8 from argparse import ArgumentParser
10 drop_capabilities='cap_sys_admin,cap_sys_boot,cap_sys_module'
12 # can set to True here, but also use the -d option
16 output = os.popen('readelf -h %s 2>&1'%f).readlines()
17 classlines = [x for x in output if ('Class' in x.split(':')[0])]
19 c = line.split(':')[1]
25 raise Exception('Could not determine architecture')
27 def get_cgroup_subdirs_for_pid(pid):
28 cgroup_info_file = '/proc/%s/cgroup'%pid
29 cgroup_lines = open(cgroup_info_file).read().splitlines()
32 for line in cgroup_lines:
34 _, cgroup_name, subdir = line.split(':')
35 subdirs[cgroup_name] = subdir
37 print "Error reading cgroup info: %s"%str(e)
43 def umount(fs_dir, opts=''):
44 output = os.popen('/bin/umount %s %s 2>&1'%(opts, fs_dir)).read()
45 return ('device is busy' not in output)
48 parser = ArgumentParser()
49 parser.add_argument("-n", "--nonet",
50 action="store_true", dest="netns", default=False,
51 help="Don't enter network namespace")
52 parser.add_argument("-m", "--nomnt",
53 action="store_true", dest="mntns", default=False,
54 help="Don't enter mount namespace")
55 parser.add_argument("-p", "--nopid",
56 action="store_true", dest="pidns", default=False,
57 help="Don't enter pid namespace")
58 parser.add_argument("-r", "--root",
59 action="store_true", dest="root", default=False,
60 help="Enter as root: be careful")
61 parser.add_argument("-i","--internal",
62 action="store_true", dest="internal", default=False,
63 help="does *not* prepend '-- -c' to arguments - or invoke lxcsu-internal")
64 parser.add_argument("-d","--debug",
65 action='store_true', dest='debug', default=False,
67 parser.add_argument("-s","--nosliceuid",
68 action='store_true', dest="nosliceuid", default=False,
69 help="do not change to slice uid inside of slice")
70 parser.add_argument("-o","--noslicehome",
71 action='store_true', dest="noslicehome", default=False,
72 help="do not change to slice home directory inside of slice")
74 if os.path.exists("/etc/lxcsu_default"):
75 defaults = parser.parse_args(file("/etc/lxcsu_default","r").read().split())
76 parser.set_defaults(**defaults.__dict__)
78 parser.add_argument ("slice_name")
79 parser.add_argument ("command_to_run",nargs="*")
81 args = parser.parse_args()
82 slice_name=args.slice_name
84 # unless we run the symlink 'lxcsu-internal', or we specify the -i option, prepend '--' '-c'
85 if sys.argv[0].find('internal')>=0: args.internal=True
87 if len(args.command_to_run)>0 and (args.command_to_run[0] == "/sbin/service"):
88 # A quick hack to support nodemanager interfaces.py when restarting
89 # networking in a slice.
90 args.nosliceuid = True
94 # no command given: enter interactive shell
95 if not args.command_to_run: args.command_to_run=['/bin/sh']
96 args.command_to_run = [ '-c' ] + [" ".join(args.command_to_run)]
98 # support for either setting debug at the top of this file, or on the command-line
104 cmd = '/usr/bin/virsh --connect lxc:/// domid %s'%slice_name
105 pidnum = int(os.popen(cmd).read().rstrip())
106 # locate the pid for the - expected - single child, that would be the init for that VM
107 child_pid = open("/proc/%s/task/%s/children"%(pidnum,pidnum)).read().strip()
110 print "Domain %s not found"%slice_name
114 if debug: print "Found pidnum",pidnum,'child_pid=',child_pid
115 cmdline = open('/proc/%s/cmdline'%pidnum).read().rstrip('\n\x00')
116 arch = getarch('/proc/%s/exe'%pid)
119 print "Domain %s not started"%slice_name
125 # Set sysctls specific to slice
127 sysctl_dir = '/etc/planetlab/vsys-attributes/%s'%slice_name
128 if (os.access(sysctl_dir,0)):
129 entries = os.listdir(sysctl_dir)
131 prefix = 'vsys_sysctl.'
132 if (e.startswith(prefix)):
133 sysctl_file = '/'.join([sysctl_dir,e])
134 sysctl_name = e[len(prefix):]
135 sysctl_val = open(sysctl_file).read()
136 sysctls.append((sysctl_file, sysctl_name, sysctl_val))
138 subdirs = get_cgroup_subdirs_for_pid(pid)
139 sysfs_root = '/sys/fs/cgroup'
141 # If the slice is frozen, then we'll get an EBUSY when trying to write to the task
142 # list for the freezer cgroup. Since the user couldn't do anything anyway, it's best
143 # in this case to error out the shell. (an alternative would be to un-freeze it,
144 # add the task, and re-freeze it)
147 for subsystem in ['cpuset','memory','blkio','cpuacct','cpuacct,cpu','freezer']:
149 current_cgroup = subsystem
151 # There seems to be a bug in the cgroup schema: cpuacct,cpu can become cpu,cpuacct
152 # We need to handle both
155 subsystem_comps = subsystem.split(',')
156 subsystem_comps.reverse()
157 subsystem_alt = ','.join(subsystem_comps)
158 tasks_path_alt = [sysfs_root, subsystem_alt, subdirs[subsystem], 'tasks']
162 tasks_path = [sysfs_root,subsystem,subdirs[subsystem],'tasks']
163 tasks_path_str = '/'.join(tasks_path)
166 f = open(tasks_path_str, 'w')
168 tasks_path_alt_str = '/'.join(tasks_path_alt)
169 f = open(tasks_path_alt_str, 'w')
171 f.write(str(os.getpid()))
172 if (subsystem=='freezer'):
176 if (not subdirs.has_key(subsystem)):
180 print "Error assigning cgroup %s (%s) for slice %s"%(current_cgroup,pid, slice_name)
184 def chcontext (path):
185 retcod = setns.chcontext (path)
187 print 'WARNING - setns(%s)=>%s (ignored)'%(path,retcod)
190 # Use child_pid and not pid to locate reference namespaces
191 ref_ns = "/proc/%s/ns/"%child_pid
192 chcontext(ref_ns+'uts')
193 chcontext(ref_ns+'ipc')
196 chcontext(ref_ns+'pid')
199 chcontext(ref_ns+'net')
202 chcontext(ref_ns+'mnt')
205 if (not os.access('/proc/self',0)):
209 for (sysctl_file, sysctl_name, sysctl_val) in sysctls:
210 for fn in ["/sbin/sysctl", "/usr/sbin/sysctl", "/bin/sysctl", "/usr/bin/sysctl"]:
211 if os.path.exists(fn):
212 os.system('%s -w %s=%s >/dev/null 2>&1'%(fn, sysctl_name,sysctl_val))
215 print "Error: image does not have a sysctl binary"
217 # cgroups is not yet LXC-safe, so we need to use the coarse grained access control
218 # strategy of unmounting the filesystem
221 for subsystem in ['cpuset','cpu,cpuacct','memory','devices','freezer','net_cls','blkio','perf_event','systemd']:
222 fs_path = '/sys/fs/cgroup/%s'%subsystem
223 if (not umount(fs_path,'-l')):
224 print 'WARNING - umount failed (ignored) with path=',fs_path
226 # Leaving these comments for historical reference
227 #print "Error disabling cgroup access"
228 #exit(1) - Don't need this because failure here implies failure in the call to umount /sys/fs/cgroup
230 if (not umount('/sys/fs/cgroup')):
231 print "Error disabling cgroup access"
236 # capsh has a --user option starting with f14
237 # so if only for f12 we need to fake this one
239 # capsh.c does essentially the following when invoked with --user:
240 # pwd = getpwnam(user); ...
241 # ngroups = MAX_GROUPS;
242 # status = getgrouplist(user, pwd->pw_gid, groups, &ngroups); ...
243 # status = setgroups(ngroups, groups); ...
244 # status = setgid(pwd->pw_gid); ...
245 # status = setuid(pwd->pw_uid); ...
246 # however we cannot simulate that ourselves because if we did in this process then
247 # capsh could not be allowed to mess with caps any more
249 def getuid (slicename):
252 return pwd.getpwnam(slicename).pw_uid
257 cap_arg = '--drop='+drop_capabilities
260 if (args.nosliceuid):
261 # we still want to drop capabilities, but don't want to switch UIDs
262 exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--','--login',]+args.command_to_run
264 uid = getuid (slice_name)
266 print "lxcsu could not spot %s in /etc/passwd - exiting"%slice_name
268 exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--uid=%s'%uid,'--','--login',]+args.command_to_run
269 # once we can drop f12, it would be nicer to instead go for
270 # exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--user=%s'%slice_name,'--','--login',]+args.command_to_run
272 exec_args = [arch,'/usr/sbin/capsh','--','--login']+args.command_to_run
274 os.environ['SHELL'] = '/bin/sh'
275 if os.path.exists('/etc/planetlab/lib/bind_public.so'):
276 os.environ['LD_PRELOAD'] = '/etc/planetlab/lib/bind_public.so'
277 if not args.noslicehome:
278 os.environ['HOME'] = '/home/%s'%slice_name
279 os.chdir("/home/%s"%(slice_name))
280 if debug: print 'lxcsu:execv:','/usr/bin/setarch',exec_args
281 os.execv('/usr/bin/setarch',exec_args)
284 _,status = os.waitpid(pid,0)
285 exit(os.WEXITSTATUS(status))
287 if __name__ == '__main__':