Skip to contentSkip to conversation

nvme: check if the namespace supports metadata in nvme_map_user_request()

PATCH Superseded Puranjay Mohan 3 participants 6 messages Aug 27, 2024 – Sep 6, 2024 NVMe

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nvme: check if the namespace supports metadata in nvme_map_user_request()

On an NVMe namespace that does not support metadata, it is possible to send an IO command with metadata through io-passthru. nvme_map_user_request() doesn't check if the namespace supports metadata before sending it forward.

Reject an IO command with metadata when the NVMe namespace doesn't support it.

---
 drivers/nvme/host/ioctl.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/nvme/host/ioctl.c +4 -0
index f1d58e70933f..4f49523b8a41 100644
@@ -4,6 +4,7 @@
  * Copyright (c) 2017-2021 Christoph Hellwig.
  */
 #include <linux/bio-integrity.h>
+#include <linux/blk-integrity.h>
 #include <linux/ptrace.h>	/* for force_successful_syscall_return */
 #include <linux/nvme_ioctl.h>
 #include <linux/io_uring/cmd.h>
@@ -122,6 +123,9 @@ static int nvme_map_user_request(struct request *req, u64 ubuffer,
 	struct bio *bio = NULL;
 	int ret;
 
+	if (meta_buffer && meta_len && bdev && !blk_get_integrity(bdev->bd_disk))
+		return -EINVAL;
+
 	if (ioucmd && (ioucmd->flags & IORING_URING_CMD_FIXED)) {
 		struct iov_iter iter;
 
Signature · 1 line
2.40.1
#1 lore raw

drivers/nvme/host/ioctl.c static int nvme_map_user_request(struct request *req, u64 ubuffer, · line 127

Quoted 2 lines · Puranjay Mohan
drivers/nvme/host/ioctl.c +2 -0
+	if (meta_buffer && meta_len && bdev && !blk_get_integrity(bdev->bd_disk))
+		return -EINVAL;

Overly long line here. If we go past my initial RFC I'd probably restructure the function a little bit, i.e. add a new

	bool has_metadata = bdev && meta_buffer && meta_len;

and then use that both for the support check and the actualy mapping below.

#2 lore raw
Quoted 11 lines · Christoph Hellwig

drivers/nvme/host/ioctl.c static int nvme_map_user_request(struct request *req, u64 ubuffer, · line 127

Quoted 2 lines · Puranjay Mohan
drivers/nvme/host/ioctl.c +2 -0
+	if (meta_buffer && meta_len && bdev && !blk_get_integrity(bdev->bd_disk))
+		return -EINVAL;

Overly long line here. If we go past my initial RFC I'd probably restructure the function a little bit, i.e. add a new

	bool has_metadata = bdev && meta_buffer && meta_len;

and then use that both for the support check and the actualy mapping below.

Sure,
I will send v2 with these changes now.

P.S. - It looks like we will need manual backports for stable kernels as this won't apply directly. I will send them after this is accepted.

Thanks,
Puranjay

maintainer
#3 lore raw
Quoted 3 lines · Puranjay Mohan

P.S. - It looks like we will need manual backports for stable kernels as this won't apply directly. I will send them after this is accepted.

BTW, the reason for your bug observation on older stable but not on newer ones looks like was originally "fixed" with:

d4aa57a1cac3c99 ("block: don't bother iter advancing a fully done bio")

So while your change is fine, the above commit possibly inadvertently fixes the inappropriate "advance" because all successful nvme completions are "fully done".

It still probably doesn't make sense to attempt metadata on request queues that didn't register with the interface though, so your patch is also fine.

maintainer
#4 lore raw
In reply to Puranjay Mohan

drivers/nvme/host/ioctl.c static int nvme_map_user_request(struct request *req, u64 ubuffer, · line 128

Quoted 7 lines · Puranjay Mohan
drivers/nvme/host/ioctl.c +3 -0
@@ -122,6 +123,9 @@ static int nvme_map_user_request(struct request *req, u64 ubuffer,
 	struct bio *bio = NULL;
 	int ret;
 
+	if (meta_buffer && meta_len && bdev && !blk_get_integrity(bdev->bd_disk))
+		return -EINVAL;
+

Should we also fail if there's no bdev? The driver won't use the requested metadata for admin commands either, so that is also an invalid user request.

#5 lore raw
Quoted 5 lines · Keith Busch
Quoted 4 lines · Puranjay Mohan
drivers/nvme/host/ioctl.c +0 -0
@@ -122,6 +123,9 @@ static int nvme_map_user_request(struct request *req, u64 ubuffer,
 	struct bio *bio = NULL;
 	int ret;
 
Quoted 7 lines

drivers/nvme/host/ioctl.c static int nvme_map_user_request(struct request *req, u64 ubuffer, · line 128

Quoted 3 lines
drivers/nvme/host/ioctl.c +3 -0
+	if (meta_buffer && meta_len && bdev && !blk_get_integrity(bdev->bd_disk))
+		return -EINVAL;
+

Should we also fail if there's no bdev? The driver won't use the requested metadata for admin commands either, so that is also an invalid user request.

Yes, the last version gets the right.