Skip to contentSkip to conversation

Set bounds on what /proc/self/make-it-fail accepts.

PATCH In review Dave Jones 4 participants 10 messages Feb 18, 2014 – Feb 19, 2014 Proc filesystem

lore
Share & tools
lore

Dave Jones

author
#0 lore raw

PATCH 1/1 Set bounds on what /proc/self/make-it-fail accepts.

/proc/self/make-it-fail is a boolean, but accepts any number, including negative ones. Change variable to unsigned, and cap upper bound at 1.

fs/proc/base.c +5 -2
index 51507065263b..b926377c354f 100644
@@ -1207,7 +1207,7 @@ static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
 	struct task_struct *task = get_proc_task(file_inode(file));
 	char buffer[PROC_NUMBUF];
 	size_t len;
-	int make_it_fail;
+	unsigned int make_it_fail;
 
 	if (!task)
 		return -ESRCH;
@@ -1224,7 +1224,7 @@ static ssize_t proc_fault_inject_write(struct file * file,
 {
 	struct task_struct *task;
 	char buffer[PROC_NUMBUF], *end;
-	int make_it_fail;
+	unsigned int make_it_fail;
 
 	if (!capable(CAP_SYS_RESOURCE))
 		return -EPERM;
@@ -1236,6 +1236,9 @@ static ssize_t proc_fault_inject_write(struct file * file,
 	make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
 	if (*end)
 		return -EINVAL;
+	if (make_it_fail > 1)
+		return -EINVAL;
+
 	task = get_proc_task(file_inode(file));
 	if (!task)
 		return -ESRCH;
#1 lore raw
Quoted 5 lines · Dave Jones

/proc/self/make-it-fail is a boolean, but accepts any number, including negative ones. Change variable to unsigned, and cap upper bound at 1.

Hmm, this would break anything that uses anything other than one to enable it, but it looks like Documentation/fault-injection/fault-injection.txt only provides an example for when it does equal one, so it's probably an ok change. I'm just wondering why non-zero is wrong? Is this an interface that will be extended to support other modes?

Adding Akinobu to the cc.

Quoted 32 lines
fs/proc/base.c +5 -2
index 51507065263b..b926377c354f 100644
@@ -1207,7 +1207,7 @@ static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
 	struct task_struct *task = get_proc_task(file_inode(file));
 	char buffer[PROC_NUMBUF];
 	size_t len;
-	int make_it_fail;
+	unsigned int make_it_fail;
 
 	if (!task)
 		return -ESRCH;
@@ -1224,7 +1224,7 @@ static ssize_t proc_fault_inject_write(struct file * file,
 {
 	struct task_struct *task;
 	char buffer[PROC_NUMBUF], *end;
-	int make_it_fail;
+	unsigned int make_it_fail;
 
 	if (!capable(CAP_SYS_RESOURCE))
 		return -EPERM;
@@ -1236,6 +1236,9 @@ static ssize_t proc_fault_inject_write(struct file * file,
 	make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
 	if (*end)
 		return -EINVAL;
+	if (make_it_fail > 1)
+		return -EINVAL;
+
 	task = get_proc_task(file_inode(file));
 	if (!task)
 		return -ESRCH;
author
#2 lore raw
Quoted 10 lines · David Rientjes
Quoted 2 lines · Dave Jones

/proc/self/make-it-fail is a boolean, but accepts any number, including negative ones. Change variable to unsigned, and cap upper bound at 1.

Hmm, this would break anything that uses anything other than one to enable it, but it looks like Documentation/fault-injection/fault-injection.txt only provides an example for when it does equal one, so it's probably an ok change. I'm just wondering why non-zero is wrong? Is this an interface that will be extended to support other modes?

"Wrong" is perhaps too strong a word, but we only ever check it for non-zero state, so it seems at best suboptimal to allow strange configurations.

When I saw I could set it to nonsense values like -1, I figured it could use some idiot proofing. The lack of any checking at all surprised me.

Future extension of this interface seems unlikely given the boolean sounding name. (Though we've done that in the past with things like the overcommit_memory sysctl,
with pretty awful end-user results).

Dave

Reviewed-by
#3 lore raw

2014-02-19 8:27 GMT+09:00 Dave Jones <[email protected]>:

Quoted 15 lines · David Rientjes
Quoted 10 lines · Dave Jones
Quoted 3 lines

/proc/self/make-it-fail is a boolean, but accepts any number, including negative ones. Change variable to unsigned, and cap upper bound at 1.

Hmm, this would break anything that uses anything other than one to enable it, but it looks like Documentation/fault-injection/fault-injection.txt only provides an example for when it does equal one, so it's probably an ok change. I'm just wondering why non-zero is wrong? Is this an interface that will be extended to support other modes?

"Wrong" is perhaps too strong a word, but we only ever check it for non-zero state, so it seems at best suboptimal to allow strange configurations.

Quoted 6 lines

When I saw I could set it to nonsense values like -1, I figured it could use some idiot proofing. The lack of any checking at all surprised me.

Future extension of this interface seems unlikely given the boolean sounding name. (Though we've done that in the past with things like the overcommit_memory sysctl,
with pretty awful end-user results).

I don't have any plans to extend /proc/self/make-it-fail to support other than 0 or 1. So I have no objection against this change.

#4 lore raw
Quoted 5 lines · Akinobu Mita

I don't have any plans to extend /proc/self/make-it-fail to support other than 0 or 1. So I have no objection against this change.

Ok, thanks. I hoped that the simple_strtol() would have been replaced by kstrtoint() since it's the preferred interface to make a lot of this simpler and then just test for val == !!val, but the minimal change is fine as well.

#5 lore raw
In reply to Dave Jones
Quoted 28 lines · On Tue, 18 Feb 2014 17:06:06 -0500 Dave Jones <[email protected]> wrote:

/proc/self/make-it-fail is a boolean, but accepts any number, including negative ones. Change variable to unsigned, and cap upper bound at 1.

fs/proc/base.c +2 -2
index 51507065263b..b926377c354f 100644
@@ -1207,7 +1207,7 @@ static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
 	struct task_struct *task = get_proc_task(file_inode(file));
 	char buffer[PROC_NUMBUF];
 	size_t len;
-	int make_it_fail;
+	unsigned int make_it_fail;
 
 	if (!task)
 		return -ESRCH;
@@ -1224,7 +1224,7 @@ static ssize_t proc_fault_inject_write(struct file * file,
 {
 	struct task_struct *task;
 	char buffer[PROC_NUMBUF], *end;
-	int make_it_fail;
+	unsigned int make_it_fail;
 
 	if (!capable(CAP_SYS_RESOURCE))
 		return -EPERM;

fs/proc/base.c static ssize_t proc_fault_inject_write(struct file * file, · line 1244

Quoted 9 lines
fs/proc/base.c +3 -0
@@ -1236,6 +1236,9 @@ static ssize_t proc_fault_inject_write(struct file * file,
 	make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
 	if (*end)
 		return -EINVAL;
+	if (make_it_fail > 1)
+		return -EINVAL;
+
 	task = get_proc_task(file_inode(file));
 	if (!task)
 		return -ESRCH;

Switching `make_it_fail' to unsigned makes the test simpler but it does rather muck up the typing in there. task_struct.make_it_fail is still an int, we should now use simple_strtoul rather than simple_strtol, proc_fault_inject_read()'s snprintf() should now use %u, etc. None of which actually matters, but still...

Rather than address all that I'm inclined to leave `make_it_fail' as an int, turning your patch into a one-liner?

fs/proc/base.c renamed from fs/proc/base.c~fault-injection-set-bounds-on-what-proc-self-make-it-fail-accepts-fix +3 -3
@@ -1207,7 +1207,7 @@ static ssize_t proc_fault_inject_read(st
 	struct task_struct *task = get_proc_task(file_inode(file));
 	char buffer[PROC_NUMBUF];
 	size_t len;
-	unsigned int make_it_fail;
+	int make_it_fail;
 
 	if (!task)
 		return -ESRCH;
@@ -1224,7 +1224,7 @@ static ssize_t proc_fault_inject_write(s
 {
 	struct task_struct *task;
 	char buffer[PROC_NUMBUF], *end;
-	unsigned int make_it_fail;
+	int make_it_fail;
 
 	if (!capable(CAP_SYS_RESOURCE))
 		return -EPERM;
@@ -1236,7 +1236,7 @@ static ssize_t proc_fault_inject_write(s
 	make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
 	if (*end)
 		return -EINVAL;
-	if (make_it_fail > 1)
+	if (make_it_fail < 0 || make_it_fail > 1)
 		return -EINVAL;
 
 	task = get_proc_task(file_inode(file));

_

author
#6 lore raw
Quoted 5 lines · Andrew Morton

Switching `make_it_fail' to unsigned makes the test simpler but it does rather muck up the typing in there. task_struct.make_it_fail is still an int, we should now use simple_strtoul rather than simple_strtol, proc_fault_inject_read()'s snprintf() should now use %u, etc. None of which actually matters, but still...

I toyed with the idea of changing task_struct.make_it_fail to unsigned too, but only realized I missed that after I'd sent out the diff.

Quoted 2 lines

Rather than address all that I'm inclined to leave `make_it_fail' as an int, turning your patch into a one-liner?

That works for me too.

Dave

#7 lore raw
Quoted 10 lines · On Wed, 19 Feb 2014 16:55:05 -0500 Dave Jones <[email protected]> wrote:
Quoted 5 lines · Andrew Morton

Switching `make_it_fail' to unsigned makes the test simpler but it does rather muck up the typing in there. task_struct.make_it_fail is still an int, we should now use simple_strtoul rather than simple_strtol, proc_fault_inject_read()'s snprintf() should now use %u, etc. None of which actually matters, but still...

I toyed with the idea of changing task_struct.make_it_fail to unsigned too, but only realized I missed that after I'd sent out the diff.

If we're touching the task_struct we could make it a bool.

Or just a single bit(field). task_struct already has a bunch of bitfields in it (strangely, they aren't contiguous). But some locking would be needed if tasks-other-than-current can modify the bit.

#8 lore raw
Quoted 6 lines · Andrew Morton

If we're touching the task_struct we could make it a bool.

Or just a single bit(field). task_struct already has a bunch of bitfields in it (strangely, they aren't contiguous). But some locking would be needed if tasks-other-than-current can modify the bit.

Or store to any adjacent bit in the word. I think the simplest solution is to use kstrtoint() and test for val == !!val since, as we all know, {simple,strict}_strto* is deprecated per checkpatch :)

author
#9 lore raw
In reply to Andrew Morton
Quoted 7 lines · Andrew Morton
Quoted 2 lines

I toyed with the idea of changing task_struct.make_it_fail to unsigned too, but only realized I missed that after I'd sent out the diff.

If we're touching the task_struct we could make it a bool.

Or just a single bit(field). task_struct already has a bunch of bitfields in it (strangely, they aren't contiguous).

afaics, asides from brk_randomized, they're contiguous, and gcc dtrt..

        unsigned int               in_execve:1;          /*   768:31  4 */
        unsigned int               in_iowait:1;          /*   768:30  4 */
        unsigned int               no_new_privs:1;       /*   768:29  4 */
        unsigned int               sched_reset_on_fork:1; /*   768:28  4 */
        unsigned int               sched_contributes_to_load:1; /*   768:27  4 */

So we could move the COMPAT_BRK ifdef and save 4 bytes for all the people still using libc5. (Or those who are for some reason averse to heap randomization).

It's not really worth doing unless you're moving a bunch of other stuff around in task_struct though, because as it is now, that struct has a bunch of alignment padding & holes, so you're not going to save anything.

The other tricky part with reorganizing that struct is that so much of it is configurable.

Dave