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