#include <stdio.h>
#include <stdlib.h>
#include "kerncompat.h"
+#include "radix-tree.h"
+#include "ctree.h"
+#include "disk-io.h"
+#include "print-tree.h"
+
+static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int level);
+static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int data_size);
+static int push_node_left(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_buffer *dst, struct btrfs_buffer
+ *src);
+static int balance_node_right(struct btrfs_trans_handle *trans, struct
+ btrfs_root *root, struct btrfs_buffer *dst_buf,
+ struct btrfs_buffer *src_buf);
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+ struct btrfs_path *path, int level, int slot);
+
+inline void btrfs_init_path(struct btrfs_path *p)
+{
+ memset(p, 0, sizeof(*p));
+}
-#define BLOCKSIZE 4096
-
-struct key {
- u64 objectid;
- u32 flags;
- u64 offset;
-} __attribute__ ((__packed__));
-
-struct header {
- u64 fsid[2]; /* FS specific uuid */
- u64 blocknum;
- u64 parentid;
- u32 csum;
- u32 ham;
- u16 nritems;
- u16 flags;
-} __attribute__ ((__packed__));
-
-#define NODEPTRS_PER_BLOCK ((BLOCKSIZE - sizeof(struct header)) / \
- (sizeof(struct key) + sizeof(u64)))
-
-#define LEVEL_BITS 3
-#define MAX_LEVEL (1 << LEVEL_BITS)
-#define node_level(f) ((f) & (MAX_LEVEL-1))
-#define is_leaf(f) (node_level(f) == 0)
-
-struct ctree_root {
- struct node *node;
-};
-
-struct item {
- struct key key;
- u16 offset;
- u16 size;
-} __attribute__ ((__packed__));
-
-#define LEAF_DATA_SIZE (BLOCKSIZE - sizeof(struct header))
-struct leaf {
- struct header header;
- union {
- struct item items[LEAF_DATA_SIZE/sizeof(struct item)];
- u8 data[BLOCKSIZE-sizeof(struct header)];
- };
-} __attribute__ ((__packed__));
-
-struct node {
- struct header header;
- struct key keys[NODEPTRS_PER_BLOCK];
- u64 blockptrs[NODEPTRS_PER_BLOCK];
-} __attribute__ ((__packed__));
-
-struct ctree_path {
- struct node *nodes[MAX_LEVEL];
- int slots[MAX_LEVEL];
-};
-
-static inline void init_path(struct ctree_path *p)
+void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
{
+ int i;
+ for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+ if (!p->nodes[i])
+ break;
+ btrfs_block_release(root, p->nodes[i]);
+ }
memset(p, 0, sizeof(*p));
}
-static inline unsigned int leaf_data_end(struct leaf *leaf)
+static int btrfs_cow_block(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_buffer *buf, struct btrfs_buffer
+ *parent, int parent_slot, struct btrfs_buffer
+ **cow_ret)
{
- unsigned int nr = leaf->header.nritems;
+ struct btrfs_buffer *cow;
+
+ if (!list_empty(&buf->dirty)) {
+ *cow_ret = buf;
+ return 0;
+ }
+ cow = btrfs_alloc_free_block(trans, root);
+ memcpy(&cow->node, &buf->node, root->blocksize);
+ btrfs_set_header_blocknr(&cow->node.header, cow->blocknr);
+ *cow_ret = cow;
+ btrfs_inc_ref(trans, root, buf);
+ if (buf == root->node) {
+ root->node = cow;
+ cow->count++;
+ if (buf != root->commit_root)
+ btrfs_free_extent(trans, root, buf->blocknr, 1, 1);
+ btrfs_block_release(root, buf);
+ } else {
+ btrfs_set_node_blockptr(&parent->node, parent_slot,
+ cow->blocknr);
+ BUG_ON(list_empty(&parent->dirty));
+ btrfs_free_extent(trans, root, buf->blocknr, 1, 1);
+ }
+ btrfs_block_release(root, buf);
+ return 0;
+}
+
+/*
+ * The leaf data grows from end-to-front in the node.
+ * this returns the address of the start of the last item,
+ * which is the stop of the leaf data stack
+ */
+static inline unsigned int leaf_data_end(struct btrfs_root *root,
+ struct btrfs_leaf *leaf)
+{
+ u32 nr = btrfs_header_nritems(&leaf->header);
if (nr == 0)
- return ARRAY_SIZE(leaf->data);
- return leaf->items[nr-1].offset;
+ return BTRFS_LEAF_DATA_SIZE(root);
+ return btrfs_item_offset(leaf->items + nr - 1);
}
-static inline int leaf_free_space(struct leaf *leaf)
+/*
+ * The space between the end of the leaf items and
+ * the start of the leaf data. IOW, how much room
+ * the leaf has left for both items and data
+ */
+int btrfs_leaf_free_space(struct btrfs_root *root, struct btrfs_leaf *leaf)
{
- int data_end = leaf_data_end(leaf);
- int nritems = leaf->header.nritems;
+ int data_end = leaf_data_end(root, leaf);
+ int nritems = btrfs_header_nritems(&leaf->header);
char *items_end = (char *)(leaf->items + nritems + 1);
- return (char *)(leaf->data + data_end) - (char *)items_end;
+ return (char *)(btrfs_leaf_data(leaf) + data_end) - (char *)items_end;
}
-int comp_keys(struct key *k1, struct key *k2)
+/*
+ * compare two keys in a memcmp fashion
+ */
+static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
{
- if (k1->objectid > k2->objectid)
+ struct btrfs_key k1;
+
+ btrfs_disk_key_to_cpu(&k1, disk);
+
+ if (k1.objectid > k2->objectid)
return 1;
- if (k1->objectid < k2->objectid)
+ if (k1.objectid < k2->objectid)
return -1;
- if (k1->flags > k2->flags)
+ if (k1.flags > k2->flags)
return 1;
- if (k1->flags < k2->flags)
+ if (k1.flags < k2->flags)
return -1;
- if (k1->offset > k2->offset)
+ if (k1.offset > k2->offset)
return 1;
- if (k1->offset < k2->offset)
+ if (k1.offset < k2->offset)
return -1;
return 0;
}
-int generic_bin_search(char *p, int item_size, struct key *key,
+
+static int check_node(struct btrfs_root *root, struct btrfs_path *path,
+ int level)
+{
+ int i;
+ struct btrfs_node *parent = NULL;
+ struct btrfs_node *node = &path->nodes[level]->node;
+ int parent_slot;
+ u32 nritems = btrfs_header_nritems(&node->header);
+
+ if (path->nodes[level + 1])
+ parent = &path->nodes[level + 1]->node;
+ parent_slot = path->slots[level + 1];
+ BUG_ON(nritems == 0);
+ if (parent) {
+ struct btrfs_disk_key *parent_key;
+ parent_key = &parent->ptrs[parent_slot].key;
+ BUG_ON(memcmp(parent_key, &node->ptrs[0].key,
+ sizeof(struct btrfs_disk_key)));
+ BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
+ btrfs_header_blocknr(&node->header));
+ }
+ BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
+ for (i = 0; nritems > 1 && i < nritems - 2; i++) {
+ struct btrfs_key cpukey;
+ btrfs_disk_key_to_cpu(&cpukey, &node->ptrs[i + 1].key);
+ BUG_ON(comp_keys(&node->ptrs[i].key, &cpukey) >= 0);
+ }
+ return 0;
+}
+
+static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
+ int level)
+{
+ int i;
+ struct btrfs_leaf *leaf = &path->nodes[level]->leaf;
+ struct btrfs_node *parent = NULL;
+ int parent_slot;
+ u32 nritems = btrfs_header_nritems(&leaf->header);
+
+ if (path->nodes[level + 1])
+ parent = &path->nodes[level + 1]->node;
+ parent_slot = path->slots[level + 1];
+ BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
+
+ if (nritems == 0)
+ return 0;
+
+ if (parent) {
+ struct btrfs_disk_key *parent_key;
+ parent_key = &parent->ptrs[parent_slot].key;
+ BUG_ON(memcmp(parent_key, &leaf->items[0].key,
+ sizeof(struct btrfs_disk_key)));
+ BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
+ btrfs_header_blocknr(&leaf->header));
+ }
+ for (i = 0; nritems > 1 && i < nritems - 2; i++) {
+ struct btrfs_key cpukey;
+ btrfs_disk_key_to_cpu(&cpukey, &leaf->items[i + 1].key);
+ BUG_ON(comp_keys(&leaf->items[i].key,
+ &cpukey) >= 0);
+ BUG_ON(btrfs_item_offset(leaf->items + i) !=
+ btrfs_item_end(leaf->items + i + 1));
+ if (i == 0) {
+ BUG_ON(btrfs_item_offset(leaf->items + i) +
+ btrfs_item_size(leaf->items + i) !=
+ BTRFS_LEAF_DATA_SIZE(root));
+ }
+ }
+ return 0;
+}
+
+static int check_block(struct btrfs_root *root, struct btrfs_path *path,
+ int level)
+{
+ if (level == 0)
+ return check_leaf(root, path, level);
+ return check_node(root, path, level);
+}
+
+/*
+ * search for key in the array p. items p are item_size apart
+ * and there are 'max' items in p
+ * the slot in the array is returned via slot, and it points to
+ * the place where you would insert key if it is not found in
+ * the array.
+ *
+ * slot may point to max if the key is bigger than all of the keys
+ */
+static int generic_bin_search(char *p, int item_size, struct btrfs_key *key,
int max, int *slot)
{
int low = 0;
int high = max;
int mid;
int ret;
- struct key *tmp;
+ struct btrfs_disk_key *tmp;
while(low < high) {
mid = (low + high) / 2;
- tmp = (struct key *)(p + mid * item_size);
+ tmp = (struct btrfs_disk_key *)(p + mid * item_size);
ret = comp_keys(tmp, key);
if (ret < 0)
return 1;
}
-int bin_search(struct node *c, struct key *key, int *slot)
+/*
+ * simple bin_search frontend that does the right thing for
+ * leaves vs nodes
+ */
+static int bin_search(struct btrfs_node *c, struct btrfs_key *key, int *slot)
{
- if (is_leaf(c->header.flags)) {
- struct leaf *l = (struct leaf *)c;
- return generic_bin_search((void *)l->items, sizeof(struct item),
- key, c->header.nritems, slot);
+ if (btrfs_is_leaf(c)) {
+ struct btrfs_leaf *l = (struct btrfs_leaf *)c;
+ return generic_bin_search((void *)l->items,
+ sizeof(struct btrfs_item),
+ key, btrfs_header_nritems(&c->header),
+ slot);
} else {
- return generic_bin_search((void *)c->keys, sizeof(struct key),
- key, c->header.nritems, slot);
+ return generic_bin_search((void *)c->ptrs,
+ sizeof(struct btrfs_key_ptr),
+ key, btrfs_header_nritems(&c->header),
+ slot);
}
return -1;
}
-void *read_block(u64 blocknum)
+static struct btrfs_buffer *read_node_slot(struct btrfs_root *root,
+ struct btrfs_buffer *parent_buf,
+ int slot)
+{
+ struct btrfs_node *node = &parent_buf->node;
+ if (slot < 0)
+ return NULL;
+ if (slot >= btrfs_header_nritems(&node->header))
+ return NULL;
+ return read_tree_block(root, btrfs_node_blockptr(node, slot));
+}
+
+static int balance_level(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int level)
{
- return (void *)blocknum;
+ struct btrfs_buffer *right_buf;
+ struct btrfs_buffer *mid_buf;
+ struct btrfs_buffer *left_buf;
+ struct btrfs_buffer *parent_buf = NULL;
+ struct btrfs_node *right = NULL;
+ struct btrfs_node *mid;
+ struct btrfs_node *left = NULL;
+ struct btrfs_node *parent = NULL;
+ int ret = 0;
+ int wret;
+ int pslot;
+ int orig_slot = path->slots[level];
+ u64 orig_ptr;
+
+ if (level == 0)
+ return 0;
+
+ mid_buf = path->nodes[level];
+ mid = &mid_buf->node;
+ orig_ptr = btrfs_node_blockptr(mid, orig_slot);
+
+ if (level < BTRFS_MAX_LEVEL - 1)
+ parent_buf = path->nodes[level + 1];
+ pslot = path->slots[level + 1];
+
+ /*
+ * deal with the case where there is only one pointer in the root
+ * by promoting the node below to a root
+ */
+ if (!parent_buf) {
+ struct btrfs_buffer *child;
+ u64 blocknr = mid_buf->blocknr;
+
+ if (btrfs_header_nritems(&mid->header) != 1)
+ return 0;
+
+ /* promote the child to a root */
+ child = read_node_slot(root, mid_buf, 0);
+ BUG_ON(!child);
+ root->node = child;
+ path->nodes[level] = NULL;
+ /* once for the path */
+ btrfs_block_release(root, mid_buf);
+ /* once for the root ptr */
+ btrfs_block_release(root, mid_buf);
+ clean_tree_block(trans, root, mid_buf);
+ return btrfs_free_extent(trans, root, blocknr, 1, 1);
+ }
+ parent = &parent_buf->node;
+
+ if (btrfs_header_nritems(&mid->header) >
+ BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
+ return 0;
+
+ left_buf = read_node_slot(root, parent_buf, pslot - 1);
+ right_buf = read_node_slot(root, parent_buf, pslot + 1);
+
+ /* first, try to make some room in the middle buffer */
+ if (left_buf) {
+ btrfs_cow_block(trans, root, left_buf, parent_buf, pslot - 1,
+ &left_buf);
+ left = &left_buf->node;
+ orig_slot += btrfs_header_nritems(&left->header);
+ wret = push_node_left(trans, root, left_buf, mid_buf);
+ if (wret < 0)
+ ret = wret;
+ }
+
+ /*
+ * then try to empty the right most buffer into the middle
+ */
+ if (right_buf) {
+ btrfs_cow_block(trans, root, right_buf, parent_buf, pslot + 1,
+ &right_buf);
+ right = &right_buf->node;
+ wret = push_node_left(trans, root, mid_buf, right_buf);
+ if (wret < 0)
+ ret = wret;
+ if (btrfs_header_nritems(&right->header) == 0) {
+ u64 blocknr = right_buf->blocknr;
+ btrfs_block_release(root, right_buf);
+ clean_tree_block(trans, root, right_buf);
+ right_buf = NULL;
+ right = NULL;
+ wret = del_ptr(trans, root, path, level + 1, pslot +
+ 1);
+ if (wret)
+ ret = wret;
+ wret = btrfs_free_extent(trans, root, blocknr, 1, 1);
+ if (wret)
+ ret = wret;
+ } else {
+ memcpy(&parent->ptrs[pslot + 1].key,
+ &right->ptrs[0].key,
+ sizeof(struct btrfs_disk_key));
+ BUG_ON(list_empty(&parent_buf->dirty));
+ }
+ }
+ if (btrfs_header_nritems(&mid->header) == 1) {
+ /*
+ * we're not allowed to leave a node with one item in the
+ * tree during a delete. A deletion from lower in the tree
+ * could try to delete the only pointer in this node.
+ * So, pull some keys from the left.
+ * There has to be a left pointer at this point because
+ * otherwise we would have pulled some pointers from the
+ * right
+ */
+ BUG_ON(!left_buf);
+ wret = balance_node_right(trans, root, mid_buf, left_buf);
+ if (wret < 0)
+ ret = wret;
+ BUG_ON(wret == 1);
+ }
+ if (btrfs_header_nritems(&mid->header) == 0) {
+ /* we've managed to empty the middle node, drop it */
+ u64 blocknr = mid_buf->blocknr;
+ btrfs_block_release(root, mid_buf);
+ clean_tree_block(trans, root, mid_buf);
+ mid_buf = NULL;
+ mid = NULL;
+ wret = del_ptr(trans, root, path, level + 1, pslot);
+ if (wret)
+ ret = wret;
+ wret = btrfs_free_extent(trans, root, blocknr, 1, 1);
+ if (wret)
+ ret = wret;
+ } else {
+ /* update the parent key to reflect our changes */
+ memcpy(&parent->ptrs[pslot].key, &mid->ptrs[0].key,
+ sizeof(struct btrfs_disk_key));
+ BUG_ON(list_empty(&parent_buf->dirty));
+ }
+
+ /* update the path */
+ if (left_buf) {
+ if (btrfs_header_nritems(&left->header) > orig_slot) {
+ left_buf->count++; // released below
+ path->nodes[level] = left_buf;
+ path->slots[level + 1] -= 1;
+ path->slots[level] = orig_slot;
+ if (mid_buf)
+ btrfs_block_release(root, mid_buf);
+ } else {
+ orig_slot -= btrfs_header_nritems(&left->header);
+ path->slots[level] = orig_slot;
+ }
+ }
+ /* double check we haven't messed things up */
+ check_block(root, path, level);
+ if (orig_ptr != btrfs_node_blockptr(&path->nodes[level]->node,
+ path->slots[level]))
+ BUG();
+
+ if (right_buf)
+ btrfs_block_release(root, right_buf);
+ if (left_buf)
+ btrfs_block_release(root, left_buf);
+ return ret;
}
-int search_slot(struct ctree_root *root, struct key *key, struct ctree_path *p)
+/*
+ * look for key in the tree. path is filled in with nodes along the way
+ * if key is found, we return zero and you can find the item in the leaf
+ * level of the path (level 0)
+ *
+ * If the key isn't found, the path points to the slot where it should
+ * be inserted, and 1 is returned. If there are other errors during the
+ * search a negative error number is returned.
+ *
+ * if ins_len > 0, nodes and leaves will be split as we walk down the
+ * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
+ * possible)
+ */
+int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_key *key, struct btrfs_path *p, int
+ ins_len, int cow)
{
- struct node *c = root->node;
+ struct btrfs_buffer *b;
+ struct btrfs_buffer *cow_buf;
+ struct btrfs_node *c;
int slot;
int ret;
int level;
- while (c) {
- level = node_level(c->header.flags);
- p->nodes[level] = c;
+
+again:
+ b = root->node;
+ b->count++;
+ while (b) {
+ level = btrfs_header_level(&b->node.header);
+ if (cow) {
+ int wret;
+ wret = btrfs_cow_block(trans, root, b, p->nodes[level +
+ 1], p->slots[level + 1],
+ &cow_buf);
+ b = cow_buf;
+ }
+ BUG_ON(!cow && ins_len);
+ c = &b->node;
+ p->nodes[level] = b;
+ ret = check_block(root, p, level);
+ if (ret)
+ return -1;
ret = bin_search(c, key, &slot);
- if (!is_leaf(c->header.flags)) {
+ if (!btrfs_is_leaf(c)) {
if (ret && slot > 0)
slot -= 1;
p->slots[level] = slot;
- c = read_block(c->blockptrs[slot]);
- continue;
+ if (ins_len > 0 && btrfs_header_nritems(&c->header) ==
+ BTRFS_NODEPTRS_PER_BLOCK(root)) {
+ int sret = split_node(trans, root, p, level);
+ BUG_ON(sret > 0);
+ if (sret)
+ return sret;
+ b = p->nodes[level];
+ c = &b->node;
+ slot = p->slots[level];
+ } else if (ins_len < 0) {
+ int sret = balance_level(trans, root, p,
+ level);
+ if (sret)
+ return sret;
+ b = p->nodes[level];
+ if (!b)
+ goto again;
+ c = &b->node;
+ slot = p->slots[level];
+ BUG_ON(btrfs_header_nritems(&c->header) == 1);
+ }
+ b = read_tree_block(root, btrfs_node_blockptr(c, slot));
} else {
+ struct btrfs_leaf *l = (struct btrfs_leaf *)c;
p->slots[level] = slot;
+ if (ins_len > 0 && btrfs_leaf_free_space(root, l) <
+ sizeof(struct btrfs_item) + ins_len) {
+ int sret = split_leaf(trans, root, p, ins_len);
+ BUG_ON(sret > 0);
+ if (sret)
+ return sret;
+ }
+ BUG_ON(root->node->count == 1);
return ret;
}
}
- return -1;
+ BUG_ON(root->node->count == 1);
+ return 1;
}
-static void fixup_low_keys(struct ctree_path *path, struct key *key,
- int level)
+/*
+ * adjust the pointers going up the tree, starting at level
+ * making sure the right key of each node is points to 'key'.
+ * This is used after shifting pointers to the left, so it stops
+ * fixing up pointers when a given leaf/node is not in slot 0 of the
+ * higher levels
+ *
+ * If this fails to write a tree block, it returns -1, but continues
+ * fixing up the blocks in ram so the tree is consistent.
+ */
+static int fixup_low_keys(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, struct btrfs_disk_key
+ *key, int level)
{
int i;
- /* adjust the pointers going up the tree */
- for (i = level; i < MAX_LEVEL; i++) {
- struct node *t = path->nodes[i];
+ int ret = 0;
+ for (i = level; i < BTRFS_MAX_LEVEL; i++) {
+ struct btrfs_node *t;
int tslot = path->slots[i];
- if (!t)
+ if (!path->nodes[i])
break;
- memcpy(t->keys + tslot, key, sizeof(*key));
+ t = &path->nodes[i]->node;
+ memcpy(&t->ptrs[tslot].key, key, sizeof(*key));
+ BUG_ON(list_empty(&path->nodes[i]->dirty));
if (tslot != 0)
break;
}
+ return ret;
}
-int __insert_ptr(struct ctree_root *root,
- struct ctree_path *path, struct key *key,
- u64 blocknr, int slot, int level)
-{
- struct node *c;
- struct node *lower;
- struct key *lower_key;
- int nritems;
- /* need a new root */
- if (!path->nodes[level]) {
- c = malloc(sizeof(struct node));
- memset(c, 0, sizeof(c));
- c->header.nritems = 2;
- c->header.flags = node_level(level);
- lower = path->nodes[level-1];
- if (is_leaf(lower->header.flags))
- lower_key = &((struct leaf *)lower)->items[0].key;
- else
- lower_key = lower->keys;
- memcpy(c->keys, lower_key, sizeof(struct key));
- memcpy(c->keys + 1, key, sizeof(struct key));
- c->blockptrs[0] = (u64)lower;
- c->blockptrs[1] = blocknr;
- root->node = c;
- path->nodes[level] = c;
- path->slots[level] = 0;
- if (c->keys[1].objectid == 0)
- BUG();
- return 0;
- }
- lower = path->nodes[level];
- nritems = lower->header.nritems;
- if (slot > nritems)
- BUG();
- if (nritems == NODEPTRS_PER_BLOCK)
- BUG();
- if (slot != nritems) {
- memmove(lower->keys + slot + 1, lower->keys + slot,
- (nritems - slot) * sizeof(struct key));
- memmove(lower->blockptrs + slot + 1, lower->blockptrs + slot,
- (nritems - slot) * sizeof(u64));
- }
- memcpy(lower->keys + slot, key, sizeof(struct key));
- lower->blockptrs[slot] = blocknr;
- lower->header.nritems++;
- if (lower->keys[1].objectid == 0)
- BUG();
- return 0;
-}
-
-int push_node_left(struct ctree_root *root, struct ctree_path *path, int level)
+/*
+ * try to push data from one node into the next node left in the
+ * tree.
+ *
+ * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
+ * error, and > 0 if there was no room in the left hand block.
+ */
+static int push_node_left(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_buffer *dst_buf, struct
+ btrfs_buffer *src_buf)
{
- int slot;
- struct node *left;
- struct node *right;
+ struct btrfs_node *src = &src_buf->node;
+ struct btrfs_node *dst = &dst_buf->node;
int push_items = 0;
- int left_nritems;
- int right_nritems;
-
- if (level == MAX_LEVEL - 1 || path->nodes[level + 1] == 0)
- return 1;
- slot = path->slots[level + 1];
- if (slot == 0)
- return 1;
+ int src_nritems;
+ int dst_nritems;
+ int ret = 0;
- left = read_block(path->nodes[level + 1]->blockptrs[slot - 1]);
- right = path->nodes[level];
- left_nritems = left->header.nritems;
- right_nritems = right->header.nritems;
- push_items = NODEPTRS_PER_BLOCK - (left_nritems + 1);
- if (push_items <= 0)
+ src_nritems = btrfs_header_nritems(&src->header);
+ dst_nritems = btrfs_header_nritems(&dst->header);
+ push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
+ if (push_items <= 0) {
return 1;
+ }
- if (right_nritems < push_items)
- push_items = right_nritems;
- memcpy(left->keys + left_nritems, right->keys,
- push_items * sizeof(struct key));
- memcpy(left->blockptrs + left_nritems, right->blockptrs,
- push_items * sizeof(u64));
- memmove(right->keys, right->keys + push_items,
- (right_nritems - push_items) * sizeof(struct key));
- memmove(right->blockptrs, right->blockptrs + push_items,
- (right_nritems - push_items) * sizeof(u64));
- right->header.nritems -= push_items;
- left->header.nritems += push_items;
-
- /* adjust the pointers going up the tree */
- fixup_low_keys(path, right->keys, level + 1);
+ if (src_nritems < push_items)
+ push_items = src_nritems;
- /* then fixup the leaf pointer in the path */
- if (path->slots[level] < push_items) {
- path->slots[level] += left_nritems;
- path->nodes[level] = (struct node*)left;
- path->slots[level + 1] -= 1;
- } else {
- path->slots[level] -= push_items;
+ memcpy(dst->ptrs + dst_nritems, src->ptrs,
+ push_items * sizeof(struct btrfs_key_ptr));
+ if (push_items < src_nritems) {
+ memmove(src->ptrs, src->ptrs + push_items,
+ (src_nritems - push_items) *
+ sizeof(struct btrfs_key_ptr));
}
- return 0;
+ btrfs_set_header_nritems(&src->header, src_nritems - push_items);
+ btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
+ BUG_ON(list_empty(&src_buf->dirty));
+ BUG_ON(list_empty(&dst_buf->dirty));
+ return ret;
}
-int push_node_right(struct ctree_root *root, struct ctree_path *path, int level)
+/*
+ * try to push data from one node into the next node right in the
+ * tree.
+ *
+ * returns 0 if some ptrs were pushed, < 0 if there was some horrible
+ * error, and > 0 if there was no room in the right hand block.
+ *
+ * this will only push up to 1/2 the contents of the left node over
+ */
+static int balance_node_right(struct btrfs_trans_handle *trans, struct
+ btrfs_root *root, struct btrfs_buffer *dst_buf,
+ struct btrfs_buffer *src_buf)
{
- int slot;
- struct node *dst;
- struct node *src;
+ struct btrfs_node *src = &src_buf->node;
+ struct btrfs_node *dst = &dst_buf->node;
int push_items = 0;
- int dst_nritems;
+ int max_push;
int src_nritems;
+ int dst_nritems;
+ int ret = 0;
- if (level == MAX_LEVEL - 1 || path->nodes[level + 1] == 0)
- return 1;
- slot = path->slots[level + 1];
- if (slot == NODEPTRS_PER_BLOCK - 1)
+ src_nritems = btrfs_header_nritems(&src->header);
+ dst_nritems = btrfs_header_nritems(&dst->header);
+ push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
+ if (push_items <= 0) {
return 1;
+ }
- if (slot >= path->nodes[level + 1]->header.nritems -1)
+ max_push = src_nritems / 2 + 1;
+ /* don't try to empty the node */
+ if (max_push > src_nritems)
return 1;
+ if (max_push < push_items)
+ push_items = max_push;
- dst = read_block(path->nodes[level + 1]->blockptrs[slot + 1]);
- src = path->nodes[level];
- dst_nritems = dst->header.nritems;
- src_nritems = src->header.nritems;
- push_items = NODEPTRS_PER_BLOCK - (dst_nritems + 1);
- if (push_items <= 0)
- return 1;
+ memmove(dst->ptrs + push_items, dst->ptrs,
+ dst_nritems * sizeof(struct btrfs_key_ptr));
+ memcpy(dst->ptrs, src->ptrs + src_nritems - push_items,
+ push_items * sizeof(struct btrfs_key_ptr));
- if (src_nritems < push_items)
- push_items = src_nritems;
- memmove(dst->keys + push_items, dst->keys,
- dst_nritems * sizeof(struct key));
- memcpy(dst->keys, src->keys + src_nritems - push_items,
- push_items * sizeof(struct key));
-
- memmove(dst->blockptrs + push_items, dst->blockptrs,
- dst_nritems * sizeof(u64));
- memcpy(dst->blockptrs, src->blockptrs + src_nritems - push_items,
- push_items * sizeof(u64));
-
- src->header.nritems -= push_items;
- dst->header.nritems += push_items;
-
- /* adjust the pointers going up the tree */
- memcpy(path->nodes[level + 1]->keys + path->slots[level + 1] + 1,
- dst->keys, sizeof(struct key));
- /* then fixup the leaf pointer in the path */
- if (path->slots[level] >= src->header.nritems) {
- path->slots[level] -= src->header.nritems;
- path->nodes[level] = (struct node*)dst;
- path->slots[level + 1] += 1;
+ btrfs_set_header_nritems(&src->header, src_nritems - push_items);
+ btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
+
+ BUG_ON(list_empty(&src_buf->dirty));
+ BUG_ON(list_empty(&dst_buf->dirty));
+ return ret;
+}
+
+/*
+ * helper function to insert a new root level in the tree.
+ * A new node is allocated, and a single item is inserted to
+ * point to the existing root
+ *
+ * returns zero on success or < 0 on failure.
+ */
+static int insert_new_root(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int level)
+{
+ struct btrfs_buffer *t;
+ struct btrfs_node *lower;
+ struct btrfs_node *c;
+ struct btrfs_disk_key *lower_key;
+
+ BUG_ON(path->nodes[level]);
+ BUG_ON(path->nodes[level-1] != root->node);
+
+ t = btrfs_alloc_free_block(trans, root);
+ c = &t->node;
+ memset(c, 0, root->blocksize);
+ btrfs_set_header_nritems(&c->header, 1);
+ btrfs_set_header_level(&c->header, level);
+ btrfs_set_header_blocknr(&c->header, t->blocknr);
+ lower = &path->nodes[level-1]->node;
+ if (btrfs_is_leaf(lower))
+ lower_key = &((struct btrfs_leaf *)lower)->items[0].key;
+ else
+ lower_key = &lower->ptrs[0].key;
+ memcpy(&c->ptrs[0].key, lower_key, sizeof(struct btrfs_disk_key));
+ btrfs_set_node_blockptr(c, 0, path->nodes[level - 1]->blocknr);
+ /* the super has an extra ref to root->node */
+ btrfs_block_release(root, root->node);
+ root->node = t;
+ t->count++;
+ path->nodes[level] = t;
+ path->slots[level] = 0;
+ return 0;
+}
+
+/*
+ * worker function to insert a single pointer in a node.
+ * the node should have enough room for the pointer already
+ *
+ * slot and level indicate where you want the key to go, and
+ * blocknr is the block the key points to.
+ *
+ * returns zero on success and < 0 on any error
+ */
+static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, struct btrfs_disk_key
+ *key, u64 blocknr, int slot, int level)
+{
+ struct btrfs_node *lower;
+ int nritems;
+
+ BUG_ON(!path->nodes[level]);
+ lower = &path->nodes[level]->node;
+ nritems = btrfs_header_nritems(&lower->header);
+ if (slot > nritems)
+ BUG();
+ if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
+ BUG();
+ if (slot != nritems) {
+ memmove(lower->ptrs + slot + 1, lower->ptrs + slot,
+ (nritems - slot) * sizeof(struct btrfs_key_ptr));
}
+ memcpy(&lower->ptrs[slot].key, key, sizeof(struct btrfs_disk_key));
+ btrfs_set_node_blockptr(lower, slot, blocknr);
+ btrfs_set_header_nritems(&lower->header, nritems + 1);
+ BUG_ON(list_empty(&path->nodes[level]->dirty));
return 0;
}
-int insert_ptr(struct ctree_root *root,
- struct ctree_path *path, struct key *key,
- u64 blocknr, int level)
+/*
+ * split the node at the specified level in path in two.
+ * The path is corrected to point to the appropriate node after the split
+ *
+ * Before splitting this tries to make some room in the node by pushing
+ * left and right, if either one works, it returns right away.
+ *
+ * returns 0 on success and < 0 on failure
+ */
+static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int level)
{
- struct node *c = path->nodes[level];
- struct node *b;
- struct node *bal[MAX_LEVEL];
- int bal_level = level;
+ struct btrfs_buffer *t;
+ struct btrfs_node *c;
+ struct btrfs_buffer *split_buffer;
+ struct btrfs_node *split;
int mid;
- int bal_start = -1;
-
- memset(bal, 0, ARRAY_SIZE(bal));
- while(c && c->header.nritems == NODEPTRS_PER_BLOCK) {
- if (push_node_left(root, path,
- node_level(c->header.flags)) == 0)
- break;
- if (push_node_right(root, path,
- node_level(c->header.flags)) == 0)
- break;
- bal_start = bal_level;
- if (bal_level == MAX_LEVEL - 1)
- BUG();
- b = malloc(sizeof(struct node));
- b->header.flags = c->header.flags;
- mid = (c->header.nritems + 1) / 2;
- memcpy(b->keys, c->keys + mid,
- (c->header.nritems - mid) * sizeof(struct key));
- memcpy(b->blockptrs, c->blockptrs + mid,
- (c->header.nritems - mid) * sizeof(u64));
- b->header.nritems = c->header.nritems - mid;
- c->header.nritems = mid;
- bal[bal_level] = b;
- if (bal_level == MAX_LEVEL - 1)
- break;
- bal_level += 1;
- c = path->nodes[bal_level];
- }
- while(bal_start > 0) {
- b = bal[bal_start];
- c = path->nodes[bal_start];
- __insert_ptr(root, path, b->keys, (u64)b,
- path->slots[bal_start + 1] + 1, bal_start + 1);
- if (path->slots[bal_start] >= c->header.nritems) {
- path->slots[bal_start] -= c->header.nritems;
- path->nodes[bal_start] = b;
- path->slots[bal_start + 1] += 1;
- }
- bal_start--;
- if (!bal[bal_start])
- break;
+ int ret;
+ int wret;
+ u32 c_nritems;
+
+ t = path->nodes[level];
+ c = &t->node;
+ if (t == root->node) {
+ /* trying to split the root, lets make a new one */
+ ret = insert_new_root(trans, root, path, level + 1);
+ if (ret)
+ return ret;
}
- return __insert_ptr(root, path, key, blocknr, path->slots[level] + 1,
- level);
+ c_nritems = btrfs_header_nritems(&c->header);
+ split_buffer = btrfs_alloc_free_block(trans, root);
+ split = &split_buffer->node;
+ btrfs_set_header_flags(&split->header, btrfs_header_flags(&c->header));
+ btrfs_set_header_level(&split->header, btrfs_header_level(&c->header));
+ btrfs_set_header_blocknr(&split->header, split_buffer->blocknr);
+ mid = (c_nritems + 1) / 2;
+ memcpy(split->ptrs, c->ptrs + mid,
+ (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
+ btrfs_set_header_nritems(&split->header, c_nritems - mid);
+ btrfs_set_header_nritems(&c->header, mid);
+ ret = 0;
+
+ BUG_ON(list_empty(&t->dirty));
+ wret = insert_ptr(trans, root, path, &split->ptrs[0].key,
+ split_buffer->blocknr, path->slots[level + 1] + 1,
+ level + 1);
+ if (wret)
+ ret = wret;
+
+ if (path->slots[level] >= mid) {
+ path->slots[level] -= mid;
+ btrfs_block_release(root, t);
+ path->nodes[level] = split_buffer;
+ path->slots[level + 1] += 1;
+ } else {
+ btrfs_block_release(root, split_buffer);
+ }
+ return ret;
}
-int leaf_space_used(struct leaf *l, int start, int nr)
+/*
+ * how many bytes are required to store the items in a leaf. start
+ * and nr indicate which items in the leaf to check. This totals up the
+ * space used both by the item structs and the item data
+ */
+static int leaf_space_used(struct btrfs_leaf *l, int start, int nr)
{
int data_len;
int end = start + nr - 1;
if (!nr)
return 0;
- data_len = l->items[start].offset + l->items[start].size;
- data_len = data_len - l->items[end].offset;
- data_len += sizeof(struct item) * nr;
+ data_len = btrfs_item_end(l->items + start);
+ data_len = data_len - btrfs_item_offset(l->items + end);
+ data_len += sizeof(struct btrfs_item) * nr;
return data_len;
}
-int push_leaf_left(struct ctree_root *root, struct ctree_path *path,
- int data_size)
+/*
+ * push some data in the path leaf to the right, trying to free up at
+ * least data_size bytes. returns zero if the push worked, nonzero otherwise
+ *
+ * returns 1 if the push failed because the other node didn't have enough
+ * room, 0 if everything worked out and < 0 if there were major errors.
+ */
+static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int data_size)
{
- struct leaf *right = (struct leaf *)path->nodes[0];
- struct leaf *left;
+ struct btrfs_buffer *left_buf = path->nodes[0];
+ struct btrfs_leaf *left = &left_buf->leaf;
+ struct btrfs_leaf *right;
+ struct btrfs_buffer *right_buf;
+ struct btrfs_buffer *upper;
int slot;
int i;
int free_space;
int push_space = 0;
int push_items = 0;
- struct item *item;
- int old_left_nritems;
+ struct btrfs_item *item;
+ u32 left_nritems;
+ u32 right_nritems;
+
+ slot = path->slots[1];
+ if (!path->nodes[1]) {
+ return 1;
+ }
+ upper = path->nodes[1];
+ if (slot >= btrfs_header_nritems(&upper->node.header) - 1) {
+ return 1;
+ }
+ right_buf = read_tree_block(root, btrfs_node_blockptr(&upper->node,
+ slot + 1));
+ right = &right_buf->leaf;
+ free_space = btrfs_leaf_free_space(root, right);
+ if (free_space < data_size + sizeof(struct btrfs_item)) {
+ btrfs_block_release(root, right_buf);
+ return 1;
+ }
+ /* cow and double check */
+ btrfs_cow_block(trans, root, right_buf, upper, slot + 1, &right_buf);
+ right = &right_buf->leaf;
+ free_space = btrfs_leaf_free_space(root, right);
+ if (free_space < data_size + sizeof(struct btrfs_item)) {
+ btrfs_block_release(root, right_buf);
+ return 1;
+ }
+
+ left_nritems = btrfs_header_nritems(&left->header);
+ for (i = left_nritems - 1; i >= 0; i--) {
+ item = left->items + i;
+ if (path->slots[0] == i)
+ push_space += data_size + sizeof(*item);
+ if (btrfs_item_size(item) + sizeof(*item) + push_space >
+ free_space)
+ break;
+ push_items++;
+ push_space += btrfs_item_size(item) + sizeof(*item);
+ }
+ if (push_items == 0) {
+ btrfs_block_release(root, right_buf);
+ return 1;
+ }
+ right_nritems = btrfs_header_nritems(&right->header);
+ /* push left to right */
+ push_space = btrfs_item_end(left->items + left_nritems - push_items);
+ push_space -= leaf_data_end(root, left);
+ /* make room in the right data area */
+ memmove(btrfs_leaf_data(right) + leaf_data_end(root, right) -
+ push_space, btrfs_leaf_data(right) + leaf_data_end(root, right),
+ BTRFS_LEAF_DATA_SIZE(root) - leaf_data_end(root, right));
+ /* copy from the left data area */
+ memcpy(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - push_space,
+ btrfs_leaf_data(left) + leaf_data_end(root, left), push_space);
+ memmove(right->items + push_items, right->items,
+ right_nritems * sizeof(struct btrfs_item));
+ /* copy the items from left to right */
+ memcpy(right->items, left->items + left_nritems - push_items,
+ push_items * sizeof(struct btrfs_item));
+
+ /* update the item pointers */
+ right_nritems += push_items;
+ btrfs_set_header_nritems(&right->header, right_nritems);
+ push_space = BTRFS_LEAF_DATA_SIZE(root);
+ for (i = 0; i < right_nritems; i++) {
+ btrfs_set_item_offset(right->items + i, push_space -
+ btrfs_item_size(right->items + i));
+ push_space = btrfs_item_offset(right->items + i);
+ }
+ left_nritems -= push_items;
+ btrfs_set_header_nritems(&left->header, left_nritems);
+
+ BUG_ON(list_empty(&left_buf->dirty));
+ BUG_ON(list_empty(&right_buf->dirty));
+ memcpy(&upper->node.ptrs[slot + 1].key,
+ &right->items[0].key, sizeof(struct btrfs_disk_key));
+ BUG_ON(list_empty(&upper->dirty));
+
+ /* then fixup the leaf pointer in the path */
+ if (path->slots[0] >= left_nritems) {
+ path->slots[0] -= left_nritems;
+ btrfs_block_release(root, path->nodes[0]);
+ path->nodes[0] = right_buf;
+ path->slots[1] += 1;
+ } else {
+ btrfs_block_release(root, right_buf);
+ }
+ return 0;
+}
+/*
+ * push some data in the path leaf to the left, trying to free up at
+ * least data_size bytes. returns zero if the push worked, nonzero otherwise
+ */
+static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int data_size)
+{
+ struct btrfs_buffer *right_buf = path->nodes[0];
+ struct btrfs_leaf *right = &right_buf->leaf;
+ struct btrfs_buffer *t;
+ struct btrfs_leaf *left;
+ int slot;
+ int i;
+ int free_space;
+ int push_space = 0;
+ int push_items = 0;
+ struct btrfs_item *item;
+ u32 old_left_nritems;
+ int ret = 0;
+ int wret;
slot = path->slots[1];
if (slot == 0) {
if (!path->nodes[1]) {
return 1;
}
- left = read_block(path->nodes[1]->blockptrs[slot - 1]);
- free_space = leaf_free_space(left);
- if (free_space < data_size + sizeof(struct item)) {
+ t = read_tree_block(root, btrfs_node_blockptr(&path->nodes[1]->node,
+ slot - 1));
+ left = &t->leaf;
+ free_space = btrfs_leaf_free_space(root, left);
+ if (free_space < data_size + sizeof(struct btrfs_item)) {
+ btrfs_block_release(root, t);
+ return 1;
+ }
+
+ /* cow and double check */
+ btrfs_cow_block(trans, root, t, path->nodes[1], slot - 1, &t);
+ left = &t->leaf;
+ free_space = btrfs_leaf_free_space(root, left);
+ if (free_space < data_size + sizeof(struct btrfs_item)) {
+ btrfs_block_release(root, t);
return 1;
}
- for (i = 0; i < right->header.nritems; i++) {
+
+ for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
item = right->items + i;
if (path->slots[0] == i)
push_space += data_size + sizeof(*item);
- if (item->size + sizeof(*item) + push_space > free_space)
+ if (btrfs_item_size(item) + sizeof(*item) + push_space >
+ free_space)
break;
push_items++;
- push_space += item->size + sizeof(*item);
+ push_space += btrfs_item_size(item) + sizeof(*item);
}
if (push_items == 0) {
+ btrfs_block_release(root, t);
return 1;
}
/* push data from right to left */
- memcpy(left->items + left->header.nritems,
- right->items, push_items * sizeof(struct item));
- push_space = LEAF_DATA_SIZE - right->items[push_items -1].offset;
- memcpy(left->data + leaf_data_end(left) - push_space,
- right->data + right->items[push_items - 1].offset,
+ memcpy(left->items + btrfs_header_nritems(&left->header),
+ right->items, push_items * sizeof(struct btrfs_item));
+ push_space = BTRFS_LEAF_DATA_SIZE(root) -
+ btrfs_item_offset(right->items + push_items -1);
+ memcpy(btrfs_leaf_data(left) + leaf_data_end(root, left) - push_space,
+ btrfs_leaf_data(right) +
+ btrfs_item_offset(right->items + push_items - 1),
push_space);
- old_left_nritems = left->header.nritems;
- for(i = old_left_nritems; i < old_left_nritems + push_items; i++) {
- left->items[i].offset -= LEAF_DATA_SIZE -
- left->items[old_left_nritems -1].offset;
+ old_left_nritems = btrfs_header_nritems(&left->header);
+ BUG_ON(old_left_nritems < 0);
+
+ for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
+ u32 ioff = btrfs_item_offset(left->items + i);
+ btrfs_set_item_offset(left->items + i, ioff -
+ (BTRFS_LEAF_DATA_SIZE(root) -
+ btrfs_item_offset(left->items +
+ old_left_nritems - 1)));
}
- left->header.nritems += push_items;
+ btrfs_set_header_nritems(&left->header, old_left_nritems + push_items);
/* fixup right node */
- push_space = right->items[push_items-1].offset - leaf_data_end(right);
- memmove(right->data + LEAF_DATA_SIZE - push_space, right->data +
- leaf_data_end(right), push_space);
+ push_space = btrfs_item_offset(right->items + push_items - 1) -
+ leaf_data_end(root, right);
+ memmove(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
+ push_space, btrfs_leaf_data(right) +
+ leaf_data_end(root, right), push_space);
memmove(right->items, right->items + push_items,
- (right->header.nritems - push_items) * sizeof(struct item));
- right->header.nritems -= push_items;
- push_space = LEAF_DATA_SIZE;
- for (i = 0; i < right->header.nritems; i++) {
- right->items[i].offset = push_space - right->items[i].size;
- push_space = right->items[i].offset;
+ (btrfs_header_nritems(&right->header) - push_items) *
+ sizeof(struct btrfs_item));
+ btrfs_set_header_nritems(&right->header,
+ btrfs_header_nritems(&right->header) -
+ push_items);
+ push_space = BTRFS_LEAF_DATA_SIZE(root);
+
+ for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
+ btrfs_set_item_offset(right->items + i, push_space -
+ btrfs_item_size(right->items + i));
+ push_space = btrfs_item_offset(right->items + i);
}
- fixup_low_keys(path, &right->items[0].key, 1);
+
+ BUG_ON(list_empty(&t->dirty));
+ BUG_ON(list_empty(&right_buf->dirty));
+
+ wret = fixup_low_keys(trans, root, path, &right->items[0].key, 1);
+ if (wret)
+ ret = wret;
/* then fixup the leaf pointer in the path */
if (path->slots[0] < push_items) {
path->slots[0] += old_left_nritems;
- path->nodes[0] = (struct node*)left;
+ btrfs_block_release(root, path->nodes[0]);
+ path->nodes[0] = t;
path->slots[1] -= 1;
} else {
+ btrfs_block_release(root, t);
path->slots[0] -= push_items;
}
- return 0;
+ BUG_ON(path->slots[0] < 0);
+ return ret;
}
-int split_leaf(struct ctree_root *root, struct ctree_path *path, int data_size)
+/*
+ * split the path's leaf in two, making sure there is at least data_size
+ * available for the resulting leaf level of the path.
+ *
+ * returns 0 if all went well and < 0 on failure.
+ */
+static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, int data_size)
{
- struct leaf *l = (struct leaf *)path->nodes[0];
- int nritems = l->header.nritems;
- int mid = (nritems + 1)/ 2;
- int slot = path->slots[0];
- struct leaf *right;
- int space_needed = data_size + sizeof(struct item);
+ struct btrfs_buffer *l_buf;
+ struct btrfs_leaf *l;
+ u32 nritems;
+ int mid;
+ int slot;
+ struct btrfs_leaf *right;
+ struct btrfs_buffer *right_buffer;
+ int space_needed = data_size + sizeof(struct btrfs_item);
int data_copy_size;
int rt_data_off;
int i;
int ret;
+ int wret;
+
+ /* first try to make some room by pushing left and right */
+ wret = push_leaf_left(trans, root, path, data_size);
+ if (wret < 0)
+ return wret;
+ if (wret) {
+ wret = push_leaf_right(trans, root, path, data_size);
+ if (wret < 0)
+ return wret;
+ }
+ l_buf = path->nodes[0];
+ l = &l_buf->leaf;
- if (push_leaf_left(root, path, data_size) == 0) {
+ /* did the pushes work? */
+ if (btrfs_leaf_free_space(root, l) >=
+ sizeof(struct btrfs_item) + data_size)
return 0;
+
+ if (!path->nodes[1]) {
+ ret = insert_new_root(trans, root, path, 1);
+ if (ret)
+ return ret;
}
- right = malloc(sizeof(struct leaf));
- memset(right, 0, sizeof(*right));
+ slot = path->slots[0];
+ nritems = btrfs_header_nritems(&l->header);
+ mid = (nritems + 1)/ 2;
+ right_buffer = btrfs_alloc_free_block(trans, root);
+ BUG_ON(!right_buffer);
+ BUG_ON(mid == nritems);
+ right = &right_buffer->leaf;
+ memset(&right->header, 0, sizeof(right->header));
if (mid <= slot) {
+ /* FIXME, just alloc a new leaf here */
if (leaf_space_used(l, mid, nritems - mid) + space_needed >
- LEAF_DATA_SIZE)
+ BTRFS_LEAF_DATA_SIZE(root))
BUG();
} else {
+ /* FIXME, just alloc a new leaf here */
if (leaf_space_used(l, 0, mid + 1) + space_needed >
- LEAF_DATA_SIZE)
+ BTRFS_LEAF_DATA_SIZE(root))
BUG();
}
- right->header.nritems = nritems - mid;
- data_copy_size = l->items[mid].offset + l->items[mid].size -
- leaf_data_end(l);
+ btrfs_set_header_nritems(&right->header, nritems - mid);
+ btrfs_set_header_blocknr(&right->header, right_buffer->blocknr);
+ btrfs_set_header_level(&right->header, 0);
+ data_copy_size = btrfs_item_end(l->items + mid) -
+ leaf_data_end(root, l);
memcpy(right->items, l->items + mid,
- (nritems - mid) * sizeof(struct item));
- memcpy(right->data + LEAF_DATA_SIZE - data_copy_size,
- l->data + leaf_data_end(l), data_copy_size);
- rt_data_off = LEAF_DATA_SIZE -
- (l->items[mid].offset + l->items[mid].size);
- for (i = 0; i < right->header.nritems; i++) {
- right->items[i].offset += rt_data_off;
- }
- l->header.nritems = mid;
- ret = insert_ptr(root, path, &right->items[0].key,
- (u64)right, 1);
+ (nritems - mid) * sizeof(struct btrfs_item));
+ memcpy(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
+ data_copy_size, btrfs_leaf_data(l) +
+ leaf_data_end(root, l), data_copy_size);
+ rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
+ btrfs_item_end(l->items + mid);
+
+ for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
+ u32 ioff = btrfs_item_offset(right->items + i);
+ btrfs_set_item_offset(right->items + i, ioff + rt_data_off);
+ }
+
+ btrfs_set_header_nritems(&l->header, mid);
+ ret = 0;
+ wret = insert_ptr(trans, root, path, &right->items[0].key,
+ right_buffer->blocknr, path->slots[1] + 1, 1);
+ if (wret)
+ ret = wret;
+ BUG_ON(list_empty(&right_buffer->dirty));
+ BUG_ON(list_empty(&l_buf->dirty));
+ BUG_ON(path->slots[0] != slot);
if (mid <= slot) {
- path->nodes[0] = (struct node *)right;
+ btrfs_block_release(root, path->nodes[0]);
+ path->nodes[0] = right_buffer;
path->slots[0] -= mid;
path->slots[1] += 1;
- }
+ } else
+ btrfs_block_release(root, right_buffer);
+ BUG_ON(path->slots[0] < 0);
return ret;
}
-int insert_item(struct ctree_root *root, struct key *key,
- void *data, int data_size)
+/*
+ * Given a key and some data, insert an item into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ */
+int btrfs_insert_empty_item(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_path *path, struct btrfs_key
+ *cpu_key, u32 data_size)
{
- int ret;
+ int ret = 0;
int slot;
- struct leaf *leaf;
- unsigned int nritems;
+ int slot_orig;
+ struct btrfs_leaf *leaf;
+ struct btrfs_buffer *leaf_buf;
+ u32 nritems;
unsigned int data_end;
- struct ctree_path path;
+ struct btrfs_disk_key disk_key;
+
+ btrfs_cpu_key_to_disk(&disk_key, cpu_key);
- init_path(&path);
- ret = search_slot(root, key, &path);
- if (ret == 0)
+ /* create a root if there isn't one */
+ if (!root->node)
+ BUG();
+ ret = btrfs_search_slot(trans, root, cpu_key, path, data_size, 1);
+ if (ret == 0) {
+ btrfs_release_path(root, path);
return -EEXIST;
+ }
+ if (ret < 0)
+ goto out;
+
+ slot_orig = path->slots[0];
+ leaf_buf = path->nodes[0];
+ leaf = &leaf_buf->leaf;
- leaf = (struct leaf *)path.nodes[0];
- if (leaf_free_space(leaf) < sizeof(struct item) + data_size)
- split_leaf(root, &path, data_size);
- leaf = (struct leaf *)path.nodes[0];
- nritems = leaf->header.nritems;
- data_end = leaf_data_end(leaf);
- if (leaf_free_space(leaf) < sizeof(struct item) + data_size)
+ nritems = btrfs_header_nritems(&leaf->header);
+ data_end = leaf_data_end(root, leaf);
+
+ if (btrfs_leaf_free_space(root, leaf) <
+ sizeof(struct btrfs_item) + data_size)
BUG();
- slot = path.slots[0];
- if (slot == 0)
- fixup_low_keys(&path, key, 1);
+ slot = path->slots[0];
+ BUG_ON(slot < 0);
if (slot != nritems) {
int i;
- unsigned int old_data = leaf->items[slot].offset +
- leaf->items[slot].size;
+ unsigned int old_data = btrfs_item_end(leaf->items + slot);
/*
* item0..itemN ... dataN.offset..dataN.size .. data0.size
*/
/* first correct the data pointers */
- for (i = slot; i < nritems; i++)
- leaf->items[i].offset -= data_size;
+ for (i = slot; i < nritems; i++) {
+ u32 ioff = btrfs_item_offset(leaf->items + i);
+ btrfs_set_item_offset(leaf->items + i,
+ ioff - data_size);
+ }
/* shift the items */
memmove(leaf->items + slot + 1, leaf->items + slot,
- (nritems - slot) * sizeof(struct item));
+ (nritems - slot) * sizeof(struct btrfs_item));
/* shift the data */
- memmove(leaf->data + data_end - data_size, leaf->data +
+ memmove(btrfs_leaf_data(leaf) + data_end - data_size,
+ btrfs_leaf_data(leaf) +
data_end, old_data - data_end);
data_end = old_data;
}
- memcpy(&leaf->items[slot].key, key, sizeof(struct key));
- leaf->items[slot].offset = data_end - data_size;
- leaf->items[slot].size = data_size;
- memcpy(leaf->data + data_end - data_size, data, data_size);
- leaf->header.nritems += 1;
- if (leaf_free_space(leaf) < 0)
+ /* setup the item for the new data */
+ memcpy(&leaf->items[slot].key, &disk_key,
+ sizeof(struct btrfs_disk_key));
+ btrfs_set_item_offset(leaf->items + slot, data_end - data_size);
+ btrfs_set_item_size(leaf->items + slot, data_size);
+ btrfs_set_header_nritems(&leaf->header, nritems + 1);
+
+ ret = 0;
+ if (slot == 0)
+ ret = fixup_low_keys(trans, root, path, &disk_key, 1);
+
+ BUG_ON(list_empty(&leaf_buf->dirty));
+ if (btrfs_leaf_free_space(root, leaf) < 0)
BUG();
- return 0;
+ check_leaf(root, path, 0);
+out:
+ return ret;
}
-int del_ptr(struct ctree_root *root, struct ctree_path *path, int level)
+/*
+ * Given a key and some data, insert an item into the tree.
+ * This does all the path init required, making room in the tree if needed.
+ */
+int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
+ *root, struct btrfs_key *cpu_key, void *data, u32
+ data_size)
{
- int slot;
- struct node *node;
- int nritems;
+ int ret = 0;
+ struct btrfs_path path;
+ u8 *ptr;
+
+ btrfs_init_path(&path);
+ ret = btrfs_insert_empty_item(trans, root, &path, cpu_key, data_size);
+ if (!ret) {
+ ptr = btrfs_item_ptr(&path.nodes[0]->leaf, path.slots[0], u8);
+ memcpy(ptr, data, data_size);
+ }
+ btrfs_release_path(root, &path);
+ return ret;
+}
- while(1) {
- node = path->nodes[level];
- if (!node)
- break;
- slot = path->slots[level];
- nritems = node->header.nritems;
-
- if (slot != nritems -1) {
- memmove(node->keys + slot, node->keys + slot + 1,
- sizeof(struct key) * (nritems - slot - 1));
- memmove(node->blockptrs + slot,
- node->blockptrs + slot + 1,
- sizeof(u64) * (nritems - slot - 1));
- }
- node->header.nritems--;
- if (node->header.nritems != 0) {
- int tslot;
- if (slot == 0)
- fixup_low_keys(path, node->keys, level + 1);
- tslot = path->slots[level+1];
- push_node_left(root, path, level);
- if (node->header.nritems) {
- push_node_right(root, path, level);
- }
- if (node->header.nritems)
- break;
- path->slots[level+1] = tslot;
- }
- if (node == root->node) {
- printf("root is now null!\n");
- root->node = NULL;
- break;
- }
- level++;
- if (!path->nodes[level])
- BUG();
- free(node);
+/*
+ * delete the pointer from a given node.
+ *
+ * If the delete empties a node, the node is removed from the tree,
+ * continuing all the way the root if required. The root is converted into
+ * a leaf if all the nodes are emptied.
+ */
+static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+ struct btrfs_path *path, int level, int slot)
+{
+ struct btrfs_node *node;
+ struct btrfs_buffer *parent = path->nodes[level];
+ u32 nritems;
+ int ret = 0;
+ int wret;
+
+ node = &parent->node;
+ nritems = btrfs_header_nritems(&node->header);
+ if (slot != nritems -1) {
+ memmove(node->ptrs + slot, node->ptrs + slot + 1,
+ sizeof(struct btrfs_key_ptr) * (nritems - slot - 1));
}
- return 0;
+ nritems--;
+ btrfs_set_header_nritems(&node->header, nritems);
+ if (nritems == 0 && parent == root->node) {
+ BUG_ON(btrfs_header_level(&root->node->node.header) != 1);
+ /* just turn the root into a leaf and break */
+ btrfs_set_header_level(&root->node->node.header, 0);
+ } else if (slot == 0) {
+ wret = fixup_low_keys(trans, root, path, &node->ptrs[0].key,
+ level + 1);
+ if (wret)
+ ret = wret;
+ }
+ BUG_ON(list_empty(&parent->dirty));
+ return ret;
}
-int del_item(struct ctree_root *root, struct ctree_path *path)
+/*
+ * delete the item at the leaf level in path. If that empties
+ * the leaf, remove it from the tree
+ */
+int btrfs_del_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+ struct btrfs_path *path)
{
int slot;
- struct leaf *leaf;
+ struct btrfs_leaf *leaf;
+ struct btrfs_buffer *leaf_buf;
int doff;
int dsize;
+ int ret = 0;
+ int wret;
+ u32 nritems;
- leaf = (struct leaf *)path->nodes[0];
+ leaf_buf = path->nodes[0];
+ leaf = &leaf_buf->leaf;
slot = path->slots[0];
- doff = leaf->items[slot].offset;
- dsize = leaf->items[slot].size;
+ doff = btrfs_item_offset(leaf->items + slot);
+ dsize = btrfs_item_size(leaf->items + slot);
+ nritems = btrfs_header_nritems(&leaf->header);
- if (slot != leaf->header.nritems - 1) {
+ if (slot != nritems - 1) {
int i;
- int data_end = leaf_data_end(leaf);
- memmove(leaf->data + data_end + dsize,
- leaf->data + data_end,
+ int data_end = leaf_data_end(root, leaf);
+ memmove(btrfs_leaf_data(leaf) + data_end + dsize,
+ btrfs_leaf_data(leaf) + data_end,
doff - data_end);
- for (i = slot + 1; i < leaf->header.nritems; i++)
- leaf->items[i].offset += dsize;
+ for (i = slot + 1; i < nritems; i++) {
+ u32 ioff = btrfs_item_offset(leaf->items + i);
+ btrfs_set_item_offset(leaf->items + i, ioff + dsize);
+ }
memmove(leaf->items + slot, leaf->items + slot + 1,
- sizeof(struct item) *
- (leaf->header.nritems - slot - 1));
- }
- leaf->header.nritems -= 1;
- if (leaf->header.nritems == 0) {
- if (leaf == (struct leaf *)root->node)
- root->node = NULL;
- else
- del_ptr(root, path, 1);
- free(leaf);
+ sizeof(struct btrfs_item) *
+ (nritems - slot - 1));
+ }
+ btrfs_set_header_nritems(&leaf->header, nritems - 1);
+ nritems--;
+ /* delete the leaf if we've emptied it */
+ if (nritems == 0) {
+ if (leaf_buf == root->node) {
+ btrfs_set_header_level(&leaf->header, 0);
+ BUG_ON(list_empty(&leaf_buf->dirty));
+ } else {
+ clean_tree_block(trans, root, leaf_buf);
+ wret = del_ptr(trans, root, path, 1, path->slots[1]);
+ if (wret)
+ ret = wret;
+ wret = btrfs_free_extent(trans, root,
+ leaf_buf->blocknr, 1, 1);
+ if (wret)
+ ret = wret;
+ }
} else {
- if (slot == 0)
- fixup_low_keys(path, &leaf->items[0].key, 1);
- if (leaf_space_used(leaf, 0, leaf->header.nritems) <
- LEAF_DATA_SIZE / 4) {
+ int used = leaf_space_used(leaf, 0, nritems);
+ if (slot == 0) {
+ wret = fixup_low_keys(trans, root, path,
+ &leaf->items[0].key, 1);
+ if (wret)
+ ret = wret;
+ }
+ BUG_ON(list_empty(&leaf_buf->dirty));
+
+ /* delete the leaf if it is mostly empty */
+ if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
/* push_leaf_left fixes the path.
* make sure the path still points to our leaf
* for possible call to del_ptr below
*/
slot = path->slots[1];
- push_leaf_left(root, path, 1);
- if (leaf->header.nritems == 0) {
- free(leaf);
- path->slots[1] = slot;
- del_ptr(root, path, 1);
+ leaf_buf->count++;
+ wret = push_leaf_left(trans, root, path, 1);
+ if (wret < 0)
+ ret = wret;
+ if (path->nodes[0] == leaf_buf &&
+ btrfs_header_nritems(&leaf->header)) {
+ wret = push_leaf_right(trans, root, path, 1);
+ if (wret < 0)
+ ret = wret;
+ }
+ if (btrfs_header_nritems(&leaf->header) == 0) {
+ u64 blocknr = leaf_buf->blocknr;
+ clean_tree_block(trans, root, leaf_buf);
+ wret = del_ptr(trans, root, path, 1, slot);
+ if (wret)
+ ret = wret;
+ btrfs_block_release(root, leaf_buf);
+ wret = btrfs_free_extent(trans, root, blocknr,
+ 1, 1);
+ if (wret)
+ ret = wret;
+ } else {
+ btrfs_block_release(root, leaf_buf);
}
}
}
- return 0;
+ return ret;
}
-void print_leaf(struct leaf *l)
+/*
+ * walk up the tree as far as required to find the next leaf.
+ * returns 0 if it found something or 1 if there are no greater leaves.
+ * returns < 0 on io errors.
+ */
+int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
{
- int i;
- int nr = l->header.nritems;
- struct item *item;
- printf("leaf %p total ptrs %d free space %d\n", l, nr,
- leaf_free_space(l));
- fflush(stdout);
- for (i = 0 ; i < nr ; i++) {
- item = l->items + i;
- printf("\titem %d key (%lu %u %lu) itemoff %d itemsize %d\n",
- i,
- item->key.objectid, item->key.flags, item->key.offset,
- item->offset, item->size);
- fflush(stdout);
- printf("\t\titem data %.*s\n", item->size, l->data+item->offset);
- fflush(stdout);
- }
-}
-void print_tree(struct node *c)
-{
- int i;
- int nr;
-
- if (!c)
- return;
- nr = c->header.nritems;
- if (is_leaf(c->header.flags)) {
- print_leaf((struct leaf *)c);
- return;
- }
- printf("node %p level %d total ptrs %d free spc %lu\n", c,
- node_level(c->header.flags), c->header.nritems,
- NODEPTRS_PER_BLOCK - c->header.nritems);
- fflush(stdout);
- for (i = 0; i < nr; i++) {
- printf("\tkey %d (%lu %u %lu) block %lx\n",
- i,
- c->keys[i].objectid, c->keys[i].flags, c->keys[i].offset,
- c->blockptrs[i]);
- fflush(stdout);
- }
- for (i = 0; i < nr; i++) {
- struct node *next = read_block(c->blockptrs[i]);
- if (is_leaf(next->header.flags) &&
- node_level(c->header.flags) != 1)
- BUG();
- if (node_level(next->header.flags) !=
- node_level(c->header.flags) - 1)
- BUG();
- print_tree(next);
- }
-
-}
-
-/* for testing only */
-int next_key(int i, int max_key) {
- return rand() % max_key;
- // return i;
-}
+ int slot;
+ int level = 1;
+ u64 blocknr;
+ struct btrfs_buffer *c;
+ struct btrfs_buffer *next = NULL;
-int main() {
- struct leaf *first_node = malloc(sizeof(struct leaf));
- struct ctree_root root;
- struct key ins;
- struct key last = { (u64)-1, 0, 0};
- char *buf;
- int i;
- int num;
- int ret;
- int run_size = 100000;
- int max_key = 100000000;
- int tree_size = 0;
- struct ctree_path path;
-
-
- srand(55);
- root.node = (struct node *)first_node;
- memset(first_node, 0, sizeof(*first_node));
- for (i = 0; i < run_size; i++) {
- buf = malloc(64);
- num = next_key(i, max_key);
- // num = i;
- sprintf(buf, "string-%d", num);
- // printf("insert %d\n", num);
- ins.objectid = num;
- ins.offset = 0;
- ins.flags = 0;
- ret = insert_item(&root, &ins, buf, strlen(buf));
- if (!ret)
- tree_size++;
- }
- srand(55);
- for (i = 0; i < run_size; i++) {
- num = next_key(i, max_key);
- ins.objectid = num;
- init_path(&path);
- ret = search_slot(&root, &ins, &path);
- if (ret) {
- print_tree(root.node);
- printf("unable to find %d\n", num);
- exit(1);
- }
- }
- printf("node %p level %d total ptrs %d free spc %lu\n", root.node,
- node_level(root.node->header.flags), root.node->header.nritems,
- NODEPTRS_PER_BLOCK - root.node->header.nritems);
- // print_tree(root.node);
- printf("all searches good\n");
- i = 0;
- srand(55);
- for (i = 0 ; i < run_size/4; i++) {
- num = next_key(i, max_key);
- ins.objectid = num;
- init_path(&path);
- ret = search_slot(&root, &ins, &path);
- if (ret)
+ while(level < BTRFS_MAX_LEVEL) {
+ if (!path->nodes[level])
+ return 1;
+ slot = path->slots[level] + 1;
+ c = path->nodes[level];
+ if (slot >= btrfs_header_nritems(&c->node.header)) {
+ level++;
continue;
- ret = del_item(&root, &path);
- if (ret != 0)
- BUG();
- tree_size--;
- }
- srand(128);
- for (i = 0; i < run_size; i++) {
- buf = malloc(64);
- num = next_key(i, max_key);
- sprintf(buf, "string-%d", num);
- ins.objectid = num;
- ret = insert_item(&root, &ins, buf, strlen(buf));
- if (!ret)
- tree_size++;
- }
- while(root.node) {
- struct leaf *leaf;
- int slot;
- ins.objectid = (u64)-1;
- init_path(&path);
- ret = search_slot(&root, &ins, &path);
- if (ret == 0)
- BUG();
-
- leaf = (struct leaf *)(path.nodes[0]);
- slot = path.slots[0];
- if (slot != leaf->header.nritems)
- BUG();
- while(path.slots[0] > 0) {
- path.slots[0] -= 1;
- slot = path.slots[0];
- leaf = (struct leaf *)(path.nodes[0]);
-
- if (comp_keys(&last, &leaf->items[slot].key) <= 0)
- BUG();
- memcpy(&last, &leaf->items[slot].key, sizeof(last));
- ret = del_item(&root, &path);
- if (ret != 0)
- BUG();
- tree_size--;
}
+ blocknr = btrfs_node_blockptr(&c->node, slot);
+ if (next)
+ btrfs_block_release(root, next);
+ next = read_tree_block(root, blocknr);
+ break;
+ }
+ path->slots[level] = slot;
+ while(1) {
+ level--;
+ c = path->nodes[level];
+ btrfs_block_release(root, c);
+ path->nodes[level] = next;
+ path->slots[level] = 0;
+ if (!level)
+ break;
+ next = read_tree_block(root,
+ btrfs_node_blockptr(&next->node, 0));
}
- print_tree(root.node);
- printf("tree size is now %d\n", tree_size);
return 0;
}
projects / platform / upstream / btrfs-progs.git / blobdiff