5.10-stable review patch. If anyone has any objections, please let me know.
------------------
From: Chao Yu chao@kernel.org
[ Upstream commit e6494977bd4a83862118a05f57a8df40256951c0 ]
syzbot reports an UBSAN issue as below:
------------[ cut here ]------------ UBSAN: array-index-out-of-bounds in fs/f2fs/node.h:381:10 index 18446744073709550692 is out of range for type '__le32[5]' (aka 'unsigned int[5]') CPU: 0 UID: 0 PID: 5318 Comm: syz.0.0 Not tainted 6.14.0-rc3-syzkaller-00060-g6537cfb395f3 #0 Call Trace: <TASK> __dump_stack lib/dump_stack.c:94 [inline] dump_stack_lvl+0x241/0x360 lib/dump_stack.c:120 ubsan_epilogue lib/ubsan.c:231 [inline] __ubsan_handle_out_of_bounds+0x121/0x150 lib/ubsan.c:429 get_nid fs/f2fs/node.h:381 [inline] f2fs_truncate_inode_blocks+0xa5e/0xf60 fs/f2fs/node.c:1181 f2fs_do_truncate_blocks+0x782/0x1030 fs/f2fs/file.c:808 f2fs_truncate_blocks+0x10d/0x300 fs/f2fs/file.c:836 f2fs_truncate+0x417/0x720 fs/f2fs/file.c:886 f2fs_file_write_iter+0x1bdb/0x2550 fs/f2fs/file.c:5093 aio_write+0x56b/0x7c0 fs/aio.c:1633 io_submit_one+0x8a7/0x18a0 fs/aio.c:2052 __do_sys_io_submit fs/aio.c:2111 [inline] __se_sys_io_submit+0x171/0x2e0 fs/aio.c:2081 do_syscall_x64 arch/x86/entry/common.c:52 [inline] do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83 entry_SYSCALL_64_after_hwframe+0x77/0x7f RIP: 0033:0x7f238798cde9
index 18446744073709550692 (decimal, unsigned long long) = 0xfffffffffffffc64 (hexadecimal, unsigned long long) = -924 (decimal, long long)
In f2fs_truncate_inode_blocks(), UBSAN detects that get_nid() tries to access .i_nid[-924], it means both offset[0] and level should zero.
The possible case should be in f2fs_do_truncate_blocks(), we try to truncate inode size to zero, however, dn.ofs_in_node is zero and dn.node_page is not an inode page, so it fails to truncate inode page, and then pass zeroed free_from to f2fs_truncate_inode_blocks(), result in this issue.
if (dn.ofs_in_node || IS_INODE(dn.node_page)) { f2fs_truncate_data_blocks_range(&dn, count); free_from += count; }
I guess the reason why dn.node_page is not an inode page could be: there are multiple nat entries share the same node block address, once the node block address was reused, f2fs_get_node_page() may load a non-inode block.
Let's add a sanity check for such condition to avoid out-of-bounds access issue.
Reported-by: syzbot+6653f10281a1badc749e@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/66fdcdf3.050a0220.40bef.0025.GAE@google.com Signed-off-by: Chao Yu chao@kernel.org Signed-off-by: Jaegeuk Kim jaegeuk@kernel.org Signed-off-by: Sasha Levin sashal@kernel.org --- fs/f2fs/node.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c index 348ad1d6199ff..57baaba17174d 100644 --- a/fs/f2fs/node.c +++ b/fs/f2fs/node.c @@ -1047,7 +1047,14 @@ int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from) trace_f2fs_truncate_inode_blocks_enter(inode, from);
level = get_node_path(inode, from, offset, noffset); - if (level < 0) { + if (level <= 0) { + if (!level) { + level = -EFSCORRUPTED; + f2fs_err(sbi, "%s: inode ino=%lx has corrupted node block, from:%lu addrs:%u", + __func__, inode->i_ino, + from, ADDRS_PER_INODE(inode)); + set_sbi_flag(sbi, SBI_NEED_FSCK); + } trace_f2fs_truncate_inode_blocks_exit(inode, level); return level; }