/*
* __backup_all --
* Backup all objects in the database.
*/
static int
__backup_all(WT_SESSION_IMPL *session, WT_CURSOR_BACKUP *cb)
{
WT_CONFIG_ITEM cval;
WT_CURSOR *cursor;
WT_DECL_RET;
const char *key, *value;
cursor = NULL;
/* Copy all of the metadata entries to the hot backup file. */
WT_RET(__wt_metadata_cursor(session, &cursor));
while ((ret = cursor->next(cursor)) == 0) {
WT_ERR(cursor->get_key(cursor, &key));
WT_ERR(cursor->get_value(cursor, &value));
WT_ERR(__wt_fprintf(cb->bfp, "%s\n%s\n", key, value));
/*
* While reading the metadata file, check there are no "sources"
* or "types" which can't support hot backup. This checks for
* a data source that's non-standard, which can't be backed up,
* but is also sanity checking: if there's an entry backed by
* anything other than a file or lsm entry, we're confused.
*/
if ((ret = __wt_config_getones(
session, value, "type", &cval)) == 0 &&
!WT_PREFIX_MATCH_LEN(cval.str, cval.len, "file") &&
!WT_PREFIX_MATCH_LEN(cval.str, cval.len, "lsm"))
WT_ERR_MSG(session, ENOTSUP,
"hot backup is not supported for objects of "
"type %.*s", (int)cval.len, cval.str);
WT_ERR_NOTFOUND_OK(ret);
if ((ret =__wt_config_getones(
session, value, "source", &cval)) == 0 &&
!WT_PREFIX_MATCH_LEN(cval.str, cval.len, "file:") &&
!WT_PREFIX_MATCH_LEN(cval.str, cval.len, "lsm:"))
WT_ERR_MSG(session, ENOTSUP,
"hot backup is not supported for objects of "
"source %.*s", (int)cval.len, cval.str);
WT_ERR_NOTFOUND_OK(ret);
}
WT_ERR_NOTFOUND_OK(ret);
WT_ERR(__wt_metadata_cursor_release(session, &cursor));
/* Build a list of the file objects that need to be copied. */
WT_WITH_HANDLE_LIST_LOCK(session, ret =
__wt_meta_btree_apply(session, __backup_list_all_append, NULL));
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __wt_metadata_update --
* Update a row in the metadata.
*/
int
__wt_metadata_update(
WT_SESSION_IMPL *session, const char *key, const char *value)
{
WT_CURSOR *cursor;
WT_DECL_RET;
WT_RET(__wt_verbose(session, WT_VERB_METADATA,
"Update: key: %s, value: %s, tracking: %s, %s" "turtle",
key, value, WT_META_TRACKING(session) ? "true" : "false",
__metadata_turtle(key) ? "" : "not "));
if (__metadata_turtle(key)) {
WT_WITH_TURTLE_LOCK(session, ret,
ret = __wt_turtle_update(session, key, value));
return (ret);
}
if (WT_META_TRACKING(session))
WT_RET(__wt_meta_track_update(session, key));
WT_RET(__wt_metadata_cursor(session, &cursor));
/* This cursor needs to have overwrite semantics. */
WT_ASSERT(session, F_ISSET(cursor, WT_CURSTD_OVERWRITE));
cursor->set_key(cursor, key);
cursor->set_value(cursor, value);
WT_ERR(cursor->insert(cursor));
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __wt_metadata_insert --
* Insert a row into the metadata.
*/
int
__wt_metadata_insert(
WT_SESSION_IMPL *session, const char *key, const char *value)
{
WT_CURSOR *cursor;
WT_DECL_RET;
WT_RET(__wt_verbose(session, WT_VERB_METADATA,
"Insert: key: %s, value: %s, tracking: %s, %s" "turtle",
key, value, WT_META_TRACKING(session) ? "true" : "false",
__metadata_turtle(key) ? "" : "not "));
if (__metadata_turtle(key))
WT_RET_MSG(session, EINVAL,
"%s: insert not supported on the turtle file", key);
WT_RET(__wt_metadata_cursor(session, &cursor));
cursor->set_key(cursor, key);
cursor->set_value(cursor, value);
WT_ERR(cursor->insert(cursor));
if (WT_META_TRACKING(session))
WT_ERR(__wt_meta_track_insert(session, key));
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __wt_meta_btree_apply --
* Apply a function to all files listed in the metadata, apart from the
* metadata file.
*/
int
__wt_meta_btree_apply(WT_SESSION_IMPL *session,
int (*func)(WT_SESSION_IMPL *, const char *[]), const char *cfg[])
{
WT_CURSOR *cursor;
WT_DECL_RET;
WT_RET(__wt_metadata_cursor(session, &cursor));
WT_SAVE_DHANDLE(session,
ret = __meta_btree_apply(session, cursor, func, cfg));
WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __wt_metadata_search --
* Return a copied row from the metadata.
* The caller is responsible for freeing the allocated memory.
*/
int
__wt_metadata_search(WT_SESSION_IMPL *session, const char *key, char **valuep)
{
WT_CURSOR *cursor;
WT_DECL_RET;
const char *value;
*valuep = NULL;
__wt_verbose(session, WT_VERB_METADATA,
"Search: key: %s, tracking: %s, %s" "turtle",
key, WT_META_TRACKING(session) ? "true" : "false",
__metadata_turtle(key) ? "" : "not ");
if (__metadata_turtle(key)) {
/*
* The returned value should only be set if ret is non-zero, but
* Coverity is convinced otherwise. The code path is used enough
* that Coverity complains a lot, add an error check to get some
* peace and quiet.
*/
if ((ret = __wt_turtle_read(session, key, valuep)) != 0)
__wt_free(session, *valuep);
return (ret);
}
/*
* All metadata reads are at read-uncommitted isolation. That's
* because once a schema-level operation completes, subsequent
* operations must see the current version of checkpoint metadata, or
* they may try to read blocks that may have been freed from a file.
* Metadata updates use non-transactional techniques (such as the
* schema and metadata locks) to protect access to in-flight updates.
*/
WT_RET(__wt_metadata_cursor(session, &cursor));
cursor->set_key(cursor, key);
WT_WITH_TXN_ISOLATION(session, WT_ISO_READ_UNCOMMITTED,
ret = cursor->search(cursor));
WT_ERR(ret);
WT_ERR(cursor->get_value(cursor, &value));
WT_ERR(__wt_strdup(session, value, valuep));
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
if (ret != 0)
__wt_free(session, *valuep);
return (ret);
}
/*
* __wt_meta_apply_all --
* Apply a function to all files listed in the metadata, apart from the
* metadata file.
*/
int
__wt_meta_apply_all(WT_SESSION_IMPL *session,
int (*file_func)(WT_SESSION_IMPL *, const char *[]),
int (*name_func)(WT_SESSION_IMPL *, const char *, bool *),
const char *cfg[])
{
WT_CURSOR *cursor;
WT_DECL_RET;
WT_ASSERT(session, F_ISSET(session, WT_SESSION_LOCKED_SCHEMA));
WT_RET(__wt_metadata_cursor(session, &cursor));
WT_SAVE_DHANDLE(session, ret =
__meta_btree_apply(session, cursor, file_func, name_func, cfg));
WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __metadata_load_bulk --
* Create any bulk-loaded file stubs.
*/
static int
__metadata_load_bulk(WT_SESSION_IMPL *session)
{
WT_CURSOR *cursor;
WT_DECL_RET;
uint32_t allocsize;
bool exist;
const char *filecfg[] = {
WT_CONFIG_BASE(session, file_meta), NULL, NULL };
const char *key, *value;
/*
* If a file was being bulk-loaded during the hot backup, it will appear
* in the metadata file, but the file won't exist. Create on demand.
*/
WT_RET(__wt_metadata_cursor(session, &cursor));
while ((ret = cursor->next(cursor)) == 0) {
WT_ERR(cursor->get_key(cursor, &key));
if (!WT_PREFIX_SKIP(key, "file:"))
continue;
/* If the file exists, it's all good. */
WT_ERR(__wt_fs_exist(session, key, &exist));
if (exist)
continue;
/*
* If the file doesn't exist, assume it's a bulk-loaded file;
* retrieve the allocation size and re-create the file.
*/
WT_ERR(cursor->get_value(cursor, &value));
filecfg[1] = value;
WT_ERR(__wt_direct_io_size_check(
session, filecfg, "allocation_size", &allocsize));
WT_ERR(__wt_block_manager_create(session, key, allocsize));
}
WT_ERR_NOTFOUND_OK(ret);
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __wt_metadata_remove --
* Remove a row from the metadata.
*/
int
__wt_metadata_remove(WT_SESSION_IMPL *session, const char *key)
{
WT_CURSOR *cursor;
WT_DECL_RET;
__wt_verbose(session, WT_VERB_METADATA,
"Remove: key: %s, tracking: %s, %s" "turtle",
key, WT_META_TRACKING(session) ? "true" : "false",
__metadata_turtle(key) ? "" : "not ");
if (__metadata_turtle(key))
WT_RET_MSG(session, EINVAL,
"%s: remove not supported on the turtle file", key);
WT_RET(__wt_metadata_cursor(session, &cursor));
cursor->set_key(cursor, key);
WT_ERR(cursor->search(cursor));
if (WT_META_TRACKING(session))
WT_ERR(__wt_meta_track_update(session, key));
WT_ERR(cursor->remove(cursor));
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __wt_metadata_search --
* Return a copied row from the metadata.
* The caller is responsible for freeing the allocated memory.
*/
int
__wt_metadata_search(WT_SESSION_IMPL *session, const char *key, char **valuep)
{
WT_CURSOR *cursor;
WT_DECL_RET;
const char *value;
*valuep = NULL;
WT_RET(__wt_verbose(session, WT_VERB_METADATA,
"Search: key: %s, tracking: %s, %s" "turtle",
key, WT_META_TRACKING(session) ? "true" : "false",
__metadata_turtle(key) ? "" : "not "));
if (__metadata_turtle(key))
return (__wt_turtle_read(session, key, valuep));
/*
* All metadata reads are at read-uncommitted isolation. That's
* because once a schema-level operation completes, subsequent
* operations must see the current version of checkpoint metadata, or
* they may try to read blocks that may have been freed from a file.
* Metadata updates use non-transactional techniques (such as the
* schema and metadata locks) to protect access to in-flight updates.
*/
WT_RET(__wt_metadata_cursor(session, &cursor));
cursor->set_key(cursor, key);
WT_WITH_TXN_ISOLATION(session, WT_ISO_READ_UNCOMMITTED,
ret = cursor->search(cursor));
WT_ERR(ret);
WT_ERR(cursor->get_value(cursor, &value));
WT_ERR(__wt_strdup(session, value, valuep));
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
return (ret);
}
/*
* __schema_open_table --
* Open a named table (internal version).
*/
static int
__schema_open_table(WT_SESSION_IMPL *session,
const char *name, size_t namelen, bool ok_incomplete, WT_TABLE **tablep)
{
WT_CONFIG cparser;
WT_CONFIG_ITEM ckey, cval;
WT_CURSOR *cursor;
WT_DECL_ITEM(buf);
WT_DECL_RET;
WT_TABLE *table;
const char *tconfig;
char *tablename;
cursor = NULL;
table = NULL;
tablename = NULL;
WT_ASSERT(session, F_ISSET(session, WT_SESSION_LOCKED_TABLE));
WT_ERR(__wt_scr_alloc(session, 0, &buf));
WT_ERR(__wt_buf_fmt(session, buf, "table:%.*s", (int)namelen, name));
WT_ERR(__wt_strndup(session, buf->data, buf->size, &tablename));
WT_ERR(__wt_metadata_cursor(session, &cursor));
cursor->set_key(cursor, tablename);
WT_ERR(cursor->search(cursor));
WT_ERR(cursor->get_value(cursor, &tconfig));
WT_ERR(__wt_calloc_one(session, &table));
table->name = tablename;
tablename = NULL;
table->name_hash = __wt_hash_city64(name, namelen);
WT_ERR(__wt_config_getones(session, tconfig, "columns", &cval));
WT_ERR(__wt_config_getones(session, tconfig, "key_format", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &table->key_format));
WT_ERR(__wt_config_getones(session, tconfig, "value_format", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &table->value_format));
WT_ERR(__wt_strdup(session, tconfig, &table->config));
/* Point to some items in the copy to save re-parsing. */
WT_ERR(__wt_config_getones(session, table->config,
"columns", &table->colconf));
/*
* Count the number of columns: tables are "simple" if the columns
* are not named.
*/
WT_ERR(__wt_config_subinit(session, &cparser, &table->colconf));
table->is_simple = true;
while ((ret = __wt_config_next(&cparser, &ckey, &cval)) == 0)
table->is_simple = false;
if (ret != WT_NOTFOUND)
goto err;
/* Check that the columns match the key and value formats. */
if (!table->is_simple)
WT_ERR(__wt_schema_colcheck(session,
table->key_format, table->value_format, &table->colconf,
&table->nkey_columns, NULL));
WT_ERR(__wt_config_getones(session, table->config,
"colgroups", &table->cgconf));
/* Count the number of column groups. */
WT_ERR(__wt_config_subinit(session, &cparser, &table->cgconf));
table->ncolgroups = 0;
while ((ret = __wt_config_next(&cparser, &ckey, &cval)) == 0)
++table->ncolgroups;
if (ret != WT_NOTFOUND)
goto err;
if (table->ncolgroups > 0 && table->is_simple)
WT_ERR_MSG(session, EINVAL,
"%s requires a table with named columns", tablename);
WT_ERR(__wt_calloc_def(session, WT_COLGROUPS(table), &table->cgroups));
WT_ERR(__wt_schema_open_colgroups(session, table));
if (!ok_incomplete && !table->cg_complete)
WT_ERR_MSG(session, EINVAL, "'%s' cannot be used "
"until all column groups are created",
table->name);
/* Copy the schema generation into the new table. */
table->schema_gen = S2C(session)->schema_gen;
*tablep = table;
if (0) {
err: WT_TRET(__wt_schema_destroy_table(session, &table));
}
WT_TRET(__wt_metadata_cursor_release(session, &cursor));
__wt_free(session, tablename);
__wt_scr_free(session, &buf);
return (ret);
}
/*
* __schema_open_index --
* Open one or more indices for a table (internal version).
*/
static int
__schema_open_index(WT_SESSION_IMPL *session,
WT_TABLE *table, const char *idxname, size_t len, WT_INDEX **indexp)
{
WT_CURSOR *cursor;
WT_DECL_ITEM(tmp);
WT_DECL_RET;
WT_INDEX *idx;
u_int i;
int cmp;
bool match;
const char *idxconf, *name, *tablename, *uri;
/* Check if we've already done the work. */
if (idxname == NULL && table->idx_complete)
return (0);
cursor = NULL;
idx = NULL;
match = false;
/* Build a search key. */
tablename = table->name;
(void)WT_PREFIX_SKIP(tablename, "table:");
WT_ERR(__wt_scr_alloc(session, 512, &tmp));
WT_ERR(__wt_buf_fmt(session, tmp, "index:%s:", tablename));
/* Find matching indices. */
WT_ERR(__wt_metadata_cursor(session, &cursor));
cursor->set_key(cursor, tmp->data);
if ((ret = cursor->search_near(cursor, &cmp)) == 0 && cmp < 0)
ret = cursor->next(cursor);
for (i = 0; ret == 0; i++, ret = cursor->next(cursor)) {
WT_ERR(cursor->get_key(cursor, &uri));
name = uri;
if (!WT_PREFIX_SKIP(name, tmp->data))
break;
/* Is this the index we are looking for? */
match = idxname == NULL || WT_STRING_MATCH(name, idxname, len);
/*
* Ensure there is space, including if we have to make room for
* a new entry in the middle of the list.
*/
WT_ERR(__wt_realloc_def(session, &table->idx_alloc,
WT_MAX(i, table->nindices) + 1, &table->indices));
/* Keep the in-memory list in sync with the metadata. */
cmp = 0;
while (table->indices[i] != NULL &&
(cmp = strcmp(uri, table->indices[i]->name)) > 0) {
/* Index no longer exists, remove it. */
__wt_free(session, table->indices[i]);
memmove(&table->indices[i], &table->indices[i + 1],
(table->nindices - i) * sizeof(WT_INDEX *));
table->indices[--table->nindices] = NULL;
}
if (cmp < 0) {
/* Make room for a new index. */
memmove(&table->indices[i + 1], &table->indices[i],
(table->nindices - i) * sizeof(WT_INDEX *));
table->indices[i] = NULL;
++table->nindices;
}
if (!match)
continue;
if (table->indices[i] == NULL) {
WT_ERR(cursor->get_value(cursor, &idxconf));
WT_ERR(__wt_calloc_one(session, &idx));
WT_ERR(__wt_strdup(session, uri, &idx->name));
WT_ERR(__wt_strdup(session, idxconf, &idx->config));
WT_ERR(__open_index(session, table, idx));
/*
* If we're checking the creation of an index before a
* table is fully created, don't save the index: it
* will need to be reopened once the table is complete.
*/
if (!table->cg_complete) {
WT_ERR(
__wt_schema_destroy_index(session, &idx));
if (idxname != NULL)
break;
continue;
}
table->indices[i] = idx;
idx = NULL;
/*
* If the slot is bigger than anything else we've seen,
* bump the number of indices.
*/
if (i >= table->nindices)
table->nindices = i + 1;
}
/* If we were looking for a single index, we're done. */
if (indexp != NULL)
*indexp = table->indices[i];
if (idxname != NULL)
break;
}
WT_ERR_NOTFOUND_OK(ret);
if (idxname != NULL && !match)
ret = WT_NOTFOUND;
/* If we did a full pass, we won't need to do it again. */
if (idxname == NULL) {
table->nindices = i;
table->idx_complete = true;
}
err: WT_TRET(__wt_metadata_cursor_release(session, &cursor));
WT_TRET(__wt_schema_destroy_index(session, &idx));
__wt_scr_free(session, &tmp);
return (ret);
}
