From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id A651617AA6 for ; Wed, 9 Aug 2023 11:41:14 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 24825C433C7; Wed, 9 Aug 2023 11:41:13 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=linuxfoundation.org; s=korg; t=1691581274; bh=sff+9mdnpdJR4zyMWjaKq+ZxK7kI4+p6ZYtrikNBBkw=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=q47RnLNdXr06xXwCYc4ehzHfGYt/3RTDIEEGh7sJUx0yp0b8B4cdtXSaydezO2nC2 ppA8VrobgYqAI+Mw0DUnLt1QtXMhcJ/hhHaAjxBnwiggrVNhsZhYuaVpuo/Gq5Lpbp jrcriiUstfLLry++Ktd0vaJEOWp0gaFwSVXbBt2Q= From: Greg Kroah-Hartman To: stable@vger.kernel.org Cc: Greg Kroah-Hartman , patches@lists.linux.dev, syzbot+1741a5d9b79989c10bdc@syzkaller.appspotmail.com, Sungjong Seo , Namjae Jeon Subject: [PATCH 5.10 167/201] exfat: release s_lock before calling dir_emit() Date: Wed, 9 Aug 2023 12:42:49 +0200 Message-ID: <20230809103649.321284501@linuxfoundation.org> X-Mailer: git-send-email 2.41.0 In-Reply-To: <20230809103643.799166053@linuxfoundation.org> References: <20230809103643.799166053@linuxfoundation.org> User-Agent: quilt/0.67 Precedence: bulk X-Mailing-List: patches@lists.linux.dev List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit From: Sungjong Seo commit ff84772fd45d486e4fc78c82e2f70ce5333543e6 upstream. 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 Tested-by: syzbot+1741a5d9b79989c10bdc@syzkaller.appspotmail.com Signed-off-by: Sungjong Seo Signed-off-by: Namjae Jeon Signed-off-by: Greg Kroah-Hartman --- fs/exfat/dir.c | 27 ++++++++++++--------------- 1 file changed, 12 insertions(+), 15 deletions(-) --- a/fs/exfat/dir.c +++ b/fs/exfat/dir.c @@ -210,7 +210,10 @@ static void exfat_free_namebuf(struct ex 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 *filp, struct dir_context *ctx) { @@ -225,11 +228,10 @@ static int exfat_iterate(struct file *fi int err = 0, fake_offset = 0; exfat_init_namebuf(nb); - mutex_lock(&EXFAT_SB(sb)->s_lock); cpos = ctx->pos; if (!dir_emit_dots(filp, ctx)) - goto unlock; + goto out; if (ctx->pos == ITER_POS_FILLED_DOTS) { cpos = 0; @@ -241,16 +243,18 @@ static int exfat_iterate(struct file *fi /* 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); @@ -273,16 +277,10 @@ get_new: 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; @@ -290,9 +288,8 @@ end_of_dir: 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