/*
* __session_compact_worker --
* Worker function to do the actual compaction call.
*/
static int
__session_compact_worker(
WT_SESSION *wt_session, const char *uri, const char *config)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, compact, config, cfg);
WT_WITH_SCHEMA_LOCK(session,
ret = __wt_schema_worker(session, uri, __wt_compact, cfg, 0));
err: API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_drop --
* WT_SESSION->drop method.
*/
static int
__session_drop(WT_SESSION *wt_session, const char *uri, const char *config)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, drop, config, cfg);
WT_WITH_SCHEMA_LOCK(session,
ret = __wt_schema_drop(session, uri, cfg));
err: /* Note: drop operations cannot be unrolled (yet?). */
API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_verify --
* WT_SESSION->verify method.
*/
static int
__session_verify(WT_SESSION *wt_session, const char *uri, const char *config)
{
WT_SESSION_IMPL *session;
int ret;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, verify, config, cfg);
ret = __wt_schema_worker(session, uri, cfg,
__wt_verify, WT_BTREE_EXCLUSIVE | WT_BTREE_VERIFY);
err:
API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_rename --
* WT_SESSION->rename method.
*/
static int
__session_rename(WT_SESSION *wt_session,
const char *uri, const char *newname, const char *config)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, rename, config, cfg);
WT_WITH_SCHEMA_LOCK(session,
ret = __wt_schema_rename(session, uri, newname, cfg));
err: API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_upgrade --
* WT_SESSION->upgrade method.
*/
static int
__session_upgrade(WT_SESSION *wt_session, const char *uri, const char *config)
{
WT_SESSION_IMPL *session;
int ret;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, upgrade, config, cfg);
ret = __wt_schema_worker(session, uri, cfg,
__wt_upgrade, WT_BTREE_EXCLUSIVE | WT_BTREE_UPGRADE);
err:
API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_rollback_transaction --
* WT_SESSION->rollback_transaction method.
*/
static int
__session_rollback_transaction(WT_SESSION *wt_session, const char *config)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, rollback_transaction, config, cfg);
WT_CSTAT_INCR(session, txn_rollback);
WT_TRET(__session_reset_cursors(session));
WT_TRET(__wt_txn_rollback(session, cfg));
err: API_END(session);
return (ret);
}
/*
* __session_create --
* WT_SESSION->create method.
*/
static int
__session_create(WT_SESSION *wt_session, const char *uri, const char *config)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, create, config, cfg);
WT_UNUSED(cfg);
/* Disallow objects in the WiredTiger name space. */
WT_ERR(__wt_schema_name_check(session, uri));
WT_WITH_SCHEMA_LOCK(session,
ret = __wt_schema_create(session, uri, config));
err: API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_checkpoint --
* WT_SESSION->checkpoint method.
*/
static int
__session_checkpoint(WT_SESSION *wt_session, const char *config)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
WT_TXN *txn;
session = (WT_SESSION_IMPL *)wt_session;
txn = &session->txn;
WT_CSTAT_INCR(session, checkpoint);
SESSION_API_CALL(session, checkpoint, config, cfg);
/*
* Checkpoints require a snapshot to write a transactionally consistent
* snapshot of the data.
*
* We can't use an application's transaction: if it has uncommitted
* changes, they will be written in the checkpoint and may appear after
* a crash.
*
* Use a real snapshot transaction: we don't want any chance of the
* snapshot being updated during the checkpoint. Eviction is prevented
* from evicting anything newer than this because we track the oldest
* transaction ID in the system that is not visible to all readers.
*/
if (F_ISSET(txn, TXN_RUNNING))
WT_ERR_MSG(session, EINVAL,
"Checkpoint not permitted in a transaction");
/*
* Reset open cursors.
*
* We do this here explicitly even though it will happen implicitly in
* the call to begin_transaction for the checkpoint, in case some
* implementation of WT_CURSOR::reset needs the schema lock.
*/
WT_ERR(__session_reset_cursors(session));
WT_WITH_SCHEMA_LOCK(session,
ret = __wt_txn_checkpoint(session, cfg));
err: API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_begin_transaction --
* WT_SESSION->begin_transaction method.
*/
static int
__session_begin_transaction(WT_SESSION *wt_session, const char *config)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, begin_transaction, config, cfg);
WT_CSTAT_INCR(session, txn_begin);
if (!F_ISSET(S2C(session), WT_CONN_TRANSACTIONAL))
WT_ERR_MSG(session, EINVAL,
"Database not configured for transactions");
if (F_ISSET(&session->txn, TXN_RUNNING))
WT_ERR_MSG(session, EINVAL, "Transaction already running");
WT_ERR(__session_reset_cursors(session));
ret = __wt_txn_begin(session, cfg);
err: API_END(session);
return (ret);
}
/*
* __session_open_cursor --
* WT_SESSION->open_cursor method.
*/
static int
__session_open_cursor(WT_SESSION *wt_session,
const char *uri, WT_CURSOR *to_dup, const char *config, WT_CURSOR **cursorp)
{
WT_SESSION_IMPL *session;
int ret;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, open_cursor, config, cfg);
if (uri != NULL && to_dup != NULL)
WT_ERR_MSG(session, EINVAL,
"should be passed either a URI or a cursor, but not both");
if (to_dup != NULL)
ret = __wt_cursor_dup(session, to_dup, config, cursorp);
else if (WT_PREFIX_MATCH(uri, "colgroup:"))
ret = __wt_curfile_open(session, uri, cfg, cursorp);
else if (WT_PREFIX_MATCH(uri, "config:"))
ret = __wt_curconfig_open(session, uri, cfg, cursorp);
else if (WT_PREFIX_MATCH(uri, "file:"))
ret = __wt_curfile_open(session, uri, cfg, cursorp);
else if (WT_PREFIX_MATCH(uri, "index:"))
ret = __wt_curindex_open(session, uri, cfg, cursorp);
else if (WT_PREFIX_MATCH(uri, "statistics:"))
ret = __wt_curstat_open(session, uri, cfg, cursorp);
else if (WT_PREFIX_MATCH(uri, "table:"))
ret = __wt_curtable_open(session, uri, cfg, cursorp);
else {
__wt_err(session, EINVAL, "Unknown cursor type '%s'", uri);
ret = EINVAL;
}
err:
API_END_NOTFOUND_MAP(session, ret);
}
/*
* __wt_session_compact --
* WT_SESSION.compact method.
*/
int
__wt_session_compact(
WT_SESSION *wt_session, const char *uri, const char *config)
{
WT_COMPACT_STATE compact;
WT_CONFIG_ITEM cval;
WT_DATA_SOURCE *dsrc;
WT_DECL_RET;
WT_SESSION_IMPL *session;
u_int i;
bool ignore_cache_size_set;
ignore_cache_size_set = false;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, compact, config, cfg);
/*
* The compaction thread should not block when the cache is full: it is
* holding locks blocking checkpoints and once the cache is full, it can
* spend a long time doing eviction.
*/
if (!F_ISSET(session, WT_SESSION_IGNORE_CACHE_SIZE)) {
ignore_cache_size_set = true;
F_SET(session, WT_SESSION_IGNORE_CACHE_SIZE);
}
/* In-memory ignores compaction operations. */
if (F_ISSET(S2C(session), WT_CONN_IN_MEMORY))
goto err;
/*
* Non-LSM object compaction requires checkpoints, which are impossible
* in transactional contexts. Disallow in all contexts (there's no
* reason for LSM to allow this, possible or not), and check now so the
* error message isn't confusing.
*/
WT_ERR(__wt_txn_context_check(session, false));
/* Disallow objects in the WiredTiger name space. */
WT_ERR(__wt_str_name_check(session, uri));
if (!WT_PREFIX_MATCH(uri, "colgroup:") &&
!WT_PREFIX_MATCH(uri, "file:") &&
!WT_PREFIX_MATCH(uri, "index:") &&
!WT_PREFIX_MATCH(uri, "lsm:") &&
!WT_PREFIX_MATCH(uri, "table:")) {
if ((dsrc = __wt_schema_get_source(session, uri)) != NULL)
ret = dsrc->compact == NULL ?
__wt_object_unsupported(session, uri) :
dsrc->compact(
dsrc, wt_session, uri, (WT_CONFIG_ARG *)cfg);
else
ret = __wt_bad_object_type(session, uri);
goto err;
}
/* Setup the session handle's compaction state structure. */
memset(&compact, 0, sizeof(WT_COMPACT_STATE));
session->compact = &compact;
/* Compaction can be time-limited. */
WT_ERR(__wt_config_gets(session, cfg, "timeout", &cval));
session->compact->max_time = (uint64_t)cval.val;
__wt_epoch(session, &session->compact->begin);
/* Find the types of data sources being compacted. */
WT_WITH_SCHEMA_LOCK(session,
ret = __wt_schema_worker(session, uri,
__compact_handle_append, __compact_uri_analyze, cfg, 0));
WT_ERR(ret);
if (session->compact->lsm_count != 0)
WT_ERR(__wt_schema_worker(
session, uri, NULL, __wt_lsm_compact, cfg, 0));
if (session->compact->file_count != 0)
WT_ERR(__compact_worker(session));
err: session->compact = NULL;
for (i = 0; i < session->op_handle_next; ++i) {
WT_WITH_DHANDLE(session, session->op_handle[i],
WT_TRET(__compact_end(session)));
WT_WITH_DHANDLE(session, session->op_handle[i],
WT_TRET(__wt_session_release_dhandle(session)));
}
__wt_free(session, session->op_handle);
session->op_handle_allocated = session->op_handle_next = 0;
/*
* Release common session resources (for example, checkpoint may acquire
* significant reconciliation structures/memory).
*/
WT_TRET(__wt_session_release_resources(session));
if (ignore_cache_size_set)
F_CLR(session, WT_SESSION_IGNORE_CACHE_SIZE);
if (ret != 0)
WT_STAT_CONN_INCR(session, session_table_compact_fail);
else
WT_STAT_CONN_INCR(session, session_table_compact_success);
API_END_RET_NOTFOUND_MAP(session, ret);
}
/*
* __session_close --
* WT_SESSION->close method.
*/
static int
__session_close(WT_SESSION *wt_session, const char *config)
{
WT_CONNECTION_IMPL *conn;
WT_CURSOR *cursor;
WT_DECL_RET;
WT_SESSION_IMPL *session;
int tret;
conn = (WT_CONNECTION_IMPL *)wt_session->connection;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, close, config, cfg);
WT_UNUSED(cfg);
/* Rollback any active transaction. */
if (F_ISSET(&session->txn, TXN_RUNNING))
WT_TRET(__session_rollback_transaction(wt_session, NULL));
/* Close all open cursors. */
while ((cursor = TAILQ_FIRST(&session->cursors)) != NULL)
WT_TRET(cursor->close(cursor));
WT_ASSERT(session, session->ncursors == 0);
/*
* Acquire the schema lock: we may be closing btree handles.
*
* Note that in some special cases, the schema may already be locked
* (e.g., if this session is an LSM tree worker and the tree is being
* dropped).
*/
WT_WITH_SCHEMA_LOCK_OPT(session, tret = __session_close_cache(session));
WT_TRET(tret);
/* Discard metadata tracking. */
__wt_meta_track_discard(session);
/* Discard scratch buffers. */
__wt_scr_discard(session);
/* Free transaction information. */
__wt_txn_destroy(session);
/* Confirm we're not holding any hazard references. */
__wt_hazard_close(session);
/* Free the reconciliation information. */
__wt_rec_destroy(session, &session->reconcile);
/* Free the eviction exclusive-lock information. */
__wt_free(session, session->excl);
/* Destroy the thread's mutex. */
if (session->cond != NULL)
(void)__wt_cond_destroy(session, session->cond);
/* The API lock protects opening and closing of sessions. */
__wt_spin_lock(session, &conn->api_lock);
/*
* Sessions are re-used, clear the structure: this code sets the active
* field to 0, which will exclude the hazard array from review by the
* eviction thread. Note: there's no serialization support around the
* review of the hazard array, which means threads checking for hazard
* references first check the active field (which may be 0) and then use
* the hazard pointer (which cannot be NULL). For this reason, clear
* the session structure carefully.
*
* We don't need to publish here, because regardless of the active field
* being non-zero, the hazard reference is always valid.
*/
WT_SESSION_CLEAR(session);
session = conn->default_session;
/*
* Decrement the count of active sessions if that's possible: a session
* being closed may or may not be at the end of the array, step toward
* the beginning of the array until we reach an active session.
*/
while (conn->sessions[conn->session_cnt - 1].active == 0)
if (--conn->session_cnt == 0)
break;
__wt_spin_unlock(session, &conn->api_lock);
err: API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_truncate --
* WT_SESSION->truncate method.
*/
static int
__session_truncate(WT_SESSION *wt_session,
const char *uri, WT_CURSOR *start, WT_CURSOR *stop, const char *config)
{
WT_SESSION_IMPL *session;
int ret;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, truncate, config, cfg);
/*
* If the URI is specified, we don't need a start/stop, if start/stop
* is specified, we don't need a URI.
*
* If no URI is specified, and both cursors are specified, start/stop
* must reference the same object.
*
* Any specified cursor must have been initialized.
*/
if ((uri == NULL && start == NULL && stop == NULL) ||
(uri != NULL && (start != NULL || stop != NULL)))
WT_ERR_MSG(session, EINVAL,
"the truncate method should be passed either a URI or "
"start/stop cursors, but not both");
if (start != NULL && stop != NULL && strcmp(start->uri, stop->uri) != 0)
WT_ERR_MSG(session, EINVAL,
"truncate method cursors must reference the same object");
if ((start != NULL && !F_ISSET(start, WT_CURSTD_KEY_SET)) ||
(stop != NULL && !F_ISSET(stop, WT_CURSTD_KEY_SET)))
WT_ERR_MSG(session, EINVAL,
"the truncate method cursors must have their keys set");
if (uri == NULL) {
/*
* From a starting/stopping cursor to the begin/end of the
* object is easy, walk the object.
*/
if (start == NULL)
for (;;) {
WT_ERR(stop->remove(stop));
if ((ret = stop->prev(stop)) != 0) {
if (ret == WT_NOTFOUND)
ret = 0;
break;
}
}
else
for (;;) {
WT_ERR(start->remove(start));
if (stop != NULL &&
start->equals(start, stop))
break;
if ((ret = start->next(start)) != 0) {
if (ret == WT_NOTFOUND)
ret = 0;
break;
}
}
} else
ret = __wt_schema_truncate(session, uri, cfg);
err:
API_END_NOTFOUND_MAP(session, ret);
}
/*
* __session_close --
* WT_SESSION->close method.
*/
static int
__session_close(WT_SESSION *wt_session, const char *config)
{
WT_BTREE_SESSION *btree_session;
WT_CONNECTION_IMPL *conn;
WT_CURSOR *cursor;
WT_SESSION_IMPL *session, **tp;
int ret;
conn = (WT_CONNECTION_IMPL *)wt_session->connection;
session = (WT_SESSION_IMPL *)wt_session;
SESSION_API_CALL(session, close, config, cfg);
WT_UNUSED(cfg);
while ((cursor = TAILQ_FIRST(&session->cursors)) != NULL)
WT_TRET(cursor->close(cursor));
while ((btree_session = TAILQ_FIRST(&session->btrees)) != NULL)
WT_TRET(__wt_session_remove_btree(session, btree_session, 0));
WT_TRET(__wt_schema_close_tables(session));
__wt_spin_lock(session, &conn->spinlock);
/* Discard scratch buffers. */
__wt_scr_discard(session);
/* Confirm we're not holding any hazard references. */
__wt_hazard_empty(session);
/* Free the reconciliation information. */
__wt_rec_destroy(session);
/* Free the eviction exclusive-lock information. */
__wt_free(session, session->excl);
/* Destroy the thread's mutex. */
if (session->cond != NULL)
(void)__wt_cond_destroy(session, session->cond);
/*
* Replace the session reference we're closing with the last entry in
* the table, then clear the last entry. As far as the walk of the
* server threads is concerned, it's OK if the session appears twice,
* or if it doesn't appear at all, so these lines can race all they
* want.
*/
for (tp = conn->sessions; *tp != session; ++tp)
;
--conn->session_cnt;
*tp = conn->sessions[conn->session_cnt];
conn->sessions[conn->session_cnt] = NULL;
/*
* Publish, making the session array entry available for re-use. There
* must be a barrier here to ensure the cleanup above completes before
* the entry is re-used.
*/
WT_PUBLISH(session->iface.connection, NULL);
session = &conn->default_session;
__wt_spin_unlock(session, &conn->spinlock);
err:
API_END_NOTFOUND_MAP(session, ret);
}
