On Tue, Feb 08, 2022 at 03:42:18PM +1100, Stephen Rothwell wrote:
> Hi all,
>
> [Cc'ing the scsi maintainers. Sorry I should have done that sooner]
>
> On Mon, 24 Jan 2022 19:43:44 -0800 Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> >
> > On Tue, Jan 25, 2022 at 02:24:30PM +1100, Stephen Rothwell wrote:
> > >
> > > After merging the kspp tree, today's linux-next build (powerpc
> > > allyesconfig) failed like this:
> > >
> > >
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c: In function 'ibmvscsis_send_messages':
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1934:44: error: array subscript 'struct viosrp_crq[0]' is partly outside array bounds of 'u64[1]' {aka 'long long unsigned int[1]'} [-Werror=array-bounds]
> > > 1934 | crq->valid = VALID_CMD_RESP_EL;
> > > | ^~
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1875:13: note: while referencing 'msg_hi'
> > > 1875 | u64 msg_hi = 0;
> > > | ^~~~~~
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1935:44: error: array subscript 'struct viosrp_crq[0]' is partly outside array bounds of 'u64[1]' {aka 'long long unsigned int[1]'} [-Werror=array-bounds]
> > > 1935 | crq->format = cmd->rsp.format;
> > > | ^~
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1875:13: note: while referencing 'msg_hi'
> > > 1875 | u64 msg_hi = 0;
> > > | ^~~~~~
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1938:52: error: array subscript 'struct viosrp_crq[0]' is partly outside array bounds of 'u64[1]' {aka 'long long unsigned int[1]'} [-Werror=array-bounds]
> > > 1938 | crq->status = VIOSRP_ADAPTER_FAIL;
> > > | ^~
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1875:13: note: while referencing 'msg_hi'
> > > 1875 | u64 msg_hi = 0;
> > > | ^~~~~~
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1940:44: error: array subscript 'struct viosrp_crq[0]' is partly outside array bounds of 'u64[1]' {aka 'long long unsigned int[1]'} [-Werror=array-bounds]
> > > 1940 | crq->IU_length = cpu_to_be16(cmd->rsp.len);
> > > | ^~
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c:1875:13: note: while referencing 'msg_hi'
> > > 1875 | u64 msg_hi = 0;
> > > | ^~~~~~
> > >
> > > Exposed by commit
> > >
> > > 4ba545781e20 ("Makefile: Enable -Warray-bounds")
> > >
> > > Probably introduced by commit
> > >
> > > 88a678bbc34c ("ibmvscsis: Initial commit of IBM VSCSI Tgt Driver")
> > >
> > > I applied the following hack for now:
> > >
> > > From: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
> > > Date: Tue, 25 Jan 2022 14:18:36 +1100
> > > Subject: [PATCH] scsi: hack for building with -Warray-bounds
> > >
> > > Signed-off-by: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
> > > ---
> > > drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c | 6 +++---
> > > 1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
> > > index 61f06f6885a5..89fcf98c61c3 100644
> > > --- a/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
> > > +++ b/drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c
> > > @@ -1872,11 +1872,11 @@ static void srp_snd_msg_failed(struct scsi_info *vscsi, long rc)
> > > */
> > > static void ibmvscsis_send_messages(struct scsi_info *vscsi)
> > > {
> > > - u64 msg_hi = 0;
> > > + u64 msg_hi[2] = { };
> > > /* note do not attempt to access the IU_data_ptr with this pointer
> > > * it is not valid
> > > */
> > > - struct viosrp_crq *crq = (struct viosrp_crq *)&msg_hi;
> > > + struct viosrp_crq *crq = (struct viosrp_crq *)msg_hi;
> > > struct ibmvscsis_cmd *cmd, *nxt;
> > > long rc = ADAPT_SUCCESS;
> > > bool retry = false;
> > > @@ -1940,7 +1940,7 @@ static void ibmvscsis_send_messages(struct scsi_info *vscsi)
> > > crq->IU_length = cpu_to_be16(cmd->rsp.len);
> > >
> > > rc = h_send_crq(vscsi->dma_dev->unit_address,
> > > - be64_to_cpu(msg_hi),
> > > + be64_to_cpu(msg_hi[0]),
> > > be64_to_cpu(cmd->rsp.tag));
> > >
> > > dev_dbg(&vscsi->dev, "send_messages: cmd %p, tag 0x%llx, rc %ld\n",
> >
> > This looks correct to me. struct viosrp_crq is 16 bytes wide. The only
> > suggestion I might make would be either avoid the bare '2':
> >
> > u64 msg_hi[sizeof(struct viosrp_crq) / sizeof(u64)] = { };
> >
> > or adjust struct viosrp_crq so the casting isn't needed at all:
> >
> >
> > truct viosrp_crq {
> > union {
> > u64 hi;
> > struct {
> > u8 valid; /* used by RPA */
> > u8 format; /* SCSI vs out-of-band */
> > u8 reserved;
> > u8 status; /* non-scsi failure? (e.g. DMA failure) */
> > __be16 timeout; /* in seconds */
> > __be16 IU_length; /* in bytes */
> > };
> > };
> > __be64 IU_data_ptr; /* the TCE for transferring data */
> > };
> >
> > struct viosrp_crq crq = { };
> > ...
> > rc = h_send_crq(vscsi->dma_dev->unit_address,
> > be64_to_cpu(crq.hi),
> > be64_to_cpu(cmd->rsp.tag));
> >
> >
> >
>
> Has there been any progress on this?
>
> Commit 88a678bbc34c mentioned above was merged in v4.8-rc1.
Thanks for the reminder! I've sent this now:
https://lore.kernel.org/lkml/20220208061231.3429486-1-keescook@xxxxxxxxxxxx
--
Kees Cook
