In tb_cfg_read_raw, req is allocated by tb_cfg_request_alloc()
with an initial reference. Before calling tb_cfg_request_sync(),
there is no refcount inc operation. tb_cfg_request_sync()
calls tb_cfg_request(..,req,..) and if the callee failed,
the initial reference of req is dropped and req is freed.
Later in tb_cfg_read_raw before the err check,
tb_cfg_request_put(req) is called again. It may cause error
in race.
My patch puts tb_cfg_request_put(req) after the err check
finished to avoid unexpected result.
Signed-off-by: Lv Yunlong <[email protected]>
---
drivers/thunderbolt/ctl.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/thunderbolt/ctl.c b/drivers/thunderbolt/ctl.c
index f1aeaff9f368..bb60269c89ab 100644
--- a/drivers/thunderbolt/ctl.c
+++ b/drivers/thunderbolt/ctl.c
@@ -890,11 +890,11 @@ struct tb_cfg_result tb_cfg_read_raw(struct tb_ctl *ctl, void *buffer,
res = tb_cfg_request_sync(ctl, req, timeout_msec);
- tb_cfg_request_put(req);
-
if (res.err != -ETIMEDOUT)
break;
+ tb_cfg_request_put(req);
+
/* Wait a bit (arbitrary time) until we send a retry */
usleep_range(10, 100);
}
--
2.25.1
Hi,
On Mon, Mar 22, 2021 at 08:15:12PM -0700, Lv Yunlong wrote:
> In tb_cfg_read_raw, req is allocated by tb_cfg_request_alloc()
> with an initial reference. Before calling tb_cfg_request_sync(),
> there is no refcount inc operation. tb_cfg_request_sync()
> calls tb_cfg_request(..,req,..) and if the callee failed,
> the initial reference of req is dropped and req is freed.
>
> Later in tb_cfg_read_raw before the err check,
> tb_cfg_request_put(req) is called again. It may cause error
> in race.
Hmm, tb_cfg_request() does tb_cfg_request_get() too and in case of error
it does tb_cfg_request_put(). So the refcount should be fine. What am I
missing?
>
> My patch puts tb_cfg_request_put(req) after the err check
> finished to avoid unexpected result.
>
> Signed-off-by: Lv Yunlong <[email protected]>
> ---
> drivers/thunderbolt/ctl.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/thunderbolt/ctl.c b/drivers/thunderbolt/ctl.c
> index f1aeaff9f368..bb60269c89ab 100644
> --- a/drivers/thunderbolt/ctl.c
> +++ b/drivers/thunderbolt/ctl.c
> @@ -890,11 +890,11 @@ struct tb_cfg_result tb_cfg_read_raw(struct tb_ctl *ctl, void *buffer,
>
> res = tb_cfg_request_sync(ctl, req, timeout_msec);
>
> - tb_cfg_request_put(req);
> -
> if (res.err != -ETIMEDOUT)
> break;
>
> + tb_cfg_request_put(req);
> +
> /* Wait a bit (arbitrary time) until we send a retry */
> usleep_range(10, 100);
> }
> --
> 2.25.1
>
On Tue, Mar 23, 2021 at 10:30:16PM +0800, [email protected] wrote:
>
>
>
> > -----原始邮件-----
> > 发件人: "Mika Westerberg" <[email protected]>
> > 发送时间: 2021-03-23 22:06:47 (星期二)
> > 收件人: "Lv Yunlong" <[email protected]>
> > 抄送: [email protected], [email protected], [email protected], [email protected], [email protected]
> > 主题: Re: [PATCH] thunderbolt: Fix a double put in tb_cfg_read_raw
> >
> > Hi,
> >
> > On Mon, Mar 22, 2021 at 08:15:12PM -0700, Lv Yunlong wrote:
> > > In tb_cfg_read_raw, req is allocated by tb_cfg_request_alloc()
> > > with an initial reference. Before calling tb_cfg_request_sync(),
> > > there is no refcount inc operation. tb_cfg_request_sync()
> > > calls tb_cfg_request(..,req,..) and if the callee failed,
> > > the initial reference of req is dropped and req is freed.
> > >
> > > Later in tb_cfg_read_raw before the err check,
> > > tb_cfg_request_put(req) is called again. It may cause error
> > > in race.
> >
> > Hmm, tb_cfg_request() does tb_cfg_request_get() too and in case of error
> > it does tb_cfg_request_put(). So the refcount should be fine. What am I
> > missing?
> >
> > >
> > > My patch puts tb_cfg_request_put(req) after the err check
> > > finished to avoid unexpected result.
> > >
> > > Signed-off-by: Lv Yunlong <[email protected]>
> > > ---
> > > drivers/thunderbolt/ctl.c | 4 ++--
> > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/thunderbolt/ctl.c b/drivers/thunderbolt/ctl.c
> > > index f1aeaff9f368..bb60269c89ab 100644
> > > --- a/drivers/thunderbolt/ctl.c
> > > +++ b/drivers/thunderbolt/ctl.c
> > > @@ -890,11 +890,11 @@ struct tb_cfg_result tb_cfg_read_raw(struct tb_ctl *ctl, void *buffer,
> > >
> > > res = tb_cfg_request_sync(ctl, req, timeout_msec);
> > >
> > > - tb_cfg_request_put(req);
> > > -
> > > if (res.err != -ETIMEDOUT)
> > > break;
> > >
> > > + tb_cfg_request_put(req);
> > > +
> > > /* Wait a bit (arbitrary time) until we send a retry */
> > > usleep_range(10, 100);
> > > }
> > > --
> > > 2.25.1
> > >
>
> I'm very sorry, i was ashamed that i had missed the tb_cfg_request_get() in tb_cfg_request().
It happens, no worries :)