Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1707367 > unrolled thread

[RFC PATCH 0/6] bsg: fix regression resulting in panics when sending commands via BSG and some sanity cleanups

Back to article view | Back to linux.kernel

Contents

  [RFC PATCH 0/6] bsg: fix regression resulting in panics when sending commands via BSG and some sanity cleanups Benjamin Block <bblock@linux.vnet.ibm.com> - 2017-08-09 16:20 +0200
    [RFC PATCH 5/6] bsg: reduce unnecessary arguments for bsg_map_hdr() Benjamin Block <bblock@linux.vnet.ibm.com> - 2017-08-09 16:20 +0200
      Re: [RFC PATCH 5/6] bsg: reduce unnecessary arguments for  bsg_map_hdr() Johannes Thumshirn <jthumshirn@suse.de> - 2017-08-10 10:30 +0200
      Re: [RFC PATCH 5/6] bsg: reduce unnecessary arguments for         bsg_map_hdr() Christoph Hellwig <hch@lst.de> - 2017-08-10 11:40 +0200
        Re: [RFC PATCH 5/6] bsg: reduce unnecessary arguments for  bsg_map_hdr() Benjamin Block <bblock@linux.vnet.ibm.com> - 2017-08-11 00:30 +0200
    [RFC PATCH 3/6] bsg: scsi-transport: add compile-tests to prevent reply-buffer overflows Benjamin Block <bblock@linux.vnet.ibm.com> - 2017-08-09 16:20 +0200
      Re: [RFC PATCH 3/6] bsg: scsi-transport: add compile-tests to         prevent reply-buffer overflows Christoph Hellwig <hch@lst.de> - 2017-08-10 11:40 +0200

#1707367 — [RFC PATCH 0/6] bsg: fix regression resulting in panics when sending commands via BSG and some sanity cleanups

Hello all,

Steffen noticed recently that we have a regression in the BSG code that
prevents us from sending any traffic over this interface. After I
researched this a bit, it turned out that this affects not only zFCP, but
likely all LLDs that implements the BSG API. This was introduced in 4.11
(details in Patch 1 of this series).

I imagine the regression happened because of some very "unfortunate"
variable- namings. I can not fix them, as they involve the base struct
request, but I tried to add some cleanups that should make the
relationships between stuff more visible in the future I hope.

Patch 1   - Regression Fix; Also tagged for stable
Patch 2-6 - Cleanups

I tagged this as RFC. Patches 2-6 are a 'nice to have' IMO, Patch 1 is
obviously necessary, and if it is OK, I can re-send it separately if
necessary. If you don't like the changes in the other patches, I don't mind
dropping them.

I am not sure about Patch 4. It certainly works, but it changes
user-visible behavior, in what I believe is within the behavior described
by the SG interface. It makes the different methods of how BSG passes
commands down to the LLDs more conform with each other - even though I
can't make them the exact same. More details in the patch description.

I rebased the series on Jens' for-next and I have function-tested the
series on s390x's zFCP with the tools provided in the zfcp HBA API library
(https://www.ibm.com/developerworks/linux/linux390/zfcp-hbaapi.html) and
some custom code to test the read/write interface of BSG.

Reviews are more than welcome :)


                                                Beste Grüße / Best regards,
                                                  - Benjamin Block

Benjamin Block (6):
  bsg: fix kernel panic resulting from missing allocation of a
    reply-buffer
  bsg: assign sense_len instead of fixed SCSI_SENSE_BUFFERSIZE
  bsg: scsi-transport: add compile-tests to prevent reply-buffer
    overflows
  bsg: refactor ioctl to use regular BSG-command infrastructure for
    SG_IO
  bsg: reduce unecessary arguments for bsg_map_hdr()
  bsg: reduce unecessary arguments for blk_complete_sgv4_hdr_rq()

 block/bsg-lib.c                     |  4 +-
 block/bsg.c                         | 90 ++++++++++++++++++++++---------------
 drivers/scsi/scsi_transport_fc.c    |  3 ++
 drivers/scsi/scsi_transport_iscsi.c |  3 ++
 include/linux/bsg-lib.h             |  2 +
 5 files changed, 65 insertions(+), 37 deletions(-)

--
2.12.2

[toc] | [next] | [standalone]

#1707368 — [RFC PATCH 5/6] bsg: reduce unnecessary arguments for bsg_map_hdr()

Since struct bsg_command is now used in every calling case, we don't
need separation of arguments anymore that are contained in the same
bsg_command.

Signed-off-by: Benjamin Block <bblock@linux.vnet.ibm.com>
---
 block/bsg.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/block/bsg.c b/block/bsg.c
index 8517361a9b3f..6ee2ffca808a 100644
--- a/block/bsg.c
+++ b/block/bsg.c
@@ -212,11 +212,12 @@ bsg_validate_sgv4_hdr(struct sg_io_v4 *hdr, int *op)
  * map sg_io_v4 to a request.
  */
 static struct request *
-bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm,
-	    u8 *reply_buffer)
+bsg_map_hdr(struct bsg_device *bd, struct bsg_command *bc,
+	    fmode_t has_write_perm)
 {
 	struct request_queue *q = bd->queue;
 	struct request *rq, *next_rq = NULL;
+	struct sg_io_v4 *hdr = &bc->hdr;
 	int ret;
 	unsigned int op, dxfer_len;
 	void __user *dxferp = NULL;
@@ -244,7 +245,7 @@ bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm,
 	if (IS_ERR(rq))
 		return rq;
 
-	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, reply_buffer,
+	ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, bc->reply_buffer,
 				   has_write_perm);
 	if (ret)
 		goto out;
@@ -633,8 +634,7 @@ static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 		/*
 		 * get a request, fill in the blanks, and add to request queue
 		 */
-		rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm,
-				 bc->reply_buffer);
+		rq = bsg_map_hdr(bd, bc, has_write_perm);
 		if (IS_ERR(rq)) {
 			ret = PTR_ERR(rq);
 			rq = NULL;
@@ -934,8 +934,7 @@ static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 			goto sg_io_out;
 		}
 
-		bc->rq = bsg_map_hdr(bd, &bc->hdr, file->f_mode & FMODE_WRITE,
-				     bc->reply_buffer);
+		bc->rq = bsg_map_hdr(bd, bc, file->f_mode & FMODE_WRITE);
 		if (IS_ERR(bc->rq)) {
 			ret = PTR_ERR(bc->rq);
 			goto sg_io_out;
-- 
2.12.2

[toc] | [prev] | [next] | [standalone]

#1708312 — Re: [RFC PATCH 5/6] bsg: reduce unnecessary arguments for bsg_map_hdr()

Looks good,
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
-- 
Johannes Thumshirn                                          Storage
jthumshirn@suse.de                                +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850

[toc] | [prev] | [next] | [standalone]

#1708372 — Re: [RFC PATCH 5/6] bsg: reduce unnecessary arguments for bsg_map_hdr()

On Wed, Aug 09, 2017 at 04:11:19PM +0200, Benjamin Block wrote:
> Since struct bsg_command is now used in every calling case, we don't
> need separation of arguments anymore that are contained in the same
> bsg_command.
> 
> Signed-off-by: Benjamin Block <bblock@linux.vnet.ibm.com>
> ---
>  block/bsg.c | 13 ++++++-------
>  1 file changed, 6 insertions(+), 7 deletions(-)
> 
> diff --git a/block/bsg.c b/block/bsg.c
> index 8517361a9b3f..6ee2ffca808a 100644
> --- a/block/bsg.c
> +++ b/block/bsg.c
> @@ -212,11 +212,12 @@ bsg_validate_sgv4_hdr(struct sg_io_v4 *hdr, int *op)
>   * map sg_io_v4 to a request.
>   */
>  static struct request *
> -bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm,
> -	    u8 *reply_buffer)
> +bsg_map_hdr(struct bsg_device *bd, struct bsg_command *bc,
> +	    fmode_t has_write_perm)

I wish we could just rename the argument to mode and pass on the
whole file->f_mode while you are cleaning up this code.  That should
be a separate patch, though.

Reviewed-by: Christoph Hellwig <hch@lst.de>

[toc] | [prev] | [next] | [standalone]

#1709078 — Re: [RFC PATCH 5/6] bsg: reduce unnecessary arguments for bsg_map_hdr()

On Thu, Aug 10, 2017 at 11:35:31AM +0200, Christoph Hellwig wrote:
> On Wed, Aug 09, 2017 at 04:11:19PM +0200, Benjamin Block wrote:
> > Since struct bsg_command is now used in every calling case, we don't
> > need separation of arguments anymore that are contained in the same
> > bsg_command.
> > 
> > Signed-off-by: Benjamin Block <bblock@linux.vnet.ibm.com>
> > ---
> >  block/bsg.c | 13 ++++++-------
> >  1 file changed, 6 insertions(+), 7 deletions(-)
> > 
> > diff --git a/block/bsg.c b/block/bsg.c
> > index 8517361a9b3f..6ee2ffca808a 100644
> > --- a/block/bsg.c
> > +++ b/block/bsg.c
> > @@ -212,11 +212,12 @@ bsg_validate_sgv4_hdr(struct sg_io_v4 *hdr, int *op)
> >   * map sg_io_v4 to a request.
> >   */
> >  static struct request *
> > -bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm,
> > -	    u8 *reply_buffer)
> > +bsg_map_hdr(struct bsg_device *bd, struct bsg_command *bc,
> > +	    fmode_t has_write_perm)
> 
> I wish we could just rename the argument to mode and pass on the
> whole file->f_mode while you are cleaning up this code.  That should
> be a separate patch, though.
> 

Hmm, I did a quick pass through the code and the only place this seems
to be used, is to pass it to blk_verify_command() if the subcommand used
in the BSG request is a SCSI Command. And this has the same semantics.
So I guess this would require adjustments to the whole stack, as this is
also used from the 'normal' SG side of the world.



                                                    Beste Grüße / Best regards,
                                                      - Benjamin Block
-- 
Linux on z Systems Development         /         IBM Systems & Technology Group
		  IBM Deutschland Research & Development GmbH 
Vorsitz. AufsR.: Martina Koederitz     /        Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: AmtsG Stuttgart, HRB 243294

[toc] | [prev] | [next] | [standalone]

#1707369 — [RFC PATCH 3/6] bsg: scsi-transport: add compile-tests to prevent reply-buffer overflows

The BSG implementations use the bsg_job's reply buffer as storage for their
own custom reply structures (e.g.: struct fc_bsg_reply or
struct iscsi_bsg_reply). The size of bsg_job's reply buffer and those of
the implementations is not dependent in any way the compiler can currently
check.

To make it easier to notice accidental violations add an explicit compile-
time check that tests whether the implementations' reply buffer is at most
as large as bsg_job's.

To do so, we have to move the size-define from bsg.c to a common header.

Signed-off-by: Benjamin Block <bblock@linux.vnet.ibm.com>
---
 block/bsg.c                         | 3 +--
 drivers/scsi/scsi_transport_fc.c    | 3 +++
 drivers/scsi/scsi_transport_iscsi.c | 3 +++
 include/linux/bsg-lib.h             | 2 ++
 4 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/block/bsg.c b/block/bsg.c
index 285b1b8126c3..b924f1c23c58 100644
--- a/block/bsg.c
+++ b/block/bsg.c
@@ -20,6 +20,7 @@
 #include <linux/uio.h>
 #include <linux/idr.h>
 #include <linux/bsg.h>
+#include <linux/bsg-lib.h>
 #include <linux/slab.h>
 
 #include <scsi/scsi.h>
@@ -74,8 +75,6 @@ static int bsg_major;
 
 static struct kmem_cache *bsg_cmd_cachep;
 
-#define BSG_COMMAND_REPLY_BUFFERSIZE	SCSI_SENSE_BUFFERSIZE
-
 /*
  * our internal command type
  */
diff --git a/drivers/scsi/scsi_transport_fc.c b/drivers/scsi/scsi_transport_fc.c
index 892fbd9800d9..ce6654b5d329 100644
--- a/drivers/scsi/scsi_transport_fc.c
+++ b/drivers/scsi/scsi_transport_fc.c
@@ -3736,6 +3736,9 @@ static int fc_bsg_dispatch(struct bsg_job *job)
 {
 	struct Scsi_Host *shost = fc_bsg_to_shost(job);
 
+	BUILD_BUG_ON(sizeof(struct fc_bsg_reply) >
+		     BSG_COMMAND_REPLY_BUFFERSIZE);
+
 	if (scsi_is_fc_rport(job->dev))
 		return fc_bsg_rport_dispatch(shost, job);
 	else
diff --git a/drivers/scsi/scsi_transport_iscsi.c b/drivers/scsi/scsi_transport_iscsi.c
index a424eaeafeb0..4e021c949ad7 100644
--- a/drivers/scsi/scsi_transport_iscsi.c
+++ b/drivers/scsi/scsi_transport_iscsi.c
@@ -1483,6 +1483,9 @@ static int iscsi_bsg_host_dispatch(struct bsg_job *job)
 	int cmdlen = sizeof(uint32_t);	/* start with length of msgcode */
 	int ret;
 
+	BUILD_BUG_ON(sizeof(struct iscsi_bsg_reply) >
+		     BSG_COMMAND_REPLY_BUFFERSIZE);
+
 	/* check if we have the msgcode value at least */
 	if (job->request_len < sizeof(uint32_t)) {
 		ret = -ENOMSG;
diff --git a/include/linux/bsg-lib.h b/include/linux/bsg-lib.h
index e34dde2da0ef..85d7c7678cc6 100644
--- a/include/linux/bsg-lib.h
+++ b/include/linux/bsg-lib.h
@@ -25,6 +25,8 @@
 
 #include <linux/blkdev.h>
 
+#define BSG_COMMAND_REPLY_BUFFERSIZE	SCSI_SENSE_BUFFERSIZE
+
 struct request;
 struct device;
 struct scatterlist;
-- 
2.12.2

[toc] | [prev] | [next] | [standalone]

#1708375 — Re: [RFC PATCH 3/6] bsg: scsi-transport: add compile-tests to prevent reply-buffer overflows

Looks fine,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web