git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH 2/4] compat: use git_mkdtemp()

From
René Scharfe <l.s.r@web.de>
Date
Dec 5, 2025, 12:11 UTC
Message-ID
<aebd0ffe-7914-4731-8f79-830bd3b5a147@web.de>
In-Reply-To
<20251203161154.GA44940@coredump.intra.peff.net>
On 12/3/25 5:11 PM, Jeff King wrote:
Show 29 quoted lines
> On Wed, Dec 03, 2025 at 11:52:30AM +0100, René Scharfe wrote:
> 
>> A file might appear at the path returned by mktemp(3) before we call
>> mkdir(2).  Use the more robust git_mkdtemp() instead, which retries a
>> number of times and doesn't need to call lstat(2).
> 
> This seems like a good idea. At least one of the mkdtemp() callers was
> using $TMPDIR, so this was a potential security-sensitive race.
> 
>> diff --git a/compat/mkdtemp.c b/compat/mkdtemp.c
>> index 1136119592..fcdd4e01e1 100644
>> --- a/compat/mkdtemp.c
>> +++ b/compat/mkdtemp.c
>> @@ -2,7 +2,5 @@
>>  
>>  char *gitmkdtemp(char *template)
>>  {
>> -	if (!*mktemp(template) || mkdir(template, 0700))
>> -		return NULL;
>> -	return template;
>> +	return git_mkdtemp(template);
>>  }
> 
> OK, so now we have gitmkdtemp() and git_mkdtemp(), which are also now
> the exact same thing. That seems overly complicated. ;)
> 
> This one is a conditionally-compiled wrapper for NO_MKDTEMP. But since
> we always have git_mkdtemp() available (as of your first patch), can't
> we just point at it directly with the macro?

A worthwhile cleanup if we stop at this point, but complicated by targeting three build systems, the CMake build being broken on macOS and me only knowing how to fake NO_MKDEMP for make, leaving half the build space untestable for me.

Show 52 quoted lines
> Like this:
> 
> diff --git a/Makefile b/Makefile
> index 237b56fc9d..8226aed443 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -1919,7 +1919,6 @@ ifdef NO_SETENV
>  endif
>  ifdef NO_MKDTEMP
>  	COMPAT_CFLAGS += -DNO_MKDTEMP
> -	COMPAT_OBJS += compat/mkdtemp.o
>  endif
>  ifdef MKDIR_WO_TRAILING_SLASH
>  	COMPAT_CFLAGS += -DMKDIR_WO_TRAILING_SLASH
> diff --git a/compat/mkdtemp.c b/compat/mkdtemp.c
> deleted file mode 100644
> index fcdd4e01e1..0000000000
> --- a/compat/mkdtemp.c
> +++ /dev/null
> @@ -1,6 +0,0 @@
> -#include "../git-compat-util.h"
> -
> -char *gitmkdtemp(char *template)
> -{
> -	return git_mkdtemp(template);
> -}
> diff --git a/compat/posix.h b/compat/posix.h
> index 067a00f33b..245386fa4a 100644
> --- a/compat/posix.h
> +++ b/compat/posix.h
> @@ -329,8 +329,7 @@ int gitsetenv(const char *, const char *, int);
>  #endif
>  
>  #ifdef NO_MKDTEMP
> -#define mkdtemp gitmkdtemp
> -char *gitmkdtemp(char *);
> +#define mkdtemp git_mkdtemp
>  #endif
>  
>  #ifdef NO_UNSETENV
> diff --git a/meson.build b/meson.build
> index f1b3615659..090b1911ca 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -1401,7 +1401,6 @@ checkfuncs = {
>    'strlcpy' : ['strlcpy.c'],
>    'strtoull' : [],
>    'setenv' : ['setenv.c'],
> -  'mkdtemp' : ['mkdtemp.c'],
>    'initgroups' : [],
>    'strtoumax' : ['strtoumax.c', 'strtoimax.c'],
>    'pread' : ['pread.c'],

We need to keep that dictionary entry to still define NO_MKDTEMP, and just empty the array of filenames.

contrib/buildsystems/CMakeLists.txt needs to be updated as well, like you do below (keep in function_checks, remove from compat_SOURCES).

At the very least this cleanup should be done in a separated patch, as it's harder than it looks.

Show 72 quoted lines
> We could even take it a step further and just always use git_mkdtemp(),
> like we were discussing elsewhere for mkstemp(). And then the makefile
> knobs can go away, too, like:
> 
> diff --git a/Makefile b/Makefile
> index 8226aed443..8ef5497c10 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -68,8 +68,6 @@ include shared.mak
>  #
>  # Define NO_UNSETENV if you don't have unsetenv in the C library.
>  #
> -# Define NO_MKDTEMP if you don't have mkdtemp in the C library.
> -#
>  # Define MKDIR_WO_TRAILING_SLASH if your mkdir() can't deal with trailing slash.
>  #
>  # Define NO_GECOS_IN_PWENT if you don't have pw_gecos in struct passwd
> diff --git a/compat/posix.h b/compat/posix.h
> index 245386fa4a..c49d67e653 100644
> --- a/compat/posix.h
> +++ b/compat/posix.h
> @@ -328,9 +328,7 @@ ssize_t git_pread(int fd, void *buf, size_t count, off_t offset);
>  int gitsetenv(const char *, const char *, int);
>  #endif
>  
> -#ifdef NO_MKDTEMP
>  #define mkdtemp git_mkdtemp
> -#endif
>  
>  #ifdef NO_UNSETENV
>  #define unsetenv gitunsetenv
> diff --git a/configure.ac b/configure.ac
> index cfb50112bf..8e61186f18 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -1140,12 +1140,6 @@ GIT_CHECK_FUNC(unsetenv,
>  [NO_UNSETENV=YesPlease])
>  GIT_CONF_SUBST([NO_UNSETENV])
>  #
> -# Define NO_MKDTEMP if you don't have mkdtemp in the C library.
> -GIT_CHECK_FUNC(mkdtemp,
> -[NO_MKDTEMP=],
> -[NO_MKDTEMP=YesPlease])
> -GIT_CONF_SUBST([NO_MKDTEMP])
> -#
>  # Define NO_INITGROUPS if you don't have initgroups in the C library.
>  GIT_CHECK_FUNC(initgroups,
>  [NO_INITGROUPS=],
> diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt
> index 479163ab5c..d28de227f5 100644
> --- a/contrib/buildsystems/CMakeLists.txt
> +++ b/contrib/buildsystems/CMakeLists.txt
> @@ -374,7 +374,7 @@ endif()
>  #function checks
>  set(function_checks
>  	strcasestr memmem strlcpy strtoimax strtoumax strtoull
> -	setenv mkdtemp poll pread memmem)
> +	setenv poll pread memmem)
>  
>  #unsetenv,hstrerror are incompatible with windows build
>  if(NOT WIN32)
> @@ -411,10 +411,6 @@ if(NOT HAVE_SETENV)
>  	list(APPEND compat_SOURCES compat/setenv.c)
>  endif()
>  
> -if(NOT HAVE_MKDTEMP)
> -	list(APPEND compat_SOURCES compat/mkdtemp.c)
> -endif()
> -
>  if(NOT HAVE_PREAD)
>  	list(APPEND compat_SOURCES compat/pread.c)
>  endif()

Right. Dropping this dependency and then deep cleaning the compat code is attractive and mostly sidesteps the build system complications. That's for a later series.

Ultimately you'd prefer banning mkdtemp(3) instead of automatically redirecting to git_mkdtemp(), though, no?

René
Previous: Jeff KingNext: Jeff King
Message 7 of 19 in “ban mktemp(3)”
  1. 0/4 ban mktemp(3)René Scharfe, Dec 3, 2025
  2. 1/4 wrapper: add git_mkdtemp()René Scharfe, Dec 3, 2025
  3. Chris TorekDec 4, 2025
  4. Junio C HamanoDec 5, 2025
  5. 2/4 compat: use git_mkdtemp()René Scharfe, Dec 3, 2025
  6. Jeff KingDec 3, 2025
  7. René ScharfeDec 5, 2025
  8. Jeff KingDec 6, 2025
  9. Junio C HamanoDec 5, 2025
  10. 3/4 compat: remove mingw_mktemp()René Scharfe, Dec 3, 2025
  11. 4/4 banned.h: ban mktemp(3)René Scharfe, Dec 3, 2025
  12. Jeff KingDec 3, 2025
  13. 0/5 ban mktemp(3)René Scharfe, Dec 6, 2025
  14. 1/5 wrapper: add git_mkdtemp()René Scharfe, Dec 6, 2025
  15. 2/5 compat: use git_mkdtemp()René Scharfe, Dec 6, 2025
  16. 3/5 compat: remove mingw_mktemp()René Scharfe, Dec 6, 2025
  17. 4/5 banned.h: ban mktemp(3)René Scharfe, Dec 6, 2025
  18. 5/5 compat: remove gitmkdtemp()René Scharfe, Dec 6, 2025
  19. Jeff KingDec 8, 2025

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.