[LTP] [PATCH] tst_find_backing_dev: fix logic in matching mount point

[Richard Palethorpe]

Hello,

Jan Stancek <jstancek@redhat.com> writes:

> On Mon, Jul 11, 2022 at 2:13 PM Richard Palethorpe <rpalethorpe@suse.de> wrote:
>>
>> Hello,
>>
>> Jan Stancek <jstancek@redhat.com> writes:
>>
>> > If backing dev is btrfs root device, then starting best_match_len
>> > from 1 creates an issue, because root (/) is never matched.
>> > Also we should check that entire mount point string is present in
>> > path we are matching against.
>> >
>> > In case there's error also dump /proc/self/mountinfo before tst_brk.
>> >
>> > This fixes test with following partition layout (TMPDIR is on /):
>> >   # cat /proc/self/mountinfo  | grep btrfs
>> >   59 1 0:29 /root / rw,relatime shared:1 - btrfs /dev/dasda2 rw,seclabel,compress=zstd:1,ssd,space_cache=v2,subvolid=257,subvol=/root
>> >   93 59 0:29 /home /home rw,relatime shared:47 - btrfs /dev/dasda2 rw,seclabel,compress=zstd:1,ssd,space_cache=v2,subvolid=256,subvol=/home
>> >
>> > Signed-off-by: Jan Stancek <jstancek@redhat.com>
>> > ---
>> >  lib/tst_device.c | 17 ++++++++++++-----
>> >  1 file changed, 12 insertions(+), 5 deletions(-)
>> >
>> > diff --git a/lib/tst_device.c b/lib/tst_device.c
>> > index c34cbe6d1f56..414bf0eea816 100644
>> > --- a/lib/tst_device.c
>> > +++ b/lib/tst_device.c
>> > @@ -506,12 +506,17 @@ unsigned long tst_dev_bytes_written(const char *dev)
>> >       return dev_bytes_written;
>> >  }
>> >
>> > -static int count_match_len(const char *first, const char *second)
>> > +static int str_starts_with(const char *str, const char *prefix)
>> >  {
>> >       int len = 0;
>> >
>> > -     while (*first && *first++ == *second++)
>> > +     while (*prefix) {
>> > +             if (!*str)
>> > +                     return 0;
>> > +             if (*str++ != *prefix++)
>> > +                     return 0;
>> >               len++;
>> > +     }
>>
>> I'm not sure this is better than the original. It's a seperate cleanup
>> in any case.
>
> The difference is that partial matches now returns 0. Previously it did not,
> it only counted number of matching characters. So for example ("/foobar" "/foo")
> appeared as better match than ("/foobar", "/")

Ah, sorry.

>
>>
>> >
>> >       return len;
>> >  }
>> > @@ -524,7 +529,7 @@ void tst_find_backing_dev(const char *path, char *dev)
>> >       char *pre = NULL;
>> >       char *next = NULL;
>> >       unsigned int dev_major, dev_minor, line_mjr, line_mnr;
>> > -     unsigned int len, best_match_len = 1;
>> > +     unsigned int len, best_match_len = 0;
>> >       char mnt_point[PATH_MAX];
>> >
>> >       if (stat(path, &buf) < 0)
>> > @@ -550,7 +555,7 @@ void tst_find_backing_dev(const char *path, char *dev)
>> >                       break;
>> >               }
>> >
>> > -             len = count_match_len(path, mnt_point);
>> > +             len = str_starts_with(path, mnt_point);
>> >               if (len > best_match_len) {
>> >                       strcpy(dev, pre);
>> >                       best_match_len = len;
>> > @@ -559,8 +564,10 @@ void tst_find_backing_dev(const char *path, char *dev)
>> >
>> >       SAFE_FCLOSE(NULL, file);
>> >
>> > -     if (!*dev)
>> > +     if (!*dev) {
>> > +             tst_system("cat /proc/self/mountinfo");
>> >               tst_brkm(TBROK, NULL, "Cannot find block device for %s", path);
>> > +     }
>> >
>> >       if (stat(dev, &buf) < 0)
>> >               tst_brkm(TWARN | TERRNO, NULL, "stat(%s) failed", dev);
>> > --
>> > 2.27.0
>>
>> Makes sense. However I suspect this function can be replaced with the
>> standard library method used in tst_stat_mount_dev. I didn't try this
>> before because I'm not sure why it scans mountinfo instead of mounts.
>
> /proc/self/mounts doesn't contain major/minor numbers, which is primarily what
> tst_find_backing_dev is using.

Does anyone know a situation where checking the mount path doesn't work?
It seems like we are just checking the device numbers first for
historical reasons and not because it is necessary.

I haven't seen any failure reports for io_control01 which only checks
the path with tst_stat_mount_dev. However this test only runs if you
have CG v2 enabled by default.

>
>>
>>
>> --
>> Thank you,
>> Richard.
>>


-- 
Thank you,
Richard.