Add cheap defense mechanisms (PR #1171)
Alejandro Colomar
maintainerHi,
Quoted 3 lines · On Sun, Feb 16, 2025 at 06:15:18PM -0800, Karlson2k wrote:
Karlson2k left a comment (shadow-maint/shadow#1171)
Doesn't use of glibc extensions break functioning with non-glibc, like musl?
Hmmm, I didn't know musl doesn't support this. It would be interesting to get them to support it. I've CCd several interested parties in this email.
Quoted 8 lines
Isn't it safe to use constructs like
``` C shadow = fopen (SGROUP_FILE, "re"); if (NULL == shadow ) shadow = fopen (SGROUP_FILE, "r"); ```?
Is 'e' only available in glibc? Do other libraries consciously not support O_CLOEXEC in fopen(3)?
I see that POSIX.1-2024 added the 'e' mode string character, so we're using standard features (yeah, very modern ones, but still standard). Is there any reason to not implement them, or is it just a matter of time and contributors?
<https://pubs.opengroup.org/onlinepubs/9799919799/functions/fopen.html>
Quoted 1 line
Or, alternatively, detect extension in
configure?
If we have to...
Have a lovely day!
Alex
Signature · 1 line
<https://www.alejandro-colomar.es/>
Alejandro Colomar
maintainerHi,
Quoted 4 lines · Alejandro Colomar
Quoted 3 lines · On Sun, Feb 16, 2025 at 06:15:18PM -0800, Karlson2k wrote:
Karlson2k left a comment (shadow-maint/shadow#1171)
Doesn't use of glibc extensions break functioning with non-glibc, like musl?
After reading musl's source code, it does support 'e'. And it does so since 2012. Which libc are you using? And which version?
alx@devuan:~/src/musl/libc/master$ grepc fopen .
./include/stdio.h:FILE *fopen(const char *__restrict, const char *__restrict);
./src/stdio/fopen.c:FILE *fopen(const char *restrict filename, const char *restrict mode)
{
FILE *f;
int fd;
int flags; /* Check for valid initial mode character */
if (!strchr("rwa", *mode)) {
errno = EINVAL;
return 0;
}
/* Compute the flags to pass to open() */
flags = __fmodeflags(mode);
fd = sys_open(filename, flags, 0666);
if (fd < 0) return 0;
if (flags & O_CLOEXEC)
__syscall(SYS_fcntl, fd, F_SETFD, FD_CLOEXEC);
f = __fdopen(fd, mode);
if (f) return f;
__syscall(SYS_close, fd);
return 0;
}
alx@devuan:~/src/musl/libc/master$ grepc __fmodeflags .
./src/stdio/__fmodeflags.c:int __fmodeflags(const char *mode)
{
int flags;
if (strchr(mode, '+')) flags = O_RDWR;
else if (*mode == 'r') flags = O_RDONLY;
else flags = O_WRONLY;
if (strchr(mode, 'x')) flags |= O_EXCL;
if (strchr(mode, 'e')) flags |= O_CLOEXEC;
if (*mode != 'r') flags |= O_CREAT;
if (*mode == 'w') flags |= O_TRUNC;
if (*mode == 'a') flags |= O_APPEND;
return flags;
}
./src/internal/stdio_impl.h:hidden int __fmodeflags(const char *);
alx@devuan:~/src/musl/libc/master$ git blame -- src/stdio/__fmodeflags.c | grep CLOEXEC
892cafff6 (Rich Felker 2012-10-24 21:16:06 -0400 11) if (strchr(mode, 'e')) flags |= O_CLOEXEC;
alx@devuan:~/src/musl/libc/master$ git show 892cafff6 | grep -e CLOEXEC -e ^diff | head -n5
diff --git a/src/internal/stdio_impl.h b/src/internal/stdio_impl.h
diff --git a/src/stdio/__fmodeflags.c b/src/stdio/__fmodeflags.c
+ if (strchr(mode, 'e')) flags |= O_CLOEXEC;
diff --git a/src/stdio/fopen.c b/src/stdio/fopen.c
- if (strchr(mode, 'e')) flags |= O_CLOEXEC;
alx@devuan:~/src/musl/libc/master$ git blame 892cafff6^ -- src/stdio/fopen.c | grep CLOEXEC
8582a6e9f (Rich Felker 2012-09-29 18:09:34 -0400 20) if (strchr(mode, 'e')) flags |= O_CLOEXEC;
alx@devuan:~/src/musl/libc/master$ git log -1 8582a6e9f
commit 8582a6e9f25dd7b87d72961f58008052a4cac473
Author: Rich Felker <[email protected]>
Date: Sat Sep 29 18:09:34 2012 -0400
add 'e' modifier (close-on-exec) to fopen and fdopen
this feature will be in the next version of POSIX, and can be used
internally immediately. there are many internal uses of fopen where
close-on-exec is needed to fix bugs.Cheers,
Alex
Signature · 1 line
<https://www.alejandro-colomar.es/>
Quoted 10 lines · Alejandro Colomar
Hi,
Quoted 3 lines · On Sun, Feb 16, 2025 at 06:15:18PM -0800, Karlson2k wrote:
Karlson2k left a comment (shadow-maint/shadow#1171)
Doesn't use of glibc extensions break functioning with non-glibc, like musl?
Hmmm, I didn't know musl doesn't support this. It would be interesting to get them to support it. I've CCd several interested parties in this email.
It's in the latest POSIX and we have supported it for a long time as POSIX-future (since 2012/release 0.9.7).
Quoted 7 lines
Quoted 7 lines
Isn't it safe to use constructs like
``` C shadow = fopen (SGROUP_FILE, "re"); if (NULL == shadow ) shadow = fopen (SGROUP_FILE, "r"); ```?
Unfortunately this doesn't work because it's UB to pass any modes but the standards-specified ones.
In any case use of fopen is just gratuitously bad for software that targets POSIX. The right way to do things is a two-step open+fdopen. This avoids needing to depend on new features to open and lets you use all the modern open flags, openat if needed, etc.
Rich
P.S. Had to omit shadow-utils <~hallyn/[email protected]> from CC because my mail software rejects / in an address... gotta fix that. Apologies.
Hi,
Quoted 10 lines · Rich Felker
Quoted 7 lines · Alejandro Colomar
Quoted 7 lines
Isn't it safe to use constructs like
``` C shadow = fopen (SGROUP_FILE, "re"); if (NULL == shadow ) shadow = fopen (SGROUP_FILE, "r"); ```?
Quoted 2 lines
Unfortunately this doesn't work because it's UB to pass any modes but the standards-specified ones.
You are right, according to POSIX, failure with EINVAL is not guaranteed.
That is not nice as the function is allowed to perform not as requested without any indication.
Quoted 4 lines
In any case use of fopen is just gratuitously bad for software that targets POSIX. The right way to do things is a two-step open+fdopen. This avoids needing to depend on new features to open and lets you use all the modern open flags, openat if needed, etc.
I agree. This would be the best approach.
While dietlibc does not support "e" mode and glibc, musl, uclibc-ng support "e" mode, checking for particular library is bad.
Checking for feature support is not reliable (function may not return error, while the flag is ignored or even used differently) and not possible when cross-compiling.
Signature · 2 lines
Best, Evgeny
Alejandro Colomar
maintainerHi Rich,
Quoted 2 lines · Rich Felker
It's in the latest POSIX and we have supported it for a long time as POSIX-future (since 2012/release 0.9.7).
Thanks! That agrees with my own research.
Quoted 10 lines
Quoted 7 lines
Quoted 7 lines
Isn't it safe to use constructs like
``` C shadow = fopen (SGROUP_FILE, "re"); if (NULL == shadow ) shadow = fopen (SGROUP_FILE, "r"); ```?
Unfortunately this doesn't work because it's UB to pass any modes but the standards-specified ones.
Makes sense. Thanks!
Quoted 4 lines
In any case use of fopen is just gratuitously bad for software that targets POSIX. The right way to do things is a two-step open+fdopen. This avoids needing to depend on new features to open and lets you use all the modern open flags, openat if needed, etc.
Hmmm, thanks for the idea! This code is old, so we didn't actually write it; we only added 'e' recently to it. If we need to write new code, I'll take into consideration doing that as a two-step process, maybe adding a wrapper around that process.
Have a lovely day!
Alex
Quoted 3 lines
P.S. Had to omit shadow-utils <~hallyn/[email protected]> from CC because my mail software rejects / in an address... gotta fix that. Apologies.
No problem. Thanks!
Signature · 1 line
<https://www.alejandro-colomar.es/>