mtd: sm_ftl: heap corruption in sm_create_sysfs_attributes()
Subsystems (2)
PATCH 1/1 mtd: sm_ftl: heap corruption in sm_create_sysfs_attributes()
We always put a NUL terminator one space past the end of the "vendor" buffer. Also I think the strnlen() limiter is off by one so I have made it smaller.
---
Static analysis stuff. Untested.
drivers/mtd/sm_ftl.c +5 -4
@@ -59,11 +59,12 @@ static struct attribute_group *sm_create_sysfs_attributes(struct sm_ftl *ftl)struct attribute_group *attr_group;struct attribute **attributes;struct sm_sysfs_attribute *vendor_attribute; +int vendor_len; +char *vendor; -int vendor_len =strnlen (ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , -SM_SMALL_PAGE -SM_CIS_VENDOR_OFFSET ); - -char *vendor =kmalloc (vendor_len,GFP_KERNEL ); + vendor_len =strnlen (ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , +SM_SMALL_PAGE -SM_CIS_VENDOR_OFFSET -1 ); + vendor =kmalloc (vendor_len +1 ,GFP_KERNEL );if (!vendor) goto error1;memcpy (vendor, ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , vendor_len);
Am 05.12.2013 12:02, schrieb Dan Carpenter:
Quoted 20 lines
We always put a NUL terminator one space past the end of the "vendor" buffer. Also I think the
strnlen()limiter is off by one so I have made it smaller.Fixes 7d17c02a01a1 ('mtd: Add new SmartMedia/xD FTL')Signed-off-by Dan Carpenter <[email protected]>---
Static analysis stuff. Untested.drivers/mtd/sm_ftl.c +2
-0@@ -59,11 +59,12 @@ static struct attribute_group *sm_create_sysfs_attributes(struct sm_ftl *ftl)struct attribute_group *attr_group;struct attribute **attributes;struct sm_sysfs_attribute *vendor_attribute; +int vendor_len; +char *vendor;
drivers/mtd/sm_ftl.c static struct attribute_group *sm_create_sysfs_attributes(struct sm_ftl *ftl) · line 70
Quoted 10 lines
drivers/mtd/sm_ftl.c +3
-4-int vendor_len =strnlen (ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , -SM_SMALL_PAGE -SM_CIS_VENDOR_OFFSET ); - -char *vendor =kmalloc (vendor_len,GFP_KERNEL ); + vendor_len =strnlen (ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , +SM_SMALL_PAGE -SM_CIS_VENDOR_OFFSET -1 ); + vendor =kmalloc (vendor_len +1 ,GFP_KERNEL );if (!vendor) goto error1;memcpy (vendor, ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , vendor_len);
Hi Dan,
is this a case for kstr(n)dup() ? like:
int maxlen=SM_SMALL_PAGE - SM_CIS_VENDOR_OFFSET-1; char *buf=ftl->cis_buffer + SM_CIS_VENDOR_OFFSET; vendor=kstrndup(buf,maxlen,GFP_KERNEL);
if (!vendor)
goto error1;
i added maxlen, buf to to help me understand the code, when the source is save kstrdup() is also an option
just my to cents,
wh
Yeah, good point. kstrndup() is cleaner. I will resend.
regards,
dan carpenter
PATCH v2 1/1 mtd: sm_ftl: heap corruption in sm_create_sysfs_attributes()
In reply to walter harmsWe always put a NUL terminator one space past the end of the "vendor" buffer. Walter Harms also pointed out that this should just use kstrndup().
---
v2: Use kstrndup().
Don't subtract 1 from the strnlen() limiter for the NUL. I think it's ok to go over?
drivers/mtd/sm_ftl.c +3 -6
@@ -59,15 +59,12 @@ static struct attribute_group *sm_create_sysfs_attributes(struct sm_ftl *ftl)struct attribute_group *attr_group;struct attribute **attributes;struct sm_sysfs_attribute *vendor_attribute; +char *vendor; -int vendor_len =strnlen (ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , -SM_SMALL_PAGE -SM_CIS_VENDOR_OFFSET ); - -char *vendor =kmalloc (vendor_len,GFP_KERNEL ); + vendor =kstrndup (ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , +SM_SMALL_PAGE -SM_CIS_VENDOR_OFFSET ,GFP_KERNEL );if (!vendor) goto error1; -memcpy (vendor, ftl->cis_buffer +SM_CIS_VENDOR_OFFSET , vendor_len); - vendor[vendor_len] =0 ;/* Initialize sysfs attributes */ vendor_attribute @@ -78,7 +75,7 @@ static struct attribute_group *sm_create_sysfs_attributes(struct sm_ftl *ftl)
sysfs_attr_init(&vendor_attribute->dev_attr.attr);
vendor_attribute->data = vendor;
- vendor_attribute->len = vendor_len;
+ vendor_attribute->len = strlen(vendor);
vendor_attribute->dev_attr.attr.name = "vendor"; vendor_attribute->dev_attr.attr.mode = S_IRUGO; vendor_attribute->dev_attr.show = sm_attr_show;
Brian Norris
AppliedQuoted 6 lines · Dan Carpenter
We always put a NUL terminator one space past the end of the "vendor" buffer. Walter Harms also pointed out that this should just use
kstrndup().Fixes 7d17c02a01a1 ('mtd: Add new SmartMedia/xD FTL')Signed-off-by Dan Carpenter <[email protected]>
Patch looks good.
Quoted 4 lines
---
v2: Usekstrndup().
Don't subtract 1 from thestrnlen()limiter for the NUL. I think it's ok to go over?
I don't think you're "going over" anything. You're just (correctly) allocating the buffer as (at most) one byte larger than (SM_SMALL_PAGE - SM_CIS_VENDOR_OFFSET), which is perfect if you're copying up to (SM_SMALL_PAGE - SM_CIS_VENDOR_OFFSET) bytes of character data.
Pushed to l2-mtd.git/next, queued for 3.15. Thanks.
Brian