PowerShell stripping double quotes from command line arguments

Viewed 41652

Recently I have been having some trouble using GnuWin32 from PowerShell whenever double quotes are involved.

Upon further investigation, it appears PowerShell is stripping double quotes from command line arguments, even when properly escaped.

PS C:\Documents and Settings\Nick> echo '"hello"'
"hello"
PS C:\Documents and Settings\Nick> echo.exe '"hello"'
hello
PS C:\Documents and Settings\Nick> echo.exe '\"hello\"'
"hello"

Notice that the double quotes are there when passed to PowerShell's echo cmdlet, but when passed as an argument to echo.exe, the double quotes are stripped unless escaped with a backslash (even though PowerShell's escape character is a backtick, not a backslash).

This seems like a bug to me. If I am passing the correct escaped strings to PowerShell, then PowerShell should take care of whatever escaping may be necessary for however it invokes the command.

What is going on here?

For now, the fix is to escape command line arguments in accordance with these rules (which seem to be used (indirectly) by the CreateProcess API call which PowerShell uses to invoke .exe files):

  • To pass a double quote, escape with a backslash: \" -> "
  • To pass a one or more backslashes followed by a double quote, escape each backslash with another backslash and escape the quote: \\\\\" -> \\"
  • If not followed by a double quote, no escaping is necessary for backslashes: \\ -> \\

Note that further escaping of double quotes may be necessary to escape the double quotes in the Windows API escaped string to PowerShell.

Here are some examples, with echo.exe from GnuWin32:

PS C:\Documents and Settings\Nick> echo.exe "\`""
"
PS C:\Documents and Settings\Nick> echo.exe "\\\\\`""
\\"
PS C:\Documents and Settings\Nick> echo.exe "\\"
\\

I imagine that this can quickly become hell if you need to pass a complicated command line parameter. Of course, none of this documented in the CreateProcess() or PowerShell documentation.

Also note that this is not necessary to pass arguments with double quotes to .NET functions or PowerShell cmdlets. For that, you need only escape your double quotes to PowerShell.

Edit: As Martin pointed out in his excellent answer, this is documented in the CommandLineToArgv() function (which the CRT uses to parse the command line arguments) documentation.

7 Answers

TL;DR

If you just want a solution for Powershell 5, see:

ConvertTo-ArgvQuoteForPoSh.ps: Powershell V5 (and C# Code) to allow escaping native command arguments

The Question I will try to answer

..., it appears PowerShell is stripping double quotes from command line arguments, even when properly escaped.

PS C:\Documents and Settings\Nick> echo.exe '"hello"'
hello 
PS C:\Documents and Settings\Nick> echo.exe '\"hello\"' 
"hello"

Notice that the double quotes are there when passed to PowerShell's echo cmdlet, but when passed as an argument to echo.exe, the double quotes are stripped unless escaped with a backslash (even though PowerShell's escape character is a backtick, not a backslash).

This seems like a bug to me. If I am passing the correct escaped strings to PowerShell, then PowerShell should take care of whatever escaping may be necessary for however it invokes the command.

What is going on here?

The Non-Powershell Background

The fact that you need to escape the quotes with backslashes \ has nothing to to with powershell, but with the CommandLineToArgvW function that is used by all msvcrt and C# programs to build the argv array from the single-string command line that the Windows process gets passed.

The details are explained at Everyone quotes command line arguments the wrong way and it basically boils down to the fact that this function historically has very uninutitive escaping rules:

  • 2n backslashes followed by a quotation mark produce n backslashes followed by begin/end quote. This does not become part of the parsed argument, but toggles the "in quotes" mode.
  • (2n) + 1 backslashes followed by a quotation mark again produce n backslashes followed by a quotation mark literal ("). This does not toggle the "in quotes" mode.
  • n backslashes not followed by a quotation mark simply produce n backslashes.

leading to the described generic escaping function (shortquote of the logic here):

CommandLine.push_back (L'"');

for (auto It = Argument.begin () ; ; ++It) {
      unsigned NumberBackslashes = 0;

      while (It != Argument.end () && *It == L'\\') {
              ++It;
              ++NumberBackslashes;
      }

      if (It == Argument.end ()) {
              // Escape all backslashes, but let the terminating
              // double quotation mark we add below be interpreted
              // as a metacharacter.
              CommandLine.append (NumberBackslashes * 2, L'\\');
              break;
      } else if (*It == L'"') {
              // Escape all backslashes and the following
              // double quotation mark.
              CommandLine.append (NumberBackslashes * 2 + 1, L'\\');
              CommandLine.push_back (*It);
      } else {
              // Backslashes aren't special here.
              CommandLine.append (NumberBackslashes, L'\\');
              CommandLine.push_back (*It);
      }
}

CommandLine.push_back (L'"');

The Powershell specifics

Now, up to Powershell 5 (including PoSh 5.1.18362.145 on Win10/1909) PoSh knows basically diddly about these rules, nor should it arguably, because these rules are not really general, because any executable you call could, in theory, use some other means to interpret the passed command line.

Which leads us to -

The Powershell Quoting Rules

What PoSh does do however is try to figure out whether the strings you pass it as arguments to the native commands need to be quoted because they contain whitespace.

PoSh - in contrast to cmd.exe - does a lot more parsing on the command you hand it, since it has to resolve variables and knows about multiple arguments.

So, given a command like

$firs  = 'whaddyaknow'
$secnd = 'it may have spaces'
$third = 'it may also have "quotes" and other \" weird \\ stuff'
EchoArgs.exe $firs $secnd $third

Powershell has to take a stance on how to create the single string CommandLine for the Win32 CreateProcess (or rather the C# Process.Start) call it will evetually have to do.

The approach Powershell takes is weird and got more complicated in PoSh V7 , and as far as I can follow, it's got to do how powershell treats unbalanced quotes in unquoted string. The long stories short is this:

Powershell will auto-quote (enclose in <">) a single argument string, if it contains spaces and the spaces don't mix with an uneven number of (unsescaped) double quotes.

The specific quoting rules of PoSh V5 make it impossible to pass a certain category of string as single argument to a child process.

PoSh V7 fixed this, so that as long as all quotes are \" escaped -- which they need to be anyway to get them through CommandLineToArgvW -- we can pass any aribtrary string from PoSh to a child executable that uses CommandLineToArgvW.

Here's the rules as C# code as extracted from the PoSh github repo for a tool class of ours:

PoSh Quoting Rules V5

    public static bool NeedQuotesPoshV5(string arg)
    {
        // bool needQuotes = false;
        int quoteCount = 0;
        for (int i = 0; i < arg.Length; i++)
        {
            if (arg[i] == '"')
            {
                quoteCount += 1;
            }
            else if (char.IsWhiteSpace(arg[i]) && (quoteCount % 2 == 0))
            {
                // needQuotes = true;
                return true;
            }
        }
        return false;
    }

PoSh Quoting Rules V7

    internal static bool NeedQuotesPoshV7(string arg)
    {
        bool followingBackslash = false;
        // bool needQuotes = false;
        int quoteCount = 0;
        for (int i = 0; i < arg.Length; i++)
        {
            if (arg[i] == '"' && !followingBackslash)
            {
                quoteCount += 1;
            }
            else if (char.IsWhiteSpace(arg[i]) && (quoteCount % 2 == 0))
            {
                // needQuotes = true;
                return true;
            }

            followingBackslash = arg[i] == '\\';
        }
        // return needQuotes;
        return false;
    }

Oh yeah, and they also added in a half baked attempt to correctly escape the and of the quoted string in V7:

if (NeedQuotes(arg))
{
      _arguments.Append('"');
      // need to escape all trailing backslashes so the native command receives it correctly
      // according to http://www.daviddeley.com/autohotkey/parameters/parameters.htm#WINCRULESDOC
      _arguments.Append(arg);
      for (int i = arg.Length - 1; i >= 0 && arg[i] == '\\'; i--)
      {
              _arguments.Append('\\');
      }

      _arguments.Append('"');

The Powershell Situation

Input to EchoArgs             | Output V5 (powershell.exe)  | Output V7 (pwsh.exe)
===================================================================================
EchoArgs.exe 'abc def'        | Arg 0 is <abc def>          | Arg 0 is <abc def>
------------------------------|-----------------------------|---------------------------
EchoArgs.exe '\"nospace\"'    | Arg 0 is <"nospace">        | Arg 0 is <"nospace">
------------------------------|-----------------------------|---------------------------
EchoArgs.exe '"\"nospace\""'  | Arg 0 is <"nospace">        | Arg 0 is <"nospace">
------------------------------|-----------------------------|---------------------------
EchoArgs.exe 'a\"bc def'      | Arg 0 is <a"bc>             | Arg 0 is <a"bc def>
                              | Arg 1 is <def>              |
------------------------------|-----------------------------|---------------------------
   ...

I'm snipping further examples here for time reasons. They shouldn't add overmuch to the answer anyways.

The Powershell Solution

To pass arbitrary Strings from Powershell to a native command using CommandLineToArgvW, we have to:

  • properly escape all quotes and Backslashes in the source argument
    • This means recognizing the special string-end handling for backslashes that V7 has. (This part is not implemented in the code below.)
  • and determine whether powershell will auto-quote our escaped string and if it won't auto-quote it, quote it ourselves.
    • and make sure that the string we quoted ourselves then doesn't get auto-quoted by powershell: This is what breaks V5.

Powershell V5 Source code for correctly escaping all arguments to any native command

I've put the full code on Gist, as it got too long to include here: ConvertTo-ArgvQuoteForPoSh.ps: Powershell V5 (and C# Code) to allow escaping native command arguments

  • Note that this code tries it's best, but for some strings with quotes in the payload and V5 you simply must add in leading space to the arguments you pass. (See code for logic details).

This seems to be fixed in recent versions of PowerShell at the time of this writing, so it is no longer something to worry about.

If you still think you see this issue, remember it may be related to something else, such as the program that invokes PowerShell, so if you cannot reproduce it when invoking PowerShell directly from a command prompt or the ISE, you should debug elsewhere.

For example, I found this question when investigating a problem of disappearing quotes when running a PowerShell script from C# code using Process.Start. The issue was actually C# Process Start needs Arguments with double quotes - they disappear.

With PowerShell 7.2.0, it is finally possible for arguments passed to native executables to behave as expected. This is currently an experimental feature and needs to be enabled manually.

Enable-ExperimentalFeature PSNativeCommandArgumentPassing

After that edit your PSProfile, for example, using notepad:

notepad.exe $PROFILE

Add $PSNativeCommandArgumentPassing = 'Standard' to the top of the file. You may instead also use $PSNativeCommandArgumentPassing = 'Windows' which uses the Legacy behaviour for some native executables. The differences are documented in this pull request.

Finally, restart PowerShell. Command arguments will no longer have quotes removed.


The new behaviour can be verified with this little C program:

#include <stdio.h>

int main(int argc, char** argv) {
    for (int i = 1; i < argc; i++) {
        puts(argv[i]);
    }
    return 0;
}

Compile it with gcc and pass in some arguments with quotes, like a JSON string.

> gcc echo-test.c
> ./a.exe '{"foo": "bar"}'

With the Legacy behaviour, the output is {foo: bar}. However, with the Standard option, the output becomes {"foo": "bar"}.

Oh dear. Clearly trying to escape double quotes to get them into PowerShell from the command line, or worse, some other language you are using to generate such a command line, or execution environments which might chain PowerShell scripts, can be a colossal waste of time.

As an attempt at a practical solution, what can we do instead? Silly-looking workarounds can sometimes be effective:

powershell Write-Host "'say ___hi___'.Replace('___', [String][Char]34)"

But it depends a lot on how this is being executed. Note that if you want that command to have the same results when pasted in PowerShell instead of run from command prompt, you need those outer double quotes! Because the hosting Powershell turns the expression into a string object which becomes just one more parameter to 'powershell.exe'

PS> powershell Write-Host 'say ___hi___'.Replace('___', [String][Char]34)

Which then, I guess, parses its arguments as Write-Host say "hi"

So the quotes you are trying so hard to reintroduce with string.Replace() will just disappear!

Related