Skip to contentSkip to conversation

dpt_i20: several use after free issues

PATCH In review Dan Carpenter 3 participants 3 messages Mar 15, 2010

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

PATCH 1/1 dpt_i20: several use after free issues

adpt_i2o_delete_hba() calls kfree() so we have to save "pHba->next" before calling it. Also inside adpt_i2o_delete_hba() itself, there was another use after free bug which I fixed by moving the kfree() down a line.

drivers/scsi/dpt_i2o.c +9 -6
index 4967643..0435d04 100644
@@ -188,7 +188,8 @@ MODULE_DEVICE_TABLE(pci,dptids);
 static int adpt_detect(struct scsi_host_template* sht)
 {
 	struct pci_dev *pDev = NULL;
-	adpt_hba* pHba;
+	adpt_hba *pHba;
+	adpt_hba *next;
 
 	PINFO("Detecting Adaptec I2O RAID controllers...\n");
 
@@ -206,7 +207,8 @@ static int adpt_detect(struct scsi_host_template* sht)
 	}
 
 	/* In INIT state, Activate IOPs */
-	for (pHba = hba_chain; pHba; pHba = pHba->next) {
+	for (pHba = hba_chain; pHba; pHba = next) {
+		next = pHba->next;
 		// Activate does get status , init outbound, and get hrt
 		if (adpt_i2o_activate_hba(pHba) < 0) {
 			adpt_i2o_delete_hba(pHba);
@@ -243,7 +245,8 @@ rebuild_sys_tab:
 	PDEBUG("HBA's in OPERATIONAL state\n");
 
 	printk("dpti: If you have a lot of devices this could take a few minutes.\n");
-	for (pHba = hba_chain; pHba; pHba = pHba->next) {
+	for (pHba = hba_chain; pHba; pHba = next) {
+		next = pHba->next;
 		printk(KERN_INFO"%s: Reading the hardware resource table.\n", pHba->name);
 		if (adpt_i2o_lct_get(pHba) < 0){
 			adpt_i2o_delete_hba(pHba);
@@ -263,7 +266,8 @@ rebuild_sys_tab:
 		adpt_sysfs_class = NULL;
 	}
 
-	for (pHba = hba_chain; pHba; pHba = pHba->next) {
+	for (pHba = hba_chain; pHba; pHba = next) {
+		next = pHba->next;
 		if (adpt_scsi_host_alloc(pHba, sht) < 0){
 			adpt_i2o_delete_hba(pHba);
 			continue;
@@ -1229,11 +1233,10 @@ static void adpt_i2o_delete_hba(adpt_hba* pHba)
 		}
 	}
 	pci_dev_put(pHba->pDev);
-	kfree(pHba);
-
 	if (adpt_sysfs_class)
 		device_destroy(adpt_sysfs_class,
 				MKDEV(DPTI_I2O_MAJOR, pHba->unit));
+	kfree(pHba);
 
 	if(hba_count <= 0){
 		unregister_chrdev(DPTI_I2O_MAJOR, DPT_DRIVER);   
#1 lore raw
Quoted 4 lines · On Mon, 15 Mar 2010 11:26:56 +0300 Dan Carpenter <[email protected]> wrote:

adpt_i2o_delete_hba() calls kfree() so we have to save "pHba->next" before calling it. Also inside adpt_i2o_delete_hba() itself, there was another use after free bug which I fixed by moving the kfree() down a line.

erk. This code should be crashing most gruesomely. I wonder why it doesn't.

#2 lore raw
Quoted 10 lines · Andrew Morton
Quoted 4 lines · On Mon, 15 Mar 2010 11:26:56 +0300 Dan Carpenter <[email protected]> wrote:

adpt_i2o_delete_hba() calls kfree() so we have to save "pHba->next" before calling it. Also inside adpt_i2o_delete_hba() itself, there was another use after free bug which I fixed by moving the kfree() down a line.

erk. This code should be crashing most gruesomely. I wonder why it doesn't.

I'm tempted to say because we have no users, but actually, for this driver, I know we do. So I think it works because most people don't have poisoning turned on in a running system and there's no sleep between the free and the use, so no opportunity to reuse the area on the percpu hot list even in an smp system.

I'll add it to -rc fixes ... when I get the current misc tree broken apart.

James