2019-12-19 03:21:40

by Jia-Ju Bai

Subject: [PATCH v2] mtd: maps: pcmciamtd: fix possible sleep-in-atomic-context bugs in pcmciamtd_set_vpp()

The driver may sleep while holding a spinlock.
The function call path (from bottom to top) in Linux 4.19 is:

drivers/pcmcia/pcmcia_resource.c, 312:
mutex_lock in pcmcia_fixup_vpp
drivers/mtd/maps/pcmciamtd.c, 309:
pcmcia_fixup_vpp in pcmciamtd_set_vpp
drivers/mtd/maps/pcmciamtd.c, 306:
_raw_spin_lock_irqsave in pcmciamtd_set_vpp

drivers/pcmcia/pcmcia_resource.c, 312:
mutex_lock in pcmcia_fixup_vpp
drivers/mtd/maps/pcmciamtd.c, 312:
pcmcia_fixup_vpp in pcmciamtd_set_vpp
drivers/mtd/maps/pcmciamtd.c, 306:
_raw_spin_lock_irqsave in pcmciamtd_set_vp

mutex_lock() may sleep at runtime.

To fix these bugs, the spinlock is replaced with a mutex.

These bugs are found by a static analysis tool STCheck written by
myself.

Signed-off-by: Jia-Ju Bai <[email protected]>
---
v2:
* Change the spinlock to a mutex.
Thank Dominik for good advice.

---

drivers/mtd/maps/pcmciamtd.c | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/mtd/maps/pcmciamtd.c b/drivers/mtd/maps/pcmciamtd.c
index 70bb403f69f7..2ac79e1cedd9 100644
--- a/drivers/mtd/maps/pcmciamtd.c
+++ b/drivers/mtd/maps/pcmciamtd.c
@@ -294,16 +294,15 @@ static void pcmcia_copy_to(struct map_info *map, unsigned long to, const void *f
}


-static DEFINE_SPINLOCK(pcmcia_vpp_lock);
+static DEFINE_MUTEX(pcmcia_vpp_lock);
static int pcmcia_vpp_refcnt;
static void pcmciamtd_set_vpp(struct map_info *map, int on)
{
struct pcmciamtd_dev *dev = (struct pcmciamtd_dev *)map->map_priv_1;
struct pcmcia_device *link = dev->p_dev;
- unsigned long flags;

pr_debug("dev = %p on = %d vpp = %d\n\n", dev, on, dev->vpp);
- spin_lock_irqsave(&pcmcia_vpp_lock, flags);
+ mutex_lock(&pcmcia_vpp_lock);
if (on) {
if (++pcmcia_vpp_refcnt == 1) /* first nested 'on' */
pcmcia_fixup_vpp(link, dev->vpp);
@@ -311,7 +310,7 @@ static void pcmciamtd_set_vpp(struct map_info *map, int on)
if (--pcmcia_vpp_refcnt == 0) /* last nested 'off' */
pcmcia_fixup_vpp(link, 0);
}
- spin_unlock_irqrestore(&pcmcia_vpp_lock, flags);
+ mutex_unlock(&pcmcia_vpp_lock);
}


--
2.17.1

2019-12-19 06:48:28

by Dominik Brodowski

Subject: Re: [PATCH v2] mtd: maps: pcmciamtd: fix possible sleep-in-atomic-context bugs in pcmciamtd_set_vpp()

On Thu, Dec 19, 2019 at 11:20:23AM +0800, Jia-Ju Bai wrote:
> The driver may sleep while holding a spinlock.
> The function call path (from bottom to top) in Linux 4.19 is:
>
> drivers/pcmcia/pcmcia_resource.c, 312:
> mutex_lock in pcmcia_fixup_vpp
> drivers/mtd/maps/pcmciamtd.c, 309:
> pcmcia_fixup_vpp in pcmciamtd_set_vpp
> drivers/mtd/maps/pcmciamtd.c, 306:
> _raw_spin_lock_irqsave in pcmciamtd_set_vpp
>
> drivers/pcmcia/pcmcia_resource.c, 312:
> mutex_lock in pcmcia_fixup_vpp
> drivers/mtd/maps/pcmciamtd.c, 312:
> pcmcia_fixup_vpp in pcmciamtd_set_vpp
> drivers/mtd/maps/pcmciamtd.c, 306:
> _raw_spin_lock_irqsave in pcmciamtd_set_vp
>
> mutex_lock() may sleep at runtime.
>
> To fix these bugs, the spinlock is replaced with a mutex.
>
> These bugs are found by a static analysis tool STCheck written by
> myself.
>
> Signed-off-by: Jia-Ju Bai <[email protected]>
Reviewed-by: Dominik Brodowski <[email protected]>

I presume this will go upstream (with CC to stable -- likely applies to all
longterm kernels still maintained) via MTD? Or should I route it via pcmcia?

Thanks,
Dominik