bdcht / amoco Goto Github PK
View Code? Open in Web Editor NEWyet another tool for analysing binaries
License: GNU General Public License v2.0
yet another tool for analysing binaries
License: GNU General Public License v2.0
>>> str(cpu_x64.disassemble('\x66\x41\xc7\x84\x55\x11\x11\x11\x11\x22\x22'))
'None'
>>> str(cpu_x64.disassemble('\x66\x41\xc7\x84\x55\x11\x11\x11\x11\x22\x22\x22\x22'))
'mov word ptr [(r13+(rdx*0x2))+0x11111111], 0x22222222'
while the first one is valid and should decode as mov word ptr [(r13+(rdx*0x2))+0x11111111], 0x2222
Proposed patch
--- a/amoco/arch/x64/spec_ia32e.py
+++ b/amoco/arch/x64/spec_ia32e.py
@@ -531,7 +532,7 @@ def ia32_ptr_iwd(obj,Mod,RM,data):
op1,data = getModRM(obj,Mod,RM,data)
REX = obj.misc['REX']
size = op1.size
- if REX: size=32
+ if REX and size==64: size = 32
if data.size<size: raise InstructionError(obj)
imm = data[0:size]
x = env.cst(imm.int(),size).signextend(op1.size)
>>> str(cpu_x64.disassemble('\xff\x35\0\0\0\0'))
'push dword ptr cs:[rip]'
>>> str(cpu_x64.disassemble('\x8f\x05\0\0\0\0'))
'pop dword ptr cs:[rip]'
Should not be dword ptr
.
The prefix 0x48 is not mandatory and is not set by compilers.
Proposed patch:
--- a/amoco/arch/x64/spec_ia32e.py
+++ b/amoco/arch/x64/spec_ia32e.py
@@ -254,6 +254,7 @@ def ia32_rm8(obj,Mod,RM,data):
@ispec_ia32("*>[ {f7} /7 ]", mnemonic = "IDIV" )
def ia32_rm32(obj,Mod,RM,data):
op1,data = getModRM(obj,Mod,RM,data)
+ if obj.mnemonic in ["PUSH", "POP"] and op1.size == 32: op1.size = 64
obj.operands = [op1]
obj.type = type_data_processing
Proposed patch:
--- a/amoco/arch/x86/spec_ia32.py
+++ b/amoco/arch/x86/spec_ia32.py
@@ -136,11 +136,14 @@ def ia32_strings(obj):
# imm8:
@ispec_ia32("16>[ {d5} ib(8) ]", mnemonic = "AAD", type=type_data_processing)
@ispec_ia32("16>[ {d4} ib(8) ]", mnemonic = "AAM", type=type_data_processing)
-@ispec_ia32("16>[ {6a} ib(8) ]", mnemonic = "PUSH", type=type_data_processing)
@ispec_ia32("16>[ {cd} ib(8) ]", mnemonic = "INT", type=type_control_flow)
def ia32_imm8(obj,ib):
obj.operands = [env.cst(ib,8)]
+@ispec_ia32("16>[ {6a} ib(8) ]", mnemonic = "PUSH", type=type_data_processing)
+def ia32_imm8_signed(obj,ib):
+ obj.operands = [env.cst(ib,8).signextend(8)]
+
@ispec_ia32("16>[ {eb} ib(8) ]", mnemonic = "JMP", type=type_control_flow)
@ispec_ia32("16>[ {e2} ib(8) ]", mnemonic = "LOOP", type=type_control_flow)
@ispec_ia32("16>[ {e1} ib(8) ]", mnemonic = "LOOPE", type=type_control_flow)
Hello,
First, amoco seems really cool. Thanks!
I have an issue, and a question
Issue:
cfg recovery seems to be a bit broken currently. I have a simple 'puts("hello world")' elf which I am using for testing. the lsweep
method recovers most of the basic blocks, as expect however others don't seem to get past the first basic block:
>>> p = amoco.system.loader.load_program('hi32')
>>> z = amoco.lforward(p)
>>> G=z.getcfg()
>>> print G.C
[<grandalf.graphs.graph_core object at 0x7fa70251bf10>, <grandalf.graphs.graph_core object at 0x7fa7024660d0>]
>>> print G.C[0].sV
0.| <node [0x8048380] at 0x7fa70251bb90>
>>> print G.C[1].sV
0.| <node [#PLT@__libc_start_main] at 0x7fa7079073d0>
1.| <node [@__libc_start_main] at 0x7fa7024def50>
Furthermore, some methods error:
>>> z = amoco.fbackward(p)
>>> z.getcfg()
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/local/lib/python2.7/dist-packages/amoco/main.py", line 267, in getcfg
for x in self.itercfg(loc): pass
File "/usr/local/lib/python2.7/dist-packages/amoco/main.py", line 289, in itercfg
if self.check_ext_target(t):
File "/usr/local/lib/python2.7/dist-packages/amoco/main.py", line 261, in check_ext_target
self.update_spool(e.v[1],t.parent)
File "/usr/local/lib/python2.7/dist-packages/amoco/main.py", line 206, in update_spool
T = self.get_targets(vtx,parent)
File "/usr/local/lib/python2.7/dist-packages/amoco/main.py", line 373, in get_targets
func.map[pc] = mpc
File "/usr/local/lib/python2.7/dist-packages/amoco/code.py", line 33, in map
self.helper(self._map)
File "/usr/local/lib/python2.7/dist-packages/amoco/code.py", line 42, in helper
if self._helper: self._helper(self,m)
AttributeError: _helper
Although interestingly, running it again does not error, it still doesn't recovery the cfg:
>>> z.getcfg()
<amoco.cfg.graph object at 0x7fa702511a90>
(I've also noticed running other cfg recoveries twice gives different results the second time)
I think this these errors stem from certain function calls not getting resolved properly, but I'm not sure:
>>> print n.data
# --- block 0x8048380 ---
0x8048380 '31ed' xor ebp, ebp
0x8048382 '5e' pop esi
0x8048383 '89e1' mov ecx, esp
0x8048385 '83e4f0' and esp, 0xfffffff0
0x8048388 '50' push eax
0x8048389 '54' push esp
0x804838a '52' push edx
0x804838b '6820850408' push #__libc_csu_fini
0x8048390 '68b0840408' push #__libc_csu_init
0x8048395 '51' push ecx
0x8048396 '56' push esi
0x8048397 '687b840408' push #main
0x804839c 'e8afffffff' call *0x8048350
>>> i = n.data.instr[-1]
>>> print i.misc['to']
0x8048350
>>> p.mmap.read(0x8048350, 4)
['\xff%(\x97']
Im happy to provide the binaries, and/or any other info that would be helpful. I spent some time looking around trying to solve it, but am still wrapping my head around how everything is set up.
Question:
Im interested in tagging the memory sections with their flags (i.e read, write execute). I originally hacked it onto 'mo', before I noticed that MemoryMap has an unused 'perm' attribute. Is this the correct spot to store that info?
Alternatively, the info is stored in p.bin.Phdr. Would it be cleaner to just query that?
Thanks!
>>> print(str(mapper().M(env.mem(env.ptr(env.cst(0,size=32))))))
M32(0x0)
>>> print(str(mapper().M(env.mem(env.ptr(env.ext('toto',size=32))))))
@toto
where I would expect M32(@toto)
.
I propose the following patch
--- a/amoco/cas/mapper.py
+++ b/amoco/cas/mapper.py
@@ -130,7 +130,7 @@ class mapper(object):
# get a memory location value (fetch) :
# k must be mem expressions
def M(self,k):
- if k.a.base._is_ext: return k.a.base
+ if k.a.base._is_ext: return k
n = self.aliasing(k)
if n>0:
f = lambda e:e[0]._is_ptr
Hi,
I got confused by the following codes, maybe this is an Amoco bug.
The code as follows:
https://gist.github.com/lieanu/f65788ff947a04d50aa0
import amoco
import amoco.system.raw
import amoco.system.core
def sym_exec_gadget_and_get_mapper(code, cpu):
'''Taken from https://github.com/0vercl0k/stuffz/blob/master/look_for_gadgets_with_equations.py'''
p = amoco.system.raw.RawExec(
amoco.system.core.DataIO(code), cpu
)
try:
blocks = list(amoco.lsweep(p).iterblocks())
except:
return None
if len(blocks) == 0:
return None
mp = amoco.cas.mapper.mapper()
for block in blocks:
if block.instr[-1].mnemonic.lower() == 'call':
p.cpu.i_RET(None, block.map)
try:
mp >>= block.map
except Exception as e:
pass
return mp
if __name__ == "__main__":
# pop rdi; ret --> "\x5f\xc3"
print "-"*20, "AMD64", "-"*20
print "Instr: ", "pop rdi; ret"
import amoco.arch.x64.cpu_x64 as amd64_cpu
cpu = amd64_cpu
print sym_exec_gadget_and_get_mapper("\x5f\xc3", cpu)
# pop eax; ret --> "\x58\xc3"
print "-"*20, "I386", "-"*20
print "Instr: ", "pop eax; ret"
import amoco.arch.x86.cpu_x86 as i386_cpu
cpu = i386_cpu
print sym_exec_gadget_and_get_mapper("\x58\xc3", cpu)
# pop rdi; ret --> "\x5f\xc3"
print "-"*20, "AMD64 again", "-"*20
print "Instr: ", "pop rdi; ret"
import amoco.arch.x64.cpu_x64 as amd64_cpu
cpu = amd64_cpu
print sym_exec_gadget_and_get_mapper("\x5f\xc3", cpu)
The outputs:
-------------------- AMD64 --------------------
Instr: pop rdi; ret
rdi <- { | [0:64]->M64(rsp) | } instructions [352]
rsp <- { | [0:64]->(rsp+0x10) | }
rip <- { | [0:64]->M64(rsp+8) | }
-------------------- I386 --------------------
Instr: pop eax; ret
eax <- { | [0:32]->M32(esp) | } instructions [352]
esp <- { | [0:32]->(esp+0x8) | }
eip <- { | [0:32]->M32(esp+4) | }
-------------------- AMD64 again --------------------
Instr: pop rdi; ret
rdi <- { | [0:64]->M64(esp) | }
esp <- { | [0:32]->(esp+0xc) | }
eip <- { | [0:32]->M32(esp+8) | }
You can see that the mapper of pop rdi; ret
after switching to amd64 cpu, is incorrect.
Regards,
lieanu
>>> from amoco.arch.x64 import cpu_x64
>>> str(cpu_x64.disassemble('\x80\xcc\x0c'))
'or spl, 0xc'
>>> str(cpu_x64.disassemble('\x40\x80\xcc\x0c'))
'or spl, 0xc'
The first should be or ah, 0xc
.
I am looking for how to correct this bug.
The function getreg
in env.py
never returns ah
, therefore one solution would be to change this function to add a third argument (the value of REX) ; another possibility is to change getModRM
in utils.py
. It depends on if we need also to change getregR
, getregRW
and getregB
.
My proposal for a patch is
--- a/amoco/arch/x64/utils.py
+++ b/amoco/arch/x64/utils.py
@@ -67,6 +67,16 @@ def getModRM(obj,Mod,RM,data):
# r/16/32 case:
if Mod==0b11:
op1 = env.getreg((B<<3)+RM,opdsz)
+ if REX is None and opdsz is 8: op1 = {
+ env.al: env.al,
+ env.cl: env.cl,
+ env.dl: env.dl,
+ env.bl: env.bl,
+ env.spl: env.ah,
+ env.bpl: env.ch,
+ env.sil: env.dh,
+ env.dil: env.bh,
+ }[op1]
return op1,data
# m/16/32 case:
if adrsz!=16 and RM==0b100:
>>> print(cpu_x86.disassemble('\x0f\xa5\xef'))
None
>>> print(cpu_x86.disassemble('\x0f\xac\xef\x01'))
None
>>> print(cpu_x86.disassemble('\x0f\xa4\x6d\xf8\x01'))
None
I would like to use something like
@ispec_ia32("*>[ {0f}{a4} /r ib(8) ]", mnemonic = "SHLD")
but it is not possible.
As mentioned in the documentation, the "variable length directive" should be "at the end of the FORMAT".
How are we supposed to parse these instructions without using ugly hacks?
>>> str(cpu_x64.disassemble('\xa5'))
'movsd '
>>> str(cpu_x64.disassemble('\x48\xa5'))
'movsd '
Proposed patch:
--- a/amoco/arch/x64/spec_ia32e.py
+++ b/amoco/arch/x64/spec_ia32e.py
@@ -134,6 +134,8 @@ def ia32_nooperand(obj):
def ia32_strings(obj):
if obj.mnemonic[-1]=='D' and obj.misc['opdsz']:
obj.mnemonic = obj.mnemonic[:-1]+'W'
+ if obj.mnemonic[-1]=='D' and obj.misc['REX']:
+ obj.mnemonic = obj.mnemonic[:-1]+'Q'
#1 operand
#----------
>>> from amoco.arch.x64 import cpu_x64
>>> from amoco.cas.mapper import mapper
>>> cpu_x64.disassemble('\x77\x08')(mapper())
>>> from amoco.arch.x86 import cpu_x86
>>> cpu_x64.disassemble('\x77\x08')(mapper())
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "amoco/arch/core.py", line 67, in __call__
i_xxx(self,map)
File "amoco/arch/x86/asm.py", line 503, in i_Jcc
fmap[eip] = tst(fmap(cond),fmap[eip]+op1,fmap[eip])
File "amoco/cas/expressions.py", line 39, in checkarg_numeric
return f(self,n)
File "amoco/cas/expressions.py", line 186, in __add__
def __add__(self,n): return oper('+',self,n)
File "amoco/cas/expressions.py", line 1201, in oper
return op(opsym,l,r).simplify()
File "amoco/cas/expressions.py", line 1216, in __init__
raise ValueError("Size mismatch %d <> %d"%(l.size,r.size))
ValueError: Size mismatch 32 <> 64
We can see that after importing x86
, decoding 64-bit instructions gets the semantics of 32-bit instructions.
Example:
>>> from amoco.arch.x64 import cpu_x64
>>> str(cpu_x64.disassemble('\xf3\x0f\x2a\xc0'))
'cvtsi2ss xmm0, rax'
should return cvtsi2ss xmm0, eax
Same for cvtsi2sd
.
For these examples, it seems to be sufficient to replace set_opdsz_64
by set_opdsz_32
in the definition in spec_sse.py
(and to create set_opdsz_32
in utils.py
).
>>> str(cpu_x64.disassemble('\xf2\x48\x0f\x2c\xf3'))
'cvttsd2si rsi, rbx'
should be cvttsd2si rsi, xmm3
Proposed patch:
--- a/amoco/arch/x64/spec_sse.py
+++ b/amoco/arch/x64/spec_sse.py
@@ -389,7 +389,9 @@ def sse_sd(obj,Mod,REG,RM,data):
@ispec_ia32("*>[ {0f}{2d} /r ]", mnemonic="CVTSD2SI")
def sse_sd(obj,Mod,REG,RM,data):
if not check_f2(obj,set_opdsz_128): raise InstructionError(obj)
- op2,data = getModRM(obj,Mod,RM,data)
+ # force REX.W=0 for op2 decoding:
+ W,R,X,B = getREX(obj)
+ op2,data = getModRM(obj,Mod,RM,data,REX=(0,R,X,B))
if op2._is_mem: op2.size = 64
op1 = getregRW(obj,REG,32)
obj.operands = [op1,op2]
Hello Axel,
Here is a repro for a bug I recently encountered:
import amoco
import amoco.system.raw
import amoco.arch.x86.cpu_x86 as cpu
def sym_exec_gadget_and_get_mapper(code):
p = amoco.system.raw.RawExec(
amoco.system.core.DataIO(code), cpu
)
blocks = list(amoco.lsweep(p).iterblocks())
assert(len(blocks) > 0)
mp = amoco.cas.mapper.mapper()
for block in blocks:
if block.instr[-1].mnemonic.lower() == 'call':
p.cpu.i_RET(None, block.map)
mp >>= block.map
return mp
code = '\xec\x27\x57\x33\xec\x67\x11\x04\xc3'
x = sym_exec_gadget_and_get_mapper(code)
# D:\Codes\rp2s\amoco\cas\smt.pyc in cast_z3_bool(x, solver)
# 102
# 103 def cast_z3_bool(x,solver=None):
# --> 104 b = x.to_smtlib(solver)
# 105 if not z3.is_bool(b):
# 106 assert b.size()==1
# TypeError: to_smtlib() takes exactly 1 argument (2 given)
Cheers,
0vercl0k
Same problem as in #34
>>> i = cpu_x86.disassemble('\xF2\x0F\xC2\xCA\x00')
>>> str(i)
'cmpsd xmm1, xmm2, 0x0'
>>> m=mapper()
>>> i(m)
>>> print(str(m))
eflags <- { | [0:1]->(((~M32(edi)[31:32])&M32(esi)[31:32])|((M32(edi)-M32(esi))[31:32]&((~M32(edi)[31:32])|M32(esi)[31:32]))) | [1:2]->eflags[1:2] | [2:3]->T1 | [3:4]->eflags[3:4] | [4:5]->T1 | [5:6]->eflags[5:6] | [6:7]->((M32(edi)-M32(esi))==0x0) | [7:8]->((M32(edi)-M32(esi))<0x0) | [8:11]->eflags[8:11] | [11:12]->((M32(edi)[31:32]^(M32(edi)-M32(esi))[31:32])&(M32(edi)[31:32]^M32(esi)[31:32])) | [12:32]->eflags[12:32] | }
eip <- { | [0:32]->(eip+0x5) | }
edi <- { | [0:32]->(df ? (edi-0x4) : (edi+0x4)) | }
esi <- { | [0:32]->(df ? (esi-0x4) : (esi+0x4)) | }
which is the semantics of cpu_x86.disassemble('\xA7')
Proposed patch, until there is an implementation of this SSE2 instruction:
--- a/amoco/arch/x86/asm.py
+++ b/amoco/arch/x86/asm.py
@@ -296,7 +296,10 @@ def i_CMPSB(i,fmap):
def i_CMPSW(i,fmap):
_cmps_(i,fmap,2)
def i_CMPSD(i,fmap):
- _cmps_(i,fmap,4)
+ if i.misc['opdsz']==128:
+ return
+ else:
+ _cmps_(i,fmap,4)
In amoco/arch/x86/asm.py
I can read
def i_SAR(i,fmap):
(...)
if count._is_cst:
if count.value==0: return
if count.value==1:
fmap[of] = bit0
else:
fmap[of] = top(1)
while the Intel instruction set reference says
The OF flag is affected only on 1-bit shifts.
Why do you set it to top?
One of the things that I'd really like to see out of Amoco is to support the basic 32-bit MIPS ISA. As it's a RISC architecture and not entirely dissimilar to ARM/SPARC, it shouldn't be too difficult.
Eventually I'll have time to do this, but I don't think it'll be until after DEFCON in August.
>>> print(str(cpu_x86.disassemble('\x74\xc0')))
jz .-64 # OK
>>> print(str(cpu_x86.disassemble('\xeb\xc0')))
jmp .+192 # should be jmp .-64
Proposed patch:
--- a/amoco/arch/x86/spec_ia32.py
+++ b/amoco/arch/x86/spec_ia32.py
@@ -146,7 +146,7 @@ def ia32_imm8(obj,ib):
@ispec_ia32("16>[ {e1} ib(8) ]", mnemonic = "LOOPE", type=type_control_flow)
@ispec_ia32("16>[ {e0} ib(8) ]", mnemonic = "LOOPNE", type=type_control_flow)
def ia32_imm_rel(obj,ib):
- obj.operands = [env.cst(ib,8)]
+ obj.operands = [env.cst(ib,8).signextend(32)]
@ispec_ia32("16>[ {e3} cb(8) ]", mnemonic = "JECXZ", type=type_control_flow)
def ia32_cb8(obj,cb):
>>> str(cpu_x64.disassemble('\x48\x63\xd2'))
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "amoco/arch/core.py", line 197, in __call__
return self(bytestring[s.mask.size/8:],**kargs)
File "amoco/arch/core.py", line 190, in __call__
i = s.decode(bytestring,e,i=self.__i,ival=b.ival)
File "amoco/arch/core.py", line 437, in decode
self.hook(obj=i,**kargs)
File "amoco/arch/x64/spec_ia32e.py", line 629, in ia32_movsxd
op2,data = getModRM(obj,Mod,RM,data,REX=(0,R,X,B))
NameError: global name 'R' is not defined
Proposed patch
--- a/amoco/arch/x64/spec_ia32e.py
+++ b/amoco/arch/x64/spec_ia32e.py
@@ -626,6 +626,7 @@ def ia32_arpl(obj,Mod,REG,RM,data,_inv):
def ia32_movsxd(obj,Mod,REG,RM,data):
op1 = getregR(obj,REG,64)
# force REX.W=0 for op2 decoding:
+ W,R,X,B = getREX(obj)
op2,data = getModRM(obj,Mod,RM,data,REX=(0,R,X,B))
obj.operands = [op1, op2]
obj.type = type_data_processing
The PE loader was stopping reading after 179 bytes in some samples I was analysing. Tracked it down to the following line in system\pe.py:567
f = open(filename,'r')
The open mode should be 'rb', otherwise the stream terminates at invalid string characters.
Sorry for not doing the whole "fork and merge" thing, I'm lazy :)
Proposed patch
--- a/amoco/arch/x64/spec_ia32e.py
+++ b/amoco/arch/x64/spec_ia32e.py
@@ -732,9 +732,11 @@ def ia32_reg_rm_8(obj,Mod,RM,REG,data):
def ia32_reg_rm_wd(obj,Mod,RM,REG,data):
op2,data = getModRM(obj,Mod,RM,data)
op1 = getregR(obj,REG,op2.size)
- if data.size<op2.size: raise InstructionError(obj)
- imm = data[0:op2.size]
- x = env.cst(imm.int(),op2.size)
+ sz = op2.size
+ if sz == 64: sz = 32
+ if data.size<sz: raise InstructionError(obj)
+ imm = data[0:sz]
+ x = env.cst(imm.int(),sz)
obj.operands = [op1, op2, x]
obj.bytes += pack(imm)
obj.type = type_data_processing
--- a/amoco/arch/x86/asm.py
+++ b/amoco/arch/x86/asm.py
@@ -1411,7 +1413,7 @@ def i_PSLLQ(i,fmap):
fmap[op1] = composer(res)
def i_PSHUFD(i,fmap):
- fmap[rip] = fmap[rip]+i.length
+ fmap[eip] = fmap[eip]+i.length
op1 = i.operands[0]
op2 = i.operands[1]
op3 = i.operands[2]
FISTPP
should be FISTTP
The typo appears three times.
>>> str(cpu_x64.disassemble('\x48\x0f\xbe\xc0'))
'movsx rax, rax'
should return movsx rax, al
Proposed patch:
--- a/amoco/arch/x64/spec_ia32e.py
+++ b/amoco/arch/x64/spec_ia32e.py
@@ -860,6 +860,7 @@ def ia32_movx(obj,Mod,RM,REG,data,_flg8):
if R==1: REG = (R<<3)+REG
op1 = env.getreg(REG,size)
obj.misc['opdsz']=8 if _flg8 else 16
+ if REX is not None: obj.misc['REX']=(0,R,X,B) # op2 not 64-bit
op2,data = getModRM(obj,Mod,RM,data)
obj.operands = [op1, op2]
obj.type = type_data_processing
Fixed in pull request #26.
code:
import sys
import amoco
import amoco.system.raw
import amoco.system.core
def sym_exec_gadget_and_get_mapper(code, cpu):
'''Taken from https://github.com/0vercl0k/stuffz/blob/master/look_for_gadgets_with_equations.py'''
p = amoco.system.raw.RawExec(
amoco.system.core.DataIO(code), cpu
)
try:
blocks = list(amoco.lsweep(p).iterblocks())
except:
return None
if len(blocks) == 0:
return None
mp = amoco.cas.mapper.mapper()
for block in blocks:
if block.instr[-1].mnemonic.lower() == 'call':
p.cpu.i_RET(None, block.map)
try:
mp >>= block.map
except Exception as e:
pass
return mp
if __name__ == "__main__":
# adc bh, bh ; call qword ptr [rsi]
import amoco.arch.x64.cpu_x64 as cpu
bytes = "10ffff16".decode("hex")
print sym_exec_gadget_and_get_mapper(bytes, cpu)
output:
Traceback (most recent call last):structions [352]
File "amoco_test2.py", line 38, in <module>
print sym_exec_gadget_and_get_mapper(bytes, cpu)
File "amoco_test2.py", line 24, in sym_exec_gadget_and_get_mapper
p.cpu.i_RET(None, block.map)
File "/usr/lib/python2.7/site-packages/amoco/code.py", line 32, in map
self._map = mapper(self.instr)
File "/usr/lib/python2.7/site-packages/amoco/cas/mapper.py", line 44, in __init__
if not instr.misc['delayed']: instr(self)
File "/usr/lib/python2.7/site-packages/amoco/arch/core.py", line 65, in __call__
i_xxx(self,map)
File "/usr/lib/python2.7/site-packages/amoco/arch/x64/asm.py", line 560, in i_ADC
fmap[af] = halfcarry(a,op2,c)
NameError: global name 'c' is not defined
>>> from amoco.arch.x64 import cpu_x64
>>> str(cpu_x64.disassemble('\x48\x0f\xbf\xf2'))
'movsx rsi, rdx'
should be movsx rsi, dx
Proposed patch:
--- a/amoco/arch/x64/spec_ia32e.py
+++ b/amoco/arch/x64/spec_ia32e.py
@@ -849,7 +852,7 @@ def ia32_movx(obj,Mod,RM,REG,data,_flg8):
if R==1: REG = (R<<3)+REG
op1 = env.getreg(REG,size)
obj.misc['opdsz']=8 if _flg8 else 16
- op2,data = getModRM(obj,Mod,RM,data)
+ op2,data = getModRM(obj,Mod,RM,data,REX=(0,R,X,B))
obj.operands = [op1, op2]
obj.type = type_data_processing
The following script
from amoco.cas.mapper import mapper
import gc
gc.enable()
gc.set_debug(gc.DEBUG_LEAK)
m=mapper()
del m
gc.collect()
print(gc.garbage)
outputs:
gc: collectable <list 0x10d8357e8>
[[[...], [...], None]]
This might not seem an issue, but it makes my software eat hundreds of megabytes of memory...
The culprit might be OrderedDict.
Cf. http://www.gossamer-threads.com/lists/python/bugs/860875
This bug is known since 2010, and not corrected on standard python 2.7 installations.
>>> cpu_x86.disassemble('\x66\x6a\x08').operands[0].size
8
>>> cpu_x64.disassemble('\x66\x6a\x08').operands[0].size
8
Should be 16
.
The Intel manual recommends that 0x66 0x68 N N is used for 16-bit push, but GNU as (at least version 2.25) generates 0x66 0x6A 0x08 when asked to assemble pushw $8
.
>>> i = cpu_x86.disassemble('\x66\x0F\xD6\x4D\xF0')
>>> str(i)
'movq qword ptr [ebp-16], xmm1'
>>> i(mapper())
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "amoco/arch/core.py", line 67, in __call__
i_xxx(self,map)
File "amoco/arch/x86/asm.py", line 1246, in i_MOVQ
fmap[op1] = op2.zeroextend(op1.size)
File "amoco/cas/mapper.py", line 204, in __setitem__
raise ValueError('size mismatch')
ValueError: size mismatch
and also
>>> str(cpu_x86.disassemble('\xF3\x0F\x7E\x55\xF0'))
'movq mm2, qword ptr [ebp-16]'
where it should be 'movq xmm2, qword ptr [ebp-16]'
and finally
>>> i = cpu_x86.disassemble('\xF3\x0F\x7E\xC1')
>>> str(i)
'movq xmm0, xmm1'
>>> m=mapper()
>>> i(m)
>>> print(str(m))
eip <- { | [0:32]->(eip+0x4) | }
xmm0 <- { | [0:128]->xmm1 | }
which should only copy the low quadword of %xmm1
and clear the high quadword of %xmm0
Hello Axel,
Here is a repro for a bug I recently encountered:
import amoco
import amoco.system.raw
import amoco.arch.x86.cpu_x86 as cpu
def sym_exec_gadget_and_get_mapper(code):
p = amoco.system.raw.RawExec(
amoco.system.core.DataIO(code), cpu
)
blocks = list(amoco.lsweep(p).iterblocks())
assert(len(blocks) > 0)
mp = amoco.cas.mapper.mapper()
for block in blocks:
if block.instr[-1].mnemonic.lower() == 'call':
p.cpu.i_RET(None, block.map)
mp >>= block.map
return mp
code = '\x01\x00\xf4\xfd\xf4\xff\xd6'
x = sym_exec_gadget_and_get_mapper(code)
# D:\Codes\rp2s\amoco\cas\expressions.pyc in checkarg_numeric(self, n)
# 31 def _checkarg_numeric(f):
# 32 def checkarg_numeric(self,n):
# ---> 33 if isinstance(n,(int,long)):
# 34 n = cst(n,self.size)
# 35 elif isinstance(n,(float)):
# RuntimeError: maximum recursion depth exceeded in __instancecheck__
Cheers,
0vercl0k
Proposed patch:
--- a/amoco/arch/x86/spec_sse.py
+++ b/amoco/arch/x86/spec_sse.py
@@ -577,6 +577,7 @@ def sse_sd(obj,Mod,REG,RM,data):
@ispec_ia32("*>[ {0f}{6a} /r ]", mnemonic="PUNPCKHDQ")
@ispec_ia32("*>[ {0f}{6b} /r ]", mnemonic="PACKSSDW")
@ispec_ia32("*>[ {0f}{6c} /r ]", mnemonic="PUNPCKLQDQ")
+@ispec_ia32("*>[ {0f}{6d} /r ]", mnemonic="PUNPCKHQDQ")
@ispec_ia32("*>[ {0f}{6f} /r ]", mnemonic="MOVDQA")
@ispec_ia32("*>[ {0f}{74} /r ]", mnemonic="PCMPEQB")
@ispec_ia32("*>[ {0f}{75} /r ]", mnemonic="PCMPEQW")
This seems like a really great project with a lot of potential uses. For one, I'd like to use it in my pwndbg debugger scripts for GDB in order to simulate execution of a single basic block (to determine the expected path of a conditional branch).
Unfortunately, there do not appear to be any examples which I can base this work off of. Since I'm running in a debugger, I can provide complete state information, and even memory access.
How would I go about simulating a short sequence of instructions, given a known starting register context?
The instruction pop esp
does not add 4 to esp
after popping. Currently, Amoco treats it as effectively: mov esp [esp]; add esp 0x4
, which is incorrect.
Thanks to @lieanu for pointing it out!
Proposed patch.
--- a/amoco/arch/x86/formats.py
+++ b/amoco/arch/x86/formats.py
@@ -11,15 +11,60 @@ def pfx(i):
def mnemo(i):
mnemo = i.mnemonic.replace('cc','')
if hasattr(i,'cond'): mnemo += i.cond[0].split('/')[0]
- return [(Token.Mnemonic,'{: <12}'.format(mnemo.lower()))]
+ if mnemo == 'RETN': mnemo = 'RET'
+ s = [(Token.Mnemonic,'{: <12}'.format(mnemo.lower()))]
+ # clang assembler does not understand 'sal'
+ if mnemo == 'SAL': s = [(Token.Mnemonic,'{: <12}'.format('shl'))]
+ # Special case: when gcc produces 'rep ret'
+ # http://mikedimmick.blogspot.fr/2008/03/what-heck-does-ret-mean.html
+ # it usually puts it on two separate lines, and old versions of
+ # GNU as don't like a true 'rep ret'
+ if mnemo == 'RET' and i.misc.get('pfx') is not None:
+ if i.misc['pfx'][0] == 'rep':
+ s = [(Token.Prefix,'rep; ')] + s
+ return s
+
+def str_intel(op):
+ if op is None:
+ return "<None>" # NEVER
+ elif op._is_ext:
+ return str(op.ref)
+ elif op._is_eqn:
+ # Special cases
+ if op.op.symbol == '+' and op.l._is_eqn and op.l.op.symbol == '+' \
+ and op.l.l._is_eqn and op.l.l.l is None and op.l.l.op.symbol == '-':
+ # clang ptr diff
+ return '(%s+%s-%s)' % (
+ str_intel(op.r),
+ str_intel(op.l.r),
+ str_intel(op.l.l.r))
+ if op.op.symbol == '+' and op.r._is_reg and str(op.r) == 'esp':
+ # ending the formula with 'esp' is not allowed by assemblers
+ return '(%s+%s)' % (str_intel(op.r), str_intel(op.l))
+ # Generic case
+ if op.l is None: return '(%s%s)' % (op.op.symbol, str_intel(op.r))
+ return '(%s%s%s)' % (str_intel(op.l), op.op.symbol, str_intel(op.r))
+ else: # _is_cst or _is_reg
+ return str(op)
def deref(op):
assert op._is_mem
- d = '%+d'%op.a.disp if op.a.disp else ''
- s = {8:'byte ptr ',16:'word ptr ', 64:'qword ptr ', 128:'xmmword ptr '}.get(op.size,'')
- s += '%s:'%op.a.seg if (op.a.seg is not '') else ''
- s += '[%s%s]'%(op.a.base,d)
- return s
+ address = str_intel(op.a.base)
+ if op.a.disp:
+ address += '%+d' % op.a.disp
+ prefix = {
+ 8: 'byte ptr ',
+ 16: 'word ptr ',
+ 32: 'dword ptr ',
+ 64: 'qword ptr ',
+ 80: 'tbyte ptr ',
+ 128:'xmmword ptr ',
+ }.get(op.size,'')
+ if op.a.seg is not '':
+ prefix += '%s:' % op.a.seg
+ if getattr(op.a.disp, '_is_ext', False):
+ prefix += '%s' % op.a.disp.ref
+ return '%s[%s]' % (prefix, address)
def opers(i):
s = []
@@ -33,8 +78,19 @@ def opers(i):
s.append((Token.Constant,'%+d'%op.value))
else:
s.append((Token.Constant,str(op)))
+ elif op._is_ext:
+ s.append((Token.Address,'OFFSET FLAT:%s'%op.ref))
elif op._is_reg:
- s.append((Token.Register,str(op)))
+ op = str(op)
+ if op.startswith('st'):
+ op = 'st(%s)'%op[2]
+ s.append((Token.Register,op))
+ elif op._is_eqn and op.l._is_ext:
+ s.append((Token.Address,'OFFSET FLAT:%s%s%s'%
+ (op.l.ref,op.op.symbol,int(op.r))))
+ else:
+ import sys
+ sys.stderr.write("TODO %s %s\n"%(op.__class__,op))
s.append((Token.Literal,', '))
if len(s)>0: s.pop()
return s
@@ -42,7 +98,9 @@ def opers(i):
def oprel(i):
to = i.misc['to']
if to is not None:
- return [(Token.Address,'*'+str(to))]
+ return [(Token.Address,str(to))]
+ if i.operands[0]._is_ext:
+ return [(Token.Address,str(i.operands[0].ref))]
if (i.address is not None) and i.operands[0]._is_cst:
v = i.address + i.operands[0].signextend(32) + i.length
i.misc['to'] = v
I am a developer on the ArchAssault project, we would like to distribute this tool. Would be possible to add a tag and a license to it, so that we can release it?
m=mapper()
for i in [
'\x8B\xBB\0\0\0\0',
'\x83\xEF\x0C',
'\x89\xBB\0\0\0\0',
'\x8B\x93\0\0\0\0',
'\xFF\x74\xBA\x2C',
'\xFF\x74\xBA\x28',
'\xFF\x74\xBA\x24',
'\xFF\x74\xBA\x20',
'\xFF\x74\xBA\x1C',
'\xFF\x74\xBA\x18',
'\xFF\x74\xBA\x14',
'\xFF\x74\xBA\x10',
'\xFF\x74\xBA\x0C',
'\xFF\x74\xBA\x08',
'\xFF\x74\xBA\x04',
'\xFF\x34\xBA',
'\x8B\x84\x24\xF0\0\0\0',
'\xFF\x50\x08',
]:
i=cpu_x86.disassemble(i)
print(str(i))
i(m)
outputs
mov edi, [ebx]
sub edi, 0xc
mov [ebx], edi
mov edx, [ebx]
push [((edi*0x4)+edx)+0x2c]
push [((edi*0x4)+edx)+0x28]
push [((edi*0x4)+edx)+0x24]
push [((edi*0x4)+edx)+0x20]
push [((edi*0x4)+edx)+0x1c]
push [((edi*0x4)+edx)+0x18]
push [((edi*0x4)+edx)+0x14]
push [((edi*0x4)+edx)+0x10]
push [((edi*0x4)+edx)+0xc]
push [((edi*0x4)+edx)+0x8]
push [((edi*0x4)+edx)+0x4]
push [((edi*0x4)+edx)]
mov eax, [esp+240]
call [eax+0x8]
(which is OK) but takes 16 seconds on my computer (which is not OK).
It looks like amoco lacks support for 64-bit imul
.
movabs rax,0xbdad7412bb99a0c1
movabs r10,0x9a899ae230f27801
imul r10
import amoco
import amoco.arch.x64.cpu_x64 as cpu
shellcode = 'H\xb8\xc1\xa0\x99\xbb\x12t\xad\xbdI\xba\x01x\xf20\xe2\x9a\x89\x9aI\xf7\xea'
loc = cpu.cst(0x8048380,64)
instr = [ ]
while len(shellcode)>0:
i = cpu.disassemble(shellcode,address=loc)
l = i.length
instr.append(i)
shellcode = shellcode[l:]
loc += l
b = amoco.code.block(instr)
print b
print b.map
/home/user/.pyenv/versions/2.7.9/lib/python2.7/site-packages/amoco/arch/x64/asm.pyc in i_IMUL(i, fmap)
915 if len(i.operands)==1:
916 src = i.operands[0]
--> 917 m,d = {8:(al,ah), 16:(ax,dx), 32:(eax,edx)}[src.size]
918 r = fmap(m**src)
919 elif len(i.operands)==2:
KeyError: 64
>>> m=mapper()
>>> i=cpu_x86.disassemble('\x0F\x95\xC2')
>>> m[env.edx]=env.mem(env.ebp)
>>> print(str(m))
edx <- { | [0:32]->M32(ebp) | }
>>> print(str(i))
setnz dl
>>> i(m)
>>> print(str(m))
edx <- { | [0:32]->M32(ebp) | }
eip <- { | [0:32]->(eip+0x3) | }
(ebp) <- ((zf==0x0) ? 0x1 : 0x0)
edx
should be modified, not (ebp)
Proposed patch
--- a/amoco/arch/x86/asm.py
+++ b/amoco/arch/x86/asm.py
@@ -569,7 +572,7 @@ def i_NOT(i,fmap):
fmap[op1] = ~fmap(op1)
def i_SETcc(i,fmap):
- op1 = fmap(i.operands[0])
+ op1 = i.operands[0]
fmap[eip] = fmap[eip]+i.length
fmap[op1] = tst(fmap(i.cond[1]),cst(1,op1.size),cst(0,op1.size))
>>> from amoco.arch.x64 import cpu_x64
>>> i=cpu_x64.disassemble('\x48\x8b\x04\xc5\0\0\0\0')
>>> str(i.operands[1])
'M64cs(((rax*0x8)+rip))'
but it should be M64cs((rax*0x8))
There is also
>>> from amoco.arch.x64 import cpu_x64
>>> i=cpu_x64.disassemble('\x64\x48\x8b\x04\x25\x28\0\0\0')
>>> str(i.operands[1])
'M64fs(rip+40)'
which should be M64fs(40)
.
The proposed patch below addresses these problems.
--- a/amoco/arch/x64/utils.py
+++ b/amoco/arch/x64/utils.py
@@ -96,12 +96,16 @@ def getModRM(obj,Mod,RM,data):
# check [disp16/32] case:
if (b is env.rbp or b is env.r13) and Mod==0:
- b=env.rip
- if seg is '': seg = env.cs
- Mod = 0b10
- if (b is env.bp) and Mod==0:
- b=env.cst(0,adrsz)
+ if s is 0 and seg is '':
+ seg = env.cs
+ bs = env.rip
+ else:
+ bs = s + env.cst(0,adrsz)
Mod = 0b10
+ elif s is 0:
+ bs = b
+ else:
+ bs = b + s
# now read displacement bytes:
if Mod==0b00:
d = 0
@@ -119,7 +123,6 @@ def getModRM(obj,Mod,RM,data):
obj.bytes += pack(d)
data = data[immsz:data.size]
d = d.int(-1)
- bs = b+s
if bs._is_cst and bs.v==0x0:
bs.v = d
bs.size = adrsz
Note that I wrote this patch such that bs = b + s
appears only when both are non-zero.
It allows me to replace this addition by the following piece of code, that I need because I want amoco to remember whether the binary conained 8b 04 18
(aka. mov (%rax,%rbx), %eax
) or 8b 04 03
(aka. mov (%rbx,%rax), %eax
).
# Instead of doing bs = b+s, which will reorder arguments, we do
# the addition manually, and change 'prop' so the many future calls
# to 'simplify' does not reorder the arguments
from amoco.cas import expressions
bs = expressions.op('+', b, s)
bs.prop |= 16
For example
>>> str(cpu_x64.disassemble('\x66\x0f\x7e\xc7'))
'movd rdi, xmm0'
should be movd edi, xmm0
Proposed patch:
--- a/amoco/arch/x64/spec_sse.py
+++ b/amoco/arch/x64/spec_sse.py
@@ -122,7 +122,7 @@ def sse_ps(obj,Mod,REG,RM,data):
@ispec_ia32("*>[ {0f}{6e} /r ]", mnemonic="MOVD", _inv=False)
@ispec_ia32("*>[ {0f}{7e} /r ]", mnemonic="MOVD", _inv=True)
def sse_pd(obj,Mod,REG,RM,data, _inv):
- if not check_nopfx(obj,set_opdsz_64): raise InstructionError(obj)
+ if not check_nopfx(obj,set_opdsz_32): raise InstructionError(obj)
REX = obj.misc['REX']
if REX is not None:
W,R,X,B = REX
@@ -889,7 +889,7 @@ def sse_pd(obj,Mod,RM,data):
@ispec_ia32("*>[ {0f}{6e} /r ]", mnemonic="MOVD", _inv=False)
@ispec_ia32("*>[ {0f}{7e} /r ]", mnemonic="MOVD", _inv=True)
def sse_pd(obj,Mod,REG,RM,data, _inv):
- if not check_66(obj,set_opdsz_64): raise InstructionError(obj)
+ if not check_66(obj,set_opdsz_32): raise InstructionError(obj)
op2,data = getModRM(obj,Mod,RM,data)
op1 = getregR(obj,REG,128)
obj.operands = [op1,op2] if not _inv else [op2,op1]
The second argument can be an imm8.
--- a/amoco/arch/x86/asm.py
+++ b/amoco/arch/x86/asm.py
@@ -1363,7 +1365,6 @@ def i_PSRLQ(i,fmap):
fmap[eip] = fmap[eip]+i.length
op1 = i.operands[0]
op2 = i.operands[1]
- assert op1.size==op2.size
src1 = fmap(op1)
src2 = fmap(op2)
val1 = (src1[i:i+64] for i in range(0,op1.size,64))
@@ -1374,7 +1376,6 @@ def i_PSLLQ(i,fmap):
fmap[eip] = fmap[eip]+i.length
op1 = i.operands[0]
op2 = i.operands[1]
- assert op1.size==op2.size
src1 = fmap(op1)
src2 = fmap(op2)
val1 = (src1[i:i+64] for i in range(0,op1.size,64))
Solved by the following patch.
Note that this patch is not sufficient to generate AT&T syntax that can be used as an input to GNU as.
--- a/amoco/arch/x86/formats.py
+++ b/amoco/arch/x86/formats.py
@@ -104,7 +161,7 @@ def opers_att(i):
s = []
for op in reversed(i.operands):
if op._is_mem:
- s.extend(deref(op))
+ s.extend(deref_att(op))
elif op._is_cst:
if i.misc['imm_ref'] is not None:
s.append((Token.Address,str(i.misc['imm_ref'])))
Hi,
i followed the example and got
>>> from amoco.arch.x86.env import *
>>> from amoco.cas import smt
>>> z = (eax^cst(0xcafebabe,32))+(ebx+(eax>>2))
>>> print z
T32
>>> print z.to_smtlib()
_top
>>> type(z)
<class 'amoco.cas.expressions.top'>
>>> z.to_smtlib().sexpr()
'_top'
It should be op. I removed "+(eax":
>>> z = (eax^cst(0xcafebabe,32))+(ebx>>2)
>>> type(z)
<class 'amoco.cas.expressions.op'>
>>> z.to_smtlib()
(eax ^ 3405691582) + LShR(ebx, 2)
>>> z.to_smtlib().sexpr()
'(bvadd (bvxor eax #xcafebabe) (bvlshr ebx #x00000002))'
>>> i = cpu_x86.disassemble('\xf2\x0f\x10\x40\x30')
>>> str(i)
'movsd xmm0, qword ptr [eax+48]'
>>> m=mapper()
>>> i(m)
>>> print(str(m))
(edi) <- M32(esi)
eip <- { | [0:32]->(eip+0x5) | }
esi <- { | [0:32]->(df ? (esi-0x4) : (esi+0x4)) | }
edi <- { | [0:32]->(df ? (edi-0x4) : (edi+0x4)) | }
which is the semantics of the movsd
instruction obtained from i = cpu_x86.disassemble('\xa5')
Proposed patch:
--- a/amoco/arch/x86/asm.py
+++ b/amoco/arch/x86/asm.py
@@ -374,6 +374,10 @@ def i_STOSD(i,fmap):
#------------------------------------------------------------------------------
def _movs_(i,fmap,l):
+ if len(i.operands):
+ # SSE2 movsd instruction
+ # no semantics available
+ return
counter = cx if i.misc['adrsz'] else ecx
loc = mem(edi,l*8)
src = fmap(mem(esi,l*8))
>>> ad=env.mem(env.cst(12,32),disp=4)
>>> str(ad.simplify())
'M32(0xc+4)'
I would have expected 'M32(0x10)'
Proposed patch:
--- a/amoco/cas/expressions.py
+++ b/amoco/cas/expressions.py
@@ -960,6 +960,9 @@ class ptr(exp):
if isinstance(self.seg,exp):
self.seg = self.seg.simplify()
if not self.base._is_def: self.disp=0
+ if self.disp and self.base._is_cst:
+ self.base += self.disp
+ self.disp = 0
return self
# default segment handler does not care about seg value:
When parsing relocatable files, some offsets for memory zones are determined by the relocation table, and therefore are not integer values, but symbols.
Amoco handles this relatively well (we can put an amoco expression in .disp) but there are a few minor issues.
One is for displaying; proposed patch is
--- a/amoco/cas/expressions.py
+++ b/amoco/cas/expressions.py
@@ -953,6 +953,7 @@ class ptr(exp):
return '%s(%s%s)'%(self.seg,self.base,d)
def disp_tostring(self,base10=True):
+ if hasattr(self.disp, '_is_cst'): return '+%s'%self.disp
if self.disp==0: return ''
if base10: return '%+d'%self.disp
c = cst(self.disp,self.size)
Another one, more tricky, is in memory zone management; proposed patch is
--- a/amoco/system/core.py
+++ b/amoco/system/core.py
@@ -197,7 +197,9 @@ class MemoryZone(object):
if i is None:
if len(self._map)==0: return [void(l*8)]
v0 = self._map[0].vaddr
- if (vaddr+l)<=v0: return [void(l*8)]
+ # Don't test if (vaddr+l)<=v0 because we need the test to be
+ # true if vaddr or v0 contain label/symbols
+ if not (v0<(vaddr+l)): return [void(l*8)]
res.append(void((v0-vaddr)*8))
l = (vaddr+l)-v0
vaddr = v0
Not sure how to minimize this, but I got an abort with the above error while attempting to classify AMD64 instructions. I'll see if I can dig into it some more.
>>> from amoco.arch.x64 import cpu_x64
>>> from amoco.cas.mapper import mapper
>>> i=cpu_x64.disassemble('\x66\x0f\xf3\xd3')
>>> str(i)
'psllq xmm2, xmm3'
>>> i(mapper())
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "amoco/arch/core.py", line 67, in __call__
i_xxx(self,map)
File "amoco/arch/x64/asm.py", line 1461, in i_PSLLQ
res = [v1<<src2.value for v1 in val1]
AttributeError: 'reg' object has no attribute 'value'
Some are missing, e.g. F3 0F 2C 45 F0
aka. cvttss2si -16(%ebp), %eax
.
Some are incorrectly parsed, e.g. F3 0F 5A 45 F0
aka. cvtss2sd -16(%ebp), %xmm0
but amoco says cvtss2sd eax, [ebp-16]
Here is the current output:
>>> from amoco.arch.x86 import cpu_x86
>>> i = cpu_x86.disassemble('\x66\x0f\xc5\xc1\x00')
>>> str(i)
'pextrw xmm0, ecx, 0x0'
While it should be pextrw eax, xmm1, 0
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.