nvmet: unique discovery subsystem
Subsystems (2)
v2 cover nvmet: unique discovery subsystem
Hi all,
here's my next attempt to support unique discovery subsystems. The main idea is to make the subsystem type configurable; if it's being set to 'discovery' it'll replace the static discovery subsystem. The admin then need to configure the subsystem as normal by linking it into the ports where the discovery subsystem should be visible. And the discovery log then includes all configured ports to all configured subsystems, as usual modified by whether the host may access this subsystem or not.
As usual, comments and reviews are welcome.
Changes to the original submission:
- Include all subsystems in the discovery log output
Hannes Reinecke (3):
nvmet: check for subsystem type in nvmet_find_get_subsys()
nvmet: make the subsystem type configurable
nvmet: include all configured ports in discovery log page for unique
discover controller
drivers/nvme/target/configfs.c | 65 ++++++++++++++++++++++++++ drivers/nvme/target/core.c | 23 ++++++--- drivers/nvme/target/discovery.c | 83 +++++++++++++++++++++++++-------- drivers/nvme/target/nvmet.h | 3 ++ 4 files changed, 148 insertions(+), 26 deletions(-)
Signature · 1 line
2.29.2
PATCH v2 1/3 nvmet: check for subsystem type in nvmet_find_get_subsys()
When looking for a subsystem we have two ways of finding the
subsystem: either looking for the subsystem NQN itself, or check the type of the subsystem when looking for a discovery controller. This patch implements this check, and also moves the check for the static discovery controller to the end such that we can return unique discovery controllers.
--- drivers/nvme/target/core.c | 20 ++++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-)
drivers/nvme/target/core.c +14 -6
@@ -1508,12 +1508,6 @@ static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,if (!port)return NULL ; -if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn)) { -if (!kref_get_unless_zero (&nvmet_disc_subsys->ref)) -return NULL ; -return nvmet_disc_subsys; - } -down_read (&nvmet_config_sem); list_for_each_entry(p, &port->subsystems, entry) { if (!strncmp(p->subsys->subsysnqn, subsysnqn,
@@ -1523,8 +1517,22 @@ static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,up_read (&nvmet_config_sem);return p->subsys; } +if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn) && +nvmet_is_disc_subsys (p->subsys)) { +if (!kref_get_unless_zero (&p->subsys->ref)) +break ; +up_read (&nvmet_config_sem); +return p->subsys; + } }up_read (&nvmet_config_sem); + +if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn)) { +if (!kref_get_unless_zero (&nvmet_disc_subsys->ref)) +return NULL ; +return nvmet_disc_subsys; + } +return NULL ; }
Signature · 1 line
2.29.2
Christoph Hellwig
maintainerRe: [PATCH 1/3] nvmet: check for subsystem type in nvmet_find_get_subsys()
Quoted 30 lines · Hannes Reinecke
When looking for a subsystem we have two ways of finding the
subsystem: either looking for the subsystem NQN itself, or check the type of the subsystem when looking for a discovery controller. This patch implements this check, and also moves the check for the static discovery controller to the end such that we can return unique discovery controllers.Signed-off-by Hannes Reinecke <[email protected]>--- drivers/nvme/target/core.c | 20 ++++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-)drivers/nvme/target/core.c +0
-6@@ -1508,12 +1508,6 @@ static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,if (!port)return NULL ; -if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn)) { -if (!kref_get_unless_zero (&nvmet_disc_subsys->ref)) -return NULL ; -return nvmet_disc_subsys; - } -down_read (&nvmet_config_sem); list_for_each_entry(p, &port->subsystems, entry) { if (!strncmp(p->subsys->subsysnqn, subsysnqn,
drivers/nvme/target/core.c static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port, · line 1521
Quoted 5 lines
drivers/nvme/target/core.c +2
-0@@ -1523,8 +1517,22 @@ static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,up_read (&nvmet_config_sem);return p->subsys; } +if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn) && +nvmet_is_disc_subsys (p->subsys)) {
I don't get this. The strcmp is on the passed in subqnen that we are looking for. So with this patch we now do the fairly expensive string compare for every subsystem in the loop, and the just retuns the first discovery subsystems when we look for the well known discovery subsystem. As-is there is only one, so this is pointless. If we add discovery subsystems with different names (as I assume the later patches do), this changes beahvior in a rather unexpected way.
Re: [PATCH 1/3] nvmet: check for subsystem type in nvmet_find_get_subsys()
Quoted 39 lines · Christoph Hellwig
Quoted 30 lines · Hannes Reinecke
When looking for a subsystem we have two ways of finding the
subsystem: either looking for the subsystem NQN itself, or check the type of the subsystem when looking for a discovery controller. This patch implements this check, and also moves the check for the static discovery controller to the end such that we can return unique discovery controllers.Signed-off-by Hannes Reinecke <[email protected]>--- drivers/nvme/target/core.c | 20 ++++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-)drivers/nvme/target/core.c +0
-6@@ -1508,12 +1508,6 @@ static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,if (!port)return NULL ; -if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn)) { -if (!kref_get_unless_zero (&nvmet_disc_subsys->ref)) -return NULL ; -return nvmet_disc_subsys; - } -down_read (&nvmet_config_sem); list_for_each_entry(p, &port->subsystems, entry) { if (!strncmp(p->subsys->subsysnqn, subsysnqn,drivers/nvme/target/core.c static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port, · line 1521
Quoted 5 lines
drivers/nvme/target/core.c +2
-0@@ -1523,8 +1517,22 @@ static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,up_read (&nvmet_config_sem);return p->subsys; } +if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn) && +nvmet_is_disc_subsys (p->subsys)) {
Quoted 7 lines
I don't get this. The strcmp is on the passed in subqnen that we are looking for. So with this patch we now do the fairly expensive string compare for every subsystem in the loop, and the just retuns the first discovery subsystems when we look for the well known discovery subsystem. As-is there is only one, so this is pointless. If we add discovery subsystems with different names (as I assume the later patches do), this changes beahvior in a rather unexpected way.
Will have to check; might be that this is a left-over from the previous attempt of handling discovery subsystems just like any other subsystem.
However, this is just part of a larger patchset to support unique discovery controller NQNs. And the overarching question here is:
Do we want to support unique discovery controller NQNs in nvmet?
Previously there was a rather strict policy of implementing only the bare necessities in nvmet, and unique discovery controller NQNs is arguably not a necessary thing. So the whole discussion gets pretty pointless if we decide _not_ to support it after all.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Christoph Hellwig
maintainerRe: [PATCH 1/3] nvmet: check for subsystem type in nvmet_find_get_subsys()
Quoted 5 lines · Hannes Reinecke
Do we want to support unique discovery controller NQNs in nvmet?
Previously there was a rather strict policy of implementing only the bare necessities in nvmet, and unique discovery controller NQNs is arguably not a necessary thing.
I don't think there has ever been a policy of only bare necessities. But more one of checking if features are useful for nvmet.
Given that we basically need unique discovery controllers to usefull support authentication it seems like we have to support them, even if the feature on its down does not look all that useful.
Re: [PATCH 1/3] nvmet: check for subsystem type in nvmet_find_get_subsys()
In reply to Hannes ReineckeQuoted 8 lines · Hannes Reinecke
Will have to check; might be that this is a left-over from the previous attempt of handling discovery subsystems just like any other subsystem.
However, this is just part of a larger patchset to support unique discovery controller NQNs. And the overarching question here is:
Do we want to support unique discovery controller NQNs in nvmet?
Previously there was a rather strict policy of implementing only the bare necessities in nvmet, and unique discovery controller NQNs is arguably not a necessary thing.
I think the question is: do we want to support unique discovery controller NQNs at all?
Quoted 1 line
So the whole discussion gets pretty pointless if we decide _not_ to support it after all.
I'm really concerned about the idea of having NO support for unique discovery controller NQNs in nvmet. If Linux is going to support unique discovery NQNs in nvme/host then I think we need to have support for unique discovery NQNs in nvme/target.
Without this, there is just no way to test anything. I want the ability to test NVMe protocol features like this w/out having to rely on external hardware, if possible.
/John
Quoted 3 lines
Cheers,
Hannes
PATCH v2 2/3 nvmet: make the subsystem type configurable
In reply to Hannes ReineckeMake the subsystem type configurable to allow for unique discovery subsystems by changing the subsystem type to 'discovery'.
--- drivers/nvme/target/configfs.c | 65 +++++++++++++++++++++++++++++++++ drivers/nvme/target/core.c | 3 ++ drivers/nvme/target/discovery.c | 20 +++++++++- drivers/nvme/target/nvmet.h | 3 ++ 4 files changed, 89 insertions(+), 2 deletions(-)
drivers/nvme/target/configfs.c +65 -0
@@ -930,6 +930,9 @@ static void nvmet_port_subsys_drop_link(struct config_item *parent, found:list_del (&p->entry); +/* Reset to static discovery controller */ +if (nvmet_disc_subsys == subsys) +nvmet_set_disc_subsys (NULL );nvmet_port_del_ctrls (port, subsys);nvmet_port_disc_changed (port, subsys);
@@ -1302,6 +1305,67 @@ static ssize_t nvmet_subsys_attr_model_store(struct config_item *item, }CONFIGFS_ATTR (nvmet_subsys_, attr_model); +static const struct nvmet_type_name_map nvmet_subsys_type_map[] = { + {NVME_NQN_DISC ,"referral" }, + {NVME_NQN_NVME ,"nvme" }, + {NVME_NQN_CURR ,"discovery" }, +}; + +static ssize_t nvmet_subsys_attr_type_show (struct config_item *item, +char *page) +{ +u8 type =to_subsys (item)->type; +int i; + +for (i =1 ; i <ARRAY_SIZE (nvmet_subsys_type_map); i++) { +if (nvmet_subsys_type_map[i].type == type) +return snprintf (page,PAGE_SIZE ,"%s\n" , + nvmet_subsys_type_map[i].name); + } + +return snprintf (page,PAGE_SIZE ,"\n" ); +} + +static ssize_t nvmet_subsys_attr_type_store (struct config_item *item, +const char *page,size_t count) +{ +struct nvmet_subsys *subsys =to_subsys (item); +int i; + +if (subsys->subsys_discovered) +return -EACCES ; + +for (i =0 ; i <ARRAY_SIZE (nvmet_subsys_type_map); i++) { +if (sysfs_streq (page, nvmet_subsys_type_map[i].name)) + goto found; + } + +pr_err ("Invalid value '%s' for subsystem type\n" , page); +return -EINVAL ; + +found: +down_write (&nvmet_config_sem); +if (nvmet_subsys_type_map[i].type ==NVME_NQN_CURR ) { +if (nvmet_has_unique_disc_subsys ()) { +pr_err ("unique discovery subsystem already present\n" ); +return -EINVAL ; + } +if (!xa_empty (&subsys->namespaces)) { +pr_err ("discovery subsystem cannot have namespaces\n" ); +return -EINVAL ; + } +nvmet_subsys_del_ctrls (nvmet_disc_subsys); +nvmet_set_disc_subsys (subsys); + }else if (subsys->type ==NVME_NQN_CURR ) +nvmet_set_disc_subsys (NULL ); + + subsys->type = nvmet_subsys_type_map[i].type; +up_write (&nvmet_config_sem); + +return count; +} +CONFIGFS_ATTR (nvmet_subsys_, attr_type); +#ifdef CONFIG_BLK_DEV_INTEGRITY static ssize_t nvmet_subsys_attr_pi_enable_show (struct config_item *item,char *page)
@@ -1331,6 +1395,7 @@ static struct configfs_attribute *nvmet_subsys_attrs[] = { &nvmet_subsys_attr_attr_cntlid_min, &nvmet_subsys_attr_attr_cntlid_max, &nvmet_subsys_attr_attr_model, + &nvmet_subsys_attr_attr_type, #ifdef CONFIG_BLK_DEV_INTEGRITY &nvmet_subsys_attr_attr_pi_enable, #endif
drivers/nvme/target/core.c +3 -0
@@ -1527,6 +1527,9 @@ static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port, }up_read (&nvmet_config_sem); +if (nvmet_has_unique_disc_subsys ()) +return NULL ; +if (!strcmp (NVME_DISC_SUBSYS_NAME , subsysnqn)) {if (!kref_get_unless_zero (&nvmet_disc_subsys->ref))return NULL ;
drivers/nvme/target/discovery.c +18 -2
@@ -9,9 +9,23 @@#include "nvmet.h" struct nvmet_subsys *nvmet_disc_subsys; +static struct nvmet_subsys *_nvmet_disc_subsys;static u64 nvmet_genctr; +bool nvmet_has_unique_disc_subsys (void ) +{ +return (_nvmet_disc_subsys != nvmet_disc_subsys); +} + +void nvmet_set_disc_subsys (struct nvmet_subsys *subsys) +{ +if (subsys) + nvmet_disc_subsys = subsys; +else + nvmet_disc_subsys = _nvmet_disc_subsys; +} +static void __nvmet_disc_changed (struct nvmet_port *port,struct nvmet_ctrl *ctrl) {
@@ -393,12 +407,14 @@ u16 nvmet_parse_discovery_cmd(struct nvmet_req *req)int __init nvmet_init_discovery (void ) { - nvmet_disc_subsys = + _nvmet_disc_subsys =nvmet_subsys_alloc (NVME_DISC_SUBSYS_NAME ,NVME_NQN_CURR ); + nvmet_disc_subsys = _nvmet_disc_subsys;return PTR_ERR_OR_ZERO (nvmet_disc_subsys); }void nvmet_exit_discovery (void ) { -nvmet_subsys_put (nvmet_disc_subsys); + nvmet_disc_subsys =NULL ; +nvmet_subsys_put (_nvmet_disc_subsys); }
drivers/nvme/target/nvmet.h +3 -0
@@ -556,6 +556,9 @@ void nvmet_exit_discovery(void);extern struct nvmet_subsys *nvmet_disc_subsys;extern struct rw_semaphore nvmet_config_sem; +bool nvmet_has_unique_disc_subsys (void ); +void nvmet_set_disc_subsys (struct nvmet_subsys *subsys); +extern u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS +1 ];extern u64 nvmet_ana_chgcnt;extern struct rw_semaphore nvmet_ana_sem;
Signature · 1 line
2.29.2
Christoph Hellwig
maintainerRe: [PATCH 2/3] nvmet: make the subsystem type configurable
Quoted 3 lines · Hannes Reinecke
Make the subsystem type configurable to allow for unique discovery subsystems by changing the subsystem type to 'discovery'.
Does it make sense to change the type? Or should we have a new top-level interface to create and configure discovery subsystems from the start?
If we had e.g.
/sys/kernel/config/nvmet/discovery-subsystems/
we could start that out with a pre-filled well known discovery subsystem but also allow disabling that for example if we really want.
Re: [PATCH 2/3] nvmet: make the subsystem type configurable
Quoted 8 lines · Christoph Hellwig
Quoted 3 lines · Hannes Reinecke
Make the subsystem type configurable to allow for unique discovery subsystems by changing the subsystem type to 'discovery'.
Does it make sense to change the type? Or should we have a new top-level interface to create and configure discovery subsystems from the start?
Quoted 6 lines
If we had e.g.
/sys/kernel/config/nvmet/discovery-subsystems/we could start that out with a pre-filled well known discovery subsystem but also allow disabling that for example if we really want.
I actually looked into that with my previous attempt, and really found no good way of having an entry which is
a) pre-filled
and
b) removable by the admin later on.
From what I've seen all pre-filled entries are static; I haven't found a way to do pre-filled dynamic entries.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Christoph Hellwig
maintainerRe: [PATCH 2/3] nvmet: make the subsystem type configurable
Quoted 7 lines · Hannes Reinecke
I actually looked into that with my previous attempt, and really found no good way of having an entry which is
a) pre-filled
and
b) removable by the admin later on.
From what I've seen all pre-filled entries are static; I haven't found a way to do pre-filled dynamic entries.
I don't think we need to remove the well known discovery subsystem. But we can allow removing all entries from it and thus reporting an empty log.
Re: [PATCH 2/3] nvmet: make the subsystem type configurable
Quoted 9 lines · Christoph Hellwig
Quoted 7 lines · Hannes Reinecke
I actually looked into that with my previous attempt, and really found no good way of having an entry which is
a) pre-filled
and
b) removable by the admin later on.
From what I've seen all pre-filled entries are static; I haven't found a
way to do pre-filled dynamic entries.
Quoted 3 lines
I don't think we need to remove the well known discovery subsystem. But we can allow removing all entries from it and thus reporting an empty log.
Hmm.
We still would need to pre-fill these entries when creating subsystems. Bringing us back to square one, namely there is no way to create dynamic pre-filled entries.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Sagi Grimberg
maintainerRe: [PATCH 2/3] nvmet: make the subsystem type configurable
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
I actually looked into that with my previous attempt, and really found no good way of having an entry which is
a) pre-filled
and
b) removable by the admin later on.
From what I've seen all pre-filled entries are static; I haven't found a way to do pre-filled dynamic entries.
Quoted 7 lines
Quoted 3 lines
I don't think we need to remove the well known discovery subsystem. But we can allow removing all entries from it and thus reporting an empty log.
Hmm.
We still would need to pre-fill these entries when creating subsystems. Bringing us back to square one, namely there is no way to create dynamic pre-filled entries.
Why do we need to remove it altogether? Why not just make these subsystems attach to port(s) like any other subsystem?
Re: [PATCH 2/3] nvmet: make the subsystem type configurable
Quoted 13 lines · Sagi Grimberg
Quoted 13 lines
Quoted 12 lines
Quoted 8 lines
I actually looked into that with my previous attempt, and really found no good way of having an entry which is
a) pre-filled
and
b) removable by the admin later on.
From what I've seen all pre-filled entries are static; I haven't found a way to do pre-filled dynamic entries.I don't think we need to remove the well known discovery subsystem. But we can allow removing all entries from it and thus reporting an empty log.
Quoted 6 lines
Quoted 3 lines
Hmm.
We still would need to pre-fill these entries when creating subsystems. Bringing us back to square one, namely there is no way to create dynamic pre-filled entries.Why do we need to remove it altogether? Why not just make these subsystems attach to port(s) like any other subsystem?
That's what I tried in my first attempt, which got rejected as it does break existing installations.
We could introduce a module option for this (and I think that's what I did in my first round).
And yes, ideally I would like to have the discovery subsystem as a 'normal' subsystem just like the others; should be trivial to expose the current one in configfs. But then changing the subsystem NQN becomes awkward:
- Either we create a new subsystem with the unique NQN, but then we have to change the type later on (as creation happens with a simple 'mkdir', and one cannot really pass arguments to that).
- Or we have a distinct discovery subsystem type (ie the idea Christoph mentioned), but then we have to cross-check the directory names against the existing one in the 'normal' subsystem directory. And, of course, we still end up with a defunct discovery subsystem if we create a unique discovery subsystem.
Alternatively: if we're going to break existing configurations anyway (or, rather, have to modify nvmetcli to handle the new discovery mechanism), why do we need to start off with a default discovery subsystem? Can't we require the user to create one?
We would be requiring a module option here, but that looks like the best approach here as we won't have to deal with stale subsystems ...
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
Re: [PATCH 2/3] nvmet: make the subsystem type configurable
Quoted 7 lines · Hannes Reinecke
That's what I tried in my first attempt, which got rejected as it does break existing installations.
We could introduce a module option for this (and I think that's what I did in my first round).
And yes, ideally I would like to have the discovery subsystem as a 'normal' subsystem just like the others; should be trivial to expose the current one in configfs. But then changing the subsystem NQN becomes awkward: > - Either we create a new subsystem with the unique NQN, but then we have to change the type later on (as creation happens with a simple 'mkdir', and one cannot really pass arguments to that).
- Or we have a distinct discovery subsystem type (ie the idea Christoph mentioned), but then we have to cross-check the
directory names against the existing one in the 'normal' subsystem directory.
I don't think we need to change the discovery subsystem. I think what we want, in the simple case, is 2 discovery subsystems: 1) the existing well known discovery controller with the well-known discovery NQN, 2) a unique discovery controller with a unique discovery NQN.
The first well known discovery controller exists today and doesn't need to be configurable (unless you're going to do something stupid like support fabric authenticate with the well known discovery NQN).
The second unique discovery controller can be configured just like any other NVM subsystem, but its discovery log pages show up in the well known discovery controller log page with Subsystem Type (SUBTYPE): 03h - as defined by TP-8014.
So the well known discovery controller always returns log page entries for three types of log pages:
NVME_NQN_DISC, "referral"NVME_NQN_NVME, "nvme"NVME_NQN_CURR, "discovery"
When there is no unique discovery controller configured the NVME_NQN_CURR log page has no entries and everything works as it does today.
When someone configures a unique discovery NQN, the controller is created and all of the NVME_NQN_NVME log page entries move
to the new unique discovery controller while the NVME_NQN_CURR log page entries on the well-known discovery controller report only the NVME_NQN_CURR and NVME_NQN_DISC log page entires.
This is one possible approach.
Quoted 1 line
And, of course, we still end up with a defunct discovery subsystem if we create a unique discovery subsystem.
Why do we have to defunct the discovery subsystem if we create a unique discover subsystem?
Quoted 3 lines
Alternatively: if we're going to break existing configurations anyway (or, rather, have to modify nvmetcli to handle the new discovery mechanism), why do we need to start off with a default discovery subsystem?
Can't we require the user to create one?
I would say yes. The user has to create the unique discovery controller and assign it a unique NQN
Quoted 1 line
We would be requiring a module option here, but that looks like the best approach here as we won't have to deal with stale subsystems ...
I don't understand why there needs to be any stale subsystems. If we add a optional subsystem (the unique discovery controller) we don't need to take anything away.
/John
Quoted 3 lines
Cheers,
Hannes
Re: [PATCH 2/3] nvmet: make the subsystem type configurable
In reply to Christoph HellwigQuoted 5 lines · Christoph Hellwig
I don't think we need to remove the well known discovery subsystem. But we can allow removing all entries from it and thus reporting an empty log.
I agree. This was exactly my vision for how things should work with TP-8013 and TP-8014
/John
PATCH v2 3/3 nvmet: include all configured ports in discovery log page for unique discover controller
In reply to Hannes ReineckeWhen an unique discovery controller is configured we should be reporting all configured ports, and not just those which are reachable from the current port.
--- drivers/nvme/target/discovery.c | 63 +++++++++++++++++++++++---------- 1 file changed, 45 insertions(+), 18 deletions(-)
drivers/nvme/target/discovery.c +45 -18
@@ -160,12 +160,24 @@ static size_t discovery_log_entries(struct nvmet_req *req)struct nvmet_ctrl *ctrl = req->sq->ctrl;struct nvmet_subsys_link *p;struct nvmet_port *r; -size_t entries =1 ; - -list_for_each_entry (p, &req->port->subsystems, entry) { -if (!nvmet_host_allowed (p->subsys, ctrl->hostnqn)) -continue ; +size_t entries =0 ; + +if (nvmet_has_unique_disc_subsys ()) { +list_for_each_entry (r, nvmet_ports, global_entry) { +list_for_each_entry (p, &r->subsystems, entry) { +if (!nvmet_host_allowed (p->subsys, + ctrl->hostnqn)) +continue ; + entries++; + } + } + }else { entries++; +list_for_each_entry (p, &req->port->subsystems, entry) { +if (!nvmet_host_allowed (p->subsys, ctrl->hostnqn)) +continue ; + entries++; + } }list_for_each_entry (r, &req->port->referrals, entry) entries++;
@@ -220,23 +232,38 @@ static void nvmet_execute_disc_get_log_page(struct nvmet_req *req) } hdr = buffer; -nvmet_set_disc_traddr (req, req->port, traddr); - -nvmet_format_discovery_entry (hdr, req->port, - nvmet_disc_subsys->subsysnqn, - traddr,NVME_NQN_CURR , numrec); - numrec++; - -list_for_each_entry (p, &req->port->subsystems, entry) { -if (!nvmet_host_allowed (p->subsys, ctrl->hostnqn)) -continue ; +if (!nvmet_has_unique_disc_subsys ()) { +nvmet_set_disc_traddr (req, req->port, traddr);nvmet_format_discovery_entry (hdr, req->port, - p->subsys->subsysnqn, traddr, -NVME_NQN_NVME , numrec); + nvmet_disc_subsys->subsysnqn, + traddr,NVME_NQN_CURR , numrec); numrec++; - } +list_for_each_entry (p, &req->port->subsystems, entry) { +if (!nvmet_host_allowed (p->subsys, ctrl->hostnqn)) +continue ; + +nvmet_format_discovery_entry (hdr, req->port, + p->subsys->subsysnqn, traddr, +NVME_NQN_NVME , numrec); + numrec++; + } + }else { +list_for_each_entry (r, nvmet_ports, global_entry) { +nvmet_set_disc_traddr (req, r, traddr); + +list_for_each_entry (p, &r->subsystems, entry) { +if (!nvmet_host_allowed (p->subsys, + ctrl->hostnqn)) +continue ; +nvmet_format_discovery_entry (hdr, r, + p->subsys->subsysnqn, traddr, + p->subsys->type, numrec); + numrec++; + } + } + }list_for_each_entry (r, &req->port->referrals, entry) { nvmet_format_discovery_entry(hdr, r, NVME_DISC_SUBSYS_NAME,
Signature · 1 line
2.29.2
Christoph Hellwig
maintainerRe: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 3 lines · Hannes Reinecke
When an unique discovery controller is configured we should be reporting all configured ports, and not just those which are reachable from the current port.
Why?
Re: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 6 lines · Christoph Hellwig
Quoted 3 lines · Hannes Reinecke
When an unique discovery controller is configured we should be reporting all configured ports, and not just those which are reachable from the current port.
Why?
To make it configurable.
Unique discovery controllers show up in configfs just like any other subsystems.
And with that we need to clarify the relationship between the discovery subsystem and the other subsystems, ie which subsystems should be presented by this discovery subsystem.
Linking the discovery subsystem into a given port makes it obvious that
a) this port will be presenting a discovery subsystem
and
b) that the discovery subsystem will be presenting all subsystems configured on that port.
The built-in mechanism for discovery subsystems was okay as long as the discovery subsystem was built-in, too. But with this patchset we're moving to an explicit configuration.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Christoph Hellwig
maintainerRe: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 13 lines · Hannes Reinecke
To make it configurable.
Unique discovery controllers show up in configfs just like any other subsystems.
And with that we need to clarify the relationship between the discovery subsystem and the other subsystems, ie which subsystems should be presented by this discovery subsystem.Linking the discovery subsystem into a given port makes it obvious that
a) this port will be presenting a discovery subsystem
and
b) that the discovery subsystem will be presenting all subsystems configured on that port.
Quoted 3 lines
The built-in mechanism for discovery subsystems was okay as long as the discovery subsystem was built-in, too.
But with this patchset we're moving to an explicit configuration.
Shouldn't we just require anything to be manually listed for this case similar to how we configure referrals for the well known discovery controller?
Re: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 14 lines · Christoph Hellwig
Quoted 13 lines · Hannes Reinecke
To make it configurable.
Unique discovery controllers show up in configfs just like any other subsystems.
And with that we need to clarify the relationship between the discovery subsystem and the other subsystems, ie which subsystems should be presented by this discovery subsystem.Linking the discovery subsystem into a given port makes it obvious that
a) this port will be presenting a discovery subsystem
and
b) that the discovery subsystem will be presenting all subsystems configured on that port.
Quoted 7 lines
Quoted 3 lines
The built-in mechanism for discovery subsystems was okay as long as the discovery subsystem was built-in, too.
But with this patchset we're moving to an explicit configuration.Shouldn't we just require anything to be manually listed for this case similar to how we configure referrals for the well known discovery controller?
Which is what I've tried with this attempt.
I did _not_ want to create a new configuration mechanism, but rather use the existing ones. And the existing mechanism we have is linking subsystems to ports.
If we want to treat discovery subsystems differently (as you proposed) we sure can have a different mechanism on how to configure it. But I wasn't sure if that's the direction we want to go.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Christoph Hellwig
maintainerRe: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
So maybe we'll take a step and think about what we actually want to support. Here is my understanding, which might as well be incorrect and/or incomplete.
1) support a uniqueue discovery controller for bidirectional authentication to work properly
The only thing we really need is to have the existing discovery log for the well known discovery subsystems to be able to also show up as a unique discovery subsystem. Either by just renaming it like your earlier patch did, or by supporting an alternative name in addition to that.
2) (optionally) make the well known discovery controller go away
Just renaming the subsystem will make the well known one go away, but while that can be useful for some use cases, I think making it always go away has the potential for breakage. So maybe we need to not rename but be able to create an alias and have a way to disable the well known discovery controller if not used.
3) Actually have a different content.
Can you explain the use cases here a bit more?
And maybe add anything else we'd like to consider. (Everyone else on the list please also chime in)
Sagi Grimberg
maintainerRe: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 5 lines
So maybe we'll take a step and think about what we actually want to support. Here is my understanding, which might as well be incorrect and/or incomplete.
Quoted 7 lines
1) support a uniqueue discovery controller for bidirectional authentication to work properly
The only thing we really need is to have the existing discovery log for the well known discovery subsystems to be able to also show up as a unique discovery subsystem. Either by just renaming it like your earlier patch did, or by supporting an alternative name in addition to that.
I think that making the discovery subsystem show up in configfs makes sense, it can be attached to port(s) just like any other subsystem.
Quoted 8 lines
2) (optionally) make the well known discovery controller go away
Just renaming the subsystem will make the well known one go away, but while that can be useful for some use cases, I think making it always go away has the potential for breakage. So maybe we need to not rename but be able to create an alias and have a way to disable the well known discovery controller if not used.
agreed.
Quoted 7 lines
3) Actually have a different content.
Can you explain the use cases here a bit more?
And maybe add anything else we'd like to consider. (Everyone else on the list please also chime in)
I don't think we should support multiple discovery subsystems on the same target. Don't know why would anyone need it. I also think that reporting subsystems that are attached to all ports is just a a routing failure waiting to happen. But if anyone really really wants it, we can keep a flag on the discovery subsystem to report on all ports if explicitly set.
Re: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
In reply to Christoph HellwigQuoted 7 lines · Christoph Hellwig
2) (optionally) make the well known discovery controller go away
Just renaming the subsystem will make the well known one go away, but while that can be useful for some use cases, I think making it always go away has the potential for breakage. So maybe we need to not rename but be able to create an alias and have a way to disable the well known discovery controller if not used.
I agree. I think making the well known discovery controller "go away" will have huge problems in a multi-vendor fabric. We really can't touch the well known discovery controller because there are too many legacy and non-TP-8010 compliant NVMe hosts on the fabric.
We have to assume that both new and old implementations will coexist. And legacy hosts will NEVER look at the new, unique discovery NQN because they won't even recognize the new NVME_NQN_CURR log page entries.
/John
Re: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 9 lines · John Meneghini
Quoted 7 lines · Christoph Hellwig
2) (optionally) make the well known discovery controller go away
Just renaming the subsystem will make the well known one go away, but while that can be useful for some use cases, I think making it always go away has the potential for breakage. So maybe we need to not rename but be able to create an alias and have a way to disable the well known discovery controller if not used.
Quoted 4 lines
I agree. I think making the well known discovery controller "go away" will have huge problems in a multi-vendor fabric. We really can't touch the well known discovery controller because there are too many legacy and non-TP-8010 compliant NVMe hosts on the fabric.
We have to assume that both new and old implementations will coexist. And legacy hosts will NEVER look at the new, unique discovery NQN because they won't even recognize the new
NVME_NQN_CURRlog page entries.
It doesn't 'go away'.
The original, static, configfs entry will be unused, true.
But the unique discovery subsystem will _still_ react to the default discovery NQN, and any controller will continue to get information when using the default discovery NQN.
And even older implementations will continue to work _if_ they just skip unknown entries.
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
Sagi Grimberg
maintainerRe: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
In reply to Hannes ReineckeQuoted 22 lines
Quoted 14 lines
Quoted 14 lines
To make it configurable.
Unique discovery controllers show up in configfs just like any other subsystems.
And with that we need to clarify the relationship between the discovery subsystem and the other subsystems, ie which subsystems should be presented by this discovery subsystem.Linking the discovery subsystem into a given port makes it obvious that
a) this port will be presenting a discovery subsystem
and
b) that the discovery subsystem will be presenting all subsystems configured on that port.Quoted 7 lines
Quoted 3 lines
The built-in mechanism for discovery subsystems was okay as long as the discovery subsystem was built-in, too.
But with this patchset we're moving to an explicit configuration.Shouldn't we just require anything to be manually listed for this case similar to how we configure referrals for the well known discovery controller?
Quoted 6 lines
Which is what I've tried with this attempt.
I did _not_ want to create a new configuration mechanism, but rather use the existing ones. And the existing mechanism we have is linking subsystems to ports.If we want to treat discovery subsystems differently (as you proposed) we sure can have a different mechanism on how to configure it. But I wasn't sure if that's the direction we want to go.
What is the concern here? that it will break existing users with introducing an additional configuration step?
Re: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 26 lines · Sagi Grimberg
Quoted 26 lines
Quoted 14 lines
Quoted 14 lines
To make it configurable.
Unique discovery controllers show up in configfs just like any other subsystems.
And with that we need to clarify the relationship between the discovery subsystem and the other subsystems, ie which subsystems should be presented by this discovery subsystem.Linking the discovery subsystem into a given port makes it obvious that
a) this port will be presenting a discovery subsystem
and
b) that the discovery subsystem will be presenting all subsystems configured on that port.Quoted 7 lines
Quoted 3 lines
The built-in mechanism for discovery subsystems was okay as long as the discovery subsystem was built-in, too.
But with this patchset we're moving to an explicit configuration.Shouldn't we just require anything to be manually listed for this case similar to how we configure referrals for the well known discovery controller?
Which is what I've tried with this attempt.
I did _not_ want to create a new configuration mechanism, but rather use the existing ones. And the existing mechanism we have is linking subsystems to ports.
Quoted 5 lines
Quoted 2 lines
If we want to treat discovery subsystems differently (as you proposed) we sure can have a different mechanism on how to configure it. But I wasn't sure if that's the direction we want to go.
What is the concern here? that it will break existing users with introducing an additional configuration step?
Yes. It would break backward compability.
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
Sagi Grimberg
maintainerRe: [PATCH 3/3] nvmet: include all configured ports in discovery log page for unique discover controller
Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 14 lines
Quoted 14 lines
To make it configurable.
Unique discovery controllers show up in configfs just like any other subsystems.
And with that we need to clarify the relationship between the discovery subsystem and the other subsystems, ie which subsystems should be presented by this discovery subsystem.Linking the discovery subsystem into a given port makes it obvious that
a) this port will be presenting a discovery subsystem
and
b) that the discovery subsystem will be presenting all subsystems configured on that port.Quoted 7 lines
Quoted 3 lines
The built-in mechanism for discovery subsystems was okay as long as the discovery subsystem was built-in, too.
But with this patchset we're moving to an explicit configuration.Shouldn't we just require anything to be manually listed for this case similar to how we configure referrals for the well known discovery controller?
Which is what I've tried with this attempt.
I did _not_ want to create a new configuration mechanism, but rather use the existing ones. And the existing mechanism we have is linking subsystems to ports.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
If we want to treat discovery subsystems differently (as you proposed) we sure can have a different mechanism on how to configure it. But I wasn't sure if that's the direction we want to go.
What is the concern here? that it will break existing users with introducing an additional configuration step?
Yes. It would break backward compability.
So maybe hide it behind a modparam that we can deprecate in the future?