The process_sysctl_arg() does not check whether val is empty before
invoking strlen(val). If the command line parameter () is incorrectly
configured and val is empty, oops is triggered.
For example:
"hung_task_panic=1" is incorrectly written as "hung_task_panic", oops is
triggered. The call stack is as follows:
Kernel command line: .... hung_task_panic
......
Call trace:
__pi_strlen+0x10/0x98
parse_args+0x278/0x344
do_sysctl_args+0x8c/0xfc
kernel_init+0x5c/0xf4
ret_from_fork+0x10/0x30
To fix it, check whether "val" is empty when "phram" is a sysctl field.
Error codes are returned in the failure branch, and error logs are
generated by parse_args().
Fixes: 3db978d480e2843 ("kernel/sysctl: support setting sysctl parameters
from kernel command line")
Signed-off-by: Xiaoming Ni <[email protected]>
---------
v3:
Return -EINVAL, When phram is the sysctl field and val is empty.
v2: https://lore.kernel.org/lkml/[email protected]/
Added log output of the failure branch based on the review comments of Kees Cook.
v1: https://lore.kernel.org/lkml/[email protected]/
---------
---
fs/proc/proc_sysctl.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
index 317899222d7f..d493a50058a5 100644
--- a/fs/proc/proc_sysctl.c
+++ b/fs/proc/proc_sysctl.c
@@ -1770,6 +1770,9 @@ static int process_sysctl_arg(char *param, char *val,
return 0;
}
+ if (!val)
+ return -EINVAL;
+
/*
* To set sysctl options, we use a temporary mount of proc, look up the
* respective sys/ file and write to it. To avoid mounting it when no
--
2.27.0
On Tue, 12 Jan 2021 11:31:55 +0800 Xiaoming Ni <[email protected]> wrote:
> The process_sysctl_arg() does not check whether val is empty before
> invoking strlen(val). If the command line parameter () is incorrectly
> configured and val is empty, oops is triggered.
>
> --- a/fs/proc/proc_sysctl.c
> +++ b/fs/proc/proc_sysctl.c
> @@ -1770,6 +1770,9 @@ static int process_sysctl_arg(char *param, char *val,
> return 0;
> }
>
> + if (!val)
> + return -EINVAL;
> +
I think v2 (return 0) was preferable. Because all the other error-out
cases in process_sysctl_arg() also do a `return 0'.
If we're going to do a separate "patch: make process_sysctl_arg()
return an errno instead of 0" then fine, we can discuss that. But it's
conceptually a different work from fixing this situation.
On 2021/1/12 12:33, Andrew Morton wrote:
> On Tue, 12 Jan 2021 11:31:55 +0800 Xiaoming Ni <[email protected]> wrote:
>
>> The process_sysctl_arg() does not check whether val is empty before
>> invoking strlen(val). If the command line parameter () is incorrectly
>> configured and val is empty, oops is triggered.
>>
>> --- a/fs/proc/proc_sysctl.c
>> +++ b/fs/proc/proc_sysctl.c
>> @@ -1770,6 +1770,9 @@ static int process_sysctl_arg(char *param, char *val,
>> return 0;
>> }
>>
>> + if (!val)
>> + return -EINVAL;
>> +
>
> I think v2 (return 0) was preferable. Because all the other error-out
> cases in process_sysctl_arg() also do a `return 0'.
https://lore.kernel.org/lkml/[email protected]/
patch4:
+++ b/fs/proc/proc_sysctl.c
@@ -1757,6 +1757,9 @@ static int process_sysctl_arg(char *param,
char *val,
loff_t pos = 0;
ssize_t wret;
+ if (!val)
+ return 0;
+
if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
param += sizeof("sysctl") - 1;
Is this the version you're talking about?
>
> If we're going to do a separate "patch: make process_sysctl_arg()
> return an errno instead of 0" then fine, we can discuss that. But it's
> conceptually a different work from fixing this situation.
> .
>
However, are the logs generated by process_sysctl_arg() clearer and more
accurate than parse_args()? Should the logs generated by
process_sysctl_arg() be deleted?
Thanks
Xiaoming Ni
On Tue, 12 Jan 2021 14:24:05 +0800 Xiaoming Ni <[email protected]> wrote:
> On 2021/1/12 12:33, Andrew Morton wrote:
> > On Tue, 12 Jan 2021 11:31:55 +0800 Xiaoming Ni <[email protected]> wrote:
> >
> >> The process_sysctl_arg() does not check whether val is empty before
> >> invoking strlen(val). If the command line parameter () is incorrectly
> >> configured and val is empty, oops is triggered.
> >>
> >> --- a/fs/proc/proc_sysctl.c
> >> +++ b/fs/proc/proc_sysctl.c
> >> @@ -1770,6 +1770,9 @@ static int process_sysctl_arg(char *param, char *val,
> >> return 0;
> >> }
> >>
> >> + if (!val)
> >> + return -EINVAL;
> >> +
> >
> > I think v2 (return 0) was preferable. Because all the other error-out
> > cases in process_sysctl_arg() also do a `return 0'.
>
> https://lore.kernel.org/lkml/[email protected]/
>
> patch4:
> +++ b/fs/proc/proc_sysctl.c
> @@ -1757,6 +1757,9 @@ static int process_sysctl_arg(char *param,
> char *val,
> loff_t pos = 0;
> ssize_t wret;
>
> + if (!val)
> + return 0;
> +
> if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
> param += sizeof("sysctl") - 1;
>
> Is this the version you're talking about?
yes, but as a separate patch. The bugfix comes first.
> >
> > If we're going to do a separate "patch: make process_sysctl_arg()
> > return an errno instead of 0" then fine, we can discuss that. But it's
> > conceptually a different work from fixing this situation.
> > .
> >
> However, are the logs generated by process_sysctl_arg() clearer and more
> accurate than parse_args()? Should the logs generated by
> process_sysctl_arg() be deleted?
I think the individual logs are very useful and should be retained.
On Mon 11-01-21 22:28:45, Andrew Morton wrote:
> On Tue, 12 Jan 2021 14:24:05 +0800 Xiaoming Ni <[email protected]> wrote:
>
> > On 2021/1/12 12:33, Andrew Morton wrote:
> > > On Tue, 12 Jan 2021 11:31:55 +0800 Xiaoming Ni <[email protected]> wrote:
> > >
> > >> The process_sysctl_arg() does not check whether val is empty before
> > >> invoking strlen(val). If the command line parameter () is incorrectly
> > >> configured and val is empty, oops is triggered.
> > >>
> > >> --- a/fs/proc/proc_sysctl.c
> > >> +++ b/fs/proc/proc_sysctl.c
> > >> @@ -1770,6 +1770,9 @@ static int process_sysctl_arg(char *param, char *val,
> > >> return 0;
> > >> }
> > >>
> > >> + if (!val)
> > >> + return -EINVAL;
> > >> +
> > >
> > > I think v2 (return 0) was preferable. Because all the other error-out
> > > cases in process_sysctl_arg() also do a `return 0'.
> >
> > https://lore.kernel.org/lkml/[email protected]/
> >
> > patch4:
> > +++ b/fs/proc/proc_sysctl.c
> > @@ -1757,6 +1757,9 @@ static int process_sysctl_arg(char *param,
> > char *val,
> > loff_t pos = 0;
> > ssize_t wret;
> >
> > + if (!val)
> > + return 0;
> > +
> > if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
> > param += sizeof("sysctl") - 1;
> >
> > Is this the version you're talking about?
>
> yes, but as a separate patch. The bugfix comes first.
>
> > >
> > > If we're going to do a separate "patch: make process_sysctl_arg()
> > > return an errno instead of 0" then fine, we can discuss that. But it's
> > > conceptually a different work from fixing this situation.
> > > .
> > >
> > However, are the logs generated by process_sysctl_arg() clearer and more
> > accurate than parse_args()? Should the logs generated by
> > process_sysctl_arg() be deleted?
>
> I think the individual logs are very useful and should be retained.
Yes, other sysfs specific error messages are likely useful. I just fail
to see why a missing value should be handled here when there is an
existing handling in the caller. Not sure whether a complete shadow
reporting in process_sysctl_arg is a deliberate decision or not.
Vlastimil?
Anyway one way or the other, all I care about is to have a reporting in
place because this shouldn't be a silent failure.
--
Michal Hocko
SUSE Labs
On 1/12/21 8:24 AM, Michal Hocko wrote:
>> > >
>> > > If we're going to do a separate "patch: make process_sysctl_arg()
>> > > return an errno instead of 0" then fine, we can discuss that. But it's
>> > > conceptually a different work from fixing this situation.
>> > > .
>> > >
>> > However, are the logs generated by process_sysctl_arg() clearer and more
>> > accurate than parse_args()? Should the logs generated by
>> > process_sysctl_arg() be deleted?
>>
>> I think the individual logs are very useful and should be retained.
>
> Yes, other sysfs specific error messages are likely useful. I just fail
> to see why a missing value should be handled here when there is an
> existing handling in the caller. Not sure whether a complete shadow
> reporting in process_sysctl_arg is a deliberate decision or not.
> Vlastimil?
Yes, it's a way to have more useful sysctl-specific reports than the generic
ones. And I think I was inspired by some other existing code, but don't remember
exactly. The options are:
1) the current sysctl-specific reports, return 0 as the values are only consumed
2) be silent and return error, invent new error codes to have generic report be
more useful for sysctl, but inevitably lose some nuances anyway
3) a mix where 2) is used for situations where generic report is sufficient
enough, 1) where not
Patch v2 went with option 1), v3 with option 3). I think it's down to
preferences. I would personally go with v2 and message similar to the existing
ones, i.e.:
"Failed to set sysctl parameter '%s': no value given\n"
Also we seem to be silently doing nothing when strlen(val) == 0, i.e.
"hung_task_panic=" was passed. Worth reporting the same error.
But v3 is fine with me as well. The generic error message works. We could just
add "if (!len) return -EINVAL" below the strlen() call.
Also please Cc: stable.
> Anyway one way or the other, all I care about is to have a reporting in
> place because this shouldn't be a silent failure.
>
On 2021/1/12 19:42, Vlastimil Babka wrote:
> On 1/12/21 8:24 AM, Michal Hocko wrote:
>>>>>
>>>>> If we're going to do a separate "patch: make process_sysctl_arg()
>>>>> return an errno instead of 0" then fine, we can discuss that. But it's
>>>>> conceptually a different work from fixing this situation.
>>>>> .
>>>>>
>>>> However, are the logs generated by process_sysctl_arg() clearer and more
>>>> accurate than parse_args()? Should the logs generated by
>>>> process_sysctl_arg() be deleted?
>>>
>>> I think the individual logs are very useful and should be retained.
>>
>> Yes, other sysfs specific error messages are likely useful. I just fail
>> to see why a missing value should be handled here when there is an
>> existing handling in the caller. Not sure whether a complete shadow
>> reporting in process_sysctl_arg is a deliberate decision or not.
>> Vlastimil?
>
> Yes, it's a way to have more useful sysctl-specific reports than the generic
> ones. And I think I was inspired by some other existing code, but don't remember
> exactly. The options are:
>
> 1) the current sysctl-specific reports, return 0 as the values are only consumed
> 2) be silent and return error, invent new error codes to have generic report be
> more useful for sysctl, but inevitably lose some nuances anyway
> 3) a mix where 2) is used for situations where generic report is sufficient
> enough, 1) where not
>
> Patch v2 went with option 1), v3 with option 3). I think it's down to
> preferences. I would personally go with v2 and message similar to the existing
> ones, i.e.:
>
> "Failed to set sysctl parameter '%s': no value given\n"
>
> Also we seem to be silently doing nothing when strlen(val) == 0, i.e.
> "hung_task_panic=" was passed. Worth reporting the same error.
>
> But v3 is fine with me as well. The generic error message works. We could just
> add "if (!len) return -EINVAL" below the strlen() call.
>
> Also please Cc: stable.
>
>> Anyway one way or the other, all I care about is to have a reporting in
>> place because this shouldn't be a silent failure.
>>
The current v2 is already in the linux-next branch and throws a new
error:
https://lore.kernel.org/lkml/[email protected]/
This bug has been mentioned in the previous discussion and has been
fixed in the current v3 patch.
https://lore.kernel.org/linux-fsdevel/202101111149.20A58E1@keescook/
What am I supposed to do now?
- Resend V3?
- Rewrite a new fix patch based on the current code of linux-next.
- Develop a new V4 patch: Use V2 to discuss how to use the Patch4
solution.
https://lore.kernel.org/linux-fsdevel/[email protected]/#t
Thanks
Xiaoming Ni
On 1/17/21 3:59 AM, Xiaoming Ni wrote:
> On 2021/1/12 19:42, Vlastimil Babka wrote:
>> On 1/12/21 8:24 AM, Michal Hocko wrote:
>>>>>>
>>>>>> If we're going to do a separate "patch: make process_sysctl_arg()
>>>>>> return an errno instead of 0" then fine, we can discuss that. But it's
>>>>>> conceptually a different work from fixing this situation.
>>>>>> .
>>>>>>
>>>>> However, are the logs generated by process_sysctl_arg() clearer and more
>>>>> accurate than parse_args()? Should the logs generated by
>>>>> process_sysctl_arg() be deleted?
>>>>
>>>> I think the individual logs are very useful and should be retained.
>>>
>>> Yes, other sysfs specific error messages are likely useful. I just fail
>>> to see why a missing value should be handled here when there is an
>>> existing handling in the caller. Not sure whether a complete shadow
>>> reporting in process_sysctl_arg is a deliberate decision or not.
>>> Vlastimil?
>>
>> Yes, it's a way to have more useful sysctl-specific reports than the generic
>> ones. And I think I was inspired by some other existing code, but don't remember
>> exactly. The options are:
>>
>> 1) the current sysctl-specific reports, return 0 as the values are only consumed
>> 2) be silent and return error, invent new error codes to have generic report be
>> more useful for sysctl, but inevitably lose some nuances anyway
>> 3) a mix where 2) is used for situations where generic report is sufficient
>> enough, 1) where not
>>
>> Patch v2 went with option 1), v3 with option 3). I think it's down to
>> preferences. I would personally go with v2 and message similar to the existing
>> ones, i.e.:
>>
>> "Failed to set sysctl parameter '%s': no value given\n"
>>
>> Also we seem to be silently doing nothing when strlen(val) == 0, i.e.
>> "hung_task_panic=" was passed. Worth reporting the same error.
>>
>> But v3 is fine with me as well. The generic error message works. We could just
>> add "if (!len) return -EINVAL" below the strlen() call.
>>
>> Also please Cc: stable.
>>
>>> Anyway one way or the other, all I care about is to have a reporting in
>>> place because this shouldn't be a silent failure.
>>>
>
>
> The current v2 is already in the linux-next branch and throws a new error:
> https://lore.kernel.org/lkml/[email protected]/
>
> This bug has been mentioned in the previous discussion and has been fixed in the
> current v3 patch.
> https://lore.kernel.org/linux-fsdevel/202101111149.20A58E1@keescook/
>
> What am I supposed to do now?
> - Resend V3?
IMHO this. But also please handle also len == 0 like below. And add
Cc: <[email protected]>
> - Rewrite a new fix patch based on the current code of linux-next.
AFAICS Andrew dropped the v2 already.
Thanks.
diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
index 317899222d7f..f424010d1a60 100644
--- a/fs/proc/proc_sysctl.c
+++ b/fs/proc/proc_sysctl.c
@@ -1770,6 +1770,12 @@ static int process_sysctl_arg(char *param, char *val,
return 0;
}
+ if (!val)
+ return -EINVAL;
+ len = strlen(val);
+ if (!len)
+ return -EINVAL;
+
/*
* To set sysctl options, we use a temporary mount of proc, look up the
* respective sys/ file and write to it. To avoid mounting it when no
@@ -1811,7 +1817,6 @@ static int process_sysctl_arg(char *param, char *val,
file, param, val);
goto out;
}
- len = strlen(val);
wret = kernel_write(file, val, len, &pos);
if (wret < 0) {
err = wret;