`strncpy` is deprecated for use on NUL-terminated destination strings
[1] and as such we should prefer more robust and less ambiguous
interfaces.
Let's opt for memcpy as we are copying strings into slices of length
`ETH_GSTRING_LEN` within the `data` buffer. Other similar get_strings()
implementations [2] [3] use memcpy().
Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings [1]
Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/opa_vnic/opa_vnic_ethtool.c#L167 [2]
Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c#L137 [3]
Link: https://github.com/KSPP/linux/issues/90
Cc: [email protected]
Signed-off-by: Justin Stitt <[email protected]>
---
Note: build-tested only.
---
drivers/net/dsa/lan9303-core.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/dsa/lan9303-core.c b/drivers/net/dsa/lan9303-core.c
index ee67adeb2cdb..665d69384b62 100644
--- a/drivers/net/dsa/lan9303-core.c
+++ b/drivers/net/dsa/lan9303-core.c
@@ -1013,8 +1013,8 @@ static void lan9303_get_strings(struct dsa_switch *ds, int port,
return;
for (u = 0; u < ARRAY_SIZE(lan9303_mib); u++) {
- strncpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
- ETH_GSTRING_LEN);
+ memcpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
+ ETH_GSTRING_LEN);
}
}
---
base-commit: cbf3a2cb156a2c911d8f38d8247814b4c07f49a2
change-id: 20231005-strncpy-drivers-net-dsa-lan9303-core-c-6386858e5c22
Best regards,
--
Justin Stitt <[email protected]>
On Wed, Oct 04, 2023 at 08:07:55PM -0700, Kees Cook wrote:
> On Thu, Oct 05, 2023 at 12:30:18AM +0000, Justin Stitt wrote:
> > `strncpy` is deprecated for use on NUL-terminated destination strings
> > [1] and as such we should prefer more robust and less ambiguous
> > interfaces.
> >
> > Let's opt for memcpy as we are copying strings into slices of length
> > `ETH_GSTRING_LEN` within the `data` buffer. Other similar get_strings()
> > implementations [2] [3] use memcpy().
> >
> > Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings [1]
> > Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/opa_vnic/opa_vnic_ethtool.c#L167 [2]
> > Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c#L137 [3]
> > Link: https://github.com/KSPP/linux/issues/90
> > Cc: [email protected]
> > Signed-off-by: Justin Stitt <[email protected]>
> > ---
> > Note: build-tested only.
> > ---
> > drivers/net/dsa/lan9303-core.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/dsa/lan9303-core.c b/drivers/net/dsa/lan9303-core.c
> > index ee67adeb2cdb..665d69384b62 100644
> > --- a/drivers/net/dsa/lan9303-core.c
> > +++ b/drivers/net/dsa/lan9303-core.c
> > @@ -1013,8 +1013,8 @@ static void lan9303_get_strings(struct dsa_switch *ds, int port,
> > return;
> >
> > for (u = 0; u < ARRAY_SIZE(lan9303_mib); u++) {
> > - strncpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
> > - ETH_GSTRING_LEN);
> > + memcpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
> > + ETH_GSTRING_LEN);
>
> This won't work because lan9303_mib entries aren't ETH_GSTRING_LEN-long
> strings; they're string pointers:
>
> static const struct lan9303_mib_desc lan9303_mib[] = {
> { .offset = LAN9303_MAC_RX_BRDCST_CNT_0, .name = "RxBroad", },
>
> So this really does need a strcpy-family function.
>
> And, I think the vnic_gstrings_stats and ipoib_gstrings_stats examples
> are actually buggy -- they're copying junk into userspace...
>
> I am reminded of this patch, which correctly uses strscpy_pad():
> https://lore.kernel.org/lkml/[email protected]/
>
> I think you want to do the same here, and use strscpy_pad(). And perhaps
> send some fixes for the other memcpy() users?
Meh, I think it's not worth fixing the memcpy() users of this. This
buggy pattern is very common, it seems:
$ git grep 'data.*ETH_GSTRING_LEN' | grep memcpy | wc -l
47
--
Kees Cook
On Thu, Oct 05, 2023 at 12:30:18AM +0000, Justin Stitt wrote:
> `strncpy` is deprecated for use on NUL-terminated destination strings
> [1] and as such we should prefer more robust and less ambiguous
> interfaces.
>
> Let's opt for memcpy as we are copying strings into slices of length
> `ETH_GSTRING_LEN` within the `data` buffer. Other similar get_strings()
> implementations [2] [3] use memcpy().
>
> Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings [1]
> Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/opa_vnic/opa_vnic_ethtool.c#L167 [2]
> Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c#L137 [3]
> Link: https://github.com/KSPP/linux/issues/90
> Cc: [email protected]
> Signed-off-by: Justin Stitt <[email protected]>
> ---
> Note: build-tested only.
> ---
> drivers/net/dsa/lan9303-core.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/dsa/lan9303-core.c b/drivers/net/dsa/lan9303-core.c
> index ee67adeb2cdb..665d69384b62 100644
> --- a/drivers/net/dsa/lan9303-core.c
> +++ b/drivers/net/dsa/lan9303-core.c
> @@ -1013,8 +1013,8 @@ static void lan9303_get_strings(struct dsa_switch *ds, int port,
> return;
>
> for (u = 0; u < ARRAY_SIZE(lan9303_mib); u++) {
> - strncpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
> - ETH_GSTRING_LEN);
> + memcpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
> + ETH_GSTRING_LEN);
This won't work because lan9303_mib entries aren't ETH_GSTRING_LEN-long
strings; they're string pointers:
static const struct lan9303_mib_desc lan9303_mib[] = {
{ .offset = LAN9303_MAC_RX_BRDCST_CNT_0, .name = "RxBroad", },
So this really does need a strcpy-family function.
And, I think the vnic_gstrings_stats and ipoib_gstrings_stats examples
are actually buggy -- they're copying junk into userspace...
I am reminded of this patch, which correctly uses strscpy_pad():
https://lore.kernel.org/lkml/[email protected]/
I think you want to do the same here, and use strscpy_pad(). And perhaps
send some fixes for the other memcpy() users?
(Have I mentioned I really dislike the get_strings() API?)
-Kees
--
Kees Cook
From: Kees Cook <[email protected]>
Date: Wed, 4 Oct 2023 22:02:00 -0700
> On Wed, Oct 04, 2023 at 08:07:55PM -0700, Kees Cook wrote:
>> On Thu, Oct 05, 2023 at 12:30:18AM +0000, Justin Stitt wrote:
>>> `strncpy` is deprecated for use on NUL-terminated destination strings
>>> [1] and as such we should prefer more robust and less ambiguous
>>> interfaces.
>>>
>>> Let's opt for memcpy as we are copying strings into slices of length
>>> `ETH_GSTRING_LEN` within the `data` buffer. Other similar get_strings()
>>> implementations [2] [3] use memcpy().
>>>
>>> Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings [1]
>>> Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/opa_vnic/opa_vnic_ethtool.c#L167 [2]
>>> Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c#L137 [3]
>>> Link: https://github.com/KSPP/linux/issues/90
>>> Cc: [email protected]
>>> Signed-off-by: Justin Stitt <[email protected]>
>>> ---
>>> Note: build-tested only.
>>> ---
>>> drivers/net/dsa/lan9303-core.c | 4 ++--
>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/net/dsa/lan9303-core.c b/drivers/net/dsa/lan9303-core.c
>>> index ee67adeb2cdb..665d69384b62 100644
>>> --- a/drivers/net/dsa/lan9303-core.c
>>> +++ b/drivers/net/dsa/lan9303-core.c
>>> @@ -1013,8 +1013,8 @@ static void lan9303_get_strings(struct dsa_switch *ds, int port,
>>> return;
>>>
>>> for (u = 0; u < ARRAY_SIZE(lan9303_mib); u++) {
>>> - strncpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
>>> - ETH_GSTRING_LEN);
>>> + memcpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
>>> + ETH_GSTRING_LEN);
>>
>> This won't work because lan9303_mib entries aren't ETH_GSTRING_LEN-long
>> strings; they're string pointers:
>>
>> static const struct lan9303_mib_desc lan9303_mib[] = {
>> { .offset = LAN9303_MAC_RX_BRDCST_CNT_0, .name = "RxBroad", },
>>
>> So this really does need a strcpy-family function.
>>
>> And, I think the vnic_gstrings_stats and ipoib_gstrings_stats examples
>> are actually buggy -- they're copying junk into userspace...
>>
>> I am reminded of this patch, which correctly uses strscpy_pad():
>> https://lore.kernel.org/lkml/[email protected]/
>>
>> I think you want to do the same here, and use strscpy_pad(). And perhaps
>> send some fixes for the other memcpy() users?
>
> Meh, I think it's not worth fixing the memcpy() users of this. This
> buggy pattern is very common, it seems:
>
> $ git grep 'data.*ETH_GSTRING_LEN' | grep memcpy | wc -l
> 47
We have ethtool_sprintf() precisely for the sake of filling the Ethtool
statistics names.
BTW this weird pattern "let's make the array of our stats names of fixed
width (ETH_GSTRING_LEN), so that we could use memcpy() instead of
strcpy()" was pretty common some time ago, no idea why, as it wastes
memory for tons of \0 padding and provokes issues like the one you
noticed here.
>
Thanks,
Olek
On Thu, Oct 5, 2023 at 7:54 AM Alexander Lobakin
<[email protected]> wrote:
>
> From: Kees Cook <[email protected]>
> Date: Wed, 4 Oct 2023 22:02:00 -0700
>
> > On Wed, Oct 04, 2023 at 08:07:55PM -0700, Kees Cook wrote:
> >> On Thu, Oct 05, 2023 at 12:30:18AM +0000, Justin Stitt wrote:
> >>> `strncpy` is deprecated for use on NUL-terminated destination strings
> >>> [1] and as such we should prefer more robust and less ambiguous
> >>> interfaces.
> >>>
> >>> Let's opt for memcpy as we are copying strings into slices of length
> >>> `ETH_GSTRING_LEN` within the `data` buffer. Other similar get_strings()
> >>> implementations [2] [3] use memcpy().
> >>>
> >>> Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings [1]
> >>> Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/opa_vnic/opa_vnic_ethtool.c#L167 [2]
> >>> Link: https://elixir.bootlin.com/linux/v6.3/source/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c#L137 [3]
> >>> Link: https://github.com/KSPP/linux/issues/90
> >>> Cc: [email protected]
> >>> Signed-off-by: Justin Stitt <[email protected]>
> >>> ---
> >>> Note: build-tested only.
> >>> ---
> >>> drivers/net/dsa/lan9303-core.c | 4 ++--
> >>> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/net/dsa/lan9303-core.c b/drivers/net/dsa/lan9303-core.c
> >>> index ee67adeb2cdb..665d69384b62 100644
> >>> --- a/drivers/net/dsa/lan9303-core.c
> >>> +++ b/drivers/net/dsa/lan9303-core.c
> >>> @@ -1013,8 +1013,8 @@ static void lan9303_get_strings(struct dsa_switch *ds, int port,
> >>> return;
> >>>
> >>> for (u = 0; u < ARRAY_SIZE(lan9303_mib); u++) {
> >>> - strncpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
> >>> - ETH_GSTRING_LEN);
> >>> + memcpy(data + u * ETH_GSTRING_LEN, lan9303_mib[u].name,
> >>> + ETH_GSTRING_LEN);
> >>
> >> This won't work because lan9303_mib entries aren't ETH_GSTRING_LEN-long
> >> strings; they're string pointers:
> >>
> >> static const struct lan9303_mib_desc lan9303_mib[] = {
> >> { .offset = LAN9303_MAC_RX_BRDCST_CNT_0, .name = "RxBroad", },
> >>
> >> So this really does need a strcpy-family function.
> >>
> >> And, I think the vnic_gstrings_stats and ipoib_gstrings_stats examples
> >> are actually buggy -- they're copying junk into userspace...
> >>
> >> I am reminded of this patch, which correctly uses strscpy_pad():
> >> https://lore.kernel.org/lkml/[email protected]/
> >>
> >> I think you want to do the same here, and use strscpy_pad(). And perhaps
> >> send some fixes for the other memcpy() users?
> >
> > Meh, I think it's not worth fixing the memcpy() users of this. This
> > buggy pattern is very common, it seems:
> >
> > $ git grep 'data.*ETH_GSTRING_LEN' | grep memcpy | wc -l
> > 47
>
> We have ethtool_sprintf() precisely for the sake of filling the Ethtool
> statistics names.
>
> BTW this weird pattern "let's make the array of our stats names of fixed
> width (ETH_GSTRING_LEN), so that we could use memcpy() instead of
> strcpy()" was pretty common some time ago, no idea why, as it wastes
> memory for tons of \0 padding and provokes issues like the one you
> noticed here.
Just sent a v2 using ethtool_sprintf(). I'd appreciate some feedback on if I
used it correctly.
>
> >
>
> Thanks,
> Olek
[v2]: https://lore.kernel.org/r/20231005-strncpy-drivers-net-dsa-lan9303-core-c-v2-1-feb452a532db@google.com
Thanks
Justin