Re: [PATCH v4 3/4] cachestat: implement cachestat syscall

[Brian Foster <bfoster@redhat.com>]

On Thu, Dec 22, 2022 at 01:50:19PM -0800, Nhat Pham wrote:
> On Tue, Dec 20, 2022 at 6:37 AM Brian Foster <bfoster@redhat.com> wrote:
> >
> > On Fri, Dec 16, 2022 at 11:21:48AM -0800, Nhat Pham wrote:
> > > Implement a new syscall that queries cache state of a file and
> > > summarizes the number of cached pages, number of dirty pages, number of
> > > pages marked for writeback, number of (recently) evicted pages, etc. in
> > > a given range.
> > >
> > > NAME
> > >     cachestat - query the page cache status of a file.
> > >
> > > SYNOPSIS
> > >     #include <sys/mman.h>
> > >
> > >     struct cachestat {
> > >         __u64 nr_cache;
> > >         __u64 nr_dirty;
> > >         __u64 nr_writeback;
> > >         __u64 nr_evicted;
> > >         __u64 nr_recently_evicted;
> > >     };
> > >
> > >     int cachestat(unsigned int fd, off_t off, size_t len,
> > >           size_t cstat_size, struct cachestat *cstat,
> > >           unsigned int flags);
> > >
> > > DESCRIPTION
> > >     cachestat() queries the number of cached pages, number of dirty
> > >     pages, number of pages marked for writeback, number of (recently)
> > >     evicted pages, in the bytes range given by `off` and `len`.
> > >
> > >     These values are returned in a cachestat struct, whose address is
> > >     given by the `cstat` argument.
> > >
> > >     The `off` and `len` arguments must be non-negative integers. If
> > >     `len` > 0, the queried range is [`off`, `off` + `len`]. If `len` ==
> > >     0, we will query in the range from `off` to the end of the file.
> > >
> > >     `cstat_size` allows users to obtain partial results. The syscall
> > >     will copy the first `csstat_size` bytes to the specified userspace
> > >     memory. `cstat_size` must be a non-negative value that is no larger
> > >     than the current size of the cachestat struct.
> > >
> > >     The `flags` argument is unused for now, but is included for future
> > >     extensibility. User should pass 0 (i.e no flag specified).
> > >
> > > RETURN VALUE
> > >     On success, cachestat returns 0. On error, -1 is returned, and errno
> > >     is set to indicate the error.
> > >
> > > ERRORS
> > >     EFAULT cstat points to an invalid address.
> > >
> > >     EINVAL invalid `cstat_size` or `flags`
> > >
> > >     EBADF  invalid file descriptor.
> > >
> > > Signed-off-by: Nhat Pham <nphamcs@gmail.com>
> > > ---
> >
> > Hi Nhat,
> >
> > Thanks for the tweaks. FWIW, this by and large looks reasonable to me.
> 
> Thanks for suggesting the tweaks! I tried refactoring it out and it
> did look cleaner to me (and one fewer file that I have to keep
> track of...)
> 
> > Just a couple more random nitty things, if you happen to care about
> > them..
> >
> > >  arch/alpha/kernel/syscalls/syscall.tbl      |   1 +
> > >  arch/arm/tools/syscall.tbl                  |   1 +
> > >  arch/ia64/kernel/syscalls/syscall.tbl       |   1 +
> > >  arch/m68k/kernel/syscalls/syscall.tbl       |   1 +
> > >  arch/microblaze/kernel/syscalls/syscall.tbl |   1 +
> > >  arch/parisc/kernel/syscalls/syscall.tbl     |   1 +
> > >  arch/powerpc/kernel/syscalls/syscall.tbl    |   1 +
> > >  arch/s390/kernel/syscalls/syscall.tbl       |   1 +
> > >  arch/sh/kernel/syscalls/syscall.tbl         |   1 +
> > >  arch/sparc/kernel/syscalls/syscall.tbl      |   1 +
> > >  arch/x86/entry/syscalls/syscall_32.tbl      |   1 +
> > >  arch/x86/entry/syscalls/syscall_64.tbl      |   1 +
> > >  arch/xtensa/kernel/syscalls/syscall.tbl     |   1 +
> > >  include/linux/fs.h                          |   3 +
> > >  include/linux/syscalls.h                    |   3 +
> > >  include/uapi/asm-generic/unistd.h           |   5 +-
> > >  include/uapi/linux/mman.h                   |   9 ++
> > >  init/Kconfig                                |  10 ++
> > >  kernel/sys_ni.c                             |   1 +
> > >  mm/filemap.c                                | 137 ++++++++++++++++++++
> > >  20 files changed, 180 insertions(+), 1 deletion(-)
> > >
> > ...
> > > diff --git a/mm/filemap.c b/mm/filemap.c
> > > index 08341616ae7a..29ffb906caa4 100644
> > > --- a/mm/filemap.c
> > > +++ b/mm/filemap.c
> > ...
> > > @@ -3949,3 +3953,136 @@ bool filemap_release_folio(struct folio *folio, gfp_t gfp)
> > >       return try_to_free_buffers(folio);
> > >  }
> > >  EXPORT_SYMBOL(filemap_release_folio);
> > > +
> > > +#ifdef CONFIG_CACHESTAT
> > > +/**
> > > + * filemap_cachestat() - compute the page cache statistics of a mapping
> > > + * @mapping: The mapping to compute the statistics for.
> > > + * @first_index:     The starting page cache index.
> > > + * @last_index:      The final page index (inclusive).
> > > + * @cs:      the cachestat struct to write the result to.
> > > + *
> > > + * This will query the page cache statistics of a mapping in the
> > > + * page range of [first_index, last_index] (inclusive). THe statistics
> > > + * queried include: number of dirty pages, number of pages marked for
> > > + * writeback, and the number of (recently) evicted pages.
> > > + */
> > > +void filemap_cachestat(struct address_space *mapping, pgoff_t first_index,
> > > +             pgoff_t last_index, struct cachestat *cs)
> > > +{
> >
> > Not sure why the internal helper needs to be wrapped in a config option?
> > Perhaps it makes more sense to define CONFIG_CACHESTAT_SYSCALL and just
> > wrap the syscall bits with that..?
> 
> No particularly strong reasons - I was just bundling the two together
> because I was not entirely sure if it's interesting or has any use
> case outside of the syscall itself.
> 
> Do you have something in mind?
> 

Not immediately, though it looks like a straightforward and potentially
useful helper. It wouldn't surprise me if it grew more in-kernel users
eventually. :)

That said, I only suggested this for cleanup/consistency reasons. I.e.,
it seems there is plenty of precedent for CONFIG_*_SYSCALL config
options, and the one or two I happened to peek at looked like they only
wrapped the syscall bits as opposed to the entire implementation (which
also seems like a clean and elegant approach to me). Somebody could
easily come along and make that change later if/when they might want to
use the helper (though it might be annoying to change the name of the
config option), so I have no strong opinion on either suggestion.

Brian

> >
> > I would also think it might make things simpler to split out all the
> > syscall bits into a separate patch from filemap_cachestat().
> 
> Same as above.
> 
> >
> > > +     XA_STATE(xas, &mapping->i_pages, first_index);
> > > +     struct folio *folio;
> > > +
> > > +     rcu_read_lock();
> > > +     xas_for_each(&xas, folio, last_index) {
> > > +             unsigned long nr_pages;
> > > +             pgoff_t folio_first_index, folio_last_index;
> > > +
> > > +             if (xas_retry(&xas, folio))
> > > +                     continue;
> > > +
> > > +             nr_pages = folio_nr_pages(folio);
> > > +             folio_first_index = folio_pgoff(folio);
> > > +             folio_last_index = folio_first_index + nr_pages - 1;
> > > +
> > > +             /* Folios might straddle the range boundaries, only count covered subpages */
> > > +             if (folio_first_index < first_index)
> > > +                     nr_pages -= first_index - folio_first_index;
> > > +
> > > +             if (folio_last_index > last_index)
> > > +                     nr_pages -= folio_last_index - last_index;
> > > +
> > > +             if (xa_is_value(folio)) {
> > > +                     /* page is evicted */
> > > +                     void *shadow = (void *)folio;
> > > +                     bool workingset; /* not used */
> > > +
> > > +                     cs->nr_evicted += nr_pages;
> > > +
> > > +#ifdef CONFIG_SWAP /* implies CONFIG_MMU */
> > > +                     if (shmem_mapping(mapping)) {
> > > +                             /* shmem file - in swap cache */
> > > +                             swp_entry_t swp = radix_to_swp_entry(folio);
> > > +
> > > +                             shadow = get_shadow_from_swap_cache(swp);
> > > +                     }
> > > +#endif
> > > +                     if (workingset_test_recent(shadow, true, &workingset))
> > > +                             cs->nr_recently_evicted += nr_pages;
> > > +
> > > +                     continue;
> > > +             }
> > > +
> > > +             /* page is in cache */
> > > +             cs->nr_cache += nr_pages;
> > > +
> > > +             if (folio_test_dirty(folio))
> > > +                     cs->nr_dirty += nr_pages;
> > > +
> > > +             if (folio_test_writeback(folio))
> > > +                     cs->nr_writeback += nr_pages;
> >
> > I'm not sure this is an issue right now (or if it will ever be for your
> > use cases), but from a filesystem perspective it is possible to have
> > large or variable sized folios in cache. At the moment I think the fs
> > (or XFS+iomap at least) manages dirty/writeback state at folio
> > granularity, but that may change in the near future if/when iomap
> > sub-page dirty tracking comes along. I suspect that means it may become
> > possible to have a large folio of some N number of pages where only a
> > subset of those pages are actually in dirty/writeback state, and thus
> > introduces some inaccuracy here because this assumes that folio state
> > applies to folio_nr_pages() worth of pages. Just something to be aware
> > of..
> >
> > Brian
> 
> Oof, I coded this with the mental framework of folio-as-a-unit, and
> assumed that the dirty/writeback state is managed at the granularity of folio.
> Thanks for bringing this up Brian! I'll have to watch out for this as iomap
> evolves (and the subpage tracking becomes a thing).
> 
> >
> > > +     }
> > > +     rcu_read_unlock();
> > > +}
> > > +EXPORT_SYMBOL(filemap_cachestat);
> > > +
> > > +/*
> > > + * The cachestat(5) system call.
> > > + *
> > > + * cachestat() returns the page cache statistics of a file in the
> > > + * bytes specified by `off` and `len`: number of cached pages,
> > > + * number of dirty pages, number of pages marked for writeback,
> > > + * number of (recently) evicted pages.
> > > + *
> > > + * `off` and `len` must be non-negative integers. If `len` > 0,
> > > + * the queried range is [`off`, `off` + `len`]. If `len` == 0,
> > > + * we will query in the range from `off` to the end of the file.
> > > + *
> > > + * `cstat_size` allows users to obtain partial results. The syscall
> > > + * will copy the first `csstat_size` bytes to the specified userspace
> > > + * memory. It also makes the cachestat struct extensible - new fields
> > > + * can be added in the future without breaking existing usage.
> > > + * `cstat_size` must be a non-negative value that is no larger than
> > > + * the current size of the cachestat struct.
> > > + *
> > > + * The `flags` argument is unused for now, but is included for future
> > > + * extensibility. User should pass 0 (i.e no flag specified).
> > > + *
> > > + * Because the status of a page can change after cachestat() checks it
> > > + * but before it returns to the application, the returned values may
> > > + * contain stale information.
> > > + *
> > > + * return values:
> > > + *  zero    - success
> > > + *  -EFAULT - cstat points to an illegal address
> > > + *  -EINVAL - invalid arguments
> > > + *  -EBADF   - invalid file descriptor
> > > + */
> > > +SYSCALL_DEFINE6(cachestat, unsigned int, fd, off_t, off, size_t, len,
> > > +             size_t, cstat_size, struct cachestat __user *, cstat,
> > > +             unsigned int, flags)
> > > +{
> > > +     struct fd f = fdget(fd);
> > > +     struct address_space *mapping;
> > > +     struct cachestat cs;
> > > +     pgoff_t first_index = off >> PAGE_SHIFT;
> > > +     pgoff_t last_index =
> > > +             len == 0 ? ULONG_MAX : (off + len - 1) >> PAGE_SHIFT;
> > > +
> > > +     if (off < 0 || cstat_size > sizeof(struct cachestat) || flags != 0)
> > > +             return -EINVAL;
> > > +
> > > +     if (!f.file)
> > > +             return -EBADF;
> > > +
> > > +     memset(&cs, 0, sizeof(struct cachestat));
> > > +     mapping = f.file->f_mapping;
> > > +     filemap_cachestat(mapping, first_index, last_index, &cs);
> > > +     fdput(f);
> > > +
> > > +     if (copy_to_user(cstat, &cs, cstat_size))
> > > +             return -EFAULT;
> > > +
> > > +     return 0;
> > > +}
> > > +#endif /* CONFIG_CACHESTAT */
> > > --
> > > 2.30.2
> > >
> >
>