Dataset Viewer
Auto-converted to Parquet Duplicate
SNIPPETS
stringlengths
2
1.64k
INTENTS
stringlengths
4
1.66k
_appendfile:
declare appendfile_label
_appendfile:
define the _appendfile label
_build:
declare _build label
_build:
define the _build label
_child:
define the _child label
_close:
declare _close label
_close:
define the _close label
_cmd:
define _cmd label
_cmd:
declare _cmd register
_cmd:
define the _cmd label
_continue:
define the _continue label
_decoder:
define _decoder_label
_dup2_loop:
define _dup2_loop label
_dup2_loop:
define the _dup2_loop label
_egg:
define the _egg label
_exec:
declare _exec _label
_exec:
define _exec label
_exec:
declare _exec label
_exec:
define the _exec label
_execline:
define _execline label
_execline:
define the _execline label
_exit:
define _exit label
_exit:
define the _exit label
_file:
_declare _file label
_file:
define _file_label
_file:
define the _file label
_ip equ 0x0100007f
define ip address as _ip = 0x0100007f
_ip equ 0x0100007f
define ip equal to 0x0100007f
_isegg:
define the _isegg label
_load_data:
define _load_data label
_load_data:
define the _load_data label
_next:
define the _next label
_param:
declare _param label
_param:
define the _param label
_params:
define _params label
_params:
define the _params label
_parent_read:
define the _parent_read label
_parent:
define the _parent label
_random:
define _random label
_random:
define the _random label
_read:
define _read label
_return:
define the _return label
_setsockopt:
define the _setsockopt label
_shell:
define the _shell label
_socket:
define the _socket label
_start:
declare _start label
_start:
_start function
_start:
_start label
_start:
_start
_start:
define the _start label
_start: \n jmp call_decoder
define the _start label and jump short to the call_decoder
_start: \n jmp call_decoder
define the _start label and jump to the call_decoder
_start: \n jmp call_decoder
define the _start label and jump to call_decoder
_start: \n jmp short call_decoder
define the _start label and jump short to the call_decoder
_start: \n jmp short call_decoder
define the _start label and jump short to call_decoder
_start: \n jmp short call_shellcode
define the _start label and jump short to the call_shellcode
_start: \n jmp short call_shellcode
define the _start label and jump short to call_shellcode
_start: \n jmp short call_shellcode
define the _start label and jump short to call_shellcode label
_start: \n jmp short enc
define the _start label and jump short to enc
_start: \n jmp short stage
define the _start label and jump short to stage
_user:
declare _user label
_user:
define the _user label
_while_loop:
define the _while_loop label
_write:
declare _write label
_write:
define _write label
_write:
define the _write label
1:
define the numeric label 1
1:
define the label 1
2:
define the label 2
2:
define the numeric label 2
3:
define the label 3
3:
define the numeric label 3
a_letter db 'd'
allocate one byte of memory for a_letter variable and initialize it to 'd'
aas
correct the result od the previous unpacked decimal substraction in al register
adc eax, 0x9c09ebea
sum hexadecimal value 0x9c09ebea with eax register. if cf is set, add 1 to eax.
add [ebp-4], esi
add esi into memory address ebp-4
add ah, bh
add the contents of the bh register into the ah register
add al, 0x2
add 0x2 to al
add al, 0x25
add 0x25 to the al register
add al, 0x3
add 0x3 to the contents in al and save the result in al
add al, 0x33
add 0x33 to the contents in al and save the result in al
add al, 0x40
add 0x40 to the al register
add al, 0x66
add 0x66 to the contents in al and save the result in al
add al, 0xa
add 0xa to the al register
add al, 0xb
add 0xb to the al register
add al, 2
add 2 to al register
add al, 2
add the value 2 to the contents of the al register
add al, 3
add 3 to al register
add al, 3
add 3 to al
add al, byte '0'
add byte '0' to al
add al, byte [edi]
add the byte in edi to the al register
add al,0x2
add hexadecimal value 0x2 to al register
add al,0x3
add the value 0x3 into al register
add al,0x33
add the hexadecimal value 0x33 to al register
add al,0x3b
add hexadecimal value 0x3b to al register
add al,0x66
add the hexadecimal value 0x66 to al register
add ax, [di + 20]
add memory[di + 20] to ax
add ax, 01
add 01 to ax
add ax, bx
add the contents of bx into ax
add bl, 0x2
load jexadecimal value 0x2 into bl register
End of preview. Expand in Data Studio

Shellcode_IA32

Shellcode_IA32 is a dataset containing more than 20 years of shellcodes from a variety of sources and is the largest collection of shellcodes in assembly available to date. We are currently extending the dataset. Up to now, we released three versions of the dataset.

Shellcode_IA32 was presented for the first time in the paper Shellcode_IA32: A Dataset for Automatic Shellcode Generation, accepted to the 1st Workshop on Natural Language Processing for Programming (NLP4Prog 2021). This version consists of 3,200 examples of instructions in assembly language for IA-32 (the 32-bit version of the x86 Intel Architecture) from publicly available security exploits. We collected assembly programs used to generate shellcode from exploit-db and from shell-storm. We enriched the dataset by adding examples of assembly programs for the IA-32 architecture from popular tutorials and books. This allowed us to understand how different authors and assembly experts comment and, thus, how to deal with the ambiguity of natural language in this specific context. Our dataset consists of 10% of instructions collected from books and guidelines, and the rest from real shellcodes.

Our focus is on Linux, the most common OS for security-critical network services. Accordingly, we added assembly instructions written with Netwide Assembler (NASM) for Linux.

Each line of Shellcode_IA32 dataset represents a snippet - intent pair. The snippet is a line or a combination of multiple lines of assembly code, built by following the NASM syntax. The intent is a comment in the English language.

We conducted an extensive experimental evaluation using the Shellcode_IA32 dataset in the journal paper Can we generate shellcodes via natural language? An empirical study, published in the Automated Software Engineering (AUSE) journal. The paper also contains further statistics on the dataset.

We further enriched the dataset (Extended_Shellcode_IA32) with additional samples of shellcodes collected from publicly available security exploits, reaching 5,900 unique pairs of assembly code snippets/English intents.

Our dataset also includes 1,374 intents (~23% of the dataset) that generate multiple lines of assembly code, separated by the newline character \n. These multi-line snippets contain many different assembly instructions (e.g., whole functions).

Downloads last month
47