have a sysctl binary"
190 # cgroups is not yet LXC-safe, so we need to use the coarse grained access control
191 # strategy of unmounting the filesystem
194 for subsystem in ['cpuset','cpu,cpuacct','memory','devices','freezer','net_cls','blkio','perf_event']:
195 fs_path = '/sys/fs/cgroup/%s'%subsystem
196 if (not umount(fs_path)):
197 print "Error disabling cgroup access"
200 if (not umount('/sys/fs/cgroup')):
201 print "Error disabling cgroup access"
206 # capsh has a --user option starting with f14
207 # so if only for f12 we need to fake this one
209 # capsh.c does essentially the following when invoked with --user:
210 # pwd = getpwnam(user); ...
211 # ngroups = MAX_GROUPS;
212 # status = getgrouplist(user, pwd->pw_gid, groups, &ngroups); ...
213 # status = setgroups(ngroups, groups); ...
214 # status = setgid(pwd->pw_gid); ...
215 # status = setuid(pwd->pw_uid); ...
216 # however we cannot simulate that ourselves because if we did in this process then
217 # capsh could not be allowed to mess with caps any more
219 def getuid (slicename):
222 return pwd.getpwnam(slicename).pw_uid
227 cap_arg = '--drop='+drop_capabilities
230 if (args.nosliceuid):
231 # we still want to drop capabilities, but don't want to switch UIDs
232 exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--','--login',]+args.command_to_run
234 uid = getuid (slice_name)
236 print "lxcsu could not spot %s in /etc/passwd - exiting"%slice_name
238 exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--uid=%s'%uid,'--','--login',]+args.command_to_run
239 # once we can drop f12, it would be nicer to instead go for
240 # exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--user=%s'%slice_name,'--','--login',]+args.command_to_run
242 exec_args = [arch,'/usr/sbin/capsh','--','--login']+args.command_to_run
244 os.environ['SHELL'] = '/bin/sh'
245 os.environ['HOME'] = '/home/%s'%slice_name
246 os.environ['LD_PRELOAD'] = '/etc/planetlab/lib/bind_public.so'
247 os.chdir("/home/%s"%(slice_name))
248 if debug: print 'lxcsu:execv:','/usr/bin/setarch',exec_args
249 os.execv('/usr/bin/setarch',exec_args)
252 _,status = os.waitpid(pid,0)
253 exit(os.WEXITSTATUS(status))
255 if __name__ == '__main__':