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 total_length = addr 155 while total_length > 0: 156 namelen = self.file.read8() 157 self.name = self.file.readstr(len = namelen) 158 len = self.file.read64() 159 total_length -= len 160 self.sizeinfo[self.name] = '0x%016x' % len 161 if self.write_memory: 162 print(self.name) 163 mkdir_p('./' + os.path.dirname(self.name)) 164 f = open('./' + self.name, "wb") 165 f.truncate(0) 166 f.truncate(len) 167 self.files[self.name] = f 168 if self.ignore_shared: 169 mr_addr = self.file.read64() 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 if flags & self.RAM_SAVE_FLAG_MULTIFD_FLUSH: 209 continue 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 224class 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 261class S390StorageAttributes(object): 262 STATTR_FLAG_EOS = 0x01 263 STATTR_FLAG_MORE = 0x02 264 STATTR_FLAG_ERROR = 0x04 265 STATTR_FLAG_DONE = 0x08 266 267 def __init__(self, file, version_id, device, section_key): 268 if version_id != 0: 269 raise Exception("Unknown storage_attributes version %d" % version_id) 270 271 self.file = file 272 self.section_key = section_key 273 274 def read(self): 275 while True: 276 addr_flags = self.file.read64() 277 flags = addr_flags & 0xfff 278 if (flags & (self.STATTR_FLAG_DONE | self.STATTR_FLAG_EOS)): 279 return 280 if (flags & self.STATTR_FLAG_ERROR): 281 raise Exception("Error in migration stream") 282 count = self.file.read64() 283 self.file.readvar(count) 284 285 def getDict(self): 286 return "" 287 288 289class ConfigurationSection(object): 290 def __init__(self, file, desc): 291 self.file = file 292 self.desc = desc 293 self.caps = [] 294 295 def parse_capabilities(self, vmsd_caps): 296 if not vmsd_caps: 297 return 298 299 ncaps = vmsd_caps.data['caps_count'].data 300 self.caps = vmsd_caps.data['capabilities'] 301 302 if type(self.caps) != list: 303 self.caps = [self.caps] 304 305 if len(self.caps) != ncaps: 306 raise Exception("Number of capabilities doesn't match " 307 "caps_count field") 308 309 def has_capability(self, cap): 310 return any([str(c) == cap for c in self.caps]) 311 312 def read(self): 313 if self.desc: 314 version_id = self.desc['version'] 315 section = VMSDSection(self.file, version_id, self.desc, 316 'configuration') 317 section.read() 318 self.parse_capabilities( 319 section.data.get("configuration/capabilities")) 320 else: 321 # backward compatibility for older streams that don't have 322 # the configuration section in the json 323 name_len = self.file.read32() 324 name = self.file.readstr(len = name_len) 325 326class VMSDFieldGeneric(object): 327 def __init__(self, desc, file): 328 self.file = file 329 self.desc = desc 330 self.data = "" 331 332 def __repr__(self): 333 return str(self.__str__()) 334 335 def __str__(self): 336 return " ".join("{0:02x}".format(c) for c in self.data) 337 338 def getDict(self): 339 return self.__str__() 340 341 def read(self): 342 size = int(self.desc['size']) 343 self.data = self.file.readvar(size) 344 return self.data 345 346class VMSDFieldCap(object): 347 def __init__(self, desc, file): 348 self.file = file 349 self.desc = desc 350 self.data = "" 351 352 def __repr__(self): 353 return self.data 354 355 def __str__(self): 356 return self.data 357 358 def read(self): 359 len = self.file.read8() 360 self.data = self.file.readstr(len) 361 362 363class VMSDFieldInt(VMSDFieldGeneric): 364 def __init__(self, desc, file): 365 super(VMSDFieldInt, self).__init__(desc, file) 366 self.size = int(desc['size']) 367 self.format = '0x%%0%dx' % (self.size * 2) 368 self.sdtype = '>i%d' % self.size 369 self.udtype = '>u%d' % self.size 370 371 def __repr__(self): 372 if self.data < 0: 373 return ('%s (%d)' % ((self.format % self.udata), self.data)) 374 else: 375 return self.format % self.data 376 377 def __str__(self): 378 return self.__repr__() 379 380 def getDict(self): 381 return self.__str__() 382 383 def read(self): 384 super(VMSDFieldInt, self).read() 385 self.sdata = int.from_bytes(self.data, byteorder='big', signed=True) 386 self.udata = int.from_bytes(self.data, byteorder='big', signed=False) 387 self.data = self.sdata 388 return self.data 389 390class VMSDFieldUInt(VMSDFieldInt): 391 def __init__(self, desc, file): 392 super(VMSDFieldUInt, self).__init__(desc, file) 393 394 def read(self): 395 super(VMSDFieldUInt, self).read() 396 self.data = self.udata 397 return self.data 398 399class VMSDFieldIntLE(VMSDFieldInt): 400 def __init__(self, desc, file): 401 super(VMSDFieldIntLE, self).__init__(desc, file) 402 self.dtype = '<i%d' % self.size 403 404class VMSDFieldBool(VMSDFieldGeneric): 405 def __init__(self, desc, file): 406 super(VMSDFieldBool, self).__init__(desc, file) 407 408 def __repr__(self): 409 return self.data.__repr__() 410 411 def __str__(self): 412 return self.data.__str__() 413 414 def getDict(self): 415 return self.data 416 417 def read(self): 418 super(VMSDFieldBool, self).read() 419 if self.data[0] == 0: 420 self.data = False 421 else: 422 self.data = True 423 return self.data 424 425class VMSDFieldStruct(VMSDFieldGeneric): 426 QEMU_VM_SUBSECTION = 0x05 427 428 def __init__(self, desc, file): 429 super(VMSDFieldStruct, self).__init__(desc, file) 430 self.data = collections.OrderedDict() 431 432 # When we see compressed array elements, unfold them here 433 new_fields = [] 434 for field in self.desc['struct']['fields']: 435 if not 'array_len' in field: 436 new_fields.append(field) 437 continue 438 array_len = field.pop('array_len') 439 field['index'] = 0 440 new_fields.append(field) 441 for i in range(1, array_len): 442 c = field.copy() 443 c['index'] = i 444 new_fields.append(c) 445 446 self.desc['struct']['fields'] = new_fields 447 448 def __repr__(self): 449 return self.data.__repr__() 450 451 def __str__(self): 452 return self.data.__str__() 453 454 def read(self): 455 for field in self.desc['struct']['fields']: 456 try: 457 reader = vmsd_field_readers[field['type']] 458 except: 459 reader = VMSDFieldGeneric 460 461 field['data'] = reader(field, self.file) 462 field['data'].read() 463 464 if 'index' in field: 465 if field['name'] not in self.data: 466 self.data[field['name']] = [] 467 a = self.data[field['name']] 468 if len(a) != int(field['index']): 469 raise Exception("internal index of data field unmatched (%d/%d)" % (len(a), int(field['index']))) 470 a.append(field['data']) 471 else: 472 self.data[field['name']] = field['data'] 473 474 if 'subsections' in self.desc['struct']: 475 for subsection in self.desc['struct']['subsections']: 476 if self.file.read8() != self.QEMU_VM_SUBSECTION: 477 raise Exception("Subsection %s not found at offset %x" % ( subsection['vmsd_name'], self.file.tell())) 478 name = self.file.readstr() 479 version_id = self.file.read32() 480 self.data[name] = VMSDSection(self.file, version_id, subsection, (name, 0)) 481 self.data[name].read() 482 483 def getDictItem(self, value): 484 # Strings would fall into the array category, treat 485 # them specially 486 if value.__class__ is ''.__class__: 487 return value 488 489 try: 490 return self.getDictOrderedDict(value) 491 except: 492 try: 493 return self.getDictArray(value) 494 except: 495 try: 496 return value.getDict() 497 except: 498 return value 499 500 def getDictArray(self, array): 501 r = [] 502 for value in array: 503 r.append(self.getDictItem(value)) 504 return r 505 506 def getDictOrderedDict(self, dict): 507 r = collections.OrderedDict() 508 for (key, value) in dict.items(): 509 r[key] = self.getDictItem(value) 510 return r 511 512 def getDict(self): 513 return self.getDictOrderedDict(self.data) 514 515vmsd_field_readers = { 516 "bool" : VMSDFieldBool, 517 "int8" : VMSDFieldInt, 518 "int16" : VMSDFieldInt, 519 "int32" : VMSDFieldInt, 520 "int32 equal" : VMSDFieldInt, 521 "int32 le" : VMSDFieldIntLE, 522 "int64" : VMSDFieldInt, 523 "uint8" : VMSDFieldUInt, 524 "uint16" : VMSDFieldUInt, 525 "uint32" : VMSDFieldUInt, 526 "uint32 equal" : VMSDFieldUInt, 527 "uint64" : VMSDFieldUInt, 528 "int64 equal" : VMSDFieldInt, 529 "uint8 equal" : VMSDFieldInt, 530 "uint16 equal" : VMSDFieldInt, 531 "float64" : VMSDFieldGeneric, 532 "timer" : VMSDFieldGeneric, 533 "buffer" : VMSDFieldGeneric, 534 "unused_buffer" : VMSDFieldGeneric, 535 "bitmap" : VMSDFieldGeneric, 536 "struct" : VMSDFieldStruct, 537 "capability": VMSDFieldCap, 538 "unknown" : VMSDFieldGeneric, 539} 540 541class VMSDSection(VMSDFieldStruct): 542 def __init__(self, file, version_id, device, section_key): 543 self.file = file 544 self.data = "" 545 self.vmsd_name = "" 546 self.section_key = section_key 547 desc = device 548 if 'vmsd_name' in device: 549 self.vmsd_name = device['vmsd_name'] 550 551 # A section really is nothing but a FieldStruct :) 552 super(VMSDSection, self).__init__({ 'struct' : desc }, file) 553 554############################################################################### 555 556class MigrationDump(object): 557 QEMU_VM_FILE_MAGIC = 0x5145564d 558 QEMU_VM_FILE_VERSION = 0x00000003 559 QEMU_VM_EOF = 0x00 560 QEMU_VM_SECTION_START = 0x01 561 QEMU_VM_SECTION_PART = 0x02 562 QEMU_VM_SECTION_END = 0x03 563 QEMU_VM_SECTION_FULL = 0x04 564 QEMU_VM_SUBSECTION = 0x05 565 QEMU_VM_VMDESCRIPTION = 0x06 566 QEMU_VM_CONFIGURATION = 0x07 567 QEMU_VM_SECTION_FOOTER= 0x7e 568 569 def __init__(self, filename): 570 self.section_classes = { 571 ( 'ram', 0 ) : [ RamSection, None ], 572 ( 's390-storage_attributes', 0 ) : [ S390StorageAttributes, None], 573 ( 'spapr/htab', 0) : ( HTABSection, None ) 574 } 575 self.filename = filename 576 self.vmsd_desc = None 577 578 def read(self, desc_only = False, dump_memory = False, write_memory = False): 579 # Read in the whole file 580 file = MigrationFile(self.filename) 581 582 # File magic 583 data = file.read32() 584 if data != self.QEMU_VM_FILE_MAGIC: 585 raise Exception("Invalid file magic %x" % data) 586 587 # Version (has to be v3) 588 data = file.read32() 589 if data != self.QEMU_VM_FILE_VERSION: 590 raise Exception("Invalid version number %d" % data) 591 592 self.load_vmsd_json(file) 593 594 # Read sections 595 self.sections = collections.OrderedDict() 596 597 if desc_only: 598 return 599 600 ramargs = {} 601 ramargs['page_size'] = self.vmsd_desc['page_size'] 602 ramargs['dump_memory'] = dump_memory 603 ramargs['write_memory'] = write_memory 604 ramargs['ignore_shared'] = False 605 self.section_classes[('ram',0)][1] = ramargs 606 607 while True: 608 section_type = file.read8() 609 if section_type == self.QEMU_VM_EOF: 610 break 611 elif section_type == self.QEMU_VM_CONFIGURATION: 612 config_desc = self.vmsd_desc.get('configuration') 613 section = ConfigurationSection(file, config_desc) 614 section.read() 615 ramargs['ignore_shared'] = section.has_capability('x-ignore-shared') 616 elif section_type == self.QEMU_VM_SECTION_START or section_type == self.QEMU_VM_SECTION_FULL: 617 section_id = file.read32() 618 name = file.readstr() 619 instance_id = file.read32() 620 version_id = file.read32() 621 section_key = (name, instance_id) 622 classdesc = self.section_classes[section_key] 623 section = classdesc[0](file, version_id, classdesc[1], section_key) 624 self.sections[section_id] = section 625 section.read() 626 elif section_type == self.QEMU_VM_SECTION_PART or section_type == self.QEMU_VM_SECTION_END: 627 section_id = file.read32() 628 self.sections[section_id].read() 629 elif section_type == self.QEMU_VM_SECTION_FOOTER: 630 read_section_id = file.read32() 631 if read_section_id != section_id: 632 raise Exception("Mismatched section footer: %x vs %x" % (read_section_id, section_id)) 633 else: 634 raise Exception("Unknown section type: %d" % section_type) 635 file.close() 636 637 def load_vmsd_json(self, file): 638 vmsd_json = file.read_migration_debug_json() 639 self.vmsd_desc = json.loads(vmsd_json, object_pairs_hook=collections.OrderedDict) 640 for device in self.vmsd_desc['devices']: 641 key = (device['name'], device['instance_id']) 642 value = ( VMSDSection, device ) 643 self.section_classes[key] = value 644 645 def getDict(self): 646 r = collections.OrderedDict() 647 for (key, value) in self.sections.items(): 648 key = "%s (%d)" % ( value.section_key[0], key ) 649 r[key] = value.getDict() 650 return r 651 652############################################################################### 653 654class JSONEncoder(json.JSONEncoder): 655 def default(self, o): 656 if isinstance(o, VMSDFieldGeneric): 657 return str(o) 658 return json.JSONEncoder.default(self, o) 659 660parser = argparse.ArgumentParser() 661parser.add_argument("-f", "--file", help='migration dump to read from', required=True) 662parser.add_argument("-m", "--memory", help='dump RAM contents as well', action='store_true') 663parser.add_argument("-d", "--dump", help='what to dump ("state" or "desc")', default='state') 664parser.add_argument("-x", "--extract", help='extract contents into individual files', action='store_true') 665args = parser.parse_args() 666 667jsonenc = JSONEncoder(indent=4, separators=(',', ': ')) 668 669if args.extract: 670 dump = MigrationDump(args.file) 671 672 dump.read(desc_only = True) 673 print("desc.json") 674 f = open("desc.json", "w") 675 f.truncate() 676 f.write(jsonenc.encode(dump.vmsd_desc)) 677 f.close() 678 679 dump.read(write_memory = True) 680 dict = dump.getDict() 681 print("state.json") 682 f = open("state.json", "w") 683 f.truncate() 684 f.write(jsonenc.encode(dict)) 685 f.close() 686elif args.dump == "state": 687 dump = MigrationDump(args.file) 688 dump.read(dump_memory = args.memory) 689 dict = dump.getDict() 690 print(jsonenc.encode(dict)) 691elif args.dump == "desc": 692 dump = MigrationDump(args.file) 693 dump.read(desc_only = True) 694 print(jsonenc.encode(dump.vmsd_desc)) 695else: 696 raise Exception("Please specify either -x, -d state or -d desc") 697