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 import numpy as np
  21 import json
  22 import os
  23 import argparse
  24 import collections
  25 import pprint
  26
  27 def mkdir_p(path):
  28     try:
  29         os.makedirs(path)
  30     except OSError:
  31         pass
  32
  33 class 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
 107 class 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
 223 class 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 == -1):
 238             # "no HPT" encoding
 239             return
 240
 241         if (header > 0):
 242             # First section, just the hash shift
 243             return
 244
 245         # Read until end marker
 246         while True:
 247             index = self.file.read32()
 248             n_valid = self.file.read16()
 249             n_invalid = self.file.read16()
 250
 251             if index == 0 and n_valid == 0 and n_invalid == 0:
 252                 break
 253
 254             self.file.readvar(n_valid * self.HASH_PTE_SIZE_64)
 255
 256     def getDict(self):
 257         return ""
 258
 259
 260 class ConfigurationSection(object):
 261     def __init__(self, file):
 262         self.file = file
 263
 264     def read(self):
 265         name_len = self.file.read32()
 266         name = self.file.readstr(len = name_len)
 267
 268 class VMSDFieldGeneric(object):
 269     def __init__(self, desc, file):
 270         self.file = file
 271         self.desc = desc
 272         self.data = ""
 273
 274     def __repr__(self):
 275         return str(self.__str__())
 276
 277     def __str__(self):
 278         return " ".join("{0:02x}".format(ord(c)) for c in self.data)
 279
 280     def getDict(self):
 281         return self.__str__()
 282
 283     def read(self):
 284         size = int(self.desc['size'])
 285         self.data = self.file.readvar(size)
 286         return self.data
 287
 288 class VMSDFieldInt(VMSDFieldGeneric):
 289     def __init__(self, desc, file):
 290         super(VMSDFieldInt, self).__init__(desc, file)
 291         self.size = int(desc['size'])
 292         self.format = '0x%%0%dx' % (self.size * 2)
 293         self.sdtype = '>i%d' % self.size
 294         self.udtype = '>u%d' % self.size
 295
 296     def __repr__(self):
 297         if self.data < 0:
 298             return ('%s (%d)' % ((self.format % self.udata), self.data))
 299         else:
 300             return self.format % self.data
 301
 302     def __str__(self):
 303         return self.__repr__()
 304
 305     def getDict(self):
 306         return self.__str__()
 307
 308     def read(self):
 309         super(VMSDFieldInt, self).read()
 310         self.sdata = np.fromstring(self.data, count=1, dtype=(self.sdtype))[0]
 311         self.udata = np.fromstring(self.data, count=1, dtype=(self.udtype))[0]
 312         self.data = self.sdata
 313         return self.data
 314
 315 class VMSDFieldUInt(VMSDFieldInt):
 316     def __init__(self, desc, file):
 317         super(VMSDFieldUInt, self).__init__(desc, file)
 318
 319     def read(self):
 320         super(VMSDFieldUInt, self).read()
 321         self.data = self.udata
 322         return self.data
 323
 324 class VMSDFieldIntLE(VMSDFieldInt):
 325     def __init__(self, desc, file):
 326         super(VMSDFieldIntLE, self).__init__(desc, file)
 327         self.dtype = '<i%d' % self.size
 328
 329 class VMSDFieldBool(VMSDFieldGeneric):
 330     def __init__(self, desc, file):
 331         super(VMSDFieldBool, self).__init__(desc, file)
 332
 333     def __repr__(self):
 334         return self.data.__repr__()
 335
 336     def __str__(self):
 337         return self.data.__str__()
 338
 339     def getDict(self):
 340         return self.data
 341
 342     def read(self):
 343         super(VMSDFieldBool, self).read()
 344         if self.data[0] == 0:
 345             self.data = False
 346         else:
 347             self.data = True
 348         return self.data
 349
 350 class VMSDFieldStruct(VMSDFieldGeneric):
 351     QEMU_VM_SUBSECTION    = 0x05
 352
 353     def __init__(self, desc, file):
 354         super(VMSDFieldStruct, self).__init__(desc, file)
 355         self.data = collections.OrderedDict()
 356
 357         # When we see compressed array elements, unfold them here
 358         new_fields = []
 359         for field in self.desc['struct']['fields']:
 360             if not 'array_len' in field:
 361                 new_fields.append(field)
 362                 continue
 363             array_len = field.pop('array_len')
 364             field['index'] = 0
 365             new_fields.append(field)
 366             for i in xrange(1, array_len):
 367                 c = field.copy()
 368                 c['index'] = i
 369                 new_fields.append(c)
 370
 371         self.desc['struct']['fields'] = new_fields
 372
 373     def __repr__(self):
 374         return self.data.__repr__()
 375
 376     def __str__(self):
 377         return self.data.__str__()
 378
 379     def read(self):
 380         for field in self.desc['struct']['fields']:
 381             try:
 382                 reader = vmsd_field_readers[field['type']]
 383             except:
 384                 reader = VMSDFieldGeneric
 385
 386             field['data'] = reader(field, self.file)
 387             field['data'].read()
 388
 389             if 'index' in field:
 390                 if field['name'] not in self.data:
 391                     self.data[field['name']] = []
 392                 a = self.data[field['name']]
 393                 if len(a) != int(field['index']):
 394                     raise Exception("internal index of data field unmatched (%d/%d)" % (len(a), int(field['index'])))
 395                 a.append(field['data'])
 396             else:
 397                 self.data[field['name']] = field['data']
 398
 399         if 'subsections' in self.desc['struct']:
 400             for subsection in self.desc['struct']['subsections']:
 401                 if self.file.read8() != self.QEMU_VM_SUBSECTION:
 402                     raise Exception("Subsection %s not found at offset %x" % ( subsection['vmsd_name'], self.file.tell()))
 403                 name = self.file.readstr()
 404                 version_id = self.file.read32()
 405                 self.data[name] = VMSDSection(self.file, version_id, subsection, (name, 0))
 406                 self.data[name].read()
 407
 408     def getDictItem(self, value):
 409        # Strings would fall into the array category, treat
 410        # them specially
 411        if value.__class__ is ''.__class__:
 412            return value
 413
 414        try:
 415            return self.getDictOrderedDict(value)
 416        except:
 417            try:
 418                return self.getDictArray(value)
 419            except:
 420                try:
 421                    return value.getDict()
 422                except:
 423                    return value
 424
 425     def getDictArray(self, array):
 426         r = []
 427         for value in array:
 428            r.append(self.getDictItem(value))
 429         return r
 430
 431     def getDictOrderedDict(self, dict):
 432         r = collections.OrderedDict()
 433         for (key, value) in dict.items():
 434             r[key] = self.getDictItem(value)
 435         return r
 436
 437     def getDict(self):
 438         return self.getDictOrderedDict(self.data)
 439
 440 vmsd_field_readers = {
 441     "bool" : VMSDFieldBool,
 442     "int8" : VMSDFieldInt,
 443     "int16" : VMSDFieldInt,
 444     "int32" : VMSDFieldInt,
 445     "int32 equal" : VMSDFieldInt,
 446     "int32 le" : VMSDFieldIntLE,
 447     "int64" : VMSDFieldInt,
 448     "uint8" : VMSDFieldUInt,
 449     "uint16" : VMSDFieldUInt,
 450     "uint32" : VMSDFieldUInt,
 451     "uint32 equal" : VMSDFieldUInt,
 452     "uint64" : VMSDFieldUInt,
 453     "int64 equal" : VMSDFieldInt,
 454     "uint8 equal" : VMSDFieldInt,
 455     "uint16 equal" : VMSDFieldInt,
 456     "float64" : VMSDFieldGeneric,
 457     "timer" : VMSDFieldGeneric,
 458     "buffer" : VMSDFieldGeneric,
 459     "unused_buffer" : VMSDFieldGeneric,
 460     "bitmap" : VMSDFieldGeneric,
 461     "struct" : VMSDFieldStruct,
 462     "unknown" : VMSDFieldGeneric,
 463 }
 464
 465 class VMSDSection(VMSDFieldStruct):
 466     def __init__(self, file, version_id, device, section_key):
 467         self.file = file
 468         self.data = ""
 469         self.vmsd_name = ""
 470         self.section_key = section_key
 471         desc = device
 472         if 'vmsd_name' in device:
 473             self.vmsd_name = device['vmsd_name']
 474
 475         # A section really is nothing but a FieldStruct :)
 476         super(VMSDSection, self).__init__({ 'struct' : desc }, file)
 477
 478 ###############################################################################
 479
 480 class MigrationDump(object):
 481     QEMU_VM_FILE_MAGIC    = 0x5145564d
 482     QEMU_VM_FILE_VERSION  = 0x00000003
 483     QEMU_VM_EOF           = 0x00
 484     QEMU_VM_SECTION_START = 0x01
 485     QEMU_VM_SECTION_PART  = 0x02
 486     QEMU_VM_SECTION_END   = 0x03
 487     QEMU_VM_SECTION_FULL  = 0x04
 488     QEMU_VM_SUBSECTION    = 0x05
 489     QEMU_VM_VMDESCRIPTION = 0x06
 490     QEMU_VM_CONFIGURATION = 0x07
 491     QEMU_VM_SECTION_FOOTER= 0x7e
 492
 493     def __init__(self, filename):
 494         self.section_classes = { ( 'ram', 0 ) : [ RamSection, None ],
 495                                  ( 'spapr/htab', 0) : ( HTABSection, None ) }
 496         self.filename = filename
 497         self.vmsd_desc = None
 498
 499     def read(self, desc_only = False, dump_memory = False, write_memory = False):
 500         # Read in the whole file
 501         file = MigrationFile(self.filename)
 502
 503         # File magic
 504         data = file.read32()
 505         if data != self.QEMU_VM_FILE_MAGIC:
 506             raise Exception("Invalid file magic %x" % data)
 507
 508         # Version (has to be v3)
 509         data = file.read32()
 510         if data != self.QEMU_VM_FILE_VERSION:
 511             raise Exception("Invalid version number %d" % data)
 512
 513         self.load_vmsd_json(file)
 514
 515         # Read sections
 516         self.sections = collections.OrderedDict()
 517
 518         if desc_only:
 519             return
 520
 521         ramargs = {}
 522         ramargs['page_size'] = self.vmsd_desc['page_size']
 523         ramargs['dump_memory'] = dump_memory
 524         ramargs['write_memory'] = write_memory
 525         self.section_classes[('ram',0)][1] = ramargs
 526
 527         while True:
 528             section_type = file.read8()
 529             if section_type == self.QEMU_VM_EOF:
 530                 break
 531             elif section_type == self.QEMU_VM_CONFIGURATION:
 532                 section = ConfigurationSection(file)
 533                 section.read()
 534             elif section_type == self.QEMU_VM_SECTION_START or section_type == self.QEMU_VM_SECTION_FULL:
 535                 section_id = file.read32()
 536                 name = file.readstr()
 537                 instance_id = file.read32()
 538                 version_id = file.read32()
 539                 section_key = (name, instance_id)
 540                 classdesc = self.section_classes[section_key]
 541                 section = classdesc[0](file, version_id, classdesc[1], section_key)
 542                 self.sections[section_id] = section
 543                 section.read()
 544             elif section_type == self.QEMU_VM_SECTION_PART or section_type == self.QEMU_VM_SECTION_END:
 545                 section_id = file.read32()
 546                 self.sections[section_id].read()
 547             elif section_type == self.QEMU_VM_SECTION_FOOTER:
 548                 read_section_id = file.read32()
 549                 if read_section_id != section_id:
 550                     raise Exception("Mismatched section footer: %x vs %x" % (read_section_id, section_id))
 551             else:
 552                 raise Exception("Unknown section type: %d" % section_type)
 553         file.close()
 554
 555     def load_vmsd_json(self, file):
 556         vmsd_json = file.read_migration_debug_json()
 557         self.vmsd_desc = json.loads(vmsd_json, object_pairs_hook=collections.OrderedDict)
 558         for device in self.vmsd_desc['devices']:
 559             key = (device['name'], device['instance_id'])
 560             value = ( VMSDSection, device )
 561             self.section_classes[key] = value
 562
 563     def getDict(self):
 564         r = collections.OrderedDict()
 565         for (key, value) in self.sections.items():
 566            key = "%s (%d)" % ( value.section_key[0], key )
 567            r[key] = value.getDict()
 568         return r
 569
 570 ###############################################################################
 571
 572 class JSONEncoder(json.JSONEncoder):
 573     def default(self, o):
 574         if isinstance(o, VMSDFieldGeneric):
 575             return str(o)
 576         return json.JSONEncoder.default(self, o)
 577
 578 parser = argparse.ArgumentParser()
 579 parser.add_argument("-f", "--file", help='migration dump to read from', required=True)
 580 parser.add_argument("-m", "--memory", help='dump RAM contents as well', action='store_true')
 581 parser.add_argument("-d", "--dump", help='what to dump ("state" or "desc")', default='state')
 582 parser.add_argument("-x", "--extract", help='extract contents into individual files', action='store_true')
 583 args = parser.parse_args()
 584
 585 jsonenc = JSONEncoder(indent=4, separators=(',', ': '))
 586
 587 if args.extract:
 588     dump = MigrationDump(args.file)
 589
 590     dump.read(desc_only = True)
 591     print "desc.json"
 592     f = open("desc.json", "wb")
 593     f.truncate()
 594     f.write(jsonenc.encode(dump.vmsd_desc))
 595     f.close()
 596
 597     dump.read(write_memory = True)
 598     dict = dump.getDict()
 599     print "state.json"
 600     f = open("state.json", "wb")
 601     f.truncate()
 602     f.write(jsonenc.encode(dict))
 603     f.close()
 604 elif args.dump == "state":
 605     dump = MigrationDump(args.file)
 606     dump.read(dump_memory = args.memory)
 607     dict = dump.getDict()
 608     print jsonenc.encode(dict)
 609 elif args.dump == "desc":
 610     dump = MigrationDump(args.file)
 611     dump.read(desc_only = True)
 612     print jsonenc.encode(dump.vmsd_desc)
 613 else:
 614     raise Exception("Please specify either -x, -d state or -d dump")