Skip to contentSkip to conversation

nvme fabrics authentication fixups

PATCH In review Christoph Hellwig 3 participants 11 messages Sep 20, 2022 – Sep 21, 2022

Subsystems (3)
lore
Share & tools
lore

Christoph Hellwig

author maintainer
#0 lore raw

Hi all,

this little series fixed a debugobjects complaint with the new authentication code, and some other bits while looking at it.

Btw: Hannes, can we add you as a co-maintainer for the auth code?

Diffstat:

 drivers/nvme/target/core.c             |    1 +
 drivers/nvme/target/fabrics-cmd-auth.c |   13 ++++---------
 drivers/nvme/target/fabrics-cmd.c      |   18 ++++++++----------
 drivers/nvme/target/nvmet.h            |    7 ++++---
 include/linux/nvme.h                   |    4 ++--
 5 files changed, 19 insertions(+), 24 deletions(-)
author maintainer
#1 lore raw

PATCH 1/3 nvmet-auth: don't try to cancel a non-initialized work_struct

Currently blktests nvme/002 trips up debugobjects if CONFIG_NVME_AUTH is enabled, but authentication is not on a queue. This is because nvmet_auth_sq_free cancels sq->auth_expired_work unconditionaly, while auth_expired_work is only ever initialized if authentication is enabled for a given controller.

Fix this by calling most of what is nvmet_init_auth unconditionally when initializing the SQ, and just do the setting of the result field in the connect command handler.

Fixes db1312dd9548 ("nvmet: implement basic In-Band Authentication")
---
 drivers/nvme/target/core.c             |  1 +
 drivers/nvme/target/fabrics-cmd-auth.c | 13 ++++---------
 drivers/nvme/target/fabrics-cmd.c      |  6 ++++--
 drivers/nvme/target/nvmet.h            |  7 ++++---
 4 files changed, 13 insertions(+), 14 deletions(-)
drivers/nvme/target/core.c +1 -0
index a1345790005f4..8e3cf0c3588ce 100644
@@ -830,6 +830,7 @@ int nvmet_sq_init(struct nvmet_sq *sq)
 	}
 	init_completion(&sq->free_done);
 	init_completion(&sq->confirm_done);
+	nvmet_auth_sq_init(sq);
 
 	return 0;
 }
drivers/nvme/target/fabrics-cmd-auth.c +4 -9
index 84601e38a335a..7970a7640e585 100644
@@ -23,17 +23,12 @@ static void nvmet_auth_expired_work(struct work_struct *work)
 	sq->dhchap_tid = -1;
 }
 
-void nvmet_init_auth(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
+void nvmet_auth_sq_init(struct nvmet_sq *sq)
 {
-	u32 result = le32_to_cpu(req->cqe->result.u32);
-
 	/* Initialize in-band authentication */
-	INIT_DELAYED_WORK(&req->sq->auth_expired_work,
-			  nvmet_auth_expired_work);
-	req->sq->authenticated = false;
-	req->sq->dhchap_step = NVME_AUTH_DHCHAP_MESSAGE_NEGOTIATE;
-	result |= (u32)NVME_CONNECT_AUTHREQ_ATR << 16;
-	req->cqe->result.u32 = cpu_to_le32(result);
+	INIT_DELAYED_WORK(&sq->auth_expired_work, nvmet_auth_expired_work);
+	sq->authenticated = false;
+	sq->dhchap_step = NVME_AUTH_DHCHAP_MESSAGE_NEGOTIATE;
 }
 
 static u16 nvmet_auth_negotiate(struct nvmet_req *req, void *d)
drivers/nvme/target/fabrics-cmd.c +4 -2
index c1dfdfb92ebf0..c7e903589d377 100644
@@ -272,7 +272,8 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 	req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
 
 	if (nvmet_has_auth(ctrl))
-		nvmet_init_auth(ctrl, req);
+		req->cqe->result.u32 |=
+			cpu_to_le32((u32)NVME_CONNECT_AUTHREQ_ATR << 16);
 out:
 	kfree(d);
 complete:
@@ -333,7 +334,8 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
 
 	pr_debug("adding queue %d to ctrl %d.\n", qid, ctrl->cntlid);
 	if (nvmet_has_auth(ctrl))
-		nvmet_init_auth(ctrl, req);
+		req->cqe->result.u32 |=
+			cpu_to_le32((u32)NVME_CONNECT_AUTHREQ_ATR << 16);
 
 out:
 	kfree(d);
drivers/nvme/target/nvmet.h +4 -3
index 6ffeeb0a1c49e..dfe3894205aa7 100644
@@ -704,7 +704,7 @@ int nvmet_auth_set_key(struct nvmet_host *host, const char *secret,
 		       bool set_ctrl);
 int nvmet_auth_set_host_hash(struct nvmet_host *host, const char *hash);
 int nvmet_setup_auth(struct nvmet_ctrl *ctrl);
-void nvmet_init_auth(struct nvmet_ctrl *ctrl, struct nvmet_req *req);
+void nvmet_auth_sq_init(struct nvmet_sq *sq);
 void nvmet_destroy_auth(struct nvmet_ctrl *ctrl);
 void nvmet_auth_sq_free(struct nvmet_sq *sq);
 int nvmet_setup_dhgroup(struct nvmet_ctrl *ctrl, u8 dhgroup_id);
@@ -726,8 +726,9 @@ static inline int nvmet_setup_auth(struct nvmet_ctrl *ctrl)
 {
 	return 0;
 }
-static inline void nvmet_init_auth(struct nvmet_ctrl *ctrl,
-				   struct nvmet_req *req) {};
+static inline void nvmet_auth_sq_init(struct nvmet_sq *sq)
+{
+}
 static inline void nvmet_destroy_auth(struct nvmet_ctrl *ctrl) {};
 static inline void nvmet_auth_sq_free(struct nvmet_sq *sq) {};
 static inline bool nvmet_check_auth_status(struct nvmet_req *req)
Signature · 1 line
2.30.2
maintainer Reviewed-by on v1 1/3
#2 lore raw

Re: [PATCH 1/3] nvmet-auth: don't try to cancel a non-initialized work_struct

Quoted 10 lines · Christoph Hellwig

Currently blktests nvme/002 trips up debugobjects if CONFIG_NVME_AUTH is enabled, but authentication is not on a queue. This is because nvmet_auth_sq_free cancels sq->auth_expired_work unconditionaly, while auth_expired_work is only ever initialized if authentication is enabled for a given controller.

Fix this by calling most of what is nvmet_init_auth unconditionally when initializing the SQ, and just do the setting of the result field in the connect command handler.

Quoted 9 lines
Fixes db1312dd9548 ("nvmet: implement basic In-Band Authentication")
---
  drivers/nvme/target/core.c             |  1 +
  drivers/nvme/target/fabrics-cmd-auth.c | 13 ++++---------
  drivers/nvme/target/fabrics-cmd.c      |  6 ++++--
  drivers/nvme/target/nvmet.h            |  7 ++++---
  4 files changed, 13 insertions(+), 14 deletions(-)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		           Kernel Storage Architect
[email protected]			                  +49 911 74053 688
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
author maintainer
#4 lore raw

PATCH 2/3 nvme: improve the NVME_CONNECT_AUTHREQ* definitions

In reply to Christoph Hellwig

Mark them as unsigned so that we don't need extra casts, and define them relative to cdword0 instead of requiring extra shifts.

---
 drivers/nvme/target/fabrics-cmd.c | 6 ++----
 include/linux/nvme.h              | 4 ++--
 2 files changed, 4 insertions(+), 6 deletions(-)
drivers/nvme/target/fabrics-cmd.c +2 -4
index c7e903589d377..618f7adca70fd 100644
@@ -272,8 +272,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 	req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
 
 	if (nvmet_has_auth(ctrl))
-		req->cqe->result.u32 |=
-			cpu_to_le32((u32)NVME_CONNECT_AUTHREQ_ATR << 16);
+		req->cqe->result.u32 |= cpu_to_le32(NVME_CONNECT_AUTHREQ_ATR);
 out:
 	kfree(d);
 complete:
@@ -334,8 +333,7 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
 
 	pr_debug("adding queue %d to ctrl %d.\n", qid, ctrl->cntlid);
 	if (nvmet_has_auth(ctrl))
-		req->cqe->result.u32 |=
-			cpu_to_le32((u32)NVME_CONNECT_AUTHREQ_ATR << 16);
+		req->cqe->result.u32 |= cpu_to_le32(NVME_CONNECT_AUTHREQ_ATR);
 
 out:
 	kfree(d);
include/linux/nvme.h +2 -2
index ae53d74f3696a..050d7d0cd81b0 100644
@@ -1482,8 +1482,8 @@ struct nvmf_connect_command {
 };
 
 enum {
-	NVME_CONNECT_AUTHREQ_ASCR	= (1 << 2),
-	NVME_CONNECT_AUTHREQ_ATR	= (1 << 1),
+	NVME_CONNECT_AUTHREQ_ASCR	= (1U << 18),
+	NVME_CONNECT_AUTHREQ_ATR	= (1U << 17),
 };
 
 struct nvmf_connect_data {
Signature · 1 line
2.30.2
maintainer Reviewed-by on v1 2/3
#5 lore raw

Re: [PATCH 2/3] nvme: improve the NVME_CONNECT_AUTHREQ* definitions

Quoted 9 lines · Christoph Hellwig

Mark them as unsigned so that we don't need extra casts, and define them relative to cdword0 instead of requiring extra shifts.

---
  drivers/nvme/target/fabrics-cmd.c | 6 ++----
  include/linux/nvme.h              | 4 ++--
  2 files changed, 4 insertions(+), 6 deletions(-)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		           Kernel Storage Architect
[email protected]			                  +49 911 74053 688
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
author maintainer
#7 lore raw

PATCH 3/3 nvmet: add helpers to set the result field for connect commands

In reply to Christoph Hellwig

The code to set the result field for the admin and I/O connect commands is not only verbose and duplicated, but also violates the aliasing rules as it accesses both the u16 and u32 members in the union.

Add a little helper to sort all that out.

Fixes db1312dd9548 ("nvmet: implement basic In-Band Authentication")
---
 drivers/nvme/target/fabrics-cmd.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
drivers/nvme/target/fabrics-cmd.c +8 -10
index 618f7adca70fd..43b5bd8bb6a52 100644
@@ -198,6 +198,12 @@ static u16 nvmet_install_queue(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
 	return ret;
 }
 
+static u32 nvmet_connect_result(struct nvmet_ctrl *ctrl)
+{
+	return (u32)ctrl->cntlid |
+		(nvmet_has_auth(ctrl) ? NVME_CONNECT_AUTHREQ_ATR : 0);
+}
+
 static void nvmet_execute_admin_connect(struct nvmet_req *req)
 {
 	struct nvmf_connect_command *c = &req->cmd->connect;
@@ -269,10 +275,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 		ctrl->cntlid, ctrl->subsys->subsysnqn, ctrl->hostnqn,
 		ctrl->pi_support ? " T10-PI is enabled" : "",
 		nvmet_has_auth(ctrl) ? " with DH-HMAC-CHAP" : "");
-	req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
-
-	if (nvmet_has_auth(ctrl))
-		req->cqe->result.u32 |= cpu_to_le32(NVME_CONNECT_AUTHREQ_ATR);
+	req->cqe->result.u32 = cpu_to_le32(nvmet_connect_result(ctrl));
 out:
 	kfree(d);
 complete:
@@ -328,13 +331,8 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
 	if (status)
 		goto out_ctrl_put;
 
-	/* pass back cntlid for successful completion */
-	req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
-
 	pr_debug("adding queue %d to ctrl %d.\n", qid, ctrl->cntlid);
-	if (nvmet_has_auth(ctrl))
-		req->cqe->result.u32 |= cpu_to_le32(NVME_CONNECT_AUTHREQ_ATR);
-
+	req->cqe->result.u32 = cpu_to_le32(nvmet_connect_result(ctrl));
 out:
 	kfree(d);
 complete:
Signature · 1 line
2.30.2
maintainer Reviewed-by on v1 3/3
#8 lore raw

Re: [PATCH 3/3] nvmet: add helpers to set the result field for connect commands

Quoted 4 lines · Christoph Hellwig

The code to set the result field for the admin and I/O connect commands is not only verbose and duplicated, but also violates the aliasing rules as it accesses both the u16 and u32 members in the union.

Quoted 8 lines

Add a little helper to sort all that out.

Fixes db1312dd9548 ("nvmet: implement basic In-Band Authentication")
---
  drivers/nvme/target/fabrics-cmd.c | 18 ++++++++----------
  1 file changed, 8 insertions(+), 10 deletions(-)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		           Kernel Storage Architect
[email protected]			                  +49 911 74053 688
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
maintainer
#10 lore raw
In reply to Christoph Hellwig
Quoted 7 lines · Christoph Hellwig

Hi all,

this little series fixed a debugobjects complaint with the new authentication code, and some other bits while looking at it.

Btw: Hannes, can we add you as a co-maintainer for the auth code?

Sure.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		           Kernel Storage Architect
[email protected]			                  +49 911 74053 688
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), GF: Felix Imendörffer