Skip to contentSkip to conversation

nvdimm/pfn_dev: Prevent the creation of zero-sized namespaces

PATCH In review Aneesh Kumar K.V 2 participants 3 messages Aug 9, 2023 libnvdimm: non-volatile memory device

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 nvdimm/pfn_dev: Prevent the creation of zero-sized namespaces

On architectures that have different page size values used for kernel direct mapping and userspace mappings, the user can end up creating zero-sized namespaces as shown below

:/sys/bus/nd/devices/region1# cat align
0x1000000
/sys/bus/nd/devices/region1# echo 0x200000 > align
/sys/bus/nd/devices/region1/dax1.0# cat supported_alignments
65536 16777216
 $ ndctl create-namespace -r region1 -m devdax -s 18M --align 64K
{
  "dev":"namespace1.0",
  "mode":"devdax",
  "map":"dev",
  "size":0,
  "uuid":"3094329a-0c66-4905-847e-357223e56ab0",
  "daxregion":{
    "id":1,
    "size":0,
    "align":65536
  },
  "align":65536
}
similarily for fsdax
 $ ndctl create-namespace -r region1 -m fsdax  -s 18M --align 64K
{
  "dev":"namespace1.0",
  "mode":"fsdax",
  "map":"dev",
  "size":0,
  "uuid":"45538a6f-dec7-405d-b1da-2a4075e06232",
  "sector_size":512,
  "align":65536,
  "blockdev":"pmem1"
}

In commit 9ffc1d19fc4a ("mm/memremap_pages: Introduce memremap_compat_align()") memremap_compat_align was added to make sure the kernel always creates namespaces with 16MB alignment. But the user can still override the region alignment and no input validation is done in the region alignment values to retain the flexibility user had before. However, the kernel ensures that only part of the namespace that can be mapped via kernel direct mapping page size is enabled. This is achieved by tracking the unmapped part of the namespace in pfn_sb->end_trunc. The kernel also ensures that the start address of the namespace is also aligned to the kernel direct mapping page size.

Depending on the user request, the kernel implements userspace mapping alignment by updating pfn device alignment attribute and this value is used to adjust the start address for userspace mappings. This is tracked in pfn_sb->dataoff. Hence the available size for userspace mapping is:

usermapping_size = size of the namespace - pfn_sb->end_trun - pfn_sb->dataoff

If the kernel finds the user mapping size zero then don't allow the creation of namespace.

After fix:

$ ndctl create-namespace -f  -r region1 -m devdax  -s 18M --align 64K
libndctl: ndctl_dax_enable: dax1.1: failed to enable
  Error: namespace1.2: failed to enable

failed to create namespace: No such device or address

And existing zero sized namespace will be marked disabled.

root@ltczz75-lp2:/home/kvaneesh# ndctl  list -N -r region1 -i
[
  {
    "dev":"namespace1.0",
    "mode":"raw",
    "size":18874368,
    "uuid":"94a90fb0-8e78-4fb6-a759-ffc62f9fa181",
    "sector_size":512,
    "state":"disabled"
  },

---
Changes from v1:
* Use resource_size() helper instead of opencoding it

 drivers/nvdimm/pfn_devs.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
drivers/nvdimm/pfn_devs.c +11 -3
index af7d9301520c..0777b1626f6c 100644
@@ -452,8 +452,9 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 	u64 checksum, offset;
 	struct resource *res;
 	enum nd_pfn_mode mode;
+	resource_size_t res_size;
 	struct nd_namespace_io *nsio;
-	unsigned long align, start_pad;
+	unsigned long align, start_pad, end_trunc;
 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
 	struct nd_namespace_common *ndns = nd_pfn->ndns;
 	const uuid_t *parent_uuid = nd_dev_to_uuid(&ndns->dev);
@@ -503,6 +504,7 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 	align = le32_to_cpu(pfn_sb->align);
 	offset = le64_to_cpu(pfn_sb->dataoff);
 	start_pad = le32_to_cpu(pfn_sb->start_pad);
+	end_trunc = le32_to_cpu(pfn_sb->end_trunc);
 	if (align == 0)
 		align = 1UL << ilog2(offset);
 	mode = le32_to_cpu(pfn_sb->mode);
@@ -584,7 +586,8 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 	 */
 	nsio = to_nd_namespace_io(&ndns->dev);
 	res = &nsio->res;
-	if (offset >= resource_size(res)) {
+	res_size = resource_size(res);
+	if (offset >= res_size) {
 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
 				dev_name(&ndns->dev));
 		return -EOPNOTSUPP;
@@ -610,6 +613,10 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 		return -EOPNOTSUPP;
 	}
 
+	if (offset >= (res_size - start_pad - end_trunc)) {
+		dev_err(&nd_pfn->dev, "bad offset with small namespace\n");
+		return -EOPNOTSUPP;
+	}
 	return 0;
 }
 EXPORT_SYMBOL(nd_pfn_validate);
@@ -810,7 +817,8 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
 	else
 		return -ENXIO;
 
-	if (offset >= size) {
+	if (offset >= (size - end_trunc)) {
+		/* This results in zero size devices */
 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
 				dev_name(&ndns->dev));
 		return -ENXIO;
Signature · 1 line
2.41.0
#1 lore raw

PATCH v2 2/2 nvdimm/pfn_dev: Avoid unnecessary endian conversion

use the local variable that already have the converted values.

No functional change in this patch.

---
 drivers/nvdimm/pfn_devs.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/nvdimm/pfn_devs.c +2 -4
index 0777b1626f6c..c662d8964802 100644
@@ -601,14 +601,12 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 		return -EOPNOTSUPP;
 	}
 
-	if (!IS_ALIGNED(res->start + le32_to_cpu(pfn_sb->start_pad),
-				memremap_compat_align())) {
+	if (!IS_ALIGNED(res->start + start_pad, memremap_compat_align())) {
 		dev_err(&nd_pfn->dev, "resource start misaligned\n");
 		return -EOPNOTSUPP;
 	}
 
-	if (!IS_ALIGNED(res->end + 1 - le32_to_cpu(pfn_sb->end_trunc),
-				memremap_compat_align())) {
+	if (!IS_ALIGNED(res->end + 1 - end_trunc, memremap_compat_align())) {
 		dev_err(&nd_pfn->dev, "resource end misaligned\n");
 		return -EOPNOTSUPP;
 	}
Signature · 1 line
2.41.0
Reviewed-by on v2 1/2
#2 lore raw
In reply to Aneesh Kumar K.V
Quoted 130 lines · Aneesh Kumar K.V

On architectures that have different page size values used for kernel direct mapping and userspace mappings, the user can end up creating zero-sized namespaces as shown below

:/sys/bus/nd/devices/region1# cat align
0x1000000
/sys/bus/nd/devices/region1# echo 0x200000 > align
/sys/bus/nd/devices/region1/dax1.0# cat supported_alignments
65536 16777216
 $ ndctl create-namespace -r region1 -m devdax -s 18M --align 64K
{
  "dev":"namespace1.0",
  "mode":"devdax",
  "map":"dev",
  "size":0,
  "uuid":"3094329a-0c66-4905-847e-357223e56ab0",
  "daxregion":{
    "id":1,
    "size":0,
    "align":65536
  },
  "align":65536
}
similarily for fsdax
 $ ndctl create-namespace -r region1 -m fsdax  -s 18M --align 64K
{
  "dev":"namespace1.0",
  "mode":"fsdax",
  "map":"dev",
  "size":0,
  "uuid":"45538a6f-dec7-405d-b1da-2a4075e06232",
  "sector_size":512,
  "align":65536,
  "blockdev":"pmem1"
}

In commit 9ffc1d19fc4a ("mm/memremap_pages: Introduce memremap_compat_align()") memremap_compat_align was added to make sure the kernel always creates namespaces with 16MB alignment. But the user can still override the region alignment and no input validation is done in the region alignment values to retain the flexibility user had before. However, the kernel ensures that only part of the namespace that can be mapped via kernel direct mapping page size is enabled. This is achieved by tracking the unmapped part of the namespace in pfn_sb->end_trunc. The kernel also ensures that the start address of the namespace is also aligned to the kernel direct mapping page size.

Depending on the user request, the kernel implements userspace mapping alignment by updating pfn device alignment attribute and this value is used to adjust the start address for userspace mappings. This is tracked in pfn_sb->dataoff. Hence the available size for userspace mapping is:

usermapping_size = size of the namespace - pfn_sb->end_trun - pfn_sb->dataoff

If the kernel finds the user mapping size zero then don't allow the creation of namespace.

After fix:

$ ndctl create-namespace -f  -r region1 -m devdax  -s 18M --align 64K
libndctl: ndctl_dax_enable: dax1.1: failed to enable
  Error: namespace1.2: failed to enable

failed to create namespace: No such device or address

And existing zero sized namespace will be marked disabled.

root@ltczz75-lp2:/home/kvaneesh# ndctl  list -N -r region1 -i
[
  {
    "dev":"namespace1.0",
    "mode":"raw",
    "size":18874368,
    "uuid":"94a90fb0-8e78-4fb6-a759-ffc62f9fa181",
    "sector_size":512,
    "state":"disabled"
  },

---
Changes from v1:
* Use resource_size() helper instead of opencoding it

 drivers/nvdimm/pfn_devs.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
drivers/nvdimm/pfn_devs.c +9 -2
index af7d9301520c..0777b1626f6c 100644
@@ -452,8 +452,9 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 	u64 checksum, offset;
 	struct resource *res;
 	enum nd_pfn_mode mode;
+	resource_size_t res_size;
 	struct nd_namespace_io *nsio;
-	unsigned long align, start_pad;
+	unsigned long align, start_pad, end_trunc;
 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
 	struct nd_namespace_common *ndns = nd_pfn->ndns;
 	const uuid_t *parent_uuid = nd_dev_to_uuid(&ndns->dev);
@@ -503,6 +504,7 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 	align = le32_to_cpu(pfn_sb->align);
 	offset = le64_to_cpu(pfn_sb->dataoff);
 	start_pad = le32_to_cpu(pfn_sb->start_pad);
+	end_trunc = le32_to_cpu(pfn_sb->end_trunc);
 	if (align == 0)
 		align = 1UL << ilog2(offset);
 	mode = le32_to_cpu(pfn_sb->mode);
@@ -584,7 +586,8 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 	 */
 	nsio = to_nd_namespace_io(&ndns->dev);
 	res = &nsio->res;
-	if (offset >= resource_size(res)) {
+	res_size = resource_size(res);
+	if (offset >= res_size) {
 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
 				dev_name(&ndns->dev));
 		return -EOPNOTSUPP;
@@ -610,6 +613,10 @@ int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 		return -EOPNOTSUPP;
 	}
 
+	if (offset >= (res_size - start_pad - end_trunc)) {
+		dev_err(&nd_pfn->dev, "bad offset with small namespace\n");
+		return -EOPNOTSUPP;
+	}
 	return 0;
 }
 EXPORT_SYMBOL(nd_pfn_validate);

drivers/nvdimm/pfn_devs.c static int nd_pfn_init(struct nd_pfn *nd_pfn) · line 824

Quoted 9 lines
drivers/nvdimm/pfn_devs.c +2 -1
@@ -810,7 +817,8 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
 	else
 		return -ENXIO;
 
-	if (offset >= size) {
+	if (offset >= (size - end_trunc)) {
+		/* This results in zero size devices */
 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
 				dev_name(&ndns->dev));
 		return -ENXIO;