Merge tag 'for-6.1-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs fixes from David Sterba:

 - fiemap fixes:
      - add missing path cache update
      - fix processing of delayed data and tree refs during backref
        walking, this could lead to reporting incorrect extent sharing

 - fix extent range locking under heavy contention to avoid deadlocks

 - make it possible to test send v3 in debugging mode

 - update links in MAINTAINERS

* tag 'for-6.1-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  MAINTAINERS: update btrfs website links and files
  btrfs: ignore fiemap path cache if we have multiple leaves for a data extent
  btrfs: fix processing of delayed tree block refs during backref walking
  btrfs: fix processing of delayed data refs during backref walking
  btrfs: delete stale comments after merge conflict resolution
  btrfs: unlock locked extent area if we have contention
  btrfs: send: update command for protocol version check
  btrfs: send: allow protocol version 3 with CONFIG_BTRFS_DEBUG
  btrfs: add missing path cache update during fiemap

MAINTAINERS

@@ -4459,13 +4459,15 @@ M:	Josef Bacik <josef@toxicpanda.com>
 M:	David Sterba <dsterba@suse.com>
 L:	linux-btrfs@vger.kernel.org
 S:	Maintained
-W:	http://btrfs.wiki.kernel.org/
-Q:	http://patchwork.kernel.org/project/linux-btrfs/list/
+W:	https://btrfs.readthedocs.io
+W:	https://btrfs.wiki.kernel.org/
+Q:	https://patchwork.kernel.org/project/linux-btrfs/list/
 C:	irc://irc.libera.chat/btrfs
 T:	git git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git
 F:	Documentation/filesystems/btrfs.rst
 F:	fs/btrfs/
 F:	include/linux/btrfs*
+F:	include/trace/events/btrfs.h
 F:	include/uapi/linux/btrfs*
 
 BTTV VIDEO4LINUX DRIVER

fs/btrfs/backref.c

@@ -138,6 +138,7 @@ struct share_check {
 	u64 root_objectid;
 	u64 inum;
 	int share_count;
+	bool have_delayed_delete_refs;
 };
 
 static inline int extent_is_shared(struct share_check *sc)
@@ -820,16 +821,11 @@ static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
 			    struct preftrees *preftrees, struct share_check *sc)
 {
 	struct btrfs_delayed_ref_node *node;
-	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 	struct btrfs_key key;
-	struct btrfs_key tmp_op_key;
 	struct rb_node *n;
 	int count;
 	int ret = 0;
 
-	if (extent_op && extent_op->update_key)
-		btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key);
-
 	spin_lock(&head->lock);
 	for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
 		node = rb_entry(n, struct btrfs_delayed_ref_node,
@@ -855,10 +851,16 @@ static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
 		case BTRFS_TREE_BLOCK_REF_KEY: {
 			/* NORMAL INDIRECT METADATA backref */
 			struct btrfs_delayed_tree_ref *ref;
+			struct btrfs_key *key_ptr = NULL;
+
+			if (head->extent_op && head->extent_op->update_key) {
+				btrfs_disk_key_to_cpu(&key, &head->extent_op->key);
+				key_ptr = &key;
+			}
 
 			ref = btrfs_delayed_node_to_tree_ref(node);
 			ret = add_indirect_ref(fs_info, preftrees, ref->root,
-					       &tmp_op_key, ref->level + 1,
+					       key_ptr, ref->level + 1,
 					       node->bytenr, count, sc,
 					       GFP_ATOMIC);
 			break;
@@ -884,13 +886,22 @@ static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
 			key.offset = ref->offset;
 
 			/*
-			 * Found a inum that doesn't match our known inum, we
-			 * know it's shared.
+			 * If we have a share check context and a reference for
+			 * another inode, we can't exit immediately. This is
+			 * because even if this is a BTRFS_ADD_DELAYED_REF
+			 * reference we may find next a BTRFS_DROP_DELAYED_REF
+			 * which cancels out this ADD reference.
+			 *
+			 * If this is a DROP reference and there was no previous
+			 * ADD reference, then we need to signal that when we
+			 * process references from the extent tree (through
+			 * add_inline_refs() and add_keyed_refs()), we should
+			 * not exit early if we find a reference for another
+			 * inode, because one of the delayed DROP references
+			 * may cancel that reference in the extent tree.
 			 */
-			if (sc && sc->inum && ref->objectid != sc->inum) {
-				ret = BACKREF_FOUND_SHARED;
-				goto out;
-			}
+			if (sc && count < 0)
+				sc->have_delayed_delete_refs = true;
 
 			ret = add_indirect_ref(fs_info, preftrees, ref->root,
 					       &key, 0, node->bytenr, count, sc,
@@ -920,7 +931,7 @@ static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
 	}
 	if (!ret)
 		ret = extent_is_shared(sc);
-out:
+
 	spin_unlock(&head->lock);
 	return ret;
 }
@@ -1023,7 +1034,8 @@ static int add_inline_refs(const struct btrfs_fs_info *fs_info,
 			key.type = BTRFS_EXTENT_DATA_KEY;
 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 
-			if (sc && sc->inum && key.objectid != sc->inum) {
+			if (sc && sc->inum && key.objectid != sc->inum &&
+			    !sc->have_delayed_delete_refs) {
 				ret = BACKREF_FOUND_SHARED;
 				break;
 			}
@@ -1033,6 +1045,7 @@ static int add_inline_refs(const struct btrfs_fs_info *fs_info,
 			ret = add_indirect_ref(fs_info, preftrees, root,
 					       &key, 0, bytenr, count,
 					       sc, GFP_NOFS);
+
 			break;
 		}
 		default:
@@ -1122,7 +1135,8 @@ static int add_keyed_refs(struct btrfs_root *extent_root,
 			key.type = BTRFS_EXTENT_DATA_KEY;
 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 
-			if (sc && sc->inum && key.objectid != sc->inum) {
+			if (sc && sc->inum && key.objectid != sc->inum &&
+			    !sc->have_delayed_delete_refs) {
 				ret = BACKREF_FOUND_SHARED;
 				break;
 			}
@@ -1522,6 +1536,9 @@ static bool lookup_backref_shared_cache(struct btrfs_backref_shared_cache *cache
 {
 	struct btrfs_backref_shared_cache_entry *entry;
 
+	if (!cache->use_cache)
+		return false;
+
 	if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
 		return false;
 
@@ -1557,6 +1574,19 @@ static bool lookup_backref_shared_cache(struct btrfs_backref_shared_cache *cache
 		return false;
 
 	*is_shared = entry->is_shared;
+	/*
+	 * If the node at this level is shared, than all nodes below are also
+	 * shared. Currently some of the nodes below may be marked as not shared
+	 * because we have just switched from one leaf to another, and switched
+	 * also other nodes above the leaf and below the current level, so mark
+	 * them as shared.
+	 */
+	if (*is_shared) {
+		for (int i = 0; i < level; i++) {
+			cache->entries[i].is_shared = true;
+			cache->entries[i].gen = entry->gen;
+		}
+	}
 
 	return true;
 }
@@ -1573,6 +1603,9 @@ static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,
 	struct btrfs_backref_shared_cache_entry *entry;
 	u64 gen;
 
+	if (!cache->use_cache)
+		return;
+
 	if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
 		return;
 
@@ -1648,6 +1681,7 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
 		.root_objectid = root->root_key.objectid,
 		.inum = inum,
 		.share_count = 0,
+		.have_delayed_delete_refs = false,
 	};
 	int level;
 
@@ -1669,6 +1703,7 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
 	/* -1 means we are in the bytenr of the data extent. */
 	level = -1;
 	ULIST_ITER_INIT(&uiter);
+	cache->use_cache = true;
 	while (1) {
 		bool is_shared;
 		bool cached;
@@ -1698,6 +1733,24 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
 		    extent_gen > btrfs_root_last_snapshot(&root->root_item))
 			break;
 
+		/*
+		 * If our data extent was not directly shared (without multiple
+		 * reference items), than it might have a single reference item
+		 * with a count > 1 for the same offset, which means there are 2
+		 * (or more) file extent items that point to the data extent -
+		 * this happens when a file extent item needs to be split and
+		 * then one item gets moved to another leaf due to a b+tree leaf
+		 * split when inserting some item. In this case the file extent
+		 * items may be located in different leaves and therefore some
+		 * of the leaves may be referenced through shared subtrees while
+		 * others are not. Since our extent buffer cache only works for
+		 * a single path (by far the most common case and simpler to
+		 * deal with), we can not use it if we have multiple leaves
+		 * (which implies multiple paths).
+		 */
+		if (level == -1 && tmp->nnodes > 1)
+			cache->use_cache = false;
+
 		if (level >= 0)
 			store_backref_shared_cache(cache, root, bytenr,
 						   level, false);
@@ -1713,6 +1766,7 @@ int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
 			break;
 		}
 		shared.share_count = 0;
+		shared.have_delayed_delete_refs = false;
 		cond_resched();
 	}
 

fs/btrfs/backref.h

@@ -29,6 +29,7 @@ struct btrfs_backref_shared_cache {
 	 * a given data extent should never exceed the maximum b+tree height.
 	 */
 	struct btrfs_backref_shared_cache_entry entries[BTRFS_MAX_LEVEL];
+	bool use_cache;
 };
 
 typedef int (iterate_extent_inodes_t)(u64 inum, u64 offset, u64 root,

fs/btrfs/block-group.c

@@ -774,10 +774,8 @@ int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait)
 
 	btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
 out:
-	/* REVIEW */
 	if (wait && caching_ctl)
 		ret = btrfs_caching_ctl_wait_done(cache, caching_ctl);
-		/* wait_event(caching_ctl->wait, space_cache_v1_done(cache)); */
 	if (caching_ctl)
 		btrfs_put_caching_control(caching_ctl);
 

fs/btrfs/extent-io-tree.c

@@ -1641,16 +1641,17 @@ int lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 	int err;
 	u64 failed_start;
 
-	while (1) {
+	err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, &failed_start,
+			       cached_state, NULL, GFP_NOFS);
+	while (err == -EEXIST) {
+		if (failed_start != start)
+			clear_extent_bit(tree, start, failed_start - 1,
+					 EXTENT_LOCKED, cached_state);
+
+		wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
 		err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
 				       &failed_start, cached_state, NULL,
 				       GFP_NOFS);
-		if (err == -EEXIST) {
-			wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
-			start = failed_start;
-		} else
-			break;
-		WARN_ON(start > end);
 	}
 	return err;
 }

fs/btrfs/send.c

@@ -348,6 +348,7 @@ static bool proto_cmd_ok(const struct send_ctx *sctx, int cmd)
 	switch (sctx->proto) {
 	case 1:	 return cmd <= BTRFS_SEND_C_MAX_V1;
 	case 2:	 return cmd <= BTRFS_SEND_C_MAX_V2;
+	case 3:	 return cmd <= BTRFS_SEND_C_MAX_V3;
 	default: return false;
 	}
 }
@@ -6469,7 +6470,9 @@ static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
 		if (ret < 0)
 			goto out;
 	}
-	if (sctx->cur_inode_needs_verity) {
+
+	if (proto_cmd_ok(sctx, BTRFS_SEND_C_ENABLE_VERITY)
+	    && sctx->cur_inode_needs_verity) {
 		ret = process_verity(sctx);
 		if (ret < 0)
 			goto out;

fs/btrfs/send.h

@@ -10,7 +10,12 @@
 #include <linux/types.h>
 
 #define BTRFS_SEND_STREAM_MAGIC "btrfs-stream"
+/* Conditional support for the upcoming protocol version. */
+#ifdef CONFIG_BTRFS_DEBUG
+#define BTRFS_SEND_STREAM_VERSION 3
+#else
 #define BTRFS_SEND_STREAM_VERSION 2
+#endif
 
 /*
  * In send stream v1, no command is larger than 64K. In send stream v2, no limit