Don't inject matchpathcon_filespec_add64() ifdef __x86_64__ (#463, Debian#1098481)

[nabijaczleweli]

As the code notes, it wants to add an

/* ABI backwards-compatible shim for non-LFS 32-bit systems */

it tries to detect these with

#if defined(_FILE_OFFSET_BITS) && _FILE_OFFSET_BITS == 64 && __BITS_PER_LONG < 64

which is correct with the added precondition that the ino_t without
-D_FILE_OFFSET_BITS=64 was actually u32 (i.e. it conflates all ILP32 systems into being non-LFS).

This is not the case on x32, for example, which is LFS; thus, the

static_assert(sizeof(unsigned long) == sizeof(__ino_t), "inode size mismatch");

assertion fails (__ino_t is the "kernel ino_t" type, which generally corresponds to the kernel's ulong, which is u64 on x32).

The correct spelling of the test for this is

#if ... && sizeof(__ino_t) == 4

but this is not statically solvable with the preprocessor.

Thus, we need to explicitly special-case this. __x86_64__ indicates one of two ABIs (LP64 (amd64) or ILP32 (x32)), both of which have ino_t=u64, and is the macro used for defining __INO_T_TYPE in the system headers, so it's the best fit here.

Fixes: commit 9395cc0 ("Always build for LFS mode on 32-bit archs.")
Closes: #463
Closes: Debian#1098481
Cc: @kainz

---

Insert -I../../libselinux/include and -L../../libselinux/src into subprograms where needed

These subprograms (sublibraries) use plain

#include <selinux/selinux.h>
cc ... -lselinux ...

&c. which includes/links to the system libselinux.

Naturally, this doesn't work if you don't have one.

These all came out of a plain build on x32.

[bachradsusi]

Hello, thanks for the patch!

In order to get the patch reviewed and accepted, it's necessary to send it to [email protected] mailing list. The best way is to use git send-email command, e.g.:

 git send-email --from='Your Name <[email protected]>' [email protected] --smtp-server=my.smtp.server --confirm=auto -1

See https://github.com/SELinuxProject/selinux/blob/main/CONTRIBUTING.md

[nabijaczleweli]

https://lore.kernel.org/selinux/jievtm5yb2pqqrn5p2idny7iipfoxnwn3gx5xtw6ycq6qqfgh5@tarta.nabijaczleweli.xyz/t/#u
https://lore.kernel.org/selinux/kwnnb3dirxubncipnklpimiv34bmetaxchzvzkoztsw62kx7zd@tarta.nabijaczleweli.xyz/t/#u

[mildsunrise]

@nabijaczleweli thank you for the patch, I was about to submit a very similar one for RV32. unfortunately this one doesn't fully fix the issue for all 32-bit LFS platforms, and I'm not sure what would be the best way to do that

[mildsunrise]

The definition is using __ino_t which is, as far as I can see, glibc specific. If using glibc-specific macros is acceptable, then the right criteria would be:

-#if defined(_FILE_OFFSET_BITS) && _FILE_OFFSET_BITS == 64 && __BITS_PER_LONG < 64
+#if defined(_FILE_OFFSET_BITS) && _FILE_OFFSET_BITS == 64 && __TIMESIZE < 64

or also

#ifndef __INO_T_MATCHES_INO64_T

[nabijaczleweli]

I think you nailed it actually yeah

[nabijaczleweli]

https://lore.kernel.org/selinux/cpbjftpok3lt54s6t7qmhzwbhl47vsz3efrxs45t5os7ztp3zg@tarta.nabijaczleweli.xyz/t/#u

[mildsunrise]

perfect, thank you :)

[mildsunrise]

hi sorry to bother you again, just wanted to note that (if I'm not mistaken) this patch breaks builds with USE_LFS=n on non-LFS 32-bit systems. I think nobody should be building with USE_LFS=n but since it's just a default, wanted to raise it just in case.

of course, the previous assertion

static_assert(sizeof(unsigned long) == sizeof(ino_t), "inode size mismatch");

was also wrong... I'd suggest a third clause, e.g.:

#if (defined(_FILE_OFFSET_BITS) && _FILE_OFFSET_BITS == 64) && !defined(__INO_T_MATCHES_INO64_T)
/*...*/
#elif (defined(_FILE_OFFSET_BITS) && _FILE_OFFSET_BITS == 64) || !defined(__INO_T_MATCHES_INO64_T)

static_assert(sizeof(uint64_t) == sizeof(ino_t), "inode size mismatch");

#else

static_assert(sizeof(uint32_t) == sizeof(ino_t), "inode size mismatch");

#endif

[nabijaczleweli]

I can repro this on i386, yeah; https://lore.kernel.org/selinux/jxjamggy4xaie53uyfuvriryqj4mtdc7gqr4gmjveyhwoukrxm@tarta.nabijaczleweli.xyz/t/#u

[mildsunrise]

perfect, thanks again!

[mildsunrise]

It seems that before this patch, libselinux didn't compile with musl for 32-bit platforms, and this patch further broke the build for 64-bit platforms too. @alyssais has posted an improved version of this patch which should fix the build for both 32-bit and 64-bit platforms :)