| Message ID | 20190925110008.GK3264@mwanda (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | libnvdimm/namespace: Fix a signedness bug in __holder_class_store() | expand |
On Wed, 2019-09-25 at 14:00 +0300, Dan Carpenter wrote: > The "ndns->claim_class" variable is an enum but in this case GCC will > treat it as an unsigned int so the error handling is never triggered. > > Fixes: 14e494542636 ("libnvdimm, btt: BTT updates for UEFI 2.7 format") > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> > --- > drivers/nvdimm/namespace_devs.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/nvdimm/namespace_devs.c b/drivers/nvdimm/namespace_devs.c > index cca0a3ba1d2c..669985527716 100644 > --- a/drivers/nvdimm/namespace_devs.c > +++ b/drivers/nvdimm/namespace_devs.c > @@ -1529,7 +1529,7 @@ static ssize_t __holder_class_store(struct device *dev, const char *buf) > return -EINVAL; > > /* btt_claim_class() could've returned an error */ > - if (ndns->claim_class < 0) > + if ((int)ndns->claim_class < 0) > return ndns->claim_class; > > return 0; Looks straightforward, and a good catch. Reviewed-by: Vishal Verma <vishal.l.verma@intel.com>
> > On Wed, 2019-09-25 at 14:00 +0300, Dan Carpenter wrote: > > The "ndns->claim_class" variable is an enum but in this case GCC will > > treat it as an unsigned int so the error handling is never triggered. > > > > Fixes: 14e494542636 ("libnvdimm, btt: BTT updates for UEFI 2.7 > > format") > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> > > --- > > drivers/nvdimm/namespace_devs.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/drivers/nvdimm/namespace_devs.c > > b/drivers/nvdimm/namespace_devs.c index cca0a3ba1d2c..669985527716 > > 100644 > > --- a/drivers/nvdimm/namespace_devs.c > > +++ b/drivers/nvdimm/namespace_devs.c > > @@ -1529,7 +1529,7 @@ static ssize_t __holder_class_store(struct device > *dev, const char *buf) > > return -EINVAL; > > > > /* btt_claim_class() could've returned an error */ > > - if (ndns->claim_class < 0) > > + if ((int)ndns->claim_class < 0) > > return ndns->claim_class; > > > > return 0; > > Looks straightforward, and a good catch. > Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> I'm not sure this is really a good fix. This leaves ndns->claim_class set to an invalid value. Is that ok? Ira
On Wed, 2019-09-25 at 11:25 -0600, Weiny, Ira wrote: > > On Wed, 2019-09-25 at 14:00 +0300, Dan Carpenter wrote: > > > The "ndns->claim_class" variable is an enum but in this case GCC will > > > treat it as an unsigned int so the error handling is never triggered. > > > > > > Fixes: 14e494542636 ("libnvdimm, btt: BTT updates for UEFI 2.7 > > > format") > > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> > > > --- > > > drivers/nvdimm/namespace_devs.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/drivers/nvdimm/namespace_devs.c > > > b/drivers/nvdimm/namespace_devs.c index cca0a3ba1d2c..669985527716 > > > 100644 > > > --- a/drivers/nvdimm/namespace_devs.c > > > +++ b/drivers/nvdimm/namespace_devs.c > > > @@ -1529,7 +1529,7 @@ static ssize_t __holder_class_store(struct device > > *dev, const char *buf) > > > return -EINVAL; > > > > > > /* btt_claim_class() could've returned an error */ > > > - if (ndns->claim_class < 0) > > > + if ((int)ndns->claim_class < 0) > > > return ndns->claim_class; > > > > > > return 0; > > > > Looks straightforward, and a good catch. > > Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> > > I'm not sure this is really a good fix. This leaves ndns->claim_class set to an invalid value. Is that ok? > Hm, it is just in a store operation for the holder_class sysfs attribute. if claim_class was negative, that store will just fail. From the userspace side, this means the 'mode' enforcement API will fail. Typically, 'ndctl' doesn't require the enforcement to succeed, since we can disambiguate namespaces even without it, so it doesn't block namespace creation etc. On the kernel side, claim_class gets used by btt_devs.c during probe, and looks like this case, as it currently stands, would fail the BTT probe (nd_btt_probe()). I think this is what we want? I guess it can be made a bit clearer by storing NVDIMM_CCLASS_UNKNOWN explicitly in holder_class_store(), but that can be a separate improvement from what this patch is addressing.
On Wed, Sep 25, 2019 at 10:49:08AM -0700, 'Vishal Verma' wrote: > On Wed, 2019-09-25 at 11:25 -0600, Weiny, Ira wrote: > > > On Wed, 2019-09-25 at 14:00 +0300, Dan Carpenter wrote: > > > > The "ndns->claim_class" variable is an enum but in this case GCC will > > > > treat it as an unsigned int so the error handling is never triggered. > > > > > > > > Fixes: 14e494542636 ("libnvdimm, btt: BTT updates for UEFI 2.7 > > > > format") > > > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> > > > > --- > > > > drivers/nvdimm/namespace_devs.c | 2 +- > > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > > > diff --git a/drivers/nvdimm/namespace_devs.c > > > > b/drivers/nvdimm/namespace_devs.c index cca0a3ba1d2c..669985527716 > > > > 100644 > > > > --- a/drivers/nvdimm/namespace_devs.c > > > > +++ b/drivers/nvdimm/namespace_devs.c > > > > @@ -1529,7 +1529,7 @@ static ssize_t __holder_class_store(struct device > > > *dev, const char *buf) > > > > return -EINVAL; > > > > > > > > /* btt_claim_class() could've returned an error */ > > > > - if (ndns->claim_class < 0) > > > > + if ((int)ndns->claim_class < 0) > > > > return ndns->claim_class; > > > > > > > > return 0; > > > > > > Looks straightforward, and a good catch. > > > Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> > > > > I'm not sure this is really a good fix. This leaves ndns->claim_class set to an invalid value. Is that ok? > > > > Hm, it is just in a store operation for the holder_class sysfs > attribute. if claim_class was negative, that store will just fail. > > From the userspace side, this means the 'mode' enforcement API will > fail. Typically, 'ndctl' doesn't require the enforcement to succeed, > since we can disambiguate namespaces even without it, so it doesn't > block namespace creation etc. > > On the kernel side, claim_class gets used by btt_devs.c during probe, > and looks like this case, as it currently stands, would fail the BTT > probe (nd_btt_probe()). I think this is what we want? > > I guess it can be made a bit clearer by storing NVDIMM_CCLASS_UNKNOWN > explicitly in holder_class_store(), but that can be a separate > improvement from what this patch is addressing. Fair enough I've sent a follow on patch. Thanks, Ira
diff --git a/drivers/nvdimm/namespace_devs.c b/drivers/nvdimm/namespace_devs.c index cca0a3ba1d2c..669985527716 100644 --- a/drivers/nvdimm/namespace_devs.c +++ b/drivers/nvdimm/namespace_devs.c @@ -1529,7 +1529,7 @@ static ssize_t __holder_class_store(struct device *dev, const char *buf) return -EINVAL; /* btt_claim_class() could've returned an error */ - if (ndns->claim_class < 0) + if ((int)ndns->claim_class < 0) return ndns->claim_class; return 0;
The "ndns->claim_class" variable is an enum but in this case GCC will treat it as an unsigned int so the error handling is never triggered. Fixes: 14e494542636 ("libnvdimm, btt: BTT updates for UEFI 2.7 format") Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> --- drivers/nvdimm/namespace_devs.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)