Received-SPF: pass (google.com: domain of linux-kernel+bounces-42531-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
From: Zhiguo Niu <zhiguo.niu@unisoc.com>
To: <jaegeuk@kernel.org>, <chao@kernel.org>
CC: <linux-f2fs-devel@lists.sourceforge.net>, <linux-kernel@vger.kernel.org>,
        <niuzhiguo84@gmail.com>, <zhiguo.niu@unisoc.com>, <ke.wang@unisoc.com>,
        <hongyu.jin@unisoc.com>
Subject: [PATCH 1/3] f2fs: correct counting methods of free_segments in __set_inuse
Date: Mon, 29 Jan 2024 18:21:49 +0800
Message-ID: <1706523709-24605-1-git-send-email-zhiguo.niu@unisoc.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain

There is a corner scenario on a small-capacity partition with 64MB size:
1. The main area has a total of 24 segments, and there are no free
segments left shown from the free_segmap bitmap and free_secmap in
free_segmap_info.
---------------------------------------------------------------------
bitmap value: ffffffffffffffff
---------------------------------------------------------------------
2. When doing gc, an out-of-bounds segment with segno=24 is allocated.
Because CONFIG_F2FS_CHECK_FS is not enabled, f2fs_bug_on in get_new_segment
just print warning log but the subsequent process continues to run.
---------------------------------------------------------------------
got_it:
    /* set it as dirty segment in free segmap */
    f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
    __set_inuse(sbi, segno);
----------------------------------------------------------------------
3. __set_inuse directly sets free_i->free_segments--,
As a result, free_i->free_segments=-1, as shown in the following
coredump information:
----------------------------------------------------------------------
  crash_arm64> struct free_segmap_info 0xffffff8084d9a000 -x
  struct free_segmap_info {
  start_segno = 0x7,
  free_segments = 0xffffffff,
  free_sections = 0x0,
----------------------------------------------------------------------
This is unreasonable and will cause free_segments and free_sections
counts mismatch if there are segments released as free.

So same counting methods like free_sections should be used to
free_segments.

Signed-off-by: Zhiguo Niu <zhiguo.niu@unisoc.com>
---
 fs/f2fs/segment.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h
index 8129be7..f2847f1 100644
--- a/fs/f2fs/segment.h
+++ b/fs/f2fs/segment.h
@@ -463,8 +463,8 @@ static inline void __set_inuse(struct f2fs_sb_info *sbi,
 	struct free_segmap_info *free_i = FREE_I(sbi);
 	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
 
-	set_bit(segno, free_i->free_segmap);
-	free_i->free_segments--;
+	if (!test_and_set_bit(segno, free_i->free_segmap))
+		free_i->free_segments--;
 	if (!test_and_set_bit(secno, free_i->free_secmap))
 		free_i->free_sections--;
 }
-- 
1.9.1