Re: [syzbot] [mm?] WARNING in folio_large_mapcount

[Shivank Garg <shivankg@amd.com>]

[-- Attachment #1: Type: text/plain, Size: 3638 bytes --]

On 5/19/2025 6:56 PM, David Hildenbrand wrote:
> On 17.05.25 10:21, syzbot wrote:
>> Hello,
>>
>> syzbot found the following issue on:
>>
>> HEAD commit:    627277ba7c23 Merge tag 'arm64_cbpf_mitigation_2025_05_08' ..
>> git tree:       upstream
>> console output: https://syzkaller.appspot.com/x/log.txt?x=1150f670580000
>> kernel config:  https://syzkaller.appspot.com/x/.config?x=5929ac65be9baf3c
>> dashboard link: https://syzkaller.appspot.com/bug?extid=2b99589e33edbe9475ca
>> compiler:       Debian clang version 20.1.2 (++20250402124445+58df0ef89dd6-1~exp1~20250402004600.97), Debian LLD 20.1.2
>>
>> Unfortunately, I don't have any reproducer for this issue yet.
>>
>> Downloadable assets:
>> disk image: https://storage.googleapis.com/syzbot-assets/0a42ae72fe0e/disk-627277ba.raw.xz
>> vmlinux: https://storage.googleapis.com/syzbot-assets/0be88297bb66/vmlinux-627277ba.xz
>> kernel image: https://storage.googleapis.com/syzbot-assets/31808a4b1210/bzImage-627277ba.xz
>>
>> IMPORTANT: if you fix the issue, please add the following tag to the commit:
>> Reported-by: syzbot+2b99589e33edbe9475ca@syzkaller.appspotmail.com
>>
>> ------------[ cut here ]------------
>> WARNING: CPU: 1 PID: 38 at ./include/linux/mm.h:1335 folio_large_mapcount+0xd0/0x110 include/linux/mm.h:1335
> 
> This should be
> 
> VM_WARN_ON_FOLIO(!folio_test_large(folio), folio);
> 
>> Modules linked in:
>> CPU: 1 UID: 0 PID: 38 Comm: khugepaged Not tainted 6.15.0-rc6-syzkaller-00025-g627277ba7c23 #0 PREEMPT(full)
>> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 05/07/2025
>> RIP: 0010:folio_large_mapcount+0xd0/0x110 include/linux/mm.h:1335
>> Code: 04 38 84 c0 75 29 8b 03 ff c0 5b 41 5e 41 5f e9 96 d2 2b 09 cc e8 d0 cb 99 ff 48 89 df 48 c7 c6 20 de 77 8b e8 a1 dc de ff 90 <0f> 0b 90 eb b6 89 d9 80 e1 07 80 c1 03 38 c1 7c cb 48 89 df e8 87
>> RSP: 0018:ffffc90000af77e0 EFLAGS: 00010246
>> RAX: e1fcb38c0ff8ce00 RBX: ffffea00014c8000 RCX: e1fcb38c0ff8ce00
>> RDX: 0000000000000001 RSI: ffffffff8d9226df RDI: ffff88801e2fbc00
>> RBP: ffffc90000af7b50 R08: ffff8880b8923e93 R09: 1ffff110171247d2
>> R10: dffffc0000000000 R11: ffffed10171247d3 R12: 1ffffd4000299000
>> R13: dffffc0000000000 R14: 0000000000000000 R15: dffffc0000000000
>> FS:  0000000000000000(0000) GS:ffff8881261fb000(0000) knlGS:0000000000000000
>> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> CR2: 00007ffe58f12dc0 CR3: 0000000030e04000 CR4: 00000000003526f0
>> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>> Call Trace:
>>   <TASK>
>>   folio_mapcount include/linux/mm.h:1369 [inline]
> 
> And here we come through
> 
> if (likely(!folio_test_large(folio))) {
>     ...
> }
> return folio_large_mapcount(folio);
> 
> 
> So the folio is split concurrently. And I think there is nothing stopping it from getting freed.
> 
> We do a xas_for_each() under RCU. So yes, this is racy.
> 
> In  collapse_file(), we re-validate everything.
> 
> We could
> 
> (A) Take proper pagecache locks
> 
> (B) Try grabbing a temporary folio reference
> 
> (C) Try snapshotting the folio
> 
> Probably, in this code, (B) might be cleanest for now? Handling it just like other code in mm/filemap.c.
> 

Hi,

I've implemented your suggestion (B) using folio_try_get().
Could you please review if my patch looks correct?

Tested it using existing selftests: sudo make -C tools/testing/selftests/mm run_tests

Other two instances of is_refcount_suitable() uses folio locking. Should we maintain
consistency with those?

Thanks,
Shivank

#syz test

[-- Attachment #2: 0001-mm-khugepaged-Fix-race-with-folio-splitting-in-hpage.patch --]
[-- Type: text/plain, Size: 2498 bytes --]

From d1c3427e80215fea992428c8b5caf5291725dd65 Mon Sep 17 00:00:00 2001
From: Shivank Garg <shivankg@amd.com>
Date: Mon, 19 May 2025 20:19:32 +0000
Subject: [PATCH] mm/khugepaged: Fix race with folio splitting in
 hpage_collapse_scan_file()

folio_mapcount() checks folio_test_large() before proceeding to
folio_large_mapcount(), but there exists a race window where a folio
could be split between these checks which triggered the
VM_WARN_ON_FOLIO(!folio_test_large(folio), folio) in
folio_large_mapcount().

Take a temporary folio reference in hpage_collapse_scan_file() to prevent
races with concurrent folio splitting/freeing. This prevent potential
incorrect large folio detection.

Reported-by: syzbot+2b99589e33edbe9475ca@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/6828470d.a70a0220.38f255.000c.GAE@google.com
Suggested-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Shivank Garg <shivankg@amd.com>
---
 mm/khugepaged.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index cc945c6ab3bd..ef4f95409723 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -2295,6 +2295,19 @@ static int hpage_collapse_scan_file(struct mm_struct *mm, unsigned long addr,
 			continue;
 		}
 
+		/* Take a reference to prevent any concurrent split or free. */
+		if (!folio_try_get(folio)) {
+			xas_reset(&xas);
+			continue;
+		}
+
+		/* Has the folio been freed or split? */
+		if (unlikely(folio != xas_reload(&xas))) {
+			folio_put(folio);
+			xas_reset(&xas);
+			continue;
+		}
+
 		if (folio_order(folio) == HPAGE_PMD_ORDER &&
 		    folio->index == start) {
 			/* Maybe PMD-mapped */
@@ -2305,23 +2318,27 @@ static int hpage_collapse_scan_file(struct mm_struct *mm, unsigned long addr,
 			 * it's safe to skip LRU and refcount checks before
 			 * returning.
 			 */
+			folio_put(folio);
 			break;
 		}
 
 		node = folio_nid(folio);
 		if (hpage_collapse_scan_abort(node, cc)) {
 			result = SCAN_SCAN_ABORT;
+			folio_put(folio);
 			break;
 		}
 		cc->node_load[node]++;
 
 		if (!folio_test_lru(folio)) {
 			result = SCAN_PAGE_LRU;
+			folio_put(folio);
 			break;
 		}
 
 		if (!is_refcount_suitable(folio)) {
 			result = SCAN_PAGE_COUNT;
+			folio_put(folio);
 			break;
 		}
 
@@ -2333,6 +2350,7 @@ static int hpage_collapse_scan_file(struct mm_struct *mm, unsigned long addr,
 		 */
 
 		present += folio_nr_pages(folio);
+		folio_put(folio);
 
 		if (need_resched()) {
 			xas_pause(&xas);
-- 
2.34.1