Re: [PATCH 4/5] mm: shmem: Convert shmem_enabled_show to use sysfs_emit_at

[Matthew Wilcox]

On Mon, Nov 02, 2020 at 03:32:59PM +0100, Greg Kroah-Hartman wrote:
> On Mon, Nov 02, 2020 at 02:08:36PM +0000, Matthew Wilcox wrote:
> > On Mon, Nov 02, 2020 at 02:33:43PM +0100, Greg Kroah-Hartman wrote:
> > > > Oh, ugh, sysfs_emit() should be able to work on a buffer that isn't
> > > > page aligned.  Greg, how about this?
> > > 
> > > How can sysfs_emit() be called on a non-page-aligned buffer?  It's being
> > > used on the buffer that was passed to the sysfs call.
> > > 
> > > And if you are writing multiple values to a single sysfs file output,
> > > well, not good...
> > 
> > See shmem_enabled_show() in mm/shmem.c (output at
> > /sys/kernel/mm/transparent_hugepage/shmem_enabled on your machine).
> > 
> > I don't claim it's a good interface, but it exists.
> 
> Ok, that's a common pattern for sysfs files, not that bad.
> 
> What's wrong with using sysfs_emit_at()?  We want sysfs_emit() to "know"
> that the buffer is PAGE_SIZE big, if you try to allow offsets in it,
> that defeats the purpose of the check.

For someone who's used to C "strings", it's pretty common to do
something like:

	buf += sprintf(buf, "foo ");
	buf += sprintf(buf, "bar ");

sysfs_emit_at instead wants me to do:

	len += sprintf(buf + len, "foo ");
	len += sprintf(buf + len, "bar ");

I don't see how the code I wrote defeats the check.  It checks that the
buffer never crosses a PAGE_SIZE boundary, which is equivalently safe.