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