Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Sungjong Seo <sj1557.seo@samsung.com>
To:     linkinjeon@kernel.org
Cc:     linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        Sungjong Seo <sj1557.seo@samsung.com>, stable@vger.kernel.org,
        syzbot+1741a5d9b79989c10bdc@syzkaller.appspotmail.com
Subject: [PATCH] exfat: release s_lock before calling dir_emit()
Date:   Fri, 14 Jul 2023 17:43:54 +0900
Message-Id: <20230714084354.1959951-1-sj1557.seo@samsung.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Type: text/plain; charset="utf-8"
CMS-TYPE: 101P
DLP-Filter: Pass
References: <CGME20230714084427epcas1p2ce3efb1524c8efae6038d1940149ae54@epcas1p2.samsung.com>
Precedence: bulk

There is a potential deadlock reported by syzbot as below:

======================================================
WARNING: possible circular locking dependency detected
6.4.0-next-20230707-syzkaller #0 Not tainted
------------------------------------------------------
syz-executor330/5073 is trying to acquire lock:
ffff8880218527a0 (&mm->mmap_lock){++++}-{3:3}, at: mmap_read_lock_killable include/linux/mmap_lock.h:151 [inline]
ffff8880218527a0 (&mm->mmap_lock){++++}-{3:3}, at: get_mmap_lock_carefully mm/memory.c:5293 [inline]
ffff8880218527a0 (&mm->mmap_lock){++++}-{3:3}, at: lock_mm_and_find_vma+0x369/0x510 mm/memory.c:5344
but task is already holding lock:
ffff888019f760e0 (&sbi->s_lock){+.+.}-{3:3}, at: exfat_iterate+0x117/0xb50 fs/exfat/dir.c:232

which lock already depends on the new lock.

Chain exists of:
  &mm->mmap_lock --> mapping.invalidate_lock#3 --> &sbi->s_lock

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&sbi->s_lock);
                               lock(mapping.invalidate_lock#3);
                               lock(&sbi->s_lock);
  rlock(&mm->mmap_lock);

Let's try to avoid above potential deadlock condition by moving dir_emit*()
out of sbi->s_lock coverage.

Fixes: ca06197382bd ("exfat: add directory operations")
Cc: stable@vger.kernel.org #v5.7+
Reported-by: syzbot+1741a5d9b79989c10bdc@syzkaller.appspotmail.com
Link: https://lore.kernel.org/lkml/00000000000078ee7e060066270b@google.com/T/#u
Signed-off-by: Sungjong Seo <sj1557.seo@samsung.com>
---
 fs/exfat/dir.c | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/fs/exfat/dir.c b/fs/exfat/dir.c
index 957574180a5e..4e3743341ce7 100644
--- a/fs/exfat/dir.c
+++ b/fs/exfat/dir.c
@@ -214,7 +214,10 @@ static void exfat_free_namebuf(struct exfat_dentry_namebuf *nb)
 	exfat_init_namebuf(nb);
 }
 
-/* skip iterating emit_dots when dir is empty */
+/*
+ * Before calling dir_emit*(), sbi->s_lock should be released
+ * because page fault can occur in dir_emit*().
+ */
 #define ITER_POS_FILLED_DOTS    (2)
 static int exfat_iterate(struct file *file, struct dir_context *ctx)
 {
@@ -229,11 +232,10 @@ static int exfat_iterate(struct file *file, struct dir_context *ctx)
 	int err = 0, fake_offset = 0;
 
 	exfat_init_namebuf(nb);
-	mutex_lock(&EXFAT_SB(sb)->s_lock);
 
 	cpos = ctx->pos;
 	if (!dir_emit_dots(file, ctx))
-		goto unlock;
+		goto out;
 
 	if (ctx->pos == ITER_POS_FILLED_DOTS) {
 		cpos = 0;
@@ -245,16 +247,18 @@ static int exfat_iterate(struct file *file, struct dir_context *ctx)
 	/* name buffer should be allocated before use */
 	err = exfat_alloc_namebuf(nb);
 	if (err)
-		goto unlock;
+		goto out;
 get_new:
+	mutex_lock(&EXFAT_SB(sb)->s_lock);
+
 	if (ei->flags == ALLOC_NO_FAT_CHAIN && cpos >= i_size_read(inode))
 		goto end_of_dir;
 
 	err = exfat_readdir(inode, &cpos, &de);
 	if (err) {
 		/*
-		 * At least we tried to read a sector.  Move cpos to next sector
-		 * position (should be aligned).
+		 * At least we tried to read a sector.
+		 * Move cpos to next sector position (should be aligned).
 		 */
 		if (err == -EIO) {
 			cpos += 1 << (sb->s_blocksize_bits);
@@ -277,16 +281,10 @@ static int exfat_iterate(struct file *file, struct dir_context *ctx)
 		inum = iunique(sb, EXFAT_ROOT_INO);
 	}
 
-	/*
-	 * Before calling dir_emit(), sb_lock should be released.
-	 * Because page fault can occur in dir_emit() when the size
-	 * of buffer given from user is larger than one page size.
-	 */
 	mutex_unlock(&EXFAT_SB(sb)->s_lock);
 	if (!dir_emit(ctx, nb->lfn, strlen(nb->lfn), inum,
 			(de.attr & ATTR_SUBDIR) ? DT_DIR : DT_REG))
-		goto out_unlocked;
-	mutex_lock(&EXFAT_SB(sb)->s_lock);
+		goto out;
 	ctx->pos = cpos;
 	goto get_new;
 
@@ -294,9 +292,8 @@ static int exfat_iterate(struct file *file, struct dir_context *ctx)
 	if (!cpos && fake_offset)
 		cpos = ITER_POS_FILLED_DOTS;
 	ctx->pos = cpos;
-unlock:
 	mutex_unlock(&EXFAT_SB(sb)->s_lock);
-out_unlocked:
+out:
 	/*
 	 * To improve performance, free namebuf after unlock sb_lock.
 	 * If namebuf is not allocated, this function do nothing
-- 
2.25.1