Skip to contentSkip to conversation

NVMe: Check for Extended metadata in ioctl path

PATCH New Sathayavathi M 2 participants 2 messages Feb 19, 2015

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 NVMe: Check for Extended metadata in ioctl path

From: Sathyavathi M <[email protected]>

In NVME_IOCTL_SUBMIT_IO the check for the extended metadata is missing. Currently, whenever an i/o request with extended metadata is issued, io.metadata has to be assigned a dummy addr. If this is not assigned then the check fails and returns error. This patch fixes this issue and also adds checks at necessary places for extended metadata.

Signed-off-by Sathyavathi M <sathya.m at samsung.com>
---
 drivers/block/nvme-core.c | 11 ++++++++---
 include/linux/nvme.h      |  1 +
 include/uapi/linux/nvme.h |  4 ++++
 3 files changed, 13 insertions(+), 3 deletions(-)
drivers/block/nvme-core.c +8 -3
index 3eaa0be..a632d78 100644
@@ -1595,13 +1595,15 @@ static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 	struct nvme_iod *iod, *meta_iod = NULL;
 	dma_addr_t meta_dma_addr;
 	void *meta, *uninitialized_var(meta_mem);
+	bool ext_lba = ns->flbas & EXT_LBA;
 
 	if (copy_from_user(&io, uio, sizeof(io)))
 		return -EFAULT;
 	length = (io.nblocks + 1) << ns->lba_shift;
 	meta_len = (io.nblocks + 1) * ns->ms;
 
-	if (meta_len && ((io.metadata & 3) || !io.metadata))
+	/* Check for unaligned or NULL metadata ptr for separate buffer */
+	if (meta_len && !(ext_lba) && ((io.metadata & 3) || !io.metadata))
 		return -EINVAL;
 
 	switch (io.opcode) {
@@ -1629,7 +1631,8 @@ static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 	c.rw.apptag = cpu_to_le16(io.apptag);
 	c.rw.appmask = cpu_to_le16(io.appmask);
 
-	if (meta_len) {
+	/* Map the Separate Buffer separately (not needed for Extenede LBA) */
+	if (meta_len && !(ext_lba)) {
 		meta_iod = nvme_map_user_pages(dev, io.opcode & 1, io.metadata,
 								meta_len);
 		if (IS_ERR(meta_iod)) {
@@ -1670,7 +1673,7 @@ static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 	else
 		status = nvme_submit_io_cmd(dev, ns, &c, NULL);
 
-	if (meta_len) {
+	if (meta_len && !(ext_lba)) {
 		if (status == NVME_SC_SUCCESS && !(io.opcode & 1)) {
 			int meta_offset = 0;
 
@@ -1862,6 +1865,7 @@ static int nvme_revalidate_disk(struct gendisk *disk)
 	if (nvme_identify(dev, ns->ns_id, 0, dma_addr))
 		goto free;
 
+	ns->flbas = id->flbas;
 	lbaf = id->flbas & 0xf;
 	ns->lba_shift = id->lbaf[lbaf].ds;
 
@@ -1963,6 +1967,7 @@ static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid,
 
 	ns->ns_id = nsid;
 	ns->disk = disk;
+	ns->flbas = id->flbas;
 	lbaf = id->flbas & 0xf;
 	ns->lba_shift = id->lbaf[lbaf].ds;
 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
include/linux/nvme.h +1 -0
index 19a5d4b..9c74cbc 100644
@@ -121,6 +121,7 @@ struct nvme_ns {
 	unsigned ns_id;
 	int lba_shift;
 	int ms;
+	u8 flbas;
 	u64 mode_select_num_blocks;
 	u32 mode_select_block_len;
 };
include/uapi/linux/nvme.h +4 -0
index 26386cf..1e3b7a6 100644
@@ -98,6 +98,10 @@ struct nvme_lbaf {
 	__u8			rp;
 };
 
+enum {
+	EXT_LBA		= 0x10
+};
+
 struct nvme_id_ns {
 	__le64			nsze;
 	__le64			ncap;
Signature · 1 line
1.8.3.2
#1 lore raw

PATCH 1/1 NVMe: Check for Extended metadata in ioctl path

Quoted 20 lines · Sathayavathi M

From: Sathyavathi M <sathya.m at samsung.com>

In NVME_IOCTL_SUBMIT_IO the check for the extended metadata is missing. Currently, whenever an i/o request with extended metadata is issued, io.metadata has to be assigned a dummy addr. If this is not assigned then the check fails and returns error. This patch fixes this issue and also adds checks at necessary places for extended metadata.

Signed-off-by Sathyavathi M <sathya.m at samsung.com>

---

drivers/block/nvme-core.c | 11 ++++++++---
include/linux/nvme.h      |  1 +
include/uapi/linux/nvme.h |  4 ++++
3 files changed, 13 insertions(+), 3 deletions(-)
drivers/block/nvme-core.c +0 -0
index 3eaa0be..a632d78 100644

drivers/block/nvme-core.c static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) · line 1603

Quoted 9 lines
drivers/block/nvme-core.c +1 -0
@@ -1595,13 +1595,15 @@ static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 	struct nvme_iod *iod, *meta_iod = NULL;
 	dma_addr_t meta_dma_addr;
 	void *meta, *uninitialized_var(meta_mem);
+	bool ext_lba = ns->flbas & EXT_LBA;
 
 	if (copy_from_user(&io, uio, sizeof(io)))
 		return -EFAULT;
 	length = (io.nblocks + 1) << ns->lba_shift;
 	meta_len = (io.nblocks + 1) * ns->ms;

If you're using extended, the 'length' you're using for get_user_pages and setup_prps is wrong. You need to add the meta_len to it since we expect the buffer has the metadata spliced into user data.

This function (and this driver for that matter) didn't expect to deal with extended lba's, and we have bigger problems if you're actually using them. There is more interest here so we should get this working, but patches are starting to clash with other things on the patch queue. It would really help if maintainer would merge, or at least comment on what's going on.

Willy:

Are you going to help us out, or wait for a coup d'etat?