slab: fix barn NULL pointer dereference on memoryless nodes

Phil reported a boot failure once sheaves become used in commits
59faa4da7cd4 ("maple_tree: use percpu sheaves for maple_node_cache") and
3accabda4da1 ("mm, vma: use percpu sheaves for vm_area_struct cache"):

 BUG: kernel NULL pointer dereference, address: 0000000000000040
 #PF: supervisor read access in kernel mode
 #PF: error_code(0x0000) - not-present page
 PGD 0 P4D 0
 Oops: Oops: 0000 [#1] SMP NOPTI
 CPU: 21 UID: 0 PID: 818 Comm: kworker/u398:0 Not tainted 6.17.0-rc3.slab+ #5 PREEMPT(voluntary)
 Hardware name: Dell Inc. PowerEdge R7425/02MJ3T, BIOS 1.26.0 07/30/2025
 RIP: 0010:__pcs_replace_empty_main+0x44/0x1d0
 Code: ec 08 48 8b 46 10 48 8b 76 08 48 85 c0 74 0b 8b 48 18 85 c9 0f 85 e5 00 00 00 65 48 63 05 e4 ee 50 02 49 8b 84 c6 e0 00 00 00 <4c> 8b 68 40 4c 89 ef e8 b0 81 ff ff 48 89 c5 48 85 c0 74 1d 48 89
 RSP: 0018:ffffd2d10950bdb0 EFLAGS: 00010246
 RAX: 0000000000000000 RBX: ffff8a775dab74b0 RCX: 00000000ffffffff
 RDX: 0000000000000cc0 RSI: ffff8a6800804000 RDI: ffff8a680004e300
 RBP: ffffd2d10950be40 R08: 0000000000000060 R09: ffffffffb9367388
 R10: 00000000000149e8 R11: ffff8a6f87a38000 R12: 0000000000000cc0
 R13: 0000000000000cc0 R14: ffff8a680004e300 R15: 00000000000000c0
 FS:  0000000000000000(0000) GS:ffff8a77a3541000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000040 CR3: 0000000e1aa24000 CR4: 00000000003506f0
 Call Trace:
  <TASK>
  ? srso_return_thunk+0x5/0x5f
  ? vm_area_alloc+0x1e/0x60
  kmem_cache_alloc_noprof+0x4ec/0x5b0
  vm_area_alloc+0x1e/0x60
  create_init_stack_vma+0x26/0x210
  alloc_bprm+0x139/0x200
  kernel_execve+0x4a/0x140
  call_usermodehelper_exec_async+0xd0/0x190
  ? __pfx_call_usermodehelper_exec_async+0x10/0x10
  ret_from_fork+0xf0/0x110
  ? __pfx_call_usermodehelper_exec_async+0x10/0x10
  ret_from_fork_asm+0x1a/0x30
  </TASK>
 Modules linked in:
 CR2: 0000000000000040
 ---[ end trace 0000000000000000 ]---
 RIP: 0010:__pcs_replace_empty_main+0x44/0x1d0
 Code: ec 08 48 8b 46 10 48 8b 76 08 48 85 c0 74 0b 8b 48 18 85 c9 0f 85 e5 00 00 00 65 48 63 05 e4 ee 50 02 49 8b 84 c6 e0 00 00 00 <4c> 8b 68 40 4c 89 ef e8 b0 81 ff ff 48 89 c5 48 85 c0 74 1d 48 89
 RSP: 0018:ffffd2d10950bdb0 EFLAGS: 00010246
 RAX: 0000000000000000 RBX: ffff8a775dab74b0 RCX: 00000000ffffffff
 RDX: 0000000000000cc0 RSI: ffff8a6800804000 RDI: ffff8a680004e300
 RBP: ffffd2d10950be40 R08: 0000000000000060 R09: ffffffffb9367388
 R10: 00000000000149e8 R11: ffff8a6f87a38000 R12: 0000000000000cc0
 R13: 0000000000000cc0 R14: ffff8a680004e300 R15: 00000000000000c0
 FS:  0000000000000000(0000) GS:ffff8a77a3541000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000040 CR3: 0000000e1aa24000 CR4: 00000000003506f0
 Kernel panic - not syncing: Fatal exception
 Kernel Offset: 0x36a00000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
 ---[ end Kernel panic - not syncing: Fatal exception ]---

And noted "this is an AMD EPYC 7401 with 8 NUMA nodes configured such
that memory is only on 2 of them."

 # numactl --hardware
 available: 8 nodes (0-7)
 node 0 cpus: 0 8 16 24 32 40 48 56 64 72 80 88
 node 0 size: 0 MB
 node 0 free: 0 MB
 node 1 cpus: 2 10 18 26 34 42 50 58 66 74 82 90
 node 1 size: 31584 MB
 node 1 free: 30397 MB
 node 2 cpus: 4 12 20 28 36 44 52 60 68 76 84 92
 node 2 size: 0 MB
 node 2 free: 0 MB
 node 3 cpus: 6 14 22 30 38 46 54 62 70 78 86 94
 node 3 size: 0 MB
 node 3 free: 0 MB
 node 4 cpus: 1 9 17 25 33 41 49 57 65 73 81 89
 node 4 size: 0 MB
 node 4 free: 0 MB
 node 5 cpus: 3 11 19 27 35 43 51 59 67 75 83 91
 node 5 size: 32214 MB
 node 5 free: 31625 MB
 node 6 cpus: 5 13 21 29 37 45 53 61 69 77 85 93
 node 6 size: 0 MB
 node 6 free: 0 MB
 node 7 cpus: 7 15 23 31 39 47 55 63 71 79 87 95
 node 7 size: 0 MB
 node 7 free: 0 MB

Linus decoded the stacktrace to get_barn() and get_node() and determined
that kmem_cache->node[numa_mem_id()] is NULL.

The problem is due to a wrong assumption that memoryless nodes only
exist on systems with CONFIG_HAVE_MEMORYLESS_NODES, where numa_mem_id()
points to the nearest node that has memory. SLUB has been allocating its
kmem_cache_node structures only on nodes with memory and so it does with
struct node_barn.

For kmem_cache_node, get_partial_node() checks if get_node() result is
not NULL, which I assumed was for protection from a bogus node id passed
to kmalloc_node() but apparently it's also for systems where
numa_mem_id() (used when no specific node is given) might return a
memoryless node.

Fix the sheaves code the same way by checking the result of get_node()
and bailing out if it's NULL. Note that cpus on such memoryless nodes
will have degraded sheaves performance, which can be improved later,
preferably by making numa_mem_id() work properly on such systems.

Fixes: 2d517aa09bbc ("slab: add opt-in caching layer of percpu sheaves")
Reported-and-tested-by: Phil Auld <pauld@redhat.com>
Closes: https://lore.kernel.org/all/20251010151116.GA436967@pauld.westford.csb/
Analyzed-by: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://lore.kernel.org/all/CAHk-%3Dwg1xK%2BBr%3DFJ5QipVhzCvq7uQVPt5Prze6HDhQQ%3DQD_BcQ@mail.gmail.com/
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>

diff --git a/mm/slub.c b/mm/slub.c
index 135c408e051528389fe4b5952bcdd84f455956d4..b1f15598fbfd2320df7d7f7a306c6a2c311d378e 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -504,10 +504,18 @@ static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
        return s->node[node];
 }
 
-/* Get the barn of the current cpu's memory node */
+/*
+ * Get the barn of the current cpu's closest memory node. It may not exist on
+ * systems with memoryless nodes but without CONFIG_HAVE_MEMORYLESS_NODES
+ */
 static inline struct node_barn *get_barn(struct kmem_cache *s)
 {
-       return get_node(s, numa_mem_id())->barn;
+       struct kmem_cache_node *n = get_node(s, numa_mem_id());
+
+       if (!n)
+               return NULL;
+
+       return n->barn;
 }
 
 /*
@@ -4982,6 +4990,10 @@ __pcs_replace_empty_main(struct kmem_cache *s, struct slub_percpu_sheaves *pcs,
        }
 
        barn = get_barn(s);
+       if (!barn) {
+               local_unlock(&s->cpu_sheaves->lock);
+               return NULL;
+       }
 
        full = barn_replace_empty_sheaf(barn, pcs->main);
 
@@ -5153,13 +5165,20 @@ next_batch:
        if (unlikely(pcs->main->size == 0)) {
 
                struct slab_sheaf *full;
+               struct node_barn *barn;
 
                if (pcs->spare && pcs->spare->size > 0) {
                        swap(pcs->main, pcs->spare);
                        goto do_alloc;
                }
 
-               full = barn_replace_empty_sheaf(get_barn(s), pcs->main);
+               barn = get_barn(s);
+               if (!barn) {
+                       local_unlock(&s->cpu_sheaves->lock);
+                       return allocated;
+               }
+
+               full = barn_replace_empty_sheaf(barn, pcs->main);
 
                if (full) {
                        stat(s, BARN_GET);
@@ -5314,6 +5333,7 @@ kmem_cache_prefill_sheaf(struct kmem_cache *s, gfp_t gfp, unsigned int size)
 {
        struct slub_percpu_sheaves *pcs;
        struct slab_sheaf *sheaf = NULL;
+       struct node_barn *barn;
 
        if (unlikely(size > s->sheaf_capacity)) {
 
@@ -5355,8 +5375,11 @@ kmem_cache_prefill_sheaf(struct kmem_cache *s, gfp_t gfp, unsigned int size)
                pcs->spare = NULL;
                stat(s, SHEAF_PREFILL_FAST);
        } else {
+               barn = get_barn(s);
+
                stat(s, SHEAF_PREFILL_SLOW);
-               sheaf = barn_get_full_or_empty_sheaf(get_barn(s));
+               if (barn)
+                       sheaf = barn_get_full_or_empty_sheaf(barn);
                if (sheaf && sheaf->size)
                        stat(s, BARN_GET);
                else
@@ -5426,7 +5449,7 @@ void kmem_cache_return_sheaf(struct kmem_cache *s, gfp_t gfp,
         * If the barn has too many full sheaves or we fail to refill the sheaf,
         * simply flush and free it.
         */
-       if (data_race(barn->nr_full) >= MAX_FULL_SHEAVES ||
+       if (!barn || data_race(barn->nr_full) >= MAX_FULL_SHEAVES ||
            refill_sheaf(s, sheaf, gfp)) {
                sheaf_flush_unused(s, sheaf);
                free_empty_sheaf(s, sheaf);
@@ -5943,10 +5966,9 @@ slab_empty:
  * put the full sheaf there.
  */
 static void __pcs_install_empty_sheaf(struct kmem_cache *s,
-               struct slub_percpu_sheaves *pcs, struct slab_sheaf *empty)
+               struct slub_percpu_sheaves *pcs, struct slab_sheaf *empty,
+               struct node_barn *barn)
 {
-       struct node_barn *barn;
-
        lockdep_assert_held(this_cpu_ptr(&s->cpu_sheaves->lock));
 
        /* This is what we expect to find if nobody interrupted us. */
@@ -5956,8 +5978,6 @@ static void __pcs_install_empty_sheaf(struct kmem_cache *s,
                return;
        }
 
-       barn = get_barn(s);
-
        /*
         * Unlikely because if the main sheaf had space, we would have just
         * freed to it. Get rid of our empty sheaf.
@@ -6002,6 +6022,11 @@ restart:
        lockdep_assert_held(this_cpu_ptr(&s->cpu_sheaves->lock));
 
        barn = get_barn(s);
+       if (!barn) {
+               local_unlock(&s->cpu_sheaves->lock);
+               return NULL;
+       }
+
        put_fail = false;
 
        if (!pcs->spare) {
@@ -6084,7 +6109,7 @@ got_empty:
        }
 
        pcs = this_cpu_ptr(s->cpu_sheaves);
-       __pcs_install_empty_sheaf(s, pcs, empty);
+       __pcs_install_empty_sheaf(s, pcs, empty, barn);
 
        return pcs;
 }
@@ -6121,8 +6146,9 @@ bool free_to_pcs(struct kmem_cache *s, void *object)
 
 static void rcu_free_sheaf(struct rcu_head *head)
 {
+       struct kmem_cache_node *n;
        struct slab_sheaf *sheaf;
-       struct node_barn *barn;
+       struct node_barn *barn = NULL;
        struct kmem_cache *s;
 
        sheaf = container_of(head, struct slab_sheaf, rcu_head);
@@ -6139,7 +6165,11 @@ static void rcu_free_sheaf(struct rcu_head *head)
         */
        __rcu_free_sheaf_prepare(s, sheaf);
 
-       barn = get_node(s, sheaf->node)->barn;
+       n = get_node(s, sheaf->node);
+       if (!n)
+               goto flush;
+
+       barn = n->barn;
 
        /* due to slab_free_hook() */
        if (unlikely(sheaf->size == 0))
@@ -6157,11 +6187,12 @@ static void rcu_free_sheaf(struct rcu_head *head)
                return;
        }
 
+flush:
        stat(s, BARN_PUT_FAIL);
        sheaf_flush_unused(s, sheaf);
 
 empty:
-       if (data_race(barn->nr_empty) < MAX_EMPTY_SHEAVES) {
+       if (barn && data_race(barn->nr_empty) < MAX_EMPTY_SHEAVES) {
                barn_put_empty_sheaf(barn, sheaf);
                return;
        }
@@ -6191,6 +6222,10 @@ bool __kfree_rcu_sheaf(struct kmem_cache *s, void *obj)
                }
 
                barn = get_barn(s);
+               if (!barn) {
+                       local_unlock(&s->cpu_sheaves->lock);
+                       goto fail;
+               }
 
                empty = barn_get_empty_sheaf(barn);
 
@@ -6304,6 +6339,8 @@ next_batch:
                goto do_free;
 
        barn = get_barn(s);
+       if (!barn)
+               goto no_empty;
 
        if (!pcs->spare) {
                empty = barn_get_empty_sheaf(barn);