scripts/analyze-migration.py

   1 #!/usr/bin/env python3
   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
  20 import json
  21 import os
  22 import argparse
  23 import collections
  24 import struct
  25 import sys
  26
  27
  28 MIN_PYTHON = (3, 2)
  29 if sys.version_info < MIN_PYTHON:
  30     sys.exit("Python %s.%s or later is required.\n" % MIN_PYTHON)
  31
  32
  33 def mkdir_p(path):
  34     try:
  35         os.makedirs(path)
  36     except OSError:
  37         pass
  38
  39
  40 class MigrationFile(object):
  41     def __init__(self, filename):
  42         self.filename = filename
  43         self.file = open(self.filename, "rb")
  44
  45     def read64(self):
  46         return int.from_bytes(self.file.read(8), byteorder='big', signed=True)
  47
  48     def read32(self):
  49         return int.from_bytes(self.file.read(4), byteorder='big', signed=True)
  50
  51     def read16(self):
  52         return int.from_bytes(self.file.read(2), byteorder='big', signed=True)
  53
  54     def read8(self):
  55         return int.from_bytes(self.file.read(1), byteorder='big', signed=True)
  56
  57     def readstr(self, len = None):
  58         return self.readvar(len).decode('utf-8')
  59
  60     def readvar(self, size = None):
  61         if size is None:
  62             size = self.read8()
  63         if size == 0:
  64             return ""
  65         value = self.file.read(size)
  66         if len(value) != size:
  67             raise Exception("Unexpected end of %s at 0x%x" % (self.filename, self.file.tell()))
  68         return value
  69
  70     def tell(self):
  71         return self.file.tell()
  72
  73     # The VMSD description is at the end of the file, after EOF. Look for
  74     # the last NULL byte, then for the beginning brace of JSON.
  75     def read_migration_debug_json(self):
  76         QEMU_VM_VMDESCRIPTION = 0x06
  77
  78         # Remember the offset in the file when we started
  79         entrypos = self.file.tell()
  80
  81         # Read the last 10MB
  82         self.file.seek(0, os.SEEK_END)
  83         endpos = self.file.tell()
  84         self.file.seek(max(-endpos, -10 * 1024 * 1024), os.SEEK_END)
  85         datapos = self.file.tell()
  86         data = self.file.read()
  87         # The full file read closed the file as well, reopen it
  88         self.file = open(self.filename, "rb")
  89
  90         # Find the last NULL byte, then the first brace after that. This should
  91         # be the beginning of our JSON data.
  92         nulpos = data.rfind(b'\0')
  93         jsonpos = data.find(b'{', nulpos)
  94
  95         # Check backwards from there and see whether we guessed right
  96         self.file.seek(datapos + jsonpos - 5, 0)
  97         if self.read8() != QEMU_VM_VMDESCRIPTION:
  98             raise Exception("No Debug Migration device found")
  99
 100         jsonlen = self.read32()
 101
 102         # Seek back to where we were at the beginning
 103         self.file.seek(entrypos, 0)
 104
 105         return data[jsonpos:jsonpos + jsonlen]
 106
 107     def close(self):
 108         self.file.close()
 109
 110 class RamSection(object):
 111     RAM_SAVE_FLAG_COMPRESS = 0x02
 112     RAM_SAVE_FLAG_MEM_SIZE = 0x04
 113     RAM_SAVE_FLAG_PAGE     = 0x08
 114     RAM_SAVE_FLAG_EOS      = 0x10
 115     RAM_SAVE_FLAG_CONTINUE = 0x20
 116     RAM_SAVE_FLAG_XBZRLE   = 0x40
 117     RAM_SAVE_FLAG_HOOK     = 0x80
 118
 119     def __init__(self, file, version_id, ramargs, section_key):
 120         if version_id != 4:
 121             raise Exception("Unknown RAM version %d" % version_id)
 122
 123         self.file = file
 124         self.section_key = section_key
 125         self.TARGET_PAGE_SIZE = ramargs['page_size']
 126         self.dump_memory = ramargs['dump_memory']
 127         self.write_memory = ramargs['write_memory']
 128         self.sizeinfo = collections.OrderedDict()
 129         self.data = collections.OrderedDict()
 130         self.data['section sizes'] = self.sizeinfo
 131         self.name = ''
 132         if self.write_memory:
 133             self.files = { }
 134         if self.dump_memory:
 135             self.memory = collections.OrderedDict()
 136             self.data['memory'] = self.memory
 137
 138     def __repr__(self):
 139         return self.data.__repr__()
 140
 141     def __str__(self):
 142         return self.data.__str__()
 143
 144     def getDict(self):
 145         return self.data
 146
 147     def read(self):
 148         # Read all RAM sections
 149         while True:
 150             addr = self.file.read64()
 151             flags = addr & (self.TARGET_PAGE_SIZE - 1)
 152             addr &= ~(self.TARGET_PAGE_SIZE - 1)
 153
 154             if flags & self.RAM_SAVE_FLAG_MEM_SIZE:
 155                 while True:
 156                     namelen = self.file.read8()
 157                     # We assume that no RAM chunk is big enough to ever
 158                     # hit the first byte of the address, so when we see
 159                     # a zero here we know it has to be an address, not the
 160                     # length of the next block.
 161                     if namelen == 0:
 162                         self.file.file.seek(-1, 1)
 163                         break
 164                     self.name = self.file.readstr(len = namelen)
 165                     len = self.file.read64()
 166                     self.sizeinfo[self.name] = '0x%016x' % len
 167                     if self.write_memory:
 168                         print(self.name)
 169                         mkdir_p('./' + os.path.dirname(self.name))
 170                         f = open('./' + self.name, "wb")
 171                         f.truncate(0)
 172                         f.truncate(len)
 173                         self.files[self.name] = f
 174                 flags &= ~self.RAM_SAVE_FLAG_MEM_SIZE
 175
 176             if flags & self.RAM_SAVE_FLAG_COMPRESS:
 177                 if flags & self.RAM_SAVE_FLAG_CONTINUE:
 178                     flags &= ~self.RAM_SAVE_FLAG_CONTINUE
 179                 else:
 180                     self.name = self.file.readstr()
 181                 fill_char = self.file.read8()
 182                 # The page in question is filled with fill_char now
 183                 if self.write_memory and fill_char != 0:
 184                     self.files[self.name].seek(addr, os.SEEK_SET)
 185                     self.files[self.name].write(chr(fill_char) * self.TARGET_PAGE_SIZE)
 186                 if self.dump_memory:
 187                     self.memory['%s (0x%016x)' % (self.name, addr)] = 'Filled with 0x%02x' % fill_char
 188                 flags &= ~self.RAM_SAVE_FLAG_COMPRESS
 189             elif flags & self.RAM_SAVE_FLAG_PAGE:
 190                 if flags & self.RAM_SAVE_FLAG_CONTINUE:
 191                     flags &= ~self.RAM_SAVE_FLAG_CONTINUE
 192                 else:
 193                     self.name = self.file.readstr()
 194
 195                 if self.write_memory or self.dump_memory:
 196                     data = self.file.readvar(size = self.TARGET_PAGE_SIZE)
 197                 else: # Just skip RAM data
 198                     self.file.file.seek(self.TARGET_PAGE_SIZE, 1)
 199
 200                 if self.write_memory:
 201                     self.files[self.name].seek(addr, os.SEEK_SET)
 202                     self.files[self.name].write(data)
 203                 if self.dump_memory:
 204                     hexdata = " ".join("{0:02x}".format(ord(c)) for c in data)
 205                     self.memory['%s (0x%016x)' % (self.name, addr)] = hexdata
 206
 207                 flags &= ~self.RAM_SAVE_FLAG_PAGE
 208             elif flags & self.RAM_SAVE_FLAG_XBZRLE:
 209                 raise Exception("XBZRLE RAM compression is not supported yet")
 210             elif flags & self.RAM_SAVE_FLAG_HOOK:
 211                 raise Exception("RAM hooks don't make sense with files")
 212
 213             # End of RAM section
 214             if flags & self.RAM_SAVE_FLAG_EOS:
 215                 break
 216
 217             if flags != 0:
 218                 raise Exception("Unknown RAM flags: %x" % flags)
 219
 220     def __del__(self):
 221         if self.write_memory:
 222             for key in self.files:
 223                 self.files[key].close()
 224
 225
 226 class HTABSection(object):
 227     HASH_PTE_SIZE_64       = 16
 228
 229     def __init__(self, file, version_id, device, section_key):
 230         if version_id != 1:
 231             raise Exception("Unknown HTAB version %d" % version_id)
 232
 233         self.file = file
 234         self.section_key = section_key
 235
 236     def read(self):
 237
 238         header = self.file.read32()
 239
 240         if (header == -1):
 241             # "no HPT" encoding
 242             return
 243
 244         if (header > 0):
 245             # First section, just the hash shift
 246             return
 247
 248         # Read until end marker
 249         while True:
 250             index = self.file.read32()
 251             n_valid = self.file.read16()
 252             n_invalid = self.file.read16()
 253
 254             if index == 0 and n_valid == 0 and n_invalid == 0:
 255                 break
 256
 257             self.file.readvar(n_valid * self.HASH_PTE_SIZE_64)
 258
 259     def getDict(self):
 260         return ""
 261
 262
 263 class ConfigurationSection(object):
 264     def __init__(self, file):
 265         self.file = file
 266
 267     def read(self):
 268         name_len = self.file.read32()
 269         name = self.file.readstr(len = name_len)
 270
 271 class VMSDFieldGeneric(object):
 272     def __init__(self, desc, file):
 273         self.file = file
 274         self.desc = desc
 275         self.data = ""
 276
 277     def __repr__(self):
 278         return str(self.__str__())
 279
 280     def __str__(self):
 281         return " ".join("{0:02x}".format(c) for c in self.data)
 282
 283     def getDict(self):
 284         return self.__str__()
 285
 286     def read(self):
 287         size = int(self.desc['size'])
 288         self.data = self.file.readvar(size)
 289         return self.data
 290
 291 class VMSDFieldInt(VMSDFieldGeneric):
 292     def __init__(self, desc, file):
 293         super(VMSDFieldInt, self).__init__(desc, file)
 294         self.size = int(desc['size'])
 295         self.format = '0x%%0%dx' % (self.size * 2)
 296         self.sdtype = '>i%d' % self.size
 297         self.udtype = '>u%d' % self.size
 298
 299     def __repr__(self):
 300         if self.data < 0:
 301             return ('%s (%d)' % ((self.format % self.udata), self.data))
 302         else:
 303             return self.format % self.data
 304
 305     def __str__(self):
 306         return self.__repr__()
 307
 308     def getDict(self):
 309         return self.__str__()
 310
 311     def read(self):
 312         super(VMSDFieldInt, self).read()
 313         self.sdata = int.from_bytes(self.data, byteorder='big', signed=True)
 314         self.udata = int.from_bytes(self.data, byteorder='big', signed=False)
 315         self.data = self.sdata
 316         return self.data
 317
 318 class VMSDFieldUInt(VMSDFieldInt):
 319     def __init__(self, desc, file):
 320         super(VMSDFieldUInt, self).__init__(desc, file)
 321
 322     def read(self):
 323         super(VMSDFieldUInt, self).read()
 324         self.data = self.udata
 325         return self.data
 326
 327 class VMSDFieldIntLE(VMSDFieldInt):
 328     def __init__(self, desc, file):
 329         super(VMSDFieldIntLE, self).__init__(desc, file)
 330         self.dtype = '<i%d' % self.size
 331
 332 class VMSDFieldBool(VMSDFieldGeneric):
 333     def __init__(self, desc, file):
 334         super(VMSDFieldBool, self).__init__(desc, file)
 335
 336     def __repr__(self):
 337         return self.data.__repr__()
 338
 339     def __str__(self):
 340         return self.data.__str__()
 341
 342     def getDict(self):
 343         return self.data
 344
 345     def read(self):
 346         super(VMSDFieldBool, self).read()
 347         if self.data[0] == 0:
 348             self.data = False
 349         else:
 350             self.data = True
 351         return self.data
 352
 353 class VMSDFieldStruct(VMSDFieldGeneric):
 354     QEMU_VM_SUBSECTION    = 0x05
 355
 356     def __init__(self, desc, file):
 357         super(VMSDFieldStruct, self).__init__(desc, file)
 358         self.data = collections.OrderedDict()
 359
 360         # When we see compressed array elements, unfold them here
 361         new_fields = []
 362         for field in self.desc['struct']['fields']:
 363             if not 'array_len' in field:
 364                 new_fields.append(field)
 365                 continue
 366             array_len = field.pop('array_len')
 367             field['index'] = 0
 368             new_fields.append(field)
 369             for i in range(1, array_len):
 370                 c = field.copy()
 371                 c['index'] = i
 372                 new_fields.append(c)
 373
 374         self.desc['struct']['fields'] = new_fields
 375
 376     def __repr__(self):
 377         return self.data.__repr__()
 378
 379     def __str__(self):
 380         return self.data.__str__()
 381
 382     def read(self):
 383         for field in self.desc['struct']['fields']:
 384             try:
 385                 reader = vmsd_field_readers[field['type']]
 386             except:
 387                 reader = VMSDFieldGeneric
 388
 389             field['data'] = reader(field, self.file)
 390             field['data'].read()
 391
 392             if 'index' in field:
 393                 if field['name'] not in self.data:
 394                     self.data[field['name']] = []
 395                 a = self.data[field['name']]
 396                 if len(a) != int(field['index']):
 397                     raise Exception("internal index of data field unmatched (%d/%d)" % (len(a), int(field['index'])))
 398                 a.append(field['data'])
 399             else:
 400                 self.data[field['name']] = field['data']
 401
 402         if 'subsections' in self.desc['struct']:
 403             for subsection in self.desc['struct']['subsections']:
 404                 if self.file.read8() != self.QEMU_VM_SUBSECTION:
 405                     raise Exception("Subsection %s not found at offset %x" % ( subsection['vmsd_name'], self.file.tell()))
 406                 name = self.file.readstr()
 407                 version_id = self.file.read32()
 408                 self.data[name] = VMSDSection(self.file, version_id, subsection, (name, 0))
 409                 self.data[name].read()
 410
 411     def getDictItem(self, value):
 412        # Strings would fall into the array category, treat
 413        # them specially
 414        if value.__class__ is ''.__class__:
 415            return value
 416
 417        try:
 418            return self.getDictOrderedDict(value)
 419        except:
 420            try:
 421                return self.getDictArray(value)
 422            except:
 423                try:
 424                    return value.getDict()
 425                except:
 426                    return value
 427
 428     def getDictArray(self, array):
 429         r = []
 430         for value in array:
 431            r.append(self.getDictItem(value))
 432         return r
 433
 434     def getDictOrderedDict(self, dict):
 435         r = collections.OrderedDict()
 436         for (key, value) in dict.items():
 437             r[key] = self.getDictItem(value)
 438         return r
 439
 440     def getDict(self):
 441         return self.getDictOrderedDict(self.data)
 442
 443 vmsd_field_readers = {
 444     "bool" : VMSDFieldBool,
 445     "int8" : VMSDFieldInt,
 446     "int16" : VMSDFieldInt,
 447     "int32" : VMSDFieldInt,
 448     "int32 equal" : VMSDFieldInt,
 449     "int32 le" : VMSDFieldIntLE,
 450     "int64" : VMSDFieldInt,
 451     "uint8" : VMSDFieldUInt,
 452     "uint16" : VMSDFieldUInt,
 453     "uint32" : VMSDFieldUInt,
 454     "uint32 equal" : VMSDFieldUInt,
 455     "uint64" : VMSDFieldUInt,
 456     "int64 equal" : VMSDFieldInt,
 457     "uint8 equal" : VMSDFieldInt,
 458     "uint16 equal" : VMSDFieldInt,
 459     "float64" : VMSDFieldGeneric,
 460     "timer" : VMSDFieldGeneric,
 461     "buffer" : VMSDFieldGeneric,
 462     "unused_buffer" : VMSDFieldGeneric,
 463     "bitmap" : VMSDFieldGeneric,
 464     "struct" : VMSDFieldStruct,
 465     "unknown" : VMSDFieldGeneric,
 466 }
 467
 468 class VMSDSection(VMSDFieldStruct):
 469     def __init__(self, file, version_id, device, section_key):
 470         self.file = file
 471         self.data = ""
 472         self.vmsd_name = ""
 473         self.section_key = section_key
 474         desc = device
 475         if 'vmsd_name' in device:
 476             self.vmsd_name = device['vmsd_name']
 477
 478         # A section really is nothing but a FieldStruct :)
 479         super(VMSDSection, self).__init__({ 'struct' : desc }, file)
 480
 481 ###############################################################################
 482
 483 class MigrationDump(object):
 484     QEMU_VM_FILE_MAGIC    = 0x5145564d
 485     QEMU_VM_FILE_VERSION  = 0x00000003
 486     QEMU_VM_EOF           = 0x00
 487     QEMU_VM_SECTION_START = 0x01
 488     QEMU_VM_SECTION_PART  = 0x02
 489     QEMU_VM_SECTION_END   = 0x03
 490     QEMU_VM_SECTION_FULL  = 0x04
 491     QEMU_VM_SUBSECTION    = 0x05
 492     QEMU_VM_VMDESCRIPTION = 0x06
 493     QEMU_VM_CONFIGURATION = 0x07
 494     QEMU_VM_SECTION_FOOTER= 0x7e
 495
 496     def __init__(self, filename):
 497         self.section_classes = { ( 'ram', 0 ) : [ RamSection, None ],
 498                                  ( 'spapr/htab', 0) : ( HTABSection, None ) }
 499         self.filename = filename
 500         self.vmsd_desc = None
 501
 502     def read(self, desc_only = False, dump_memory = False, write_memory = False):
 503         # Read in the whole file
 504         file = MigrationFile(self.filename)
 505
 506         # File magic
 507         data = file.read32()
 508         if data != self.QEMU_VM_FILE_MAGIC:
 509             raise Exception("Invalid file magic %x" % data)
 510
 511         # Version (has to be v3)
 512         data = file.read32()
 513         if data != self.QEMU_VM_FILE_VERSION:
 514             raise Exception("Invalid version number %d" % data)
 515
 516         self.load_vmsd_json(file)
 517
 518         # Read sections
 519         self.sections = collections.OrderedDict()
 520
 521         if desc_only:
 522             return
 523
 524         ramargs = {}
 525         ramargs['page_size'] = self.vmsd_desc['page_size']
 526         ramargs['dump_memory'] = dump_memory
 527         ramargs['write_memory'] = write_memory
 528         self.section_classes[('ram',0)][1] = ramargs
 529
 530         while True:
 531             section_type = file.read8()
 532             if section_type == self.QEMU_VM_EOF:
 533                 break
 534             elif section_type == self.QEMU_VM_CONFIGURATION:
 535                 section = ConfigurationSection(file)
 536                 section.read()
 537             elif section_type == self.QEMU_VM_SECTION_START or section_type == self.QEMU_VM_SECTION_FULL:
 538                 section_id = file.read32()
 539                 name = file.readstr()
 540                 instance_id = file.read32()
 541                 version_id = file.read32()
 542                 section_key = (name, instance_id)
 543                 classdesc = self.section_classes[section_key]
 544                 section = classdesc[0](file, version_id, classdesc[1], section_key)
 545                 self.sections[section_id] = section
 546                 section.read()
 547             elif section_type == self.QEMU_VM_SECTION_PART or section_type == self.QEMU_VM_SECTION_END:
 548                 section_id = file.read32()
 549                 self.sections[section_id].read()
 550             elif section_type == self.QEMU_VM_SECTION_FOOTER:
 551                 read_section_id = file.read32()
 552                 if read_section_id != section_id:
 553                     raise Exception("Mismatched section footer: %x vs %x" % (read_section_id, section_id))
 554             else:
 555                 raise Exception("Unknown section type: %d" % section_type)
 556         file.close()
 557
 558     def load_vmsd_json(self, file):
 559         vmsd_json = file.read_migration_debug_json()
 560         self.vmsd_desc = json.loads(vmsd_json, object_pairs_hook=collections.OrderedDict)
 561         for device in self.vmsd_desc['devices']:
 562             key = (device['name'], device['instance_id'])
 563             value = ( VMSDSection, device )
 564             self.section_classes[key] = value
 565
 566     def getDict(self):
 567         r = collections.OrderedDict()
 568         for (key, value) in self.sections.items():
 569            key = "%s (%d)" % ( value.section_key[0], key )
 570            r[key] = value.getDict()
 571         return r
 572
 573 ###############################################################################
 574
 575 class JSONEncoder(json.JSONEncoder):
 576     def default(self, o):
 577         if isinstance(o, VMSDFieldGeneric):
 578             return str(o)
 579         return json.JSONEncoder.default(self, o)
 580
 581 parser = argparse.ArgumentParser()
 582 parser.add_argument("-f", "--file", help='migration dump to read from', required=True)
 583 parser.add_argument("-m", "--memory", help='dump RAM contents as well', action='store_true')
 584 parser.add_argument("-d", "--dump", help='what to dump ("state" or "desc")', default='state')
 585 parser.add_argument("-x", "--extract", help='extract contents into individual files', action='store_true')
 586 args = parser.parse_args()
 587
 588 jsonenc = JSONEncoder(indent=4, separators=(',', ': '))
 589
 590 if args.extract:
 591     dump = MigrationDump(args.file)
 592
 593     dump.read(desc_only = True)
 594     print("desc.json")
 595     f = open("desc.json", "wb")
 596     f.truncate()
 597     f.write(jsonenc.encode(dump.vmsd_desc))
 598     f.close()
 599
 600     dump.read(write_memory = True)
 601     dict = dump.getDict()
 602     print("state.json")
 603     f = open("state.json", "wb")
 604     f.truncate()
 605     f.write(jsonenc.encode(dict))
 606     f.close()
 607 elif args.dump == "state":
 608     dump = MigrationDump(args.file)
 609     dump.read(dump_memory = args.memory)
 610     dict = dump.getDict()
 611     print(jsonenc.encode(dict))
 612 elif args.dump == "desc":
 613     dump = MigrationDump(args.file)
 614     dump.read(desc_only = True)
 615     print(jsonenc.encode(dump.vmsd_desc))
 616 else:
 617     raise Exception("Please specify either -x, -d state or -d dump")