Python Script to disable ASLR and make nv fatbins read-only to reduce memory commit

fixNvPe.py

# Simple script to disable ASLR and make .nv_fatb sections read-only
# Requires: pefile  ( python -m pip install pefile )
# Usage:  fixNvPe.py --input path/to/*.dll

import argparse
import pefile
import glob
import os
import shutil

def main(args):
    failures = []
    for file in glob.glob( args.input, recursive=args.recursive ):
        print(f"\n---\nChecking {file}...")
        pe = pefile.PE(file, fast_load=True)
        nvbSect = [ section for section in pe.sections if section.Name.decode().startswith(".nv_fatb")]
        if len(nvbSect) == 1:
            sect = nvbSect[0]
            size = sect.Misc_VirtualSize
            aslr = pe.OPTIONAL_HEADER.IMAGE_DLLCHARACTERISTICS_DYNAMIC_BASE
            writable = 0 != ( sect.Characteristics & pefile.SECTION_CHARACTERISTICS['IMAGE_SCN_MEM_WRITE'] )
            print(f"Found NV FatBin! Size: {size/1024/1024:0.2f}MB  ASLR: {aslr}  Writable: {writable}")
            if (writable or aslr) and size > 0:
                print("- Modifying DLL")
                if args.backup:
                    bakFile = f"{file}_bak"
                    print(f"- Backing up [{file}] -> [{bakFile}]")
                    if os.path.exists( bakFile ):
                        print( f"- Warning: Backup file already exists ({bakFile}), not modifying file! Delete the 'bak' to allow modification")
                        failures.append( file )
                        continue
                    try:
                        shutil.copy2( file, bakFile)
                    except Exception as e:
                        print( f"- Failed to create backup! [{str(e)}], not modifying file!")
                        failures.append( file )
                        continue
                # Disable ASLR for DLL, and disable writing for section
                pe.OPTIONAL_HEADER.DllCharacteristics &= ~pefile.DLL_CHARACTERISTICS['IMAGE_DLLCHARACTERISTICS_DYNAMIC_BASE']
                sect.Characteristics = sect.Characteristics & ~pefile.SECTION_CHARACTERISTICS['IMAGE_SCN_MEM_WRITE']
                try:
                    newFile = f"{file}_mod"
                    print( f"- Writing modified DLL to [{newFile}]")
                    pe.write( newFile )
                    pe.close()
                    print( f"- Moving modified DLL to [{file}]")
                    os.remove( file )
                    shutil.move( newFile, file )
                except Exception as e:
                    print( f"- Failed to write modified DLL! [{str(e)}]")
                    failures.append( file )
                    continue

    print("\n\nDone!")
    if len(failures) > 0:
        print("***WARNING**** These files needed modification but failed: ")
        for failure in failures:
            print( f" - {failure}")







def parseArgs():
    parser = argparse.ArgumentParser( description="Disable ASLR and make .nv_fatb sections read-only", formatter_class=argparse.ArgumentDefaultsHelpFormatter )
    parser.add_argument('--input', help="Glob to parse", default="*.dll")
    parser.add_argument('--backup', help="Backup modified files", default=True, required=False)
    parser.add_argument('--recursive', '-r', default=False, action='store_true', help="Recurse into subdirectories")

    return parser.parse_args()


###############################
# program entry point
#
if __name__ == "__main__":
    args = parseArgs()
    main( args )

I've been running https://github.com/lucidrains/lightweight-gan mostly successfully with random crashes due to illegal memory accesses and BSoD. I can't really test if it's caused by this script or not, but it's something to keep an eye on.

But, great work on this (hopefully) temporary patch! Hopefully nvidia will get it together (even if we know they won't) and make an official fix for this.

pyd files are also dll files you may want to include this is your script, for example:

• D:\Users**\Miniconda3\Lib\site-packages\torchvision_C.pyd
• D:\Users**\Miniconda3\pkgs\torchvision-0.12.0.dev20220130-py39_cu113\Lib\site-packages\torchvision_C.pyd
  I also couldn't seem to get glob to go recursive so I modified to pathlib path
  `# Simple script to disable ASLR and make .nv_fatb sections read-only

Requires: pefile ( python -m pip install pefile )

Usage: fixNvPe.py --input path/to/*.dll

import argparse
import pefile
import os
import shutil
from pathlib import Path

def main(args):
failures = []
extension = args.ext if type(args.ext) == list else [args.ext]
for ext in extension:
for file in Path(args.input).rglob(f'.{ext}') if args.recursive else Path(args.input).glob(f'.{ext}'):
print(f"\n---\nChecking {file}...")
pe = pefile.PE(file, fast_load=True)
nvbSect = [ section for section in pe.sections if section.Name.decode().startswith(".nv_fatb")]
if len(nvbSect) == 1:
sect = nvbSect[0]
size = sect.Misc_VirtualSize
aslr = pe.OPTIONAL_HEADER.IMAGE_DLLCHARACTERISTICS_DYNAMIC_BASE
writable = 0 != ( sect.Characteristics & pefile.SECTION_CHARACTERISTICS['IMAGE_SCN_MEM_WRITE'] )
print(f"Found NV FatBin! Size: {size/1024/1024:0.2f}MB ASLR: {aslr} Writable: {writable}")
if (writable or aslr) and size > 0:
print("- Modifying DLL")
if args.backup:
bakFile = f"{file}_bak"
print(f"- Backing up [{file}] -> [{bakFile}]")
if os.path.exists( bakFile ):
print( f"- Warning: Backup file already exists ({bakFile}), not modifying file! Delete the 'bak' to allow modification")
failures.append( file )
continue
try:
shutil.copy2( file, bakFile)
except Exception as e:
print( f"- Failed to create backup! [{str(e)}], not modifying file!")
failures.append( file )
continue
# Disable ASLR for DLL, and disable writing for section
pe.OPTIONAL_HEADER.DllCharacteristics &= ~pefile.DLL_CHARACTERISTICS['IMAGE_DLLCHARACTERISTICS_DYNAMIC_BASE']
sect.Characteristics = sect.Characteristics & ~pefile.SECTION_CHARACTERISTICS['IMAGE_SCN_MEM_WRITE']
try:
newFile = f"{file}_mod"
print( f"- Writing modified DLL to [{newFile}]")
pe.write( newFile )
pe.close()
print( f"- Moving modified DLL to [{file}]")
os.remove( file )
shutil.move( newFile, file )
except Exception as e:
print( f"- Failed to write modified DLL! [{str(e)}]")
failures.append( file )
continue

print("\n\nDone!")
if len(failures) > 0:
    print("***WARNING**** These files needed modification but failed: ")
    for failure in failures:
        print( f" - {failure}")

def parseArgs():
parser = argparse.ArgumentParser( description="Disable ASLR and make .nv_fatb sections read-only", formatter_class=argparse.ArgumentDefaultsHelpFormatter )
parser.add_argument('--ext', help="extension", default=["dll","pyd","exe"])
parser.add_argument('--input', help="Glob to parse", default="*")
parser.add_argument('--backup', help="Backup modified files", default=True, required=False)
parser.add_argument('--recursive', '-r', default=False, action='store_true', help="Recurse into subdirectories")

return parser.parse_args()

###############################

program entry point

if name == "main":
args = parseArgs()
main( args )`

Hi @cobryan05, is this code available for redistribution under an explicit license?

Hi @cobryan05, is this code available for redistribution under an explicit license?

You can consider it public domain. Do whatever you'd like with it

Cheers! Thanks for the thoughtful write-up of the problem and the code 💙

Hi @cobryan05 , i have some question after using your script. I don't know what is that error about sdh.dll...

this is run with yolov5 and i try to run val.py

hi,i have some question after using your script. I don't know what is that error about sdh.dll...

@KlausRyu Your error is that you are running out of memory. My script will significantly reduce memory usage for files that have an .nv_fatb section, but that's it. It's definitely still possible you just simply don't have enough memory, even with the reduction.

Make sure you actually ran the script. It looks like in your case you should have run
python fixNvPe.py --input C:\Users\Admin\anaconda3\envs\newyolo\lib\site-packages\torch\lib\*.dll

but if that has successfully ran and you still hit that error, then you just don't have enough memory. Increase your page file size, decrease your number of workers, or add more RAM?

@cobryan05 Is it successful? cuz that have some files look like failed install
or my torch got some problem?

No, it wasn't successful. Read the log. "Not modifying file. Delete the 'bak' to allow modification" Perhaps you ran it once, then upgraded Torch and it overwrote the DLLs, so you have files that need modification but it won't modify them since it won't overwrite the backup.

No, it wasn't successful. Read the log. "Not modifying file. Delete the 'bak' to allow modification" Perhaps you ran it once, then upgraded Torch and it overwrote the DLLs, so you have files that need modification but it won't modify them since it won't overwrite the backup.

So any solutions to solve this problem?

The log says:
"Warning: Backup file already exists (C:\User\Admin\anaconda3\envs\newyolo\lib\site-packages\torch\lib\torch_cuda_cu.dll_bak), not modifying file! Delete the 'bak' file to allow modification."

This is telling you that the file C:\User\Admin\anaconda3\envs\newyolo\lib\site-packages\torch\lib\torch_cuda_cu.dll_bak already exists, so it is not modifying the C:\User\Admin\anaconda3\envs\newyolo\lib\site-packages\torch\lib\torch_cuda_cu.dll. You must delete the file that ends with '_bak' to allow modifying the file.

Thank you! Works like a charm for paging file errors running https://github.com/minimaxir/aitextgen.

Hi @cobryan05 , I'm getting this output that says :
" Failed to write modified DLL! [[WinError 5] Access is denied: .. "

Can I know what should I do to enable it to be modified?

@szan12 For it to be getting 'access denied' in your User directory, I would assume that it means the file is in use. Try restarting your computer and then running it, or try typing "taskkill /f /im python.exe" in your command prompt before running it (this will forcefully close any python process you have running). If that still fails, try running from a cmd prompt that is "Run as administrator", but that shouldn't be necessary in the 'user' directory

i can't remove the folder's read-only attributes, when i removed, and it just automatically recovery the attributes.

Thank you for sharing codes and troubleshooting. Yet, this didn't work for me. I am surprised I can't train a FCN8s on NVIDIA 3090 with 24 Gb of VRAM. I could have done that in an older server with two 12 Gb cards. The main difference is that the old server ran Keras on linux. So I'll have to change OS and maybe go back to keras/TF.

Best regards!