[JFFS2][XATTR] using 'delete marker' for xdatum/xref deletion

- When xdatum is removed, a new xdatum with 'delete marker' is
  written. (version==0xffffffff means 'delete marker')
- When xref is removed, a new xref with 'delete marker' is written.
  (odd-numbered xseqno means 'delete marker')

- delete_xattr_(datum/xref)_delay() are new deletion functions
  are added. We can only use them if we can detect the target
  obsolete xdatum/xref as a orphan or errir one.
  (e.g when inode deletion, or detecting crc error)

[1/3] jffs2-xattr-v6-01-delete_marker.patch

Signed-off-by: KaiGai Kohei <kaigai@ak.jp.nec.com>
Signed-off-by: David Woodhouse <dwmw2@infradead.org>

9 files changed, 583 insertions, 339 deletions

diff --git a/fs/jffs2/erase.c b/fs/jffs2/erase.c
index 1862e8bc101..1644e340885 100644
--- a/fs/jffs2/erase.c
+++ b/fs/jffs2/erase.c
@@ -230,7 +230,6 @@ static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
 			   at the end of the linked list. Stash it and continue
 			   from the beginning of the list */
 			ic = (struct jffs2_inode_cache *)(*prev);
-			BUG_ON(ic->class != RAWNODE_CLASS_INODE_CACHE);
 			prev = &ic->nodes;
 			continue;
 		}
@@ -254,7 +253,8 @@ static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
 
 	/* PARANOIA */
 	if (!ic) {
-		printk(KERN_WARNING "inode_cache not found in remove_node_refs()!!\n");
+		JFFS2_WARNING("inode_cache/xattr_datum/xattr_ref"
+			      " not found in remove_node_refs()!!\n");
 		return;
 	}
 
@@ -279,8 +279,19 @@ static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
 		printk("\n");
 	});
 
-	if (ic->nodes == (void *)ic && ic->nlink == 0)
-		jffs2_del_ino_cache(c, ic);
+	switch (ic->class) {
+#ifdef CONFIG_JFFS2_FS_XATTR
+		case RAWNODE_CLASS_XATTR_DATUM:
+			jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
+			break;
+		case RAWNODE_CLASS_XATTR_REF:
+			jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
+			break;
+#endif
+		default:
+			if (ic->nodes == (void *)ic && ic->nlink == 0)
+				jffs2_del_ino_cache(c, ic);
+	}
 }
 
 void jffs2_free_jeb_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
diff --git a/fs/jffs2/gc.c b/fs/jffs2/gc.c
index 477c526d638..f59b147661c 100644
--- a/fs/jffs2/gc.c
+++ b/fs/jffs2/gc.c
@@ -275,13 +275,12 @@ int jffs2_garbage_collect_pass(struct jffs2_sb_info *c)
 	 * We can decide whether this node is inode or xattr by ic->class.     */
 	if (ic->class == RAWNODE_CLASS_XATTR_DATUM
 	    || ic->class == RAWNODE_CLASS_XATTR_REF) {
-		BUG_ON(raw->next_in_ino != (void *)ic);
 		spin_unlock(&c->erase_completion_lock);
 
 		if (ic->class == RAWNODE_CLASS_XATTR_DATUM) {
-			ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
+			ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic, raw);
 		} else {
-			ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
+			ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic, raw);
 		}
 		goto release_sem;
 	}
diff --git a/fs/jffs2/jffs2_fs_sb.h b/fs/jffs2/jffs2_fs_sb.h
index 935fec1b120..b98594992ee 100644
--- a/fs/jffs2/jffs2_fs_sb.h
+++ b/fs/jffs2/jffs2_fs_sb.h
@@ -119,8 +119,11 @@ struct jffs2_sb_info {
 #ifdef CONFIG_JFFS2_FS_XATTR
 #define XATTRINDEX_HASHSIZE	(57)
 	uint32_t highest_xid;
+	uint32_t highest_xseqno;
 	struct list_head xattrindex[XATTRINDEX_HASHSIZE];
 	struct list_head xattr_unchecked;
+	struct list_head xattr_dead_list;
+	struct jffs2_xattr_ref *xref_dead_list;
 	struct jffs2_xattr_ref *xref_temp;
 	struct rw_semaphore xattr_sem;
 	uint32_t xdatum_mem_usage;
diff --git a/fs/jffs2/malloc.c b/fs/jffs2/malloc.c
index 4889d0700c0..8310c95478e 100644
--- a/fs/jffs2/malloc.c
+++ b/fs/jffs2/malloc.c
@@ -291,6 +291,7 @@ struct jffs2_xattr_datum *jffs2_alloc_xattr_datum(void)
 
 	memset(xd, 0, sizeof(struct jffs2_xattr_datum));
 	xd->class = RAWNODE_CLASS_XATTR_DATUM;
+	xd->node = (void *)xd;
 	INIT_LIST_HEAD(&xd->xindex);
 	return xd;
 }
@@ -309,6 +310,7 @@ struct jffs2_xattr_ref *jffs2_alloc_xattr_ref(void)
 
 	memset(ref, 0, sizeof(struct jffs2_xattr_ref));
 	ref->class = RAWNODE_CLASS_XATTR_REF;
+	ref->node = (void *)ref;
 	return ref;
 }
 
diff --git a/fs/jffs2/nodemgmt.c b/fs/jffs2/nodemgmt.c
index 8bedfd2ff68..01594a2256e 100644
--- a/fs/jffs2/nodemgmt.c
+++ b/fs/jffs2/nodemgmt.c
@@ -684,19 +684,26 @@ void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref
 		spin_lock(&c->erase_completion_lock);
 
 		ic = jffs2_raw_ref_to_ic(ref);
-		/* It seems we should never call jffs2_mark_node_obsolete() for
-		   XATTR nodes.... yet. Make sure we notice if/when we change
-		   that :) */
-		BUG_ON(ic->class != RAWNODE_CLASS_INODE_CACHE);
 		for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
 			;
 
 		*p = ref->next_in_ino;
 		ref->next_in_ino = NULL;
 
-		if (ic->nodes == (void *)ic && ic->nlink == 0)
-			jffs2_del_ino_cache(c, ic);
-
+		switch (ic->class) {
+#ifdef CONFIG_JFFS2_FS_XATTR
+			case RAWNODE_CLASS_XATTR_DATUM:
+				jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
+				break;
+			case RAWNODE_CLASS_XATTR_REF:
+				jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
+				break;
+#endif
+			default:
+				if (ic->nodes == (void *)ic && ic->nlink == 0)
+					jffs2_del_ino_cache(c, ic);
+				break;
+		}
 		spin_unlock(&c->erase_completion_lock);
 	}
 
diff --git a/fs/jffs2/scan.c b/fs/jffs2/scan.c
index 61618080b86..79638f56c5e 100644
--- a/fs/jffs2/scan.c
+++ b/fs/jffs2/scan.c
@@ -317,20 +317,25 @@ static int jffs2_scan_xattr_node(struct jffs2_sb_info *c, struct jffs2_erasebloc
 				 struct jffs2_summary *s)
 {
 	struct jffs2_xattr_datum *xd;
-	uint32_t totlen, crc;
+	uint32_t xid, version, totlen, crc;
 	int err;
 
 	crc = crc32(0, rx, sizeof(struct jffs2_raw_xattr) - 4);
 	if (crc != je32_to_cpu(rx->node_crc)) {
-		if (je32_to_cpu(rx->node_crc) != 0xffffffff)
-			JFFS2_WARNING("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
-				      ofs, je32_to_cpu(rx->node_crc), crc);
+		JFFS2_WARNING("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
+			      ofs, je32_to_cpu(rx->node_crc), crc);
 		if ((err = jffs2_scan_dirty_space(c, jeb, je32_to_cpu(rx->totlen))))
 			return err;
 		return 0;
 	}
 
-	totlen = PAD(sizeof(*rx) + rx->name_len + 1 + je16_to_cpu(rx->value_len));
+	xid = je32_to_cpu(rx->xid);
+	version = je32_to_cpu(rx->version);
+
+	totlen = sizeof(struct jffs2_raw_xattr);
+	if (version != XDATUM_DELETE_MARKER)
+		totlen += rx->name_len + 1 + je16_to_cpu(rx->value_len);
+	totlen = PAD(totlen);
 	if (totlen != je32_to_cpu(rx->totlen)) {
 		JFFS2_WARNING("node length mismatch at %#08x, read=%u, calc=%u\n",
 			      ofs, je32_to_cpu(rx->totlen), totlen);
@@ -339,22 +344,24 @@ static int jffs2_scan_xattr_node(struct jffs2_sb_info *c, struct jffs2_erasebloc
 		return 0;
 	}
 
-	xd = jffs2_setup_xattr_datum(c, je32_to_cpu(rx->xid), je32_to_cpu(rx->version));
-	if (IS_ERR(xd)) {
-		if (PTR_ERR(xd) == -EEXIST) {
-			if ((err = jffs2_scan_dirty_space(c, jeb, PAD(je32_to_cpu(rx->totlen)))))
-				return err;
-			return 0;
-		}
+	xd = jffs2_setup_xattr_datum(c, xid, version);
+	if (IS_ERR(xd))
 		return PTR_ERR(xd);
-	}
-	xd->xprefix = rx->xprefix;
-	xd->name_len = rx->name_len;
-	xd->value_len = je16_to_cpu(rx->value_len);
-	xd->data_crc = je32_to_cpu(rx->data_crc);
 
-	xd->node = jffs2_link_node_ref(c, jeb, ofs | REF_PRISTINE, totlen, NULL);
-	/* FIXME */ xd->node->next_in_ino = (void *)xd;
+	if (xd->version > version) {
+		struct jffs2_raw_node_ref *raw
+			= jffs2_link_node_ref(c, jeb, ofs | REF_PRISTINE, totlen, NULL);
+		raw->next_in_ino = xd->node->next_in_ino;
+		xd->node->next_in_ino = raw;
+	} else {
+		xd->version = version;
+		xd->xprefix = rx->xprefix;
+		xd->name_len = rx->name_len;
+		xd->value_len = je16_to_cpu(rx->value_len);
+		xd->data_crc = je32_to_cpu(rx->data_crc);
+
+		jffs2_link_node_ref(c, jeb, ofs | REF_PRISTINE, totlen, (void *)xd);
+	}
 
 	if (jffs2_sum_active())
 		jffs2_sum_add_xattr_mem(s, rx, ofs - jeb->offset);
@@ -373,9 +380,8 @@ static int jffs2_scan_xref_node(struct jffs2_sb_info *c, struct jffs2_eraseblock
 
 	crc = crc32(0, rr, sizeof(*rr) - 4);
 	if (crc != je32_to_cpu(rr->node_crc)) {
-		if (je32_to_cpu(rr->node_crc) != 0xffffffff)
-			JFFS2_WARNING("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
-				      ofs, je32_to_cpu(rr->node_crc), crc);
+		JFFS2_WARNING("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
+			      ofs, je32_to_cpu(rr->node_crc), crc);
 		if ((err = jffs2_scan_dirty_space(c, jeb, PAD(je32_to_cpu(rr->totlen)))))
 			return err;
 		return 0;
@@ -395,6 +401,7 @@ static int jffs2_scan_xref_node(struct jffs2_sb_info *c, struct jffs2_eraseblock
 		return -ENOMEM;
 
 	/* BEFORE jffs2_build_xattr_subsystem() called, 
+	 * and AFTER xattr_ref is marked as a dead xref,
 	 * ref->xid is used to store 32bit xid, xd is not used
 	 * ref->ino is used to store 32bit inode-number, ic is not used
 	 * Thoes variables are declared as union, thus using those
@@ -404,11 +411,13 @@ static int jffs2_scan_xref_node(struct jffs2_sb_info *c, struct jffs2_eraseblock
 	 */
 	ref->ino = je32_to_cpu(rr->ino);
 	ref->xid = je32_to_cpu(rr->xid);
+	ref->xseqno = je32_to_cpu(rr->xseqno);
+	if (ref->xseqno > c->highest_xseqno)
+		c->highest_xseqno = (ref->xseqno & ~XREF_DELETE_MARKER);
 	ref->next = c->xref_temp;
 	c->xref_temp = ref;
 
-	ref->node = jffs2_link_node_ref(c, jeb, ofs | REF_PRISTINE, PAD(je32_to_cpu(rr->totlen)), NULL);
-	/* FIXME */ ref->node->next_in_ino = (void *)ref;
+	jffs2_link_node_ref(c, jeb, ofs | REF_PRISTINE, PAD(je32_to_cpu(rr->totlen)), (void *)ref);
 
 	if (jffs2_sum_active())
 		jffs2_sum_add_xref_mem(s, rr, ofs - jeb->offset);
diff --git a/fs/jffs2/summary.c b/fs/jffs2/summary.c
index 0b02fc79e4d..c430f1d217e 100644
--- a/fs/jffs2/summary.c
+++ b/fs/jffs2/summary.c
@@ -310,8 +310,6 @@ int jffs2_sum_add_kvec(struct jffs2_sb_info *c, const struct kvec *invecs,
 #ifdef CONFIG_JFFS2_FS_XATTR
 		case JFFS2_NODETYPE_XATTR: {
 			struct jffs2_sum_xattr_mem *temp;
-			if (je32_to_cpu(node->x.version) == 0xffffffff)
-				return 0;
 			temp = kmalloc(sizeof(struct jffs2_sum_xattr_mem), GFP_KERNEL);
 			if (!temp)
 				goto no_mem;
@@ -327,10 +325,6 @@ int jffs2_sum_add_kvec(struct jffs2_sb_info *c, const struct kvec *invecs,
 		}
 		case JFFS2_NODETYPE_XREF: {
 			struct jffs2_sum_xref_mem *temp;
-
-			if (je32_to_cpu(node->r.ino) == 0xffffffff
-			    && je32_to_cpu(node->r.xid) == 0xffffffff)
-				return 0;
 			temp = kmalloc(sizeof(struct jffs2_sum_xref_mem), GFP_KERNEL);
 			if (!temp)
 				goto no_mem;
@@ -483,22 +477,20 @@ static int jffs2_sum_process_sum_data(struct jffs2_sb_info *c, struct jffs2_eras
 
 				xd = jffs2_setup_xattr_datum(c, je32_to_cpu(spx->xid),
 								je32_to_cpu(spx->version));
-				if (IS_ERR(xd)) {
-					if (PTR_ERR(xd) == -EEXIST) {
-						/* a newer version of xd exists */
-						if ((err = jffs2_scan_dirty_space(c, jeb, je32_to_cpu(spx->totlen))))
-							return err;
-						sp += JFFS2_SUMMARY_XATTR_SIZE;
-						break;
-					}
-					JFFS2_NOTICE("allocation of xattr_datum failed\n");
+				if (IS_ERR(xd))
 					return PTR_ERR(xd);
+				if (xd->version > je32_to_cpu(spx->version)) {
+					/* node is not the newest one */
+					struct jffs2_raw_node_ref *raw
+						= sum_link_node_ref(c, jeb, je32_to_cpu(spx->offset) | REF_UNCHECKED,
+								    PAD(je32_to_cpu(spx->totlen)), NULL);
+					raw->next_in_ino = xd->node->next_in_ino;
+					xd->node->next_in_ino = raw;
+				} else {
+					xd->version = je32_to_cpu(spx->version);
+					sum_link_node_ref(c, jeb, je32_to_cpu(spx->offset) | REF_UNCHECKED,
+							  PAD(je32_to_cpu(spx->totlen)), (void *)xd);
 				}
-
-				xd->node = sum_link_node_ref(c, jeb, je32_to_cpu(spx->offset) | REF_UNCHECKED,
-							     PAD(je32_to_cpu(spx->totlen)), NULL);
-				/* FIXME */ xd->node->next_in_ino = (void *)xd;
-
 				*pseudo_random += je32_to_cpu(spx->xid);
 				sp += JFFS2_SUMMARY_XATTR_SIZE;
 
@@ -519,14 +511,11 @@ static int jffs2_sum_process_sum_data(struct jffs2_sb_info *c, struct jffs2_eras
 					JFFS2_NOTICE("allocation of xattr_datum failed\n");
 					return -ENOMEM;
 				}
-				ref->ino = 0xfffffffe;
-				ref->xid = 0xfffffffd;
 				ref->next = c->xref_temp;
 				c->xref_temp = ref;
 
-				ref->node = sum_link_node_ref(c, jeb, je32_to_cpu(spr->offset) | REF_UNCHECKED,
-							      PAD(sizeof(struct jffs2_raw_xref)), NULL);
-				/* FIXME */ ref->node->next_in_ino = (void *)ref;
+				sum_link_node_ref(c, jeb, je32_to_cpu(spr->offset) | REF_UNCHECKED,
+						  PAD(sizeof(struct jffs2_raw_xref)), (void *)ref);
 
 				*pseudo_random += ref->node->flash_offset;
 				sp += JFFS2_SUMMARY_XREF_SIZE;
diff --git a/fs/jffs2/xattr.c b/fs/jffs2/xattr.c
index 2d82e250be3..03871ab7c26 100644
--- a/fs/jffs2/xattr.c
+++ b/fs/jffs2/xattr.c
@@ -23,18 +23,15 @@
  * xattr_datum_hashkey(xprefix, xname, xvalue, xsize)
  *   is used to calcurate xdatum hashkey. The reminder of hashkey into XATTRINDEX_HASHSIZE is
  *   the index of the xattr name/value pair cache (c->xattrindex).
+ * is_xattr_datum_unchecked(c, xd)
+ *   returns 1, if xdatum contains any unchecked raw nodes. if all raw nodes are not
+ *   unchecked, it returns 0.
  * unload_xattr_datum(c, xd)
  *   is used to release xattr name/value pair and detach from c->xattrindex.
  * reclaim_xattr_datum(c)
  *   is used to reclaim xattr name/value pairs on the xattr name/value pair cache when
  *   memory usage by cache is over c->xdatum_mem_threshold. Currentry, this threshold 
  *   is hard coded as 32KiB.
- * delete_xattr_datum_node(c, xd)
- *   is used to delete a jffs2 node is dominated by xdatum. When EBS(Erase Block Summary) is
- *   enabled, it overwrites the obsolete node by myself.
- * delete_xattr_datum(c, xd)
- *   is used to delete jffs2_xattr_datum object. It must be called with 0-value of reference
- *   counter. (It means how many jffs2_xattr_ref object refers this xdatum.)
  * do_verify_xattr_datum(c, xd)
  *   is used to load the xdatum informations without name/value pair from the medium.
  *   It's necessary once, because those informations are not collected during mounting
@@ -53,8 +50,13 @@
  *   is used to write xdatum to medium. xd->version will be incremented.
  * create_xattr_datum(c, xprefix, xname, xvalue, xsize)
  *   is used to create new xdatum and write to medium.
+ * delete_xattr_datum_delay(c, xd)
+ *   is used to delete a xdatum without 'delete marker'. It has a possibility to detect
+ *   orphan xdatum on next mounting.
+ * delete_xattr_datum(c, xd)
+ *   is used to delete a xdatum with 'delete marker'. Calling jffs2_reserve_space() is
+ *   necessary before this function.
  * -------------------------------------------------- */
-
 static uint32_t xattr_datum_hashkey(int xprefix, const char *xname, const char *xvalue, int xsize)
 {
 	int name_len = strlen(xname);
@@ -62,6 +64,22 @@ static uint32_t xattr_datum_hashkey(int xprefix, const char *xname, const char *
 	return crc32(xprefix, xname, name_len) ^ crc32(xprefix, xvalue, xsize);
 }
 
+static int is_xattr_datum_unchecked(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
+{
+	struct jffs2_raw_node_ref *raw;
+	int rc = 0;
+
+	spin_lock(&c->erase_completion_lock);
+	for (raw=xd->node; raw != (void *)xd; raw=raw->next_in_ino) {
+		if (ref_flags(raw) == REF_UNCHECKED) {
+			rc = 1;
+			break;
+		}
+	}
+	spin_unlock(&c->erase_completion_lock);
+	return rc;
+}
+
 static void unload_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
 {
 	/* must be called under down_write(xattr_sem) */
@@ -107,80 +125,39 @@ static void reclaim_xattr_datum(struct jffs2_sb_info *c)
 		     before, c->xdatum_mem_usage, before - c->xdatum_mem_usage);
 }
 
-static void delete_xattr_datum_node(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
-{
-	/* must be called under down_write(xattr_sem) */
-	struct jffs2_raw_xattr rx;
-	size_t length;
-	int rc;
-
-	if (!xd->node) {
-		JFFS2_WARNING("xdatum (xid=%u) is removed twice.\n", xd->xid);
-		return;
-	}
-	if (jffs2_sum_active()) {
-		memset(&rx, 0xff, sizeof(struct jffs2_raw_xattr));
-		rc = jffs2_flash_read(c, ref_offset(xd->node),
-				      sizeof(struct jffs2_unknown_node),
-				      &length, (char *)&rx);
-		if (rc || length != sizeof(struct jffs2_unknown_node)) {
-			JFFS2_ERROR("jffs2_flash_read()=%d, req=%zu, read=%zu at %#08x\n",
-				    rc, sizeof(struct jffs2_unknown_node),
-				    length, ref_offset(xd->node));
-		}
-		rc = jffs2_flash_write(c, ref_offset(xd->node), sizeof(rx),
-				       &length, (char *)&rx);
-		if (rc || length != sizeof(struct jffs2_raw_xattr)) {
-			JFFS2_ERROR("jffs2_flash_write()=%d, req=%zu, wrote=%zu ar %#08x\n",
-				    rc, sizeof(rx), length, ref_offset(xd->node));
-		}
-	}
-	spin_lock(&c->erase_completion_lock);
-	xd->node->next_in_ino = NULL;
-	spin_unlock(&c->erase_completion_lock);
-	jffs2_mark_node_obsolete(c, xd->node);
-	xd->node = NULL;
-}
-
-static void delete_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
-{
-	/* must be called under down_write(xattr_sem) */
-	BUG_ON(xd->refcnt);
-
-	unload_xattr_datum(c, xd);
-	if (xd->node) {
-		delete_xattr_datum_node(c, xd);
-		xd->node = NULL;
-	}
-	jffs2_free_xattr_datum(xd);
-}
-
 static int do_verify_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
 {
 	/* must be called under down_write(xattr_sem) */
 	struct jffs2_eraseblock *jeb;
+	struct jffs2_raw_node_ref *raw;
 	struct jffs2_raw_xattr rx;
 	size_t readlen;
-	uint32_t crc, totlen;
+	uint32_t crc, offset, totlen;
 	int rc;
 
-	BUG_ON(!xd->node);
-	BUG_ON(ref_flags(xd->node) != REF_UNCHECKED);
+	spin_lock(&c->erase_completion_lock);
+	offset = ref_offset(xd->node);
+	if (ref_flags(xd->node) == REF_PRISTINE)
+		goto complete;
+	spin_unlock(&c->erase_completion_lock);
 
-	rc = jffs2_flash_read(c, ref_offset(xd->node), sizeof(rx), &readlen, (char *)&rx);
+	rc = jffs2_flash_read(c, offset, sizeof(rx), &readlen, (char *)&rx);
 	if (rc || readlen != sizeof(rx)) {
 		JFFS2_WARNING("jffs2_flash_read()=%d, req=%zu, read=%zu at %#08x\n",
-			      rc, sizeof(rx), readlen, ref_offset(xd->node));
+			      rc, sizeof(rx), readlen, offset);
 		return rc ? rc : -EIO;
 	}
 	crc = crc32(0, &rx, sizeof(rx) - 4);
 	if (crc != je32_to_cpu(rx.node_crc)) {
-		if (je32_to_cpu(rx.node_crc) != 0xffffffff)
-			JFFS2_ERROR("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
-				    ref_offset(xd->node), je32_to_cpu(rx.hdr_crc), crc);
+		JFFS2_ERROR("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
+			    offset, je32_to_cpu(rx.hdr_crc), crc);
+		xd->flags |= JFFS2_XFLAGS_INVALID;
 		return EIO;
 	}
-	totlen = PAD(sizeof(rx) + rx.name_len + 1 + je16_to_cpu(rx.value_len));
+	totlen = sizeof(rx);
+	if (xd->version != XDATUM_DELETE_MARKER)
+		totlen += rx.name_len + 1 + je16_to_cpu(rx.value_len);
+	totlen = PAD(totlen);
 	if (je16_to_cpu(rx.magic) != JFFS2_MAGIC_BITMASK
 	    || je16_to_cpu(rx.nodetype) != JFFS2_NODETYPE_XATTR
 	    || je32_to_cpu(rx.totlen) != totlen
@@ -188,11 +165,12 @@ static int do_verify_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_dat
 	    || je32_to_cpu(rx.version) != xd->version) {
 		JFFS2_ERROR("inconsistent xdatum at %#08x, magic=%#04x/%#04x, "
 			    "nodetype=%#04x/%#04x, totlen=%u/%u, xid=%u/%u, version=%u/%u\n",
-			    ref_offset(xd->node), je16_to_cpu(rx.magic), JFFS2_MAGIC_BITMASK,
+			    offset, je16_to_cpu(rx.magic), JFFS2_MAGIC_BITMASK,
 			    je16_to_cpu(rx.nodetype), JFFS2_NODETYPE_XATTR,
 			    je32_to_cpu(rx.totlen), totlen,
 			    je32_to_cpu(rx.xid), xd->xid,
 			    je32_to_cpu(rx.version), xd->version);
+		xd->flags |= JFFS2_XFLAGS_INVALID;
 		return EIO;
 	}
 	xd->xprefix = rx.xprefix;
@@ -200,14 +178,17 @@ static int do_verify_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_dat
 	xd->value_len = je16_to_cpu(rx.value_len);
 	xd->data_crc = je32_to_cpu(rx.data_crc);
 
-	/* This JFFS2_NODETYPE_XATTR node is checked */
-	jeb = &c->blocks[ref_offset(xd->node) / c->sector_size];
-	totlen = PAD(je32_to_cpu(rx.totlen));
-
 	spin_lock(&c->erase_completion_lock);
-	c->unchecked_size -= totlen; c->used_size += totlen;
-	jeb->unchecked_size -= totlen; jeb->used_size += totlen;
-	xd->node->flash_offset = ref_offset(xd->node) | REF_PRISTINE;
+ complete:
+	for (raw=xd->node; raw != (void *)xd; raw=raw->next_in_ino) {
+		jeb = &c->blocks[ref_offset(raw) / c->sector_size];
+		totlen = PAD(ref_totlen(c, jeb, raw));
+		if (ref_flags(raw) == REF_UNCHECKED) {
+			c->unchecked_size -= totlen; c->used_size += totlen;
+			jeb->unchecked_size -= totlen; jeb->used_size += totlen;
+		}
+		raw->flash_offset = ref_offset(raw) | ((xd->node==raw) ? REF_PRISTINE : REF_NORMAL);
+	}
 	spin_unlock(&c->erase_completion_lock);
 
 	/* unchecked xdatum is chained with c->xattr_unchecked */
@@ -227,7 +208,6 @@ static int do_load_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum
 	uint32_t crc, length;
 	int i, ret, retry = 0;
 
-	BUG_ON(!xd->node);
 	BUG_ON(ref_flags(xd->node) != REF_PRISTINE);
 	BUG_ON(!list_empty(&xd->xindex));
  retry:
@@ -253,6 +233,7 @@ static int do_load_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum
 			      " at %#08x, read: 0x%08x calculated: 0x%08x\n",
 			      ref_offset(xd->node), xd->data_crc, crc);
 		kfree(data);
+		xd->flags |= JFFS2_XFLAGS_INVALID;
 		return EIO;
 	}
 
@@ -286,16 +267,13 @@ static int load_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *x
 	 * rc > 0 : Unrecoverable error, this node should be deleted.
 	 */
 	int rc = 0;
-	BUG_ON(xd->xname);
-	if (!xd->node)
+
+	if (xd->xname)
+		return 0;
+	if (xd->flags & JFFS2_XFLAGS_INVALID)
 		return EIO;
-	if (unlikely(ref_flags(xd->node) != REF_PRISTINE)) {
+	if (unlikely(is_xattr_datum_unchecked(c, xd)))
 		rc = do_verify_xattr_datum(c, xd);
-		if (rc > 0) {
-			list_del_init(&xd->xindex);
-			delete_xattr_datum_node(c, xd);
-		}
-	}
 	if (!rc)
 		rc = do_load_xattr_datum(c, xd);
 	return rc;
@@ -304,36 +282,43 @@ static int load_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *x
 static int save_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
 {
 	/* must be called under down_write(xattr_sem) */
-	struct jffs2_raw_node_ref *raw;
 	struct jffs2_raw_xattr rx;
 	struct kvec vecs[2];
 	size_t length;
-	int rc, totlen;
+	int rc, totlen, nvecs = 1;
 	uint32_t phys_ofs = write_ofs(c);
 
-	BUG_ON(!xd->xname);
+	BUG_ON(is_xattr_datum_dead(xd) || (xd->flags & JFFS2_XFLAGS_INVALID)
+	       ? !!xd->xname : !xd->xname);
 
 	vecs[0].iov_base = ℞
-	vecs[0].iov_len = PAD(sizeof(rx));
-	vecs[1].iov_base = xd->xname;
-	vecs[1].iov_len = xd->name_len + 1 + xd->value_len;
-	totlen = vecs[0].iov_len + vecs[1].iov_len;
-
+	vecs[0].iov_len = totlen = sizeof(rx);
+	if (!is_xattr_datum_dead(xd) && !(xd->flags & JFFS2_XFLAGS_INVALID)) {
+		nvecs++;
+		vecs[1].iov_base = xd->xname;
+		vecs[1].iov_len = xd->name_len + 1 + xd->value_len;
+		totlen += vecs[1].iov_len;
+	}
 	/* Setup raw-xattr */
+	memset(&rx, 0, sizeof(rx));
 	rx.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
 	rx.nodetype = cpu_to_je16(JFFS2_NODETYPE_XATTR);
 	rx.totlen = cpu_to_je32(PAD(totlen));
 	rx.hdr_crc = cpu_to_je32(crc32(0, &rx, sizeof(struct jffs2_unknown_node) - 4));
 
 	rx.xid = cpu_to_je32(xd->xid);
-	rx.version = cpu_to_je32(++xd->version);
-	rx.xprefix = xd->xprefix;
-	rx.name_len = xd->name_len;
-	rx.value_len = cpu_to_je16(xd->value_len);
-	rx.data_crc = cpu_to_je32(crc32(0, vecs[1].iov_base, vecs[1].iov_len));
+	if (!is_xattr_datum_dead(xd) && !(xd->flags & JFFS2_XFLAGS_INVALID)) {
+		rx.version = cpu_to_je32(++xd->version);
+		rx.xprefix = xd->xprefix;
+		rx.name_len = xd->name_len;
+		rx.value_len = cpu_to_je16(xd->value_len);
+		rx.data_crc = cpu_to_je32(crc32(0, vecs[1].iov_base, vecs[1].iov_len));
+	} else {
+		rx.version = cpu_to_je32(XDATUM_DELETE_MARKER);
+	}
 	rx.node_crc = cpu_to_je32(crc32(0, &rx, sizeof(struct jffs2_raw_xattr) - 4));
 
-	rc = jffs2_flash_writev(c, vecs, 2, phys_ofs, &length, 0);
+	rc = jffs2_flash_writev(c, vecs, nvecs, phys_ofs, &length, 0);
 	if (rc || totlen != length) {
 		JFFS2_WARNING("jffs2_flash_writev()=%d, req=%u, wrote=%zu, at %#08x\n",
 			      rc, totlen, length, phys_ofs);
@@ -343,14 +328,8 @@ static int save_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *x
 
 		return rc;
 	}
-
 	/* success */
-	raw = jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, PAD(totlen), NULL);
-	/* FIXME */ raw->next_in_ino = (void *)xd;
-
-	if (xd->node)
-		delete_xattr_datum_node(c, xd);
-	xd->node = raw;
+	jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, PAD(totlen), (void *)xd);
 
 	dbg_xattr("success on saving xdatum (xid=%u, version=%u, xprefix=%u, xname='%s')\n",
 		  xd->xid, xd->version, xd->xprefix, xd->xname);
@@ -426,6 +405,39 @@ static struct jffs2_xattr_datum *create_xattr_datum(struct jffs2_sb_info *c,
 	return xd;
 }
 
+static void delete_xattr_datum_delay(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
+{
+	/* must be called under down_write(xattr_sem) */
+	BUG_ON(xd->refcnt);
+
+	unload_xattr_datum(c, xd);
+	set_xattr_datum_dead(xd);
+	spin_lock(&c->erase_completion_lock);
+	list_add(&xd->xindex, &c->xattr_dead_list);
+	spin_unlock(&c->erase_completion_lock);
+	JFFS2_NOTICE("xdatum(xid=%u) was removed without delete marker. "
+		     "An orphan xdatum may be detected on next mounting.\n", xd->xid);
+}
+
+static void delete_xattr_datum(struct jffs2_sb_info *c, struct jffs2_xattr_datum *xd)
+{
+	/* must be called under jffs2_reserve_space() and down_write(xattr_sem) */
+	int rc;
+	BUG_ON(xd->refcnt);
+
+	unload_xattr_datum(c, xd);
+	set_xattr_datum_dead(xd);	
+	rc = save_xattr_datum(c, xd);
+	if (rc) {
+		JFFS2_NOTICE("xdatum(xid=%u) was removed without delete marker. "
+			     "An orphan xdatum may be detected on next mounting.\n",
+			     xd->xid);
+	}
+	spin_lock(&c->erase_completion_lock);
+	list_add(&xd->xindex, &c->xattr_dead_list);
+	spin_unlock(&c->erase_completion_lock);
+}
+
 /* -------- xref related functions ------------------
  * verify_xattr_ref(c, ref)
  *   is used to load xref information from medium. Because summary data does not
@@ -450,25 +462,29 @@ static struct jffs2_xattr_datum *create_xattr_datum(struct jffs2_sb_info *c,
 static int verify_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref)
 {
 	struct jffs2_eraseblock *jeb;
+	struct jffs2_raw_node_ref *raw;
 	struct jffs2_raw_xref rr;
 	size_t readlen;
-	uint32_t crc, totlen;
+	uint32_t crc, offset, totlen;
 	int rc;
 
-	BUG_ON(ref_flags(ref->node) != REF_UNCHECKED);
+	spin_lock(&c->erase_completion_lock);
+	if (ref_flags(ref->node) != REF_UNCHECKED)
+		goto complete;
+	offset = ref_offset(ref->node);
+	spin_unlock(&c->erase_completion_lock);
 
-	rc = jffs2_flash_read(c, ref_offset(ref->node), sizeof(rr), &readlen, (char *)&rr);
+	rc = jffs2_flash_read(c, offset, sizeof(rr), &readlen, (char *)&rr);
 	if (rc || sizeof(rr) != readlen) {
 		JFFS2_WARNING("jffs2_flash_read()=%d, req=%zu, read=%zu, at %#08x\n",
-			      rc, sizeof(rr), readlen, ref_offset(ref->node));
+			      rc, sizeof(rr), readlen, offset);
 		return rc ? rc : -EIO;
 	}
 	/* obsolete node */
 	crc = crc32(0, &rr, sizeof(rr) - 4);
 	if (crc != je32_to_cpu(rr.node_crc)) {
-		if (je32_to_cpu(rr.node_crc) != 0xffffffff)
-			JFFS2_ERROR("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
-				    ref_offset(ref->node), je32_to_cpu(rr.node_crc), crc);
+		JFFS2_ERROR("node CRC failed at %#08x, read=%#08x, calc=%#08x\n",
+			    offset, je32_to_cpu(rr.node_crc), crc);
 		return EIO;
 	}
 	if (je16_to_cpu(rr.magic) != JFFS2_MAGIC_BITMASK
@@ -476,22 +492,28 @@ static int verify_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref
 	    || je32_to_cpu(rr.totlen) != PAD(sizeof(rr))) {
 		JFFS2_ERROR("inconsistent xref at %#08x, magic=%#04x/%#04x, "
 			    "nodetype=%#04x/%#04x, totlen=%u/%zu\n",
-			    ref_offset(ref->node), je16_to_cpu(rr.magic), JFFS2_MAGIC_BITMASK,
+			    offset, je16_to_cpu(rr.magic), JFFS2_MAGIC_BITMASK,
 			    je16_to_cpu(rr.nodetype), JFFS2_NODETYPE_XREF,
 			    je32_to_cpu(rr.totlen), PAD(sizeof(rr)));
 		return EIO;
 	}
 	ref->ino = je32_to_cpu(rr.ino);
 	ref->xid = je32_to_cpu(rr.xid);
-
-	/* fixup superblock/eraseblock info */
-	jeb = &c->blocks[ref_offset(ref->node) / c->sector_size];
-	totlen = PAD(sizeof(rr));
+	ref->xseqno = je32_to_cpu(rr.xseqno);
+	if (ref->xseqno > c->highest_xseqno)
+		c->highest_xseqno = (ref->xseqno & ~XREF_DELETE_MARKER);
 
 	spin_lock(&c->erase_completion_lock);
-	c->unchecked_size -= totlen; c->used_size += totlen;
-	jeb->unchecked_size -= totlen; jeb->used_size += totlen;
-	ref->node->flash_offset = ref_offset(ref->node) | REF_PRISTINE;
+ complete:
+	for (raw=ref->node; raw != (void *)ref; raw=raw->next_in_ino) {
+		jeb = &c->blocks[ref_offset(raw) / c->sector_size];
+		totlen = PAD(ref_totlen(c, jeb, raw));
+		if (ref_flags(raw) == REF_UNCHECKED) {
+			c->unchecked_size -= totlen; c->used_size += totlen;
+			jeb->unchecked_size -= totlen; jeb->used_size += totlen;
+		}
+		raw->flash_offset = ref_offset(raw) | ((ref->node==raw) ? REF_PRISTINE : REF_NORMAL);
+	}
 	spin_unlock(&c->erase_completion_lock);
 
 	dbg_xattr("success on verifying xref (ino=%u, xid=%u) at %#08x\n",
@@ -499,58 +521,12 @@ static int verify_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref
 	return 0;
 }
 
-static void delete_xattr_ref_node(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref)
-{
-	struct jffs2_raw_xref rr;
-	size_t length;
-	int rc;
-
-	if (jffs2_sum_active()) {
-		memset(&rr, 0xff, sizeof(rr));
-		rc = jffs2_flash_read(c, ref_offset(ref->node),
-				      sizeof(struct jffs2_unknown_node),
-				      &length, (char *)&rr);
-		if (rc || length != sizeof(struct jffs2_unknown_node)) {
-			JFFS2_ERROR("jffs2_flash_read()=%d, req=%zu, read=%zu at %#08x\n",
-				    rc, sizeof(struct jffs2_unknown_node),
-				    length, ref_offset(ref->node));
-		}
-		rc = jffs2_flash_write(c, ref_offset(ref->node), sizeof(rr),
-				       &length, (char *)&rr);
-		if (rc || length != sizeof(struct jffs2_raw_xref)) {
-			JFFS2_ERROR("jffs2_flash_write()=%d, req=%zu, wrote=%zu at %#08x\n",
-				    rc, sizeof(rr), length, ref_offset(ref->node));
-		}
-	}
-	spin_lock(&c->erase_completion_lock);
-	ref->node->next_in_ino = NULL;
-	spin_unlock(&c->erase_completion_lock);
-	jffs2_mark_node_obsolete(c, ref->node);
-	ref->node = NULL;
-}
-
-static void delete_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref)
-{
-	/* must be called under down_write(xattr_sem) */
-	struct jffs2_xattr_datum *xd;
-
-	BUG_ON(!ref->node);
-	delete_xattr_ref_node(c, ref);
-
-	xd = ref->xd;
-	xd->refcnt--;
-	if (!xd->refcnt)
-		delete_xattr_datum(c, xd);
-	jffs2_free_xattr_ref(ref);
-}
-
 static int save_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref)
 {
 	/* must be called under down_write(xattr_sem) */
-	struct jffs2_raw_node_ref *raw;
 	struct jffs2_raw_xref rr;
 	size_t length;
-	uint32_t phys_ofs = write_ofs(c);
+	uint32_t xseqno, phys_ofs = write_ofs(c);
 	int ret;
 
 	rr.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
@@ -558,8 +534,16 @@ static int save_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref)
 	rr.totlen = cpu_to_je32(PAD(sizeof(rr)));
 	rr.hdr_crc = cpu_to_je32(crc32(0, &rr, sizeof(struct jffs2_unknown_node) - 4));
 
-	rr.ino = cpu_to_je32(ref->ic->ino);
-	rr.xid = cpu_to_je32(ref->xd->xid);
+	xseqno = (c->highest_xseqno += 2);
+	if (is_xattr_ref_dead(ref)) {
+		xseqno |= XREF_DELETE_MARKER;
+		rr.ino = cpu_to_je32(ref->ino);
+		rr.xid = cpu_to_je32(ref->xid);
+	} else {
+		rr.ino = cpu_to_je32(ref->ic->ino);
+		rr.xid = cpu_to_je32(ref->xd->xid);
+	}
+	rr.xseqno = cpu_to_je32(xseqno);
 	rr.node_crc = cpu_to_je32(crc32(0, &rr, sizeof(rr) - 4));
 
 	ret = jffs2_flash_write(c, phys_ofs, sizeof(rr), &length, (char *)&rr);
@@ -572,12 +556,9 @@ static int save_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref)
 
 		return ret;
 	}
-
-	raw = jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, PAD(sizeof(rr)), NULL);
-	/* FIXME */ raw->next_in_ino = (void *)ref;
-	if (ref->node)
-		delete_xattr_ref_node(c, ref);
-	ref->node = raw;
+	/* success */
+	ref->xseqno = xseqno;
+	jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, PAD(sizeof(rr)), (void *)ref);
 
 	dbg_xattr("success on saving xref (ino=%u, xid=%u)\n", ref->ic->ino, ref->xd->xid);
 
@@ -610,22 +591,116 @@ static struct jffs2_xattr_ref *create_xattr_ref(struct jffs2_sb_info *c, struct
 	return ref; /* success */
 }
 
+static void delete_xattr_ref_delay(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref)
+{
+	/* must be called under down_write(xattr_sem) */
+	struct jffs2_xattr_datum *xd;
+
+	set_xattr_ref_dead(ref);
+	xd = ref->xd;
+	ref->ino = ref->ic->ino;
+	ref->xid = ref->xd->xid;
+	spin_lock(&c->erase_completion_lock);
+	ref->next = c->xref_dead_list;
+	c->xref_dead_list = ref;
+	spin_unlock(&c->erase_completion_lock);
+
+	JFFS2_NOTICE("xref(ino=%u, xid=%u) was removed without delete marker. "
+		     "An orphan xref may be detected on next mounting.\n",
+		     ref->ino, ref->xid);
+
+	if (!--xd->refcnt)
+		delete_xattr_datum_delay(c, xd);
+}
+
+static int delete_xattr_ref(struct jffs2_sb_info *c, struct jffs2_xattr_ref *ref, int enforce)
+{
+	/* must be called under jffs2_reserve_space() and down_write(xattr_sem) */
+	struct jffs2_inode_cache *ic;
+	struct jffs2_xattr_datum *xd;
+	uint32_t length;
+	int rc;
+
+	set_xattr_ref_dead(ref);
+	ic = ref->ic;
+	xd = ref->xd;
+	ref->ino = ic->ino;
+	ref->xid = xd->xid;
+	rc = save_xattr_ref(c, ref);
+	if (rc) {
+		if (!enforce) {
+			clr_xattr_ref_dead(ref);
+			ref->ic = ic;
+			ref->xd = xd;
+			return rc;
+		}
+		JFFS2_WARNING("could not write delete marker of xref(ino=%u, xid=%u). "
+			      "An orphan xref may be detected on next mounting.\n",
+			      ref->ic->ino, ref->xd->xid);
+	}
+	spin_lock(&c->erase_comp