static void _root_split(struct tree *t,
struct node *new_root,
struct node *old_root)
{
struct node *a;
struct node *b;
struct msg *split_key = NULL;
__DEBUG("root split begin, old NID %"PRIu64" , height %d"
, old_root->nid
, old_root->height);
if (old_root->height > 0 || old_root->n_children > 2)
_node_split(t, old_root, &a, &b, &split_key);
else
_leaf_and_lmb_split(t, old_root, &a, &b, &split_key);
/* swap two roots */
_root_swap(new_root, old_root);
msgcpy(&new_root->pivots[0], split_key);
new_root->parts[0].child_nid = a->nid;
new_root->parts[1].child_nid = b->nid;
msgfree(split_key);
cache_unpin(t->cf, b->cpair, make_cpair_attr(b));
node_set_dirty(old_root);
node_set_dirty(new_root);
t->hdr->height++;
status_increment(&t->e->status->tree_root_new_nums);
__DEBUG("root split end, old NID %"PRIu64, old_root->nid);
}
void arrange_container(struct sway_container *container) {
if (config->reloading) {
return;
}
if (container->view) {
view_autoconfigure(container->view);
node_set_dirty(&container->node);
return;
}
struct wlr_box box;
container_get_box(container, &box);
arrange_children(container->children, container->layout, &box);
node_set_dirty(&container->node);
}
/*
* EFFECT:
* - split leaf&lmb into two leaves:a & b
* a&b are both the half of the lmb
*
* PROCESS:
* - leaf:
* +-----------------------------------+
* | 0 | 1 | 2 | 3 | 4 | 5 |
* +-----------------------------------+
*
* - split:
* root
* +--------+
* | 2 |
* +--------+
* / \
* +-----------------+ +------------------+
* | 0 | 1 | 2 | | 3 | 4 | 5 |
* +-----------------+ +------------------+
* nodea nodeb
*
* ENTER:
* - leaf is already locked (L_WRITE)
* EXITS:
* - a is locked
* - b is locked
*/
void leaf_split(void *tree,
struct node *node,
struct node **a,
struct node **b,
struct msg **split_key)
{
struct partition *pa;
struct partition *pb;
struct node *leafa;
struct node *leafb;
struct lmb *mb;
struct lmb *mba;
struct lmb *mbb;
struct msg *sp_key = NULL;
struct buftree *t = (struct buftree*)tree;
leafa = node;
pa = &leafa->parts[0];
/* split lmb of leaf to mba & mbb */
mb = pa->msgbuf;
lmb_split(mb, &mba, &mbb, &sp_key);
lmb_free(mb);
/* reset leafa buffer */
pa->msgbuf = mba;
/* new leafb */
NID nid = hdr_next_nid(t->hdr);
leaf_new(t->hdr,
nid,
0,
1,
&leafb);
leaf_msgbuf_init(leafb);
cache_put_and_pin(t->cf, nid, leafb);
pb = &leafb->parts[0];
lmb_free(pb->msgbuf);
pb->msgbuf = mbb;
/* set dirty */
node_set_dirty(leafa);
node_set_dirty(leafb);
*a = leafa;
*b = leafb;
*split_key = sp_key;
}
/*
* EFFECT:
* - flush in background thread
* ENTER:
* - parent is already locked
* EXIT:
* - nodes are all unlocked
*/
void tree_flush_node_on_background(struct tree *t, struct node *parent)
{
LOG;
int childnum;
enum reactivity re;
struct node *child;
struct partition *part;
nassert(parent->height > 0);
childnum = node_find_heaviest_idx(parent);
part = &parent->parts[childnum];
/* pin the child */
if (cache_get_and_pin(t->cf, part->child_nid, (void**)&child, L_WRITE) != NESS_OK) {
__ERROR("cache get node error, nid [%" PRIu64 "]", part->child_nid);
return;
}
re = get_reactivity(t, child);
if (re == STABLE) {
/* detach buffer from parent */
struct nmb *buf = part->ptr.u.nonleaf->buffer;
node_set_dirty(parent);
part->ptr.u.nonleaf->buffer = nmb_new(t->e);
/* flush it in background thread */
_place_node_and_buffer_on_background(t, child, buf);
cache_unpin(t->cf, parent->cpair, make_cpair_attr(parent));
} else {
/* the child is reactive, we deal it in main thread */
_child_maybe_reactivity(t, parent, child);
}
}
/*
* EFFECT:
* - do flush in a background thread
* PROCESS:
* - if buf is NULL, we will do _flush_some_child
* - if buf is NOT NULL, we will do _flush_buffer_to_child
* ENTER:
* - fe->node is already locked
* EXIT:
* - nodes are unlocked
*/
static void _flush_node_func(void *fe)
{
enum reactivity re;
struct flusher_extra *extra = (struct flusher_extra*)fe;
struct tree *t = extra->tree;
struct node *n = extra->node;
struct nmb *buf = extra->buffer;
node_set_dirty(n);
if (buf) {
_flush_buffer_to_child(t, n, buf);
nmb_free(buf);
/* check the child node */
re = get_reactivity(t, n);
if (re == FLUSHBLE)
_flush_some_child(t, n);
else
cache_unpin(t->cf, n->cpair, make_cpair_attr(n));
} else {
/* we want flush some buffer from n */
_flush_some_child(t, n);
}
xfree(extra);
}
/*
* apply a msg to leaf basement
* neither the leaf type is LE_CLEAN or LE_MVCC
* basement will always keep multi-snapshot
* the difference between LE_CLEAN and LE_MVCC is gc affects.
*
*/
int leaf_apply_msg(struct node *leaf, struct bt_cmd *cmd)
{
write_lock(&leaf->u.l.le->rwlock);
switch (cmd->type & 0xff) {
case MSG_INSERT:
case MSG_DELETE:
case MSG_UPDATE: {
/* TODO: do msg update, node->node_op->update */
}
case MSG_COMMIT:
case MSG_ABORT:
default:
basement_put(leaf->u.l.le->bsm,
cmd->msn,
cmd->type,
cmd->key,
cmd->val,
&cmd->xidpair);
}
leaf->msn = cmd->msn > leaf->msn ? cmd->msn : leaf->msn;
node_set_dirty(leaf);
write_unlock(&leaf->u.l.le->rwlock);
return NESS_OK;
}
/*
* EFFECT:
* - put cmd to leaf
* ENTER:
* - leaf is already locked(L_WRITE)
* EXITS:
* - leaf is locked
*/
void leaf_put_cmd(struct node *leaf, struct bt_cmd *cmd)
{
LOG;
leaf_apply_msg(leaf, cmd);
leaf->msn = cmd->msn > leaf->msn ? cmd->msn : leaf->msn;
node_set_dirty(leaf);
}
/*
*
* PROCESS:
* - if the layout as follows(3 pivots, 4 partitions):
*
* +--------+--------+--------+
* | 15 | 17 | 19 | +∞
* +--------+--------+--------+\
* pidx0 pidx1 pidx2 pidx3
*
*
* - so if spk is 16, pidx = 1, after added(4 pivtos, 5 partitions):
*
* +--------+--------+--------+--------+
* | 15 | [16] | 17 | 19 | +∞
* +--------+--------+--------+--------+\
* pidx0 [pidx1] pidx2 pidx3 pidx4
*
*
* ENTER:
* - parent is already locked(L_WRITE)
* - node a is already locked(L_WRITE)
* - node b is already locked(L_WRITE)
*/
static void _add_pivot_to_parent(struct tree *t,
struct node *parent,
int child_num,
struct node *a,
struct node *b,
struct msg *spk)
{
int i;
int pidx;
struct child_pointer *cptr;
pidx = child_num;
parent->pivots = xrealloc(parent->pivots, parent->n_children * PIVOT_SIZE);
parent->parts = xrealloc(parent->parts, (parent->n_children + 1) * PART_SIZE);
/* slide pivots */
for (i = (parent->n_children - 1); i > pidx; i--)
parent->pivots[i] = parent->pivots[i - 1];
/* slide parts */
for (i = parent->n_children; i > pidx; i--)
parent->parts[i] = parent->parts[i - 1];
/* new part */
msgcpy(&parent->pivots[pidx], spk);
parent->parts[pidx].child_nid = a->nid;
cptr = &parent->parts[pidx].ptr;
cptr->u.nonleaf = create_nonleaf(t->e);
parent->parts[pidx + 1].child_nid = b->nid;
parent->n_children += 1;
node_set_dirty(parent);
status_increment(&t->e->status->tree_add_pivots_nums);
}
/*
* apply a msg to leaf msgbuf
* neither the leaf type is LE_CLEAN or LE_MVCC
* msgbuf will always keep multi-snapshot
* the difference between LE_CLEAN and LE_MVCC is gc affects.
*
*/
int leaf_put(struct node *leaf, struct bt_cmd *cmd)
{
int ret = NESS_ERR;
struct lmb *buffer = (struct lmb*)leaf->parts[0].msgbuf;
switch (cmd->type & 0xff) {
case MSG_INSERT:
lmb_put(buffer,
cmd->msn,
cmd->type,
cmd->key,
cmd->val,
&cmd->xidpair);
return NESS_OK;
break;
case MSG_DELETE:
break;
case MSG_UPDATE:
break;
case MSG_COMMIT:
break;
case MSG_ABORT:
break;
default:
break;
}
leaf->msn = cmd->msn > leaf->msn ? cmd->msn : leaf->msn;
node_set_dirty(leaf);
return ret;
}
void leaf_new(struct hdr *hdr,
NID nid,
uint32_t height,
uint32_t children,
struct node **n)
{
struct node *node;
nassert(height == 0);
nassert(children == 1);
node = xcalloc(1, sizeof(*node));
nassert(node);
node->nid = nid;
node->height = height;
ness_mutex_init(&node->attr.mtx);
ness_mutex_init(&node->mtx);
ness_rwlock_init(&node->rwlock, &node->mtx);
node->n_children = children;
node->layout_version = hdr->layout_version;
if (children > 0) {
node->pivots = xcalloc(children - 1, PIVOT_SIZE);
node->parts = xcalloc(children, PART_SIZE);
}
node->opts = hdr->opts;
node->i = &leaf_operations;
node_set_dirty(node);
*n = node;
}
void arrange_workspace(struct sway_workspace *workspace) {
if (config->reloading) {
return;
}
if (!workspace->output) {
// Happens when there are no outputs connected
return;
}
struct sway_output *output = workspace->output;
struct wlr_box *area = &output->usable_area;
sway_log(SWAY_DEBUG, "Usable area for ws: %dx%d@%d,%d",
area->width, area->height, area->x, area->y);
workspace_remove_gaps(workspace);
bool first_arrange = workspace->width == 0 && workspace->height == 0;
double prev_x = workspace->x;
double prev_y = workspace->y;
workspace->width = area->width;
workspace->height = area->height;
workspace->x = output->wlr_output->lx + area->x;
workspace->y = output->wlr_output->ly + area->y;
// Adjust any floating containers
double diff_x = workspace->x - prev_x;
double diff_y = workspace->y - prev_y;
if (!first_arrange && (diff_x != 0 || diff_y != 0)) {
for (int i = 0; i < workspace->floating->length; ++i) {
struct sway_container *floater = workspace->floating->items[i];
container_floating_translate(floater, diff_x, diff_y);
double center_x = floater->x + floater->width / 2;
double center_y = floater->y + floater->height / 2;
struct wlr_box workspace_box;
workspace_get_box(workspace, &workspace_box);
if (!wlr_box_contains_point(&workspace_box, center_x, center_y)) {
container_floating_move_to_center(floater);
}
}
}
workspace_add_gaps(workspace);
node_set_dirty(&workspace->node);
sway_log(SWAY_DEBUG, "Arranging workspace '%s' at %f, %f", workspace->name,
workspace->x, workspace->y);
if (workspace->fullscreen) {
struct sway_container *fs = workspace->fullscreen;
fs->x = output->lx;
fs->y = output->ly;
fs->width = output->width;
fs->height = output->height;
arrange_container(fs);
} else {
struct wlr_box box;
workspace_get_box(workspace, &box);
arrange_children(workspace->tiling, workspace->layout, &box);
arrange_floating(workspace->floating);
}
}
/*
* EFFECT:
* - alloc node header
*/
struct node *node_alloc_empty(NID nid, int height) {
struct node *node;
node = xcalloc(1, sizeof(*node));
node->nid = nid;
node->height = height;
node->node_op = &nop;
mutex_init(&node->attr.mtx);
node_set_dirty(node);
return node;
}
/*
* EFFECT:
* - put cmd to nonleaf
* ENTER:
* - node is already locked (L_READ)
* EXITS:
* - node is locked
*/
void nonleaf_put_cmd(struct node *node, struct bt_cmd *cmd)
{
uint32_t pidx;
struct nmb *buffer;
pidx = node_partition_idx(node, cmd->key);
buffer = node->parts[pidx].ptr.u.nonleaf->buffer;
nassert(buffer);
nmb_put(buffer,
cmd-> msn,
cmd->type,
cmd->key,
cmd->val,
&cmd->xidpair);
node->msn = cmd->msn > node->msn ? cmd->msn : node->msn;
node_set_dirty(node);
}
/*
* +-----------------------------------------------+
* | 5 | 7 | 9 |
* +-----------------------------------------------+
* |
* +---------------+
* | 60 | 61 | 62 |
* +---------------+
*
* +---------------------------------------------------------------+
* | 5 | 60 | 7 | 9 |
* +---------------------------------------------------------------+
* | |
* +--------+ +---------+
* | 60 | | 61 | 62 |
* +--------+ +---------+
*
*
* ENTER:
* - node is already locked(L_WRITE)
* EXITS:
* - a is locked(L_WRITE)
* - b is locked(L_WRITE)
*/
static void _node_split(struct tree *t,
struct node *node,
struct node **a,
struct node **b,
struct msg **split_key)
{
int i;
int pivots_old;
int pivots_in_a;
int pivots_in_b;
struct node *nodea;
struct node *nodeb;
struct msg *spk;
__DEBUG("nonleaf split begin, NID %"PRIu64""
", nodesz %d"
", nodec %d"
", children %d"
, node->nid
, node_size(node)
, node_count(node)
, node->n_children);
nodea = node;
pivots_old = node->n_children - 1;
nassert(pivots_old > 2);
pivots_in_a = pivots_old / 2;
pivots_in_b = pivots_old - pivots_in_a;
/* node a */
nodea->n_children = pivots_in_a + 1;
/* node b */
NID nid = hdr_next_nid(t->hdr);
node_create_light(nid, node->height > 0 ? 1 : 0, pivots_in_b + 1, t->hdr->version, t->e, &nodeb);
cache_put_and_pin(t->cf, nid, nodeb);
for (i = 0; i < (pivots_in_b); i++)
nodeb->pivots[i] = nodea->pivots[pivots_in_a + i];
for (i = 0; i < (pivots_in_b + 1); i++)
nodeb->parts[i] = nodea->parts[pivots_in_a + i];
/* the rightest partition of nodea */
struct child_pointer *ptr = &nodea->parts[pivots_in_a].ptr;
if (nodea->height > 0)
ptr->u.nonleaf = create_nonleaf(t->e);
else
ptr->u.leaf = create_leaf(t->e);
/* split key */
spk = msgdup(&node->pivots[pivots_in_a - 1]);
node_set_dirty(nodea);
node_set_dirty(nodeb);
__DEBUG("nonleaf split end, nodea NID %"PRIu64""
", nodesz %d"
", nodec %d"
", children %d"
, nodea->nid
, node_size(nodea)
, node_count(nodea)
, nodea->n_children);
__DEBUG("nonleaf split end, nodeb NID %"PRIu64""
", nodesz %d"
", nodec %d"
", children %d"
, nodeb->nid
, node_size(nodeb)
, node_count(nodeb)
, nodeb->n_children);
*a = nodea;
*b = nodeb;
*split_key = spk;
}
struct tree *tree_open(const char *dbname,
struct env *e,
struct tree_callback *tcb) {
int fd;
int flag;
mode_t mode;
int is_create = 0;
struct tree *t;
struct node *root;
struct cache_file *cf;
t = xcalloc(1, sizeof(*t));
t->e = e;
mode = S_IRWXU | S_IRWXG | S_IRWXO;
flag = O_RDWR | O_BINARY;
if (e->use_directio)
fd = ness_os_open_direct(dbname, flag, mode);
else
fd = ness_os_open(dbname, flag, mode);
if (fd == -1) {
if (e->use_directio)
fd = ness_os_open(dbname, flag | O_CREAT, mode);
else
fd = ness_os_open_direct(dbname, flag | O_CREAT, mode);
if (fd == -1)
goto ERR;
is_create = 1;
}
t->fd = fd;
t->hdr = hdr_new(e);
/* tree header */
if (!is_create) {
tcb->fetch_hdr_cb(fd, t->hdr);
}
/* create cache file */
cf = cache_file_create(e->cache, t->fd, t->hdr, tcb);
t->cf = cf;
/* tree root node */
if (is_create) {
NID nid = hdr_next_nid(t->hdr);
node_create(nid, 0, 1, t->hdr->version, t->e, &root);
cache_put_and_pin(cf, nid, root);
root->isroot = 1;
node_set_dirty(root);
cache_unpin(cf, root->cpair, make_cpair_attr(root));
t->hdr->root_nid = root->nid;
__DEBUG("create new root, NID %"PRIu64, root->nid);
} else {
/* get the root node */
if (cache_get_and_pin(cf, t->hdr->root_nid, (void**)&root, L_READ) != NESS_OK)
__PANIC("get root from cache error [%" PRIu64 "]", t->hdr->root_nid);
root->isroot = 1;
cache_unpin(cf, root->cpair, make_cpair_attr(root));
__DEBUG("fetch root, NID %"PRIu64, root->nid);
}
return t;
ERR:
xfree(t);
return NESS_ERR;
}
void _flush_buffer_to_child(struct tree *t, struct node *child, struct nmb *buf)
{
struct mb_iter iter;
mb_iter_init(&iter, buf->pma);
while (mb_iter_next(&iter)) {
/* TODO(BohuTANG): check msn */
struct nmb_values nvalues;
nmb_get_values(&iter, &nvalues);
struct bt_cmd cmd = {
.msn = nvalues.msn,
.type = nvalues.type,
.key = &nvalues.key,
.val = &nvalues.val,
.xidpair = nvalues.xidpair
};
node_put_cmd(t, child, &cmd);
}
}
void _flush_some_child(struct tree *t, struct node *parent);
/*
* PROCESS:
* - check child reactivity
* - if FISSIBLE: split child
* - if FLUSHBLE: flush buffer from child
* ENTER:
* - parent is already locked
* - child is already locked
* EXIT:
* - parent is unlocked
* - no nodes are locked
*/
void _child_maybe_reactivity(struct tree *t, struct node *parent, struct node *child)
{
enum reactivity re = get_reactivity(t, child);
switch (re) {
case STABLE:
cache_unpin(t->cf, child->cpair, make_cpair_attr(child));
cache_unpin(t->cf, parent->cpair, make_cpair_attr(parent));
break;
case FISSIBLE:
node_split_child(t, parent, child);
cache_unpin(t->cf, child->cpair, make_cpair_attr(child));
cache_unpin(t->cf, parent->cpair, make_cpair_attr(parent));
break;
case FLUSHBLE:
cache_unpin(t->cf, parent->cpair, make_cpair_attr(parent));
_flush_some_child(t, child);
break;
}
}
/*
* PROCESS:
* - pick a heaviest child of parent
* - flush from parent to child
* - maybe split/flush child recursively
* ENTER:
* - parent is already locked
* EXIT:
* - parent is unlocked
* - no nodes are locked
*/
void _flush_some_child(struct tree *t, struct node *parent)
{
int childnum;
enum reactivity re;
struct node *child;
struct partition *part;
struct nmb *buffer;
struct timespec t1, t2;
childnum = node_find_heaviest_idx(parent);
nassert(childnum < parent->n_children);
part = &parent->parts[childnum];
buffer = part->ptr.u.nonleaf->buffer;
if (cache_get_and_pin(t->cf, part->child_nid, (void**)&child, L_WRITE) != NESS_OK) {
__ERROR("cache get node error, nid [%" PRIu64 "]", part->child_nid);
return;
}
ngettime(&t1);
re = get_reactivity(t, child);
if (re == STABLE) {
node_set_dirty(parent);
part->ptr.u.nonleaf->buffer = nmb_new(t->e);
_flush_buffer_to_child(t, child, buffer);
nmb_free(buffer);
}
ngettime(&t2);
status_add(&t->e->status->tree_flush_child_costs, time_diff_ms(t1, t2));
status_increment(&t->e->status->tree_flush_child_nums);
_child_maybe_reactivity(t, parent, child);
}
/*
* EFFECT:
* - split leaf&lmb into two leaves:a & b
* a&b are both the half of the lmb
*
* PROCESS:
* - leaf:
* +-----------------------------------+
* | 0 | 1 | 2 | 3 | 4 | 5 |
* +-----------------------------------+
*
* - split:
* root
* +--------+
* | 2 |
* +--------+
* / \
* +-----------------+ +------------------+
* | 0 | 1 | 2 | | 3 | 4 | 5 |
* +-----------------+ +------------------+
* nodea nodeb
*
* ENTER:
* - leaf is already locked (L_WRITE)
* EXITS:
* - a is locked
* - b is locked
*/
static void _leaf_and_lmb_split(struct tree *t,
struct node *leaf,
struct node **a,
struct node **b,
struct msg **split_key)
{
struct child_pointer *cptra;
struct child_pointer *cptrb;
struct node *leafa;
struct node *leafb;
struct lmb *mb;
struct lmb *mba;
struct lmb *mbb;
struct msg *sp_key = NULL;
__DEBUG("leaf split begin, NID %"PRIu64""
", nodesz %d"
", nodec %d"
", children %d"
, leaf->nid
, node_size(leaf)
, node_count(leaf)
, leaf->n_children);
leafa = leaf;
cptra = &leafa->parts[0].ptr;
/* split lmb of leaf to mba & mbb */
mb = cptra->u.leaf->buffer;
lmb_split(mb, &mba, &mbb, &sp_key);
lmb_free(mb);
/* reset leafa buffer */
cptra->u.leaf->buffer = mba;
/* new leafb */
NID nid = hdr_next_nid(t->hdr);
node_create(nid, 0, 1, t->hdr->version, t->e, &leafb);
cache_put_and_pin(t->cf, nid, leafb);
cptrb = &leafb->parts[0].ptr;
lmb_free(cptrb->u.leaf->buffer);
cptrb->u.leaf->buffer = mbb;
/* set dirty */
node_set_dirty(leafa);
node_set_dirty(leafb);
__DEBUG("leaf split end, leafa NID %"PRIu64""
", nodesz %d"
", nodec %d"
", children %d"
, leafa->nid
, node_size(leafa)
, node_count(leafa)
, leafa->n_children);
__DEBUG("leaf split end, leafb NID %"PRIu64""
", nodesz %d"
", nodec %d"
", children %d"
, leafb->nid
, node_size(leafb)
, node_count(leafb)
, leafb->n_children);
*a = leafa;
*b = leafb;
*split_key = sp_key;
status_increment(&t->e->status->tree_leaf_split_nums);
}
