gfs2: Close timing window with GLF_INVALIDATE_IN_PROGRESS
authorBob Peterson <rpeterso@redhat.com>
Fri, 15 Nov 2019 15:45:41 +0000 (10:45 -0500)
committerAndreas Gruenbacher <agruenba@redhat.com>
Fri, 15 Nov 2019 17:21:59 +0000 (18:21 +0100)
This patch closes a timing window in which two processes compete
and overlap in the execution of do_xmote for the same glock:

             Process A                              Process B
   ------------------------------------   -----------------------------
1. Grabs gl_lockref and calls do_xmote
2.                                        Grabs gl_lockref but is blocked
3. Sets GLF_INVALIDATE_IN_PROGRESS
4. Unlocks gl_lockref
5.                                        Calls do_xmote
6. Call glops->go_sync
7. test_and_clear_bit GLF_DIRTY
8. Call gfs2_log_flush                    Call glops->go_sync
9. (slow IO, so it blocks a long time)    test_and_clear_bit GLF_DIRTY
                                          It's not dirty (step 7) returns
10.                                       Tests GLF_INVALIDATE_IN_PROGRESS
11.                                       Calls go_inval (rgrp_go_inval)
12.                                       gfs2_rgrp_relse does brelse
13.                                       truncate_inode_pages_range
14.                                       Calls lm_lock UN

In step 14 we've just told dlm to give the glock to another node
when, in fact, process A has not finished the IO and synced all
buffer_heads to disk and make sure their revokes are done.

This patch fixes the problem by changing the GLF_INVALIDATE_IN_PROGRESS
to use test_and_set_bit, and if the bit is already set, process B just
ignores it and trusts that process A will do the do_xmote in the proper
order.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
fs/gfs2/glock.c

index faa88bd..b7123de 100644 (file)
@@ -558,7 +558,14 @@ __acquires(&gl->gl_lockref.lock)
        GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
        if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
            glops->go_inval) {
-               set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
+               /*
+                * If another process is already doing the invalidate, let that
+                * finish first.  The glock state machine will get back to this
+                * holder again later.
+                */
+               if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
+                                    &gl->gl_flags))
+                       return;
                do_error(gl, 0); /* Fail queued try locks */
        }
        gl->gl_req = target;