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 2.1 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 json 21import os 22import argparse 23import collections 24import struct 25import sys 26 27 28def mkdir_p(path): 29 try: 30 os.makedirs(path) 31 except OSError: 32 pass 33 34 35class 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=False) 42 43 def read32(self): 44 return int.from_bytes(self.file.read(4), byteorder='big', signed=False) 45 46 def read16(self): 47 return int.from_bytes(self.file.read(2), byteorder='big', signed=False) 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 106class 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 RAM_SAVE_FLAG_COMPRESS_PAGE = 0x100 115 RAM_SAVE_FLAG_MULTIFD_FLUSH = 0x200 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.ignore_shared = ramargs['ignore_shared'] 127 self.sizeinfo = collections.OrderedDict() 128 self.data = collections.OrderedDict() 129 self.data['section sizes'] = self.sizeinfo 130 self.name = '' 131 if self.write_memory: 132 self.files = { } 133 if self.dump_memory: 134 self.memory = collections.OrderedDict() 135 self.data['memory'] = self.memory 136 137 def __repr__(self): 138 return self.data.__repr__() 139 140 def __str__(self): 141 return self.data.__str__() 142 143 def getDict(self): 144 return self.data 145 146 def read(self): 147 # Read all RAM sections 148 while True: 149 addr = self.file.read64() 150 flags = addr & (self.TARGET_PAGE_SIZE - 1) 151 addr &= ~(self.TARGET_PAGE_SIZE - 1) 152 153 if flags & self.RAM_SAVE_FLAG_MEM_SIZE: 154 while True: 155 namelen = self.file.read8() 156 # We assume that no RAM chunk is big enough to ever 157 # hit the first byte of the address, so when we see 158 # a zero here we know it has to be an address, not the 159 # length of the next block. 160 if namelen == 0: 161 self.file.file.seek(-1, 1) 162 break 163 self.name = self.file.readstr(len = namelen) 164 len = self.file.read64() 165 self.sizeinfo[self.name] = '0x%016x' % len 166 if self.write_memory: 167 print(self.name) 168 mkdir_p('./' + os.path.dirname(self.name)) 169 f = open('./' + self.name, "wb") 170 f.truncate(0) 171 f.truncate(len) 172 self.files[self.name] = f 173 if self.ignore_shared: 174 mr_addr = self.file.read64() 175 flags &= ~self.RAM_SAVE_FLAG_MEM_SIZE 176 177 if flags & self.RAM_SAVE_FLAG_COMPRESS: 178 if flags & self.RAM_SAVE_FLAG_CONTINUE: 179 flags &= ~self.RAM_SAVE_FLAG_CONTINUE 180 else: 181 self.name = self.file.readstr() 182 fill_char = self.file.read8() 183 # The page in question is filled with fill_char now 184 if self.write_memory and fill_char != 0: 185 self.files[self.name].seek(addr, os.SEEK_SET) 186 self.files[self.name].write(chr(fill_char) * self.TARGET_PAGE_SIZE) 187 if self.dump_memory: 188 self.memory['%s (0x%016x)' % (self.name, addr)] = 'Filled with 0x%02x' % fill_char 189 flags &= ~self.RAM_SAVE_FLAG_COMPRESS 190 elif flags & self.RAM_SAVE_FLAG_PAGE: 191 if flags & self.RAM_SAVE_FLAG_CONTINUE: 192 flags &= ~self.RAM_SAVE_FLAG_CONTINUE 193 else: 194 self.name = self.file.readstr() 195 196 if self.write_memory or self.dump_memory: 197 data = self.file.readvar(size = self.TARGET_PAGE_SIZE) 198 else: # Just skip RAM data 199 self.file.file.seek(self.TARGET_PAGE_SIZE, 1) 200 201 if self.write_memory: 202 self.files[self.name].seek(addr, os.SEEK_SET) 203 self.files[self.name].write(data) 204 if self.dump_memory: 205 hexdata = " ".join("{0:02x}".format(ord(c)) for c in data) 206 self.memory['%s (0x%016x)' % (self.name, addr)] = hexdata 207 208 flags &= ~self.RAM_SAVE_FLAG_PAGE 209 elif flags & self.RAM_SAVE_FLAG_XBZRLE: 210 raise Exception("XBZRLE RAM compression is not supported yet") 211 elif flags & self.RAM_SAVE_FLAG_HOOK: 212 raise Exception("RAM hooks don't make sense with files") 213 if flags & self.RAM_SAVE_FLAG_MULTIFD_FLUSH: 214 continue 215 216 # End of RAM section 217 if flags & self.RAM_SAVE_FLAG_EOS: 218 break 219 220 if flags != 0: 221 raise Exception("Unknown RAM flags: %x" % flags) 222 223 def __del__(self): 224 if self.write_memory: 225 for key in self.files: 226 self.files[key].close() 227 228 229class HTABSection(object): 230 HASH_PTE_SIZE_64 = 16 231 232 def __init__(self, file, version_id, device, section_key): 233 if version_id != 1: 234 raise Exception("Unknown HTAB version %d" % version_id) 235 236 self.file = file 237 self.section_key = section_key 238 239 def read(self): 240 241 header = self.file.read32() 242 243 if (header == -1): 244 # "no HPT" encoding 245 return 246 247 if (header > 0): 248 # First section, just the hash shift 249 return 250 251 # Read until end marker 252 while True: 253 index = self.file.read32() 254 n_valid = self.file.read16() 255 n_invalid = self.file.read16() 256 257 if index == 0 and n_valid == 0 and n_invalid == 0: 258 break 259 260 self.file.readvar(n_valid * self.HASH_PTE_SIZE_64) 261 262 def getDict(self): 263 return "" 264 265 266class ConfigurationSection(object): 267 def __init__(self, file, desc): 268 self.file = file 269 self.desc = desc 270 self.caps = [] 271 272 def parse_capabilities(self, vmsd_caps): 273 if not vmsd_caps: 274 return 275 276 ncaps = vmsd_caps.data['caps_count'].data 277 self.caps = vmsd_caps.data['capabilities'] 278 279 if type(self.caps) != list: 280 self.caps = [self.caps] 281 282 if len(self.caps) != ncaps: 283 raise Exception("Number of capabilities doesn't match " 284 "caps_count field") 285 286 def has_capability(self, cap): 287 return any([str(c) == cap for c in self.caps]) 288 289 def read(self): 290 if self.desc: 291 version_id = self.desc['version'] 292 section = VMSDSection(self.file, version_id, self.desc, 293 'configuration') 294 section.read() 295 self.parse_capabilities( 296 section.data.get("configuration/capabilities")) 297 else: 298 # backward compatibility for older streams that don't have 299 # the configuration section in the json 300 name_len = self.file.read32() 301 name = self.file.readstr(len = name_len) 302 303class VMSDFieldGeneric(object): 304 def __init__(self, desc, file): 305 self.file = file 306 self.desc = desc 307 self.data = "" 308 309 def __repr__(self): 310 return str(self.__str__()) 311 312 def __str__(self): 313 return " ".join("{0:02x}".format(c) for c in self.data) 314 315 def getDict(self): 316 return self.__str__() 317 318 def read(self): 319 size = int(self.desc['size']) 320 self.data = self.file.readvar(size) 321 return self.data 322 323class VMSDFieldCap(object): 324 def __init__(self, desc, file): 325 self.file = file 326 self.desc = desc 327 self.data = "" 328 329 def __repr__(self): 330 return self.data 331 332 def __str__(self): 333 return self.data 334 335 def read(self): 336 len = self.file.read8() 337 self.data = self.file.readstr(len) 338 339 340class VMSDFieldInt(VMSDFieldGeneric): 341 def __init__(self, desc, file): 342 super(VMSDFieldInt, self).__init__(desc, file) 343 self.size = int(desc['size']) 344 self.format = '0x%%0%dx' % (self.size * 2) 345 self.sdtype = '>i%d' % self.size 346 self.udtype = '>u%d' % self.size 347 348 def __repr__(self): 349 if self.data < 0: 350 return ('%s (%d)' % ((self.format % self.udata), self.data)) 351 else: 352 return self.format % self.data 353 354 def __str__(self): 355 return self.__repr__() 356 357 def getDict(self): 358 return self.__str__() 359 360 def read(self): 361 super(VMSDFieldInt, self).read() 362 self.sdata = int.from_bytes(self.data, byteorder='big', signed=True) 363 self.udata = int.from_bytes(self.data, byteorder='big', signed=False) 364 self.data = self.sdata 365 return self.data 366 367class VMSDFieldUInt(VMSDFieldInt): 368 def __init__(self, desc, file): 369 super(VMSDFieldUInt, self).__init__(desc, file) 370 371 def read(self): 372 super(VMSDFieldUInt, self).read() 373 self.data = self.udata 374 return self.data 375 376class VMSDFieldIntLE(VMSDFieldInt): 377 def __init__(self, desc, file): 378 super(VMSDFieldIntLE, self).__init__(desc, file) 379 self.dtype = '<i%d' % self.size 380 381class VMSDFieldBool(VMSDFieldGeneric): 382 def __init__(self, desc, file): 383 super(VMSDFieldBool, self).__init__(desc, file) 384 385 def __repr__(self): 386 return self.data.__repr__() 387 388 def __str__(self): 389 return self.data.__str__() 390 391 def getDict(self): 392 return self.data 393 394 def read(self): 395 super(VMSDFieldBool, self).read() 396 if self.data[0] == 0: 397 self.data = False 398 else: 399 self.data = True 400 return self.data 401 402class VMSDFieldStruct(VMSDFieldGeneric): 403 QEMU_VM_SUBSECTION = 0x05 404 405 def __init__(self, desc, file): 406 super(VMSDFieldStruct, self).__init__(desc, file) 407 self.data = collections.OrderedDict() 408 409 # When we see compressed array elements, unfold them here 410 new_fields = [] 411 for field in self.desc['struct']['fields']: 412 if not 'array_len' in field: 413 new_fields.append(field) 414 continue 415 array_len = field.pop('array_len') 416 field['index'] = 0 417 new_fields.append(field) 418 for i in range(1, array_len): 419 c = field.copy() 420 c['index'] = i 421 new_fields.append(c) 422 423 self.desc['struct']['fields'] = new_fields 424 425 def __repr__(self): 426 return self.data.__repr__() 427 428 def __str__(self): 429 return self.data.__str__() 430 431 def read(self): 432 for field in self.desc['struct']['fields']: 433 try: 434 reader = vmsd_field_readers[field['type']] 435 except: 436 reader = VMSDFieldGeneric 437 438 field['data'] = reader(field, self.file) 439 field['data'].read() 440 441 if 'index' in field: 442 if field['name'] not in self.data: 443 self.data[field['name']] = [] 444 a = self.data[field['name']] 445 if len(a) != int(field['index']): 446 raise Exception("internal index of data field unmatched (%d/%d)" % (len(a), int(field['index']))) 447 a.append(field['data']) 448 else: 449 self.data[field['name']] = field['data'] 450 451 if 'subsections' in self.desc['struct']: 452 for subsection in self.desc['struct']['subsections']: 453 if self.file.read8() != self.QEMU_VM_SUBSECTION: 454 raise Exception("Subsection %s not found at offset %x" % ( subsection['vmsd_name'], self.file.tell())) 455 name = self.file.readstr() 456 version_id = self.file.read32() 457 self.data[name] = VMSDSection(self.file, version_id, subsection, (name, 0)) 458 self.data[name].read() 459 460 def getDictItem(self, value): 461 # Strings would fall into the array category, treat 462 # them specially 463 if value.__class__ is ''.__class__: 464 return value 465 466 try: 467 return self.getDictOrderedDict(value) 468 except: 469 try: 470 return self.getDictArray(value) 471 except: 472 try: 473 return value.getDict() 474 except: 475 return value 476 477 def getDictArray(self, array): 478 r = [] 479 for value in array: 480 r.append(self.getDictItem(value)) 481 return r 482 483 def getDictOrderedDict(self, dict): 484 r = collections.OrderedDict() 485 for (key, value) in dict.items(): 486 r[key] = self.getDictItem(value) 487 return r 488 489 def getDict(self): 490 return self.getDictOrderedDict(self.data) 491 492vmsd_field_readers = { 493 "bool" : VMSDFieldBool, 494 "int8" : VMSDFieldInt, 495 "int16" : VMSDFieldInt, 496 "int32" : VMSDFieldInt, 497 "int32 equal" : VMSDFieldInt, 498 "int32 le" : VMSDFieldIntLE, 499 "int64" : VMSDFieldInt, 500 "uint8" : VMSDFieldUInt, 501 "uint16" : VMSDFieldUInt, 502 "uint32" : VMSDFieldUInt, 503 "uint32 equal" : VMSDFieldUInt, 504 "uint64" : VMSDFieldUInt, 505 "int64 equal" : VMSDFieldInt, 506 "uint8 equal" : VMSDFieldInt, 507 "uint16 equal" : VMSDFieldInt, 508 "float64" : VMSDFieldGeneric, 509 "timer" : VMSDFieldGeneric, 510 "buffer" : VMSDFieldGeneric, 511 "unused_buffer" : VMSDFieldGeneric, 512 "bitmap" : VMSDFieldGeneric, 513 "struct" : VMSDFieldStruct, 514 "capability": VMSDFieldCap, 515 "unknown" : VMSDFieldGeneric, 516} 517 518class VMSDSection(VMSDFieldStruct): 519 def __init__(self, file, version_id, device, section_key): 520 self.file = file 521 self.data = "" 522 self.vmsd_name = "" 523 self.section_key = section_key 524 desc = device 525 if 'vmsd_name' in device: 526 self.vmsd_name = device['vmsd_name'] 527 528 # A section really is nothing but a FieldStruct :) 529 super(VMSDSection, self).__init__({ 'struct' : desc }, file) 530 531############################################################################### 532 533class MigrationDump(object): 534 QEMU_VM_FILE_MAGIC = 0x5145564d 535 QEMU_VM_FILE_VERSION = 0x00000003 536 QEMU_VM_EOF = 0x00 537 QEMU_VM_SECTION_START = 0x01 538 QEMU_VM_SECTION_PART = 0x02 539 QEMU_VM_SECTION_END = 0x03 540 QEMU_VM_SECTION_FULL = 0x04 541 QEMU_VM_SUBSECTION = 0x05 542 QEMU_VM_VMDESCRIPTION = 0x06 543 QEMU_VM_CONFIGURATION = 0x07 544 QEMU_VM_SECTION_FOOTER= 0x7e 545 546 def __init__(self, filename): 547 self.section_classes = { ( 'ram', 0 ) : [ RamSection, None ], 548 ( 'spapr/htab', 0) : ( HTABSection, None ) } 549 self.filename = filename 550 self.vmsd_desc = None 551 552 def read(self, desc_only = False, dump_memory = False, write_memory = False): 553 # Read in the whole file 554 file = MigrationFile(self.filename) 555 556 # File magic 557 data = file.read32() 558 if data != self.QEMU_VM_FILE_MAGIC: 559 raise Exception("Invalid file magic %x" % data) 560 561 # Version (has to be v3) 562 data = file.read32() 563 if data != self.QEMU_VM_FILE_VERSION: 564 raise Exception("Invalid version number %d" % data) 565 566 self.load_vmsd_json(file) 567 568 # Read sections 569 self.sections = collections.OrderedDict() 570 571 if desc_only: 572 return 573 574 ramargs = {} 575 ramargs['page_size'] = self.vmsd_desc['page_size'] 576 ramargs['dump_memory'] = dump_memory 577 ramargs['write_memory'] = write_memory 578 ramargs['ignore_shared'] = False 579 self.section_classes[('ram',0)][1] = ramargs 580 581 while True: 582 section_type = file.read8() 583 if section_type == self.QEMU_VM_EOF: 584 break 585 elif section_type == self.QEMU_VM_CONFIGURATION: 586 config_desc = self.vmsd_desc.get('configuration') 587 section = ConfigurationSection(file, config_desc) 588 section.read() 589 ramargs['ignore_shared'] = section.has_capability('x-ignore-shared') 590 elif section_type == self.QEMU_VM_SECTION_START or section_type == self.QEMU_VM_SECTION_FULL: 591 section_id = file.read32() 592 name = file.readstr() 593 instance_id = file.read32() 594 version_id = file.read32() 595 section_key = (name, instance_id) 596 classdesc = self.section_classes[section_key] 597 section = classdesc[0](file, version_id, classdesc[1], section_key) 598 self.sections[section_id] = section 599 section.read() 600 elif section_type == self.QEMU_VM_SECTION_PART or section_type == self.QEMU_VM_SECTION_END: 601 section_id = file.read32() 602 self.sections[section_id].read() 603 elif section_type == self.QEMU_VM_SECTION_FOOTER: 604 read_section_id = file.read32() 605 if read_section_id != section_id: 606 raise Exception("Mismatched section footer: %x vs %x" % (read_section_id, section_id)) 607 else: 608 raise Exception("Unknown section type: %d" % section_type) 609 file.close() 610 611 def load_vmsd_json(self, file): 612 vmsd_json = file.read_migration_debug_json() 613 self.vmsd_desc = json.loads(vmsd_json, object_pairs_hook=collections.OrderedDict) 614 for device in self.vmsd_desc['devices']: 615 key = (device['name'], device['instance_id']) 616 value = ( VMSDSection, device ) 617 self.section_classes[key] = value 618 619 def getDict(self): 620 r = collections.OrderedDict() 621 for (key, value) in self.sections.items(): 622 key = "%s (%d)" % ( value.section_key[0], key ) 623 r[key] = value.getDict() 624 return r 625 626############################################################################### 627 628class JSONEncoder(json.JSONEncoder): 629 def default(self, o): 630 if isinstance(o, VMSDFieldGeneric): 631 return str(o) 632 return json.JSONEncoder.default(self, o) 633 634parser = argparse.ArgumentParser() 635parser.add_argument("-f", "--file", help='migration dump to read from', required=True) 636parser.add_argument("-m", "--memory", help='dump RAM contents as well', action='store_true') 637parser.add_argument("-d", "--dump", help='what to dump ("state" or "desc")', default='state') 638parser.add_argument("-x", "--extract", help='extract contents into individual files', action='store_true') 639args = parser.parse_args() 640 641jsonenc = JSONEncoder(indent=4, separators=(',', ': ')) 642 643if args.extract: 644 dump = MigrationDump(args.file) 645 646 dump.read(desc_only = True) 647 print("desc.json") 648 f = open("desc.json", "w") 649 f.truncate() 650 f.write(jsonenc.encode(dump.vmsd_desc)) 651 f.close() 652 653 dump.read(write_memory = True) 654 dict = dump.getDict() 655 print("state.json") 656 f = open("state.json", "w") 657 f.truncate() 658 f.write(jsonenc.encode(dict)) 659 f.close() 660elif args.dump == "state": 661 dump = MigrationDump(args.file) 662 dump.read(dump_memory = args.memory) 663 dict = dump.getDict() 664 print(jsonenc.encode(dict)) 665elif args.dump == "desc": 666 dump = MigrationDump(args.file) 667 dump.read(desc_only = True) 668 print(jsonenc.encode(dump.vmsd_desc)) 669else: 670 raise Exception("Please specify either -x, -d state or -d desc") 671