1 files changed, 207 insertions, 105 deletions

diff --git a/fs/ext4/block_validity.c b/fs/ext4/block_validity.c
index 3f11656bd72e..e8c5525afc67 100644
--- a/fs/ext4/block_validity.c
+++ b/fs/ext4/block_validity.c
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: GPL-2.0
 /*
  *  linux/fs/ext4/block_validity.c
  *
@@ -16,7 +17,6 @@
 #include <linux/swap.h>
 #include <linux/pagemap.h>
 #include <linux/blkdev.h>
-#include <linux/mutex.h>
 #include <linux/slab.h>
 #include "ext4.h"
 
@@ -24,6 +24,7 @@ struct ext4_system_zone {
 	struct rb_node	node;
 	ext4_fsblk_t	start_blk;
 	unsigned int	count;
+	u32		ino;
 };
 
 static struct kmem_cache *ext4_system_zone_cachep;
@@ -38,29 +39,40 @@ int __init ext4_init_system_zone(void)
 
 void ext4_exit_system_zone(void)
 {
+	rcu_barrier();
 	kmem_cache_destroy(ext4_system_zone_cachep);
 }
 
 static inline int can_merge(struct ext4_system_zone *entry1,
 		     struct ext4_system_zone *entry2)
 {
-	if ((entry1->start_blk + entry1->count) == entry2->start_blk)
+	if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
+	    entry1->ino == entry2->ino)
 		return 1;
 	return 0;
 }
 
+static void release_system_zone(struct ext4_system_blocks *system_blks)
+{
+	struct ext4_system_zone	*entry, *n;
+
+	rbtree_postorder_for_each_entry_safe(entry, n,
+				&system_blks->root, node)
+		kmem_cache_free(ext4_system_zone_cachep, entry);
+}
+
 /*
  * Mark a range of blocks as belonging to the "system zone" --- that
  * is, filesystem metadata blocks which should never be used by
  * inodes.
  */
-static int add_system_zone(struct ext4_sb_info *sbi,
+static int add_system_zone(struct ext4_system_blocks *system_blks,
 			   ext4_fsblk_t start_blk,
-			   unsigned int count)
+			   unsigned int count, u32 ino)
 {
-	struct ext4_system_zone *new_entry = NULL, *entry;
-	struct rb_node **n = &sbi->system_blks.rb_node, *node;
-	struct rb_node *parent = NULL, *new_node = NULL;
+	struct ext4_system_zone *new_entry, *entry;
+	struct rb_node **n = &system_blks->root.rb_node, *node;
+	struct rb_node *parent = NULL, *new_node;
 
 	while (*n) {
 		parent = *n;
@@ -69,30 +81,21 @@ static int add_system_zone(struct ext4_sb_info *sbi,
 			n = &(*n)->rb_left;
 		else if (start_blk >= (entry->start_blk + entry->count))
 			n = &(*n)->rb_right;
-		else {
-			if (start_blk + count > (entry->start_blk +
-						 entry->count))
-				entry->count = (start_blk + count -
-						entry->start_blk);
-			new_node = *n;
-			new_entry = rb_entry(new_node, struct ext4_system_zone,
-					     node);
-			break;
-		}
+		else	/* Unexpected overlap of system zones. */
+			return -EFSCORRUPTED;
 	}
 
-	if (!new_entry) {
-		new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
-					     GFP_KERNEL);
-		if (!new_entry)
-			return -ENOMEM;
-		new_entry->start_blk = start_blk;
-		new_entry->count = count;
-		new_node = &new_entry->node;
-
-		rb_link_node(new_node, parent, n);
-		rb_insert_color(new_node, &sbi->system_blks);
-	}
+	new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
+				     GFP_KERNEL);
+	if (!new_entry)
+		return -ENOMEM;
+	new_entry->start_blk = start_blk;
+	new_entry->count = count;
+	new_entry->ino = ino;
+	new_node = &new_entry->node;
+
+	rb_link_node(new_node, parent, n);
+	rb_insert_color(new_node, &system_blks->root);
 
 	/* Can we merge to the left? */
 	node = rb_prev(new_node);
@@ -101,7 +104,7 @@ static int add_system_zone(struct ext4_sb_info *sbi,
 		if (can_merge(entry, new_entry)) {
 			new_entry->start_blk = entry->start_blk;
 			new_entry->count += entry->count;
-			rb_erase(node, &sbi->system_blks);
+			rb_erase(node, &system_blks->root);
 			kmem_cache_free(ext4_system_zone_cachep, entry);
 		}
 	}
@@ -112,7 +115,7 @@ static int add_system_zone(struct ext4_sb_info *sbi,
 		entry = rb_entry(node, struct ext4_system_zone, node);
 		if (can_merge(new_entry, entry)) {
 			new_entry->count += entry->count;
-			rb_erase(node, &sbi->system_blks);
+			rb_erase(node, &system_blks->root);
 			kmem_cache_free(ext4_system_zone_cachep, entry);
 		}
 	}
@@ -123,114 +126,197 @@ static void debug_print_tree(struct ext4_sb_info *sbi)
 {
 	struct rb_node *node;
 	struct ext4_system_zone *entry;
+	struct ext4_system_blocks *system_blks;
 	int first = 1;
 
 	printk(KERN_INFO "System zones: ");
-	node = rb_first(&sbi->system_blks);
+	rcu_read_lock();
+	system_blks = rcu_dereference(sbi->s_system_blks);
+	node = rb_first(&system_blks->root);
 	while (node) {
 		entry = rb_entry(node, struct ext4_system_zone, node);
-		printk("%s%llu-%llu", first ? "" : ", ",
+		printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
 		       entry->start_blk, entry->start_blk + entry->count - 1);
 		first = 0;
 		node = rb_next(node);
 	}
-	printk("\n");
+	rcu_read_unlock();
+	printk(KERN_CONT "\n");
+}
+
+static int ext4_protect_reserved_inode(struct super_block *sb,
+				       struct ext4_system_blocks *system_blks,
+				       u32 ino)
+{
+	struct inode *inode;
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct ext4_map_blocks map;
+	u32 i = 0, num;
+	int err = 0, n;
+
+	if ((ino < EXT4_ROOT_INO) ||
+	    (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
+		return -EINVAL;
+	inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
+	if (IS_ERR(inode))
+		return PTR_ERR(inode);
+	num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+	while (i < num) {
+		cond_resched();
+		map.m_lblk = i;
+		map.m_len = num - i;
+		n = ext4_map_blocks(NULL, inode, &map, 0);
+		if (n < 0) {
+			err = n;
+			break;
+		}
+		if (n == 0) {
+			i++;
+		} else {
+			err = add_system_zone(system_blks, map.m_pblk, n, ino);
+			if (err < 0) {
+				if (err == -EFSCORRUPTED) {
+					EXT4_ERROR_INODE_ERR(inode, -err,
+						"blocks %llu-%llu from inode overlap system zone",
+						map.m_pblk,
+						map.m_pblk + map.m_len - 1);
+				}
+				break;
+			}
+			i += n;
+		}
+	}
+	iput(inode);
+	return err;
 }
 
+static void ext4_destroy_system_zone(struct rcu_head *rcu)
+{
+	struct ext4_system_blocks *system_blks;
+
+	system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
+	release_system_zone(system_blks);
+	kfree(system_blks);
+}
+
+/*
+ * Build system zone rbtree which is used for block validity checking.
+ *
+ * The update of system_blks pointer in this function is protected by
+ * sb->s_umount semaphore. However we have to be careful as we can be
+ * racing with ext4_inode_block_valid() calls reading system_blks rbtree
+ * protected only by RCU. That's why we first build the rbtree and then
+ * swap it in place.
+ */
 int ext4_setup_system_zone(struct super_block *sb)
 {
 	ext4_group_t ngroups = ext4_get_groups_count(sb);
 	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct ext4_system_blocks *system_blks;
 	struct ext4_group_desc *gdp;
 	ext4_group_t i;
-	int flex_size = ext4_flex_bg_size(sbi);
 	int ret;
 
-	if (!test_opt(sb, BLOCK_VALIDITY)) {
-		if (EXT4_SB(sb)->system_blks.rb_node)
-			ext4_release_system_zone(sb);
-		return 0;
-	}
-	if (EXT4_SB(sb)->system_blks.rb_node)
-		return 0;
+	system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
+	if (!system_blks)
+		return -ENOMEM;
 
 	for (i=0; i < ngroups; i++) {
-		if (ext4_bg_has_super(sb, i) &&
-		    ((i < 5) || ((i % flex_size) == 0)))
-			add_system_zone(sbi, ext4_group_first_block_no(sb, i),
-					ext4_bg_num_gdb(sb, i) + 1);
+		unsigned int meta_blks = ext4_num_base_meta_blocks(sb, i);
+
+		cond_resched();
+		if (meta_blks != 0) {
+			ret = add_system_zone(system_blks,
+					ext4_group_first_block_no(sb, i),
+					meta_blks, 0);
+			if (ret)
+				goto err;
+		}
 		gdp = ext4_get_group_desc(sb, i, NULL);
-		ret = add_system_zone(sbi, ext4_block_bitmap(sb, gdp), 1);
+		ret = add_system_zone(system_blks,
+				ext4_block_bitmap(sb, gdp), 1, 0);
+		if (ret)
+			goto err;
+		ret = add_system_zone(system_blks,
+				ext4_inode_bitmap(sb, gdp), 1, 0);
 		if (ret)
-			return ret;
-		ret = add_system_zone(sbi, ext4_inode_bitmap(sb, gdp), 1);
+			goto err;
+		ret = add_system_zone(system_blks,
+				ext4_inode_table(sb, gdp),
+				sbi->s_itb_per_group, 0);
 		if (ret)
-			return ret;
-		ret = add_system_zone(sbi, ext4_inode_table(sb, gdp),
-				sbi->s_itb_per_group);
+			goto err;
+	}
+	if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
+		ret = ext4_protect_reserved_inode(sb, system_blks,
+				le32_to_cpu(sbi->s_es->s_journal_inum));
 		if (ret)
-			return ret;
+			goto err;
 	}
 
+	/*
+	 * System blks rbtree complete, announce it once to prevent racing
+	 * with ext4_inode_block_valid() accessing the rbtree at the same
+	 * time.
+	 */
+	rcu_assign_pointer(sbi->s_system_blks, system_blks);
+
 	if (test_opt(sb, DEBUG))
-		debug_print_tree(EXT4_SB(sb));
+		debug_print_tree(sbi);
 	return 0;
+err:
+	release_system_zone(system_blks);
+	kfree(system_blks);
+	return ret;
 }
 
-/* Called when the filesystem is unmounted */
+/*
+ * Called when the filesystem is unmounted or when remounting it with
+ * noblock_validity specified.
+ *
+ * The update of system_blks pointer in this function is protected by
+ * sb->s_umount semaphore. However we have to be careful as we can be
+ * racing with ext4_inode_block_valid() calls reading system_blks rbtree
+ * protected only by RCU. So we first clear the system_blks pointer and
+ * then free the rbtree only after RCU grace period expires.
+ */
 void ext4_release_system_zone(struct super_block *sb)
 {
-	struct rb_node	*n = EXT4_SB(sb)->system_blks.rb_node;
-	struct rb_node	*parent;
-	struct ext4_system_zone	*entry;
+	struct ext4_system_blocks *system_blks;
 
-	while (n) {
-		/* Do the node's children first */
-		if (n->rb_left) {
-			n = n->rb_left;
-			continue;
-		}
-		if (n->rb_right) {
-			n = n->rb_right;
-			continue;
-		}
-		/*
-		 * The node has no children; free it, and then zero
-		 * out parent's link to it.  Finally go to the
-		 * beginning of the loop and try to free the parent
-		 * node.
-		 */
-		parent = rb_parent(n);
-		entry = rb_entry(n, struct ext4_system_zone, node);
-		kmem_cache_free(ext4_system_zone_cachep, entry);
-		if (!parent)
-			EXT4_SB(sb)->system_blks = RB_ROOT;
-		else if (parent->rb_left == n)
-			parent->rb_left = NULL;
-		else if (parent->rb_right == n)
-			parent->rb_right = NULL;
-		n = parent;
-	}
-	EXT4_SB(sb)->system_blks = RB_ROOT;
+	system_blks = rcu_dereference_protected(EXT4_SB(sb)->s_system_blks,
+					lockdep_is_held(&sb->s_umount));
+	rcu_assign_pointer(EXT4_SB(sb)->s_system_blks, NULL);
+
+	if (system_blks)
+		call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
 }
 
-/*
- * Returns 1 if the passed-in block region (start_blk,
- * start_blk+count) is valid; 0 if some part of the block region
- * overlaps with filesystem metadata blocks.
- */
-int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
-			  unsigned int count)
+int ext4_sb_block_valid(struct super_block *sb, struct inode *inode,
+				ext4_fsblk_t start_blk, unsigned int count)
 {
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct ext4_system_blocks *system_blks;
 	struct ext4_system_zone *entry;
-	struct rb_node *n = sbi->system_blks.rb_node;
+	struct rb_node *n;
+	int ret = 1;
 
 	if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 	    (start_blk + count < start_blk) ||
-	    (start_blk + count > ext4_blocks_count(sbi->s_es))) {
-		sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
+	    (start_blk + count > ext4_blocks_count(sbi->s_es)))
 		return 0;
-	}
+
+	/*
+	 * Lock the system zone to prevent it being released concurrently
+	 * when doing a remount which inverse current "[no]block_validity"
+	 * mount option.
+	 */
+	rcu_read_lock();
+	system_blks = rcu_dereference(sbi->s_system_blks);
+	if (system_blks == NULL)
+		goto out_rcu;
+
+	n = system_blks->root.rb_node;
 	while (n) {
 		entry = rb_entry(n, struct ext4_system_zone, node);
 		if (start_blk + count - 1 < entry->start_blk)
@@ -238,29 +324,45 @@ int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
 		else if (start_blk >= (entry->start_blk + entry->count))
 			n = n->rb_right;
 		else {
-			sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
-			return 0;
+			ret = 0;
+			if (inode)
+				ret = (entry->ino == inode->i_ino);
+			break;
 		}
 	}
-	return 1;
+out_rcu:
+	rcu_read_unlock();
+	return ret;
+}
+
+/*
+ * Returns 1 if the passed-in block region (start_blk,
+ * start_blk+count) is valid; 0 if some part of the block region
+ * overlaps with some other filesystem metadata blocks.
+ */
+int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
+			  unsigned int count)
+{
+	return ext4_sb_block_valid(inode->i_sb, inode, start_blk, count);
 }
 
 int ext4_check_blockref(const char *function, unsigned int line,
 			struct inode *inode, __le32 *p, unsigned int max)
 {
-	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 	__le32 *bref = p;
 	unsigned int blk;
+	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
+
+	if (journal && inode == journal->j_inode)
+		return 0;
 
 	while (bref < p+max) {
 		blk = le32_to_cpu(*bref++);
 		if (blk &&
-		    unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
-						    blk, 1))) {
-			es->s_last_error_block = cpu_to_le64(blk);
+		    unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
 			ext4_error_inode(inode, function, line, blk,
 					 "invalid block");
-			return -EIO;
+			return -EFSCORRUPTED;
 		}
 	}
 	return 0;