X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=python%2Fvserver.py;h=aeb05d6b596b28ffdee2b83bd3053856a5161e43;hb=c8a6d235b30fa22308f7a857ba7aaea3f15ef3cc;hp=2d72bd030e4fed57cb9c3bb2c32538d16619a6c1;hpb=173c27d970bd29594d0485f66474af5f5be37201;p=util-vserver.git diff --git a/python/vserver.py b/python/vserver.py index 2d72bd0..aeb05d6 100644 --- a/python/vserver.py +++ b/python/vserver.py @@ -4,31 +4,35 @@ import errno import fcntl import os import re +import pwd +import signal import sys import time import traceback import mountimpl -import linuxcaps -import passfdimpl +import runcmd import utmp import vserverimpl, vduimpl - -from util_vserver_vars import * - -CAP_SAFE = (linuxcaps.CAP_CHOWN | - linuxcaps.CAP_DAC_OVERRIDE | - linuxcaps.CAP_DAC_READ_SEARCH | - linuxcaps.CAP_FOWNER | - linuxcaps.CAP_FSETID | - linuxcaps.CAP_KILL | - linuxcaps.CAP_SETGID | - linuxcaps.CAP_SETUID | - linuxcaps.CAP_SETPCAP | - linuxcaps.CAP_SYS_TTY_CONFIG | - linuxcaps.CAP_LEASE | - linuxcaps.CAP_SYS_CHROOT | - linuxcaps.CAP_SYS_PTRACE) +import cpulimit, bwlimit + +from vserverimpl import VS_SCHED_CPU_GUARANTEED as SCHED_CPU_GUARANTEED +from vserverimpl import DLIMIT_INF +from vserverimpl import VC_LIM_KEEP + +from vserverimpl import RLIMIT_CPU +from vserverimpl import RLIMIT_RSS +from vserverimpl import RLIMIT_NPROC +from vserverimpl import RLIMIT_NOFILE +from vserverimpl import RLIMIT_MEMLOCK +from vserverimpl import RLIMIT_AS +from vserverimpl import RLIMIT_LOCKS +from vserverimpl import RLIMIT_SIGPENDING +from vserverimpl import RLIMIT_MSGQUEUE +from vserverimpl import VLIMIT_NSOCK +from vserverimpl import VLIMIT_OPENFD +from vserverimpl import VLIMIT_ANON +from vserverimpl import VLIMIT_SHMEM # # these are the flags taken from the kernel linux/vserver/legacy.h @@ -42,27 +46,106 @@ FLAGS_HIDEINFO = 32 FLAGS_ULIMIT = 64 FLAGS_NAMESPACE = 128 +RLIMITS = {"CPU": RLIMIT_CPU, + "RSS": RLIMIT_RSS, + "NPROC": RLIMIT_NPROC, + "NOFILE": RLIMIT_NOFILE, + "MEMLOCK": RLIMIT_MEMLOCK, + "AS": RLIMIT_AS, + "LOCKS": RLIMIT_LOCKS, + "SIGPENDING": RLIMIT_SIGPENDING, + "MSGQUEUE": RLIMIT_MSGQUEUE, + "NSOCK": VLIMIT_NSOCK, + "OPENFD": VLIMIT_OPENFD, + "ANON": VLIMIT_ANON, + "SHMEM": VLIMIT_SHMEM} + +class NoSuchVServer(Exception): pass + - class VServer: INITSCRIPTS = [('/etc/rc.vinit', 'start'), ('/etc/rc.d/rc', '%(runlevel)d')] - def __init__(self, name): + def __init__(self, name, vm_id = None, vm_running = False): self.name = name - self.config = self.__read_config_file("/etc/vservers.conf") - self.config.update(self.__read_config_file("/etc/vservers/%s.conf" % - self.name)) - self.flags = 0 - flags = self.config["S_FLAGS"].split(" ") - if "lock" in flags: - self.flags |= FLAGS_LOCK - if "nproc" in flags: - self.flags |= FLAGS_NPROC - self.remove_caps = ~CAP_SAFE - self.ctx = int(self.config["S_CONTEXT"]) + self.rlimits_changed = False + self.config_file = "/etc/vservers/%s.conf" % name + self.dir = "%s/%s" % (vserverimpl.VSERVER_BASEDIR, name) + if not (os.path.isdir(self.dir) and + os.access(self.dir, os.R_OK | os.W_OK | os.X_OK)): + raise NoSuchVServer, "no such vserver: " + name + self.config = {} + for config_file in ["/etc/vservers.conf", self.config_file]: + try: + self.config.update(self.__read_config_file(config_file)) + except IOError, ex: + if ex.errno != errno.ENOENT: + raise + self.remove_caps = ~vserverimpl.CAP_SAFE; + if vm_id == None: + vm_id = int(self.config['S_CONTEXT']) + self.ctx = vm_id + self.vm_running = vm_running + + def have_limits_changed(self): + return self.rlimits_changed + + def set_rlimit_limit(self,type,hard,soft,minimum): + """Generic set resource limit function for vserver""" + global RLIMITS + changed = False + try: + old_hard, old_soft, old_minimum = self.get_rlimit_limit(type) + if old_hard != VC_LIM_KEEP and old_hard <> hard: changed = True + if old_soft != VC_LIM_KEEP and old_soft <> soft: changed = True + if old_minimum != VC_LIM_KEEP and old_minimum <> minimum: changed = True + self.rlimits_changed = self.rlimits_changed or changed + except OSError, e: + if self.is_running(): print "Unexpected error with getrlimit for running context %d" % self.ctx + + resource_type = RLIMITS[type] + try: + ret = vserverimpl.setrlimit(self.ctx,resource_type,hard,soft,minimum) + except OSError, e: + if self.is_running(): print "Unexpected error with setrlimit for running context %d" % self.ctx + + def set_rlimit_config(self,type,hard,soft,minimum): + """Generic set resource limit function for vserver""" + resources = {} + if hard <> VC_LIM_KEEP: + resources["VS_%s_HARD"%type] = hard + if soft <> VC_LIM_KEEP: + resources["VS_%s_SOFT"%type] = soft + if minimum <> VC_LIM_KEEP: + resources["VS_%s_MINIMUM"%type] = minimum + if len(resources)>0: + self.update_resources(resources) + self.set_rlimit_limit(type,hard,soft,minimum) + + def get_rlimit_limit(self,type): + """Generic get resource configuration function for vserver""" + global RLIMITS + resource_type = RLIMITS[type] + try: + ret = vserverimpl.getrlimit(self.ctx,resource_type) + except OSError, e: + print "Unexpected error with getrlimit for context %d" % self.ctx + ret = self.get_rlimit_config(type) + return ret + + def get_rlimit_config(self,type): + """Generic get resource configuration function for vserver""" + hard = int(self.config.get("VS_%s_HARD"%type,VC_LIM_KEEP)) + soft = int(self.config.get("VS_%s_SOFT"%type,VC_LIM_KEEP)) + minimum = int(self.config.get("VS_%s_MINIMUM"%type,VC_LIM_KEEP)) + return (hard,soft,minimum) + + def set_WHITELISTED_config(self,whitelisted): + resources = {'VS_WHITELISTED': whitelisted} + self.update_resources(resources) config_var_re = re.compile(r"^ *([A-Z_]+)=(.*)\n?$", re.MULTILINE) @@ -77,93 +160,163 @@ class VServer: config[key] = val.strip('"') return config - def __do_chroot(self): + def __update_config_file(self, filename, newvars): + + # read old file, apply changes + f = open(filename, "r") + data = f.read() + f.close() + todo = newvars.copy() + changed = False + offset = 0 + for m in self.config_var_re.finditer(data): + (key, val) = m.groups() + newval = todo.pop(key, None) + if newval != None: + data = data[:offset+m.start(2)] + str(newval) + data[offset+m.end(2):] + offset += len(str(newval)) - (m.end(2)-m.start(2)) + changed = True + for (newkey, newval) in todo.items(): + data += "%s=%s\n" % (newkey, newval) + changed = True + + if not changed: + return - return os.chroot("%s/%s" % (VROOTDIR, self.name)) + # write new file + newfile = filename + ".new" + f = open(newfile, "w") + f.write(data) + f.close() - def set_dlimit(self, blocktotal): - path = "%s/%s" % (VROOTDIR, self.name) - inodes, blockcount, size = vduimpl.vdu(path) - vserverimpl.setdlimit(path, self.ctx, blockcount>>1, blocktotal, inodes, -1, 2) + # replace old file with new + os.rename(newfile, filename) - def get_dlimit(self): - path = "%s/%s" % (VROOTDIR, 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 + def __do_chroot(self): - return blocktotal + os.chroot(self.dir) + os.chdir("/") - def open(self, filename, mode = "r", bufsize = -1): + def chroot_call(self, fn, *args): - (sendsock, recvsock) = passfdimpl.socketpair() - child_pid = os.fork() - if child_pid == 0: + cwd_fd = os.open(".", os.O_RDONLY) + try: + root_fd = os.open("/", os.O_RDONLY) try: - # child process self.__do_chroot() - f = open(filename, mode) - passfdimpl.sendmsg(f.fileno(), sendsock) - os._exit(0) - except EnvironmentError, ex: - (result, errmsg) = (ex.errno, ex.strerror) - except Exception, ex: - (result, errmsg) = (255, str(ex)) - os.write(sendsock, errmsg) - os._exit(result) + result = fn(*args) + finally: + os.fchdir(root_fd) + os.chroot(".") + os.fchdir(cwd_fd) + os.close(root_fd) + finally: + os.close(cwd_fd) + return result + + def set_disklimit(self, block_limit): + # block_limit is in kB + if block_limit == 0: + try: + vserverimpl.unsetdlimit(self.dir, self.ctx) + except OSError, e: + print "Unexpected error with unsetdlimit for context %d" % self.ctx + return - # parent process + if self.vm_running: + block_usage = vserverimpl.DLIMIT_KEEP + inode_usage = vserverimpl.DLIMIT_KEEP + else: + # init_disk_info() must have been called to get usage values + block_usage = self.disk_blocks + inode_usage = self.disk_inodes - # XXX - need this since a lambda can't raise an exception - def __throw(ex): - raise ex - os.close(sendsock) - throw = lambda : __throw(Exception(errmsg)) - while True: - try: - (pid, status) = os.waitpid(child_pid, 0) - if os.WIFEXITED(status): - result = os.WEXITSTATUS(status) - if result != 255: - errmsg = os.strerror(result) - throw = lambda : __throw(IOError(result, errmsg)) - else: - errmsg = "unexpected exception in child" - else: - result = -1 - errmsg = "child killed" - break - except OSError, ex: - if ex.errno != errno.EINTR: - os.close(recvsock) - raise ex - fcntl.fcntl(recvsock, fcntl.F_SETFL, os.O_NONBLOCK) try: - (fd, errmsg) = passfdimpl.recvmsg(recvsock) + vserverimpl.setdlimit(self.dir, + self.ctx, + block_usage, + block_limit, + inode_usage, + vserverimpl.DLIMIT_INF, # inode limit + 2) # %age reserved for root + except OSError, e: + print "Unexpected error with setdlimit for context %d" % self.ctx + + + resources = {'VS_DISK_MAX': block_limit} + self.update_resources(resources) + + def is_running(self): + return vserverimpl.isrunning(self.ctx) + + def get_disklimit(self): + + try: + (self.disk_blocks, block_limit, self.disk_inodes, inode_limit, + reserved) = vserverimpl.getdlimit(self.dir, self.ctx) except OSError, ex: - if ex.errno != errno.EAGAIN: - throw = lambda : __throw(ex) - fd = 0 - os.close(recvsock) - if not fd: - throw() + if ex.errno != errno.ESRCH: + raise + # get here if no vserver disk limit has been set for xid + block_limit = -1 + + return block_limit - return os.fdopen(fd, mode, bufsize) + def set_sched_config(self, cpu_share, sched_flags): - def __do_chcontext(self, state_file = None): + """ Write current CPU scheduler parameters to the vserver + configuration file. This method does not modify the kernel CPU + scheduling parameters for this context. """ - vserverimpl.chcontext(self.ctx, self.remove_caps) - if not state_file: + if cpu_share == int(self.config.get("CPULIMIT", -1)): return - print >>state_file, "S_CONTEXT=%d" % self.ctx - print >>state_file, "S_PROFILE=%s" % self.config.get("S_PROFILE", "") - state_file.close() + cpu_guaranteed = sched_flags & SCHED_CPU_GUARANTEED + cpu_config = { "CPULIMIT": cpu_share, "CPUGUARANTEED": cpu_guaranteed } + self.update_resources(cpu_config) + if self.vm_running: + self.set_sched(cpu_share, sched_flags) + + def set_sched(self, cpu_share, sched_flags = 0): + """ Update kernel CPU scheduling parameters for this context. """ + vserverimpl.setsched(self.ctx, cpu_share, sched_flags) + + def get_sched(self): + # have no way of querying scheduler right now on a per vserver basis + return (-1, False) + + def set_bwlimit(self, minrate = bwlimit.bwmin, maxrate = None, + exempt_min = None, exempt_max = None, + share = None, dev = "eth0"): + + if minrate is None: + bwlimit.off(self.ctx, dev) + else: + bwlimit.on(self.ctx, dev, share, + minrate, maxrate, exempt_min, exempt_max) + + def get_bwlimit(self, dev = "eth0"): + + result = bwlimit.get(self.ctx) + # result of bwlimit.get is (ctx, share, minrate, maxrate) + if result: + result = result[1:] + return result + + def open(self, filename, mode = "r", bufsize = -1): + + return self.chroot_call(open, filename, mode, bufsize) + + def __do_chcontext(self, state_file): + + if state_file: + print >>state_file, "S_CONTEXT=%u" % self.ctx + print >>state_file, "S_PROFILE=" + state_file.close() + + if vserverimpl.chcontext(self.ctx): + self.set_resources() + vserverimpl.setup_done(self.ctx) def __prep(self, runlevel, log): @@ -212,6 +365,8 @@ class VServer: self.__do_chcontext(state_file) def start(self, wait, runlevel = 3): + self.vm_running = True + self.rlimits_changed = False child_pid = os.fork() if child_pid == 0: @@ -239,38 +394,18 @@ class VServer: # execute each init script in turn # 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 + self.__do_chcontext(state_file) + for cmd in self.INITSCRIPTS + [None]: + try: + # enter vserver context + arg_subst = { 'runlevel': runlevel } + cmd_args = [cmd[0]] + map(lambda x: x % arg_subst, + cmd[1:]) + print >>log, "executing '%s'" % " ".join(cmd_args) + os.spawnvp(os.P_WAIT,cmd[0],*cmd_args) + except: + traceback.print_exc() + os._exit(1) # we get here due to an exception in the top-level child process except Exception, ex: @@ -279,3 +414,40 @@ class VServer: # parent process return child_pid + + def set_resources(self): + + """ Called when vserver context is entered for first time, + should be overridden by subclass. """ + + pass + + def update_resources(self, resources): + + self.config.update(resources) + + # write new values to configuration file + self.__update_config_file(self.config_file, resources) + + def init_disk_info(self): + + (self.disk_inodes, self.disk_blocks, size) = vduimpl.vdu(self.dir) + + return size + + def stop(self, signal = signal.SIGKILL): + vserverimpl.killall(self.ctx, signal) + self.vm_running = False + self.rlimits_changed = False + + + +def create(vm_name, static = False, ctor = VServer): + + options = [] + if static: + options += ['--static'] + runcmd.run('vuseradd', options + [vm_name]) + vm_id = pwd.getpwnam(vm_name)[2] + + return ctor(vm_name, vm_id)