Currently there are no strict checks while setting SO_TXTIME
from userspace. With the recent development in skb->tstamp_type
clockid with unsupported clocks results in warn_on_once, which causes
unnecessary aborts in some systems which enables panic on warns.
Add validation in setsockopt to support only CLOCK_REALTIME,
CLOCK_MONOTONIC and CLOCK_TAI to be set from userspace.
Link: https://lore.kernel.org/netdev/[email protected]/
Link: https://lore.kernel.org/lkml/[email protected]/
Fixes: 1693c5db6ab8 ("net: Add additional bit to support clockid_t timestamp type")
Reported-by: [email protected]
Closes: https://syzkaller.appspot.com/bug?extid=d7b227731ec589e7f4f0
Reported-by: [email protected]
Closes: https://syzkaller.appspot.com/bug?extid=30a35a2e9c5067cc43fa
Signed-off-by: Abhishek Chauhan <[email protected]>
---
net/core/sock.c | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/net/core/sock.c b/net/core/sock.c
index 8629f9aecf91..f8374be9d8c9 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -1083,6 +1083,17 @@ bool sockopt_capable(int cap)
}
EXPORT_SYMBOL(sockopt_capable);
+static int sockopt_validate_clockid(int value)
+{
+ switch (value) {
+ case CLOCK_REALTIME:
+ case CLOCK_MONOTONIC:
+ case CLOCK_TAI:
+ return 0;
+ }
+ return -EINVAL;
+}
+
/*
* This is meant for all protocols to use and covers goings on
* at the socket level. Everything here is generic.
@@ -1497,6 +1508,11 @@ int sk_setsockopt(struct sock *sk, int level, int optname,
ret = -EPERM;
break;
}
+
+ ret = sockopt_validate_clockid(sk_txtime.clockid);
+ if (ret)
+ break;
+
sock_valbool_flag(sk, SOCK_TXTIME, true);
sk->sk_clockid = sk_txtime.clockid;
sk->sk_txtime_deadline_mode =
--
2.25.1
On 5/28/24 3:49 PM, Abhishek Chauhan wrote:
> Currently there are no strict checks while setting SO_TXTIME
> from userspace. With the recent development in skb->tstamp_type
> clockid with unsupported clocks results in warn_on_once, which causes
> unnecessary aborts in some systems which enables panic on warns.
>
> Add validation in setsockopt to support only CLOCK_REALTIME,
> CLOCK_MONOTONIC and CLOCK_TAI to be set from userspace.
>
> Link: https://lore.kernel.org/netdev/[email protected]/
> Link: https://lore.kernel.org/lkml/[email protected]/
> Fixes: 1693c5db6ab8 ("net: Add additional bit to support clockid_t timestamp type")
Patch lgtm. This should target for net-next instead of net. The Fixes patch is
in net-next only.
Acked-by: Martin KaFai Lau <[email protected]>
On 5/28/2024 6:15 PM, Martin KaFai Lau wrote:
> On 5/28/24 3:49 PM, Abhishek Chauhan wrote:
>> Currently there are no strict checks while setting SO_TXTIME
>> from userspace. With the recent development in skb->tstamp_type
>> clockid with unsupported clocks results in warn_on_once, which causes
>> unnecessary aborts in some systems which enables panic on warns.
>>
>> Add validation in setsockopt to support only CLOCK_REALTIME,
>> CLOCK_MONOTONIC and CLOCK_TAI to be set from userspace.
>>
>> Link: https://lore.kernel.org/netdev/[email protected]/
>> Link: https://lore.kernel.org/lkml/[email protected]/
>> Fixes: 1693c5db6ab8 ("net: Add additional bit to support clockid_t timestamp type")
>
> Patch lgtm. This should target for net-next instead of net. The Fixes patch is in net-next only.
>
Thanks Martin. Let me raise a patch on net-next and add your acked-by to it as well.
> Acked-by: Martin KaFai Lau <[email protected]>
>
minor: double space before userspace
Abhishek Chauhan wrote:
> Currently there are no strict checks while setting SO_TXTIME
> from userspace. With the recent development in skb->tstamp_type
> clockid with unsupported clocks results in warn_on_once, which causes
> unnecessary aborts in some systems which enables panic on warns.
>
> Add validation in setsockopt to support only CLOCK_REALTIME,
> CLOCK_MONOTONIC and CLOCK_TAI to be set from userspace.
>
> Link: https://lore.kernel.org/netdev/[email protected]/
> Link: https://lore.kernel.org/lkml/[email protected]/
These discussions can be found directly from the referenced commit?
If any, I'd like to the conversation we had that arrived at this
approach.
> Fixes: 1693c5db6ab8 ("net: Add additional bit to support clockid_t timestamp type")
> Reported-by: [email protected]
> Closes: https://syzkaller.appspot.com/bug?extid=d7b227731ec589e7f4f0
> Reported-by: [email protected]
> Closes: https://syzkaller.appspot.com/bug?extid=30a35a2e9c5067cc43fa
> Signed-off-by: Abhishek Chauhan <[email protected]>
> ---
> net/core/sock.c | 16 ++++++++++++++++
> 1 file changed, 16 insertions(+)
>
> diff --git a/net/core/sock.c b/net/core/sock.c
> index 8629f9aecf91..f8374be9d8c9 100644
> --- a/net/core/sock.c
> +++ b/net/core/sock.c
> @@ -1083,6 +1083,17 @@ bool sockopt_capable(int cap)
> }
> EXPORT_SYMBOL(sockopt_capable);
>
> +static int sockopt_validate_clockid(int value)
sock_txtime.clockid has type __kernel_clockid_t.
> +{
> + switch (value) {
> + case CLOCK_REALTIME:
> + case CLOCK_MONOTONIC:
> + case CLOCK_TAI:
> + return 0;
> + }
> + return -EINVAL;
> +}
> +
> /*
> * This is meant for all protocols to use and covers goings on
> * at the socket level. Everything here is generic.
> @@ -1497,6 +1508,11 @@ int sk_setsockopt(struct sock *sk, int level, int optname,
> ret = -EPERM;
> break;
> }
> +
> + ret = sockopt_validate_clockid(sk_txtime.clockid);
> + if (ret)
> + break;
> +
> sock_valbool_flag(sk, SOCK_TXTIME, true);
> sk->sk_clockid = sk_txtime.clockid;
> sk->sk_txtime_deadline_mode =
> --
> 2.25.1
>
On 5/29/2024 6:58 AM, Willem de Bruijn wrote:
> minor: double space before userspace
>
> Abhishek Chauhan wrote:
>> Currently there are no strict checks while setting SO_TXTIME
>> from userspace. With the recent development in skb->tstamp_type
>> clockid with unsupported clocks results in warn_on_once, which causes
>> unnecessary aborts in some systems which enables panic on warns.
>>
>> Add validation in setsockopt to support only CLOCK_REALTIME,
>> CLOCK_MONOTONIC and CLOCK_TAI to be set from userspace.
>>
>> Link: https://lore.kernel.org/netdev/[email protected]/
>> Link: https://lore.kernel.org/lkml/[email protected]/
>
> These discussions can be found directly from the referenced commit?
> If any, I'd like to the conversation we had that arrived at this
> approach.
>
Not Directly but from the patch series.
1. First link is for why we introduced skb->tstamp_type
2. Second link points to the series were we discussed on two approach to solve the problem
one being limit the skclockid to just TAI,MONO and REALTIME.
>> Fixes: 1693c5db6ab8 ("net: Add additional bit to support clockid_t timestamp type")
>> Reported-by: [email protected]
>> Closes: https://syzkaller.appspot.com/bug?extid=d7b227731ec589e7f4f0
>> Reported-by: [email protected]
>> Closes: https://syzkaller.appspot.com/bug?extid=30a35a2e9c5067cc43fa
>> Signed-off-by: Abhishek Chauhan <[email protected]>
>> ---
>> net/core/sock.c | 16 ++++++++++++++++
>> 1 file changed, 16 insertions(+)
>>
>> diff --git a/net/core/sock.c b/net/core/sock.c
>> index 8629f9aecf91..f8374be9d8c9 100644
>> --- a/net/core/sock.c
>> +++ b/net/core/sock.c
>> @@ -1083,6 +1083,17 @@ bool sockopt_capable(int cap)
>> }
>> EXPORT_SYMBOL(sockopt_capable);
>>
>> +static int sockopt_validate_clockid(int value)
>
> sock_txtime.clockid has type __kernel_clockid_t.
>
__kernel_clockid_t is typedef of int.
>> +{
>> + switch (value) {
>> + case CLOCK_REALTIME:
>> + case CLOCK_MONOTONIC:
>> + case CLOCK_TAI:
>> + return 0;
>> + }
>> + return -EINVAL;
>> +}
>> +
>> /*
>> * This is meant for all protocols to use and covers goings on
>> * at the socket level. Everything here is generic.
>> @@ -1497,6 +1508,11 @@ int sk_setsockopt(struct sock *sk, int level, int optname,
>> ret = -EPERM;
>> break;
>> }
>> +
>> + ret = sockopt_validate_clockid(sk_txtime.clockid);
>> + if (ret)
>> + break;
>> +
>> sock_valbool_flag(sk, SOCK_TXTIME, true);
>> sk->sk_clockid = sk_txtime.clockid;
>> sk->sk_txtime_deadline_mode =
>> --
>> 2.25.1
>>
>
>
Abhishek Chauhan (ABC) wrote:
>
>
> On 5/29/2024 6:58 AM, Willem de Bruijn wrote:
> > minor: double space before userspace
> >
> > Abhishek Chauhan wrote:
> >> Currently there are no strict checks while setting SO_TXTIME
> >> from userspace. With the recent development in skb->tstamp_type
> >> clockid with unsupported clocks results in warn_on_once, which causes
> >> unnecessary aborts in some systems which enables panic on warns.
> >>
> >> Add validation in setsockopt to support only CLOCK_REALTIME,
> >> CLOCK_MONOTONIC and CLOCK_TAI to be set from userspace.
> >>
> >> Link: https://lore.kernel.org/netdev/[email protected]/
> >> Link: https://lore.kernel.org/lkml/[email protected]/
> >
> > These discussions can be found directly from the referenced commit?
> > If any, I'd like to the conversation we had that arrived at this
> > approach.
> >
> Not Directly but from the patch series.
> 1. First link is for why we introduced skb->tstamp_type
> 2. Second link points to the series were we discussed on two approach to solve the problem
> one being limit the skclockid to just TAI,MONO and REALTIME.
Ah, I missed that.
Perhaps point directly to the start of that follow-up conversation?
https://lore.kernel.org/lkml/[email protected]/
>
>
> >> Fixes: 1693c5db6ab8 ("net: Add additional bit to support clockid_t timestamp type")
> >> Reported-by: [email protected]
> >> Closes: https://syzkaller.appspot.com/bug?extid=d7b227731ec589e7f4f0
> >> Reported-by: [email protected]
> >> Closes: https://syzkaller.appspot.com/bug?extid=30a35a2e9c5067cc43fa
> >> Signed-off-by: Abhishek Chauhan <[email protected]>
> >> ---
> >> net/core/sock.c | 16 ++++++++++++++++
> >> 1 file changed, 16 insertions(+)
> >>
> >> diff --git a/net/core/sock.c b/net/core/sock.c
> >> index 8629f9aecf91..f8374be9d8c9 100644
> >> --- a/net/core/sock.c
> >> +++ b/net/core/sock.c
> >> @@ -1083,6 +1083,17 @@ bool sockopt_capable(int cap)
> >> }
> >> EXPORT_SYMBOL(sockopt_capable);
> >>
> >> +static int sockopt_validate_clockid(int value)
> >
> > sock_txtime.clockid has type __kernel_clockid_t.
> >
>
> __kernel_clockid_t is typedef of int.
It is now, but the stricter type definition exists for a reason.
Try to keep the strict types where possible. Besides aiding
syntactic checks, it also helps self document code.
On 5/29/2024 9:00 AM, Willem de Bruijn wrote:
> Abhishek Chauhan (ABC) wrote:
>>
>>
>> On 5/29/2024 6:58 AM, Willem de Bruijn wrote:
>>> minor: double space before userspace
>>>
>>> Abhishek Chauhan wrote:
>>>> Currently there are no strict checks while setting SO_TXTIME
>>>> from userspace. With the recent development in skb->tstamp_type
>>>> clockid with unsupported clocks results in warn_on_once, which causes
>>>> unnecessary aborts in some systems which enables panic on warns.
>>>>
>>>> Add validation in setsockopt to support only CLOCK_REALTIME,
>>>> CLOCK_MONOTONIC and CLOCK_TAI to be set from userspace.
>>>>
>>>> Link: https://lore.kernel.org/netdev/[email protected]/
>>>> Link: https://lore.kernel.org/lkml/[email protected]/
>>>
>>> These discussions can be found directly from the referenced commit?
>>> If any, I'd like to the conversation we had that arrived at this
>>> approach.
>>>
>> Not Directly but from the patch series.
>> 1. First link is for why we introduced skb->tstamp_type
>> 2. Second link points to the series were we discussed on two approach to solve the problem
>> one being limit the skclockid to just TAI,MONO and REALTIME.
>
> Ah, I missed that.
> Perhaps point directly to the start of that follow-up conversation?
> Thanks Willem, Let me do that when i raise the net-next patch.
> https://lore.kernel.org/lkml/[email protected]/
>
>>
>>
>>>> Fixes: 1693c5db6ab8 ("net: Add additional bit to support clockid_t timestamp type")
>>>> Reported-by: [email protected]
>>>> Closes: https://syzkaller.appspot.com/bug?extid=d7b227731ec589e7f4f0
>>>> Reported-by: [email protected]
>>>> Closes: https://syzkaller.appspot.com/bug?extid=30a35a2e9c5067cc43fa
>>>> Signed-off-by: Abhishek Chauhan <[email protected]>
>>>> ---
>>>> net/core/sock.c | 16 ++++++++++++++++
>>>> 1 file changed, 16 insertions(+)
>>>>
>>>> diff --git a/net/core/sock.c b/net/core/sock.c
>>>> index 8629f9aecf91..f8374be9d8c9 100644
>>>> --- a/net/core/sock.c
>>>> +++ b/net/core/sock.c
>>>> @@ -1083,6 +1083,17 @@ bool sockopt_capable(int cap)
>>>> }
>>>> EXPORT_SYMBOL(sockopt_capable);
>>>>
>>>> +static int sockopt_validate_clockid(int value)
>>>
>>> sock_txtime.clockid has type __kernel_clockid_t.
>>>
>>
>> __kernel_clockid_t is typedef of int.
>> It is now, but the stricter type definition exists for a reason.
> Try to keep the strict types where possible. Besides aiding
> syntactic checks, it also helps self document code.
Okay i see what you are saying. Makes sense. I will change it to __kernel_clockid_t