blob: 2fc1bf0d4a4b3af9db6195da66ba4c0062d8c8c1 [file] [log] [blame]
Aviv Keshet07f16242013-10-10 07:54:19 -07001import os, time, socket, shutil, glob, logging, traceback, tempfile
Cheng-Yi Chianga155e7e2015-08-20 20:42:04 +08002import subprocess
3
Simran Basi3b858a22015-03-17 16:23:24 -07004from multiprocessing import Lock
Aviv Keshet53a216a2013-08-27 13:58:46 -07005from autotest_lib.client.common_lib import autotemp, error
jadmanski31c49b72008-10-27 20:44:48 +00006from autotest_lib.server import utils, autotest
mblighe8b93af2009-01-30 00:45:53 +00007from autotest_lib.server.hosts import remote
mblighefccc1b2010-01-11 19:08:42 +00008from autotest_lib.client.common_lib.global_config import global_config
jadmanskica7da372008-10-21 16:26:52 +00009
Aviv Keshet53a216a2013-08-27 13:58:46 -070010# pylint: disable-msg=C0111
jadmanskica7da372008-10-21 16:26:52 +000011
mblighb86bfa12010-02-12 20:22:21 +000012get_value = global_config.get_config_value
13enable_master_ssh = get_value('AUTOSERV', 'enable_master_ssh', type=bool,
14 default=False)
mblighefccc1b2010-01-11 19:08:42 +000015
16
Fang Deng96667ca2013-08-01 17:46:18 -070017class AbstractSSHHost(remote.RemoteHost):
mblighbc9402b2009-12-29 01:15:34 +000018 """
19 This class represents a generic implementation of most of the
jadmanskica7da372008-10-21 16:26:52 +000020 framework necessary for controlling a host via ssh. It implements
21 almost all of the abstract Host methods, except for the core
mblighbc9402b2009-12-29 01:15:34 +000022 Host.run method.
23 """
jadmanskica7da372008-10-21 16:26:52 +000024
jadmanskif6562912008-10-21 17:59:01 +000025 def _initialize(self, hostname, user="root", port=22, password="",
26 *args, **dargs):
27 super(AbstractSSHHost, self)._initialize(hostname=hostname,
28 *args, **dargs)
Dan Shic07b8932014-12-11 15:22:30 -080029 # IP address is retrieved only on demand. Otherwise the host
30 # initialization will fail for host is not online.
31 self._ip = None
jadmanskica7da372008-10-21 16:26:52 +000032 self.user = user
33 self.port = port
34 self.password = password
showard6eafb492010-01-15 20:29:06 +000035 self._use_rsync = None
Fang Deng3af66202013-08-16 15:19:25 -070036 self.known_hosts_file = tempfile.mkstemp()[1]
jadmanskica7da372008-10-21 16:26:52 +000037
mblighefccc1b2010-01-11 19:08:42 +000038 """
39 Master SSH connection background job, socket temp directory and socket
40 control path option. If master-SSH is enabled, these fields will be
41 initialized by start_master_ssh when a new SSH connection is initiated.
42 """
43 self.master_ssh_job = None
44 self.master_ssh_tempdir = None
45 self.master_ssh_option = ''
46
Simran Basi3b858a22015-03-17 16:23:24 -070047 # Create a Lock to protect against race conditions.
48 self._lock = Lock()
49
showard6eafb492010-01-15 20:29:06 +000050
Dan Shic07b8932014-12-11 15:22:30 -080051 @property
52 def ip(self):
53 """@return IP address of the host.
54 """
55 if not self._ip:
56 self._ip = socket.getaddrinfo(self.hostname, None)[0][4][0]
57 return self._ip
58
59
Fang Deng96667ca2013-08-01 17:46:18 -070060 def make_ssh_command(self, user="root", port=22, opts='',
61 hosts_file='/dev/null',
62 connect_timeout=30, alive_interval=300):
63 base_command = ("/usr/bin/ssh -a -x %s -o StrictHostKeyChecking=no "
64 "-o UserKnownHostsFile=%s -o BatchMode=yes "
65 "-o ConnectTimeout=%d -o ServerAliveInterval=%d "
66 "-l %s -p %d")
67 assert isinstance(connect_timeout, (int, long))
68 assert connect_timeout > 0 # can't disable the timeout
69 return base_command % (opts, hosts_file, connect_timeout,
70 alive_interval, user, port)
71
72
showard6eafb492010-01-15 20:29:06 +000073 def use_rsync(self):
74 if self._use_rsync is not None:
75 return self._use_rsync
76
mblighc9892c02010-01-06 19:02:16 +000077 # Check if rsync is available on the remote host. If it's not,
78 # don't try to use it for any future file transfers.
showard6eafb492010-01-15 20:29:06 +000079 self._use_rsync = self._check_rsync()
80 if not self._use_rsync:
Ilja H. Friedel04be2bd2014-05-07 21:29:59 -070081 logging.warning("rsync not available on remote host %s -- disabled",
mblighc9892c02010-01-06 19:02:16 +000082 self.hostname)
Eric Lie0493a42010-11-15 13:05:43 -080083 return self._use_rsync
mblighc9892c02010-01-06 19:02:16 +000084
85
86 def _check_rsync(self):
87 """
88 Check if rsync is available on the remote host.
89 """
90 try:
91 self.run("rsync --version", stdout_tee=None, stderr_tee=None)
92 except error.AutoservRunError:
93 return False
94 return True
95
jadmanskica7da372008-10-21 16:26:52 +000096
showard56176ec2009-10-28 19:52:30 +000097 def _encode_remote_paths(self, paths, escape=True):
mblighbc9402b2009-12-29 01:15:34 +000098 """
99 Given a list of file paths, encodes it as a single remote path, in
100 the style used by rsync and scp.
101 """
showard56176ec2009-10-28 19:52:30 +0000102 if escape:
103 paths = [utils.scp_remote_escape(path) for path in paths]
104 return '%s@%s:"%s"' % (self.user, self.hostname, " ".join(paths))
jadmanskica7da372008-10-21 16:26:52 +0000105
jadmanskica7da372008-10-21 16:26:52 +0000106
mbligh45561782009-05-11 21:14:34 +0000107 def _make_rsync_cmd(self, sources, dest, delete_dest, preserve_symlinks):
mblighbc9402b2009-12-29 01:15:34 +0000108 """
109 Given a list of source paths and a destination path, produces the
jadmanskid7b79ed2009-01-07 17:19:48 +0000110 appropriate rsync command for copying them. Remote paths must be
mblighbc9402b2009-12-29 01:15:34 +0000111 pre-encoded.
112 """
Fang Deng96667ca2013-08-01 17:46:18 -0700113 ssh_cmd = self.make_ssh_command(user=self.user, port=self.port,
114 opts=self.master_ssh_option,
115 hosts_file=self.known_hosts_file)
jadmanskid7b79ed2009-01-07 17:19:48 +0000116 if delete_dest:
117 delete_flag = "--delete"
118 else:
119 delete_flag = ""
mbligh45561782009-05-11 21:14:34 +0000120 if preserve_symlinks:
121 symlink_flag = ""
122 else:
123 symlink_flag = "-L"
Dan Shi06d7fbf2014-02-12 12:34:41 -0800124 command = ("rsync %s %s --timeout=1800 --rsh='%s' -az --no-o --no-g "
David Hendricksb8904182014-06-02 15:22:49 -0700125 "%s \"%s\"")
mbligh45561782009-05-11 21:14:34 +0000126 return command % (symlink_flag, delete_flag, ssh_cmd,
David Hendricksb8904182014-06-02 15:22:49 -0700127 " ".join(['"%s"' % p for p in sources]), dest)
jadmanskid7b79ed2009-01-07 17:19:48 +0000128
129
Eric Li861b2d52011-02-04 14:50:35 -0800130 def _make_ssh_cmd(self, cmd):
131 """
132 Create a base ssh command string for the host which can be used
133 to run commands directly on the machine
134 """
Fang Deng96667ca2013-08-01 17:46:18 -0700135 base_cmd = self.make_ssh_command(user=self.user, port=self.port,
136 opts=self.master_ssh_option,
137 hosts_file=self.known_hosts_file)
Eric Li861b2d52011-02-04 14:50:35 -0800138
139 return '%s %s "%s"' % (base_cmd, self.hostname, utils.sh_escape(cmd))
140
jadmanskid7b79ed2009-01-07 17:19:48 +0000141 def _make_scp_cmd(self, sources, dest):
mblighbc9402b2009-12-29 01:15:34 +0000142 """
143 Given a list of source paths and a destination path, produces the
jadmanskid7b79ed2009-01-07 17:19:48 +0000144 appropriate scp command for encoding it. Remote paths must be
mblighbc9402b2009-12-29 01:15:34 +0000145 pre-encoded.
146 """
mblighc0649d62010-01-15 18:15:58 +0000147 command = ("scp -rq %s -o StrictHostKeyChecking=no "
lmraf676f32010-02-04 03:36:26 +0000148 "-o UserKnownHostsFile=%s -P %d %s '%s'")
Fang Deng3af66202013-08-16 15:19:25 -0700149 return command % (self.master_ssh_option, self.known_hosts_file,
mblighefccc1b2010-01-11 19:08:42 +0000150 self.port, " ".join(sources), dest)
jadmanskid7b79ed2009-01-07 17:19:48 +0000151
152
153 def _make_rsync_compatible_globs(self, path, is_local):
mblighbc9402b2009-12-29 01:15:34 +0000154 """
155 Given an rsync-style path, returns a list of globbed paths
jadmanskid7b79ed2009-01-07 17:19:48 +0000156 that will hopefully provide equivalent behaviour for scp. Does not
157 support the full range of rsync pattern matching behaviour, only that
158 exposed in the get/send_file interface (trailing slashes).
159
160 The is_local param is flag indicating if the paths should be
mblighbc9402b2009-12-29 01:15:34 +0000161 interpreted as local or remote paths.
162 """
jadmanskid7b79ed2009-01-07 17:19:48 +0000163
164 # non-trailing slash paths should just work
165 if len(path) == 0 or path[-1] != "/":
166 return [path]
167
168 # make a function to test if a pattern matches any files
169 if is_local:
showard56176ec2009-10-28 19:52:30 +0000170 def glob_matches_files(path, pattern):
171 return len(glob.glob(path + pattern)) > 0
jadmanskid7b79ed2009-01-07 17:19:48 +0000172 else:
showard56176ec2009-10-28 19:52:30 +0000173 def glob_matches_files(path, pattern):
174 result = self.run("ls \"%s\"%s" % (utils.sh_escape(path),
175 pattern),
176 stdout_tee=None, ignore_status=True)
jadmanskid7b79ed2009-01-07 17:19:48 +0000177 return result.exit_status == 0
178
179 # take a set of globs that cover all files, and see which are needed
180 patterns = ["*", ".[!.]*"]
showard56176ec2009-10-28 19:52:30 +0000181 patterns = [p for p in patterns if glob_matches_files(path, p)]
jadmanskid7b79ed2009-01-07 17:19:48 +0000182
183 # convert them into a set of paths suitable for the commandline
jadmanskid7b79ed2009-01-07 17:19:48 +0000184 if is_local:
showard56176ec2009-10-28 19:52:30 +0000185 return ["\"%s\"%s" % (utils.sh_escape(path), pattern)
186 for pattern in patterns]
jadmanskid7b79ed2009-01-07 17:19:48 +0000187 else:
showard56176ec2009-10-28 19:52:30 +0000188 return [utils.scp_remote_escape(path) + pattern
189 for pattern in patterns]
jadmanskid7b79ed2009-01-07 17:19:48 +0000190
191
192 def _make_rsync_compatible_source(self, source, is_local):
mblighbc9402b2009-12-29 01:15:34 +0000193 """
194 Applies the same logic as _make_rsync_compatible_globs, but
jadmanskid7b79ed2009-01-07 17:19:48 +0000195 applies it to an entire list of sources, producing a new list of
mblighbc9402b2009-12-29 01:15:34 +0000196 sources, properly quoted.
197 """
jadmanskid7b79ed2009-01-07 17:19:48 +0000198 return sum((self._make_rsync_compatible_globs(path, is_local)
199 for path in source), [])
jadmanskica7da372008-10-21 16:26:52 +0000200
201
mblighfeac0102009-04-28 18:31:12 +0000202 def _set_umask_perms(self, dest):
mblighbc9402b2009-12-29 01:15:34 +0000203 """
204 Given a destination file/dir (recursively) set the permissions on
205 all the files and directories to the max allowed by running umask.
206 """
mblighfeac0102009-04-28 18:31:12 +0000207
208 # now this looks strange but I haven't found a way in Python to _just_
209 # get the umask, apparently the only option is to try to set it
210 umask = os.umask(0)
211 os.umask(umask)
212
213 max_privs = 0777 & ~umask
214
215 def set_file_privs(filename):
Chris Masone567d0d92011-12-19 09:38:30 -0800216 """Sets mode of |filename|. Assumes |filename| exists."""
217 file_stat = os.stat(filename)
mblighfeac0102009-04-28 18:31:12 +0000218
219 file_privs = max_privs
220 # if the original file permissions do not have at least one
221 # executable bit then do not set it anywhere
222 if not file_stat.st_mode & 0111:
223 file_privs &= ~0111
224
225 os.chmod(filename, file_privs)
226
227 # try a bottom-up walk so changes on directory permissions won't cut
228 # our access to the files/directories inside it
229 for root, dirs, files in os.walk(dest, topdown=False):
230 # when setting the privileges we emulate the chmod "X" behaviour
231 # that sets to execute only if it is a directory or any of the
232 # owner/group/other already has execute right
233 for dirname in dirs:
234 os.chmod(os.path.join(root, dirname), max_privs)
235
Chris Masone567d0d92011-12-19 09:38:30 -0800236 # Filter out broken symlinks as we go.
237 for filename in filter(os.path.exists, files):
mblighfeac0102009-04-28 18:31:12 +0000238 set_file_privs(os.path.join(root, filename))
239
240
241 # now set privs for the dest itself
242 if os.path.isdir(dest):
243 os.chmod(dest, max_privs)
244 else:
245 set_file_privs(dest)
246
247
mbligh45561782009-05-11 21:14:34 +0000248 def get_file(self, source, dest, delete_dest=False, preserve_perm=True,
249 preserve_symlinks=False):
jadmanskica7da372008-10-21 16:26:52 +0000250 """
251 Copy files from the remote host to a local path.
252
253 Directories will be copied recursively.
254 If a source component is a directory with a trailing slash,
255 the content of the directory will be copied, otherwise, the
256 directory itself and its content will be copied. This
257 behavior is similar to that of the program 'rsync'.
258
259 Args:
260 source: either
261 1) a single file or directory, as a string
262 2) a list of one or more (possibly mixed)
263 files or directories
264 dest: a file or a directory (if source contains a
265 directory or more than one element, you must
266 supply a directory dest)
mbligh89e258d2008-10-24 13:58:08 +0000267 delete_dest: if this is true, the command will also clear
268 out any old files at dest that are not in the
269 source
mblighfeac0102009-04-28 18:31:12 +0000270 preserve_perm: tells get_file() to try to preserve the sources
271 permissions on files and dirs
mbligh45561782009-05-11 21:14:34 +0000272 preserve_symlinks: try to preserve symlinks instead of
273 transforming them into files/dirs on copy
jadmanskica7da372008-10-21 16:26:52 +0000274
275 Raises:
276 AutoservRunError: the scp command failed
277 """
Simran Basi882f15b2013-10-29 14:59:34 -0700278 logging.debug('get_file. source: %s, dest: %s, delete_dest: %s,'
279 'preserve_perm: %s, preserve_symlinks:%s', source, dest,
280 delete_dest, preserve_perm, preserve_symlinks)
mblighefccc1b2010-01-11 19:08:42 +0000281 # Start a master SSH connection if necessary.
282 self.start_master_ssh()
283
jadmanskica7da372008-10-21 16:26:52 +0000284 if isinstance(source, basestring):
285 source = [source]
jadmanskid7b79ed2009-01-07 17:19:48 +0000286 dest = os.path.abspath(dest)
jadmanskica7da372008-10-21 16:26:52 +0000287
mblighc9892c02010-01-06 19:02:16 +0000288 # If rsync is disabled or fails, try scp.
showard6eafb492010-01-15 20:29:06 +0000289 try_scp = True
290 if self.use_rsync():
Simran Basi882f15b2013-10-29 14:59:34 -0700291 logging.debug('Using Rsync.')
mblighc9892c02010-01-06 19:02:16 +0000292 try:
293 remote_source = self._encode_remote_paths(source)
294 local_dest = utils.sh_escape(dest)
295 rsync = self._make_rsync_cmd([remote_source], local_dest,
296 delete_dest, preserve_symlinks)
297 utils.run(rsync)
showard6eafb492010-01-15 20:29:06 +0000298 try_scp = False
mblighc9892c02010-01-06 19:02:16 +0000299 except error.CmdError, e:
Ilja H. Friedel04be2bd2014-05-07 21:29:59 -0700300 logging.warning("trying scp, rsync failed: %s", e)
mblighc9892c02010-01-06 19:02:16 +0000301
302 if try_scp:
Simran Basi882f15b2013-10-29 14:59:34 -0700303 logging.debug('Trying scp.')
jadmanskid7b79ed2009-01-07 17:19:48 +0000304 # scp has no equivalent to --delete, just drop the entire dest dir
305 if delete_dest and os.path.isdir(dest):
306 shutil.rmtree(dest)
307 os.mkdir(dest)
jadmanskica7da372008-10-21 16:26:52 +0000308
jadmanskid7b79ed2009-01-07 17:19:48 +0000309 remote_source = self._make_rsync_compatible_source(source, False)
310 if remote_source:
showard56176ec2009-10-28 19:52:30 +0000311 # _make_rsync_compatible_source() already did the escaping
312 remote_source = self._encode_remote_paths(remote_source,
313 escape=False)
jadmanskid7b79ed2009-01-07 17:19:48 +0000314 local_dest = utils.sh_escape(dest)
jadmanski2583a432009-02-10 23:59:11 +0000315 scp = self._make_scp_cmd([remote_source], local_dest)
jadmanskid7b79ed2009-01-07 17:19:48 +0000316 try:
317 utils.run(scp)
318 except error.CmdError, e:
Simran Basi882f15b2013-10-29 14:59:34 -0700319 logging.debug('scp failed: %s', e)
jadmanskid7b79ed2009-01-07 17:19:48 +0000320 raise error.AutoservRunError(e.args[0], e.args[1])
jadmanskica7da372008-10-21 16:26:52 +0000321
mblighfeac0102009-04-28 18:31:12 +0000322 if not preserve_perm:
323 # we have no way to tell scp to not try to preserve the
324 # permissions so set them after copy instead.
325 # for rsync we could use "--no-p --chmod=ugo=rwX" but those
326 # options are only in very recent rsync versions
327 self._set_umask_perms(dest)
328
jadmanskica7da372008-10-21 16:26:52 +0000329
mbligh45561782009-05-11 21:14:34 +0000330 def send_file(self, source, dest, delete_dest=False,
331 preserve_symlinks=False):
jadmanskica7da372008-10-21 16:26:52 +0000332 """
333 Copy files from a local path to the remote host.
334
335 Directories will be copied recursively.
336 If a source component is a directory with a trailing slash,
337 the content of the directory will be copied, otherwise, the
338 directory itself and its content will be copied. This
339 behavior is similar to that of the program 'rsync'.
340
341 Args:
342 source: either
343 1) a single file or directory, as a string
344 2) a list of one or more (possibly mixed)
345 files or directories
346 dest: a file or a directory (if source contains a
347 directory or more than one element, you must
348 supply a directory dest)
mbligh89e258d2008-10-24 13:58:08 +0000349 delete_dest: if this is true, the command will also clear
350 out any old files at dest that are not in the
351 source
mbligh45561782009-05-11 21:14:34 +0000352 preserve_symlinks: controls if symlinks on the source will be
353 copied as such on the destination or transformed into the
354 referenced file/directory
jadmanskica7da372008-10-21 16:26:52 +0000355
356 Raises:
357 AutoservRunError: the scp command failed
358 """
Simran Basi882f15b2013-10-29 14:59:34 -0700359 logging.debug('send_file. source: %s, dest: %s, delete_dest: %s,'
360 'preserve_symlinks:%s', source, dest,
361 delete_dest, preserve_symlinks)
mblighefccc1b2010-01-11 19:08:42 +0000362 # Start a master SSH connection if necessary.
363 self.start_master_ssh()
364
jadmanskica7da372008-10-21 16:26:52 +0000365 if isinstance(source, basestring):
366 source = [source]
jadmanski2583a432009-02-10 23:59:11 +0000367 remote_dest = self._encode_remote_paths([dest])
jadmanskica7da372008-10-21 16:26:52 +0000368
mukesh agrawal0d3616c2015-07-17 15:47:36 -0700369 local_sources = [utils.sh_escape(path) for path in source]
370 if not local_sources:
371 raise error.TestError('source |%s| yielded an empty list' % (
372 source))
373 if any([local_source.find('\x00') != -1 for
374 local_source in local_sources]):
375 raise error.TestError('one or more sources include NUL char')
376
mblighc9892c02010-01-06 19:02:16 +0000377 # If rsync is disabled or fails, try scp.
showard6eafb492010-01-15 20:29:06 +0000378 try_scp = True
379 if self.use_rsync():
Simran Basi882f15b2013-10-29 14:59:34 -0700380 logging.debug('Using Rsync.')
mblighc9892c02010-01-06 19:02:16 +0000381 try:
mblighc9892c02010-01-06 19:02:16 +0000382 rsync = self._make_rsync_cmd(local_sources, remote_dest,
383 delete_dest, preserve_symlinks)
384 utils.run(rsync)
showard6eafb492010-01-15 20:29:06 +0000385 try_scp = False
mblighc9892c02010-01-06 19:02:16 +0000386 except error.CmdError, e:
Ilja H. Friedel04be2bd2014-05-07 21:29:59 -0700387 logging.warning("trying scp, rsync failed: %s", e)
mblighc9892c02010-01-06 19:02:16 +0000388
389 if try_scp:
Simran Basi882f15b2013-10-29 14:59:34 -0700390 logging.debug('Trying scp.')
jadmanskid7b79ed2009-01-07 17:19:48 +0000391 # scp has no equivalent to --delete, just drop the entire dest dir
392 if delete_dest:
showard27160152009-07-15 14:28:42 +0000393 is_dir = self.run("ls -d %s/" % dest,
jadmanskid7b79ed2009-01-07 17:19:48 +0000394 ignore_status=True).exit_status == 0
395 if is_dir:
396 cmd = "rm -rf %s && mkdir %s"
mbligh5a0ca532009-08-03 16:44:34 +0000397 cmd %= (dest, dest)
jadmanskid7b79ed2009-01-07 17:19:48 +0000398 self.run(cmd)
jadmanskica7da372008-10-21 16:26:52 +0000399
jadmanski2583a432009-02-10 23:59:11 +0000400 local_sources = self._make_rsync_compatible_source(source, True)
401 if local_sources:
402 scp = self._make_scp_cmd(local_sources, remote_dest)
jadmanskid7b79ed2009-01-07 17:19:48 +0000403 try:
404 utils.run(scp)
405 except error.CmdError, e:
Simran Basi882f15b2013-10-29 14:59:34 -0700406 logging.debug('scp failed: %s', e)
jadmanskid7b79ed2009-01-07 17:19:48 +0000407 raise error.AutoservRunError(e.args[0], e.args[1])
mukesh agrawal0d3616c2015-07-17 15:47:36 -0700408 else:
409 logging.debug('skipping scp for empty source list')
jadmanskid7b79ed2009-01-07 17:19:48 +0000410
jadmanskica7da372008-10-21 16:26:52 +0000411
beeps46dadc92013-11-07 14:07:10 -0800412 def ssh_ping(self, timeout=60, base_cmd='true'):
beepsadd66d32013-03-04 17:21:51 -0800413 """
414 Pings remote host via ssh.
415
416 @param timeout: Time in seconds before giving up.
417 Defaults to 60 seconds.
beeps46dadc92013-11-07 14:07:10 -0800418 @param base_cmd: The base command to run with the ssh ping.
419 Defaults to true.
beepsadd66d32013-03-04 17:21:51 -0800420 @raise AutoservSSHTimeout: If the ssh ping times out.
421 @raise AutoservSshPermissionDeniedError: If ssh ping fails due to
422 permissions.
423 @raise AutoservSshPingHostError: For other AutoservRunErrors.
424 """
jadmanskica7da372008-10-21 16:26:52 +0000425 try:
beeps46dadc92013-11-07 14:07:10 -0800426 self.run(base_cmd, timeout=timeout, connect_timeout=timeout)
jadmanskica7da372008-10-21 16:26:52 +0000427 except error.AutoservSSHTimeout:
mblighd0e94982009-07-11 00:15:18 +0000428 msg = "Host (ssh) verify timed out (timeout = %d)" % timeout
jadmanskica7da372008-10-21 16:26:52 +0000429 raise error.AutoservSSHTimeout(msg)
mbligh9d738d62009-03-09 21:17:10 +0000430 except error.AutoservSshPermissionDeniedError:
431 #let AutoservSshPermissionDeniedError be visible to the callers
432 raise
jadmanskica7da372008-10-21 16:26:52 +0000433 except error.AutoservRunError, e:
mblighc971c5f2009-06-08 16:48:54 +0000434 # convert the generic AutoservRunError into something more
435 # specific for this context
436 raise error.AutoservSshPingHostError(e.description + '\n' +
437 repr(e.result_obj))
jadmanskica7da372008-10-21 16:26:52 +0000438
439
beeps46dadc92013-11-07 14:07:10 -0800440 def is_up(self, timeout=60, base_cmd='true'):
jadmanskica7da372008-10-21 16:26:52 +0000441 """
beeps46dadc92013-11-07 14:07:10 -0800442 Check if the remote host is up by ssh-ing and running a base command.
jadmanskica7da372008-10-21 16:26:52 +0000443
beepsadd66d32013-03-04 17:21:51 -0800444 @param timeout: timeout in seconds.
beeps46dadc92013-11-07 14:07:10 -0800445 @param base_cmd: a base command to run with ssh. The default is 'true'.
beepsadd66d32013-03-04 17:21:51 -0800446 @returns True if the remote host is up before the timeout expires,
447 False otherwise.
jadmanskica7da372008-10-21 16:26:52 +0000448 """
449 try:
beeps46dadc92013-11-07 14:07:10 -0800450 self.ssh_ping(timeout=timeout, base_cmd=base_cmd)
jadmanskica7da372008-10-21 16:26:52 +0000451 except error.AutoservError:
452 return False
453 else:
454 return True
455
456
457 def wait_up(self, timeout=None):
458 """
459 Wait until the remote host is up or the timeout expires.
460
461 In fact, it will wait until an ssh connection to the remote
462 host can be established, and getty is running.
463
jadmanskic0354912010-01-12 15:57:29 +0000464 @param timeout time limit in seconds before returning even
465 if the host is not up.
jadmanskica7da372008-10-21 16:26:52 +0000466
beepsadd66d32013-03-04 17:21:51 -0800467 @returns True if the host was found to be up before the timeout expires,
468 False otherwise
jadmanskica7da372008-10-21 16:26:52 +0000469 """
470 if timeout:
beeps46dadc92013-11-07 14:07:10 -0800471 current_time = int(time.time())
472 end_time = current_time + timeout
jadmanskica7da372008-10-21 16:26:52 +0000473
beepsadd66d32013-03-04 17:21:51 -0800474 while not timeout or current_time < end_time:
475 if self.is_up(timeout=end_time - current_time):
jadmanskica7da372008-10-21 16:26:52 +0000476 try:
477 if self.are_wait_up_processes_up():
jadmanski7ebac3d2010-06-17 16:06:31 +0000478 logging.debug('Host %s is now up', self.hostname)
jadmanskica7da372008-10-21 16:26:52 +0000479 return True
480 except error.AutoservError:
481 pass
482 time.sleep(1)
beeps46dadc92013-11-07 14:07:10 -0800483 current_time = int(time.time())
jadmanskica7da372008-10-21 16:26:52 +0000484
jadmanski7ebac3d2010-06-17 16:06:31 +0000485 logging.debug('Host %s is still down after waiting %d seconds',
486 self.hostname, int(timeout + time.time() - end_time))
jadmanskica7da372008-10-21 16:26:52 +0000487 return False
488
489
jadmanskic0354912010-01-12 15:57:29 +0000490 def wait_down(self, timeout=None, warning_timer=None, old_boot_id=None):
jadmanskica7da372008-10-21 16:26:52 +0000491 """
492 Wait until the remote host is down or the timeout expires.
493
jadmanskic0354912010-01-12 15:57:29 +0000494 If old_boot_id is provided, this will wait until either the machine
495 is unpingable or self.get_boot_id() returns a value different from
496 old_boot_id. If the boot_id value has changed then the function
497 returns true under the assumption that the machine has shut down
498 and has now already come back up.
jadmanskica7da372008-10-21 16:26:52 +0000499
jadmanskic0354912010-01-12 15:57:29 +0000500 If old_boot_id is None then until the machine becomes unreachable the
501 method assumes the machine has not yet shut down.
jadmanskica7da372008-10-21 16:26:52 +0000502
beepsadd66d32013-03-04 17:21:51 -0800503 Based on this definition, the 4 possible permutations of timeout
504 and old_boot_id are:
505 1. timeout and old_boot_id: wait timeout seconds for either the
506 host to become unpingable, or the boot id
507 to change. In the latter case we've rebooted
508 and in the former case we've only shutdown,
509 but both cases return True.
510 2. only timeout: wait timeout seconds for the host to become unpingable.
511 If the host remains pingable throughout timeout seconds
512 we return False.
513 3. only old_boot_id: wait forever until either the host becomes
514 unpingable or the boot_id changes. Return true
515 when either of those conditions are met.
516 4. not timeout, not old_boot_id: wait forever till the host becomes
517 unpingable.
518
jadmanskic0354912010-01-12 15:57:29 +0000519 @param timeout Time limit in seconds before returning even
520 if the host is still up.
521 @param warning_timer Time limit in seconds that will generate
522 a warning if the host is not down yet.
523 @param old_boot_id A string containing the result of self.get_boot_id()
524 prior to the host being told to shut down. Can be None if this is
525 not available.
526
527 @returns True if the host was found to be down, False otherwise
jadmanskica7da372008-10-21 16:26:52 +0000528 """
mblighe5e3cf22010-05-27 23:33:14 +0000529 #TODO: there is currently no way to distinguish between knowing
530 #TODO: boot_id was unsupported and not knowing the boot_id.
beeps46dadc92013-11-07 14:07:10 -0800531 current_time = int(time.time())
jadmanskica7da372008-10-21 16:26:52 +0000532 if timeout:
mbligh2ed998f2009-04-08 21:03:47 +0000533 end_time = current_time + timeout
jadmanskica7da372008-10-21 16:26:52 +0000534
mbligh2ed998f2009-04-08 21:03:47 +0000535 if warning_timer:
536 warn_time = current_time + warning_timer
537
jadmanskic0354912010-01-12 15:57:29 +0000538 if old_boot_id is not None:
539 logging.debug('Host %s pre-shutdown boot_id is %s',
540 self.hostname, old_boot_id)
541
beepsadd66d32013-03-04 17:21:51 -0800542 # Impose semi real-time deadline constraints, since some clients
543 # (eg: watchdog timer tests) expect strict checking of time elapsed.
544 # Each iteration of this loop is treated as though it atomically
545 # completes within current_time, this is needed because if we used
546 # inline time.time() calls instead then the following could happen:
547 #
548 # while not timeout or time.time() < end_time: [23 < 30]
549 # some code. [takes 10 secs]
550 # try:
551 # new_boot_id = self.get_boot_id(timeout=end_time - time.time())
552 # [30 - 33]
553 # The last step will lead to a return True, when in fact the machine
554 # went down at 32 seconds (>30). Hence we need to pass get_boot_id
555 # the same time that allowed us into that iteration of the loop.
mbligh2ed998f2009-04-08 21:03:47 +0000556 while not timeout or current_time < end_time:
jadmanskic0354912010-01-12 15:57:29 +0000557 try:
beeps46dadc92013-11-07 14:07:10 -0800558 new_boot_id = self.get_boot_id(timeout=end_time-current_time)
mblighdbc7e4a2010-01-15 20:34:20 +0000559 except error.AutoservError:
jadmanskic0354912010-01-12 15:57:29 +0000560 logging.debug('Host %s is now unreachable over ssh, is down',
561 self.hostname)
jadmanskica7da372008-10-21 16:26:52 +0000562 return True
jadmanskic0354912010-01-12 15:57:29 +0000563 else:
564 # if the machine is up but the boot_id value has changed from
565 # old boot id, then we can assume the machine has gone down
566 # and then already come back up
567 if old_boot_id is not None and old_boot_id != new_boot_id:
568 logging.debug('Host %s now has boot_id %s and so must '
569 'have rebooted', self.hostname, new_boot_id)
570 return True
mbligh2ed998f2009-04-08 21:03:47 +0000571
572 if warning_timer and current_time > warn_time:
Scott Zawalskic86fdeb2013-10-23 10:24:04 -0400573 self.record("INFO", None, "shutdown",
mbligh2ed998f2009-04-08 21:03:47 +0000574 "Shutdown took longer than %ds" % warning_timer)
575 # Print the warning only once.
576 warning_timer = None
mbligha4464402009-04-17 20:13:41 +0000577 # If a machine is stuck switching runlevels
578 # This may cause the machine to reboot.
579 self.run('kill -HUP 1', ignore_status=True)
mbligh2ed998f2009-04-08 21:03:47 +0000580
jadmanskica7da372008-10-21 16:26:52 +0000581 time.sleep(1)
beeps46dadc92013-11-07 14:07:10 -0800582 current_time = int(time.time())
jadmanskica7da372008-10-21 16:26:52 +0000583
584 return False
jadmanskif6562912008-10-21 17:59:01 +0000585
mbligha0a27592009-01-24 01:41:36 +0000586
jadmanskif6562912008-10-21 17:59:01 +0000587 # tunable constants for the verify & repair code
mblighb86bfa12010-02-12 20:22:21 +0000588 AUTOTEST_GB_DISKSPACE_REQUIRED = get_value("SERVER",
589 "gb_diskspace_required",
Fang Deng6b05f5b2013-03-20 13:42:11 -0700590 type=float,
591 default=20.0)
mbligha0a27592009-01-24 01:41:36 +0000592
jadmanskif6562912008-10-21 17:59:01 +0000593
showardca572982009-09-18 21:20:01 +0000594 def verify_connectivity(self):
595 super(AbstractSSHHost, self).verify_connectivity()
jadmanskif6562912008-10-21 17:59:01 +0000596
showardb18134f2009-03-20 20:52:18 +0000597 logging.info('Pinging host ' + self.hostname)
jadmanskif6562912008-10-21 17:59:01 +0000598 self.ssh_ping()
mbligh2ba7ab02009-08-24 22:09:26 +0000599 logging.info("Host (ssh) %s is alive", self.hostname)
jadmanskif6562912008-10-21 17:59:01 +0000600
jadmanski80deb752009-01-21 17:14:16 +0000601 if self.is_shutting_down():
mblighc971c5f2009-06-08 16:48:54 +0000602 raise error.AutoservHostIsShuttingDownError("Host is shutting down")
jadmanski80deb752009-01-21 17:14:16 +0000603
mblighb49b5232009-02-12 21:54:49 +0000604
showardca572982009-09-18 21:20:01 +0000605 def verify_software(self):
606 super(AbstractSSHHost, self).verify_software()
jadmanskif6562912008-10-21 17:59:01 +0000607 try:
showardad812bf2009-10-20 23:49:56 +0000608 self.check_diskspace(autotest.Autotest.get_install_dir(self),
609 self.AUTOTEST_GB_DISKSPACE_REQUIRED)
jadmanskif6562912008-10-21 17:59:01 +0000610 except error.AutoservHostError:
611 raise # only want to raise if it's a space issue
showardad812bf2009-10-20 23:49:56 +0000612 except autotest.AutodirNotFoundError:
showardca572982009-09-18 21:20:01 +0000613 # autotest dir may not exist, etc. ignore
614 logging.debug('autodir space check exception, this is probably '
615 'safe to ignore\n' + traceback.format_exc())
mblighefccc1b2010-01-11 19:08:42 +0000616
617
618 def close(self):
619 super(AbstractSSHHost, self).close()
620 self._cleanup_master_ssh()
Fang Deng3af66202013-08-16 15:19:25 -0700621 os.remove(self.known_hosts_file)
mblighefccc1b2010-01-11 19:08:42 +0000622
623
624 def _cleanup_master_ssh(self):
625 """
626 Release all resources (process, temporary directory) used by an active
627 master SSH connection.
628 """
629 # If a master SSH connection is running, kill it.
630 if self.master_ssh_job is not None:
Aviv Keshet46250752013-08-27 15:52:06 -0700631 logging.debug('Nuking master_ssh_job.')
mblighefccc1b2010-01-11 19:08:42 +0000632 utils.nuke_subprocess(self.master_ssh_job.sp)
633 self.master_ssh_job = None
634
635 # Remove the temporary directory for the master SSH socket.
636 if self.master_ssh_tempdir is not None:
Aviv Keshet46250752013-08-27 15:52:06 -0700637 logging.debug('Cleaning master_ssh_tempdir.')
mblighefccc1b2010-01-11 19:08:42 +0000638 self.master_ssh_tempdir.clean()
639 self.master_ssh_tempdir = None
640 self.master_ssh_option = ''
641
642
Aviv Keshet0749a822013-10-17 09:53:26 -0700643 def start_master_ssh(self, timeout=5):
mblighefccc1b2010-01-11 19:08:42 +0000644 """
645 Called whenever a slave SSH connection needs to be initiated (e.g., by
646 run, rsync, scp). If master SSH support is enabled and a master SSH
647 connection is not active already, start a new one in the background.
648 Also, cleanup any zombie master SSH connections (e.g., dead due to
649 reboot).
Aviv Keshet0749a822013-10-17 09:53:26 -0700650
651 timeout: timeout in seconds (default 5) to wait for master ssh
652 connection to be established. If timeout is reached, a
653 warning message is logged, but no other action is taken.
mblighefccc1b2010-01-11 19:08:42 +0000654 """
655 if not enable_master_ssh:
656 return
657
Simran Basi3b858a22015-03-17 16:23:24 -0700658 # Multiple processes might try in parallel to clean up the old master
659 # ssh connection and create a new one, therefore use a lock to protect
660 # against race conditions.
661 with self._lock:
662 # If a previously started master SSH connection is not running
663 # anymore, it needs to be cleaned up and then restarted.
664 if self.master_ssh_job is not None:
665 socket_path = os.path.join(self.master_ssh_tempdir.name,
666 'socket')
667 if (not os.path.exists(socket_path) or
668 self.master_ssh_job.sp.poll() is not None):
669 logging.info("Master ssh connection to %s is down.",
670 self.hostname)
671 self._cleanup_master_ssh()
mblighefccc1b2010-01-11 19:08:42 +0000672
Simran Basi3b858a22015-03-17 16:23:24 -0700673 # Start a new master SSH connection.
674 if self.master_ssh_job is None:
675 # Create a shared socket in a temp location.
676 self.master_ssh_tempdir = autotemp.tempdir(
677 unique_id='ssh-master')
678 self.master_ssh_option = ("-o ControlPath=%s/socket" %
679 self.master_ssh_tempdir.name)
mblighefccc1b2010-01-11 19:08:42 +0000680
Simran Basi3b858a22015-03-17 16:23:24 -0700681 # Start the master SSH connection in the background.
682 master_cmd = self.ssh_command(
683 options="-N -o ControlMaster=yes")
684 logging.info("Starting master ssh connection '%s'", master_cmd)
685 self.master_ssh_job = utils.BgJob(master_cmd,
686 nickname='master-ssh',
687 no_pipes=True)
688 # To prevent a race between the the master ssh connection
689 # startup and its first attempted use, wait for socket file to
690 # exist before returning.
691 end_time = time.time() + timeout
692 socket_file_path = os.path.join(self.master_ssh_tempdir.name,
693 'socket')
694 while time.time() < end_time:
695 if os.path.exists(socket_file_path):
696 break
697 time.sleep(.2)
698 else:
699 logging.info('Timed out waiting for master-ssh connection '
700 'to be established.')
mbligh0a883702010-04-21 01:58:34 +0000701
702
703 def clear_known_hosts(self):
704 """Clears out the temporary ssh known_hosts file.
705
706 This is useful if the test SSHes to the machine, then reinstalls it,
707 then SSHes to it again. It can be called after the reinstall to
708 reduce the spam in the logs.
709 """
710 logging.info("Clearing known hosts for host '%s', file '%s'.",
Fang Deng3af66202013-08-16 15:19:25 -0700711 self.hostname, self.known_hosts_file)
mbligh0a883702010-04-21 01:58:34 +0000712 # Clear out the file by opening it for writing and then closing.
Fang Deng3af66202013-08-16 15:19:25 -0700713 fh = open(self.known_hosts_file, "w")
mbligh0a883702010-04-21 01:58:34 +0000714 fh.close()
Prashanth B98509c72014-04-04 16:01:34 -0700715
716
717 def collect_logs(self, remote_src_dir, local_dest_dir, ignore_errors=True):
718 """Copy log directories from a host to a local directory.
719
720 @param remote_src_dir: A destination directory on the host.
721 @param local_dest_dir: A path to a local destination directory.
722 If it doesn't exist it will be created.
723 @param ignore_errors: If True, ignore exceptions.
724
725 @raises OSError: If there were problems creating the local_dest_dir and
726 ignore_errors is False.
727 @raises AutoservRunError, AutotestRunError: If something goes wrong
728 while copying the directories and ignore_errors is False.
729 """
730 locally_created_dest = False
731 if (not os.path.exists(local_dest_dir)
732 or not os.path.isdir(local_dest_dir)):
733 try:
734 os.makedirs(local_dest_dir)
735 locally_created_dest = True
736 except OSError as e:
737 logging.warning('Unable to collect logs from host '
738 '%s: %s', self.hostname, e)
739 if not ignore_errors:
740 raise
741 return
742 try:
743 self.get_file(
744 remote_src_dir, local_dest_dir, preserve_symlinks=True)
745 except (error.AutotestRunError, error.AutoservRunError,
746 error.AutoservSSHTimeout) as e:
747 logging.warning('Collection of %s to local dir %s from host %s '
748 'failed: %s', remote_src_dir, local_dest_dir,
749 self.hostname, e)
750 if locally_created_dest:
751 shutil.rmtree(local_dest_dir, ignore_errors=ignore_errors)
752 if not ignore_errors:
753 raise
Cheng-Yi Chianga155e7e2015-08-20 20:42:04 +0800754
755
756 def _create_ssh_tunnel(self, port, local_port):
757 """Create an ssh tunnel from local_port to port.
758
759 @param port: remote port on the host.
760 @param local_port: local forwarding port.
761
762 @return: the tunnel process.
763 """
764 tunnel_options = '-n -N -q -L %d:localhost:%d' % (local_port, port)
765 ssh_cmd = self.make_ssh_command(opts=tunnel_options)
766 tunnel_cmd = '%s %s' % (ssh_cmd, self.hostname)
767 logging.debug('Full tunnel command: %s', tunnel_cmd)
768 tunnel_proc = subprocess.Popen(tunnel_cmd, shell=True, close_fds=True)
769 logging.debug('Started ssh tunnel, local = %d'
770 ' remote = %d, pid = %d',
771 local_port, port, tunnel_proc.pid)
772 return tunnel_proc