Btrfs: early extent mapping support

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

5 files changed, 236 insertions, 33 deletions

diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index 63360212a075..df065dd2dce7 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -4,8 +4,8 @@ CFLAGS= -g -Wall
 .c.o:
         $(CC) $(CFLAGS) -c $<
 
-ctree: ctree.o disk-io.h ctree.h disk-io.o radix-tree.o radix-tree.h
-        gcc $(CFLAGS) -o ctree ctree.o disk-io.o radix-tree.o
+ctree: ctree.o disk-io.h ctree.h disk-io.o radix-tree.o radix-tree.h mkfs.o
+        gcc $(CFLAGS) -o ctree ctree.o disk-io.o radix-tree.o mkfs.o
 
 clean:
         rm ctree *.o
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 6b64f49a0279..2177744dedd3 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -5,6 +5,8 @@
 #include "ctree.h"
 #include "disk-io.h"
 
+static int refill_alloc_extent(struct ctree_root *root);
+
 static inline void init_path(struct ctree_path *p)
 {
         memset(p, 0, sizeof(*p));
@@ -29,7 +31,7 @@ static inline unsigned int leaf_data_end(struct leaf *leaf)
 {
         unsigned int nr = leaf->header.nritems;
         if (nr == 0)
-                return ARRAY_SIZE(leaf->data);
+                return sizeof(leaf->data);
         return leaf->items[nr-1].offset;
 }
 
@@ -421,7 +423,7 @@ int insert_ptr(struct ctree_root *root,
          * due to splitting.  Once we've done all the splitting required
          * do the inserts based on the data in the bal array.
          */
-        memset(bal, 0, ARRAY_SIZE(bal));
+        memset(bal, 0, sizeof(bal));
         while(t && t->node.header.nritems == NODEPTRS_PER_BLOCK) {
                 c = &t->node;
                 if (push_node_left(root, path,
@@ -756,6 +758,7 @@ int insert_item(struct ctree_root *root, struct key *key,
         if (leaf_free_space(leaf) < 0)
                 BUG();
         release_path(root, &path);
+        refill_alloc_extent(root);
         return 0;
 }
 
@@ -884,6 +887,135 @@ int del_item(struct ctree_root *root, struct ctree_path *path)
         return 0;
 }
 
+int next_leaf(struct ctree_root *root, struct ctree_path *path)
+{
+        int slot;
+        int level = 1;
+        u64 blocknr;
+        struct tree_buffer *c;
+        struct tree_buffer *next;
+
+        while(level < MAX_LEVEL) {
+                if (!path->nodes[level])
+                        return -1;
+                slot = path->slots[level] + 1;
+                c = path->nodes[level];
+                if (slot >= c->node.header.nritems) {
+                        level++;
+                        continue;
+                }
+                blocknr = c->node.blockptrs[slot];
+                next = read_tree_block(root, blocknr);
+                break;
+        }
+        path->slots[level] = slot;
+        while(1) {
+                level--;
+                c = path->nodes[level];
+                tree_block_release(root, c);
+                path->nodes[level] = next;
+                path->slots[level] = 0;
+                if (!level)
+                        break;
+                next = read_tree_block(root, next->node.blockptrs[0]);
+        }
+        return 0;
+}
+
+int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
+                 u64 search_end, u64 owner, 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 = 0;
+        struct leaf *l;
+        struct extent_item extent_item;
+
+        init_path(&path);
+        ins->objectid = search_start;
+        ins->offset = 0;
+        ins->flags = 0;
+
+        ret = search_slot(root, ins, &path);
+        while (1) {
+                l = &path.nodes[0]->leaf;
+                slot = path.slots[0];
+                if (!l) {
+                        // FIXME allocate root
+                }
+                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;
+                                hole_size = search_end - search_start;
+                                goto insert;
+                        }
+                        ins->objectid = last_block;
+                        ins->offset = num_blocks;
+                        hole_size = search_end - last_block;
+                        goto insert;
+                }
+                key = &l->items[slot].key;
+                if (start_found) {
+                        hole_size = key->objectid - last_block;
+                        if (hole_size > num_blocks) {
+                                ins->objectid = last_block;
+                                ins->offset = num_blocks;
+                                goto insert;
+                        }
+                } else
+                        start_found = 1;
+                last_block = key->objectid + key->offset;
+                path.slots[0]++;
+                printf("last block is not %lu\n", last_block);
+        }
+        // FIXME -ENOSPC
+insert:
+        extent_item.refs = 1;
+        extent_item.owner = owner;
+        ret = insert_item(root, ins, &extent_item, sizeof(extent_item));
+        return ret;
+}
+
+static int refill_alloc_extent(struct ctree_root *root)
+{
+        struct alloc_extent *ae = root->alloc_extent;
+        struct key key;
+        int ret;
+        int min_blocks = MAX_LEVEL * 2;
+
+        printf("refill alloc root %p, numused %lu total %lu\n", root, ae->num_used, ae->num_blocks);
+        if (ae->num_blocks > ae->num_used && ae->num_blocks - ae->num_used >
+            min_blocks)
+                return 0;
+        ae = root->reserve_extent;
+        if (ae->num_blocks > ae->num_used) {
+                if (root->alloc_extent->num_blocks == 0) {
+                        /* we should swap reserve/alloc_extent when alloc
+                         * fills up
+                         */
+                        BUG();
+                }
+                if (ae->num_blocks - ae->num_used < min_blocks)
+                        BUG();
+                return 0;
+        }
+        // FIXME, this recurses
+        ret = alloc_extent(root->extent_root,
+                           min_blocks * 2, 0, (unsigned long)-1, 0, &key);
+        ae->blocknr = key.objectid;
+        ae->num_blocks = key.offset;
+        ae->num_used = 0;
+        return ret;
+}
+
 void print_leaf(struct leaf *l)
 {
         int i;
@@ -948,8 +1080,8 @@ void print_tree(struct ctree_root *root, struct tree_buffer *t)
 
 /* for testing only */
 int next_key(int i, int max_key) {
-        return rand() % max_key;
-        // return i;
+        // return rand() % max_key;
+        return i;
 }
 
 int main() {
@@ -960,7 +1092,7 @@ int main() {
         int i;
         int num;
         int ret;
-        int run_size = 25000;
+        int run_size = 256;
         int max_key = 100000000;
         int tree_size = 0;
         struct ctree_path path;
@@ -980,10 +1112,20 @@ int main() {
                 ins.objectid = num;
                 ins.offset = 0;
                 ins.flags = 0;
+                printf("insert %d\n", i);
                 ret = insert_item(root, &ins, buf, strlen(buf));
                 if (!ret)
                         tree_size++;
+                printf("done insert %d\n", i);
         }
+        printf("root used: %lu\n", root->alloc_extent->num_used);
+        printf("root tree\n");
+        print_tree(root, root->node);
+        printf("map tree\n");
+        printf("map used: %lu\n", root->extent_root->alloc_extent->num_used);
+        print_tree(root->extent_root, root->extent_root->node);
+        exit(1);
+
         close_ctree(root);
         root = open_ctree("dbfile");
         printf("starting search\n");
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 586bf1866042..b737925be314 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -1,7 +1,7 @@
 #ifndef __CTREE__
 #define __CTREE__
 
-#define CTREE_BLOCKSIZE 4096
+#define CTREE_BLOCKSIZE 256
 
 struct key {
         u64 objectid;
@@ -22,18 +22,41 @@ struct header {
 #define NODEPTRS_PER_BLOCK ((CTREE_BLOCKSIZE - sizeof(struct header)) / \
                             (sizeof(struct key) + sizeof(u64)))
 
-#define LEVEL_BITS 3
-#define MAX_LEVEL (1 << LEVEL_BITS)
+#define MAX_LEVEL 8
 #define node_level(f) ((f) & (MAX_LEVEL-1))
 #define is_leaf(f) (node_level(f) == 0)
 
 struct tree_buffer;
+
+struct alloc_extent {
+        u64 blocknr;
+        u64 num_blocks;
+        u64 num_used;
+} __attribute__ ((__packed__));
+
 struct ctree_root {
         struct tree_buffer *node;
+        struct ctree_root *extent_root;
+        struct alloc_extent *alloc_extent;
+        struct alloc_extent *reserve_extent;
         int fp;
         struct radix_tree_root cache_radix;
+        struct alloc_extent ai1;
+        struct alloc_extent ai2;
 };
 
+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 */
+        u32 csum;
+        u32 ham;
+        struct alloc_extent alloc_extent;
+        struct alloc_extent reserve_extent;
+        u64 snapuuid[2]; /* root specific uuid */
+} __attribute__ ((__packed__));
+
 struct item {
         struct key key;
         u16 offset;
@@ -55,6 +78,11 @@ struct node {
         u64 blockptrs[NODEPTRS_PER_BLOCK];
 } __attribute__ ((__packed__));
 
+struct extent_item {
+        u32 refs;
+        u64 owner;
+} __attribute__ ((__packed__));
+
 struct ctree_path {
         struct tree_buffer *nodes[MAX_LEVEL];
         int slots[MAX_LEVEL];
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 8d51a07051d9..653f18aab330 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -12,25 +12,27 @@
 
 static int allocated_blocks = 0;
 
-struct ctree_header {
-        u64 root_block;
-} __attribute__ ((__packed__));
-
 static int get_free_block(struct ctree_root *root, u64 *block)
 {
         struct stat st;
         int ret;
 
+        if (root->alloc_extent->num_used >= root->alloc_extent->num_blocks)
+                return -1;
+
+        *block = root->alloc_extent->blocknr + root->alloc_extent->num_used;
+        root->alloc_extent->num_used += 1;
+        if (root->alloc_extent->num_used >= root->alloc_extent->num_blocks) {
+                struct alloc_extent *ae = root->alloc_extent;
+                root->alloc_extent = root->reserve_extent;
+                root->reserve_extent = ae;
+                ae->num_blocks = 0;
+        }
         st.st_size = 0;
         ret = fstat(root->fp, &st);
-        if (st.st_size > sizeof(struct ctree_header)) {
-                *block = (st.st_size -
-                        sizeof(struct ctree_header)) / CTREE_BLOCKSIZE;
-        } else {
-                *block = 0;
-        }
-        ret = ftruncate(root->fp, sizeof(struct ctree_header) + (*block + 1) *
-                        CTREE_BLOCKSIZE);
+        if (st.st_size < (*block + 1) * CTREE_BLOCKSIZE)
+                ret = ftruncate(root->fp,
+                                (*block + 1) * CTREE_BLOCKSIZE);
         return ret;
 }
 
@@ -72,7 +74,7 @@ struct tree_buffer *alloc_free_block(struct ctree_root *root)
 
 struct tree_buffer *read_tree_block(struct ctree_root *root, u64 blocknr)
 {
-        loff_t offset = blocknr * CTREE_BLOCKSIZE + sizeof(struct ctree_header);
+        loff_t offset = blocknr * CTREE_BLOCKSIZE;
         struct tree_buffer *buf;
         int ret;
 
@@ -101,7 +103,7 @@ struct tree_buffer *read_tree_block(struct ctree_root *root, u64 blocknr)
 int write_tree_block(struct ctree_root *root, struct tree_buffer *buf)
 {
         u64 blocknr = buf->blocknr;
-        loff_t offset = blocknr * CTREE_BLOCKSIZE + sizeof(struct ctree_header);
+        loff_t offset = blocknr * CTREE_BLOCKSIZE;
         int ret;
 
         if (buf->blocknr != buf->node.header.blocknr)
@@ -114,11 +116,32 @@ int write_tree_block(struct ctree_root *root, struct tree_buffer *buf)
         return 0;
 }
 
+struct ctree_super_block {
+        struct ctree_root_info root_info;
+        struct ctree_root_info extent_info;
+} __attribute__ ((__packed__));
+
+static int __setup_root(struct ctree_root *root, struct ctree_root *extent_root,
+                        struct ctree_root_info *info, int fp)
+{
+        root->fp = fp;
+        root->node = read_tree_block(root, info->tree_root);
+        root->extent_root = extent_root;
+        memcpy(&root->ai1, &info->alloc_extent, sizeof(info->alloc_extent));
+        memcpy(&root->ai2, &info->reserve_extent, sizeof(info->reserve_extent));
+        root->alloc_extent = &root->ai1;
+        root->reserve_extent = &root->ai2;
+        INIT_RADIX_TREE(&root->cache_radix, GFP_KERNEL);
+        printf("setup done reading root %p, used %lu\n", root, root->alloc_extent->num_used);
+        return 0;
+}
+
 struct ctree_root *open_ctree(char *filename)
 {
         struct ctree_root *root = malloc(sizeof(struct ctree_root));
+        struct ctree_root *extent_root = malloc(sizeof(struct ctree_root));
+        struct ctree_super_block super;
         int fp;
-        u64 root_block;
         int ret;
 
         fp = open(filename, O_CREAT | O_RDWR);
@@ -126,14 +149,20 @@ struct ctree_root *open_ctree(char *filename)
                 free(root);
                 return NULL;
         }
-        root->fp = fp;
-        INIT_RADIX_TREE(&root->cache_radix, GFP_KERNEL);
-        ret = pread(fp, &root_block, sizeof(u64), 0);
-        if (ret == sizeof(u64)) {
-                printf("reading root node at block %lu\n", root_block);
-                root->node = read_tree_block(root, root_block);
-        } else
-                root->node = NULL;
+        ret = pread(fp, &super, sizeof(struct ctree_super_block),
+                     CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
+        if (ret == 0) {
+                ret = mkfs(fp);
+                if (ret)
+                        return NULL;
+                ret = pread(fp, &super, sizeof(struct ctree_super_block),
+                             CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
+                if (ret != sizeof(struct ctree_super_block))
+                        return NULL;
+        }
+        BUG_ON(ret < 0);
+        __setup_root(root, extent_root, &super.root_info, fp);
+        __setup_root(extent_root, extent_root, &super.extent_info, fp);
         return root;
 }
 
@@ -160,6 +189,7 @@ int update_root_block(struct ctree_root *root)
 
 void tree_block_release(struct ctree_root *root, struct tree_buffer *buf)
 {
+        return;
         buf->count--;
         if (buf->count == 0) {
                 if (!radix_tree_lookup(&root->cache_radix, buf->blocknr))
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index ee95fa05c4a3..41acb65eee19 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -17,5 +17,8 @@ int close_ctree(struct ctree_root *root);
 void tree_block_release(struct ctree_root *root, struct tree_buffer *buf);
 struct tree_buffer *alloc_free_block(struct ctree_root *root);
 int update_root_block(struct ctree_root *root);
+int mkfs(int fd);
+
+#define CTREE_SUPER_INFO_OFFSET(bs) (16 * (bs))
 
 #endif