[PATCH] nvme: duplicate NQNs check if devices needed

[PATCH] nvme: duplicate NQNs check if devices needed

[Liguang Zhang]

If a device provides an NQN it is expected to be globally unique.
Unfortunately some firmware revisions for Intel P4500/P4510 and some
other old devices did not satisfy this requirement. If devices have the
same subnqn field, after a kernel upgrade from 4.19 to 5.10, if a system
has >1 affected device then only one device is enabled. Error like this:

nvme nvme2: Duplicate cntlid 0 with nvme0, rejecting
nvme nvme3: Duplicate cntlid 1 with nvme1, rejecting

In these circumstances, if subnqn did not begin with "nqn.", we assume
that devices did not need subnqn check.

Signed-off-by: Liguang Zhang <zhangliguang@linux.alibaba.com>
---
 drivers/nvme/host/core.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 4c63564adeaa..80abb4526932 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -2587,6 +2587,9 @@ static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
 
 	lockdep_assert_held(&nvme_subsystems_lock);
 
+	if (strncmp(subsysnqn, "nqn.", 4))
+		return NULL;
+
 	/*
 	 * Fail matches for discovery subsystems. This results
 	 * in each discovery controller bound to a unique subsystem.
-- 
2.19.1.6.gb485710b

Re: [PATCH] nvme: duplicate NQNs check if devices needed

[Keith Busch]

On Mon, Nov 29, 2021 at 07:18:54PM +0800, Liguang Zhang wrote:
> @@ -2587,6 +2587,9 @@ static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
>  
>  	lockdep_assert_held(&nvme_subsystems_lock);
>  
> +	if (strncmp(subsysnqn, "nqn.", 4))
> +		return NULL;

This seems like an arbitrary way to decide not to check for duplicates.
Shouldn't we just add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk for the
broken controllers?

Re: [PATCH] nvme: duplicate NQNs check if devices needed

[Chaitanya Kulkarni]

Liguang,

On 11/29/21 8:34 AM, Keith Busch wrote:
> On Mon, Nov 29, 2021 at 07:18:54PM +0800, Liguang Zhang wrote:
>> @@ -2587,6 +2587,9 @@ static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
>>   
>>   	lockdep_assert_held(&nvme_subsystems_lock);
>>   
>> +	if (strncmp(subsysnqn, "nqn.", 4))
>> +		return NULL;
> 
> This seems like an arbitrary way to decide not to check for duplicates.
> Shouldn't we just add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk for the
> broken controllers?
> 

Please add t he quirk for a specific controller(s) as mentioned
by Keith.

-ck

Re: [PATCH] nvme: duplicate NQNs check if devices needed

[luanshi]

Hi Keith,

在 2021/11/30 0:34, Keith Busch 写道:
> On Mon, Nov 29, 2021 at 07:18:54PM +0800, Liguang Zhang wrote:
>> @@ -2587,6 +2587,9 @@ static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
>>   
>>   	lockdep_assert_held(&nvme_subsystems_lock);
>>   
>> +	if (strncmp(subsysnqn, "nqn.", 4))
>> +		return NULL;
> This seems like an arbitrary way to decide not to check for duplicates.
> Shouldn't we just add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk for the
> broken controllers?

Yeah, add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk is a way to resolve the 
problem. I do not

have enough controller in my hand, I don't make sure all the controller 
are covered. And there are some

Weird devices from partners are not registered in pci id table:

https://pci-ids.ucw.cz/

"nqn." prefix is a required naming, so i used this way.


Thanks,

Liguang

Re: [PATCH] nvme: duplicate NQNs check if devices needed

[Christoph Hellwig]

On Tue, Nov 30, 2021 at 01:04:55PM +0800, luanshi wrote:
> Hi Keith,
>
> 在 2021/11/30 0:34, Keith Busch 写道:
>> On Mon, Nov 29, 2021 at 07:18:54PM +0800, Liguang Zhang wrote:
>>> @@ -2587,6 +2587,9 @@ static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
>>>     	lockdep_assert_held(&nvme_subsystems_lock);
>>>   +	if (strncmp(subsysnqn, "nqn.", 4))
>>> +		return NULL;
>> This seems like an arbitrary way to decide not to check for duplicates.
>> Shouldn't we just add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk for the
>> broken controllers?
>
> Yeah, add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk is a way to resolve the 
> problem. I do not
>
> have enough controller in my hand, I don't make sure all the controller are 
> covered. And there are some

What does this broken device report in the subnqn field?

Re: [PATCH] nvme: duplicate NQNs check if devices needed

[luanshi]

Hi Christoph,

在 2021/11/30 14:40, Christoph Hellwig 写道:
> On Tue, Nov 30, 2021 at 01:04:55PM +0800, luanshi wrote:
>> Hi Keith,
>>
>> 在 2021/11/30 0:34, Keith Busch 写道:
>>> On Mon, Nov 29, 2021 at 07:18:54PM +0800, Liguang Zhang wrote:
>>>> @@ -2587,6 +2587,9 @@ static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
>>>>      	lockdep_assert_held(&nvme_subsystems_lock);
>>>>    +	if (strncmp(subsysnqn, "nqn.", 4))
>>>> +		return NULL;
>>> This seems like an arbitrary way to decide not to check for duplicates.
>>> Shouldn't we just add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk for the
>>> broken controllers?
>> Yeah, add the NVME_QUIRK_IGNORE_DEV_SUBNQN quirk is a way to resolve the
>> problem. I do not
>>
>> have enough controller in my hand, I don't make sure all the controller are
>> covered. And there are some
> What does this broken device report in the subnqn field?
In function nvme_init_subnqn, i add some printk info,

#git diff drivers/nvme/host/core.c
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index e1e574ecf031..d043f4506c37 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -2708,6 +2708,7 @@ static void nvme_init_subnqn(struct nvme_subsystem 
*subsys, struct nvme_ctrl *ct

         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
+             printk("%s: nqnlen=%d, subnqn=%s\n", __func__, nqnlen, 
id->subnqn);
                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
                         return;

I found that:

nqnlen is 1, subnqn filed is filled by space.

nvme_init_subnqn: nqnlen=1 subnqn=


Thanks,

Liguang