In ext4, there is a race condition between changing inode journal mode
and ext4_writepages(). While ext4_writepages() is executed on a
non-journalled mode inode, the inode's journal mode could be enabled
by ioctl() and then, some pages dirtied after switching the journal
mode will be still exposed to ext4_writepages() in non-journaled mode.
To resolve this problem, we use fs-wide per-cpu rw semaphore by Jan
Kara's suggestion because we don't want to waste ext4_inode_info's
space for this extra rare case.
Signed-off-by: NDaeho Jeong <daeho.jeong@samsung.com>
Signed-off-by: NTheodore Ts'o <tytso@mit.edu>
Reviewed-by: NJan Kara <jack@suse.cz>

Based on Peter Zijlstra's earlier patch.

Change percpu_down_read() to use __down_read(), this way we can
do rwsem_acquire_read() unconditionally at the start to make this
code more symmetric and clean.

Originally-From: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Signed-off-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: NJosh Triplett <josh@joshtriplett.org>

Update the comments broken by the previous change.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Signed-off-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: NJosh Triplett <josh@joshtriplett.org>

Currently down_write/up_write calls synchronize_sched_expedited()
twice, which is evil.  Change this code to rely on rcu-sync primitives.
This avoids the _expedited "big hammer", and this can be faster in
the contended case or even in the case when a single thread does
down_write/up_write in a loop.

Of course, a single down_write() will take more time, but otoh it
will be much more friendly to the whole system.

To simplify the review this patch doesn't update the comments, fixed
by the next change.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Signed-off-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: NJosh Triplett <josh@joshtriplett.org>

This is the temporary ugly hack which will be reverted later. We only
need it to ensure that the next patch will not break "change sb_writers
to use percpu_rw_semaphore" patches routed via the VFS tree.

The alloc_super()->destroy_super() error path assumes that it is safe
to call percpu_free_rwsem() after kzalloc() without percpu_init_rwsem(),
so let's not disappoint it.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Signed-off-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: NJosh Triplett <josh@joshtriplett.org>

This commit exports percpu_down_read(), percpu_down_write(),
__percpu_init_rwsem(), percpu_up_read(), and percpu_up_write() to allow
locktorture to test them when built as a module.
Signed-off-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: NJosh Triplett <josh@joshtriplett.org>

Add percpu_down_read_trylock(), it will have the user soon.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>

Suggested-by: NIngo Molnar <mingo@kernel.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-52bjmtty46we26hbfd9sc9iy@git.kernel.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

Add lockdep annotations.  Not only this can help to find the potential
problems, we do not want the false warnings if, say, the task takes two
different percpu_rw_semaphore's for reading.  IOW, at least ->rw_sem
should not use a single class.

This patch exposes this internal lock to lockdep so that it represents the
whole percpu_rw_semaphore.  This way we do not need to add another "fake"
->lockdep_map and lock_class_key.  More importantly, this also makes the
output from lockdep much more understandable if it finds the problem.

In short, with this patch from lockdep pov percpu_down_read() and
percpu_up_read() acquire/release ->rw_sem for reading, this matches the
actual semantics.  This abuses __up_read() but I hope this is fine and in
fact I'd like to have down_read_no_lockdep() as well,
percpu_down_read_recursive_readers() will need it.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

percpu_rw_semaphore->writer_mutex was only added to simplify the initial
rewrite, the only thing it protects is clear_fast_ctr() which otherwise
could be called by multiple writers.  ->rw_sem is enough to serialize the
writers.

Kill this mutex and add "atomic_t write_ctr" instead.  The writers
increment/decrement this counter, the readers check it is zero instead of
mutex_is_locked().

Move atomic_add(clear_fast_ctr(), slow_read_ctr) under down_write() to
avoid the race with other writers.  This is a bit sub-optimal, only the
first writer needs this and we do not need to exclude the readers at this
stage.  But this is simple, we do not want another internal lock until we
add more features.

And this speeds up the write-contended case.  Before this patch the racing
writers sleep in synchronize_sched_expedited() sequentially, with this
patch multiple synchronize_sched_expedited's can "overlap" with each
other.  Note: we can do more optimizations, this is only the first step.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Currently the writer does msleep() plus synchronize_sched() 3 times to
acquire/release the semaphore, and during this time the readers are
blocked completely.  Even if the "write" section was not actually started
or if it was already finished.

With this patch down_write/up_write does synchronize_sched() twice and
down_read/up_read are still possible during this time, just they use the
slow path.

percpu_down_write() first forces the readers to use rw_semaphore and
increment the "slow" counter to take the lock for reading, then it
takes that rw_semaphore for writing and blocks the readers.

Also.  With this patch the code relies on the documented behaviour of
synchronize_sched(), it doesn't try to pair synchronize_sched() with
barrier.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Reviewed-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>