Re: [PATCH] codetag: avoid race at alloc_slab_obj_exts

[Vlastimil Babka]

On 5/27/24 8:30 PM, Thadeu Lima de Souza Cascardo wrote:
> When CONFIG_MEM_ALLOC_PROFILING_DEBUG is enabled, the following warning may
> be noticed:
> 
> [   48.299584] ------------[ cut here ]------------
> [   48.300092] alloc_tag was not set
> [   48.300528] WARNING: CPU: 2 PID: 1361 at include/linux/alloc_tag.h:130 alloc_tagging_slab_free_hook+0x84/0xc7
> [   48.301305] Modules linked in:
> [   48.301553] CPU: 2 PID: 1361 Comm: systemd-udevd Not tainted 6.10.0-rc1-00003-gac8755535862 #176
> [   48.302196] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
> [   48.302752] RIP: 0010:alloc_tagging_slab_free_hook+0x84/0xc7
> [   48.303169] Code: 8d 1c c4 48 85 db 74 4d 48 83 3b 00 75 1e 80 3d 65 02 86 04 00 75 15 48 c7 c7 11 48 1d 85 c6 05 55 02 86 04 01 e8 64 44 a5 ff <0f> 0b 48 8b 03 48 85 c0 74 21 48 83 f8 01 74 14 48 8b 50 20 48 f7
> [   48.304411] RSP: 0018:ffff8880111b7d40 EFLAGS: 00010282
> [   48.304916] RAX: 0000000000000000 RBX: ffff88800fcc9008 RCX: 0000000000000000
> [   48.305455] RDX: 0000000080000000 RSI: ffff888014060000 RDI: ffffed1002236f97
> [   48.305979] RBP: 0000000000001100 R08: fffffbfff0aa73a1 R09: 0000000000000000
> [   48.306473] R10: ffffffff814515e5 R11: 0000000000000003 R12: ffff88800fcc9000
> [   48.306943] R13: ffff88800b2e5cc0 R14: ffff8880111b7d90 R15: 0000000000000000
> [   48.307529] FS:  00007faf5d1908c0(0000) GS:ffff88806cf00000(0000) knlGS:0000000000000000
> [   48.308223] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [   48.308710] CR2: 000058fb220c9118 CR3: 00000000110cc000 CR4: 0000000000750ef0
> [   48.309274] PKRU: 55555554
> [   48.309804] Call Trace:
> [   48.310029]  <TASK>
> [   48.310290]  ? show_regs+0x84/0x8d
> [   48.310722]  ? alloc_tagging_slab_free_hook+0x84/0xc7
> [   48.311298]  ? __warn+0x13b/0x2ff
> [   48.311580]  ? alloc_tagging_slab_free_hook+0x84/0xc7
> [   48.311987]  ? report_bug+0x2ce/0x3ab
> [   48.312292]  ? handle_bug+0x8c/0x107
> [   48.312563]  ? exc_invalid_op+0x34/0x6f
> [   48.312842]  ? asm_exc_invalid_op+0x1a/0x20
> [   48.313173]  ? this_cpu_in_panic+0x1c/0x72
> [   48.313503]  ? alloc_tagging_slab_free_hook+0x84/0xc7
> [   48.313880]  ? putname+0x143/0x14e
> [   48.314152]  kmem_cache_free+0xe9/0x214
> [   48.314454]  putname+0x143/0x14e
> [   48.314712]  do_unlinkat+0x413/0x45e
> [   48.315001]  ? __pfx_do_unlinkat+0x10/0x10
> [   48.315388]  ? __check_object_size+0x4d7/0x525
> [   48.315744]  ? __sanitizer_cov_trace_pc+0x20/0x4a
> [   48.316167]  ? __sanitizer_cov_trace_pc+0x20/0x4a
> [   48.316757]  ? getname_flags+0x4ed/0x500
> [   48.317261]  __x64_sys_unlink+0x42/0x4a
> [   48.317741]  do_syscall_64+0xe2/0x149
> [   48.318171]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
> [   48.318602] RIP: 0033:0x7faf5d8850ab
> [   48.318891] Code: fd ff ff e8 27 dd 01 00 0f 1f 80 00 00 00 00 f3 0f 1e fa b8 5f 00 00 00 0f 05 c3 0f 1f 40 00 f3 0f 1e fa b8 57 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 05 c3 0f 1f 40 00 48 8b 15 41 2d 0e 00 f7 d8
> [   48.320649] RSP: 002b:00007ffc44982b38 EFLAGS: 00000246 ORIG_RAX: 0000000000000057
> [   48.321182] RAX: ffffffffffffffda RBX: 00005ba344a44680 RCX: 00007faf5d8850ab
> [   48.321667] RDX: 0000000000000000 RSI: 00005ba344a44430 RDI: 00007ffc44982b40
> [   48.322139] RBP: 00007ffc44982c00 R08: 0000000000000000 R09: 0000000000000007
> [   48.322598] R10: 00005ba344a44430 R11: 0000000000000246 R12: 0000000000000000
> [   48.323071] R13: 00007ffc44982b40 R14: 0000000000000000 R15: 0000000000000000
> [   48.323596]  </TASK>
> 
> This is due to a race when two objects are allocated from the same slab,
> which did not have an obj_exts allocated for.
> 
> In such a case, the two threads will notice the NULL obj_exts and after one
> assigns slab->obj_exts, the second one will happily do the exchange if it
> reads this new assigned value.
> 
> In order to avoid that, verify that the read obj_exts does not point to an
> allocated obj_exts before doing the exchange.
> 
> Fixes: 09c46563ff6d ("codetag: debug: introduce OBJEXTS_ALLOC_FAIL to mark failed slab_ext allocations")
> Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@xxxxxxxxxx>
> Cc: Suren Baghdasaryan <surenb@xxxxxxxxxx>

Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
Thanks!

> ---
>  mm/slub.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 0809760cf789..1373ac365a46 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -1952,7 +1952,7 @@ int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
>  #ifdef CONFIG_MEMCG
>  	new_exts |= MEMCG_DATA_OBJEXTS;
>  #endif
> -	old_exts = slab->obj_exts;
> +	old_exts = READ_ONCE(slab->obj_exts);
>  	handle_failed_objexts_alloc(old_exts, vec, objects);
>  	if (new_slab) {
>  		/*
> @@ -1961,7 +1961,8 @@ int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
>  		 * be simply assigned.
>  		 */
>  		slab->obj_exts = new_exts;
> -	} else if (cmpxchg(&slab->obj_exts, old_exts, new_exts) != old_exts) {
> +	} else if ((old_exts & ~OBJEXTS_FLAGS_MASK) ||
> +		   cmpxchg(&slab->obj_exts, old_exts, new_exts) != old_exts) {
>  		/*
>  		 * If the slab is already in use, somebody can allocate and
>  		 * assign slabobj_exts in parallel. In this case the existing