All symbols created by nolibc are also visible to user code.
Syscall constants are expected to come from the kernel headers and
should not be made up by nolibc.
Refactor the logic to avoid defining syscall numbers.
Also the new code is easier to understand.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Acked-by: Willy Tarreau <w@1wt.eu>
Link: https://lore.kernel.org/r/20230917-nolibc-syscall-nr-v2-3-03863d509b9a@weissschuh.net
This allows nolic to work with `-nostdinc` avoiding any reliance on
system headers.
The implementation has been lifted from musl libc 1.2.4.
There is already an implementation of stdarg.h in include/linux/stdarg.h
but that is GPL licensed and therefore not suitable for nolibc.
The used compiler builtins have been validated to be at least available
since GCC 4.1.2 and clang 3.0.0.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Having __sysret() as an inline function has the unfortunate effect of
adding casts and large constants comparisons after the syscall returns
that significantly inflate some light code that's otherwise syscall-
heavy. Even nolibc-test grew by ~1%.
Let's switch back to a macro for this, and use it only with signed
arguments. Note that it is also possible to design a slightly more
complex macro covering unsigned and pointers but we only have 3 such
syscalls so it is pointless, and these were just addressed not to use
this macro anymore. Now for the argument (the local variable containing
the syscall return value), any negative value is an error, that results
in -1 being returned and errno to be assigned the opposite value.
This may be revisited again in the future if really needed but for now
let's get back to something sane.
Fixes: 428905da6e ("tools/nolibc: sys.h: add a syscall return helper")
Link: https://lore.kernel.org/lkml/20230806095846.GB10627@1wt.eu/
Link: https://lore.kernel.org/lkml/ZNKOJY+g66nkIyvv@1wt.eu/
Cc: Zhangjin Wu <falcon@tinylab.org>
Cc: David Laight <David.Laight@ACULAB.COM>
Cc: Thomas Weißschuh <thomas@t-8ch.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
The __sysret() function causes some undesirable casts so we'll revert
it. In order to keep it simple it will now only support integer return
values like in the past, so we must basically revert the changes that
were made to these 3 syscalls which return a pointer so that they
simply rely on their own test and the SET_ERRNO() macro.
Fixes: 4201cfce15 ("tools/nolibc: clean up sbrk() routine")
Fixes: 924e9539ae ("tools/nolibc: clean up mmap() routine")
Fixes: d27447bc2e ("tools/nolibc: sys.h: apply __sysret() helper")
Link: https://lore.kernel.org/lkml/20230806095846.GB10627@1wt.eu/
Link: https://lore.kernel.org/lkml/ZNKOJY+g66nkIyvv@1wt.eu/
Cc: Zhangjin Wu <falcon@tinylab.org>
Cc: David Laight <David.Laight@ACULAB.COM>
Cc: Thomas Weißschuh <thomas@t-8ch.de>
Signed-off-by: Willy Tarreau <w@1wt.eu>
getauxval() returns an unsigned long but the overall type of the ternary
operator needs to be signed.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
It's documented as returning int which is also implemented by glibc and
musl, so adopt that return type.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
According to manual page [1], posix spec [2] and source code like
arch/mips/kernel/syscall.c, for historic reasons, the sys_pipe() syscall
on some architectures has an unusual calling convention. It returns
results in two registers which means there is no need for it to do
verify the validity of a userspace pointer argument. Historically that
used to be expensive in Linux. These days the performance advantage is
negligible.
Nolibc doesn't support the unusual calling convention above, luckily
Linux provides a generic sys_pipe2() with an additional flags argument
from 2.6.27. If flags is 0, then pipe2() is the same as pipe(). So here
we use sys_pipe2() to implement the pipe().
pipe2() is also provided to allow users to use flags argument on demand.
[1]: https://man7.org/linux/man-pages/man2/pipe.2.html
[2]: https://pubs.opengroup.org/onlinepubs/9699919799/functions/pipe.html
Suggested-by: Zhangjin Wu <falcon@tinylab.org>
Link: https://lore.kernel.org/all/20230729100401.GA4577@1wt.eu/
Signed-off-by: Yuan Tan <tanyuan@tinylab.org>
Reviewed-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
The statx manpage [1] shows that it has been supported from Linux 4.11
and glibc 2.28, the Linux support can be checked for all of the
architectures with this command:
$ git grep -r statx v4.11 arch/ include/uapi/asm-generic/unistd.h \
| grep -E "aarch64|arm|mips|s390|x86|:include/uapi"
Besides riscv and loongarch, all of the nolibc supported architectures
have added sys_statx from Linux v4.11. riscv is mainlined to v4.15,
loongarch is mainlined to v5.19, both of them use the generic unistd.h,
so, they have added sys_statx from their first mainline versions.
The current oldest stable branch is v4.14, only reserving sys_statx
still preserves compatibility with all of the supported stable branches,
So, let's remove the old arch related and dependent sys_stat support
completely.
This is friendly to the future new architecture porting.
[1]: https://man7.org/linux/man-pages/man2/statx.2.html
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
a reverse operation of mkdir() is meaningful, add rmdir() here.
required by nolibc-test to remove /proc while CONFIG_PROC_FS is not
enabled.
Reviewed-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Both glibc and musl provide RB_ flags via <sys/reboot.h> for reboot(),
they don't need to include <linux/reboot.h>, let nolibc provide RB_
flags too.
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Fix up the error reported by scripts/checkpatch.pl:
ERROR: do not use assignment in if condition
#95: FILE: tools/include/nolibc/sys.h:95:
+ if ((ret = sys_brk(0)) && (sys_brk(ret + inc) == ret + inc))
Apply the new generic __sysret() to merge the SET_ERRNO() and return
lines.
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Do several cleanups together:
- Since all supported architectures have my_syscall6() now, remove the
#ifdef check.
- Move the mmap() related macros to tools/include/nolibc/types.h and
reuse most of them from <linux/mman.h>
- Apply the new generic __sysret() to convert the calling of sys_map()
to oneline code
Note, since MAP_FAILED is -1 on Linux, so we can use the generic
__sysret() which returns -1 upon error and still satisfy user land that
checks for MAP_FAILED.
Suggested-by: Willy Tarreau <w@1wt.eu>
Link: https://lore.kernel.org/lkml/20230702192347.GJ16233@1wt.eu/
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
No official reference states the errno range, here aligns with musl and
glibc and uses [-MAX_ERRNO, -1] instead of all negative ones.
- musl: src/internal/syscall_ret.c
- glibc: sysdeps/unix/sysv/linux/sysdep.h
The MAX_ERRNO used by musl and glibc is 4095, just like the one nolibc
defined in tools/include/nolibc/errno.h.
Suggested-by: Willy Tarreau <w@1wt.eu>
Link: https://lore.kernel.org/lkml/ZKKdD%2Fp4UkEavru6@1wt.eu/
Suggested-by: David Laight <David.Laight@ACULAB.COM>
Link: https://lore.kernel.org/linux-riscv/94dd5170929f454fbc0a10a2eb3b108d@AcuMS.aculab.com/
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Use __sysret() to shrink most of the library routines to oneline code.
Removed 266 lines of duplicated code.
Reviewed-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Most of the library routines share the same syscall return logic:
In general, a 0 return value indicates success. A -1 return value
indicates an error, and an error number is stored in errno. [1]
Let's add a __sysret() helper for the above logic to simplify the coding
and shrink the code lines too.
Thomas suggested to use inline function instead of macro for __sysret().
Willy suggested to make __sysret() be always inline.
[1]: https://man7.org/linux/man-pages/man2/syscall.2.html
Suggested-by: Willy Tarreau <w@1wt.eu>
Link: https://lore.kernel.org/linux-riscv/ZH1+hkhiA2+ItSvX@1wt.eu/
Suggested-by: Thomas Weißschuh <linux@weissschuh.net>
Link: https://lore.kernel.org/linux-riscv/ea4e7442-7223-4211-ba29-70821e907888@t-8ch.de/
Reviewed-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Compiling nolibc for rv32 got such errors:
nolibc/sysroot/riscv/include/sys.h: In function ‘sys_gettimeofday’:
nolibc/sysroot/riscv/include/sys.h:557:21: error: ‘__NR_gettimeofday’ undeclared (first use in this function); did you mean ‘sys_gettimeofday’?
557 | return my_syscall2(__NR_gettimeofday, tv, tz);
| ^~~~~~~~~~~~~~~~~
nolibc/sysroot/riscv/include/sys.h: In function ‘sys_lseek’:
nolibc/sysroot/riscv/include/sys.h:675:21: error: ‘__NR_lseek’ undeclared (first use in this function)
675 | return my_syscall3(__NR_lseek, fd, offset, whence);
| ^~~~~~~~~~
nolibc/sysroot/riscv/include/sys.h: In function ‘sys_wait4’:
nolibc/sysroot/riscv/include/sys.h:1341:21: error: ‘__NR_wait4’ undeclared (first use in this function)
1341 | return my_syscall4(__NR_wait4, pid, status, options, rusage);
If a syscall macro is not supported by a target platform, wrap it with
'#ifdef' and 'return -ENOSYS' for the '#else' branch, which lets the
other syscalls work as-is and allows developers to fix up the test
failures reported by nolibc-test one by one later.
This wraps all of the failed syscall macros with '#ifdef' and 'return
-ENOSYS' for the '#else' branch, so, all of the undeclared failures are
fixed.
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/linux-riscv/5e7d2adf-e96f-41ca-a4c6-5c87a25d4c9c@app.fastmail.com/
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Compiling nolibc for rv32 got such errors:
In file included from nolibc/sysroot/riscv/include/nolibc.h:99,
from nolibc/sysroot/riscv/include/errno.h:26,
from nolibc/sysroot/riscv/include/stdio.h:14,
from tools/testing/selftests/nolibc/nolibc-test.c:12:
nolibc/sysroot/riscv/include/sys.h:946:2: error: #error Neither __NR_ppoll nor __NR_poll defined, cannot implement sys_poll()
946 | #error Neither __NR_ppoll nor __NR_poll defined, cannot implement sys_poll()
| ^~~~~
nolibc/sysroot/riscv/include/sys.h:1062:2: error: #error None of __NR_select, __NR_pselect6, nor __NR__newselect defined, cannot implement sys_select()
1062 | #error None of __NR_select, __NR_pselect6, nor __NR__newselect defined, cannot implement sys_select()
If a syscall is not supported by a target platform, 'return -ENOSYS' is
better than '#error', which lets the other syscalls work as-is and
allows developers to fix up the test failures reported by nolibc-test
one by one later.
This converts all of the '#error' to 'return -ENOSYS', so, all of the
'#error' failures are fixed.
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/linux-riscv/5e7d2adf-e96f-41ca-a4c6-5c87a25d4c9c@app.fastmail.com/
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
In function ‘open’:
nolibc/sysroot/arm/include/sys.h:919:23: warning: ‘mode_t’ {aka ‘short unsigned int’} is promoted to ‘int’ when passed through ‘...’
919 | mode = va_arg(args, mode_t);
| ^
nolibc/sysroot/arm/include/sys.h:919:23: note: (so you should pass ‘int’ not ‘mode_t’ {aka ‘short unsigned int’} to ‘va_arg’)
nolibc/sysroot/arm/include/sys.h:919:23: note: if this code is reached, the program will abort
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Reviewed-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The ppoll and ppoll_time64 syscalls have 5 arguments, but we only
provide 4, align with kernel and add the missing sigsetsize argument.
Because the sigmask is NULL, the last sigsetsize argument is ignored,
keep it as 0 here is safe enough.
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Keep backwards compatibility through unions.
The compatibility macros like
#define st_atime st_atim.tv_sec
as documented in stat(3type) don't work for nolibc because it would
break with other stat-like structures that contain the field st_atime.
The stx_atime, stx_mtime, stx_ctime are in type of 'struct
statx_timestamp', which is incompatible with 'struct timespec', should
be converted explicitly.
/* include/uapi/linux/stat.h */
struct statx_timestamp {
__s64 tv_sec;
__u32 tv_nsec;
__s32 __reserved;
};
/* include/uapi/linux/time.h */
struct timespec {
__kernel_old_time_t tv_sec; /* seconds */
long tv_nsec; /* nanoseconds */
};
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Link: https://lore.kernel.org/linux-riscv/3a3edd48-1ace-4c89-89e8-9c594dd1b3c9@t-8ch.de/
Co-authored-by: Zhangjin Wu <falcon@tinylab.org>
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
[wt: squashed Zhangjin & Thomas' patches into one to preserve "bisectability"]
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
It will be used to disable core dumps from the child spawned to validate
the stack protector functionality.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The same constants and some more have been exposed to userspace via
linux/reboot.h for a long time.
To avoid conflicts and trim down nolibc a bit drop the custom
definitions.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
On s390 the first two arguments to the clone() syscall are swapped,
as documented in clone(2).
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Cc: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Most of nolibc is already using C89 comments.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This is useful for users and will also be used by a future testcase.
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
LoongArch and RISC-V 32-bit only have statx(). ARC, Hexagon, Nios2 and
OpenRISC have statx() and stat64() but not stat() or newstat(). Add
statx() and make stat() rely on statx() if necessary to make them happy.
We may just use statx() for all architectures in the future.
Signed-off-by: Feiyang Chen <chenfeiyang@loongson.cn>
Acked-by: Huacai Chen <chenhuacai@loongson.cn>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Include linux/fcntl.h for O_* and AT_*. asm/fcntl.h is included
by linux/fcntl.h, so it can be safely removed.
Signed-off-by: Feiyang Chen <chenfeiyang@loongson.cn>
Acked-by: Huacai Chen <chenhuacai@loongson.cn>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This can be useful to avoid attempting some privileged operations,
starting from the nolibc-test tool that gets two failures when not
privileged.
We call getuid32() and geteuid32() when they are defined, and fall
back to getuid() and geteuid() otherwise.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This function returns the page size used by the running kernel. The
page size value is taken from the auxiliary vector at 'AT_PAGESZ' key.
'getpagesize(2)' is assumed as a syscall becuase the manpage placement
of this function is in entry 2 ('man 2 getpagesize') despite there is
no real 'getpagesize(2)' syscall in the Linux syscall table. Define
this function in 'sys.h'.
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The historic nolibc code did not include asm/fcntl.h and had to define
the various O_RDWR etc macros in each arch-specific file (since such
values differ between certain archs). This was found at least once to
induce bugs due to wrong definitions. Let's get rid of all of them and
include asm/nolibc.h from sys.h instead. This was verified to work
properly on all supported architectures.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Use arch-x86_64 as a template. Not really different, but
we have our own mmap syscall which takes a structure instead
of discrete arguments.
Signed-off-by: Sven Schnelle <svens@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
After the nolibc includes were split to facilitate portability from
standard libcs, programs that include only what they need may miss
some symbols which are needed by libgcc. This is the case for raise()
which is needed by the divide by zero code in some architectures for
example.
Regardless, being able to include only the apparently needed files is
convenient.
Instead of trying to move all exported definitions to a single file,
since this can change over time, this patch takes another approach
consisting in including the nolibc header at the end of all standard
include files. This way their types and functions are already known
at the moment of inclusion, and including any single one of them is
sufficient to bring all the required ones.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
__NR_mmap2 was used for i386 but it's also needed for other archs such
as RISCV32 or ARM. Let's decide to use it based on the __NR_mmap2
definition as it's not defined on other archs.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
We return -ENOSYS when there's no syscall6() operation, but we must cast
it to void* to avoid a warning.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Implement mmap() and munmap(). Currently, they are only available for
architecures that have my_syscall6 macro. For architectures that don't
have, this function will return -1 with errno set to ENOSYS (Function
not implemented).
This has been tested on x86 and i386.
Notes for i386:
1) The common mmap() syscall implementation uses __NR_mmap2 instead
of __NR_mmap.
2) The offset must be shifted-right by 12-bit.
Acked-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This is essentially for completeness as it's not the most often used
in regtests.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This allows us to provide a minimal errno.h to ease porting applications
that use it.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
These syscalls never fail so there is no need to extract and set errno
for them.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Let's pass a vararg to open() so that it remains compatible with existing
code. The arg is only dereferenced when flags contain O_CREAT. The function
is generally not inlined anymore, causing an extra call (total 16 extra
bytes) but it's still optimized for constant propagation, limiting the
excess to no more than 16 bytes in practice when open() is called without
O_CREAT, and ~40 with O_CREAT, which remains reasonable.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The syscall definitions were moved to sys.h. They were arranged
in a more easily maintainable order, whereby the sys_xxx() and xxx()
functions were grouped together, which also enlights the occasional
mappings such as wait relying on wait4().
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
