Re: [PATCH v7 3/6] Uprobes: Support SDT markers having reference count (semaphore)

[Oleg Nesterov <oleg@redhat.com>]

Hi Ravi,

I was going to give up and ack this series, but it seems I noticed
a bug...

On 07/31, Ravi Bangoria wrote:
>
> +static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
> +{
> +	struct delayed_uprobe *du;
> +
> +	if (delayed_uprobe_check(uprobe, mm))
> +		return 0;
> +
> +	du  = kzalloc(sizeof(*du), GFP_KERNEL);
> +	if (!du)
> +		return -ENOMEM;
> +
> +	du->uprobe = uprobe;
> +	du->mm = mm;

I am surprised I didn't notice this before...

So
	du->mm = mm;

is fine, mm can't go away, uprobe_clear_state() does delayed_uprobe_remove(NULL,mm).

But
	du->uprobe = uprobe;

doesn't look right, uprobe can go away and it can be freed, its memory can be reused.
We can't rely on remove_breakpoint(), the application can unmap the probed page/vma.
Yes we do not care about the application in this case, say, the next uprobe_mmap() can
wrongly increment the counter, we do not care although this can lead to hard-to-debug
problems. But, if nothing else, the kernel can crash if the freed memory is unmapped.
So I think put_uprobe() should do delayed_uprobe_remove(uprobe, NULL) before kfree()
and delayed_uprobe_remove() should be updated to handle the mm==NULL case.

Also. delayed_uprobe_add() should check the list and avoid duplicates. Otherwise the
trivial

	for (;;)
		munmap(mmap(uprobed_file));

will eat the memory until uprobe is unregistered.


> +static bool valid_ref_ctr_vma(struct uprobe *uprobe,
> +			      struct vm_area_struct *vma)
> +{
> +	unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
> +
> +	return uprobe->ref_ctr_offset &&
> +		vma->vm_file &&
> +		file_inode(vma->vm_file) == uprobe->inode &&
> +		vma->vm_flags & VM_WRITE &&
> +		!(vma->vm_flags & VM_SHARED) &&

		vma->vm_flags & (VM_WRITE|VM_SHARED) == VM_WRITE &&

looks a bit better to me, but I won't insist.

> +static int delayed_uprobe_install(struct vm_area_struct *vma)
> +{
> +	struct list_head *pos, *q;
> +	struct delayed_uprobe *du;
> +	unsigned long vaddr;
> +	int ret = 0, err = 0;
> +
> +	mutex_lock(&delayed_uprobe_lock);
> +	list_for_each_safe(pos, q, &delayed_uprobe_list) {
> +		du = list_entry(pos, struct delayed_uprobe, list);
> +
> +		if (!valid_ref_ctr_vma(du->uprobe, vma))
> +			continue;
> +
> +		vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
> +		ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
> +		/* Record an error and continue. */
> +		err = ret & !err ? ret : err;

I try to avoid the cosmetic nits, but I simply can't look at this line ;)

		if (ret && !err)
			err = ret;

> @@ -1072,7 +1281,14 @@ int uprobe_mmap(struct vm_area_struct *vma)
>  	struct uprobe *uprobe, *u;
>  	struct inode *inode;
>
> -	if (no_uprobe_events() || !valid_vma(vma, true))
> +	if (no_uprobe_events())
> +		return 0;
> +
> +	if (vma->vm_flags & VM_WRITE &&
> +	    test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
> +		delayed_uprobe_install(vma);

OK, so you also added the VM_WRITE check and I agree. But then I think we
should also check VM_SHARED, just like valid_ref_ctr_vma() does?

Oleg.