X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=tools.py;h=02e2ab9b7193b17f8f9aef7e9e6ccd3f9f65215a;hb=9e6b9c1ea9e020c55c85b433bac47231d63e9ffd;hp=49441bda94a42af81f0e922565d97f37dc826911;hpb=2f100cd2a78a0f91022bc4a895007975181cf353;p=nodemanager.git

diff --git a/tools.py b/tools.py
index 49441bd..02e2ab9 100644
--- a/tools.py
+++ b/tools.py
@@ -1,18 +1,45 @@
 """A few things that didn't seem to fit anywhere else."""
 
-import cPickle
-import errno
-import os
+import os, os.path
 import pwd
 import tempfile
-import threading
 import fcntl
+import errno
+import threading
+import subprocess
+import shutil
+import sys
+import signal
 
 import logger
 
+PID_FILE = '/var/run/nodemanager.pid'
+
+####################
+def get_default_if():
+    interface = get_if_from_hwaddr(get_hwaddr_from_plnode())
+    if not interface: interface = "eth0"
+    return interface
 
-PID_FILE = '/var/run/nm.pid'
+def get_hwaddr_from_plnode():
+    try:
+        for line in open("/usr/boot/plnode.txt", 'r').readlines():
+            if line.startswith("NET_DEVICE"):
+                return line.split("=")[1].strip().strip('"')
+    except:
+        pass
+    return None
 
+def get_if_from_hwaddr(hwaddr):
+    import sioc
+    devs = sioc.gifconf()
+    for dev in devs:
+        dev_hwaddr = sioc.gifhwaddr(dev)
+        if dev_hwaddr == hwaddr: return dev
+    return None
+
+####################
+# daemonizing
 def as_daemon_thread(run):
     """Call function <run> with no arguments in its own thread."""
     thr = threading.Thread(target=run)
@@ -33,11 +60,11 @@ def daemon():
     os.setsid()
     if os.fork() != 0: os._exit(0)
     os.chdir('/')
-    os.umask(0)
+    os.umask(0022)
     devnull = os.open(os.devnull, os.O_RDWR)
     os.dup2(devnull, 0)
     # xxx fixme - this is just to make sure that nothing gets stupidly lost - should use devnull
-    crashlog = os.open('/var/log/nm.daemon', os.O_RDWR | os.O_APPEND | os.O_CREAT, 0644)
+    crashlog = os.open('/var/log/nodemanager.daemon', os.O_RDWR | os.O_APPEND | os.O_CREAT, 0644)
     os.dup2(crashlog, 1)
     os.dup2(crashlog, 2)
 
@@ -57,12 +84,16 @@ def fork_as(su, function, *args):
         except:
             os.seteuid(os.getuid())  # undo su so we can write the log file
             os.setegid(os.getgid())
-            logger.log_exc()
+            logger.log_exc("tools: fork_as")
         os._exit(0)
     else: os.waitpid(child_pid, 0)
 
+####################
+# manage files
 def pid_file():
-    """We use a pid file to ensure that only one copy of NM is running at a given time.  If successful, this function will write a pid file containing the pid of the current process.  The return value is the pid of the other running process, or None otherwise."""
+    """We use a pid file to ensure that only one copy of NM is running at a given time.
+If successful, this function will write a pid file containing the pid of the current process.
+The return value is the pid of the other running process, or None otherwise."""
     other_pid = None
     if os.access(PID_FILE, os.F_OK):  # check for a pid file
         handle = open(PID_FILE)  # pid file exists, read it
@@ -80,7 +111,7 @@ def pid_file():
 
 def write_file(filename, do_write, **kw_args):
     """Write file <filename> atomically by opening a temporary file, using <do_write> to write that file, and then renaming the temporary file."""
-    os.rename(write_temp_file(do_write, **kw_args), filename)
+    shutil.move(write_temp_file(do_write, **kw_args), filename)
 
 def write_temp_file(do_write, mode=None, uidgid=None):
     fd, temporary_filename = tempfile.mkstemp()
@@ -91,9 +122,65 @@ def write_temp_file(do_write, mode=None, uidgid=None):
     finally: f.close()
     return temporary_filename
 
+# replace a target file with a new contents - checks for changes
+# can handle chmod if requested
+# can also remove resulting file if contents are void, if requested
+# performs atomically:
+#    writes in a tmp file, which is then renamed (from sliverauth originally)
+# returns True if a change occurred, or the file is deleted
+def replace_file_with_string (target, new_contents, chmod=None, remove_if_empty=False):
+    try:
+        current=file(target).read()
+    except:
+        current=""
+    if current==new_contents:
+        # if turns out to be an empty string, and remove_if_empty is set,
+        # then make sure to trash the file if it exists
+        if remove_if_empty and not new_contents and os.path.isfile(target):
+            logger.verbose("tools.replace_file_with_string: removing file %s"%target)
+            try: os.unlink(target)
+            finally: return True
+        return False
+    # overwrite target file: create a temp in the same directory
+    path=os.path.dirname(target) or '.'
+    fd, name = tempfile.mkstemp('','repl',path)
+    os.write(fd,new_contents)
+    os.close(fd)
+    if os.path.exists(target):
+        os.unlink(target)
+    shutil.move(name,target)
+    if chmod: os.chmod(target,chmod)
+    return True
+
+
+####################
+# utilities functions to get (cached) information from the node
+
+# get node_id from /etc/planetlab/node_id and cache it
+_node_id=None
+def node_id():
+    global _node_id
+    if _node_id is None:
+        try:
+            _node_id=int(file("/etc/planetlab/node_id").read())
+        except:
+            _node_id=""
+    return _node_id
+
+_root_context_arch=None
+def root_context_arch():
+    global _root_context_arch
+    if not _root_context_arch:
+        sp=subprocess.Popen(["uname","-i"],stdout=subprocess.PIPE)
+        (_root_context_arch,_)=sp.communicate()
+        _root_context_arch=_root_context_arch.strip()
+    return _root_context_arch
+
 
+####################
 class NMLock:
     def __init__(self, file):
+        logger.log("tools: Lock %s initialized." % file, 2)
         self.fd = os.open(file, os.O_RDWR|os.O_CREAT, 0600)
         flags = fcntl.fcntl(self.fd, fcntl.F_GETFD)
         flags |= fcntl.FD_CLOEXEC
@@ -101,6 +188,162 @@ class NMLock:
     def __del__(self):
         os.close(self.fd)
     def acquire(self):
-        fcntl.lockf(self.fd, fcntl.LOCK_EX)
+        logger.log("tools: Lock acquired.", 2)
+        fcntl.lockf(self.fd, fcntl.LOCK_SH)
     def release(self):
+        logger.log("tools: Lock released.", 2)
         fcntl.lockf(self.fd, fcntl.LOCK_UN)
+
+####################
+# Utilities for getting the IP address of a LXC/Openvswitch slice. Do this by
+# running ifconfig inside of the slice's context.
+
+def get_sliver_process(slice_name, process_cmdline):
+    """ Utility function to find a process inside of an LXC sliver. Returns
+        (cgroup_fn, pid). cgroup_fn is the filename of the cgroup file for
+        the process, for example /proc/2592/cgroup. Pid is the process id of
+        the process. If the process is not found then (None, None) is returned.
+    """
+    try:
+        cmd = 'grep %s /proc/*/cgroup | grep freezer'%slice_name
+        output = os.popen(cmd).readlines()
+    except:
+        # the slice couldn't be found
+        logger.log("get_sliver_process: couldn't find slice %s" % slice_name)
+        return (None, None)
+
+    cgroup_fn = None
+    pid = None
+    for e in output:
+        try:
+            l = e.rstrip()
+            path = l.split(':')[0]
+            comp = l.rsplit(':')[-1]
+            slice_name_check = comp.rsplit('/')[-1]
+
+            if (slice_name_check == slice_name):
+                slice_path = path
+                pid = slice_path.split('/')[2]
+                cmdline = open('/proc/%s/cmdline'%pid).read().rstrip('\n\x00')
+                if (cmdline == process_cmdline):
+                    cgroup_fn = slice_path
+                    break
+        except:
+            break
+
+    if (not cgroup_fn) or (not pid):
+        logger.log("get_sliver_process: process %s not running in slice %s" % (process_cmdline, slice_name))
+        return (None, None)
+
+    return (cgroup_fn, pid)
+
+def get_sliver_ifconfig(slice_name, device="eth0"):
+    """ return the output of "ifconfig" run from inside the sliver.
+
+        side effects: adds "/usr/sbin" to sys.path
+    """
+
+    # See if setns is installed. If it's not then we're probably not running
+    # LXC.
+    if not os.path.exists("/usr/sbin/setns.so"):
+        return None
+
+    # setns is part of lxcsu and is installed to /usr/sbin
+    if not "/usr/sbin" in sys.path:
+        sys.path.append("/usr/sbin")
+    import setns
+
+    (cgroup_fn, pid) = get_sliver_process(slice_name, "/sbin/init")
+    if (not cgroup_fn) or (not pid):
+        return None
+
+    path = '/proc/%s/ns/net'%pid
+
+    result = None
+    try:
+        setns.chcontext(path)
+
+        args = ["/sbin/ifconfig", device]
+        sub = subprocess.Popen(args, stderr = subprocess.PIPE, stdout = subprocess.PIPE)
+        sub.wait()
+
+        if (sub.returncode != 0):
+            logger.log("get_slice_ifconfig: error in ifconfig: %s" % sub.stderr.read())
+
+        result = sub.stdout.read()
+    finally:
+        setns.chcontext("/proc/1/ns/net")
+
+    return result
+
+def get_sliver_ip(slice_name):
+    ifconfig = get_sliver_ifconfig(slice_name)
+    if not ifconfig:
+        return None
+
+    for line in ifconfig.split("\n"):
+        if "inet addr:" in line:
+            # example: '          inet addr:192.168.122.189  Bcast:192.168.122.255  Mask:255.255.255.0'
+            parts = line.strip().split()
+            if len(parts)>=2 and parts[1].startswith("addr:"):
+                return parts[1].split(":")[1]
+
+    return None
+
+### this returns the kind of virtualization on the node
+# either 'vs' or 'lxc'
+# also caches it in /etc/planetlab/virt for next calls
+# could be promoted to core nm if need be
+virt_stamp="/etc/planetlab/virt"
+def get_node_virt ():
+    try:
+        return file(virt_stamp).read().strip()
+    except:
+        pass
+    logger.log("Computing virt..")
+    try: 
+        if subprocess.call ([ 'vserver', '--help' ]) ==0: virt='vs'
+        else:                                             virt='lxc'      
+    except:
+        virt='lxc'
+    with file(virt_stamp,"w") as f:
+        f.write(virt)
+    return virt
+
+### this return True or False to indicate that systemctl is present on that box
+# cache result in memory as _has_systemctl
+_has_systemctl=None
+def has_systemctl ():
+    global _has_systemctl
+    if _has_systemctl is None:
+        _has_systemctl = (subprocess.call([ 'systemctl', '--help' ]) == 0)
+    return _has_systemctl
+
+# how to run a command in a slice
+# now this is a painful matter
+# the problem is with capsh that forces a bash command to be injected in its exec'ed command
+# so because lxcsu uses capsh, you cannot exec anything else than bash
+# bottom line is, what actually needs to be called is
+# vs:  vserver exec slicename command and its arguments
+# lxc: lxcsu slicename "command and its arguments"
+# which, OK, is no big deal as long as the command is simple enough, 
+# but do not stretch it with arguments that have spaces or need quoting as that will become a nightmare
+def command_in_slice (slicename, argv):
+    virt=get_node_virt()
+    if virt=='vs':
+        return [ 'vserver', slicename, 'exec', ] + argv
+    elif virt=='lxc':
+        # wrap up argv in a single string for -c
+        return [ 'lxcsu', slicename, ] + [ " ".join(argv) ]
+    logger.log("command_in_slice: WARNING: could not find a valid virt")
+    return argv
+
+####################
+def init_signals ():
+    def handler (signum, frame):
+        logger.log("Received signal %d - exiting"%signum)
+        os._exit(1)
+    signal.signal(signal.SIGHUP,handler)
+    signal.signal(signal.SIGQUIT,handler)
+    signal.signal(signal.SIGINT,handler)
+    signal.signal(signal.SIGTERM,handler)