/*
* __ckpt_version_chk --
* Check the version major/minor numbers.
*/
static int
__ckpt_version_chk(
WT_SESSION_IMPL *session, const char *fname, const char *config)
{
WT_CONFIG_ITEM a, v;
int majorv, minorv;
WT_RET(__wt_config_getones(session, config, "version", &v));
WT_RET(__wt_config_subgets(session, &v, "major", &a));
majorv = (int)a.val;
WT_RET(__wt_config_subgets(session, &v, "minor", &a));
minorv = (int)a.val;
if (majorv < WT_BTREE_MAJOR_VERSION_MIN ||
majorv > WT_BTREE_MAJOR_VERSION_MAX ||
(majorv == WT_BTREE_MAJOR_VERSION_MIN &&
minorv < WT_BTREE_MINOR_VERSION_MIN) ||
(majorv == WT_BTREE_MAJOR_VERSION_MAX &&
minorv > WT_BTREE_MINOR_VERSION_MAX))
WT_RET_MSG(session, EACCES,
"%s is an unsupported WiredTiger source file version %d.%d"
"; this WiredTiger build only supports versions from %d.%d "
"to %d.%d",
fname,
majorv, minorv,
WT_BTREE_MAJOR_VERSION_MIN,
WT_BTREE_MINOR_VERSION_MIN,
WT_BTREE_MAJOR_VERSION_MAX,
WT_BTREE_MINOR_VERSION_MAX);
return (0);
}
/*
* __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);
}
/*
* __ckpt_last_addr --
* Return the cookie associated with the file's last checkpoint.
*/
static int
__ckpt_last_addr(
WT_SESSION_IMPL *session, const char *config, WT_ITEM *addr)
{
WT_CONFIG ckptconf;
WT_CONFIG_ITEM a, k, v;
int64_t found;
WT_RET(__wt_config_getones(session, config, "checkpoint", &v));
WT_RET(__wt_config_subinit(session, &ckptconf, &v));
for (found = 0; __wt_config_next(&ckptconf, &k, &v) == 0;) {
/* Ignore checkpoints before the ones we've already seen. */
WT_RET(__wt_config_subgets(session, &v, "order", &a));
if (found && a.val < found)
continue;
found = a.val;
/*
* Copy out the address; our caller wants the raw cookie, not
* the hex.
*/
WT_RET(__wt_config_subgets(session, &v, "addr", &a));
if (a.len != 0)
WT_RET(__wt_nhex_to_raw(session, a.str, a.len, addr));
}
return (found ? 0 : WT_NOTFOUND);
}
/*
* __snap_get_last --
* Return the cookie associated with the file's last snapshot.
*/
static int
__snap_get_last(
WT_SESSION_IMPL *session, const char *config, WT_ITEM *addr)
{
WT_CONFIG snapconf;
WT_CONFIG_ITEM a, k, v;
int64_t found;
WT_RET(__wt_config_getones(session, config, "snapshot", &v));
WT_RET(__wt_config_subinit(session, &snapconf, &v));
for (found = 0; __wt_config_next(&snapconf, &k, &v) == 0;) {
if (found) {
WT_RET(__wt_config_subgets(session, &v, "order", &a));
if (a.val < found)
continue;
}
WT_RET(__wt_config_subgets(session, &v, "addr", &a));
if (a.len == 0)
WT_RET(EINVAL);
/* Our caller wants the raw cookie, not the hex. */
WT_RET(__wt_nhex_to_raw(session, a.str, a.len, addr));
WT_RET(__wt_config_subgets(session, &v, "order", &a));
found = a.val;
}
return (found ? 0 : WT_NOTFOUND);
}
/*
* __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");
}
/*
* __config_parser_get --
* WT_CONFIG_PARSER->search method.
*/
static int
__config_parser_get(WT_CONFIG_PARSER *wt_config_parser,
const char *key, WT_CONFIG_ITEM *cval)
{
WT_CONFIG_PARSER_IMPL *config_parser;
config_parser = (WT_CONFIG_PARSER_IMPL *)wt_config_parser;
return (__wt_config_subgets(config_parser->session,
&config_parser->config_item, key, cval));
}
/*
* __snap_get_name --
* Return the cookie associated with a file's named snapshot.
*/
static int
__snap_get_name(WT_SESSION_IMPL *session,
const char *name, const char *config, WT_ITEM *addr)
{
WT_CONFIG snapconf;
WT_CONFIG_ITEM a, k, v;
WT_RET(__wt_config_getones(session, config, "snapshot", &v));
WT_RET(__wt_config_subinit(session, &snapconf, &v));
while (__wt_config_next(&snapconf, &k, &v) == 0)
if (strlen(name) == k.len && strncmp(name, k.str, k.len) == 0) {
WT_RET(__wt_config_subgets(session, &v, "addr", &a));
WT_RET(__wt_nhex_to_raw(session, a.str, a.len, addr));
return (0);
}
return (WT_NOTFOUND);
}
/*
* __conn_verbose_config --
* Set verbose configuration.
*/
static int
__conn_verbose_config(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_CONFIG_ITEM cval, sval;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
static struct {
const char *name;
uint32_t flag;
} *ft, verbtypes[] = {
{ "block", WT_VERB_block },
{ "shared_cache",WT_VERB_shared_cache },
{ "ckpt", WT_VERB_ckpt },
{ "evict", WT_VERB_evict },
{ "evictserver",WT_VERB_evictserver },
{ "fileops", WT_VERB_fileops },
{ "hazard", WT_VERB_hazard },
{ "lsm", WT_VERB_lsm },
{ "mutex", WT_VERB_mutex },
{ "read", WT_VERB_read },
{ "reconcile", WT_VERB_reconcile },
{ "salvage", WT_VERB_salvage },
{ "verify", WT_VERB_verify },
{ "write", WT_VERB_write },
{ NULL, 0 }
};
conn = S2C(session);
if ((ret = __wt_config_gets(session, cfg, "verbose", &cval)) != 0)
return (ret == WT_NOTFOUND ? 0 : ret);
for (ft = verbtypes; ft->name != NULL; ft++) {
if ((ret = __wt_config_subgets(
session, &cval, ft->name, &sval)) == 0 && sval.val != 0)
FLD_SET(conn->verbose, ft->flag);
else
FLD_CLR(conn->verbose, ft->flag);
WT_RET_NOTFOUND_OK(ret);
}
return (0);
}
/*
* __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);
}
/*
* __ckpt_last --
* Return the information associated with the file's last checkpoint.
*/
static int
__ckpt_last(WT_SESSION_IMPL *session, const char *config, WT_CKPT *ckpt)
{
WT_CONFIG ckptconf;
WT_CONFIG_ITEM a, k, v;
int64_t found;
WT_RET(__wt_config_getones(session, config, "checkpoint", &v));
WT_RET(__wt_config_subinit(session, &ckptconf, &v));
for (found = 0; __wt_config_next(&ckptconf, &k, &v) == 0;) {
/* Ignore checkpoints before the ones we've already seen. */
WT_RET(__wt_config_subgets(session, &v, "order", &a));
if (found) {
if (a.val < found)
continue;
__wt_meta_checkpoint_free(session, ckpt);
}
found = a.val;
WT_RET(__ckpt_load(session, &k, &v, ckpt));
}
return (found ? 0 : WT_NOTFOUND);
}
/*
* __ckpt_named_addr --
* Return the cookie associated with a file's named checkpoint.
*/
static int
__ckpt_named_addr(WT_SESSION_IMPL *session,
const char *checkpoint, const char *config, WT_ITEM *addr)
{
WT_CONFIG ckptconf;
WT_CONFIG_ITEM a, k, v;
WT_RET(__wt_config_getones(session, config, "checkpoint", &v));
WT_RET(__wt_config_subinit(session, &ckptconf, &v));
/*
* Take the first match: there should never be more than a single
* checkpoint of any name.
*/
while (__wt_config_next(&ckptconf, &k, &v) == 0)
if (WT_STRING_MATCH(checkpoint, k.str, k.len)) {
WT_RET(__wt_config_subgets(session, &v, "addr", &a));
if (a.len != 0)
WT_RET(__wt_nhex_to_raw(
session, a.str, a.len, addr));
return (0);
}
return (WT_NOTFOUND);
}
/*
* wiredtiger_open --
* Main library entry point: open a new connection to a WiredTiger
* database.
*/
int
wiredtiger_open(const char *home, WT_EVENT_HANDLER *event_handler,
const char *config, WT_CONNECTION **wt_connp)
{
static WT_CONNECTION stdc = {
__conn_close,
__conn_reconfigure,
__conn_get_home,
__conn_is_new,
__conn_open_session,
__conn_load_extension,
__conn_add_data_source,
__conn_add_collator,
__conn_add_compressor,
__conn_add_extractor
};
static struct {
const char *name;
uint32_t flag;
} *ft, directio_types[] = {
{ "data", WT_DIRECTIO_DATA },
{ "log", WT_DIRECTIO_LOG },
{ NULL, 0 }
};
WT_CONFIG subconfig;
WT_CONFIG_ITEM cval, skey, sval;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
WT_ITEM *cbuf, expath, exconfig;
WT_SESSION_IMPL *session;
const char *cfg[] =
{ __wt_confdfl_wiredtiger_open, config, NULL, NULL, NULL };
int exist;
*wt_connp = NULL;
session = NULL;
cbuf = NULL;
WT_CLEAR(expath);
WT_CLEAR(exconfig);
WT_RET(__wt_library_init());
WT_RET(__wt_calloc_def(NULL, 1, &conn));
conn->iface = stdc;
/*
* Immediately link the structure into the connection structure list:
* the only thing ever looked at on that list is the database name,
* and a NULL value is fine.
*/
__wt_spin_lock(NULL, &__wt_process.spinlock);
TAILQ_INSERT_TAIL(&__wt_process.connqh, conn, q);
__wt_spin_unlock(NULL, &__wt_process.spinlock);
session = conn->default_session = &conn->dummy_session;
session->iface.connection = &conn->iface;
session->name = "wiredtiger_open";
__wt_event_handler_set(session, event_handler);
/* Remaining basic initialization of the connection structure. */
WT_ERR(__wt_connection_init(conn));
/* Check the configuration strings. */
WT_ERR(__wt_config_check(
session, __wt_confchk_wiredtiger_open, config, 0));
/* Get the database home. */
WT_ERR(__conn_home(session, home, cfg));
/* Make sure no other thread of control already owns this database. */
WT_ERR(__conn_single(session, cfg));
/* Read the database-home configuration file. */
WT_ERR(__conn_config_file(session, cfg, &cbuf));
/* Read the environment variable configuration. */
WT_ERR(__conn_config_env(session, cfg));
WT_ERR(__wt_config_gets(session, cfg, "hazard_max", &cval));
conn->hazard_max = (uint32_t)cval.val;
WT_ERR(__wt_config_gets(session, cfg, "session_max", &cval));
conn->session_size = (uint32_t)cval.val + WT_NUM_INTERNAL_SESSIONS;
WT_ERR(__wt_config_gets(session, cfg, "lsm_merge", &cval));
if (cval.val)
F_SET(conn, WT_CONN_LSM_MERGE);
WT_ERR(__wt_config_gets(session, cfg, "sync", &cval));
if (cval.val)
F_SET(conn, WT_CONN_SYNC);
WT_ERR(__wt_config_gets(session, cfg, "transactional", &cval));
if (cval.val)
F_SET(conn, WT_CONN_TRANSACTIONAL);
/* Configure verbose flags. */
WT_ERR(__conn_verbose_config(session, cfg));
WT_ERR(__wt_conn_cache_pool_config(session, cfg));
WT_ERR(__wt_config_gets(session, cfg, "logging", &cval));
if (cval.val != 0)
WT_ERR(__wt_open(
session, WT_LOG_FILENAME, 1, 0, 0, &conn->log_fh));
/* Configure direct I/O and buffer alignment. */
WT_ERR(__wt_config_gets(session, cfg, "buffer_alignment", &cval));
if (cval.val == -1)
conn->buffer_alignment = WT_BUFFER_ALIGNMENT_DEFAULT;
else
conn->buffer_alignment = (size_t)cval.val;
#ifndef HAVE_POSIX_MEMALIGN
if (conn->buffer_alignment != 0)
WT_ERR_MSG(session, EINVAL,
"buffer_alignment requires posix_memalign");
#endif
/*
* Configuration: direct_io, mmap, statistics.
*/
WT_ERR(__wt_config_gets(session, cfg, "direct_io", &cval));
for (ft = directio_types; ft->name != NULL; ft++) {
ret = __wt_config_subgets(session, &cval, ft->name, &sval);
if (ret == 0) {
if (sval.val)
FLD_SET(conn->direct_io, ft->flag);
} else if (ret != WT_NOTFOUND)
goto err;
}
WT_ERR(__wt_config_gets(session, cfg, "mmap", &cval));
conn->mmap = cval.val == 0 ? 0 : 1;
WT_ERR(__wt_config_gets(session, cfg, "statistics", &cval));
conn->statistics = cval.val == 0 ? 0 : 1;
/* Load any extensions referenced in the config. */
WT_ERR(__wt_config_gets(session, cfg, "extensions", &cval));
WT_ERR(__wt_config_subinit(session, &subconfig, &cval));
while ((ret = __wt_config_next(&subconfig, &skey, &sval)) == 0) {
WT_ERR(__wt_buf_fmt(
session, &expath, "%.*s", (int)skey.len, skey.str));
if (sval.len > 0)
WT_ERR(__wt_buf_fmt(session, &exconfig,
"entry=%.*s\n", (int)sval.len, sval.str));
WT_ERR(conn->iface.load_extension(&conn->iface,
expath.data, (sval.len > 0) ? exconfig.data : NULL));
}
WT_ERR_NOTFOUND_OK(ret);
/*
* Open the connection; if that fails, the connection handle has been
* destroyed by the time the open function returns.
*/
if ((ret = __wt_connection_open(conn, cfg)) != 0) {
conn = NULL;
WT_ERR(ret);
}
/* Open the default session. */
WT_ERR(__wt_open_session(conn, 1, NULL, NULL, &conn->default_session));
session = conn->default_session;
/*
* Check on the turtle and metadata files, creating them if necessary
* (which avoids application threads racing to create the metadata file
* later).
*/
WT_ERR(__wt_meta_turtle_init(session, &exist));
if (!exist) {
/*
* We're single-threaded, but acquire the schema lock
* regardless: the lower level code checks that it is
* appropriately synchronized.
*/
WT_WITH_SCHEMA_LOCK(session,
ret = __wt_schema_create(session, WT_METADATA_URI, NULL));
WT_ERR(ret);
}
WT_ERR(__wt_metadata_open(session));
/* If there's a hot-backup file, load it. */
WT_ERR(__wt_metadata_load_backup(session));
/*
* XXX LSM initialization.
* This is structured so that it could be moved to an extension.
*/
WT_ERR(__wt_lsm_init(&conn->iface, NULL));
STATIC_ASSERT(offsetof(WT_CONNECTION_IMPL, iface) == 0);
*wt_connp = &conn->iface;
/*
* Destroying the connection on error will destroy our session handle,
* cleanup using the session handle first, then discard the connection.
*/
err: if (cbuf != NULL)
__wt_buf_free(session, cbuf);
__wt_buf_free(session, &expath);
__wt_buf_free(session, &exconfig);
if (ret != 0 && conn != NULL)
WT_TRET(__wt_connection_destroy(conn));
/* Let the server threads proceed. */
if (ret == 0)
conn->connection_initialized = 1;
return (ret);
}
/*
* __wt_conn_statistics_config --
* Set statistics configuration.
*/
int
__wt_conn_statistics_config(WT_SESSION_IMPL *session, const char *cfg[])
{
WT_CONFIG_ITEM cval, sval;
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
uint32_t flags;
int set;
conn = S2C(session);
WT_RET(__wt_config_gets(session, cfg, "statistics", &cval));
flags = 0;
set = 0;
if ((ret = __wt_config_subgets(
session, &cval, "none", &sval)) == 0 && sval.val != 0) {
flags = 0;
++set;
}
WT_RET_NOTFOUND_OK(ret);
if ((ret = __wt_config_subgets(
session, &cval, "fast", &sval)) == 0 && sval.val != 0) {
LF_SET(WT_STAT_TYPE_FAST);
++set;
}
WT_RET_NOTFOUND_OK(ret);
if ((ret = __wt_config_subgets(
session, &cval, "all", &sval)) == 0 && sval.val != 0) {
LF_SET(
WT_STAT_TYPE_ALL | WT_STAT_TYPE_CACHE_WALK |
WT_STAT_TYPE_FAST | WT_STAT_TYPE_TREE_WALK);
++set;
}
WT_RET_NOTFOUND_OK(ret);
if (set > 1)
WT_RET_MSG(session, EINVAL,
"Only one of all, fast, none configuration values should "
"be specified");
/*
* Now that we've parsed general statistics categories, process
* sub-categories.
*/
if ((ret = __wt_config_subgets(
session, &cval, "cache_walk", &sval)) == 0 && sval.val != 0)
/*
* Configuring cache walk statistics implies fast statistics.
* Keep that knowledge internal for now - it may change in the
* future.
*/
LF_SET(WT_STAT_TYPE_FAST | WT_STAT_TYPE_CACHE_WALK);
WT_RET_NOTFOUND_OK(ret);
if ((ret = __wt_config_subgets(
session, &cval, "tree_walk", &sval)) == 0 && sval.val != 0)
/*
* Configuring tree walk statistics implies fast statistics.
* Keep that knowledge internal for now - it may change in the
* future.
*/
LF_SET(WT_STAT_TYPE_FAST | WT_STAT_TYPE_TREE_WALK);
WT_RET_NOTFOUND_OK(ret);
if ((ret = __wt_config_subgets(
session, &cval, "clear", &sval)) == 0 && sval.val != 0) {
if (!LF_ISSET(WT_STAT_TYPE_ALL | WT_STAT_TYPE_CACHE_WALK |
WT_STAT_TYPE_FAST | WT_STAT_TYPE_TREE_WALK))
WT_RET_MSG(session, EINVAL,
"the value \"clear\" can only be specified if "
"statistics are enabled");
LF_SET(WT_STAT_CLEAR);
}
WT_RET_NOTFOUND_OK(ret);
/* Configuring statistics clears any existing values. */
conn->stat_flags = flags;
return (0);
}
/*
* __create_colgroup --
* Create a column group.
*/
static int
__create_colgroup(WT_SESSION_IMPL *session,
const char *name, int exclusive, const char *config)
{
WT_CONFIG_ITEM cval;
WT_DECL_RET;
WT_ITEM confbuf, fmt, namebuf;
WT_TABLE *table;
size_t tlen;
const char **cfgp, *cfg[4] =
{ WT_CONFIG_BASE(session, colgroup_meta), config, NULL, NULL };
const char *sourcecfg[] = { config, NULL, NULL };
const char *cgname, *source, *sourceconf, *tablename;
char *cgconf, *oldconf;
sourceconf = NULL;
cgconf = oldconf = NULL;
WT_CLEAR(fmt);
WT_CLEAR(confbuf);
WT_CLEAR(namebuf);
tablename = name;
if (!WT_PREFIX_SKIP(tablename, "colgroup:"))
return (EINVAL);
cgname = strchr(tablename, ':');
if (cgname != NULL) {
tlen = (size_t)(cgname - tablename);
++cgname;
} else
tlen = strlen(tablename);
if ((ret =
__wt_schema_get_table(session, tablename, tlen, 1, &table)) != 0)
WT_RET_MSG(session, (ret == WT_NOTFOUND) ? ENOENT : ret,
"Can't create '%s' for non-existent table '%.*s'",
name, (int)tlen, tablename);
/* Make sure the column group is referenced from the table. */
if (cgname != NULL && (ret =
__wt_config_subgets(session, &table->cgconf, cgname, &cval)) != 0)
WT_ERR_MSG(session, EINVAL,
"Column group '%s' not found in table '%.*s'",
cgname, (int)tlen, tablename);
/* Find the first NULL entry in the cfg stack. */
for (cfgp = &cfg[1]; *cfgp; cfgp++)
;
/* Add the source to the colgroup config before collapsing. */
if (__wt_config_getones(
session, config, "source", &cval) == 0 && cval.len != 0) {
WT_ERR(__wt_buf_fmt(
session, &namebuf, "%.*s", (int)cval.len, cval.str));
source = namebuf.data;
} else {
WT_ERR(__wt_schema_colgroup_source(
session, table, cgname, config, &namebuf));
source = namebuf.data;
WT_ERR(__wt_buf_fmt(
session, &confbuf, "source=\"%s\"", source));
*cfgp++ = confbuf.data;
}
/* Calculate the key/value formats: these go into the source config. */
WT_ERR(__wt_buf_fmt(session, &fmt, "key_format=%s", table->key_format));
if (cgname == NULL)
WT_ERR(__wt_buf_catfmt
(session, &fmt, ",value_format=%s", table->value_format));
else {
if (__wt_config_getones(session, config, "columns", &cval) != 0)
WT_ERR_MSG(session, EINVAL,
"No 'columns' configuration for '%s'", name);
WT_ERR(__wt_buf_catfmt(session, &fmt, ",value_format="));
WT_ERR(__wt_struct_reformat(session,
table, cval.str, cval.len, NULL, 1, &fmt));
}
sourcecfg[1] = fmt.data;
WT_ERR(__wt_config_merge(session, sourcecfg, NULL, &sourceconf));
WT_ERR(__wt_schema_create(session, source, sourceconf));
WT_ERR(__wt_config_collapse(session, cfg, &cgconf));
if ((ret = __wt_metadata_insert(session, name, cgconf)) != 0) {
/*
* If the entry already exists in the metadata, we're done.
* This is an error for exclusive creates but okay otherwise.
*/
if (ret == WT_DUPLICATE_KEY)
ret = exclusive ? EEXIST : 0;
goto err;
}
WT_ERR(__wt_schema_open_colgroups(session, table));
err: __wt_free(session, cgconf);
__wt_free(session, sourceconf);
__wt_free(session, oldconf);
__wt_buf_free(session, &confbuf);
__wt_buf_free(session, &fmt);
__wt_buf_free(session, &namebuf);
__wt_schema_release_table(session, table);
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_meta_ckptlist_get --
* Load all available checkpoint information for a file.
*/
int
__wt_meta_ckptlist_get(
WT_SESSION_IMPL *session, const char *fname, WT_CKPT **ckptbasep)
{
WT_CKPT *ckpt, *ckptbase;
WT_CONFIG ckptconf;
WT_CONFIG_ITEM a, k, v;
WT_DECL_RET;
WT_ITEM *buf;
size_t allocated, slot;
const char *config;
char timebuf[64];
*ckptbasep = NULL;
buf = NULL;
ckptbase = NULL;
allocated = slot = 0;
config = NULL;
/* Retrieve the metadata information for the file. */
WT_RET(__wt_metadata_read(session, fname, &config));
/* Load any existing checkpoints into the array. */
WT_ERR(__wt_scr_alloc(session, 0, &buf));
if (__wt_config_getones(session, config, "checkpoint", &v) == 0 &&
__wt_config_subinit(session, &ckptconf, &v) == 0)
for (; __wt_config_next(&ckptconf, &k, &v) == 0; ++slot) {
if (slot * sizeof(WT_CKPT) == allocated)
WT_ERR(__wt_realloc(session, &allocated,
(slot + 50) * sizeof(WT_CKPT), &ckptbase));
ckpt = &ckptbase[slot];
/*
* Copy the name, address (raw and hex), order and time
* into the slot. If there's no address, it's a fake.
*/
WT_ERR(
__wt_strndup(session, k.str, k.len, &ckpt->name));
WT_ERR(__wt_config_subgets(session, &v, "addr", &a));
WT_ERR(
__wt_buf_set(session, &ckpt->addr, a.str, a.len));
if (a.len == 0)
F_SET(ckpt, WT_CKPT_FAKE);
else
WT_ERR(__wt_nhex_to_raw(
session, a.str, a.len, &ckpt->raw));
WT_ERR(__wt_config_subgets(session, &v, "order", &a));
if (a.val == 0)
goto format;
ckpt->order = a.val;
WT_ERR(__wt_config_subgets(session, &v, "time", &a));
if (a.len == 0)
goto format;
if (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_ERR(__wt_config_subgets(session, &v, "size", &a));
ckpt->ckpt_size = (uint64_t)a.val;
}
/*
* Allocate an extra slot for a new value, plus a slot to mark the end.
*
* This isn't very clean, but there's necessary cooperation between the
* schema layer (that maintains the list of checkpoints), the btree
* layer (that knows when the root page is written, creating a new
* checkpoint), and the block manager (which actually creates the
* checkpoint). All of that cooperation is handled in the WT_CKPT
* structure referenced from the WT_BTREE structure.
*/
if ((slot + 2) * sizeof(WT_CKPT) > allocated)
WT_ERR(__wt_realloc(session, &allocated,
(slot + 2) * sizeof(WT_CKPT), &ckptbase));
/* Sort in creation-order. */
qsort(ckptbase, slot, sizeof(WT_CKPT), __ckpt_compare_order);
/* Return the array to our caller. */
*ckptbasep = ckptbase;
if (0) {
format: WT_ERR_MSG(session, WT_ERROR, "corrupted checkpoint list");
err: __wt_meta_ckptlist_free(session, ckptbase);
}
__wt_free(session, config);
__wt_scr_free(&buf);
return (ret);
}
