Btrfs: Add fsx-style randomized tree tester

Add debug-tree command to print the tree
Add extent-tree.c to the repo
Comment ctree.h

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/Makefile

@@ -6,11 +6,17 @@ objects = ctree.o disk-io.o radix-tree.o mkfs.o extent-tree.o print-tree.o
 #.c.o:
 #	$(CC) $(CFLAGS) -c $<
 
-ctree : $(objects)
-	gcc $(CFLAGS) -o ctree $(objects)
+all: tester debug-tree
+
+debug-tree: $(objects) debug-tree.o
+	gcc $(CFLAGS) -o debug-tree $(objects) debug-tree.o
+
+tester: $(objects) random-test.o
+	gcc $(CFLAGS) -o tester $(objects) random-test.o
 
 $(objects) : $(headers)
 
 clean :
 	rm ctree *.o
 
+

fs/btrfs/ctree.c

@@ -1047,14 +1047,14 @@ int next_leaf(struct ctree_root *root, struct ctree_path *path)
 	return 0;
 }
 
+/* some sample code to insert,search & delete items */
+#if 0
 /* for testing only */
 int next_key(int i, int max_key) {
 	return rand() % max_key;
 	//return i;
 }
-
 int main() {
-	struct ctree_root *root;
 	struct key ins;
 	struct key last = { (u64)-1, 0, 0};
 	char *buf;
@@ -1066,6 +1066,7 @@ int main() {
 	int tree_size = 0;
 	struct ctree_path path;
 	struct ctree_super_block super;
+	struct ctree_root *root;
 
 	radix_tree_init();
 
@@ -1207,3 +1208,4 @@ int main() {
 	close_ctree(root);
 	return 0;
 }
+#endif

fs/btrfs/ctree.h

@@ -1,22 +1,36 @@
 #ifndef __CTREE__
 #define __CTREE__
 
-#define CTREE_BLOCKSIZE 4096
+#define CTREE_BLOCKSIZE 1024
 
+/*
+ * the key defines the order in the tree, and so it also defines (optimal)
+ * block layout.  objectid corresonds to the inode number.  The flags
+ * tells us things about the object, and is a kind of stream selector.
+ * so for a given inode, keys with flags of 1 might refer to the inode
+ * data, flags of 2 may point to file data in the btree and flags == 3
+ * may point to extents.
+ *
+ * offset is the starting byte offset for this key in the stream.
+ */
 struct key {
 	u64 objectid;
 	u32 flags;
 	u64 offset;
 } __attribute__ ((__packed__));
 
+/*
+ * every tree block (leaf or node) starts with this header.
+ */
 struct header {
 	u64 fsid[2]; /* FS specific uuid */
-	u64 blocknr;
-	u64 parentid;
+	u64 blocknr; /* which block this node is supposed to live in */
+	u64 parentid; /* objectid of the tree root */
 	u32 csum;
 	u32 ham;
 	u16 nritems;
 	u16 flags;
+	/* generation flags to be added */
 } __attribute__ ((__packed__));
 
 #define NODEPTRS_PER_BLOCK ((CTREE_BLOCKSIZE - sizeof(struct header)) / \
@@ -28,6 +42,11 @@ struct header {
 
 struct tree_buffer;
 
+/*
+ * in ram representation of the tree.  extent_root is used for all allocations
+ * and for the extent tree extent_root root.  current_insert is used
+ * only for the extent tree.
+ */
 struct ctree_root {
 	struct tree_buffer *node;
 	struct ctree_root *extent_root;
@@ -36,27 +55,46 @@ struct ctree_root {
 	struct radix_tree_root cache_radix;
 };
 
+/*
+ * describes a tree on disk
+ */
 struct ctree_root_info {
 	u64 fsid[2]; /* FS specific uuid */
 	u64 blocknr; /* blocknr of this block */
 	u64 objectid; /* inode number of this root */
-	u64 tree_root; /* the tree root */
+	u64 tree_root; /* the tree root block */
 	u32 csum;
 	u32 ham;
 	u64 snapuuid[2]; /* root specific uuid */
 } __attribute__ ((__packed__));
 
+/*
+ * the super block basically lists the main trees of the FS
+ * it currently lacks any block count etc etc
+ */
 struct ctree_super_block {
 	struct ctree_root_info root_info;
 	struct ctree_root_info extent_info;
 } __attribute__ ((__packed__));
 
+/*
+ * A leaf is full of items.  The exact type of item is defined by
+ * the key flags parameter.  offset and size tell us where to find
+ * the item in the leaf (relative to the start of the data area)
+ */
 struct item {
 	struct key key;
 	u16 offset;
 	u16 size;
 } __attribute__ ((__packed__));
 
+/*
+ * leaves have an item area and a data area:
+ * [item0, item1....itemN] [free space] [dataN...data1, data0]
+ *
+ * The data is separate from the items to get the keys closer together
+ * during searches.
+ */
 #define LEAF_DATA_SIZE (CTREE_BLOCKSIZE - sizeof(struct header))
 struct leaf {
 	struct header header;
@@ -66,17 +104,33 @@ struct leaf {
 	};
 } __attribute__ ((__packed__));
 
+/*
+ * all non-leaf blocks are nodes, they hold only keys and pointers to
+ * other blocks
+ */
 struct node {
 	struct header header;
 	struct key keys[NODEPTRS_PER_BLOCK];
 	u64 blockptrs[NODEPTRS_PER_BLOCK];
 } __attribute__ ((__packed__));
 
+/*
+ * items in the extent btree are used to record the objectid of the
+ * owner of the block and the number of references
+ */
 struct extent_item {
 	u32 refs;
 	u64 owner;
 } __attribute__ ((__packed__));
 
+/*
+ * ctree_paths remember the path taken from the root down to the leaf.
+ * level 0 is always the leaf, and nodes[1...MAX_LEVEL] will point
+ * to any other levels that are present.
+ *
+ * The slots array records the index of the item or block pointer
+ * used while walking the tree.
+ */
 struct ctree_path {
 	struct tree_buffer *nodes[MAX_LEVEL];
 	int slots[MAX_LEVEL];

fs/btrfs/debug-tree.c

@@ -0,0 +1,19 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include "kerncompat.h"
+#include "radix-tree.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+
+int main() {
+	struct ctree_super_block super;
+	struct ctree_root *root;
+	radix_tree_init();
+	root = open_ctree("dbfile", &super);
+	printf("root tree\n");
+	print_tree(root, root->node);
+	printf("map tree\n");
+	print_tree(root->extent_root, root->extent_root->node);
+	return 0;
+}

fs/btrfs/extent-tree.c

@@ -0,0 +1,296 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include "kerncompat.h"
+#include "radix-tree.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+
+/*
+ * pending extents are blocks that we're trying to allocate in the extent
+ * map while trying to grow the map because of other allocations.  To avoid
+ * recursing, they are tagged in the radix tree and cleaned up after
+ * other allocations are done.  The pending tag is also used in the same
+ * manner for deletes.
+ */
+#define CTREE_EXTENT_PENDING 0
+
+/*
+ * find all the blocks marked as pending in the radix tree and remove
+ * them from the extent map
+ */
+static int del_pending_extents(struct ctree_root *extent_root)
+{
+	int ret;
+	struct key key;
+	struct tree_buffer *gang[4];
+	int i;
+	struct ctree_path path;
+
+	while(1) {
+		ret = radix_tree_gang_lookup_tag(&extent_root->cache_radix,
+						 (void **)gang, 0,
+						 ARRAY_SIZE(gang),
+						 CTREE_EXTENT_PENDING);
+		if (!ret)
+			break;
+		for (i = 0; i < ret; i++) {
+			key.objectid = gang[i]->blocknr;
+			key.flags = 0;
+			key.offset = 1;
+			init_path(&path);
+			ret = search_slot(extent_root, &key, &path, 0);
+			if (ret) {
+				print_tree(extent_root, extent_root->node);
+				printf("unable to find %lu\n", key.objectid);
+				BUG();
+				// FIXME undo it and return sane
+				return ret;
+			}
+			ret = del_item(extent_root, &path);
+			if (ret) {
+				BUG();
+				return ret;
+			}
+			release_path(extent_root, &path);
+			radix_tree_tag_clear(&extent_root->cache_radix,
+						gang[i]->blocknr,
+						CTREE_EXTENT_PENDING);
+			tree_block_release(extent_root, gang[i]);
+		}
+	}
+	return 0;
+}
+
+/*
+ * remove an extent from the root, returns 0 on success
+ */
+int free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
+{
+	struct ctree_path path;
+	struct key key;
+	struct ctree_root *extent_root = root->extent_root;
+	struct tree_buffer *t;
+	int pending_ret;
+	int ret;
+	key.objectid = blocknr;
+	key.flags = 0;
+	key.offset = num_blocks;
+	if (root == extent_root) {
+		t = read_tree_block(root, key.objectid);
+		radix_tree_tag_set(&root->cache_radix, key.objectid,
+				   CTREE_EXTENT_PENDING);
+		return 0;
+	}
+	init_path(&path);
+	ret = search_slot(extent_root, &key, &path, 0);
+	if (ret) {
+		print_tree(extent_root, extent_root->node);
+		printf("failed to find %lu\n", key.objectid);
+		BUG();
+	}
+	ret = del_item(extent_root, &path);
+	if (ret)
+		BUG();
+	release_path(extent_root, &path);
+	pending_ret = del_pending_extents(root->extent_root);
+	return ret ? ret : pending_ret;
+}
+
+/*
+ * walks the btree of allocated extents and find a hole of a given size.
+ * The key ins is changed to record the hole:
+ * ins->objectid == block start
+ * ins->flags = 0
+ * ins->offset == number of blocks
+ * Any available blocks before search_start are skipped.
+ */
+int find_free_extent(struct ctree_root *orig_root, u64 num_blocks,
+		     u64 search_start, u64 search_end, struct key *ins)
+{
+	struct ctree_path path;
+	struct key *key;
+	int ret;
+	u64 hole_size = 0;
+	int slot = 0;
+	u64 last_block;
+	int start_found;
+	struct leaf *l;
+	struct ctree_root * root = orig_root->extent_root;
+
+check_failed:
+	init_path(&path);
+	ins->objectid = search_start;
+	ins->offset = 0;
+	ins->flags = 0;
+	start_found = 0;
+	ret = search_slot(root, ins, &path, 0);
+	while (1) {
+		l = &path.nodes[0]->leaf;
+		slot = path.slots[0];
+		if (slot >= l->header.nritems) {
+			ret = next_leaf(root, &path);
+			if (ret == 0)
+				continue;
+			if (!start_found) {
+				ins->objectid = search_start;
+				ins->offset = num_blocks;
+				start_found = 1;
+				goto check_pending;
+			}
+			ins->objectid = last_block > search_start ?
+					last_block : search_start;
+			ins->offset = num_blocks;
+			goto check_pending;
+		}
+		key = &l->items[slot].key;
+		if (key->objectid >= search_start) {
+			if (start_found) {
+				hole_size = key->objectid - last_block;
+				if (hole_size > num_blocks) {
+					ins->objectid = last_block;
+					ins->offset = num_blocks;
+					goto check_pending;
+				}
+			} else
+				start_found = 1;
+			last_block = key->objectid + key->offset;
+		}
+		path.slots[0]++;
+	}
+	// FIXME -ENOSPC
+check_pending:
+	/* we have to make sure we didn't find an extent that has already
+	 * been allocated by the map tree or the original allocation
+	 */
+	release_path(root, &path);
+	BUG_ON(ins->objectid < search_start);
+	if (orig_root->extent_root == orig_root) {
+		BUG_ON(num_blocks != 1);
+		if ((root->current_insert.objectid <= ins->objectid &&
+		    root->current_insert.objectid +
+		    root->current_insert.offset > ins->objectid) ||
+		   (root->current_insert.objectid > ins->objectid &&
+		    root->current_insert.objectid <= ins->objectid +
+		    ins->offset) ||
+		   radix_tree_tag_get(&root->cache_radix, ins->objectid,
+				      CTREE_EXTENT_PENDING)) {
+			search_start = ins->objectid + 1;
+			goto check_failed;
+		}
+	}
+	if (ins->offset != 1)
+		BUG();
+	return 0;
+}
+
+/*
+ * insert all of the pending extents reserved during the original
+ * allocation.  (CTREE_EXTENT_PENDING).  Returns zero if it all worked out
+ */
+static int insert_pending_extents(struct ctree_root *extent_root)
+{
+	int ret;
+	struct key key;
+	struct extent_item item;
+	struct tree_buffer *gang[4];
+	int i;
+
+	// FIXME -ENOSPC
+	item.refs = 1;
+	item.owner = extent_root->node->node.header.parentid;
+	while(1) {
+		ret = radix_tree_gang_lookup_tag(&extent_root->cache_radix,
+						 (void **)gang, 0,
+						 ARRAY_SIZE(gang),
+						 CTREE_EXTENT_PENDING);
+		if (!ret)
+			break;
+		for (i = 0; i < ret; i++) {
+			key.objectid = gang[i]->blocknr;
+			key.flags = 0;
+			key.offset = 1;
+			ret = insert_item(extent_root, &key, &item,
+					  sizeof(item));
+			if (ret) {
+				BUG();
+				// FIXME undo it and return sane
+				return ret;
+			}
+			radix_tree_tag_clear(&extent_root->cache_radix,
+					     gang[i]->blocknr,
+					     CTREE_EXTENT_PENDING);
+			tree_block_release(extent_root, gang[i]);
+		}
+	}
+	return 0;
+}
+
+/*
+ * finds a free extent and does all the dirty work required for allocation
+ * returns the key for the extent through ins, and a tree buffer for
+ * the first block of the extent through buf.
+ *
+ * returns 0 if everything worked, non-zero otherwise.
+ */
+int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
+			 u64 search_end, u64 owner, struct key *ins,
+			 struct tree_buffer **buf)
+{
+	int ret;
+	int pending_ret;
+	struct extent_item extent_item;
+	extent_item.refs = 1;
+	extent_item.owner = owner;
+
+	ret = find_free_extent(root, num_blocks, search_start, search_end, ins);
+	if (ret)
+		return ret;
+	if (root != root->extent_root) {
+		memcpy(&root->extent_root->current_insert, ins, sizeof(*ins));
+		ret = insert_item(root->extent_root, ins, &extent_item,
+				  sizeof(extent_item));
+		memset(&root->extent_root->current_insert, 0,
+		       sizeof(struct key));
+		pending_ret = insert_pending_extents(root->extent_root);
+		if (ret)
+			return ret;
+		if (pending_ret)
+			return pending_ret;
+		*buf = find_tree_block(root, ins->objectid);
+		return 0;
+	}
+	/* we're allocating an extent for the extent tree, don't recurse */
+	BUG_ON(ins->offset != 1);
+	*buf = find_tree_block(root, ins->objectid);
+	BUG_ON(!*buf);
+	radix_tree_tag_set(&root->cache_radix, ins->objectid,
+			   CTREE_EXTENT_PENDING);
+	(*buf)->count++;
+	return 0;
+
+}
+
+/*
+ * helper function to allocate a block for a given tree
+ * returns the tree buffer or NULL.
+ */
+struct tree_buffer *alloc_free_block(struct ctree_root *root)
+{
+	struct key ins;
+	int ret;
+	struct tree_buffer *buf = NULL;
+
+	ret = alloc_extent(root, 1, 0, (unsigned long)-1,
+			   root->node->node.header.parentid,
+			   &ins, &buf);
+
+	if (ret) {
+		BUG();
+		return NULL;
+	}
+	if (root != root->extent_root)
+		BUG_ON(radix_tree_tag_get(&root->extent_root->cache_radix,
+					  buf->blocknr, CTREE_EXTENT_PENDING));
+	return buf;
+}

fs/btrfs/random-test.c

@@ -0,0 +1,295 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <signal.h>
+#include "kerncompat.h"
+#include "radix-tree.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+
+int keep_running = 1;
+
+static int setup_key(struct radix_tree_root *root, struct key *key, int exists)
+{
+	int num = rand();
+	unsigned long res[2];
+	int ret;
+
+	key->flags = 0;
+	key->offset = 0;
+again:
+	ret = radix_tree_gang_lookup(root, (void **)res, num, 2);
+	if (exists) {
+		if (ret == 0)
+			return -1;
+		num = res[0];
+	} else if (ret != 0 && num == res[0]) {
+		num++;
+		if (ret > 1 && num == res[1]) {
+			num++;
+			goto again;
+		}
+	}
+	key->objectid = num;
+	return 0;
+}
+
+static int ins_one(struct ctree_root *root, struct radix_tree_root *radix)
+{
+	struct ctree_path path;
+	struct key key;
+	int ret;
+	char buf[128];
+	init_path(&path);
+	ret = setup_key(radix, &key, 0);
+	sprintf(buf, "str-%lu\n", key.objectid);
+	ret = insert_item(root, &key, buf, strlen(buf));
+	if (ret)
+		goto error;
+	radix_tree_preload(GFP_KERNEL);
+	ret = radix_tree_insert(radix, key.objectid,
+					(void *)key.objectid);
+	radix_tree_preload_end();
+	if (ret)
+		goto error;
+	return ret;
+error:
+	printf("failed to insert %lu\n", key.objectid);
+	return -1;
+}
+
+static int insert_dup(struct ctree_root *root, struct radix_tree_root *radix)
+{
+	struct ctree_path path;
+	struct key key;
+	int ret;
+	char buf[128];
+	init_path(&path);
+	ret = setup_key(radix, &key, 1);
+	if (ret < 0)
+		return 0;
+	sprintf(buf, "str-%lu\n", key.objectid);
+	ret = insert_item(root, &key, buf, strlen(buf));
+	if (ret != -EEXIST) {
+		printf("insert on %lu gave us %d\n", key.objectid, ret);
+		return 1;
+	}
+	return 0;
+}
+
+static int del_one(struct ctree_root *root, struct radix_tree_root *radix)
+{
+	struct ctree_path path;
+	struct key key;
+	int ret;
+	unsigned long *ptr;
+	init_path(&path);
+	ret = setup_key(radix, &key, 1);
+	if (ret < 0)
+		return 0;
+	ret = search_slot(root, &key, &path, -1);
+	if (ret)
+		goto error;
+	ret = del_item(root, &path);
+	release_path(root, &path);
+	if (ret != 0)
+		goto error;
+	ptr = radix_tree_delete(radix, key.objectid);
+	if (!ptr)
+		goto error;
+	return 0;
+error:
+	printf("failed to delete %lu\n", key.objectid);
+	return -1;
+}
+
+static int lookup_item(struct ctree_root *root, struct radix_tree_root *radix)
+{
+	struct ctree_path path;
+	struct key key;
+	int ret;
+	init_path(&path);
+	ret = setup_key(radix, &key, 1);
+	if (ret < 0)
+		return 0;
+	ret = search_slot(root, &key, &path, 0);
+	release_path(root, &path);
+	if (ret)
+		goto error;
+	return 0;
+error:
+	printf("unable to find key %lu\n", key.objectid);
+	return -1;
+}
+
+static int lookup_enoent(struct ctree_root *root, struct radix_tree_root *radix)
+{
+	struct ctree_path path;
+	struct key key;
+	int ret;
+	init_path(&path);
+	ret = setup_key(radix, &key, 0);
+	if (ret < 0)
+		return ret;
+	ret = search_slot(root, &key, &path, 0);
+	release_path(root, &path);
+	if (ret == 0)
+		goto error;
+	return 0;
+error:
+	printf("able to find key that should not exist %lu\n", key.objectid);
+	return -1;
+}
+
+int (*ops[])(struct ctree_root *root, struct radix_tree_root *radix) =
+{ ins_one, insert_dup, del_one, lookup_item, lookup_enoent };
+
+static int fill_radix(struct ctree_root *root, struct radix_tree_root *radix)
+{
+	struct ctree_path path;
+	struct key key;
+	u64 found;
+	int ret;
+	int slot;
+	int i;
+	key.offset = 0;
+	key.flags = 0;
+	key.objectid = (unsigned long)-1;
+	while(1) {
+		init_path(&path);
+		ret = search_slot(root, &key, &path, 0);
+		slot = path.slots[0];
+		if (ret != 0) {
+			if (slot == 0) {
+				release_path(root, &path);
+				break;
+			}
+			slot -= 1;
+		}
+		for (i = slot; i >= 0; i--) {
+			found = path.nodes[0]->leaf.items[i].key.objectid;
+			radix_tree_preload(GFP_KERNEL);
+			ret = radix_tree_insert(radix, found, (void *)found);
+			if (ret) {
+				fprintf(stderr,
+					"failed to insert %lu into radix\n",
+					found);
+				exit(1);
+			}
+
+			radix_tree_preload_end();
+		}
+		release_path(root, &path);
+		key.objectid = found - 1;
+		if (key.objectid > found)
+			break;
+	}
+	return 0;
+}
+
+void sigstopper(int ignored)
+{
+	keep_running = 0;
+	fprintf(stderr, "caught exit signal, stopping\n");
+}
+
+int print_usage(void)
+{
+	printf("usage: tester [-ih] [-c count] [-f count]\n");
+	printf("\t -c count -- iteration count after filling\n");
+	printf("\t -f count -- run this many random inserts before starting\n");
+	printf("\t -i       -- only do initial fill\n");
+	printf("\t -h       -- this help text\n");
+	exit(1);
+}
+int main(int ac, char **av)
+{
+	RADIX_TREE(radix, GFP_KERNEL);
+	struct ctree_super_block super;
+	struct ctree_root *root;
+	int i;
+	int ret;
+	int count;
+	int op;
+	int iterations = 20000;
+	int init_fill_count = 800000;
+	int err = 0;
+	int initial_only = 0;
+	radix_tree_init();
+	root = open_ctree("dbfile", &super);
+	fill_radix(root, &radix);
+
+	signal(SIGTERM, sigstopper);
+	signal(SIGINT, sigstopper);
+
+	for (i = 1 ; i < ac ; i++) {
+		if (strcmp(av[i], "-i") == 0) {
+			initial_only = 1;
+		} else if (strcmp(av[i], "-c") == 0) {
+			iterations = atoi(av[i+1]);
+			i++;
+		} else if (strcmp(av[i], "-f") == 0) {
+			init_fill_count = atoi(av[i+1]);
+			i++;
+		} else {
+			print_usage();
+		}
+	}
+	for (i = 0; i < init_fill_count; i++) {
+		ret = ins_one(root, &radix);
+		if (ret) {
+			printf("initial fill failed\n");
+			err = ret;
+			goto out;
+		}
+		if (i % 10000 == 0) {
+			printf("initial fill %d level %d count %d\n", i,
+				node_level(root->node->node.header.flags),
+				root->node->node.header.nritems);
+		}
+		if (keep_running == 0) {
+			err = 0;
+			goto out;
+		}
+	}
+	if (initial_only == 1) {
+		goto out;
+	}
+	for (i = 0; i < iterations; i++) {
+		op = rand() % ARRAY_SIZE(ops);
+		count = rand() % 128;
+		if (i % 2000 == 0) {
+			printf("%d\n", i);
+			fflush(stdout);
+		}
+		if (i && i % 5000 == 0) {
+			printf("open & close, root level %d nritems %d\n",
+				node_level(root->node->node.header.flags),
+				root->node->node.header.nritems);
+			write_ctree_super(root, &super);
+			close_ctree(root);
+			root = open_ctree("dbfile", &super);
+		}
+		while(count--) {
+			ret = ops[op](root, &radix);
+			if (ret) {
+				fprintf(stderr, "op %d failed %d:%d\n",
+					op, i, iterations);
+				print_tree(root, root->node);
+				fprintf(stderr, "op %d failed %d:%d\n",
+					op, i, iterations);
+				err = ret;
+				goto out;
+			}
+			if (keep_running == 0) {
+				err = 0;
+				goto out;
+			}
+		}
+	}
+out:
+	write_ctree_super(root, &super);
+	close_ctree(root);
+	return err;
+}
+