Re: [RFC] Bluetooth: Change foreach_rcu to foreach_safe

[Ulisses Furquim <ulisses@profusion.mobi>]

Hi Andrei,

On Wed, Jan 25, 2012 at 11:10 AM, Emeltchenko Andrei
<Andrei.Emeltchenko.news@gmail.com> wrote:
> From: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
>
> When receiving l2cap packets and putting hciX down following
> trace can be seen without this patch. If there is only one element
> in RCU list after the element was deleted list_for_each_entry_rcu
> iterates again over some junk data.
>
> [   95.571834] Bluetooth: unknown link type 127
> [   95.578349] BUG: unable to handle kernel paging request at 20002000
> [   95.580236] IP: [<20002000>] 0x20001fff
> [   95.580763] *pde = 00000000
> [   95.581196] Oops: 0000 [#1] SMP
> ...
> [   95.582298] Pid: 3355, comm: hciconfig Tainted: G           O 3.2.0-rc2niko+ #62 innotek GmbH VirtualBox
> [   95.582298] EIP: 0060:[<20002000>] EFLAGS: 00210206 CPU: 0
> [   95.582298] EIP is at 0x20002000
> [   95.582298] EAX: ea4bc000 EBX: ea4bc000 ECX: 20002000 EDX: 00000016
> [   95.582298] ESI: f49516f8 EDI: f4951008 EBP: ea4f1e6c ESP: ea4f1e50
> [   95.582298]  DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
> [   95.582298] Process hciconfig (pid: 3355, ti=ea4f0000 task=ea4a8000 task.ti=ea4f0000)
> [   95.582298] Stack:
> [   95.582298]  f8231ab6 f825518d f8255178 0000007f ea4bc000 f4951000 f495163c ea4f1e98
> [   95.582298]  f822bcb1 f4951000 ea4f1e98 f822d679 0000000c f4951068 ea4a8000 f4951000
> [   95.582298]  ffffffed 00000000 ea4f1ea8 f822e1da 400448ca ea4f6800 ea4f1ee4 f824102f
> [   95.582298] Call Trace:
> [   95.582298]  [<f8231ab6>] ? hci_conn_hash_flush+0x76/0xf0 [bluetooth]
> [   95.582298]  [<f822bcb1>] hci_dev_do_close+0xc1/0x2e0 [bluetooth]
> [   95.582298]  [<f822d679>] ? hci_dev_get+0x69/0xb0 [bluetooth]
> [   95.582298]  [<f822e1da>] hci_dev_close+0x2a/0x50 [bluetooth]
> [   95.582298]  [<f824102f>] hci_sock_ioctl+0x1af/0x3f0 [bluetooth]
> [   95.582298]  [<c11153ea>] ? handle_pte_fault+0x8a/0x8f0
> [   95.582298]  [<c146becf>] sock_ioctl+0x5f/0x260
> [   95.582298]  [<c146be70>] ? sock_fasync+0x90/0x90
> [   95.582298]  [<c1152b33>] do_vfs_ioctl+0x83/0x5b0
> [   95.582298]  [<c1563f87>] ? do_page_fault+0x297/0x500
> [   95.582298]  [<c1563cf0>] ? spurious_fault+0xd0/0xd0
> [   95.582298]  [<c107165b>] ? up_read+0x1b/0x30
> [   95.582298]  [<c1563f87>] ? do_page_fault+0x297/0x500
> [   95.582298]  [<c100aa9f>] ? init_fpu+0xef/0x160
> [   95.582298]  [<c15617c0>] ? do_debug+0x180/0x180
> [   95.582298]  [<c100a958>] ? fpu_finit+0x28/0x80
> [   95.582298]  [<c11530e7>] sys_ioctl+0x87/0x90
> [   95.582298]  [<c156795f>] sysenter_do_call+0x12/0x38
> ...
>
> Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> ---
>  net/bluetooth/hci_conn.c |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/net/bluetooth/hci_conn.c b/net/bluetooth/hci_conn.c
> index 1de6efb..cdb6a3f 100644
> --- a/net/bluetooth/hci_conn.c
> +++ b/net/bluetooth/hci_conn.c
> @@ -802,11 +802,11 @@ timer:
>  void hci_conn_hash_flush(struct hci_dev *hdev)
>  {
>        struct hci_conn_hash *h = &hdev->conn_hash;
> -       struct hci_conn *c;
> +       struct hci_conn *c, *n;
>
>        BT_DBG("hdev %s", hdev->name);
>
> -       list_for_each_entry_rcu(c, &h->list, list) {
> +       list_for_each_entry_safe(c, n, &h->list, list) {
>                c->state = BT_CLOSED;
>
>                hci_proto_disconn_cfm(c, HCI_ERROR_LOCAL_HOST_TERM);

It looks correct to me. And I also checked that all updaters of this
RCU protected list grab hdev lock so we don't need any locks.

Regards,

-- 
Ulisses Furquim
ProFUSION embedded systems
http://profusion.mobi
Mobile: +55 19 9250 0942
Skype: ulissesffs