Skip to contentSkip to conversation

fix potential null pointer deref in quota

PATCH In review Jesper Juhl 3 participants 5 messages Mar 18, 2006 – Mar 26, 2006 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fix potential null pointer deref in quota

The coverity checker noticed that we may pass a NULL super_block to do_quotactl() that dereferences it.
Dereferencing NULL pointers is bad medicine, better check and fail gracefully.

---

 fs/quota.c |    3 +++
 1 files changed, 3 insertions(+)
fs/quota.c +3 -0
@@ -231,6 +231,9 @@ static int do_quotactl(struct super_bloc
 {
 	int ret;
 
+	if (!sb)
+		return -ENODEV;
+
 	switch (cmd) {
 		case Q_QUOTAON: {
 			char *pathname;
#1 lore raw
Quoted 18 lines · Jesper Juhl

The coverity checker noticed that we may pass a NULL super_block to
do_quotactl() that dereferences it.
Dereferencing NULL pointers is bad medicine, better check and fail gracefully.

---

  fs/quota.c |    3 +++
  1 files changed, 3 insertions(+)

--- linux-2.6.16-rc6-orig/fs/quota.c 2006-03-12 14:19:02.000000000 +0100
+++ linux-2.6.16-rc6/fs/quota.c 2006-03-18 23:03:32.000000000 +0100
@@ -231,6 +231,9 @@ static int do_quotactl(struct super_bloc
{
int ret;

Quoted 6 lines

+ if (!sb)
+ return -ENODEV;
+
switch (cmd) {
case Q_QUOTAON: {
char *pathname;

I'd have thought that check_quotactl_valid() would be the appropriate place for this check. Jan, can you please sort out what we need to do here?

#2 lore raw
Quoted 7 lines · Andrew Morton
Quoted 6 lines · Jesper Juhl

The coverity checker noticed that we may pass a NULL super_block to
do_quotactl() that dereferences it.
Dereferencing NULL pointers is bad medicine, better check and fail gracefully.

[snip]

Quoted 4 lines

I'd have thought that check_quotactl_valid() would be the appropriate place for this check. Jan, can you please sort out what we need to do here?

You may well be right. I openly admit that this is the first time I've ever stuck my head in the quota code. I picked the location I did simply because I thought making the function resistant to being passed invalid data was the sane thing to do, but there may well be a more logical place to fix it.

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html

reviewer
#3 lore raw
In reply to Jesper Juhl

Hello,

Quoted 4 lines

The coverity checker noticed that we may pass a NULL super_block to do_quotactl() that dereferences it.
Dereferencing NULL pointers is bad medicine, better check and fail gracefully.

Umm, when do you think we can dereference NULL pointer to a super_block?
check_quotactl_valid() allows sb==NULL only in the case of Q_SYNC command. And that is handled in do_quotactl() correctly even if sb==NULL...

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SuSE CR Labs
#4 lore raw
Quoted 10 lines · Jan Kara

Hello,

Quoted 4 lines

The coverity checker noticed that we may pass a NULL super_block to do_quotactl() that dereferences it.
Dereferencing NULL pointers is bad medicine, better check and fail gracefully.

Umm, when do you think we can dereference NULL pointer to a super_block?
check_quotactl_valid() allows sb==NULL only in the case of Q_SYNC command. And that is handled in do_quotactl() correctly even if sb==NULL...

Reading the code again it looks like you are right and that this is a false positive by the coverity checker.

This is what the checker had to say :

346 asmlinkage long sys_quotactl(unsigned int cmd, const char __user
*special, qid_t id, void __user *addr)
347 {
348 uint cmds, type;

Event assign_zero: Variable "sb" assigned value 0.
Also see events:
[var_deref_model][var_deref_model][var_deref_model][var_deref_model][var_deref_model][var_deref_model][var_deref_model][var_deref_model]

349  		struct super_block *sb = NULL;
350  		struct block_device *bdev;
351  		char *tmp;
352  		int ret;
353  	
354  		cmds = cmd >> SUBCMDSHIFT;
355  		type = cmd & SUBCMDMASK;
356  	

At conditional (1): "cmds != 8388609" taking false path At conditional (2): "special != 0" taking false path

357  		if (cmds != Q_SYNC || special) {
358  			tmp = getname(special);
359  			if (IS_ERR(tmp))
360  				return PTR_ERR(tmp);
361  			bdev = lookup_bdev(tmp);
362  			putname(tmp);
363  			if (IS_ERR(bdev))
364  				return PTR_ERR(bdev);
365  			sb = get_super(bdev);
366  			bdput(bdev);
367  			if (!sb)
368  				return -ENODEV;
369  		}
370  	
371  		ret = check_quotactl_valid(sb, type, cmds, id);

At conditional (3): "ret >= 0" taking true path

372 if (ret >= 0)

Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Event var_deref_model: Variable "sb" tracked as NULL was passed to a
function that dereferences it. [model]
Also see events:
[assign_zero][var_deref_model][var_deref_model][var_deref_model][var_deref_model][var_deref_model][var_deref_model][var_deref_model]

373  			ret = do_quotactl(sb, type, cmds, id, addr);
374  		if (sb)
375  			drop_super(sb);
376  	
377  		return ret;
378  	}

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html