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 def mkdir_p(path):
  29     try:
  30         os.makedirs(path)
  31     except OSError:
  32         pass
  33
  34
  35 class MigrationFile(object):
  36     def __init__(self, filename):
  37         self.filename = filename
  38         self.file = open(self.filename, "rb")
  39
  40     def read64(self):
  41         return int.from_bytes(self.file.read(8), byteorder='big', signed=True)
  42
  43     def read32(self):
  44         return int.from_bytes(self.file.read(4), byteorder='big', signed=True)
  45
  46     def read16(self):
  47         return int.from_bytes(self.file.read(2), byteorder='big', signed=True)
  48
  49     def read8(self):
  50         return int.from_bytes(self.file.read(1), byteorder='big', signed=True)
  51
  52     def readstr(self, len = None):
  53         return self.readvar(len).decode('utf-8')
  54
  55     def readvar(self, size = None):
  56         if size is None:
  57             size = self.read8()
  58         if size == 0:
  59             return ""
  60         value = self.file.read(size)
  61         if len(value) != size:
  62             raise Exception("Unexpected end of %s at 0x%x" % (self.filename, self.file.tell()))
  63         return value
  64
  65     def tell(self):
  66         return self.file.tell()
  67
  68     # The VMSD description is at the end of the file, after EOF. Look for
  69     # the last NULL byte, then for the beginning brace of JSON.
  70     def read_migration_debug_json(self):
  71         QEMU_VM_VMDESCRIPTION = 0x06
  72
  73         # Remember the offset in the file when we started
  74         entrypos = self.file.tell()
  75
  76         # Read the last 10MB
  77         self.file.seek(0, os.SEEK_END)
  78         endpos = self.file.tell()
  79         self.file.seek(max(-endpos, -10 * 1024 * 1024), os.SEEK_END)
  80         datapos = self.file.tell()
  81         data = self.file.read()
  82         # The full file read closed the file as well, reopen it
  83         self.file = open(self.filename, "rb")
  84
  85         # Find the last NULL byte, then the first brace after that. This should
  86         # be the beginning of our JSON data.
  87         nulpos = data.rfind(b'\0')
  88         jsonpos = data.find(b'{', nulpos)
  89
  90         # Check backwards from there and see whether we guessed right
  91         self.file.seek(datapos + jsonpos - 5, 0)
  92         if self.read8() != QEMU_VM_VMDESCRIPTION:
  93             raise Exception("No Debug Migration device found")
  94
  95         jsonlen = self.read32()
  96
  97         # Seek back to where we were at the beginning
  98         self.file.seek(entrypos, 0)
  99
 100         # explicit decode() needed for Python 3.5 compatibility
 101         return data[jsonpos:jsonpos + jsonlen].decode("utf-8")
 102
 103     def close(self):
 104         self.file.close()
 105
 106 class RamSection(object):
 107     RAM_SAVE_FLAG_COMPRESS = 0x02
 108     RAM_SAVE_FLAG_MEM_SIZE = 0x04
 109     RAM_SAVE_FLAG_PAGE     = 0x08
 110     RAM_SAVE_FLAG_EOS      = 0x10
 111     RAM_SAVE_FLAG_CONTINUE = 0x20
 112     RAM_SAVE_FLAG_XBZRLE   = 0x40
 113     RAM_SAVE_FLAG_HOOK     = 0x80
 114
 115     def __init__(self, file, version_id, ramargs, section_key):
 116         if version_id != 4:
 117             raise Exception("Unknown RAM version %d" % version_id)
 118
 119         self.file = file
 120         self.section_key = section_key
 121         self.TARGET_PAGE_SIZE = ramargs['page_size']
 122         self.dump_memory = ramargs['dump_memory']
 123         self.write_memory = ramargs['write_memory']
 124         self.sizeinfo = collections.OrderedDict()
 125         self.data = collections.OrderedDict()
 126         self.data['section sizes'] = self.sizeinfo
 127         self.name = ''
 128         if self.write_memory:
 129             self.files = { }
 130         if self.dump_memory:
 131             self.memory = collections.OrderedDict()
 132             self.data['memory'] = self.memory
 133
 134     def __repr__(self):
 135         return self.data.__repr__()
 136
 137     def __str__(self):
 138         return self.data.__str__()
 139
 140     def getDict(self):
 141         return self.data
 142
 143     def read(self):
 144         # Read all RAM sections
 145         while True:
 146             addr = self.file.read64()
 147             flags = addr & (self.TARGET_PAGE_SIZE - 1)
 148             addr &= ~(self.TARGET_PAGE_SIZE - 1)
 149
 150             if flags & self.RAM_SAVE_FLAG_MEM_SIZE:
 151                 while True:
 152                     namelen = self.file.read8()
 153                     # We assume that no RAM chunk is big enough to ever
 154                     # hit the first byte of the address, so when we see
 155                     # a zero here we know it has to be an address, not the
 156                     # length of the next block.
 157                     if namelen == 0:
 158                         self.file.file.seek(-1, 1)
 159                         break
 160                     self.name = self.file.readstr(len = namelen)
 161                     len = self.file.read64()
 162                     self.sizeinfo[self.name] = '0x%016x' % len
 163                     if self.write_memory:
 164                         print(self.name)
 165                         mkdir_p('./' + os.path.dirname(self.name))
 166                         f = open('./' + self.name, "wb")
 167                         f.truncate(0)
 168                         f.truncate(len)
 169                         self.files[self.name] = f
 170                 flags &= ~self.RAM_SAVE_FLAG_MEM_SIZE
 171
 172             if flags & self.RAM_SAVE_FLAG_COMPRESS:
 173                 if flags & self.RAM_SAVE_FLAG_CONTINUE:
 174                     flags &= ~self.RAM_SAVE_FLAG_CONTINUE
 175                 else:
 176                     self.name = self.file.readstr()
 177                 fill_char = self.file.read8()
 178                 # The page in question is filled with fill_char now
 179                 if self.write_memory and fill_char != 0:
 180                     self.files[self.name].seek(addr, os.SEEK_SET)
 181                     self.files[self.name].write(chr(fill_char) * self.TARGET_PAGE_SIZE)
 182                 if self.dump_memory:
 183                     self.memory['%s (0x%016x)' % (self.name, addr)] = 'Filled with 0x%02x' % fill_char
 184                 flags &= ~self.RAM_SAVE_FLAG_COMPRESS
 185             elif flags & self.RAM_SAVE_FLAG_PAGE:
 186                 if flags & self.RAM_SAVE_FLAG_CONTINUE:
 187                     flags &= ~self.RAM_SAVE_FLAG_CONTINUE
 188                 else:
 189                     self.name = self.file.readstr()
 190
 191                 if self.write_memory or self.dump_memory:
 192                     data = self.file.readvar(size = self.TARGET_PAGE_SIZE)
 193                 else: # Just skip RAM data
 194                     self.file.file.seek(self.TARGET_PAGE_SIZE, 1)
 195
 196                 if self.write_memory:
 197                     self.files[self.name].seek(addr, os.SEEK_SET)
 198                     self.files[self.name].write(data)
 199                 if self.dump_memory:
 200                     hexdata = " ".join("{0:02x}".format(ord(c)) for c in data)
 201                     self.memory['%s (0x%016x)' % (self.name, addr)] = hexdata
 202
 203                 flags &= ~self.RAM_SAVE_FLAG_PAGE
 204             elif flags & self.RAM_SAVE_FLAG_XBZRLE:
 205                 raise Exception("XBZRLE RAM compression is not supported yet")
 206             elif flags & self.RAM_SAVE_FLAG_HOOK:
 207                 raise Exception("RAM hooks don't make sense with files")
 208
 209             # End of RAM section
 210             if flags & self.RAM_SAVE_FLAG_EOS:
 211                 break
 212
 213             if flags != 0:
 214                 raise Exception("Unknown RAM flags: %x" % flags)
 215
 216     def __del__(self):
 217         if self.write_memory:
 218             for key in self.files:
 219                 self.files[key].close()
 220
 221
 222 class HTABSection(object):
 223     HASH_PTE_SIZE_64       = 16
 224
 225     def __init__(self, file, version_id, device, section_key):
 226         if version_id != 1:
 227             raise Exception("Unknown HTAB version %d" % version_id)
 228
 229         self.file = file
 230         self.section_key = section_key
 231
 232     def read(self):
 233
 234         header = self.file.read32()
 235
 236         if (header == -1):
 237             # "no HPT" encoding
 238             return
 239
 240         if (header > 0):
 241             # First section, just the hash shift
 242             return
 243
 244         # Read until end marker
 245         while True:
 246             index = self.file.read32()
 247             n_valid = self.file.read16()
 248             n_invalid = self.file.read16()
 249
 250             if index == 0 and n_valid == 0 and n_invalid == 0:
 251                 break
 252
 253             self.file.readvar(n_valid * self.HASH_PTE_SIZE_64)
 254
 255     def getDict(self):
 256         return ""
 257
 258
 259 class ConfigurationSection(object):
 260     def __init__(self, file):
 261         self.file = file
 262
 263     def read(self):
 264         name_len = self.file.read32()
 265         name = self.file.readstr(len = name_len)
 266
 267 class VMSDFieldGeneric(object):
 268     def __init__(self, desc, file):
 269         self.file = file
 270         self.desc = desc
 271         self.data = ""
 272
 273     def __repr__(self):
 274         return str(self.__str__())
 275
 276     def __str__(self):
 277         return " ".join("{0:02x}".format(c) for c in self.data)
 278
 279     def getDict(self):
 280         return self.__str__()
 281
 282     def read(self):
 283         size = int(self.desc['size'])
 284         self.data = self.file.readvar(size)
 285         return self.data
 286
 287 class VMSDFieldInt(VMSDFieldGeneric):
 288     def __init__(self, desc, file):
 289         super(VMSDFieldInt, self).__init__(desc, file)
 290         self.size = int(desc['size'])
 291         self.format = '0x%%0%dx' % (self.size * 2)
 292         self.sdtype = '>i%d' % self.size
 293         self.udtype = '>u%d' % self.size
 294
 295     def __repr__(self):
 296         if self.data < 0:
 297             return ('%s (%d)' % ((self.format % self.udata), self.data))
 298         else:
 299             return self.format % self.data
 300
 301     def __str__(self):
 302         return self.__repr__()
 303
 304     def getDict(self):
 305         return self.__str__()
 306
 307     def read(self):
 308         super(VMSDFieldInt, self).read()
 309         self.sdata = int.from_bytes(self.data, byteorder='big', signed=True)
 310         self.udata = int.from_bytes(self.data, byteorder='big', signed=False)
 311         self.data = self.sdata
 312         return self.data
 313
 314 class VMSDFieldUInt(VMSDFieldInt):
 315     def __init__(self, desc, file):
 316         super(VMSDFieldUInt, self).__init__(desc, file)
 317
 318     def read(self):
 319         super(VMSDFieldUInt, self).read()
 320         self.data = self.udata
 321         return self.data
 322
 323 class VMSDFieldIntLE(VMSDFieldInt):
 324     def __init__(self, desc, file):
 325         super(VMSDFieldIntLE, self).__init__(desc, file)
 326         self.dtype = '<i%d' % self.size
 327
 328 class VMSDFieldBool(VMSDFieldGeneric):
 329     def __init__(self, desc, file):
 330         super(VMSDFieldBool, self).__init__(desc, file)
 331
 332     def __repr__(self):
 333         return self.data.__repr__()
 334
 335     def __str__(self):
 336         return self.data.__str__()
 337
 338     def getDict(self):
 339         return self.data
 340
 341     def read(self):
 342         super(VMSDFieldBool, self).read()
 343         if self.data[0] == 0:
 344             self.data = False
 345         else:
 346             self.data = True
 347         return self.data
 348
 349 class VMSDFieldStruct(VMSDFieldGeneric):
 350     QEMU_VM_SUBSECTION    = 0x05
 351
 352     def __init__(self, desc, file):
 353         super(VMSDFieldStruct, self).__init__(desc, file)
 354         self.data = collections.OrderedDict()
 355
 356         # When we see compressed array elements, unfold them here
 357         new_fields = []
 358         for field in self.desc['struct']['fields']:
 359             if not 'array_len' in field:
 360                 new_fields.append(field)
 361                 continue
 362             array_len = field.pop('array_len')
 363             field['index'] = 0
 364             new_fields.append(field)
 365             for i in range(1, array_len):
 366                 c = field.copy()
 367                 c['index'] = i
 368                 new_fields.append(c)
 369
 370         self.desc['struct']['fields'] = new_fields
 371
 372     def __repr__(self):
 373         return self.data.__repr__()
 374
 375     def __str__(self):
 376         return self.data.__str__()
 377
 378     def read(self):
 379         for field in self.desc['struct']['fields']:
 380             try:
 381                 reader = vmsd_field_readers[field['type']]
 382             except:
 383                 reader = VMSDFieldGeneric
 384
 385             field['data'] = reader(field, self.file)
 386             field['data'].read()
 387
 388             if 'index' in field:
 389                 if field['name'] not in self.data:
 390                     self.data[field['name']] = []
 391                 a = self.data[field['name']]
 392                 if len(a) != int(field['index']):
 393                     raise Exception("internal index of data field unmatched (%d/%d)" % (len(a), int(field['index'])))
 394                 a.append(field['data'])
 395             else:
 396                 self.data[field['name']] = field['data']
 397
 398         if 'subsections' in self.desc['struct']:
 399             for subsection in self.desc['struct']['subsections']:
 400                 if self.file.read8() != self.QEMU_VM_SUBSECTION:
 401                     raise Exception("Subsection %s not found at offset %x" % ( subsection['vmsd_name'], self.file.tell()))
 402                 name = self.file.readstr()
 403                 version_id = self.file.read32()
 404                 self.data[name] = VMSDSection(self.file, version_id, subsection, (name, 0))
 405                 self.data[name].read()
 406
 407     def getDictItem(self, value):
 408        # Strings would fall into the array category, treat
 409        # them specially
 410        if value.__class__ is ''.__class__:
 411            return value
 412
 413        try:
 414            return self.getDictOrderedDict(value)
 415        except:
 416            try:
 417                return self.getDictArray(value)
 418            except:
 419                try:
 420                    return value.getDict()
 421                except:
 422                    return value
 423
 424     def getDictArray(self, array):
 425         r = []
 426         for value in array:
 427            r.append(self.getDictItem(value))
 428         return r
 429
 430     def getDictOrderedDict(self, dict):
 431         r = collections.OrderedDict()
 432         for (key, value) in dict.items():
 433             r[key] = self.getDictItem(value)
 434         return r
 435
 436     def getDict(self):
 437         return self.getDictOrderedDict(self.data)
 438
 439 vmsd_field_readers = {
 440     "bool" : VMSDFieldBool,
 441     "int8" : VMSDFieldInt,
 442     "int16" : VMSDFieldInt,
 443     "int32" : VMSDFieldInt,
 444     "int32 equal" : VMSDFieldInt,
 445     "int32 le" : VMSDFieldIntLE,
 446     "int64" : VMSDFieldInt,
 447     "uint8" : VMSDFieldUInt,
 448     "uint16" : VMSDFieldUInt,
 449     "uint32" : VMSDFieldUInt,
 450     "uint32 equal" : VMSDFieldUInt,
 451     "uint64" : VMSDFieldUInt,
 452     "int64 equal" : VMSDFieldInt,
 453     "uint8 equal" : VMSDFieldInt,
 454     "uint16 equal" : VMSDFieldInt,
 455     "float64" : VMSDFieldGeneric,
 456     "timer" : VMSDFieldGeneric,
 457     "buffer" : VMSDFieldGeneric,
 458     "unused_buffer" : VMSDFieldGeneric,
 459     "bitmap" : VMSDFieldGeneric,
 460     "struct" : VMSDFieldStruct,
 461     "unknown" : VMSDFieldGeneric,
 462 }
 463
 464 class VMSDSection(VMSDFieldStruct):
 465     def __init__(self, file, version_id, device, section_key):
 466         self.file = file
 467         self.data = ""
 468         self.vmsd_name = ""
 469         self.section_key = section_key
 470         desc = device
 471         if 'vmsd_name' in device:
 472             self.vmsd_name = device['vmsd_name']
 473
 474         # A section really is nothing but a FieldStruct :)
 475         super(VMSDSection, self).__init__({ 'struct' : desc }, file)
 476
 477 ###############################################################################
 478
 479 class MigrationDump(object):
 480     QEMU_VM_FILE_MAGIC    = 0x5145564d
 481     QEMU_VM_FILE_VERSION  = 0x00000003
 482     QEMU_VM_EOF           = 0x00
 483     QEMU_VM_SECTION_START = 0x01
 484     QEMU_VM_SECTION_PART  = 0x02
 485     QEMU_VM_SECTION_END   = 0x03
 486     QEMU_VM_SECTION_FULL  = 0x04
 487     QEMU_VM_SUBSECTION    = 0x05
 488     QEMU_VM_VMDESCRIPTION = 0x06
 489     QEMU_VM_CONFIGURATION = 0x07
 490     QEMU_VM_SECTION_FOOTER= 0x7e
 491
 492     def __init__(self, filename):
 493         self.section_classes = { ( 'ram', 0 ) : [ RamSection, None ],
 494                                  ( 'spapr/htab', 0) : ( HTABSection, None ) }
 495         self.filename = filename
 496         self.vmsd_desc = None
 497
 498     def read(self, desc_only = False, dump_memory = False, write_memory = False):
 499         # Read in the whole file
 500         file = MigrationFile(self.filename)
 501
 502         # File magic
 503         data = file.read32()
 504         if data != self.QEMU_VM_FILE_MAGIC:
 505             raise Exception("Invalid file magic %x" % data)
 506
 507         # Version (has to be v3)
 508         data = file.read32()
 509         if data != self.QEMU_VM_FILE_VERSION:
 510             raise Exception("Invalid version number %d" % data)
 511
 512         self.load_vmsd_json(file)
 513
 514         # Read sections
 515         self.sections = collections.OrderedDict()
 516
 517         if desc_only:
 518             return
 519
 520         ramargs = {}
 521         ramargs['page_size'] = self.vmsd_desc['page_size']
 522         ramargs['dump_memory'] = dump_memory
 523         ramargs['write_memory'] = write_memory
 524         self.section_classes[('ram',0)][1] = ramargs
 525
 526         while True:
 527             section_type = file.read8()
 528             if section_type == self.QEMU_VM_EOF:
 529                 break
 530             elif section_type == self.QEMU_VM_CONFIGURATION:
 531                 section = ConfigurationSection(file)
 532                 section.read()
 533             elif section_type == self.QEMU_VM_SECTION_START or section_type == self.QEMU_VM_SECTION_FULL:
 534                 section_id = file.read32()
 535                 name = file.readstr()
 536                 instance_id = file.read32()
 537                 version_id = file.read32()
 538                 section_key = (name, instance_id)
 539                 classdesc = self.section_classes[section_key]
 540                 section = classdesc[0](file, version_id, classdesc[1], section_key)
 541                 self.sections[section_id] = section
 542                 section.read()
 543             elif section_type == self.QEMU_VM_SECTION_PART or section_type == self.QEMU_VM_SECTION_END:
 544                 section_id = file.read32()
 545                 self.sections[section_id].read()
 546             elif section_type == self.QEMU_VM_SECTION_FOOTER:
 547                 read_section_id = file.read32()
 548                 if read_section_id != section_id:
 549                     raise Exception("Mismatched section footer: %x vs %x" % (read_section_id, section_id))
 550             else:
 551                 raise Exception("Unknown section type: %d" % section_type)
 552         file.close()
 553
 554     def load_vmsd_json(self, file):
 555         vmsd_json = file.read_migration_debug_json()
 556         self.vmsd_desc = json.loads(vmsd_json, object_pairs_hook=collections.OrderedDict)
 557         for device in self.vmsd_desc['devices']:
 558             key = (device['name'], device['instance_id'])
 559             value = ( VMSDSection, device )
 560             self.section_classes[key] = value
 561
 562     def getDict(self):
 563         r = collections.OrderedDict()
 564         for (key, value) in self.sections.items():
 565            key = "%s (%d)" % ( value.section_key[0], key )
 566            r[key] = value.getDict()
 567         return r
 568
 569 ###############################################################################
 570
 571 class JSONEncoder(json.JSONEncoder):
 572     def default(self, o):
 573         if isinstance(o, VMSDFieldGeneric):
 574             return str(o)
 575         return json.JSONEncoder.default(self, o)
 576
 577 parser = argparse.ArgumentParser()
 578 parser.add_argument("-f", "--file", help='migration dump to read from', required=True)
 579 parser.add_argument("-m", "--memory", help='dump RAM contents as well', action='store_true')
 580 parser.add_argument("-d", "--dump", help='what to dump ("state" or "desc")', default='state')
 581 parser.add_argument("-x", "--extract", help='extract contents into individual files', action='store_true')
 582 args = parser.parse_args()
 583
 584 jsonenc = JSONEncoder(indent=4, separators=(',', ': '))
 585
 586 if args.extract:
 587     dump = MigrationDump(args.file)
 588
 589     dump.read(desc_only = True)
 590     print("desc.json")
 591     f = open("desc.json", "wb")
 592     f.truncate()
 593     f.write(jsonenc.encode(dump.vmsd_desc))
 594     f.close()
 595
 596     dump.read(write_memory = True)
 597     dict = dump.getDict()
 598     print("state.json")
 599     f = open("state.json", "wb")
 600     f.truncate()
 601     f.write(jsonenc.encode(dict))
 602     f.close()
 603 elif args.dump == "state":
 604     dump = MigrationDump(args.file)
 605     dump.read(dump_memory = args.memory)
 606     dict = dump.getDict()
 607     print(jsonenc.encode(dict))
 608 elif args.dump == "desc":
 609     dump = MigrationDump(args.file)
 610     dump.read(desc_only = True)
 611     print(jsonenc.encode(dump.vmsd_desc))
 612 else:
 613     raise Exception("Please specify either -x, -d state or -d dump")