[PATCH 0/3] uring-passthrough for admin commands

[PATCH 0/3] uring-passthrough for admin commands

[Kanchan Joshi]

During the review rounds and LSFMM, it emerged that uring-passthrough
for admin-commands will also be useful. So this series wires that up.

Prepared against for-next (linux-block) on top of
commit cef3f8e4ed04 (Merge branch 'for-5.19/block'...).

Patch 1 is prep.
Patch 2 adds new opcode for admin uring-passthrough, and enables it over
ns char device (/dev/ngXnY). It reuses the code of io-command
passthrough, with the only difference that commands are issued on admin
queue.

Patch 3 enables the same for controller char device (/dev/nvmeX).

Kanchan Joshi (3):
  nvme: helper for uring-passthrough checks
  nvme: enable uring-passthrough for admin commands
  nvme: admin command uring-passthrough on controller node

 drivers/nvme/host/core.c        |  1 +
 drivers/nvme/host/ioctl.c       | 83 +++++++++++++++++++++++++++++----
 drivers/nvme/host/nvme.h        |  1 +
 include/uapi/linux/nvme_ioctl.h |  1 +
 4 files changed, 77 insertions(+), 9 deletions(-)

-- 
2.25.1

[PATCH 1/3] nvme: helper for uring-passthrough checks

[Kanchan Joshi]

Factor out a helper consolidating the error checks, and fix typo in a
comment too. This is in preparation to support admin commands on this
path.

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 drivers/nvme/host/ioctl.c | 24 ++++++++++++++++--------
 1 file changed, 16 insertions(+), 8 deletions(-)

diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c
index 7b0e2c9cdcae..114b490592b0 100644
--- a/drivers/nvme/host/ioctl.c
+++ b/drivers/nvme/host/ioctl.c
@@ -556,22 +556,30 @@ long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 	return __nvme_ioctl(ns, cmd, (void __user *)arg);
 }
 
-static int nvme_ns_uring_cmd(struct nvme_ns *ns, struct io_uring_cmd *ioucmd,
-			     unsigned int issue_flags)
+static int nvme_uring_cmd_checks(unsigned int issue_flags)
 {
-	struct nvme_ctrl *ctrl = ns->ctrl;
-	int ret;
-
-	BUILD_BUG_ON(sizeof(struct nvme_uring_cmd_pdu) > sizeof(ioucmd->pdu));
-
 	/* IOPOLL not supported yet */
 	if (issue_flags & IO_URING_F_IOPOLL)
 		return -EOPNOTSUPP;
 
-	/* NVMe passthrough requires bit SQE/CQE support */
+	/* NVMe passthrough requires big SQE/CQE support */
 	if ((issue_flags & (IO_URING_F_SQE128|IO_URING_F_CQE32)) !=
 	    (IO_URING_F_SQE128|IO_URING_F_CQE32))
 		return -EOPNOTSUPP;
+	return 0;
+}
+
+static int nvme_ns_uring_cmd(struct nvme_ns *ns, struct io_uring_cmd *ioucmd,
+			     unsigned int issue_flags)
+{
+	struct nvme_ctrl *ctrl = ns->ctrl;
+	int ret;
+
+	BUILD_BUG_ON(sizeof(struct nvme_uring_cmd_pdu) > sizeof(ioucmd->pdu));
+
+	ret = nvme_uring_cmd_checks(issue_flags);
+	if (ret)
+		return ret;
 
 	switch (ioucmd->cmd_op) {
 	case NVME_URING_CMD_IO:
-- 
2.25.1

[PATCH 2/3] nvme: enable uring-passthrough for admin commands

[Kanchan Joshi]

Add a new opcode NVME_URING_CMD_ADMIN that userspace can use.
Wire up support for the case when this is issued over generic char
device node (/dev/ngXnY).

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 drivers/nvme/host/ioctl.c       | 54 +++++++++++++++++++++++++++++++--
 include/uapi/linux/nvme_ioctl.h |  1 +
 2 files changed, 53 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c
index 114b490592b0..0f46dc7381a0 100644
--- a/drivers/nvme/host/ioctl.c
+++ b/drivers/nvme/host/ioctl.c
@@ -569,6 +569,31 @@ static int nvme_uring_cmd_checks(unsigned int issue_flags)
 	return 0;
 }
 
+static inline bool is_ctrl_uring_cmd(unsigned int cmd)
+{
+	return (cmd == NVME_URING_CMD_ADMIN);
+}
+
+static int nvme_ctrl_uring_cmd(struct nvme_ctrl *ctrl,
+			struct io_uring_cmd *ioucmd, unsigned int issue_flags)
+{
+	int ret;
+
+	ret = nvme_uring_cmd_checks(issue_flags);
+	if (ret)
+		return ret;
+
+	switch (ioucmd->cmd_op) {
+	case NVME_URING_CMD_ADMIN:
+		ret = nvme_uring_cmd_io(ctrl, NULL, ioucmd, issue_flags, false);
+		break;
+	default:
+		ret = -ENOTTY;
+	}
+
+	return ret;
+}
+
 static int nvme_ns_uring_cmd(struct nvme_ns *ns, struct io_uring_cmd *ioucmd,
 			     unsigned int issue_flags)
 {
@@ -577,6 +602,9 @@ static int nvme_ns_uring_cmd(struct nvme_ns *ns, struct io_uring_cmd *ioucmd,
 
 	BUILD_BUG_ON(sizeof(struct nvme_uring_cmd_pdu) > sizeof(ioucmd->pdu));
 
+	if (is_ctrl_uring_cmd(ioucmd->cmd_op))
+		return nvme_ctrl_uring_cmd(ctrl, ioucmd, issue_flags);
+
 	ret = nvme_uring_cmd_checks(issue_flags);
 	if (ret)
 		return ret;
@@ -670,6 +698,22 @@ long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
 	return ret;
 }
 
+static int nvme_ns_head_ctrl_uring_cmd(struct nvme_ns *ns,
+		struct io_uring_cmd *ioucmd, unsigned int issue_flags,
+		struct nvme_ns_head *head, int srcu_idx)
+	__releases(&head->srcu)
+{
+	struct nvme_ctrl *ctrl = ns->ctrl;
+	int ret;
+
+	nvme_get_ctrl(ctrl);
+	srcu_read_unlock(&head->srcu, srcu_idx);
+	ret = nvme_ctrl_uring_cmd(ctrl, ioucmd, issue_flags);
+
+	nvme_put_ctrl(ctrl);
+	return ret;
+}
+
 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
 		unsigned int issue_flags)
 {
@@ -679,8 +723,14 @@ int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
 	struct nvme_ns *ns = nvme_find_path(head);
 	int ret = -EINVAL;
 
-	if (ns)
-		ret = nvme_ns_uring_cmd(ns, ioucmd, issue_flags);
+	if (!ns)
+		goto out_unlock;
+	if (is_ctrl_uring_cmd(ioucmd->cmd_op))
+		return nvme_ns_head_ctrl_uring_cmd(ns, ioucmd, issue_flags,
+				head, srcu_idx);
+
+	ret = nvme_ns_uring_cmd(ns, ioucmd, issue_flags);
+out_unlock:
 	srcu_read_unlock(&head->srcu, srcu_idx);
 	return ret;
 }
diff --git a/include/uapi/linux/nvme_ioctl.h b/include/uapi/linux/nvme_ioctl.h
index 0b1876aa5a59..6bcda1be6456 100644
--- a/include/uapi/linux/nvme_ioctl.h
+++ b/include/uapi/linux/nvme_ioctl.h
@@ -108,5 +108,6 @@ struct nvme_uring_cmd {
 /* io_uring async commands: */
 #define NVME_URING_CMD_IO	_IOWR('N', 0x80, struct nvme_uring_cmd)
 #define NVME_URING_CMD_IO_VEC	_IOWR('N', 0x81, struct nvme_uring_cmd)
+#define NVME_URING_CMD_ADMIN	_IOWR('N', 0x82, struct nvme_uring_cmd)
 
 #endif /* _UAPI_LINUX_NVME_IOCTL_H */
-- 
2.25.1

[PATCH 3/3] nvme: admin command uring-passthrough on controller node

[Kanchan Joshi]

Enable uring-passthrough for admin commands when issued on controller
(/dev/nvmeX) node.

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 drivers/nvme/host/core.c  | 1 +
 drivers/nvme/host/ioctl.c | 7 +++++++
 drivers/nvme/host/nvme.h  | 1 +
 3 files changed, 9 insertions(+)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 510e3860358b..23479cf866cc 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -3145,6 +3145,7 @@ static const struct file_operations nvme_dev_fops = {
 	.release	= nvme_dev_release,
 	.unlocked_ioctl	= nvme_dev_ioctl,
 	.compat_ioctl	= compat_ptr_ioctl,
+	.uring_cmd	= nvme_dev_uring_cmd,
 };
 
 static ssize_t nvme_sysfs_reset(struct device *dev,
diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c
index 0f46dc7381a0..833b285b1b04 100644
--- a/drivers/nvme/host/ioctl.c
+++ b/drivers/nvme/host/ioctl.c
@@ -736,6 +736,13 @@ int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
 }
 #endif /* CONFIG_NVME_MULTIPATH */
 
+int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
+{
+	struct nvme_ctrl *ctrl = ioucmd->file->private_data;
+
+	return nvme_ctrl_uring_cmd(ctrl, ioucmd, issue_flags);
+}
+
 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
 {
 	struct nvme_ns *ns;
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index 086ccbdd7003..26d35c557588 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -787,6 +787,7 @@ int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
 		unsigned int issue_flags);
 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
+int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
 
 extern const struct attribute_group *nvme_ns_id_attr_groups[];
 extern const struct pr_ops nvme_pr_ops;
-- 
2.25.1

Re: [PATCH 2/3] nvme: enable uring-passthrough for admin commands

[Christoph Hellwig]

On Tue, May 17, 2022 at 11:01:46AM +0530, Kanchan Joshi wrote:
> Add a new opcode NVME_URING_CMD_ADMIN that userspace can use.
> Wire up support for the case when this is issued over generic char
> device node (/dev/ngXnY).

I would really prefer not adding this to the per-namespaces nodes.
The do something controller specific on the namespace node patter is
a massive pain both in terms of code and semantics.

Re: [PATCH 1/3] nvme: helper for uring-passthrough checks

[Christoph Hellwig]

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH 2/3] nvme: enable uring-passthrough for admin commands

[Kanchan Joshi]

On Tue, May 17, 2022 at 1:45 PM Christoph Hellwig <hch@lst.de> wrote:
>
> On Tue, May 17, 2022 at 11:01:46AM +0530, Kanchan Joshi wrote:
> > Add a new opcode NVME_URING_CMD_ADMIN that userspace can use.
> > Wire up support for the case when this is issued over generic char
> > device node (/dev/ngXnY).
>
> I would really prefer not adding this to the per-namespaces nodes.
> The do something controller specific on the namespace node patter is
> a massive pain both in terms of code and semantics.

Both block (nvme0n1) and char (ng0n1) nodes allow admin-cmds to be
sent in the same way (with sync passthrough) as this patch does. So
same semantics when we support this for async/uring passthrough, IMHO.

To do passthrough read/write (and other stuff) via ns-generic
interface, uring-application will need information (such as lba
format) which can be obtained only with admin-commands. And if we
don't support what this patch does, it will have to either -

(a) get that information with sync-passthrough from /dev/ng0n1. Since
we clearly separated async-passthrough with separate opcode/struct, it
will need more bookkeeping to keep both sync and async dispatch
engines alive.

or

(b) maintain one more fd (for /dev/nvme0 node) and use that for async
dispatch whenever admin-cmd needs to be sent

Given the implications, is it not better to have this up on ns-char
node and applications get to use everything (io and admin cmds) in
async manner with async-specific opcode/struct only?

I will kill this if you still think it is a bad idea somehow, and move
the code of this to patch 3 (enable only /dev/nvme0). LMK.

Re: [PATCH 2/3] nvme: enable uring-passthrough for admin commands

[Christoph Hellwig]

Doing the controller based ioctls on a per-namespace node has been really
painful for the existing ioctl.  So I'd much prefer to avoid that, but
I'd be happy to hear more opinions.

Re: [PATCH 2/3] nvme: enable uring-passthrough for admin commands

[Kanchan Joshi]

On Wed, May 18, 2022 at 12:59 PM Christoph Hellwig <hch@lst.de> wrote:
>
> Doing the controller based ioctls on a per-namespace node has been really
> painful for the existing ioctl.  So I'd much prefer to avoid that, but
> I'd be happy to hear more opinions.

Not sure if we may get opinions before merge-window expiration.
So I am thinking to respin with only the controller node (and defer ns
one to 5.20). Hope that is fine?