Bug: 33346201
Test: trigger LMK and verify tgid is there
Change-Id: I047abc3aa541522766d2a84ebb4c77caf57a18a3
Signed-off-by: Joel Fernandes <joelaf@google.com>
This reverts commit 9d19f72b43.
This fixes CVE-2017-0710.
SELinux allows more fine grained control: We grant processes that need
access to smaps CAP_SYS_PTRACE but prohibit them from using ptrace
attach().
Bug: 34951864
Bug: 36468447
Change-Id: I8ea67f8771ec212950bc251ee750bd8a7e7c0643
Signed-off-by: Daniel Mentz <danielmentz@google.com>
Mark as rw when adding as linear target to allow changes
to the underlying filesystem through adb disable verity
and adb remount.
(Cherry-picked from
https://partner-android-review.googlesource.com/#/c/613573/
79a3032bb62da65a5d724eb70c8bdc662945d475)
BUG: 28845874
Signed-off-by: Badhri Jagan Sridharan <Badhri@google.com>
Change-Id: If41e9cad8e0f054f4778c09a6e2f0cb8af6fddaf
The mmap_sem lock does not appear to be protecting
anything, and has been removed in Samsung's more
recent versions of sdcardfs.
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Change-Id: I76ff3e33002716b8384fc8be368028ed63dffe4e
Bug: 63785372
Because we have disabled RT priority inheritance for
the regular binder domain, the following can happen:
1) thread A (prio 98) calls into thread B
2) because RT prio inheritance is disabled, thread B
runs at the lowest nice (prio 100) instead
3) thread B calls back into A; A will run at prio 100
for the duration of the transaction
4) When thread A is done with the call from B, we will
try to restore the prio back to 98. But, we fail
because the process doesn't hold CAP_SYS_NICE,
neither is RLIMIT_RT_PRIO set.
While the proper fix going forward will be to
correctly apply CAP_SYS_NICE or RLIMIT_RT_PRIO,
for now it seems reasonable to not check permissions
on the restore path.
Change-Id: Ibede5960c9b7bb786271c001e405de50be64d944
Signed-off-by: Martijn Coenen <maco@android.com>
The BINDER_GET_NODE_DEBUG_INFO ioctl will return debug info on
a node. Each successive call reusing the previous return value
will return the next node. The data will be used by
libmemunreachable to mark the pointers with kernel references
as reachable.
Bug: 28275695
Change-Id: Idbbafa648a33822dc023862cd92b51a595cf7c1c
Signed-off-by: Colin Cross <ccross@android.com>
Signed-off-by: Martijn Coenen <maco@android.com>
sugov_update_commit() calls trace_cpu_frequency() to record the
current CPU frequency if it has not changed in the fast switch case
to prevent utilities from getting confused (they may report that the
CPU is idle if the frequency has not been recorded for too long, for
example).
However, that may cause the tracepoint to be triggered quite often
for no real reason (if the frequency doesn't change, we will not
modify the last update time stamp and governor computations may
run again shortly when that happens), so don't do that (arguably, it
is done to work around a utilities bug anyway).
That allows code duplication in sugov_update_commit() to be reduced
somewhat too.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
(cherry picked from commit 38d4ea229d)
(conflicts with sugov_up_down_rate_limit resolved)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: Ia019dda29b8c1c4cf3553da75c88d066eb5674e9
The way the schedutil governor uses the PELT metric causes it to
underestimate the CPU utilization in some cases.
That can be easily demonstrated by running kernel compilation on
a Sandy Bridge Intel processor, running turbostat in parallel with
it and looking at the values written to the MSR_IA32_PERF_CTL
register. Namely, the expected result would be that when all CPUs
were 100% busy, all of them would be requested to run in the maximum
P-state, but observation shows that this clearly isn't the case.
The CPUs run in the maximum P-state for a while and then are
requested to run slower and go back to the maximum P-state after
a while again. That causes the actual frequency of the processor to
visibly oscillate below the sustainable maximum in a jittery fashion
which clearly is not desirable.
That has been attributed to CPU utilization metric updates on task
migration that cause the total utilization value for the CPU to be
reduced by the utilization of the migrated task. If that happens,
the schedutil governor may see a CPU utilization reduction and will
attempt to reduce the CPU frequency accordingly right away. That
may be premature, though, for example if the system is generally
busy and there are other runnable tasks waiting to be run on that
CPU already.
This is unlikely to be an issue on systems where cpufreq policies are
shared between multiple CPUs, because in those cases the policy
utilization is computed as the maximum of the CPU utilization values
over the whole policy and if that turns out to be low, reducing the
frequency for the policy most likely is a good idea anyway. On
systems with one CPU per policy, however, it may affect performance
adversely and even lead to increased energy consumption in some cases.
On those systems it may be addressed by taking another utilization
metric into consideration, like whether or not the CPU whose
frequency is about to be reduced has been idle recently, because if
that's not the case, the CPU is likely to be busy in the near future
and its frequency should not be reduced.
To that end, use the counter of idle calls in the timekeeping code.
Namely, make the schedutil governor look at that counter for the
current CPU every time before its frequency is about to be reduced.
If the counter has not changed since the previous iteration of the
governor computations for that CPU, the CPU has been busy for all
that time and its frequency should not be decreased, so if the new
frequency would be lower than the one set previously, the governor
will skip the frequency update.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Joel Fernandes <joelaf@google.com>
(cherry picked from commit b7eaf1aab9)
(simple CPUFREQ_RT_DL vs CPUFREQ_DL usage conflicts)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: I531ec02c052944ee07a904dc2a25c59948ee762b
The loop in sugov_next_freq_shared() contains an if block to skip the
loop for the current CPU. This turns out to be an unnecessary
conditional in the scheduler's hot-path for every CPU in the policy.
It would be better to drop the conditional and make the loop treat all
the CPUs in the same way. That would eliminate the need of calling
sugov_iowait_boost() at the top of the routine.
To keep the code optimized to return early if the current CPU has RT/DL
flags set, move the flags check to sugov_update_shared() instead in
order to avoid the function call entirely.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
(cherry picked from commit cba1dfb57b)
(modified for SCHED_CPUFREQ_DL vs SCHED_CPUFREQ_RT)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: Ie046fdc8eda46821356750edd0fb6f7d077af363
sugov_start() only initializes struct sugov_cpu per-CPU structures
for shared policies, but it should do that for single-CPU policies too.
That in particular makes the IO-wait boost mechanism work in the
cases when cpufreq policies correspond to individual CPUs.
Fixes: 21ca6d2c52 (cpufreq: schedutil: Add iowait boosting)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Cc: 4.9+ <stable@vger.kernel.org> # 4.9+
(cherry picked from commit 4296f23ed4)
(we use SCHED_CPUFREQ_DL instead of SCHED_CPUFREQ_RT in cpu->flags)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: I5b837a0ee4432115d85caa1a9808ea61e1e1b07f
get_next_freq() uses sg_cpu only to get sg_policy, which the callers of
get_next_freq() already have. Pass sg_policy instead of sg_cpu to
get_next_freq(), to make it more efficient.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
(cherry picked from commit 655cb1ebff)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: Ia210058da32930a6cdb18258aa679cd1a44a747e
cached_raw_freq applies to the entire cpufreq policy and not individual
CPUs. Apart from wasting per-cpu memory, it is actually wrong to keep it
in struct sugov_cpu as we may end up comparing next_freq with a stale
cached_raw_freq of a random CPU.
Move cached_raw_freq to struct sugov_policy.
Fixes: 5cbea46984 (cpufreq: schedutil: map raw required frequency to driver frequency)
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
(cherry-picked from 6c4f0fa643)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: Ie91420f710819b383947f9031da9be1f3bb7f636
Execute the irq-work specific initialization/exit code only when the
fast path isn't available.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
(cherry picked from commit 21ef57297b)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: Icfd68f455ef71846d799fcd2d8ec6aa1bf59573e
The fast_switch_enabled flag will be used by both sugov_policy_alloc()
and sugov_policy_free() with a later patch.
Prepare for that by moving the calls to enable and disable it to the
beginning of sugov_init() and end of sugov_exit().
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
(cherry picked from commit 4a71ce4348)
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
Change-Id: Ia174f423ca02d59360657ac2e77a5098ce5cf99c
Allows a binder node to specify whether it wants to
inherit real-time scheduling policy from a caller.
Change-Id: I375b6094bf441c19f19cba06d5a6be02cd07d714
Signed-off-by: Martijn Coenen <maco@android.com>
By raising the priority of a thread selected for
a transaction *before* we wake it up.
Delay restoring the priority when doing a reply
until after we wake-up the process receiving
the reply.
Change-Id: Ic332e4e0ed7d2d3ca6ab1034da4629c9eadd3405
Signed-off-by: Martijn Coenen <maco@google.com>
This change adds flags to flat_binder_object.flags
to allow indicating a minimum scheduling policy for
the node. It also clarifies the valid value range
for the priority bits in the flags.
Internally, we use the priority map that the kernel
uses, e.g. [0..99] for real-time policies and [100..139]
for the SCHED_NORMAL/SCHED_BATCH policies.
Bug: 34461621
Bug: 37293077
Change-Id: I12438deecb53df432da18c6fc77460768ae726d2
Signed-off-by: Martijn Coenen <maco@google.com>
Instead of pushing new transactions to the process
waitqueue, select a thread that is waiting on proc
work to handle the transaction. This will make it
easier to improve priority inheritance in future
patches, by setting the priority before we wake up
a thread.
If we can't find a waiting thread, submit the work
to the proc waitqueue instead as we did previously.
Change-Id: I23cbfcca867bed7b86007e22137d0a8fad4b4001
Signed-off-by: Martijn Coenen <maco@google.com>
Removes the process waitqueue, so that threads
can only wait on the thread waitqueue. Whenever
there is process work to do, pick a thread and
wake it up.
This also fixes an issue with using epoll(),
since we no longer have to block on different
waitqueues.
Bug: 34461621
Change-Id: I2950b9de6fa078ee72d53c667a03cbaf587f0849
Signed-off-by: Martijn Coenen <maco@google.com>
(from https://patchwork.kernel.org/patch/9817765/)
A race existed where one thread could register
a death notification for a node, while another
thread was cleaning up that node and sending
out death notifications for its references,
causing simultaneous access to ref->death
because different locks were held.
Test: boots, manual testing
Change-Id: Iff73312f34f70374f417beba4c4c82dd33cac119
Signed-off-by: Martijn Coenen <maco@google.com>
(from https://patchwork.kernel.org/patch/9817761/)
When printing transactions there were several race conditions
that could cause a stale pointer to be deferenced. Fixed by
reading the pointer once and using it if valid (which is
safe). The transaction buffer also needed protection via proc
lock, so it is only printed if we are holding the correct lock.
Bug: 36650912
Test: tested manually
Change-Id: I78240f99cc1a070d70a841c0d84d4306e2fd528d
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817771/)
Use proc->outer_lock to protect the binder_ref structure.
The outer lock allows functions operating on the binder_ref
to do nested acquires of node and inner locks as necessary
to attach refs to nodes atomically.
Binder refs must never be accesssed without holding the
outer lock.
Change-Id: Icf6add0eddf70473b39239960b2d9a524775b53a
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817763/)
Use the inner lock to protect thread accounting fields in
proc structure: max_threads, requested_threads,
requested_threads_started and ready_threads.
Change-Id: I5a17eb68812702f803d4e2806e7887de0b3af18e
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817779/)
This makes future changes to priority inheritance
easier, since we want to be able to look at a thread's
transaction stack when selecting a thread to inherit
priority for.
It also allows us to take just a single lock in a
few paths, where we used to take two in succession.
Change-Id: Idb1b6e9faa5c669978b2b3011fe326be8aece586
Signed-off-by: Martijn Coenen <maco@google.com>
(from https://patchwork.kernel.org/patch/9817775/)
proc->threads will need to be accessed with higher
locks of other processes held so use proc->inner_lock
to protect it. proc->tmp_ref now needs to be protected
by proc->inner_lock.
Change-Id: I176cfeca16bf7c9b34b428c16405f93db81d2ff8
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817783/)
When locks for binder_ref handling are added, proc->nodes
will need to be modified while holding the outer lock
Change-Id: I17b39e981c55130c14a62fe49900eceff6e3642b
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817769/)
node->node_lock is used to protect elements of node. No
need to acquire for fields that are invariant: debug_id,
ptr, cookie.
Change-Id: Ib7738e52fa7689767f17136e18cc05ff548b5717
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817769/)
The todo lists in the proc, thread, and node structures
are accessed by other procs/threads to place work
items on the queue.
The todo lists are protected by the new proc->inner_lock.
No locks should ever be nested under these locks. As the
name suggests, an outer lock will be introduced in
a later patch.
Change-Id: I7720bacf5ebae4af177e22fcab0900d54c94c11a
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817789/)
For correct behavior we need to hold the inner lock when
dequeuing and processing node work in binder_thread_read.
We now hold the inner lock when we enter the switch statement
and release it after processing anything that might be
affected by other threads.
We also need to hold the inner lock to protect the node
weak/strong ref tracking fields as long as node->proc
is non-NULL (if it is NULL then we are guaranteed that
we don't have any node work queued).
This means that other functions that manipulate these fields
must hold the inner lock. Refactored these functions to use
the inner lock.
Change-Id: I02c5cfdd3ab6dadea7f07f2a275faf3e27be77ad
Test: tested manually
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817791/)
There are 3 main spinlocks which must be acquired in this
order:
1) proc->outer_lock : protects most fields of binder_proc,
binder_thread, and binder_ref structures. binder_proc_lock()
and binder_proc_unlock() are used to acq/rel.
2) node->lock : protects most fields of binder_node.
binder_node_lock() and binder_node_unlock() are
used to acq/rel
3) proc->inner_lock : protects the thread and node lists
(proc->threads, proc->nodes) and all todo lists associated
with the binder_proc (proc->todo, thread->todo,
proc->delivered_death and node->async_todo).
binder_inner_proc_lock() and binder_inner_proc_unlock()
are used to acq/rel
Any lock under procA must never be nested under any lock at the same
level or below on procB.
Functions that require a lock held on entry indicate which lock
in the suffix of the function name:
foo_olocked() : requires node->outer_lock
foo_nlocked() : requires node->lock
foo_ilocked() : requires proc->inner_lock
foo_iolocked(): requires proc->outer_lock and proc->inner_lock
foo_nilocked(): requires node->lock and proc->inner_lock
Change-Id: Ied42674486092a0e3bdde64356e45b2494844558
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817795/)
When obtaining a node via binder_get_node(),
binder_get_node_from_ref() or binder_new_node(),
increment node->tmp_refs to take a
temporary reference on the node to ensure the node
persists while being used. binder_put_node() must
be called to remove the temporary reference.
Change-Id: I962b39d5cd80b2d7e4786bb87236ede7914e2db7
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817781/)
Once locks are added, binder_ref's will only be accessed
safely with the proc lock held. Refactor the inc/dec paths
to make them atomic with the binder_get_ref* paths and
node inc/dec. For example, instead of:
ref = binder_get_ref(proc, handle, strong);
...
binder_dec_ref(ref, strong);
we now have:
ret = binder_dec_ref_for_handle(proc, handle, strong, &rdata);
Since the actual ref is no longer exposed to callers, a
new struct binder_ref_data is introduced which can be used
to return a copy of ref state.
Change-Id: I7de22107f8ebc967cee63251d584fceb4ea56250
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817787/)
binder_thread and binder_proc may be accessed by other
threads when processing transaction. Therefore they
must be prevented from being freed while a transaction
is in progress that references them.
This is done by introducing a temporary reference
counter for threads and procs that indicates that the
object is in use and must not be freed. binder_thread_dec_tmpref()
and binder_proc_dec_tmpref() are used to decrement
the temporary reference.
It is safe to free a binder_thread if there
is no reference and it has been released
(indicated by thread->is_dead).
It is safe to free a binder_proc if it has no
remaining threads and no reference.
A spinlock is added to the binder_transaction
to safely access and set references for t->from
and for debug code to safely access t->to_thread
and t->to_proc.
Change-Id: I0a00a0294c3e93aea8b3f141c6f18e77ad244078
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817787/)
When initiating a transaction, the target_node must
have a strong ref on it. Then we take a second
strong ref to make sure the node survives until the
transaction is complete.
Change-Id: If7429cb43eda520ab89d45df6c19327cee97c60c
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817805/)
Since errors are tracked in the return_error/return_error2
fields of the binder_thread object and BR_TRANSACTION_COMPLETEs
can be tracked either in those fields or via the thread todo
work list, it is possible for errors to be reported ahead
of the associated txn complete.
Use the thread todo work list for errors to guarantee
order. Also changed binder_send_failed_reply to pop
the transaction even if it failed to send a reply.
Bug: 37218618
Test: tested manually
Change-Id: I196cfaeed09fdcd697f8ab25eea4e04241fdb08f
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://lkml.org/lkml/2017/6/29/754)
binder_pop_transaction needs to be split into 2 pieces to
to allow the proc lock to be held on entry to dequeue the
transaction stack, but no lock when kfree'ing the transaction.
Split into binder_pop_transaction_locked and binder_free_transaction
(the actual locks are still to be added).
Change-Id: I848ae994cc27b3cd083cff2dbd1071762784f4a3
Test: tested manually
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817807/)
The log->next index for the transaction log was
not protected when incremented. This led to a
case where log->next++ resulted in an index
larger than ARRAY_SIZE(log->entry) and eventually
a bad access to memory.
Fixed by making the log index an atomic64 and
converting to an array by using "% ARRAY_SIZE(log->entry)"
Also added "complete" field to the log entry which is
written last to tell the print code whether the
entry is complete
Bug: 62038227
Test: tested manually
Change-Id: I1bb1c1a332a6ac458a626f5bedd05022b56b91f2
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817815/)
Adds protection against malicious user code freeing
the same buffer at the same time which could cause
a crash. Cannot happen under normal use.
Bug: 36650912
Change-Id: I43e078cbf31c0789aaff5ceaf8f1a94c75f79d45
Test: tested manually
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817819/)
node is always non-NULL in binder_get_ref_for_node so the
conditional and else clause are not needed
Change-Id: I23f011ba59e1869d9577e6bf28e1f1dd38f45713
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817799/)
The looper member of struct binder_thread is a bitmask
of control bits. All of the existing bits are modified
by the affected thread except for BINDER_LOOPER_STATE_NEED_RETURN
which can be modified in binder_deferred_flush() by
another thread.
To avoid adding a spinlock around all read-mod-writes to
modify a bit, the BINDER_LOOPER_STATE_NEED_RETURN flag
is replaced by a separate field in struct binder_thread.
Bug: 33250092 32225111
Change-Id: Ia4cefbdbd683c6cb17c323ba7d278de5f2ca0745
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817813/)
Currently, the transaction complete work item is queued
after the transaction. This means that it is possible
for the transaction to be handled and a reply to be
enqueued in the current thread before the transaction
complete is enqueued, which violates the protocol
with userspace who may not expect the transaction
complete. Fixed by always enqueing the transaction
complete first.
Also, once the transaction is enqueued, it is unsafe
to access since it might be freed. Currently,
t->flags is accessed to determine whether a sync
wake is needed. Changed to access tr->flags
instead.
Change-Id: I6c01566e167a39cf17c9027c3817618182e56975
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817757/)
In binder_thread_read, the BINDER_WORK_NODE command is used
to communicate the references on the node to userspace. It
can take a couple of iterations in the loop to construct
the list of commands for user space. When locking is added,
the lock would need to be release on each iteration which
means the state could change. The work item is not dequeued
during this process which prevents a simpler queue management
that can just dequeue up front and handle the work item.
Fixed by changing the BINDER_WORK_NODE algorithm in
binder_thread_read to determine which commands to send
to userspace atomically in 1 pass so it stays consistent
with the kernel view.
The work item is now dequeued immediately since only
1 pass is needed.
Change-Id: I9b4109997b2d53ba661867b14d7336cd076be06d
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817751/)
Add additional information to determine the cause of binder
failures. Adds the following to failed transaction log and
kernel messages:
return_error : value returned for transaction
return_error_param : errno returned by binder allocator
return_error_line : line number where error detected
Also, return BR_DEAD_REPLY if an allocation error indicates
a dead proc (-ESRCH)
Bug: 36406078
Change-Id: Ifc8881fa5adfcced3f2d67f9030fbd3efa3e2cab
Test: tested manually
Signed-off-by: Todd Kjos <tkjos@google.com>
(from https://patchwork.kernel.org/patch/9817755/)
Use atomics for stats to avoid needing to lock for
increments/decrements
Bug: 33250092 32225111
Change-Id: I13e69b7f0485ccf16673e25091455781e1933a98
Signed-off-by: Todd Kjos <tkjos@google.com>
