Hi Pauli,
On Tue, Dec 3, 2024 at 5:48 PM Pauli Virtanen <pav@xxxxxx> wrote:
>
> Hi,
>
> ti, 2024-12-03 kello 16:14 -0500, Luiz Augusto von Dentz kirjoitti:
> > From: Luiz Augusto von Dentz <luiz.von.dentz@xxxxxxxxx>
> >
> > This reworks hci_cb_list to not use mutex hci_cb_list_lock to avoid bugs
> > like the bellow:
> >
> > BUG: sleeping function called from invalid context at kernel/locking/mutex.c:585
> > in_atomic(): 0, irqs_disabled(): 0, non_block: 0, pid: 5070, name: kworker/u9:2
> > preempt_count: 0, expected: 0
> > RCU nest depth: 1, expected: 0
> > 4 locks held by kworker/u9:2/5070:
> > #0: ffff888015be3948 ((wq_completion)hci0#2){+.+.}-{0:0}, at: process_one_work kernel/workqueue.c:3229 [inline]
> > #0: ffff888015be3948 ((wq_completion)hci0#2){+.+.}-{0:0}, at: process_scheduled_works+0x8e0/0x1770 kernel/workqueue.c:3335
> > #1: ffffc90003b6fd00 ((work_completion)(&hdev->rx_work)){+.+.}-{0:0}, at: process_one_work kernel/workqueue.c:3230 [inline]
> > #1: ffffc90003b6fd00 ((work_completion)(&hdev->rx_work)){+.+.}-{0:0}, at: process_scheduled_works+0x91b/0x1770 kernel/workqueue.c:3335
> > #2: ffff8880665d0078 (&hdev->lock){+.+.}-{3:3}, at: hci_le_create_big_complete_evt+0xcf/0xae0 net/bluetooth/hci_event.c:6914
> > #3: ffffffff8e132020 (rcu_read_lock){....}-{1:2}, at: rcu_lock_acquire include/linux/rcupdate.h:298 [inline]
> > #3: ffffffff8e132020 (rcu_read_lock){....}-{1:2}, at: rcu_read_lock include/linux/rcupdate.h:750 [inline]
> > #3: ffffffff8e132020 (rcu_read_lock){....}-{1:2}, at: hci_le_create_big_complete_evt+0xdb/0xae0 net/bluetooth/hci_event.c:6915
> > CPU: 0 PID: 5070 Comm: kworker/u9:2 Not tainted 6.8.0-syzkaller-08073-g480e035fc4c7 #0
> > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 03/27/2024
> > Workqueue: hci0 hci_rx_work
> > Call Trace:
> > <TASK>
> > __dump_stack lib/dump_stack.c:88 [inline]
> > dump_stack_lvl+0x241/0x360 lib/dump_stack.c:114
> > __might_resched+0x5d4/0x780 kernel/sched/core.c:10187
> > __mutex_lock_common kernel/locking/mutex.c:585 [inline]
> > __mutex_lock+0xc1/0xd70 kernel/locking/mutex.c:752
> > hci_connect_cfm include/net/bluetooth/hci_core.h:2004 [inline]
> > hci_le_create_big_complete_evt+0x3d9/0xae0 net/bluetooth/hci_event.c:6939
> > hci_event_func net/bluetooth/hci_event.c:7514 [inline]
> > hci_event_packet+0xa53/0x1540 net/bluetooth/hci_event.c:7569
> > hci_rx_work+0x3e8/0xca0 net/bluetooth/hci_core.c:4171
> > process_one_work kernel/workqueue.c:3254 [inline]
> > process_scheduled_works+0xa00/0x1770 kernel/workqueue.c:3335
> > worker_thread+0x86d/0xd70 kernel/workqueue.c:3416
> > kthread+0x2f0/0x390 kernel/kthread.c:388
> > ret_from_fork+0x4b/0x80 arch/x86/kernel/process.c:147
> > ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:243
> > </TASK>
> >
> > Reported-by: syzbot+2fb0835e0c9cefc34614@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Closes: https://syzkaller.appspot.com/bug?extid=2fb0835e0c9cefc34614
> > Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@xxxxxxxxx>
> > ---
> > include/net/bluetooth/hci_core.h | 89 ++++++++++++++++++++++----------
> > net/bluetooth/hci_core.c | 9 ++--
> > 2 files changed, 65 insertions(+), 33 deletions(-)
> >
> > diff --git a/include/net/bluetooth/hci_core.h b/include/net/bluetooth/hci_core.h
> > index ea798f07c5a2..95f11f04e24a 100644
> > --- a/include/net/bluetooth/hci_core.h
> > +++ b/include/net/bluetooth/hci_core.h
> > @@ -804,7 +804,6 @@ struct hci_conn_params {
> > extern struct list_head hci_dev_list;
> > extern struct list_head hci_cb_list;
> > extern rwlock_t hci_dev_list_lock;
> > -extern struct mutex hci_cb_list_lock;
> >
> > #define hci_dev_set_flag(hdev, nr) set_bit((nr), (hdev)->dev_flags)
> > #define hci_dev_clear_flag(hdev, nr) clear_bit((nr), (hdev)->dev_flags)
> > @@ -2029,12 +2028,18 @@ static inline void hci_connect_cfm(struct hci_conn *conn, __u8 status)
> > {
> > struct hci_cb *cb;
> >
> > - mutex_lock(&hci_cb_list_lock);
> > - list_for_each_entry(cb, &hci_cb_list, list) {
> > - if (cb->connect_cfm)
> > - cb->connect_cfm(conn, status);
> > + rcu_read_lock();
> > + list_for_each_entry_rcu(cb, &hci_cb_list, list) {
> > + if (cb->connect_cfm) {
> > + struct hci_cb cpy = *cb;
> > +
> > + /* Callback may block so release RCU read lock */
> > + rcu_read_unlock();
> > + cpy.connect_cfm(conn, status);
> > + rcu_read_lock();
>
> This looks like incorrect RCU usage
>
> [CPU 1] rcu_read_unlock()
> [CPU 2] hci_unregister_cb(cb)
> [CPU 2] hci_unregister_cb(next cb)
> [CPU 1] rcu_read_lock()
> [CPU 1] list_for_each_entry_rcu -> iterates to "next cb" not in list
>
> If all hci_cb weren't static, it'd also UAF (maybe it is for rfcomm?).
>
>
> hci_le_create_big_complete_evt() also does this (and maybe crashes if
> ev->status != 0 ?), so maybe it is simples to fix it.
I guess you are right, it seems incorrect to have
rcu_read_unlock/relock inside list_for_each_entry_rcu, that said I
wonder why the entry is not accessed via rcu_dereference, anyway the
only alternative I can think of is to copy to a local stack list which
we can then run with list_for_each_entry_safe, or we create some
mechanism to differ actions that otherwise could block/sleep while
holding rcu_read_lock.
>
>
> > + }
> > }
> > - mutex_unlock(&hci_cb_list_lock);
> > + rcu_read_unlock();
> >
> > if (conn->connect_cfm_cb)
> > conn->connect_cfm_cb(conn, status);
> > @@ -2044,12 +2049,18 @@ static inline void hci_disconn_cfm(struct hci_conn *conn, __u8 reason)
> > {
> > struct hci_cb *cb;
> >
> > - mutex_lock(&hci_cb_list_lock);
> > + rcu_read_lock();
> > list_for_each_entry(cb, &hci_cb_list, list) {
> > - if (cb->disconn_cfm)
> > - cb->disconn_cfm(conn, reason);
> > + if (cb->disconn_cfm) {
> > + struct hci_cb cpy = *cb;
> > +
> > + /* Callback may block so release RCU read lock */
> > + rcu_read_unlock();
> > + cpy.disconn_cfm(conn, reason);
> > + rcu_read_lock();
> > + }
> > }
> > - mutex_unlock(&hci_cb_list_lock);
> > + rcu_read_unlock();
> >
> > if (conn->disconn_cfm_cb)
> > conn->disconn_cfm_cb(conn, reason);
> > @@ -2065,12 +2076,18 @@ static inline void hci_auth_cfm(struct hci_conn *conn, __u8 status)
> >
> > encrypt = test_bit(HCI_CONN_ENCRYPT, &conn->flags) ? 0x01 : 0x00;
> >
> > - mutex_lock(&hci_cb_list_lock);
> > + rcu_read_lock();
> > list_for_each_entry(cb, &hci_cb_list, list) {
> > - if (cb->security_cfm)
> > - cb->security_cfm(conn, status, encrypt);
> > + if (cb->security_cfm) {
> > + struct hci_cb cpy = *cb;
> > +
> > + /* Callback may block so release RCU read lock */
> > + rcu_read_unlock();
> > + cpy.security_cfm(conn, status, encrypt);
> > + rcu_read_lock();
> > + }
> > }
> > - mutex_unlock(&hci_cb_list_lock);
> > + rcu_read_unlock();
> >
> > if (conn->security_cfm_cb)
> > conn->security_cfm_cb(conn, status);
> > @@ -2105,12 +2122,18 @@ static inline void hci_encrypt_cfm(struct hci_conn *conn, __u8 status)
> > conn->sec_level = conn->pending_sec_level;
> > }
> >
> > - mutex_lock(&hci_cb_list_lock);
> > + rcu_read_lock();
> > list_for_each_entry(cb, &hci_cb_list, list) {
> > - if (cb->security_cfm)
> > - cb->security_cfm(conn, status, encrypt);
> > + if (cb->security_cfm) {
> > + struct hci_cb cpy = *cb;
> > +
> > + /* Callback may block so release RCU read lock */
> > + rcu_read_unlock();
> > + cpy.security_cfm(conn, status, encrypt);
> > + rcu_read_lock();
> > + }
> > }
> > - mutex_unlock(&hci_cb_list_lock);
> > + rcu_read_unlock();
> >
> > if (conn->security_cfm_cb)
> > conn->security_cfm_cb(conn, status);
> > @@ -2120,12 +2143,18 @@ static inline void hci_key_change_cfm(struct hci_conn *conn, __u8 status)
> > {
> > struct hci_cb *cb;
> >
> > - mutex_lock(&hci_cb_list_lock);
> > + rcu_read_lock();
> > list_for_each_entry(cb, &hci_cb_list, list) {
> > - if (cb->key_change_cfm)
> > - cb->key_change_cfm(conn, status);
> > + if (cb->key_change_cfm) {
> > + struct hci_cb cpy = *cb;
> > +
> > + /* Callback may block so release RCU read lock */
> > + rcu_read_unlock();
> > + cpy.key_change_cfm(conn, status);
> > + rcu_read_lock();
> > + }
> > }
> > - mutex_unlock(&hci_cb_list_lock);
> > + rcu_read_unlock();
> > }
> >
> > static inline void hci_role_switch_cfm(struct hci_conn *conn, __u8 status,
> > @@ -2133,12 +2162,18 @@ static inline void hci_role_switch_cfm(struct hci_conn *conn, __u8 status,
> > {
> > struct hci_cb *cb;
> >
> > - mutex_lock(&hci_cb_list_lock);
> > - list_for_each_entry(cb, &hci_cb_list, list) {
> > - if (cb->role_switch_cfm)
> > - cb->role_switch_cfm(conn, status, role);
> > + rcu_read_lock();
> > + list_for_each_entry_rcu(cb, &hci_cb_list, list) {
> > + if (cb->role_switch_cfm) {
> > + struct hci_cb cpy = *cb;
> > +
> > + /* Callback may block so release RCU read lock */
> > + rcu_read_unlock();
> > + cpy.role_switch_cfm(conn, status, role);
> > + rcu_read_lock();
> > + }
> > }
> > - mutex_unlock(&hci_cb_list_lock);
> > + rcu_read_unlock();
> > }
> >
> > static inline bool hci_bdaddr_is_rpa(bdaddr_t *bdaddr, u8 addr_type)
> > diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
> > index f9e19f9cb5a3..25d180d225c1 100644
> > --- a/net/bluetooth/hci_core.c
> > +++ b/net/bluetooth/hci_core.c
> > @@ -2993,9 +2993,7 @@ int hci_register_cb(struct hci_cb *cb)
> > {
> > BT_DBG("%p name %s", cb, cb->name);
> >
> > - mutex_lock(&hci_cb_list_lock);
> > - list_add_tail(&cb->list, &hci_cb_list);
> > - mutex_unlock(&hci_cb_list_lock);
> > + list_add_tail_rcu(&cb->list, &hci_cb_list);
> >
> > return 0;
> > }
> > @@ -3005,9 +3003,8 @@ int hci_unregister_cb(struct hci_cb *cb)
> > {
> > BT_DBG("%p name %s", cb, cb->name);
> >
> > - mutex_lock(&hci_cb_list_lock);
> > - list_del(&cb->list);
> > - mutex_unlock(&hci_cb_list_lock);
> > + list_del_rcu(&cb->list);
> > + synchronize_rcu();
> >
> > return 0;
> > }
>
--
Luiz Augusto von Dentz
