There is a bug observed when rtw89_core_tx_kick_off_and_wait() tries to
access already freed skb_data:
BUG: KFENCE: use-after-free write in rtw89_core_tx_kick_off_and_wait drivers/net/wireless/realtek/rtw89/core.c:1110
CPU: 6 UID: 0 PID: 41377 Comm: kworker/u64:24 Not tainted 6.17.0-rc1+ #1 PREEMPT(lazy)
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS edk2-20250523-14.fc42 05/23/2025
Workqueue: events_unbound cfg80211_wiphy_work [cfg80211]
Use-after-free write at 0x0000000020309d9d (in kfence-#251):
rtw89_core_tx_kick_off_and_wait drivers/net/wireless/realtek/rtw89/core.c:1110
rtw89_core_scan_complete drivers/net/wireless/realtek/rtw89/core.c:5338
rtw89_hw_scan_complete_cb drivers/net/wireless/realtek/rtw89/fw.c:7979
rtw89_chanctx_proceed_cb drivers/net/wireless/realtek/rtw89/chan.c:3165
rtw89_chanctx_proceed drivers/net/wireless/realtek/rtw89/chan.h:141
rtw89_hw_scan_complete drivers/net/wireless/realtek/rtw89/fw.c:8012
rtw89_mac_c2h_scanofld_rsp drivers/net/wireless/realtek/rtw89/mac.c:5059
rtw89_fw_c2h_work drivers/net/wireless/realtek/rtw89/fw.c:6758
process_one_work kernel/workqueue.c:3241
worker_thread kernel/workqueue.c:3400
kthread kernel/kthread.c:463
ret_from_fork arch/x86/kernel/process.c:154
ret_from_fork_asm arch/x86/entry/entry_64.S:258
kfence-#251: 0x0000000056e2393d-0x000000009943cb62, size=232, cache=skbuff_head_cache
allocated by task 41377 on cpu 6 at 77869.159548s (0.009551s ago):
__alloc_skb net/core/skbuff.c:659
__netdev_alloc_skb net/core/skbuff.c:734
ieee80211_nullfunc_get net/mac80211/tx.c:5844
rtw89_core_send_nullfunc drivers/net/wireless/realtek/rtw89/core.c:3431
rtw89_core_scan_complete drivers/net/wireless/realtek/rtw89/core.c:5338
rtw89_hw_scan_complete_cb drivers/net/wireless/realtek/rtw89/fw.c:7979
rtw89_chanctx_proceed_cb drivers/net/wireless/realtek/rtw89/chan.c:3165
rtw89_chanctx_proceed drivers/net/wireless/realtek/rtw89/chan.c:3194
rtw89_hw_scan_complete drivers/net/wireless/realtek/rtw89/fw.c:8012
rtw89_mac_c2h_scanofld_rsp drivers/net/wireless/realtek/rtw89/mac.c:5059
rtw89_fw_c2h_work drivers/net/wireless/realtek/rtw89/fw.c:6758
process_one_work kernel/workqueue.c:3241
worker_thread kernel/workqueue.c:3400
kthread kernel/kthread.c:463
ret_from_fork arch/x86/kernel/process.c:154
ret_from_fork_asm arch/x86/entry/entry_64.S:258
freed by task 1045 on cpu 9 at 77869.168393s (0.001557s ago):
ieee80211_tx_status_skb net/mac80211/status.c:1117
rtw89_pci_release_txwd_skb drivers/net/wireless/realtek/rtw89/pci.c:564
rtw89_pci_release_tx_skbs.isra.0 drivers/net/wireless/realtek/rtw89/pci.c:651
rtw89_pci_release_tx drivers/net/wireless/realtek/rtw89/pci.c:676
rtw89_pci_napi_poll drivers/net/wireless/realtek/rtw89/pci.c:4238
__napi_poll net/core/dev.c:7495
net_rx_action net/core/dev.c:7557 net/core/dev.c:7684
handle_softirqs kernel/softirq.c:580
do_softirq.part.0 kernel/softirq.c:480
__local_bh_enable_ip kernel/softirq.c:407
rtw89_pci_interrupt_threadfn drivers/net/wireless/realtek/rtw89/pci.c:927
irq_thread_fn kernel/irq/manage.c:1133
irq_thread kernel/irq/manage.c:1257
kthread kernel/kthread.c:463
ret_from_fork arch/x86/kernel/process.c:154
ret_from_fork_asm arch/x86/entry/entry_64.S:258
It is a consequence of a race between the waiting and the signalling side
of the completion:
Thread 1 Thread 2
rtw89_core_tx_kick_off_and_wait()
rcu_assign_pointer(skb_data->wait, wait)
/* start waiting */
wait_for_completion_timeout()
rtw89_pci_tx_status()
rtw89_core_tx_wait_complete()
rcu_read_lock()
/* signals completion and
* proceeds further
*/
complete(&wait->completion)
rcu_read_unlock()
...
/* frees skb_data */
ieee80211_tx_status_ni()
/* returns (exit status doesn't matter) */
wait_for_completion_timeout()
...
/* accesses the already freed skb_data */
rcu_assign_pointer(skb_data->wait, NULL)
The signalling side might proceed and free the underlying skb even before
the waiting side is fully awoken and run to execution.
RCU synchronization here would work well if the signalling side didn't go
on and release skb on its own. Thus the waiting side should be told
somehow about what is happening on the completion side.
It seems the only correct way is to use standard locking primitives with
owner tracking, like was originally published in one [1] of the versions
of the patch mentioned in Fixes.
[1]: https://lore.kernel.org/linux-wireless/20230404025259.15503-3-pkshih@realtek.com/
Found by Linux Verification Center (linuxtesting.org).
Fixes: 1ae5ca615285 ("wifi: rtw89: add function to wait for completion of TX skbs")
Cc: stable@vger.kernel.org
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
---
The bug is tricky because the waiter-completer interaction isn't simple
here. I've tried to come up with something that wouldn't require taking
additional locks at rtw89_core_tx_wait_complete() but these ideas don't
eliminate the possible race entirely, to my mind.
Though one solution that _works_ currently is to get rid of
'struct rtw89_tx_wait_info' and replace it with the only field it is
used for - 'bool tx_done'. Then it can be stored at
'struct ieee80211_tx_info::status::status_driver_data' directly without
the need for allocating an extra dynamic object and tracking its lifecycle.
I didn't post this since then the structure won't be expandable for new
fields and that's probably the reason for why it wasn't done in this manner
initially.
drivers/net/wireless/realtek/rtw89/core.c | 15 ++++++++---
drivers/net/wireless/realtek/rtw89/core.h | 32 ++++++++++++++---------
drivers/net/wireless/realtek/rtw89/pci.c | 6 +++--
3 files changed, 36 insertions(+), 17 deletions(-)
diff --git a/drivers/net/wireless/realtek/rtw89/core.c b/drivers/net/wireless/realtek/rtw89/core.c
index 57590f5577a3..826540319027 100644
--- a/drivers/net/wireless/realtek/rtw89/core.c
+++ b/drivers/net/wireless/realtek/rtw89/core.c
@@ -1088,6 +1088,7 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev, struct sk_buff *sk
struct rtw89_tx_skb_data *skb_data = RTW89_TX_SKB_CB(skb);
struct rtw89_tx_wait_info *wait;
unsigned long time_left;
+ bool free_wait = true;
int ret = 0;
wait = kzalloc(sizeof(*wait), GFP_KERNEL);
@@ -1097,7 +1098,8 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev, struct sk_buff *sk
}
init_completion(&wait->completion);
- rcu_assign_pointer(skb_data->wait, wait);
+ spin_lock_init(&wait->owner_lock);
+ skb_data->wait = wait;
rtw89_core_tx_kick_off(rtwdev, qsel);
time_left = wait_for_completion_timeout(&wait->completion,
@@ -1107,8 +1109,15 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev, struct sk_buff *sk
else if (!wait->tx_done)
ret = -EAGAIN;
- rcu_assign_pointer(skb_data->wait, NULL);
- kfree_rcu(wait, rcu_head);
+ spin_lock_bh(&wait->owner_lock);
+ if (time_left == 0 && wait->owner != RTW89_TX_WAIT_OWNER_WAIT) {
+ free_wait = false;
+ wait->owner = RTW89_TX_WAIT_OWNER_COMPLETE;
+ }
+ spin_unlock_bh(&wait->owner_lock);
+
+ if (free_wait)
+ kfree(wait);
return ret;
}
diff --git a/drivers/net/wireless/realtek/rtw89/core.h b/drivers/net/wireless/realtek/rtw89/core.h
index 43e10278e14d..0117f24324d5 100644
--- a/drivers/net/wireless/realtek/rtw89/core.h
+++ b/drivers/net/wireless/realtek/rtw89/core.h
@@ -3506,14 +3506,21 @@ struct rtw89_phy_rate_pattern {
bool enable;
};
+enum rtw89_tx_wait_owner {
+ RTW89_TX_WAIT_OWNER_UNDET,
+ RTW89_TX_WAIT_OWNER_WAIT,
+ RTW89_TX_WAIT_OWNER_COMPLETE,
+};
+
struct rtw89_tx_wait_info {
- struct rcu_head rcu_head;
struct completion completion;
bool tx_done;
+ spinlock_t owner_lock; /* lock to access owner */
+ enum rtw89_tx_wait_owner owner;
};
struct rtw89_tx_skb_data {
- struct rtw89_tx_wait_info __rcu *wait;
+ struct rtw89_tx_wait_info *wait;
u8 hci_priv[];
};
@@ -7259,22 +7266,23 @@ static inline struct sk_buff *rtw89_alloc_skb_for_rx(struct rtw89_dev *rtwdev,
}
static inline void rtw89_core_tx_wait_complete(struct rtw89_dev *rtwdev,
- struct rtw89_tx_skb_data *skb_data,
+ struct rtw89_tx_wait_info *wait,
bool tx_done)
{
- struct rtw89_tx_wait_info *wait;
-
- rcu_read_lock();
-
- wait = rcu_dereference(skb_data->wait);
- if (!wait)
- goto out;
+ bool free_wait = true;
wait->tx_done = tx_done;
+
+ spin_lock_bh(&wait->owner_lock);
complete(&wait->completion);
+ if (wait->owner != RTW89_TX_WAIT_OWNER_COMPLETE) {
+ free_wait = false;
+ wait->owner = RTW89_TX_WAIT_OWNER_WAIT;
+ }
+ spin_unlock_bh(&wait->owner_lock);
-out:
- rcu_read_unlock();
+ if (free_wait)
+ kfree(wait);
}
static inline bool rtw89_is_mlo_1_1(struct rtw89_dev *rtwdev)
diff --git a/drivers/net/wireless/realtek/rtw89/pci.c b/drivers/net/wireless/realtek/rtw89/pci.c
index a669f2f843aa..d9d4558b21ea 100644
--- a/drivers/net/wireless/realtek/rtw89/pci.c
+++ b/drivers/net/wireless/realtek/rtw89/pci.c
@@ -462,9 +462,11 @@ static void rtw89_pci_tx_status(struct rtw89_dev *rtwdev,
struct sk_buff *skb, u8 tx_status)
{
struct rtw89_tx_skb_data *skb_data = RTW89_TX_SKB_CB(skb);
+ struct rtw89_tx_wait_info *wait = skb_data->wait;
struct ieee80211_tx_info *info;
- rtw89_core_tx_wait_complete(rtwdev, skb_data, tx_status == RTW89_TX_DONE);
+ if (wait)
+ rtw89_core_tx_wait_complete(rtwdev, wait, tx_status == RTW89_TX_DONE);
info = IEEE80211_SKB_CB(skb);
ieee80211_tx_info_clear_status(info);
@@ -1387,7 +1389,7 @@ static int rtw89_pci_txwd_submit(struct rtw89_dev *rtwdev,
}
tx_data->dma = dma;
- rcu_assign_pointer(skb_data->wait, NULL);
+ skb_data->wait = NULL;
txwp_len = sizeof(*txwp_info);
txwd_len = chip->txwd_body_size;
--
2.50.1Fedor Pchelkin <pchelkin@ispras.ru> wrote:
>
> There is a bug observed when rtw89_core_tx_kick_off_and_wait() tries to access already
> freed skb_data:
>
> BUG: KFENCE: use-after-free write in rtw89_core_tx_kick_off_and_wait
> drivers/net/wireless/realtek/rtw89/core.c:1110
>
> CPU: 6 UID: 0 PID: 41377 Comm: kworker/u64:24 Not tainted 6.17.0-rc1+ #1 PREEMPT(lazy)
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS edk2-20250523-14.fc42
> 05/23/2025
> Workqueue: events_unbound cfg80211_wiphy_work [cfg80211]
>
> Use-after-free write at 0x0000000020309d9d (in kfence-#251):
> rtw89_core_tx_kick_off_and_wait drivers/net/wireless/realtek/rtw89/core.c:1110
> rtw89_core_scan_complete drivers/net/wireless/realtek/rtw89/core.c:5338
> rtw89_hw_scan_complete_cb drivers/net/wireless/realtek/rtw89/fw.c:7979
> rtw89_chanctx_proceed_cb drivers/net/wireless/realtek/rtw89/chan.c:3165
> rtw89_chanctx_proceed drivers/net/wireless/realtek/rtw89/chan.h:141
> rtw89_hw_scan_complete drivers/net/wireless/realtek/rtw89/fw.c:8012
> rtw89_mac_c2h_scanofld_rsp drivers/net/wireless/realtek/rtw89/mac.c:5059
> rtw89_fw_c2h_work drivers/net/wireless/realtek/rtw89/fw.c:6758
> process_one_work kernel/workqueue.c:3241 worker_thread kernel/workqueue.c:3400
> kthread kernel/kthread.c:463 ret_from_fork arch/x86/kernel/process.c:154
> ret_from_fork_asm arch/x86/entry/entry_64.S:258
>
> kfence-#251: 0x0000000056e2393d-0x000000009943cb62, size=232,
> cache=skbuff_head_cache
>
> allocated by task 41377 on cpu 6 at 77869.159548s (0.009551s ago):
> __alloc_skb net/core/skbuff.c:659
> __netdev_alloc_skb net/core/skbuff.c:734 ieee80211_nullfunc_get
> net/mac80211/tx.c:5844 rtw89_core_send_nullfunc
> drivers/net/wireless/realtek/rtw89/core.c:3431
> rtw89_core_scan_complete drivers/net/wireless/realtek/rtw89/core.c:5338
> rtw89_hw_scan_complete_cb drivers/net/wireless/realtek/rtw89/fw.c:7979
> rtw89_chanctx_proceed_cb drivers/net/wireless/realtek/rtw89/chan.c:3165
> rtw89_chanctx_proceed drivers/net/wireless/realtek/rtw89/chan.c:3194
> rtw89_hw_scan_complete drivers/net/wireless/realtek/rtw89/fw.c:8012
> rtw89_mac_c2h_scanofld_rsp drivers/net/wireless/realtek/rtw89/mac.c:5059
> rtw89_fw_c2h_work drivers/net/wireless/realtek/rtw89/fw.c:6758
> process_one_work kernel/workqueue.c:3241 worker_thread kernel/workqueue.c:3400
> kthread kernel/kthread.c:463 ret_from_fork arch/x86/kernel/process.c:154
> ret_from_fork_asm arch/x86/entry/entry_64.S:258
>
> freed by task 1045 on cpu 9 at 77869.168393s (0.001557s ago):
> ieee80211_tx_status_skb net/mac80211/status.c:1117 rtw89_pci_release_txwd_skb
> drivers/net/wireless/realtek/rtw89/pci.c:564
> rtw89_pci_release_tx_skbs.isra.0 drivers/net/wireless/realtek/rtw89/pci.c:651
> rtw89_pci_release_tx drivers/net/wireless/realtek/rtw89/pci.c:676
> rtw89_pci_napi_poll drivers/net/wireless/realtek/rtw89/pci.c:4238
> __napi_poll net/core/dev.c:7495
> net_rx_action net/core/dev.c:7557 net/core/dev.c:7684 handle_softirqs
> kernel/softirq.c:580
> do_softirq.part.0 kernel/softirq.c:480
> __local_bh_enable_ip kernel/softirq.c:407 rtw89_pci_interrupt_threadfn
> drivers/net/wireless/realtek/rtw89/pci.c:927
> irq_thread_fn kernel/irq/manage.c:1133
> irq_thread kernel/irq/manage.c:1257
> kthread kernel/kthread.c:463
> ret_from_fork arch/x86/kernel/process.c:154 ret_from_fork_asm
> arch/x86/entry/entry_64.S:258
>
> It is a consequence of a race between the waiting and the signalling side of the completion:
>
> Thread 1 Thread 2
> rtw89_core_tx_kick_off_and_wait()
> rcu_assign_pointer(skb_data->wait, wait)
> /* start waiting */
> wait_for_completion_timeout()
> rtw89_pci_tx_status()
> rtw89_core_tx_wait_complete()
> rcu_read_lock()
> /* signals completion and
> * proceeds further
> */
>
> complete(&wait->completion)
> rcu_read_unlock()
> ...
> /* frees skb_data */
> ieee80211_tx_status_ni()
> /* returns (exit status doesn't matter) */
> wait_for_completion_timeout()
> ...
> /* accesses the already freed skb_data */
> rcu_assign_pointer(skb_data->wait, NULL)
>
> The signalling side might proceed and free the underlying skb even before the waiting side is
> fully awoken and run to execution.
>
> RCU synchronization here would work well if the signalling side didn't go on and release skb
> on its own. Thus the waiting side should be told somehow about what is happening on the
> completion side.
I reread the flow and am thinking about it a bit.
Actually, only when signaling side doesn't run on time, waiting side should update skb_data->wait.
(see code comments below)
>
> It seems the only correct way is to use standard locking primitives with owner tracking, like
> was originally published in one [1] of the versions of the patch mentioned in Fixes.
>
> [1]: https://lore.kernel.org/linux-wireless/20230404025259.15503-3-pkshih@realtek.com/
>
> Found by Linux Verification Center (linuxtesting.org).
>
> Fixes: 1ae5ca615285 ("wifi: rtw89: add function to wait for completion of TX skbs")
> Cc: stable@vger.kernel.org
> Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
> ---
>
> The bug is tricky because the waiter-completer interaction isn't simple here. I've tried to
> come up with something that wouldn't require taking additional locks at
> rtw89_core_tx_wait_complete() but these ideas don't eliminate the possible race entirely, to
> my mind.
Thank you for finding the potential race condition.
>
> Though one solution that _works_ currently is to get rid of 'struct rtw89_tx_wait_info' and
> replace it with the only field it is used for - 'bool tx_done'. Then it can be stored at 'struct
> ieee80211_tx_info::status::status_driver_data' directly without the need for allocating an
> extra dynamic object and tracking its lifecycle.
> I didn't post this since then the structure won't be expandable for new fields and that's
> probably the reason for why it wasn't done in this manner initially.
With a busy waiting on tx waiting side ?
If so, it would be unacceptable.
>
> drivers/net/wireless/realtek/rtw89/core.c | 15 ++++++++---
> drivers/net/wireless/realtek/rtw89/core.h | 32 ++++++++++++++---------
> drivers/net/wireless/realtek/rtw89/pci.c | 6 +++--
> 3 files changed, 36 insertions(+), 17 deletions(-)
>
> diff --git a/drivers/net/wireless/realtek/rtw89/core.c
> b/drivers/net/wireless/realtek/rtw89/core.c
> index 57590f5577a3..826540319027 100644
> --- a/drivers/net/wireless/realtek/rtw89/core.c
> +++ b/drivers/net/wireless/realtek/rtw89/core.c
> @@ -1088,6 +1088,7 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev,
> struct sk_buff *sk
> struct rtw89_tx_skb_data *skb_data = RTW89_TX_SKB_CB(skb);
> struct rtw89_tx_wait_info *wait;
> unsigned long time_left;
> + bool free_wait = true;
> int ret = 0;
>
> wait = kzalloc(sizeof(*wait), GFP_KERNEL); @@ -1097,7 +1098,8 @@ int
> rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev, struct sk_buff *sk
> }
>
> init_completion(&wait->completion);
> - rcu_assign_pointer(skb_data->wait, wait);
> + spin_lock_init(&wait->owner_lock);
> + skb_data->wait = wait;
>
> rtw89_core_tx_kick_off(rtwdev, qsel);
> time_left = wait_for_completion_timeout(&wait->completion,
> @@ -1107,8 +1109,15 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev,
> struct sk_buff *sk
> else if (!wait->tx_done)
> ret = -EAGAIN;
>
> - rcu_assign_pointer(skb_data->wait, NULL);
> - kfree_rcu(wait, rcu_head);
Please consider the following.
(moving "rcu_assign_pointer(skb_data->wait, NULL)" to be under "if (time_left == 0)")
if (time_left == 0) {
rcu_assign_pointer(skb_data->wait, NULL);
ret = -ETIMEDOUT;
} else if (!wait->tx_done) {
ret = -EAGAIN;
}
kfree_rcu(wait, rcu_head);
If completing side does run as expected (potential racing mentioned in this patch),
there is no real need to assign NULL back.
Thanks for the feedback, Zong-Zhe!
On Thu, 21. Aug 04:01, Zong-Zhe Yang wrote:
> Fedor Pchelkin <pchelkin@ispras.ru> wrote:
> > Though one solution that _works_ currently is to get rid of 'struct rtw89_tx_wait_info' and
> > replace it with the only field it is used for - 'bool tx_done'. Then it can be stored at 'struct
> > ieee80211_tx_info::status::status_driver_data' directly without the need for allocating an
> > extra dynamic object and tracking its lifecycle.
> > I didn't post this since then the structure won't be expandable for new fields and that's
> > probably the reason for why it wasn't done in this manner initially.
>
> With a busy waiting on tx waiting side ?
> If so, it would be unacceptable.
Ohh, I forgot about the need for async completion here. Nevermind that
solution, sorry.
>
> >
> > drivers/net/wireless/realtek/rtw89/core.c | 15 ++++++++---
> > drivers/net/wireless/realtek/rtw89/core.h | 32 ++++++++++++++---------
> > drivers/net/wireless/realtek/rtw89/pci.c | 6 +++--
> > 3 files changed, 36 insertions(+), 17 deletions(-)
> >
> > diff --git a/drivers/net/wireless/realtek/rtw89/core.c
> > b/drivers/net/wireless/realtek/rtw89/core.c
> > index 57590f5577a3..826540319027 100644
> > --- a/drivers/net/wireless/realtek/rtw89/core.c
> > +++ b/drivers/net/wireless/realtek/rtw89/core.c
> > @@ -1088,6 +1088,7 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev,
> > struct sk_buff *sk
> > struct rtw89_tx_skb_data *skb_data = RTW89_TX_SKB_CB(skb);
> > struct rtw89_tx_wait_info *wait;
> > unsigned long time_left;
> > + bool free_wait = true;
> > int ret = 0;
> >
> > wait = kzalloc(sizeof(*wait), GFP_KERNEL); @@ -1097,7 +1098,8 @@ int
> > rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev, struct sk_buff *sk
> > }
> >
> > init_completion(&wait->completion);
> > - rcu_assign_pointer(skb_data->wait, wait);
> > + spin_lock_init(&wait->owner_lock);
> > + skb_data->wait = wait;
> >
> > rtw89_core_tx_kick_off(rtwdev, qsel);
> > time_left = wait_for_completion_timeout(&wait->completion,
> > @@ -1107,8 +1109,15 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev,
> > struct sk_buff *sk
> > else if (!wait->tx_done)
> > ret = -EAGAIN;
> >
> > - rcu_assign_pointer(skb_data->wait, NULL);
> > - kfree_rcu(wait, rcu_head);
>
> Please consider the following.
> (moving "rcu_assign_pointer(skb_data->wait, NULL)" to be under "if (time_left == 0)")
>
There is still a tiny race window. Suppose wait_for_completion_timeout()
exits with a timeout, so time_left is 0. If completing side goes on in
parallel just after that, it has a chance to proceed and free skb_data
before the below if (time_left == 0) fragment is executed.
> if (time_left == 0) {
> rcu_assign_pointer(skb_data->wait, NULL);
> ret = -ETIMEDOUT;
> } else if (!wait->tx_done) {
> ret = -EAGAIN;
> }
>
> kfree_rcu(wait, rcu_head);
>
> If completing side does run as expected (potential racing mentioned in this patch),
> there is no real need to assign NULL back.
Actually the race happens regardless of wait_for_completion_timeout() exit
status, it's briefly mentioned in the race diagram inside commit message
(but the diagram can show only one possible concurrency scenario). I agree
this may be improved and described more explicitly though.
As for the patch itself, currently I can't see another way of fixing that
other than introducing locks on both waiting and completing side.
Fedor Pchelkin <pchelkin@ispras.ru> wrote:
>
> [...]
>
> > >
> > > drivers/net/wireless/realtek/rtw89/core.c | 15 ++++++++---
> > > drivers/net/wireless/realtek/rtw89/core.h | 32
> > > ++++++++++++++--------- drivers/net/wireless/realtek/rtw89/pci.c |
> > > 6 +++--
> > > 3 files changed, 36 insertions(+), 17 deletions(-)
> > >
> > > diff --git a/drivers/net/wireless/realtek/rtw89/core.c
> > > b/drivers/net/wireless/realtek/rtw89/core.c
> > > index 57590f5577a3..826540319027 100644
> > > --- a/drivers/net/wireless/realtek/rtw89/core.c
> > > +++ b/drivers/net/wireless/realtek/rtw89/core.c
> > > @@ -1088,6 +1088,7 @@ int rtw89_core_tx_kick_off_and_wait(struct
> > > rtw89_dev *rtwdev, struct sk_buff *sk
> > > struct rtw89_tx_skb_data *skb_data = RTW89_TX_SKB_CB(skb);
> > > struct rtw89_tx_wait_info *wait;
> > > unsigned long time_left;
> > > + bool free_wait = true;
> > > int ret = 0;
> > >
> > > wait = kzalloc(sizeof(*wait), GFP_KERNEL); @@ -1097,7
> > > +1098,8 @@ int rtw89_core_tx_kick_off_and_wait(struct rtw89_dev *rtwdev, struct
> sk_buff *sk
> > > }
> > >
> > > init_completion(&wait->completion);
> > > - rcu_assign_pointer(skb_data->wait, wait);
> > > + spin_lock_init(&wait->owner_lock);
> > > + skb_data->wait = wait;
> > >
> > > rtw89_core_tx_kick_off(rtwdev, qsel);
> > > time_left = wait_for_completion_timeout(&wait->completion,
> > > @@ -1107,8 +1109,15 @@ int rtw89_core_tx_kick_off_and_wait(struct
> > > rtw89_dev *rtwdev, struct sk_buff *sk
> > > else if (!wait->tx_done)
> > > ret = -EAGAIN;
> > >
> > > - rcu_assign_pointer(skb_data->wait, NULL);
> > > - kfree_rcu(wait, rcu_head);
> >
> > Please consider the following.
> > (moving "rcu_assign_pointer(skb_data->wait, NULL)" to be under "if
> > (time_left == 0)")
> >
>
> There is still a tiny race window. Suppose wait_for_completion_timeout() exits with a timeout,
> so time_left is 0. If completing side goes on in parallel just after that, it has a chance to
> proceed and free skb_data before the below if (time_left == 0) fragment is executed.
Okay, logically it sounds right.
>
> > if (time_left == 0) {
> > rcu_assign_pointer(skb_data->wait, NULL);
> > ret = -ETIMEDOUT;
> > } else if (!wait->tx_done) {
> > ret = -EAGAIN;
> > }
> >
> > kfree_rcu(wait, rcu_head);
> >
> > If completing side does run as expected (potential racing mentioned in
> > this patch), there is no real need to assign NULL back.
>
> Actually the race happens regardless of wait_for_completion_timeout() exit status, it's briefly
> mentioned in the race diagram inside commit message (but the diagram can show only one
> possible concurrency scenario). I agree this may be improved and described more explicitly
> though.
Will appreciate to see that in next version. Thanks.
>
> As for the patch itself, currently I can't see another way of fixing that other than introducing
> locks on both waiting and completing side.
I took some time on thinking this. The following is another idea.
The skb, which are sent by tx_wait_complete, are owned by driver.
They don't come from stack, so we don't need to do ieee80211_tx_status_ni.
Based on above, some rough points of the new idea are listed below.
1.
Let rtw89_core_tx_wait_complete
return true/false to indicate whether tx_wait or not
2.
Add some new field into rtw89_tx_wait_info
e.g. list_head, skb, finished
3.
Add a list_head to rtwdev
Add a work func doing things as
for each wait in rtwdev->XXX_list:
if !wait->finished:
wait_for_completion()
free wait->skb
free wait
4.
[waiting side] [completing side]
wait_for_completion_timeout() ...
... /* make complete the last step */
... if (rtw89_core_tx_wait_complete)
... return;
...
// not assign NULL back to skb_data->wait
if time_left != 0:
wait-> finished = true
wait->skb = skb
add wait to rtwdev->XXX_list
queue above work
Please help evaluate the new idea.
Thank you.
© 2016 - 2025 Red Hat, Inc.