Re: [PATCH v7 2/6] mm/vmalloc: fix vrealloc() grow-in-place check

[Shivam Kalra]

On 24/03/26 16:31, Alice Ryhl wrote:
> On Tue, Mar 24, 2026 at 03:30:27PM +0530, Shivam Kalra via B4 Relay wrote:
>> From: Shivam Kalra <shivamkalra98@xxxxxxxxxxx>
>>
>> Fix the grow-in-place check in vrealloc() to compare the requested size
>> against the actual physical page count (vm->nr_pages) rather than the
>> virtual area size (alloced_size, derived from get_vm_area_size()).
>>
>> The virtual reservation size (get_vm_area_size()) does not decrease when
>> pages are freed during a shrink operation. Consequently, without this fix,
>> a subsequent grow-in-place operation after a shrink would incorrectly
>> succeed and attempt to access freed pages. Correcting this check is a
>> prerequisite for the upcoming vrealloc() shrink functionality.
>>
>> Signed-off-by: Shivam Kalra <shivamkalra98@xxxxxxxxxxx>
> 
> This commit is titled 'fix', but is it actually a fix for a bug that
> exists today? Or is it just a refactor to support the follow-up
> shrinking logic?
> 
> Alice
You're right, this is not a fix for a bug that exists today. Currently
get_vm_area_size() and vm->nr_pages << PAGE_SHIFT always return the same
value because no code frees tail pages from a live vmalloc allocation.

The two expressions only diverge once the shrink patch (patch 5/6) is
applied, which reduces nr_pages while keeping vm->size unchanged.

I'll retitle this to something like:

mm/vmalloc: base vrealloc() grow-in-place check on nr_pages

and rephrase the commit message to describe it as a preparatory change
rather than a fix.

Thanks for catching this.