2 # NEPI, a framework to manage network experiments
3 # Copyright (C) 2013 INRIA
5 # This program is free software: you can redistribute it and/or modify
6 # it under the terms of the GNU General Public License as published by
7 # the Free Software Foundation, either version 3 of the License, or
8 # (at your option) any later version.
10 # This program is distributed in the hope that it will be useful,
11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 # GNU General Public License for more details.
15 # You should have received a copy of the GNU General Public License
16 # along with this program. If not, see <http://www.gnu.org/licenses/>.
18 # Author: Alina Quereilhac <alina.quereilhac@inria.fr>
19 # Claudio Freire <claudio-daniel.freire@inria.fr>
36 logger = logging.getLogger("sshfuncs")
38 def log(msg, level, out = None, err = None):
40 msg += " - OUT: %s " % out
43 msg += " - ERROR: %s " % err
45 logger.log(level, msg)
48 if hasattr(os, "devnull"):
51 DEV_NULL = "/dev/null"
53 SHELL_SAFE = re.compile('^[-a-zA-Z0-9_=+:.,/]*$')
57 Special value that when given to rspawn in stderr causes stderr to
58 redirect to whatever stdout was redirected to.
63 Codes for status of remote spawned process
65 # Process is still running
71 # Process hasn't started running yet (this should be very rare)
74 hostbyname_cache = dict()
75 hostbyname_cache_lock = threading.Lock()
77 def gethostbyname(host):
78 global hostbyname_cache
79 global hostbyname_cache_lock
81 hostbyname = hostbyname_cache.get(host)
83 with hostbyname_cache_lock:
84 hostbyname = socket.gethostbyname(host)
85 hostbyname_cache[host] = hostbyname
87 msg = " Added hostbyname %s - %s " % (host, hostbyname)
88 log(msg, logging.DEBUG)
92 OPENSSH_HAS_PERSIST = None
94 def openssh_has_persist():
95 """ The ssh_config options ControlMaster and ControlPersist allow to
96 reuse a same network connection for multiple ssh sessions. In this
97 way limitations on number of open ssh connections can be bypassed.
98 However, older versions of openSSH do not support this feature.
99 This function is used to determine if ssh connection persist features
102 global OPENSSH_HAS_PERSIST
103 if OPENSSH_HAS_PERSIST is None:
104 proc = subprocess.Popen(["ssh","-v"],
105 stdout = subprocess.PIPE,
106 stderr = subprocess.STDOUT,
107 stdin = open("/dev/null","r") )
108 out,err = proc.communicate()
111 vre = re.compile(r'OpenSSH_(?:[6-9]|5[.][8-9]|5[.][1-9][0-9]|[1-9][0-9]).*', re.I)
112 OPENSSH_HAS_PERSIST = bool(vre.match(out))
113 return OPENSSH_HAS_PERSIST
115 def make_server_key_args(server_key, host, port):
116 """ Returns a reference to a temporary known_hosts file, to which
117 the server key has been added.
119 Make sure to hold onto the temp file reference until the process is
122 :param server_key: the server public key
123 :type server_key: str
125 :param host: the hostname
128 :param port: the ssh port
133 host = '%s:%s' % (host, str(port))
135 # Create a temporary server key file
136 tmp_known_hosts = tempfile.NamedTemporaryFile()
138 hostbyname = gethostbyname(host)
140 # Add the intended host key
141 tmp_known_hosts.write('%s,%s %s\n' % (host, hostbyname, server_key))
143 # If we're not in strict mode, add user-configured keys
144 if os.environ.get('NEPI_STRICT_AUTH_MODE',"").lower() not in ('1','true','on'):
145 user_hosts_path = '%s/.ssh/known_hosts' % (os.environ.get('HOME',""),)
146 if os.access(user_hosts_path, os.R_OK):
147 f = open(user_hosts_path, "r")
148 tmp_known_hosts.write(f.read())
151 tmp_known_hosts.flush()
153 return tmp_known_hosts
155 def make_control_path(agent, forward_x11):
156 ctrl_path = "/tmp/nepi_ssh"
164 ctrl_path += "-%r@%h:%p"
169 """ Escapes strings so that they are safe to use as command-line
171 if SHELL_SAFE.match(s):
172 # safe string - no escaping needed
175 # unsafe string - escape
177 if (32 <= ord(c) < 127 or c in ('\r','\n','\t')) and c not in ("'",'"'):
180 return "'$'\\x%02x''" % (ord(c),)
181 s = ''.join(map(escp,s))
184 def eintr_retry(func):
185 """Retries a function invocation when a EINTR occurs"""
187 @functools.wraps(func)
189 retry = kw.pop("_retry", False)
190 for i in xrange(0 if retry else 4):
192 return func(*p, **kw)
193 except (select.error, socket.error), args:
194 if args[0] == errno.EINTR:
199 if e.errno == errno.EINTR:
204 return func(*p, **kw)
207 def rexec(command, host, user,
220 err_on_timeout = True,
221 connect_timeout = 30,
225 strict_host_checking = True):
227 Executes a remote command, returns ((stdout,stderr),process)
230 tmp_known_hosts = None
232 hostip = gethostbyname(host)
236 # Don't bother with localhost. Makes test easier
237 '-o', 'NoHostAuthenticationForLocalhost=yes',
238 '-o', 'ConnectTimeout=%d' % (int(connect_timeout),),
239 '-o', 'ConnectionAttempts=3',
240 '-o', 'ServerAliveInterval=30',
241 '-o', 'TCPKeepAlive=yes',
242 '-o', 'Batchmode=yes',
243 '-l', user, hostip or host]
245 if persistent and openssh_has_persist():
247 '-o', 'ControlMaster=auto',
248 '-o', 'ControlPath=%s' % (make_control_path(agent, forward_x11),),
249 '-o', 'ControlPersist=60' ])
251 if not strict_host_checking:
252 # Do not check for Host key. Unsafe.
253 args.extend(['-o', 'StrictHostKeyChecking=no'])
257 proxycommand = 'ProxyCommand=ssh %s@%s -W %%h:%%p' % (gwuser, gw)
259 proxycommand = 'ProxyCommand=ssh %%r@%s -W %%h:%%p' % gw
260 args.extend(['-o', proxycommand])
266 args.append('-p%d' % port)
269 args.extend(('-i', identity))
279 # Create a temporary server key file
280 tmp_known_hosts = make_server_key_args(server_key, host, port)
281 args.extend(['-o', 'UserKnownHostsFile=%s' % (tmp_known_hosts.name,)])
284 command = "sudo " + command
288 for x in xrange(retry):
289 # connects to the remote host and starts a remote connection
290 proc = subprocess.Popen(args,
292 stdout = subprocess.PIPE,
293 stdin = subprocess.PIPE,
294 stderr = subprocess.PIPE)
296 # attach tempfile object to the process, to make sure the file stays
297 # alive until the process is finished with it
298 proc._known_hosts = tmp_known_hosts
300 # by default, rexec calls _communicate which will block
301 # until the process has exit. The argument block == False
302 # forces to rexec to return immediately, without blocking
305 out, err = _communicate(proc, stdin, timeout, err_on_timeout)
307 err = proc.stderr.read()
308 out = proc.stdout.read()
310 msg = " rexec - host %s - command %s " % (host, " ".join(args))
311 log(msg, logging.DEBUG, out, err)
316 if err.strip().startswith('ssh: ') or err.strip().startswith('mux_client_hello_exchange: '):
317 # SSH error, can safely retry
320 # Probably timed out or plain failed but can retry
325 msg = "SLEEPING %d ... ATEMPT %d - host %s - command %s " % (
326 t, x, host, " ".join(args))
327 log(msg, logging.DEBUG)
332 except RuntimeError, e:
333 msg = " rexec EXCEPTION - host %s - command %s - TIMEOUT -> %s" % (host, " ".join(args), e.args)
334 log(msg, logging.DEBUG, out, err)
340 return ((out, err), proc)
342 def rcopy(source, dest,
351 strict_host_checking = True):
353 Copies from/to remote sites.
355 Source and destination should have the user and host encoded
358 Source can be a list of files to copy to a single destination,
359 (in which case it is advised that the destination be a folder),
360 a single file in a string or a semi-colon separated list of files
364 # Parse destination as <user>@<server>:<path>
365 if isinstance(dest, str) and ':' in dest:
366 remspec, path = dest.split(':',1)
367 elif isinstance(source, str) and ':' in source:
368 remspec, path = source.split(':',1)
370 raise ValueError, "Both endpoints cannot be local"
371 user,host = remspec.rsplit('@',1)
374 tmp_known_hosts = None
376 args = ['scp', '-q', '-p', '-C',
377 # Speed up transfer using blowfish cypher specification which is
378 # faster than the default one (3des)
380 # Don't bother with localhost. Makes test easier
381 '-o', 'NoHostAuthenticationForLocalhost=yes',
382 '-o', 'ConnectTimeout=60',
383 '-o', 'ConnectionAttempts=3',
384 '-o', 'ServerAliveInterval=30',
385 '-o', 'TCPKeepAlive=yes' ]
388 args.append('-P%d' % port)
392 proxycommand = 'ProxyCommand=ssh %s@%s -W %%h:%%p' % (gwuser, gw)
394 proxycommand = 'ProxyCommand=ssh %%r@%s -W %%h:%%p' % gw
395 args.extend(['-o', proxycommand])
401 args.extend(('-i', identity))
404 # Create a temporary server key file
405 tmp_known_hosts = make_server_key_args(server_key, host, port)
406 args.extend(['-o', 'UserKnownHostsFile=%s' % (tmp_known_hosts.name,)])
408 if not strict_host_checking:
409 # Do not check for Host key. Unsafe.
410 args.extend(['-o', 'StrictHostKeyChecking=no'])
412 if openssh_has_persist():
414 '-o', 'ControlMaster=auto',
415 '-o', 'ControlPath=%s' % (make_control_path(agent, False),)
418 if isinstance(dest, str):
419 dest = map(str.strip, dest.split(";"))
421 if isinstance(source, str):
422 source = map(str.strip, source.split(";"))
428 for x in xrange(retry):
429 # connects to the remote host and starts a remote connection
430 proc = subprocess.Popen(args,
431 stdout = subprocess.PIPE,
432 stdin = subprocess.PIPE,
433 stderr = subprocess.PIPE)
435 # attach tempfile object to the process, to make sure the file stays
436 # alive until the process is finished with it
437 proc._known_hosts = tmp_known_hosts
440 (out, err) = proc.communicate()
441 eintr_retry(proc.wait)()
442 msg = " rcopy - host %s - command %s " % (host, " ".join(args))
443 log(msg, logging.DEBUG, out, err)
447 msg = "SLEEPING %d ... ATEMPT %d - host %s - command %s " % (
448 t, x, host, " ".join(args))
449 log(msg, logging.DEBUG)
455 except RuntimeError, e:
456 msg = " rcopy EXCEPTION - host %s - command %s - TIMEOUT -> %s" % (host, " ".join(args), e.args)
457 log(msg, logging.DEBUG, out, err)
463 return ((out, err), proc)
465 def rspawn(command, pidfile,
466 stdout = '/dev/null',
482 Spawn a remote command such that it will continue working asynchronously in
485 :param command: The command to run, it should be a single line.
488 :param pidfile: Path to a file where to store the pid and ppid of the
492 :param stdout: Path to file to redirect standard output.
493 The default value is /dev/null
496 :param stderr: Path to file to redirect standard error.
497 If the special STDOUT value is used, stderr will
498 be redirected to the same file as stdout
501 :param stdin: Path to a file with input to be piped into the command's standard input
504 :param home: Path to working directory folder.
505 It is assumed to exist unless the create_home flag is set.
508 :param create_home: Flag to force creation of the home folder before
510 :type create_home: bool
512 :param sudo: Flag forcing execution with sudo user
517 (stdout, stderr), process
519 Of the spawning process, which only captures errors at spawning time.
520 Usually only useful for diagnostics.
522 # Start process in a "daemonized" way, using nohup and heavy
523 # stdin/out redirection to avoid connection issues
527 stderr = ' ' + stderr
529 daemon_command = '{ { %(command)s > %(stdout)s 2>%(stderr)s < %(stdin)s & } ; echo $! 1 > %(pidfile)s ; }' % {
531 'pidfile' : shell_escape(pidfile),
537 cmd = "%(create)s%(gohome)s rm -f %(pidfile)s ; %(sudo)s nohup bash -c %(command)s " % {
538 'command' : shell_escape(daemon_command),
539 'sudo' : 'sudo -S' if sudo else '',
540 'pidfile' : shell_escape(pidfile),
541 'gohome' : 'cd %s ; ' % (shell_escape(home),) if home else '',
542 'create' : 'mkdir -p %s ; ' % (shell_escape(home),) if create_home and home else '',
545 (out,err),proc = rexec(
554 server_key = server_key,
559 raise RuntimeError, "Failed to set up application on host %s: %s %s" % (host, out,err,)
561 return ((out, err), proc)
574 Returns the pid and ppid of a process from a remote file where the
575 information was stored.
577 :param home: Path to directory where the pidfile is located
580 :param pidfile: Name of file containing the pid information
585 A (pid, ppid) tuple useful for calling rstatus and rkill,
586 or None if the pidfile isn't valid yet (can happen when process is staring up)
589 (out,err),proc = rexec(
590 "cat %(pidfile)s" % {
600 server_key = server_key
608 return map(int,out.strip().split(' ',1))
610 # Ignore, many ways to fail that don't matter that much
614 def rstatus(pid, ppid,
624 Returns a code representing the the status of a remote process
626 :param pid: Process id of the process
629 :param ppid: Parent process id of process
632 :rtype: int (One of NOT_STARTED, RUNNING, FINISHED)
635 (out,err),proc = rexec(
636 # Check only by pid. pid+ppid does not always work (especially with sudo)
637 " (( ps --pid %(pid)d -o pid | grep -c %(pid)d && echo 'wait') || echo 'done' ) | tail -n 1" % {
648 server_key = server_key
652 return ProcStatus.NOT_STARTED
656 if err.strip().find("Error, do this: mount -t proc none /proc") >= 0:
659 status = (out.strip() == 'wait')
661 return ProcStatus.NOT_STARTED
662 return ProcStatus.RUNNING if status else ProcStatus.FINISHED
677 Sends a kill signal to a remote process.
679 First tries a SIGTERM, and if the process does not end in 10 seconds,
682 :param pid: Process id of process to be killed
685 :param ppid: Parent process id of process to be killed
688 :param sudo: Flag indicating if sudo should be used to kill the process
692 subkill = "$(ps --ppid %(pid)d -o pid h)" % { 'pid' : pid }
694 SUBKILL="%(subkill)s" ;
695 %(sudo)s kill -- -%(pid)d $SUBKILL || /bin/true
696 %(sudo)s kill %(pid)d $SUBKILL || /bin/true
697 for x in 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 ; do
699 if [ `ps --pid %(pid)d -o pid | grep -c %(pid)d` == '0' ]; then
702 %(sudo)s kill -- -%(pid)d $SUBKILL || /bin/true
703 %(sudo)s kill %(pid)d $SUBKILL || /bin/true
707 if [ `ps --pid %(pid)d -o pid | grep -c %(pid)d` != '0' ]; then
708 %(sudo)s kill -9 -- -%(pid)d $SUBKILL || /bin/true
709 %(sudo)s kill -9 %(pid)d $SUBKILL || /bin/true
713 cmd = "( %s ) >/dev/null 2>/dev/null </dev/null &" % (cmd,)
715 (out,err),proc = rexec(
719 'sudo' : 'sudo -S' if sudo else '',
729 server_key = server_key
732 # wait, don't leave zombies around
735 return (out, err), proc
738 def _communicate(proc, input, timeout=None, err_on_timeout=True):
741 stdout = None # Return
742 stderr = None # Return
746 if timeout is not None:
747 timelimit = time.time() + timeout
748 killtime = timelimit + 4
749 bailtime = timelimit + 4
752 # Flush stdio buffer. This might block, if the user has
753 # been writing to .stdin in an uncontrolled fashion.
756 write_set.append(proc.stdin)
761 read_set.append(proc.stdout)
765 read_set.append(proc.stderr)
769 while read_set or write_set:
770 if timeout is not None:
771 curtime = time.time()
772 if timeout is None or curtime > timelimit:
773 if curtime > bailtime:
775 elif curtime > killtime:
776 signum = signal.SIGKILL
778 signum = signal.SIGTERM
780 os.kill(proc.pid, signum)
783 select_timeout = timelimit - curtime + 0.1
787 if select_timeout > 1.0:
791 rlist, wlist, xlist = select.select(read_set, write_set, [], select_timeout)
792 except select.error,e:
798 if not rlist and not wlist and not xlist and proc.poll() is not None:
799 # timeout and process exited, say bye
802 if proc.stdin in wlist:
803 # When select has indicated that the file is writable,
804 # we can write up to PIPE_BUF bytes without risk
805 # blocking. POSIX defines PIPE_BUF >= 512
806 bytes_written = os.write(proc.stdin.fileno(),
807 buffer(input, input_offset, 512))
808 input_offset += bytes_written
810 if input_offset >= len(input):
812 write_set.remove(proc.stdin)
814 if proc.stdout in rlist:
815 data = os.read(proc.stdout.fileno(), 1024)
818 read_set.remove(proc.stdout)
821 if proc.stderr in rlist:
822 data = os.read(proc.stderr.fileno(), 1024)
825 read_set.remove(proc.stderr)
828 # All data exchanged. Translate lists into strings.
829 if stdout is not None:
830 stdout = ''.join(stdout)
831 if stderr is not None:
832 stderr = ''.join(stderr)
834 # Translate newlines, if requested. We cannot let the file
835 # object do the translation: It is based on stdio, which is
836 # impossible to combine with select (unless forcing no
838 if proc.universal_newlines and hasattr(file, 'newlines'):
840 stdout = proc._translate_newlines(stdout)
842 stderr = proc._translate_newlines(stderr)
844 if killed and err_on_timeout:
845 errcode = proc.poll()
846 raise RuntimeError, ("Operation timed out", errcode, stdout, stderr)
852 return (stdout, stderr)