*/
#include "includes.h"
-#include "lib/events/events.h"
+#include "lib/tevent/tevent.h"
#include "lib/tdb/include/tdb.h"
#include "system/network.h"
#include "system/filesys.h"
+#include "system/dir.h"
+#include "system/time.h"
#include "../include/ctdb_private.h"
#include "db_wrap.h"
#include "lib/util/dlinklist.h"
+#include <ctype.h>
+
+#define PERSISTENT_HEALTH_TDB "persistent_health.tdb"
/*
this is the dummy null procedure that all databases support
struct lock_fetch_state *state = talloc_get_type(p, struct lock_fetch_state);
if (!state->ignore_generation &&
state->generation != state->ctdb->vnn_map->generation) {
- DEBUG(0,("Discarding previous generation lockwait packet\n"));
+ DEBUG(DEBUG_NOTICE,("Discarding previous generation lockwait packet\n"));
talloc_free(state->hdr);
return;
}
state->recv_pkt(state->recv_context, state->hdr);
- DEBUG(2,(__location__ " PACKET REQUEUED\n"));
+ DEBUG(DEBUG_INFO,(__location__ " PACKET REQUEUED\n"));
}
/* now the contended path */
h = ctdb_lockwait(ctdb_db, key, lock_fetch_callback, state);
if (h == NULL) {
- tdb_chainunlock(tdb, key);
return -1;
}
if (ret == 0) {
ret = ctdb_ltdb_fetch(ctdb_db, key, header, hdr, data);
if (ret != 0) {
- ctdb_ltdb_unlock(ctdb_db, key);
+ int uret;
+ uret = ctdb_ltdb_unlock(ctdb_db, key);
+ if (uret != 0) {
+ DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", uret));
+ }
}
}
return ret;
struct tdb_context *tdb = ctdb_db->ltdb->tdb;
int count = tdb_traverse_read(tdb, NULL, NULL);
if (count != 0) {
- DEBUG(0,(__location__ " tdb '%s' not empty on attach! aborting\n",
+ DEBUG(DEBUG_ALERT,(__location__ " tdb '%s' not empty on attach! aborting\n",
ctdb_db->db_path));
ctdb_fatal(ctdb_db->ctdb, "database not empty on attach");
}
}
+int ctdb_load_persistent_health(struct ctdb_context *ctdb,
+ struct ctdb_db_context *ctdb_db)
+{
+ struct tdb_context *tdb = ctdb->db_persistent_health->tdb;
+ char *old;
+ char *reason = NULL;
+ TDB_DATA key;
+ TDB_DATA val;
+
+ key.dptr = discard_const_p(uint8_t, ctdb_db->db_name);
+ key.dsize = strlen(ctdb_db->db_name);
+
+ old = ctdb_db->unhealthy_reason;
+ ctdb_db->unhealthy_reason = NULL;
+
+ val = tdb_fetch(tdb, key);
+ if (val.dsize > 0) {
+ reason = talloc_strndup(ctdb_db,
+ (const char *)val.dptr,
+ val.dsize);
+ if (reason == NULL) {
+ DEBUG(DEBUG_ALERT,(__location__ " talloc_strndup(%d) failed\n",
+ (int)val.dsize));
+ ctdb_db->unhealthy_reason = old;
+ free(val.dptr);
+ return -1;
+ }
+ }
+
+ if (val.dptr) {
+ free(val.dptr);
+ }
+
+ talloc_free(old);
+ ctdb_db->unhealthy_reason = reason;
+ return 0;
+}
+
+int ctdb_update_persistent_health(struct ctdb_context *ctdb,
+ struct ctdb_db_context *ctdb_db,
+ const char *given_reason,/* NULL means healthy */
+ int num_healthy_nodes)
+{
+ struct tdb_context *tdb = ctdb->db_persistent_health->tdb;
+ int ret;
+ TDB_DATA key;
+ TDB_DATA val;
+ char *new_reason = NULL;
+ char *old_reason = NULL;
+
+ ret = tdb_transaction_start(tdb);
+ if (ret != 0) {
+ DEBUG(DEBUG_ALERT,(__location__ " tdb_transaction_start('%s') failed: %d - %s\n",
+ tdb_name(tdb), ret, tdb_errorstr(tdb)));
+ return -1;
+ }
+
+ ret = ctdb_load_persistent_health(ctdb, ctdb_db);
+ if (ret != 0) {
+ DEBUG(DEBUG_ALERT,(__location__ " ctdb_load_persistent_health('%s') failed: %d\n",
+ ctdb_db->db_name, ret));
+ return -1;
+ }
+ old_reason = ctdb_db->unhealthy_reason;
+
+ key.dptr = discard_const_p(uint8_t, ctdb_db->db_name);
+ key.dsize = strlen(ctdb_db->db_name);
+
+ if (given_reason) {
+ new_reason = talloc_strdup(ctdb_db, given_reason);
+ if (new_reason == NULL) {
+ DEBUG(DEBUG_ALERT,(__location__ " talloc_strdup(%s) failed\n",
+ given_reason));
+ return -1;
+ }
+ } else if (old_reason && num_healthy_nodes == 0) {
+ /*
+ * If the reason indicates ok, but there where no healthy nodes
+ * available, that it means, we have not recovered valid content
+ * of the db. So if there's an old reason, prefix it with
+ * "NO-HEALTHY-NODES - "
+ */
+ const char *prefix;
+
+#define _TMP_PREFIX "NO-HEALTHY-NODES - "
+ ret = strncmp(_TMP_PREFIX, old_reason, strlen(_TMP_PREFIX));
+ if (ret != 0) {
+ prefix = _TMP_PREFIX;
+ } else {
+ prefix = "";
+ }
+ new_reason = talloc_asprintf(ctdb_db, "%s%s",
+ prefix, old_reason);
+ if (new_reason == NULL) {
+ DEBUG(DEBUG_ALERT,(__location__ " talloc_asprintf(%s%s) failed\n",
+ prefix, old_reason));
+ return -1;
+ }
+#undef _TMP_PREFIX
+ }
+
+ if (new_reason) {
+ val.dptr = discard_const_p(uint8_t, new_reason);
+ val.dsize = strlen(new_reason);
+
+ ret = tdb_store(tdb, key, val, TDB_REPLACE);
+ if (ret != 0) {
+ tdb_transaction_cancel(tdb);
+ DEBUG(DEBUG_ALERT,(__location__ " tdb_store('%s', %s, %s) failed: %d - %s\n",
+ tdb_name(tdb), ctdb_db->db_name, new_reason,
+ ret, tdb_errorstr(tdb)));
+ talloc_free(new_reason);
+ return -1;
+ }
+ DEBUG(DEBUG_ALERT,("Updated db health for db(%s) to: %s\n",
+ ctdb_db->db_name, new_reason));
+ } else if (old_reason) {
+ ret = tdb_delete(tdb, key);
+ if (ret != 0) {
+ tdb_transaction_cancel(tdb);
+ DEBUG(DEBUG_ALERT,(__location__ " tdb_delete('%s', %s) failed: %d - %s\n",
+ tdb_name(tdb), ctdb_db->db_name,
+ ret, tdb_errorstr(tdb)));
+ talloc_free(new_reason);
+ return -1;
+ }
+ DEBUG(DEBUG_NOTICE,("Updated db health for db(%s): OK\n",
+ ctdb_db->db_name));
+ }
+
+ ret = tdb_transaction_commit(tdb);
+ if (ret != TDB_SUCCESS) {
+ DEBUG(DEBUG_ALERT,(__location__ " tdb_transaction_commit('%s') failed: %d - %s\n",
+ tdb_name(tdb), ret, tdb_errorstr(tdb)));
+ talloc_free(new_reason);
+ return -1;
+ }
+
+ talloc_free(old_reason);
+ ctdb_db->unhealthy_reason = new_reason;
+
+ return 0;
+}
+
+static int ctdb_backup_corrupted_tdb(struct ctdb_context *ctdb,
+ struct ctdb_db_context *ctdb_db)
+{
+ time_t now = time(NULL);
+ char *new_path;
+ char *new_reason;
+ int ret;
+ struct tm *tm;
+
+ tm = gmtime(&now);
+
+ /* formatted like: foo.tdb.0.corrupted.20091204160825.0Z */
+ new_path = talloc_asprintf(ctdb_db, "%s.corrupted."
+ "%04u%02u%02u%02u%02u%02u.0Z",
+ ctdb_db->db_path,
+ tm->tm_year+1900, tm->tm_mon+1,
+ tm->tm_mday, tm->tm_hour, tm->tm_min,
+ tm->tm_sec);
+ if (new_path == NULL) {
+ DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
+ return -1;
+ }
+
+ new_reason = talloc_asprintf(ctdb_db,
+ "ERROR - Backup of corrupted TDB in '%s'",
+ new_path);
+ if (new_reason == NULL) {
+ DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
+ return -1;
+ }
+ ret = ctdb_update_persistent_health(ctdb, ctdb_db, new_reason, 0);
+ talloc_free(new_reason);
+ if (ret != 0) {
+ DEBUG(DEBUG_CRIT,(__location__
+ ": ctdb_backup_corrupted_tdb(%s) not implemented yet\n",
+ ctdb_db->db_path));
+ return -1;
+ }
+
+ ret = rename(ctdb_db->db_path, new_path);
+ if (ret != 0) {
+ DEBUG(DEBUG_CRIT,(__location__
+ ": ctdb_backup_corrupted_tdb(%s) rename to %s failed: %d - %s\n",
+ ctdb_db->db_path, new_path,
+ errno, strerror(errno)));
+ talloc_free(new_path);
+ return -1;
+ }
+
+ DEBUG(DEBUG_CRIT,(__location__
+ ": ctdb_backup_corrupted_tdb(%s) renamed to %s\n",
+ ctdb_db->db_path, new_path));
+ talloc_free(new_path);
+ return 0;
+}
+
+int ctdb_recheck_persistent_health(struct ctdb_context *ctdb)
+{
+ struct ctdb_db_context *ctdb_db;
+ int ret;
+ int ok = 0;
+ int fail = 0;
+
+ for (ctdb_db = ctdb->db_list; ctdb_db; ctdb_db = ctdb_db->next) {
+ if (!ctdb_db->persistent) {
+ continue;
+ }
+
+ ret = ctdb_load_persistent_health(ctdb, ctdb_db);
+ if (ret != 0) {
+ DEBUG(DEBUG_ALERT,(__location__
+ " load persistent health for '%s' failed\n",
+ ctdb_db->db_path));
+ return -1;
+ }
+
+ if (ctdb_db->unhealthy_reason == NULL) {
+ ok++;
+ DEBUG(DEBUG_INFO,(__location__
+ " persistent db '%s' healthy\n",
+ ctdb_db->db_path));
+ continue;
+ }
+
+ fail++;
+ DEBUG(DEBUG_ALERT,(__location__
+ " persistent db '%s' unhealthy: %s\n",
+ ctdb_db->db_path,
+ ctdb_db->unhealthy_reason));
+ }
+ DEBUG((fail!=0)?DEBUG_ALERT:DEBUG_NOTICE,
+ ("ctdb_recheck_presistent_health: OK[%d] FAIL[%d]\n",
+ ok, fail));
+
+ if (fail != 0) {
+ return -1;
+ }
+
+ return 0;
+}
+
+
/*
- a client has asked to attach a new database
+ mark a database - as healthy
*/
-int32_t ctdb_control_db_attach(struct ctdb_context *ctdb, TDB_DATA indata,
- TDB_DATA *outdata)
+int32_t ctdb_control_db_set_healthy(struct ctdb_context *ctdb, TDB_DATA indata)
{
- const char *db_name = (const char *)indata.dptr;
- struct ctdb_db_context *ctdb_db, *tmp_db;
+ uint32_t db_id = *(uint32_t *)indata.dptr;
+ struct ctdb_db_context *ctdb_db;
int ret;
+ bool may_recover = false;
- /* see if we already have this name */
- for (tmp_db=ctdb->db_list;tmp_db;tmp_db=tmp_db->next) {
- if (strcmp(db_name, tmp_db->db_name) == 0) {
- /* this is not an error */
- outdata->dptr = (uint8_t *)&tmp_db->db_id;
- outdata->dsize = sizeof(tmp_db->db_id);
- return 0;
- }
+ ctdb_db = find_ctdb_db(ctdb, db_id);
+ if (!ctdb_db) {
+ DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", db_id));
+ return -1;
+ }
+
+ if (ctdb_db->unhealthy_reason) {
+ may_recover = true;
+ }
+
+ ret = ctdb_update_persistent_health(ctdb, ctdb_db, NULL, 1);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR,(__location__
+ " ctdb_update_persistent_health(%s) failed\n",
+ ctdb_db->db_name));
+ return -1;
}
+ if (may_recover && !ctdb->done_startup) {
+ DEBUG(DEBUG_ERR, (__location__ " db %s become healthy - force recovery for startup\n",
+ ctdb_db->db_name));
+ ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
+ }
+
+ return 0;
+}
+
+int32_t ctdb_control_db_get_health(struct ctdb_context *ctdb,
+ TDB_DATA indata,
+ TDB_DATA *outdata)
+{
+ uint32_t db_id = *(uint32_t *)indata.dptr;
+ struct ctdb_db_context *ctdb_db;
+ int ret;
+
+ ctdb_db = find_ctdb_db(ctdb, db_id);
+ if (!ctdb_db) {
+ DEBUG(DEBUG_ERR,(__location__ " Unknown db 0x%x\n", db_id));
+ return -1;
+ }
+
+ ret = ctdb_load_persistent_health(ctdb, ctdb_db);
+ if (ret != 0) {
+ DEBUG(DEBUG_ERR,(__location__
+ " ctdb_load_persistent_health(%s) failed\n",
+ ctdb_db->db_name));
+ return -1;
+ }
+
+ *outdata = tdb_null;
+ if (ctdb_db->unhealthy_reason) {
+ outdata->dptr = (uint8_t *)ctdb_db->unhealthy_reason;
+ outdata->dsize = strlen(ctdb_db->unhealthy_reason)+1;
+ }
+
+ return 0;
+}
+
+/*
+ attach to a database, handling both persistent and non-persistent databases
+ return 0 on success, -1 on failure
+ */
+static int ctdb_local_attach(struct ctdb_context *ctdb, const char *db_name,
+ bool persistent, const char *unhealthy_reason,
+ bool jenkinshash)
+{
+ struct ctdb_db_context *ctdb_db, *tmp_db;
+ int ret;
+ struct TDB_DATA key;
+ unsigned tdb_flags;
+ int mode = 0600;
+ int remaining_tries = 0;
+
ctdb_db = talloc_zero(ctdb, struct ctdb_db_context);
CTDB_NO_MEMORY(ctdb, ctdb_db);
+ ctdb_db->priority = 1;
ctdb_db->ctdb = ctdb;
ctdb_db->db_name = talloc_strdup(ctdb_db, db_name);
CTDB_NO_MEMORY(ctdb, ctdb_db->db_name);
- ctdb_db->db_id = ctdb_hash(&indata);
-
- outdata->dptr = (uint8_t *)&ctdb_db->db_id;
- outdata->dsize = sizeof(ctdb_db->db_id);
+ key.dsize = strlen(db_name)+1;
+ key.dptr = discard_const(db_name);
+ ctdb_db->db_id = ctdb_hash(&key);
+ ctdb_db->persistent = persistent;
/* check for hash collisions */
for (tmp_db=ctdb->db_list;tmp_db;tmp_db=tmp_db->next) {
if (tmp_db->db_id == ctdb_db->db_id) {
- DEBUG(0,("db_id 0x%x hash collision. name1='%s' name2='%s'\n",
+ DEBUG(DEBUG_CRIT,("db_id 0x%x hash collision. name1='%s' name2='%s'\n",
tmp_db->db_id, db_name, tmp_db->db_name));
talloc_free(ctdb_db);
return -1;
}
}
- if (ctdb->db_directory == NULL) {
- ctdb->db_directory = VARDIR "/ctdb";
+ if (persistent) {
+ if (unhealthy_reason) {
+ ret = ctdb_update_persistent_health(ctdb, ctdb_db,
+ unhealthy_reason, 0);
+ if (ret != 0) {
+ DEBUG(DEBUG_ALERT,(__location__ " ctdb_update_persistent_health('%s','%s') failed: %d\n",
+ ctdb_db->db_name, unhealthy_reason, ret));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+ }
+
+ if (ctdb->max_persistent_check_errors > 0) {
+ remaining_tries = 1;
+ }
+ if (ctdb->done_startup) {
+ remaining_tries = 0;
+ }
+
+ ret = ctdb_load_persistent_health(ctdb, ctdb_db);
+ if (ret != 0) {
+ DEBUG(DEBUG_ALERT,(__location__ " ctdb_load_persistent_health('%s') failed: %d\n",
+ ctdb_db->db_name, ret));
+ talloc_free(ctdb_db);
+ return -1;
+ }
}
- /* make sure the db directory exists */
- if (mkdir(ctdb->db_directory, 0700) == -1 && errno != EEXIST) {
- DEBUG(0,(__location__ " Unable to create ctdb directory '%s'\n",
- ctdb->db_directory));
+ if (ctdb_db->unhealthy_reason && remaining_tries == 0) {
+ DEBUG(DEBUG_ALERT,(__location__ "ERROR: tdb %s is marked as unhealthy: %s\n",
+ ctdb_db->db_name, ctdb_db->unhealthy_reason));
talloc_free(ctdb_db);
return -1;
}
+ if (ctdb_db->unhealthy_reason) {
+ /* this is just a warning, but we want that in the log file! */
+ DEBUG(DEBUG_ALERT,(__location__ "Warning: tdb %s is marked as unhealthy: %s\n",
+ ctdb_db->db_name, ctdb_db->unhealthy_reason));
+ }
+
/* open the database */
ctdb_db->db_path = talloc_asprintf(ctdb_db, "%s/%s.%u",
- ctdb->db_directory,
- db_name, ctdb->vnn);
+ persistent?ctdb->db_directory_persistent:ctdb->db_directory,
+ db_name, ctdb->pnn);
+
+ tdb_flags = persistent? TDB_DEFAULT : TDB_CLEAR_IF_FIRST | TDB_NOSYNC;
+ if (ctdb->valgrinding) {
+ tdb_flags |= TDB_NOMMAP;
+ }
+ tdb_flags |= TDB_DISALLOW_NESTING;
+ if (jenkinshash) {
+ tdb_flags |= TDB_INCOMPATIBLE_HASH;
+ }
+again:
ctdb_db->ltdb = tdb_wrap_open(ctdb, ctdb_db->db_path,
ctdb->tunable.database_hash_size,
- TDB_CLEAR_IF_FIRST, O_CREAT|O_RDWR, 0666);
+ tdb_flags,
+ O_CREAT|O_RDWR, mode);
if (ctdb_db->ltdb == NULL) {
- DEBUG(0,("Failed to open tdb '%s'\n", ctdb_db->db_path));
- talloc_free(ctdb_db);
- return -1;
+ struct stat st;
+ int saved_errno = errno;
+
+ if (!persistent) {
+ DEBUG(DEBUG_CRIT,("Failed to open tdb '%s': %d - %s\n",
+ ctdb_db->db_path,
+ saved_errno,
+ strerror(saved_errno)));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+ if (remaining_tries == 0) {
+ DEBUG(DEBUG_CRIT,(__location__
+ "Failed to open persistent tdb '%s': %d - %s\n",
+ ctdb_db->db_path,
+ saved_errno,
+ strerror(saved_errno)));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+ ret = stat(ctdb_db->db_path, &st);
+ if (ret != 0) {
+ DEBUG(DEBUG_CRIT,(__location__
+ "Failed to open persistent tdb '%s': %d - %s\n",
+ ctdb_db->db_path,
+ saved_errno,
+ strerror(saved_errno)));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+ ret = ctdb_backup_corrupted_tdb(ctdb, ctdb_db);
+ if (ret != 0) {
+ DEBUG(DEBUG_CRIT,(__location__
+ "Failed to open persistent tdb '%s': %d - %s\n",
+ ctdb_db->db_path,
+ saved_errno,
+ strerror(saved_errno)));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+ remaining_tries--;
+ mode = st.st_mode;
+ goto again;
}
- ctdb_check_db_empty(ctdb_db);
+ if (!persistent) {
+ ctdb_check_db_empty(ctdb_db);
+ } else {
+ ret = tdb_check(ctdb_db->ltdb->tdb, NULL, NULL);
+ if (ret != 0) {
+ int fd;
+ struct stat st;
+
+ DEBUG(DEBUG_CRIT,("tdb_check(%s) failed: %d - %s\n",
+ ctdb_db->db_path, ret,
+ tdb_errorstr(ctdb_db->ltdb->tdb)));
+ if (remaining_tries == 0) {
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+ fd = tdb_fd(ctdb_db->ltdb->tdb);
+ ret = fstat(fd, &st);
+ if (ret != 0) {
+ DEBUG(DEBUG_CRIT,(__location__
+ "Failed to fstat() persistent tdb '%s': %d - %s\n",
+ ctdb_db->db_path,
+ errno,
+ strerror(errno)));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+ /* close the TDB */
+ talloc_free(ctdb_db->ltdb);
+ ctdb_db->ltdb = NULL;
+
+ ret = ctdb_backup_corrupted_tdb(ctdb, ctdb_db);
+ if (ret != 0) {
+ DEBUG(DEBUG_CRIT,("Failed to backup corrupted tdb '%s'\n",
+ ctdb_db->db_path));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+ remaining_tries--;
+ mode = st.st_mode;
+ goto again;
+ }
+ }
DLIST_ADD(ctdb->db_list, ctdb_db);
+ /* setting this can help some high churn databases */
+ tdb_set_max_dead(ctdb_db->ltdb->tdb, ctdb->tunable.database_max_dead);
+
/*
all databases support the "null" function. we need this in
order to do forced migration of records
*/
ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_null_func, CTDB_NULL_FUNC);
if (ret != 0) {
- DEBUG(0,("Failed to setup null function for '%s'\n", ctdb_db->db_name));
+ DEBUG(DEBUG_CRIT,("Failed to setup null function for '%s'\n", ctdb_db->db_name));
talloc_free(ctdb_db);
return -1;
}
*/
ret = ctdb_daemon_set_call(ctdb, ctdb_db->db_id, ctdb_fetch_func, CTDB_FETCH_FUNC);
if (ret != 0) {
- DEBUG(0,("Failed to setup fetch function for '%s'\n", ctdb_db->db_name));
+ DEBUG(DEBUG_CRIT,("Failed to setup fetch function for '%s'\n", ctdb_db->db_name));
talloc_free(ctdb_db);
return -1;
}
+
+ ret = ctdb_vacuum_init(ctdb_db);
+ if (ret != 0) {
+ DEBUG(DEBUG_CRIT,("Failed to setup vacuuming for "
+ "database '%s'\n", ctdb_db->db_name));
+ talloc_free(ctdb_db);
+ return -1;
+ }
+
+
+ DEBUG(DEBUG_INFO,("Attached to database '%s'\n", ctdb_db->db_path));
+ /* success */
+ return 0;
+}
+
+
+struct ctdb_deferred_attach_context {
+ struct ctdb_deferred_attach_context *next, *prev;
+ struct ctdb_context *ctdb;
+ struct ctdb_req_control *c;
+};
+
+
+static int ctdb_deferred_attach_destructor(struct ctdb_deferred_attach_context *da_ctx)
+{
+ DLIST_REMOVE(da_ctx->ctdb->deferred_attach, da_ctx);
+
+ return 0;
+}
+
+static void ctdb_deferred_attach_timeout(struct event_context *ev, struct timed_event *te, struct timeval t, void *private_data)
+{
+ struct ctdb_deferred_attach_context *da_ctx = talloc_get_type(private_data, struct ctdb_deferred_attach_context);
+ struct ctdb_context *ctdb = da_ctx->ctdb;
+
+ ctdb_request_control_reply(ctdb, da_ctx->c, NULL, -1, NULL);
+ talloc_free(da_ctx);
+}
+
+static void ctdb_deferred_attach_callback(struct event_context *ev, struct timed_event *te, struct timeval t, void *private_data)
+{
+ struct ctdb_deferred_attach_context *da_ctx = talloc_get_type(private_data, struct ctdb_deferred_attach_context);
+ struct ctdb_context *ctdb = da_ctx->ctdb;
+
+ /* This talloc-steals the packet ->c */
+ ctdb_input_pkt(ctdb, (struct ctdb_req_header *)da_ctx->c);
+ talloc_free(da_ctx);
+}
+
+int ctdb_process_deferred_attach(struct ctdb_context *ctdb)
+{
+ struct ctdb_deferred_attach_context *da_ctx;
+
+ /* call it from the main event loop as soon as the current event
+ finishes.
+ */
+ while ((da_ctx = ctdb->deferred_attach) != NULL) {
+ DLIST_REMOVE(ctdb->deferred_attach, da_ctx);
+ event_add_timed(ctdb->ev, ctdb, timeval_current_ofs(1,0), ctdb_deferred_attach_callback, da_ctx);
+ }
+
+ return 0;
+}
+
+/*
+ a client has asked to attach a new database
+ */
+int32_t ctdb_control_db_attach(struct ctdb_context *ctdb, TDB_DATA indata,
+ TDB_DATA *outdata, uint64_t tdb_flags,
+ bool persistent, uint32_t client_id,
+ struct ctdb_req_control *c,
+ bool *async_reply)
+{
+ const char *db_name = (const char *)indata.dptr;
+ struct ctdb_db_context *db;
+ struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
+ struct ctdb_client *client = NULL;
+
+ /* dont allow any local clients to attach while we are in recovery mode
+ * except for the recovery daemon.
+ * allow all attach from the network since these are always from remote
+ * recovery daemons.
+ */
+ if (client_id != 0) {
+ client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
+ }
+ if (client != NULL) {
+ /* If the node is inactive it is not part of the cluster
+ and we should not allow clients to attach to any
+ databases
+ */
+ if (node->flags & NODE_FLAGS_INACTIVE) {
+ DEBUG(DEBUG_ERR,("DB Attach to database %s refused since node is inactive (disconnected or banned)\n", db_name));
+ return -1;
+ }
+
+ if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE
+ && client->pid != ctdb->recoverd_pid) {
+ struct ctdb_deferred_attach_context *da_ctx = talloc(client, struct ctdb_deferred_attach_context);
+
+ if (da_ctx == NULL) {
+ DEBUG(DEBUG_ERR,("DB Attach to database %s deferral for client with pid:%d failed due to OOM.\n", db_name, client->pid));
+ return -1;
+ }
+
+ da_ctx->ctdb = ctdb;
+ da_ctx->c = talloc_steal(da_ctx, c);
+ talloc_set_destructor(da_ctx, ctdb_deferred_attach_destructor);
+ DLIST_ADD(ctdb->deferred_attach, da_ctx);
+
+ event_add_timed(ctdb->ev, da_ctx, timeval_current_ofs(ctdb->tunable.deferred_attach_timeout, 0), ctdb_deferred_attach_timeout, da_ctx);
+
+ DEBUG(DEBUG_ERR,("DB Attach to database %s deferred for client with pid:%d since node is in recovery mode.\n", db_name, client->pid));
+ *async_reply = true;
+ return 0;
+ }
+ }
+
+ /* the client can optionally pass additional tdb flags, but we
+ only allow a subset of those on the database in ctdb. Note
+ that tdb_flags is passed in via the (otherwise unused)
+ srvid to the attach control */
+ tdb_flags &= (TDB_NOSYNC|TDB_INCOMPATIBLE_HASH);
+
+ /* see if we already have this name */
+ db = ctdb_db_handle(ctdb, db_name);
+ if (db) {
+ outdata->dptr = (uint8_t *)&db->db_id;
+ outdata->dsize = sizeof(db->db_id);
+ tdb_add_flags(db->ltdb->tdb, tdb_flags);
+ return 0;
+ }
+
+ if (ctdb_local_attach(ctdb, db_name, persistent, NULL, (tdb_flags&TDB_INCOMPATIBLE_HASH)?true:false) != 0) {
+ return -1;
+ }
+
+ db = ctdb_db_handle(ctdb, db_name);
+ if (!db) {
+ DEBUG(DEBUG_ERR,("Failed to find db handle for name '%s'\n", db_name));
+ return -1;
+ }
+
+ /* remember the flags the client has specified */
+ tdb_add_flags(db->ltdb->tdb, tdb_flags);
+
+ outdata->dptr = (uint8_t *)&db->db_id;
+ outdata->dsize = sizeof(db->db_id);
+
+ /* Try to ensure it's locked in mem */
+ ctdb_lockdown_memory(ctdb);
+
/* tell all the other nodes about this database */
- ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, 0,
- CTDB_CONTROL_DB_ATTACH, 0, CTDB_CTRL_FLAG_NOREPLY,
+ ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, tdb_flags,
+ persistent?CTDB_CONTROL_DB_ATTACH_PERSISTENT:
+ CTDB_CONTROL_DB_ATTACH,
+ 0, CTDB_CTRL_FLAG_NOREPLY,
indata, NULL, NULL);
- DEBUG(1,("Attached to database '%s'\n", ctdb_db->db_path));
-
/* success */
return 0;
}
+
+/*
+ attach to all existing persistent databases
+ */
+static int ctdb_attach_persistent(struct ctdb_context *ctdb,
+ const char *unhealthy_reason)
+{
+ DIR *d;
+ struct dirent *de;
+
+ /* open the persistent db directory and scan it for files */
+ d = opendir(ctdb->db_directory_persistent);
+ if (d == NULL) {
+ return 0;
+ }
+
+ while ((de=readdir(d))) {
+ char *p, *s, *q;
+ size_t len = strlen(de->d_name);
+ uint32_t node;
+ int invalid_name = 0;
+
+ s = talloc_strdup(ctdb, de->d_name);
+ CTDB_NO_MEMORY(ctdb, s);
+
+ /* only accept names ending in .tdb */
+ p = strstr(s, ".tdb.");
+ if (len < 7 || p == NULL) {
+ talloc_free(s);
+ continue;
+ }
+
+ /* only accept names ending with .tdb. and any number of digits */
+ q = p+5;
+ while (*q != 0 && invalid_name == 0) {
+ if (!isdigit(*q++)) {
+ invalid_name = 1;
+ }
+ }
+ if (invalid_name == 1 || sscanf(p+5, "%u", &node) != 1 || node != ctdb->pnn) {
+ DEBUG(DEBUG_ERR,("Ignoring persistent database '%s'\n", de->d_name));
+ talloc_free(s);
+ continue;
+ }
+ p[4] = 0;
+
+ if (ctdb_local_attach(ctdb, s, true, unhealthy_reason, 0) != 0) {
+ DEBUG(DEBUG_ERR,("Failed to attach to persistent database '%s'\n", de->d_name));
+ closedir(d);
+ talloc_free(s);
+ return -1;
+ }
+
+ DEBUG(DEBUG_INFO,("Attached to persistent database %s\n", s));
+
+ talloc_free(s);
+ }
+ closedir(d);
+ return 0;
+}
+
+int ctdb_attach_databases(struct ctdb_context *ctdb)
+{
+ int ret;
+ char *persistent_health_path = NULL;
+ char *unhealthy_reason = NULL;
+ bool first_try = true;
+
+ if (ctdb->db_directory == NULL) {
+ ctdb->db_directory = VARDIR "/ctdb";
+ }
+ if (ctdb->db_directory_persistent == NULL) {
+ ctdb->db_directory_persistent = VARDIR "/ctdb/persistent";
+ }
+ if (ctdb->db_directory_state == NULL) {
+ ctdb->db_directory_state = VARDIR "/ctdb/state";
+ }
+
+ /* make sure the db directory exists */
+ ret = mkdir(ctdb->db_directory, 0700);
+ if (ret == -1 && errno != EEXIST) {
+ DEBUG(DEBUG_CRIT,(__location__ " Unable to create ctdb directory '%s'\n",
+ ctdb->db_directory));
+ return -1;
+ }
+
+ /* make sure the persistent db directory exists */
+ ret = mkdir(ctdb->db_directory_persistent, 0700);
+ if (ret == -1 && errno != EEXIST) {
+ DEBUG(DEBUG_CRIT,(__location__ " Unable to create ctdb persistent directory '%s'\n",
+ ctdb->db_directory_persistent));
+ return -1;
+ }
+
+ /* make sure the internal state db directory exists */
+ ret = mkdir(ctdb->db_directory_state, 0700);
+ if (ret == -1 && errno != EEXIST) {
+ DEBUG(DEBUG_CRIT,(__location__ " Unable to create ctdb state directory '%s'\n",
+ ctdb->db_directory_state));
+ return -1;
+ }
+
+ persistent_health_path = talloc_asprintf(ctdb, "%s/%s.%u",
+ ctdb->db_directory_state,
+ PERSISTENT_HEALTH_TDB,
+ ctdb->pnn);
+ if (persistent_health_path == NULL) {
+ DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
+ return -1;
+ }
+
+again:
+
+ ctdb->db_persistent_health = tdb_wrap_open(ctdb, persistent_health_path,
+ 0, TDB_DISALLOW_NESTING,
+ O_CREAT | O_RDWR, 0600);
+ if (ctdb->db_persistent_health == NULL) {
+ struct tdb_wrap *tdb;
+
+ if (!first_try) {
+ DEBUG(DEBUG_CRIT,("Failed to open tdb '%s': %d - %s\n",
+ persistent_health_path,
+ errno,
+ strerror(errno)));
+ talloc_free(persistent_health_path);
+ talloc_free(unhealthy_reason);
+ return -1;
+ }
+ first_try = false;
+
+ unhealthy_reason = talloc_asprintf(ctdb, "WARNING - '%s' %s - %s",
+ persistent_health_path,
+ "was cleared after a failure",
+ "manual verification needed");
+ if (unhealthy_reason == NULL) {
+ DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
+ talloc_free(persistent_health_path);
+ return -1;
+ }
+
+ DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - retrying after CLEAR_IF_FIRST\n",
+ persistent_health_path));
+ tdb = tdb_wrap_open(ctdb, persistent_health_path,
+ 0, TDB_CLEAR_IF_FIRST | TDB_DISALLOW_NESTING,
+ O_CREAT | O_RDWR, 0600);
+ if (tdb) {
+ DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - with CLEAR_IF_FIRST: %d - %s\n",
+ persistent_health_path,
+ errno,
+ strerror(errno)));
+ talloc_free(persistent_health_path);
+ talloc_free(unhealthy_reason);
+ return -1;
+ }
+
+ talloc_free(tdb);
+ goto again;
+ }
+ ret = tdb_check(ctdb->db_persistent_health->tdb, NULL, NULL);
+ if (ret != 0) {
+ struct tdb_wrap *tdb;
+
+ talloc_free(ctdb->db_persistent_health);
+ ctdb->db_persistent_health = NULL;
+
+ if (!first_try) {
+ DEBUG(DEBUG_CRIT,("tdb_check('%s') failed\n",
+ persistent_health_path));
+ talloc_free(persistent_health_path);
+ talloc_free(unhealthy_reason);
+ return -1;
+ }
+ first_try = false;
+
+ unhealthy_reason = talloc_asprintf(ctdb, "WARNING - '%s' %s - %s",
+ persistent_health_path,
+ "was cleared after a failure",
+ "manual verification needed");
+ if (unhealthy_reason == NULL) {
+ DEBUG(DEBUG_CRIT,(__location__ " talloc_asprintf() failed\n"));
+ talloc_free(persistent_health_path);
+ return -1;
+ }
+
+ DEBUG(DEBUG_CRIT,("tdb_check('%s') failed - retrying after CLEAR_IF_FIRST\n",
+ persistent_health_path));
+ tdb = tdb_wrap_open(ctdb, persistent_health_path,
+ 0, TDB_CLEAR_IF_FIRST | TDB_DISALLOW_NESTING,
+ O_CREAT | O_RDWR, 0600);
+ if (tdb) {
+ DEBUG(DEBUG_CRIT,("Failed to open tdb '%s' - with CLEAR_IF_FIRST: %d - %s\n",
+ persistent_health_path,
+ errno,
+ strerror(errno)));
+ talloc_free(persistent_health_path);
+ talloc_free(unhealthy_reason);
+ return -1;
+ }
+
+ talloc_free(tdb);
+ goto again;
+ }
+ talloc_free(persistent_health_path);
+
+ ret = ctdb_attach_persistent(ctdb, unhealthy_reason);
+ talloc_free(unhealthy_reason);
+ if (ret != 0) {
+ return ret;
+ }
+
+ return 0;
+}
+
/*
called when a broadcast seqnum update comes in
*/
int32_t ctdb_ltdb_update_seqnum(struct ctdb_context *ctdb, uint32_t db_id, uint32_t srcnode)
{
struct ctdb_db_context *ctdb_db;
- if (srcnode == ctdb->vnn) {
+ if (srcnode == ctdb->pnn) {
/* don't update ourselves! */
return 0;
}
ctdb_db = find_ctdb_db(ctdb, db_id);
if (!ctdb_db) {
- DEBUG(0,("Unknown db_id 0x%x in ctdb_ltdb_update_seqnum\n", db_id));
+ DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_ltdb_update_seqnum\n", db_id));
+ return -1;
+ }
+
+ if (ctdb_db->unhealthy_reason) {
+ DEBUG(DEBUG_ERR,("db(%s) unhealty in ctdb_ltdb_update_seqnum: %s\n",
+ ctdb_db->db_name, ctdb_db->unhealthy_reason));
return -1;
}
ctdb_db->seqnum = new_seqnum;
/* setup a new timer */
- ctdb_db->te =
+ ctdb_db->seqnum_update =
event_add_timed(ctdb->ev, ctdb_db,
- timeval_current_ofs(ctdb->tunable.seqnum_frequency, 0),
+ timeval_current_ofs(ctdb->tunable.seqnum_interval/1000, (ctdb->tunable.seqnum_interval%1000)*1000),
ctdb_ltdb_seqnum_check, ctdb_db);
}
struct ctdb_db_context *ctdb_db;
ctdb_db = find_ctdb_db(ctdb, db_id);
if (!ctdb_db) {
- DEBUG(0,("Unknown db_id 0x%x in ctdb_ltdb_enable_seqnum\n", db_id));
+ DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_ltdb_enable_seqnum\n", db_id));
return -1;
}
- if (ctdb_db->te == NULL) {
- ctdb_db->te =
+ if (ctdb_db->seqnum_update == NULL) {
+ ctdb_db->seqnum_update =
event_add_timed(ctdb->ev, ctdb_db,
- timeval_current_ofs(ctdb->tunable.seqnum_frequency, 0),
+ timeval_current_ofs(ctdb->tunable.seqnum_interval/1000, (ctdb->tunable.seqnum_interval%1000)*1000),
ctdb_ltdb_seqnum_check, ctdb_db);
}
return 0;
}
+int32_t ctdb_control_set_db_priority(struct ctdb_context *ctdb, TDB_DATA indata)
+{
+ struct ctdb_db_priority *db_prio = (struct ctdb_db_priority *)indata.dptr;
+ struct ctdb_db_context *ctdb_db;
+
+ ctdb_db = find_ctdb_db(ctdb, db_prio->db_id);
+ if (!ctdb_db) {
+ DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_set_db_priority\n", db_prio->db_id));
+ return 0;
+ }
+
+ if ((db_prio->priority<1) || (db_prio->priority>NUM_DB_PRIORITIES)) {
+ DEBUG(DEBUG_ERR,("Trying to set invalid priority : %u\n", db_prio->priority));
+ return 0;
+ }
+
+ ctdb_db->priority = db_prio->priority;
+ DEBUG(DEBUG_INFO,("Setting DB priority to %u for db 0x%08x\n", db_prio->priority, db_prio->db_id));
+
+ return 0;
+}
+
+
git.samba.org / sahlberg / ctdb.git / blobdiff