Re: [PATCH v2 3/4][next] scsi: megaraid_sas: Replace one-element array with flexible-array member in MR_DRV_RAID_MAP

From: Kees Cook
Date: Thu Jun 23 2022 - 11:19:58 EST


On Thu, Jun 23, 2022 at 05:14:01AM +0200, Gustavo A. R. Silva wrote:
> On Thu, Jun 23, 2022 at 03:45:33AM +0200, Gustavo A. R. Silva wrote:
> > On Wed, Jun 22, 2022 at 03:26:38PM -0700, Kees Cook wrote:
> > > On Wed, Aug 04, 2021 at 11:20:04PM -0500, Gustavo A. R. Silva wrote:
> > > > Replace one-element array with a flexible-array member in struct
> > > > MR_DRV_RAID_MAP and use the flex_array_size() helper.
> > > >
> > > > This helps with the ongoing efforts to globally enable -Warray-bounds
> > > > and get us closer to being able to tighten the FORTIFY_SOURCE routines
> > > > on memcpy().
> > > >
> > > > Link: https://en.wikipedia.org/wiki/Flexible_array_member
> > > > Link: https://www.kernel.org/doc/html/v5.10/process/deprecated.html#zero-length-and-one-element-arrays
> > > > Link: https://github.com/KSPP/linux/issues/79
> > > > Link: https://github.com/KSPP/linux/issues/109
> > > > Signed-off-by: Gustavo A. R. Silva <gustavoars@xxxxxxxxxx>
> > >
> > > I'd really like to see this fixed. :) I'm running into this 1-element
> > > array problem now with UBSAN_BOUNDS:
> >
> > Wow; another forgoten patch from the times we didn't have Patchwork. :)
> >
> > >
> > > [ 10.011173] UBSAN: array-index-out-of-bounds in /build/linux-WLUive/linux-5.15.0/drivers/scsi/megaraid/megaraid_sas_fp.c:103:32
> > > [ 10.087824] index 1 is out of range for type 'MR_LD_SPAN_MAP [1]'
> > >
> > > and I'm not the only one:
> > >
> > > https://bugzilla.kernel.org/show_bug.cgi?id=215943
> >
> > It's actually great that other people are running into these issues now.
> > That could only means that we should fixed ASAP. :)
> >
> > We also have this other series that hasn't been applied yet:
> >
> > https://lore.kernel.org/linux-hardening/cover.1645513670.git.gustavoars@xxxxxxxxxx/
> >
> > >
> > > > ---
> > > > Changes in v2:
> > > > - None.
> > > >
> > > > drivers/scsi/megaraid/megaraid_sas_fp.c | 6 +++---
> > > > drivers/scsi/megaraid/megaraid_sas_fusion.h | 2 +-
> > > > 2 files changed, 4 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/drivers/scsi/megaraid/megaraid_sas_fp.c b/drivers/scsi/megaraid/megaraid_sas_fp.c
> > > > index da1cad1ee123..9cb36ef96c2c 100644
> > > > --- a/drivers/scsi/megaraid/megaraid_sas_fp.c
> > > > +++ b/drivers/scsi/megaraid/megaraid_sas_fp.c
> > > > @@ -229,8 +229,8 @@ static int MR_PopulateDrvRaidMap(struct megasas_instance *instance, u64 map_id)
> > > > le32_to_cpu(desc_table->raid_map_desc_offset));
> > > > memcpy(pDrvRaidMap->ldSpanMap,
> > > > fw_map_dyn->ld_span_map,
> > > > - sizeof(struct MR_LD_SPAN_MAP) *
> > > > - le32_to_cpu(desc_table->raid_map_desc_elements));
> > > > + flex_array_size(pDrvRaidMap, ldSpanMap,
> > > > + le32_to_cpu(desc_table->raid_map_desc_elements)));
> > > > break;
> > > > default:
> > > > dev_dbg(&instance->pdev->dev, "wrong number of desctableElements %d\n",
> > > > @@ -254,7 +254,7 @@ static int MR_PopulateDrvRaidMap(struct megasas_instance *instance, u64 map_id)
> > > > pDrvRaidMap->ldTgtIdToLd[i] =
> > > > (u16)fw_map_ext->ldTgtIdToLd[i];
> > > > memcpy(pDrvRaidMap->ldSpanMap, fw_map_ext->ldSpanMap,
> > > > - sizeof(struct MR_LD_SPAN_MAP) * ld_count);
> > > > + flex_array_size(pDrvRaidMap, ldSpanMap, ld_count));
> > > > memcpy(pDrvRaidMap->arMapInfo, fw_map_ext->arMapInfo,
> > > > sizeof(struct MR_ARRAY_INFO) * MAX_API_ARRAYS_EXT);
> > > > memcpy(pDrvRaidMap->devHndlInfo, fw_map_ext->devHndlInfo,
> > > > diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.h b/drivers/scsi/megaraid/megaraid_sas_fusion.h
> > > > index 9adb8b30f422..5fe2f7a6eebe 100644
> > > > --- a/drivers/scsi/megaraid/megaraid_sas_fusion.h
> > > > +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.h
> > > > @@ -1182,7 +1182,7 @@ struct MR_DRV_RAID_MAP {
> > > > devHndlInfo[MAX_RAIDMAP_PHYSICAL_DEVICES_DYN];
> > > > u16 ldTgtIdToLd[MAX_LOGICAL_DRIVES_DYN];
> > > > struct MR_ARRAY_INFO arMapInfo[MAX_API_ARRAYS_DYN];
> > > > - struct MR_LD_SPAN_MAP ldSpanMap[1];
> > > > + struct MR_LD_SPAN_MAP ldSpanMap[];
> > > >
> > > > };
> > > >
> > >
> > > I think this patch is incomplete, and the wrapping struct needs to be
> > > adjusted too:
> > >
> > > @@ -1193,7 +1193,7 @@ struct MR_DRV_RAID_MAP {
> > > struct MR_DRV_RAID_MAP_ALL {
> > >
> > > struct MR_DRV_RAID_MAP raidMap;
> > > - struct MR_LD_SPAN_MAP ldSpanMap[MAX_LOGICAL_DRIVES_DYN - 1];
> > > + struct MR_LD_SPAN_MAP ldSpanMap[MAX_LOGICAL_DRIVES_DYN];
> > > } __packed;
>
> BTW, I'd really like to get some input from the maintainers of this
> code. :)

Agreed, though if we get 0 output from a diffoscope of the object files,
I think it's safe to carry such patches in our trees. This is how I've
been testing:

$ make allmodconfig
$ ./scripts/config -d GCOV_KERNEL -d KCOV -d GCC_PLUGINS -d IKHEADERS -d KASAN -d UBSAN
$ make olddefconfig
$ make the/path/to/code.o
$ cp the/path/to/code.o the/path/to/code.before
$ *apply patch*
$ make the/path/to/code.o
$ cp the/path/to/code.o the/path/to/code.after
$ diffoscope the/path/to/code.before the/path/to/code.after


--
Kees Cook