X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lxcsu;h=f46bcfb8fa201a996b8e3da295e2bbc37582b716;hb=51380fc5bdc153460ea678b67ac96cd85847c8ea;hp=5cba780cf851fb9e45251aca444866a6a0f2d068;hpb=a48641c75412f0bbd8416af26f94ca1b5792294b;p=lxc-userspace.git diff --git a/lxcsu b/lxcsu index 5cba780..f46bcfb 100755 --- a/lxcsu +++ b/lxcsu @@ -12,129 +12,6 @@ drop_capabilities='cap_sys_admin,cap_sys_boot,cap_sys_module' # can set to True here, but also use the -d option debug = False -#################### should go into a separate libvirtsystemd.py -# but we want to keep packaging simple for now - -# reproducing libvirt's systemd-oriented escaping mechanism -# http://code.metager.de/source/xref/lib/virt/src/util/virsystemd.c -# (see original code at the end of this file) - -def virSystemdEscapeName (name): - result='' - def ESCAPE(c,s): - # replace hex's output '0x..' into '\x..' - return s+hex(ord(c)).replace('0','\\',1) - VALID_CHARS = \ - "0123456789" + \ - "abcdefghijklmnopqrstuvwxyz" + \ - "ABCDEFGHIJKLMNOPQRSTUVWXYZ" + \ - ":-_.\\" - for c in name: - if c=='/': - result += '-' - elif c in '-\\' or c not in VALID_CHARS: - result=ESCAPE(c,result) - else: - result += c - return result - -#35static void virSystemdEscapeName(virBufferPtr buf, -#36 const char *name) -#37{ -#38 static const char hextable[16] = "0123456789abcdef"; -#39 -#40#define ESCAPE(c) \ -#41 do { \ -#42 virBufferAddChar(buf, '\\'); \ -#43 virBufferAddChar(buf, 'x'); \ -#44 virBufferAddChar(buf, hextable[(c >> 4) & 15]); \ -#45 virBufferAddChar(buf, hextable[c & 15]); \ -#46 } while (0) -#47 -#48#define VALID_CHARS \ -#49 "0123456789" \ -#50 "abcdefghijklmnopqrstuvwxyz" \ -#51 "ABCDEFGHIJKLMNOPQRSTUVWXYZ" \ -#52 ":-_.\\" -#53 -#54 if (*name == '.') { -#55 ESCAPE(*name); -#56 name++; -#57 } -#58 -#59 while (*name) { -#60 if (*name == '/') -#61 virBufferAddChar(buf, '-'); -#62 else if (*name == '-' || -#63 *name == '\\' || -#64 !strchr(VALID_CHARS, *name)) -#65 ESCAPE(*name); -#66 else -#67 virBufferAddChar(buf, *name); -#68 name++; -#69 } -#70 -#71#undef ESCAPE -#72#undef VALID_CHARS -#73} - -def virSystemdMakeScopeName (name, drivername, partition): - result='' - result += virSystemdEscapeName (partition) - result += '-' - result += virSystemdEscapeName (drivername) - result += '\\x2d' - result += virSystemdEscapeName (name) - result += '.scope' - return result - -#76char *virSystemdMakeScopeName(const char *name, -#77 const char *drivername, -#78 const char *partition) -#79{ -#80 virBuffer buf = VIR_BUFFER_INITIALIZER; -#81 -#82 if (*partition == '/') -#83 partition++; -#84 -#85 virSystemdEscapeName(&buf, partition); -#86 virBufferAddChar(&buf, '-'); -#87 virSystemdEscapeName(&buf, drivername); -#88 virBufferAddLit(&buf, "\\x2d"); -#89 virSystemdEscapeName(&buf, name); -#90 virBufferAddLit(&buf, ".scope"); -#91 -#92 if (virBufferError(&buf)) { -#93 virReportOOMError(); -#94 return NULL; -#95 } -#96 -#97 return virBufferContentAndReset(&buf); -#98} - -### our own additions -import os.path -def find_first_dir (candidates): - for candidate in candidates: - if os.path.isdir(candidate): return candidate - raise Exception,"Cannot find valid dir among\n" + "\n".join([" ->"+c for c in candidates]) - -def find_sysfs_scope (subsystem, slice_name): - subsystem1=subsystem - subsystem2=subsystem - if subsystem=='cpuacct': - subsystem2='cpu,cpuacct' - candidates = [ - # for f16 and our locally brewed libvirt 1.0.4 - "/sys/fs/cgroup/%s/libvirt/lxc/%s"%(subsystem1, slice_name), - # f20 and libvirt 1.1.3 - "/sys/fs/cgroup/%s/machine.slice/%s"%(subsystem2, - virSystemdMakeScopeName(slice_name,'lxc','machine')), - ] - return find_first_dir (candidates) - -#################### end of libvirtsystemd.py - def getarch(f): output = os.popen('readelf -h %s 2>&1'%f).readlines() classlines = [x for x in output if ('Class' in x.split(':')[0])] @@ -170,29 +47,29 @@ def umount(fs_dir, opts=''): def main (): parser = ArgumentParser() parser.add_argument("-n", "--nonet", - action="store_true", dest="netns", default=False, - help="Don't enter network namespace") + action="store_true", dest="no_netns", default=False, + help="Don't enter network namespace") parser.add_argument("-m", "--nomnt", - action="store_true", dest="mntns", default=False, - help="Don't enter mount namespace") + action="store_true", dest="no_mntns", default=False, + help="Don't enter mount namespace") parser.add_argument("-p", "--nopid", - action="store_true", dest="pidns", default=False, - help="Don't enter pid namespace") + action="store_true", dest="no_pidns", default=False, + help="Don't enter pid namespace") parser.add_argument("-r", "--root", - action="store_true", dest="root", default=False, - help="Enter as root: be careful") + action="store_true", dest="root", default=False, + help="Enter as root: be careful") parser.add_argument("-i","--internal", - action="store_true", dest="internal", default=False, - help="does *not* prepend '-- -c' to arguments - or invoke lxcsu-internal") + action="store_true", dest="internal", default=False, + help="does *not* prepend '-- -c' to arguments - or invoke lxcsu-internal") parser.add_argument("-d","--debug", - action='store_true', dest='debug', default=False, - help="debug option") + action='store_true', dest='debug', default=False, + help="debug option") parser.add_argument("-s","--nosliceuid", - action='store_true', dest="nosliceuid", default=False, - help="do not change to slice uid inside of slice") + action='store_true', dest="nosliceuid", default=False, + help="do not change to slice uid inside of slice") parser.add_argument("-o","--noslicehome", - action='store_true', dest="noslicehome", default=False, - help="do not change to slice home directory inside of slice") + action='store_true', dest="noslicehome", default=False, + help="do not change to slice home directory inside of slice") if os.path.exists("/etc/lxcsu_default"): defaults = parser.parse_args(file("/etc/lxcsu_default","r").read().split()) @@ -214,48 +91,54 @@ def main (): # plain lxcsu if not args.internal: - # no command given: enter interactive shell - if not args.command_to_run: args.command_to_run=['/bin/sh'] - args.command_to_run = [ '-c' ] + [" ".join(args.command_to_run)] + # no command given: enter interactive shell + if not args.command_to_run: args.command_to_run=['/bin/sh'] + args.command_to_run = [ '-c' ] + [" ".join(args.command_to_run)] # support for either setting debug at the top of this file, or on the command-line if args.debug: global debug - debug=True + debug=True try: cmd = '/usr/bin/virsh --connect lxc:/// domid %s'%slice_name - pidnum = int(os.popen(cmd).read().rstrip()) + # convert to int as a minimal raincheck + driver_pid = int(os.popen(cmd).read().strip()) + # locate the pid for the - expected - single child, that would be the init for that VM + #init_pid = int(open("/proc/%s/task/%s/children"%(driver_pid,driver_pid)).read().strip()) + init_pid = int(os.popen('pgrep -P %s'%driver_pid).readlines()[0].strip()) + # Thierry: I am changing the code below to use child_pid instead of driver_pid + # for the namespace handling features, that I was able to check + # I've left the other ones as they were, i.e. using driver_pid, but I suspect + # they chould be changed as well + except: print "Domain %s not found"%slice_name - exit(1) - - pid = '%s'%pidnum - if debug: print "Found pidnum",pidnum - cmdline = open('/proc/%s/cmdline'%pidnum).read().rstrip('\n\x00') - arch = getarch('/proc/%s/exe'%pid) + exit(1) - if (not pid): + if not driver_pid or not init_pid: print "Domain %s not started"%slice_name - exit(1) + exit(1) - if arch is None: - arch = 'x86_64' + if debug: print "Found driver_pid",driver_pid,'and init_pid=',init_pid + # xxx probably init_pid here too + arch = getarch('/proc/%s/exe'%driver_pid) # Set sysctls specific to slice sysctls = [] sysctl_dir = '/etc/planetlab/vsys-attributes/%s'%slice_name if (os.access(sysctl_dir,0)): entries = os.listdir(sysctl_dir) - for e in entries: + for e in entries: prefix = 'vsys_sysctl.' - if (e.startswith(prefix)): - sysctl_file = '/'.join([sysctl_dir,e]) - sysctl_name = e[len(prefix):] - sysctl_val = open(sysctl_file).read() - sysctls.append((sysctl_file, sysctl_name, sysctl_val)) - - subdirs = get_cgroup_subdirs_for_pid(pid) + if (e.startswith(prefix)): + sysctl_file = '/'.join([sysctl_dir,e]) + sysctl_name = e[len(prefix):] + sysctl_val = open(sysctl_file).read() + sysctls.append((sysctl_file, sysctl_name, sysctl_val)) + + # xxx probably init_pid here too + subdirs = get_cgroup_subdirs_for_pid(driver_pid) sysfs_root = '/sys/fs/cgroup' # If the slice is frozen, then we'll get an EBUSY when trying to write to the task @@ -297,63 +180,68 @@ def main (): pass else: if debug: print e - print "Error assigning cgroup %s (%s) for slice %s"%(current_cgroup,pid, slice_name) + print "Error assigning cgroup %s (pid=%s) for slice %s"%(current_cgroup,driver_pid, slice_name) exit(1) - setns.chcontext('/proc/%s/ns/uts'%pid) - setns.chcontext('/proc/%s/ns/ipc'%pid) - - if (not args.pidns): - setns.chcontext('/proc/%s/ns/pid'%pid) + def chcontext (path): + retcod = setns.chcontext (path) + if retcod != 0: + print 'WARNING - setns(%s)=>%s (ignored)'%(path,retcod) + return retcod - if (not args.netns): - setns.chcontext('/proc/%s/ns/net'%pid) + # Use init_pid and not driver_pid to locate reference namespaces + ref_ns = "/proc/%s/ns/"%init_pid - if (not args.mntns): - setns.chcontext('/proc/%s/ns/mnt'%pid) + if True: chcontext(ref_ns+'uts') + if True: chcontext(ref_ns+'ipc') + + if (not args.no_pidns): chcontext(ref_ns+'pid') + if (not args.no_netns): chcontext(ref_ns+'net') + if (not args.no_mntns): chcontext(ref_ns+'mnt') proc_mounted = False if (not os.access('/proc/self',0)): - proc_mounted = True - setns.proc_mount() + proc_mounted = True + setns.proc_mount() for (sysctl_file, sysctl_name, sysctl_val) in sysctls: - for fn in ["/sbin/sysctl", "/usr/sbin/sysctl", "/bin/sysctl", "/usr/bin/sysctl"]: - if os.path.exists(fn): - os.system('%s -w %s=%s >/dev/null 2>&1'%(fn, sysctl_name,sysctl_val)) - break - else: - print "Error: image does not have a sysctl binary" + for fn in ["/sbin/sysctl", "/usr/sbin/sysctl", "/bin/sysctl", "/usr/bin/sysctl"]: + if os.path.exists(fn): + os.system('%s -w %s=%s >/dev/null 2>&1'%(fn, sysctl_name,sysctl_val)) + break + else: + print "Error: image does not have a sysctl binary" # cgroups is not yet LXC-safe, so we need to use the coarse grained access control # strategy of unmounting the filesystem umount_result = True for subsystem in ['cpuset','cpu,cpuacct','memory','devices','freezer','net_cls','blkio','perf_event','systemd']: - fs_path = '/sys/fs/cgroup/%s'%subsystem - if (not umount(fs_path,'-l')): - pass - # Leaving these comments for historical reference - #print "Error disabling cgroup access" - #exit(1) - Don't need this because failure here implies failure in the call to umount /sys/fs/cgroup + fs_path = '/sys/fs/cgroup/%s'%subsystem + if (not umount(fs_path,'-l')): + print 'WARNING - umount failed (ignored) with path=',fs_path + pass + # Leaving these comments for historical reference + #print "Error disabling cgroup access" + #exit(1) - Don't need this because failure here implies failure in the call to umount /sys/fs/cgroup if (not umount('/sys/fs/cgroup')): print "Error disabling cgroup access" exit(1) - pid = os.fork() + fork_pid = os.fork() # capsh has a --user option starting with f14 # so if only for f12 we need to fake this one # # capsh.c does essentially the following when invoked with --user: - # pwd = getpwnam(user); ... - # ngroups = MAX_GROUPS; - # status = getgrouplist(user, pwd->pw_gid, groups, &ngroups); ... - # status = setgroups(ngroups, groups); ... - # status = setgid(pwd->pw_gid); ... - # status = setuid(pwd->pw_uid); ... + # pwd = getpwnam(user); ... + # ngroups = MAX_GROUPS; + # status = getgrouplist(user, pwd->pw_gid, groups, &ngroups); ... + # status = setgroups(ngroups, groups); ... + # status = setgid(pwd->pw_gid); ... + # status = setuid(pwd->pw_uid); ... # however we cannot simulate that ourselves because if we did in this process then # capsh could not be allowed to mess with caps any more @@ -364,7 +252,7 @@ def main (): except: return - if (pid == 0): + if (fork_pid == 0): cap_arg = '--drop='+drop_capabilities if (not args.root): @@ -392,8 +280,8 @@ def main (): os.execv('/usr/bin/setarch',exec_args) else: setns.proc_umount() - _,status = os.waitpid(pid,0) + _,status = os.waitpid(fork_pid,0) exit(os.WEXITSTATUS(status)) if __name__ == '__main__': - main() + main()