/*
* __wt_schema_drop --
* Process a WT_SESSION::drop operation for all supported types.
*/
int
__wt_schema_drop(WT_SESSION_IMPL *session, const char *uri, const char *cfg[])
{
WT_CONFIG_ITEM cval;
WT_DATA_SOURCE *dsrc;
WT_DECL_RET;
bool force;
WT_RET(__wt_config_gets_def(session, cfg, "force", 0, &cval));
force = cval.val != 0;
WT_RET(__wt_meta_track_on(session));
/* Paranoia: clear any handle from our caller. */
session->dhandle = NULL;
if (WT_PREFIX_MATCH(uri, "colgroup:"))
ret = __drop_colgroup(session, uri, force, cfg);
else if (WT_PREFIX_MATCH(uri, "file:"))
ret = __drop_file(session, uri, force, cfg);
else if (WT_PREFIX_MATCH(uri, "index:"))
ret = __drop_index(session, uri, force, cfg);
else if (WT_PREFIX_MATCH(uri, "lsm:"))
ret = __wt_lsm_tree_drop(session, uri, cfg);
else if (WT_PREFIX_MATCH(uri, "table:"))
ret = __drop_table(session, uri, cfg);
else if ((dsrc = __wt_schema_get_source(session, uri)) != NULL)
ret = dsrc->drop == NULL ?
__wt_object_unsupported(session, uri) :
dsrc->drop(
dsrc, &session->iface, uri, (WT_CONFIG_ARG *)cfg);
else
ret = __wt_bad_object_type(session, uri);
/*
* Map WT_NOTFOUND to ENOENT, based on the assumption WT_NOTFOUND means
* there was no metadata entry. Map ENOENT to zero if force is set.
*/
if (ret == WT_NOTFOUND || ret == ENOENT)
ret = force ? 0 : ENOENT;
/* Bump the schema generation so that stale data is ignored. */
++S2C(session)->schema_gen;
WT_TRET(__wt_meta_track_off(session, true, ret != 0));
return (ret);
}
/*
* __wt_schema_rename --
* WT_SESSION::rename.
*/
int
__wt_schema_rename(WT_SESSION_IMPL *session,
const char *uri, const char *newuri, const char *cfg[])
{
WT_DATA_SOURCE *dsrc;
WT_DECL_RET;
const char *p, *t;
/* Disallow renames to/from the WiredTiger name space. */
WT_RET(__wt_schema_name_check(session, uri));
WT_RET(__wt_schema_name_check(session, newuri));
/* The target type must match the source type. */
for (p = uri, t = newuri; *p == *t && *p != ':'; ++p, ++t)
;
if (*p != ':' || *t != ':')
WT_RET_MSG(session, EINVAL,
"rename target type must match URI: %s to %s", uri, newuri);
/*
* We track rename operations, if we fail in the middle, we want to
* back it all out.
*/
WT_RET(__wt_meta_track_on(session));
if (WT_PREFIX_MATCH(uri, "file:"))
ret = __rename_file(session, uri, newuri);
else if (WT_PREFIX_MATCH(uri, "lsm:"))
ret = __wt_lsm_tree_rename(session, uri, newuri, cfg);
else if (WT_PREFIX_MATCH(uri, "table:"))
ret = __rename_table(session, uri, newuri, cfg);
else if ((ret = __wt_schema_get_source(session, uri, &dsrc)) == 0)
ret = dsrc->rename == NULL ?
__wt_object_unsupported(session, uri) :
dsrc->rename(dsrc, &session->iface,
uri, newuri, (WT_CONFIG_ARG *)cfg);
/* Bump the schema generation so that stale data is ignored. */
++S2C(session)->schema_gen;
WT_TRET(__wt_meta_track_off(session, ret != 0));
/* If we didn't find a metadata entry, map that error to ENOENT. */
return (ret == WT_NOTFOUND ? ENOENT : ret);
}
/*
* __wt_schema_create --
* Process a WT_SESSION::create operation for all supported types.
*/
int
__wt_schema_create(
WT_SESSION_IMPL *session, const char *uri, const char *config)
{
WT_CONFIG_ITEM cval;
WT_DATA_SOURCE *dsrc;
WT_DECL_RET;
int exclusive;
exclusive = (
__wt_config_getones(session, config, "exclusive", &cval) == 0 &&
cval.val != 0);
/*
* We track create operations: if we fail in the middle of creating a
* complex object, we want to back it all out.
*/
WT_RET(__wt_meta_track_on(session));
if (WT_PREFIX_MATCH(uri, "colgroup:"))
ret = __create_colgroup(session, uri, exclusive, config);
else if (WT_PREFIX_MATCH(uri, "file:"))
ret = __create_file(session, uri, exclusive, config);
else if (WT_PREFIX_MATCH(uri, "lsm:"))
ret = __wt_lsm_tree_create(session, uri, exclusive, config);
else if (WT_PREFIX_MATCH(uri, "index:"))
ret = __create_index(session, uri, exclusive, config);
else if (WT_PREFIX_MATCH(uri, "table:"))
ret = __create_table(session, uri, exclusive, config);
else if ((dsrc = __wt_schema_get_source(session, uri)) != NULL)
ret = dsrc->create == NULL ?
__wt_object_unsupported(session, uri) :
__create_data_source(session, uri, config, dsrc);
else
ret = __wt_bad_object_type(session, uri);
session->dhandle = NULL;
WT_TRET(__wt_meta_track_off(session, 1, ret != 0));
return (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);
}
/*
* __wt_schema_worker --
* Get Btree handles for the object and cycle through calls to an
* underlying worker function with each handle.
*/
int
__wt_schema_worker(WT_SESSION_IMPL *session,
const char *uri,
int (*file_func)(WT_SESSION_IMPL *, const char *[]),
int (*name_func)(WT_SESSION_IMPL *, const char *, bool *),
const char *cfg[], uint32_t open_flags)
{
WT_COLGROUP *colgroup;
WT_DATA_SOURCE *dsrc;
WT_DECL_RET;
WT_INDEX *idx;
WT_SESSION *wt_session;
WT_TABLE *table;
u_int i;
bool skip;
table = NULL;
skip = false;
if (name_func != NULL)
WT_ERR(name_func(session, uri, &skip));
/* If the callback said to skip this object, we're done. */
if (skip)
return (0);
/* Get the btree handle(s) and call the underlying function. */
if (WT_PREFIX_MATCH(uri, "file:")) {
if (file_func != NULL)
WT_ERR(__wt_exclusive_handle_operation(session,
uri, file_func, cfg, open_flags));
} else if (WT_PREFIX_MATCH(uri, "colgroup:")) {
WT_ERR(__wt_schema_get_colgroup(
session, uri, false, NULL, &colgroup));
WT_ERR(__wt_schema_worker(session,
colgroup->source, file_func, name_func, cfg, open_flags));
} else if (WT_PREFIX_MATCH(uri, "index:")) {
idx = NULL;
WT_ERR(__wt_schema_get_index(session, uri, false, false, &idx));
WT_ERR(__wt_schema_worker(session, idx->source,
file_func, name_func, cfg, open_flags));
} else if (WT_PREFIX_MATCH(uri, "lsm:")) {
WT_ERR(__wt_lsm_tree_worker(session,
uri, file_func, name_func, cfg, open_flags));
} else if (WT_PREFIX_MATCH(uri, "table:")) {
/*
* Note: we would like to use open_flags here (e.g., to lock
* the table exclusive during schema-changing operations), but
* that is currently problematic because we get the table again
* in order to discover column groups and indexes.
*/
WT_ERR(__wt_schema_get_table_uri(
session, uri, false, 0, &table));
/*
* We could make a recursive call for each colgroup or index
* URI, but since we have already opened the table, we can take
* a short cut and skip straight to the sources. If we have a
* name function, it needs to know about the intermediate URIs.
*/
for (i = 0; i < WT_COLGROUPS(table); i++) {
colgroup = table->cgroups[i];
skip = false;
if (name_func != NULL)
WT_ERR(name_func(
session, colgroup->name, &skip));
if (!skip)
WT_ERR(__wt_schema_worker(
session, colgroup->source,
file_func, name_func, cfg, open_flags));
}
WT_ERR(__wt_schema_open_indices(session, table));
for (i = 0; i < table->nindices; i++) {
idx = table->indices[i];
skip = false;
if (name_func != NULL)
WT_ERR(name_func(session, idx->name, &skip));
if (!skip)
WT_ERR(__wt_schema_worker(session, idx->source,
file_func, name_func, cfg, open_flags));
}
} else if ((dsrc = __wt_schema_get_source(session, uri)) != NULL) {
wt_session = (WT_SESSION *)session;
if (file_func == __wt_salvage && dsrc->salvage != NULL)
WT_ERR(dsrc->salvage(
dsrc, wt_session, uri, (WT_CONFIG_ARG *)cfg));
else if (file_func == __wt_verify && dsrc->verify != NULL)
WT_ERR(dsrc->verify(
dsrc, wt_session, uri, (WT_CONFIG_ARG *)cfg));
else if (file_func == __wt_checkpoint)
;
else if (file_func == __wt_checkpoint_get_handles)
;
else if (file_func == __wt_checkpoint_sync)
;
else
WT_ERR(__wt_object_unsupported(session, uri));
} else
WT_ERR(__wt_bad_object_type(session, uri));
err: if (table != NULL)
WT_TRET(__wt_schema_release_table(session, table));
return (ret);
}
/*
* __wt_schema_worker --
* Get Btree handles for the object and cycle through calls to an
* underlying worker function with each handle.
*/
int
__wt_schema_worker(WT_SESSION_IMPL *session,
const char *uri,
int (*file_func)(WT_SESSION_IMPL *, const char *[]),
int (*name_func)(WT_SESSION_IMPL *, const char *),
const char *cfg[], uint32_t open_flags)
{
WT_COLGROUP *colgroup;
WT_DATA_SOURCE *dsrc;
WT_DECL_RET;
WT_INDEX *idx;
WT_SESSION *wt_session;
WT_TABLE *table;
const char *tablename;
u_int i;
table = NULL;
tablename = uri;
if (name_func != NULL)
WT_ERR(name_func(session, uri));
/* Get the btree handle(s) and call the underlying function. */
if (WT_PREFIX_MATCH(uri, "file:")) {
if (file_func != NULL) {
WT_ERR(__wt_session_get_btree_ckpt(
session, uri, cfg, open_flags));
ret = file_func(session, cfg);
WT_TRET(__wt_session_release_btree(session));
}
} else if (WT_PREFIX_MATCH(uri, "colgroup:")) {
WT_ERR(__wt_schema_get_colgroup(session, uri, NULL, &colgroup));
WT_ERR(__wt_schema_worker(session, colgroup->source,
file_func, name_func, cfg, open_flags));
} else if (WT_PREFIX_SKIP(tablename, "index:")) {
idx = NULL;
WT_ERR(__wt_schema_get_index(session, uri, NULL, &idx));
WT_ERR(__wt_schema_worker(session, idx->source,
file_func, name_func, cfg, open_flags));
} else if (WT_PREFIX_MATCH(uri, "lsm:")) {
WT_ERR(__wt_lsm_tree_worker(
session, uri, file_func, name_func, cfg, open_flags));
} else if (WT_PREFIX_SKIP(tablename, "table:")) {
WT_ERR(__wt_schema_get_table(session,
tablename, strlen(tablename), 0, &table));
WT_ASSERT(session, session->dhandle == NULL);
/*
* We could make a recursive call for each colgroup or index
* URI, but since we have already opened the table, we can take
* a short cut and skip straight to the sources. If we have a
* name function, it needs to know about the intermediate URIs.
*/
for (i = 0; i < WT_COLGROUPS(table); i++) {
colgroup = table->cgroups[i];
if (name_func != NULL)
WT_ERR(name_func(session, colgroup->name));
WT_ERR(__wt_schema_worker(session, colgroup->source,
file_func, name_func, cfg, open_flags));
}
WT_ERR(__wt_schema_open_indices(session, table));
for (i = 0; i < table->nindices; i++) {
idx = table->indices[i];
if (name_func != NULL)
WT_ERR(name_func(session, idx->name));
WT_ERR(__wt_schema_worker(session, idx->source,
file_func, name_func, cfg, open_flags));
}
} else if ((dsrc = __wt_schema_get_source(session, uri)) != NULL) {
wt_session = (WT_SESSION *)session;
if (file_func == __wt_compact && dsrc->compact != NULL)
WT_ERR(dsrc->compact(
dsrc, wt_session, uri, (WT_CONFIG_ARG *)cfg));
else if (file_func == __wt_salvage && dsrc->salvage != NULL)
WT_ERR(dsrc->salvage(
dsrc, wt_session, uri, (WT_CONFIG_ARG *)cfg));
else if (file_func == __wt_verify && dsrc->verify != NULL)
WT_ERR(dsrc->verify(
dsrc, wt_session, uri, (WT_CONFIG_ARG *)cfg));
else
WT_ERR(__wt_object_unsupported(session, uri));
} else
WT_ERR(__wt_bad_object_type(session, uri));
err: if (table != NULL)
__wt_schema_release_table(session, table);
return (ret);
}
