static inline int
si_getresult(siread *q, sv *v, int compare)
{
int rc;
if (compare) {
rc = sr_compare(q->r->scheme, sv_pointer(v), sv_size(v),
q->key, q->keysize);
if (ssunlikely(rc != 0))
return 0;
}
if (q->prefix) {
rc = sr_compareprefix(q->r->scheme,
q->prefix,
q->prefixsize,
sv_pointer(v), sv_size(v));
if (ssunlikely(! rc))
return 0;
}
if (ssunlikely(q->has))
return sv_lsn(v) > q->vlsn;
if (ssunlikely(sv_is(v, SVDELETE)))
return 2;
rc = si_readdup(q, v);
if (ssunlikely(rc == -1))
return -1;
return 1;
}
int sx_get(sx *t, sxindex *index, sv *key, sv *result)
{
sxmanager *m = t->manager;
ssrbnode *n = NULL;
int rc = sx_match(&index->i, index->scheme,
sv_pointer(key),
sv_size(key), &n);
if (! (rc == 0 && n)) {
rc = 0;
goto done;
}
sxv *head = sscast(n, sxv, node);
sxv *v = sx_vmatch(head, t->id);
if (v == NULL) {
rc = 0;
goto done;
}
if (ssunlikely((v->v->flags & SVDELETE) > 0)) {
rc = 2;
goto done;
}
sv vv;
sv_init(&vv, &sv_vif, v->v, NULL);
svv *ret = sv_vdup(m->a, &vv);
if (ssunlikely(ret == NULL)) {
rc = sr_oom(index->r->e);
} else {
sv_init(result, &sv_vif, ret, NULL);
rc = 1;
}
done:
return rc;
}
int sd_buildadd(sdbuild *b, sr *r, sv *v, uint32_t flags)
{
/* prepare metadata reference */
int rc = sr_bufensure(&b->k, r->a, sizeof(sdv));
if (srunlikely(rc == -1))
return sr_error(r->e, "%s", "memory allocation failed");
sdpageheader *h = sd_buildheader(b);
sdv *sv = (sdv*)b->k.p;
sv->lsn = sv_lsn(v);
sv->flags = sv_flags(v) | flags;
sv->size = sv_size(v);
sv->offset = sr_bufused(&b->v) - sd_buildref(b)->v;
/* copy object */
rc = sr_bufensure(&b->v, r->a, sv->size);
if (srunlikely(rc == -1))
return sr_error(r->e, "%s", "memory allocation failed");
memcpy(b->v.p, sv_pointer(v), sv->size);
sr_bufadvance(&b->v, sv->size);
sr_bufadvance(&b->k, sizeof(sdv));
/* update page header */
h->count++;
h->size += sv->size + sizeof(sdv);
if (sv->lsn > h->lsnmax)
h->lsnmax = sv->lsn;
if (sv->lsn < h->lsnmin)
h->lsnmin = sv->lsn;
if (sv->flags & SVDUP) {
h->countdup++;
if (sv->lsn < h->lsnmindup)
h->lsnmindup = sv->lsn;
}
return 0;
}
int
songvec_delete(struct songvec *sv, const struct song *del)
{
size_t i;
g_mutex_lock(nr_lock);
for (i = 0; i < sv->nr; ++i) {
if (sv->base[i] != del)
continue;
/* we _don't_ call song_free() here */
if (!--sv->nr) {
g_free(sv->base);
sv->base = NULL;
} else {
memmove(&sv->base[i], &sv->base[i + 1],
(sv->nr - i) * sizeof(struct song *));
sv->base = g_realloc(sv->base, sv_size(sv));
}
g_mutex_unlock(nr_lock);
return i;
}
g_mutex_unlock(nr_lock);
return -1; /* not found */
}
sxstate sx_setstmt(sxmanager *m, sxindex *index, sv *v)
{
sr_seq(m->seq, SR_TSNNEXT);
ssrbnode *n = NULL;
int rc = sx_match(&index->i, index->scheme, sv_pointer(v), sv_size(v), &n);
if (rc == 0 && n)
return SXLOCK;
return SXCOMMIT;
}
void
songvec_add(struct songvec *sv, struct song *add)
{
g_mutex_lock(nr_lock);
++sv->nr;
sv->base = g_realloc(sv->base, sv_size(sv));
sv->base[sv->nr - 1] = add;
g_mutex_unlock(nr_lock);
}
ss_t *ss_cpy_sub(ss_t **s, const ss_t *src, const sv_t *offs, const size_t nth)
{
ASSERT_RETURN_IF(!s, ss_void);
ASSERT_RETURN_IF((!src || !offs), ss_clear(s)); /* BEHAVIOR: empty */
const size_t elems = sv_size(offs) / 2;
ASSERT_RETURN_IF(nth >= elems, ss_clear(s)); /* BEHAVIOR: empty */
const size_t off = (size_t)sv_u_at(offs, nth * 2);
const size_t size = (size_t)sv_u_at(offs, nth * 2 + 1);
return ss_cpy_substr(s, src, off, size);
}
static int
si_readcommited_branch(sr *r, sibranch *b, sv *v)
{
ssiter i;
ss_iterinit(sd_indexiter, &i);
ss_iteropen(sd_indexiter, &i, r, &b->index, SS_GTE,
sv_pointer(v), sv_size(v));
sdindexpage *page = ss_iterof(sd_indexiter, &i);
if (page == NULL)
return 0;
return page->lsnmax >= sv_lsn(v);
}
static inline int
se_txwrite(setx *t, sev *o, uint8_t flags)
{
se *e = se_of(&t->o);
sedb *db = se_cast(o->o.parent, sedb*, SEDB);
/* validate req */
if (ssunlikely(t->t.state == SXPREPARE)) {
sr_error(&e->error, "%s", "transaction is in 'prepare' state (read-only)");
goto error;
}
/* validate database status */
int status = se_status(&db->status);
switch (status) {
case SE_SHUTDOWN:
if (ssunlikely(! se_dbvisible(db, t->t.id))) {
sr_error(&e->error, "%s", "database is invisible for the transaction");
goto error;
}
break;
case SE_RECOVER:
case SE_ONLINE: break;
default: goto error;
}
if (flags == SVUPDATE && !sf_updatehas(&db->scheme.fmt_update))
flags = 0;
/* prepare object */
svv *v;
int rc = se_dbv(db, o, 0, &v);
if (ssunlikely(rc == -1))
goto error;
v->flags = flags;
v->log = o->log;
sv vp;
sv_init(&vp, &sv_vif, v, NULL);
so_destroy(&o->o);
/* ensure quota */
int size = sizeof(svv) + sv_size(&vp);
ss_quota(&e->quota, SS_QADD, size);
/* concurrent index only */
rc = sx_set(&t->t, &db->coindex, v);
if (ssunlikely(rc == -1)) {
ss_quota(&e->quota, SS_QREMOVE, size);
return -1;
}
return 0;
error:
so_destroy(&o->o);
return -1;
}
static inline int
sd_buildadd_raw(sdbuild *b, sr *r, sv *v)
{
uint64_t lsn = sv_lsn(v);
uint32_t size = sv_size(v);
uint32_t sizemeta = ss_leb128size(size) + ss_leb128size(lsn);
int rc = ss_bufensure(&b->v, r->a, sizemeta + size);
if (ssunlikely(rc == -1))
return sr_oom(r->e);
ss_bufadvance(&b->v, ss_leb128write(b->v.p, size));
ss_bufadvance(&b->v, ss_leb128write(b->v.p, lsn));
memcpy(b->v.p, sv_pointer(v), size);
ss_bufadvance(&b->v, size);
return 0;
}
int se_reqread(sereq *r)
{
sereqarg *arg = &r->arg;
sedb *db = (sedb*)r->db;
uint32_t keysize;
void *key;
if (sslikely(arg->v.v)) {
keysize = sv_size(&arg->v);
key = sv_pointer(&arg->v);
} else {
keysize = 0;
key = NULL;
}
char *prefix;
uint32_t prefixsize;
if (arg->vprefix.v) {
void *vptr = sv_vpointer(arg->vprefix.v);
prefix = sf_key(vptr, 0);
prefixsize = sf_keysize(vptr, 0);
} else {
prefix = NULL;
prefixsize = 0;
}
if (sslikely(arg->vlsn_generate))
arg->vlsn = sr_seq(db->r.seq, SR_LSN);
sitx x;
si_begin(&x, &db->index, 1);
siread q;
si_readopen(&q, &x, arg->cache,
arg->order,
arg->vlsn,
prefix,
prefixsize, key, keysize);
if (arg->update)
si_readupdate(&q, &arg->vup, arg->update_eq);
if (arg->cache_only)
si_readcache_only(&q);
if (arg->has)
si_readhas(&q);
r->rc = si_read(&q);
r->read_disk = q.read_disk;
r->read_cache = q.read_cache;
r->v = q.result.v;
si_readclose(&q);
si_commit(&x);
return r->rc;
}
static void
svv_kv(stc *cx)
{
sra a;
sr_aopen(&a, &sr_stda);
srkey cmp;
sr_keyinit(&cmp);
srkeypart *part = sr_keyadd(&cmp, &a);
t( sr_keypart_setname(part, &a, "key") == 0 );
t( sr_keypart_set(part, &a, "u32") == 0 );
sr r;
sr_init(&r, NULL, &a, NULL, SR_FKV, &cmp, NULL, NULL, NULL);
uint32_t key = 123;
uint32_t value = 321;
srformatv pv;
pv.key = (char*)&key;
pv.r.size = sizeof(key);
pv.r.offset = 0;
svv *vv = sv_vbuild(&r, &pv, 1, (char*)&value, sizeof(value));
t( vv != NULL );
vv->flags = SVSET;
vv->lsn = 10;
sv v;
sv_init(&v, &sv_vif, vv, NULL);
t( sv_flags(&v) == SVSET );
t( sv_lsn(&v) == 10 );
sv_lsnset(&v, 8);
t( sv_lsn(&v) == 8 );
t( *(uint32_t*)sr_formatkey(sv_pointer(&v), 0) == key );
t( sr_formatkey_size(sv_pointer(&v), 0) == sizeof(key) );
t( *(uint32_t*)sr_formatvalue(SR_FKV, &cmp, sv_pointer(&v)) == value );
t( sr_formatvalue_size(SR_FKV, &cmp, sv_pointer(&v), sv_size(&v) ) == sizeof(value) );
sv_vfree(&a, vv);
sr_keyfree(&cmp, &a);
}
int si_readcommited(si *index, sr *r, sv *v)
{
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, r, index, SS_GTE,
sv_pointer(v), sv_size(v));
sinode *node;
node = ss_iterof(si_iter, &i);
assert(node != NULL);
sibranch *b = node->branch;
int rc;
while (b) {
rc = si_readcommited_branch(r, b, v);
if (rc)
return 1;
b = b->next;
}
rc = si_readcommited_branch(r, &node->self, v);
return rc;
}
int sd_buildadd(sdbuild *b, sr *r, sv *v, uint32_t flags)
{
/* prepare object metadata */
int rc = ss_bufensure(&b->m, r->a, sizeof(sdv));
if (ssunlikely(rc == -1))
return sr_oom(r->e);
sdpageheader *h = sd_buildheader(b);
sdv *sv = (sdv*)b->m.p;
sv->flags = sv_flags(v) | flags;
sv->offset = ss_bufused(&b->v) - sd_buildref(b)->v;
ss_bufadvance(&b->m, sizeof(sdv));
/* copy object */
switch (r->fmt_storage) {
case SF_SKEYVALUE:
rc = sd_buildadd_keyvalue(b, r, v);
break;
case SF_SRAW:
rc = sd_buildadd_raw(b, r, v);
break;
}
if (ssunlikely(rc == -1))
return -1;
/* update page header */
h->count++;
uint32_t size = sizeof(sdv) + sv_size(v) +
sizeof(sfref) * r->scheme->count;
if (size > b->vmax)
b->vmax = size;
uint64_t lsn = sv_lsn(v);
if (lsn > h->lsnmax)
h->lsnmax = lsn;
if (lsn < h->lsnmin)
h->lsnmin = lsn;
if (sv->flags & SVDUP) {
h->countdup++;
if (lsn < h->lsnmindup)
h->lsnmindup = lsn;
}
return 0;
}
int sx_get(sx *x, sxindex *index, sv *key, sv *result)
{
sxmanager *m = x->manager;
ssrbnode *n = NULL;
int rc;
rc = sx_match(&index->i, index->r->scheme,
sv_pointer(key),
sv_size(key),
&n);
if (! (rc == 0 && n))
goto add;
sxv *head = sscast(n, sxv, node);
sxv *v = sx_vmatch(head, x->id);
if (v == NULL)
goto add;
if (ssunlikely((v->v->flags & SVGET) > 0))
return 0;
if (ssunlikely((v->v->flags & SVDELETE) > 0))
return 2;
sv vv;
sv_init(&vv, &sv_vif, v->v, NULL);
svv *ret = sv_vdup(m->r, &vv);
if (ssunlikely(ret == NULL)) {
rc = sr_oom(m->r->e);
} else {
sv_init(result, &sv_vif, ret, NULL);
rc = 1;
}
return rc;
add:
/* track a start of the latest read sequence in the
* transactional log */
if (x->log_read == -1)
x->log_read = sv_logcount(&x->log);
rc = sx_set(x, index, key->v);
if (ssunlikely(rc == -1))
return -1;
sv_vref((svv*)key->v);
return 0;
}
static inline void
si_qrangebranch(siquery *q, sinode *n, sibranch *b, svmerge *m)
{
sicachebranch *cb = si_cachefollow(q->cache);
assert(cb->branch == b);
/* iterate cache */
if (ss_iterhas(si_read, &cb->i)) {
svmergesrc *s = sv_mergeadd(m, &cb->i);
q->index->read_cache++;
s->ptr = cb;
return;
}
if (cb->open) {
return;
}
cb->open = 1;
sireadarg arg = {
.scheme = q->index->scheme,
.index = q->index,
.n = n,
.b = b,
.buf = &cb->buf_a,
.buf_xf = &cb->buf_b,
.buf_read = &q->index->readbuf,
.index_iter = &cb->index_iter,
.page_iter = &cb->page_iter,
.vlsn = q->vlsn,
.has = 0,
.mmap_copy = 1,
.o = q->order,
.r = q->r
};
ss_iterinit(si_read, &cb->i);
int rc = ss_iteropen(si_read, &cb->i, &arg, q->key, q->keysize);
if (ssunlikely(rc == -1))
return;
if (ssunlikely(! ss_iterhas(si_read, &cb->i)))
return;
svmergesrc *s = sv_mergeadd(m, &cb->i);
s->ptr = cb;
}
static inline int
si_qrange(siquery *q)
{
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, q->r, q->index, q->order, q->key, q->keysize);
sinode *node;
next_node:
node = ss_iterof(si_iter, &i);
if (ssunlikely(node == NULL))
return 0;
/* prepare sources */
svmerge *m = &q->merge;
int count = node->branch_count + 2 + 1;
int rc = sv_mergeprepare(m, q->r, count);
if (ssunlikely(rc == -1)) {
sr_errorreset(q->r->e);
return -1;
}
/* external source (update) */
svmergesrc *s;
sv upbuf_reserve;
ssbuf upbuf;
if (ssunlikely(q->update_v && q->update_v->v)) {
ss_bufinit_reserve(&upbuf, &upbuf_reserve, sizeof(upbuf_reserve));
ss_bufadd(&upbuf, NULL, (void*)&q->update_v, sizeof(sv*));
s = sv_mergeadd(m, NULL);
ss_iterinit(ss_bufiterref, &s->src);
ss_iteropen(ss_bufiterref, &s->src, &upbuf, sizeof(sv*));
}
/* in-memory indexes */
svindex *second;
svindex *first = si_nodeindex_priority(node, &second);
if (first->count) {
s = sv_mergeadd(m, NULL);
ss_iterinit(sv_indexiter, &s->src);
ss_iteropen(sv_indexiter, &s->src, q->r, first, q->order,
q->key, q->keysize);
}
if (ssunlikely(second && second->count)) {
s = sv_mergeadd(m, NULL);
ss_iterinit(sv_indexiter, &s->src);
ss_iteropen(sv_indexiter, &s->src, q->r, second, q->order,
q->key, q->keysize);
}
/* cache and branches */
rc = si_cachevalidate(q->cache, node);
if (ssunlikely(rc == -1)) {
sr_oom(q->r->e);
return -1;
}
sibranch *b = node->branch;
while (b) {
si_qrangebranch(q, node, b, m);
b = b->next;
}
/* merge and filter data stream */
ssiter j;
ss_iterinit(sv_mergeiter, &j);
ss_iteropen(sv_mergeiter, &j, q->r, m, q->order);
ssiter k;
ss_iterinit(sv_readiter, &k);
ss_iteropen(sv_readiter, &k, q->r, &j, &q->index->u, q->vlsn, 0);
sv *v = ss_iterof(sv_readiter, &k);
if (ssunlikely(v == NULL)) {
sv_mergereset(&q->merge);
ss_iternext(si_iter, &i);
goto next_node;
}
rc = 1;
/* convert update search to SS_EQ */
if (q->update_eq) {
rc = sr_compare(q->r->scheme, sv_pointer(v), sv_size(v),
q->key, q->keysize);
rc = rc == 0;
}
/* do prefix search */
if (q->prefix && rc) {
rc = sr_compareprefix(q->r->scheme, q->prefix, q->prefixsize,
sv_pointer(v),
sv_size(v));
}
if (sslikely(rc == 1)) {
if (ssunlikely(si_querydup(q, v) == -1))
return -1;
}
/* skip a possible duplicates from data sources */
ss_iternext(sv_readiter, &k);
return rc;
}
static inline int
si_rangebranch(siread *q, sinode *n, sibranch *b, svmerge *m)
{
sicachebranch *c = si_cachefollow(q->cache);
assert(c->branch == b);
/* iterate cache */
if (ss_iterhas(sd_read, &c->i)) {
svmergesrc *s = sv_mergeadd(m, &c->i);
si_readstat(q, 1, n, 1);
s->ptr = c;
return 1;
}
if (c->open) {
return 1;
}
if (q->cache_only) {
return 2;
}
c->open = 1;
/* choose compression type */
int compression;
ssfilterif *compression_if;
if (! si_branchis_root(b)) {
compression = q->index->scheme->compression_branch;
compression_if = q->index->scheme->compression_branch_if;
} else {
compression = q->index->scheme->compression;
compression_if = q->index->scheme->compression_if;
}
sdreadarg arg = {
.index = &b->index,
.buf = &c->buf_a,
.buf_xf = &c->buf_b,
.buf_read = &q->index->readbuf,
.index_iter = &c->index_iter,
.page_iter = &c->page_iter,
.use_memory = n->in_memory,
.use_mmap = q->index->scheme->mmap,
.use_mmap_copy = 1,
.use_compression = compression,
.compression_if = compression_if,
.has = 0,
.has_vlsn = 0,
.o = q->order,
.memory = &b->copy,
.mmap = &n->map,
.file = &n->file,
.r = q->r
};
ss_iterinit(sd_read, &c->i);
int rc = ss_iteropen(sd_read, &c->i, &arg, q->key, q->keysize);
int reads = sd_read_stat(&c->i);
si_readstat(q, 0, n, reads);
if (ssunlikely(rc == -1))
return -1;
if (ssunlikely(! ss_iterhas(sd_read, &c->i)))
return 0;
svmergesrc *s = sv_mergeadd(m, &c->i);
s->ptr = c;
return 1;
}
static inline int
si_range(siread *q)
{
assert(q->has == 0);
ssiter i;
ss_iterinit(si_iter, &i);
ss_iteropen(si_iter, &i, q->r, q->index, q->order, q->key, q->keysize);
sinode *node;
next_node:
node = ss_iterof(si_iter, &i);
if (ssunlikely(node == NULL))
return 0;
si_txtrack(q->x, node);
/* prepare sources */
svmerge *m = &q->merge;
int count = node->branch_count + 2 + 1;
int rc = sv_mergeprepare(m, q->r, count);
if (ssunlikely(rc == -1)) {
sr_errorreset(q->r->e);
return -1;
}
/* external source (upsert) */
svmergesrc *s;
sv upbuf_reserve;
ssbuf upbuf;
if (ssunlikely(q->upsert_v && q->upsert_v->v)) {
ss_bufinit_reserve(&upbuf, &upbuf_reserve, sizeof(upbuf_reserve));
ss_bufadd(&upbuf, NULL, (void*)&q->upsert_v, sizeof(sv*));
s = sv_mergeadd(m, NULL);
ss_iterinit(ss_bufiterref, &s->src);
ss_iteropen(ss_bufiterref, &s->src, &upbuf, sizeof(sv*));
}
/* in-memory indexes */
svindex *second;
svindex *first = si_nodeindex_priority(node, &second);
if (first->count) {
s = sv_mergeadd(m, NULL);
ss_iterinit(sv_indexiter, &s->src);
ss_iteropen(sv_indexiter, &s->src, q->r, first, q->order,
q->key, q->keysize);
}
if (ssunlikely(second && second->count)) {
s = sv_mergeadd(m, NULL);
ss_iterinit(sv_indexiter, &s->src);
ss_iteropen(sv_indexiter, &s->src, q->r, second, q->order,
q->key, q->keysize);
}
/* cache and branches */
rc = si_cachevalidate(q->cache, node);
if (ssunlikely(rc == -1)) {
sr_oom(q->r->e);
return -1;
}
sibranch *b = node->branch;
while (b) {
rc = si_rangebranch(q, node, b, m);
if (ssunlikely(rc == -1 || rc == 2))
return rc;
b = b->next;
}
/* merge and filter data stream */
ssiter j;
ss_iterinit(sv_mergeiter, &j);
ss_iteropen(sv_mergeiter, &j, q->r, m, q->order);
ssiter k;
ss_iterinit(sv_readiter, &k);
ss_iteropen(sv_readiter, &k, q->r, &j, &q->index->u, q->vlsn, 0);
sv *v = ss_iterof(sv_readiter, &k);
if (ssunlikely(v == NULL)) {
sv_mergereset(&q->merge);
ss_iternext(si_iter, &i);
goto next_node;
}
rc = 1;
/* convert upsert search to SS_EQ */
if (q->upsert_eq) {
rc = sr_compare(q->r->scheme, sv_pointer(v), sv_size(v),
q->key, q->keysize);
rc = rc == 0;
}
/* do prefix search */
if (q->prefix && rc) {
rc = sr_compareprefix(q->r->scheme, q->prefix, q->prefixsize,
sv_pointer(v),
sv_size(v));
}
if (sslikely(rc == 1)) {
if (ssunlikely(si_readdup(q, v) == -1))
return -1;
}
/* skip a possible duplicates from data sources */
ss_iternext(sv_readiter, &k);
return rc;
}