[LTP] [PATCH 2/2] Use SAFE_RUNCMD()

[Li Wang]

On Mon, Mar 23, 2020 at 9:42 PM Li Wang <liwang@redhat.com> wrote:

> Hi Petr, Xu,
>
> On Mon, Mar 23, 2020 at 7:37 PM Petr Vorel <pvorel@suse.cz> wrote:
>
>> Hi Li, Xu,
>>
>> > >       testcases/kernel/syscalls/add_key/add_key05.c   | 15
>> ++-------------
>> > >       testcases/kernel/syscalls/quotactl/quotactl01.c | 14
>> ++------------
>> > >       testcases/kernel/syscalls/quotactl/quotactl06.c | 12
>> +-----------
>>
>> > > Apart from the three, do you consider converting to SAFE_RUNCMD for
>> the
>> > > rest testcases?
>> > > (it seems not too much work remaining since only a few test case uses
>> > > tst_run_cmd)
>> > At the beginning, I have the same idea. But after seeing code, I think
>> we
>> > should not because these cases have many sub tests(only few test
>> deponds on
>> > the result of the cmd execution.
>>
>> > > kernel/syscalls/setpriority/setpriority01.c
>> > One year ago has a commit db82b596(setpriority01: Skip only PRIO_USER
>> when
>> > unable to add test user). It only affects PRIO_USER sub test.
>> + 1. I didn't want to break the case when useradd is not available
>> (android or
>> some custom embedded linux) or there is no password file (root mounted as
>> ro -
>> custom embedded linux).
>>
>
> That's right. Thanks for the clarification.
>
>>
>> BTW I also avoid handling adding user as I want to implement better
>> handling
>> user and group in LTP (adding a flag), see:
>> https://github.com/linux-test-project/ltp/issues/468
>
>
> Good plan.
>
>
>>
>>
>> Feel free to commend this plan.
>> This patchset is kind of preparation for it.
>>
>> > > kernel/syscalls/copy_file_range/copy_file_range02.c
>> > only affect test6 and test7
>> >  6) Try to copy contents to a file chattred with +i
>> >  *    flag -> EPERM
>> >  * 7) Try to copy contents to a swapfile ->ETXTBSY
>> Yes, it'd be bad to break all tests due it.
>>
>> Here is also problem with swapoff (or maybe chattr, mkswap, swapon; I
>> don't
>> remember), which returns exit code 255 on error, so it's not possible to
>> distinguish this from the case whether command is not available (any
>> idea, how
>> to fix it?).
>>
>
> Maybe we could achieve a tst_cmd_available(char *cmd) in the C version?
> which uses popen() to open a process like: "whereis/which command" and do
> string parse in the result to see the path(/usr/bin/cmd, /usr/sbin/cmd) of
> the bin if it has been found.
>

Or, simply to use access() if we gonna take care of embedded Linux, is this
reliable?

int tst_cmd_available(char *cmd)
{
    int ret = 0;
    char path[PATH_MAX];

    snprintf(path, PATH_MAX, "/usr/bin/%s", cmd);
    if (!access(path, X_OK)) {
        ret = 1;
        goto out;
    }

    snprintf(path, PATH_MAX, "/usr/sbin/%s", cmd);
    if (!access(path, X_OK)) {
        ret = 1;
        goto out;
    }

    snprintf(path, PATH_MAX, "/usr/local/bin/%s", cmd);
    if (!access(path, X_OK)) {
        ret = 1;
        goto out;
    }

    snprintf(path, PATH_MAX, "/usr/local/sbin/%s", cmd);
    if (!access(path, X_OK)) {
        ret = 1;
        goto out;
    }

out:
    return ret;
}
-- 
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200323/6d7d726d/attachment-0001.htm>