diff --git a/cts/CTStests.py b/cts/CTStests.py index 88abffcdf3..58f084fa10 100644 --- a/cts/CTStests.py +++ b/cts/CTStests.py @@ -1,3109 +1,3111 @@ """ Test-specific classes for Pacemaker's Cluster Test Suite (CTS) """ # Pacemaker targets compatibility with Python 2.7 and 3.2+ from __future__ import print_function, unicode_literals, absolute_import, division __copyright__ = """Copyright 2000, 2001 Alan Robertson Add RecourceRecover testcase Zhao Kai """ __license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY" # # SPECIAL NOTE: # # Tests may NOT implement any cluster-manager-specific code in them. # EXTEND the ClusterManager object to provide the base capabilities # the test needs if you need to do something that the current CM classes # do not. Otherwise you screw up the whole point of the object structure # in CTS. # # Thank you. # import os import re import time import subprocess import tempfile from stat import * from cts import CTS from cts.CTSaudits import * from cts.CTSvars import * from cts.patterns import PatternSelector from cts.logging import LogFactory from cts.remote import RemoteFactory, input_wrapper from cts.watcher import LogWatcher from cts.environment import EnvFactory AllTestClasses = [ ] class CTSTest(object): ''' A Cluster test. We implement the basic set of properties and behaviors for a generic cluster test. Cluster tests track their own statistics. We keep each of the kinds of counts we track as separate {name,value} pairs. ''' def __init__(self, cm): #self.name="the unnamed test" self.Stats = {"calls":0 , "success":0 , "failure":0 , "skipped":0 , "auditfail":0} # if not issubclass(cm.__class__, ClusterManager): # raise ValueError("Must be a ClusterManager object") self.CM = cm self.Env = EnvFactory().getInstance() self.rsh = RemoteFactory().getInstance() self.logger = LogFactory() self.templates = PatternSelector(cm["Name"]) self.Audits = [] self.timeout = 120 self.passed = 1 self.is_loop = 0 self.is_unsafe = 0 self.is_docker_unsafe = 0 self.is_experimental = 0 self.is_container = 0 self.is_valgrind = 0 self.benchmark = 0 # which tests to benchmark self.timer = {} # timers def log(self, args): self.logger.log(args) def debug(self, args): self.logger.debug(args) def has_key(self, key): return key in self.Stats def __setitem__(self, key, value): self.Stats[key] = value def __getitem__(self, key): if str(key) == "0": raise ValueError("Bad call to 'foo in X', should reference 'foo in X.Stats' instead") if key in self.Stats: return self.Stats[key] return None def log_mark(self, msg): self.debug("MARK: test %s %s %d" % (self.name,msg,time.time())) return def get_timer(self,key = "test"): try: return self.timer[key] except: return 0 def set_timer(self,key = "test"): self.timer[key] = time.time() return self.timer[key] def log_timer(self,key = "test"): elapsed = 0 if key in self.timer: elapsed = time.time() - self.timer[key] s = key == "test" and self.name or "%s:%s" % (self.name,key) self.debug("%s runtime: %.2f" % (s, elapsed)) del self.timer[key] return elapsed def incr(self, name): '''Increment (or initialize) the value associated with the given name''' if not name in self.Stats: self.Stats[name] = 0 self.Stats[name] = self.Stats[name]+1 # Reset the test passed boolean if name == "calls": self.passed = 1 def failure(self, reason="none"): '''Increment the failure count''' self.passed = 0 self.incr("failure") self.logger.log(("Test %s" % self.name).ljust(35) + " FAILED: %s" % reason) return None def success(self): '''Increment the success count''' self.incr("success") return 1 def skipped(self): '''Increment the skipped count''' self.incr("skipped") return 1 def __call__(self, node): '''Perform the given test''' raise ValueError("Abstract Class member (__call__)") self.incr("calls") return self.failure() def audit(self): passed = 1 if len(self.Audits) > 0: for audit in self.Audits: if not audit(): self.logger.log("Internal %s Audit %s FAILED." % (self.name, audit.name())) self.incr("auditfail") passed = 0 return passed def setup(self, node): '''Setup the given test''' return self.success() def teardown(self, node): '''Tear down the given test''' return self.success() def create_watch(self, patterns, timeout, name=None): if not name: name = self.name return LogWatcher(self.Env["LogFileName"], patterns, name, timeout, kind=self.Env["LogWatcher"], hosts=self.Env["nodes"]) def local_badnews(self, prefix, watch, local_ignore=[]): errcount = 0 if not prefix: prefix = "LocalBadNews:" ignorelist = [] ignorelist.append(" CTS: ") ignorelist.append(prefix) ignorelist.extend(local_ignore) while errcount < 100: match = watch.look(0) if match: add_err = 1 for ignore in ignorelist: if add_err == 1 and re.search(ignore, match): add_err = 0 if add_err == 1: self.logger.log(prefix + " " + match) errcount = errcount + 1 else: break else: self.logger.log("Too many errors!") watch.end() return errcount def is_applicable(self): return self.is_applicable_common() def is_applicable_common(self): '''Return TRUE if we are applicable in the current test configuration''' #raise ValueError("Abstract Class member (is_applicable)") if self.is_loop and not self.Env["loop-tests"]: return 0 elif self.is_unsafe and not self.Env["unsafe-tests"]: return 0 elif self.is_valgrind and not self.Env["valgrind-tests"]: return 0 elif self.is_experimental and not self.Env["experimental-tests"]: return 0 elif self.is_docker_unsafe and self.Env["docker"]: return 0 elif self.is_container and not self.Env["container-tests"]: return 0 elif self.Env["benchmark"] and self.benchmark == 0: return 0 return 1 def find_ocfs2_resources(self, node): self.r_o2cb = None self.r_ocfs2 = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rtype == "o2cb" and r.parent != "NA": self.debug("Found o2cb: %s" % self.r_o2cb) self.r_o2cb = r.parent if re.search("^Constraint", line): c = AuditConstraint(self.CM, line) if c.type == "rsc_colocation" and c.target == self.r_o2cb: self.r_ocfs2.append(c.rsc) self.debug("Found ocfs2 filesystems: %s" % repr(self.r_ocfs2)) return len(self.r_ocfs2) def canrunnow(self, node): '''Return TRUE if we can meaningfully run right now''' return 1 def errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [] class StopTest(CTSTest): '''Stop (deactivate) the cluster manager on a node''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "Stop" def __call__(self, node): '''Perform the 'stop' test. ''' self.incr("calls") if self.CM.ShouldBeStatus[node] != "up": return self.skipped() patterns = [] # Technically we should always be able to notice ourselves stopping patterns.append(self.templates["Pat:We_stopped"] % node) # Any active node needs to notice this one left # (note that this won't work if we have multiple partitions) for other in self.Env["nodes"]: if self.CM.ShouldBeStatus[other] == "up" and other != node: patterns.append(self.templates["Pat:They_stopped"] %(other, self.CM.key_for_node(node))) #self.debug("Checking %s will notice %s left"%(other, node)) watch = self.create_watch(patterns, self.Env["DeadTime"]) watch.setwatch() if node == self.CM.OurNode: self.incr("us") else: if self.CM.upcount() <= 1: self.incr("all") else: self.incr("them") self.CM.StopaCM(node) watch_result = watch.lookforall() failreason = None UnmatchedList = "||" if watch.unmatched: (rc, output) = self.rsh(node, "/bin/ps axf", None) for line in output: self.debug(line) (rc, output) = self.rsh(node, "/usr/sbin/dlm_tool dump", None) for line in output: self.debug(line) for regex in watch.unmatched: self.logger.log ("ERROR: Shutdown pattern not found: %s" % (regex)) UnmatchedList += regex + "||"; failreason = "Missing shutdown pattern" self.CM.cluster_stable(self.Env["DeadTime"]) if not watch.unmatched or self.CM.upcount() == 0: return self.success() if len(watch.unmatched) >= self.CM.upcount(): return self.failure("no match against (%s)" % UnmatchedList) if failreason == None: return self.success() else: return self.failure(failreason) # # We don't register StopTest because it's better when called by # another test... # class StartTest(CTSTest): '''Start (activate) the cluster manager on a node''' def __init__(self, cm, debug=None): CTSTest.__init__(self,cm) self.name = "start" self.debug = debug def __call__(self, node): '''Perform the 'start' test. ''' self.incr("calls") if self.CM.upcount() == 0: self.incr("us") else: self.incr("them") if self.CM.ShouldBeStatus[node] != "down": return self.skipped() elif self.CM.StartaCM(node): return self.success() else: return self.failure("Startup %s on node %s failed" % (self.Env["Name"], node)) # # We don't register StartTest because it's better when called by # another test... # class FlipTest(CTSTest): '''If it's running, stop it. If it's stopped start it. Overthrow the status quo... ''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Flip" self.start = StartTest(cm) self.stop = StopTest(cm) def __call__(self, node): '''Perform the 'Flip' test. ''' self.incr("calls") if self.CM.ShouldBeStatus[node] == "up": self.incr("stopped") ret = self.stop(node) type = "up->down" # Give the cluster time to recognize it's gone... time.sleep(self.Env["StableTime"]) elif self.CM.ShouldBeStatus[node] == "down": self.incr("started") ret = self.start(node) type = "down->up" else: return self.skipped() self.incr(type) if ret: return self.success() else: return self.failure("%s failure" % type) # Register FlipTest as a good test to run AllTestClasses.append(FlipTest) class RestartTest(CTSTest): '''Stop and restart a node''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Restart" self.start = StartTest(cm) self.stop = StopTest(cm) self.benchmark = 1 def __call__(self, node): '''Perform the 'restart' test. ''' self.incr("calls") self.incr("node:" + node) ret1 = 1 if self.CM.StataCM(node): self.incr("WasStopped") if not self.start(node): return self.failure("start (setup) failure: "+node) self.set_timer() if not self.stop(node): return self.failure("stop failure: "+node) if not self.start(node): return self.failure("start failure: "+node) return self.success() # Register RestartTest as a good test to run AllTestClasses.append(RestartTest) class StonithdTest(CTSTest): def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "Stonithd" self.startall = SimulStartLite(cm) self.benchmark = 1 def __call__(self, node): self.incr("calls") if len(self.Env["nodes"]) < 2: return self.skipped() ret = self.startall(None) if not ret: return self.failure("Setup failed") is_dc = self.CM.is_node_dc(node) watchpats = [] watchpats.append(self.templates["Pat:FenceOpOK"] % node) watchpats.append(self.templates["Pat:NodeFenced"] % node) if self.Env["at-boot"] == 0: self.debug("Expecting %s to stay down" % node) self.CM.ShouldBeStatus[node] = "down" else: self.debug("Expecting %s to come up again %d" % (node, self.Env["at-boot"])) watchpats.append("%s.* S_STARTING -> S_PENDING" % node) watchpats.append("%s.* S_PENDING -> S_NOT_DC" % node) watch = self.create_watch(watchpats, 30 + self.Env["DeadTime"] + self.Env["StableTime"] + self.Env["StartTime"]) watch.setwatch() origin = self.Env.RandomGen.choice(self.Env["nodes"]) rc = self.rsh(origin, "stonith_admin --reboot %s -VVVVVV" % node) if rc == 194: # 194 - 256 = -62 = Timer expired # # Look for the patterns, usually this means the required # device was running on the node to be fenced - or that # the required devices were in the process of being loaded # and/or moved # # Effectively the node committed suicide so there will be # no confirmation, but pacemaker should be watching and # fence the node again self.logger.log("Fencing command on %s to fence %s timed out" % (origin, node)) elif origin != node and rc != 0: self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting for fenced node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.logger.log("Fencing command on %s failed to fence %s (rc=%d)" % (origin, node, rc)) elif origin == node and rc != 255: # 255 == broken pipe, ie. the node was fenced as expected self.logger.log("Locally originated fencing returned %d" % rc) self.set_timer("fence") matched = watch.lookforall() self.log_timer("fence") self.set_timer("reform") if watch.unmatched: self.logger.log("Patterns not found: " + repr(watch.unmatched)) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting for fenced node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.debug("Waiting for the cluster to re-stabilize with all nodes") is_stable = self.CM.cluster_stable(self.Env["StartTime"]) if not matched: return self.failure("Didn't find all expected patterns") elif not is_stable: return self.failure("Cluster did not become stable") self.log_timer("reform") return self.success() def errorstoignore(self): return [ self.templates["Pat:Fencing_start"] % ".*", self.templates["Pat:Fencing_ok"] % ".*", r"error.*: Resource .*stonith::.* is active on 2 nodes attempting recovery", r"error.*: Operation reboot of .*by .* for stonith_admin.*: Timer expired", ] def is_applicable(self): if not self.is_applicable_common(): return 0 if "DoFencing" in list(self.Env.keys()): return self.Env["DoFencing"] return 1 AllTestClasses.append(StonithdTest) class StartOnebyOne(CTSTest): '''Start all the nodes ~ one by one''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "StartOnebyOne" self.stopall = SimulStopLite(cm) self.start = StartTest(cm) self.ns = CTS.NodeStatus(cm.Env) def __call__(self, dummy): '''Perform the 'StartOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Test setup failed") failed = [] self.set_timer() for node in self.Env["nodes"]: if not self.start(node): failed.append(node) if len(failed) > 0: return self.failure("Some node failed to start: " + repr(failed)) return self.success() # Register StartOnebyOne as a good test to run AllTestClasses.append(StartOnebyOne) class SimulStart(CTSTest): '''Start all the nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStart" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) def __call__(self, dummy): '''Perform the 'SimulStart' test. ''' self.incr("calls") # We ignore the "node" parameter... # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Setup failed") if not self.startall(None): return self.failure("Startall failed") return self.success() # Register SimulStart as a good test to run AllTestClasses.append(SimulStart) class SimulStop(CTSTest): '''Stop all the nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStop" self.startall = SimulStartLite(cm) self.stopall = SimulStopLite(cm) def __call__(self, dummy): '''Perform the 'SimulStop' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") if not self.stopall(None): return self.failure("Stopall failed") return self.success() # Register SimulStop as a good test to run AllTestClasses.append(SimulStop) class StopOnebyOne(CTSTest): '''Stop all the nodes in order''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "StopOnebyOne" self.startall = SimulStartLite(cm) self.stop = StopTest(cm) def __call__(self, dummy): '''Perform the 'StopOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") failed = [] self.set_timer() for node in self.Env["nodes"]: if not self.stop(node): failed.append(node) if len(failed) > 0: return self.failure("Some node failed to stop: " + repr(failed)) return self.success() # Register StopOnebyOne as a good test to run AllTestClasses.append(StopOnebyOne) class RestartOnebyOne(CTSTest): '''Restart all the nodes in order''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RestartOnebyOne" self.startall = SimulStartLite(cm) def __call__(self, dummy): '''Perform the 'RestartOnebyOne' test. ''' self.incr("calls") # We ignore the "node" parameter... # Start up all the nodes... ret = self.startall(None) if not ret: return self.failure("Setup failed") did_fail = [] self.set_timer() self.restart = RestartTest(self.CM) for node in self.Env["nodes"]: if not self.restart(node): did_fail.append(node) if did_fail: return self.failure("Could not restart %d nodes: %s" % (len(did_fail), repr(did_fail))) return self.success() # Register StopOnebyOne as a good test to run AllTestClasses.append(RestartOnebyOne) class PartialStart(CTSTest): '''Start a node - but tell it to stop before it finishes starting up''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "PartialStart" self.startall = SimulStartLite(cm) self.stopall = SimulStopLite(cm) self.stop = StopTest(cm) #self.is_unsafe = 1 def __call__(self, node): '''Perform the 'PartialStart' test. ''' self.incr("calls") ret = self.stopall(None) if not ret: return self.failure("Setup failed") # FIXME! This should use the CM class to get the pattern # then it would be applicable in general watchpats = [] watchpats.append("pacemaker-controld.*Connecting to cluster infrastructure") watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() self.CM.StartaCMnoBlock(node) ret = watch.lookforall() if not ret: self.logger.log("Patterns not found: " + repr(watch.unmatched)) return self.failure("Setup of %s failed" % node) ret = self.stop(node) if not ret: return self.failure("%s did not stop in time" % node) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # We might do some fencing in the 2-node case if we make it up far enough return [ r"Executing reboot fencing operation", r"Requesting fencing \([^)]+\) of node ", ] # Register StopOnebyOne as a good test to run AllTestClasses.append(PartialStart) class StandbyTest(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Standby" self.benchmark = 1 self.start = StartTest(cm) self.startall = SimulStartLite(cm) # make sure the node is active # set the node to standby mode # check resources, none resource should be running on the node # set the node to active mode # check resouces, resources should have been migrated back (SHOULD THEY?) def __call__(self, node): self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Start all nodes failed") self.debug("Make sure node %s is active" % node) if self.CM.StandbyStatus(node) != "off": if not self.CM.SetStandbyMode(node, "off"): return self.failure("can't set node %s to active mode" % node) self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "off": return self.failure("standby status of %s is [%s] but we expect [off]" % (node, status)) self.debug("Getting resources running on node %s" % node) rsc_on_node = self.CM.active_resources(node) watchpats = [] watchpats.append(r"State transition .* -> S_POLICY_ENGINE") watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() self.debug("Setting node %s to standby mode" % node) if not self.CM.SetStandbyMode(node, "on"): return self.failure("can't set node %s to standby mode" % node) self.set_timer("on") ret = watch.lookforall() if not ret: self.logger.log("Patterns not found: " + repr(watch.unmatched)) self.CM.SetStandbyMode(node, "off") return self.failure("cluster didn't react to standby change on %s" % node) self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "on": return self.failure("standby status of %s is [%s] but we expect [on]" % (node, status)) self.log_timer("on") self.debug("Checking resources") bad_run = self.CM.active_resources(node) if len(bad_run) > 0: rc = self.failure("%s set to standby, %s is still running on it" % (node, repr(bad_run))) self.debug("Setting node %s to active mode" % node) self.CM.SetStandbyMode(node, "off") return rc self.debug("Setting node %s to active mode" % node) if not self.CM.SetStandbyMode(node, "off"): return self.failure("can't set node %s to active mode" % node) self.set_timer("off") self.CM.cluster_stable() status = self.CM.StandbyStatus(node) if status != "off": return self.failure("standby status of %s is [%s] but we expect [off]" % (node, status)) self.log_timer("off") return self.success() AllTestClasses.append(StandbyTest) class ValgrindTest(CTSTest): '''Check for memory leaks''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Valgrind" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) self.is_valgrind = 1 self.is_loop = 1 def setup(self, node): self.incr("calls") ret = self.stopall(None) if not ret: return self.failure("Stop all nodes failed") # @TODO Edit /etc/sysconfig/pacemaker on all nodes to enable valgrind, # and clear any valgrind logs from previous runs. For now, we rely on # the user to do this manually. ret = self.startall(None) if not ret: return self.failure("Start all nodes failed") return self.success() def teardown(self, node): # Return all nodes to normal # @TODO Edit /etc/sysconfig/pacemaker on all nodes to disable valgrind ret = self.stopall(None) if not ret: return self.failure("Stop all nodes failed") return self.success() def find_leaks(self): # Check for leaks + # (no longer used but kept in case feature is restored) leaked = [] self.stop = StopTest(self.CM) for node in self.Env["nodes"]: rc = self.stop(node) if not rc: self.failure("Couldn't shut down %s" % node) rc = self.rsh(node, "grep -e indirectly.*lost:.*[1-9] -e definitely.*lost:.*[1-9] -e (ERROR|error).*SUMMARY:.*[1-9].*errors %s" % self.logger.logPat, 0) if rc != 1: leaked.append(node) self.failure("Valgrind errors detected on %s" % node) (rc, output) = self.rsh(node, "grep -e lost: -e SUMMARY: %s" % self.logger.logPat, None) for line in output: self.logger.log(line) (rc, output) = self.rsh(node, "cat %s" % self.logger.logPat, None) for line in output: self.debug(line) self.rsh(node, "rm -f %s" % self.logger.logPat, None) return leaked def __call__(self, node): - leaked = self.find_leaks() - if len(leaked) > 0: - return self.failure("Nodes %s leaked" % repr(leaked)) + #leaked = self.find_leaks() + #if len(leaked) > 0: + # return self.failure("Nodes %s leaked" % repr(leaked)) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"pacemaker-based.*: \*\*\*\*\*\*\*\*\*\*\*\*\*", r"pacemaker-based.*: .* avoid confusing Valgrind", r"HA_VALGRIND_ENABLED", ] class StandbyLoopTest(ValgrindTest): '''Check for memory leaks by putting a node in and out of standby for an hour''' + # @TODO This is not a useful test for memory leaks def __init__(self, cm): ValgrindTest.__init__(self,cm) self.name = "StandbyLoop" def __call__(self, node): lpc = 0 delay = 2 failed = 0 done = time.time() + self.Env["loop-minutes"] * 60 while time.time() <= done and not failed: lpc = lpc + 1 time.sleep(delay) if not self.CM.SetStandbyMode(node, "on"): self.failure("can't set node %s to standby mode" % node) failed = lpc time.sleep(delay) if not self.CM.SetStandbyMode(node, "off"): self.failure("can't set node %s to active mode" % node) failed = lpc leaked = self.find_leaks() if failed: return self.failure("Iteration %d failed" % failed) elif len(leaked) > 0: return self.failure("Nodes %s leaked" % repr(leaked)) return self.success() -AllTestClasses.append(StandbyLoopTest) +#AllTestClasses.append(StandbyLoopTest) class BandwidthTest(CTSTest): # Tests should not be cluster-manager-specific # If you need to find out cluster manager configuration to do this, then # it should be added to the generic cluster manager API. '''Test the bandwidth which the cluster uses''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "Bandwidth" self.start = StartTest(cm) self.__setitem__("min",0) self.__setitem__("max",0) self.__setitem__("totalbandwidth",0) (handle, self.tempfile) = tempfile.mkstemp(".cts") os.close(handle) self.startall = SimulStartLite(cm) def __call__(self, node): '''Perform the Bandwidth test''' self.incr("calls") if self.CM.upcount() < 1: return self.skipped() Path = self.CM.InternalCommConfig() if "ip" not in Path["mediatype"]: return self.skipped() port = Path["port"][0] port = int(port) ret = self.startall(None) if not ret: return self.failure("Test setup failed") time.sleep(5) # We get extra messages right after startup. fstmpfile = "/var/run/band_estimate" dumpcmd = "tcpdump -p -n -c 102 -i any udp port %d > %s 2>&1" \ % (port, fstmpfile) rc = self.rsh(node, dumpcmd) if rc == 0: farfile = "root@%s:%s" % (node, fstmpfile) self.rsh.cp(farfile, self.tempfile) Bandwidth = self.countbandwidth(self.tempfile) if not Bandwidth: self.logger.log("Could not compute bandwidth.") return self.success() intband = int(Bandwidth + 0.5) self.logger.log("...bandwidth: %d bits/sec" % intband) self.Stats["totalbandwidth"] = self.Stats["totalbandwidth"] + Bandwidth if self.Stats["min"] == 0: self.Stats["min"] = Bandwidth if Bandwidth > self.Stats["max"]: self.Stats["max"] = Bandwidth if Bandwidth < self.Stats["min"]: self.Stats["min"] = Bandwidth self.rsh(node, "rm -f %s" % fstmpfile) os.unlink(self.tempfile) return self.success() else: return self.failure("no response from tcpdump command [%d]!" % rc) def countbandwidth(self, file): fp = open(file, "r") fp.seek(0) count = 0 sum = 0 while 1: line = fp.readline() if not line: return None if re.search("udp",line) or re.search("UDP,", line): count = count + 1 linesplit = line.split(" ") for j in range(len(linesplit)-1): if linesplit[j] == "udp": break if linesplit[j] == "length:": break try: sum = sum + int(linesplit[j+1]) except ValueError: self.logger.log("Invalid tcpdump line: %s" % line) return None T1 = linesplit[0] timesplit = T1.split(":") time2split = timesplit[2].split(".") time1 = (int(timesplit[0])*60+int(timesplit[1]))*60+int(time2split[0])+int(time2split[1])*0.000001 break while count < 100: line = fp.readline() if not line: return None if re.search("udp",line) or re.search("UDP,", line): count = count+1 linessplit = line.split(" ") for j in range(len(linessplit)-1): if linessplit[j] == "udp": break if linesplit[j] == "length:": break try: sum = int(linessplit[j+1]) + sum except ValueError: self.logger.log("Invalid tcpdump line: %s" % line) return None T2 = linessplit[0] timesplit = T2.split(":") time2split = timesplit[2].split(".") time2 = (int(timesplit[0])*60+int(timesplit[1]))*60+int(time2split[0])+int(time2split[1])*0.000001 time = time2-time1 if (time <= 0): return 0 return int((sum*8)/time) def is_applicable(self): '''BandwidthTest never applicable''' return 0 AllTestClasses.append(BandwidthTest) ################################################################### class MaintenanceMode(CTSTest): ################################################################### def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "MaintenanceMode" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.max = 30 #self.is_unsafe = 1 self.benchmark = 1 self.action = "asyncmon" self.interval = 0 self.rid = "maintenanceDummy" def toggleMaintenanceMode(self, node, action): pats = [] pats.append(self.templates["Pat:DC_IDLE"]) # fail the resource right after turning Maintenance mode on # verify it is not recovered until maintenance mode is turned off if action == "On": pats.append(r"schedulerd.*:\s+warning:.*Processing failed %s of %s on" % (self.action, self.rid)) else: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.rid)) pats.append(self.templates["Pat:RscOpOK"] % ("start", self.rid)) watch = self.create_watch(pats, 60) watch.setwatch() self.debug("Turning maintenance mode %s" % action) self.rsh(node, self.templates["MaintenanceMode%s" % (action)]) if (action == "On"): self.rsh(node, "crm_resource -V -F -r %s -H %s &>/dev/null" % (self.rid, node)) self.set_timer("recover%s" % (action)) watch.lookforall() self.log_timer("recover%s" % (action)) if watch.unmatched: self.debug("Failed to find patterns when turning maintenance mode %s" % action) return repr(watch.unmatched) return "" def insertMaintenanceDummy(self, node): pats = [] pats.append(("%s.*" % node) + (self.templates["Pat:RscOpOK"] % ("start", self.rid))) watch = self.create_watch(pats, 60) watch.setwatch() self.CM.AddDummyRsc(node, self.rid) self.set_timer("addDummy") watch.lookforall() self.log_timer("addDummy") if watch.unmatched: self.debug("Failed to find patterns when adding maintenance dummy resource") return repr(watch.unmatched) return "" def removeMaintenanceDummy(self, node): pats = [] pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.rid)) watch = self.create_watch(pats, 60) watch.setwatch() self.CM.RemoveDummyRsc(node, self.rid) self.set_timer("removeDummy") watch.lookforall() self.log_timer("removeDummy") if watch.unmatched: self.debug("Failed to find patterns when removing maintenance dummy resource") return repr(watch.unmatched) return "" def managedRscList(self, node): rscList = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if tmp.managed(): rscList.append(tmp.id) return rscList def verifyResources(self, node, rscList, managed): managedList = list(rscList) managed_str = "managed" if not managed: managed_str = "unmanaged" (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if managed and not tmp.managed(): continue elif not managed and tmp.managed(): continue elif managedList.count(tmp.id): managedList.remove(tmp.id) if len(managedList) == 0: self.debug("Found all %s resources on %s" % (managed_str, node)) return True self.logger.log("Could not find all %s resources on %s. %s" % (managed_str, node, managedList)) return False def __call__(self, node): '''Perform the 'MaintenanceMode' test. ''' self.incr("calls") verify_managed = False verify_unmanaged = False failPat = "" ret = self.startall(None) if not ret: return self.failure("Setup failed") # get a list of all the managed resources. We use this list # after enabling maintenance mode to verify all managed resources # become un-managed. After maintenance mode is turned off, we use # this list to verify all the resources become managed again. managedResources = self.managedRscList(node) if len(managedResources) == 0: self.logger.log("No managed resources on %s" % node) return self.skipped() # insert a fake resource we can fail during maintenance mode # so we can verify recovery does not take place until after maintenance # mode is disabled. failPat = failPat + self.insertMaintenanceDummy(node) # toggle maintenance mode ON, then fail dummy resource. failPat = failPat + self.toggleMaintenanceMode(node, "On") # verify all the resources are now unmanaged if self.verifyResources(node, managedResources, False): verify_unmanaged = True # Toggle maintenance mode OFF, verify dummy is recovered. failPat = failPat + self.toggleMaintenanceMode(node, "Off") # verify all the resources are now managed again if self.verifyResources(node, managedResources, True): verify_managed = True # Remove our maintenance dummy resource. failPat = failPat + self.removeMaintenanceDummy(node) self.CM.cluster_stable() if failPat != "": return self.failure("Unmatched patterns: %s" % (failPat)) elif verify_unmanaged is False: return self.failure("Failed to verify resources became unmanaged during maintenance mode") elif verify_managed is False: return self.failure("Failed to verify resources switched back to managed after disabling maintenance mode") return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"Updating failcount for %s" % self.rid, r"schedulerd.*: Recover %s\s*\(.*\)" % self.rid, r"Unknown operation: fail", self.templates["Pat:RscOpOK"] % (self.action, self.rid), r"(ERROR|error).*: Action %s_%s_%d .* initiated outside of a transition" % (self.rid, self.action, self.interval), ] AllTestClasses.append(MaintenanceMode) class ResourceRecover(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "ResourceRecover" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.max = 30 self.rid = None self.rid_alt = None #self.is_unsafe = 1 self.benchmark = 1 # these are the values used for the new LRM API call self.action = "asyncmon" self.interval = 0 def __call__(self, node): '''Perform the 'ResourceRecover' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed") resourcelist = self.CM.active_resources(node) # if there are no resourcelist, return directly if len(resourcelist) == 0: self.logger.log("No active resources on %s" % node) return self.skipped() self.rid = self.Env.RandomGen.choice(resourcelist) self.rid_alt = self.rid rsc = None (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): tmp = AuditResource(self.CM, line) if tmp.id == self.rid: rsc = tmp # Handle anonymous clones that get renamed self.rid = rsc.clone_id break if not rsc: return self.failure("Could not find %s in the resource list" % self.rid) self.debug("Shooting %s aka. %s" % (rsc.clone_id, rsc.id)) pats = [] pats.append(r"schedulerd.*:\s+warning:.*Processing failed %s of (%s|%s) on" % (self.action, rsc.id, rsc.clone_id)) if rsc.managed(): pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.rid)) if rsc.unique(): pats.append(self.templates["Pat:RscOpOK"] % ("start", self.rid)) else: # Anonymous clones may get restarted with a different clone number pats.append(self.templates["Pat:RscOpOK"] % ("start", ".*")) watch = self.create_watch(pats, 60) watch.setwatch() self.rsh(node, "crm_resource -V -F -r %s -H %s &>/dev/null" % (self.rid, node)) self.set_timer("recover") watch.lookforall() self.log_timer("recover") self.CM.cluster_stable() recovered = self.CM.ResourceLocation(self.rid) if watch.unmatched: return self.failure("Patterns not found: %s" % repr(watch.unmatched)) elif rsc.unique() and len(recovered) > 1: return self.failure("%s is now active on more than one node: %s"%(self.rid, repr(recovered))) elif len(recovered) > 0: self.debug("%s is running on: %s" % (self.rid, repr(recovered))) elif rsc.managed(): return self.failure("%s was not recovered and is inactive" % self.rid) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"Updating failcount for %s" % self.rid, r"schedulerd.*: Recover (%s|%s)\s*\(.*\)" % (self.rid, self.rid_alt), r"Unknown operation: fail", self.templates["Pat:RscOpOK"] % (self.action, self.rid), r"(ERROR|error).*: Action %s_%s_%d .* initiated outside of a transition" % (self.rid, self.action, self.interval), ] AllTestClasses.append(ResourceRecover) class ComponentFail(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "ComponentFail" # TODO make this work correctly in docker. self.is_docker_unsafe = 1 self.startall = SimulStartLite(cm) self.complist = cm.Components() self.patterns = [] self.okerrpatterns = [] self.is_unsafe = 1 def __call__(self, node): '''Perform the 'ComponentFail' test. ''' self.incr("calls") self.patterns = [] self.okerrpatterns = [] # start all nodes ret = self.startall(None) if not ret: return self.failure("Setup failed") if not self.CM.cluster_stable(self.Env["StableTime"]): return self.failure("Setup failed - unstable") node_is_dc = self.CM.is_node_dc(node, None) # select a component to kill chosen = self.Env.RandomGen.choice(self.complist) while chosen.dc_only == 1 and node_is_dc == 0: chosen = self.Env.RandomGen.choice(self.complist) self.debug("...component %s (dc=%d,boot=%d)" % (chosen.name, node_is_dc,chosen.triggersreboot)) self.incr(chosen.name) if chosen.name != "corosync": self.patterns.append(self.templates["Pat:ChildKilled"] %(node, chosen.name)) self.patterns.append(self.templates["Pat:ChildRespawn"] %(node, chosen.name)) self.patterns.extend(chosen.pats) if node_is_dc: self.patterns.extend(chosen.dc_pats) if chosen.name == "pacemaker-fenced": # Ignore actions for STONITH resources (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rclass == "stonith": self.okerrpatterns.append(self.templates["Pat:Fencing_recover"] % r.id) # supply a copy so self.patterns doesn't end up empty tmpPats = [] tmpPats.extend(self.patterns) self.patterns.extend(chosen.badnews_ignore) # Look for STONITH ops, depending on Env["at-boot"] we might need to change the nodes status stonithPats = [] stonithPats.append(self.templates["Pat:Fencing_ok"] % node) stonith = self.create_watch(stonithPats, 0) stonith.setwatch() # set the watch for stable watch = self.create_watch( tmpPats, self.Env["DeadTime"] + self.Env["StableTime"] + self.Env["StartTime"]) watch.setwatch() # kill the component chosen.kill(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() self.debug("Waiting for any fenced node to come back up") self.CM.ns.WaitForAllNodesToComeUp(self.Env["nodes"], 600) self.debug("Waiting for the cluster to re-stabilize with all nodes") self.CM.cluster_stable(self.Env["StartTime"]) self.debug("Checking if %s was shot" % node) shot = stonith.look(60) if shot: self.debug("Found: " + repr(shot)) self.okerrpatterns.append(self.templates["Pat:Fencing_start"] % node) if self.Env["at-boot"] == 0: self.CM.ShouldBeStatus[node] = "down" # If fencing occurred, chances are many (if not all) the expected logs # will not be sent - or will be lost when the node reboots return self.success() # check for logs indicating a graceful recovery matched = watch.lookforall(allow_multiple_matches=1) if watch.unmatched: self.logger.log("Patterns not found: " + repr(watch.unmatched)) self.debug("Waiting for the cluster to re-stabilize with all nodes") is_stable = self.CM.cluster_stable(self.Env["StartTime"]) if not matched: return self.failure("Didn't find all expected %s patterns" % chosen.name) elif not is_stable: return self.failure("Cluster did not become stable after killing %s" % chosen.name) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # Note that okerrpatterns refers to the last time we ran this test # The good news is that this works fine for us... self.okerrpatterns.extend(self.patterns) return self.okerrpatterns AllTestClasses.append(ComponentFail) class SplitBrainTest(CTSTest): '''It is used to test split-brain. when the path between the two nodes break check the two nodes both take over the resource''' def __init__(self,cm): CTSTest.__init__(self,cm) self.name = "SplitBrain" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.is_experimental = 1 def isolate_partition(self, partition): other_nodes = [] other_nodes.extend(self.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.logger.log("Node "+node+" not in " + repr(self.Env["nodes"]) + " from " +repr(partition)) if len(other_nodes) == 0: return 1 self.debug("Creating partition: " + repr(partition)) self.debug("Everyone else: " + repr(other_nodes)) for node in partition: if not self.CM.isolate_node(node, other_nodes): self.logger.log("Could not isolate %s" % node) return 0 return 1 def heal_partition(self, partition): other_nodes = [] other_nodes.extend(self.Env["nodes"]) for node in partition: try: other_nodes.remove(node) except ValueError: self.logger.log("Node "+node+" not in " + repr(self.Env["nodes"])) if len(other_nodes) == 0: return 1 self.debug("Healing partition: " + repr(partition)) self.debug("Everyone else: " + repr(other_nodes)) for node in partition: self.CM.unisolate_node(node, other_nodes) def __call__(self, node): '''Perform split-brain test''' self.incr("calls") self.passed = 1 partitions = {} ret = self.startall(None) if not ret: return self.failure("Setup failed") while 1: # Retry until we get multiple partitions partitions = {} p_max = len(self.Env["nodes"]) for node in self.Env["nodes"]: p = self.Env.RandomGen.randint(1, p_max) if not p in partitions: partitions[p] = [] partitions[p].append(node) p_max = len(list(partitions.keys())) if p_max > 1: break # else, try again self.debug("Created %d partitions" % p_max) for key in list(partitions.keys()): self.debug("Partition["+str(key)+"]:\t"+repr(partitions[key])) # Disabling STONITH to reduce test complexity for now self.rsh(node, "crm_attribute -V -n stonith-enabled -v false") for key in list(partitions.keys()): self.isolate_partition(partitions[key]) count = 30 while count > 0: if len(self.CM.find_partitions()) != p_max: time.sleep(10) else: break else: self.failure("Expected partitions were not created") # Target number of partitions formed - wait for stability if not self.CM.cluster_stable(): self.failure("Partitioned cluster not stable") # Now audit the cluster state self.CM.partitions_expected = p_max if not self.audit(): self.failure("Audits failed") self.CM.partitions_expected = 1 # And heal them again for key in list(partitions.keys()): self.heal_partition(partitions[key]) # Wait for a single partition to form count = 30 while count > 0: if len(self.CM.find_partitions()) != 1: time.sleep(10) count -= 1 else: break else: self.failure("Cluster did not reform") # Wait for it to have the right number of members count = 30 while count > 0: members = [] partitions = self.CM.find_partitions() if len(partitions) > 0: members = partitions[0].split() if len(members) != len(self.Env["nodes"]): time.sleep(10) count -= 1 else: break else: self.failure("Cluster did not completely reform") # Wait up to 20 minutes - the delay is more preferable than # trying to continue with in a messed up state if not self.CM.cluster_stable(1200): self.failure("Reformed cluster not stable") if self.Env["continue"] == 1: answer = "Y" else: try: answer = input_wrapper('Continue? [nY]') except EOFError as e: answer = "n" if answer and answer == "n": raise ValueError("Reformed cluster not stable") # Turn fencing back on if self.Env["DoFencing"]: self.rsh(node, "crm_attribute -V -D -n stonith-enabled") self.CM.cluster_stable() if self.passed: return self.success() return self.failure("See previous errors") def errorstoignore(self): '''Return list of errors which are 'normal' and should be ignored''' return [ r"Another DC detected:", r"(ERROR|error).*: .*Application of an update diff failed", r"pacemaker-controld.*:.*not in our membership list", r"CRIT:.*node.*returning after partition", ] def is_applicable(self): if not self.is_applicable_common(): return 0 return len(self.Env["nodes"]) > 2 AllTestClasses.append(SplitBrainTest) class Reattach(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "Reattach" self.startall = SimulStartLite(cm) self.restart1 = RestartTest(cm) self.stopall = SimulStopLite(cm) self.is_unsafe = 0 # Handled by canrunnow() def _is_managed(self, node): is_managed = self.rsh(node, "crm_attribute -t rsc_defaults -n is-managed -q -G -d true", 1) is_managed = is_managed[:-1] # Strip off the newline return is_managed == "true" def _set_unmanaged(self, node): self.debug("Disable resource management") self.rsh(node, "crm_attribute -t rsc_defaults -n is-managed -v false") def _set_managed(self, node): self.debug("Re-enable resource management") self.rsh(node, "crm_attribute -t rsc_defaults -n is-managed -D") def setup(self, node): attempt = 0 if not self.startall(None): return None # Make sure we are really _really_ stable and that all # resources, including those that depend on transient node # attributes, are started while not self.CM.cluster_stable(double_check=True): if attempt < 5: attempt += 1 self.debug("Not stable yet, re-testing") else: self.logger.log("Cluster is not stable") return None return 1 def teardown(self, node): # Make sure 'node' is up start = StartTest(self.CM) start(node) if not self._is_managed(node): self.logger.log("Attempting to re-enable resource management on %s" % node) self._set_managed(node) self.CM.cluster_stable() if not self._is_managed(node): self.logger.log("Could not re-enable resource management") return 0 return 1 def canrunnow(self, node): '''Return TRUE if we can meaningfully run right now''' if self.find_ocfs2_resources(node): self.logger.log("Detach/Reattach scenarios are not possible with OCFS2 services present") return 0 return 1 def __call__(self, node): self.incr("calls") pats = [] # Conveniently, the scheduler will display this message when disabling # management, even if fencing is not enabled, so we can rely on it. managed = self.create_watch(["Delaying fencing operations"], 60) managed.setwatch() self._set_unmanaged(node) if not managed.lookforall(): self.logger.log("Patterns not found: " + repr(managed.unmatched)) return self.failure("Resource management not disabled") pats = [] pats.append(self.templates["Pat:RscOpOK"] % ("start", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("stop", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("promote", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("demote", ".*")) pats.append(self.templates["Pat:RscOpOK"] % ("migrate", ".*")) watch = self.create_watch(pats, 60, "ShutdownActivity") watch.setwatch() self.debug("Shutting down the cluster") ret = self.stopall(None) if not ret: self._set_managed(node) return self.failure("Couldn't shut down the cluster") self.debug("Bringing the cluster back up") ret = self.startall(None) time.sleep(5) # allow ping to update the CIB if not ret: self._set_managed(node) return self.failure("Couldn't restart the cluster") if self.local_badnews("ResourceActivity:", watch): self._set_managed(node) return self.failure("Resources stopped or started during cluster restart") watch = self.create_watch(pats, 60, "StartupActivity") watch.setwatch() # Re-enable resource management (and verify it happened). self._set_managed(node) self.CM.cluster_stable() if not self._is_managed(node): return self.failure("Could not re-enable resource management") # Ignore actions for STONITH resources ignore = [] (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rclass == "stonith": self.debug("Ignoring start actions for %s" % r.id) ignore.append(self.templates["Pat:RscOpOK"] % ("start", r.id)) if self.local_badnews("ResourceActivity:", watch, ignore): return self.failure("Resources stopped or started after resource management was re-enabled") return ret def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"resource( was|s were) active at shutdown", ] def is_applicable(self): return 1 AllTestClasses.append(Reattach) class SpecialTest1(CTSTest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SpecialTest1" self.startall = SimulStartLite(cm) self.restart1 = RestartTest(cm) self.stopall = SimulStopLite(cm) def __call__(self, node): '''Perform the 'SpecialTest1' test for Andrew. ''' self.incr("calls") # Shut down all the nodes... ret = self.stopall(None) if not ret: return self.failure("Could not stop all nodes") # Test config recovery when the other nodes come up self.rsh(node, "rm -f "+CTSvars.CRM_CONFIG_DIR+"/cib*") # Start the selected node ret = self.restart1(node) if not ret: return self.failure("Could not start "+node) # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Could not start the remaining nodes") return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' # Errors that occur as a result of the CIB being wiped return [ r"error.*: v1 patchset error, patch failed to apply: Application of an update diff failed", r"error.*: Resource start-up disabled since no STONITH resources have been defined", r"error.*: Either configure some or disable STONITH with the stonith-enabled option", r"error.*: NOTE: Clusters with shared data need STONITH to ensure data integrity", ] AllTestClasses.append(SpecialTest1) class HAETest(CTSTest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "HAETest" self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) self.is_loop = 1 def setup(self, node): # Start all remaining nodes ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") return self.success() def wait_on_state(self, node, resource, expected_clones, attempts=240): while attempts > 0: active = 0 (rc, lines) = self.rsh(node, "crm_resource -r %s -W -Q" % resource, stdout=None) # Hack until crm_resource does the right thing if rc == 0 and lines: active = len(lines) if len(lines) == expected_clones: return 1 elif rc == 1: self.debug("Resource %s is still inactive" % resource) elif rc == 234: self.logger.log("Unknown resource %s" % resource) return 0 elif rc == 246: self.logger.log("Cluster is inactive") return 0 elif rc != 0: self.logger.log("Call to crm_resource failed, rc=%d" % rc) return 0 else: self.debug("Resource %s is active on %d times instead of %d" % (resource, active, expected_clones)) attempts -= 1 time.sleep(1) return 0 def find_dlm(self, node): self.r_dlm = None (rc, lines) = self.rsh(node, "crm_resource -c", None) for line in lines: if re.search("^Resource", line): r = AuditResource(self.CM, line) if r.rtype == "controld" and r.parent != "NA": self.debug("Found dlm: %s" % self.r_dlm) self.r_dlm = r.parent return 1 return 0 def find_hae_resources(self, node): self.r_dlm = None self.r_o2cb = None self.r_ocfs2 = [] if self.find_dlm(node): self.find_ocfs2_resources(node) def is_applicable(self): if not self.is_applicable_common(): return 0 if self.Env["Schema"] == "hae": return 1 return None class HAERoleTest(HAETest): def __init__(self, cm): '''Lars' mount/unmount test for the HA extension. ''' HAETest.__init__(self,cm) self.name = "HAERoleTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_resource -V -r %s -p target-role -v %s --meta" % (resource, target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 delay = 2 done = time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "Stopped") if not self.wait_on_state(node, self.r_dlm, 0): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "Started") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAERoleTest) class HAEStandbyTest(HAETest): '''Set up a custom test to cause quorum failure issues for Andrew''' def __init__(self, cm): HAETest.__init__(self,cm) self.name = "HAEStandbyTest" def change_state(self, node, resource, target): rc = self.rsh(node, "crm_standby -V -l reboot -v %s" % (target)) return rc def __call__(self, node): self.incr("calls") lpc = 0 failed = 0 done = time.time() + self.Env["loop-minutes"]*60 self.find_hae_resources(node) clone_max = len(self.Env["nodes"]) while time.time() <= done and not failed: lpc = lpc + 1 self.change_state(node, self.r_dlm, "true") if not self.wait_on_state(node, self.r_dlm, clone_max-1): self.failure("%s did not go down correctly" % self.r_dlm) failed = lpc self.change_state(node, self.r_dlm, "false") if not self.wait_on_state(node, self.r_dlm, clone_max): self.failure("%s did not come up correctly" % self.r_dlm) failed = lpc if not self.wait_on_state(node, self.r_o2cb, clone_max): self.failure("%s did not come up correctly" % self.r_o2cb) failed = lpc for fs in self.r_ocfs2: if not self.wait_on_state(node, fs, clone_max): self.failure("%s did not come up correctly" % fs) failed = lpc if failed: return self.failure("iteration %d failed" % failed) return self.success() AllTestClasses.append(HAEStandbyTest) class NearQuorumPointTest(CTSTest): ''' This test brings larger clusters near the quorum point (50%). In addition, it will test doing starts and stops at the same time. Here is how I think it should work: - loop over the nodes and decide randomly which will be up and which will be down Use a 50% probability for each of up/down. - figure out what to do to get into that state from the current state - in parallel, bring up those going up and bring those going down. ''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "NearQuorumPoint" def __call__(self, dummy): '''Perform the 'NearQuorumPoint' test. ''' self.incr("calls") startset = [] stopset = [] stonith = self.CM.prepare_fencing_watcher("NearQuorumPoint") #decide what to do with each node for node in self.Env["nodes"]: action = self.Env.RandomGen.choice(["start","stop"]) #action = self.Env.RandomGen.choice(["start","stop","no change"]) if action == "start" : startset.append(node) elif action == "stop" : stopset.append(node) self.debug("start nodes:" + repr(startset)) self.debug("stop nodes:" + repr(stopset)) #add search patterns watchpats = [ ] for node in stopset: if self.CM.ShouldBeStatus[node] == "up": watchpats.append(self.templates["Pat:We_stopped"] % node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": #watchpats.append(self.templates["Pat:NonDC_started"] % node) watchpats.append(self.templates["Pat:Local_started"] % node) else: for stopping in stopset: if self.CM.ShouldBeStatus[stopping] == "up": watchpats.append(self.templates["Pat:They_stopped"] % (node, self.CM.key_for_node(stopping))) if len(watchpats) == 0: return self.skipped() if len(startset) != 0: watchpats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() #begin actions for node in stopset: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) for node in startset: if self.CM.ShouldBeStatus[node] == "down": self.CM.StartaCMnoBlock(node) #get the result if watch.lookforall(): self.CM.cluster_stable() self.CM.fencing_cleanup("NearQuorumPoint", stonith) return self.success() self.logger.log("Warn: Patterns not found: " + repr(watch.unmatched)) #get the "bad" nodes upnodes = [] for node in stopset: if self.CM.StataCM(node) == 1: upnodes.append(node) downnodes = [] for node in startset: if self.CM.StataCM(node) == 0: downnodes.append(node) self.CM.fencing_cleanup("NearQuorumPoint", stonith) if upnodes == [] and downnodes == []: self.CM.cluster_stable() # Make sure they're completely down with no residule for node in stopset: self.rsh(node, self.templates["StopCmd"]) return self.success() if len(upnodes) > 0: self.logger.log("Warn: Unstoppable nodes: " + repr(upnodes)) if len(downnodes) > 0: self.logger.log("Warn: Unstartable nodes: " + repr(downnodes)) return self.failure() def is_applicable(self): return 1 AllTestClasses.append(NearQuorumPointTest) class RollingUpgradeTest(CTSTest): '''Perform a rolling upgrade of the cluster''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RollingUpgrade" self.start = StartTest(cm) self.stop = StopTest(cm) self.stopall = SimulStopLite(cm) self.startall = SimulStartLite(cm) def setup(self, node): # Start all remaining nodes ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.downgrade(node, None): return self.failure("Couldn't downgrade %s" % node) ret = self.startall(None) if not ret: return self.failure("Couldn't start all nodes") return self.success() def teardown(self, node): # Stop everything ret = self.stopall(None) if not ret: return self.failure("Couldn't stop all nodes") for node in self.Env["nodes"]: if not self.upgrade(node, None): return self.failure("Couldn't upgrade %s" % node) return self.success() def install(self, node, version, start=1, flags="--force"): target_dir = "/tmp/rpm-%s" % version src_dir = "%s/%s" % (self.Env["rpm-dir"], version) self.logger.log("Installing %s on %s with %s" % (version, node, flags)) if not self.stop(node): return self.failure("stop failure: "+node) rc = self.rsh(node, "mkdir -p %s" % target_dir) rc = self.rsh(node, "rm -f %s/*.rpm" % target_dir) (rc, lines) = self.rsh(node, "ls -1 %s/*.rpm" % src_dir, None) for line in lines: line = line[:-1] rc = self.rsh.cp("%s" % (line), "%s:%s/" % (node, target_dir)) rc = self.rsh(node, "rpm -Uvh %s %s/*.rpm" % (flags, target_dir)) if start and not self.start(node): return self.failure("start failure: "+node) return self.success() def upgrade(self, node, start=1): return self.install(node, self.Env["current-version"], start) def downgrade(self, node, start=1): return self.install(node, self.Env["previous-version"], start, "--force --nodeps") def __call__(self, node): '''Perform the 'Rolling Upgrade' test. ''' self.incr("calls") for node in self.Env["nodes"]: if self.upgrade(node): return self.failure("Couldn't upgrade %s" % node) self.CM.cluster_stable() return self.success() def is_applicable(self): if not self.is_applicable_common(): return None if not "rpm-dir" in list(self.Env.keys()): return None if not "current-version" in list(self.Env.keys()): return None if not "previous-version" in list(self.Env.keys()): return None return 1 # Register RestartTest as a good test to run AllTestClasses.append(RollingUpgradeTest) class BSC_AddResource(CTSTest): '''Add a resource to the cluster''' def __init__(self, cm): CTSTest.__init__(self, cm) self.name = "AddResource" self.resource_offset = 0 self.cib_cmd = """cibadmin -C -o %s -X '%s' """ def __call__(self, node): self.incr("calls") self.resource_offset = self.resource_offset + 1 r_id = "bsc-rsc-%s-%d" % (node, self.resource_offset) start_pat = "pacemaker-controld.*%s_start_0.*confirmed.*ok" patterns = [] patterns.append(start_pat % r_id) watch = self.create_watch(patterns, self.Env["DeadTime"]) watch.setwatch() ip = self.NextIP() if not self.make_ip_resource(node, r_id, "ocf", "IPaddr", ip): return self.failure("Make resource %s failed" % r_id) failed = 0 watch_result = watch.lookforall() if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Pattern not found: %s" % (regex)) failed = 1 if failed: return self.failure("Resource pattern(s) not found") if not self.CM.cluster_stable(self.Env["DeadTime"]): return self.failure("Unstable cluster") return self.success() def NextIP(self): ip = self.Env["IPBase"] if ":" in ip: fields = ip.rpartition(":") fields[2] = str(hex(int(fields[2], 16)+1)) print(str(hex(int(f[2], 16)+1))) else: fields = ip.rpartition('.') fields[2] = str(int(fields[2])+1) ip = fields[0] + fields[1] + fields[3]; self.Env["IPBase"] = ip return ip.strip() def make_ip_resource(self, node, id, rclass, type, ip): self.logger.log("Creating %s::%s:%s (%s) on %s" % (rclass,type,id,ip,node)) rsc_xml=""" """ % (id, rclass, type, id, id, ip) node_constraint = """ """ % (id, id, id, id, node) rc = 0 (rc, lines) = self.rsh(node, self.cib_cmd % ("constraints", node_constraint), None) if rc != 0: self.logger.log("Constraint creation failed: %d" % rc) return None (rc, lines) = self.rsh(node, self.cib_cmd % ("resources", rsc_xml), None) if rc != 0: self.logger.log("Resource creation failed: %d" % rc) return None return 1 def is_applicable(self): if self.Env["DoBSC"]: return 1 return None AllTestClasses.append(BSC_AddResource) class SimulStopLite(CTSTest): '''Stop any active nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStopLite" def __call__(self, dummy): '''Perform the 'SimulStopLite' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... watchpats = [ ] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.incr("WasStarted") watchpats.append(self.templates["Pat:We_stopped"] % node) if len(watchpats) == 0: return self.success() # Stop all the nodes - at about the same time... watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() self.set_timer() for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "up": self.CM.StopaCMnoBlock(node) if watch.lookforall(): # Make sure they're completely down with no residule for node in self.Env["nodes"]: self.rsh(node, self.templates["StopCmd"]) return self.success() did_fail = 0 up_nodes = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 1: did_fail = 1 up_nodes.append(node) if did_fail: return self.failure("Active nodes exist: " + repr(up_nodes)) self.logger.log("Warn: All nodes stopped but CTS didnt detect: " + repr(watch.unmatched)) return self.failure("Missing log message: "+repr(watch.unmatched)) def is_applicable(self): '''SimulStopLite is a setup test and never applicable''' return 0 class SimulStartLite(CTSTest): '''Start any stopped nodes ~ simultaneously''' def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "SimulStartLite" def __call__(self, dummy): '''Perform the 'SimulStartList' setup work. ''' self.incr("calls") self.debug("Setup: " + self.name) # We ignore the "node" parameter... node_list = [] for node in self.Env["nodes"]: if self.CM.ShouldBeStatus[node] == "down": self.incr("WasStopped") node_list.append(node) self.set_timer() while len(node_list) > 0: # Repeat until all nodes come up watchpats = [ ] uppat = self.templates["Pat:NonDC_started"] if self.CM.upcount() == 0: uppat = self.templates["Pat:Local_started"] watchpats.append(self.templates["Pat:DC_IDLE"]) for node in node_list: watchpats.append(uppat % node) watchpats.append(self.templates["Pat:InfraUp"] % node) watchpats.append(self.templates["Pat:PacemakerUp"] % node) # Start all the nodes - at about the same time... watch = self.create_watch(watchpats, self.Env["DeadTime"]+10) watch.setwatch() stonith = self.CM.prepare_fencing_watcher(self.name) for node in node_list: self.CM.StartaCMnoBlock(node) watch.lookforall() node_list = self.CM.fencing_cleanup(self.name, stonith) if node_list == None: return self.failure("Cluster did not stabilize") # Remove node_list messages from watch.unmatched for node in node_list: self.logger.debug("Dealing with stonith operations for %s" % repr(node_list)) if watch.unmatched: try: watch.unmatched.remove(uppat % node) except: self.debug("Already matched: %s" % (uppat % node)) try: watch.unmatched.remove(self.templates["Pat:InfraUp"] % node) except: self.debug("Already matched: %s" % (self.templates["Pat:InfraUp"] % node)) try: watch.unmatched.remove(self.templates["Pat:PacemakerUp"] % node) except: self.debug("Already matched: %s" % (self.templates["Pat:PacemakerUp"] % node)) if watch.unmatched: for regex in watch.unmatched: self.logger.log ("Warn: Startup pattern not found: %s" %(regex)) if not self.CM.cluster_stable(): return self.failure("Cluster did not stabilize") did_fail = 0 unstable = [] for node in self.Env["nodes"]: if self.CM.StataCM(node) == 0: did_fail = 1 unstable.append(node) if did_fail: return self.failure("Unstarted nodes exist: " + repr(unstable)) unstable = [] for node in self.Env["nodes"]: if not self.CM.node_stable(node): did_fail = 1 unstable.append(node) if did_fail: return self.failure("Unstable cluster nodes exist: " + repr(unstable)) return self.success() def is_applicable(self): '''SimulStartLite is a setup test and never applicable''' return 0 def TestList(cm, audits): result = [] for testclass in AllTestClasses: bound_test = testclass(cm) if bound_test.is_applicable(): bound_test.Audits = audits result.append(bound_test) return result class RemoteLXC(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = "RemoteLXC" self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.num_containers = 2 self.is_container = 1 self.is_docker_unsafe = 1 self.failed = 0 self.fail_string = "" def start_lxc_simple(self, node): # restore any artifacts laying around from a previous test. self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") # generate the containers, put them in the config, add some resources to them pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc1")) pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc2")) pats.append(self.templates["Pat:RscOpOK"] % ("start", "lxc-ms")) pats.append(self.templates["Pat:RscOpOK"] % ("promote", "lxc-ms")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -g -a -m -s -c %d &>/dev/null" % self.num_containers) self.set_timer("remoteSimpleInit") watch.lookforall() self.log_timer("remoteSimpleInit") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 def cleanup_lxc_simple(self, node): pats = [ ] # if the test failed, attempt to clean up the cib and libvirt environment # as best as possible if self.failed == 1: # restore libvirt and cib self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") return watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("stop", "container1")) pats.append(self.templates["Pat:RscOpOK"] % ("stop", "container2")) self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -p &>/dev/null") self.set_timer("remoteSimpleCleanup") watch.lookforall() self.log_timer("remoteSimpleCleanup") if watch.unmatched: self.fail_string = "Unmatched patterns: %s" % (repr(watch.unmatched)) self.failed = 1 # cleanup libvirt self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -s -R &>/dev/null") def __call__(self, node): '''Perform the 'RemoteLXC' test. ''' self.incr("calls") ret = self.startall(None) if not ret: return self.failure("Setup failed, start all nodes failed.") rc = self.rsh(node, "/usr/share/pacemaker/tests/cts/lxc_autogen.sh -v &>/dev/null") if rc == 1: self.log("Environment test for lxc support failed.") return self.skipped() self.start_lxc_simple(node) self.cleanup_lxc_simple(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed == 1: return self.failure(self.fail_string) return self.success() def errorstoignore(self): '''Return list of errors which should be ignored''' return [ r"Updating failcount for ping", r"schedulerd.*: Recover (ping|lxc-ms|container)\s*\(.*\)", # The orphaned lxc-ms resource causes an expected transition error # that is a result of the scheduler not having knowledge that the # promotable resource used to be a clone. As a result, it looks like that # resource is running in multiple locations when it shouldn't... But in # this instance we know why this error is occurring and that it is expected. r"Calculated [Tt]ransition .*pe-error", r"Resource lxc-ms .* is active on 2 nodes attempting recovery", r"Unknown operation: fail", r"VirtualDomain.*ERROR: Unable to determine emulator", ] AllTestClasses.append(RemoteLXC) class RemoteDriver(CTSTest): def __init__(self, cm): CTSTest.__init__(self,cm) self.name = self.__class__.__name__ self.is_docker_unsafe = 1 self.start = StartTest(cm) self.startall = SimulStartLite(cm) self.stop = StopTest(cm) self.remote_rsc = "remote-rsc" self.cib_cmd = """cibadmin -C -o %s -X '%s' """ self.reset() def reset(self): self.pcmk_started = 0 self.failed = False self.fail_string = "" self.remote_node_added = 0 self.remote_rsc_added = 0 self.remote_use_reconnect_interval = self.Env.RandomGen.choice([True,False]) def fail(self, msg): """ Mark test as failed. """ self.failed = True # Always log the failure. self.logger.log(msg) # Use first failure as test status, as it's likely to be most useful. if not self.fail_string: self.fail_string = msg def get_othernode(self, node): for othernode in self.Env["nodes"]: if othernode == node: # we don't want to try and use the cib that we just shutdown. # find a cluster node that is not our soon to be remote-node. continue else: return othernode def del_rsc(self, node, rsc): othernode = self.get_othernode(node) rc = self.rsh(othernode, "crm_resource -D -r %s -t primitive" % (rsc)) if rc != 0: self.fail("Removal of resource '%s' failed" % rsc) def add_rsc(self, node, rsc_xml): othernode = self.get_othernode(node) rc = self.rsh(othernode, self.cib_cmd % ("resources", rsc_xml)) if rc != 0: self.fail("resource creation failed") def add_primitive_rsc(self, node): rsc_xml = """ """ % (self.remote_rsc) self.add_rsc(node, rsc_xml) if not self.failed: self.remote_rsc_added = 1 def add_connection_rsc(self, node): if self.remote_use_reconnect_interval: # use reconnect interval and make sure to set cluster-recheck-interval as well. rsc_xml = """ """ % (self.remote_node, node) self.rsh(self.get_othernode(node), self.templates["SetCheckInterval"] % ("45s")) else: # not using reconnect interval rsc_xml = """ """ % (self.remote_node, node) self.add_rsc(node, rsc_xml) if not self.failed: self.remote_node_added = 1 def stop_pcmk_remote(self, node): # disable pcmk remote for i in range(10): rc = self.rsh(node, "service pacemaker_remote stop") if rc != 0: time.sleep(6) else: break def start_pcmk_remote(self, node): for i in range(10): rc = self.rsh(node, "service pacemaker_remote start") if rc != 0: time.sleep(6) else: self.pcmk_started = 1 break def kill_pcmk_remote(self, node): """ Simulate a Pacemaker Remote daemon failure. """ # We kill the process to prevent a graceful stop, # then stop it to prevent the OS from restarting it. self.rsh(node, "killall -9 pacemaker-remoted") self.stop_pcmk_remote(node) def start_metal(self, node): pcmk_started = 0 # make sure the resource doesn't already exist for some reason self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_rsc)) self.rsh(node, "crm_resource -D -r %s -t primitive" % (self.remote_node)) if not self.stop(node): self.fail("Failed to shutdown cluster node %s" % node) return self.start_pcmk_remote(node) if self.pcmk_started == 0: self.fail("Failed to start pacemaker_remote on node %s" % node) return # Convert node to baremetal now that it has shutdown the cluster stack pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) self.add_connection_rsc(node) self.set_timer("remoteMetalInit") watch.lookforall() self.log_timer("remoteMetalInit") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) def migrate_connection(self, node): if self.failed: return pats = [ ] pats.append(self.templates["Pat:RscOpOK"] % ("migrate_to", self.remote_node)) pats.append(self.templates["Pat:RscOpOK"] % ("migrate_from", self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(pats, 120) watch.setwatch() (rc, lines) = self.rsh(node, "crm_resource -M -r %s" % (self.remote_node), None) if rc != 0: self.fail("failed to move remote node connection resource") return self.set_timer("remoteMetalMigrate") watch.lookforall() self.log_timer("remoteMetalMigrate") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return def fail_rsc(self, node): if self.failed: return watchpats = [ ] watchpats.append(self.templates["Pat:RscRemoteOpOK"] % ("stop", self.remote_rsc, self.remote_node)) watchpats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) watchpats.append(self.templates["Pat:DC_IDLE"]) watch = self.create_watch(watchpats, 120) watch.setwatch() self.debug("causing dummy rsc to fail.") rc = self.rsh(node, "rm -f /var/run/resource-agents/Dummy*") self.set_timer("remoteRscFail") watch.lookforall() self.log_timer("remoteRscFail") if watch.unmatched: self.fail("Unmatched patterns during rsc fail: %s" % watch.unmatched) def fail_connection(self, node): if self.failed: return watchpats = [ ] watchpats.append(self.templates["Pat:FenceOpOK"] % self.remote_node) watchpats.append(self.templates["Pat:NodeFenced"] % self.remote_node) watch = self.create_watch(watchpats, 120) watch.setwatch() # force stop the pcmk remote daemon. this will result in fencing self.debug("Force stopped active remote node") self.kill_pcmk_remote(node) self.debug("Waiting for remote node to be fenced.") self.set_timer("remoteMetalFence") watch.lookforall() self.log_timer("remoteMetalFence") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return self.debug("Waiting for the remote node to come back up") self.CM.ns.WaitForNodeToComeUp(node, 120); pats = [ ] watch = self.create_watch(pats, 240) watch.setwatch() pats.append(self.templates["Pat:RscOpOK"] % ("start", self.remote_node)) if self.remote_rsc_added == 1: pats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) # start the remote node again watch it integrate back into cluster. self.start_pcmk_remote(node) if self.pcmk_started == 0: self.fail("Failed to start pacemaker_remote on node %s" % node) return self.debug("Waiting for remote node to rejoin cluster after being fenced.") self.set_timer("remoteMetalRestart") watch.lookforall() self.log_timer("remoteMetalRestart") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) return def add_dummy_rsc(self, node): if self.failed: return # verify we can put a resource on the remote node pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() pats.append(self.templates["Pat:RscRemoteOpOK"] % ("start", self.remote_rsc, self.remote_node)) pats.append(self.templates["Pat:DC_IDLE"]) # Add a resource that must live on remote-node self.add_primitive_rsc(node) # force that rsc to prefer the remote node. (rc, line) = self.CM.rsh(node, "crm_resource -M -r %s -N %s -f" % (self.remote_rsc, self.remote_node), None) if rc != 0: self.fail("Failed to place remote resource on remote node.") return self.set_timer("remoteMetalRsc") watch.lookforall() self.log_timer("remoteMetalRsc") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) def test_attributes(self, node): if self.failed: return # This verifies permanent attributes can be set on a remote-node. It also # verifies the remote-node can edit its own cib node section remotely. (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -v testval -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to set remote-node attribute. rc:%s output:%s" % (rc, line)) return (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -q -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to get remote-node attribute") return (rc, line) = self.CM.rsh(node, "crm_attribute -l forever -n testattr -D -N %s" % (self.remote_node), None) if rc != 0: self.fail("Failed to delete remote-node attribute") return def cleanup_metal(self, node): if self.pcmk_started == 0: return pats = [ ] watch = self.create_watch(pats, 120) watch.setwatch() if self.remote_rsc_added == 1: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.remote_rsc)) if self.remote_node_added == 1: pats.append(self.templates["Pat:RscOpOK"] % ("stop", self.remote_node)) self.set_timer("remoteMetalCleanup") if self.remote_use_reconnect_interval: self.debug("Cleaning up re-check interval") self.rsh(self.get_othernode(node), self.templates["ClearCheckInterval"]) if self.remote_rsc_added == 1: # Remove dummy resource added for remote node tests self.debug("Cleaning up dummy rsc put on remote node") self.rsh(node, "crm_resource -U -r %s" % self.remote_rsc) self.del_rsc(node, self.remote_rsc) if self.remote_node_added == 1: # Remove remote node's connection resource self.debug("Cleaning up remote node connection resource") self.rsh(node, "crm_resource -U -r %s" % (self.remote_node)) self.del_rsc(node, self.remote_node) watch.lookforall() self.log_timer("remoteMetalCleanup") if watch.unmatched: self.fail("Unmatched patterns: %s" % watch.unmatched) self.stop_pcmk_remote(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.remote_node_added == 1: # Remove remote node itself self.debug("Cleaning up node entry for remote node") self.rsh(self.get_othernode(node), "crm_node --force --remove %s" % self.remote_node) def setup_env(self, node): self.remote_node = "remote-%s" % (node) # we are assuming if all nodes have a key, that it is # the right key... If any node doesn't have a remote # key, we regenerate it everywhere. if self.rsh.exists_on_all("/etc/pacemaker/authkey", self.Env["nodes"]): return # create key locally (handle, keyfile) = tempfile.mkstemp(".cts") os.close(handle) devnull = open(os.devnull, 'wb') subprocess.check_call(["dd", "if=/dev/urandom", "of=%s" % keyfile, "bs=4096", "count=1"], stdout=devnull, stderr=devnull) devnull.close() # sync key throughout the cluster for node in self.Env["nodes"]: self.rsh(node, "mkdir -p --mode=0750 /etc/pacemaker") self.rsh.cp(keyfile, "root@%s:/etc/pacemaker/authkey" % node) self.rsh(node, "chgrp haclient /etc/pacemaker /etc/pacemaker/authkey") self.rsh(node, "chmod 0640 /etc/pacemaker/authkey") os.unlink(keyfile) def is_applicable(self): if not self.is_applicable_common(): return False for node in self.Env["nodes"]: rc = self.rsh(node, "which pacemaker-remoted >/dev/null 2>&1") if rc != 0: return False return True def start_new_test(self, node): self.incr("calls") self.reset() ret = self.startall(None) if not ret: return self.failure("setup failed: could not start all nodes") self.setup_env(node) self.start_metal(node) self.add_dummy_rsc(node) return True def __call__(self, node): return self.failure("This base class is not meant to be called directly.") def errorstoignore(self): '''Return list of errors which should be ignored''' return [ """is running on remote.*which isn't allowed""", """Connection terminated""", """Failed to send remote""", ] # RemoteDriver is just a base class for other tests, so it is not added to AllTestClasses class RemoteBasic(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteBaremetal' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.test_attributes(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() AllTestClasses.append(RemoteBasic) class RemoteStonithd(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteStonithd' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.fail_connection(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def is_applicable(self): if not RemoteDriver.is_applicable(self): return False if "DoFencing" in list(self.Env.keys()): return self.Env["DoFencing"] return True def errorstoignore(self): ignore_pats = [ r"Lost connection to Pacemaker Remote node", r"Software caused connection abort", r"pacemaker-controld.*:\s+error.*: Operation remote-.*_monitor", r"pacemaker-controld.*:\s+error.*: Result of monitor operation for remote-.*", r"schedulerd.*:\s+Recover remote-.*\s*\(.*\)", r"Calculated [Tt]ransition .*pe-error", r"error.*: Resource .*ocf::.* is active on 2 nodes attempting recovery", ] ignore_pats.extend(RemoteDriver.errorstoignore(self)) return ignore_pats AllTestClasses.append(RemoteStonithd) class RemoteMigrate(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteMigrate' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) self.migrate_connection(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() AllTestClasses.append(RemoteMigrate) class RemoteRscFailure(RemoteDriver): def __call__(self, node): '''Perform the 'RemoteRscFailure' test. ''' if not self.start_new_test(node): return self.failure(self.fail_string) # This is an important step. We are migrating the connection # before failing the resource. This verifies that the migration # has properly maintained control over the remote-node. self.migrate_connection(node) self.fail_rsc(node) self.cleanup_metal(node) self.debug("Waiting for the cluster to recover") self.CM.cluster_stable() if self.failed: return self.failure(self.fail_string) return self.success() def errorstoignore(self): ignore_pats = [ r"schedulerd.*: Recover remote-rsc\s*\(.*\)", r"Dummy.*: No process state file found", ] ignore_pats.extend(RemoteDriver.errorstoignore(self)) return ignore_pats AllTestClasses.append(RemoteRscFailure) # vim:ts=4:sw=4:et: diff --git a/daemons/controld/controld_transition.c b/daemons/controld/controld_transition.c index 3bd7c6fde2..5f164abd86 100644 --- a/daemons/controld/controld_transition.c +++ b/daemons/controld/controld_transition.c @@ -1,213 +1,215 @@ /* * Copyright 2004-2018 Andrew Beekhof * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include extern crm_graph_functions_t te_graph_fns; stonith_t *stonith_api = NULL; static void global_cib_callback(const xmlNode * msg, int callid, int rc, xmlNode * output) { } static crm_graph_t * create_blank_graph(void) { crm_graph_t *a_graph = unpack_graph(NULL, NULL); a_graph->complete = TRUE; a_graph->abort_reason = "DC Takeover"; a_graph->completion_action = tg_restart; return a_graph; } /* A_TE_START, A_TE_STOP, O_TE_RESTART */ void do_te_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { gboolean init_ok = TRUE; if (action & A_TE_STOP) { if (transition_graph) { destroy_graph(transition_graph); transition_graph = NULL; } if (fsa_cib_conn) { fsa_cib_conn->cmds->del_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, te_update_diff); } clear_bit(fsa_input_register, R_TE_CONNECTED); crm_info("Transitioner is now inactive"); } if ((action & A_TE_START) == 0) { return; } else if (is_set(fsa_input_register, R_TE_CONNECTED)) { crm_debug("The transitioner is already active"); return; } else if ((action & A_TE_START) && cur_state == S_STOPPING) { crm_info("Ignoring request to start the transitioner while shutting down"); return; } - te_uuid = crm_generate_uuid(); - crm_info("Registering TE UUID: %s", te_uuid); + if (te_uuid == NULL) { + te_uuid = crm_generate_uuid(); + crm_info("Registering TE UUID: %s", te_uuid); + } if (fsa_cib_conn == NULL) { crm_err("Could not set CIB callbacks"); init_ok = FALSE; } else { if (fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, te_update_diff) != pcmk_ok) { crm_err("Could not set CIB notification callback"); init_ok = FALSE; } if (fsa_cib_conn->cmds->set_op_callback(fsa_cib_conn, global_cib_callback) != pcmk_ok) { crm_err("Could not set CIB global callback"); init_ok = FALSE; } } if (init_ok) { set_graph_functions(&te_graph_fns); if (transition_graph) { destroy_graph(transition_graph); } /* create a blank one */ crm_debug("Transitioner is now active"); transition_graph = create_blank_graph(); set_bit(fsa_input_register, R_TE_CONNECTED); } } /* A_TE_INVOKE, A_TE_CANCEL */ void do_te_invoke(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { if (AM_I_DC == FALSE || (fsa_state != S_TRANSITION_ENGINE && (action & A_TE_INVOKE))) { crm_notice("No need to invoke the TE (%s) in state %s", fsa_action2string(action), fsa_state2string(fsa_state)); return; } if (action & A_TE_CANCEL) { crm_debug("Cancelling the transition: %s", transition_graph->complete ? "inactive" : "active"); abort_transition(INFINITY, tg_restart, "Peer Cancelled", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_HALT) { crm_debug("Halting the transition: %s", transition_graph->complete ? "inactive" : "active"); abort_transition(INFINITY, tg_stop, "Peer Halt", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_INVOKE) { const char *value = NULL; xmlNode *graph_data = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *ref = crm_element_value(input->msg, XML_ATTR_REFERENCE); const char *graph_file = crm_element_value(input->msg, F_CRM_TGRAPH); const char *graph_input = crm_element_value(input->msg, F_CRM_TGRAPH_INPUT); if (graph_file == NULL && input->xml == NULL) { crm_log_xml_err(input->msg, "Bad command"); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); return; } if (transition_graph->complete == FALSE) { crm_info("Another transition is already active"); abort_transition(INFINITY, tg_restart, "Transition Active", NULL); return; } if (fsa_pe_ref == NULL || safe_str_neq(fsa_pe_ref, ref)) { crm_info("Transition is redundant: %s vs. %s", crm_str(fsa_pe_ref), crm_str(ref)); abort_transition(INFINITY, tg_restart, "Transition Redundant", NULL); } graph_data = input->xml; if (graph_data == NULL && graph_file != NULL) { graph_data = filename2xml(graph_file); } if (is_timer_started(transition_timer)) { crm_debug("The transitioner wait for a transition timer"); return; } CRM_CHECK(graph_data != NULL, crm_err("Input raised by %s is invalid", msg_data->origin); crm_log_xml_err(input->msg, "Bad command"); return); destroy_graph(transition_graph); transition_graph = unpack_graph(graph_data, graph_input); if (transition_graph == NULL) { CRM_CHECK(transition_graph != NULL,); transition_graph = create_blank_graph(); return; } crm_info("Processing graph %d (ref=%s) derived from %s", transition_graph->id, ref, graph_input); te_reset_job_counts(); value = crm_element_value(graph_data, "failed-stop-offset"); if (value) { free(failed_stop_offset); failed_stop_offset = strdup(value); } value = crm_element_value(graph_data, "failed-start-offset"); if (value) { free(failed_start_offset); failed_start_offset = strdup(value); } trigger_graph(); print_graph(LOG_TRACE, transition_graph); if (graph_data != input->xml) { free_xml(graph_data); } } } diff --git a/daemons/execd/execd_commands.c b/daemons/execd/execd_commands.c index 10f000f8c9..04dc41f875 100644 --- a/daemons/execd/execd_commands.c +++ b/daemons/execd/execd_commands.c @@ -1,1675 +1,1686 @@ /* * Copyright 2012-2018 David Vossel * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include "pacemaker-execd.h" #ifdef HAVE_SYS_TIMEB_H # include #endif #define EXIT_REASON_MAX_LEN 128 GHashTable *rsc_list = NULL; typedef struct lrmd_cmd_s { int timeout; guint interval_ms; int start_delay; int timeout_orig; int call_id; int exec_rc; int lrmd_op_status; int call_opts; /* Timer ids, must be removed on cmd destruction. */ int delay_id; int stonith_recurring_id; int rsc_deleted; int service_flags; char *client_id; char *origin; char *rsc_id; char *action; char *real_action; char *exit_reason; char *output; char *userdata_str; #ifdef HAVE_SYS_TIMEB_H /* Recurring and systemd operations may involve more than one executor * command per operation, so they need info about the original and the most * recent. */ struct timeb t_first_run; /* Timestamp of when op first ran */ struct timeb t_run; /* Timestamp of when op most recently ran */ struct timeb t_first_queue; /* Timestamp of when op first was queued */ struct timeb t_queue; /* Timestamp of when op most recently was queued */ struct timeb t_rcchange; /* Timestamp of last rc change */ #endif int first_notify_sent; int last_notify_rc; int last_notify_op_status; int last_pid; GHashTable *params; } lrmd_cmd_t; static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc); static gboolean lrmd_rsc_dispatch(gpointer user_data); static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id); static void log_finished(lrmd_cmd_t * cmd, int exec_time, int queue_time) { char pid_str[32] = { 0, }; int log_level = LOG_INFO; if (cmd->last_pid) { snprintf(pid_str, 32, "%d", cmd->last_pid); } if (safe_str_eq(cmd->action, "monitor")) { log_level = LOG_DEBUG; } #ifdef HAVE_SYS_TIMEB_H do_crm_log(log_level, "finished - rsc:%s action:%s call_id:%d %s%s exit-code:%d exec-time:%dms queue-time:%dms", cmd->rsc_id, cmd->action, cmd->call_id, cmd->last_pid ? "pid:" : "", pid_str, cmd->exec_rc, exec_time, queue_time); #else do_crm_log(log_level, "finished - rsc:%s action:%s call_id:%d %s%s exit-code:%d", cmd->rsc_id, cmd->action, cmd->call_id, cmd->last_pid ? "pid:" : "", pid_str, cmd->exec_rc); #endif } static void log_execute(lrmd_cmd_t * cmd) { int log_level = LOG_INFO; if (safe_str_eq(cmd->action, "monitor")) { log_level = LOG_DEBUG; } do_crm_log(log_level, "executing - rsc:%s action:%s call_id:%d", cmd->rsc_id, cmd->action, cmd->call_id); } static const char * normalize_action_name(lrmd_rsc_t * rsc, const char *action) { if (safe_str_eq(action, "monitor") && (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_LSB) || safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_SERVICE) || safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_SYSTEMD))) { return "status"; } return action; } static lrmd_rsc_t * build_rsc_from_xml(xmlNode * msg) { xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR); lrmd_rsc_t *rsc = NULL; rsc = calloc(1, sizeof(lrmd_rsc_t)); crm_element_value_int(msg, F_LRMD_CALLOPTS, &rsc->call_opts); rsc->rsc_id = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ID); rsc->class = crm_element_value_copy(rsc_xml, F_LRMD_CLASS); rsc->provider = crm_element_value_copy(rsc_xml, F_LRMD_PROVIDER); rsc->type = crm_element_value_copy(rsc_xml, F_LRMD_TYPE); rsc->work = mainloop_add_trigger(G_PRIORITY_HIGH, lrmd_rsc_dispatch, rsc); return rsc; } static lrmd_cmd_t * create_lrmd_cmd(xmlNode * msg, crm_client_t * client) { int call_options = 0; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR); lrmd_cmd_t *cmd = NULL; cmd = calloc(1, sizeof(lrmd_cmd_t)); crm_element_value_int(msg, F_LRMD_CALLOPTS, &call_options); cmd->call_opts = call_options; cmd->client_id = strdup(client->id); crm_element_value_int(msg, F_LRMD_CALLID, &cmd->call_id); crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &cmd->interval_ms); crm_element_value_int(rsc_xml, F_LRMD_TIMEOUT, &cmd->timeout); crm_element_value_int(rsc_xml, F_LRMD_RSC_START_DELAY, &cmd->start_delay); cmd->timeout_orig = cmd->timeout; cmd->origin = crm_element_value_copy(rsc_xml, F_LRMD_ORIGIN); cmd->action = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ACTION); cmd->userdata_str = crm_element_value_copy(rsc_xml, F_LRMD_RSC_USERDATA_STR); cmd->rsc_id = crm_element_value_copy(rsc_xml, F_LRMD_RSC_ID); cmd->params = xml2list(rsc_xml); if (safe_str_eq(g_hash_table_lookup(cmd->params, "CRM_meta_on_fail"), "block")) { crm_debug("Setting flag to leave pid group on timeout and only kill action pid for " CRM_OP_FMT, cmd->rsc_id, cmd->action, cmd->interval_ms); cmd->service_flags |= SVC_ACTION_LEAVE_GROUP; } return cmd; } static void free_lrmd_cmd(lrmd_cmd_t * cmd) { if (cmd->stonith_recurring_id) { g_source_remove(cmd->stonith_recurring_id); } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->params) { g_hash_table_destroy(cmd->params); } free(cmd->origin); free(cmd->action); free(cmd->real_action); free(cmd->userdata_str); free(cmd->rsc_id); free(cmd->output); free(cmd->exit_reason); free(cmd->client_id); free(cmd); } static gboolean stonith_recurring_op_helper(gpointer data) { lrmd_cmd_t *cmd = data; lrmd_rsc_t *rsc; cmd->stonith_recurring_id = 0; if (!cmd->rsc_id) { return FALSE; } rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); CRM_ASSERT(rsc != NULL); /* take it out of recurring_ops list, and put it in the pending ops * to be executed */ rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef HAVE_SYS_TIMEB_H ftime(&cmd->t_queue); if (cmd->t_first_queue.time == 0) { cmd->t_first_queue = cmd->t_queue; } #endif mainloop_set_trigger(rsc->work); return FALSE; } static gboolean start_delay_helper(gpointer data) { lrmd_cmd_t *cmd = data; lrmd_rsc_t *rsc = NULL; cmd->delay_id = 0; rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL; if (rsc) { mainloop_set_trigger(rsc->work); } return FALSE; } static gboolean merge_recurring_duplicate(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { GListPtr gIter = NULL; lrmd_cmd_t * dup = NULL; gboolean dup_pending = FALSE; if (cmd->interval_ms == 0) { return 0; } for (gIter = rsc->pending_ops; gIter != NULL; gIter = gIter->next) { dup = gIter->data; if (safe_str_eq(cmd->action, dup->action) && (cmd->interval_ms == dup->interval_ms)) { dup_pending = TRUE; goto merge_dup; } } /* if dup is in recurring_ops list, that means it has already executed * and is in the interval loop. we can't just remove it in this case. */ for (gIter = rsc->recurring_ops; gIter != NULL; gIter = gIter->next) { dup = gIter->data; if (safe_str_eq(cmd->action, dup->action) && (cmd->interval_ms == dup->interval_ms)) { goto merge_dup; } } return FALSE; merge_dup: /* This should not occur. If it does, we need to investigate how something * like this is possible in the controller. */ crm_warn("Duplicate recurring op entry detected (" CRM_OP_FMT "), merging with previous op entry", rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); /* merge */ dup->first_notify_sent = 0; free(dup->userdata_str); dup->userdata_str = cmd->userdata_str; cmd->userdata_str = NULL; dup->call_id = cmd->call_id; if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { /* if we are waiting for the next interval, kick it off now */ if (dup_pending == TRUE) { g_source_remove(cmd->stonith_recurring_id); cmd->stonith_recurring_id = 0; stonith_recurring_op_helper(cmd); } } else if (dup_pending == FALSE) { /* if we've already handed this to the service lib, kick off an early execution */ services_action_kick(rsc->rsc_id, normalize_action_name(rsc, dup->action), dup->interval_ms); } free_lrmd_cmd(cmd); return TRUE; } static void schedule_lrmd_cmd(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { gboolean dup_processed = FALSE; CRM_CHECK(cmd != NULL, return); CRM_CHECK(rsc != NULL, return); crm_trace("Scheduling %s on %s", cmd->action, rsc->rsc_id); dup_processed = merge_recurring_duplicate(rsc, cmd); if (dup_processed) { /* duplicate recurring cmd found, cmds merged */ return; } /* The controller expects the executor to automatically cancel * recurring operations before a resource stops. */ if (safe_str_eq(cmd->action, "stop")) { cancel_all_recurring(rsc, NULL); } rsc->pending_ops = g_list_append(rsc->pending_ops, cmd); #ifdef HAVE_SYS_TIMEB_H ftime(&cmd->t_queue); if (cmd->t_first_queue.time == 0) { cmd->t_first_queue = cmd->t_queue; } #endif mainloop_set_trigger(rsc->work); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } } static xmlNode * create_lrmd_reply(const char *origin, int rc, int call_id) { xmlNode *reply = create_xml_node(NULL, T_LRMD_REPLY); crm_xml_add(reply, F_LRMD_ORIGIN, origin); crm_xml_add_int(reply, F_LRMD_RC, rc); crm_xml_add_int(reply, F_LRMD_CALLID, call_id); return reply; } static void send_client_notify(gpointer key, gpointer value, gpointer user_data) { xmlNode *update_msg = user_data; crm_client_t *client = value; int rc; if (client == NULL) { crm_err("Asked to send event to NULL client"); return; } else if (client->name == NULL) { crm_trace("Asked to send event to client with no name"); return; } rc = lrmd_server_send_notify(client, update_msg); if ((rc <= 0) && (rc != -ENOTCONN)) { crm_warn("Could not notify client %s/%s: %s " CRM_XS " rc=%d", client->name, client->id, (rc? pcmk_strerror(rc) : "no data sent"), rc); } } #ifdef HAVE_SYS_TIMEB_H /*! * \internal * \brief Return difference between two times in milliseconds * * \param[in] now More recent time (or NULL to use current time) * \param[in] old Earlier time * * \return milliseconds difference (or 0 if old is NULL or has time zero) */ static int time_diff_ms(struct timeb *now, struct timeb *old) { struct timeb local_now = { 0, }; if (now == NULL) { ftime(&local_now); now = &local_now; } if ((old == NULL) || (old->time == 0)) { return 0; } return difftime(now->time, old->time) * 1000 + now->millitm - old->millitm; } /*! * \internal * \brief Reset a command's operation times to their original values. * * Reset a command's run and queued timestamps to the timestamps of the original * command, so we report the entire time since then and not just the time since * the most recent command (for recurring and systemd operations). * * /param[in] cmd Executor command object to reset * * /note It's not obvious what the queued time should be for a systemd * start/stop operation, which might go like this: * initial command queued 5ms, runs 3s * monitor command queued 10ms, runs 10s * monitor command queued 10ms, runs 10s * Is the queued time for that operation 5ms, 10ms or 25ms? The current * implementation will report 5ms. If it's 25ms, then we need to * subtract 20ms from the total exec time so as not to count it twice. * We can implement that later if it matters to anyone ... */ static void cmd_original_times(lrmd_cmd_t * cmd) { cmd->t_run = cmd->t_first_run; cmd->t_queue = cmd->t_first_queue; } #endif static void send_cmd_complete_notify(lrmd_cmd_t * cmd) { int exec_time = 0; int queue_time = 0; xmlNode *notify = NULL; #ifdef HAVE_SYS_TIMEB_H exec_time = time_diff_ms(NULL, &cmd->t_run); queue_time = time_diff_ms(&cmd->t_run, &cmd->t_queue); #endif log_finished(cmd, exec_time, queue_time); /* if the first notify result for a cmd has already been sent earlier, and the * the option to only send notifies on result changes is set. Check to see * if the last result is the same as the new one. If so, suppress this update */ if (cmd->first_notify_sent && (cmd->call_opts & lrmd_opt_notify_changes_only)) { if (cmd->last_notify_rc == cmd->exec_rc && cmd->last_notify_op_status == cmd->lrmd_op_status) { /* only send changes */ return; } } cmd->first_notify_sent = 1; cmd->last_notify_rc = cmd->exec_rc; cmd->last_notify_op_status = cmd->lrmd_op_status; notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __FUNCTION__); crm_xml_add_int(notify, F_LRMD_TIMEOUT, cmd->timeout); crm_xml_add_ms(notify, F_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(notify, F_LRMD_RSC_START_DELAY, cmd->start_delay); crm_xml_add_int(notify, F_LRMD_EXEC_RC, cmd->exec_rc); crm_xml_add_int(notify, F_LRMD_OP_STATUS, cmd->lrmd_op_status); crm_xml_add_int(notify, F_LRMD_CALLID, cmd->call_id); crm_xml_add_int(notify, F_LRMD_RSC_DELETED, cmd->rsc_deleted); #ifdef HAVE_SYS_TIMEB_H crm_xml_add_int(notify, F_LRMD_RSC_RUN_TIME, cmd->t_run.time); crm_xml_add_int(notify, F_LRMD_RSC_RCCHANGE_TIME, cmd->t_rcchange.time); crm_xml_add_int(notify, F_LRMD_RSC_EXEC_TIME, exec_time); crm_xml_add_int(notify, F_LRMD_RSC_QUEUE_TIME, queue_time); #endif crm_xml_add(notify, F_LRMD_OPERATION, LRMD_OP_RSC_EXEC); crm_xml_add(notify, F_LRMD_RSC_ID, cmd->rsc_id); if(cmd->real_action) { crm_xml_add(notify, F_LRMD_RSC_ACTION, cmd->real_action); } else { crm_xml_add(notify, F_LRMD_RSC_ACTION, cmd->action); } crm_xml_add(notify, F_LRMD_RSC_USERDATA_STR, cmd->userdata_str); crm_xml_add(notify, F_LRMD_RSC_OUTPUT, cmd->output); crm_xml_add(notify, F_LRMD_RSC_EXIT_REASON, cmd->exit_reason); if (cmd->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; xmlNode *args = create_xml_node(notify, XML_TAG_ATTRS); g_hash_table_iter_init(&iter, cmd->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { hash2smartfield((gpointer) key, (gpointer) value, args); } } if (cmd->client_id && (cmd->call_opts & lrmd_opt_notify_orig_only)) { crm_client_t *client = crm_client_get_by_id(cmd->client_id); if (client) { send_client_notify(client->id, client, notify); } } else if (client_connections != NULL) { g_hash_table_foreach(client_connections, send_client_notify, notify); } free_xml(notify); } static void send_generic_notify(int rc, xmlNode * request) { if (client_connections != NULL) { int call_id = 0; xmlNode *notify = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); const char *op = crm_element_value(request, F_LRMD_OPERATION); crm_element_value_int(request, F_LRMD_CALLID, &call_id); notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __FUNCTION__); crm_xml_add_int(notify, F_LRMD_RC, rc); crm_xml_add_int(notify, F_LRMD_CALLID, call_id); crm_xml_add(notify, F_LRMD_OPERATION, op); crm_xml_add(notify, F_LRMD_RSC_ID, rsc_id); g_hash_table_foreach(client_connections, send_client_notify, notify); free_xml(notify); } } static void cmd_reset(lrmd_cmd_t * cmd) { cmd->lrmd_op_status = 0; cmd->last_pid = 0; #ifdef HAVE_SYS_TIMEB_H memset(&cmd->t_run, 0, sizeof(cmd->t_run)); memset(&cmd->t_queue, 0, sizeof(cmd->t_queue)); #endif free(cmd->exit_reason); cmd->exit_reason = NULL; free(cmd->output); cmd->output = NULL; } static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc) { crm_trace("Resource operation rsc:%s action:%s completed (%p %p)", cmd->rsc_id, cmd->action, rsc ? rsc->active : NULL, cmd); if (rsc && (rsc->active == cmd)) { rsc->active = NULL; mainloop_set_trigger(rsc->work); } if (!rsc) { cmd->rsc_deleted = 1; } /* reset original timeout so client notification has correct information */ cmd->timeout = cmd->timeout_orig; send_cmd_complete_notify(cmd); if (cmd->interval_ms && (cmd->lrmd_op_status == PCMK_LRM_OP_CANCELLED)) { if (rsc) { rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd); rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else if (cmd->interval_ms == 0) { if (rsc) { rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd); } free_lrmd_cmd(cmd); } else { /* Clear all the values pertaining just to the last iteration of a recurring op. */ cmd_reset(cmd); } } static int ocf2uniform_rc(int rc) { if (rc < 0 || rc > PCMK_OCF_FAILED_MASTER) { return PCMK_OCF_UNKNOWN_ERROR; } return rc; } static int stonith2uniform_rc(const char *action, int rc) { if (rc == -ENODEV) { if (safe_str_eq(action, "stop")) { rc = PCMK_OCF_OK; } else if (safe_str_eq(action, "start")) { rc = PCMK_OCF_NOT_INSTALLED; } else { rc = PCMK_OCF_NOT_RUNNING; } } else if (rc != 0) { rc = PCMK_OCF_UNKNOWN_ERROR; } return rc; } #if SUPPORT_NAGIOS static int nagios2uniform_rc(const char *action, int rc) { if (rc < 0) { return PCMK_OCF_UNKNOWN_ERROR; } switch (rc) { case NAGIOS_STATE_OK: return PCMK_OCF_OK; case NAGIOS_INSUFFICIENT_PRIV: return PCMK_OCF_INSUFFICIENT_PRIV; case NAGIOS_NOT_INSTALLED: return PCMK_OCF_NOT_INSTALLED; case NAGIOS_STATE_WARNING: case NAGIOS_STATE_CRITICAL: case NAGIOS_STATE_UNKNOWN: case NAGIOS_STATE_DEPENDENT: default: return PCMK_OCF_UNKNOWN_ERROR; } return PCMK_OCF_UNKNOWN_ERROR; } #endif static int get_uniform_rc(const char *standard, const char *action, int rc) { if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_OCF)) { return ocf2uniform_rc(rc); } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_STONITH)) { return stonith2uniform_rc(action, rc); } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD)) { return rc; } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_UPSTART)) { return rc; #if SUPPORT_NAGIOS } else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_NAGIOS)) { return nagios2uniform_rc(action, rc); #endif } else { return services_get_ocf_exitcode(action, rc); } } static int action_get_uniform_rc(svc_action_t * action) { lrmd_cmd_t *cmd = action->cb_data; return get_uniform_rc(action->standard, cmd->action, action->rc); } void notify_of_new_client(crm_client_t *new_client) { crm_client_t *client = NULL; GHashTableIter iter; xmlNode *notify = NULL; char *key = NULL; notify = create_xml_node(NULL, T_LRMD_NOTIFY); crm_xml_add(notify, F_LRMD_ORIGIN, __FUNCTION__); crm_xml_add(notify, F_LRMD_OPERATION, LRMD_OP_NEW_CLIENT); g_hash_table_iter_init(&iter, client_connections); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & client)) { if (safe_str_eq(client->id, new_client->id)) { continue; } send_client_notify((gpointer) key, (gpointer) client, (gpointer) notify); } free_xml(notify); } static char * parse_exit_reason(const char *output) { const char *cur = NULL; const char *last = NULL; static int cookie_len = 0; char *eol = NULL; size_t reason_len = EXIT_REASON_MAX_LEN; if (output == NULL) { return NULL; } if (!cookie_len) { cookie_len = strlen(PCMK_OCF_REASON_PREFIX); } cur = strstr(output, PCMK_OCF_REASON_PREFIX); for (; cur != NULL; cur = strstr(cur, PCMK_OCF_REASON_PREFIX)) { /* skip over the cookie delimiter string */ cur += cookie_len; last = cur; } if (last == NULL) { return NULL; } // Truncate everything after a new line, and limit reason string size eol = strchr(last, '\n'); if (eol) { reason_len = QB_MIN(reason_len, eol - last); } return strndup(last, reason_len); } void client_disconnect_cleanup(const char *client_id) { GHashTableIter iter; lrmd_rsc_t *rsc = NULL; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) { if (rsc->call_opts & lrmd_opt_drop_recurring) { /* This client is disconnecting, drop any recurring operations * it may have initiated on the resource */ cancel_all_recurring(rsc, client_id); } } } static void action_complete(svc_action_t * action) { lrmd_rsc_t *rsc; lrmd_cmd_t *cmd = action->cb_data; const char *rclass = NULL; bool goagain = false; if (!cmd) { crm_err("Completed executor action (%s) does not match any known operations", action->id); return; } #ifdef HAVE_SYS_TIMEB_H if (cmd->exec_rc != action->rc) { ftime(&cmd->t_rcchange); } #endif cmd->last_pid = action->pid; cmd->exec_rc = action_get_uniform_rc(action); cmd->lrmd_op_status = action->status; rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL; if (rsc && safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_SERVICE)) { rclass = resources_find_service_class(rsc->type); } else if(rsc) { rclass = rsc->class; } if (safe_str_eq(rclass, PCMK_RESOURCE_CLASS_SYSTEMD)) { if(cmd->exec_rc == PCMK_OCF_OK && safe_str_eq(cmd->action, "start")) { /* systemd I curse thee! * * systemd returns from start actions after the start _begins_ * not after it completes. * * So we have to jump through a few hoops so that we don't * report 'complete' to the rest of pacemaker until, you know, * it's actually done. */ goagain = true; cmd->real_action = cmd->action; cmd->action = strdup("monitor"); } else if(cmd->exec_rc == PCMK_OCF_OK && safe_str_eq(cmd->action, "stop")) { goagain = true; cmd->real_action = cmd->action; cmd->action = strdup("monitor"); } else if(cmd->real_action) { /* Ok, so this is the follow up monitor action to check if start actually completed */ if(cmd->lrmd_op_status == PCMK_LRM_OP_DONE && cmd->exec_rc == PCMK_OCF_PENDING) { goagain = true; } else if(cmd->exec_rc == PCMK_OCF_OK && safe_str_eq(cmd->real_action, "stop")) { goagain = true; } else { #ifdef HAVE_SYS_TIMEB_H int time_sum = time_diff_ms(NULL, &cmd->t_first_run); int timeout_left = cmd->timeout_orig - time_sum; crm_debug("%s %s is now complete (elapsed=%dms, remaining=%dms): %s (%d)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, services_ocf_exitcode_str(cmd->exec_rc), cmd->exec_rc); cmd_original_times(cmd); #endif if(cmd->lrmd_op_status == PCMK_LRM_OP_DONE && cmd->exec_rc == PCMK_OCF_NOT_RUNNING && safe_str_eq(cmd->real_action, "stop")) { cmd->exec_rc = PCMK_OCF_OK; } } } } #if SUPPORT_NAGIOS if (rsc && safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS)) { if (safe_str_eq(cmd->action, "monitor") && (cmd->interval_ms == 0) && cmd->exec_rc == PCMK_OCF_OK) { /* Successfully executed --version for the nagios plugin */ cmd->exec_rc = PCMK_OCF_NOT_RUNNING; } else if (safe_str_eq(cmd->action, "start") && cmd->exec_rc != PCMK_OCF_OK) { goagain = true; } } #endif /* Wrapping this section in ifdef implies that systemd resources are not * fully supported on platforms without sys/timeb.h. Since timeb is * obsolete, we should eventually prefer a clock_gettime() implementation * (wrapped in its own ifdef) with timeb as a fallback. */ if(goagain) { #ifdef HAVE_SYS_TIMEB_H int time_sum = time_diff_ms(NULL, &cmd->t_first_run); int timeout_left = cmd->timeout_orig - time_sum; int delay = cmd->timeout_orig / 10; if(delay >= timeout_left && timeout_left > 20) { delay = timeout_left/2; } delay = QB_MIN(2000, delay); if (delay < timeout_left) { cmd->start_delay = delay; cmd->timeout = timeout_left; if(cmd->exec_rc == PCMK_OCF_OK) { crm_debug("%s %s may still be in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->real_action, time_sum, timeout_left, delay); } else if(cmd->exec_rc == PCMK_OCF_PENDING) { crm_info("%s %s is still in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->action, time_sum, timeout_left, delay); } else { crm_notice("%s %s failed '%s' (%d): re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)", cmd->rsc_id, cmd->action, services_ocf_exitcode_str(cmd->exec_rc), cmd->exec_rc, time_sum, timeout_left, delay); } cmd_reset(cmd); if(rsc) { rsc->active = NULL; } schedule_lrmd_cmd(rsc, cmd); /* Don't finalize cmd, we're not done with it yet */ return; } else { crm_notice("Giving up on %s %s (rc=%d): timeout (elapsed=%dms, remaining=%dms)", cmd->rsc_id, cmd->real_action?cmd->real_action:cmd->action, cmd->exec_rc, time_sum, timeout_left); cmd->lrmd_op_status = PCMK_LRM_OP_TIMEOUT; cmd->exec_rc = PCMK_OCF_TIMEOUT; cmd_original_times(cmd); } #endif } if (action->stderr_data) { cmd->output = strdup(action->stderr_data); cmd->exit_reason = parse_exit_reason(action->stderr_data); } else if (action->stdout_data) { cmd->output = strdup(action->stdout_data); } cmd_finalize(cmd, rsc); } static void stonith_action_complete(lrmd_cmd_t * cmd, int rc) { bool recurring = (cmd->interval_ms > 0); lrmd_rsc_t *rsc = NULL; cmd->exec_rc = get_uniform_rc(PCMK_RESOURCE_CLASS_STONITH, cmd->action, rc); rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id); if (cmd->lrmd_op_status == PCMK_LRM_OP_CANCELLED) { recurring = FALSE; /* do nothing */ } else if (rc == -ENODEV && safe_str_eq(cmd->action, "monitor")) { - /* Not registered == inactive */ - cmd->lrmd_op_status = PCMK_LRM_OP_DONE; + // The device is not registered with the fencer + + if (recurring) { + /* If we get here, the fencer somehow lost the registration of a + * previously active device (possibly due to crash and respawn). In + * that case, we need to indicate that the recurring monitor needs + * to be cancelled. + */ + cmd->lrmd_op_status = PCMK_LRM_OP_CANCELLED; + recurring = FALSE; + } else { + cmd->lrmd_op_status = PCMK_LRM_OP_DONE; + } cmd->exec_rc = PCMK_OCF_NOT_RUNNING; } else if (rc) { /* Attempt to map return codes to op status if possible */ switch (rc) { case -EPROTONOSUPPORT: cmd->lrmd_op_status = PCMK_LRM_OP_NOTSUPPORTED; break; case -ETIME: cmd->lrmd_op_status = PCMK_LRM_OP_TIMEOUT; break; default: /* TODO: This looks wrong. Status should be _DONE and exec_rc set to an error */ cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; } } else { /* command successful */ cmd->lrmd_op_status = PCMK_LRM_OP_DONE; if (safe_str_eq(cmd->action, "start") && rsc) { rsc->stonith_started = 1; } } if (recurring && rsc) { if (cmd->stonith_recurring_id) { g_source_remove(cmd->stonith_recurring_id); } cmd->stonith_recurring_id = g_timeout_add(cmd->interval_ms, stonith_recurring_op_helper, cmd); } cmd_finalize(cmd, rsc); } static void lrmd_stonith_callback(stonith_t * stonith, stonith_callback_data_t * data) { stonith_action_complete(data->userdata, data->rc); } void stonith_connection_failed(void) { GHashTableIter iter; GList *cmd_list = NULL; GList *cmd_iter = NULL; lrmd_rsc_t *rsc = NULL; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) { if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { if (rsc->active) { cmd_list = g_list_append(cmd_list, rsc->active); } if (rsc->recurring_ops) { cmd_list = g_list_concat(cmd_list, rsc->recurring_ops); } if (rsc->pending_ops) { cmd_list = g_list_concat(cmd_list, rsc->pending_ops); } rsc->pending_ops = rsc->recurring_ops = NULL; } } if (!cmd_list) { return; } crm_err("Connection to fencer failed, finalizing %d pending operations", g_list_length(cmd_list)); for (cmd_iter = cmd_list; cmd_iter; cmd_iter = cmd_iter->next) { stonith_action_complete(cmd_iter->data, -ENOTCONN); } g_list_free(cmd_list); } static int lrmd_rsc_execute_stonith(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { int rc = 0; int do_monitor = 0; stonith_t *stonith_api = get_stonith_connection(); if (!stonith_api) { cmd->exec_rc = get_uniform_rc(PCMK_RESOURCE_CLASS_STONITH, cmd->action, -ENOTCONN); cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; cmd_finalize(cmd, rsc); return -EUNATCH; } if (safe_str_eq(cmd->action, "start")) { char *key = NULL; char *value = NULL; stonith_key_value_t *device_params = NULL; if (cmd->params) { GHashTableIter iter; g_hash_table_iter_init(&iter, cmd->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { device_params = stonith_key_value_add(device_params, key, value); } } /* Stonith automatically registers devices from the IPC when changes * occur, but to avoid a possible race condition between stonith * receiving the IPC update and the executor requesting that resource, * the executor still registers the device as well. Stonith knows how to * handle duplicate device registrations correctly. */ rc = stonith_api->cmds->register_device(stonith_api, st_opt_sync_call, cmd->rsc_id, rsc->provider, rsc->type, device_params); stonith_key_value_freeall(device_params, 1, 1); if (rc == 0) { do_monitor = 1; } } else if (safe_str_eq(cmd->action, "stop")) { rc = stonith_api->cmds->remove_device(stonith_api, st_opt_sync_call, cmd->rsc_id); rsc->stonith_started = 0; } else if (safe_str_eq(cmd->action, "monitor")) { if (cmd->interval_ms > 0) { do_monitor = 1; } else { rc = rsc->stonith_started ? 0 : -ENODEV; } } if (!do_monitor) { goto cleanup_stonith_exec; } rc = stonith_api->cmds->monitor(stonith_api, 0, cmd->rsc_id, cmd->timeout / 1000); rc = stonith_api->cmds->register_callback(stonith_api, rc, 0, 0, cmd, "lrmd_stonith_callback", lrmd_stonith_callback); /* don't cleanup yet, we will find out the result of the monitor later */ if (rc > 0) { rsc->active = cmd; return rc; } else if (rc == 0) { rc = -1; } cleanup_stonith_exec: stonith_action_complete(cmd, rc); return rc; } static int lrmd_rsc_execute_service_lib(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd) { svc_action_t *action = NULL; GHashTable *params_copy = NULL; CRM_ASSERT(rsc); CRM_ASSERT(cmd); crm_trace("Creating action, resource:%s action:%s class:%s provider:%s agent:%s", rsc->rsc_id, cmd->action, rsc->class, rsc->provider, rsc->type); #if SUPPORT_NAGIOS /* Recurring operations are cancelled anyway for a stop operation */ if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS) && safe_str_eq(cmd->action, "stop")) { cmd->exec_rc = PCMK_OCF_OK; goto exec_done; } #endif params_copy = crm_str_table_dup(cmd->params); action = resources_action_create(rsc->rsc_id, rsc->class, rsc->provider, rsc->type, normalize_action_name(rsc, cmd->action), cmd->interval_ms, cmd->timeout, params_copy, cmd->service_flags); if (!action) { crm_err("Failed to create action, action:%s on resource %s", cmd->action, rsc->rsc_id); cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; goto exec_done; } action->cb_data = cmd; /* 'cmd' may not be valid after this point if * services_action_async() returned TRUE * * Upstart and systemd both synchronously determine monitor/status * results and call action_complete (which may free 'cmd') if necessary. */ if (services_action_async(action, action_complete)) { return TRUE; } cmd->exec_rc = action->rc; if(action->status != PCMK_LRM_OP_DONE) { cmd->lrmd_op_status = action->status; } else { cmd->lrmd_op_status = PCMK_LRM_OP_ERROR; } services_action_free(action); action = NULL; exec_done: cmd_finalize(cmd, rsc); return TRUE; } static gboolean lrmd_rsc_execute(lrmd_rsc_t * rsc) { lrmd_cmd_t *cmd = NULL; CRM_CHECK(rsc != NULL, return FALSE); if (rsc->active) { crm_trace("%s is still active", rsc->rsc_id); return TRUE; } if (rsc->pending_ops) { GList *first = rsc->pending_ops; cmd = first->data; if (cmd->delay_id) { crm_trace ("Command %s %s was asked to run too early, waiting for start_delay timeout of %dms", cmd->rsc_id, cmd->action, cmd->start_delay); return TRUE; } rsc->pending_ops = g_list_remove_link(rsc->pending_ops, first); g_list_free_1(first); #ifdef HAVE_SYS_TIMEB_H if (cmd->t_first_run.time == 0) { ftime(&cmd->t_first_run); } ftime(&cmd->t_run); #endif } if (!cmd) { crm_trace("Nothing further to do for %s", rsc->rsc_id); return TRUE; } rsc->active = cmd; /* only one op at a time for a rsc */ if (cmd->interval_ms) { rsc->recurring_ops = g_list_append(rsc->recurring_ops, cmd); } log_execute(cmd); if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { lrmd_rsc_execute_stonith(rsc, cmd); } else { lrmd_rsc_execute_service_lib(rsc, cmd); } return TRUE; } static gboolean lrmd_rsc_dispatch(gpointer user_data) { return lrmd_rsc_execute(user_data); } void free_rsc(gpointer data) { GListPtr gIter = NULL; lrmd_rsc_t *rsc = data; int is_stonith = safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH); gIter = rsc->pending_ops; while (gIter != NULL) { GListPtr next = gIter->next; lrmd_cmd_t *cmd = gIter->data; /* command was never executed */ cmd->lrmd_op_status = PCMK_LRM_OP_CANCELLED; cmd_finalize(cmd, NULL); gIter = next; } /* frees list, but not list elements. */ g_list_free(rsc->pending_ops); gIter = rsc->recurring_ops; while (gIter != NULL) { GListPtr next = gIter->next; lrmd_cmd_t *cmd = gIter->data; if (is_stonith) { cmd->lrmd_op_status = PCMK_LRM_OP_CANCELLED; /* If a stonith command is in-flight, just mark it as cancelled; * it is not safe to finalize/free the cmd until the stonith api * says it has either completed or timed out. */ if (rsc->active != cmd) { cmd_finalize(cmd, NULL); } } else { /* This command is already handed off to service library, * let service library cancel it and tell us via the callback * when it is cancelled. The rsc can be safely destroyed * even if we are waiting for the cancel result */ services_action_cancel(rsc->rsc_id, normalize_action_name(rsc, cmd->action), cmd->interval_ms); } gIter = next; } /* frees list, but not list elements. */ g_list_free(rsc->recurring_ops); free(rsc->rsc_id); free(rsc->class); free(rsc->provider); free(rsc->type); mainloop_destroy_trigger(rsc->work); free(rsc); } static xmlNode * process_lrmd_signon(crm_client_t *client, xmlNode *request, int call_id) { xmlNode *reply = NULL; int rc = pcmk_ok; const char *is_ipc_provider = crm_element_value(request, F_LRMD_IS_IPC_PROVIDER); const char *protocol_version = crm_element_value(request, F_LRMD_PROTOCOL_VERSION); if (compare_version(protocol_version, LRMD_MIN_PROTOCOL_VERSION) < 0) { crm_err("Cluster API version must be greater than or equal to %s, not %s", LRMD_MIN_PROTOCOL_VERSION, protocol_version); rc = -EPROTO; } reply = create_lrmd_reply(__FUNCTION__, rc, call_id); crm_xml_add(reply, F_LRMD_OPERATION, CRM_OP_REGISTER); crm_xml_add(reply, F_LRMD_CLIENTID, client->id); crm_xml_add(reply, F_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); if (crm_is_true(is_ipc_provider)) { // This is a remote connection from a cluster node's controller #ifdef SUPPORT_REMOTE ipc_proxy_add_provider(client); #endif } return reply; } static int process_lrmd_rsc_register(crm_client_t * client, uint32_t id, xmlNode * request) { int rc = pcmk_ok; lrmd_rsc_t *rsc = build_rsc_from_xml(request); lrmd_rsc_t *dup = g_hash_table_lookup(rsc_list, rsc->rsc_id); if (dup && safe_str_eq(rsc->class, dup->class) && safe_str_eq(rsc->provider, dup->provider) && safe_str_eq(rsc->type, dup->type)) { crm_warn("Can't add, RSC '%s' already present in the rsc list (%d active resources)", rsc->rsc_id, g_hash_table_size(rsc_list)); free_rsc(rsc); return rc; } g_hash_table_replace(rsc_list, rsc->rsc_id, rsc); crm_info("Added '%s' to the rsc list (%d active resources)", rsc->rsc_id, g_hash_table_size(rsc_list)); return rc; } static xmlNode * process_lrmd_get_rsc_info(xmlNode *request, int call_id) { int rc = pcmk_ok; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); xmlNode *reply = NULL; lrmd_rsc_t *rsc = NULL; if (rsc_id == NULL) { rc = -ENODEV; } else { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); rc = -ENODEV; } } reply = create_lrmd_reply(__FUNCTION__, rc, call_id); if (rsc) { crm_xml_add(reply, F_LRMD_RSC_ID, rsc->rsc_id); crm_xml_add(reply, F_LRMD_CLASS, rsc->class); crm_xml_add(reply, F_LRMD_PROVIDER, rsc->provider); crm_xml_add(reply, F_LRMD_TYPE, rsc->type); } return reply; } static int process_lrmd_rsc_unregister(crm_client_t * client, uint32_t id, xmlNode * request) { int rc = pcmk_ok; lrmd_rsc_t *rsc = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); if (!rsc_id) { return -ENODEV; } if (!(rsc = g_hash_table_lookup(rsc_list, rsc_id))) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); return pcmk_ok; } if (rsc->active) { /* let the caller know there are still active ops on this rsc to watch for */ crm_trace("Operation still in progress: %p", rsc->active); rc = -EINPROGRESS; } g_hash_table_remove(rsc_list, rsc_id); return rc; } static int process_lrmd_rsc_exec(crm_client_t * client, uint32_t id, xmlNode * request) { lrmd_rsc_t *rsc = NULL; lrmd_cmd_t *cmd = NULL; xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); int call_id; if (!rsc_id) { return -EINVAL; } if (!(rsc = g_hash_table_lookup(rsc_list, rsc_id))) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); return -ENODEV; } cmd = create_lrmd_cmd(request, client); call_id = cmd->call_id; /* Don't reference cmd after handing it off to be scheduled. * The cmd could get merged and freed. */ schedule_lrmd_cmd(rsc, cmd); return call_id; } static int cancel_op(const char *rsc_id, const char *action, guint interval_ms) { GListPtr gIter = NULL; lrmd_rsc_t *rsc = g_hash_table_lookup(rsc_list, rsc_id); /* How to cancel an action. * 1. Check pending ops list, if it hasn't been handed off * to the service library or stonith recurring list remove * it there and that will stop it. * 2. If it isn't in the pending ops list, then it's either a * recurring op in the stonith recurring list, or the service * library's recurring list. Stop it there * 3. If not found in any lists, then this operation has either * been executed already and is not a recurring operation, or * never existed. */ if (!rsc) { return -ENODEV; } for (gIter = rsc->pending_ops; gIter != NULL; gIter = gIter->next) { lrmd_cmd_t *cmd = gIter->data; if (safe_str_eq(cmd->action, action) && (cmd->interval_ms == interval_ms)) { cmd->lrmd_op_status = PCMK_LRM_OP_CANCELLED; cmd_finalize(cmd, rsc); return pcmk_ok; } } if (safe_str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH)) { /* The service library does not handle stonith operations. * We have to handle recurring stonith operations ourselves. */ for (gIter = rsc->recurring_ops; gIter != NULL; gIter = gIter->next) { lrmd_cmd_t *cmd = gIter->data; if (safe_str_eq(cmd->action, action) && (cmd->interval_ms == interval_ms)) { cmd->lrmd_op_status = PCMK_LRM_OP_CANCELLED; if (rsc->active != cmd) { cmd_finalize(cmd, rsc); } return pcmk_ok; } } } else if (services_action_cancel(rsc_id, normalize_action_name(rsc, action), interval_ms) == TRUE) { /* The service library will tell the action_complete callback function * this action was cancelled, which will destroy the cmd and remove * it from the recurring_op list. Do not do that in this function * if the service library says it cancelled it. */ return pcmk_ok; } return -EOPNOTSUPP; } static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id) { GList *cmd_list = NULL; GList *cmd_iter = NULL; /* Notice a copy of each list is created when concat is called. * This prevents odd behavior from occurring when the cmd_list * is iterated through later on. It is possible the cancel_op * function may end up modifying the recurring_ops and pending_ops * lists. If we did not copy those lists, our cmd_list iteration * could get messed up.*/ if (rsc->recurring_ops) { cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->recurring_ops)); } if (rsc->pending_ops) { cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->pending_ops)); } if (!cmd_list) { return; } for (cmd_iter = cmd_list; cmd_iter; cmd_iter = cmd_iter->next) { lrmd_cmd_t *cmd = cmd_iter->data; if (cmd->interval_ms == 0) { continue; } if (client_id && safe_str_neq(cmd->client_id, client_id)) { continue; } cancel_op(rsc->rsc_id, cmd->action, cmd->interval_ms); } /* frees only the copied list data, not the cmds */ g_list_free(cmd_list); } static int process_lrmd_rsc_cancel(crm_client_t * client, uint32_t id, xmlNode * request) { xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR); const char *rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); const char *action = crm_element_value(rsc_xml, F_LRMD_RSC_ACTION); guint interval_ms = 0; crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &interval_ms); if (!rsc_id || !action) { return -EINVAL; } return cancel_op(rsc_id, action, interval_ms); } static void add_recurring_op_xml(xmlNode *reply, lrmd_rsc_t *rsc) { xmlNode *rsc_xml = create_xml_node(reply, F_LRMD_RSC); crm_xml_add(rsc_xml, F_LRMD_RSC_ID, rsc->rsc_id); for (GList *item = rsc->recurring_ops; item != NULL; item = item->next) { lrmd_cmd_t *cmd = item->data; xmlNode *op_xml = create_xml_node(rsc_xml, T_LRMD_RSC_OP); crm_xml_add(op_xml, F_LRMD_RSC_ACTION, (cmd->real_action? cmd->real_action : cmd->action)); crm_xml_add_ms(op_xml, F_LRMD_RSC_INTERVAL, cmd->interval_ms); crm_xml_add_int(op_xml, F_LRMD_TIMEOUT, cmd->timeout_orig); } } static xmlNode * process_lrmd_get_recurring(xmlNode *request, int call_id) { int rc = pcmk_ok; const char *rsc_id = NULL; lrmd_rsc_t *rsc = NULL; xmlNode *reply = NULL; xmlNode *rsc_xml = NULL; // Resource ID is optional rsc_xml = first_named_child(request, F_LRMD_CALLDATA); if (rsc_xml) { rsc_xml = first_named_child(rsc_xml, F_LRMD_RSC); } if (rsc_xml) { rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID); } // If resource ID is specified, resource must exist if (rsc_id != NULL) { rsc = g_hash_table_lookup(rsc_list, rsc_id); if (rsc == NULL) { crm_info("Resource '%s' not found (%d active resources)", rsc_id, g_hash_table_size(rsc_list)); rc = -ENODEV; } } reply = create_lrmd_reply(__FUNCTION__, rc, call_id); // If resource ID is not specified, check all resources if (rsc_id == NULL) { GHashTableIter iter; char *key = NULL; g_hash_table_iter_init(&iter, rsc_list); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &rsc)) { add_recurring_op_xml(reply, rsc); } } else if (rsc) { add_recurring_op_xml(reply, rsc); } return reply; } void process_lrmd_message(crm_client_t * client, uint32_t id, xmlNode * request) { int rc = pcmk_ok; int call_id = 0; const char *op = crm_element_value(request, F_LRMD_OPERATION); int do_reply = 0; int do_notify = 0; xmlNode *reply = NULL; crm_trace("Processing %s operation from %s", op, client->id); crm_element_value_int(request, F_LRMD_CALLID, &call_id); if (crm_str_eq(op, CRM_OP_IPC_FWD, TRUE)) { #ifdef SUPPORT_REMOTE ipc_proxy_forward_client(client, request); #endif do_reply = 1; } else if (crm_str_eq(op, CRM_OP_REGISTER, TRUE)) { reply = process_lrmd_signon(client, request, call_id); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_REG, TRUE)) { rc = process_lrmd_rsc_register(client, id, request); do_notify = 1; do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_INFO, TRUE)) { reply = process_lrmd_get_rsc_info(request, call_id); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_UNREG, TRUE)) { rc = process_lrmd_rsc_unregister(client, id, request); /* don't notify anyone about failed un-registers */ if (rc == pcmk_ok || rc == -EINPROGRESS) { do_notify = 1; } do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_EXEC, TRUE)) { rc = process_lrmd_rsc_exec(client, id, request); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_RSC_CANCEL, TRUE)) { rc = process_lrmd_rsc_cancel(client, id, request); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_POKE, TRUE)) { do_notify = 1; do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_CHECK, TRUE)) { xmlNode *data = get_message_xml(request, F_LRMD_CALLDATA); const char *timeout = crm_element_value(data, F_LRMD_WATCHDOG); CRM_LOG_ASSERT(data != NULL); check_sbd_timeout(timeout); } else if (crm_str_eq(op, LRMD_OP_ALERT_EXEC, TRUE)) { rc = process_lrmd_alert_exec(client, id, request); do_reply = 1; } else if (crm_str_eq(op, LRMD_OP_GET_RECURRING, TRUE)) { reply = process_lrmd_get_recurring(request, call_id); do_reply = 1; } else { rc = -EOPNOTSUPP; do_reply = 1; crm_err("Unknown %s from %s", op, client->name); crm_log_xml_warn(request, "UnknownOp"); } crm_debug("Processed %s operation from %s: rc=%d, reply=%d, notify=%d", op, client->id, rc, do_reply, do_notify); if (do_reply) { int send_rc = pcmk_ok; if (reply == NULL) { reply = create_lrmd_reply(__FUNCTION__, rc, call_id); } send_rc = lrmd_server_send_reply(client, id, reply); free_xml(reply); if (send_rc < 0) { crm_warn("Reply to client %s failed: %s " CRM_XS " %d", client->name, pcmk_strerror(send_rc), send_rc); } } if (do_notify) { send_generic_notify(rc, request); } } diff --git a/include/crm_internal.h b/include/crm_internal.h index 63dca974ad..70cfac515c 100644 --- a/include/crm_internal.h +++ b/include/crm_internal.h @@ -1,354 +1,353 @@ /* * Copyright 2006-2018 Andrew Beekhof * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef CRM_INTERNAL__H # define CRM_INTERNAL__H # include # include # include # include # include # include # include # include # include /* Dynamic loading of libraries */ void *find_library_function(void **handle, const char *lib, const char *fn, int fatal); void *convert_const_pointer(const void *ptr); /* For ACLs */ char *uid2username(uid_t uid); -void determine_request_user(const char *user, xmlNode * request, const char *field); const char *crm_acl_get_set_user(xmlNode * request, const char *field, const char *peer_user); # if ENABLE_ACL # include static inline gboolean is_privileged(const char *user) { if (user == NULL) { return FALSE; } else if (strcmp(user, CRM_DAEMON_USER) == 0) { return TRUE; } else if (strcmp(user, "root") == 0) { return TRUE; } return FALSE; } # endif /* CLI option processing*/ # ifdef HAVE_GETOPT_H # include # else # define no_argument 0 # define required_argument 1 # endif # define pcmk_option_default 0x00000 # define pcmk_option_hidden 0x00001 # define pcmk_option_paragraph 0x00002 # define pcmk_option_example 0x00004 struct crm_option { /* Fields from 'struct option' in getopt.h */ /* name of long option */ const char *name; /* * one of no_argument, required_argument, and optional_argument: * whether option takes an argument */ int has_arg; /* if not NULL, set *flag to val when option found */ int *flag; /* if flag not NULL, value to set *flag to; else return value */ int val; /* Custom fields */ const char *desc; long flags; }; void crm_set_options(const char *short_options, const char *usage, struct crm_option *long_options, const char *app_desc); int crm_get_option(int argc, char **argv, int *index); int crm_get_option_long(int argc, char **argv, int *index, const char **longname); crm_exit_t crm_help(char cmd, crm_exit_t exit_code); /* Cluster Option Processing */ typedef struct pe_cluster_option_s { const char *name; const char *alt_name; const char *type; const char *values; const char *default_value; gboolean(*is_valid) (const char *); const char *description_short; const char *description_long; } pe_cluster_option; const char *cluster_option(GHashTable * options, gboolean(*validate) (const char *), const char *name, const char *old_name, const char *def_value); const char *get_cluster_pref(GHashTable * options, pe_cluster_option * option_list, int len, const char *name); void config_metadata(const char *name, const char *version, const char *desc_short, const char *desc_long, pe_cluster_option * option_list, int len); void verify_all_options(GHashTable * options, pe_cluster_option * option_list, int len); gboolean check_time(const char *value); gboolean check_timer(const char *value); gboolean check_boolean(const char *value); gboolean check_number(const char *value); gboolean check_positive_number(const char *value); gboolean check_quorum(const char *value); gboolean check_script(const char *value); gboolean check_utilization(const char *value); long crm_get_sbd_timeout(void); long crm_auto_watchdog_timeout(void); gboolean check_sbd_timeout(const char *value); void crm_args_fini(void); /* char2score */ extern int node_score_red; extern int node_score_green; extern int node_score_yellow; /* Assorted convenience functions */ void crm_make_daemon(const char *name, gboolean daemonize, const char *pidfile); // printf-style format to create operation ID from resource, action, interval #define CRM_OP_FMT "%s_%s_%u" static inline long long crm_clear_bit(const char *function, int line, const char *target, long long word, long long bit) { long long rc = (word & ~bit); if (rc == word) { /* Unchanged */ } else if (target) { crm_trace("Bit 0x%.8llx for %s cleared by %s:%d", bit, target, function, line); } else { crm_trace("Bit 0x%.8llx cleared by %s:%d", bit, function, line); } return rc; } static inline long long crm_set_bit(const char *function, int line, const char *target, long long word, long long bit) { long long rc = (word | bit); if (rc == word) { /* Unchanged */ } else if (target) { crm_trace("Bit 0x%.8llx for %s set by %s:%d", bit, target, function, line); } else { crm_trace("Bit 0x%.8llx set by %s:%d", bit, function, line); } return rc; } # define set_bit(word, bit) word = crm_set_bit(__FUNCTION__, __LINE__, NULL, word, bit) # define clear_bit(word, bit) word = crm_clear_bit(__FUNCTION__, __LINE__, NULL, word, bit) char *generate_hash_key(const char *crm_msg_reference, const char *sys); /*! remote tcp/tls helper functions */ typedef struct crm_remote_s crm_remote_t; int crm_remote_send(crm_remote_t * remote, xmlNode * msg); int crm_remote_ready(crm_remote_t * remote, int total_timeout /*ms */ ); gboolean crm_remote_recv(crm_remote_t * remote, int total_timeout /*ms */ , int *disconnected); xmlNode *crm_remote_parse_buffer(crm_remote_t * remote); int crm_remote_tcp_connect(const char *host, int port); int crm_remote_tcp_connect_async(const char *host, int port, int timeout, /*ms */ int *timer_id, void *userdata, void (*callback) (void *userdata, int sock)); int crm_remote_accept(int ssock); void crm_sockaddr2str(void *sa, char *s); # ifdef HAVE_GNUTLS_GNUTLS_H /*! * \internal * \brief Initiate the client handshake after establishing the tcp socket. * \note This is a blocking function, it will block until the entire handshake * is complete or until the timeout period is reached. * \retval 0 success * \retval negative, failure */ int crm_initiate_client_tls_handshake(crm_remote_t * remote, int timeout_ms); /*! * \internal * \brief Create client or server session for anon DH encryption credentials * \param sock, the socket the session will use for transport * \param type, GNUTLS_SERVER or GNUTLS_CLIENT * \param credentials, gnutls_anon_server_credentials_t or gnutls_anon_client_credentials_t * * \retval gnutls_session_t * on success * \retval NULL on failure */ void *crm_create_anon_tls_session(int sock, int type, void *credentials); /*! * \internal * \brief Create client or server session for PSK credentials * \param sock, the socket the session will use for transport * \param type, GNUTLS_SERVER or GNUTLS_CLIENT * \param credentials, gnutls_psk_server_credentials_t or gnutls_osk_client_credentials_t * * \retval gnutls_session_t * on success * \retval NULL on failure */ void *create_psk_tls_session(int csock, int type, void *credentials); # endif const char *daemon_option(const char *option); void set_daemon_option(const char *option, const char *value); gboolean daemon_option_enabled(const char *daemon, const char *option); void strip_text_nodes(xmlNode * xml); void pcmk_panic(const char *origin); void sysrq_init(void); pid_t pcmk_locate_sbd(void); # define crm_config_err(fmt...) { crm_config_error = TRUE; crm_err(fmt); } # define crm_config_warn(fmt...) { crm_config_warning = TRUE; crm_warn(fmt); } # define F_ATTRD_KEY "attr_key" # define F_ATTRD_ATTRIBUTE "attr_name" # define F_ATTRD_REGEX "attr_regex" # define F_ATTRD_TASK "task" # define F_ATTRD_VALUE "attr_value" # define F_ATTRD_SET "attr_set" # define F_ATTRD_IS_REMOTE "attr_is_remote" # define F_ATTRD_IS_PRIVATE "attr_is_private" # define F_ATTRD_SECTION "attr_section" # define F_ATTRD_DAMPEN "attr_dampening" # define F_ATTRD_HOST "attr_host" # define F_ATTRD_HOST_ID "attr_host_id" # define F_ATTRD_USER "attr_user" # define F_ATTRD_WRITER "attr_writer" # define F_ATTRD_VERSION "attr_version" # define F_ATTRD_RESOURCE "attr_resource" # define F_ATTRD_OPERATION "attr_clear_operation" # define F_ATTRD_INTERVAL "attr_clear_interval" /* attrd operations */ # define ATTRD_OP_PEER_REMOVE "peer-remove" # define ATTRD_OP_UPDATE "update" # define ATTRD_OP_UPDATE_BOTH "update-both" # define ATTRD_OP_UPDATE_DELAY "update-delay" # define ATTRD_OP_QUERY "query" # define ATTRD_OP_REFRESH "refresh" # define ATTRD_OP_FLUSH "flush" # define ATTRD_OP_SYNC "sync" # define ATTRD_OP_SYNC_RESPONSE "sync-response" # define ATTRD_OP_CLEAR_FAILURE "clear-failure" # define PCMK_ENV_PHYSICAL_HOST "physical_host" # if SUPPORT_COROSYNC # include # include typedef struct qb_ipc_request_header cs_ipc_header_request_t; typedef struct qb_ipc_response_header cs_ipc_header_response_t; # else typedef struct { int size __attribute__ ((aligned(8))); int id __attribute__ ((aligned(8))); } __attribute__ ((aligned(8))) cs_ipc_header_request_t; typedef struct { int size __attribute__ ((aligned(8))); int id __attribute__ ((aligned(8))); int error __attribute__ ((aligned(8))); } __attribute__ ((aligned(8))) cs_ipc_header_response_t; # endif void attrd_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb); void stonith_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb); qb_ipcs_service_t * crmd_ipc_server_init(struct qb_ipcs_service_handlers *cb); void cib_ipc_servers_init(qb_ipcs_service_t **ipcs_ro, qb_ipcs_service_t **ipcs_rw, qb_ipcs_service_t **ipcs_shm, struct qb_ipcs_service_handlers *ro_cb, struct qb_ipcs_service_handlers *rw_cb); void cib_ipc_servers_destroy(qb_ipcs_service_t *ipcs_ro, qb_ipcs_service_t *ipcs_rw, qb_ipcs_service_t *ipcs_shm); static inline void *realloc_safe(void *ptr, size_t size) { void *ret = realloc(ptr, size); if (ret == NULL) { free(ptr); /* make coverity happy */ abort(); } return ret; } const char *crm_xml_add_last_written(xmlNode *xml_node); void crm_xml_dump(xmlNode * data, int options, char **buffer, int *offset, int *max, int depth); void crm_buffer_add_char(char **buffer, int *offset, int *max, char c); gboolean crm_digest_verify(xmlNode *input, const char *expected); /* cross-platform compatibility functions */ char *crm_compat_realpath(const char *path); /* IPC Proxy Backend Shared Functions */ typedef struct remote_proxy_s { char *node_name; char *session_id; gboolean is_local; crm_ipc_t *ipc; mainloop_io_t *source; uint32_t last_request_id; lrmd_t *lrm; } remote_proxy_t; remote_proxy_t *remote_proxy_new( lrmd_t *lrmd, struct ipc_client_callbacks *proxy_callbacks, const char *node_name, const char *session_id, const char *channel); int remote_proxy_check(lrmd_t *lrmd, GHashTable *hash); void remote_proxy_cb(lrmd_t *lrmd, const char *node_name, xmlNode *msg); void remote_proxy_ack_shutdown(lrmd_t *lrmd); void remote_proxy_nack_shutdown(lrmd_t *lrmd); int remote_proxy_dispatch(const char *buffer, ssize_t length, gpointer userdata); void remote_proxy_disconnected(gpointer data); void remote_proxy_free(gpointer data); void remote_proxy_relay_event(remote_proxy_t *proxy, xmlNode *msg); void remote_proxy_relay_response(remote_proxy_t *proxy, xmlNode *msg, int msg_id); #endif /* CRM_INTERNAL__H */ diff --git a/lib/common/utils.c b/lib/common/utils.c index 2bf4aeedbd..955d313453 100644 --- a/lib/common/utils.c +++ b/lib/common/utils.c @@ -1,1407 +1,1386 @@ /* * Copyright 2004-2018 Andrew Beekhof * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef MAXLINE # define MAXLINE 512 #endif #ifdef HAVE_GETOPT_H # include #endif #ifndef PW_BUFFER_LEN # define PW_BUFFER_LEN 500 #endif CRM_TRACE_INIT_DATA(common); gboolean crm_config_error = FALSE; gboolean crm_config_warning = FALSE; char *crm_system_name = NULL; int node_score_red = 0; int node_score_green = 0; int node_score_yellow = 0; static struct crm_option *crm_long_options = NULL; static const char *crm_app_description = NULL; static char *crm_short_options = NULL; static const char *crm_app_usage = NULL; gboolean check_time(const char *value) { if (crm_get_msec(value) < 5000) { return FALSE; } return TRUE; } gboolean check_timer(const char *value) { if (crm_get_msec(value) < 0) { return FALSE; } return TRUE; } gboolean check_boolean(const char *value) { int tmp = FALSE; if (crm_str_to_boolean(value, &tmp) != 1) { return FALSE; } return TRUE; } gboolean check_number(const char *value) { errno = 0; if (value == NULL) { return FALSE; } else if (safe_str_eq(value, CRM_MINUS_INFINITY_S)) { } else if (safe_str_eq(value, CRM_INFINITY_S)) { } else { crm_int_helper(value, NULL); } if (errno != 0) { return FALSE; } return TRUE; } gboolean check_positive_number(const char* value) { if (safe_str_eq(value, CRM_INFINITY_S) || (crm_int_helper(value, NULL))) { return TRUE; } return FALSE; } gboolean check_quorum(const char *value) { if (safe_str_eq(value, "stop")) { return TRUE; } else if (safe_str_eq(value, "freeze")) { return TRUE; } else if (safe_str_eq(value, "ignore")) { return TRUE; } else if (safe_str_eq(value, "suicide")) { return TRUE; } return FALSE; } gboolean check_script(const char *value) { struct stat st; if(safe_str_eq(value, "/dev/null")) { return TRUE; } if(stat(value, &st) != 0) { crm_err("Script %s does not exist", value); return FALSE; } if(S_ISREG(st.st_mode) == 0) { crm_err("Script %s is not a regular file", value); return FALSE; } if( (st.st_mode & (S_IXUSR | S_IXGRP )) == 0) { crm_err("Script %s is not executable", value); return FALSE; } return TRUE; } gboolean check_utilization(const char *value) { char *end = NULL; long number = strtol(value, &end, 10); if(end && end[0] != '%') { return FALSE; } else if(number < 0) { return FALSE; } return TRUE; } void crm_args_fini() { free(crm_short_options); crm_short_options = NULL; } int char2score(const char *score) { int score_f = 0; if (score == NULL) { } else if (safe_str_eq(score, CRM_MINUS_INFINITY_S)) { score_f = -CRM_SCORE_INFINITY; } else if (safe_str_eq(score, CRM_INFINITY_S)) { score_f = CRM_SCORE_INFINITY; } else if (safe_str_eq(score, CRM_PLUS_INFINITY_S)) { score_f = CRM_SCORE_INFINITY; } else if (safe_str_eq(score, "red")) { score_f = node_score_red; } else if (safe_str_eq(score, "yellow")) { score_f = node_score_yellow; } else if (safe_str_eq(score, "green")) { score_f = node_score_green; } else { score_f = crm_parse_int(score, NULL); if (score_f > 0 && score_f > CRM_SCORE_INFINITY) { score_f = CRM_SCORE_INFINITY; } else if (score_f < 0 && score_f < -CRM_SCORE_INFINITY) { score_f = -CRM_SCORE_INFINITY; } } return score_f; } char * score2char_stack(int score, char *buf, size_t len) { if (score >= CRM_SCORE_INFINITY) { strncpy(buf, CRM_INFINITY_S, 9); } else if (score <= -CRM_SCORE_INFINITY) { strncpy(buf, CRM_MINUS_INFINITY_S , 10); } else { return crm_itoa_stack(score, buf, len); } return buf; } char * score2char(int score) { if (score >= CRM_SCORE_INFINITY) { return strdup(CRM_INFINITY_S); } else if (score <= -CRM_SCORE_INFINITY) { return strdup(CRM_MINUS_INFINITY_S); } return crm_itoa(score); } const char * cluster_option(GHashTable * options, gboolean(*validate) (const char *), const char *name, const char *old_name, const char *def_value) { const char *value = NULL; char *new_value = NULL; CRM_ASSERT(name != NULL); if (options) { value = g_hash_table_lookup(options, name); if ((value == NULL) && old_name) { value = g_hash_table_lookup(options, old_name); if (value != NULL) { crm_config_warn("Support for legacy name '%s' for cluster option '%s'" " is deprecated and will be removed in a future release", old_name, name); // Inserting copy with current name ensures we only warn once new_value = strdup(value); g_hash_table_insert(options, strdup(name), new_value); value = new_value; } } if (value && validate && (validate(value) == FALSE)) { crm_config_err("Resetting cluster option '%s' to default: value '%s' is invalid", name, value); value = NULL; } if (value) { return value; } } // No value found, use default value = def_value; if (value == NULL) { crm_trace("No value or default provided for cluster option '%s'", name); return NULL; } if (validate) { CRM_CHECK(validate(value) != FALSE, crm_err("Bug: default value for cluster option '%s' is invalid", name); return NULL); } crm_trace("Using default value '%s' for cluster option '%s'", value, name); if (options) { new_value = strdup(value); g_hash_table_insert(options, strdup(name), new_value); value = new_value; } return value; } const char * get_cluster_pref(GHashTable * options, pe_cluster_option * option_list, int len, const char *name) { const char *value = NULL; for (int lpc = 0; lpc < len; lpc++) { if (safe_str_eq(name, option_list[lpc].name)) { value = cluster_option(options, option_list[lpc].is_valid, option_list[lpc].name, option_list[lpc].alt_name, option_list[lpc].default_value); return value; } } CRM_CHECK(FALSE, crm_err("Bug: looking for unknown option '%s'", name)); return NULL; } void config_metadata(const char *name, const char *version, const char *desc_short, const char *desc_long, pe_cluster_option * option_list, int len) { int lpc = 0; fprintf(stdout, "" "\n" "\n" " %s\n" " %s\n" " %s\n" " \n", name, version, desc_long, desc_short); for (lpc = 0; lpc < len; lpc++) { if (option_list[lpc].description_long == NULL && option_list[lpc].description_short == NULL) { continue; } fprintf(stdout, " \n" " %s\n" " \n" " %s%s%s\n" " \n", option_list[lpc].name, option_list[lpc].description_short, option_list[lpc].type, option_list[lpc].default_value, option_list[lpc].description_long ? option_list[lpc]. description_long : option_list[lpc].description_short, option_list[lpc].values ? " Allowed values: " : "", option_list[lpc].values ? option_list[lpc].values : ""); } fprintf(stdout, " \n\n"); } void verify_all_options(GHashTable * options, pe_cluster_option * option_list, int len) { int lpc = 0; for (lpc = 0; lpc < len; lpc++) { cluster_option(options, option_list[lpc].is_valid, option_list[lpc].name, option_list[lpc].alt_name, option_list[lpc].default_value); } } char * generate_hash_key(const char *crm_msg_reference, const char *sys) { char *hash_key = crm_concat(sys ? sys : "none", crm_msg_reference, '_'); crm_trace("created hash key: (%s)", hash_key); return hash_key; } int crm_user_lookup(const char *name, uid_t * uid, gid_t * gid) { int rc = pcmk_ok; char *buffer = NULL; struct passwd pwd; struct passwd *pwentry = NULL; buffer = calloc(1, PW_BUFFER_LEN); rc = getpwnam_r(name, &pwd, buffer, PW_BUFFER_LEN, &pwentry); if (pwentry) { if (uid) { *uid = pwentry->pw_uid; } if (gid) { *gid = pwentry->pw_gid; } crm_trace("User %s has uid=%d gid=%d", name, pwentry->pw_uid, pwentry->pw_gid); } else { rc = rc? -rc : -EINVAL; crm_info("User %s lookup: %s", name, pcmk_strerror(rc)); } free(buffer); return rc; } static int crm_version_helper(const char *text, char **end_text) { int atoi_result = -1; CRM_ASSERT(end_text != NULL); errno = 0; if (text != NULL && text[0] != 0) { atoi_result = (int)strtol(text, end_text, 10); if (errno == EINVAL) { crm_err("Conversion of '%s' %c failed", text, text[0]); atoi_result = -1; } } return atoi_result; } /* * version1 < version2 : -1 * version1 = version2 : 0 * version1 > version2 : 1 */ int compare_version(const char *version1, const char *version2) { int rc = 0; int lpc = 0; char *ver1_copy = NULL, *ver2_copy = NULL; char *rest1 = NULL, *rest2 = NULL; if (version1 == NULL && version2 == NULL) { return 0; } else if (version1 == NULL) { return -1; } else if (version2 == NULL) { return 1; } ver1_copy = strdup(version1); ver2_copy = strdup(version2); rest1 = ver1_copy; rest2 = ver2_copy; while (1) { int digit1 = 0; int digit2 = 0; lpc++; if (rest1 == rest2) { break; } if (rest1 != NULL) { digit1 = crm_version_helper(rest1, &rest1); } if (rest2 != NULL) { digit2 = crm_version_helper(rest2, &rest2); } if (digit1 < digit2) { rc = -1; break; } else if (digit1 > digit2) { rc = 1; break; } if (rest1 != NULL && rest1[0] == '.') { rest1++; } if (rest1 != NULL && rest1[0] == 0) { rest1 = NULL; } if (rest2 != NULL && rest2[0] == '.') { rest2++; } if (rest2 != NULL && rest2[0] == 0) { rest2 = NULL; } } free(ver1_copy); free(ver2_copy); if (rc == 0) { crm_trace("%s == %s (%d)", version1, version2, lpc); } else if (rc < 0) { crm_trace("%s < %s (%d)", version1, version2, lpc); } else if (rc > 0) { crm_trace("%s > %s (%d)", version1, version2, lpc); } return rc; } gboolean do_stderr = FALSE; #ifndef NUMCHARS # define NUMCHARS "0123456789." #endif #ifndef WHITESPACE # define WHITESPACE " \t\n\r\f" #endif guint crm_parse_interval_spec(const char *input) { long long msec = 0; if (input == NULL) { return 0; } else if (input[0] != 'P') { long long tmp = crm_get_msec(input); if(tmp > 0) { msec = tmp; } } else { crm_time_t *period_s = crm_time_parse_duration(input); msec = 1000 * crm_time_get_seconds(period_s); crm_time_free(period_s); } return (msec <= 0)? 0 : ((msec >= G_MAXUINT)? G_MAXUINT : (guint) msec); } long long crm_get_msec(const char *input) { const char *cp = input; const char *units; long long multiplier = 1000; long long divisor = 1; long long msec = -1; char *end_text = NULL; /* double dret; */ if (input == NULL) { return msec; } cp += strspn(cp, WHITESPACE); units = cp + strspn(cp, NUMCHARS); units += strspn(units, WHITESPACE); if (strchr(NUMCHARS, *cp) == NULL) { return msec; } if (strncasecmp(units, "ms", 2) == 0 || strncasecmp(units, "msec", 4) == 0) { multiplier = 1; divisor = 1; } else if (strncasecmp(units, "us", 2) == 0 || strncasecmp(units, "usec", 4) == 0) { multiplier = 1; divisor = 1000; } else if (strncasecmp(units, "s", 1) == 0 || strncasecmp(units, "sec", 3) == 0) { multiplier = 1000; divisor = 1; } else if (strncasecmp(units, "m", 1) == 0 || strncasecmp(units, "min", 3) == 0) { multiplier = 60 * 1000; divisor = 1; } else if (strncasecmp(units, "h", 1) == 0 || strncasecmp(units, "hr", 2) == 0) { multiplier = 60 * 60 * 1000; divisor = 1; } else if (*units != EOS && *units != '\n' && *units != '\r') { return msec; } msec = crm_int_helper(cp, &end_text); if (msec > LLONG_MAX/multiplier) { /* arithmetics overflow while multiplier/divisor mutually exclusive */ return LLONG_MAX; } msec *= multiplier; msec /= divisor; /* dret += 0.5; */ /* msec = (long long)dret; */ return msec; } extern bool crm_is_daemon; /* coverity[+kill] */ void crm_abort(const char *file, const char *function, int line, const char *assert_condition, gboolean do_core, gboolean do_fork) { int rc = 0; int pid = 0; int status = 0; /* Implied by the parent's error logging below */ /* crm_write_blackbox(0); */ if(crm_is_daemon == FALSE) { /* This is a command line tool - do not fork */ /* crm_add_logfile(NULL); * Record it to a file? */ crm_enable_stderr(TRUE); /* Make sure stderr is enabled so we can tell the caller */ do_fork = FALSE; /* Just crash if needed */ } if (do_core == FALSE) { crm_err("%s: Triggered assert at %s:%d : %s", function, file, line, assert_condition); return; } else if (do_fork) { pid = fork(); } else { crm_err("%s: Triggered fatal assert at %s:%d : %s", function, file, line, assert_condition); } if (pid == -1) { crm_crit("%s: Cannot create core for non-fatal assert at %s:%d : %s", function, file, line, assert_condition); return; } else if(pid == 0) { /* Child process */ abort(); return; } /* Parent process */ crm_err("%s: Forked child %d to record non-fatal assert at %s:%d : %s", function, pid, file, line, assert_condition); crm_write_blackbox(SIGTRAP, NULL); do { rc = waitpid(pid, &status, 0); if(rc == pid) { return; /* Job done */ } } while(errno == EINTR); if (errno == ECHILD) { /* crm_mon does this */ crm_trace("Cannot wait on forked child %d - SIGCHLD is probably set to SIG_IGN", pid); return; } crm_perror(LOG_ERR, "Cannot wait on forked child %d", pid); } void crm_make_daemon(const char *name, gboolean daemonize, const char *pidfile) { int rc; long pid; const char *devnull = "/dev/null"; if (daemonize == FALSE) { return; } /* Check before we even try... */ rc = crm_pidfile_inuse(pidfile, 1, name); if(rc < pcmk_ok && rc != -ENOENT) { pid = crm_read_pidfile(pidfile); crm_err("%s: already running [pid %ld in %s]", name, pid, pidfile); printf("%s: already running [pid %ld in %s]\n", name, pid, pidfile); crm_exit(CRM_EX_ERROR); } pid = fork(); if (pid < 0) { fprintf(stderr, "%s: could not start daemon\n", name); crm_perror(LOG_ERR, "fork"); crm_exit(CRM_EX_OSERR); } else if (pid > 0) { crm_exit(CRM_EX_OK); } rc = crm_lock_pidfile(pidfile, name); if(rc < pcmk_ok) { crm_err("Could not lock '%s' for %s: %s (%d)", pidfile, name, pcmk_strerror(rc), rc); printf("Could not lock '%s' for %s: %s (%d)\n", pidfile, name, pcmk_strerror(rc), rc); crm_exit(CRM_EX_ERROR); } umask(S_IWGRP | S_IWOTH | S_IROTH); close(STDIN_FILENO); (void)open(devnull, O_RDONLY); /* Stdin: fd 0 */ close(STDOUT_FILENO); (void)open(devnull, O_WRONLY); /* Stdout: fd 1 */ close(STDERR_FILENO); (void)open(devnull, O_WRONLY); /* Stderr: fd 2 */ } char * crm_meta_name(const char *field) { int lpc = 0; int max = 0; char *crm_name = NULL; CRM_CHECK(field != NULL, return NULL); crm_name = crm_concat(CRM_META, field, '_'); /* Massage the names so they can be used as shell variables */ max = strlen(crm_name); for (; lpc < max; lpc++) { switch (crm_name[lpc]) { case '-': crm_name[lpc] = '_'; break; } } return crm_name; } const char * crm_meta_value(GHashTable * hash, const char *field) { char *key = NULL; const char *value = NULL; key = crm_meta_name(field); if (key) { value = g_hash_table_lookup(hash, key); free(key); } return value; } static struct option * crm_create_long_opts(struct crm_option *long_options) { struct option *long_opts = NULL; #ifdef HAVE_GETOPT_H int index = 0, lpc = 0; /* * A previous, possibly poor, choice of '?' as the short form of --help * means that getopt_long() returns '?' for both --help and for "unknown option" * * This dummy entry allows us to differentiate between the two in crm_get_option() * and exit with the correct error code */ long_opts = realloc_safe(long_opts, (index + 1) * sizeof(struct option)); long_opts[index].name = "__dummmy__"; long_opts[index].has_arg = 0; long_opts[index].flag = 0; long_opts[index].val = '_'; index++; for (lpc = 0; long_options[lpc].name != NULL; lpc++) { if (long_options[lpc].name[0] == '-') { continue; } long_opts = realloc_safe(long_opts, (index + 1) * sizeof(struct option)); /*fprintf(stderr, "Creating %d %s = %c\n", index, * long_options[lpc].name, long_options[lpc].val); */ long_opts[index].name = long_options[lpc].name; long_opts[index].has_arg = long_options[lpc].has_arg; long_opts[index].flag = long_options[lpc].flag; long_opts[index].val = long_options[lpc].val; index++; } /* Now create the list terminator */ long_opts = realloc_safe(long_opts, (index + 1) * sizeof(struct option)); long_opts[index].name = NULL; long_opts[index].has_arg = 0; long_opts[index].flag = 0; long_opts[index].val = 0; #endif return long_opts; } void crm_set_options(const char *short_options, const char *app_usage, struct crm_option *long_options, const char *app_desc) { if (short_options) { crm_short_options = strdup(short_options); } else if (long_options) { int lpc = 0; int opt_string_len = 0; char *local_short_options = NULL; for (lpc = 0; long_options[lpc].name != NULL; lpc++) { if (long_options[lpc].val && long_options[lpc].val != '-' && long_options[lpc].val < UCHAR_MAX) { local_short_options = realloc_safe(local_short_options, opt_string_len + 4); local_short_options[opt_string_len++] = long_options[lpc].val; /* getopt(3) says: Two colons mean an option takes an optional arg; */ if (long_options[lpc].has_arg == optional_argument) { local_short_options[opt_string_len++] = ':'; } if (long_options[lpc].has_arg >= required_argument) { local_short_options[opt_string_len++] = ':'; } local_short_options[opt_string_len] = 0; } } crm_short_options = local_short_options; crm_trace("Generated short option string: '%s'", local_short_options); } if (long_options) { crm_long_options = long_options; } if (app_desc) { crm_app_description = app_desc; } if (app_usage) { crm_app_usage = app_usage; } } int crm_get_option(int argc, char **argv, int *index) { return crm_get_option_long(argc, argv, index, NULL); } int crm_get_option_long(int argc, char **argv, int *index, const char **longname) { #ifdef HAVE_GETOPT_H static struct option *long_opts = NULL; if (long_opts == NULL && crm_long_options) { long_opts = crm_create_long_opts(crm_long_options); } *index = 0; if (long_opts) { int flag = getopt_long(argc, argv, crm_short_options, long_opts, index); switch (flag) { case 0: if (long_opts[*index].val) { return long_opts[*index].val; } else if (longname) { *longname = long_opts[*index].name; } else { crm_notice("Unhandled option --%s", long_opts[*index].name); return flag; } case -1: /* End of option processing */ break; case ':': crm_trace("Missing argument"); crm_help('?', CRM_EX_USAGE); break; case '?': crm_help('?', (*index? CRM_EX_OK : CRM_EX_USAGE)); break; } return flag; } #endif if (crm_short_options) { return getopt(argc, argv, crm_short_options); } return -1; } crm_exit_t crm_help(char cmd, crm_exit_t exit_code) { int i = 0; FILE *stream = (exit_code ? stderr : stdout); if (cmd == 'v' || cmd == '$') { fprintf(stream, "Pacemaker %s\n", PACEMAKER_VERSION); fprintf(stream, "Written by Andrew Beekhof\n"); goto out; } if (cmd == '!') { fprintf(stream, "Pacemaker %s (Build: %s): %s\n", PACEMAKER_VERSION, BUILD_VERSION, CRM_FEATURES); goto out; } fprintf(stream, "%s - %s\n", crm_system_name, crm_app_description); if (crm_app_usage) { fprintf(stream, "Usage: %s %s\n", crm_system_name, crm_app_usage); } if (crm_long_options) { fprintf(stream, "Options:\n"); for (i = 0; crm_long_options[i].name != NULL; i++) { if (crm_long_options[i].flags & pcmk_option_hidden) { } else if (crm_long_options[i].flags & pcmk_option_paragraph) { fprintf(stream, "%s\n\n", crm_long_options[i].desc); } else if (crm_long_options[i].flags & pcmk_option_example) { fprintf(stream, "\t#%s\n\n", crm_long_options[i].desc); } else if (crm_long_options[i].val == '-' && crm_long_options[i].desc) { fprintf(stream, "%s\n", crm_long_options[i].desc); } else { /* is val printable as char ? */ if (crm_long_options[i].val && crm_long_options[i].val <= UCHAR_MAX) { fprintf(stream, " -%c,", crm_long_options[i].val); } else { fputs(" ", stream); } fprintf(stream, " --%s%s\t%s\n", crm_long_options[i].name, crm_long_options[i].has_arg == optional_argument ? "[=value]" : crm_long_options[i].has_arg == required_argument ? "=value" : "", crm_long_options[i].desc ? crm_long_options[i].desc : ""); } } } else if (crm_short_options) { fprintf(stream, "Usage: %s - %s\n", crm_system_name, crm_app_description); for (i = 0; crm_short_options[i] != 0; i++) { int has_arg = no_argument /* 0 */; if (crm_short_options[i + 1] == ':') { if (crm_short_options[i + 2] == ':') has_arg = optional_argument /* 2 */; else has_arg = required_argument /* 1 */; } fprintf(stream, " -%c %s\n", crm_short_options[i], has_arg == optional_argument ? "[value]" : has_arg == required_argument ? "{value}" : ""); i += has_arg; } } fprintf(stream, "\nReport bugs to %s\n", PACKAGE_BUGREPORT); out: return crm_exit(exit_code); } void cib_ipc_servers_init(qb_ipcs_service_t **ipcs_ro, qb_ipcs_service_t **ipcs_rw, qb_ipcs_service_t **ipcs_shm, struct qb_ipcs_service_handlers *ro_cb, struct qb_ipcs_service_handlers *rw_cb) { *ipcs_ro = mainloop_add_ipc_server(cib_channel_ro, QB_IPC_NATIVE, ro_cb); *ipcs_rw = mainloop_add_ipc_server(cib_channel_rw, QB_IPC_NATIVE, rw_cb); *ipcs_shm = mainloop_add_ipc_server(cib_channel_shm, QB_IPC_SHM, rw_cb); if (*ipcs_ro == NULL || *ipcs_rw == NULL || *ipcs_shm == NULL) { crm_err("Failed to create the CIB manager: exiting and inhibiting respawn"); crm_warn("Verify pacemaker and pacemaker_remote are not both enabled"); crm_exit(CRM_EX_FATAL); } } void cib_ipc_servers_destroy(qb_ipcs_service_t *ipcs_ro, qb_ipcs_service_t *ipcs_rw, qb_ipcs_service_t *ipcs_shm) { qb_ipcs_destroy(ipcs_ro); qb_ipcs_destroy(ipcs_rw); qb_ipcs_destroy(ipcs_shm); } qb_ipcs_service_t * crmd_ipc_server_init(struct qb_ipcs_service_handlers *cb) { return mainloop_add_ipc_server(CRM_SYSTEM_CRMD, QB_IPC_NATIVE, cb); } void attrd_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb) { *ipcs = mainloop_add_ipc_server(T_ATTRD, QB_IPC_NATIVE, cb); if (*ipcs == NULL) { crm_err("Failed to create pacemaker-attrd server: exiting and inhibiting respawn"); crm_warn("Verify pacemaker and pacemaker_remote are not both enabled."); crm_exit(CRM_EX_FATAL); } } void stonith_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb) { *ipcs = mainloop_add_ipc_server("stonith-ng", QB_IPC_NATIVE, cb); if (*ipcs == NULL) { crm_err("Failed to create fencer: exiting and inhibiting respawn."); crm_warn("Verify pacemaker and pacemaker_remote are not both enabled."); crm_exit(CRM_EX_FATAL); } } bool pcmk_acl_required(const char *user) { #if ENABLE_ACL if(user == NULL || strlen(user) == 0) { crm_trace("no user set"); return FALSE; } else if (strcmp(user, CRM_DAEMON_USER) == 0) { return FALSE; } else if (strcmp(user, "root") == 0) { return FALSE; } crm_trace("acls required for %s", user); return TRUE; #else crm_trace("acls not supported"); return FALSE; #endif } #if ENABLE_ACL char * uid2username(uid_t uid) { struct passwd *pwent = getpwuid(uid); if (pwent == NULL) { crm_perror(LOG_ERR, "Cannot get password entry of uid: %d", uid); return NULL; } else { return strdup(pwent->pw_name); } } const char * crm_acl_get_set_user(xmlNode * request, const char *field, const char *peer_user) { /* field is only checked for backwards compatibility */ static const char *effective_user = NULL; const char *requested_user = NULL; const char *user = NULL; if(effective_user == NULL) { effective_user = uid2username(geteuid()); } requested_user = crm_element_value(request, XML_ACL_TAG_USER); if(requested_user == NULL) { requested_user = crm_element_value(request, field); } if (is_privileged(effective_user) == FALSE) { /* We're not running as a privileged user, set or overwrite any existing value for $XML_ACL_TAG_USER */ user = effective_user; } else if(peer_user == NULL && requested_user == NULL) { /* No user known or requested, use 'effective_user' and make sure one is set for the request */ user = effective_user; } else if(peer_user == NULL) { /* No user known, trusting 'requested_user' */ user = requested_user; } else if (is_privileged(peer_user) == FALSE) { /* The peer is not a privileged user, set or overwrite any existing value for $XML_ACL_TAG_USER */ user = peer_user; } else if (requested_user == NULL) { /* Even if we're privileged, make sure there is always a value set */ user = peer_user; } else { /* Legal delegation to 'requested_user' */ user = requested_user; } // This requires pointer comparison, not string comparison if(user != crm_element_value(request, XML_ACL_TAG_USER)) { crm_xml_add(request, XML_ACL_TAG_USER, user); } if(field != NULL && user != crm_element_value(request, field)) { crm_xml_add(request, field, user); } return requested_user; } - -void -determine_request_user(const char *user, xmlNode * request, const char *field) -{ - /* Get our internal validation out of the way first */ - CRM_CHECK(user != NULL && request != NULL && field != NULL, return); - - /* If our peer is a privileged user, we might be doing something on behalf of someone else */ - if (is_privileged(user) == FALSE) { - /* We're not a privileged user, set or overwrite any existing value for $field */ - crm_xml_replace(request, field, user); - - } else if (crm_element_value(request, field) == NULL) { - /* Even if we're privileged, make sure there is always a value set */ - crm_xml_replace(request, field, user); - -/* } else { Legal delegation */ - } - - crm_trace("Processing msg as user '%s'", crm_element_value(request, field)); -} #endif void * find_library_function(void **handle, const char *lib, const char *fn, gboolean fatal) { char *error; void *a_function; if (*handle == NULL) { *handle = dlopen(lib, RTLD_LAZY); } if (!(*handle)) { crm_err("%sCould not open %s: %s", fatal ? "Fatal: " : "", lib, dlerror()); if (fatal) { crm_exit(CRM_EX_FATAL); } return NULL; } a_function = dlsym(*handle, fn); if (a_function == NULL) { error = dlerror(); crm_err("%sCould not find %s in %s: %s", fatal ? "Fatal: " : "", fn, lib, error); if (fatal) { crm_exit(CRM_EX_FATAL); } } return a_function; } void * convert_const_pointer(const void *ptr) { /* Worst function ever */ return (void *)ptr; } #ifdef HAVE_UUID_UUID_H # include #endif char * crm_generate_uuid(void) { unsigned char uuid[16]; char *buffer = malloc(37); /* Including NUL byte */ uuid_generate(uuid); uuid_unparse(uuid, buffer); return buffer; } /*! * \brief Get name to be used as identifier for cluster messages * * \param[in] name Actual system name to check * * \return Non-NULL cluster message identifier corresponding to name * * \note The Pacemaker daemons were renamed in version 2.0.0, but the old names * must continue to be used as the identifier for cluster messages, so * that mixed-version clusters are possible during a rolling upgrade. */ const char * pcmk_message_name(const char *name) { if (name == NULL) { return "unknown"; } else if (!strcmp(name, "pacemaker-attrd")) { return "attrd"; } else if (!strcmp(name, "pacemaker-based")) { return CRM_SYSTEM_CIB; } else if (!strcmp(name, "pacemaker-controld")) { return CRM_SYSTEM_CRMD; } else if (!strcmp(name, "pacemaker-execd")) { return CRM_SYSTEM_LRMD; } else if (!strcmp(name, "pacemaker-fenced")) { return "stonith-ng"; } else if (!strcmp(name, "pacemaker-schedulerd")) { return CRM_SYSTEM_PENGINE; } else { return name; } } /*! * \brief Check whether a string represents a cluster daemon name * * \param[in] name String to check * * \return TRUE if name is standard client name used by daemons, FALSE otherwise */ bool crm_is_daemon_name(const char *name) { name = pcmk_message_name(name); return (!strcmp(name, CRM_SYSTEM_CRMD) || !strcmp(name, CRM_SYSTEM_STONITHD) || !strcmp(name, "stonith-ng") || !strcmp(name, "attrd") || !strcmp(name, CRM_SYSTEM_CIB) || !strcmp(name, CRM_SYSTEM_MCP) || !strcmp(name, CRM_SYSTEM_DC) || !strcmp(name, CRM_SYSTEM_TENGINE) || !strcmp(name, CRM_SYSTEM_LRMD)); } #include char * crm_md5sum(const char *buffer) { int lpc = 0, len = 0; char *digest = NULL; unsigned char raw_digest[MD5_DIGEST_SIZE]; if (buffer == NULL) { buffer = ""; } len = strlen(buffer); crm_trace("Beginning digest of %d bytes", len); digest = malloc(2 * MD5_DIGEST_SIZE + 1); if(digest) { md5_buffer(buffer, len, raw_digest); for (lpc = 0; lpc < MD5_DIGEST_SIZE; lpc++) { sprintf(digest + (2 * lpc), "%02x", raw_digest[lpc]); } digest[(2 * MD5_DIGEST_SIZE)] = 0; crm_trace("Digest %s.", digest); } else { crm_err("Could not create digest"); } return digest; } #ifdef HAVE_GNUTLS_GNUTLS_H void crm_gnutls_global_init(void) { signal(SIGPIPE, SIG_IGN); gnutls_global_init(); } #endif char * crm_generate_ra_key(const char *standard, const char *provider, const char *type) { if (!standard && !provider && !type) { return NULL; } return crm_strdup_printf("%s%s%s:%s", (standard? standard : ""), (provider? ":" : ""), (provider? provider : ""), (type? type : "")); } /*! * \brief Check whether a resource standard requires a provider to be specified * * \param[in] standard Standard name * * \return TRUE if standard requires a provider, FALSE otherwise */ bool crm_provider_required(const char *standard) { CRM_CHECK(standard != NULL, return FALSE); /* @TODO * - this should probably be case-sensitive, but isn't, * for backward compatibility * - it might be nice to keep standards' capabilities (supports provider, * can be promotable, etc.) as structured data somewhere */ if (!strcasecmp(standard, PCMK_RESOURCE_CLASS_OCF)) { return TRUE; } return FALSE; } /*! * \brief Parse a "standard[:provider]:type" agent specification * * \param[in] spec Agent specification * \param[out] standard Newly allocated memory containing agent standard (or NULL) * \param[out] provider Newly allocated memory containing agent provider (or NULL) * \param[put] type Newly allocated memory containing agent type (or NULL) * * \return pcmk_ok if the string could be parsed, -EINVAL otherwise * * \note It is acceptable for the type to contain a ':' if the standard supports * that. For example, systemd supports the form "systemd:UNIT@A:B". * \note It is the caller's responsibility to free the returned values. */ int crm_parse_agent_spec(const char *spec, char **standard, char **provider, char **type) { char *colon; CRM_CHECK(spec && standard && provider && type, return -EINVAL); *standard = NULL; *provider = NULL; *type = NULL; colon = strchr(spec, ':'); if ((colon == NULL) || (colon == spec)) { return -EINVAL; } *standard = strndup(spec, colon - spec); spec = colon + 1; if (crm_provider_required(*standard)) { colon = strchr(spec, ':'); if ((colon == NULL) || (colon == spec)) { free(*standard); return -EINVAL; } *provider = strndup(spec, colon - spec); spec = colon + 1; } if (*spec == '\0') { free(*standard); free(*provider); return -EINVAL; } *type = strdup(spec); return pcmk_ok; } /*! * \brief Get the local hostname * * \return Newly allocated string with name, or NULL (and set errno) on error */ char * pcmk_hostname() { struct utsname hostinfo; return (uname(&hostinfo) < 0)? NULL : strdup(hostinfo.nodename); }