tree 9411715142e45d893e46154b1a5f686d6d3625a5
parent 59a5912569f51e802dfd6f63269c72e57b530d67
author Barret Rhoden <brho@cs.berkeley.edu> 1532096456 -0400
committer Barret Rhoden <brho@cs.berkeley.edu> 1532096754 -0400

Don't chan_release() from an RCU callback

RCU callbacks are not allowed to block.  chan_release(), which is triggered
by decreffing a chan, can be triggered from an RCU callback.

Here's an example of a callchain that would have panicked:

 #01 [<0xffffffffc200b7c2>] in sem_down  (block here)
 #02 [<0xffffffffc200c221>] in cv_wait
 #03 [<0xffffffffc204ee05>] in rendez_sleep (this actually panics now)
 #04 [<0xffffffffc2039fa4>] in __qbread
 #05 [<0xffffffffc207c014>] in doread
 #06 [<0xffffffffc207c901>] in mntrpcread
 #07 [<0xffffffffc207d1a5>] in mountio
 #08 [<0xffffffffc207d3b5>] in mountrpc
 #09 [<0xffffffffc207dbfd>] in mntclunk
 #10 [<0xffffffffc2031448>] in chan_release
 #11 [<0xffffffffc2030bcb>] in kref_put
 #12 [<0xffffffffc2078be0>] in gtfs_tf_free
 #13 [<0xffffffffc2042125>] in __tf_free
 #14 [<0xffffffffc204f7e2>] in rcu_mgmt_ktask
 #15 [<0xffffffffc200acc0>] in __ktask_wrapper
 #16 [<0xffffffffc205981f>] in process_routine_kmsg
 #17 [<0xffffffffc20534a8>] in __smp_idle

Signed-off-by: Barret Rhoden <brho@cs.berkeley.edu>