import fcntl
import os
import re
+import sys
+import time
import traceback
+import mountimpl
import linuxcaps
import passfdimpl
-import vserverimpl
+import utmp
+import vserverimpl, vduimpl
+import cpulimit, bwlimit
from util_vserver_vars import *
FLAGS_ULIMIT = 64
FLAGS_NAMESPACE = 128
+# default values for new vserver scheduler
+SCHED_TOKENS_MIN = 50
+SCHED_TOKENS_MAX = 100
class VServer:
- INITSCRIPTS = [('/etc/rc.vinit', 'start'), ('/etc/rc.d/rc', '3')]
+ INITSCRIPTS = [('/etc/rc.vinit', 'start'),
+ ('/etc/rc.d/rc', '%(runlevel)d')]
def __init__(self, name):
def __do_chroot(self):
- return os.chroot("%s/%s" % (VROOTDIR, self.name))
+ return os.chroot("%s/%s" % (DEFAULT_VSERVERDIR, self.name))
- def open(self, filename, mode = "r"):
+ def set_disklimit(self, blocktotal):
+ path = "%s/%s" % (DEFAULT_VSERVERDIR, self.name)
+ inodes, blockcount, size = vduimpl.vdu(path)
+ blockcount = blockcount >> 1
+
+ if blocktotal > blockcount:
+ vserverimpl.setdlimit(path, self.ctx, blockcount>>1, \
+ blocktotal, inodes, -1, 2)
+ else:
+ # should raise some error value
+ print "block limit (%d) ignored for vserver %s" %(blocktotal,self.name)
+
+ def get_disklimit(self):
+ path = "%s/%s" % (DEFAULT_VSERVERDIR, self.name)
+ try:
+ blocksused, blocktotal, inodesused, inodestotal, reserved = \
+ vserverimpl.getdlimit(path,self.ctx)
+ except OSError, ex:
+ if ex.errno == 3:
+ # get here if no vserver disk limit has been set for xid
+ # set blockused to -1 to indicate no limit
+ blocktotal = -1
+
+ return blocktotal
+
+ def set_sched(self, shares, besteffort = True):
+ # for the old CKRM scheduler
+ if cpulimit.checkckrm() is True:
+ cpulimit.cpuinit()
+ cpulimit.vs2ckrm_on(self.name)
+ try:
+ cpulimit.cpulimit(self.name,shares)
+ except OSError, ex:
+ if ex.errno == 22:
+ print "invalid shares argument"
+ # should re-raise exception?!
+
+ # for the new vserver scheduler
+ else:
+ global SCHED_TOKENS_MIN, SCHED_TOKENS_MAX
+ tokensmin = SCHED_TOKENS_MIN
+ tokensmax = SCHED_TOKENS_MAX
+
+ if besteffort is True:
+ # magic "interval" value for Andy's scheduler to denote besteffort
+ interval = 1000
+ fillrate = shares
+ else:
+ interval = 1001
+ fillrate = shares
+
+ try:
+ cpuguaranteed = 0 # need to set this from the conf file
+ vserverimpl.setsched(self.ctx,fillrate,interval,tokensmin,tokensmax,cpuguaranteed)
+ except OSError, ex:
+ if ex.errno == 22:
+ print "kernel does not support vserver scheduler"
+ else:
+ raise ex
+
+ def get_sched(self):
+ # have no way of querying scheduler right now on a per vserver basis
+ return (-1, False)
+
+ def set_memlimit(self, limit):
+ ret = vserverimpl.setrlimit(self.ctx,5,limit)
+ return ret
+
+ def get_memlimit(self):
+ ret = vserverimpl.getrlimit(self.ctx,5)
+ return ret
+
+ def set_tasklimit(self, limit):
+ ret = vserverimpl.setrlimit(self.ctx,6,limit)
+ return ret
+
+ def get_tasklimit(self):
+ ret = vserverimpl.getrlimit(self.ctx,6)
+ return ret
+
+ def set_bwlimit(self, eth, limit, cap, minrate, maxrate):
+ if cap == "-1":
+ bwlimit.off(self.ctx,eth)
+ else:
+ bwlimit.on(self.ctx, eth, limit, cap, minrate, maxrate)
+
+ def get_bwlimit(self, eth):
+ # not implemented yet
+ bwlimit = -1
+ cap = "unknown"
+ minrate = "unknown"
+ maxrate = "unknown"
+ return (bwlimit, cap, minrate, maxrate)
+
+ def open(self, filename, mode = "r", bufsize = -1):
(sendsock, recvsock) = passfdimpl.socketpair()
child_pid = os.fork()
if not fd:
throw()
- return os.fdopen(fd)
+ return os.fdopen(fd, mode, bufsize)
- def enter(self):
+ def __do_chcontext(self, state_file = None):
- state_file = open("/var/run/vservers/%s.ctx" % self.name, "w")
- self.__do_chroot()
- vserverimpl.chcontext(self.ctx, self.remove_caps)
+ vserverimpl.chcontext(self.ctx)
+ if not state_file:
+ return
print >>state_file, "S_CONTEXT=%d" % self.ctx
print >>state_file, "S_PROFILE=%s" % self.config.get("S_PROFILE", "")
state_file.close()
- def start(self):
+ def __prep(self, runlevel, log):
+
+ """ Perform all the crap that the vserver script does before
+ actually executing the startup scripts. """
+
+ # remove /var/run and /var/lock/subsys files
+ # but don't remove utmp from the top-level /var/run
+ RUNDIR = "/var/run"
+ LOCKDIR = "/var/lock/subsys"
+ filter_fn = lambda fs: filter(lambda f: f != 'utmp', fs)
+ garbage = reduce((lambda (out, ff), (dir, subdirs, files):
+ (out + map((dir + "/").__add__, ff(files)),
+ lambda fs: fs)),
+ list(os.walk(RUNDIR)),
+ ([], filter_fn))[0]
+ garbage += filter(os.path.isfile, map((LOCKDIR + "/").__add__,
+ os.listdir(LOCKDIR)))
+ for f in garbage:
+ os.unlink(f)
+
+ # set the initial runlevel
+ f = open(RUNDIR + "/utmp", "w")
+ utmp.set_runlevel(f, runlevel)
+ f.close()
+
+ # mount /proc and /dev/pts
+ self.__do_mount("none", "/proc", "proc")
+ # XXX - magic mount options
+ self.__do_mount("none", "/dev/pts", "devpts", 0, "gid=5,mode=0620")
+
+ def __do_mount(self, *mount_args):
+
+ try:
+ mountimpl.mount(*mount_args)
+ except OSError, ex:
+ if ex.errno == errno.EBUSY:
+ # assume already mounted
+ return
+ raise ex
+
+ def enter(self):
+
+ state_file = open("/var/run/vservers/%s.ctx" % self.name, "w")
+ self.__do_chroot()
+ self.__do_chcontext(state_file)
+
+ def start(self, wait, runlevel = 3):
child_pid = os.fork()
if child_pid == 0:
# get a new session
os.setsid()
- # enter vserver context
- self.enter()
+ # open state file to record vserver info
+ state_file = open("/var/run/vservers/%s.ctx" % self.name, "w")
# use /dev/null for stdin, /var/log/boot.log for stdout/err
os.close(0)
os.close(1)
os.open("/dev/null", os.O_RDONLY)
- os.open("/var/log/boot.log",
- os.O_WRONLY | os.O_CREAT | os.O_TRUNC)
+ self.__do_chroot()
+ log = open("/var/log/boot.log", "w", 0)
os.dup2(1, 2)
- # write same output that vserver script does
- os.write(1, "Starting the virtual server %s\n" % self.name)
- os.write(1, "Server %s is not running\n" % self.name)
+ print >>log, ("%s: starting the virtual server %s" %
+ (time.asctime(time.gmtime()), self.name))
+
+ # perform pre-init cleanup
+ self.__prep(runlevel, log)
# execute each init script in turn
- # XXX - we don't support all the possible scripts
- for cmd in self.INITSCRIPTS:
- os.spawnl(os.P_WAIT, *cmd)
+ # XXX - we don't support all scripts that vserver script does
+ cmd_pid = 0
+ for cmd in self.INITSCRIPTS + [None]:
+ # wait for previous command to terminate, unless it
+ # is the last one and the caller has specified to wait
+ if cmd_pid and (cmd != None or wait):
+ try:
+ os.waitpid(cmd_pid, 0)
+ except:
+ print >>log, "error waiting for %s:" % cmd_pid
+ traceback.print_exc()
+
+ # end of list
+ if cmd == None:
+ os._exit(0)
+
+ # fork and exec next command
+ cmd_pid = os.fork()
+ if cmd_pid == 0:
+ try:
+ # enter vserver context
+ self.__do_chcontext(state_file)
+ arg_subst = { 'runlevel': runlevel }
+ cmd_args = [cmd[0]] + map(lambda x: x % arg_subst,
+ cmd[1:])
+ print >>log, "executing '%s'" % " ".join(cmd_args)
+ os.execl(cmd[0], *cmd_args)
+ except:
+ traceback.print_exc()
+ os._exit(1)
+ else:
+ # don't want to write state_file multiple times
+ state_file = None
+
+ # we get here due to an exception in the top-level child process
except Exception, ex:
traceback.print_exc()
os._exit(0)