/*
* __wt_snapshot_drop --
* Snapshot the tree, dropping one or more snapshots.
*/
int
__wt_snapshot_drop(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_CONFIG_ITEM cval, sval;
WT_DECL_RET;
char *name;
name = NULL;
WT_RET(__wt_config_gets(session, cfg, "snapshot", &cval));
if (cval.type != ITEM_STRUCT) {
WT_RET(__wt_strndup(session, cval.str, cval.len, &name));
ret = __snapshot_worker(session, name, 0, SNAPSHOT_DROP);
} else if (__wt_config_subgets(session, &cval, "all", &sval) == 0 &&
sval.val != 0)
ret = __snapshot_worker(session, name, 0, SNAPSHOT_DROP_ALL);
else if (__wt_config_subgets(session, &cval, "from", &sval) == 0 &&
sval.len != 0) {
WT_RET(__wt_strndup(session, sval.str, sval.len, &name));
ret = __snapshot_worker(session, name, 0, SNAPSHOT_DROP_FROM);
} else if (__wt_config_subgets(session, &cval, "to", &sval) == 0 &&
sval.len != 0) {
WT_RET(__wt_strndup(session, sval.str, sval.len, &name));
ret = __snapshot_worker(session, name, 0, SNAPSHOT_DROP_TO);
} else
WT_RET_MSG(session, EINVAL,
"Unexpected value for 'snapshot' key: %.*s",
(int)cval.len, cval.str);
__wt_free(session, name);
return (ret);
}
/*
* __conn_load_extension --
* WT_CONNECTION->load_extension method.
*/
static int
__conn_load_extension(
WT_CONNECTION *wt_conn, const char *path, const char *config)
{
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
WT_DLH *dlh;
WT_SESSION_IMPL *session;
int (*load)(WT_CONNECTION *, WT_CONFIG_ARG *);
const char *init_name, *terminate_name;
dlh = NULL;
init_name = terminate_name = NULL;
conn = (WT_CONNECTION_IMPL *)wt_conn;
CONNECTION_API_CALL(conn, session, load_extension, config, cfg);
WT_ERR(__wt_config_gets(session, cfg, "entry", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &init_name));
/*
* This assumes the underlying shared libraries are reference counted,
* that is, that re-opening a shared library simply increments a ref
* count, and closing it simply decrements the ref count, and the last
* close discards the reference entirely -- in other words, we do not
* check to see if we've already opened this shared library.
*
* Fill in the extension structure and call the load function.
*/
WT_ERR(__wt_dlopen(session, path, &dlh));
WT_ERR(__wt_dlsym(session, dlh, init_name, 1, &load));
WT_ERR(load(wt_conn, (WT_CONFIG_ARG *)cfg));
/* Remember the unload function for when we close. */
WT_ERR(__wt_config_gets(session, cfg, "terminate", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &terminate_name));
WT_ERR(__wt_dlsym(session, dlh, terminate_name, 0, &dlh->terminate));
/* Link onto the environment's list of open libraries. */
__wt_spin_lock(session, &conn->api_lock);
TAILQ_INSERT_TAIL(&conn->dlhqh, dlh, q);
__wt_spin_unlock(session, &conn->api_lock);
dlh = NULL;
err: if (dlh != NULL)
WT_TRET(__wt_dlclose(session, dlh));
__wt_free(session, init_name);
__wt_free(session, terminate_name);
API_END_NOTFOUND_MAP(session, ret);
}
/*
* __wt_nfilename --
* Build a file name in a scratch buffer. If the name is already an
* absolute path duplicate it, otherwise generate a path relative to the
* connection home directory.
*/
int
__wt_nfilename(
WT_SESSION_IMPL *session, const char *name, size_t namelen, char **path)
{
size_t len;
char *buf;
*path = NULL;
/*
* Needs to work with a NULL session handle - since this is called via
* the exists API which is used by the test utilities.
*/
if (session == NULL || __wt_absolute_path(name))
WT_RET(__wt_strndup(session, name, namelen, path));
else {
len = strlen(S2C(session)->home) + 1 + namelen + 1;
WT_RET(__wt_calloc(session, 1, len, &buf));
snprintf(buf, len, "%s%s%.*s", S2C(session)->home,
__wt_path_separator(), (int)namelen, name);
*path = buf;
}
return (0);
}
/*
* __logmgr_config --
* Parse and setup the logging server options.
*/
static int
__logmgr_config(WT_SESSION_IMPL *session, const char **cfg, int *runp)
{
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
conn = S2C(session);
/*
* The logging configuration is off by default.
*/
WT_RET(__wt_config_gets(session, cfg, "log.enabled", &cval));
*runp = cval.val != 0;
/*
* Setup a log path, compression and encryption even if logging is
* disabled in case we are going to print a log.
*/
conn->log_compressor = NULL;
WT_RET(__wt_config_gets_none(session, cfg, "log.compressor", &cval));
WT_RET(__wt_compressor_config(session, &cval, &conn->log_compressor));
WT_RET(__wt_config_gets(session, cfg, "log.path", &cval));
WT_RET(__wt_strndup(session, cval.str, cval.len, &conn->log_path));
/* We are done if logging isn't enabled. */
if (*runp == 0)
return (0);
WT_RET(__wt_config_gets(session, cfg, "log.archive", &cval));
if (cval.val != 0)
FLD_SET(conn->log_flags, WT_CONN_LOG_ARCHIVE);
WT_RET(__wt_config_gets(session, cfg, "log.file_max", &cval));
conn->log_file_max = (wt_off_t)cval.val;
WT_STAT_FAST_CONN_SET(session, log_max_filesize, conn->log_file_max);
WT_RET(__wt_config_gets(session, cfg, "log.prealloc", &cval));
/*
* If pre-allocation is configured, set the initial number to one.
* We'll adapt as load dictates.
*/
if (cval.val != 0) {
FLD_SET(conn->log_flags, WT_CONN_LOG_PREALLOC);
conn->log_prealloc = 1;
}
WT_RET(__wt_config_gets_def(session, cfg, "log.recover", 0, &cval));
if (cval.len != 0 && WT_STRING_MATCH("error", cval.str, cval.len))
FLD_SET(conn->log_flags, WT_CONN_LOG_RECOVER_ERR);
WT_RET(__logmgr_sync_cfg(session, cfg));
return (0);
}
/*
* __ckpt_load --
* Load a single checkpoint's information into a WT_CKPT structure.
*/
static int
__ckpt_load(WT_SESSION_IMPL *session,
WT_CONFIG_ITEM *k, WT_CONFIG_ITEM *v, WT_CKPT *ckpt)
{
WT_CONFIG_ITEM a;
char timebuf[64];
/*
* Copy the name, address (raw and hex), order and time into the slot.
* If there's no address, it's a fake.
*/
WT_RET(__wt_strndup(session, k->str, k->len, &ckpt->name));
WT_RET(__wt_config_subgets(session, v, "addr", &a));
WT_RET(__wt_buf_set(session, &ckpt->addr, a.str, a.len));
if (a.len == 0)
F_SET(ckpt, WT_CKPT_FAKE);
else
WT_RET(__wt_nhex_to_raw(session, a.str, a.len, &ckpt->raw));
WT_RET(__wt_config_subgets(session, v, "order", &a));
if (a.len == 0)
goto format;
ckpt->order = a.val;
WT_RET(__wt_config_subgets(session, v, "time", &a));
if (a.len == 0 || a.len > sizeof(timebuf) - 1)
goto format;
memcpy(timebuf, a.str, a.len);
timebuf[a.len] = '\0';
if (sscanf(timebuf, "%" SCNuMAX, &ckpt->sec) != 1)
goto format;
WT_RET(__wt_config_subgets(session, v, "size", &a));
ckpt->ckpt_size = (uint64_t)a.val;
WT_RET(__wt_config_subgets(session, v, "write_gen", &a));
if (a.len == 0)
goto format;
/*
* The largest value a WT_CONFIG_ITEM can handle is signed: this value
* appears on disk and I don't want to sign it there, so I'm casting it
* here instead.
*/
ckpt->write_gen = (uint64_t)a.val;
return (0);
format:
WT_RET_MSG(session, WT_ERROR, "corrupted checkpoint list");
}
/*
* __wt_config_collapse --
* Collapse a set of configuration strings into newly allocated memory.
*
* This function takes a NULL-terminated list of configuration strings (where
* the first one contains all the defaults and the values are in order from
* least to most preferred, that is, the default values are least preferred),
* and collapses them into newly allocated memory. The algorithm is to walk
* the first of the configuration strings, and for each entry, search all of
* the configuration strings for a final value, keeping the last value found.
*
* Notes:
* Any key not appearing in the first configuration string is discarded
* from the final result, because we'll never search for it.
*
* Nested structures aren't parsed. For example, imagine a configuration
* string contains "key=(k2=v2,k3=v3)", and a subsequent string has
* "key=(k4=v4)", the result will be "key=(k4=v4)", as we search for and
* use the final value of "key", regardless of field overlap or missing
* fields in the nested value.
*/
int
__wt_config_collapse(
WT_SESSION_IMPL *session, const char **cfg, char **config_ret)
{
WT_CONFIG cparser;
WT_CONFIG_ITEM k, v;
WT_DECL_ITEM(tmp);
WT_DECL_RET;
*config_ret = NULL;
WT_RET(__wt_scr_alloc(session, 0, &tmp));
__wt_config_init(session, &cparser, cfg[0]);
while ((ret = __wt_config_next(&cparser, &k, &v)) == 0) {
if (k.type != WT_CONFIG_ITEM_STRING &&
k.type != WT_CONFIG_ITEM_ID)
WT_ERR_MSG(session, EINVAL,
"Invalid configuration key found: '%s'", k.str);
WT_ERR(__wt_config_get(session, cfg, &k, &v));
/* Include the quotes around string keys/values. */
if (k.type == WT_CONFIG_ITEM_STRING) {
--k.str;
k.len += 2;
}
if (v.type == WT_CONFIG_ITEM_STRING) {
--v.str;
v.len += 2;
}
WT_ERR(__wt_buf_catfmt(session, tmp, "%.*s=%.*s,",
(int)k.len, k.str, (int)v.len, v.str));
}
/* We loop until error, and the expected error is WT_NOTFOUND. */
if (ret != WT_NOTFOUND)
goto err;
/*
* If the caller passes us no valid configuration strings, we get here
* with no bytes to copy -- that's OK, the underlying string copy can
* handle empty strings.
*
* Strip any trailing comma.
*/
if (tmp->size != 0)
--tmp->size;
ret = __wt_strndup(session, tmp->data, tmp->size, config_ret);
err: __wt_scr_free(session, &tmp);
return (ret);
}
/*
* __wt_lsm_tree_chunk_name --
* Get the URI of the file for a given chunk.
*/
int
__wt_lsm_tree_chunk_name(WT_SESSION_IMPL *session,
WT_LSM_TREE *lsm_tree, uint32_t id, const char **retp)
{
WT_DECL_ITEM(tmp);
WT_DECL_RET;
WT_RET(__wt_scr_alloc(session, 0, &tmp));
WT_ERR(__wt_buf_fmt(
session, tmp, "file:%s-%06" PRIu32 ".lsm", lsm_tree->filename, id));
WT_ERR(__wt_strndup(session, tmp->data, tmp->size, retp));
err: __wt_scr_free(&tmp);
return (ret);
}
/*
* __conn_load_extension --
* WT_CONNECTION->load_extension method.
*/
static int
__conn_load_extension(
WT_CONNECTION *wt_conn, const char *path, const char *config)
{
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
WT_DLH *dlh;
WT_SESSION_IMPL *session;
int (*entry)(WT_SESSION *, WT_EXTENSION_API *, const char *);
const char *entry_name;
dlh = NULL;
conn = (WT_CONNECTION_IMPL *)wt_conn;
CONNECTION_API_CALL(conn, session, load_extension, config, cfg);
entry_name = NULL;
WT_ERR(__wt_config_gets(session, cfg, "entry", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &entry_name));
/*
* This assumes the underlying shared libraries are reference counted,
* that is, that re-opening a shared library simply increments a ref
* count, and closing it simply decrements the ref count, and the last
* close discards the reference entirely -- in other words, we do not
* check to see if we've already opened this shared library.
*/
WT_ERR(__wt_dlopen(session, path, &dlh));
WT_ERR(__wt_dlsym(session, dlh, entry_name, &entry));
/* Call the entry function. */
WT_ERR(entry(&session->iface, &__api, config));
/* Link onto the environment's list of open libraries. */
__wt_spin_lock(session, &conn->api_lock);
TAILQ_INSERT_TAIL(&conn->dlhqh, dlh, q);
__wt_spin_unlock(session, &conn->api_lock);
if (0) {
err: if (dlh != NULL)
WT_TRET(__wt_dlclose(session, dlh));
}
__wt_free(session, entry_name);
API_END_NOTFOUND_MAP(session, ret);
}
/*
* __config_merge_format --
* Take the sorted array of entries, and format them into allocated memory.
*/
static int
__config_merge_format(
WT_SESSION_IMPL *session, WT_CONFIG_MERGE *cp, const char **config_ret)
{
WT_DECL_ITEM(build);
WT_DECL_RET;
size_t entries;
WT_RET(__wt_scr_alloc(session, 4 * 1024, &build));
entries = 0;
WT_ERR(__config_merge_format_next(session, "", 0, &entries, cp, build));
__strip_comma(build);
ret = __wt_strndup(session, build->data, build->size, config_ret);
err: __wt_scr_free(session, &build);
return (ret);
}
/*
* __ckpt_last_name --
* Return the name associated with the file's last unnamed checkpoint.
*/
static int
__ckpt_last_name(
WT_SESSION_IMPL *session, const char *config, const char **namep)
{
WT_CONFIG ckptconf;
WT_CONFIG_ITEM a, k, v;
WT_DECL_RET;
int64_t found;
*namep = NULL;
WT_ERR(__wt_config_getones(session, config, "checkpoint", &v));
WT_ERR(__wt_config_subinit(session, &ckptconf, &v));
for (found = 0; __wt_config_next(&ckptconf, &k, &v) == 0;) {
/*
* We only care about unnamed checkpoints; applications may not
* use any matching prefix as a checkpoint name, the comparison
* is pretty simple.
*/
if (k.len < strlen(WT_CHECKPOINT) ||
strncmp(k.str, WT_CHECKPOINT, strlen(WT_CHECKPOINT)) != 0)
continue;
/* Ignore checkpoints before the ones we've already seen. */
WT_ERR(__wt_config_subgets(session, &v, "order", &a));
if (found && a.val < found)
continue;
if (*namep != NULL)
__wt_free(session, *namep);
WT_ERR(__wt_strndup(session, k.str, k.len, namep));
found = a.val;
}
if (!found)
ret = WT_NOTFOUND;
if (0) {
err: __wt_free(session, namep);
}
return (ret);
}
/*
* __wt_snapshot --
* Snapshot the tree.
*/
int
__wt_snapshot(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_CONFIG_ITEM cval;
WT_DECL_RET;
char *name;
name = NULL;
/* This may be a named snapshot, check the configuration. */
if ((ret = __wt_config_gets(
session, cfg, "snapshot", &cval)) != 0 && ret != WT_NOTFOUND)
WT_RET(ret);
if (cval.len != 0)
WT_RET(__wt_strndup(session, cval.str, cval.len, &name));
ret = __snapshot_worker(session, name, 0, SNAPSHOT);
__wt_free(session, name);
return (ret);
}
/*
* __wt_nfilename --
* Build a file name in a scratch buffer. If the name is already an
* absolute path duplicate it, otherwise generate a path relative to the
* connection home directory.
*/
int
__wt_nfilename(WT_SESSION_IMPL *session,
const char *name, size_t namelen, const char **path)
{
WT_CONNECTION_IMPL *conn;
size_t len;
char *buf;
conn = S2C(session);
*path = NULL;
if (__wt_absolute_path(name))
WT_RET(__wt_strndup(session, name, namelen, path));
else {
len = strlen(conn->home) + 1 + namelen + 1;
WT_RET(__wt_calloc(session, 1, len, &buf));
snprintf(buf, len, "%s/%.*s", conn->home, (int)namelen, name);
*path = buf;
}
return (0);
}
/*
* __huffman_confchk_file --
* Check for a Huffman configuration file and return the file name.
*/
static int
__huffman_confchk_file(
WT_SESSION_IMPL *session, WT_CONFIG_ITEM *v, int *is_utf8p, FILE **fpp)
{
FILE *fp;
WT_DECL_RET;
size_t len;
char *fname;
/* Look for a prefix and file name. */
len = 0;
if (is_utf8p != NULL)
*is_utf8p = 0;
if (WT_PREFIX_MATCH(v->str, "utf8")) {
if (is_utf8p != NULL)
*is_utf8p = 1;
len = strlen("utf8");
} else if (WT_PREFIX_MATCH(v->str, "utf16"))
len = strlen("utf16");
if (len == 0 || len >= v->len)
WT_RET_MSG(session, EINVAL,
"illegal Huffman configuration: %.*s", (int)v->len, v->str);
/* Check the file exists. */
WT_RET(__wt_strndup(session, v->str + len, v->len - len, &fname));
WT_ERR(__wt_fopen(session,
fname, WT_FHANDLE_READ, WT_FOPEN_FIXED, &fp));
/* Optionally return the file handle. */
if (fpp == NULL)
(void)__wt_fclose(&fp, WT_FHANDLE_READ);
else
*fpp = fp;
err: __wt_free(session, fname);
return (ret);
}
/*
* __ckpt_server_config --
* Parse and setup the checkpoint server options.
*/
static int
__ckpt_server_config(WT_SESSION_IMPL *session, const char **cfg, int *startp)
{
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
WT_DECL_ITEM(tmp);
WT_DECL_RET;
conn = S2C(session);
/*
* The checkpoint configuration requires a wait time -- if it's not set,
* we're not running at all.
*/
WT_RET(__wt_config_gets(session, cfg, "checkpoint.wait", &cval));
if (cval.val == 0) {
*startp = 0;
return (0);
}
conn->ckpt_usecs = (long)cval.val * 1000000;
*startp = 1;
WT_RET(__wt_config_gets(session, cfg, "checkpoint.name", &cval));
if (!WT_STRING_MATCH(WT_CHECKPOINT, cval.str, cval.len)) {
WT_RET(__wt_scr_alloc(session, cval.len + 20, &tmp));
strcpy((char *)tmp->data, "name=");
strncat((char *)tmp->data, cval.str, cval.len);
ret = __wt_strndup(session,
tmp->data, strlen("name=") + cval.len, &conn->ckpt_config);
__wt_scr_free(&tmp);
WT_RET(ret);
}
return (0);
}
/*
* __wt_schema_open_colgroups --
* Open the column groups for a table.
*/
int
__wt_schema_open_colgroups(WT_SESSION_IMPL *session, WT_TABLE *table)
{
WT_COLGROUP *colgroup;
WT_CONFIG cparser;
WT_CONFIG_ITEM ckey, cval;
WT_DECL_RET;
WT_DECL_ITEM(buf);
char *cgconfig;
u_int i;
WT_ASSERT(session, F_ISSET(session, WT_SESSION_LOCKED_TABLE));
if (table->cg_complete)
return (0);
colgroup = NULL;
cgconfig = NULL;
WT_RET(__wt_scr_alloc(session, 0, &buf));
WT_ERR(__wt_config_subinit(session, &cparser, &table->cgconf));
/* Open each column group. */
for (i = 0; i < WT_COLGROUPS(table); i++) {
if (table->ncolgroups > 0)
WT_ERR(__wt_config_next(&cparser, &ckey, &cval));
else
WT_CLEAR(ckey);
/*
* Always open from scratch: we may have failed part of the way
* through opening a table, or column groups may have changed.
*/
__wt_schema_destroy_colgroup(session, &table->cgroups[i]);
WT_ERR(__wt_buf_init(session, buf, 0));
WT_ERR(__wt_schema_colgroup_name(session, table,
ckey.str, ckey.len, buf));
if ((ret = __wt_metadata_search(
session, buf->data, &cgconfig)) != 0) {
/* It is okay if the table is incomplete. */
if (ret == WT_NOTFOUND)
ret = 0;
goto err;
}
WT_ERR(__wt_calloc_one(session, &colgroup));
WT_ERR(__wt_strndup(
session, buf->data, buf->size, &colgroup->name));
colgroup->config = cgconfig;
cgconfig = NULL;
WT_ERR(__wt_config_getones(session,
colgroup->config, "columns", &colgroup->colconf));
WT_ERR(__wt_config_getones(
session, colgroup->config, "source", &cval));
WT_ERR(__wt_strndup(
session, cval.str, cval.len, &colgroup->source));
table->cgroups[i] = colgroup;
colgroup = NULL;
}
if (!table->is_simple) {
WT_ERR(__wt_table_check(session, table));
WT_ERR(__wt_buf_init(session, buf, 0));
WT_ERR(__wt_struct_plan(session,
table, table->colconf.str, table->colconf.len, true, buf));
WT_ERR(__wt_strndup(
session, buf->data, buf->size, &table->plan));
}
table->cg_complete = true;
err: __wt_scr_free(session, &buf);
__wt_schema_destroy_colgroup(session, &colgroup);
if (cgconfig != NULL)
__wt_free(session, cgconfig);
return (ret);
}
/*
* __open_index --
* Open an index.
*/
static int
__open_index(WT_SESSION_IMPL *session, WT_TABLE *table, WT_INDEX *idx)
{
WT_CONFIG colconf;
WT_CONFIG_ITEM ckey, cval, metadata;
WT_DECL_ITEM(buf);
WT_DECL_ITEM(plan);
WT_DECL_RET;
u_int npublic_cols, i;
WT_ERR(__wt_scr_alloc(session, 0, &buf));
/* Get the data source from the index config. */
WT_ERR(__wt_config_getones(session, idx->config, "source", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &idx->source));
WT_ERR(__wt_config_getones(session, idx->config, "immutable", &cval));
if (cval.val)
F_SET(idx, WT_INDEX_IMMUTABLE);
/*
* Compatibility: we didn't always maintain collator information in
* index metadata, cope when it isn't found.
*/
WT_CLEAR(cval);
WT_ERR_NOTFOUND_OK(__wt_config_getones(
session, idx->config, "collator", &cval));
if (cval.len != 0) {
WT_CLEAR(metadata);
WT_ERR_NOTFOUND_OK(__wt_config_getones(
session, idx->config, "app_metadata", &metadata));
WT_ERR(__wt_collator_config(
session, idx->name, &cval, &metadata,
&idx->collator, &idx->collator_owned));
}
WT_ERR(__wt_extractor_config(
session, idx->name, idx->config, &idx->extractor,
&idx->extractor_owned));
WT_ERR(__wt_config_getones(session, idx->config, "key_format", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &idx->key_format));
/*
* The key format for an index is somewhat subtle: the application
* specifies a set of columns that it will use for the key, but the
* engine usually adds some hidden columns in order to derive the
* primary key. These hidden columns are part of the file's key.
*
* The file's key_format is stored persistently, we need to calculate
* the index cursor key format (which will usually omit some of those
* keys).
*/
WT_ERR(__wt_buf_init(session, buf, 0));
WT_ERR(__wt_config_getones(
session, idx->config, "columns", &idx->colconf));
/* Start with the declared index columns. */
WT_ERR(__wt_config_subinit(session, &colconf, &idx->colconf));
for (npublic_cols = 0;
(ret = __wt_config_next(&colconf, &ckey, &cval)) == 0;
++npublic_cols)
WT_ERR(__wt_buf_catfmt(
session, buf, "%.*s,", (int)ckey.len, ckey.str));
if (ret != WT_NOTFOUND)
goto err;
/*
* If we didn't find any columns, the index must have an extractor.
* We don't rely on this unconditionally because it was only added to
* the metadata after version 2.3.1.
*/
if (npublic_cols == 0) {
WT_ERR(__wt_config_getones(
session, idx->config, "index_key_columns", &cval));
npublic_cols = (u_int)cval.val;
WT_ASSERT(session, npublic_cols != 0);
for (i = 0; i < npublic_cols; i++)
WT_ERR(__wt_buf_catfmt(session, buf, "\"bad col\","));
}
/*
* Now add any primary key columns from the table that are not
* already part of the index key.
*/
WT_ERR(__wt_config_subinit(session, &colconf, &table->colconf));
for (i = 0; i < table->nkey_columns &&
(ret = __wt_config_next(&colconf, &ckey, &cval)) == 0;
i++) {
/*
* If the primary key column is already in the secondary key,
* don't add it again.
*/
if (__wt_config_subgetraw(
session, &idx->colconf, &ckey, &cval) == 0)
continue;
WT_ERR(__wt_buf_catfmt(
session, buf, "%.*s,", (int)ckey.len, ckey.str));
}
WT_ERR_NOTFOUND_OK(ret);
/*
* If the table doesn't yet have its column groups, don't try to
* calculate a plan: we are just checking that the index creation is
* sane.
*/
if (!table->cg_complete)
goto err;
WT_ERR(__wt_scr_alloc(session, 0, &plan));
WT_ERR(__wt_struct_plan(
session, table, buf->data, buf->size, false, plan));
WT_ERR(__wt_strndup(session, plan->data, plan->size, &idx->key_plan));
/* Set up the cursor key format (the visible columns). */
WT_ERR(__wt_buf_init(session, buf, 0));
WT_ERR(__wt_struct_truncate(session,
idx->key_format, npublic_cols, buf));
WT_ERR(__wt_strndup(
session, buf->data, buf->size, &idx->idxkey_format));
/*
* Add a trailing padding byte to the format. This ensures that there
* will be no special optimization of the last column, so the primary
* key columns can be simply appended.
*/
WT_ERR(__wt_buf_catfmt(session, buf, "x"));
WT_ERR(__wt_strndup(session, buf->data, buf->size, &idx->exkey_format));
/* By default, index cursor values are the table value columns. */
/* TODO Optimize to use index columns in preference to table lookups. */
WT_ERR(__wt_buf_init(session, plan, 0));
WT_ERR(__wt_struct_plan(session,
table, table->colconf.str, table->colconf.len, true, plan));
WT_ERR(__wt_strndup(session, plan->data, plan->size, &idx->value_plan));
err: __wt_scr_free(session, &buf);
__wt_scr_free(session, &plan);
return (ret);
}
/*
* __wt_lsm_tree_create --
* Create an LSM tree structure for the given name.
*/
int
__wt_lsm_tree_create(WT_SESSION_IMPL *session,
const char *uri, int exclusive, const char *config)
{
WT_CONFIG_ITEM cval;
WT_DECL_ITEM(buf);
WT_DECL_RET;
WT_LSM_TREE *lsm_tree;
const char *cfg[] =
{ WT_CONFIG_BASE(session, session_create), config, NULL };
char *tmpconfig;
/* If the tree is open, it already exists. */
if ((ret = __wt_lsm_tree_get(session, uri, 0, &lsm_tree)) == 0) {
__wt_lsm_tree_release(session, lsm_tree);
return (exclusive ? EEXIST : 0);
}
WT_RET_NOTFOUND_OK(ret);
/*
* If the tree has metadata, it already exists.
*
* !!!
* Use a local variable: we don't care what the existing configuration
* is, but we don't want to overwrite the real config.
*/
if (__wt_metadata_search(session, uri, &tmpconfig) == 0) {
__wt_free(session, tmpconfig);
return (exclusive ? EEXIST : 0);
}
WT_RET_NOTFOUND_OK(ret);
WT_RET(__wt_config_gets(session, cfg, "key_format", &cval));
if (WT_STRING_MATCH("r", cval.str, cval.len))
WT_RET_MSG(session, EINVAL,
"LSM trees cannot be configured as column stores");
WT_RET(__wt_calloc_def(session, 1, &lsm_tree));
WT_ERR(__lsm_tree_set_name(session, lsm_tree, uri));
WT_ERR(__wt_config_gets(session, cfg, "key_format", &cval));
WT_ERR(__wt_strndup(
session, cval.str, cval.len, &lsm_tree->key_format));
WT_ERR(__wt_config_gets(session, cfg, "value_format", &cval));
WT_ERR(__wt_strndup(
session, cval.str, cval.len, &lsm_tree->value_format));
WT_ERR(__wt_config_gets(session, cfg, "collator", &cval));
WT_ERR(__wt_strndup(
session, cval.str, cval.len, &lsm_tree->collator_name));
WT_ERR(__wt_config_gets(session, cfg, "lsm.auto_throttle", &cval));
if (cval.val)
F_SET(lsm_tree, WT_LSM_TREE_THROTTLE);
else
F_CLR(lsm_tree, WT_LSM_TREE_THROTTLE);
WT_ERR(__wt_config_gets(session, cfg, "lsm.bloom", &cval));
FLD_SET(lsm_tree->bloom,
(cval.val == 0 ? WT_LSM_BLOOM_OFF : WT_LSM_BLOOM_MERGED));
WT_ERR(__wt_config_gets(session, cfg, "lsm.bloom_oldest", &cval));
if (cval.val != 0)
FLD_SET(lsm_tree->bloom, WT_LSM_BLOOM_OLDEST);
if (FLD_ISSET(lsm_tree->bloom, WT_LSM_BLOOM_OFF) &&
FLD_ISSET(lsm_tree->bloom, WT_LSM_BLOOM_OLDEST))
WT_ERR_MSG(session, EINVAL,
"Bloom filters can only be created on newest and oldest "
"chunks if bloom filters are enabled");
WT_ERR(__wt_config_gets(session, cfg, "lsm.bloom_config", &cval));
if (cval.type == WT_CONFIG_ITEM_STRUCT) {
cval.str++;
cval.len -= 2;
}
WT_ERR(__wt_strndup(
session, cval.str, cval.len, &lsm_tree->bloom_config));
WT_ERR(__wt_config_gets(session, cfg, "lsm.bloom_bit_count", &cval));
lsm_tree->bloom_bit_count = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm.bloom_hash_count", &cval));
lsm_tree->bloom_hash_count = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm.chunk_max", &cval));
lsm_tree->chunk_max = (uint64_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm.chunk_size", &cval));
lsm_tree->chunk_size = (uint64_t)cval.val;
if (lsm_tree->chunk_size > lsm_tree->chunk_max)
WT_ERR_MSG(session, EINVAL,
"Chunk size (chunk_size) must be smaller than or equal to "
"the maximum chunk size (chunk_max)");
WT_ERR(__wt_config_gets(session, cfg, "lsm.merge_max", &cval));
lsm_tree->merge_max = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm.merge_min", &cval));
lsm_tree->merge_min = (uint32_t)cval.val;
if (lsm_tree->merge_min > lsm_tree->merge_max)
WT_ERR_MSG(session, EINVAL,
"LSM merge_min must be less than or equal to merge_max");
/*
* Set up the config for each chunk.
*
* Make the memory_page_max double the chunk size, so application
* threads don't immediately try to force evict the chunk when the
* worker thread clears the NO_EVICTION flag.
*/
WT_ERR(__wt_scr_alloc(session, 0, &buf));
WT_ERR(__wt_buf_fmt(session, buf,
"%s,key_format=u,value_format=u,memory_page_max=%" PRIu64,
config, 2 * lsm_tree->chunk_max));
WT_ERR(__wt_strndup(
session, buf->data, buf->size, &lsm_tree->file_config));
/* Create the first chunk and flush the metadata. */
WT_ERR(__wt_lsm_meta_write(session, lsm_tree));
/* Discard our partially populated handle. */
ret = __lsm_tree_discard(session, lsm_tree);
lsm_tree = NULL;
/*
* Open our new tree and add it to the handle cache. Don't discard on
* error: the returned handle is NULL on error, and the metadata
* tracking macros handle cleaning up on failure.
*/
if (ret == 0)
ret = __lsm_tree_open(session, uri, &lsm_tree);
if (ret == 0)
__wt_lsm_tree_release(session, lsm_tree);
if (0) {
err: WT_TRET(__lsm_tree_discard(session, lsm_tree));
}
__wt_scr_free(&buf);
return (ret);
}
/*
* __wt_lsm_meta_read --
* Read the metadata for an LSM tree.
*/
int
__wt_lsm_meta_read(WT_SESSION_IMPL *session, WT_LSM_TREE *lsm_tree)
{
WT_CONFIG cparser, lparser;
WT_CONFIG_ITEM ck, cv, fileconf, lk, lv, metadata;
WT_DECL_RET;
WT_LSM_CHUNK *chunk;
char *lsmconfig;
u_int nchunks;
chunk = NULL; /* -Wconditional-uninitialized */
/* LSM trees inherit the merge setting from the connection. */
if (F_ISSET(S2C(session), WT_CONN_LSM_MERGE))
F_SET(lsm_tree, WT_LSM_TREE_MERGES);
WT_RET(__wt_metadata_search(session, lsm_tree->name, &lsmconfig));
WT_ERR(__wt_config_init(session, &cparser, lsmconfig));
while ((ret = __wt_config_next(&cparser, &ck, &cv)) == 0) {
if (WT_STRING_MATCH("key_format", ck.str, ck.len)) {
__wt_free(session, lsm_tree->key_format);
WT_ERR(__wt_strndup(session,
cv.str, cv.len, &lsm_tree->key_format));
} else if (WT_STRING_MATCH("value_format", ck.str, ck.len)) {
__wt_free(session, lsm_tree->value_format);
WT_ERR(__wt_strndup(session,
cv.str, cv.len, &lsm_tree->value_format));
} else if (WT_STRING_MATCH("collator", ck.str, ck.len)) {
if (cv.len == 0 ||
WT_STRING_MATCH("none", cv.str, cv.len))
continue;
/*
* Extract the application-supplied metadata (if any)
* from the file configuration.
*/
WT_ERR(__wt_config_getones(
session, lsmconfig, "file_config", &fileconf));
WT_CLEAR(metadata);
WT_ERR_NOTFOUND_OK(__wt_config_subgets(
session, &fileconf, "app_metadata", &metadata));
WT_ERR(__wt_collator_config(session, lsm_tree->name,
&cv, &metadata,
&lsm_tree->collator, &lsm_tree->collator_owned));
WT_ERR(__wt_strndup(session,
cv.str, cv.len, &lsm_tree->collator_name));
} else if (WT_STRING_MATCH("bloom_config", ck.str, ck.len)) {
__wt_free(session, lsm_tree->bloom_config);
/* Don't include the brackets. */
WT_ERR(__wt_strndup(session,
cv.str + 1, cv.len - 2, &lsm_tree->bloom_config));
} else if (WT_STRING_MATCH("file_config", ck.str, ck.len)) {
__wt_free(session, lsm_tree->file_config);
/* Don't include the brackets. */
WT_ERR(__wt_strndup(session,
cv.str + 1, cv.len - 2, &lsm_tree->file_config));
} else if (WT_STRING_MATCH("auto_throttle", ck.str, ck.len)) {
if (cv.val)
F_SET(lsm_tree, WT_LSM_TREE_THROTTLE);
else
F_CLR(lsm_tree, WT_LSM_TREE_THROTTLE);
} else if (WT_STRING_MATCH("bloom", ck.str, ck.len))
lsm_tree->bloom = (uint32_t)cv.val;
else if (WT_STRING_MATCH("bloom_bit_count", ck.str, ck.len))
lsm_tree->bloom_bit_count = (uint32_t)cv.val;
else if (WT_STRING_MATCH("bloom_hash_count", ck.str, ck.len))
lsm_tree->bloom_hash_count = (uint32_t)cv.val;
else if (WT_STRING_MATCH("chunk_count_limit", ck.str, ck.len)) {
lsm_tree->chunk_count_limit = (uint32_t)cv.val;
if (cv.val != 0)
F_CLR(lsm_tree, WT_LSM_TREE_MERGES);
} else if (WT_STRING_MATCH("chunk_max", ck.str, ck.len))
lsm_tree->chunk_max = (uint64_t)cv.val;
else if (WT_STRING_MATCH("chunk_size", ck.str, ck.len))
lsm_tree->chunk_size = (uint64_t)cv.val;
else if (WT_STRING_MATCH("merge_max", ck.str, ck.len))
lsm_tree->merge_max = (uint32_t)cv.val;
else if (WT_STRING_MATCH("merge_min", ck.str, ck.len))
lsm_tree->merge_min = (uint32_t)cv.val;
else if (WT_STRING_MATCH("last", ck.str, ck.len))
lsm_tree->last = (u_int)cv.val;
else if (WT_STRING_MATCH("chunks", ck.str, ck.len)) {
WT_ERR(__wt_config_subinit(session, &lparser, &cv));
for (nchunks = 0; (ret =
__wt_config_next(&lparser, &lk, &lv)) == 0; ) {
if (WT_STRING_MATCH("id", lk.str, lk.len)) {
WT_ERR(__wt_realloc_def(session,
&lsm_tree->chunk_alloc,
nchunks + 1, &lsm_tree->chunk));
WT_ERR(
__wt_calloc_one(session, &chunk));
lsm_tree->chunk[nchunks++] = chunk;
chunk->id = (uint32_t)lv.val;
WT_ERR(__wt_lsm_tree_chunk_name(session,
lsm_tree, chunk->id, &chunk->uri));
F_SET(chunk,
WT_LSM_CHUNK_ONDISK |
WT_LSM_CHUNK_STABLE);
} else if (WT_STRING_MATCH(
"bloom", lk.str, lk.len)) {
WT_ERR(__wt_lsm_tree_bloom_name(
session, lsm_tree,
chunk->id, &chunk->bloom_uri));
F_SET(chunk, WT_LSM_CHUNK_BLOOM);
continue;
} else if (WT_STRING_MATCH(
"chunk_size", lk.str, lk.len)) {
chunk->size = (uint64_t)lv.val;
continue;
} else if (WT_STRING_MATCH(
"count", lk.str, lk.len)) {
chunk->count = (uint64_t)lv.val;
continue;
} else if (WT_STRING_MATCH(
"generation", lk.str, lk.len)) {
chunk->generation = (uint32_t)lv.val;
continue;
}
}
WT_ERR_NOTFOUND_OK(ret);
lsm_tree->nchunks = nchunks;
} else if (WT_STRING_MATCH("old_chunks", ck.str, ck.len)) {
WT_ERR(__wt_config_subinit(session, &lparser, &cv));
for (nchunks = 0; (ret =
__wt_config_next(&lparser, &lk, &lv)) == 0; ) {
if (WT_STRING_MATCH("bloom", lk.str, lk.len)) {
WT_ERR(__wt_strndup(session,
lv.str, lv.len, &chunk->bloom_uri));
F_SET(chunk, WT_LSM_CHUNK_BLOOM);
continue;
}
WT_ERR(__wt_realloc_def(session,
&lsm_tree->old_alloc, nchunks + 1,
&lsm_tree->old_chunks));
WT_ERR(__wt_calloc_one(session, &chunk));
lsm_tree->old_chunks[nchunks++] = chunk;
WT_ERR(__wt_strndup(session,
lk.str, lk.len, &chunk->uri));
F_SET(chunk, WT_LSM_CHUNK_ONDISK);
}
WT_ERR_NOTFOUND_OK(ret);
lsm_tree->nold_chunks = nchunks;
}
/*
* Ignore any other values: the metadata entry might have been
* created by a future release, with unknown options.
*/
}
WT_ERR_NOTFOUND_OK(ret);
/*
* If the default merge_min was not overridden, calculate it now. We
* do this here so that trees created before merge_min was added get a
* sane value.
*/
if (lsm_tree->merge_min < 2)
lsm_tree->merge_min = WT_MAX(2, lsm_tree->merge_max / 2);
err: __wt_free(session, lsmconfig);
return (ret);
}
/*
* __wt_lsm_tree_create --
* Create an LSM tree structure for the given name.
*/
int
__wt_lsm_tree_create(WT_SESSION_IMPL *session,
const char *uri, int exclusive, const char *config)
{
WT_CONFIG_ITEM cval;
WT_DECL_ITEM(buf);
WT_DECL_RET;
WT_LSM_TREE *lsm_tree;
const char *cfg[] = API_CONF_DEFAULTS(session, create, config);
const char *tmpconfig;
/* If the tree is open, it already exists. */
if ((ret = __wt_lsm_tree_get(session, uri, 0, &lsm_tree)) == 0) {
__wt_lsm_tree_release(session, lsm_tree);
return (exclusive ? EEXIST : 0);
}
WT_RET_NOTFOUND_OK(ret);
/*
* If the tree has metadata, it already exists.
*
* !!!
* Use a local variable: we don't care what the existing configuration
* is, but we don't want to overwrite the real config.
*/
if (__wt_metadata_read(session, uri, &tmpconfig) == 0) {
__wt_free(session, tmpconfig);
return (exclusive ? EEXIST : 0);
}
WT_RET_NOTFOUND_OK(ret);
WT_RET(__wt_config_gets(session, cfg, "key_format", &cval));
if (WT_STRING_MATCH("r", cval.str, cval.len))
WT_RET_MSG(session, EINVAL,
"LSM trees cannot be configured as column stores");
WT_RET(__wt_calloc_def(session, 1, &lsm_tree));
WT_RET(__lsm_tree_set_name(session, lsm_tree, uri));
WT_ERR(__wt_config_gets(session, cfg, "key_format", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len,
&lsm_tree->key_format));
WT_ERR(__wt_config_gets(session, cfg, "value_format", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len,
&lsm_tree->value_format));
WT_ERR(__wt_config_gets(session, cfg, "lsm_bloom", &cval));
FLD_SET(lsm_tree->bloom,
(cval.val == 0 ? WT_LSM_BLOOM_OFF : WT_LSM_BLOOM_MERGED));
WT_ERR(__wt_config_gets(session, cfg, "lsm_bloom_newest", &cval));
if (cval.val != 0)
FLD_SET(lsm_tree->bloom, WT_LSM_BLOOM_NEWEST);
WT_ERR(__wt_config_gets(session, cfg, "lsm_bloom_oldest", &cval));
if (cval.val != 0)
FLD_SET(lsm_tree->bloom, WT_LSM_BLOOM_OLDEST);
if (FLD_ISSET(lsm_tree->bloom, WT_LSM_BLOOM_OFF) &&
(FLD_ISSET(lsm_tree->bloom, WT_LSM_BLOOM_NEWEST) ||
FLD_ISSET(lsm_tree->bloom, WT_LSM_BLOOM_OLDEST)))
WT_ERR_MSG(session, EINVAL,
"Bloom filters can only be created on newest and oldest "
"chunks if bloom filters are enabled");
WT_ERR(__wt_config_gets(session, cfg, "lsm_bloom_config", &cval));
if (cval.type == ITEM_STRUCT) {
cval.str++;
cval.len -= 2;
}
WT_ERR(__wt_strndup(session, cval.str, cval.len,
&lsm_tree->bloom_config));
WT_ERR(__wt_config_gets(session, cfg, "lsm_bloom_bit_count", &cval));
lsm_tree->bloom_bit_count = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm_bloom_hash_count", &cval));
lsm_tree->bloom_hash_count = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm_chunk_size", &cval));
lsm_tree->chunk_size = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm_merge_max", &cval));
lsm_tree->merge_max = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "lsm_merge_threads", &cval));
lsm_tree->merge_threads = (uint32_t)cval.val;
/* Sanity check that api_data.py is in sync with lsm.h */
WT_ASSERT(session, lsm_tree->merge_threads <= WT_LSM_MAX_WORKERS);
WT_ERR(__wt_scr_alloc(session, 0, &buf));
WT_ERR(__wt_buf_fmt(session, buf,
"%s,key_format=u,value_format=u", config));
lsm_tree->file_config = __wt_buf_steal(session, buf, NULL);
/* Create the first chunk and flush the metadata. */
WT_ERR(__wt_lsm_meta_write(session, lsm_tree));
/* Discard our partially populated handle. */
ret = __lsm_tree_discard(session, lsm_tree);
lsm_tree = NULL;
/*
* Open our new tree and add it to the handle cache. Don't discard on
* error: the returned handle is NULL on error, and the metadata
* tracking macros handle cleaning up on failure.
*/
if (ret == 0)
ret = __lsm_tree_open(session, uri, &lsm_tree);
if (ret == 0)
__wt_lsm_tree_release(session, lsm_tree);
if (0) {
err: WT_TRET(__lsm_tree_discard(session, lsm_tree));
}
__wt_scr_free(&buf);
return (ret);
}
/*
* __split_ref_deepen_move --
* Move a WT_REF from a parent to a child in service of a split to deepen
* the tree, including updating the accounting information.
*/
static int
__split_ref_deepen_move(WT_SESSION_IMPL *session,
WT_PAGE *parent, WT_REF *ref, size_t *parent_decrp, size_t *child_incrp)
{
WT_ADDR *addr;
WT_CELL_UNPACK unpack;
WT_DECL_RET;
WT_IKEY *ikey;
size_t size;
void *key;
/*
* Instantiate row-store keys, and column- and row-store addresses in
* the WT_REF structures referenced by a page that's being split (and
* deepening the tree). The WT_REF structures aren't moving, but the
* index references are moving from the page we're splitting to a set
* of child pages, and so we can no longer reference the block image
* that remains with the page being split.
*
* No locking is required to update the WT_REF structure because we're
* the only thread splitting the parent page, and there's no way for
* readers to race with our updates of single pointers. The changes
* have to be written before the page goes away, of course, our caller
* owns that problem.
*
* Row-store keys, first.
*/
if (parent->type == WT_PAGE_ROW_INT) {
if ((ikey = __wt_ref_key_instantiated(ref)) == NULL) {
__wt_ref_key(parent, ref, &key, &size);
WT_RET(__wt_row_ikey(session, 0, key, size, ref));
ikey = ref->key.ikey;
} else {
WT_RET(__split_ovfl_key_cleanup(session, parent, ref));
*parent_decrp += sizeof(WT_IKEY) + ikey->size;
}
*child_incrp += sizeof(WT_IKEY) + ikey->size;
}
/*
* If there's no address (the page has never been written), or the
* address has been instantiated, there's no work to do. Otherwise,
* get the address from the on-page cell.
*/
addr = ref->addr;
if (addr != NULL && !__wt_off_page(parent, addr)) {
__wt_cell_unpack((WT_CELL *)ref->addr, &unpack);
WT_RET(__wt_calloc_one(session, &addr));
if ((ret = __wt_strndup(
session, unpack.data, unpack.size, &addr->addr)) != 0) {
__wt_free(session, addr);
return (ret);
}
addr->size = (uint8_t)unpack.size;
addr->type =
unpack.raw == WT_CELL_ADDR_INT ? WT_ADDR_INT : WT_ADDR_LEAF;
ref->addr = addr;
}
/* And finally, the WT_REF itself. */
WT_MEM_TRANSFER(*parent_decrp, *child_incrp, sizeof(WT_REF));
return (0);
}
/*
* __logmgr_config --
* Parse and setup the logging server options.
*/
static int
__logmgr_config(
WT_SESSION_IMPL *session, const char **cfg, bool *runp, bool reconfig)
{
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
bool enabled;
conn = S2C(session);
WT_RET(__wt_config_gets(session, cfg, "log.enabled", &cval));
enabled = cval.val != 0;
/*
* If we're reconfiguring, enabled must match the already
* existing setting.
*
* If it is off and the user it turning it on, or it is on
* and the user is turning it off, return an error.
*/
if (reconfig &&
((enabled && !FLD_ISSET(conn->log_flags, WT_CONN_LOG_ENABLED)) ||
(!enabled && FLD_ISSET(conn->log_flags, WT_CONN_LOG_ENABLED))))
return (EINVAL);
/* Logging is incompatible with in-memory */
if (enabled) {
WT_RET(__wt_config_gets(session, cfg, "in_memory", &cval));
if (cval.val != 0)
WT_RET_MSG(session, EINVAL,
"In memory configuration incompatible with "
"log=(enabled=true)");
}
*runp = enabled;
/*
* Setup a log path and compression even if logging is disabled in case
* we are going to print a log. Only do this on creation. Once a
* compressor or log path are set they cannot be changed.
*/
if (!reconfig) {
conn->log_compressor = NULL;
WT_RET(__wt_config_gets_none(
session, cfg, "log.compressor", &cval));
WT_RET(__wt_compressor_config(
session, &cval, &conn->log_compressor));
WT_RET(__wt_config_gets(session, cfg, "log.path", &cval));
WT_RET(__wt_strndup(
session, cval.str, cval.len, &conn->log_path));
}
/* We are done if logging isn't enabled. */
if (!*runp)
return (0);
WT_RET(__wt_config_gets(session, cfg, "log.archive", &cval));
if (cval.val != 0)
FLD_SET(conn->log_flags, WT_CONN_LOG_ARCHIVE);
if (!reconfig) {
/*
* Ignore if the user tries to change the file size. The
* amount of memory allocated to the log slots may be based
* on the log file size at creation and we don't want to
* re-allocate that memory while running.
*/
WT_RET(__wt_config_gets(session, cfg, "log.file_max", &cval));
conn->log_file_max = (wt_off_t)cval.val;
WT_STAT_FAST_CONN_SET(session,
log_max_filesize, conn->log_file_max);
}
/*
* If pre-allocation is configured, set the initial number to a few.
* We'll adapt as load dictates.
*/
WT_RET(__wt_config_gets(session, cfg, "log.prealloc", &cval));
if (cval.val != 0)
conn->log_prealloc = 1;
/*
* Note that it is meaningless to reconfigure this value during
* runtime. It only matters on create before recovery runs.
*/
WT_RET(__wt_config_gets_def(session, cfg, "log.recover", 0, &cval));
if (cval.len != 0 && WT_STRING_MATCH("error", cval.str, cval.len))
FLD_SET(conn->log_flags, WT_CONN_LOG_RECOVER_ERR);
WT_RET(__wt_config_gets(session, cfg, "log.zero_fill", &cval));
if (cval.val != 0) {
if (F_ISSET(conn, WT_CONN_READONLY))
WT_RET_MSG(session, EINVAL,
"Read-only configuration incompatible with "
"zero-filling log files");
FLD_SET(conn->log_flags, WT_CONN_LOG_ZERO_FILL);
}
WT_RET(__logmgr_sync_cfg(session, cfg));
if (conn->log_cond != NULL)
WT_RET(__wt_cond_auto_signal(session, conn->log_cond));
return (0);
}
/*
* __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);
}
/*
* __wt_curindex_open --
* WT_SESSION->open_cursor method for index cursors.
*/
int
__wt_curindex_open(WT_SESSION_IMPL *session,
const char *uri, WT_CURSOR *owner, const char *cfg[], WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__wt_cursor_get_key, /* get-key */
__curindex_get_value, /* get-value */
__wt_cursor_set_key, /* set-key */
__curindex_set_value, /* set-value */
__curindex_compare, /* compare */
__wt_cursor_equals, /* equals */
__curindex_next, /* next */
__curindex_prev, /* prev */
__curindex_reset, /* reset */
__curindex_search, /* search */
__curindex_search_near, /* search-near */
__wt_cursor_notsup, /* insert */
__wt_cursor_notsup, /* update */
__wt_cursor_notsup, /* remove */
__wt_cursor_reconfigure_notsup, /* reconfigure */
__curindex_close); /* close */
WT_CURSOR_INDEX *cindex;
WT_CURSOR *cursor;
WT_DECL_ITEM(tmp);
WT_DECL_RET;
WT_INDEX *idx;
WT_TABLE *table;
const char *columns, *idxname, *tablename;
size_t namesize;
tablename = uri;
if (!WT_PREFIX_SKIP(tablename, "index:") ||
(idxname = strchr(tablename, ':')) == NULL)
WT_RET_MSG(session, EINVAL, "Invalid cursor URI: '%s'", uri);
namesize = (size_t)(idxname - tablename);
++idxname;
if ((ret = __wt_schema_get_table(session,
tablename, namesize, false, &table)) != 0) {
if (ret == WT_NOTFOUND)
WT_RET_MSG(session, EINVAL,
"Cannot open cursor '%s' on unknown table", uri);
return (ret);
}
columns = strchr(idxname, '(');
if (columns == NULL)
namesize = strlen(idxname);
else
namesize = (size_t)(columns - idxname);
if ((ret = __wt_schema_open_index(
session, table, idxname, namesize, &idx)) != 0) {
__wt_schema_release_table(session, table);
return (ret);
}
WT_RET(__wt_calloc_one(session, &cindex));
cursor = &cindex->iface;
*cursor = iface;
cursor->session = &session->iface;
cindex->table = table;
cindex->index = idx;
cindex->key_plan = idx->key_plan;
cindex->value_plan = idx->value_plan;
cursor->internal_uri = idx->name;
cursor->key_format = idx->idxkey_format;
cursor->value_format = table->value_format;
/*
* XXX
* A very odd corner case is an index with a recno key.
* The only way to get here is by creating an index on a column store
* using only the primary's recno as the index key. Disallow that for
* now.
*/
if (WT_CURSOR_RECNO(cursor))
WT_ERR_MSG(session, WT_ERROR,
"Column store indexes based on a record number primary "
"key are not supported");
/* Handle projections. */
if (columns != NULL) {
WT_ERR(__wt_scr_alloc(session, 0, &tmp));
WT_ERR(__wt_struct_reformat(session, table,
columns, strlen(columns), NULL, false, tmp));
WT_ERR(__wt_strndup(
session, tmp->data, tmp->size, &cursor->value_format));
WT_ERR(__wt_buf_init(session, tmp, 0));
WT_ERR(__wt_struct_plan(session, table,
columns, strlen(columns), false, tmp));
WT_ERR(__wt_strndup(
session, tmp->data, tmp->size, &cindex->value_plan));
}
WT_ERR(__wt_cursor_init(
cursor, cursor->internal_uri, owner, cfg, cursorp));
WT_ERR(__wt_open_cursor(
session, idx->source, cursor, cfg, &cindex->child));
/* Open the column groups needed for this index cursor. */
WT_ERR(__curindex_open_colgroups(session, cindex, cfg));
if (F_ISSET(cursor, WT_CURSTD_DUMP_JSON))
__wt_json_column_init(
cursor, table->key_format, &idx->colconf, &table->colconf);
if (0) {
err: WT_TRET(__curindex_close(cursor));
*cursorp = NULL;
}
__wt_scr_free(session, &tmp);
return (ret);
}
/*
* __wt_schema_open_table --
* Open a named table.
*/
int
__wt_schema_open_table(WT_SESSION_IMPL *session,
const char *name, size_t namelen, WT_TABLE **tablep)
{
WT_CONFIG cparser;
WT_CONFIG_ITEM ckey, cval;
WT_CURSOR *cursor;
WT_DECL_RET;
WT_ITEM buf;
WT_TABLE *table;
const char *tconfig;
char *tablename;
cursor = NULL;
table = NULL;
WT_CLEAR(buf);
WT_RET(__wt_buf_fmt(session, &buf, "table:%.*s", (int)namelen, name));
tablename = __wt_buf_steal(session, &buf, NULL);
WT_ERR(__wt_metadata_cursor(session, NULL, &cursor));
cursor->set_key(cursor, tablename);
WT_ERR(cursor->search(cursor));
WT_ERR(cursor->get_value(cursor, &tconfig));
WT_ERR(__wt_calloc_def(session, 1, &table));
table->name = tablename;
tablename = NULL;
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 = 1;
while ((ret = __wt_config_next(&cparser, &ckey, &cval)) == 0)
table->is_simple = 0;
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;
WT_ERR(__wt_calloc_def(session, WT_COLGROUPS(table), &table->cgroups));
WT_ERR(__wt_schema_open_colgroups(session, table));
*tablep = table;
if (0) {
err: if (table != NULL)
__wt_schema_destroy_table(session, table);
}
if (cursor != NULL)
WT_TRET(cursor->close(cursor));
__wt_free(session, tablename);
return (ret);
}
/*
* __wt_strdup --
* ANSI strdup function.
*/
int
__wt_strdup(WT_SESSION_IMPL *session, const char *str, void *retp)
{
return (__wt_strndup(
session, str, (str == NULL) ? 0 : strlen(str), retp));
}
/*
* __logmgr_config --
* Parse and setup the logging server options.
*/
static int
__logmgr_config(
WT_SESSION_IMPL *session, const char **cfg, bool *runp, bool reconfig)
{
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
bool enabled;
/*
* A note on reconfiguration: the standard "is this configuration string
* allowed" checks should fail if reconfiguration has invalid strings,
* for example, "log=(enabled)", or "statistics_log=(path=XXX)", because
* the connection reconfiguration method doesn't allow those strings.
* Additionally, the base configuration values during reconfiguration
* are the currently configured values (so we don't revert to default
* values when repeatedly reconfiguring), and configuration processing
* of a currently set value should not change the currently set value.
*
* In this code path, log server reconfiguration does not stop/restart
* the log server, so there's no point in re-evaluating configuration
* strings that cannot be reconfigured, risking bugs in configuration
* setup, and depending on evaluation of currently set values to always
* result in the currently set value. Skip tests for any configuration
* strings which don't make sense during reconfiguration, but don't
* worry about error reporting because it should never happen.
*/
conn = S2C(session);
WT_RET(__wt_config_gets(session, cfg, "log.enabled", &cval));
enabled = cval.val != 0;
/*
* If we're reconfiguring, enabled must match the already
* existing setting.
*
* If it is off and the user it turning it on, or it is on
* and the user is turning it off, return an error.
*
* See above: should never happen.
*/
if (reconfig &&
((enabled && !FLD_ISSET(conn->log_flags, WT_CONN_LOG_ENABLED)) ||
(!enabled && FLD_ISSET(conn->log_flags, WT_CONN_LOG_ENABLED))))
return (EINVAL);
/* Logging is incompatible with in-memory */
if (enabled) {
WT_RET(__wt_config_gets(session, cfg, "in_memory", &cval));
if (cval.val != 0)
WT_RET_MSG(session, EINVAL,
"In memory configuration incompatible with "
"log=(enabled=true)");
}
*runp = enabled;
/*
* Setup a log path and compression even if logging is disabled in case
* we are going to print a log. Only do this on creation. Once a
* compressor or log path are set they cannot be changed.
*
* See above: should never happen.
*/
if (!reconfig) {
conn->log_compressor = NULL;
WT_RET(__wt_config_gets_none(
session, cfg, "log.compressor", &cval));
WT_RET(__wt_compressor_config(
session, &cval, &conn->log_compressor));
WT_RET(__wt_config_gets(session, cfg, "log.path", &cval));
WT_RET(__wt_strndup(
session, cval.str, cval.len, &conn->log_path));
}
/* We are done if logging isn't enabled. */
if (!*runp)
return (0);
WT_RET(__wt_config_gets(session, cfg, "log.archive", &cval));
if (cval.val != 0)
FLD_SET(conn->log_flags, WT_CONN_LOG_ARCHIVE);
/*
* The file size cannot be reconfigured. The amount of memory allocated
* to the log slots may be based on the log file size at creation and we
* don't want to re-allocate that memory while running.
*
* See above: should never happen.
*/
if (!reconfig) {
WT_RET(__wt_config_gets(session, cfg, "log.file_max", &cval));
conn->log_file_max = (wt_off_t)cval.val;
WT_STAT_FAST_CONN_SET(session,
log_max_filesize, conn->log_file_max);
}
/*
* If pre-allocation is configured, set the initial number to a few.
* We'll adapt as load dictates.
*/
WT_RET(__wt_config_gets(session, cfg, "log.prealloc", &cval));
if (cval.val != 0)
conn->log_prealloc = 1;
/*
* Note it's meaningless to reconfigure this value during runtime, it
* only matters on create before recovery runs.
*
* See above: should never happen.
*/
if (!reconfig) {
WT_RET(__wt_config_gets_def(
session, cfg, "log.recover", 0, &cval));
if (WT_STRING_MATCH("error", cval.str, cval.len))
FLD_SET(conn->log_flags, WT_CONN_LOG_RECOVER_ERR);
}
WT_RET(__wt_config_gets(session, cfg, "log.zero_fill", &cval));
if (cval.val != 0) {
if (F_ISSET(conn, WT_CONN_READONLY))
WT_RET_MSG(session, EINVAL,
"Read-only configuration incompatible with "
"zero-filling log files");
FLD_SET(conn->log_flags, WT_CONN_LOG_ZERO_FILL);
}
WT_RET(__logmgr_sync_cfg(session, cfg));
if (conn->log_cond != NULL)
WT_RET(__wt_cond_auto_signal(session, conn->log_cond));
return (0);
}
/*
* __wt_conn_optrack_setup --
* Set up operation logging.
*/
int
__wt_conn_optrack_setup(WT_SESSION_IMPL *session,
const char *cfg[], bool reconfig)
{
WT_CONFIG_ITEM cval;
WT_CONNECTION_IMPL *conn;
WT_DECL_ITEM(buf);
WT_DECL_RET;
conn = S2C(session);
/* Once an operation tracking path has been set it can't be changed. */
if (!reconfig) {
WT_RET(__wt_config_gets(session,
cfg, "operation_tracking.path", &cval));
WT_RET(__wt_strndup(session,
cval.str, cval.len, &conn->optrack_path));
}
WT_RET(__wt_config_gets(session,
cfg, "operation_tracking.enabled", &cval));
if (cval.val == 0) {
if (F_ISSET(conn, WT_CONN_OPTRACK)) {
WT_RET(__wt_conn_optrack_teardown(session, reconfig));
F_CLR(conn, WT_CONN_OPTRACK);
}
return (0);
}
if (F_ISSET(conn, WT_CONN_READONLY))
/* Operation tracking isn't supported in read-only mode */
WT_RET_MSG(session, EINVAL,
"Operation tracking is incompatible with read only "
"configuration.");
if (F_ISSET(conn, WT_CONN_OPTRACK))
/* Already enabled, nothing else to do */
return (0);
/*
* Operation tracking files will include the ID of the creating process
* in their name, so we can distinguish between log files created by
* different WiredTiger processes in the same directory. We cache the
* process id for future use.
*/
conn->optrack_pid = __wt_process_id();
/*
* Open the file in the same directory that will hold a map of
* translations between function names and function IDs. If the file
* exists, remove it.
*/
WT_RET(__wt_scr_alloc(session, 0, &buf));
WT_ERR(__wt_filename_construct(session, conn->optrack_path,
"optrack-map", conn->optrack_pid, UINT32_MAX, buf));
WT_ERR(__wt_open(session,
(const char *)buf->data, WT_FS_OPEN_FILE_TYPE_REGULAR,
WT_FS_OPEN_CREATE, &conn->optrack_map_fh));
WT_ERR(__wt_spin_init(session,
&conn->optrack_map_spinlock, "optrack map spinlock"));
WT_ERR(__wt_malloc(session, WT_OPTRACK_BUFSIZE,
&conn->dummy_session.optrack_buf));
/* Set operation tracking on */
F_SET(conn, WT_CONN_OPTRACK);
err: __wt_scr_free(session, &buf);
return (ret);
}
/*
* __wt_curjoin_open --
* Initialize a join cursor.
*
* Join cursors are read-only.
*/
int
__wt_curjoin_open(WT_SESSION_IMPL *session,
const char *uri, WT_CURSOR *owner, const char *cfg[], WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__curjoin_get_key, /* get-key */
__curjoin_get_value, /* get-value */
__wt_cursor_set_key_notsup, /* set-key */
__wt_cursor_set_value_notsup, /* set-value */
__wt_cursor_compare_notsup, /* compare */
__wt_cursor_equals_notsup, /* equals */
__curjoin_next, /* next */
__wt_cursor_notsup, /* prev */
__curjoin_reset, /* reset */
__wt_cursor_notsup, /* search */
__wt_cursor_search_near_notsup, /* search-near */
__wt_cursor_notsup, /* insert */
__wt_cursor_modify_notsup, /* modify */
__wt_cursor_notsup, /* update */
__wt_cursor_notsup, /* remove */
__wt_cursor_notsup, /* reserve */
__wt_cursor_reconfigure_notsup, /* reconfigure */
__wt_cursor_notsup, /* cache */
__wt_cursor_reopen_notsup, /* reopen */
__curjoin_close); /* close */
WT_CURSOR *cursor;
WT_CURSOR_JOIN *cjoin;
WT_DECL_ITEM(tmp);
WT_DECL_RET;
WT_TABLE *table;
size_t size;
const char *tablename, *columns;
WT_STATIC_ASSERT(offsetof(WT_CURSOR_JOIN, iface) == 0);
if (owner != NULL)
WT_RET_MSG(session, EINVAL,
"unable to initialize a join cursor with existing owner");
tablename = uri;
if (!WT_PREFIX_SKIP(tablename, "join:table:"))
return (
__wt_unexpected_object_type(session, uri, "join:table:"));
columns = strchr(tablename, '(');
if (columns == NULL)
size = strlen(tablename);
else
size = WT_PTRDIFF(columns, tablename);
WT_RET(__wt_schema_get_table(
session, tablename, size, false, 0, &table));
WT_RET(__wt_calloc_one(session, &cjoin));
cursor = (WT_CURSOR *)cjoin;
*cursor = iface;
cursor->session = (WT_SESSION *)session;
cursor->key_format = table->key_format;
cursor->value_format = table->value_format;
cjoin->table = table;
/* Handle projections. */
WT_ERR(__wt_scr_alloc(session, 0, &tmp));
if (columns != NULL) {
WT_ERR(__wt_struct_reformat(session, table,
columns, strlen(columns), NULL, false, tmp));
WT_ERR(__wt_strndup(
session, tmp->data, tmp->size, &cursor->value_format));
WT_ERR(__wt_strdup(session, columns, &cjoin->projection));
}
WT_ERR(__wt_cursor_init(cursor, uri, owner, cfg, cursorp));
if (0) {
err: WT_TRET(__curjoin_close(cursor));
*cursorp = NULL;
}
__wt_scr_free(session, &tmp);
return (ret);
}
/*
* __config_merge_scan --
* Walk a configuration string, inserting entries into the merged array.
*/
static int
__config_merge_scan(WT_SESSION_IMPL *session,
const char *key, const char *value, WT_CONFIG_MERGE *cp)
{
WT_CONFIG cparser;
WT_CONFIG_ITEM k, v;
WT_DECL_ITEM(kb);
WT_DECL_ITEM(vb);
WT_DECL_RET;
size_t len;
WT_ERR(__wt_scr_alloc(session, 0, &kb));
WT_ERR(__wt_scr_alloc(session, 0, &vb));
WT_ERR(__wt_config_init(session, &cparser, value));
while ((ret = __wt_config_next(&cparser, &k, &v)) == 0) {
if (k.type != WT_CONFIG_ITEM_STRING &&
k.type != WT_CONFIG_ITEM_ID)
WT_ERR_MSG(session, EINVAL,
"Invalid configuration key found: '%s'\n", k.str);
/* Include the quotes around string keys/values. */
if (k.type == WT_CONFIG_ITEM_STRING) {
--k.str;
k.len += 2;
}
if (v.type == WT_CONFIG_ITEM_STRING) {
--v.str;
v.len += 2;
}
/*
* !!!
* We're using a JSON quote character to separate the names we
* create for nested structures. That's not completely safe as
* it's possible to quote characters in JSON such that a quote
* character appears as a literal character in a key name. In
* a few cases, applications can create their own key namespace
* (for example, shared library extension names), and therefore
* it's possible for an application to confuse us. Error if we
* we ever see a key with a magic character.
*/
for (len = 0; len < k.len; ++len)
if (k.str[len] == SEPC)
WT_ERR_MSG(session, EINVAL,
"key %.*s contains a '%c' separator "
"character",
(int)k.len, (char *)k.str, SEPC);
/* Build the key/value strings. */
WT_ERR(__wt_buf_fmt(session,
kb, "%s%s%.*s",
key == NULL ? "" : key,
key == NULL ? "" : SEP,
(int)k.len, k.str));
WT_ERR(__wt_buf_fmt(session,
vb, "%.*s", (int)v.len, v.str));
/*
* If the value is a structure, recursively parse it.
*
* !!!
* Don't merge unless the structure has field names. WiredTiger
* stores checkpoint LSNs in the metadata file using nested
* structures without field names: "checkpoint_lsn=(1,0)", not
* "checkpoint_lsn=(file=1,offset=0)". The value type is still
* WT_CONFIG_ITEM_STRUCT, so we check for a field name in the
* value.
*/
if (v.type == WT_CONFIG_ITEM_STRUCT &&
strchr(vb->data, '=') != NULL) {
WT_ERR(__config_merge_scan(
session, kb->data, vb->data, cp));
continue;
}
/* Insert the value into the array. */
WT_ERR(__wt_realloc_def(session,
&cp->entries_allocated,
cp->entries_next + 1, &cp->entries));
WT_ERR(__wt_strndup(session,
kb->data, kb->size, &cp->entries[cp->entries_next].k));
WT_ERR(__wt_strndup(session,
vb->data, vb->size, &cp->entries[cp->entries_next].v));
cp->entries[cp->entries_next].gen = cp->entries_next;
++cp->entries_next;
}
WT_ERR_NOTFOUND_OK(ret);
err: __wt_scr_free(session, &kb);
__wt_scr_free(session, &vb);
return (ret);
}
/*
* __wt_curds_open --
* Initialize a data-source cursor.
*/
int
__wt_curds_open(
WT_SESSION_IMPL *session, const char *uri, WT_CURSOR *owner,
const char *cfg[], WT_DATA_SOURCE *dsrc, WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__wt_cursor_get_key, /* get-key */
__wt_cursor_get_value, /* get-value */
__wt_cursor_set_key, /* set-key */
__wt_cursor_set_value, /* set-value */
__curds_compare, /* compare */
__wt_cursor_equals, /* equals */
__curds_next, /* next */
__curds_prev, /* prev */
__curds_reset, /* reset */
__curds_search, /* search */
__curds_search_near, /* search-near */
__curds_insert, /* insert */
__wt_cursor_modify_notsup, /* modify */
__curds_update, /* update */
__curds_remove, /* remove */
__curds_reserve, /* reserve */
__wt_cursor_reconfigure_notsup, /* reconfigure */
__curds_close); /* close */
WT_CONFIG_ITEM cval, metadata;
WT_CURSOR *cursor, *source;
WT_CURSOR_DATA_SOURCE *data_source;
WT_DECL_RET;
char *metaconf;
WT_STATIC_ASSERT(offsetof(WT_CURSOR_DATA_SOURCE, iface) == 0);
data_source = NULL;
metaconf = NULL;
WT_RET(__wt_calloc_one(session, &data_source));
cursor = &data_source->iface;
*cursor = iface;
cursor->session = &session->iface;
/*
* XXX
* The underlying data-source may require the object's key and value
* formats. This isn't a particularly elegant way of getting that
* information to the data-source, this feels like a layering problem
* to me.
*/
WT_ERR(__wt_metadata_search(session, uri, &metaconf));
WT_ERR(__wt_config_getones(session, metaconf, "key_format", &cval));
WT_ERR(__wt_strndup(session, cval.str, cval.len, &cursor->key_format));
WT_ERR(__wt_config_getones(session, metaconf, "value_format", &cval));
WT_ERR(
__wt_strndup(session, cval.str, cval.len, &cursor->value_format));
WT_ERR(__wt_cursor_init(cursor, uri, owner, cfg, cursorp));
/* Data-source cursors may have a custom collator. */
ret = __wt_config_getones(session, metaconf, "collator", &cval);
if (ret == 0 && cval.len != 0) {
WT_CLEAR(metadata);
WT_ERR_NOTFOUND_OK(__wt_config_getones(
session, metaconf, "app_metadata", &metadata));
WT_ERR(__wt_collator_config(session, uri, &cval, &metadata,
&data_source->collator, &data_source->collator_owned));
}
WT_ERR_NOTFOUND_OK(ret);
WT_ERR(dsrc->open_cursor(dsrc,
&session->iface, uri, (WT_CONFIG_ARG *)cfg, &data_source->source));
source = data_source->source;
source->session = (WT_SESSION *)session;
memset(&source->q, 0, sizeof(source->q));
source->recno = WT_RECNO_OOB;
memset(source->raw_recno_buf, 0, sizeof(source->raw_recno_buf));
memset(&source->key, 0, sizeof(source->key));
memset(&source->value, 0, sizeof(source->value));
source->saved_err = 0;
source->flags = 0;
if (0) {
err: WT_TRET(__curds_close(cursor));
*cursorp = NULL;
}
__wt_free(session, metaconf);
return (ret);
}
