git.onelab.eu Git - lxc-userspace.git/blob

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 #################### should go into a separate libvirtsystemd.py

15 # but we want to keep packaging simple for now

16

17 # reproducing libvirt's systemd-oriented escaping mechanism

18 # http://code.metager.de/source/xref/lib/virt/src/util/virsystemd.c

19 # (see original code at the end of this file)

20

21 def virSystemdEscapeName (name):

22 result=''

23 def ESCAPE(c,s):

24 # replace hex's output '0x..' into '\x..'

25 return s+hex(ord(c)).replace('0','\\',1)

26 VALID_CHARS = \

27 "0123456789" + \

28 "abcdefghijklmnopqrstuvwxyz" + \

29 "ABCDEFGHIJKLMNOPQRSTUVWXYZ" + \

30 ":-_.\\"

31 for c in name:

32 if c=='/':

33 result += '-'

34 elif c in '-\\' or c not in VALID_CHARS:

35 result=ESCAPE(c,result)

36 else:

37 result += c

38 return result

39

40 #35static void virSystemdEscapeName(virBufferPtr buf,

41 #36 const char *name)

42 #37{

43 #38 static const char hextable[16] = "0123456789abcdef";

44 #39

45 #40#define ESCAPE(c) \

46 #41 do { \

47 #42 virBufferAddChar(buf, '\\'); \

48 #43 virBufferAddChar(buf, 'x'); \

49 #44 virBufferAddChar(buf, hextable[(c >> 4) & 15]); \

50 #45 virBufferAddChar(buf, hextable[c & 15]); \

51 #46 } while (0)

52 #47

53 #48#define VALID_CHARS \

54 #49 "0123456789" \

55 #50 "abcdefghijklmnopqrstuvwxyz" \

56 #51 "ABCDEFGHIJKLMNOPQRSTUVWXYZ" \

57 #52 ":-_.\\"

58 #53

59 #54 if (*name == '.') {

60 #55 ESCAPE(*name);

61 #56 name++;

62 #57 }

63 #58

64 #59 while (*name) {

65 #60 if (*name == '/')

66 #61 virBufferAddChar(buf, '-');

67 #62 else if (*name == '-' ||

68 #63 *name == '\\' ||

69 #64 !strchr(VALID_CHARS, *name))

70 #65 ESCAPE(*name);

71 #66 else

72 #67 virBufferAddChar(buf, *name);

73 #68 name++;

74 #69 }

75 #70

76 #71#undef ESCAPE

77 #72#undef VALID_CHARS

78 #73}

79

80 def virSystemdMakeScopeName (name, drivername, partition):

81 result=''

82 result += virSystemdEscapeName (partition)

83 result += '-'

84 result += virSystemdEscapeName (drivername)

85 result += '\\x2d'

86 result += virSystemdEscapeName (name)

87 result += '.scope'

88 return result

89

90 #76char *virSystemdMakeScopeName(const char *name,

91 #77 const char *drivername,

92 #78 const char *partition)

93 #79{

94 #80 virBuffer buf = VIR_BUFFER_INITIALIZER;

95 #81

96 #82 if (*partition == '/')

97 #83 partition++;

98 #84

99 #85 virSystemdEscapeName(&buf, partition);

100 #86 virBufferAddChar(&buf, '-');

101 #87 virSystemdEscapeName(&buf, drivername);

102 #88 virBufferAddLit(&buf, "\\x2d");

103 #89 virSystemdEscapeName(&buf, name);

104 #90 virBufferAddLit(&buf, ".scope");

105 #91

106 #92 if (virBufferError(&buf)) {

107 #93 virReportOOMError();

108 #94 return NULL;

109 #95 }

110 #96

111 #97 return virBufferContentAndReset(&buf);

112 #98}

113

114 ### our own additions

115 import os.path

116 def find_first_dir (candidates):

117 for candidate in candidates:

118 if os.path.isdir(candidate): return candidate

119 raise Exception,"Cannot find valid dir among %s"%'\n'.join(candidates)

120

121 def find_sysfs_scope (subsystem, slice_name):

122 subsystem1=subsystem

123 subsystem2=subsystem

124 if subsystem=='cpuacct':

125 subsystem2='cpu,cpuacct'

126 candidates = [

127 # for f16 and our locally brewed libvirt 1.0.4

128 "/sys/fs/cgroup/%s/libvirt/lxc/%s"%(subsystem1, slice_name),

129 # f20 and libvirt 1.1.3

130 "/sys/fs/cgroup/%s/machine.slice/%s"%(subsystem2,

131 virSystemdMakeScopeName(slice_name,'lxc','machine')),

132 ]

133 return find_first_dir (candidates)

134

135 #################### end of libvirtsystemd.py

136

137 def getarch(f):

138 output = os.popen('readelf -h %s 2>&1'%f).readlines()

139 classlines = [x for x in output if ('Class' in x.split(':')[0])]

140 line = classlines[0]

141 c = line.split(':')[1]

142 if ('ELF64' in c):

143 return 'x86_64'

144 elif ('ELF32' in c):

145 return 'i686'

146 else:

147 raise Exception('Could not determine architecture')

148

149 def umount(fs_dir, opts=''):

150 output = os.popen('/bin/umount %s %s 2>&1'%(opts, fs_dir)).read()

151 return ('device is busy' not in output)

152

153 def main ():

154 parser = ArgumentParser()

155 parser.add_argument("-n", "--nonet",

156 action="store_true", dest="netns", default=False,

157 help="Don't enter network namespace")

158 parser.add_argument("-m", "--nomnt",

159 action="store_true", dest="mntns", default=False,

160 help="Don't enter mount namespace")

161 parser.add_argument("-p", "--nopid",

162 action="store_true", dest="pidns", default=False,

163 help="Don't enter pid namespace")

164 parser.add_argument("-r", "--root",

165 action="store_true", dest="root", default=False,

166 help="Enter as root: be careful")

167 parser.add_argument("-i","--internal",

168 action="store_true", dest="internal", default=False,

169 help="does *not* prepend '-- -c' to arguments - or invoke lxcsu-internal")

170 parser.add_argument("-d","--debug",

171 action='store_true', dest='debug', default=False,

172 help="debug option")

173 parser.add_argument("-s","--nosliceuid",

174 action='store_true', dest="nosliceuid", default=False,

175 help="do not change to slice uid inside of slice")

176 parser.add_argument("-o","--noslicehome",

177 action='store_true', dest="noslicehome", default=False,

178 help="do not change to slice home directory inside of slice")

179

180 if os.path.exists("/etc/lxcsu_default"):

181 defaults = parser.parse_args(file("/etc/lxcsu_default","r").read().split())

182 parser.set_defaults(**defaults.__dict__)

183

184 parser.add_argument ("slice_name")

185 parser.add_argument ("command_to_run",nargs="*")

186

187 args = parser.parse_args()

188 slice_name=args.slice_name

189

190 # unless we run the symlink 'lxcsu-internal', or we specify the -i option, prepend '--' '-c'

191 if sys.argv[0].find('internal')>=0: args.internal=True

192

193 if len(args.command_to_run)>0 and (args.command_to_run[0] == "/sbin/service"):

194 # A quick hack to support nodemanager interfaces.py when restarting

195 # networking in a slice.

196 args.nosliceuid = True

197

198 # plain lxcsu

199 if not args.internal:

200 # no command given: enter interactive shell

201 if not args.command_to_run: args.command_to_run=['/bin/sh']

202 args.command_to_run = [ '-c' ] + [" ".join(args.command_to_run)]

203

204 # support for either setting debug at the top of this file, or on the command-line

205 if args.debug:

206 global debug

207 debug=True

208

209 try:

210 cmd = '/usr/bin/virsh --connect lxc:/// domid %s'%slice_name

211 pidnum = int(os.popen(cmd).read().rstrip())

212 except:

213 print "Error finding slice %s"%slice_name

214 exit(1)

215

216 pid = '%s'%pidnum

217 if debug: print "Found pidnum",pidnum

218 cmdline = open('/proc/%s/cmdline'%pidnum).read().rstrip('\n\x00')

219 arch = getarch('/proc/%s/exe'%pid)

220

221 if (not pid):

222 print "Not started: %s"%slice_name

223 exit(1)

224

225 if arch is None:

226 arch = 'x86_64'

227

228 # Set sysctls specific to slice

229 sysctls = []

230 sysctl_dir = '/etc/planetlab/vsys-attributes/%s'%slice_name

231 if (os.access(sysctl_dir,0)):

232 entries = os.listdir(sysctl_dir)

233 for e in entries:

234 prefix = 'vsys_sysctl.'

235 if (e.startswith(prefix)):

236 sysctl_file = '/'.join([sysctl_dir,e])

237 sysctl_name = e[len(prefix):]

238 sysctl_val = open(sysctl_file).read()

239 sysctls.append((sysctl_file, sysctl_name, sysctl_val))

240

241 # Enter cgroups

242 try:

243 for subsystem in ['cpuset','memory','blkio']:

244 open( find_sysfs_scope (subsystem, slice_name)+"/tasks", 'w').write(str(os.getpid()))

245

246 except Exception,e:

247 if debug: print e

248 print "Error assigning resources: %s"%slice_name

249 exit(1)

250

251 try:

252 open ( find_sysfs_scope ('cpuacct', slice_name)+"/tasks", 'w').write(str(os.getpid()))

253 except Exception,e:

254 if debug: print e

255 print "Error assigning cpuacct: %s" % slice_name

256 exit(1)

257

258 # If the slice is frozen, then we'll get an EBUSY when trying to write to the task

259 # list for the freezer cgroup. Since the user couldn't do anything anyway, it's best

260 # in this case to error out the shell. (an alternative would be to un-freeze it,

261 # add the task, and re-freeze it)

262 try:

263 f=open( find_sysfs_scope ('freezer', slice_name)+"/tasks", 'w')

264 f.write(str(os.getpid()))

265 # note: we need to call f.close() explicitly, or we'll get an exception in

266 # the object destructor, which will not be caught

267 f.close()

268 except Exception,e:

269 if debug: print e

270 print "Error adding task to freezer cgroup. Slice is probably frozen: %s" % slice_name

271 exit(1)

272

273 setns.chcontext('/proc/%s/ns/uts'%pid)

274 setns.chcontext('/proc/%s/ns/ipc'%pid)

275

276 if (not args.pidns):

277 setns.chcontext('/proc/%s/ns/pid'%pid)

278

279 if (not args.netns):

280 setns.chcontext('/proc/%s/ns/net'%pid)

281

282 if (not args.mntns):

283 setns.chcontext('/proc/%s/ns/mnt'%pid)

284

285 proc_mounted = False

286 if (not os.access('/proc/self',0)):

287 proc_mounted = True

288 setns.proc_mount()

289

290 for (sysctl_file, sysctl_name, sysctl_val) in sysctls:

291 for fn in ["/sbin/sysctl", "/usr/sbin/sysctl", "/bin/sysctl", "/usr/bin/sysctl"]:

292 if os.path.exists(fn):

293 os.system('%s -w %s=%s >/dev/null 2>&1'%(fn, sysctl_name,sysctl_val))

294 break

295 else:

296 print "Error: image does not have a sysctl binary"

297

298 # cgroups is not yet LXC-safe, so we need to use the coarse grained access control

299 # strategy of unmounting the filesystem

300

301 umount_result = True

302 for subsystem in ['cpuset','cpu,cpuacct','memory','devices','freezer','net_cls','blkio','perf_event','systemd']:

303 fs_path = '/sys/fs/cgroup/%s'%subsystem

304 if (not umount(fs_path,'-l')):

305 pass

306 # Leaving these comments for historical reference

307 #print "Error disabling cgroup access"

308 #exit(1) - Don't need this because failure here implies failure in the call to umount /sys/fs/cgroup

309

310 if (not umount('/sys/fs/cgroup')):

311 print "Error disabling cgroup access"

312 exit(1)

313

314 pid = os.fork()

315

316 # capsh has a --user option starting with f14

317 # so if only for f12 we need to fake this one

318 #

319 # capsh.c does essentially the following when invoked with --user:

320 # pwd = getpwnam(user); ...

321 # ngroups = MAX_GROUPS;

322 # status = getgrouplist(user, pwd->pw_gid, groups, &ngroups); ...

323 # status = setgroups(ngroups, groups); ...

324 # status = setgid(pwd->pw_gid); ...

325 # status = setuid(pwd->pw_uid); ...

326 # however we cannot simulate that ourselves because if we did in this process then

327 # capsh could not be allowed to mess with caps any more

328

329 def getuid (slicename):

330 import pwd

331 try:

332 return pwd.getpwnam(slicename).pw_uid

333 except:

334 return

335

336 if (pid == 0):

337 cap_arg = '--drop='+drop_capabilities

338

339 if (not args.root):

340 if (args.nosliceuid):

341 # we still want to drop capabilities, but don't want to switch UIDs

342 exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--','--login',]+args.command_to_run

343 else:

344 uid = getuid (slice_name)

345 if not uid:

346 print "lxcsu could not spot %s in /etc/passwd - exiting"%slice_name

347 exit(1)

348 exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--uid=%s'%uid,'--','--login',]+args.command_to_run

349 # once we can drop f12, it would be nicer to instead go for

350 # exec_args = [arch,'/usr/sbin/capsh',cap_arg,'--user=%s'%slice_name,'--','--login',]+args.command_to_run

351 else:

352 exec_args = [arch,'/usr/sbin/capsh','--','--login']+args.command_to_run

353

354 os.environ['SHELL'] = '/bin/sh'

355 if os.path.exists('/etc/planetlab/lib/bind_public.so'):

356 os.environ['LD_PRELOAD'] = '/etc/planetlab/lib/bind_public.so'

357 if not args.noslicehome:

358 os.environ['HOME'] = '/home/%s'%slice_name

359 os.chdir("/home/%s"%(slice_name))

360 if debug: print 'lxcsu:execv:','/usr/bin/setarch',exec_args

361 os.execv('/usr/bin/setarch',exec_args)

362 else:

363 setns.proc_umount()

364 _,status = os.waitpid(pid,0)

365 exit(os.WEXITSTATUS(status))

366

367 if __name__ == '__main__':

368 main()