Re: [PATCH v6 4/6] mm/vmalloc: protect /proc/vmallocinfo readers with READ_ONCE()

[Uladzislau Rezki <urezki@gmail.com>]

On Sat, Mar 21, 2026 at 11:35:49PM +0530, Shivam Kalra via B4 Relay wrote:
> From: Shivam Kalra <shivamkalra98@zohomail.in>
> 
> The /proc/vmallocinfo readers, specifically show_numa_info() and
> vmalloc_info_show(), currently read v->nr_pages and the v->pages
> array without any concurrent protection.
> 
> In preparation for vrealloc() shrink support, where v->nr_pages can
> be decreased and entries in the v->pages array can be nulled out
> concurrently, these readers must be protected to prevent use-after-free
> or NULL pointer dereferences.
> 
> Update show_numa_info() to use READ_ONCE(v->nr_pages) and
> READ_ONCE(v->pages[nr]), explicitly checking for NULL before
> dereferencing the page. Similarly, update vmalloc_info_show() to
> read nr_pages safely to avoid parsing a torn or inconsistent value.
> 
> Signed-off-by: Shivam Kalra <shivamkalra98@zohomail.in>
> ---
>  mm/vmalloc.c | 17 ++++++++++++-----
>  1 file changed, 12 insertions(+), 5 deletions(-)
> 
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 64f5d1088281..7658fdc087d2 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -5204,7 +5204,7 @@ bool vmalloc_dump_obj(void *object)
>  static void show_numa_info(struct seq_file *m, struct vm_struct *v,
>  				 unsigned int *counters)
>  {
> -	unsigned int nr;
> +	unsigned int nr, nr_pages;
>  	unsigned int step = 1U << vm_area_page_order(v);
>  
>  	if (!counters)
> @@ -5212,8 +5212,13 @@ static void show_numa_info(struct seq_file *m, struct vm_struct *v,
>  
>  	memset(counters, 0, nr_node_ids * sizeof(unsigned int));
>  
> -	for (nr = 0; nr < v->nr_pages; nr += step)
> -		counters[page_to_nid(v->pages[nr])] += step;
> +	nr_pages = READ_ONCE(v->nr_pages);
> +	for (nr = 0; nr < nr_pages; nr += step) {
> +		struct page *page = READ_ONCE(v->pages[nr]);
> +
> +		if (page)
> +			counters[page_to_nid(page)] += step;
> +	}
>  	for_each_node_state(nr, N_HIGH_MEMORY)
>  		if (counters[nr])
>  			seq_printf(m, " N%u=%u", nr, counters[nr]);
> @@ -5241,6 +5246,7 @@ static int vmalloc_info_show(struct seq_file *m, void *p)
>  	struct vmap_area *va;
>  	struct vm_struct *v;
>  	unsigned int *counters;
> +	unsigned int nr_pages;
>  
>  	if (IS_ENABLED(CONFIG_NUMA))
>  		counters = kmalloc_array(nr_node_ids, sizeof(unsigned int), GFP_KERNEL);
> @@ -5270,8 +5276,9 @@ static int vmalloc_info_show(struct seq_file *m, void *p)
>  			if (v->caller)
>  				seq_printf(m, " %pS", v->caller);
>  
> -			if (v->nr_pages)
> -				seq_printf(m, " pages=%d", v->nr_pages);
> +			nr_pages = READ_ONCE(v->nr_pages);
> +			if (nr_pages)
> +				seq_printf(m, " pages=%d", nr_pages);
>  
>  			if (v->phys_addr)
>  				seq_printf(m, " phys=%pa", &v->phys_addr);
> 
It is protected by the vn->busy.lock. When you update the page counter
in the vrealloc, we should do it under the lock if i do not miss anything.

--
Uladzislau Rezki