nvme: restrict authentication to the admin queue
Subsystems (3)
v2 cover nvme: restrict authentication to the admin queue
From: Hannes Reinecke <[email protected]>
Hi all,
with secure concatenation the spec got more explicit to state that it would be perfectly fine to implement authentication on the admin queue only. But once a partner implemented that he found that re-authentication was failing as we continue to start authentication on all queues.
So these two patches implement this functionalify, the first one on the target (to have a testbed to test against), and the second one to the host to have it fixed.
Patches are on top of my 'secure-concat.v14' branch on kernel.org.
As usual, comments and reviews are welcome.
Changes to the original submission:
- Rebased to nvme-6.14
Hannes Reinecke (2):
nvmet: Implement 'admin_only' authentication
nvme: Do not re-authenticate queues with no prior authentication
drivers/nvme/host/auth.c | 12 ++++++++++++ drivers/nvme/target/auth.c | 11 +++++++---- drivers/nvme/target/configfs.c | 24 ++++++++++++++++++++++++ drivers/nvme/target/fabrics-cmd-auth.c | 7 +++++++ drivers/nvme/target/fabrics-cmd.c | 4 ++-- drivers/nvme/target/nvmet.h | 2 ++ 6 files changed, 54 insertions(+), 6 deletions(-)
Signature · 1 line
2.35.3
PATCH v2 1/2 nvmet: Implement 'admin_only' authentication
From: Hannes Reinecke <[email protected]>
The spec allows for authentication to run on admin queues only, and secure concatenation even requires it. So add a configfs attribute 'dhchap_admin_only' to the target configuration to allow for testing independently of secure concatenation.
--- drivers/nvme/target/auth.c | 11 +++++++---- drivers/nvme/target/configfs.c | 24 ++++++++++++++++++++++++ drivers/nvme/target/fabrics-cmd-auth.c | 7 +++++++ drivers/nvme/target/fabrics-cmd.c | 4 ++-- drivers/nvme/target/nvmet.h | 2 ++ 5 files changed, 42 insertions(+), 6 deletions(-)
drivers/nvme/target/auth.c +7 -4
@@ -190,6 +190,8 @@ u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq) ctrl->shash_id = host->dhchap_hash_id; } + ctrl->dh_admin_only = host->dhchap_admin_only; +/* Skip the 'DHHC-1:XX:' prefix */ nvme_auth_free_key (ctrl->host_key); ctrl->host_key = nvme_auth_extract_key(host->dhchap_secret + 10,
@@ -280,10 +282,11 @@ void nvmet_destroy_auth(struct nvmet_ctrl *ctrl)bool nvmet_check_auth_status (struct nvmet_req *req) { -if (req->sq->ctrl->host_key && - !req->sq->authenticated) -return false; -return true; +if (!req->sq->ctrl->host_key) +return true; +if (req->sq->qid && req->sq->ctrl->dh_admin_only) +return true; +return req->sq->authenticated; } int nvmet_auth_host_hash(struct nvmet_req *req, u8 *response,
drivers/nvme/target/configfs.c +24 -0
@@ -2219,11 +2219,34 @@ static ssize_t nvmet_host_dhchap_dhgroup_store(struct config_item *item,CONFIGFS_ATTR (nvmet_host_, dhchap_dhgroup); +static ssize_t nvmet_host_dhchap_admin_only_show (struct config_item *item, +char *page) +{ +struct nvmet_host *host =to_host (item); + +return sprintf (page,"%d\n" , host->dhchap_admin_only); +} + +static ssize_t nvmet_host_dhchap_admin_only_store (struct config_item *item, +const char *page,size_t count) +{ +struct nvmet_host *host =to_host (item); +bool val; + +if (kstrtobool (page, &val)) +return -EINVAL ; + host->dhchap_admin_only = val; +return count; +} + +CONFIGFS_ATTR (nvmet_host_, dhchap_admin_only); +static struct configfs_attribute *nvmet_host_attrs[] = { &nvmet_host_attr_dhchap_key, &nvmet_host_attr_dhchap_ctrl_key, &nvmet_host_attr_dhchap_hash, &nvmet_host_attr_dhchap_dhgroup, + &nvmet_host_attr_dhchap_admin_only,NULL , };#endif /* CONFIG_NVME_TARGET_AUTH */
@@ -2263,6 +2286,7 @@ static struct config_group *nvmet_hosts_make_group(struct config_group *group,#ifdef CONFIG_NVME_TARGET_AUTH /* Default to SHA256 */ host->dhchap_hash_id =NVME_AUTH_HASH_SHA256 ; + host->dhchap_admin_only = false;#endif config_group_init_type_name (&host->group, name, &nvmet_host_type);
drivers/nvme/target/fabrics-cmd-auth.c +7 -0
@@ -62,6 +62,7 @@ static u8 nvmet_auth_negotiate(struct nvmet_req *req, void *d)return NVME_AUTH_DHCHAP_FAILURE_CONCAT_MISMATCH ; } ctrl->concat = true; + ctrl->dh_admin_only = true; }if (data->napd !=1 )
@@ -253,6 +254,12 @@ void nvmet_execute_auth_send(struct nvmet_req *req) offsetof(struct nvmf_auth_send_command , tl); goto done; } +if (req->sq->qid && ctrl->dh_admin_only) { +pr_debug ("%s: ctrl %d qid %d reject authentication on I/O queues\n" , + __func__, ctrl->cntlid, req->sq->qid); + status =NVME_SC_INVALID_OPCODE |NVME_STATUS_DNR ; + goto done; + }if (!nvmet_check_transfer_len (req, tl)) {pr_debug ("%s: transfer length mismatch (%u)\n" , __func__, tl);return ;
drivers/nvme/target/fabrics-cmd.c +2 -2
@@ -239,8 +239,8 @@ static u32 nvmet_connect_result(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq)bool needs_auth =nvmet_has_auth (ctrl, sq);key_serial_t keyid =nvmet_queue_tls_keyid (sq); -/* Do not authenticate I/O queues for secure concatenation */ -if (ctrl->concat && sq->qid) +/* Disable authentication on I/O queues if requested */ +if (ctrl->dh_admin_only && sq->qid) needs_auth = false;if (keyid)
drivers/nvme/target/nvmet.h +2 -0
@@ -301,6 +301,7 @@ struct nvmet_ctrl {u8 dh_gid;u8 *dh_key;size_t dh_keysize; +bool dh_admin_only;#endif #ifdef CONFIG_NVME_TARGET_TCP_TLS struct key *tls_key;
@@ -379,6 +380,7 @@ struct nvmet_host {u8 dhchap_ctrl_key_hash;u8 dhchap_hash_id;u8 dhchap_dhgroup_id; +bool dhchap_admin_only; };static inline struct nvmet_host *to_host (struct config_item *item)
Signature · 1 line
2.35.3
PATCH v2 2/2 nvme: Do not re-authenticate queues with no prior authentication
In reply to [email protected]From: Hannes Reinecke <[email protected]>
When sending 'connect' the queues can figure out from the return code whether authentication is required or not. But reauthentication doesn't disconnect the queues, so this check is not available.
Rather we need to check whether the queue had been authenticated initially to figure out if we need to reauthenticate.
--- drivers/nvme/host/auth.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
drivers/nvme/host/auth.c +12 -0
@@ -31,6 +31,7 @@ struct nvme_dhchap_queue_context {u32 s1;u32 s2;bool bi_directional; +bool authenticated;u16 transaction;u8 status;u8 dhgroup_id;
@@ -927,12 +928,14 @@ static void nvme_queue_auth_work(struct work_struct *work) }if (!ret) { chap->error =0 ; + chap->authenticated = true;if (ctrl->opts->concat && (ret =nvme_auth_secure_concat (ctrl, chap))) {dev_warn (ctrl->device,"%s: qid %d failed to enable secure concatenation\n" , __func__, chap->qid); chap->error = ret; + chap->authenticated = false; }return ; }
@@ -1021,6 +1024,14 @@ static void nvme_ctrl_auth_work(struct work_struct *work)return ;for (q =1 ; q < ctrl->queue_count; q++) { +/* +* Skip re-authentication if the queue had +* not been authenticated initially. +*/ +if (!ctrl->dhchap_ctxs[q].authenticated) { + ctrl->dhchap_ctxs[q].error =0 ; +continue ; + } ret =nvme_auth_negotiate (ctrl, q);if (ret) { dev_warn(ctrl->device,
@@ -1074,6 +1085,7 @@ int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl) chap = &ctrl->dhchap_ctxs[i]; chap->qid = i; chap->ctrl = ctrl; + chap->authenticated = false;INIT_WORK (&chap->auth_work, nvme_queue_auth_work); }
Signature · 1 line
2.35.3
Sagi Grimberg
maintainerRe: [PATCH 1/2] nvmet: Implement 'admin_only' authentication
In reply to [email protected]Quoted 6 lines · On 24/01/2025 13:47, [email protected] wrote:
From: Hannes Reinecke <[email protected]>
The spec allows for authentication to run on admin queues only, and secure concatenation even requires it. So add a configfs attribute '
dhchap_admin_only' to the target configuration to allow for testing independently of secure concatenation.
Why do we want it conditionally? why not always?
Hannes Reinecke
maintainerRe: [PATCH 1/2] nvmet: Implement 'admin_only' authentication
Quoted 4 lines · Sagi Grimberg
On 24/01/2025 13:47, [email protected] wrote:
Quoted 8 lines
Quoted 6 lines
From: Hannes Reinecke <[email protected]>
The spec allows for authentication to run on admin queues only, and secure concatenation even requires it. So add a configfs attribute '
dhchap_admin_only' to the target configuration to allow for testing independently of secure concatenation.Why do we want it conditionally? why not always?
Because we did support it originally, so I thought to play it safe. Plus it'll cause a regression with the host implementation if applied on its own.
But if you say so ...
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
Re: [PATCH 2/2] nvme: Do not re-authenticate queues with no prior authentication
In reply to [email protected]Sagi Grimberg
maintainerRe: [PATCH 1/2] nvmet: Implement 'admin_only' authentication
In reply to Hannes ReineckeQuoted 13 lines · Sagi Grimberg
Quoted 11 lines · On 24/01/2025 13:47, [email protected] wrote:
Quoted 6 lines
From: Hannes Reinecke <[email protected]>
The spec allows for authentication to run on admin queues only, and secure concatenation even requires it. So add a configfs attribute '
dhchap_admin_only' to the target configuration to allow for testing independently of secure concatenation.
Quoted 7 lines
Quoted 1 line
Why do we want it conditionally? why not always?
Because we did support it originally, so I thought to play it safe. Plus it'll cause a regression with the host implementation if applied on its own.
But if you say so ...
I'd like to avoid the extra configuration if possible, however we cannot break existing hosts.
Is there a way to not authenticate I/O queues but still allow it?
Hannes Reinecke
maintainerRe: [PATCH 1/2] nvmet: Implement 'admin_only' authentication
Quoted 19 lines · Sagi Grimberg
Quoted 19 lines · Sagi Grimberg
Quoted 12 lines · On 24/01/2025 13:47, [email protected] wrote:
Quoted 6 lines
From: Hannes Reinecke <[email protected]>
The spec allows for authentication to run on admin queues only, and secure concatenation even requires it. So add a configfs attribute '
dhchap_admin_only' to the target configuration to allow for testing independently of secure concatenation.Why do we want it conditionally? why not always?
Because we did support it originally, so I thought to play it safe. Plus it'll cause a regression with the host implementation if applied on its own.
Quoted 6 lines
Quoted 1 line
But if you say so ...
I'd like to avoid the extra configuration if possible, however we cannot break existing hosts.
Is there a way to not authenticate I/O queues but still allow it?
Not really. Authentication is driven by the AUTHREQ_ATR / AUTHREQ_ASCR return status code in the connect response. And that is sent by the target, so the target has to decide whether it wants authentication on the queue.
Sadly there is no status for 'whatever, I don't care' ...
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
Hannes Reinecke
maintainerQuoted 8 lines · Kamaljit Singh
Hi Hannes,
One of the key TLS handshake issues we faced is with SHA384 and I see that you had requested this gnutls fix ~2 years ago. I’m curious how you got around that.
https://gitlab.com/gnutls/gnutls/-/issues/386 <https://gitlab.com/ gnutls/gnutls/-/issues/386>
We had to hardcode gnutls v3.8.9 to change its default SHA256 to SHA384 for using the appropriate binder length.
Yeah, that is quite annoying.
Guess I'll have to bite the bullet and fix gnutls. Especially as even the maintainer acknowledged that it's an issue.
BTW, the secure concatenation fixes are now merged with the nvme-6.15 branch at git.infradead.org/nvme.git, and I've opened a pull request for blktest:
https://github.com/osandov/blktests/pull/158
which should test possible combinations.
No test for the binder hash, I'm afraid; if you have ideas how to test
this I'd be happy to take them.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich