Skip to contentSkip to conversation

scsi: target: pscsi: Fix possible null-pointer dereference in pscsi_complete_cmd()

PATCH Superseded Tuo Li 2 participants 3 messages Aug 7, 2021 – Aug 9, 2021

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Tuo Li

author
#0 lore raw

PATCH 1/1 scsi: target: pscsi: Fix possible null-pointer dereference in pscsi_complete_cmd()

The return value of transport_kmap_data_sg() is assigned to the variable buf:
buf = transport_kmap_data_sg(cmd);

And then it is checked:
if (!buf) {

This indicates that buf can be NULL. However, it is dereferenced in the following statements:
if (!(buf[3] & 0x80))

    buf[3] |= 0x80;
  if (!(buf[2] & 0x80))
	buf[2] |= 0x80;

To fix these possible null-pointer dereferences, dereference buf only when it is not NULL.

---
 drivers/target/target_core_pscsi.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/target/target_core_pscsi.c +7 -7
index 2629d2ef3970..560815729182 100644
@@ -620,14 +620,14 @@ static void pscsi_complete_cmd(struct se_cmd *cmd, u8 scsi_status,
 			buf = transport_kmap_data_sg(cmd);
 			if (!buf) {
 				; /* XXX: TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE */
-			}
-
-			if (cdb[0] == MODE_SENSE_10) {
-				if (!(buf[3] & 0x80))
-					buf[3] |= 0x80;
 			} else {
-				if (!(buf[2] & 0x80))
-					buf[2] |= 0x80;
+				if (cdb[0] == MODE_SENSE_10) {
+					if (!(buf[3] & 0x80))
+						buf[3] |= 0x80;
+				} else {
+					if (!(buf[2] & 0x80))
+						buf[2] |= 0x80;
+				}
 			}
 
 			transport_kunmap_data_sg(cmd);
Signature · 1 line
2.25.1
#1 lore raw
Quoted 40 lines · Tuo Li

The return value of transport_kmap_data_sg() is assigned to the variable buf:

   buf = transport_kmap_data_sg(cmd);

And then it is checked:

   if (!buf) {

This indicates that buf can be NULL. However, it is dereferenced in the following statements:

   if (!(buf[3] & 0x80))
     buf[3] |= 0x80;
   if (!(buf[2] & 0x80))
	buf[2] |= 0x80;

To fix these possible null-pointer dereferences, dereference buf only when it is not NULL.

---
  drivers/target/target_core_pscsi.c | 14 +++++++-------
  1 file changed, 7 insertions(+), 7 deletions(-)
drivers/target/target_core_pscsi.c +0 -7
index 2629d2ef3970..560815729182 100644
@@ -620,14 +620,14 @@ static void pscsi_complete_cmd(struct se_cmd *cmd, u8 scsi_status,
  			buf = transport_kmap_data_sg(cmd);
  			if (!buf) {
  				; /* XXX: TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE */
-			}
-
-			if (cdb[0] == MODE_SENSE_10) {
-				if (!(buf[3] & 0x80))
-					buf[3] |= 0x80;
  			} else {
-				if (!(buf[2] & 0x80))
-					buf[2] |= 0x80;

drivers/target/target_core_pscsi.c static void pscsi_complete_cmd(struct se_cmd *cmd, u8 scsi_status, · line 633

Quoted 10 lines
drivers/target/target_core_pscsi.c +7 -0
+				if (cdb[0] == MODE_SENSE_10) {
+					if (!(buf[3] & 0x80))
+						buf[3] |= 0x80;
+				} else {
+					if (!(buf[2] & 0x80))
+						buf[2] |= 0x80;
+				}
  			}
 
  			transport_kunmap_data_sg(cmd);

I'm wondering whether we should better put the transport_kunmap_data_sg into the else-branch of the if (!buf)? AFAICS, calling it after transport_kmap_data_sg failed does not cause problems, but I feel it would be cleaner.

Otherwise it looks good to me.

author
#2 lore raw

Thanks for your feedback. We will prepare a V2 patch and put the transport_kunmap_data_sg() into the else-branch of the if (!buf).

Best wishes,
Tuo Li

Quoted 58 lines · Bodo Stroesser
Quoted 41 lines · Tuo Li

The return value of transport_kmap_data_sg() is assigned to the variable buf:
   buf = transport_kmap_data_sg(cmd);

And then it is checked:
   if (!buf) {

This indicates that buf can be NULL. However, it is dereferenced in the following statements:
   if (!(buf[3] & 0x80))
     buf[3] |= 0x80;
   if (!(buf[2] & 0x80))
    buf[2] |= 0x80;

To fix these possible null-pointer dereferences, dereference buf only when it is not NULL.

---
  drivers/target/target_core_pscsi.c | 14 +++++++-------

  1 file changed, 7 insertions(+), 7 deletions(-)
drivers/target/target_core_pscsi.c +0 -0
index 2629d2ef3970..560815729182 100644
@@ -620,14 +620,14 @@ static void pscsi_complete_cmd(struct se_cmd 
 *cmd, u8 scsi_status,
               buf = transport_kmap_data_sg(cmd);
               if (!buf) {

                  ; /* XXX: TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE */ -            }
-
-            if (cdb[0] == MODE_SENSE_10) {
-                if (!(buf[3] & 0x80))
-                    buf[3] |= 0x80;
              } else {
-                if (!(buf[2] & 0x80))
-                    buf[2] |= 0x80;
+                if (cdb[0] == MODE_SENSE_10) {

Quoted 9 lines

+                    if (!(buf[3] & 0x80))
+                        buf[3] |= 0x80;
+                } else {
+                    if (!(buf[2] & 0x80))
+                        buf[2] |= 0x80;
+                }
              }
                transport_kunmap_data_sg(cmd);

I'm wondering whether we should better put the transport_kunmap_data_sg into the else-branch of the if (!buf)? AFAICS, calling it after transport_kmap_data_sg failed does not cause problems, but I feel it would be cleaner.

Otherwise it looks good to me.