Re: [LTP] [PATCH v4] getcwd01: Use syscall directly check invalid argument

[Petr Vorel <pvorel@suse.cz>]

Hi Wei,

> Fixes: #1084

> User space wrap getcwd with different implementation, for example
> glibc will directly input parameter into kernel in normal situation
> but uclibc-ng and musl will malloc buffer when buffer is NULL, so for
> uclibc and musl the parameter size will be ignored. Use system call
> directly check invalid argument can be a solution.

For the sake of the correctness: there is no malloc() in musl [1] (nor in the
mirror source you posted to #1084), that's only in uclibc-ng [2] and glibc [3].

The reason why musl failed was already described by Richie and Cyril in #1084:
musl ignores the size parameter when buffer is NULL and allocates it with PATH_MAX
and passes this size to kernel.

Therefore I reword the commit message.

[1] https://git.musl-libc.org/cgit/musl/tree/src/unistd/getcwd.c
[2] https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/tree/libc/sysdeps/linux/common/getcwd.c#n38
[3] https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/getcwd.c;h=5b0b6879ed28f278f07ce494f9be30f504757daa;hb=HEAD#l47

...
> -	tst_res(TPASS | TTERRNO, "getcwd() failed as expected");
> +	TST_EXP_FAIL2(syscall(__NR_getcwd, tc->buf, tc->size), tc->exp_err);
While syscalls() would work everywhere, it's better is LTP wrapper tst_syscall()
(it TCONF in case when syscall is not implemented which is I admit nearly
impossible).

I used that and merged.
Thank you!

NOTE: we should implement .test_variants, where you just skip affected NULL
buffer test (it's enough to test it with raw syscall). Please send a patch or
let me know that you don't plan to do it and I'll do it.

Kind regards,
Petr

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp