1 /* Copyright (c) 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
31 #include "ovsdb-error.h"
33 #include "socket-util.h"
36 #include "transaction.h"
41 VLOG_DEFINE_THIS_MODULE(ovsdb_file);
43 /* Minimum number of milliseconds between database compactions. */
44 #define COMPACT_MIN_MSEC (10 * 60 * 1000) /* 10 minutes. */
46 /* Minimum number of milliseconds between trying to compact the database if
47 * compacting fails. */
48 #define COMPACT_RETRY_MSEC (60 * 1000) /* 1 minute. */
50 /* A transaction being converted to JSON for writing to a file. */
51 struct ovsdb_file_txn {
52 struct json *json; /* JSON for the whole transaction. */
53 struct json *table_json; /* JSON for 'table''s transaction. */
54 struct ovsdb_table *table; /* Table described in 'table_json'. */
57 static void ovsdb_file_txn_init(struct ovsdb_file_txn *);
58 static void ovsdb_file_txn_add_row(struct ovsdb_file_txn *,
59 const struct ovsdb_row *old,
60 const struct ovsdb_row *new,
61 const unsigned long int *changed);
62 static struct ovsdb_error *ovsdb_file_txn_commit(struct json *,
67 static struct ovsdb_error *ovsdb_file_open__(const char *file_name,
68 const struct ovsdb_schema *,
69 bool read_only, struct ovsdb **,
70 struct ovsdb_file **);
71 static struct ovsdb_error *ovsdb_file_txn_from_json(
72 struct ovsdb *, const struct json *, bool converting,
73 long long int *date, struct ovsdb_txn **);
74 static struct ovsdb_error *ovsdb_file_create(struct ovsdb *,
76 const char *file_name,
77 long long int oldest_commit,
78 unsigned int n_transactions,
79 struct ovsdb_file **filep);
81 /* Opens database 'file_name' and stores a pointer to the new database in
82 * '*dbp'. If 'read_only' is false, then the database will be locked and
83 * changes to the database will be written to disk. If 'read_only' is true,
84 * the database will not be locked and changes to the database will persist
85 * only as long as the "struct ovsdb".
87 * If 'filep' is nonnull and 'read_only' is false, then on success sets
88 * '*filep' to an ovsdb_file that represents the open file. This ovsdb_file
89 * persists until '*dbp' is destroyed.
91 * On success, returns NULL. On failure, returns an ovsdb_error (which the
92 * caller must destroy) and sets '*dbp' and '*filep' to NULL. */
94 ovsdb_file_open(const char *file_name, bool read_only,
95 struct ovsdb **dbp, struct ovsdb_file **filep)
97 return ovsdb_file_open__(file_name, NULL, read_only, dbp, filep);
100 /* Opens database 'file_name' with an alternate schema. The specified 'schema'
101 * is used to interpret the data in 'file_name', ignoring the schema actually
102 * stored in the file. Data in the file for tables or columns that do not
103 * exist in 'schema' are ignored, but the ovsdb file format must otherwise be
104 * observed, including column constraints.
106 * This function can be useful for upgrading or downgrading databases to
107 * "almost-compatible" formats.
109 * The database will not be locked. Changes to the database will persist only
110 * as long as the "struct ovsdb".
112 * On success, stores a pointer to the new database in '*dbp' and returns a
113 * null pointer. On failure, returns an ovsdb_error (which the caller must
114 * destroy) and sets '*dbp' to NULL. */
116 ovsdb_file_open_as_schema(const char *file_name,
117 const struct ovsdb_schema *schema,
120 return ovsdb_file_open__(file_name, schema, true, dbp, NULL);
123 static struct ovsdb_error *
124 ovsdb_file_open_log(const char *file_name, enum ovsdb_log_open_mode open_mode,
125 struct ovsdb_log **logp, struct ovsdb_schema **schemap)
127 struct ovsdb_schema *schema = NULL;
128 struct ovsdb_log *log = NULL;
129 struct ovsdb_error *error;
130 struct json *json = NULL;
132 assert(logp || schemap);
134 error = ovsdb_log_open(file_name, open_mode, -1, &log);
139 error = ovsdb_log_read(log, &json);
143 error = ovsdb_io_error(EOF, "%s: database file contains no schema",
149 error = ovsdb_schema_from_json(json, &schema);
151 error = ovsdb_wrap_error(error,
152 "failed to parse \"%s\" as ovsdb schema",
162 ovsdb_log_close(log);
170 ovsdb_log_close(log);
181 static struct ovsdb_error *
182 ovsdb_file_open__(const char *file_name,
183 const struct ovsdb_schema *alternate_schema,
184 bool read_only, struct ovsdb **dbp,
185 struct ovsdb_file **filep)
187 enum ovsdb_log_open_mode open_mode;
188 long long int oldest_commit;
189 unsigned int n_transactions;
190 struct ovsdb_schema *schema = NULL;
191 struct ovsdb_error *error;
192 struct ovsdb_log *log;
194 struct ovsdb *db = NULL;
196 /* In read-only mode there is no ovsdb_file so 'filep' must be null. */
197 assert(!(read_only && filep));
199 open_mode = read_only ? OVSDB_LOG_READ_ONLY : OVSDB_LOG_READ_WRITE;
200 error = ovsdb_file_open_log(file_name, open_mode, &log,
201 alternate_schema ? NULL : &schema);
206 db = ovsdb_create(schema ? schema : ovsdb_schema_clone(alternate_schema));
208 oldest_commit = LLONG_MAX;
210 while ((error = ovsdb_log_read(log, &json)) == NULL && json) {
211 struct ovsdb_txn *txn;
214 error = ovsdb_file_txn_from_json(db, json, alternate_schema != NULL,
222 if (date < oldest_commit) {
223 oldest_commit = date;
226 ovsdb_txn_commit(txn, false);
229 /* Log error but otherwise ignore it. Probably the database just got
230 * truncated due to power failure etc. and we should use its current
232 char *msg = ovsdb_error_to_string(error);
233 VLOG_WARN("%s", msg);
236 ovsdb_error_destroy(error);
240 struct ovsdb_file *file;
242 error = ovsdb_file_create(db, log, file_name, oldest_commit,
243 n_transactions, &file);
251 ovsdb_log_close(log);
263 ovsdb_log_close(log);
267 static struct ovsdb_error *
268 ovsdb_file_update_row_from_json(struct ovsdb_row *row, bool converting,
269 const struct json *json)
271 struct ovsdb_table_schema *schema = row->table->schema;
272 struct ovsdb_error *error;
273 struct shash_node *node;
275 if (json->type != JSON_OBJECT) {
276 return ovsdb_syntax_error(json, NULL, "row must be JSON object");
279 SHASH_FOR_EACH (node, json_object(json)) {
280 const char *column_name = node->name;
281 const struct ovsdb_column *column;
282 struct ovsdb_datum datum;
284 column = ovsdb_table_schema_get_column(schema, column_name);
289 return ovsdb_syntax_error(json, "unknown column",
290 "No column %s in table %s.",
291 column_name, schema->name);
294 error = ovsdb_datum_from_json(&datum, &column->type, node->data, NULL);
298 ovsdb_datum_swap(&row->fields[column->index], &datum);
299 ovsdb_datum_destroy(&datum, &column->type);
305 static struct ovsdb_error *
306 ovsdb_file_txn_row_from_json(struct ovsdb_txn *txn, struct ovsdb_table *table,
308 const struct uuid *row_uuid, struct json *json)
310 const struct ovsdb_row *row = ovsdb_table_get_row(table, row_uuid);
311 if (json->type == JSON_NULL) {
313 return ovsdb_syntax_error(NULL, NULL, "transaction deletes "
314 "row "UUID_FMT" that does not exist",
315 UUID_ARGS(row_uuid));
317 ovsdb_txn_row_delete(txn, row);
320 return ovsdb_file_update_row_from_json(ovsdb_txn_row_modify(txn, row),
323 struct ovsdb_error *error;
324 struct ovsdb_row *new;
326 new = ovsdb_row_create(table);
327 *ovsdb_row_get_uuid_rw(new) = *row_uuid;
328 error = ovsdb_file_update_row_from_json(new, converting, json);
330 ovsdb_row_destroy(new);
333 ovsdb_txn_row_insert(txn, new);
339 static struct ovsdb_error *
340 ovsdb_file_txn_table_from_json(struct ovsdb_txn *txn,
341 struct ovsdb_table *table,
342 bool converting, struct json *json)
344 struct shash_node *node;
346 if (json->type != JSON_OBJECT) {
347 return ovsdb_syntax_error(json, NULL, "object expected");
350 SHASH_FOR_EACH (node, json->u.object) {
351 const char *uuid_string = node->name;
352 struct json *txn_row_json = node->data;
353 struct ovsdb_error *error;
354 struct uuid row_uuid;
356 if (!uuid_from_string(&row_uuid, uuid_string)) {
357 return ovsdb_syntax_error(json, NULL, "\"%s\" is not a valid UUID",
361 error = ovsdb_file_txn_row_from_json(txn, table, converting,
362 &row_uuid, txn_row_json);
371 /* Converts 'json' to an ovsdb_txn for 'db', storing the new transaction in
372 * '*txnp'. Returns NULL if successful, otherwise an error.
374 * If 'converting' is true, then unknown table and column names are ignored
375 * (which can ease upgrading and downgrading schemas); otherwise, they are
378 * If successful, the date associated with the transaction, as the number of
379 * milliseconds since the epoch, is stored in '*date'. If the transaction does
380 * not include a date, LLONG_MAX is stored. */
381 static struct ovsdb_error *
382 ovsdb_file_txn_from_json(struct ovsdb *db, const struct json *json,
383 bool converting, long long int *date,
384 struct ovsdb_txn **txnp)
386 struct ovsdb_error *error;
387 struct shash_node *node;
388 struct ovsdb_txn *txn;
393 if (json->type != JSON_OBJECT) {
394 return ovsdb_syntax_error(json, NULL, "object expected");
397 txn = ovsdb_txn_create(db);
398 SHASH_FOR_EACH (node, json->u.object) {
399 const char *table_name = node->name;
400 struct json *node_json = node->data;
401 struct ovsdb_table *table;
403 table = shash_find_data(&db->tables, table_name);
405 if (!strcmp(table_name, "_date")
406 && node_json->type == JSON_INTEGER) {
408 *date = json_integer(node_json);
411 } else if (!strcmp(table_name, "_comment") || converting) {
415 error = ovsdb_syntax_error(json, "unknown table",
416 "No table named %s.", table_name);
420 error = ovsdb_file_txn_table_from_json(txn, table, converting,
430 ovsdb_txn_abort(txn);
434 static struct ovsdb_error *
435 ovsdb_file_save_copy__(const char *file_name, int locking,
436 const char *comment, const struct ovsdb *db,
437 struct ovsdb_log **logp)
439 const struct shash_node *node;
440 struct ovsdb_file_txn ftxn;
441 struct ovsdb_error *error;
442 struct ovsdb_log *log;
445 error = ovsdb_log_open(file_name, OVSDB_LOG_CREATE, locking, &log);
451 json = ovsdb_schema_to_json(db->schema);
452 error = ovsdb_log_write(log, json);
459 ovsdb_file_txn_init(&ftxn);
460 SHASH_FOR_EACH (node, &db->tables) {
461 const struct ovsdb_table *table = node->data;
462 const struct ovsdb_row *row;
464 HMAP_FOR_EACH (row, hmap_node, &table->rows) {
465 ovsdb_file_txn_add_row(&ftxn, NULL, row, NULL);
468 error = ovsdb_file_txn_commit(ftxn.json, comment, true, log);
479 ovsdb_log_close(log);
486 /* Saves a snapshot of 'db''s current contents as 'file_name'. If 'comment' is
487 * nonnull, then it is added along with the data contents and can be viewed
488 * with "ovsdb-tool show-log".
490 * 'locking' is passed along to ovsdb_log_open() untouched. */
492 ovsdb_file_save_copy(const char *file_name, int locking,
493 const char *comment, const struct ovsdb *db)
495 return ovsdb_file_save_copy__(file_name, locking, comment, db, NULL);
498 /* Opens database 'file_name', reads its schema, and closes it. On success,
499 * stores the schema into '*schemap' and returns NULL; the caller then owns the
500 * schema. On failure, returns an ovsdb_error (which the caller must destroy)
501 * and sets '*dbp' to NULL. */
503 ovsdb_file_read_schema(const char *file_name, struct ovsdb_schema **schemap)
505 assert(schemap != NULL);
506 return ovsdb_file_open_log(file_name, OVSDB_LOG_READ_ONLY, NULL, schemap);
509 /* Replica implementation. */
512 struct ovsdb_replica replica;
514 struct ovsdb_log *log;
516 long long int oldest_commit;
517 long long int next_compact;
518 unsigned int n_transactions;
521 static const struct ovsdb_replica_class ovsdb_file_class;
523 static struct ovsdb_error *
524 ovsdb_file_create(struct ovsdb *db, struct ovsdb_log *log,
525 const char *file_name,
526 long long int oldest_commit,
527 unsigned int n_transactions,
528 struct ovsdb_file **filep)
530 long long int now = time_msec();
531 struct ovsdb_file *file;
534 /* Use the absolute name of the file because ovsdb-server opens its
535 * database before daemonize() chdirs to "/". */
536 abs_name = abs_file_name(NULL, file_name);
539 return ovsdb_io_error(0, "could not determine current "
540 "working directory");
543 file = xmalloc(sizeof *file);
544 ovsdb_replica_init(&file->replica, &ovsdb_file_class);
547 file->file_name = abs_name;
548 file->oldest_commit = MIN(oldest_commit, now);
549 file->next_compact = file->oldest_commit + COMPACT_MIN_MSEC;
550 file->n_transactions = n_transactions;
551 ovsdb_add_replica(db, &file->replica);
557 static struct ovsdb_file *
558 ovsdb_file_cast(struct ovsdb_replica *replica)
560 assert(replica->class == &ovsdb_file_class);
561 return CONTAINER_OF(replica, struct ovsdb_file, replica);
565 ovsdb_file_change_cb(const struct ovsdb_row *old,
566 const struct ovsdb_row *new,
567 const unsigned long int *changed,
570 struct ovsdb_file_txn *ftxn = ftxn_;
571 ovsdb_file_txn_add_row(ftxn, old, new, changed);
575 static struct ovsdb_error *
576 ovsdb_file_commit(struct ovsdb_replica *replica,
577 const struct ovsdb_txn *txn, bool durable)
579 struct ovsdb_file *file = ovsdb_file_cast(replica);
580 struct ovsdb_file_txn ftxn;
581 struct ovsdb_error *error;
583 ovsdb_file_txn_init(&ftxn);
584 ovsdb_txn_for_each_change(txn, ovsdb_file_change_cb, &ftxn);
586 /* Nothing to commit. */
590 error = ovsdb_file_txn_commit(ftxn.json, ovsdb_txn_get_comment(txn),
595 file->n_transactions++;
597 /* If it has been at least COMPACT_MIN_MSEC millseconds since the last time
598 * we compacted (or at least COMPACT_RETRY_MSEC since the last time we
599 * tried), and if there are at least 100 transactions in the database, and
600 * if the database is at least 1 MB, then compact the database. */
601 if (time_msec() >= file->next_compact
602 && file->n_transactions >= 100
603 && ovsdb_log_get_offset(file->log) >= 10 * 1024 * 1024)
605 error = ovsdb_file_compact(file);
607 char *s = ovsdb_error_to_string(error);
608 ovsdb_error_destroy(error);
609 VLOG_WARN("%s: compacting database failed (%s), retrying in "
610 "60 seconds", file->file_name, s);
613 file->next_compact = time_msec() + COMPACT_RETRY_MSEC;
621 ovsdb_file_compact(struct ovsdb_file *file)
623 struct ovsdb_log *new_log = NULL;
624 struct lockfile *tmp_lock = NULL;
625 struct ovsdb_error *error;
626 char *tmp_name = NULL;
627 char *comment = NULL;
630 comment = xasprintf("compacting database online "
631 "(%.3f seconds old, %u transactions, %llu bytes)",
632 (time_msec() - file->oldest_commit) / 1000.0,
633 file->n_transactions,
634 (unsigned long long) ovsdb_log_get_offset(file->log));
635 VLOG_INFO("%s: %s", file->file_name, comment);
637 /* Commit the old version, so that we can be assured that we'll eventually
638 * have either the old or the new version. */
639 error = ovsdb_log_commit(file->log);
644 /* Lock temporary file. */
645 tmp_name = xasprintf("%s.tmp", file->file_name);
646 retval = lockfile_lock(tmp_name, 0, &tmp_lock);
648 error = ovsdb_io_error(retval, "could not get lock on %s", tmp_name);
652 /* Remove temporary file. (It might not exist.) */
653 if (unlink(tmp_name) < 0 && errno != ENOENT) {
654 error = ovsdb_io_error(errno, "failed to remove %s", tmp_name);
659 error = ovsdb_file_save_copy__(tmp_name, false, comment, file->db,
665 /* Replace original by temporary. */
666 if (rename(tmp_name, file->file_name)) {
667 error = ovsdb_io_error(errno, "failed to rename \"%s\" to \"%s\"",
668 tmp_name, file->file_name);
671 fsync_parent_dir(file->file_name);
675 ovsdb_log_close(file->log);
677 file->oldest_commit = time_msec();
678 file->next_compact = file->oldest_commit + COMPACT_MIN_MSEC;
679 file->n_transactions = 1;
681 ovsdb_log_close(new_log);
687 lockfile_unlock(tmp_lock);
695 ovsdb_file_destroy(struct ovsdb_replica *replica)
697 struct ovsdb_file *file = ovsdb_file_cast(replica);
699 ovsdb_log_close(file->log);
700 free(file->file_name);
704 static const struct ovsdb_replica_class ovsdb_file_class = {
710 ovsdb_file_txn_init(struct ovsdb_file_txn *ftxn)
713 ftxn->table_json = NULL;
718 ovsdb_file_txn_add_row(struct ovsdb_file_txn *ftxn,
719 const struct ovsdb_row *old,
720 const struct ovsdb_row *new,
721 const unsigned long int *changed)
726 row = json_null_create();
728 struct shash_node *node;
730 row = old ? NULL : json_object_create();
731 SHASH_FOR_EACH (node, &new->table->schema->columns) {
732 const struct ovsdb_column *column = node->data;
733 const struct ovsdb_type *type = &column->type;
734 unsigned int idx = column->index;
736 if (idx != OVSDB_COL_UUID && column->persistent
738 ? bitmap_is_set(changed, idx)
739 : !ovsdb_datum_is_default(&new->fields[idx], type)))
742 row = json_object_create();
744 json_object_put(row, column->name,
745 ovsdb_datum_to_json(&new->fields[idx], type));
751 struct ovsdb_table *table = new ? new->table : old->table;
752 char uuid[UUID_LEN + 1];
754 if (table != ftxn->table) {
755 /* Create JSON object for transaction overall. */
757 ftxn->json = json_object_create();
760 /* Create JSON object for transaction on this table. */
761 ftxn->table_json = json_object_create();
763 json_object_put(ftxn->json, table->schema->name, ftxn->table_json);
766 /* Add row to transaction for this table. */
767 snprintf(uuid, sizeof uuid,
768 UUID_FMT, UUID_ARGS(ovsdb_row_get_uuid(new ? new : old)));
769 json_object_put(ftxn->table_json, uuid, row);
773 static struct ovsdb_error *
774 ovsdb_file_txn_commit(struct json *json, const char *comment,
775 bool durable, struct ovsdb_log *log)
777 struct ovsdb_error *error;
780 json = json_object_create();
783 json_object_put_string(json, "_comment", comment);
785 json_object_put(json, "_date", json_integer_create(time_wall()));
787 error = ovsdb_log_write(log, json);
790 return ovsdb_wrap_error(error, "writing transaction failed");
794 error = ovsdb_log_commit(log);
796 return ovsdb_wrap_error(error, "committing transaction failed");