]> git.baikalelectronics.ru Git - kernel.git/commit
idr: make idr_remove_all() do removal -before- free_layer()
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Tue, 10 Mar 2009 19:55:52 +0000 (12:55 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 10 Mar 2009 22:55:11 +0000 (15:55 -0700)
commitf3560e763af0a504fe678ca9f09ae2a148def2c7
tree8743554370742f55d993bfd6b5a281f81bc4dd0b
parent8d02be58afd8226e79441b9c1a11cdd9b8e92d60
idr: make idr_remove_all() do removal -before- free_layer()

Fix a problem in the IDR system, where an idr_remove_all() hands a data
element to call_rcu() (via free_layer()) before making that data element
inaccessible to new readers.  This is very bad, and results in readers
still having a reference to this data element at the end of the grace
period.

Tests on large machines that concurrently map and unmap user-space memory
within the same multithreaded process result in crashes within about five
minutes.  Applying this patch increases the kernel's longevity to the
three-to-eight-hour range.

There appear to be other similar problems in idr_get_empty_slot() and
sub_remove(), but I fixed the easy one in idr_remove_all() first.  It is
therefore no surprise that failures still occur.

Located-by: Milton Miller II <miltonm@austin.ibm.com>
Tested-by: Milton Miller II <miltonm@austin.ibm.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
lib/idr.c