Re: [PATCH v4] mm/hugetlb_cma: Fix null nodemask dereference in hugetlb_cma_alloc_frozen_folio

On Thu, Aug 06, 2026 at 09:11:29PM -0700, Sourav Panda wrote:
> On Thu, Aug 6, 2026 at 4:07 AM Usama Arif <[email protected]> wrote:
> >
> >
> >
> > On 06/08/2026 06:33, Sourav Panda wrote:
> > > On Wed, Aug 5, 2026 at 7:47 PM Anshuman Khandual
> > > <[email protected]> wrote:
> > >>
> > >> On Wed, Aug 05, 2026 at 01:24:42PM -0700, Usama Arif wrote:
> > >>> On Sun, 26 Jul 2026 07:29:34 +0000 Sourav Panda <[email protected]> wrote:
> > >>>
> > >>>> alloc_buddy_hugetlb_folio_with_mpol() can pass a NULL nodemask to
> > >>>> alloc_fresh_hugetlb_folio() as a fallback to allocate from all
> > >>>> nodes. If order is gigantic, alloc_fresh_hugetlb_folio() propagates
> > >>>> the NULL nodemask down to hugetlb_cma_alloc_frozen_folio() via
> > >>>> alloc_gigantic_frozen_folio().
> > >>>>
> > >>>> hugetlb_cma_alloc_frozen_folio() blindly dereferences the nodemask in
> > >>>> node_isset(nid, *nodemask) and for_each_node_mask(node, *nodemask),
> > >>>> leading to a null pointer dereference kernel panic.
> > >>>>
> > >>>> Fix this by checking if nodemask is NULL in
> > >>>> hugetlb_cma_alloc_frozen_folio() and defaulting it to
> > >>>> node_states[N_MEMORY]. This allows hugetlb_cma allocations to fall
> > >>>> back to any node with memory, keeping behavior consistent with
> > >>>> alloc_contig_frozen_pages() and alloc_buddy_frozen_folio().
> > >>>>
> > >>>> >From a userspace perspective, this bug allows an unprivileged user to
> > >>>> crash the kernel (trigger a panic) by requesting a gigantic hugepage
> > >>>> allocation with MPOL_PREFERRED_MANY on a system where CMA is only
> > >>>> configured on a subset of NUMA nodes.
> > >>>>
> > >>>> This can be reproduced by booting a VM with two NUMA nodes, restricting
> > >>>> CMA to Node 1 (e.g., hugetlb_cma=1:1G default_hugepagesz=1G
> > >>>> hugepagesz=1G hugepages=0), and running a program that allocates a
> > >>>> 1GB hugepage area without reserving, restricts allocation to Node 0
> > >>>> using mbind() with MPOL_PREFERRED_MANY, and triggers a page fault:
> > >>>>
> > >>>>   void *ptr = mmap(NULL, 1UL << 30, PROT_READ | PROT_WRITE,
> > >>>>                    MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB |
> > >>>>                    MAP_HUGE_1GB | MAP_NORESERVE, -1, 0);
> > >>>>   unsigned long nodemask = 1; /* Node 0 */
> > >>>>   mbind(ptr, 1UL << 30, MPOL_PREFERRED_MANY, &nodemask,
> > >>>>         sizeof(nodemask) * 8, 0);
> > >>>>   memset(ptr, 0, 1UL << 30); /* Trigger fault */
> > >>>>
> > >>>> This results in a NULL pointer dereference:
> > >>>>
> > >>>>   BUG: kernel NULL pointer dereference, address: 0000000000000000
> > >>>>   #PF: supervisor read access in kernel mode
> > >>>>   #PF: error_code(0x0000) - not-present page
> > >>>>   Oops: Oops: 0000 [#1] SMP NOPTI
> > >>>>   RIP: 0010:hugetlb_cma_alloc_frozen_folio+0x75/0x120
> > >>>>   Call Trace:
> > >>>>    <TASK>
> > >>>>    only_alloc_fresh_hugetlb_folio.isra.0+0x2c/0x160
> > >>>>    alloc_surplus_hugetlb_folio+0x6d/0x100
> > >>>>    alloc_hugetlb_folio+0x3c5/0x660
> > >>>>    hugetlb_no_page+0x3d9/0x650
> > >>>>
> > >>>> Fixes: eb02f14c4a2b ("mm/hugetlb: allow overcommitting gigantic hugepages")
> > >>>> Cc: [email protected]
> > >>>> Signed-off-by: Sourav Panda <[email protected]>
> > >>>> ---
> > >>>> Changes in v4:
> > >>>> - Reverted the alloc_fresh_hugetlb_folio() cpuset snapshot approach from v3.
> > >>>>   As Muchun Song pointed out, snapshotting cpuset_current_mems_allowed does
> > >>>>   not prevent false-positive allocation failures without complex retry loops,
> > >>>>   and alloc_contig_frozen_pages() / alloc_buddy_frozen_folio() already handle
> > >>>>   NULL nodemasks safely internally.
> > >>>> - Handled NULL nodemask directly inside hugetlb_cma_alloc_frozen_folio()
> > >>>>   by defaulting nodemask to node_states[N_MEMORY] (Option 2), keeping
> > >>>>   HugeTLB allocators clean and consistent.
> > >>>> - v3: https://lore.kernel.org/linux-mm/[email protected]/
> > >>>> - v2: https://lore.kernel.org/linux-mm/[email protected]/
> > >>>> - v1: https://lore.kernel.org/linux-mm/[email protected]/
> > >>>>
> > >>>>  mm/hugetlb_cma.c | 5 ++++-
> > >>>>  1 file changed, 4 insertions(+), 1 deletion(-)
> > >>>>
> > >>>> diff --git a/mm/hugetlb_cma.c b/mm/hugetlb_cma.c
> > >>>> index 39344d6c78d8..5744de0ceeb7 100644
> > >>>> --- a/mm/hugetlb_cma.c
> > >>>> +++ b/mm/hugetlb_cma.c
> > >>>> @@ -34,7 +34,10 @@ struct folio *hugetlb_cma_alloc_frozen_folio(int order, gfp_t gfp_mask,
> > >>>>     if (!hugetlb_cma_size)
> > >>>>             return NULL;
> > >>>>
> > >>>> -   if (hugetlb_cma[nid])
> > >>>> +   if (!nodemask)
> > >>>> +           nodemask = &node_states[N_MEMORY];
> > >>>
> > >>> Hi Sourav,
> > >>>
> > >>> hmm what if cpuset.mems only allows allocation from node 0, and hugetlb CMA
> > >>> is only available on node 1. This will no now allow allocating a gigantic page
> > >>> on node 1 when its explicitly not allowed?
> > >>
> > >> That's fair point but should not the user be also responsible in provding a right
> > >> nodemask containing CMA memory if it prefers avoiding node_states[N_MEMORY] based
> > >> fallback mechanism in kernel ?
> > >>
> >
> > I don't think the task can be made responsible for providing a CMA-aware fallback mask
> > here.  Userspace supplies the MPOL_PREFERRED_MANY preference, but the kernel itself
> > changes the second attempt to a NULL nodemask.  That permits fallback outside the
> > preferred policy nodes; it does not permit fallback outside the task's hardwall cpuset.
> >
> > >
> > > We have precedence for three options, which makes selecting one difficult :)
> > >
> > > 1) nodemask = &cpuset_current_mems_allowed is currently being used in:
> > >        only_alloc_fresh_hugetlb_folio --> alloc_buddy_frozen_folio -->
> > > __alloc_frozen_pages_noprof --> prepare_alloc_pages.
> >
> > Here I think using &cpuset_current_mems_allowed is part of the page allocator's wider cpuset handling.
> >
> > > 2) Some places use cookies with cpuset_current_mems_allowed to prevent
> > > torn writes. But this adds complexity.
> > >        An example would be dequeue_hugetlb_folio_nodemask()
> > > 3) node_states[N_MEMORY] is sprinkled all over the kernel (simplest).
> > >
> >
> > I would prefer option 2.  A NULL mempolicy mask should fall back to
> > cpuset_current_mems_allowed, not all N_MEMORY nodes.
> >
> 
> Sounds good Usama :)
> 
> I will send a patch that replaces nodemask = &node_states[N_MEMORY];
> with the below:
> 
> + unsigned int cpuset_mems_cookie;
> +
> + do {
> + cpuset_mems_cookie = read_mems_allowed_begin();
> + local_node_mask = cpuset_current_mems_allowed;
> + } while (read_mems_allowed_retry(cpuset_mems_cookie));
> +
> + nodemask = &local_node_mask;

Although complexity increases with the above mechanism but probably
it might be better in terms of being compliant with task's hardwall
cpuset as explained by Usama earlier.

> 
> Will send it out tomorrow! Let me know if anyone has any objections.

I would say let's wait for a day or two before sending the respin.

> 
> >
> > >>>
> > >>> Would it not be better to check cpuset as well?
> > >>>
> > >>> Thanks,
> > >>> Usama
> > >>>
> > >>>> +
> > >>>> +   if (hugetlb_cma[nid] && node_isset(nid, *nodemask))
> > >>>>             page = cma_alloc_frozen_compound(hugetlb_cma[nid], order);
> > >>>>
> > >>>>     if (!page && !(gfp_mask & __GFP_THISNODE)) {
> > >>>> --
> > >>>> 2.55.0
> > >>>>
> >