witness
—
lock validation facility
options WITNESS
options WITNESS_COUNT
options WITNESS_KDB
options WITNESS_NO_VNODE
options WITNESS_SKIPSPIN
The witness
module keeps track of the locks acquired and
released by each thread. It also keeps track of the order in which locks are
acquired with respect to each other. Each time a lock is acquired,
witness
uses these two lists to verify that a lock is
not being acquired in the wrong order. If a lock order violation is detected,
then a message is output to the kernel console or log detailing the locks
involved and the locations in question. Witness can also be configured to drop
into the kernel debugger when an order violation occurs.
The witness
code also checks various other
conditions such as verifying that one does not recurse on a non-recursive
lock, or attempt an upgrade on a shared lock held by another thread. If any
of these checks fail, then the kernel will panic.
The WITNESS_COUNT
kernel option controls
the maximum number of witness
entries that are
tracked in the kernel. The maximum number of entries can be queried via the
debug.witness.count sysctl. It can also be set from
the
loader(8)
via the debug.witness.count environment variable.
The WITNESS_NO_VNODE
kernel option tells
witness
to ignore locking issues between
vnode(9)
objects.
The flag that controls whether or not the kernel debugger is
entered when a lock order violation is detected can be set in a variety of
ways. By default, the flag is off, but if the
WITNESS_KDB
kernel option is specified, then the
flag will default to on. It can also be set from the
loader(8)
via the debug.witness.kdb environment variable or
after the kernel has booted via the debug.witness.kdb
sysctl. If the flag is set to zero, then the debugger will not be entered.
If the flag is non-zero, then the debugger will be entered.
The witness
code can also be configured to
skip all checks on spin mutexes. By default, this flag defaults to off, but
it can be turned on by specifying the
WITNESS_SKIPSPIN
kernel option. The flag can also be
set via the
loader(8)
environment variable debug.witness.skipspin. If the
variable is set to a non-zero value, then spin mutexes are skipped. Once the
kernel has booted, the status of this flag can be examined but not set via
the read-only sysctl debug.witness.skipspin.
The sysctl debug.witness.watch specifies the
level of witness involvement in the system. A value of 1 specifies that
witness is enabled. A value of 0 specifies that witness is disabled, but
that can be enabled again. This will maintain a small amount of overhead in
the system. A value of -1 specifies that witness is disabled permanently and
cannot be enabled again. The sysctl
debug.witness.watch can be set via
loader(8).
The sysctl debug.witness.output_channel
specifies the output channel used to display warnings emitted by
witness
. The possible values are
‘console
’, indicating that warnings
are to be printed to the system console,
‘log
’, indicating that warnings are to
be logged via
log(9),
and ‘none
’. This sysctl can be set via
loader(8).
The witness
code also provides three extra
ddb(4)
commands if both witness
and
ddb(4) are
compiled into the kernel:
show
locks
[thread]
- Outputs the list of locks held by a thread to the kernel console along
with the filename and line number at which each lock was last acquired by
the thread. The optional thread argument may be
either a TID, PID, or pointer to a thread structure. If
thread is not specified, then the locks held by the
current thread are displayed.
show
all locks
- Outputs the list of locks held by all threads in the system to the kernel
console.
show
witness
- Dump the current order list to the kernel console. The code first displays
the lock order tree for all of the sleep locks. Then it displays the lock
order tree for all of the spin locks. Finally, it displays a list of locks
that have not yet been acquired.
The witness
code first appeared in
BSD/OS 5.0 and was imported from there into
FreeBSD 5.0.