linux-toradex.git/ipc, branch v3.5-rc6

ipc: shm: restore MADV_REMOVE functionality on shared memory segments

2012-06-07T21:43:55+00:00

Commit 17cf28afea2a ("mm/fs: remove truncate_range") removed the
truncate_range inode operation in favour of the fallocate file
operation.

When using SYSV IPC shared memory segments, calling madvise with the
MADV_REMOVE advice on an area of shared memory will attempt to invoke
the .fallocate function for the shm_file_operations, which is NULL and
therefore returns -EOPNOTSUPP to userspace.  The previous behaviour
would inherit the inode_operations from the underlying tmpfs file and
invoke truncate_range there.

This patch restores the previous behaviour by wrapping the underlying
fallocate function in shm_fallocate, as we do for fsync.

[hughd@google.com: use -ENOTSUPP in shm_fallocate()]
Signed-off-by: Will Deacon 
Acked-by: Hugh Dickins 
Signed-off-by: Hugh Dickins 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs

2012-06-01T17:34:35+00:00

Pull vfs changes from Al Viro.
 "A lot of misc stuff.  The obvious groups:
   * Miklos' atomic_open series; kills the damn abuse of
     ->d_revalidate() by NFS, which was the major stumbling block for
     all work in that area.
   * ripping security_file_mmap() and dealing with deadlocks in the
     area; sanitizing the neighborhood of vm_mmap()/vm_munmap() in
     general.
   * ->encode_fh() switched to saner API; insane fake dentry in
     mm/cleancache.c gone.
   * assorted annotations in fs (endianness, __user)
   * parts of Artem's ->s_dirty work (jff2 and reiserfs parts)
   * ->update_time() work from Josef.
   * other bits and pieces all over the place.

  Normally it would've been in two or three pull requests, but
  signal.git stuff had eaten a lot of time during this cycle ;-/"

Fix up trivial conflicts in Documentation/filesystems/vfs.txt (the
'truncate_range' inode method was removed by the VM changes, the VFS
update adds an 'update_time()' method), and in fs/btrfs/ulist.[ch] (due
to sparse fix added twice, with other changes nearby).

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (95 commits)
  nfs: don't open in ->d_revalidate
  vfs: retry last component if opening stale dentry
  vfs: nameidata_to_filp(): don't throw away file on error
  vfs: nameidata_to_filp(): inline __dentry_open()
  vfs: do_dentry_open(): don't put filp
  vfs: split __dentry_open()
  vfs: do_last() common post lookup
  vfs: do_last(): add audit_inode before open
  vfs: do_last(): only return EISDIR for O_CREAT
  vfs: do_last(): check LOOKUP_DIRECTORY
  vfs: do_last(): make ENOENT exit RCU safe
  vfs: make follow_link check RCU safe
  vfs: do_last(): use inode variable
  vfs: do_last(): inline walk_component()
  vfs: do_last(): make exit RCU safe
  vfs: split do_lookup()
  Btrfs: move over to use ->update_time
  fs: introduce inode operation ->update_time
  reiserfs: get rid of resierfs_sync_super
  reiserfs: mark the superblock as dirty a bit later
  ...

switch aio and shm to do_mmap_pgoff(), make do_mmap() static

2012-06-01T14:37:17+00:00

after all, 0 bytes and 0 pages is the same thing...

Signed-off-by: Al Viro

take security_mmap_file() outside of ->mmap_sem

2012-06-01T14:37:01+00:00

Signed-off-by: Al Viro

ipc/mqueue: add rbtree node caching support

2012-06-01T00:49:31+00:00

When I wrote the first patch that added the rbtree support for message
queue insertion, it sped up the case where the queue was very full
drastically from the original code.  It, however, slowed down the case
where the queue was empty (not drastically though).

This patch caches the last freed rbtree node struct so we can quickly
reuse it when we get a new message.  This is the common path for any queue
that very frequently goes from 0 to 1 then back to 0 messages in queue.

Andrew Morton didn't like that we were doing a GFP_ATOMIC allocation in
msg_insert, so this patch attempts to speculatively allocate a new node
struct outside of the spin lock when we know we need it, but will still
fall back to a GFP_ATOMIC allocation if it has to.

Once I added the caching, the necessary various ret = ; spin_unlock
gyrations in mq_timedsend were getting pretty ugly, so this also slightly
refactors that function to streamline the flow of the code and the
function exit.

Finally, while working on getting performance back I made sure that all of
the node structs were always fully initialized when they were first used,
rendering the use of kzalloc unnecessary and a waste of CPU cycles.

The net result of all of this is:

1) We will avoid a GFP_ATOMIC allocation when possible, but fall back
   on it when necessary.

2) We will speculatively allocate a node struct using GFP_KERNEL if our
   cache is empty (and save the struct to our cache if it's still empty
   after we have obtained the spin lock).

3) The performance of the common queue empty case has significantly
   improved and is now much more in line with the older performance for
   this case.

The performance changes are:

            Old mqueue      new mqueue      new mqueue + caching
queue empty
send/recv   305/288ns       349/318ns       310/322ns

I don't think we'll ever be able to get the recv performance back, but
that's because the old recv performance was a direct result and
consequence of the old methods abysmal send performance.  The recv path
simply must do more so that the send path does not incur such a penalty
under higher queue depths.

As it turns out, the new caching code also sped up the various queue full
cases relative to my last patch.  That could be because of the difference
between the syscall path in 3.3.4-rc5 and 3.3.4-rc6, or because of the
change in code flow in the mq_timedsend routine.  Regardless, I'll take
it.  It wasn't huge, and I *would* say it was within the margin for error,
but after many repeated runs what I'm seeing is that the old numbers trend
slightly higher (about 10 to 20ns depending on which test is the one
running).

[akpm@linux-foundation.org: checkpatch fixes]
Signed-off-by: Doug Ledford 
Cc: Frederic Weisbecker 
Cc: Manfred Spraul 
Cc: Stephen Rothwell 
Cc: KOSAKI Motohiro 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

ipc/mqueue: strengthen checks on mqueue creation

2012-06-01T00:49:31+00:00

We already check the mq attr struct if it's passed in, but now that the
admin can set system wide defaults separate from maximums, it's actually
possible to set the defaults to something that would overflow.  So, if
there is no attr struct passed in to the open call, check the default
values.

While we are at it, simplify mq_attr_ok() by making it return 0 or an
error condition, so that way if we add more tests to it later, we have the
option of what error should be returned instead of the calling location
having to pick a possibly inaccurate error code.

[akpm@linux-foundation.org: s/ENOMEM/EOVERFLOW/]
Signed-off-by: Doug Ledford 
Cc: Stephen Rothwell 
Cc: Manfred Spraul 
Acked-by: KOSAKI Motohiro 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

ipc/mqueue: correct mq_attr_ok test

2012-06-01T00:49:31+00:00

While working on the other parts of the mqueue stuff, I noticed that the
calculation for overflow in mq_attr_ok didn't actually match reality (this
is especially true since my last patch which changed how we account memory
slightly).

In particular, we used to test for overflow using:
  msgs * msgsize + msgs * sizeof(struct msg_msg *)

That was never really correct because each message we allocate via
load_msg() is actually a struct msg_msg followed by the data for the
message (and if struct msg_msg + data exceeds PAGE_SIZE we end up
allocating struct msg_msgseg structs too, but accounting for them would
get really tedious, so let's ignore those...they're only a pointer in size
anyway).  This patch updates the calculation to be more accurate in
regards to maximum possible memory consumption by the mqueue.

[akpm@linux-foundation.org: add a local to simplify overflow-checking expression]
Signed-off-by: Doug Ledford 
Cc: Stephen Rothwell 
Cc: Manfred Spraul 
Acked-by: KOSAKI Motohiro 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

ipc/mqueue: improve performance of send/recv

2012-06-01T00:49:31+00:00

The existing implementation of the POSIX message queue send and recv
functions is, well, abysmal.  Even worse than abysmal.  I submitted a
patch to increase the maximum POSIX message queue limit to 65536 due to
customer needs, however, upon looking over the send/recv implementation, I
realized that my customer needs help with that too even if they don't know
it.  The basic problem is that, given the fairly typical use case scenario
for a large queue of queueing lots of messages all at the same priority (I
verified with my customer that this is indeed what their app does), the
msg_insert routine is basically a frikkin' bubble sort.  I mean, whoa,
that's *so* middle school.

OK, OK, to not slam the original author too much, I'm sure they didn't
envision a queue depth of 50,000+ messages.  No one would think that
moving elements in an array, one at a time, and dereferencing each pointer
in that array to check priority of the message being pointed too, again
one at a time, for 50,000+ times would be good.  So let's assume that, as
is typical, the users have found a way to break our code simply by using
it in a way we didn't envision.  Fair enough.

"So, just how broken is it?", you ask.  I wondered the same thing, so I
wrote an app to let me know.  It's my next patch.  It gave me some
interesting results.  Here's what it tested:

Interference with other apps - In continuous mode, the app just sits there
and hits a message queue forever, while you go do something productive on
another terminal using other CPUs.  You then measure how long it takes you
to do that something productive.  Then you restart the app in fake
continuous mode, and it sits in a tight loop on a CPU while you repeat
your tests.  The whole point of this is to keep one CPU tied up (so it
can't be used in your other work) but in one case tied up hitting the
mqueue code so we can see the effect of walking that 65,528 element array
one pointer at a time on the global CPU cache.  If it's bad, then it will
slow down your app on the other CPUs just by polluting cache mercilessly.
In the fake case, it will be in a tight loop, but not polluting cache.
Testing the mqueue subsystem directly - Here we just run a number of tests
to see how the mqueue subsystem performs under different conditions.  A
couple conditions are known to be worst case for the old system, and some
routines, so this tests all of them.

So, on to the results already:

Subsystem/Test                  Old                         New

Time to compile linux
kernel (make -j12 on a
6 core CPU)
  Running mqueue test     user 49m10.744s             user 45m26.294s
			   sys  5m51.924s              sys  4m59.894s
			 total 55m02.668s            total 50m26.188s

  Running fake test       user 45m32.686s             user 45m18.552s
                           sys  5m12.465s              sys  4m56.468s
                         total 50m45.151s            total 50m15.020s

  % slowdown from mqueue
    cache thrashing            ~8%                         ~.5%

Avg time to send/recv (in nanoseconds per message)
  when queue empty            305/288                    349/318
  when queue full (65528 messages)
    constant priority      526589/823                    362/314
    increasing priority    403105/916                    495/445
    decreasing priority     73420/594                    482/409
    random priority        280147/920                    546/436

Time to fill/drain queue (65528 messages, in seconds)
  constant priority         17.37/.12                    .13/.12
  increasing priority        4.14/.14                    .21/.18
  decreasing priority       12.93/.13                    .21/.18
  random priority            8.88/.16                    .22/.17

So, I think the results speak for themselves.  It's possible this
implementation could be improved by cacheing at least one priority level
in the node tree (that would bring the queue empty performance more in
line with the old implementation), but this works and is *so* much better
than what we had, especially for the common case of a single priority in
use, that further refinements can be in follow on patches.

[akpm@linux-foundation.org: fix typo in comment, remove stray semicolon]
[levinsasha928@gmail.com: use correct gfp flags in msg_insert]
Signed-off-by: Doug Ledford 
Cc: Stephen Rothwell 
Cc: Manfred Spraul 
Acked-by: KOSAKI Motohiro 
Signed-off-by: Sasha Levin 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mqueue: separate mqueue default value from maximum value

2012-06-01T00:49:31+00:00

Commit b231cca4381e ("message queues: increase range limits") changed
mqueue default value when attr parameter is specified NULL from hard
coded value to fs.mqueue.{msg,msgsize}_max sysctl value.

This made large side effect.  When user need to use two mqueue
applications 1) using !NULL attr parameter and it require big message
size and 2) using NULL attr parameter and only need small size message,
app (1) require to raise fs.mqueue.msgsize_max and app (2) consume large
memory size even though it doesn't need.

Doug Ledford propsed to switch back it to static hard coded value.
However it also has a compatibility problem.  Some applications might
started depend on the default value is tunable.

The solution is to separate default value from maximum value.

Signed-off-by: KOSAKI Motohiro 
Signed-off-by: Doug Ledford 
Acked-by: Doug Ledford 
Acked-by: Joe Korty 
Cc: Amerigo Wang 
Acked-by: Serge E. Hallyn 
Cc: Jiri Slaby 
Cc: Manfred Spraul 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mqueue: don't use kmalloc with KMALLOC_MAX_SIZE

2012-06-01T00:49:31+00:00

KMALLOC_MAX_SIZE is not a good threshold.  It is extremely high and
problematic.  Unfortunately, some silly drivers depend on this and we
can't change it.  But any new code needn't use such extreme ugly high
order allocations.  It brings us awful fragmentation issues and system
slowdown.

Signed-off-by: KOSAKI Motohiro 
Acked-by: Doug Ledford 
Acked-by: Joe Korty 
Cc: Amerigo Wang 
Cc: Serge E. Hallyn 
Cc: Jiri Slaby 
Cc: Joe Korty 
Cc: Manfred Spraul 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds