Skip to contentSkip to conversation

drm: A couple of fixes for drm_copy_field() helper function

PATCH In review Javier Martinez Canillas 1 participant 4 messages Jul 5, 2022 – Jul 15, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v2 cover drm: A couple of fixes for drm_copy_field() helper function

Hello,

Peter Robinson reported me a kernel bug in one of his aarch64 test boards and even though I was not able to reproduce it, I think that figured out what the problem was. It seems the cause is that a DRM driver doesn't set some of the struct drm fields copied to userspace via DRM_IOCTL_VERSION.

Even though this is a driver bug, we can make drm_copy_field() more robust and warn about it instead of leading to an attempt to copy a NULL pointer.

While looking at this, I also found that a variable in drm_copy_field() is not using the correct type. So I included that change in the patch-set too.

Best regards,
Javier

Changes in v2:
- Add Peter Robinson Tested-by and Thomas Zimmermann Reviewed-by tags.
- Just warn if a value isn't set and report it as a string of length 0. (Thomas Zimmermann).

Javier Martinez Canillas (2):
drm: Use size_t type for len variable in drm_copy_field()
drm: Prevent drm_copy_field() to attempt copying a NULL pointer

 drivers/gpu/drm/drm_ioctl.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
Signature · 1 line
2.36.1
#1 lore raw

PATCH v2 1/2 drm: Use size_t type for len variable in drm_copy_field()

The strlen() function returns a size_t which is an unsigned int on 32-bit arches and an unsigned long on 64-bit arches. But in the drm_copy_field() function, the strlen() return value is assigned to an 'int len' variable.

Later, the len variable is passed as copy_from_user() third argument that is an unsigned long parameter as well.

In theory, this can lead to an integer overflow via type conversion. Since the assignment happens to a signed int lvalue instead of a size_t lvalue.

In practice though, that's unlikely since the values copied are set by DRM drivers and not controlled by userspace. But using a size_t for len is the correct thing to do anyways.

---

Changes in v2:
- Add Peter Robinson Tested-by and Thomas Zimmermann Reviewed-by tags.

 drivers/gpu/drm/drm_ioctl.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_ioctl.c +1 -1
index 8faad23dc1d8..e1b9a03e619c 100644
@@ -472,7 +472,7 @@ EXPORT_SYMBOL(drm_invalid_op);
  */
 static int drm_copy_field(char __user *buf, size_t *buf_len, const char *value)
 {
-	int len;
+	size_t len;
 
 	/* don't overflow userbuf */
 	len = strlen(value);
Signature · 1 line
2.36.1
#2 lore raw

PATCH v2 2/2 drm: Prevent drm_copy_field() to attempt copying a NULL pointer

In reply to Javier Martinez Canillas

There are some struct drm_driver fields that are required by drivers since drm_copy_field() attempts to copy them to user-space via DRM_IOCTL_VERSION.

But it can be possible that a driver has a bug and did not set some of the fields, which leads to drm_copy_field() attempting to copy a NULL pointer:

[ +10.395966] Unable to handle kernel access to user memory outside uaccess routines at virtual address 0000000000000000

[  +0.010955] Mem abort info:
[  +0.002835]   ESR = 0x0000000096000004
[  +0.003872]   EC = 0x25: DABT (current EL), IL = 32 bits
[  +0.005395]   SET = 0, FnV = 0
[  +0.003113]   EA = 0, S1PTW = 0
[  +0.003182]   FSC = 0x04: level 0 translation fault
[  +0.004964] Data abort info:
[  +0.002919]   ISV = 0, ISS = 0x00000004
[  +0.003886]   CM = 0, WnR = 0
[  +0.003040] user pgtable: 4k pages, 48-bit VAs, pgdp=0000000115dad000
[  +0.006536] [0000000000000000] pgd=0000000000000000, p4d=0000000000000000
[  +0.006925] Internal error: Oops: 96000004 [#1] SMP
...
[  +0.011113] pstate: 80400005 (Nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[  +0.007061] pc : __pi_strlen+0x14/0x150
[  +0.003895] lr : drm_copy_field+0x30/0x1a4
[  +0.004156] sp : ffff8000094b3a50
[  +0.003355] x29: ffff8000094b3a50 x28: ffff8000094b3b70 x27: 0000000000000040
[  +0.007242] x26: ffff443743c2ba00 x25: 0000000000000000 x24: 0000000000000040
[  +0.007243] x23: ffff443743c2ba00 x22: ffff8000094b3b70 x21: 0000000000000000
[  +0.007241] x20: 0000000000000000 x19: ffff8000094b3b90 x18: 0000000000000000
[  +0.007241] x17: 0000000000000000 x16: 0000000000000000 x15: 0000aaab14b9af40
[  +0.007241] x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
[  +0.007239] x11: 0000000000000000 x10: 0000000000000000 x9 : ffffa524ad67d4d8
[  +0.007242] x8 : 0101010101010101 x7 : 7f7f7f7f7f7f7f7f x6 : 6c6e6263606e7141
[  +0.007239] x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000
[  +0.007241] x2 : 0000000000000000 x1 : ffff8000094b3b90 x0 : 0000000000000000
[  +0.007240] Call trace:
[  +0.002475]  __pi_strlen+0x14/0x150
[  +0.003537]  drm_version+0x84/0xac
[  +0.003448]  drm_ioctl_kernel+0xa8/0x16c
[  +0.003975]  drm_ioctl+0x270/0x580
[  +0.003448]  __arm64_sys_ioctl+0xb8/0xfc
[  +0.003978]  invoke_syscall+0x78/0x100
[  +0.003799]  el0_svc_common.constprop.0+0x4c/0xf4
[  +0.004767]  do_el0_svc+0x38/0x4c
[  +0.003357]  el0_svc+0x34/0x100
[  +0.003185]  el0t_64_sync_handler+0x11c/0x150
[  +0.004418]  el0t_64_sync+0x190/0x194
[  +0.003716] Code: 92402c04 b200c3e8 f13fc09f 5400088c (a9400c02)
[  +0.006180] ---[ end trace 0000000000000000 ]---

---

Changes in v2:
- Just warn if a value isn't set and report it as a string of length 0. (Thomas Zimmermann).

 drivers/gpu/drm/drm_ioctl.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/drm_ioctl.c +6 -0
index e1b9a03e619c..ca2a6e6101dc 100644
@@ -474,6 +474,12 @@ static int drm_copy_field(char __user *buf, size_t *buf_len, const char *value)
 {
 	size_t len;
 
+	/* don't attempt to copy a NULL pointer */
+	if (WARN_ONCE(!value, "BUG: the value to copy was not set!")) {
+		*buf_len = 0;
+		return 0;
+	}
+
 	/* don't overflow userbuf */
 	len = strlen(value);
 	if (len > *buf_len)
Signature · 1 line
2.36.1
#3 lore raw
In reply to Javier Martinez Canillas
Quoted 7 lines · Javier Martinez Canillas

Hello,

Peter Robinson reported me a kernel bug in one of his aarch64 test boards and even though I was not able to reproduce it, I think that figured out what the problem was. It seems the cause is that a DRM driver doesn't set some of the struct drm fields copied to userspace via DRM_IOCTL_VERSION.

Quoted 6 lines

Even though this is a driver bug, we can make drm_copy_field() more robust and warn about it instead of leading to an attempt to copy a NULL pointer.

While looking at this, I also found that a variable in drm_copy_field() is not using the correct type. So I included that change in the patch-set too.

Pushed these to drm-misc (drm-misc-next). Thanks!

Signature · 5 lines
Best regards,

Javier Martinez Canillas
Linux Engineering
Red Hat