Clean up orphaned indirdep dependency structures after disk failure.

During forcible unmount after a disk failure there is a bug that
causes one or more indirdep dependency structures to fail to be
deallocated. Until we manage to track down why they fail to get
cleaned up, this code tracks them down and eliminates them so that
the unmount can succeed.

Reported by:  Peter Holm
Help from:    kib
Reviewed by:  Chuck Silvers
Tested by:    Peter Holm
MFC after:    7 days
Sponsored by: Netflix

2 files changed, 47 insertions, 11 deletions

diff --git a/sys/ufs/ffs/ffs_softdep.c b/sys/ufs/ffs/ffs_softdep.c
index da80f28bc814..e518cc5c5deb 100644
--- a/sys/ufs/ffs/ffs_softdep.c
+++ b/sys/ufs/ffs/ffs_softdep.c
@@ -1233,9 +1233,7 @@ workitem_free(item, type)
 	    ump->um_fs->fs_fsmnt, TYPENAME(item->wk_type)));
 	atomic_subtract_long(&dep_current[item->wk_type], 1);
 	ump->softdep_curdeps[item->wk_type] -= 1;
-#ifdef INVARIANTS
 	LIST_REMOVE(item, wk_all);
-#endif
 	free(item, DtoM(type));
 }
 
@@ -1262,9 +1260,7 @@ workitem_alloc(item, type, mp)
 	ump->softdep_curdeps[type] += 1;
 	ump->softdep_deps++;
 	ump->softdep_accdeps++;
-#ifdef INVARIANTS
 	LIST_INSERT_HEAD(&ump->softdep_alldeps[type], item, wk_all);
-#endif
 	FREE_LOCK(ump);
 }
 
@@ -1293,10 +1289,8 @@ workitem_reassign(item, newtype)
 	dep_total[newtype]++;
 	FREE_GBLLOCK(&lk);
 	item->wk_type = newtype;
-#ifdef INVARIANTS
 	LIST_REMOVE(item, wk_all);
 	LIST_INSERT_HEAD(&ump->softdep_alldeps[newtype], item, wk_all);
-#endif
 }
 
 /*
@@ -2713,10 +2707,8 @@ softdep_mount(devvp, mp, fs, cred)
 	sdp->sd_indirhashsize = i - 1;
 	for (i = 0; i <= sdp->sd_indirhashsize; i++)
 		TAILQ_INIT(&sdp->sd_indirhash[i]);
-#ifdef INVARIANTS
 	for (i = 0; i <= D_LAST; i++)
 		LIST_INIT(&sdp->sd_alldeps[i]);
-#endif
 	ACQUIRE_GBLLOCK(&lk);
 	TAILQ_INSERT_TAIL(&softdepmounts, sdp, sd_next);
 	FREE_GBLLOCK(&lk);
@@ -14793,9 +14785,12 @@ softdep_check_suspend(struct mount *mp,
 		      int secondary_writes,
 		      int secondary_accwrites)
 {
+	struct buf *bp;
 	struct bufobj *bo;
 	struct ufsmount *ump;
 	struct inodedep *inodedep;
+	struct indirdep *indirdep;
+	struct worklist *wk, *nextwk;
 	int error, unlinked;
 
 	bo = &devvp->v_bufobj;
@@ -14872,8 +14867,51 @@ softdep_check_suspend(struct mount *mp,
 	}
 
 	/*
+	 * XXX Check for orphaned indirdep dependency structures.
+	 *
+	 * During forcible unmount after a disk failure there is a
+	 * bug that causes one or more indirdep dependency structures
+	 * to fail to be deallocated. We check for them here and clean
+	 * them up so that the unmount can succeed.
+	 */
+	if ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0 && ump->softdep_deps > 0 &&
+	    ump->softdep_deps == ump->softdep_curdeps[D_INDIRDEP]) {
+		LIST_FOREACH_SAFE(wk, &ump->softdep_alldeps[D_INDIRDEP],
+		    wk_all, nextwk) {
+			indirdep = WK_INDIRDEP(wk);
+			if ((indirdep->ir_state & (GOINGAWAY | DEPCOMPLETE)) !=
+			    (GOINGAWAY | DEPCOMPLETE) ||
+			    !TAILQ_EMPTY(&indirdep->ir_trunc) ||
+			    !LIST_EMPTY(&indirdep->ir_completehd) ||
+			    !LIST_EMPTY(&indirdep->ir_writehd) ||
+			    !LIST_EMPTY(&indirdep->ir_donehd) ||
+			    !LIST_EMPTY(&indirdep->ir_deplisthd) ||
+			    indirdep->ir_saveddata != NULL ||
+			    indirdep->ir_savebp == NULL) {
+				printf("%s: skipping orphaned indirdep %p\n",
+				    __FUNCTION__, indirdep);
+				continue;
+			}
+			printf("%s: freeing orphaned indirdep %p\n",
+			    __FUNCTION__, indirdep);
+			bp = indirdep->ir_savebp;
+			indirdep->ir_savebp = NULL;
+			free_indirdep(indirdep);
+			FREE_LOCK(ump);
+			brelse(bp);
+			while (!TRY_ACQUIRE_LOCK(ump)) {
+				BO_UNLOCK(bo);
+				ACQUIRE_LOCK(ump);
+				FREE_LOCK(ump);
+				BO_LOCK(bo);
+			}
+		}
+	}
+
+	/*
 	 * Reasons for needing more work before suspend:
 	 * - Dirty buffers on devvp.
+	 * - Dependency structures still exist
 	 * - Softdep activity occurred after start of vnode sync loop
 	 * - Secondary writes occurred after start of vnode sync loop
 	 */
diff --git a/sys/ufs/ffs/softdep.h b/sys/ufs/ffs/softdep.h
index 3493aadafc98..41728be3ec0f 100644
--- a/sys/ufs/ffs/softdep.h
+++ b/sys/ufs/ffs/softdep.h
@@ -213,10 +213,10 @@ struct worklist {
 	struct mount		*wk_mp;		/* Mount we live in */
 	unsigned int		wk_type:8,	/* type of request */
 				wk_state:24;	/* state flags */
+	LIST_ENTRY(worklist)	wk_all;		/* list of deps of this type */
 #ifdef INVARIANTS
 	const char		*wk_func;	/* func where added / removed */
 	int			wk_line;	/* line where added / removed */
-	LIST_ENTRY(worklist)	wk_all;		/* list of deps of this type */
 #endif
 };
 #define	WK_DATA(wk) ((void *)(wk))
@@ -1075,9 +1075,7 @@ struct mount_softdeps {
 	TAILQ_ENTRY(mount_softdeps) sd_next;	/* List of softdep filesystem */
 	struct	ufsmount *sd_ump;		/* our ufsmount structure */
 	u_long	sd_curdeps[D_LAST + 1];		/* count of current deps */
-#ifdef INVARIANTS
 	struct	workhead sd_alldeps[D_LAST + 1];/* Lists of all deps */
-#endif
 };
 /*
  * Flags for communicating with the syncer thread.