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 string or a string list of colon-separeted files.
363 # Parse destination as <user>@<server>:<path>
364 if isinstance(dest, str) and ':' in dest:
365 remspec, path = dest.split(':',1)
366 elif isinstance(source, str) and ':' in source:
367 remspec, path = source.split(':',1)
369 raise ValueError, "Both endpoints cannot be local"
370 user,host = remspec.rsplit('@',1)
373 tmp_known_hosts = None
375 args = ['scp', '-q', '-p', '-C',
376 # Speed up transfer using blowfish cypher specification which is
377 # faster than the default one (3des)
379 # Don't bother with localhost. Makes test easier
380 '-o', 'NoHostAuthenticationForLocalhost=yes',
381 '-o', 'ConnectTimeout=60',
382 '-o', 'ConnectionAttempts=3',
383 '-o', 'ServerAliveInterval=30',
384 '-o', 'TCPKeepAlive=yes' ]
387 args.append('-P%d' % port)
391 proxycommand = 'ProxyCommand=ssh %s@%s -W %%h:%%p' % (gwuser, gw)
393 proxycommand = 'ProxyCommand=ssh %%r@%s -W %%h:%%p' % gw
394 args.extend(['-o', proxycommand])
400 args.extend(('-i', identity))
403 # Create a temporary server key file
404 tmp_known_hosts = make_server_key_args(server_key, host, port)
405 args.extend(['-o', 'UserKnownHostsFile=%s' % (tmp_known_hosts.name,)])
407 if not strict_host_checking:
408 # Do not check for Host key. Unsafe.
409 args.extend(['-o', 'StrictHostKeyChecking=no'])
411 if openssh_has_persist():
413 '-o', 'ControlMaster=auto',
414 '-o', 'ControlPath=%s' % (make_control_path(agent, False),)
417 if isinstance(dest, str):
418 dest = map(str.strip, dest.split(";"))
420 if isinstance(source, str):
421 source = map(str.strip, source.split(";"))
427 for x in xrange(retry):
428 # connects to the remote host and starts a remote connection
429 proc = subprocess.Popen(args,
430 stdout = subprocess.PIPE,
431 stdin = subprocess.PIPE,
432 stderr = subprocess.PIPE)
434 # attach tempfile object to the process, to make sure the file stays
435 # alive until the process is finished with it
436 proc._known_hosts = tmp_known_hosts
439 (out, err) = proc.communicate()
440 eintr_retry(proc.wait)()
441 msg = " rcopy - host %s - command %s " % (host, " ".join(args))
442 log(msg, logging.DEBUG, out, err)
446 msg = "SLEEPING %d ... ATEMPT %d - host %s - command %s " % (
447 t, x, host, " ".join(args))
448 log(msg, logging.DEBUG)
454 except RuntimeError, e:
455 msg = " rcopy EXCEPTION - host %s - command %s - TIMEOUT -> %s" % (host, " ".join(args), e.args)
456 log(msg, logging.DEBUG, out, err)
462 return ((out, err), proc)
464 def rspawn(command, pidfile,
465 stdout = '/dev/null',
481 Spawn a remote command such that it will continue working asynchronously in
484 :param command: The command to run, it should be a single line.
487 :param pidfile: Path to a file where to store the pid and ppid of the
491 :param stdout: Path to file to redirect standard output.
492 The default value is /dev/null
495 :param stderr: Path to file to redirect standard error.
496 If the special STDOUT value is used, stderr will
497 be redirected to the same file as stdout
500 :param stdin: Path to a file with input to be piped into the command's standard input
503 :param home: Path to working directory folder.
504 It is assumed to exist unless the create_home flag is set.
507 :param create_home: Flag to force creation of the home folder before
509 :type create_home: bool
511 :param sudo: Flag forcing execution with sudo user
516 (stdout, stderr), process
518 Of the spawning process, which only captures errors at spawning time.
519 Usually only useful for diagnostics.
521 # Start process in a "daemonized" way, using nohup and heavy
522 # stdin/out redirection to avoid connection issues
526 stderr = ' ' + stderr
528 daemon_command = '{ { %(command)s > %(stdout)s 2>%(stderr)s < %(stdin)s & } ; echo $! 1 > %(pidfile)s ; }' % {
530 'pidfile' : shell_escape(pidfile),
536 cmd = "%(create)s%(gohome)s rm -f %(pidfile)s ; %(sudo)s nohup bash -c %(command)s " % {
537 'command' : shell_escape(daemon_command),
538 'sudo' : 'sudo -S' if sudo else '',
539 'pidfile' : shell_escape(pidfile),
540 'gohome' : 'cd %s ; ' % (shell_escape(home),) if home else '',
541 'create' : 'mkdir -p %s ; ' % (shell_escape(home),) if create_home and home else '',
544 (out,err),proc = rexec(
553 server_key = server_key,
558 raise RuntimeError, "Failed to set up application on host %s: %s %s" % (host, out,err,)
560 return ((out, err), proc)
573 Returns the pid and ppid of a process from a remote file where the
574 information was stored.
576 :param home: Path to directory where the pidfile is located
579 :param pidfile: Name of file containing the pid information
584 A (pid, ppid) tuple useful for calling rstatus and rkill,
585 or None if the pidfile isn't valid yet (can happen when process is staring up)
588 (out,err),proc = rexec(
589 "cat %(pidfile)s" % {
599 server_key = server_key
607 return map(int,out.strip().split(' ',1))
609 # Ignore, many ways to fail that don't matter that much
613 def rstatus(pid, ppid,
623 Returns a code representing the the status of a remote process
625 :param pid: Process id of the process
628 :param ppid: Parent process id of process
631 :rtype: int (One of NOT_STARTED, RUNNING, FINISHED)
634 (out,err),proc = rexec(
635 # Check only by pid. pid+ppid does not always work (especially with sudo)
636 " (( ps --pid %(pid)d -o pid | grep -c %(pid)d && echo 'wait') || echo 'done' ) | tail -n 1" % {
647 server_key = server_key
651 return ProcStatus.NOT_STARTED
655 if err.strip().find("Error, do this: mount -t proc none /proc") >= 0:
658 status = (out.strip() == 'wait')
660 return ProcStatus.NOT_STARTED
661 return ProcStatus.RUNNING if status else ProcStatus.FINISHED
676 Sends a kill signal to a remote process.
678 First tries a SIGTERM, and if the process does not end in 10 seconds,
681 :param pid: Process id of process to be killed
684 :param ppid: Parent process id of process to be killed
687 :param sudo: Flag indicating if sudo should be used to kill the process
691 subkill = "$(ps --ppid %(pid)d -o pid h)" % { 'pid' : pid }
693 SUBKILL="%(subkill)s" ;
694 %(sudo)s kill -- -%(pid)d $SUBKILL || /bin/true
695 %(sudo)s kill %(pid)d $SUBKILL || /bin/true
696 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
698 if [ `ps --pid %(pid)d -o pid | grep -c %(pid)d` == '0' ]; then
701 %(sudo)s kill -- -%(pid)d $SUBKILL || /bin/true
702 %(sudo)s kill %(pid)d $SUBKILL || /bin/true
706 if [ `ps --pid %(pid)d -o pid | grep -c %(pid)d` != '0' ]; then
707 %(sudo)s kill -9 -- -%(pid)d $SUBKILL || /bin/true
708 %(sudo)s kill -9 %(pid)d $SUBKILL || /bin/true
712 cmd = "( %s ) >/dev/null 2>/dev/null </dev/null &" % (cmd,)
714 (out,err),proc = rexec(
718 'sudo' : 'sudo -S' if sudo else '',
728 server_key = server_key
731 # wait, don't leave zombies around
734 return (out, err), proc
737 def _communicate(proc, input, timeout=None, err_on_timeout=True):
740 stdout = None # Return
741 stderr = None # Return
745 if timeout is not None:
746 timelimit = time.time() + timeout
747 killtime = timelimit + 4
748 bailtime = timelimit + 4
751 # Flush stdio buffer. This might block, if the user has
752 # been writing to .stdin in an uncontrolled fashion.
755 write_set.append(proc.stdin)
760 read_set.append(proc.stdout)
764 read_set.append(proc.stderr)
768 while read_set or write_set:
769 if timeout is not None:
770 curtime = time.time()
771 if timeout is None or curtime > timelimit:
772 if curtime > bailtime:
774 elif curtime > killtime:
775 signum = signal.SIGKILL
777 signum = signal.SIGTERM
779 os.kill(proc.pid, signum)
782 select_timeout = timelimit - curtime + 0.1
786 if select_timeout > 1.0:
790 rlist, wlist, xlist = select.select(read_set, write_set, [], select_timeout)
791 except select.error,e:
797 if not rlist and not wlist and not xlist and proc.poll() is not None:
798 # timeout and process exited, say bye
801 if proc.stdin in wlist:
802 # When select has indicated that the file is writable,
803 # we can write up to PIPE_BUF bytes without risk
804 # blocking. POSIX defines PIPE_BUF >= 512
805 bytes_written = os.write(proc.stdin.fileno(),
806 buffer(input, input_offset, 512))
807 input_offset += bytes_written
809 if input_offset >= len(input):
811 write_set.remove(proc.stdin)
813 if proc.stdout in rlist:
814 data = os.read(proc.stdout.fileno(), 1024)
817 read_set.remove(proc.stdout)
820 if proc.stderr in rlist:
821 data = os.read(proc.stderr.fileno(), 1024)
824 read_set.remove(proc.stderr)
827 # All data exchanged. Translate lists into strings.
828 if stdout is not None:
829 stdout = ''.join(stdout)
830 if stderr is not None:
831 stderr = ''.join(stderr)
833 # Translate newlines, if requested. We cannot let the file
834 # object do the translation: It is based on stdio, which is
835 # impossible to combine with select (unless forcing no
837 if proc.universal_newlines and hasattr(file, 'newlines'):
839 stdout = proc._translate_newlines(stdout)
841 stderr = proc._translate_newlines(stderr)
843 if killed and err_on_timeout:
844 errcode = proc.poll()
845 raise RuntimeError, ("Operation timed out", errcode, stdout, stderr)
851 return (stdout, stderr)