source: trunk/packages/sipb-xen-www/code/main.py @ 642

#!/usr/bin/python
"""Main CGI script for web interface"""

import base64
import cPickle
import cgi
import datetime
import hmac
import sha
import simplejson
import sys
import time
import urllib
import random
from StringIO import StringIO

def revertStandardError():
    """Move stderr to stdout, and return the contents of the old stderr."""
    errio = sys.stderr
    if not isinstance(errio, StringIO):
        return ''
    sys.stderr = sys.stdout
    errio.seek(0)
    return errio.read()

def printError():
    """Revert stderr to stdout, and print the contents of stderr"""
    if isinstance(sys.stderr, StringIO):
        print revertStandardError()

if __name__ == '__main__':
    import atexit
    atexit.register(printError)

import templates
from Cheetah.Template import Template
import sipb_xen_database
from sipb_xen_database import Machine, CDROM, ctx, connect, MachineAccess, Type, Autoinstall
import validation
import cache_acls
from webcommon import InvalidInput, CodeError, State
import controls
from getafsgroups import getAfsGroupMembers

def pathSplit(path):
    if path.startswith('/'):
        path = path[1:]
    i = path.find('/')
    if i == -1:
        i = len(path)
    return path[:i], path[i:]

class Checkpoint:
    def __init__(self):
        self.start_time = time.time()
        self.checkpoints = []

    def checkpoint(self, s):
        self.checkpoints.append((s, time.time()))

    def __str__(self):
        return ('Timing info:\n%s\n' %
                '\n'.join(['%s: %s' % (d, t - self.start_time) for
                           (d, t) in self.checkpoints]))

checkpoint = Checkpoint()

def jquote(string):
    return "'" + string.replace('\\', '\\\\').replace("'", "\\'").replace('\n', '\\n') + "'"

def helppopup(subj):
    """Return HTML code for a (?) link to a specified help topic"""
    return ('<span class="helplink"><a href="help?' +
            cgi.escape(urllib.urlencode(dict(subject=subj, simple='true')))
            +'" target="_blank" ' +
            'onclick="return helppopup(' + cgi.escape(jquote(subj)) + ')">(?)</a></span>')

def makeErrorPre(old, addition):
    if addition is None:
        return
    if old:
        return old[:-6]  + '\n----\n' + str(addition) + '</pre>'
    else:
        return '<p>STDERR:</p><pre>' + str(addition) + '</pre>'

Template.sipb_xen_database = sipb_xen_database
Template.helppopup = staticmethod(helppopup)
Template.err = None

class JsonDict:
    """Class to store a dictionary that will be converted to JSON"""
    def __init__(self, **kws):
        self.data = kws
        if 'err' in kws:
            err = kws['err']
            del kws['err']
            self.addError(err)

    def __str__(self):
        return simplejson.dumps(self.data)

    def addError(self, text):
        """Add stderr text to be displayed on the website."""
        self.data['err'] = \
            makeErrorPre(self.data.get('err'), text)

class Defaults:
    """Class to store default values for fields."""
    memory = 256
    disk = 4.0
    cdrom = ''
    autoinstall = ''
    name = ''
    description = ''
    type = 'linux-hvm'

    def __init__(self, max_memory=None, max_disk=None, **kws):
        if max_memory is not None:
            self.memory = min(self.memory, max_memory)
        if max_disk is not None:
            self.max_disk = min(self.disk, max_disk)
        for key in kws:
            setattr(self, key, kws[key])



DEFAULT_HEADERS = {'Content-Type': 'text/html'}

def invalidInput(op, username, fields, err, emsg):
    """Print an error page when an InvalidInput exception occurs"""
    d = dict(op=op, user=username, err_field=err.err_field,
             err_value=str(err.err_value), stderr=emsg,
             errorMessage=str(err))
    return templates.invalid(searchList=[d])

def hasVnc(status):
    """Does the machine with a given status list support VNC?"""
    if status is None:
        return False
    for l in status:
        if l[0] == 'device' and l[1][0] == 'vfb':
            d = dict(l[1][1:])
            return 'location' in d
    return False

def parseCreate(username, state, fields):
    kws = dict([(kw, fields.getfirst(kw)) for kw in 'name description owner memory disksize vmtype cdrom autoinstall'.split()])
    validate = validation.Validate(username, state, strict=True, **kws)
    return dict(contact=username, name=validate.name, description=validate.description, memory=validate.memory,
                disksize=validate.disksize, owner=validate.owner, machine_type=validate.vmtype,
                cdrom=getattr(validate, 'cdrom', None),
                autoinstall=getattr(validate, 'autoinstall', None))

def create(username, state, path, fields):
    """Handler for create requests."""
    try:
        parsed_fields = parseCreate(username, state, fields)
        machine = controls.createVm(username, state, **parsed_fields)
    except InvalidInput, err:
        pass
    else:
        err = None
    state.clear() #Changed global state
    d = getListDict(username, state)
    d['err'] = err
    if err:
        for field in fields.keys():
            setattr(d['defaults'], field, fields.getfirst(field))
    else:
        d['new_machine'] = parsed_fields['name']
    return templates.list(searchList=[d])


def getListDict(username, state):
    """Gets the list of local variables used by list.tmpl."""
    checkpoint.checkpoint('Starting')
    machines = state.machines
    checkpoint.checkpoint('Got my machines')
    on = {}
    has_vnc = {}
    xmlist = state.xmlist
    checkpoint.checkpoint('Got uptimes')
    can_clone = 'ice3' not in state.xmlist_raw
    for m in machines:
        if m not in xmlist:
            has_vnc[m] = 'Off'
            m.uptime = None
        else:
            m.uptime = xmlist[m]['uptime']
            if xmlist[m]['console']:
                has_vnc[m] = True
            elif m.type.hvm:
                has_vnc[m] = "WTF?"
            else:
                has_vnc[m] = "ParaVM"+helppopup("ParaVM Console")
    max_memory = validation.maxMemory(username, state)
    max_disk = validation.maxDisk(username)
    checkpoint.checkpoint('Got max mem/disk')
    defaults = Defaults(max_memory=max_memory,
                        max_disk=max_disk,
                        owner=username,
                        cdrom='gutsy-i386')
    checkpoint.checkpoint('Got defaults')
    def sortkey(machine):
        return (machine.owner != username, machine.owner, machine.name)
    machines = sorted(machines, key=sortkey)
    d = dict(user=username,
             cant_add_vm=validation.cantAddVm(username, state),
             max_memory=max_memory,
             max_disk=max_disk,
             defaults=defaults,
             machines=machines,
             has_vnc=has_vnc,
             can_clone=can_clone)
    return d

def listVms(username, state, path, fields):
    """Handler for list requests."""
    checkpoint.checkpoint('Getting list dict')
    d = getListDict(username, state)
    checkpoint.checkpoint('Got list dict')
    return templates.list(searchList=[d])

def vnc(username, state, path, fields):
    """VNC applet page.

    Note that due to same-domain restrictions, the applet connects to
    the webserver, which needs to forward those requests to the xen
    server.  The Xen server runs another proxy that (1) authenticates
    and (2) finds the correct port for the VM.

    You might want iptables like:

    -t nat -A PREROUTING -s ! 18.181.0.60 -i eth1 -p tcp -m tcp \
      --dport 10003 -j DNAT --to-destination 18.181.0.60:10003
    -t nat -A POSTROUTING -d 18.181.0.60 -o eth1 -p tcp -m tcp \
      --dport 10003 -j SNAT --to-source 18.187.7.142
    -A FORWARD -d 18.181.0.60 -i eth1 -o eth1 -p tcp -m tcp \
      --dport 10003 -j ACCEPT

    Remember to enable iptables!
    echo 1 > /proc/sys/net/ipv4/ip_forward
    """
    machine = validation.Validate(username, state, machine_id=fields.getfirst('machine_id')).machine

    TOKEN_KEY = "0M6W0U1IXexThi5idy8mnkqPKEq1LtEnlK/pZSn0cDrN"

    data = {}
    data["user"] = username
    data["machine"] = machine.name
    data["expires"] = time.time()+(5*60)
    pickled_data = cPickle.dumps(data)
    m = hmac.new(TOKEN_KEY, digestmod=sha)
    m.update(pickled_data)
    token = {'data': pickled_data, 'digest': m.digest()}
    token = cPickle.dumps(token)
    token = base64.urlsafe_b64encode(token)

    status = controls.statusInfo(machine)
    has_vnc = hasVnc(status)

    d = dict(user=username,
             on=status,
             has_vnc=has_vnc,
             machine=machine,
             hostname=state.environ.get('SERVER_NAME', 'localhost'),
             authtoken=token)
    return templates.vnc(searchList=[d])

def getHostname(nic):
    """Find the hostname associated with a NIC.

    XXX this should be merged with the similar logic in DNS and DHCP.
    """
    if nic.hostname and '.' in nic.hostname:
        return nic.hostname
    elif nic.machine:
        return nic.machine.name + '.xvm.mit.edu'
    else:
        return None


def getNicInfo(data_dict, machine):
    """Helper function for info, get data on nics for a machine.

    Modifies data_dict to include the relevant data, and returns a list
    of (key, name) pairs to display "name: data_dict[key]" to the user.
    """
    data_dict['num_nics'] = len(machine.nics)
    nic_fields_template = [('nic%s_hostname', 'NIC %s Hostname'),
                           ('nic%s_mac', 'NIC %s MAC Addr'),
                           ('nic%s_ip', 'NIC %s IP'),
                           ]
    nic_fields = []
    for i in range(len(machine.nics)):
        nic_fields.extend([(x % i, y % i) for x, y in nic_fields_template])
        if not i:
            data_dict['nic%s_hostname' % i] = getHostname(machine.nics[i])
        data_dict['nic%s_mac' % i] = machine.nics[i].mac_addr
        data_dict['nic%s_ip' % i] = machine.nics[i].ip
    if len(machine.nics) == 1:
        nic_fields = [(x, y.replace('NIC 0 ', '')) for x, y in nic_fields]
    return nic_fields

def getDiskInfo(data_dict, machine):
    """Helper function for info, get data on disks for a machine.

    Modifies data_dict to include the relevant data, and returns a list
    of (key, name) pairs to display "name: data_dict[key]" to the user.
    """
    data_dict['num_disks'] = len(machine.disks)
    disk_fields_template = [('%s_size', '%s size')]
    disk_fields = []
    for disk in machine.disks:
        name = disk.guest_device_name
        disk_fields.extend([(x % name, y % name) for x, y in
                            disk_fields_template])
        data_dict['%s_size' % name] = "%0.1f GiB" % (disk.size / 1024.)
    return disk_fields

def command(username, state, path, fields):
    """Handler for running commands like boot and delete on a VM."""
    back = fields.getfirst('back')
    try:
        d = controls.commandResult(username, state, fields)
        if d['command'] == 'Delete VM':
            back = 'list'
    except InvalidInput, err:
        if not back:
            raise
        print >> sys.stderr, err
        result = err
    else:
        result = 'Success!'
        if not back:
            return templates.command(searchList=[d])
    if back == 'list':
        state.clear() #Changed global state
        d = getListDict(username, state)
        d['result'] = result
        return templates.list(searchList=[d])
    elif back == 'info':
        machine = validation.Validate(username, state, machine_id=fields.getfirst('machine_id')).machine
        return ({'Status': '303 See Other',
                 'Location': 'info?machine_id=%d' % machine.machine_id},
                "You shouldn't see this message.")
    else:
        raise InvalidInput('back', back, 'Not a known back page.')

def modifyDict(username, state, fields):
    """Modify a machine as specified by CGI arguments.

    Return a list of local variables for modify.tmpl.
    """
    olddisk = {}
    transaction = ctx.current.create_transaction()
    try:
        kws = dict([(kw, fields.getfirst(kw)) for kw in 'machine_id owner admin contact name description memory vmtype disksize'.split()])
        validate = validation.Validate(username, state, **kws)
        machine = validate.machine
        oldname = machine.name

        if hasattr(validate, 'memory'):
            machine.memory = validate.memory

        if hasattr(validate, 'vmtype'):
            machine.type = validate.vmtype

        if hasattr(validate, 'disksize'):
            disksize = validate.disksize
            disk = machine.disks[0]
            if disk.size != disksize:
                olddisk[disk.guest_device_name] = disksize
                disk.size = disksize
                ctx.current.save(disk)

        update_acl = False
        if hasattr(validate, 'owner') and validate.owner != machine.owner:
            machine.owner = validate.owner
            update_acl = True
        if hasattr(validate, 'name'):
            machine.name = validate.name
        if hasattr(validate, 'description'):
            machine.description = validate.description
        if hasattr(validate, 'admin') and validate.admin != machine.administrator:
            machine.administrator = validate.admin
            update_acl = True
        if hasattr(validate, 'contact'):
            machine.contact = validate.contact

        ctx.current.save(machine)
        if update_acl:
            print >> sys.stderr, machine, machine.administrator
            cache_acls.refreshMachine(machine)
        transaction.commit()
    except:
        transaction.rollback()
        raise
    for diskname in olddisk:
        controls.resizeDisk(oldname, diskname, str(olddisk[diskname]))
    if hasattr(validate, 'name'):
        controls.renameMachine(machine, oldname, validate.name)
    return dict(user=username,
                command="modify",
                machine=machine)

def modify(username, state, path, fields):
    """Handler for modifying attributes of a machine."""
    try:
        modify_dict = modifyDict(username, state, fields)
    except InvalidInput, err:
        result = None
        machine = validation.Validate(username, state, machine_id=fields.getfirst('machine_id')).machine
    else:
        machine = modify_dict['machine']
        result = 'Success!'
        err = None
    info_dict = infoDict(username, state, machine)
    info_dict['err'] = err
    if err:
        for field in fields.keys():
            setattr(info_dict['defaults'], field, fields.getfirst(field))
    info_dict['result'] = result
    return templates.info(searchList=[info_dict])


def helpHandler(username, state, path, fields):
    """Handler for help messages."""
    simple = fields.getfirst('simple')
    subjects = fields.getlist('subject')

    help_mapping = {'ParaVM Console': """
ParaVM machines do not support local console access over VNC.  To
access the serial console of these machines, you can SSH with Kerberos
to console.xvm.mit.edu, using the name of the machine as your
username.""",
                    'HVM/ParaVM': """
HVM machines use the virtualization features of the processor, while
ParaVM machines use Xen's emulation of virtualization features.  You
want an HVM virtualized machine.""",
                    'CPU Weight': """
Don't ask us!  We're as mystified as you are.""",
                    'Owner': """
The owner field is used to determine <a
href="help?subject=Quotas">quotas</a>.  It must be the name of a
locker that you are an AFS administrator of.  In particular, you or an
AFS group you are a member of must have AFS rlidwka bits on the
locker.  You can check who administers the LOCKER locker using the
commands 'attach LOCKER; fs la /mit/LOCKER' on Athena.)  See also <a
href="help?subject=Administrator">administrator</a>.""",
                    'Administrator': """
The administrator field determines who can access the console and
power on and off the machine.  This can be either a user or a moira
group.""",
                    'Quotas': """
Quotas are determined on a per-locker basis.  Each locker may have a
maximum of 512 megabytes of active ram, 50 gigabytes of disk, and 4
active machines.""",
                    'Console': """
<strong>Framebuffer:</strong> At a Linux boot prompt in your VM, try
setting <tt>fb=false</tt> to disable the framebuffer.  If you don't,
your machine will run just fine, but the applet's display of the
console will suffer artifacts.
"""
                    }

    if not subjects:
        subjects = sorted(help_mapping.keys())

    d = dict(user=username,
             simple=simple,
             subjects=subjects,
             mapping=help_mapping)

    return templates.help(searchList=[d])


def badOperation(u, s, p, e):
    """Function called when accessing an unknown URI."""
    return ({'Status': '404 Not Found'}, 'Invalid operation.')

def infoDict(username, state, machine):
    """Get the variables used by info.tmpl."""
    status = controls.statusInfo(machine)
    checkpoint.checkpoint('Getting status info')
    has_vnc = hasVnc(status)
    if status is None:
        main_status = dict(name=machine.name,
                           memory=str(machine.memory))
        uptime = None
        cputime = None
    else:
        main_status = dict(status[1:])
        start_time = float(main_status.get('start_time', 0))
        uptime = datetime.timedelta(seconds=int(time.time()-start_time))
        cpu_time_float = float(main_status.get('cpu_time', 0))
        cputime = datetime.timedelta(seconds=int(cpu_time_float))
    checkpoint.checkpoint('Status')
    display_fields = """name uptime memory state cpu_weight on_reboot 
     on_poweroff on_crash on_xend_start on_xend_stop bootloader""".split()
    display_fields = [('name', 'Name'),
                      ('description', 'Description'),
                      ('owner', 'Owner'),
                      ('administrator', 'Administrator'),
                      ('contact', 'Contact'),
                      ('type', 'Type'),
                      'NIC_INFO',
                      ('uptime', 'uptime'),
                      ('cputime', 'CPU usage'),
                      ('memory', 'RAM'),
                      'DISK_INFO',
                      ('state', 'state (xen format)'),
                      ('cpu_weight', 'CPU weight'+helppopup('CPU Weight')),
                      ('on_reboot', 'Action on VM reboot'),
                      ('on_poweroff', 'Action on VM poweroff'),
                      ('on_crash', 'Action on VM crash'),
                      ('on_xend_start', 'Action on Xen start'),
                      ('on_xend_stop', 'Action on Xen stop'),
                      ('bootloader', 'Bootloader options'),
                      ]
    fields = []
    machine_info = {}
    machine_info['name'] = machine.name
    machine_info['description'] = machine.description
    machine_info['type'] = machine.type.hvm and 'HVM' or 'ParaVM'
    machine_info['owner'] = machine.owner
    machine_info['administrator'] = machine.administrator
    machine_info['contact'] = machine.contact

    nic_fields = getNicInfo(machine_info, machine)
    nic_point = display_fields.index('NIC_INFO')
    display_fields = (display_fields[:nic_point] + nic_fields +
                      display_fields[nic_point+1:])

    disk_fields = getDiskInfo(machine_info, machine)
    disk_point = display_fields.index('DISK_INFO')
    display_fields = (display_fields[:disk_point] + disk_fields +
                      display_fields[disk_point+1:])

    main_status['memory'] += ' MiB'
    for field, disp in display_fields:
        if field in ('uptime', 'cputime') and locals()[field] is not None:
            fields.append((disp, locals()[field]))
        elif field in machine_info:
            fields.append((disp, machine_info[field]))
        elif field in main_status:
            fields.append((disp, main_status[field]))
        else:
            pass
            #fields.append((disp, None))

    checkpoint.checkpoint('Got fields')


    max_mem = validation.maxMemory(machine.owner, state, machine, False)
    checkpoint.checkpoint('Got mem')
    max_disk = validation.maxDisk(machine.owner, machine)
    defaults = Defaults()
    for name in 'machine_id name description administrator owner memory contact'.split():
        setattr(defaults, name, getattr(machine, name))
    defaults.type = machine.type.type_id
    defaults.disk = "%0.2f" % (machine.disks[0].size/1024.)
    checkpoint.checkpoint('Got defaults')
    d = dict(user=username,
             on=status is not None,
             machine=machine,
             defaults=defaults,
             has_vnc=has_vnc,
             uptime=str(uptime),
             ram=machine.memory,
             max_mem=max_mem,
             max_disk=max_disk,
             owner_help=helppopup("Owner"),
             fields = fields)
    return d

def info(username, state, path, fields):
    """Handler for info on a single VM."""
    machine = validation.Validate(username, state, machine_id=fields.getfirst('machine_id')).machine
    d = infoDict(username, state, machine)
    checkpoint.checkpoint('Got infodict')
    return templates.info(searchList=[d])

def unauthFront(_, _2, _3, fields):
    """Information for unauth'd users."""
    return templates.unauth(searchList=[{'simple' : True}])

def overlord(username, state, path, fields):
    if path == '':
        return ({'Status': '303 See Other',
                 'Location': 'overlord/'},
                "You shouldn't see this message.")
    if not username in getAfsGroupMembers('system:xvm', 'athena.mit.edu'):
        raise InvalidInput('username', username, 'Not an overlord.')
    newstate = State(username, overlord=True)
    newstate.environ = state.environ
    return handler(username, newstate, path, fields)

def throwError(_, __, ___, ____):
    """Throw an error, to test the error-tracing mechanisms."""
    raise RuntimeError("test of the emergency broadcast system")

mapping = dict(list=listVms,
               vnc=vnc,
               command=command,
               modify=modify,
               info=info,
               create=create,
               help=helpHandler,
               unauth=unauthFront,
               overlord=overlord,
               errortest=throwError)

def printHeaders(headers):
    """Print a dictionary as HTTP headers."""
    for key, value in headers.iteritems():
        print '%s: %s' % (key, value)
    print

def send_error_mail(subject, body):
    import subprocess

    to = 'xvm@mit.edu'
    mail = """To: %s
From: root@xvm.mit.edu
Subject: %s

%s
""" % (to, subject, body)
    p = subprocess.Popen(['/usr/sbin/sendmail', to], stdin=subprocess.PIPE)
    p.stdin.write(mail)
    p.stdin.close()
    p.wait()

def show_error(op, username, fields, err, emsg, traceback):
    """Print an error page when an exception occurs"""
    d = dict(op=op, user=username, fields=fields,
             errorMessage=str(err), stderr=emsg, traceback=traceback)
    details = templates.error_raw(searchList=[d])
    if username not in ('price', 'ecprice', 'andersk'): #add yourself at will
        send_error_mail('xvm error on %s for %s: %s' % (op, username, err),
                        details)
    d['details'] = details
    return templates.error(searchList=[d])

def getUser(environ):
    """Return the current user based on the SSL environment variables"""
    return environ.get('REMOTE_USER', None)

def handler(username, state, path, fields):
    operation, path = pathSplit(path)
    if not operation:
        operation = 'list'
    print 'Starting', operation
    fun = mapping.get(operation, badOperation)
    return fun(username, state, path, fields)

class App:
    def __init__(self, environ, start_response):
        self.environ = environ
        self.start = start_response

        self.username = getUser(environ)
        self.state = State(self.username)
        self.state.environ = environ

        random.seed() #sigh

    def __iter__(self):
        start_time = time.time()
        sipb_xen_database.clear_cache()
        sys.stderr = StringIO()
        fields = cgi.FieldStorage(fp=self.environ['wsgi.input'], environ=self.environ)
        operation = self.environ.get('PATH_INFO', '')
        if not operation:
            self.start("301 Moved Permanently", [('Location', './')])
            return
        if self.username is None:
            operation = 'unauth'

        try:
            checkpoint.checkpoint('Before')
            output = handler(self.username, self.state, operation, fields)
            checkpoint.checkpoint('After')

            headers = dict(DEFAULT_HEADERS)
            if isinstance(output, tuple):
                new_headers, output = output
                headers.update(new_headers)
            e = revertStandardError()
            if e:
                if isinstance(output, basestring):
                    sys.stderr = StringIO()
                    x = str(output)
                    print >> sys.stderr, x
                    print >> sys.stderr, 'XXX'
                    print >> sys.stderr, e
                    raise Exception()
                output.addError(e)
            output_string =  str(output)
            checkpoint.checkpoint('output as a string')
        except Exception, err:
            if not fields.has_key('js'):
                if isinstance(err, InvalidInput):
                    self.start('200 OK', [('Content-Type', 'text/html')])
                    e = revertStandardError()
                    yield str(invalidInput(operation, self.username, fields,
                                           err, e))
                    return
            import traceback
            self.start('500 Internal Server Error',
                       [('Content-Type', 'text/html')])
            e = revertStandardError()
            s = show_error(operation, self.username, fields,
                           err, e, traceback.format_exc())
            yield str(s)
            return
        status = headers.setdefault('Status', '200 OK')
        del headers['Status']
        self.start(status, headers.items())
        yield output_string
        if fields.has_key('timedebug'):
            yield '<pre>%s</pre>' % cgi.escape(str(checkpoint))

def constructor():
    connect('postgres://sipb-xen@sipb-xen-dev.mit.edu/sipb_xen')
    return App

def main():
    from flup.server.fcgi_fork import WSGIServer
    WSGIServer(constructor()).run()

if __name__ == '__main__':
    main()