nilfs2: do not force clear folio if buffer is referenced

commit ca76bb226bf47ff04c782cacbd299f12ddee1ec1 upstream.

Patch series "nilfs2: protect busy buffer heads from being force-cleared".

This series fixes the buffer head state inconsistency issues reported by
syzbot that occurs when the filesystem is corrupted and falls back to
read-only, and the associated buffer head use-after-free issue.

This patch (of 2):

Syzbot has reported that after nilfs2 detects filesystem corruption and
falls back to read-only, inconsistencies in the buffer state may occur.

One of the inconsistencies is that when nilfs2 calls mark_buffer_dirty()
to set a data or metadata buffer as dirty, but it detects that the buffer
is not in the uptodate state:

 WARNING: CPU: 0 PID: 6049 at fs/buffer.c:1177 mark_buffer_dirty+0x2e5/0x520
  fs/buffer.c:1177
 ...
 Call Trace:
  <TASK>
  nilfs_palloc_commit_alloc_entry+0x4b/0x160 fs/nilfs2/alloc.c:598
  nilfs_ifile_create_inode+0x1dd/0x3a0 fs/nilfs2/ifile.c:73
  nilfs_new_inode+0x254/0x830 fs/nilfs2/inode.c:344
  nilfs_mkdir+0x10d/0x340 fs/nilfs2/namei.c:218
  vfs_mkdir+0x2f9/0x4f0 fs/namei.c:4257
  do_mkdirat+0x264/0x3a0 fs/namei.c:4280
  __do_sys_mkdirat fs/namei.c:4295 [inline]
  __se_sys_mkdirat fs/namei.c:4293 [inline]
  __x64_sys_mkdirat+0x87/0xa0 fs/namei.c:4293
  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

The other is when nilfs_btree_propagate(), which propagates the dirty
state to the ancestor nodes of a b-tree that point to a dirty buffer,
detects that the origin buffer is not dirty, even though it should be:

 WARNING: CPU: 0 PID: 5245 at fs/nilfs2/btree.c:2089
  nilfs_btree_propagate+0xc79/0xdf0 fs/nilfs2/btree.c:2089
 ...
 Call Trace:
  <TASK>
  nilfs_bmap_propagate+0x75/0x120 fs/nilfs2/bmap.c:345
  nilfs_collect_file_data+0x4d/0xd0 fs/nilfs2/segment.c:587
  nilfs_segctor_apply_buffers+0x184/0x340 fs/nilfs2/segment.c:1006
  nilfs_segctor_scan_file+0x28c/0xa50 fs/nilfs2/segment.c:1045
  nilfs_segctor_collect_blocks fs/nilfs2/segment.c:1216 [inline]
  nilfs_segctor_collect fs/nilfs2/segment.c:1540 [inline]
  nilfs_segctor_do_construct+0x1c28/0x6b90 fs/nilfs2/segment.c:2115
  nilfs_segctor_construct+0x181/0x6b0 fs/nilfs2/segment.c:2479
  nilfs_segctor_thread_construct fs/nilfs2/segment.c:2587 [inline]
  nilfs_segctor_thread+0x69e/0xe80 fs/nilfs2/segment.c:2701
  kthread+0x2f0/0x390 kernel/kthread.c:389
  ret_from_fork+0x4b/0x80 arch/x86/kernel/process.c:147
  ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:244
  </TASK>

Both of these issues are caused by the callbacks that handle the
page/folio write requests, forcibly clear various states, including the
working state of the buffers they hold, at unexpected times when they
detect read-only fallback.

Fix these issues by checking if the buffer is referenced before clearing
the page/folio state, and skipping the clear if it is.

[[email protected]: adjusted for page/folio conversion]
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Ryusuke Konishi <[email protected]>
Reported-by: [email protected]
Closes: https://syzkaller.appspot.com/bug?extid=b2b14916b77acf8626d7
Reported-by: [email protected]
Link: https://syzkaller.appspot.com/bug?extid=d98fd19acd08b36ff422
Fixes: 8c26c4e ("nilfs2: fix issue with flush kernel thread after remount in RO mode because of driver's internal error or metadata corruption")
Tested-by: [email protected]
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: konis

fs/nilfs2/page.c

@@ -388,31 +388,54 @@ void nilfs_clear_dirty_pages(struct address_space *mapping)
 /**
  * nilfs_clear_dirty_page - discard dirty page
  * @page: dirty page that will be discarded
+ *
+ * nilfs_clear_dirty_page() clears working states including dirty state for
+ * the page and its buffers.  If the page has buffers, clear only if it is
+ * confirmed that none of the buffer heads are busy (none have valid
+ * references and none are locked).
  */
 void nilfs_clear_dirty_page(struct page *page)
 {
 	BUG_ON(!PageLocked(page));
 
-	ClearPageUptodate(page);
-	ClearPageMappedToDisk(page);
-	ClearPageChecked(page);
-
 	if (page_has_buffers(page)) {
-		struct buffer_head *bh, *head;
+		struct buffer_head *bh, *head = page_buffers(page);
 		const unsigned long clear_bits =
 			(BIT(BH_Uptodate) | BIT(BH_Dirty) | BIT(BH_Mapped) |
 			 BIT(BH_Async_Write) | BIT(BH_NILFS_Volatile) |
 			 BIT(BH_NILFS_Checked) | BIT(BH_NILFS_Redirected) |
 			 BIT(BH_Delay));
+		bool busy, invalidated = false;
 
-		bh = head = page_buffers(page);
+recheck_buffers:
+		busy = false;
+		bh = head;
+		do {
+			if (atomic_read(&bh->b_count) | buffer_locked(bh)) {
+				busy = true;
+				break;
+			}
+		} while (bh = bh->b_this_page, bh != head);
+
+		if (busy) {
+			if (invalidated)
+				return;
+			invalidate_bh_lrus();
+			invalidated = true;
+			goto recheck_buffers;
+		}
+
+		bh = head;
 		do {
 			lock_buffer(bh);
 			set_mask_bits(&bh->b_state, clear_bits, 0);
 			unlock_buffer(bh);
 		} while (bh = bh->b_this_page, bh != head);
 	}
 
+	ClearPageUptodate(page);
+	ClearPageMappedToDisk(page);
+	ClearPageChecked(page);
 	__nilfs_clear_page_dirty(page);
 }