4 Functionality common to all account classes.
6 Each subclass of Account must provide five methods:
7 (*) create() and destroy(), which are static;
8 (*) configure(), start(), and stop(), which are not.
10 configure(), which takes a record as its only argument, does
11 things like set up ssh keys. In addition, an Account subclass must
12 provide static member variables SHELL, which contains the unique shell
13 that it uses; and TYPE, a string that is used by the account creation
14 code. For no particular reason, TYPE is divided hierarchically by
15 periods; at the moment the only convention is that all sliver accounts
16 have type that begins with sliver.
18 There are any number of race conditions that may result from the fact
19 that account names are not unique over time. Moreover, it's a bad
20 idea to perform lengthy operations while holding the database lock.
21 In order to deal with both of these problems, we use a worker thread
22 for each account name that ever exists. On 32-bit systems with large
23 numbers of accounts, this may cause the NM process to run out of
24 *virtual* memory! This problem may be remedied by decreasing the
37 # shell path -> account class association
39 # account type -> account class association
42 # these semaphores are acquired before creating/destroying an account
43 create_sem = threading.Semaphore(1)
44 destroy_sem = threading.Semaphore(1)
46 def register_class(acct_class):
48 Call once for each account class. This method adds the class
49 to the dictionaries used to look up account classes
52 shell_acct_class[acct_class.SHELL] = acct_class
53 type_acct_class[acct_class.TYPE] = acct_class
56 # private account name -> worker object association and associated lock
57 name_worker_lock = threading.Lock()
61 return [pw_ent for pw_ent in pwd.getpwall() if pw_ent[6] in shell_acct_class]
64 """Return the names of all accounts on the system with recognized shells."""
65 return [pw_ent[0] for pw_ent in allpwents()]
69 Return the worker object for a particular username.
70 If no such object exists, create it first.
72 name_worker_lock.acquire()
74 if name not in name_worker: name_worker[name] = Worker(name)
75 return name_worker[name]
76 finally: name_worker_lock.release()
81 Base class for all types of account
84 def __init__(self, name):
87 logger.verbose('account: Initing account {}'.format(name))
90 def create(name, vref = None):
97 def configure(self, rec):
99 Write <rec['keys']> to my authorized_keys file.
101 new_keys = rec['keys']
102 logger.verbose('account: configuring {} with {} keys'.format(self.name, len(new_keys)))
103 if new_keys != self.keys:
104 # get the unix account info
105 gid = grp.getgrnam("slices")[2]
106 pw_info = pwd.getpwnam(self.name)
110 # write out authorized_keys file and conditionally create
111 # the .ssh subdir if need be.
112 dot_ssh = os.path.join(pw_dir, '.ssh')
113 if not os.path.isdir(dot_ssh):
114 if not os.path.isdir(pw_dir):
115 logger.verbose('account: WARNING: homedir {} does not exist for {}!'
116 .format(pw_dir, self.name))
118 os.chown(pw_dir, uid, gid)
121 auth_keys = os.path.join(dot_ssh, 'authorized_keys')
122 tools.write_file(auth_keys, lambda f: f.write(new_keys))
124 # set access permissions and ownership properly
125 os.chmod(dot_ssh, 0700)
126 os.chown(dot_ssh, uid, gid)
127 os.chmod(auth_keys, 0600)
128 os.chown(auth_keys, uid, gid)
130 # set self.keys to new_keys only when all of the above ops succeed
133 logger.log('account: {}: installed ssh keys'.format(self.name))
135 def start(self, delay=0):
139 def is_running(self):
141 def needs_reimage(self, target_slicefamily):
142 stampname = "/vservers/{}/etc/slicefamily".format(self.name)
144 with open(stampname) as f:
145 current_slicefamily = f.read().strip()
146 if current_slicefamily != target_slicefamily:
147 logger.info("slice {} : new slice family - would require reimaging"
153 logger.verbose("Missing slicefamily {} - considered OK"
157 ### this used to be a plain method but because it needs to be invoked by destroy
158 # which is a static method, they need to become static as well
159 # needs to be done before sliver starts (checked with vs and lxc)
161 def mount_ssh_dir (slicename): return Account._manage_ssh_dir (slicename, do_mount=True)
163 def umount_ssh_dir (slicename): return Account._manage_ssh_dir (slicename, do_mount=False)
165 # bind mount / umount root side dir to sliver side
167 def _manage_ssh_dir (slicename, do_mount):
168 logger.log ("_manage_ssh_dir, requested to "+("mount" if do_mount else "umount")+" ssh dir for "+ slicename)
170 root_ssh = "/home/{}/.ssh".format(slicename)
171 sliver_ssh = "/vservers/{}/home/{}/.ssh".format(slicename, slicename)
172 def is_mounted (root_ssh):
173 with open('/proc/mounts') as mountsfile:
174 for mount_line in mountsfile.readlines():
175 if mount_line.find (root_ssh) >= 0:
179 # any of both might not exist yet
180 for path in [root_ssh, sliver_ssh]:
181 if not os.path.exists (path):
183 if not os.path.isdir (path):
185 if not is_mounted(root_ssh):
186 command = ['mount', '--bind', '-o', 'ro', root_ssh, sliver_ssh]
187 mounted = logger.log_call (command)
188 msg = "OK" if mounted else "WARNING: FAILED"
189 logger.log("_manage_ssh_dir: mounted {} into slice {} - {}"
190 .format(root_ssh, slicename, msg))
192 if is_mounted (sliver_ssh):
193 command = ['umount', sliver_ssh]
194 umounted = logger.log_call(command)
195 msg = "OK" if umounted else "WARNING: FAILED"
196 logger.log("_manage_ssh_dir: umounted {} - {}"
197 .format(sliver_ssh, msg))
199 logger.log_exc("_manage_ssh_dir failed", name=slicename)
203 def __init__(self, name):
204 self.name = name # username
205 self._acct = None # the account object currently associated with this worker
207 def ensure_created(self, rec):
209 Check account type is still valid. If not, recreate sliver.
210 If still valid, check if running and configure/start if not.
212 logger.log_data_in_file(rec, "/var/lib/nodemanager/{}.rec.txt".format(rec['name']),
213 'raw rec captured in ensure_created', logger.LOG_VERBOSE)
214 curr_class = self._get_class()
215 next_class = type_acct_class[rec['type']]
216 if next_class != curr_class:
217 self._destroy(curr_class)
219 try: next_class.create(self.name, rec)
220 finally: create_sem.release()
221 if not isinstance(self._acct, next_class):
222 self._acct = next_class(rec)
223 logger.verbose("account.Worker.ensure_created: {}, running={}"
224 .format(self.name, self.is_running()))
226 # reservation_alive is set on reservable nodes, and its value is a boolean
227 if 'reservation_alive' in rec:
229 if rec['reservation_alive']:
230 # this sliver has the lease, it is safe to start it
231 if not self.is_running():
233 else: self.configure(rec)
235 # not having the lease, do not start it
237 # usual nodes - preserve old code
238 # xxx it's not clear what to do when a sliver changes type/class
239 # in a reservable node
241 if not self.is_running() or self.needs_reimage(ref['vref']) or next_class != curr_class:
246 def ensure_destroyed(self):
247 self._destroy(self._get_class())
249 # take rec as an arg here for api_calls
250 def start(self, rec, d = 0):
251 self._acct.configure(rec)
252 self._acct.start(delay=d)
254 def configure(self, rec):
255 self._acct.configure(rec)
260 def is_running(self):
261 if self._acct and self._acct.is_running():
264 logger.verbose("Worker.is_running ({}) - no account or not running".format(self.name))
267 def needs_reimage(self, target_slicefamily):
269 logger.verbose("Worker.needs_reimage ({}) - no account".format(self.name))
271 elif self._acct.needs_reimage(target_slicefamily):
272 logger.info("Worker.needs_reimage ({}) - account needs reimage (tmp: DRY RUN)".format(self.name))
277 def _destroy(self, curr_class):
280 destroy_sem.acquire()
281 try: curr_class.destroy(self.name)
282 finally: destroy_sem.release()
284 def _get_class(self):
285 try: shell = pwd.getpwnam(self.name)[6]
286 except KeyError: return None
287 return shell_acct_class[shell]