Dipankar Sarma a écrit :
> The way we do file struct accounting is not very suitable for batched
> freeing. For scalability reasons, file accounting was constructor/destructor
> based. This meant that nr_files was decremented only when
> the object was removed from the slab cache. This is
> susceptible to slab fragmentation. With RCU based file structure,
> consequent batched freeing and a test program like Serge's,
> we just speed this up and end up with a very fragmented slab -
> 
> llm22:~ # cat /proc/sys/fs/file-nr
> 587730  0       758844
> 
> At the same time, I see only a 2000+ objects in filp cache.
> The following patch I fixes this problem. 
> 
> This patch changes the file counting by removing the filp_count_lock.
> Instead we use a separate atomic_t, nr_files, for now and all
> accesses to it are through get_nr_files() api. In the sysctl
> handler for nr_files, we populate files_stat.nr_files before returning
> to user.
> 
> Counting files as an when they are created and destroyed (as opposed
> to inside slab) allows us to correctly count open files with RCU.
> 
> Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com>
> ---

Well...

I am using a patch that seems sligthly better : It removes the filp_count_lock 
as yours but introduces a percpu variable, and a lazy nr_files . (Its value 
can be off with a delta of +/- 16*num_possible_cpus()

Pros :
  - No more delay caused by the slab dtor hack (like your patch)
  - No more ping pongs in SMP/NUMA machines because of the nr_files being 
constantly modified.
  - litle memory footprint of NR_CPUS*4 bytes
  - No more cli/sti games (because of filp ctor/dtor spinlock_bh)
  - Moves files_stat definition in a new file (include/linux/files_stat.h)

Cons : The lazy nr_files value in SMP, but who cares ? Do we want the exact 
value displayed in /proc/sys/fs/file-nr ?

Thank you

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>