[PATCH] mm/truncate: fix out-of-bounds when doing a right-aligned split

[Zhang Yi <yi.zhang@huaweicloud.com>]

From: Zhang Yi <yi.zhang@huawei.com>

When performing a right split on a folio, the split_at2 may point to a
not-present page if the offset + length equals the original folio size,
which will trigger the following error:

 BUG: unable to handle page fault for address: ffffea0006000008
 #PF: supervisor read access in kernel mode
 #PF: error_code(0x0000) - not-present page
 PGD 143ffb9067 P4D 143ffb9067 PUD 143ffb8067 PMD 0
 Oops: Oops: 0000 [#1] SMP PTI
 CPU: 0 UID: 0 PID: 502640 Comm: fsx Not tainted 6.15.0-rc3-gc6156189fc6b #889 PR
 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-2.fc40 04/01/4
 RIP: 0010:truncate_inode_partial_folio+0x208/0x620
 Code: ff 03 48 01 da e8 78 7e 13 00 48 83 05 10 b5 5a 0c 01 85 c0 0f 85 1c 02 001
 RSP: 0018:ffffc90005bafab0 EFLAGS: 00010286
 RAX: 0000000000000000 RBX: ffffea0005ffff00 RCX: 0000000000000002
 RDX: 000000000000000c RSI: 0000000000013975 RDI: ffffc90005bafa30
 RBP: ffffea0006000000 R08: 0000000000000000 R09: 00000000000009bf
 R10: 00000000000007e0 R11: 0000000000000000 R12: 0000000000001633
 R13: 0000000000000000 R14: ffffea0005ffff00 R15: fffffffffffffffe
 FS:  00007f9f9a161740(0000) GS:ffff8894971fd000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: ffffea0006000008 CR3: 000000017c2ae000 CR4: 00000000000006f0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 Call Trace:
  <TASK>
  truncate_inode_pages_range+0x226/0x720
  truncate_pagecache+0x57/0x90
  ...

Fix this issue by skipping the split if truncation aligns with the folio
size, make sure the split page number lies within the folio.

Fixes: 7460b470a131 ("mm/truncate: use folio_split() in truncate operation")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
---
 mm/truncate.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/mm/truncate.c b/mm/truncate.c
index 5d98054094d1..f2aaf99f2990 100644
--- a/mm/truncate.c
+++ b/mm/truncate.c
@@ -191,6 +191,7 @@ int truncate_inode_folio(struct address_space *mapping, struct folio *folio)
 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end)
 {
 	loff_t pos = folio_pos(folio);
+	size_t size = folio_size(folio);
 	unsigned int offset, length;
 	struct page *split_at, *split_at2;
 
@@ -198,14 +199,13 @@ bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end)
 		offset = start - pos;
 	else
 		offset = 0;
-	length = folio_size(folio);
-	if (pos + length <= (u64)end)
-		length = length - offset;
+	if (pos + size <= (u64)end)
+		length = size - offset;
 	else
 		length = end + 1 - pos - offset;
 
 	folio_wait_writeback(folio);
-	if (length == folio_size(folio)) {
+	if (length == size) {
 		truncate_inode_folio(folio->mapping, folio);
 		return true;
 	}
@@ -224,16 +224,20 @@ bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end)
 		return true;
 
 	split_at = folio_page(folio, PAGE_ALIGN_DOWN(offset) / PAGE_SIZE);
-	split_at2 = folio_page(folio,
-			PAGE_ALIGN_DOWN(offset + length) / PAGE_SIZE);
-
 	if (!try_folio_split(folio, split_at, NULL)) {
 		/*
 		 * try to split at offset + length to make sure folios within
 		 * the range can be dropped, especially to avoid memory waste
 		 * for shmem truncate
 		 */
-		struct folio *folio2 = page_folio(split_at2);
+		struct folio *folio2;
+
+		if (offset + length == size)
+			goto no_split;
+
+		split_at2 = folio_page(folio,
+				PAGE_ALIGN_DOWN(offset + length) / PAGE_SIZE);
+		folio2 = page_folio(split_at2);
 
 		if (!folio_try_get(folio2))
 			goto no_split;
-- 
2.46.1