Skip to contentSkip to conversation

Potential bug in queue_count_set

DISCUSSION Maximilian Luz 3 participants 6 messages Apr 10, 2019 – Apr 12, 2019 NVMe

lore
Share & tools
lore
#0 lore raw

Hi all,

I just stumbled (more or less randomly) over this piece of code (/drivers/nvme/host/pci.c, v5.1-rc4):

147: static int queue_count_set(const char *val, const struct kernel_param *kp) 148: {
149: int n = 0, ret;
150:
151: ret = kstrtoint(val, 10, &n);
152: if (ret)
153: return ret;
154: if (n > num_possible_cpus())
155: n = num_possible_cpus();
156:
157: return param_set_int(val, kp);
158: }

Link with context: https://github.com/torvalds/linux/blob/v5.1-rc4/drivers/nvme/host/pci.c#L154-L155

It looks like lines 154, 155 don't achieve anything as n is not being used afterwards. Sorry if this is already known or a non-issue, just wanted to let you guys know.

Maximilian

#1 lore raw
Quoted 19 lines · Maximilian Luz

Hi all,

I just stumbled (more or less randomly) over this piece of code (/drivers/nvme/host/pci.c, v5.1-rc4):

147: static int queue_count_set(const char *val, const struct kernel_param *kp)
148: {
149:???? int n = 0, ret;
150:
151:???? ret = kstrtoint(val, 10, &n);
152:???? if (ret)
153:???????? return ret;
154:???? if (n > num_possible_cpus())
155:???????? n = num_possible_cpus();
156:
157:???? return param_set_int(val, kp);
158: }

Quoted 8 lines

Link with context:
https://github.com/torvalds/linux/blob/v5.1-rc4/drivers/nvme/host/pci.c#L154-L155

It looks like lines 154, 155 don't achieve anything as n is not being used afterwards. Sorry if this is already known or a non-issue, just wanted to let you guys know.

It looks like module parameter "write_queues" is set to the actual given value (val) without considering reduced number of queues in case given value is greater than nr_cpus.

I think that might need sprintf() or something like this to make the actual value reduced in that case.

Quoted 6 lines

Maximilian

_______________________________________________
Linux-nvme mailing list
Linux-nvme at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

#2 lore raw
Quoted 7 lines
Quoted 6 lines

It looks like lines 154, 155 don't achieve anything as n is not being used afterwards. Sorry if this is already known or a non-issue, just wanted to let you guys know.

Quoted 6 lines

It looks like module parameter "write_queues" is set to the actual given value (val) without considering reduced number of queues in case given value is greater than nr_cpus.

I think that might need sprintf() or something like this to make the actual value reduced in that case.

Sorry, the "val" is const so that we cannot go through it with sprintf(). My bad. This function might return an error if invalid number is given.

Sorry for the confusion.

Have posted patches for it.
http://lists.infradead.org/pipermail/linux-nvme/2019-April/023369.html

Thanks,

#3 lore raw
Quoted 16 lines
Quoted 8 lines
Quoted 7 lines

It looks like lines 154, 155 don't achieve anything as n is not being used afterwards. Sorry if this is already known or a non-issue, just wanted to let you guys know.

Quoted 6 lines

It looks like module parameter "write_queues" is set to the actual given value (val) without considering reduced number of queues in case given value is greater than nr_cpus.

I think that might need sprintf() or something like this to make the actual value reduced in that case.

Quoted 7 lines

Sorry, the "val" is const so that we cannot go through it with sprintf(). My bad. This function might return an error if invalid number is given.

Sorry for the confusion.

Have posted patches for it.
http://lists.infradead.org/pipermail/linux-nvme/2019-April/023369.html

I honestly know too little about module-parameters to give a qualified answer here, but could returning -EINVAL here break things?

From what I can tell it should be possible to bypass the param_set_int
function and set *(kp->arg) directly, which would reproduce the param_set_int functionality (kernel/params.c:223).

Alternatively one could argue that after a successful kstrtoint call, we can strncpy the input into a buffer with fixed length of 11 (10 digits + null, if I'm not mistaken) and use that as input for param_set_int.

#4 lore raw
Quoted 34 lines

-----Original Message-----
From: Linux-nvme [mailto:linux-nvme-bounces at lists.infradead.org] On Behalf Of Maximilian Luz
Sent: Friday, April 12, 2019 2:40 AM
To: Minwoo Im; Keith Busch; Jens Axboe; Christoph Hellwig; Sagi Grimberg
Cc: linux-nvme at lists.infradead.org
Subject: Re: Potential bug in queue_count_set

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

It looks like lines 154, 155 don't achieve anything as n is not

being

Quoted 5 lines
Quoted 5 lines
Quoted 3 lines

used afterwards. Sorry if this is already known or a non-issue, just wanted to let you guys know.

It looks like module parameter "write_queues" is set to the actual

given

Quoted 7 lines
Quoted 5 lines

value (val) without considering reduced number of queues in case given value is greater than nr_cpus.

I think that might need sprintf() or something like this to make the actual value reduced in that case.

Quoted 7 lines

Sorry, the "val" is const so that we cannot go through it with sprintf(). My bad. This function might return an error if invalid number is given.

Sorry for the confusion.

Have posted patches for it.
http://lists.infradead.org/pipermail/linux-nvme/2019-April/023369.html

Quoted 6 lines

I honestly know too little about module-parameters to give a qualified answer here, but could returning -EINVAL here break things?

From what I can tell it should be possible to bypass the param_set_int
function and set *(kp->arg) directly, which would reproduce the param_set_int functionality (kernel/params.c:223).

Got what you concern about. We can make kp->arg have string from "n". But Anyway we have to give "val" which might be meaningless to param_set_int() even we modified kp->arg.

Quoted 4 lines

Alternatively one could argue that after a successful kstrtoint call, we can strncpy the input into a buffer with fixed length of 11 (10 digits + null, if I'm not mistaken) and use that as input for param_set_int.

Right. We can prepare a local buffer to receive the input numeric string. But I'm not pretty sure what was the intention of Jens's commit 3b6592f70. I think I need to have a discussion about it, If user gives more than nr_possible_cpus, then:
- It should be reduced to a possible value by automatically without any error.
- It should return an error to let user know.

Let's see how they think about it. Also we can have a discussion on that PATCH mail-loop.

Thanks,
Minwoo Im

#5 lore raw
Quoted 8 lines · Minwoo Im

I think I need to have a discussion about it, If user gives more than nr_possible_cpus, then:
- It should be reduced to a possible value by automatically without any error.
- It should return an error to let user know.

Let's see how they think about it. Also we can have a discussion on that PATCH mail-loop.

Sounds good.

Maximilian