linux-kernel - Re: [PATCH 3/4] cachestat: implement cachestat syscall

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Mon, 21 Nov 2022 10:55:46 -0500
From:   Johannes Weiner <hannes@...xchg.org>
To:     Brian Foster <bfoster@...hat.com>
Cc:     Nhat Pham <nphamcs@...il.com>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 3/4] cachestat: implement cachestat syscall

On Mon, Nov 21, 2022 at 09:45:49AM -0500, Brian Foster wrote:
> On Tue, Nov 15, 2022 at 10:29:00AM -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 {
> >         unsigned long nr_cache;
> >         unsigned long nr_dirty;
> >         unsigned long nr_writeback;
> >         unsigned long nr_evicted;
> >         unsigned long nr_recently_evicted;
> >     };
> > 
> >     int cachestat(unsigned int fd, off_t off, size_t len,
> >         struct cachestat *cstat);
> > 
> 
> Do you have a strong use case for a user specified range vs. just
> checking the entire file? If not, have you considered whether it might
> be worth expanding statx() to include this data? That call is already
> designed to include "extended" file status and avoids the need for a new
> syscall. For example, the fields could be added individually with
> multiple flags, or the entire struct tied to a new STATX_CACHE flag or
> some such.

Whole-file stats are only useful for data that is structured in
directory trees. It doesn't work for structured files. For example,
understanding (and subsequently advising/influencing) the readahead
and dirty flushing in certain sections of a larger database file.

Fadvise/madvise/sync_file_range etc. give the user the ability to
influence cache behavior in sub-ranges, so it makes sense to also
allow querying at that granularity.

> > 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` argument must be a non-negative integers, If `off` + `len`
> >     >= `off`, the queried range is [`off`, `off` + `len`]. Otherwise, we
> >     will query in the range from `off` to the end of the file.
> > 
> 
> (off + len < off) is an error condition on some (most?) other syscalls.
> At least some calls (i.e. fadvise(), sync_file_range()) use len == 0 to
> explicitly specify "to EOF."

Good point, it would make sense to stick to that precedent.