strcpy() performs no bounds checking on the destination buffer. This
could result in linear overflows beyond the end of the buffer, leading
to all kinds of misbehaviors. So, remove all the uses and add
devm_kstrdup() or devm_kasprintf() instead.
This patch is an effort to clean up the proliferation of str*()
functions in the kernel and a previous step in the path to remove
the strcpy function from the kernel entirely [1].
[1] https://github.com/KSPP/linux/issues/88
Signed-off-by: Len Baker <[email protected]>
---
The previous versions can be found in:
v1
https://lore.kernel.org/linux-hardening/[email protected]/
v2
https://lore.kernel.org/linux-hardening/[email protected]/
v3
https://lore.kernel.org/linux-hardening/[email protected]/
Changelog v1 -> v2
- Modify the commit changelog to inform that the motivation of this
patch is to remove the strcpy() function from the kernel entirely
(Jonathan Cameron).
Changelog v2 -> v3
- Rewrite the code using devm_kstrdup() and devm_kasprintf() functions
(Andy Shevchenko).
- Rebase against v5.14-rc5.
Changelog v3 -> v4
- Reorder the variables (Andy Shevchenko).
- Get the device in the definition block (Andy Shevchenko).
- Reword the comment (Andy Shevchenko).
- Change the conditions in the "if" statement to clarify the "0" check
(Andy Shevchenko).
drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c | 30 +++++++++++++---------
1 file changed, 18 insertions(+), 12 deletions(-)
diff --git a/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c b/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
index f282e9cc34c5..7eceae0012c9 100644
--- a/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
+++ b/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
@@ -261,6 +261,7 @@ int inv_mpu_magn_set_rate(const struct inv_mpu6050_state *st, int fifo_rate)
*/
int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
{
+ struct device *dev = regmap_get_device(st->map);
const char *orient;
char *str;
int i;
@@ -281,19 +282,24 @@ int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
/* z <- -z */
for (i = 0; i < 3; ++i) {
orient = st->orientation.rotation[6 + i];
- /* use length + 2 for adding minus sign if needed */
- str = devm_kzalloc(regmap_get_device(st->map),
- strlen(orient) + 2, GFP_KERNEL);
- if (str == NULL)
+
+ /*
+ * The value is negated according to one of the following
+ * rules:
+ *
+ * 1) Drop leading minus.
+ * 2) Leave 0 as is.
+ * 3) Add leading minus.
+ */
+ if (orient[0] == '-')
+ str = devm_kstrdup(dev, orient + 1, GFP_KERNEL);
+ else if (orient[0] == '0' && orient[1] == '\0')
+ str = devm_kstrdup(dev, orient, GFP_KERNEL);
+ else
+ str = devm_kasprintf(dev, GFP_KERNEL, "-%s", orient);
+ if (!str)
return -ENOMEM;
- if (strcmp(orient, "0") == 0) {
- strcpy(str, orient);
- } else if (orient[0] == '-') {
- strcpy(str, &orient[1]);
- } else {
- str[0] = '-';
- strcpy(&str[1], orient);
- }
+
st->magn_orient.rotation[6 + i] = str;
}
break;
--
2.25.1
On Sat, Aug 14, 2021 at 4:55 PM Len Baker <[email protected]> wrote:
>
> strcpy() performs no bounds checking on the destination buffer. This
> could result in linear overflows beyond the end of the buffer, leading
> to all kinds of misbehaviors. So, remove all the uses and add
> devm_kstrdup() or devm_kasprintf() instead.
>
> This patch is an effort to clean up the proliferation of str*()
> functions in the kernel and a previous step in the path to remove
> the strcpy function from the kernel entirely [1].
>
> [1] https://github.com/KSPP/linux/issues/88
Thank you very much for doing this!
Now I like the result,
Reviewed-by: Andy Shevchenko <[email protected]>
> Signed-off-by: Len Baker <[email protected]>
> ---
> The previous versions can be found in:
>
> v1
> https://lore.kernel.org/linux-hardening/[email protected]/
>
> v2
> https://lore.kernel.org/linux-hardening/[email protected]/
>
> v3
> https://lore.kernel.org/linux-hardening/[email protected]/
>
> Changelog v1 -> v2
> - Modify the commit changelog to inform that the motivation of this
> patch is to remove the strcpy() function from the kernel entirely
> (Jonathan Cameron).
>
> Changelog v2 -> v3
> - Rewrite the code using devm_kstrdup() and devm_kasprintf() functions
> (Andy Shevchenko).
> - Rebase against v5.14-rc5.
>
> Changelog v3 -> v4
> - Reorder the variables (Andy Shevchenko).
> - Get the device in the definition block (Andy Shevchenko).
> - Reword the comment (Andy Shevchenko).
> - Change the conditions in the "if" statement to clarify the "0" check
> (Andy Shevchenko).
>
> drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c | 30 +++++++++++++---------
> 1 file changed, 18 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c b/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
> index f282e9cc34c5..7eceae0012c9 100644
> --- a/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
> +++ b/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
> @@ -261,6 +261,7 @@ int inv_mpu_magn_set_rate(const struct inv_mpu6050_state *st, int fifo_rate)
> */
> int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
> {
> + struct device *dev = regmap_get_device(st->map);
> const char *orient;
> char *str;
> int i;
> @@ -281,19 +282,24 @@ int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
> /* z <- -z */
> for (i = 0; i < 3; ++i) {
> orient = st->orientation.rotation[6 + i];
> - /* use length + 2 for adding minus sign if needed */
> - str = devm_kzalloc(regmap_get_device(st->map),
> - strlen(orient) + 2, GFP_KERNEL);
> - if (str == NULL)
> +
> + /*
> + * The value is negated according to one of the following
> + * rules:
> + *
> + * 1) Drop leading minus.
> + * 2) Leave 0 as is.
> + * 3) Add leading minus.
> + */
> + if (orient[0] == '-')
> + str = devm_kstrdup(dev, orient + 1, GFP_KERNEL);
> + else if (orient[0] == '0' && orient[1] == '\0')
> + str = devm_kstrdup(dev, orient, GFP_KERNEL);
> + else
> + str = devm_kasprintf(dev, GFP_KERNEL, "-%s", orient);
> + if (!str)
> return -ENOMEM;
> - if (strcmp(orient, "0") == 0) {
> - strcpy(str, orient);
> - } else if (orient[0] == '-') {
> - strcpy(str, &orient[1]);
> - } else {
> - str[0] = '-';
> - strcpy(&str[1], orient);
> - }
> +
> st->magn_orient.rotation[6 + i] = str;
> }
> break;
> --
> 2.25.1
>
--
With Best Regards,
Andy Shevchenko
Hi Andy,
On Sat, Aug 14, 2021 at 10:36:18PM +0300, Andy Shevchenko wrote:
> On Sat, Aug 14, 2021 at 4:55 PM Len Baker <[email protected]> wrote:
> >
> > strcpy() performs no bounds checking on the destination buffer. This
> > could result in linear overflows beyond the end of the buffer, leading
> > to all kinds of misbehaviors. So, remove all the uses and add
> > devm_kstrdup() or devm_kasprintf() instead.
> >
> > This patch is an effort to clean up the proliferation of str*()
> > functions in the kernel and a previous step in the path to remove
> > the strcpy function from the kernel entirely [1].
> >
> > [1] https://github.com/KSPP/linux/issues/88
>
> Thank you very much for doing this!
> Now I like the result,
> Reviewed-by: Andy Shevchenko <[email protected]>
Thank you too Andy (and folks) for your help on this.
Regards,
Len
On Sun, 15 Aug 2021 10:19:49 +0200
Len Baker <[email protected]> wrote:
> Hi Andy,
>
> On Sat, Aug 14, 2021 at 10:36:18PM +0300, Andy Shevchenko wrote:
> > On Sat, Aug 14, 2021 at 4:55 PM Len Baker <[email protected]> wrote:
> > >
> > > strcpy() performs no bounds checking on the destination buffer. This
> > > could result in linear overflows beyond the end of the buffer, leading
> > > to all kinds of misbehaviors. So, remove all the uses and add
> > > devm_kstrdup() or devm_kasprintf() instead.
> > >
> > > This patch is an effort to clean up the proliferation of str*()
> > > functions in the kernel and a previous step in the path to remove
> > > the strcpy function from the kernel entirely [1].
> > >
> > > [1] https://github.com/KSPP/linux/issues/88
> >
> > Thank you very much for doing this!
> > Now I like the result,
Agreed and applied to the togreg branch of iio.git, pushed out as testing
for 0-day to poke at it and see if we missed anything.
Thanks,
Jonathan
> > Reviewed-by: Andy Shevchenko <[email protected]>
>
> Thank you too Andy (and folks) for your help on this.
>
> Regards,
> Len
On Sat, 2021-08-14 at 22:36 +0300, Andy Shevchenko wrote:
> On Sat, Aug 14, 2021 at 4:55 PM Len Baker <[email protected]> wrote:
> >
> > strcpy() performs no bounds checking on the destination buffer. This
> > could result in linear overflows beyond the end of the buffer, leading
> > to all kinds of misbehaviors. So, remove all the uses and add
> > devm_kstrdup() or devm_kasprintf() instead.
> >
> > This patch is an effort to clean up the proliferation of str*()
> > functions in the kernel and a previous step in the path to remove
> > the strcpy function from the kernel entirely [1].
> >
> > [1] https://github.com/KSPP/linux/issues/88
>
> Thank you very much for doing this!
> Now I like the result,
> Reviewed-by: Andy Shevchenko <[email protected]>
[]
> > diff --git a/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c b/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
[]
> > @@ -261,6 +261,7 @@ int inv_mpu_magn_set_rate(const struct inv_mpu6050_state *st, int fifo_rate)
> > ??*/
> > ?int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
> > ?{
> > + struct device *dev = regmap_get_device(st->map);
> > ????????const char *orient;
> > ????????char *str;
> > ????????int i;
> > @@ -281,19 +282,24 @@ int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
> > ????????????????/* z <- -z */
> > ????????????????for (i = 0; i < 3; ++i) {
> > ????????????????????????orient = st->orientation.rotation[6 + i];
> > - /* use length + 2 for adding minus sign if needed */
> > - str = devm_kzalloc(regmap_get_device(st->map),
> > - strlen(orient) + 2, GFP_KERNEL);
> > - if (str == NULL)
> > +
> > + /*
> > + * The value is negated according to one of the following
> > + * rules:
> > + *
> > + * 1) Drop leading minus.
> > + * 2) Leave 0 as is.
> > + * 3) Add leading minus.
> > + */
> > + if (orient[0] == '-')
> > + str = devm_kstrdup(dev, orient + 1, GFP_KERNEL);
> > + else if (orient[0] == '0' && orient[1] == '\0')
bikeshed:
I think this change is less intelligible than the original strcmp.
And separately, perhaps for loop would be more readable as
for (i = 6; i < 9; i++)
converting the 6 + i uses to just i.
> > + str = devm_kstrdup(dev, orient, GFP_KERNEL);
> > + else
> > + str = devm_kasprintf(dev, GFP_KERNEL, "-%s", orient);
> > + if (!str)
> > ????????????????????????????????return -ENOMEM;
> > - if (strcmp(orient, "0") == 0) {
> > - strcpy(str, orient);
> > - } else if (orient[0] == '-') {
> > - strcpy(str, &orient[1]);
> > - } else {
> > - str[0] = '-';
> > - strcpy(&str[1], orient);
> > - }
> > +
> > ????????????????????????st->magn_orient.rotation[6 + i] = str;
> > ????????????????}
> > ????????????????break;
On Sun, 15 Aug 2021 15:45:55 +0100
Jonathan Cameron <[email protected]> wrote:
> On Sun, 15 Aug 2021 10:19:49 +0200
> Len Baker <[email protected]> wrote:
>
> > Hi Andy,
> >
> > On Sat, Aug 14, 2021 at 10:36:18PM +0300, Andy Shevchenko wrote:
> > > On Sat, Aug 14, 2021 at 4:55 PM Len Baker <[email protected]> wrote:
> > > >
> > > > strcpy() performs no bounds checking on the destination buffer. This
> > > > could result in linear overflows beyond the end of the buffer, leading
> > > > to all kinds of misbehaviors. So, remove all the uses and add
> > > > devm_kstrdup() or devm_kasprintf() instead.
> > > >
> > > > This patch is an effort to clean up the proliferation of str*()
> > > > functions in the kernel and a previous step in the path to remove
> > > > the strcpy function from the kernel entirely [1].
> > > >
> > > > [1] https://github.com/KSPP/linux/issues/88
> > >
> > > Thank you very much for doing this!
> > > Now I like the result,
> Agreed and applied to the togreg branch of iio.git, pushed out as testing
> for 0-day to poke at it and see if we missed anything.
Dropped it for now so that Joe's comment can be addressed / discussed.
Jonathan
>
> Thanks,
>
> Jonathan
> > > Reviewed-by: Andy Shevchenko <[email protected]>
>
> >
> > Thank you too Andy (and folks) for your help on this.
> >
> > Regards,
> > Len
>
Hi Joe,
On Sun, Aug 15, 2021 at 08:06:45AM -0700, Joe Perches wrote:
> On Sat, 2021-08-14 at 22:36 +0300, Andy Shevchenko wrote:
> > On Sat, Aug 14, 2021 at 4:55 PM Len Baker <[email protected]> wrote:
> > >
> > > strcpy() performs no bounds checking on the destination buffer. This
> > > could result in linear overflows beyond the end of the buffer, leading
> > > to all kinds of misbehaviors. So, remove all the uses and add
> > > devm_kstrdup() or devm_kasprintf() instead.
> > >
> > > This patch is an effort to clean up the proliferation of str*()
> > > functions in the kernel and a previous step in the path to remove
> > > the strcpy function from the kernel entirely [1].
> > >
> > > [1] https://github.com/KSPP/linux/issues/88
> >
> > Thank you very much for doing this!
> > Now I like the result,
> > Reviewed-by: Andy Shevchenko <[email protected]>
> []
> > > diff --git a/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c b/drivers/iio/imu/inv_mpu6050/inv_mpu_magn.c
> []
> > > @@ -261,6 +261,7 @@ int inv_mpu_magn_set_rate(const struct inv_mpu6050_state *st, int fifo_rate)
> > > ??*/
> > > ?int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
> > > ?{
> > > + struct device *dev = regmap_get_device(st->map);
> > > ????????const char *orient;
> > > ????????char *str;
> > > ????????int i;
> > > @@ -281,19 +282,24 @@ int inv_mpu_magn_set_orient(struct inv_mpu6050_state *st)
> > > ????????????????/* z <- -z */
> > > ????????????????for (i = 0; i < 3; ++i) {
> > > ????????????????????????orient = st->orientation.rotation[6 + i];
> > > - /* use length + 2 for adding minus sign if needed */
> > > - str = devm_kzalloc(regmap_get_device(st->map),
> > > - strlen(orient) + 2, GFP_KERNEL);
> > > - if (str == NULL)
> > > +
> > > + /*
> > > + * The value is negated according to one of the following
> > > + * rules:
> > > + *
> > > + * 1) Drop leading minus.
> > > + * 2) Leave 0 as is.
> > > + * 3) Add leading minus.
> > > + */
> > > + if (orient[0] == '-')
> > > + str = devm_kstrdup(dev, orient + 1, GFP_KERNEL);
> > > + else if (orient[0] == '0' && orient[1] == '\0')
>
> bikeshed:
>
> I think this change is less intelligible than the original strcmp.
So, if I understand correctly you suggest to change the above line for:
else if (strcmp(orient, "0") == 0)
And yes, this line will be more readable ;)
>
> And separately, perhaps for loop would be more readable as
>
> for (i = 6; i < 9; i++)
>
> converting the 6 + i uses to just i.
Ok, no problem. If there are no objections I will send a new version with
these 2 changes.
Regards,
Len
>
> > > + str = devm_kstrdup(dev, orient, GFP_KERNEL);
> > > + else
> > > + str = devm_kasprintf(dev, GFP_KERNEL, "-%s", orient);
> > > + if (!str)
> > > ????????????????????????????????return -ENOMEM;
> > > - if (strcmp(orient, "0") == 0) {
> > > - strcpy(str, orient);
> > > - } else if (orient[0] == '-') {
> > > - strcpy(str, &orient[1]);
> > > - } else {
> > > - str[0] = '-';
> > > - strcpy(&str[1], orient);
> > > - }
> > > +
> > > ????????????????????????st->magn_orient.rotation[6 + i] = str;
> > > ????????????????}
> > > ????????????????break;
>
>
On Sun, 2021-08-15 at 18:36 +0200, Len Baker wrote:
> Hi Joe,
Hello Len.
Don't take this advice too seriously, it's just bikeshedding
but it seems to me an unexpected use of a strcmp equivalent
in a non performance sensitive path.
> On Sun, Aug 15, 2021 at 08:06:45AM -0700, Joe Perches wrote:
[]
> > bikeshed:
> >
> > I think this change is less intelligible than the original strcmp.
>
> So, if I understand correctly you suggest to change the above line for:
> ??????????????????????????????else if (strcmp(orient, "0") == 0)
Yes.
In kernel sources it's about 2:1 in favor of '!strcmp()' over 'strcmp() == 0'
$ git grep -P '\!\s*strcmp\b' | wc -l
3457
$ git grep -P '\bstrcmp\s*\([^\)]+\)\s*==\s*0\b' | wc -l
1719
And it's your choice to use one or the other or just your V4 patch.
btw, according to godbolt:
gcc -O2 doesn't call strcmp and produces the same object code as your
byte comparisons. clang 11 calls strcmp regardless of optimization level.
Hi,
On Sun, Aug 15, 2021 at 09:58:02AM -0700, Joe Perches wrote:
> On Sun, 2021-08-15 at 18:36 +0200, Len Baker wrote:
> > Hi Joe,
>
> Hello Len.
>
> Don't take this advice too seriously, it's just bikeshedding
> but it seems to me an unexpected use of a strcmp equivalent
> in a non performance sensitive path.
>
> > On Sun, Aug 15, 2021 at 08:06:45AM -0700, Joe Perches wrote:
> []
> > > bikeshed:
> > >
> > > I think this change is less intelligible than the original strcmp.
> >
> > So, if I understand correctly you suggest to change the above line for:
> > ??????????????????????????????else if (strcmp(orient, "0") == 0)
>
> Yes.
>
> In kernel sources it's about 2:1 in favor of '!strcmp()' over 'strcmp() == 0'
>
> $ git grep -P '\!\s*strcmp\b' | wc -l
> 3457
> $ git grep -P '\bstrcmp\s*\([^\)]+\)\s*==\s*0\b' | wc -l
> 1719
>
> And it's your choice to use one or the other or just your V4 patch.
I will increase the !strcmp() statistics ;)
>
> btw, according to godbolt:
>
> gcc -O2 doesn't call strcmp and produces the same object code as your
> byte comparisons. clang 11 calls strcmp regardless of optimization level.
>
>
Thanks for the feedback,
Len