Bug 179511

Summary:	fs: ocfs2: uptodate.c: set_buffer_uptodate(bh) is not clearing if __ocfs2_set_buffer_uptodate fails, which it can.
Product:	File System	Reporter:	Michael (michael_allen_bond)
Component:	ocfs2	Assignee:	io_other
Status:	NEW ---
Severity:	low	CC:	szg00000
Priority:	P1
Hardware:	All
OS:	Linux
Kernel Version:	Linux 4.9	Subsystem:
Regression:	No	Bisected commit-id:

Description Michael 2016-10-21 10:03:51 UTC

In ocfs2_set_new_buffer_uptodate, the slow path in ocfs2_set_buffer_uptodate may fail, but ocfs2_set_new_buffer_uptodate will return without calling  clear_buffer_uptodate(bh). This may never occur in runtime.

Based on static analysis using PALLAS.

Comment 1 Michael 2016-10-27 19:01:03 UTC

Specifically:

set_buffer_uptodate(bh);

ocfs2_metadata_cache_io_lock(ci);
ocfs2_set_buffer_uptodate(ci, bh);
ocfs2_metadata_cache_io_unlock(ci);

If these two functions are correlated (via <bh>), should they be in the locked portion together?

If set_buffer_uptodate is cleared in another thread, how does that affect ocfs2_set_buffer_uptodate?