Skip to contentSkip to conversation

staging/sep: Fix smatch false positive about potential NULL dereference in sep_main.c

PATCH In review Peter Huewe 2 participants 3 messages Feb 19, 2013 Staging

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 staging/sep: Fix smatch false positive about potential NULL dereference in sep_main.c

Smatch complains about a potential NULL pointer dereference:

sep_main.c:2312 sep_construct_dma_tables_from_lli() error: potential NULL dereference 'info_out_entry_ptr'.

info_out_entry_ptr is initialized with NULL and if info_in_entry_ptr is not NULL it gets derefenced.
However info_out_entry_ptr is only NULL in the first iteration of the while loop and in this case info_in_entry_ptr is also NULL (as indicated by the comment /* If info entry is null - this is the first table built */ -> this is a false positive.

Nevertheless we add a check for info_out_entry_ptr to silence this warning and make it more robust in regard to code changes.

---
 drivers/staging/sep/sep_main.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/staging/sep/sep_main.c +1 -1
index 30e8d25..366d56b 100644
@@ -2276,7 +2276,7 @@ static int sep_construct_dma_tables_from_lli(
 			table_data_size);
 
 		/* If info entry is null - this is the first table built */
-		if (info_in_entry_ptr == NULL) {
+		if (info_in_entry_ptr == NULL || info_out_entry_ptr == NULL) {
 			/* Set the output parameters to physical addresses */
 			*lli_table_in_ptr =
 			sep_shared_area_virt_to_bus(sep, dma_in_lli_table_ptr);
Signature · 1 line
1.7.8.6
#1 lore raw

PATCH 2/2 staging/sep: Check pointers before dereferencing (fix smatch warning)

smatch complains about two dereferenced before check issues:

sep_main.c:2898 sep_free_dma_tables_and_dcb() warn: variable dereferenced before check 'dma_ctx' (see line 2885)
sep_main.c:2898 sep_free_dma_tables_and_dcb() warn: variable dereferenced before check '*dma_ctx' (see line 2885)

-> Move the checks to the top, but keep the semantics.

---
 drivers/staging/sep/sep_main.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
drivers/staging/sep/sep_main.c +3 -2
index 366d56b..f5b7341 100644
@@ -2880,6 +2880,8 @@ static int sep_free_dma_tables_and_dcb(struct sep_device *sep, bool isapplet,
 
 	dev_dbg(&sep->pdev->dev, "[PID%d] sep_free_dma_tables_and_dcb\n",
 					current->pid);
+	if (!dma_ctx || !*dma_ctx) /* nothing to be done here*/
+		return 0;
 
 	if (((*dma_ctx)->secure_dma == false) && (isapplet == true)) {
 		dev_dbg(&sep->pdev->dev, "[PID%d] handling applet\n",
@@ -2895,8 +2897,7 @@ static int sep_free_dma_tables_and_dcb(struct sep_device *sep, bool isapplet,
 		 * Go over each DCB and see if
 		 * tail pointer must be updated
 		 */
-		for (i = 0; dma_ctx && *dma_ctx &&
-			i < (*dma_ctx)->nr_dcb_creat; i++, dcb_table_ptr++) {
+		for (i = 0; i < (*dma_ctx)->nr_dcb_creat; i++, dcb_table_ptr++) {
 			if (dcb_table_ptr->out_vr_tail_pt) {
 				pt_hold = (unsigned long)dcb_table_ptr->
 					out_vr_tail_pt;
Signature · 1 line
1.7.8.6
#2 lore raw
In reply to Peter Huewe
Quoted 12 lines · Peter Huewe

Smatch complains about a potential NULL pointer dereference:

sep_main.c:2312 sep_construct_dma_tables_from_lli() error: potential NULL dereference 'info_out_entry_ptr'.

info_out_entry_ptr is initialized with NULL and if info_in_entry_ptr is not NULL it gets derefenced.
However info_out_entry_ptr is only NULL in the first iteration of the while loop and in this case info_in_entry_ptr is also NULL (as indicated by the comment /* If info entry is null - this is the first table built */ -> this is a false positive.

Quoted 3 lines

Nevertheless we add a check for info_out_entry_ptr to silence this warning and make it more robust in regard to code changes.

Smatch doesn't handle loops very well. Of course, all along I've wanted to fix this, but it's a bit complicated so it could be another year or two before it actually happens.

Generally, as a philosophy, I always say never to change the code for false positives. It should be Smatch which changes.

Also the other thing is that with Smatch I deliberately allow more false positives than GCC does. It's a trade off between being ambitious in looking for bugs and being annoying to users.

When Smatch looks at this code it sees the else side as impossible to reach. Perhaps I should add a hack in that if the code is in an impossible to reach place then don't print a warning... It would be better to just fix loop handling... I'm not sure.

regards,
dan carpenter