Re: [PATCH v4] virtio-blk: handle block_device_operations callbacks after hot unplug

2020-05-04 Thread Michael S. Tsirkin
On Mon, May 04, 2020 at 01:48:34PM +0200, Cornelia Huck wrote:
> On Thu, 30 Apr 2020 15:04:42 +0100
> Stefan Hajnoczi  wrote:
> 
> > A userspace process holding a file descriptor to a virtio_blk device can
> > still invoke block_device_operations after hot unplug.  This leads to a
> > use-after-free accessing vblk->vdev in virtblk_getgeo() when
> > ioctl(HDIO_GETGEO) is invoked:
> > 
> >   BUG: unable to handle kernel NULL pointer dereference at 0090
> >   IP: [] virtio_check_driver_offered_feature+0x10/0x90 
> > [virtio]
> >   PGD 80003a92f067 PUD 3a930067 PMD 0
> >   Oops:  [#1] SMP
> >   CPU: 0 PID: 1310 Comm: hdio-getgeo Tainted: G   OE    
> >  3.10.0-1062.el7.x86_64 #1
> >   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 
> > rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
> >   task: 9be5fbfb8000 ti: 9be5fa89 task.ti: 9be5fa89
> >   RIP: 0010:[]  [] 
> > virtio_check_driver_offered_feature+0x10/0x90 [virtio]
> >   RSP: 0018:9be5fa893dc8  EFLAGS: 00010246
> >   RAX: 9be5fc3f3400 RBX: 9be5fa893e30 RCX: 
> >   RDX:  RSI: 0004 RDI: 9be5fbc10b40
> >   RBP: 9be5fa893dc8 R08: 0301 R09: 0301
> >   R10:  R11:  R12: 9be5fdc24680
> >   R13: 9be5fbc10b40 R14: 9be5fbc10480 R15: 
> >   FS:  7f1bfb968740() GS:9be5ffc0() 
> > knlGS:
> >   CS:  0010 DS:  ES:  CR0: 80050033
> >   CR2: 0090 CR3: 3a894000 CR4: 00360ff0
> >   DR0:  DR1:  DR2: 
> >   DR3:  DR6: fffe0ff0 DR7: 0400
> >   Call Trace:
> >[] virtblk_getgeo+0x47/0x110 [virtio_blk]
> >[] ? handle_mm_fault+0x39d/0x9b0
> >[] blkdev_ioctl+0x1f5/0xa20
> >[] block_ioctl+0x41/0x50
> >[] do_vfs_ioctl+0x3a0/0x5a0
> >[] SyS_ioctl+0xa1/0xc0
> > 
> > A related problem is that virtblk_remove() leaks the vd_index_ida index
> > when something still holds a reference to vblk->disk during hot unplug.
> > This causes virtio-blk device names to be lost (vda, vdb, etc).
> > 
> > Fix these issues by protecting vblk->vdev with a mutex and reference
> > counting vblk so the vd_index_ida index can be removed in all cases.
> > 
> > Fixes: 48e4043d4529523cbc7fa8dd745bd8e2c45ce1d3
> >("virtio: add virtio disk geometry feature")
> 
> Should be
> 
> Fixes: 48e4043d4529 ("virtio: add virtio disk geometry feature")


Yes - it was reported on linux-next so I fixed it up when applying.

> > Reported-by: Lance Digby 
> > Signed-off-by: Stefan Hajnoczi 
> > ---
> > v4:
> >  * Clarify vdev_mutex usage [Stefano and Michael]
> > 
> >  drivers/block/virtio_blk.c | 86 ++
> >  1 file changed, 78 insertions(+), 8 deletions(-)
> > 
> > diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> > index 93468b7c6701..9d21bf0f155e 100644
> > --- a/drivers/block/virtio_blk.c
> > +++ b/drivers/block/virtio_blk.c
> > @@ -33,6 +33,15 @@ struct virtio_blk_vq {
> >  } cacheline_aligned_in_smp;
> >  
> >  struct virtio_blk {
> > +   /*
> > +* This mutex must be held by anything that may run after
> > +* virtblk_remove() sets vblk->vdev to NULL.
> > +*
> > +* blk-mq, virtqueue processing, and sysfs attribute code paths are
> > +* shut down before vblk->vdev is set to NULL and therefore do not need
> > +* to hold this mutex.
> > +*/
> > +   struct mutex vdev_mutex;
> > struct virtio_device *vdev;
> >  
> > /* The disk structure for the kernel. */
> > @@ -44,6 +53,13 @@ struct virtio_blk {
> > /* Process context for config space updates */
> > struct work_struct config_work;
> >  
> > +   /*
> > +* Tracks references from block_device_operations open/release and
> > +* virtio_driver probe/remove so this object can be freed once no
> > +* longer in use.
> > +*/
> > +   refcount_t refs;
> 
> Using a struct kref might be more idiomatic.
> 
> > +
> > /* What host tells us, plus 2 for header & tailer. */
> > unsigned int sg_elems;
> >  
> > @@ -295,10 +311,55 @@ static int virtblk_get_id(struct gendisk *disk, char 
> > *id_str)
> > return err;
> >  }
> >  
> > +static void virtblk_get(struct virtio_blk *vblk)
> > +{
> > +   refcount_inc(>refs);
> 
> Should the code even be able to grab a ref if !vblk->vdev?
> 
> > +}
> > +
> > +static void virtblk_put(struct virtio_blk *vblk)
> > +{
> > +   if (refcount_dec_and_test(>refs)) {
> > +   ida_simple_remove(_index_ida, vblk->index);
> > +   mutex_destroy(>vdev_mutex);
> > +   kfree(vblk);
> 
> I think that's where putting these cleanups into a release() funtion
> would be more idiomatic.
> 
> > +   }
> > +}
> 
> (...)
> 
> Looks sane to me.



Re: [PATCH v4] virtio-blk: handle block_device_operations callbacks after hot unplug

2020-05-04 Thread Cornelia Huck
On Thu, 30 Apr 2020 15:04:42 +0100
Stefan Hajnoczi  wrote:

> A userspace process holding a file descriptor to a virtio_blk device can
> still invoke block_device_operations after hot unplug.  This leads to a
> use-after-free accessing vblk->vdev in virtblk_getgeo() when
> ioctl(HDIO_GETGEO) is invoked:
> 
>   BUG: unable to handle kernel NULL pointer dereference at 0090
>   IP: [] virtio_check_driver_offered_feature+0x10/0x90 
> [virtio]
>   PGD 80003a92f067 PUD 3a930067 PMD 0
>   Oops:  [#1] SMP
>   CPU: 0 PID: 1310 Comm: hdio-getgeo Tainted: G   OE     
> 3.10.0-1062.el7.x86_64 #1
>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 
> rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
>   task: 9be5fbfb8000 ti: 9be5fa89 task.ti: 9be5fa89
>   RIP: 0010:[]  [] 
> virtio_check_driver_offered_feature+0x10/0x90 [virtio]
>   RSP: 0018:9be5fa893dc8  EFLAGS: 00010246
>   RAX: 9be5fc3f3400 RBX: 9be5fa893e30 RCX: 
>   RDX:  RSI: 0004 RDI: 9be5fbc10b40
>   RBP: 9be5fa893dc8 R08: 0301 R09: 0301
>   R10:  R11:  R12: 9be5fdc24680
>   R13: 9be5fbc10b40 R14: 9be5fbc10480 R15: 
>   FS:  7f1bfb968740() GS:9be5ffc0() knlGS:
>   CS:  0010 DS:  ES:  CR0: 80050033
>   CR2: 0090 CR3: 3a894000 CR4: 00360ff0
>   DR0:  DR1:  DR2: 
>   DR3:  DR6: fffe0ff0 DR7: 0400
>   Call Trace:
>[] virtblk_getgeo+0x47/0x110 [virtio_blk]
>[] ? handle_mm_fault+0x39d/0x9b0
>[] blkdev_ioctl+0x1f5/0xa20
>[] block_ioctl+0x41/0x50
>[] do_vfs_ioctl+0x3a0/0x5a0
>[] SyS_ioctl+0xa1/0xc0
> 
> A related problem is that virtblk_remove() leaks the vd_index_ida index
> when something still holds a reference to vblk->disk during hot unplug.
> This causes virtio-blk device names to be lost (vda, vdb, etc).
> 
> Fix these issues by protecting vblk->vdev with a mutex and reference
> counting vblk so the vd_index_ida index can be removed in all cases.
> 
> Fixes: 48e4043d4529523cbc7fa8dd745bd8e2c45ce1d3
>("virtio: add virtio disk geometry feature")

Should be

Fixes: 48e4043d4529 ("virtio: add virtio disk geometry feature")

> Reported-by: Lance Digby 
> Signed-off-by: Stefan Hajnoczi 
> ---
> v4:
>  * Clarify vdev_mutex usage [Stefano and Michael]
> 
>  drivers/block/virtio_blk.c | 86 ++
>  1 file changed, 78 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 93468b7c6701..9d21bf0f155e 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -33,6 +33,15 @@ struct virtio_blk_vq {
>  } cacheline_aligned_in_smp;
>  
>  struct virtio_blk {
> + /*
> +  * This mutex must be held by anything that may run after
> +  * virtblk_remove() sets vblk->vdev to NULL.
> +  *
> +  * blk-mq, virtqueue processing, and sysfs attribute code paths are
> +  * shut down before vblk->vdev is set to NULL and therefore do not need
> +  * to hold this mutex.
> +  */
> + struct mutex vdev_mutex;
>   struct virtio_device *vdev;
>  
>   /* The disk structure for the kernel. */
> @@ -44,6 +53,13 @@ struct virtio_blk {
>   /* Process context for config space updates */
>   struct work_struct config_work;
>  
> + /*
> +  * Tracks references from block_device_operations open/release and
> +  * virtio_driver probe/remove so this object can be freed once no
> +  * longer in use.
> +  */
> + refcount_t refs;

Using a struct kref might be more idiomatic.

> +
>   /* What host tells us, plus 2 for header & tailer. */
>   unsigned int sg_elems;
>  
> @@ -295,10 +311,55 @@ static int virtblk_get_id(struct gendisk *disk, char 
> *id_str)
>   return err;
>  }
>  
> +static void virtblk_get(struct virtio_blk *vblk)
> +{
> + refcount_inc(>refs);

Should the code even be able to grab a ref if !vblk->vdev?

> +}
> +
> +static void virtblk_put(struct virtio_blk *vblk)
> +{
> + if (refcount_dec_and_test(>refs)) {
> + ida_simple_remove(_index_ida, vblk->index);
> + mutex_destroy(>vdev_mutex);
> + kfree(vblk);

I think that's where putting these cleanups into a release() funtion
would be more idiomatic.

> + }
> +}

(...)

Looks sane to me.



Re: [PATCH v4] virtio-blk: handle block_device_operations callbacks after hot unplug

2020-04-30 Thread Stefano Garzarella
On Thu, Apr 30, 2020 at 03:04:42PM +0100, Stefan Hajnoczi wrote:
> A userspace process holding a file descriptor to a virtio_blk device can
> still invoke block_device_operations after hot unplug.  This leads to a
> use-after-free accessing vblk->vdev in virtblk_getgeo() when
> ioctl(HDIO_GETGEO) is invoked:
> 
>   BUG: unable to handle kernel NULL pointer dereference at 0090
>   IP: [] virtio_check_driver_offered_feature+0x10/0x90 
> [virtio]
>   PGD 80003a92f067 PUD 3a930067 PMD 0
>   Oops:  [#1] SMP
>   CPU: 0 PID: 1310 Comm: hdio-getgeo Tainted: G   OE     
> 3.10.0-1062.el7.x86_64 #1
>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 
> rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
>   task: 9be5fbfb8000 ti: 9be5fa89 task.ti: 9be5fa89
>   RIP: 0010:[]  [] 
> virtio_check_driver_offered_feature+0x10/0x90 [virtio]
>   RSP: 0018:9be5fa893dc8  EFLAGS: 00010246
>   RAX: 9be5fc3f3400 RBX: 9be5fa893e30 RCX: 
>   RDX:  RSI: 0004 RDI: 9be5fbc10b40
>   RBP: 9be5fa893dc8 R08: 0301 R09: 0301
>   R10:  R11:  R12: 9be5fdc24680
>   R13: 9be5fbc10b40 R14: 9be5fbc10480 R15: 
>   FS:  7f1bfb968740() GS:9be5ffc0() knlGS:
>   CS:  0010 DS:  ES:  CR0: 80050033
>   CR2: 0090 CR3: 3a894000 CR4: 00360ff0
>   DR0:  DR1:  DR2: 
>   DR3:  DR6: fffe0ff0 DR7: 0400
>   Call Trace:
>[] virtblk_getgeo+0x47/0x110 [virtio_blk]
>[] ? handle_mm_fault+0x39d/0x9b0
>[] blkdev_ioctl+0x1f5/0xa20
>[] block_ioctl+0x41/0x50
>[] do_vfs_ioctl+0x3a0/0x5a0
>[] SyS_ioctl+0xa1/0xc0
> 
> A related problem is that virtblk_remove() leaks the vd_index_ida index
> when something still holds a reference to vblk->disk during hot unplug.
> This causes virtio-blk device names to be lost (vda, vdb, etc).
> 
> Fix these issues by protecting vblk->vdev with a mutex and reference
> counting vblk so the vd_index_ida index can be removed in all cases.
> 
> Fixes: 48e4043d4529523cbc7fa8dd745bd8e2c45ce1d3
>("virtio: add virtio disk geometry feature")
> Reported-by: Lance Digby 
> Signed-off-by: Stefan Hajnoczi 
> ---
> v4:
>  * Clarify vdev_mutex usage [Stefano and Michael]
> 
>  drivers/block/virtio_blk.c | 86 ++
>  1 file changed, 78 insertions(+), 8 deletions(-)

Reviewed-by: Stefano Garzarella 

Thanks,
Stefano

> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 93468b7c6701..9d21bf0f155e 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -33,6 +33,15 @@ struct virtio_blk_vq {
>  } cacheline_aligned_in_smp;
>  
>  struct virtio_blk {
> + /*
> +  * This mutex must be held by anything that may run after
> +  * virtblk_remove() sets vblk->vdev to NULL.
> +  *
> +  * blk-mq, virtqueue processing, and sysfs attribute code paths are
> +  * shut down before vblk->vdev is set to NULL and therefore do not need
> +  * to hold this mutex.
> +  */
> + struct mutex vdev_mutex;
>   struct virtio_device *vdev;
>  
>   /* The disk structure for the kernel. */
> @@ -44,6 +53,13 @@ struct virtio_blk {
>   /* Process context for config space updates */
>   struct work_struct config_work;
>  
> + /*
> +  * Tracks references from block_device_operations open/release and
> +  * virtio_driver probe/remove so this object can be freed once no
> +  * longer in use.
> +  */
> + refcount_t refs;
> +
>   /* What host tells us, plus 2 for header & tailer. */
>   unsigned int sg_elems;
>  
> @@ -295,10 +311,55 @@ static int virtblk_get_id(struct gendisk *disk, char 
> *id_str)
>   return err;
>  }
>  
> +static void virtblk_get(struct virtio_blk *vblk)
> +{
> + refcount_inc(>refs);
> +}
> +
> +static void virtblk_put(struct virtio_blk *vblk)
> +{
> + if (refcount_dec_and_test(>refs)) {
> + ida_simple_remove(_index_ida, vblk->index);
> + mutex_destroy(>vdev_mutex);
> + kfree(vblk);
> + }
> +}
> +
> +static int virtblk_open(struct block_device *bd, fmode_t mode)
> +{
> + struct virtio_blk *vblk = bd->bd_disk->private_data;
> + int ret = 0;
> +
> + mutex_lock(>vdev_mutex);
> +
> + if (vblk->vdev)
> + virtblk_get(vblk);
> + else
> + ret = -ENXIO;
> +
> + mutex_unlock(>vdev_mutex);
> + return ret;
> +}
> +
> +static void virtblk_release(struct gendisk *disk, fmode_t mode)
> +{
> + struct virtio_blk *vblk = disk->private_data;
> +
> + virtblk_put(vblk);
> +}
> +
>  /* We provide getgeo only to please some old bootloader/partitioning tools */
>  static int virtblk_getgeo(struct 

[PATCH v4] virtio-blk: handle block_device_operations callbacks after hot unplug

2020-04-30 Thread Stefan Hajnoczi
A userspace process holding a file descriptor to a virtio_blk device can
still invoke block_device_operations after hot unplug.  This leads to a
use-after-free accessing vblk->vdev in virtblk_getgeo() when
ioctl(HDIO_GETGEO) is invoked:

  BUG: unable to handle kernel NULL pointer dereference at 0090
  IP: [] virtio_check_driver_offered_feature+0x10/0x90 
[virtio]
  PGD 80003a92f067 PUD 3a930067 PMD 0
  Oops:  [#1] SMP
  CPU: 0 PID: 1310 Comm: hdio-getgeo Tainted: G   OE     
3.10.0-1062.el7.x86_64 #1
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 
rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
  task: 9be5fbfb8000 ti: 9be5fa89 task.ti: 9be5fa89
  RIP: 0010:[]  [] 
virtio_check_driver_offered_feature+0x10/0x90 [virtio]
  RSP: 0018:9be5fa893dc8  EFLAGS: 00010246
  RAX: 9be5fc3f3400 RBX: 9be5fa893e30 RCX: 
  RDX:  RSI: 0004 RDI: 9be5fbc10b40
  RBP: 9be5fa893dc8 R08: 0301 R09: 0301
  R10:  R11:  R12: 9be5fdc24680
  R13: 9be5fbc10b40 R14: 9be5fbc10480 R15: 
  FS:  7f1bfb968740() GS:9be5ffc0() knlGS:
  CS:  0010 DS:  ES:  CR0: 80050033
  CR2: 0090 CR3: 3a894000 CR4: 00360ff0
  DR0:  DR1:  DR2: 
  DR3:  DR6: fffe0ff0 DR7: 0400
  Call Trace:
   [] virtblk_getgeo+0x47/0x110 [virtio_blk]
   [] ? handle_mm_fault+0x39d/0x9b0
   [] blkdev_ioctl+0x1f5/0xa20
   [] block_ioctl+0x41/0x50
   [] do_vfs_ioctl+0x3a0/0x5a0
   [] SyS_ioctl+0xa1/0xc0

A related problem is that virtblk_remove() leaks the vd_index_ida index
when something still holds a reference to vblk->disk during hot unplug.
This causes virtio-blk device names to be lost (vda, vdb, etc).

Fix these issues by protecting vblk->vdev with a mutex and reference
counting vblk so the vd_index_ida index can be removed in all cases.

Fixes: 48e4043d4529523cbc7fa8dd745bd8e2c45ce1d3
   ("virtio: add virtio disk geometry feature")
Reported-by: Lance Digby 
Signed-off-by: Stefan Hajnoczi 
---
v4:
 * Clarify vdev_mutex usage [Stefano and Michael]

 drivers/block/virtio_blk.c | 86 ++
 1 file changed, 78 insertions(+), 8 deletions(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 93468b7c6701..9d21bf0f155e 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -33,6 +33,15 @@ struct virtio_blk_vq {
 } cacheline_aligned_in_smp;
 
 struct virtio_blk {
+   /*
+* This mutex must be held by anything that may run after
+* virtblk_remove() sets vblk->vdev to NULL.
+*
+* blk-mq, virtqueue processing, and sysfs attribute code paths are
+* shut down before vblk->vdev is set to NULL and therefore do not need
+* to hold this mutex.
+*/
+   struct mutex vdev_mutex;
struct virtio_device *vdev;
 
/* The disk structure for the kernel. */
@@ -44,6 +53,13 @@ struct virtio_blk {
/* Process context for config space updates */
struct work_struct config_work;
 
+   /*
+* Tracks references from block_device_operations open/release and
+* virtio_driver probe/remove so this object can be freed once no
+* longer in use.
+*/
+   refcount_t refs;
+
/* What host tells us, plus 2 for header & tailer. */
unsigned int sg_elems;
 
@@ -295,10 +311,55 @@ static int virtblk_get_id(struct gendisk *disk, char 
*id_str)
return err;
 }
 
+static void virtblk_get(struct virtio_blk *vblk)
+{
+   refcount_inc(>refs);
+}
+
+static void virtblk_put(struct virtio_blk *vblk)
+{
+   if (refcount_dec_and_test(>refs)) {
+   ida_simple_remove(_index_ida, vblk->index);
+   mutex_destroy(>vdev_mutex);
+   kfree(vblk);
+   }
+}
+
+static int virtblk_open(struct block_device *bd, fmode_t mode)
+{
+   struct virtio_blk *vblk = bd->bd_disk->private_data;
+   int ret = 0;
+
+   mutex_lock(>vdev_mutex);
+
+   if (vblk->vdev)
+   virtblk_get(vblk);
+   else
+   ret = -ENXIO;
+
+   mutex_unlock(>vdev_mutex);
+   return ret;
+}
+
+static void virtblk_release(struct gendisk *disk, fmode_t mode)
+{
+   struct virtio_blk *vblk = disk->private_data;
+
+   virtblk_put(vblk);
+}
+
 /* We provide getgeo only to please some old bootloader/partitioning tools */
 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 {
struct virtio_blk *vblk = bd->bd_disk->private_data;
+   int ret = 0;
+
+   mutex_lock(>vdev_mutex);
+
+   if (!vblk->vdev) {
+   ret = -ENXIO;
+   goto out;
+   }
 
/* see if the host passed in geometry config