linux.git/fs/file.c, branch v4.9

fs/file: more unsigned file descriptors

2016-09-27T22:47:38+00:00

Propagate unsignedness for grand total of 149 bytes:

	$ ./scripts/bloat-o-meter ../vmlinux-000 ../obj/vmlinux
	add/remove: 0/0 grow/shrink: 0/10 up/down: 0/-149 (-149)
	function                                     old     new   delta
	set_close_on_exec                             99      98      -1
	put_files_struct                             201     200      -1
	get_close_on_exec                             59      58      -1
	do_prlimit                                   498     497      -1
	do_execveat_common.isra                     1662    1661      -1
	__close_fd                                   178     173      -5
	do_dup2                                      219     204     -15
	seq_show                                     685     660     -25
	__alloc_fd                                   384     357     -27
	dup_fd                                       718     646     -72

It mostly comes from converting "unsigned int" to "long" for bit operations.

Signed-off-by: Alexey Dobriyan 
Signed-off-by: Al Viro

give readdir(2)/getdents(2)/etc. uniform exclusion with lseek()

2016-05-02T23:49:28+00:00

same as read() on regular files has, and for the same reason.

Signed-off-by: Al Viro

kmemcg: account certain kmem allocations to memcg

2016-01-15T00:00:49+00:00

Mark those kmem allocations that are known to be easily triggered from
userspace as __GFP_ACCOUNT/SLAB_ACCOUNT, which makes them accounted to
memcg.  For the list, see below:

 - threadinfo
 - task_struct
 - task_delay_info
 - pid
 - cred
 - mm_struct
 - vm_area_struct and vm_region (nommu)
 - anon_vma and anon_vma_chain
 - signal_struct
 - sighand_struct
 - fs_struct
 - files_struct
 - fdtable and fdtable->full_fds_bits
 - dentry and external_name
 - inode for all filesystems. This is the most tedious part, because
   most filesystems overwrite the alloc_inode method.

The list is far from complete, so feel free to add more objects.
Nevertheless, it should be close to "account everything" approach and
keep most workloads within bounds.  Malevolent users will be able to
breach the limit, but this was possible even with the former "account
everything" approach (simply because it did not account everything in
fact).

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Vladimir Davydov 
Acked-by: Johannes Weiner 
Acked-by: Michal Hocko 
Cc: Tejun Heo 
Cc: Greg Thelen 
Cc: Christoph Lameter 
Cc: Pekka Enberg 
Cc: David Rientjes 
Cc: Joonsoo Kim 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

fs/file.c: __const_max is actually __const_min :-)

2015-12-07T02:17:11+00:00

7f4b36f9bb930 "get rid of files_defer_init()" inexplicably changed a
min() to a __const_max() - but the __const_max macro actually gives
the minimum... So no functional change, just less confusing naming.

Signed-off-by: Rasmus Villemoes 
Signed-off-by: Al Viro

vfs: clear remainder of 'full_fds_bits' in dup_fd()

2015-11-06T07:05:32+00:00

This fixes a bug from commit f3f86e33dc3d ("vfs: Fix pathological
performance case for __alloc_fd()").

v2: refactor to share fd bitmap copying code
Signed-off-by: Eric Biggers 
Signed-off-by: Linus Torvalds

vfs: conditionally clear close-on-exec flag

2015-10-31T23:14:51+00:00

We clear the close-on-exec flag when opening and closing files, and the
bit was almost always already clear before.  Avoid dirtying the
cacheline if the clearning isn't necessary.  That avoids unnecessary
cacheline dirtying and bouncing in multi-socket environments.

Eric Dumazet has a file descriptor benchmark that goes 4% faster from
this on his two-socket machine.  It's probably partly superlinear
improvement due to getting slightly less spinlock contention on the
file_lock spinlock due to less work in the critical section.

Tested-by: Eric Dumazet 
Signed-off-by: Linus Torvalds

vfs: Fix pathological performance case for __alloc_fd()

2015-10-31T23:12:10+00:00

Al Viro points out that:
> >     * [Linux-specific aside] our __alloc_fd() can degrade quite badly
> > with some use patterns.  The cacheline pingpong in the bitmap is probably
> > inevitable, unless we accept considerably heavier memory footprint,
> > but we also have a case when alloc_fd() takes O(n) and it's _not_ hard
> > to trigger - close(3);open(...); will have the next open() after that
> > scanning the entire in-use bitmap.

And Eric Dumazet has a somewhat realistic multithreaded microbenchmark
that opens and closes a lot of sockets with minimal work per socket.

This patch largely fixes it.  We keep a 2nd-level bitmap of the open
file bitmaps, showing which words are already full.  So then we can
traverse that second-level bitmap to efficiently skip already allocated
file descriptors.

On his benchmark, this improves performance by up to an order of
magnitude, by avoiding the excessive open file bitmap scanning.

Tested-and-acked-by: Eric Dumazet 
Cc: Al Viro 
Signed-off-by: Linus Torvalds

fs/file.c: __fget() and dup2() atomicity rules

2015-07-01T06:31:08+00:00

__fget() does lockless fetch of pointer from the descriptor
table, attempts to grab a reference and treats "it was already
zero" as "it's already gone from the table, we just hadn't
seen the store, let's fail".  Unfortunately, that breaks the
atomicity of dup2() - __fget() might see the old pointer,
notice that it's been already dropped and treat that as
"it's closed".  What we should be getting is either the
old file or new one, depending whether we come before or after
dup2().

Dmitry had following test failing sometimes :

int fd;
void *Thread(void *x) {
  char buf;
  int n = read(fd, &buf, 1);
  if (n != 1)
    exit(printf("read failed: n=%d errno=%d\n", n, errno));
  return 0;
}

int main()
{
  fd = open("/dev/urandom", O_RDONLY);
  int fd2 = open("/dev/urandom", O_RDONLY);
  if (fd == -1 || fd2 == -1)
    exit(printf("open failed\n"));
  pthread_t th;
  pthread_create(&th, 0, Thread, 0);
  if (dup2(fd2, fd) == -1)
    exit(printf("dup2 failed\n"));
  pthread_join(th, 0);
  if (close(fd) == -1)
    exit(printf("close failed\n"));
  if (close(fd2) == -1)
    exit(printf("close failed\n"));
  printf("DONE\n");
  return 0;
}

Signed-off-by: Eric Dumazet 
Reported-by: Dmitry Vyukov 
Signed-off-by: Al Viro

fs/file.c: don't acquire files->file_lock in fd_install()

2015-07-01T06:30:09+00:00

Mateusz Guzik reported :

 Currently obtaining a new file descriptor results in locking fdtable
 twice - once in order to reserve a slot and second time to fill it.

Holding the spinlock in __fd_install() is needed in case a resize is
done, or to prevent a resize.

Mateusz provided an RFC patch and a micro benchmark :
  http://people.redhat.com/~mguzik/pipebench.c

A resize is an unlikely operation in a process lifetime,
as table size is at least doubled at every resize.

We can use RCU instead of the spinlock.

__fd_install() must wait if a resize is in progress.

The resize must block new __fd_install() callers from starting,
and wait that ongoing install are finished (synchronize_sched())

resize should be attempted by a single thread to not waste resources.

rcu_sched variant is used, as __fd_install() and expand_fdtable() run
from process context.

It gives us a ~30% speedup using pipebench on a dual Intel(R) Xeon(R)
CPU E5-2696 v2 @ 2.50GHz

Signed-off-by: Eric Dumazet 
Reported-by: Mateusz Guzik 
Acked-by: Mateusz Guzik 
Tested-by: Mateusz Guzik 
Signed-off-by: Al Viro

mm: rcu-protected get_mm_exe_file()

2015-04-17T13:04:07+00:00

This patch removes mm->mmap_sem from mm->exe_file read side.
Also it kills dup_mm_exe_file() and moves exe_file duplication into
dup_mmap() where both mmap_sems are locked.

[akpm@linux-foundation.org: fix comment typo]
Signed-off-by: Konstantin Khlebnikov 
Cc: Davidlohr Bueso 
Cc: Al Viro 
Cc: Oleg Nesterov 
Cc: "Paul E. McKenney" 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds