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>
21 ## TODO: This code needs reviewing !!!
38 logger = logging.getLogger("sshfuncs")
40 def log(msg, level, out = None, err = None):
42 msg += " - OUT: %s " % out
45 msg += " - ERROR: %s " % err
47 logger.log(level, msg)
50 if hasattr(os, "devnull"):
53 DEV_NULL = "/dev/null"
55 SHELL_SAFE = re.compile('^[-a-zA-Z0-9_=+:.,/]*$')
59 Special value that when given to rspawn in stderr causes stderr to
60 redirect to whatever stdout was redirected to.
65 Codes for status of remote spawned process
67 # Process is still running
73 # Process hasn't started running yet (this should be very rare)
76 hostbyname_cache = dict()
77 hostbyname_cache_lock = threading.Lock()
79 def gethostbyname(host):
80 global hostbyname_cache
81 global hostbyname_cache_lock
83 hostbyname = hostbyname_cache.get(host)
85 with hostbyname_cache_lock:
86 hostbyname = socket.gethostbyname(host)
87 hostbyname_cache[host] = hostbyname
89 msg = " Added hostbyname %s - %s " % (host, hostbyname)
90 log(msg, logging.DEBUG)
94 OPENSSH_HAS_PERSIST = None
96 def openssh_has_persist():
97 """ The ssh_config options ControlMaster and ControlPersist allow to
98 reuse a same network connection for multiple ssh sessions. In this
99 way limitations on number of open ssh connections can be bypassed.
100 However, older versions of openSSH do not support this feature.
101 This function is used to determine if ssh connection persist features
104 global OPENSSH_HAS_PERSIST
105 if OPENSSH_HAS_PERSIST is None:
106 proc = subprocess.Popen(["ssh","-v"],
107 stdout = subprocess.PIPE,
108 stderr = subprocess.STDOUT,
109 stdin = open("/dev/null","r") )
110 out,err = proc.communicate()
113 vre = re.compile(r'OpenSSH_(?:[6-9]|5[.][8-9]|5[.][1-9][0-9]|[1-9][0-9]).*', re.I)
114 OPENSSH_HAS_PERSIST = bool(vre.match(out))
115 return OPENSSH_HAS_PERSIST
117 def make_server_key_args(server_key, host, port):
118 """ Returns a reference to a temporary known_hosts file, to which
119 the server key has been added.
121 Make sure to hold onto the temp file reference until the process is
124 :param server_key: the server public key
125 :type server_key: str
127 :param host: the hostname
130 :param port: the ssh port
135 host = '%s:%s' % (host, str(port))
137 # Create a temporary server key file
138 tmp_known_hosts = tempfile.NamedTemporaryFile()
140 hostbyname = gethostbyname(host)
142 # Add the intended host key
143 tmp_known_hosts.write('%s,%s %s\n' % (host, hostbyname, server_key))
145 # If we're not in strict mode, add user-configured keys
146 if os.environ.get('NEPI_STRICT_AUTH_MODE',"").lower() not in ('1','true','on'):
147 user_hosts_path = '%s/.ssh/known_hosts' % (os.environ.get('HOME',""),)
148 if os.access(user_hosts_path, os.R_OK):
149 f = open(user_hosts_path, "r")
150 tmp_known_hosts.write(f.read())
153 tmp_known_hosts.flush()
155 return tmp_known_hosts
157 def make_control_path(agent, forward_x11):
158 ctrl_path = "/tmp/nepi_ssh"
166 ctrl_path += "-%r@%h:%p"
171 """ Escapes strings so that they are safe to use as command-line
173 if SHELL_SAFE.match(s):
174 # safe string - no escaping needed
177 # unsafe string - escape
179 if (32 <= ord(c) < 127 or c in ('\r','\n','\t')) and c not in ("'",'"'):
182 return "'$'\\x%02x''" % (ord(c),)
183 s = ''.join(map(escp,s))
186 def eintr_retry(func):
187 """Retries a function invocation when a EINTR occurs"""
189 @functools.wraps(func)
191 retry = kw.pop("_retry", False)
192 for i in xrange(0 if retry else 4):
194 return func(*p, **kw)
195 except (select.error, socket.error), args:
196 if args[0] == errno.EINTR:
201 if e.errno == errno.EINTR:
206 return func(*p, **kw)
209 def rexec(command, host, user,
217 connect_timeout = 30,
222 strict_host_checking = True):
224 Executes a remote command, returns ((stdout,stderr),process)
227 tmp_known_hosts = None
228 hostip = gethostbyname(host)
231 # Don't bother with localhost. Makes test easier
232 '-o', 'NoHostAuthenticationForLocalhost=yes',
233 '-o', 'ConnectTimeout=%d' % (int(connect_timeout),),
234 '-o', 'ConnectionAttempts=3',
235 '-o', 'ServerAliveInterval=30',
236 '-o', 'TCPKeepAlive=yes',
237 '-l', user, hostip or host]
239 if persistent and openssh_has_persist():
241 '-o', 'ControlMaster=auto',
242 '-o', 'ControlPath=%s' % (make_control_path(agent, forward_x11),),
243 '-o', 'ControlPersist=60' ])
245 if not strict_host_checking:
246 # Do not check for Host key. Unsafe.
247 args.extend(['-o', 'StrictHostKeyChecking=no'])
253 args.append('-p%d' % port)
256 args.extend(('-i', identity))
266 # Create a temporary server key file
267 tmp_known_hosts = make_server_key_args(server_key, host, port)
268 args.extend(['-o', 'UserKnownHostsFile=%s' % (tmp_known_hosts.name,)])
271 command = "sudo " + command
275 log_msg = " rexec - host %s - command %s " % (host, " ".join(args))
277 stdout = stderr = stdin = subprocess.PIPE
279 stdout = stderr = stdin = None
281 return _retry_rexec(args, log_msg,
287 tmp_known_hosts = tmp_known_hosts,
290 def rcopy(source, dest,
297 strict_host_checking = True):
299 Copies from/to remote sites.
301 Source and destination should have the user and host encoded
304 Source can be a list of files to copy to a single destination,
305 in which case it is advised that the destination be a folder.
308 # Parse destination as <user>@<server>:<path>
309 if isinstance(dest, basestring) and ':' in dest:
310 remspec, path = dest.split(':',1)
311 elif isinstance(source, basestring) and ':' in source:
312 remspec, path = source.split(':',1)
314 raise ValueError, "Both endpoints cannot be local"
315 user,host = remspec.rsplit('@',1)
318 tmp_known_hosts = None
320 args = ['scp', '-q', '-p', '-C',
321 # Speed up transfer using blowfish cypher specification which is
322 # faster than the default one (3des)
324 # Don't bother with localhost. Makes test easier
325 '-o', 'NoHostAuthenticationForLocalhost=yes',
326 '-o', 'ConnectTimeout=60',
327 '-o', 'ConnectionAttempts=3',
328 '-o', 'ServerAliveInterval=30',
329 '-o', 'TCPKeepAlive=yes' ]
332 args.append('-P%d' % port)
338 args.extend(('-i', identity))
341 # Create a temporary server key file
342 tmp_known_hosts = make_server_key_args(server_key, host, port)
343 args.extend(['-o', 'UserKnownHostsFile=%s' % (tmp_known_hosts.name,)])
345 if not strict_host_checking:
346 # Do not check for Host key. Unsafe.
347 args.extend(['-o', 'StrictHostKeyChecking=no'])
349 if isinstance(source, list):
352 if openssh_has_persist():
354 '-o', 'ControlMaster=auto',
355 '-o', 'ControlPath=%s' % (make_control_path(agent, False),)
361 log_msg = " rcopy - host %s - command %s " % (host, " ".join(args))
363 return _retry_rexec(args, log_msg, env = None, retry = retry,
364 tmp_known_hosts = tmp_known_hosts,
367 def rspawn(command, pidfile,
368 stdout = '/dev/null',
382 Spawn a remote command such that it will continue working asynchronously in
385 :param command: The command to run, it should be a single line.
388 :param pidfile: Path to a file where to store the pid and ppid of the
392 :param stdout: Path to file to redirect standard output.
393 The default value is /dev/null
396 :param stderr: Path to file to redirect standard error.
397 If the special STDOUT value is used, stderr will
398 be redirected to the same file as stdout
401 :param stdin: Path to a file with input to be piped into the command's standard input
404 :param home: Path to working directory folder.
405 It is assumed to exist unless the create_home flag is set.
408 :param create_home: Flag to force creation of the home folder before
410 :type create_home: bool
412 :param sudo: Flag forcing execution with sudo user
417 (stdout, stderr), process
419 Of the spawning process, which only captures errors at spawning time.
420 Usually only useful for diagnostics.
422 # Start process in a "daemonized" way, using nohup and heavy
423 # stdin/out redirection to avoid connection issues
427 stderr = ' ' + stderr
429 daemon_command = '{ { %(command)s > %(stdout)s 2>%(stderr)s < %(stdin)s & } ; echo $! 1 > %(pidfile)s ; }' % {
431 'pidfile' : shell_escape(pidfile),
437 cmd = "%(create)s%(gohome)s rm -f %(pidfile)s ; %(sudo)s nohup bash -c %(command)s " % {
438 'command' : shell_escape(daemon_command),
439 'sudo' : 'sudo -S' if sudo else '',
440 'pidfile' : shell_escape(pidfile),
441 'gohome' : 'cd %s ; ' % (shell_escape(home),) if home else '',
442 'create' : 'mkdir -p %s ; ' % (shell_escape(home),) if create_home and home else '',
445 (out,err),proc = rexec(
452 server_key = server_key,
457 raise RuntimeError, "Failed to set up application on host %s: %s %s" % (host, out,err,)
459 return ((out, err), proc)
470 Returns the pid and ppid of a process from a remote file where the
471 information was stored.
473 :param home: Path to directory where the pidfile is located
476 :param pidfile: Name of file containing the pid information
481 A (pid, ppid) tuple useful for calling rstatus and rkill,
482 or None if the pidfile isn't valid yet (can happen when process is staring up)
485 (out,err),proc = rexec(
486 "cat %(pidfile)s" % {
494 server_key = server_key
502 return map(int,out.strip().split(' ',1))
504 # Ignore, many ways to fail that don't matter that much
508 def rstatus(pid, ppid,
516 Returns a code representing the the status of a remote process
518 :param pid: Process id of the process
521 :param ppid: Parent process id of process
524 :rtype: int (One of NOT_STARTED, RUNNING, FINISHED)
527 (out,err),proc = rexec(
528 # Check only by pid. pid+ppid does not always work (especially with sudo)
529 " (( ps --pid %(pid)d -o pid | grep -c %(pid)d && echo 'wait') || echo 'done' ) | tail -n 1" % {
538 server_key = server_key
542 return ProcStatus.NOT_STARTED
546 if err.strip().find("Error, do this: mount -t proc none /proc") >= 0:
549 status = (out.strip() == 'wait')
551 return ProcStatus.NOT_STARTED
552 return ProcStatus.RUNNING if status else ProcStatus.FINISHED
565 Sends a kill signal to a remote process.
567 First tries a SIGTERM, and if the process does not end in 10 seconds,
570 :param pid: Process id of process to be killed
573 :param ppid: Parent process id of process to be killed
576 :param sudo: Flag indicating if sudo should be used to kill the process
580 subkill = "$(ps --ppid %(pid)d -o pid h)" % { 'pid' : pid }
582 SUBKILL="%(subkill)s" ;
583 %(sudo)s kill -- -%(pid)d $SUBKILL || /bin/true
584 %(sudo)s kill %(pid)d $SUBKILL || /bin/true
585 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
587 if [ `ps --pid %(pid)d -o pid | grep -c %(pid)d` == '0' ]; then
590 %(sudo)s kill -- -%(pid)d $SUBKILL || /bin/true
591 %(sudo)s kill %(pid)d $SUBKILL || /bin/true
595 if [ `ps --pid %(pid)d -o pid | grep -c %(pid)d` != '0' ]; then
596 %(sudo)s kill -9 -- -%(pid)d $SUBKILL || /bin/true
597 %(sudo)s kill -9 %(pid)d $SUBKILL || /bin/true
601 cmd = "( %s ) >/dev/null 2>/dev/null </dev/null &" % (cmd,)
603 (out,err),proc = rexec(
607 'sudo' : 'sudo -S' if sudo else '',
615 server_key = server_key
618 # wait, don't leave zombies around
621 return (out, err), proc
623 def _retry_rexec(args,
625 stdout = subprocess.PIPE,
626 stdin = subprocess.PIPE,
627 stderr = subprocess.PIPE,
630 tmp_known_hosts = None,
633 for x in xrange(retry):
634 # connects to the remote host and starts a remote connection
635 proc = subprocess.Popen(args,
641 # attach tempfile object to the process, to make sure the file stays
642 # alive until the process is finished with it
643 proc._known_hosts = tmp_known_hosts
645 # The argument block == False forces to rexec to return immediately,
650 (out, err) = proc.communicate()
652 out = proc.stdout.read()
653 if proc.poll() and stderr:
654 err = proc.stderr.read()
656 log(log_msg, logging.DEBUG, out, err)
661 if err.strip().startswith('ssh: ') or err.strip().startswith('mux_client_hello_exchange: '):
662 # SSH error, can safely retry
665 # Probably timed out or plain failed but can retry
670 msg = "SLEEPING %d ... ATEMPT %d - command %s " % (
671 t, x, " ".join(args))
672 log(msg, logging.DEBUG)
677 except RuntimeError, e:
678 msg = " rexec EXCEPTION - TIMEOUT -> %s \n %s" % ( e.args, log_msg )
679 log(msg, logging.DEBUG, out, err)
685 return ((out, err), proc)