1 /* Copyright (c) 2009, 2010 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__(const char *file_name,
125 const struct ovsdb_schema *alternate_schema,
126 bool read_only, struct ovsdb **dbp,
127 struct ovsdb_file **filep)
129 enum ovsdb_log_open_mode open_mode;
130 long long int oldest_commit;
131 unsigned int n_transactions;
132 struct ovsdb_schema *schema = NULL;
133 struct ovsdb_error *error;
134 struct ovsdb_log *log = NULL;
136 struct ovsdb *db = NULL;
138 /* In read-only mode there is no ovsdb_file so 'filep' must be null. */
139 assert(!(read_only && filep));
141 open_mode = read_only ? OVSDB_LOG_READ_ONLY : OVSDB_LOG_READ_WRITE;
142 error = ovsdb_log_open(file_name, open_mode, -1, &log);
147 error = ovsdb_log_read(log, &json);
151 error = ovsdb_io_error(EOF, "%s: database file contains no schema",
156 if (alternate_schema) {
157 schema = ovsdb_schema_clone(alternate_schema);
159 error = ovsdb_schema_from_json(json, &schema);
162 error = ovsdb_wrap_error(error,
163 "failed to parse \"%s\" as ovsdb schema",
170 db = ovsdb_create(schema);
173 oldest_commit = LLONG_MAX;
175 while ((error = ovsdb_log_read(log, &json)) == NULL && json) {
176 struct ovsdb_txn *txn;
179 error = ovsdb_file_txn_from_json(db, json, alternate_schema != NULL,
187 if (date < oldest_commit) {
188 oldest_commit = date;
191 ovsdb_txn_commit(txn, false);
194 /* Log error but otherwise ignore it. Probably the database just got
195 * truncated due to power failure etc. and we should use its current
197 char *msg = ovsdb_error_to_string(error);
198 VLOG_WARN("%s", msg);
201 ovsdb_error_destroy(error);
205 struct ovsdb_file *file;
207 error = ovsdb_file_create(db, log, file_name, oldest_commit,
208 n_transactions, &file);
216 ovsdb_log_close(log);
228 ovsdb_schema_destroy(schema);
229 ovsdb_log_close(log);
233 static struct ovsdb_error *
234 ovsdb_file_update_row_from_json(struct ovsdb_row *row, bool converting,
235 const struct json *json)
237 struct ovsdb_table_schema *schema = row->table->schema;
238 struct ovsdb_error *error;
239 struct shash_node *node;
241 if (json->type != JSON_OBJECT) {
242 return ovsdb_syntax_error(json, NULL, "row must be JSON object");
245 SHASH_FOR_EACH (node, json_object(json)) {
246 const char *column_name = node->name;
247 const struct ovsdb_column *column;
248 struct ovsdb_datum datum;
250 column = ovsdb_table_schema_get_column(schema, column_name);
255 return ovsdb_syntax_error(json, "unknown column",
256 "No column %s in table %s.",
257 column_name, schema->name);
260 error = ovsdb_datum_from_json(&datum, &column->type, node->data, NULL);
264 ovsdb_datum_swap(&row->fields[column->index], &datum);
265 ovsdb_datum_destroy(&datum, &column->type);
271 static struct ovsdb_error *
272 ovsdb_file_txn_row_from_json(struct ovsdb_txn *txn, struct ovsdb_table *table,
274 const struct uuid *row_uuid, struct json *json)
276 const struct ovsdb_row *row = ovsdb_table_get_row(table, row_uuid);
277 if (json->type == JSON_NULL) {
279 return ovsdb_syntax_error(NULL, NULL, "transaction deletes "
280 "row "UUID_FMT" that does not exist",
281 UUID_ARGS(row_uuid));
283 ovsdb_txn_row_delete(txn, row);
286 return ovsdb_file_update_row_from_json(ovsdb_txn_row_modify(txn, row),
289 struct ovsdb_error *error;
290 struct ovsdb_row *new;
292 new = ovsdb_row_create(table);
293 *ovsdb_row_get_uuid_rw(new) = *row_uuid;
294 error = ovsdb_file_update_row_from_json(new, converting, json);
296 ovsdb_row_destroy(new);
299 ovsdb_txn_row_insert(txn, new);
305 static struct ovsdb_error *
306 ovsdb_file_txn_table_from_json(struct ovsdb_txn *txn,
307 struct ovsdb_table *table,
308 bool converting, struct json *json)
310 struct shash_node *node;
312 if (json->type != JSON_OBJECT) {
313 return ovsdb_syntax_error(json, NULL, "object expected");
316 SHASH_FOR_EACH (node, json->u.object) {
317 const char *uuid_string = node->name;
318 struct json *txn_row_json = node->data;
319 struct ovsdb_error *error;
320 struct uuid row_uuid;
322 if (!uuid_from_string(&row_uuid, uuid_string)) {
323 return ovsdb_syntax_error(json, NULL, "\"%s\" is not a valid UUID",
327 error = ovsdb_file_txn_row_from_json(txn, table, converting,
328 &row_uuid, txn_row_json);
337 /* Converts 'json' to an ovsdb_txn for 'db', storing the new transaction in
338 * '*txnp'. Returns NULL if successful, otherwise an error.
340 * If 'converting' is true, then unknown table and column names are ignored
341 * (which can ease upgrading and downgrading schemas); otherwise, they are
344 * If successful, the date associated with the transaction, as the number of
345 * milliseconds since the epoch, is stored in '*date'. If the transaction does
346 * not include a date, LLONG_MAX is stored. */
347 static struct ovsdb_error *
348 ovsdb_file_txn_from_json(struct ovsdb *db, const struct json *json,
349 bool converting, long long int *date,
350 struct ovsdb_txn **txnp)
352 struct ovsdb_error *error;
353 struct shash_node *node;
354 struct ovsdb_txn *txn;
359 if (json->type != JSON_OBJECT) {
360 return ovsdb_syntax_error(json, NULL, "object expected");
363 txn = ovsdb_txn_create(db);
364 SHASH_FOR_EACH (node, json->u.object) {
365 const char *table_name = node->name;
366 struct json *node_json = node->data;
367 struct ovsdb_table *table;
369 table = shash_find_data(&db->tables, table_name);
371 if (!strcmp(table_name, "_date")
372 && node_json->type == JSON_INTEGER) {
374 *date = json_integer(node_json);
377 } else if (!strcmp(table_name, "_comment") || converting) {
381 error = ovsdb_syntax_error(json, "unknown table",
382 "No table named %s.", table_name);
386 error = ovsdb_file_txn_table_from_json(txn, table, converting,
396 ovsdb_txn_abort(txn);
400 static struct ovsdb_error *
401 ovsdb_file_save_copy__(const char *file_name, int locking,
402 const char *comment, const struct ovsdb *db,
403 struct ovsdb_log **logp)
405 const struct shash_node *node;
406 struct ovsdb_file_txn ftxn;
407 struct ovsdb_error *error;
408 struct ovsdb_log *log;
411 error = ovsdb_log_open(file_name, OVSDB_LOG_CREATE, locking, &log);
417 json = ovsdb_schema_to_json(db->schema);
418 error = ovsdb_log_write(log, json);
425 ovsdb_file_txn_init(&ftxn);
426 SHASH_FOR_EACH (node, &db->tables) {
427 const struct ovsdb_table *table = node->data;
428 const struct ovsdb_row *row;
430 HMAP_FOR_EACH (row, hmap_node, &table->rows) {
431 ovsdb_file_txn_add_row(&ftxn, NULL, row, NULL);
434 error = ovsdb_file_txn_commit(ftxn.json, comment, true, log);
445 ovsdb_log_close(log);
452 /* Saves a snapshot of 'db''s current contents as 'file_name'. If 'comment' is
453 * nonnull, then it is added along with the data contents and can be viewed
454 * with "ovsdb-tool show-log".
456 * 'locking' is passed along to ovsdb_log_open() untouched. */
458 ovsdb_file_save_copy(const char *file_name, int locking,
459 const char *comment, const struct ovsdb *db)
461 return ovsdb_file_save_copy__(file_name, locking, comment, db, NULL);
464 /* Replica implementation. */
467 struct ovsdb_replica replica;
469 struct ovsdb_log *log;
471 long long int oldest_commit;
472 long long int next_compact;
473 unsigned int n_transactions;
476 static const struct ovsdb_replica_class ovsdb_file_class;
478 static struct ovsdb_error *
479 ovsdb_file_create(struct ovsdb *db, struct ovsdb_log *log,
480 const char *file_name,
481 long long int oldest_commit,
482 unsigned int n_transactions,
483 struct ovsdb_file **filep)
485 long long int now = time_msec();
486 struct ovsdb_file *file;
489 /* Use the absolute name of the file because ovsdb-server opens its
490 * database before daemonize() chdirs to "/". */
491 abs_name = abs_file_name(NULL, file_name);
494 return ovsdb_io_error(0, "could not determine current "
495 "working directory");
498 file = xmalloc(sizeof *file);
499 ovsdb_replica_init(&file->replica, &ovsdb_file_class);
502 file->file_name = abs_name;
503 file->oldest_commit = MIN(oldest_commit, now);
504 file->next_compact = file->oldest_commit + COMPACT_MIN_MSEC;
505 file->n_transactions = n_transactions;
506 ovsdb_add_replica(db, &file->replica);
512 static struct ovsdb_file *
513 ovsdb_file_cast(struct ovsdb_replica *replica)
515 assert(replica->class == &ovsdb_file_class);
516 return CONTAINER_OF(replica, struct ovsdb_file, replica);
520 ovsdb_file_change_cb(const struct ovsdb_row *old,
521 const struct ovsdb_row *new,
522 const unsigned long int *changed,
525 struct ovsdb_file_txn *ftxn = ftxn_;
526 ovsdb_file_txn_add_row(ftxn, old, new, changed);
530 static struct ovsdb_error *
531 ovsdb_file_commit(struct ovsdb_replica *replica,
532 const struct ovsdb_txn *txn, bool durable)
534 struct ovsdb_file *file = ovsdb_file_cast(replica);
535 struct ovsdb_file_txn ftxn;
536 struct ovsdb_error *error;
538 ovsdb_file_txn_init(&ftxn);
539 ovsdb_txn_for_each_change(txn, ovsdb_file_change_cb, &ftxn);
541 /* Nothing to commit. */
545 error = ovsdb_file_txn_commit(ftxn.json, ovsdb_txn_get_comment(txn),
550 file->n_transactions++;
552 /* If it has been at least COMPACT_MIN_MSEC millseconds since the last time
553 * we compacted (or at least COMPACT_RETRY_MSEC since the last time we
554 * tried), and if there are at least 100 transactions in the database, and
555 * if the database is at least 1 MB, then compact the database. */
556 if (time_msec() >= file->next_compact
557 && file->n_transactions >= 100
558 && ovsdb_log_get_offset(file->log) >= 10 * 1024 * 1024)
560 error = ovsdb_file_compact(file);
562 char *s = ovsdb_error_to_string(error);
563 ovsdb_error_destroy(error);
564 VLOG_WARN("%s: compacting database failed (%s), retrying in "
565 "60 seconds", file->file_name, s);
568 file->next_compact = time_msec() + COMPACT_RETRY_MSEC;
576 ovsdb_file_compact(struct ovsdb_file *file)
578 struct ovsdb_log *new_log = NULL;
579 struct lockfile *tmp_lock = NULL;
580 struct ovsdb_error *error;
581 char *tmp_name = NULL;
582 char *comment = NULL;
585 comment = xasprintf("compacting database online "
586 "(%.3f seconds old, %u transactions, %llu bytes)",
587 (time_msec() - file->oldest_commit) / 1000.0,
588 file->n_transactions,
589 (unsigned long long) ovsdb_log_get_offset(file->log));
590 VLOG_INFO("%s: %s", file->file_name, comment);
592 /* Commit the old version, so that we can be assured that we'll eventually
593 * have either the old or the new version. */
594 error = ovsdb_log_commit(file->log);
599 /* Lock temporary file. */
600 tmp_name = xasprintf("%s.tmp", file->file_name);
601 retval = lockfile_lock(tmp_name, 0, &tmp_lock);
603 error = ovsdb_io_error(retval, "could not get lock on %s", tmp_name);
607 /* Remove temporary file. (It might not exist.) */
608 if (unlink(tmp_name) < 0 && errno != ENOENT) {
609 error = ovsdb_io_error(errno, "failed to remove %s", tmp_name);
614 error = ovsdb_file_save_copy__(tmp_name, false, comment, file->db,
620 /* Replace original by temporary. */
621 if (rename(tmp_name, file->file_name)) {
622 error = ovsdb_io_error(errno, "failed to rename \"%s\" to \"%s\"",
623 tmp_name, file->file_name);
626 fsync_parent_dir(file->file_name);
630 ovsdb_log_close(file->log);
632 file->oldest_commit = time_msec();
633 file->next_compact = file->oldest_commit + COMPACT_MIN_MSEC;
634 file->n_transactions = 1;
636 ovsdb_log_close(new_log);
642 lockfile_unlock(tmp_lock);
650 ovsdb_file_destroy(struct ovsdb_replica *replica)
652 struct ovsdb_file *file = ovsdb_file_cast(replica);
654 ovsdb_log_close(file->log);
655 free(file->file_name);
659 static const struct ovsdb_replica_class ovsdb_file_class = {
665 ovsdb_file_txn_init(struct ovsdb_file_txn *ftxn)
668 ftxn->table_json = NULL;
673 ovsdb_file_txn_add_row(struct ovsdb_file_txn *ftxn,
674 const struct ovsdb_row *old,
675 const struct ovsdb_row *new,
676 const unsigned long int *changed)
681 row = json_null_create();
683 struct shash_node *node;
685 row = old ? NULL : json_object_create();
686 SHASH_FOR_EACH (node, &new->table->schema->columns) {
687 const struct ovsdb_column *column = node->data;
688 const struct ovsdb_type *type = &column->type;
689 unsigned int idx = column->index;
691 if (idx != OVSDB_COL_UUID && column->persistent
693 ? bitmap_is_set(changed, idx)
694 : !ovsdb_datum_is_default(&new->fields[idx], type)))
697 row = json_object_create();
699 json_object_put(row, column->name,
700 ovsdb_datum_to_json(&new->fields[idx], type));
706 struct ovsdb_table *table = new ? new->table : old->table;
707 char uuid[UUID_LEN + 1];
709 if (table != ftxn->table) {
710 /* Create JSON object for transaction overall. */
712 ftxn->json = json_object_create();
715 /* Create JSON object for transaction on this table. */
716 ftxn->table_json = json_object_create();
718 json_object_put(ftxn->json, table->schema->name, ftxn->table_json);
721 /* Add row to transaction for this table. */
722 snprintf(uuid, sizeof uuid,
723 UUID_FMT, UUID_ARGS(ovsdb_row_get_uuid(new ? new : old)));
724 json_object_put(ftxn->table_json, uuid, row);
728 static struct ovsdb_error *
729 ovsdb_file_txn_commit(struct json *json, const char *comment,
730 bool durable, struct ovsdb_log *log)
732 struct ovsdb_error *error;
735 json = json_object_create();
738 json_object_put_string(json, "_comment", comment);
740 json_object_put(json, "_date", json_integer_create(time_wall()));
742 error = ovsdb_log_write(log, json);
745 return ovsdb_wrap_error(error, "writing transaction failed");
749 error = ovsdb_log_commit(log);
751 return ovsdb_wrap_error(error, "committing transaction failed");