While this code is executed with the wait_lock held, a reader can
acquire the lock without holding wait_lock. The writer side loops
checking the value with the atomic_cond_read_acquire(), but only truly
acquires the lock when the compare-and-exchange is completed
successfully which isn’t ordered. This exposes the window between the
acquire and the cmpxchg to an A-B-A problem which allows reads following
the lock acquisition to observe values speculatively before the write
lock is truly acquired.
We've seen a problem in epoll where the reader does a xchg while
holding the read lock, but the writer can see a value change out from under it.
Writer | Reader 2
--------------------------------------------------------------------------------
ep_scan_ready_list() |
|- write_lock_irq() |
|- queued_write_lock_slowpath() |
|- atomic_cond_read_acquire() |
| read_lock_irqsave(&ep->lock, flags);
--> (observes value before unlock)| chain_epi_lockless()
| | epi->next = xchg(&ep->ovflist, epi);
| | read_unlock_irqrestore(&ep->lock, flags);
| |
| atomic_cmpxchg_relaxed() |
|-- READ_ONCE(ep->ovflist); |
A core can order the read of the ovflist ahead of the
atomic_cmpxchg_relaxed(). Switching the cmpxchg to use acquire semantics
addresses this issue at which point the atomic_cond_read can be switched
to use relaxed semantics.
Fixes: b519b56e378ee ("locking/qrwlock: Use atomic_cond_read_acquire() when spinning in qrwlock")
Signed-off-by: Ali Saidi <[email protected]>
Cc: [email protected]
Acked-by: Will Deacon <[email protected]>
Tested-by: Steve Capper <[email protected]>
Reviewed-by: Steve Capper <[email protected]>
---
kernel/locking/qrwlock.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/kernel/locking/qrwlock.c b/kernel/locking/qrwlock.c
index 4786dd271b45..10770f6ac4d9 100644
--- a/kernel/locking/qrwlock.c
+++ b/kernel/locking/qrwlock.c
@@ -73,8 +73,8 @@ void queued_write_lock_slowpath(struct qrwlock *lock)
/* When no more readers or writers, set the locked flag */
do {
- atomic_cond_read_acquire(&lock->cnts, VAL == _QW_WAITING);
- } while (atomic_cmpxchg_relaxed(&lock->cnts, _QW_WAITING,
+ atomic_cond_read_relaxed(&lock->cnts, VAL == _QW_WAITING);
+ } while (atomic_cmpxchg_acquire(&lock->cnts, _QW_WAITING,
_QW_LOCKED) != _QW_WAITING);
unlock:
arch_spin_unlock(&lock->wait_lock);
--
2.24.4.AMZN
On 4/15/21 1:27 PM, Ali Saidi wrote:
> While this code is executed with the wait_lock held, a reader can
> acquire the lock without holding wait_lock. The writer side loops
> checking the value with the atomic_cond_read_acquire(), but only truly
> acquires the lock when the compare-and-exchange is completed
> successfully which isn’t ordered. This exposes the window between the
> acquire and the cmpxchg to an A-B-A problem which allows reads following
> the lock acquisition to observe values speculatively before the write
> lock is truly acquired.
>
> We've seen a problem in epoll where the reader does a xchg while
> holding the read lock, but the writer can see a value change out from under it.
>
> Writer | Reader 2
> --------------------------------------------------------------------------------
> ep_scan_ready_list() |
> |- write_lock_irq() |
> |- queued_write_lock_slowpath() |
> |- atomic_cond_read_acquire() |
> | read_lock_irqsave(&ep->lock, flags);
> --> (observes value before unlock)| chain_epi_lockless()
> | | epi->next = xchg(&ep->ovflist, epi);
> | | read_unlock_irqrestore(&ep->lock, flags);
> | |
> | atomic_cmpxchg_relaxed() |
> |-- READ_ONCE(ep->ovflist); |
>
> A core can order the read of the ovflist ahead of the
> atomic_cmpxchg_relaxed(). Switching the cmpxchg to use acquire semantics
> addresses this issue at which point the atomic_cond_read can be switched
> to use relaxed semantics.
>
> Fixes: b519b56e378ee ("locking/qrwlock: Use atomic_cond_read_acquire() when spinning in qrwlock")
> Signed-off-by: Ali Saidi <[email protected]>
> Cc: [email protected]
> Acked-by: Will Deacon <[email protected]>
> Tested-by: Steve Capper <[email protected]>
> Reviewed-by: Steve Capper <[email protected]>
>
> ---
> kernel/locking/qrwlock.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/locking/qrwlock.c b/kernel/locking/qrwlock.c
> index 4786dd271b45..10770f6ac4d9 100644
> --- a/kernel/locking/qrwlock.c
> +++ b/kernel/locking/qrwlock.c
> @@ -73,8 +73,8 @@ void queued_write_lock_slowpath(struct qrwlock *lock)
>
> /* When no more readers or writers, set the locked flag */
> do {
> - atomic_cond_read_acquire(&lock->cnts, VAL == _QW_WAITING);
> - } while (atomic_cmpxchg_relaxed(&lock->cnts, _QW_WAITING,
> + atomic_cond_read_relaxed(&lock->cnts, VAL == _QW_WAITING);
> + } while (atomic_cmpxchg_acquire(&lock->cnts, _QW_WAITING,
> _QW_LOCKED) != _QW_WAITING);
> unlock:
> arch_spin_unlock(&lock->wait_lock);
Acked-by: Waiman Long <[email protected]>
I've edited the thing to look like so. I'll go queue it for
locking/urgent.
---
Subject: locking/qrwlock: Fix ordering in queued_write_lock_slowpath()
From: Ali Saidi <[email protected]>
Date: Thu, 15 Apr 2021 17:27:11 +0000
From: Ali Saidi <[email protected]>
While this code is executed with the wait_lock held, a reader can
acquire the lock without holding wait_lock. The writer side loops
checking the value with the atomic_cond_read_acquire(), but only truly
acquires the lock when the compare-and-exchange is completed
successfully which isn’t ordered. This exposes the window between the
acquire and the cmpxchg to an A-B-A problem which allows reads
following the lock acquisition to observe values speculatively before
the write lock is truly acquired.
We've seen a problem in epoll where the reader does a xchg while
holding the read lock, but the writer can see a value change out from
under it.
Writer | Reader
--------------------------------------------------------------------------------
ep_scan_ready_list() |
|- write_lock_irq() |
|- queued_write_lock_slowpath() |
|- atomic_cond_read_acquire() |
| read_lock_irqsave(&ep->lock, flags);
--> (observes value before unlock) | chain_epi_lockless()
| | epi->next = xchg(&ep->ovflist, epi);
| | read_unlock_irqrestore(&ep->lock, flags);
| |
| atomic_cmpxchg_relaxed() |
|-- READ_ONCE(ep->ovflist); |
A core can order the read of the ovflist ahead of the
atomic_cmpxchg_relaxed(). Switching the cmpxchg to use acquire
semantics addresses this issue at which point the atomic_cond_read can
be switched to use relaxed semantics.
Fixes: b519b56e378ee ("locking/qrwlock: Use atomic_cond_read_acquire() when spinning in qrwlock")
Signed-off-by: Ali Saidi <[email protected]>
[peterz: use try_cmpxchg()]
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Steve Capper <[email protected]>
Acked-by: Will Deacon <[email protected]>
Tested-by: Steve Capper <[email protected]>
---
kernel/locking/qrwlock.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
--- a/kernel/locking/qrwlock.c
+++ b/kernel/locking/qrwlock.c
@@ -60,6 +60,8 @@ EXPORT_SYMBOL(queued_read_lock_slowpath)
*/
void queued_write_lock_slowpath(struct qrwlock *lock)
{
+ int cnts;
+
/* Put the writer into the wait queue */
arch_spin_lock(&lock->wait_lock);
@@ -73,9 +75,8 @@ void queued_write_lock_slowpath(struct q
/* When no more readers or writers, set the locked flag */
do {
- atomic_cond_read_acquire(&lock->cnts, VAL == _QW_WAITING);
- } while (atomic_cmpxchg_relaxed(&lock->cnts, _QW_WAITING,
- _QW_LOCKED) != _QW_WAITING);
+ cnts = atomic_cond_read_relaxed(&lock->cnts, VAL == _QW_WAITING);
+ } while (!atomic_try_cmpxchg_acquire(&lock->cnts, &cnts, _QW_LOCKED));
unlock:
arch_spin_unlock(&lock->wait_lock);
}