LXR qemu/scripts/analyze-migration.py

   1#!/usr/bin/env python
   2#
   3#  Migration Stream Analyzer
   4#
   5#  Copyright (c) 2015 Alexander Graf <agraf@suse.de>
   6#
   7# This library is free software; you can redistribute it and/or
   8# modify it under the terms of the GNU Lesser General Public
   9# License as published by the Free Software Foundation; either
  10# version 2 of the License, or (at your option) any later version.
  11#
  12# This library is distributed in the hope that it will be useful,
  13# but WITHOUT ANY WARRANTY; without even the implied warranty of
  14# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15# Lesser General Public License for more details.
  16#
  17# You should have received a copy of the GNU Lesser General Public
  18# License along with this library; if not, see <http://www.gnu.org/licenses/>.
  19
  20import numpy as np
  21import json
  22import os
  23import argparse
  24import collections
  25import pprint
  26
  27def mkdir_p(path):
  28    try:
  29        os.makedirs(path)
  30    except OSError:
  31        pass
  32
  33class MigrationFile(object):
  34    def __init__(self, filename):
  35        self.filename = filename
  36        self.file = open(self.filename, "rb")
  37
  38    def read64(self):
  39        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i8')[0])
  40
  41    def read32(self):
  42        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i4')[0])
  43
  44    def read16(self):
  45        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i2')[0])
  46
  47    def read8(self):
  48        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i1')[0])
  49
  50    def readstr(self, len = None):
  51        if len is None:
  52            len = self.read8()
  53        if len == 0:
  54            return ""
  55        return np.fromfile(self.file, count=1, dtype=('S%d' % len))[0]
  56
  57    def readvar(self, size = None):
  58        if size is None:
  59            size = self.read8()
  60        if size == 0:
  61            return ""
  62        value = self.file.read(size)
  63        if len(value) != size:
  64            raise Exception("Unexpected end of %s at 0x%x" % (self.filename, self.file.tell()))
  65        return value
  66
  67    def tell(self):
  68        return self.file.tell()
  69
  70    # The VMSD description is at the end of the file, after EOF. Look for
  71    # the last NULL byte, then for the beginning brace of JSON.
  72    def read_migration_debug_json(self):
  73        QEMU_VM_VMDESCRIPTION = 0x06
  74
  75        # Remember the offset in the file when we started
  76        entrypos = self.file.tell()
  77
  78        # Read the last 10MB
  79        self.file.seek(0, os.SEEK_END)
  80        endpos = self.file.tell()
  81        self.file.seek(max(-endpos, -10 * 1024 * 1024), os.SEEK_END)
  82        datapos = self.file.tell()
  83        data = self.file.read()
  84        # The full file read closed the file as well, reopen it
  85        self.file = open(self.filename, "rb")
  86
  87        # Find the last NULL byte, then the first brace after that. This should
  88        # be the beginning of our JSON data.
  89        nulpos = data.rfind("\0")
  90        jsonpos = data.find("{", nulpos)
  91
  92        # Check backwards from there and see whether we guessed right
  93        self.file.seek(datapos + jsonpos - 5, 0)
  94        if self.read8() != QEMU_VM_VMDESCRIPTION:
  95            raise Exception("No Debug Migration device found")
  96
  97        jsonlen = self.read32()
  98
  99        # Seek back to where we were at the beginning
 100        self.file.seek(entrypos, 0)
 101
 102        return data[jsonpos:jsonpos + jsonlen]
 103
 104    def close(self):
 105        self.file.close()
 106
 107class RamSection(object):
 108    RAM_SAVE_FLAG_COMPRESS = 0x02
 109    RAM_SAVE_FLAG_MEM_SIZE = 0x04
 110    RAM_SAVE_FLAG_PAGE     = 0x08
 111    RAM_SAVE_FLAG_EOS      = 0x10
 112    RAM_SAVE_FLAG_CONTINUE = 0x20
 113    RAM_SAVE_FLAG_XBZRLE   = 0x40
 114    RAM_SAVE_FLAG_HOOK     = 0x80
 115
 116    def __init__(self, file, version_id, ramargs, section_key):
 117        if version_id != 4:
 118            raise Exception("Unknown RAM version %d" % version_id)
 119
 120        self.file = file
 121        self.section_key = section_key
 122        self.TARGET_PAGE_SIZE = ramargs['page_size']
 123        self.dump_memory = ramargs['dump_memory']
 124        self.write_memory = ramargs['write_memory']
 125        self.sizeinfo = collections.OrderedDict()
 126        self.data = collections.OrderedDict()
 127        self.data['section sizes'] = self.sizeinfo
 128        self.name = ''
 129        if self.write_memory:
 130            self.files = { }
 131        if self.dump_memory:
 132            self.memory = collections.OrderedDict()
 133            self.data['memory'] = self.memory
 134
 135    def __repr__(self):
 136        return self.data.__repr__()
 137
 138    def __str__(self):
 139        return self.data.__str__()
 140
 141    def getDict(self):
 142        return self.data
 143
 144    def read(self):
 145        # Read all RAM sections
 146        while True:
 147            addr = self.file.read64()
 148            flags = addr & (self.TARGET_PAGE_SIZE - 1)
 149            addr &= ~(self.TARGET_PAGE_SIZE - 1)
 150
 151            if flags & self.RAM_SAVE_FLAG_MEM_SIZE:
 152                while True:
 153                    namelen = self.file.read8()
 154                    # We assume that no RAM chunk is big enough to ever
 155                    # hit the first byte of the address, so when we see
 156                    # a zero here we know it has to be an address, not the
 157                    # length of the next block.
 158                    if namelen == 0:
 159                        self.file.file.seek(-1, 1)
 160                        break
 161                    self.name = self.file.readstr(len = namelen)
 162                    len = self.file.read64()
 163                    self.sizeinfo[self.name] = '0x%016x' % len
 164                    if self.write_memory:
 165                        print self.name
 166                        mkdir_p('./' + os.path.dirname(self.name))
 167                        f = open('./' + self.name, "wb")
 168                        f.truncate(0)
 169                        f.truncate(len)
 170                        self.files[self.name] = f
 171                flags &= ~self.RAM_SAVE_FLAG_MEM_SIZE
 172
 173            if flags & self.RAM_SAVE_FLAG_COMPRESS:
 174                if flags & self.RAM_SAVE_FLAG_CONTINUE:
 175                    flags &= ~self.RAM_SAVE_FLAG_CONTINUE
 176                else:
 177                    self.name = self.file.readstr()
 178                fill_char = self.file.read8()
 179                # The page in question is filled with fill_char now
 180                if self.write_memory and fill_char != 0:
 181                    self.files[self.name].seek(addr, os.SEEK_SET)
 182                    self.files[self.name].write(chr(fill_char) * self.TARGET_PAGE_SIZE)
 183                if self.dump_memory:
 184                    self.memory['%s (0x%016x)' % (self.name, addr)] = 'Filled with 0x%02x' % fill_char
 185                flags &= ~self.RAM_SAVE_FLAG_COMPRESS
 186            elif flags & self.RAM_SAVE_FLAG_PAGE:
 187                if flags & self.RAM_SAVE_FLAG_CONTINUE:
 188                    flags &= ~self.RAM_SAVE_FLAG_CONTINUE
 189                else:
 190                    self.name = self.file.readstr()
 191
 192                if self.write_memory or self.dump_memory:
 193                    data = self.file.readvar(size = self.TARGET_PAGE_SIZE)
 194                else: # Just skip RAM data
 195                    self.file.file.seek(self.TARGET_PAGE_SIZE, 1)
 196
 197                if self.write_memory:
 198                    self.files[self.name].seek(addr, os.SEEK_SET)
 199                    self.files[self.name].write(data)
 200                if self.dump_memory:
 201                    hexdata = " ".join("{0:02x}".format(ord(c)) for c in data)
 202                    self.memory['%s (0x%016x)' % (self.name, addr)] = hexdata
 203
 204                flags &= ~self.RAM_SAVE_FLAG_PAGE
 205            elif flags & self.RAM_SAVE_FLAG_XBZRLE:
 206                raise Exception("XBZRLE RAM compression is not supported yet")
 207            elif flags & self.RAM_SAVE_FLAG_HOOK:
 208                raise Exception("RAM hooks don't make sense with files")
 209
 210            # End of RAM section
 211            if flags & self.RAM_SAVE_FLAG_EOS:
 212                break
 213
 214            if flags != 0:
 215                raise Exception("Unknown RAM flags: %x" % flags)
 216
 217    def __del__(self):
 218        if self.write_memory:
 219            for key in self.files:
 220                self.files[key].close()
 221
 222
 223class HTABSection(object):
 224    HASH_PTE_SIZE_64       = 16
 225
 226    def __init__(self, file, version_id, device, section_key):
 227        if version_id != 1:
 228            raise Exception("Unknown HTAB version %d" % version_id)
 229
 230        self.file = file
 231        self.section_key = section_key
 232
 233    def read(self):
 234
 235        header = self.file.read32()
 236
 237        if (header > 0):
 238            # First section, just the hash shift
 239            return
 240
 241        # Read until end marker
 242        while True:
 243            index = self.file.read32()
 244            n_valid = self.file.read16()
 245            n_invalid = self.file.read16()
 246
 247            if index == 0 and n_valid == 0 and n_invalid == 0:
 248                break
 249
 250            self.file.readvar(n_valid * self.HASH_PTE_SIZE_64)
 251
 252    def getDict(self):
 253        return ""
 254
 255
 256class ConfigurationSection(object):
 257    def __init__(self, file):
 258        self.file = file
 259
 260    def read(self):
 261        name_len = self.file.read32()
 262        name = self.file.readstr(len = name_len)
 263
 264class VMSDFieldGeneric(object):
 265    def __init__(self, desc, file):
 266        self.file = file
 267        self.desc = desc
 268        self.data = ""
 269
 270    def __repr__(self):
 271        return str(self.__str__())
 272
 273    def __str__(self):
 274        return " ".join("{0:02x}".format(ord(c)) for c in self.data)
 275
 276    def getDict(self):
 277        return self.__str__()
 278
 279    def read(self):
 280        size = int(self.desc['size'])
 281        self.data = self.file.readvar(size)
 282        return self.data
 283
 284class VMSDFieldInt(VMSDFieldGeneric):
 285    def __init__(self, desc, file):
 286        super(VMSDFieldInt, self).__init__(desc, file)
 287        self.size = int(desc['size'])
 288        self.format = '0x%%0%dx' % (self.size * 2)
 289        self.sdtype = '>i%d' % self.size
 290        self.udtype = '>u%d' % self.size
 291
 292    def __repr__(self):
 293        if self.data < 0:
 294            return ('%s (%d)' % ((self.format % self.udata), self.data))
 295        else:
 296            return self.format % self.data
 297
 298    def __str__(self):
 299        return self.__repr__()
 300
 301    def getDict(self):
 302        return self.__str__()
 303
 304    def read(self):
 305        super(VMSDFieldInt, self).read()
 306        self.sdata = np.fromstring(self.data, count=1, dtype=(self.sdtype))[0]
 307        self.udata = np.fromstring(self.data, count=1, dtype=(self.udtype))[0]
 308        self.data = self.sdata
 309        return self.data
 310
 311class VMSDFieldUInt(VMSDFieldInt):
 312    def __init__(self, desc, file):
 313        super(VMSDFieldUInt, self).__init__(desc, file)
 314
 315    def read(self):
 316        super(VMSDFieldUInt, self).read()
 317        self.data = self.udata
 318        return self.data
 319
 320class VMSDFieldIntLE(VMSDFieldInt):
 321    def __init__(self, desc, file):
 322        super(VMSDFieldIntLE, self).__init__(desc, file)
 323        self.dtype = '<i%d' % self.size
 324
 325class VMSDFieldBool(VMSDFieldGeneric):
 326    def __init__(self, desc, file):
 327        super(VMSDFieldBool, self).__init__(desc, file)
 328
 329    def __repr__(self):
 330        return self.data.__repr__()
 331
 332    def __str__(self):
 333        return self.data.__str__()
 334
 335    def getDict(self):
 336        return self.data
 337
 338    def read(self):
 339        super(VMSDFieldBool, self).read()
 340        if self.data[0] == 0:
 341            self.data = False
 342        else:
 343            self.data = True
 344        return self.data
 345
 346class VMSDFieldStruct(VMSDFieldGeneric):
 347    QEMU_VM_SUBSECTION    = 0x05
 348
 349    def __init__(self, desc, file):
 350        super(VMSDFieldStruct, self).__init__(desc, file)
 351        self.data = collections.OrderedDict()
 352
 353        # When we see compressed array elements, unfold them here
 354        new_fields = []
 355        for field in self.desc['struct']['fields']:
 356            if not 'array_len' in field:
 357                new_fields.append(field)
 358                continue
 359            array_len = field.pop('array_len')
 360            field['index'] = 0
 361            new_fields.append(field)
 362            for i in xrange(1, array_len):
 363                c = field.copy()
 364                c['index'] = i
 365                new_fields.append(c)
 366
 367        self.desc['struct']['fields'] = new_fields
 368
 369    def __repr__(self):
 370        return self.data.__repr__()
 371
 372    def __str__(self):
 373        return self.data.__str__()
 374
 375    def read(self):
 376        for field in self.desc['struct']['fields']:
 377            try:
 378                reader = vmsd_field_readers[field['type']]
 379            except:
 380                reader = VMSDFieldGeneric
 381
 382            field['data'] = reader(field, self.file)
 383            field['data'].read()
 384
 385            if 'index' in field:
 386                if field['name'] not in self.data:
 387                    self.data[field['name']] = []
 388                a = self.data[field['name']]
 389                if len(a) != int(field['index']):
 390                    raise Exception("internal index of data field unmatched (%d/%d)" % (len(a), int(field['index'])))
 391                a.append(field['data'])
 392            else:
 393                self.data[field['name']] = field['data']
 394
 395        if 'subsections' in self.desc['struct']:
 396            for subsection in self.desc['struct']['subsections']:
 397                if self.file.read8() != self.QEMU_VM_SUBSECTION:
 398                    raise Exception("Subsection %s not found at offset %x" % ( subsection['vmsd_name'], self.file.tell()))
 399                name = self.file.readstr()
 400                version_id = self.file.read32()
 401                self.data[name] = VMSDSection(self.file, version_id, subsection, (name, 0))
 402                self.data[name].read()
 403
 404    def getDictItem(self, value):
 405       # Strings would fall into the array category, treat
 406       # them specially
 407       if value.__class__ is ''.__class__:
 408           return value
 409
 410       try:
 411           return self.getDictOrderedDict(value)
 412       except:
 413           try:
 414               return self.getDictArray(value)
 415           except:
 416               try:
 417                   return value.getDict()
 418               except:
 419                   return value
 420
 421    def getDictArray(self, array):
 422        r = []
 423        for value in array:
 424           r.append(self.getDictItem(value))
 425        return r
 426
 427    def getDictOrderedDict(self, dict):
 428        r = collections.OrderedDict()
 429        for (key, value) in dict.items():
 430            r[key] = self.getDictItem(value)
 431        return r
 432
 433    def getDict(self):
 434        return self.getDictOrderedDict(self.data)
 435
 436vmsd_field_readers = {
 437    "bool" : VMSDFieldBool,
 438    "int8" : VMSDFieldInt,
 439    "int16" : VMSDFieldInt,
 440    "int32" : VMSDFieldInt,
 441    "int32 equal" : VMSDFieldInt,
 442    "int32 le" : VMSDFieldIntLE,
 443    "int64" : VMSDFieldInt,
 444    "uint8" : VMSDFieldUInt,
 445    "uint16" : VMSDFieldUInt,
 446    "uint32" : VMSDFieldUInt,
 447    "uint32 equal" : VMSDFieldUInt,
 448    "uint64" : VMSDFieldUInt,
 449    "int64 equal" : VMSDFieldInt,
 450    "uint8 equal" : VMSDFieldInt,
 451    "uint16 equal" : VMSDFieldInt,
 452    "float64" : VMSDFieldGeneric,
 453    "timer" : VMSDFieldGeneric,
 454    "buffer" : VMSDFieldGeneric,
 455    "unused_buffer" : VMSDFieldGeneric,
 456    "bitmap" : VMSDFieldGeneric,
 457    "struct" : VMSDFieldStruct,
 458    "unknown" : VMSDFieldGeneric,
 459}
 460
 461class VMSDSection(VMSDFieldStruct):
 462    def __init__(self, file, version_id, device, section_key):
 463        self.file = file
 464        self.data = ""
 465        self.vmsd_name = ""
 466        self.section_key = section_key
 467        desc = device
 468        if 'vmsd_name' in device:
 469            self.vmsd_name = device['vmsd_name']
 470
 471        # A section really is nothing but a FieldStruct :)
 472        super(VMSDSection, self).__init__({ 'struct' : desc }, file)
 473
 474###############################################################################
 475
 476class MigrationDump(object):
 477    QEMU_VM_FILE_MAGIC    = 0x5145564d
 478    QEMU_VM_FILE_VERSION  = 0x00000003
 479    QEMU_VM_EOF           = 0x00
 480    QEMU_VM_SECTION_START = 0x01
 481    QEMU_VM_SECTION_PART  = 0x02
 482    QEMU_VM_SECTION_END   = 0x03
 483    QEMU_VM_SECTION_FULL  = 0x04
 484    QEMU_VM_SUBSECTION    = 0x05
 485    QEMU_VM_VMDESCRIPTION = 0x06
 486    QEMU_VM_CONFIGURATION = 0x07
 487    QEMU_VM_SECTION_FOOTER= 0x7e
 488
 489    def __init__(self, filename):
 490        self.section_classes = { ( 'ram', 0 ) : [ RamSection, None ],
 491                                 ( 'spapr/htab', 0) : ( HTABSection, None ) }
 492        self.filename = filename
 493        self.vmsd_desc = None
 494
 495    def read(self, desc_only = False, dump_memory = False, write_memory = False):
 496        # Read in the whole file
 497        file = MigrationFile(self.filename)
 498
 499        # File magic
 500        data = file.read32()
 501        if data != self.QEMU_VM_FILE_MAGIC:
 502            raise Exception("Invalid file magic %x" % data)
 503
 504        # Version (has to be v3)
 505        data = file.read32()
 506        if data != self.QEMU_VM_FILE_VERSION:
 507            raise Exception("Invalid version number %d" % data)
 508
 509        self.load_vmsd_json(file)
 510
 511        # Read sections
 512        self.sections = collections.OrderedDict()
 513
 514        if desc_only:
 515            return
 516
 517        ramargs = {}
 518        ramargs['page_size'] = self.vmsd_desc['page_size']
 519        ramargs['dump_memory'] = dump_memory
 520        ramargs['write_memory'] = write_memory
 521        self.section_classes[('ram',0)][1] = ramargs
 522
 523        while True:
 524            section_type = file.read8()
 525            if section_type == self.QEMU_VM_EOF:
 526                break
 527            elif section_type == self.QEMU_VM_CONFIGURATION:
 528                section = ConfigurationSection(file)
 529                section.read()
 530            elif section_type == self.QEMU_VM_SECTION_START or section_type == self.QEMU_VM_SECTION_FULL:
 531                section_id = file.read32()
 532                name = file.readstr()
 533                instance_id = file.read32()
 534                version_id = file.read32()
 535                section_key = (name, instance_id)
 536                classdesc = self.section_classes[section_key]
 537                section = classdesc[0](file, version_id, classdesc[1], section_key)
 538                self.sections[section_id] = section
 539                section.read()
 540            elif section_type == self.QEMU_VM_SECTION_PART or section_type == self.QEMU_VM_SECTION_END:
 541                section_id = file.read32()
 542                self.sections[section_id].read()
 543            elif section_type == self.QEMU_VM_SECTION_FOOTER:
 544                read_section_id = file.read32()
 545                if read_section_id != section_id:
 546                    raise Exception("Mismatched section footer: %x vs %x" % (read_section_id, section_id))
 547            else:
 548                raise Exception("Unknown section type: %d" % section_type)
 549        file.close()
 550
 551    def load_vmsd_json(self, file):
 552        vmsd_json = file.read_migration_debug_json()
 553        self.vmsd_desc = json.loads(vmsd_json, object_pairs_hook=collections.OrderedDict)
 554        for device in self.vmsd_desc['devices']:
 555            key = (device['name'], device['instance_id'])
 556            value = ( VMSDSection, device )
 557            self.section_classes[key] = value
 558
 559    def getDict(self):
 560        r = collections.OrderedDict()
 561        for (key, value) in self.sections.items():
 562           key = "%s (%d)" % ( value.section_key[0], key )
 563           r[key] = value.getDict()
 564        return r
 565
 566###############################################################################
 567
 568class JSONEncoder(json.JSONEncoder):
 569    def default(self, o):
 570        if isinstance(o, VMSDFieldGeneric):
 571            return str(o)
 572        return json.JSONEncoder.default(self, o)
 573
 574parser = argparse.ArgumentParser()
 575parser.add_argument("-f", "--file", help='migration dump to read from', required=True)
 576parser.add_argument("-m", "--memory", help='dump RAM contents as well', action='store_true')
 577parser.add_argument("-d", "--dump", help='what to dump ("state" or "desc")', default='state')
 578parser.add_argument("-x", "--extract", help='extract contents into individual files', action='store_true')
 579args = parser.parse_args()
 580
 581jsonenc = JSONEncoder(indent=4, separators=(',', ': '))
 582
 583if args.extract:
 584    dump = MigrationDump(args.file)
 585
 586    dump.read(desc_only = True)
 587    print "desc.json"
 588    f = open("desc.json", "wb")
 589    f.truncate()
 590    f.write(jsonenc.encode(dump.vmsd_desc))
 591    f.close()
 592
 593    dump.read(write_memory = True)
 594    dict = dump.getDict()
 595    print "state.json"
 596    f = open("state.json", "wb")
 597    f.truncate()
 598    f.write(jsonenc.encode(dict))
 599    f.close()
 600elif args.dump == "state":
 601    dump = MigrationDump(args.file)
 602    dump.read(dump_memory = args.memory)
 603    dict = dump.getDict()
 604    print jsonenc.encode(dict)
 605elif args.dump == "desc":
 606    dump = MigrationDump(args.file)
 607    dump.read(desc_only = True)
 608    print jsonenc.encode(dump.vmsd_desc)
 609else:
 610    raise Exception("Please specify either -x, -d state or -d dump")
 611