Re: [PATCH 2/2] grub2: use stat instead of udevadm for partition lookup

[Daniel Kiper]

On Tue, Jul 13, 2021 at 09:10:39PM +0200, Petr Vorel wrote:
> Hi all,
>
> > On 7/13/21 1:48 PM, Daniel Kiper wrote:
> > > On Thu, Jul 08, 2021 at 05:55:58PM +0200, Petr Vorel wrote:
> > >> From: Jeff Mahoney <jeffm@suse.com>
>
> > >> sysfs_partition_path calls udevadm to resolve the sysfs path for
> > >> a block device. That can be accomplished by stating the device node
> > >> and using the major/minor to follow the symlinks in /sys/dev/block/.
>
> > >> This cuts the execution time of grub2-mkconfig from 10s to 2s on
> > >> my system.
>
> > >> Signed-off-by: Jeff Mahoney <jeffm@suse.com>
> > >> [ pvorel: include grub/osdep/major.h ]
> > >> Signed-off-by: Petr Vorel <pvorel@suse.cz>
> > >> ---
> > >>  grub-core/osdep/linux/hostdisk.c | 8 ++++++++
> > >>  1 file changed, 8 insertions(+)
>
> > >> diff --git a/grub-core/osdep/linux/hostdisk.c 
> > >> b/grub-core/osdep/linux/hostdisk.c
> > >> index da62f924e..43dc4b0ba 100644
> > >> --- a/grub-core/osdep/linux/hostdisk.c
> > >> +++ b/grub-core/osdep/linux/hostdisk.c
> > >> @@ -31,6 +31,7 @@
> > >>  #include <grub/misc.h>
> > >>  #include <grub/i18n.h>
> > >>  #include <grub/list.h>
> > >> +#include <grub/osdep/major.h>
>
> > >>  #include <stdio.h>
> > >>  #include <stdlib.h>
> > >> @@ -105,6 +106,13 @@ sysfs_partition_path (const char *dev, const char 
> > >> *entry)
> > >>    char *buf = NULL;
> > >>    size_t len = 0;
> > >>    char *path = NULL;
> > >> +  struct stat st;
> > >> +  int ret;
> > >> +
> > >> +  ret = stat(dev, &st);
>
> > > Missing space between "stat" and "(".
>
> > >> +  if (ret == 0 && S_ISBLK(st.st_mode))
>
> > > Same for S_ISBLK...
>
> > >> +    return xasprintf ("/sys/dev/block/%u:%u/%s",
> > >> +                      major (st.st_rdev), minor (st.st_rdev), entry);
>
> > >>    argv[0] = "udevadm";
> > >>    argv[1] = "info";
>
> > > Do we really need udevadm fallback mechanism? If something went wrong
> > > here for us I do not expect it will work for udevadm either.
>
> > I suspect not.  'udevadm info --query path --name <dev>' does
> > essentially the same thing but with many more steps and the cost
> > starting an external program for every block device.  I left it in with
> > the assumption that it was probably there for a reason and perhaps
> > udevadm on early udev systems did something more special than just doing
> > stat + sysfs path building.  I didn't spend any time on validating that
> > assumption.
>
> As it'd be wort to remove udevadm fallback, I'll stop being lazy and setup VM 
> to
> test.

Please do. I prefer to remove this fallback because it will be mostly
dead code after your fix.

Daniel