Skip to contentSkip to conversation

scsi: ufs: should not override buffer lengh

PATCH In review Jaegeuk Kim 3 participants 6 messages Jan 11, 2021 SCSI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 scsi: ufs: should not override buffer lengh

From: Jaegeuk Kim <[email protected]>

Kernel stack violation when getting unit_descriptor/wb_buf_alloc_units from rpmb lun. The reason is the unit descriptor length is different per LU.

The lengh of Normal LU is 45, while the one of rpmb LU is 35.

int ufshcd_read_desc_param(struct ufs_hba *hba, ...)

{
	param_offset=41;
	param_size=4;
	buff_len=45;
	...
	buff_len=35 by rpmb LU;

	if (is_kmalloc) {
		/* Make sure we don't copy more data than available */
		if (param_offset + param_size > buff_len)
			param_size = buff_len - param_offset;
			--> param_size = 250;
		memcpy(param_read_buf, &desc_buf[param_offset], param_size);
		--> memcpy(param_read_buf, desc_buf+41, 250);
[  141.868974][ T9174] Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: wb_buf_alloc_units_show+0x11c/0x11c
	}

}

---
 drivers/scsi/ufs/ufshcd.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/scsi/ufs/ufshcd.c +6 -2
index 2a715f13fe1d..722697b57777 100644
@@ -3293,8 +3293,12 @@ int ufshcd_read_desc_param(struct ufs_hba *hba,
 
 	if (is_kmalloc) {
 		/* Make sure we don't copy more data than available */
-		if (param_offset + param_size > buff_len)
-			param_size = buff_len - param_offset;
+		if (param_offset + param_size > buff_len) {
+			if (buff_len > param_offset)
+				param_size = buff_len - param_offset;
+			else
+				param_size = 0;
+		}
 		memcpy(param_read_buf, &desc_buf[param_offset], param_size);
 	}
 out:
Signature · 1 line
2.30.0.284.gd98b1dd5eaa7-goog
#1 lore raw

Hi Jaegeuk,

I think the problem is that func ufshcd_read_desc_param() is not expecting one access unsupported descriptors on all W-LUs, not just RPMB LU.

If we can get the right buf_len from func ufshcd_map_desc_id_to_length(), the issue won't happen. - https://lore.kernel.org/patchwork/patch/1323421/.

What do you think if we update ufshcd_map_desc_id_to_length(add one param - desc_index) so that it can tell the correct buf_len in case of W-LUs?

Thanks,
Can Guo.

Quoted 52 lines · Jaegeuk Kim

From: Jaegeuk Kim <[email protected]>

Kernel stack violation when getting unit_descriptor/wb_buf_alloc_units from rpmb lun. The reason is the unit descriptor length is different per LU.

The lengh of Normal LU is 45, while the one of rpmb LU is 35.

int ufshcd_read_desc_param(struct ufs_hba *hba, ...)

{
	param_offset=41;
	param_size=4;
	buff_len=45;
	...
	buff_len=35 by rpmb LU;

	if (is_kmalloc) {
		/* Make sure we don't copy more data than available */
		if (param_offset + param_size > buff_len)
			param_size = buff_len - param_offset;
			--> param_size = 250;
		memcpy(param_read_buf, &desc_buf[param_offset], param_size);
		--> memcpy(param_read_buf, desc_buf+41, 250);
[  141.868974][ T9174] Kernel panic - not syncing: stack-protector:
Kernel stack is corrupted in: wb_buf_alloc_units_show+0x11c/0x11c
	}
}
---
 drivers/scsi/ufs/ufshcd.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/scsi/ufs/ufshcd.c +6 -2
index 2a715f13fe1d..722697b57777 100644
@@ -3293,8 +3293,12 @@ int ufshcd_read_desc_param(struct ufs_hba *hba,
 
 	if (is_kmalloc) {
 		/* Make sure we don't copy more data than available */
-		if (param_offset + param_size > buff_len)
-			param_size = buff_len - param_offset;
+		if (param_offset + param_size > buff_len) {
+			if (buff_len > param_offset)
+				param_size = buff_len - param_offset;
+			else
+				param_size = 0;
+		}
 		memcpy(param_read_buf, &desc_buf[param_offset], param_size);
 	}
 out:
#2 lore raw

Sorry, typo corrected.

Hi Jaegeuk,

I think the problem is that func ufshcd_read_desc_param() is not expecting one access unsupported descriptors on RPMB LU.

If we can get the right buf_len from func ufshcd_map_desc_id_to_length(), the issue won't happen. - https://lore.kernel.org/patchwork/patch/1323421/.

What do you think if we update ufshcd_map_desc_id_to_length(add one param - desc_index) so that it can tell the correct buf_len in case of RPMB LU?

Thanks,
Can Guo.

Quoted 53 lines · Jaegeuk Kim
Quoted 52 lines

From: Jaegeuk Kim <[email protected]>

Kernel stack violation when getting unit_descriptor/wb_buf_alloc_units from rpmb lun. The reason is the unit descriptor length is different per LU.

The lengh of Normal LU is 45, while the one of rpmb LU is 35.

int ufshcd_read_desc_param(struct ufs_hba *hba, ...)

{
	param_offset=41;
	param_size=4;
	buff_len=45;
	...
	buff_len=35 by rpmb LU;

	if (is_kmalloc) {
		/* Make sure we don't copy more data than available */
		if (param_offset + param_size > buff_len)
			param_size = buff_len - param_offset;
			--> param_size = 250;
		memcpy(param_read_buf, &desc_buf[param_offset], param_size);
		--> memcpy(param_read_buf, desc_buf+41, 250);
[  141.868974][ T9174] Kernel panic - not syncing: stack-protector:
Kernel stack is corrupted in: wb_buf_alloc_units_show+0x11c/0x11c
	}
}
---
 drivers/scsi/ufs/ufshcd.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/scsi/ufs/ufshcd.c +6 -2
index 2a715f13fe1d..722697b57777 100644
@@ -3293,8 +3293,12 @@ int ufshcd_read_desc_param(struct ufs_hba *hba,
 
 	if (is_kmalloc) {
 		/* Make sure we don't copy more data than available */
-		if (param_offset + param_size > buff_len)
-			param_size = buff_len - param_offset;
+		if (param_offset + param_size > buff_len) {
+			if (buff_len > param_offset)
+				param_size = buff_len - param_offset;
+			else
+				param_size = 0;
+		}
 		memcpy(param_read_buf, &desc_buf[param_offset], param_size);
 	}
 out:
#3 lore raw
Quoted 8 lines

Sorry, typo corrected.

Hi Jaegeuk,

I think the problem is that func ufshcd_read_desc_param() is not expecting
one access unsupported descriptors on RPMB LU.

Correct.
This is about wb introducing a new constraint: wb buffer is only allowed in lu 0..7. And this is why, IMHO, the fix should be in ufs_is_valid_unit_desc_lun, To include param offset, as it is only called in contingency of ufshcd_read_desc_param.

Thanks,
Avri

#4 lore raw
Quoted 15 lines · Avri Altman
Quoted 8 lines

Sorry, typo corrected.

Hi Jaegeuk,

I think the problem is that func ufshcd_read_desc_param() is not expecting
one access unsupported descriptors on RPMB LU.

Correct.
This is about wb introducing a new constraint: wb buffer is only allowed in lu 0..7.
And this is why, IMHO, the fix should be in ufs_is_valid_unit_desc_lun, To include param offset, as it is only called in contingency of ufshcd_read_desc_param.

Quoted 2 lines

Thanks,
Avri

Yeah... Fixing it in ufs-sysfs.c also works. Anyways, the math in ufshcd_read_desc_param is already complex. Let's fix it somewhere close to the source/initiator.

Thanks,
Can Guo.

#5 lore raw
Quoted 16 lines · Can Guo
Quoted 15 lines · Avri Altman
Quoted 8 lines

Sorry, typo corrected.

Hi Jaegeuk,

I think the problem is that func ufshcd_read_desc_param() is not expecting
one access unsupported descriptors on RPMB LU.

Correct.
This is about wb introducing a new constraint: wb buffer is only allowed in lu 0..7.
And this is why, IMHO, the fix should be in ufs_is_valid_unit_desc_lun, To include param offset, as it is only called in contingency of ufshcd_read_desc_param.

Quoted 6 lines
Quoted 2 lines

Thanks,
Avri

Yeah... Fixing it in ufs-sysfs.c also works. Anyways, the math in ufshcd_read_desc_param is already complex. Let's fix it somewhere close to the source/initiator.

Thank you, Can and Avri.
I think fixing the lun check makese sense. Let me post v2. :)

Quoted 3 lines

Thanks,
Can Guo.