patchwincred: fix line split of secret blob content

From: Marc Becker <becm@gmx.de>

operate on immutable blob data (wcsncpy_s still had invalid target size) split on newline character to avoid bleed-over on multi-line content

Signed-off-by: Marc Becker <becm@gmx.de>
---
    wincred: fix line split of secret blob content

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-2251%2Fbecm%2Ffix-wincred-secret-linesplit-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-2251/becm/fix-wincred-secret-linesplit-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/2251

 .../wincred/git-credential-wincred.c          | 86 ++++++++++++-------
 1 file changed, 55 insertions(+), 31 deletions(-)

"Marc Becker via GitGitGadget" <gitgitgadget@gmail.com> writes:

This needs a bit more work to make it more readable than a bulleted list of lowercase fragments.

When in doubt, keep in mind that the usual way to compose a log message of this project is to:

 - Give an observation on how the current system works in the
   present tense (so no need to say "Currently X is Y", or
   "Previously X was Y" to describe the state before your change;
   just "X is Y" is enough), and discuss what you perceive as a
   problem in it.

 - Propose a solution (optional---often, problem description
   trivially leads to an obvious solution in reader's minds).

 - Give commands to somebody editing the codebase to "make it so",
   instead of saying "This commit does X".

in this order.

 - It mentions wcsncpy_s having an invalid target size, but does not
   explain why it was invalid or the consequences. Is the issue that
   wcsncpy_s expects the buffer size in wide characters, but was
   being passed a size in bytes, which obviously cannot always
   agree?

 - It mentions "bleed-over on multi-line content", but does not
   describe the observable symptoms.  Is the issue that when the
   password is empty, the skipping by wcstok_s delimiter would cause
   the oauth_refresh_token line to be erroneously parsed as the
   password?

 - The final sentence should be an imperative command to the
   codebase, e.g., "Parse the blob in-place without copying and
   split lines manually using wmemchr()."

The original was already bad, but this makes it even worse to have the code nested too deeply. Would separating out the body of the for loop into a separate helper function, or perhaps standard tricks like this

	for (...) {
		if (!match_cred(...))
			continue;
		... rest of the loop dedented by one tab stop ...
	}

make it readable?

			/* our single line comment should look like this */

I do not do Windows and I do not often deal with wchar_t, so I do not know how much practitioners of code like this one cares, but would it be better to make the fact clear that we are not dealing with a regular 'char' by writing a wchar_t literal like this as L'\n'? This is not a correctness suggestion, but a readability one. Having a function prototype would coerse the parameter types, so you may end up passing L'\n' either way.

Here, "end" is of LPCWSTR type, aka "wchar_t *". So is "blob". The difference would give us how many wide characters are in there. That is not necessarily number of bytes starting at &blob[0].

This CR is also side, right?

Value is also "wchar_t *", so klen counts the length in wchar_t, which may be wider than a byte. So is

klen that counts number of wchar_t letters in refresh[] string.

So, is the memcmp() used to check if early part of blob[] match the refresh[] as a whole correct, or is it only checking an early half (or one fourth, depending on how much wider your wchar_t is compared to char) of the string?

From: Marc Becker <becm@gmx.de>

Invalid target size check (bytes instead of characters) for `wcsncpy_s` already led to memory corruption; d22a4884 just hid the error by making sure the target is always big enough. Single invocation of `wcstok_s` only cuts out first hit delimiter. Consecutive items would always start with a line feed character if separation consists of CRLF. Only exception is 1st item (due to following bug). Advancement to end of password line is missing. Password value is reused as extended credential item but likely filtered out due to value mismatch with accepted key.

Line split needs to be deterministic and code should be split up into smaller blocks.

Create separate methods for writing credential items and the complete credential blob content to tighten code in main credential loop. Reduce variable scope and nesting level. Use early continue/return to improve code readability. Use `wmemchr` to reliably detect wide-character-newline in immutable blob data without need to create a further copy.

Signed-off-by: Marc Becker <becm@gmx.de>
---
    wincred: fix line split of secret blob content
    
    Changes since v1:
    
     * move credential blob dissection and output to separate methods
       (maintainer request)
     * mark character constants as wide char
     * consistently treat sizes/lengths as (wide) character count

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-2251%2Fbecm%2Ffix-wincred-secret-linesplit-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-2251/becm/fix-wincred-secret-linesplit-v2
Pull-Request: https://github.com/gitgitgadget/git/pull/2251

Range-diff vs v1:

 1:  fb80cfc32a ! 1:  01a1039f70 wincred: fix line split of secret blob content
     @@ Metadata
      Author: Marc Becker <becm@gmx.de>
      
       ## Commit message ##
     -    wincred: fix line split of secret blob content
     +    wincred: refactor credential blob processing
      
     -    operate on immutable blob data (wcsncpy_s still had invalid target size)
     -    split on newline character to avoid bleed-over on multi-line content
     +    Invalid target size check (bytes instead of characters) for `wcsncpy_s`
     +    already led to memory corruption; d22a4884 just hid the error by making
     +    sure the target is always big enough.
     +    Single invocation of `wcstok_s` only cuts out first hit delimiter.
     +    Consecutive items would always start with a line feed character if
     +    separation consists of CRLF.
     +    Only exception is 1st item (due to following bug).
     +    Advancement to end of password line is missing. Password value is reused
     +    as extended credential item but likely filtered out due to value
     +    mismatch with accepted key.
     +
     +    Line split needs to be deterministic and code should be split up into
     +    smaller blocks.
     +
     +    Create separate methods for writing credential items and the complete
     +    credential blob content to tighten code in main credential loop.
     +    Reduce variable scope and nesting level. Use early continue/return to
     +    improve code readability.
     +    Use `wmemchr` to reliably detect wide-character-newline in immutable
     +    blob data without need to create a further copy.
      
          Signed-off-by: Marc Becker <becm@gmx.de>
      
     @@ contrib/credential/wincred/git-credential-wincred.c
       #include <wincred.h>
       
       /* common helpers */
     +@@ contrib/credential/wincred/git-credential-wincred.c: static void write_item(const char *what, LPCWSTR wbuf, int wlen)
     + 	free(buf);
     + }
     + 
     ++/*
     ++ * Write known credential item.
     ++ */
     ++static void write_credential_item(LPCWSTR data, int wlen)
     ++{
     ++	static const LPCWSTR refresh_token = L"oauth_refresh_token";
     ++	LPCWSTR value;
     ++	DWORD klen;
     ++	DWORD vlen;
     ++
     ++	/* find key/value separator for credential item */
     ++	if ((value = wmemchr(data, L'=', wlen)) == NULL)
     ++		return;
     ++	klen = value++ - data;
     ++	vlen = wlen - klen - 1;
     ++
     ++	/* write items known to git credential protocol */
     ++	if (klen == wcslen(refresh_token) && wmemcmp(data, refresh_token, klen) == 0)
     ++		write_item("oauth_refresh_token", value, vlen);
     ++}
     ++
     ++/*
     ++ * Write single credential block
     ++ * consisting of password and further (accepted) credential items.
     ++ */
     ++static void write_credential(const CREDENTIALW *cred)
     ++{
     ++	LPCWSTR end;
     ++	DWORD length;
     ++	LPCWSTR blob = (LPCWSTR)cred->CredentialBlob;
     ++	DWORD wlen = cred->CredentialBlobSize / sizeof(WCHAR);
     ++
     ++	/* check if content is single line */
     ++	if ((end = wmemchr(blob, L'\n', wlen)) == NULL) {
     ++		write_item("password", blob, wlen);
     ++		return;
     ++	}
     ++	/* determine current line length and remaining data size */
     ++	length = end++ - blob;
     ++	wlen -= length + 1;
     ++
     ++	/* skip carriage return at line end */
     ++	if (length && blob[length - 1] == L'\r')
     ++		--length;
     ++	write_item("password", blob, length);
     ++
     ++	while (1) {
     ++		/* key/value content starts on next line */
     ++		blob = end;
     ++
     ++		/* find line end */
     ++		if ((end = wmemchr(blob, L'\n', wlen)) == NULL) {
     ++			write_credential_item(blob, wlen);
     ++			return;
     ++		}
     ++		/* determine current line length and remaining data size */
     ++		length = end++ - blob;
     ++		wlen -= length + 1;
     ++
     ++		// skip carriage return at line end
     ++		if (length && blob[length - 1] == L'\r')
     ++			--length;
     ++		write_credential_item(blob, length);
     ++	}
     ++}
     ++
     + /*
     +  * Match an (optional) expected string and a delimiter in the target string,
     +  * consuming the matched text by updating the target pointer.
      @@ contrib/credential/wincred/git-credential-wincred.c: static void get_credential(void)
       {
       	CREDENTIALW **creds;
     @@ contrib/credential/wincred/git-credential-wincred.c: static void get_credential(
       
      -	/* search for the first credential that matches username */
      -	for (i = 0; i < num_creds; ++i)
     -+	/* search for the first credential that matches target and username */
     -+	for (int i = 0; i < num_creds; ++i) {
     - 		if (match_cred(creds[i], 0)) {
     +-		if (match_cred(creds[i], 0)) {
      -			write_item("username", creds[i]->UserName,
      -				creds[i]->UserName ? wcslen(creds[i]->UserName) : 0);
      -			if (creds[i]->CredentialBlobSize > 0) {
     @@ contrib/credential/wincred/git-credential-wincred.c: static void get_credential(
      -					line = wcstok_s(NULL, L"\r\n", &remaining_lines);
      -				}
      -				free(secret);
     -+			LPCWSTR username = creds[i]->UserName;
     -+			LPCWSTR blob = (LPCWSTR)creds[i]->CredentialBlob;
     -+			LPCWSTR end;
     -+			DWORD wlen;
     -+
     -+			write_item("username", username, username ? wcslen(username) : 0);
     -+
     -+			wlen = creds[i]->CredentialBlobSize / sizeof(WCHAR);
     -+
     -+			// check if content is single line
     -+			if ((end = wmemchr(blob, '\n', wlen)) == NULL) {
     -+				write_item("password", blob, wlen);
     - 			} else {
     +-			} else {
      -				write_item("password",
      -						(LPCWSTR)creds[i]->CredentialBlob,
      -						creds[i]->CredentialBlobSize / sizeof(WCHAR));
     -+				DWORD length = end++ - blob;
     +-			}
     +-			for (int j = 0; j < creds[i]->AttributeCount; j++) {
     +-				attr = creds[i]->Attributes + j;
     +-				if (!wcscmp(attr->Keyword, L"git_password_expiry_utc")) {
     +-					write_item("password_expiry_utc", (LPCWSTR)attr->Value,
     +-					attr->ValueSize / sizeof(WCHAR));
     +-					break;
     +-				}
     ++	/* search for the first credential that matches target and username */
     ++	for (int i = 0; i < num_creds; ++i) {
     ++		LPCWSTR username;
      +
     -+				// correct remaining size and drop carriage return at line end
     -+				wlen -= length + 1;
     -+				if (length && blob[length - 1] == '\r') {
     -+					--length;
     -+				}
     -+				write_item("password", blob, length);
     ++		if (!match_cred(creds[i], 0))
     ++			continue;
      +
     -+				// key/value content starting on next line
     -+				blob = end;
     -+				do {
     -+					LPCWSTR value;
     ++		username = creds[i]->UserName;
     ++		write_item("username", username, username ? wcslen(username) : 0);
      +
     -+					// find line end
     -+					if ((end = wmemchr(blob, '\n', wlen)) == NULL) {
     -+						length = wlen;
     -+					} else {
     -+						length = end++ - blob;
     -+						// correct remaining size and drop carriage return at line end
     -+						wlen -= length + 1;
     -+						if (length && blob[length - 1] == '\r') {
     -+							--length;
     -+						}
     -+					}
     -+					// find key/value separator for extended credential info
     -+					if ((value = wmemchr(blob, '=', length)) != NULL) {
     -+						static const LPCWSTR refresh = L"oauth_refresh_token";
     -+						DWORD klen = value - blob;
     ++		write_credential(creds[i]);
      +
     -+						// write entries known to git credential protocol
     -+						if (klen == wcslen(refresh) && memcmp(blob, refresh, klen) == 0) {
     -+							write_item("oauth_refresh_token", value + 1, length - klen - 1);
     -+						}
     -+					}
     -+				} while ((blob = end));
     - 			}
     - 			for (int j = 0; j < creds[i]->AttributeCount; j++) {
     --				attr = creds[i]->Attributes + j;
     -+				CREDENTIAL_ATTRIBUTEW *attr = creds[i]->Attributes + j;
     ++		for (int j = 0; j < creds[i]->AttributeCount; j++) {
     ++			CREDENTIAL_ATTRIBUTEW *attr = creds[i]->Attributes + j;
      +
     - 				if (!wcscmp(attr->Keyword, L"git_password_expiry_utc")) {
     --					write_item("password_expiry_utc", (LPCWSTR)attr->Value,
     --					attr->ValueSize / sizeof(WCHAR));
     -+					write_item("password_expiry_utc", (LPCWSTR)attr->Value, attr->ValueSize / sizeof(WCHAR));
     - 					break;
     - 				}
     ++			if (!wcscmp(attr->Keyword, L"git_password_expiry_utc")) {
     ++				write_item("password_expiry_utc", (LPCWSTR)attr->Value, attr->ValueSize / sizeof(WCHAR));
     ++				break;
       			}
     - 			break;
     +-			break;
       		}
      -
     ++		break;
      +	}
       	CredFree(creds);
       }

 .../wincred/git-credential-wincred.c          | 126 ++++++++++++------
 1 file changed, 87 insertions(+), 39 deletions(-)