diff mbox

[4/4] Add migration stream analyzation script

Message ID 1419289424-110278-5-git-send-email-agraf@suse.de
State New
Headers show

Commit Message

Alexander Graf Dec. 22, 2014, 11:03 p.m. UTC
This patch adds a python tool to the scripts directory that can read
a dumped migration stream if it contains the JSON description of the
device states. I constructs a human readable JSON stream out of it.

It's very simple to use:

  $ qemu-system-x86_64
    (qemu) migrate "exec:cat > mig"
  $ ./scripts/analyze_migration.py -f mig

Signed-off-by: Alexander Graf <agraf@suse.de>

---

v1 -> v2:

  - Remove support for multiple vmsd versions
  - Add support for HTAB
  - Add support for array_len
  - Move to new naming schema for fields
  - Use dynamic page size
---
 scripts/analyze-migration.py | 529 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 529 insertions(+)
 create mode 100755 scripts/analyze-migration.py
diff mbox

Patch

diff --git a/scripts/analyze-migration.py b/scripts/analyze-migration.py
new file mode 100755
index 0000000..2878011
--- /dev/null
+++ b/scripts/analyze-migration.py
@@ -0,0 +1,529 @@ 
+#!/usr/bin/env python
+#
+#  Migration Stream Analyzer
+#
+#  Copyright (c) 2013 Alexander Graf <agraf@suse.de>
+#
+# This library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2 of the License, or (at your option) any later version.
+#
+# This library is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with this library; if not, see <http://www.gnu.org/licenses/>.
+
+import numpy as np
+import json
+import os
+import argparse
+import collections
+import pprint
+
+class MigrationFile(object):
+    def __init__(self, filename):
+        self.filename = filename
+        self.file = open(self.filename, "rb")
+
+    def read64(self):
+        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i8')[0])
+
+    def read32(self):
+        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i4')[0])
+
+    def read16(self):
+        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i2')[0])
+
+    def read8(self):
+        return np.asscalar(np.fromfile(self.file, count=1, dtype='>i1')[0])
+
+    def readstr(self, len = None):
+        if len is None:
+            len = self.read8()
+        if len == 0:
+            return ""
+        return np.fromfile(self.file, count=1, dtype=('S%d' % len))[0]
+
+    def readvar(self, size = None):
+        if size is None:
+            size = self.read8()
+        if size == 0:
+            return ""
+        value = self.file.read(size)
+        if len(value) != size:
+            raise Exception("Unexpected end of %s at 0x%x" % (self.filename, self.file.tell()))
+        return value
+
+    # The VMSD description is at the end of the file, after EOF. Look for
+    # the last NULL byte, then for the beginning brace of JSON.
+    def read_migration_debug_json(self):
+        QEMU_VM_VMDESCRIPTION = 0x06
+
+        # Remember the offset in the file when we started
+        entrypos = self.file.tell()
+
+        # Read the last 10MB
+        self.file.seek(0, os.SEEK_END)
+        endpos = self.file.tell()
+        self.file.seek(max(-endpos, -10 * 1024 * 1024), os.SEEK_END)
+        datapos = self.file.tell()
+        data = self.file.read()
+        # The full file read closed the file as well, reopen it
+        self.file = open(self.filename, "rb")
+
+        # Find the last NULL byte, then the first brace after that. This should
+        # be the beginning of our JSON data.
+        nulpos = data.rfind("\0")
+        jsonpos = data.find("{", nulpos)
+
+        # Check backwards from there and see whether we guessed right
+        self.file.seek(datapos + jsonpos - 5, 0)
+        if self.read8() != QEMU_VM_VMDESCRIPTION:
+            raise Exception("No Debug Migration device found")
+
+        jsonlen = self.read32()
+
+        # Seek back to where we were at the beginning
+        self.file.seek(entrypos, 0)
+
+        return data[jsonpos:jsonpos + jsonlen]
+
+    def close(self):
+        self.file.close()
+
+
+class RamSection(object):
+    RAM_SAVE_FLAG_COMPRESS = 0x02
+    RAM_SAVE_FLAG_MEM_SIZE = 0x04
+    RAM_SAVE_FLAG_PAGE     = 0x08
+    RAM_SAVE_FLAG_EOS      = 0x10
+    RAM_SAVE_FLAG_CONTINUE = 0x20
+    RAM_SAVE_FLAG_XBZRLE   = 0x40
+    RAM_SAVE_FLAG_HOOK     = 0x80
+    blocks = []
+
+    def __init__(self, file, version_id, device, section_key):
+        if version_id != 4:
+            raise Exception("Unknown RAM version %d" % version_id)
+
+        self.file = file
+        self.section_key = section_key
+        self.TARGET_PAGE_SIZE = device
+
+    def read(self):
+        # Read all RAM sections
+        while True:
+            addr = self.file.read64()
+            flags = addr & (self.TARGET_PAGE_SIZE - 1)
+            addr &= ~(self.TARGET_PAGE_SIZE - 1)
+
+            if flags & self.RAM_SAVE_FLAG_MEM_SIZE:
+                while True:
+                    namelen = self.file.read8()
+                    # We assume that no RAM chunk is big enough to ever
+                    # hit the first byte of the address, so when we see
+                    # a zero here we know it has to be an address, not the
+                    # length of the next block.
+                    if namelen == 0:
+                        self.file.file.seek(-1, 1)
+                        break
+                    name = self.file.readstr(len = namelen)
+                    len = self.file.read64()
+                    self.blocks.append((name, len))
+                flags &= ~self.RAM_SAVE_FLAG_MEM_SIZE
+
+            if flags & self.RAM_SAVE_FLAG_COMPRESS:
+                if flags & self.RAM_SAVE_FLAG_CONTINUE:
+                    flags &= ~self.RAM_SAVE_FLAG_CONTINUE
+                else:
+                    name = self.file.readstr()
+                fill_char = self.file.read8()
+                # The page in question would be filled with fill_char now
+                flags &= ~self.RAM_SAVE_FLAG_COMPRESS
+            elif flags & self.RAM_SAVE_FLAG_PAGE:
+                if flags & self.RAM_SAVE_FLAG_CONTINUE:
+                    flags &= ~self.RAM_SAVE_FLAG_CONTINUE
+                else:
+                    name = self.file.readstr()
+                # Just skip RAM data for now
+                self.file.file.seek(self.TARGET_PAGE_SIZE, 1)
+                flags &= ~self.RAM_SAVE_FLAG_PAGE
+            elif flags & self.RAM_SAVE_FLAG_XBZRLE:
+                raise Exception("XBZRLE RAM compression is not supported yet")
+            elif flags & self.RAM_SAVE_FLAG_HOOK:
+                raise Exception("RAM hooks don't make sense with files")
+
+            # End of RAM section
+            if flags & self.RAM_SAVE_FLAG_EOS:
+                break
+
+            if flags != 0:
+                raise Exception("Unknown RAM flags: %x" % flags)
+
+    def getDict(self):
+        return ""
+
+class HTABSection(object):
+    HASH_PTE_SIZE_64       = 16
+
+    def __init__(self, file, version_id, device, section_key):
+        if version_id != 1:
+            raise Exception("Unknown HTAB version %d" % version_id)
+
+        self.file = file
+        self.section_key = section_key
+
+    def read(self):
+
+        header = self.file.read32()
+
+        if (header > 0):
+            # First section, just the hash shift
+            return
+
+        # Read until end marker
+        while True:
+            index = self.file.read32()
+            n_valid = self.file.read16()
+            n_invalid = self.file.read16()
+
+            if index == 0 and n_valid == 0 and n_invalid == 0:
+                break
+
+            self.file.readvar(n_valid * HASH_PTE_SIZE_64)
+
+    def getDict(self):
+        return ""
+
+class VMSDFieldGeneric(object):
+    def __init__(self, desc, file):
+        self.file = file
+        self.desc = desc
+        self.data = ""
+
+    def __repr__(self):
+        return str(self.__str__())
+
+    def __str__(self):
+        return " ".join("{0:02x}".format(ord(c)) for c in self.data)
+
+    def getDict(self):
+        return self.__str__()
+
+    def read(self):
+        size = int(self.desc['size'])
+        self.data = self.file.readvar(size)
+        return self.data
+
+class VMSDFieldInt(VMSDFieldGeneric):
+    def __init__(self, desc, file):
+        super(VMSDFieldInt, self).__init__(desc, file)
+        self.size = int(desc['size'])
+        self.format = '0x%%0%dx' % (self.size * 2)
+        self.sdtype = '>i%d' % self.size
+        self.udtype = '>u%d' % self.size
+
+    def __repr__(self):
+        if self.data < 0:
+            return ('%s (%d)' % ((self.format % self.udata), self.data))
+        else:
+            return self.format % self.data
+
+    def __str__(self):
+        return self.__repr__()
+
+    def getDict(self):
+        return self.__str__()
+
+    def read(self):
+        super(VMSDFieldInt, self).read()
+        self.sdata = np.fromstring(self.data, count=1, dtype=(self.sdtype))[0]
+        self.udata = np.fromstring(self.data, count=1, dtype=(self.udtype))[0]
+        self.data = self.sdata
+        return self.data
+
+class VMSDFieldUInt(VMSDFieldInt):
+    def __init__(self, desc, file):
+        super(VMSDFieldUInt, self).__init__(desc, file)
+
+    def read(self):
+        super(VMSDFieldUInt, self).read()
+        self.data = self.udata
+        return self.data
+
+class VMSDFieldIntLE(VMSDFieldInt):
+    def __init__(self, desc, file):
+        super(VMSDFieldIntLE, self).__init__(desc, file)
+        self.dtype = '<i%d' % self.size
+
+class VMSDFieldBool(VMSDFieldGeneric):
+    def __init__(self, desc, file):
+        super(VMSDFieldBool, self).__init__(desc, file)
+
+    def __repr__(self):
+        return self.data.__repr__()
+
+    def __str__(self):
+        return self.data.__str__()
+
+    def getDict(self):
+        return self.data
+
+    def read(self):
+        super(VMSDFieldBool, self).read()
+        if self.data[0] == 0:
+            self.data = False
+        else:
+            self.data = True
+        return self.data
+
+class VMSDFieldStruct(VMSDFieldGeneric):
+    QEMU_VM_SUBSECTION    = 0x05
+
+    def __init__(self, desc, file):
+        super(VMSDFieldStruct, self).__init__(desc, file)
+        self.data = collections.OrderedDict()
+
+        # When we see compressed array elements, unfold them here
+        new_fields = []
+        for field in self.desc['struct']['fields']:
+            if not 'array_len' in field:
+                new_fields.append(field)
+                continue
+            array_len = field.pop('array_len')
+            field['index'] = 0
+            new_fields.append(field)
+            for i in xrange(1, array_len):
+                c = field.copy()
+                c['index'] = i
+                new_fields.append(c)
+
+        self.desc['struct']['fields'] = new_fields
+
+    def __repr__(self):
+        return self.data.__repr__()
+
+    def __str__(self):
+        return self.data.__str__()
+
+    def read(self):
+        for field in self.desc['struct']['fields']:
+            try:
+                reader = vmsd_field_readers[field['type']]
+            except:
+                reader = VMSDFieldGeneric
+
+            field['data'] = reader(field, self.file)
+            field['data'].read()
+
+            if 'index' in field:
+                if field['name'] not in self.data:
+                    self.data[field['name']] = []
+                a = self.data[field['name']]
+                if len(a) != int(field['index']):
+                    raise Exception("internal index of data field unmatched (%d/%d)" % (len(a), int(field['index'])))
+                a.append(field['data'])
+            else:
+                self.data[field['name']] = field['data']
+
+        if 'subsections' in self.desc['struct']:
+            for subsection in self.desc['struct']['subsections']:
+                if self.file.read8() != self.QEMU_VM_SUBSECTION:
+                    raise Exception("Subsection %s not found" % subsection['vmsd_name'])
+                name = self.file.readstr()
+                version_id = self.file.read32()
+                self.data[name] = VMSDSection(self.file, version_id, subsection, (name, 0))
+                self.data[name].read()
+
+    def getDictItem(self, value):
+       # Strings would fall into the array category, treat
+       # them specially
+       if value.__class__ is ''.__class__:
+           return value
+
+       try:
+           return self.getDictOrderedDict(value)
+       except:
+           try:
+               return self.getDictArray(value)
+           except:
+               try:
+                   return value.getDict()
+               except:
+                   return value
+
+    def getDictArray(self, array):
+        r = []
+        for value in array:
+           r.append(self.getDictItem(value))
+        return r
+
+    def getDictOrderedDict(self, dict):
+        r = collections.OrderedDict()
+        for (key, value) in dict.items():
+            r[key] = self.getDictItem(value)
+        return r
+
+    def getDict(self):
+        return self.getDictOrderedDict(self.data)
+
+vmsd_field_readers = {
+    "bool" : VMSDFieldBool,
+    "int8" : VMSDFieldInt,
+    "int16" : VMSDFieldInt,
+    "int32" : VMSDFieldInt,
+    "int32 equal" : VMSDFieldInt,
+    "int32 le" : VMSDFieldIntLE,
+    "int64" : VMSDFieldInt,
+    "uint8" : VMSDFieldUInt,
+    "uint16" : VMSDFieldUInt,
+    "uint32" : VMSDFieldUInt,
+    "uint32 equal" : VMSDFieldUInt,
+    "uint64" : VMSDFieldUInt,
+    "int64 equal" : VMSDFieldInt,
+    "uint8 equal" : VMSDFieldInt,
+    "uint16 equal" : VMSDFieldInt,
+    "float64" : VMSDFieldGeneric,
+    "timer" : VMSDFieldGeneric,
+    "buffer" : VMSDFieldGeneric,
+    "unused_buffer" : VMSDFieldGeneric,
+    "bitmap" : VMSDFieldGeneric,
+    "struct" : VMSDFieldStruct,
+    "unknown" : VMSDFieldGeneric,
+}
+
+class VMSDSection(VMSDFieldStruct):
+    def __init__(self, file, version_id, device, section_key):
+        self.file = file
+        self.data = ""
+        self.vmsd_name = ""
+        self.section_key = section_key
+        desc = device
+        if 'vmsd_name' in device:
+            self.vmsd_name = device['vmsd_name']
+
+        # A section really is nothing but a FieldStruct :)
+        super(VMSDSection, self).__init__({ 'struct' : desc }, file)
+
+###############################################################################
+
+class MigrationDump(object):
+    QEMU_VM_FILE_MAGIC    = 0x5145564d
+    QEMU_VM_FILE_VERSION  = 0x00000003
+    QEMU_VM_EOF           = 0x00
+    QEMU_VM_SECTION_START = 0x01
+    QEMU_VM_SECTION_PART  = 0x02
+    QEMU_VM_SECTION_END   = 0x03
+    QEMU_VM_SECTION_FULL  = 0x04
+    QEMU_VM_SUBSECTION    = 0x05
+    QEMU_VM_VMDESCRIPTION = 0x06
+
+    def __init__(self, filename):
+        self.section_classes = { ( 'ram', 0 ) : [ RamSection, None ],
+                                 ( 'spapr/htab', 0) : ( HTABSection, None ) }
+        self.filename = filename
+        self.vmsd_desc = None
+
+    def read(self, desc_only = False):
+        # Read in the whole file
+        file = MigrationFile(self.filename)
+
+        # File magic
+        data = file.read32()
+        if data != self.QEMU_VM_FILE_MAGIC:
+            raise Exception("Invalid file magic %x" % data)
+
+        # Version (has to be v3)
+        data = file.read32()
+        if data != self.QEMU_VM_FILE_VERSION:
+            raise Exception("Invalid version number %d" % data)
+
+        self.load_vmsd_json(file)
+
+        # Read sections
+        self.sections = collections.OrderedDict()
+
+        if desc_only:
+            return
+
+        self.section_classes[('ram',0)][1] = self.vmsd_desc['page_size']
+
+        while True:
+            section_type = file.read8()
+            if section_type == self.QEMU_VM_EOF:
+                break
+            elif section_type == self.QEMU_VM_SECTION_START or section_type == self.QEMU_VM_SECTION_FULL:
+                section_id = file.read32()
+                name = file.readstr()
+                instance_id = file.read32()
+                version_id = file.read32()
+                section_key = (name, instance_id)
+                classdesc = self.section_classes[section_key]
+                section = classdesc[0](file, version_id, classdesc[1], section_key)
+                self.sections[section_id] = section
+                section.read()
+            elif section_type == self.QEMU_VM_SECTION_PART or section_type == self.QEMU_VM_SECTION_END:
+                section_id = file.read32()
+                self.sections[section_id].read()
+            else:
+                raise Exception("Unknown section type: %d" % section_type)
+        file.close()
+
+    def load_vmsd_json(self, file):
+        vmsd_json = file.read_migration_debug_json()
+        self.vmsd_desc = json.loads(vmsd_json, object_pairs_hook=collections.OrderedDict)
+        for device in self.vmsd_desc['devices']:
+            key = (device['name'], device['instance_id'])
+            value = ( VMSDSection, device )
+            self.section_classes[key] = value
+
+    def getDict(self):
+        r = collections.OrderedDict()
+        for (key, value) in self.sections.items():
+           key = "%s (%d)" % ( value.section_key[0], key )
+           r[key] = value.getDict()
+        return r
+
+###############################################################################
+
+class JSONEncoder(json.JSONEncoder):
+    def default(self, o):
+        if isinstance(o, VMSDFieldGeneric):
+            return str(o)
+        return json.JSONEncoder.default(self, o)
+
+parser = argparse.ArgumentParser()
+parser.add_argument("-f", "--file", help='migration dump to read from', required=True)
+parser.add_argument("-s", "--descriptionfile", help='migration dump to read vmstate description from')
+parser.add_argument("-d", "--dump", help='what to dump ("state" or "desc")', default='state')
+args = parser.parse_args()
+
+jsonenc = JSONEncoder(indent=4, separators=(',', ': '))
+
+if args.dump == "state":
+    dump = MigrationDump(args.file)
+    if args.descriptionfile:
+        # Fetch the vmstate description from the file passed through -s
+        desc_dump = MigrationDump(args.descriptionfile)
+        desc_dump.read(desc_only = True)
+        # and override all section readers and vmsd description in our
+        # data migration file with the ones from the vmstate description
+        # migration file
+        dump.vmsd_desc = desc_dump.vmsd_desc
+        dump.section_classes = desc_dump.section_classes
+    dump.read()
+    dict = dump.getDict()
+    print jsonenc.encode(dict)
+elif args.dump == "desc":
+    if args.descriptionfile:
+        dump = MigrationDump(args.descriptionfile)
+    else:
+        dump = MigrationDump(args.file)
+    dump.read(desc_only = True)
+    print jsonenc.encode(dump.vmsd_desc)
+else:
+    raise Exception("Unknown dump type \"%s\", available: \"state\", \"desc\"" % args.dump)