/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License, Version 1.0 only * (the "License"). You may not use this file except in compliance * with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * db_dictionary.cc * * Copyright 2004 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ #include "db_headers.h" #include "db_entry.h" #include "db_dictionary.h" #include "db_dictlog.h" #include "db_vers.h" #include "nisdb_mt.h" #include "nisdb_rw.h" #include "ldap_parse.h" #include "ldap_map.h" #include "nis_hashitem.h" #include "ldap_util.h" #include "nis_db.h" #include #include #include #include #ifdef TDRPC #include #endif #include #include #include typedef bool_t (*db_func)(XDR *, db_table_desc *); extern db_dictionary *InUseDictionary; extern db_dictionary *FreeDictionary; /* *************** dictionary version ****************** */ #define DB_MAGIC 0x12340000 #define DB_MAJOR 0 #define DB_MINOR 10 #define DB_VERSION_0_9 (DB_MAGIC|(DB_MAJOR<<8)|9) #define DB_ORIG_VERSION DB_VERSION_0_9 #define DB_CURRENT_VERSION (DB_MAGIC|DB_MAJOR<<8|DB_MINOR) vers db_update_version; /* Note 'global' for all dbs. */ #define INMEMORY_ONLY 1 /* * Checks for valid version. For now, there are two: * DB_VERSION_ORIG was the ORIGINAL one with major = 0, minor = 9 * DB_CURRENT_VERSION is the latest one with changes in the database format * for entry objects and the change in the dictionary format. * * Our current implementation can support both versions. */ static inline bool_t db_valid_version(u_int vers) { return ((vers == DB_CURRENT_VERSION) || (vers == DB_ORIG_VERSION)); } static char * db_version_str(u_int vers) { static char vstr[128]; u_int d_major = (vers&0x0000ff00)>>8; u_int d_minor = (vers&0x000000ff); sprintf(vstr, "SunSoft, SSM, Version %d.%d", d_major, d_minor); return (vstr); } /* * Special XDR version that checks for a valid version number. * If we don't find an acceptable version, exit immediately instead * of continuing to xdr rest of dictionary, which might lead to * a core dump if the formats between versions are incompatible. * In the future, there might be a switch to determine versions * and their corresponding XDR routines for the rest of the dictionary. */ extern "C" { bool_t xdr_db_dict_version(XDR *xdrs, db_dict_version *objp) { if (xdrs->x_op == XDR_DECODE) { if (!xdr_u_int(xdrs, (u_int*) objp) || !db_valid_version(((u_int) *objp))) { syslog(LOG_ERR, "db_dictionary: invalid dictionary format! Expecting %s", db_version_str(DB_CURRENT_VERSION)); fprintf(stderr, "db_dictionary: invalid dictionary format! Expecting %s\n", db_version_str(DB_CURRENT_VERSION)); exit(1); } } else if (!xdr_u_int(xdrs, (u_int*) objp)) return (FALSE); return (TRUE); } void make_zero(vers* v) { v->zero(); } }; /* ******************* dictionary data structures *************** */ /* Delete contents of single db_table_desc pointed to by 'current.' */ static void delete_table_desc(db_table_desc *current) { if (current->table_name != NULL) delete current->table_name; if (current->scheme != NULL) delete current->scheme; if (current->database != NULL) delete current->database; delete current; } /* Create new table descriptor using given table name and table_object. */ db_status db_dictionary::create_table_desc(char *tab, table_obj* zdesc, db_table_desc** answer) { db_table_desc *newtab; if ((newtab = new db_table_desc) == NULL) { FATAL3( "db_dictionary::add_table: could not allocate space for new table", DB_MEMORY_LIMIT, DB_MEMORY_LIMIT); } newtab->database = NULL; newtab->table_name = NULL; newtab->next = NULL; if ((newtab->scheme = new db_scheme(zdesc)) == NULL) { delete_table_desc(newtab); FATAL3( "db_dictionary::add_table: could not allocate space for scheme", DB_MEMORY_LIMIT, DB_MEMORY_LIMIT); } if (newtab->scheme->numkeys() == 0) { WARNING( "db_dictionary::add_table: could not translate table_obj to scheme"); delete_table_desc(newtab); return (DB_BADOBJECT); } if ((newtab->table_name = strdup(tab)) == NULL) { delete_table_desc(newtab); FATAL3( "db_dictionary::add_table: could not allocate space for table name", DB_MEMORY_LIMIT, DB_MEMORY_LIMIT); } if (answer) *answer = newtab; return (DB_SUCCESS); } /* Delete list of db_table_desc pointed to by 'head.' */ static void delete_bucket(db_table_desc *head) { db_table_desc * nextone, *current; for (current = head; current != NULL; current = nextone) { nextone = current->next; // remember next delete_table_desc(current); } } static void delete_dictionary(db_dict_desc *dict) { db_table_desc* bucket; int i; if (dict) { if (dict->tables.tables_val) { /* delete each bucket */ for (i = 0; i < dict->tables.tables_len; i++) bucket = dict->tables.tables_val[i]; if (bucket) delete_bucket(bucket); /* delete table */ delete dict->tables.tables_val; } /* delete dictionary */ delete dict; } } /* Relocate bucket starting with this entry to new hashtable 'new_tab'. */ static void relocate_bucket(db_table_desc* bucket, db_table_desc_p *new_tab, unsigned long hashsize) { db_table_desc_p np, next_np, *hp; for (np = bucket; np != NULL; np = next_np) { next_np = np->next; hp = &new_tab[np->hashval % hashsize]; np->next = *hp; *hp = np; } } /* * Return pointer to entry with same hash value and table_name * as those supplied. Returns NULL if not found. */ static db_status enumerate_bucket(db_table_desc* bucket, db_status(*func)(db_table_desc *)) { db_table_desc_p np; db_status status; for (np = bucket; np != NULL; np = np->next) { status = (func)(np); if (status != DB_SUCCESS) return (status); } return (DB_SUCCESS); } /* * Return pointer to entry with same hash value and table_name * as those supplied. Returns NULL if not found. */ static db_table_desc_p search_bucket(db_table_desc* bucket, unsigned long hval, char *target) { db_table_desc_p np; for (np = bucket; np != NULL; np = np->next) { if (np->hashval == hval && strcmp(np->table_name, target) == 0) { break; } } return (np); } /* * Remove entry with the specified hashvalue and target table name. * Returns 'TRUE' if successful, FALSE otherwise. * If the entry being removed is at the head of the list, then * the head is updated to reflect the removal. The storage for the * entry is freed if desired. */ static bool_t remove_from_bucket(db_table_desc_p bucket, db_table_desc_p *head, unsigned long hval, char *target, bool_t free_storage) { db_table_desc_p np, dp; /* Search for it in the bucket */ for (dp = np = bucket; np != NULL; np = np->next) { if (np->hashval == hval && strcmp(np->table_name, target) == 0) { break; } else { dp = np; } } if (np == NULL) return (FALSE); // cannot delete if it is not there if (dp == np) { *head = np->next; // deleting head of bucket } else { dp->next = np->next; // deleting interior link } if (free_storage) delete_table_desc(np); return (TRUE); } /* * Add given entry to the bucket pointed to by 'bucket'. * If an entry with the same table_name is found, no addition * is done. The entry is added to the head of the bucket. */ static bool_t add_to_bucket(db_table_desc_p bucket, db_table_desc **head, db_table_desc_p td) { db_table_desc_p curr, prev; register char *target_name; unsigned long target_hval; target_name = td->table_name; target_hval = td->hashval; /* Search for it in the bucket */ for (prev = curr = bucket; curr != NULL; curr = curr->next) { if (curr->hashval == target_hval && strcmp(curr->table_name, target_name) == 0) { break; } else { prev = curr; } } if (curr != NULL) return (FALSE); /* duplicates not allowed */ curr = *head; *head = td; td->next = curr; return (TRUE); } /* Print bucket starting with this entry. */ static void print_bucket(db_table_desc *head) { db_table_desc *np; for (np = head; np != NULL; np = np->next) { printf("%s: %d\n", np->table_name, np->hashval); } } static db_status print_table(db_table_desc *tbl) { if (tbl == NULL) return (DB_BADTABLE); printf("%s: %d\n", tbl->table_name, tbl->hashval); return (DB_SUCCESS); } static int hashsizes[] = { /* hashtable sizes */ 11, 53, 113, 337, 977, 2053, 4073, 8011, 16001, 0 }; // prevents wrap around numbers from being passed #define CALLOC_LIMIT 536870911 /* Returns the next size to use for the hash table */ static unsigned int get_next_hashsize(long unsigned oldsize) { long unsigned newsize, n; if (oldsize == 0) newsize = hashsizes[0]; else { for (n = 0; newsize = hashsizes[n++]; ) if (oldsize == newsize) { newsize = hashsizes[n]; /* get next size */ break; } if (newsize == 0) newsize = oldsize * 2 + 1; /* just double */ } return (newsize); } /* * Grow the current hashtable upto the next size. * The contents of the existing hashtable is copied to the new one and * relocated according to its hashvalue relative to the new size. * Old table is deleted after the relocation. */ static void grow_dictionary(db_dict_desc_p dd) { unsigned int oldsize, i, new_size; db_table_desc_p * oldtab, *newtab; oldsize = dd->tables.tables_len; oldtab = dd->tables.tables_val; new_size = get_next_hashsize(oldsize); if (new_size > CALLOC_LIMIT) { FATAL("db_dictionary::grow: table size exceeds calloc limit", DB_MEMORY_LIMIT); } if ((newtab = (db_table_desc_p*) calloc((unsigned int) new_size, sizeof (db_table_desc_p))) == NULL) { FATAL("db_dictionary::grow: cannot allocate space", DB_MEMORY_LIMIT); } if (oldtab != NULL) { // must transfer contents of old to new for (i = 0; i < oldsize; i++) { relocate_bucket(oldtab[i], newtab, new_size); } delete oldtab; // delete old hashtable } dd->tables.tables_val = newtab; dd->tables.tables_len = new_size; } #define HASHSHIFT 3 #define HASHMASK 0x1f static u_int get_hashval(char *value) { int i, len; u_int hval = 0; len = strlen(value); for (i = 0; i < len; i++) { hval = ((hval<tables.tables_len; i++) { bucket = dd->tables.tables_val[i]; if (bucket) { status = enumerate_bucket(bucket, func); if (status != DB_SUCCESS) return (status); } } return (DB_SUCCESS); } /* * Look up target table_name in hashtable and return its db_table_desc. * Return NULL if not found. */ static db_table_desc * search_dictionary(db_dict_desc *dd, char *target) { register unsigned long hval; unsigned long bucket; if (target == NULL || dd == NULL || dd->tables.tables_len == 0) return (NULL); hval = get_hashval(target); bucket = hval % dd->tables.tables_len; db_table_desc_p fst = dd->tables.tables_val[bucket]; if (fst != NULL) return (search_bucket(fst, hval, target)); else return (NULL); } /* * Remove the entry with the target table_name from the dictionary. * If successful, return DB_SUCCESS; otherwise DB_NOTUNIQUE if target * is null; DB_NOTFOUND if entry is not found. * If successful, decrement count of number of entries in hash table. */ static db_status remove_from_dictionary(db_dict_desc *dd, char *target, bool_t remove_storage) { register unsigned long hval; unsigned long bucket; register db_table_desc *fst; if (target == NULL) return (DB_NOTUNIQUE); if (dd == NULL || dd->tables.tables_len == 0) return (DB_NOTFOUND); hval = get_hashval(target); bucket = hval % dd->tables.tables_len; fst = dd->tables.tables_val[bucket]; if (fst == NULL) return (DB_NOTFOUND); if (remove_from_bucket(fst, &dd->tables.tables_val[bucket], hval, target, remove_storage)) { --(dd->count); return (DB_SUCCESS); } else return (DB_NOTFOUND); } /* * Add a new db_table_desc to the dictionary. * Return DB_NOTUNIQUE, if entry with identical table_name * already exists. If entry is added, return DB_SUCCESS. * Increment count of number of entries in index table and grow table * if number of entries equals size of table. * * Inputs: db_dict_desc_p dd pointer to dictionary to add to. * db_table_desc *td pointer to table entry to be added. The * db_table_desc.next field will be altered * without regard to it's current setting. * This means that if next points to a list of * table entries, they may be either linked into * the dictionary unexpectly or cut off (leaked). */ static db_status add_to_dictionary(db_dict_desc_p dd, db_table_desc *td) { register unsigned long hval; char *target; if (dd == NULL) return (DB_NOTFOUND); if (td == NULL) return (DB_NOTFOUND); target = td->table_name; if (target == NULL) return (DB_NOTUNIQUE); hval = get_hashval(target); if (dd->tables.tables_val == NULL) grow_dictionary(dd); db_table_desc_p fst; unsigned long bucket; bucket = hval % dd->tables.tables_len; fst = dd->tables.tables_val[bucket]; td->hashval = hval; if (fst == NULL) { /* Empty bucket */ dd->tables.tables_val[bucket] = td; } else if (!add_to_bucket(fst, &dd->tables.tables_val[bucket], td)) { return (DB_NOTUNIQUE); } /* increase hash table size if number of entries equals table size */ if (++(dd->count) > dd->tables.tables_len) grow_dictionary(dd); return (DB_SUCCESS); } /* ******************* pickling routines for dictionary ******************* */ /* Does the actual writing to/from file specific for db_dict_desc structure. */ static bool_t transfer_aux(XDR* x, pptr tbl) { return (xdr_db_dict_desc_p(x, (db_dict_desc_p *) tbl)); } class pickle_dict_desc: public pickle_file { public: pickle_dict_desc(char *f, pickle_mode m) : pickle_file(f, m) {} /* Transfers db_dict_desc structure pointed to by dp to/from file. */ int transfer(db_dict_desc_p * dp) { return (pickle_file::transfer((pptr) dp, &transfer_aux)); } }; /* ************************ dictionary methods *************************** */ db_dictionary::db_dictionary() { dictionary = NULL; initialized = FALSE; filename = NULL; tmpfilename = NULL; logfilename = NULL; logfile = NULL; logfile_opened = FALSE; changed = FALSE; INITRW(dict); READLOCKOK(dict); } /* * This routine clones an entire hash bucket chain. If you clone a * data dictionary entry with the ->next pointer set, you will get a * clone of that entry, as well as the entire linked list. This can cause * pain if you then pass the cloned bucket to routines such as * add_to_dictionary(), which do not expect nor handle dictionary hash * entries with the ->next pointer set. You might either get duplicate * entires or lose entries. If you wish to clone the entire bucket chain * and add it to a new dictionary, loop through the db_table_desc->next list * and call add_to_dictionary() for each item. */ int db_dictionary::db_clone_bucket(db_table_desc *bucket, db_table_desc **clone) { u_long size; XDR xdrs; char *bufin = NULL; READLOCK(this, DB_LOCK_ERROR, "r db_dictionary::db_clone_bucket"); db_func use_this = xdr_db_table_desc; size = xdr_sizeof((xdrproc_t) use_this, (void *) bucket); bufin = (char *) calloc(1, (size_t) size * sizeof (char)); if (!bufin) { READUNLOCK(this, DB_MEMORY_LIMIT, "db_dictionary::insert_modified_table: out of memory"); FATAL3("db_dictionary::insert_modified_table: out of memory", DB_MEMORY_LIMIT, 0); } xdrmem_create(&xdrs, bufin, (size_t) size, XDR_ENCODE); if (!xdr_db_table_desc(&xdrs, bucket)) { free(bufin); xdr_destroy(&xdrs); READUNLOCK(this, DB_MEMORY_LIMIT, "db_dictionary::insert_modified_table: xdr encode failed"); FATAL3( "db_dictionary::insert_modified_table: xdr encode failed.", DB_MEMORY_LIMIT, 0); } *clone = (db_table_desc *) calloc(1, (size_t) size * sizeof (char)); if (!*clone) { xdr_destroy(&xdrs); free(bufin); READUNLOCK(this, DB_MEMORY_LIMIT, "db_dictionary::insert_modified_table: out of memory"); FATAL3("db_dictionary::insert_modified_table: out of memory", DB_MEMORY_LIMIT, 0); } xdrmem_create(&xdrs, bufin, (size_t) size, XDR_DECODE); if (!xdr_db_table_desc(&xdrs, *clone)) { free(bufin); free(*clone); xdr_destroy(&xdrs); READUNLOCK(this, DB_MEMORY_LIMIT, "db_dictionary::insert_modified_table: xdr encode failed"); FATAL3( "db_dictionary::insert_modified_table: xdr encode failed.", DB_MEMORY_LIMIT, 0); } free(bufin); xdr_destroy(&xdrs); READUNLOCK(this, DB_LOCK_ERROR, "ru db_dictionary::db_clone_bucket"); return (1); } int db_dictionary::change_table_name(db_table_desc *clone, char *tok, char *repl) { char *newname; char *loc_end, *loc_beg; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::change_table_name"); while (clone) { /* * Special case for a tok="". This is used for the * nisrestore(1M), when restoring a replica in another * domain. This routine is used to change the datafile * names in the data.dict (see bugid #4031273). This will not * effect massage_dict(), since it never generates an empty * string for tok. */ if (strlen(tok) == 0) { strcat(clone->table_name, repl); clone = clone->next; continue; } newname = (char *) calloc(1, sizeof (char) * strlen(clone->table_name) + strlen(repl) - strlen(tok) + 1); if (!newname) { WRITEUNLOCK(this, DB_MEMORY_LIMIT, "db_dictionary::change_table_name: out of memory"); FATAL3("db_dictionary::change_table_name: out of memory.", DB_MEMORY_LIMIT, 0); } if (loc_beg = strstr(clone->table_name, tok)) { loc_end = loc_beg + strlen(tok); int s = loc_beg - clone->table_name; memcpy(newname, clone->table_name, s); strcat(newname + s, repl); strcat(newname, loc_end); free(clone->table_name); clone->table_name = newname; } else { free(newname); } clone = clone->next; } WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::change_table_name"); return (1); } #ifdef curdict #undef curdict #endif /* * A function to initialize the temporary dictionary from the real * dictionary. */ bool_t db_dictionary::inittemp(char *dictname, db_dictionary& curdict) { int status; db_table_desc_p *newtab; db_shutdown(); WRITELOCK(this, FALSE, "w db_dictionary::inittemp"); if (initialized) { /* Someone else got in between db_shutdown() and lock() */ WRITEUNLOCK(this, FALSE, "wu db_dictionary::inittemp"); return (TRUE); } pickle_dict_desc f(dictname, PICKLE_READ); filename = strdup(dictname); if (filename == NULL) { WRITEUNLOCK(this, FALSE, "db_dictionary::inittemp: could not allocate space"); FATAL3("db_dictionary::inittemp: could not allocate space", DB_MEMORY_LIMIT, FALSE); } int len = strlen(filename); tmpfilename = new char[len+5]; if (tmpfilename == NULL) { delete filename; WRITEUNLOCK(this, FALSE, "db_dictionary::inittemp: could not allocate space"); FATAL3("db_dictionary::inittemp: could not allocate space", DB_MEMORY_LIMIT, FALSE); } logfilename = new char[len+5]; if (logfilename == NULL) { delete filename; delete tmpfilename; WRITEUNLOCK(this, FALSE, "db_dictionary::inittemp: cannot allocate space"); FATAL3("db_dictionary::inittemp: cannot allocate space", DB_MEMORY_LIMIT, FALSE); } sprintf(tmpfilename, "%s.tmp", filename); sprintf(logfilename, "%s.log", filename); unlink(tmpfilename); /* get rid of partial checkpoints */ dictionary = NULL; if ((status = f.transfer(&dictionary)) < 0) { initialized = FALSE; } else if (status == 1) { /* no dictionary exists, create one */ dictionary = new db_dict_desc; if (dictionary == NULL) { WRITEUNLOCK(this, FALSE, "db_dictionary::inittemp: could not allocate space"); FATAL3( "db_dictionary::inittemp: could not allocate space", DB_MEMORY_LIMIT, FALSE); } dictionary->tables.tables_len = curdict.dictionary->tables.tables_len; if ((newtab = (db_table_desc_p *) calloc( (unsigned int) dictionary->tables.tables_len, sizeof (db_table_desc_p))) == NULL) { WRITEUNLOCK(this, FALSE, "db_dictionary::inittemp: cannot allocate space"); FATAL3( "db_dictionary::inittemp: cannot allocate space", DB_MEMORY_LIMIT, 0); } dictionary->tables.tables_val = newtab; dictionary->count = 0; dictionary->impl_vers = curdict.dictionary->impl_vers; initialized = TRUE; } else /* dictionary loaded successfully */ initialized = TRUE; if (initialized == TRUE) { changed = FALSE; reset_log(); } WRITEUNLOCK(this, FALSE, "wu db_dictionary::inittemp"); return (initialized); } /* * This method replaces the token string specified with the replacment * string specified. It assumes that at least one and only one instance of * the token exists. It is the responsibility of the caller to ensure that * the above assumption stays valid. */ db_status db_dictionary::massage_dict(char *newdictname, char *tok, char *repl) { int retval; u_int i, tbl_count; db_status status; db_table_desc *bucket, *np, *clone, *next_np; char tail[NIS_MAXNAMELEN]; db_dictionary *tmpptr; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::massage_dict"); if (dictionary == NULL) { WRITEUNLOCK(this, DB_INTERNAL_ERROR, "db_dictionary::massage_dict: uninitialized dictionary file"); FATAL3( "db_dictionary::massage_dict: uninitialized dictionary file.", DB_INTERNAL_ERROR, DB_INTERNAL_ERROR); } if ((tbl_count = dictionary->count) == 0) { WRITEUNLOCK(this, DB_SUCCESS, "wu db_dictionary::massage_dict"); return (DB_SUCCESS); } /* First checkpoint */ if ((status = checkpoint()) != DB_SUCCESS) { WRITEUNLOCK(this, status, "wu db_dictionary::massage_dict"); return (status); } #ifdef DEBUG enumerate_dictionary(dictionary, &print_table); #endif /* Initialize the free dictionary so that we can start populating it */ FreeDictionary->inittemp(newdictname, *this); for (i = 0; i < dictionary->tables.tables_len; i++) { bucket = dictionary->tables.tables_val[i]; if (bucket) { np = bucket; while (np != NULL) { next_np = np->next; retval = db_clone_bucket(np, &clone); if (retval != 1) { WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::massage_dict"); return (DB_INTERNAL_ERROR); } if (change_table_name(clone, tok, repl) == -1) { delete_table_desc(clone); WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::massage_dict"); return (DB_INTERNAL_ERROR); } /* * We know we don't have a log file, so we will * just add to the in-memory database and dump * all of it once we are done. */ status = add_to_dictionary (FreeDictionary->dictionary, clone); if (status != DB_SUCCESS) { delete_table_desc(clone); WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::massage_dict"); return (DB_INTERNAL_ERROR); } status = remove_from_dictionary(dictionary, np->table_name, TRUE); if (status != DB_SUCCESS) { delete_table_desc(clone); WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::massage_dict"); return (DB_INTERNAL_ERROR); } np = next_np; } } } if (FreeDictionary->dump() != DB_SUCCESS) { WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::massage_dict"); FATAL3( "db_dictionary::massage_dict: Unable to dump new dictionary.", DB_INTERNAL_ERROR, DB_INTERNAL_ERROR); } /* * Now, shutdown the inuse dictionary and update the FreeDictionary * and InUseDictionary pointers as well. Also, delete the old dictionary * file. */ unlink(filename); /* There shouldn't be a tmpfile or logfile */ db_shutdown(); tmpptr = InUseDictionary; InUseDictionary = FreeDictionary; FreeDictionary = tmpptr; WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::massage_dict"); return (DB_SUCCESS); } db_status db_dictionary::merge_dict(db_dictionary& tempdict, char *tok, char *repl) { db_status dbstat = DB_SUCCESS; db_table_desc *tbl = NULL, *clone = NULL, *next_td = NULL; int retval, i; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::merge_dict"); for (i = 0; i < tempdict.dictionary->tables.tables_len; ++i) { tbl = tempdict.dictionary->tables.tables_val[i]; if (!tbl) continue; retval = db_clone_bucket(tbl, &clone); if (retval != 1) { WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::merge_dict"); return (DB_INTERNAL_ERROR); } while (clone) { next_td = clone->next; clone->next = NULL; if ((tok) && (change_table_name(clone, tok, repl) == -1)) { delete_table_desc(clone); if (next_td) delete_table_desc(next_td); WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::merge_dict"); return (DB_INTERNAL_ERROR); } dbstat = add_to_dictionary(dictionary, clone); if (dbstat == DB_NOTUNIQUE) { /* Overide */ dbstat = remove_from_dictionary(dictionary, clone->table_name, TRUE); if (dbstat != DB_SUCCESS) { WRITEUNLOCK(this, dbstat, "wu db_dictionary::merge_dict"); return (dbstat); } dbstat = add_to_dictionary(dictionary, clone); } else { if (dbstat != DB_SUCCESS) { WRITEUNLOCK(this, dbstat, "wu db_dictionary::merge_dict"); return (dbstat); } } clone = next_td; } } /* * If we were successful in merging the dictionaries, then mark the * dictionary changed, so that it will be properly checkpointed and * dumped to disk. */ if (dbstat == DB_SUCCESS) changed = TRUE; WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::merge_dict"); return (dbstat); } int db_dictionary::copyfile(char *infile, char *outfile) { db_table_desc *tbl = NULL; db *dbase; int ret; READLOCK(this, DB_LOCK_ERROR, "r db_dictionary::copyfile"); /* * We need to hold the read-lock until the dump() is done. * However, we must avoid the lock migration (read -> write) * that would happen in find_table() if the db must be loaded. * Hence, look first look for an already loaded db. */ dbase = find_table(infile, &tbl, TRUE, TRUE, FALSE); if (dbase == NULL) { /* Release the read-lock, and try again, allowing load */ READUNLOCK(this, DB_LOCK_ERROR, "ru db_dictionary::copyfile"); dbase = find_table(infile, &tbl, TRUE, TRUE, TRUE); if (dbase == NULL) return (DB_NOTFOUND); /* * Read-lock again, and get a 'tbl' we can use since we're * still holding the lock. */ READLOCK(this, DB_LOCK_ERROR, "r db_dictionary::copyfile"); dbase = find_table(infile, &tbl, TRUE, TRUE, FALSE); if (dbase == NULL) { READUNLOCK(this, DB_NOTFOUND, "ru db_dictionary::copyfile"); return (DB_NOTFOUND); } } ret = tbl->database->dump(outfile) ? DB_SUCCESS : DB_INTERNAL_ERROR; READUNLOCK(this, ret, "ru db_dictionary::copyfile"); return (ret); } bool_t db_dictionary::extract_entries(db_dictionary& tempdict, char **fs, int fscnt) { int i, retval; db_table_desc *tbl, *clone; db_table_desc tbl_ent; db_status dbstat; READLOCK(this, FALSE, "r db_dictionary::extract_entries"); for (i = 0; i < fscnt; ++i) { tbl = find_table_desc(fs[i]); if (!tbl) { syslog(LOG_DEBUG, "extract_entries: no dictionary entry for %s", fs[i]); READUNLOCK(this, FALSE, "ru db_dictionary::extract_entries"); return (FALSE); } else { tbl_ent.table_name = tbl->table_name; tbl_ent.hashval = tbl->hashval; tbl_ent.scheme = tbl->scheme; tbl_ent.database = tbl->database; tbl_ent.next = NULL; } retval = db_clone_bucket(&tbl_ent, &clone); if (retval != 1) { syslog(LOG_DEBUG, "extract_entries: unable to clone entry for %s", fs[i]); READUNLOCK(this, FALSE, "ru db_dictionary::extract_entries"); return (FALSE); } dbstat = add_to_dictionary(tempdict.dictionary, clone); if (dbstat != DB_SUCCESS) { delete_table_desc(clone); READUNLOCK(this, FALSE, "ru db_dictionary::extract_entries"); return (FALSE); } } if (tempdict.dump() != DB_SUCCESS) { READUNLOCK(this, FALSE, "ru db_dictionary::extract_entries"); return (FALSE); } READUNLOCK(this, FALSE, "ru db_dictionary::extract_entries"); return (TRUE); } /* * Initialize dictionary from contents in 'file'. * If there is already information in this dictionary, it is removed. * Therefore, regardless of whether the load from the file succeeds, * the contents of this dictionary will be altered. Returns * whether table has been initialized successfully. */ bool_t db_dictionary::init(char *file) { int status; WRITELOCK(this, FALSE, "w db_dictionary::init"); db_shutdown(); pickle_dict_desc f(file, PICKLE_READ); filename = strdup(file); if (filename == NULL) { WRITEUNLOCK(this, FALSE, "db_dictionary::init: could not allocate space"); FATAL3("db_dictionary::init: could not allocate space", DB_MEMORY_LIMIT, FALSE); } int len = strlen(filename); tmpfilename = new char[len+5]; if (tmpfilename == NULL) { delete filename; WRITEUNLOCK(this, FALSE, "db_dictionary::init: could not allocate space"); FATAL3("db_dictionary::init: could not allocate space", DB_MEMORY_LIMIT, FALSE); } logfilename = new char[len+5]; if (logfilename == NULL) { delete filename; delete tmpfilename; WRITEUNLOCK(this, FALSE, "db_dictionary::init: cannot allocate space"); FATAL3("db_dictionary::init: cannot allocate space", DB_MEMORY_LIMIT, FALSE); } sprintf(tmpfilename, "%s.tmp", filename); sprintf(logfilename, "%s.log", filename); unlink(tmpfilename); /* get rid of partial checkpoints */ dictionary = NULL; /* load dictionary */ if ((status = f.transfer(&dictionary)) < 0) { initialized = FALSE; } else if (status == 1) { /* no dictionary exists, create one */ dictionary = new db_dict_desc; if (dictionary == NULL) { WRITEUNLOCK(this, FALSE, "db_dictionary::init: could not allocate space"); FATAL3("db_dictionary::init: could not allocate space", DB_MEMORY_LIMIT, FALSE); } dictionary->tables.tables_len = 0; dictionary->tables.tables_val = NULL; dictionary->count = 0; dictionary->impl_vers = DB_CURRENT_VERSION; initialized = TRUE; } else /* dictionary loaded successfully */ initialized = TRUE; if (initialized == TRUE) { int num_changes = 0; changed = FALSE; reset_log(); if ((num_changes = incorporate_log(logfilename)) < 0) syslog(LOG_ERR, "incorporation of dictionary logfile '%s' failed", logfilename); changed = (num_changes > 0); } WRITEUNLOCK(this, initialized, "wu db_dictionary::init"); return (initialized); } /* * Execute log entry 'j' on the dictionary identified by 'dict' if the * version of j is later than that of the dictionary. If 'j' is executed, * 'count' is incremented and the dictionary's verison is updated to * that of 'j'. * Returns TRUE always for valid log entries; FALSE otherwise. */ static bool_t apply_log_entry(db_dictlog_entry *j, char *dictchar, int *count) { db_dictionary *dict = (db_dictionary*) dictchar; WRITELOCK(dict, FALSE, "w apply_log_entry"); if (db_update_version.earlier_than(j->get_version())) { ++ *count; #ifdef DEBUG j->print(); #endif /* DEBUG */ switch (j->get_action()) { case DB_ADD_TABLE: dict->add_table_aux(j->get_table_name(), j->get_table_object(), INMEMORY_ONLY); // ignore status break; case DB_REMOVE_TABLE: dict->delete_table_aux(j->get_table_name(), INMEMORY_ONLY); // ignore status break; default: WARNING("db::apply_log_entry: unknown action_type"); WRITEUNLOCK(dict, FALSE, "wu apply_log_entry"); return (FALSE); } db_update_version.assign(j->get_version()); } WRITEUNLOCK(dict, TRUE, "wu apply_log_entry"); return (TRUE); } int db_dictionary::incorporate_log(char *file_name) { db_dictlog f(file_name, PICKLE_READ); int ret; WRITELOCK(this, -1, "w db_dictionary::incorporate_log"); setNoWriteThrough(); ret = f.execute_on_log(&(apply_log_entry), (char *) this); clearNoWriteThrough(); WRITEUNLOCK(this, -1, "wu db_dictionary::incorporate_log"); return (ret); } /* Frees memory of filename and tables. Has no effect on disk storage. */ db_status db_dictionary::db_shutdown() { WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::db_shutdown"); if (!initialized) { WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::db_shutdown"); return (DB_SUCCESS); /* DB_NOTFOUND? */ } if (filename) { delete filename; filename = NULL; } if (tmpfilename) { delete tmpfilename; tmpfilename = NULL; } if (logfilename) { delete logfilename; logfilename = NULL; } if (dictionary) { delete_dictionary(dictionary); dictionary = NULL; } initialized = FALSE; changed = FALSE; reset_log(); WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::db_shutdown"); return (DB_SUCCESS); } /* * Dump contents of this dictionary (minus the database representations) * to its file. Returns 0 if operation succeeds, -1 otherwise. */ int db_dictionary::dump() { int status; READLOCK(this, -1, "r db_dictionary::dump"); if (!initialized) { READUNLOCK(this, -1, "ru db_dictionary::dump"); return (-1); } unlink(tmpfilename); /* get rid of partial dumps */ pickle_dict_desc f(tmpfilename, PICKLE_WRITE); status = f.transfer(&dictionary); /* dump table descs */ if (status != 0) { WARNING("db_dictionary::dump: could not write out dictionary"); } else if (rename(tmpfilename, filename) < 0) { WARNING_M("db_dictionary::dump: could not rename temp file: "); status = -1; } READUNLOCK(this, -1, "ru db_dictionary::dump"); return (status); } /* * Write out in-memory copy of dictionary to file. * 1. Update major version. * 2. Dump contents to temporary file. * 3. Rename temporary file to real dictionary file. * 4. Remove log file. * A checkpoint is done only if it has changed since the previous checkpoint. * Returns DB_SUCCESS if checkpoint was successful; error code otherwise */ db_status db_dictionary::checkpoint() { WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::checkpoint"); if (changed == FALSE) { WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::checkpoint"); return (DB_SUCCESS); } vers *oldv = new vers(db_update_version); // copy vers * newv = db_update_version.nextmajor(); // get next version db_update_version.assign(newv); // update version delete newv; if (dump() != 0) { WARNING_M( "db_dictionary::checkpoint: could not dump dictionary: "); db_update_version.assign(oldv); // rollback delete oldv; WRITEUNLOCK(this, DB_INTERNAL_ERROR, "wu db_dictionary::checkpoint"); return (DB_INTERNAL_ERROR); } unlink(logfilename); /* should do atomic rename and log delete */ reset_log(); /* should check for what? */ delete oldv; changed = FALSE; WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::checkpoint"); return (DB_SUCCESS); } /* close existing logfile and delete its structure */ int db_dictionary::reset_log() { WRITELOCK(this, -1, "w db_dictionary::reset_log"); /* try to close old log file */ /* doesnot matter since we do synchronous writes only */ if (logfile != NULL) { if (logfile_opened == TRUE) { if (logfile->close() < 0) { WARNING_M( "db_dictionary::reset_log: could not close log file: "); } } delete logfile; logfile = NULL; } logfile_opened = FALSE; WRITEUNLOCK(this, -1, "wu db_dictionary::reset_log"); return (0); } /* close existing logfile, but leave its structure if exists */ int db_dictionary::close_log() { WRITELOCK(this, -1, "w db_dictionary::close_log"); if (logfile != NULL && logfile_opened == TRUE) { logfile->close(); } logfile_opened = FALSE; WRITEUNLOCK(this, -1, "wu db_dictionary::close_log"); return (0); } /* open logfile, creating its structure if it does not exist */ int db_dictionary::open_log() { WRITELOCK(this, -1, "w db_dictionary::open_log"); if (logfile == NULL) { if ((logfile = new db_dictlog(logfilename, PICKLE_APPEND)) == NULL) { WRITEUNLOCK(this, -1, "wu db_dictionary::open_log"); FATAL3( "db_dictionary::reset_log: cannot allocate space", DB_MEMORY_LIMIT, -1); } } if (logfile_opened == TRUE) { WRITEUNLOCK(this, -1, "wu db_dictionary::open_log"); return (0); } if ((logfile->open()) == FALSE) { WARNING_M("db_dictionary::open_log: could not open log file: "); delete logfile; logfile = NULL; WRITEUNLOCK(this, -1, "wu db_dictionary::open_log"); return (-1); } logfile_opened = TRUE; WRITEUNLOCK(this, -1, "wu db_dictionary::open_log"); return (0); } /* * closes any open log files for all tables in dictionary or 'tab'. * "tab" is an optional argument. */ static int close_standby_list(); db_status db_dictionary::db_standby(char *tab) { db_table_desc *tbl; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::db_standby"); if (!initialized) { WRITEUNLOCK(this, DB_BADDICTIONARY, "wu db_dictionary::db_standby"); return (DB_BADDICTIONARY); } if (tab == NULL) { close_log(); // close dictionary log close_standby_list(); WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::db_standby"); return (DB_SUCCESS); } if ((tbl = find_table_desc(tab)) == NULL) { WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::db_standby"); return (DB_BADTABLE); } if (tbl->database != NULL) tbl->database->close_log(); WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::db_standby"); return (DB_SUCCESS); } /* * Returns db_table_desc of table name 'tab'. 'prev', if supplied, * is set to the entry located ahead of 'tab's entry in the dictionary. */ db_table_desc* db_dictionary::find_table_desc(char *tab) { db_table_desc *ret; READLOCK(this, NULL, "r db_dictionary::find_table_desc"); if (initialized) ret = search_dictionary(dictionary, tab); else ret = NULL; READUNLOCK(this, ret, "r db_dictionary::find_table_desc"); return (ret); } db_table_desc * db_dictionary::find_table_desc(char *tab, bool_t searchDeferred) { db_table_desc *ret = NULL; READLOCK(this, NULL, "r db_dictionary::find_table_desc_d"); /* If desired, look in the deferred dictionary first */ if (initialized && searchDeferred && deferred.dictionary != NULL) ret = search_dictionary(deferred.dictionary, tab); /* No result yet => search the "normal" dictionary */ if (ret == NULL) ret = find_table_desc(tab); READUNLOCK(this, ret, "r db_dictionary::find_table_desc_d"); return (ret); } db * db_dictionary::find_table(char *tab, db_table_desc **where) { /* Most operations should use the deferred dictionary if it exists */ return (find_table(tab, where, TRUE, TRUE, TRUE)); } db * db_dictionary::find_table(char *tab, db_table_desc **where, bool_t searchDeferred) { return (find_table(tab, where, searchDeferred, TRUE, TRUE)); } db * db_dictionary::find_table(char *tab, db_table_desc **where, bool_t searchDeferred, bool_t doLDAP, bool_t doLoad) { db *res; int lstat; db_status dstat; const char *myself = "db_dictionary::find_table"; res = find_table_noLDAP(tab, where, searchDeferred, doLoad); /* If found, or shouldn't try LDAP, we're done */ if (res != 0 || !doLDAP) return (res); /* See if we can retrieve the object from LDAP */ dstat = dbCreateFromLDAP(tab, &lstat); if (dstat != DB_SUCCESS) { if (dstat == DB_NOTFOUND) { if (lstat != LDAP_SUCCESS) { logmsg(MSG_NOTIMECHECK, LOG_INFO, "%s: LDAP error for \"%s\": %s", myself, NIL(tab), ldap_err2string(lstat)); } } else { logmsg(MSG_NOTIMECHECK, LOG_INFO, "%s: DB error %d for \"%s\"", myself, dstat, NIL(tab)); } return (0); } /* Try the dictionary again */ res = find_table_noLDAP(tab, where, searchDeferred, doLoad); return (res); } /* * Return database structure of table named by 'tab'. * If 'where' is set, set it to the table_desc of 'tab.' * If the database is loaded in from stable store if it has not been loaded. * If it cannot be loaded, it is initialized using the scheme stored in * the table_desc. NULL is returned if the initialization fails. */ db * db_dictionary::find_table_noLDAP(char *tab, db_table_desc **where, bool_t searchDeferred, bool_t doLoad) { if (!initialized) return (NULL); db_table_desc* tbl; db *dbase = NULL; int lret; READLOCK(this, NULL, "r db_dictionary::find_table"); tbl = find_table_desc(tab, searchDeferred); if (tbl == NULL) { READUNLOCK(this, NULL, "ru db_dictionary::find_table"); return (NULL); // not found } if (tbl->database != NULL || !doLoad) { if (tbl->database && where) *where = tbl; READUNLOCK(this, NULL, "ru db_dictionary::find_table"); return (tbl->database); // return handle } READUNLOCK(this, NULL, "ru db_dictionary::find_table"); WRITELOCK(this, NULL, "w db_dictionary::find_table"); /* Re-check; some other thread might have loaded the db */ if (tbl->database != NULL) { if (where) *where = tbl; WRITEUNLOCK(this, NULL, "wu db_dictionary::find_table"); return (tbl->database); // return handle } // need to load in/init database dbase = new db(tab); if (dbase == NULL) { WRITEUNLOCK(this, NULL, "db_dictionary::find_table: could not allocate space"); FATAL3("db_dictionary::find_table: could not allocate space", DB_MEMORY_LIMIT, NULL); } /* * Lock the newly created 'dbase', so we can release the general * db_dictionary lock. */ WRITELOCKNR(dbase, lret, "w dbase db_dictionary::find_table"); if (lret != 0) { WRITEUNLOCK(this, NULL, "db_dictionary::find_table: could not lock dbase"); FATAL3("db_dictionary::find_table: could not lock dbase", DB_LOCK_ERROR, NULL); } /* Assign tbl->database, and then release the 'this' lock */ tbl->database = dbase; WRITEUNLOCK(this, NULL, "wu db_dictionary::find_table"); if (dbase->load()) { // try to load in database if (where) *where = tbl; WRITEUNLOCK(dbase, dbase, "wu dbase db_dictionary::find_table"); return (dbase); } delete dbase; tbl->database = NULL; WARNING("db_dictionary::find_table: could not load database"); return (NULL); } /* Log action to be taken on the dictionary and update db_update_version. */ db_status db_dictionary::log_action(int action, char *tab, table_obj *tobj) { WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::log_action"); vers *newv = db_update_version.nextminor(); db_dictlog_entry le(action, newv, tab, tobj); if (open_log() < 0) { delete newv; WRITEUNLOCK(this, DB_STORAGE_LIMIT, "wu db_dictionary::log_action"); return (DB_STORAGE_LIMIT); } if (logfile->append(&le) < 0) { WARNING_M("db::log_action: could not add log entry: "); close_log(); delete newv; WRITEUNLOCK(this, DB_STORAGE_LIMIT, "wu db_dictionary::log_action"); return (DB_STORAGE_LIMIT); } db_update_version.assign(newv); delete newv; changed = TRUE; WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::log_action"); return (DB_SUCCESS); } // For a complete 'delete' operation, we want the following behaviour: // 1. If there is an entry in the log, the physical table exists and is // stable. // 2. If there is no entry in the log, the physical table may or may not // exist. db_status db_dictionary::delete_table_aux(char *tab, int mode) { db_status ret; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::delete_table_aux"); if (!initialized) { WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::delete_table_aux"); return (DB_BADDICTIONARY); } db_table_desc *tbl; if ((tbl = find_table_desc(tab)) == NULL) { // table not found WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::delete_table_aux"); return (DB_NOTFOUND); } if (mode != INMEMORY_ONLY) { int need_free = 0; // Update log. db_status status = log_action(DB_REMOVE_TABLE, tab); if (status != DB_SUCCESS) { WRITEUNLOCK(this, status, "wu db_dictionary::delete_table_aux"); return (status); } // Remove physical structures db *dbase = tbl->database; if (dbase == NULL) { // need to get desc to access files dbase = new db(tab); need_free = 1; } if (dbase == NULL) { WARNING( "db_dictionary::delete_table: could not create db structure"); WRITEUNLOCK(this, DB_MEMORY_LIMIT, "wu db_dictionary::delete_table_aux"); return (DB_MEMORY_LIMIT); } dbase->remove_files(); // remove physical files if (need_free) delete dbase; } // Remove in-memory structures ret = remove_from_dictionary(dictionary, tab, TRUE); WRITEUNLOCK(this, ret, "wu db_dictionary::delete_table_aux"); return (ret); } /* * Delete table with given name 'tab' from dictionary. * Returns error code if table does not exist or if dictionary has not been * initialized. Dictionary is updated to stable store if deletion is * successful. Fatal error occurs if dictionary cannot be saved. * Returns DB_SUCCESS if dictionary has been updated successfully. * Note that the files associated with the table are also removed. */ db_status db_dictionary::delete_table(char *tab) { return (delete_table_aux(tab, !INMEMORY_ONLY)); } // For a complete 'add' operation, we want the following behaviour: // 1. If there is an entry in the log, then the physical table exists and // has been initialized properly. // 2. If there is no entry in the log, the physical table may or may not // exist. In this case, we don't really care because we cannot get at // it. The next time we add a table with the same name to the dictionary, // it will be initialized properly. // This mode is used when the table is first created. // // For an INMEMORY_ONLY operation, only the internal structure is created and // updated. This mode is used when the database gets loaded and the internal // dictionary gets updated from the log entries. db_status db_dictionary::add_table_aux(char *tab, table_obj* tobj, int mode) { db_status ret; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::add_table_aux"); if (!initialized) { WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::add_table_aux"); return (DB_BADDICTIONARY); } if (find_table_desc(tab) != NULL) { WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::add_table_aux"); return (DB_NOTUNIQUE); // table already exists } // create data structures for table db_table_desc *new_table = 0; db_status status = create_table_desc(tab, tobj, &new_table); if (status != DB_SUCCESS) { WRITEUNLOCK(this, DB_LOCK_ERROR, "wu db_dictionary::add_table_aux"); return (status); } if (mode != INMEMORY_ONLY) { // create physical structures for table new_table->database = new db(tab); if (new_table->database == NULL) { delete_table_desc(new_table); WRITEUNLOCK(this, DB_MEMORY_LIMIT, "db_dictionary::add_table: could not allocate space for db"); FATAL3( "db_dictionary::add_table: could not allocate space for db", DB_MEMORY_LIMIT, DB_MEMORY_LIMIT); } if (new_table->database->init(new_table->scheme) == 0) { WARNING( "db_dictionary::add_table: could not initialize database from scheme"); new_table->database->remove_files(); delete_table_desc(new_table); WRITEUNLOCK(this, DB_STORAGE_LIMIT, "wu db_dictionary::add_table_aux"); return (DB_STORAGE_LIMIT); } // update 'external' copy of dictionary status = log_action(DB_ADD_TABLE, tab, tobj); if (status != DB_SUCCESS) { new_table->database->remove_files(); delete_table_desc(new_table); WRITEUNLOCK(this, status, "wu db_dictionary::add_table_aux"); return (status); } } // finally, update in-memory copy of dictionary ret = add_to_dictionary(dictionary, new_table); WRITEUNLOCK(this, ret, "wu db_dictionary::add_table_aux"); return (ret); } /* * Add table with given name 'tab' and description 'zdesc' to dictionary. * Returns error code if table already exists, or if no memory can be found * to store the descriptor, or if dictionary has not been intialized. * Dictionary is updated to stable store if addition is successful. * Fatal error occurs if dictionary cannot be saved. * Returns DB_SUCCESS if dictionary has been updated successfully. */ db_status db_dictionary::add_table(char *tab, table_obj* tobj) { return (add_table_aux(tab, tobj, !INMEMORY_ONLY)); } /* * Translate given NIS attribute list to a db_query structure. * Return FALSE if dictionary has not been initialized, or * table does not have a scheme (which should be a fatal error?). */ db_query* db_dictionary::translate_to_query(db_table_desc* tbl, int numattrs, nis_attr* attrlist) { READLOCK(this, NULL, "r db_dictionary::translate_to_query"); if (!initialized || tbl->scheme == NULL || numattrs == 0 || attrlist == NULL) { READUNLOCK(this, NULL, "ru db_dictionary::translate_to_query"); return (NULL); } db_query *q = new db_query(tbl->scheme, numattrs, attrlist); if (q == NULL) { READUNLOCK(this, NULL, "db_dictionary::translate: could not allocate space"); FATAL3("db_dictionary::translate: could not allocate space", DB_MEMORY_LIMIT, NULL); } if (q->size() == 0) { delete q; READUNLOCK(this, NULL, "ru db_dictionary::translate_to_query"); return (NULL); } READUNLOCK(this, NULL, "ru db_dictionary::translate_to_query"); return (q); } static db_table_names gt_answer; static int gt_posn; static db_status get_table_name(db_table_desc* tbl) { if (tbl) return (DB_BADTABLE); if (gt_posn < gt_answer.db_table_names_len) gt_answer.db_table_names_val[gt_posn++] = strdup(tbl->table_name); else return (DB_BADTABLE); return (DB_SUCCESS); } /* * Return the names of tables in this dictionary. * XXX This routine is used only for testing only; * if to be used for real, need to free memory sensibly, or * caller of get_table_names should have freed them. */ db_table_names* db_dictionary::get_table_names() { READLOCK(this, NULL, "r db_dictionary::get_table_names"); gt_answer.db_table_names_len = dictionary->count; gt_answer.db_table_names_val = new db_table_namep[dictionary->count]; gt_posn = 0; if ((gt_answer.db_table_names_val) == NULL) { READUNLOCK(this, NULL, "db_dictionary::get_table_names: could not allocate space for names"); FATAL3( "db_dictionary::get_table_names: could not allocate space for names", DB_MEMORY_LIMIT, NULL); } enumerate_dictionary(dictionary, &get_table_name); READUNLOCK(this, NULL, "ru db_dictionary::get_table_names"); return (>_answer); } static db_status db_checkpoint_aux(db_table_desc *current) { db *dbase; int status; if (current == NULL) return (DB_BADTABLE); if (current->database == NULL) { /* need to load it in */ dbase = new db(current->table_name); if (dbase == NULL) { FATAL3( "db_dictionary::db_checkpoint: could not allocate space", DB_MEMORY_LIMIT, DB_MEMORY_LIMIT); } if (dbase->load() == 0) { syslog(LOG_ERR, "db_dictionary::db_checkpoint: could not load table %s", current->table_name); delete dbase; return (DB_BADTABLE); } status = dbase->checkpoint(); delete dbase; // unload } else status = current->database->checkpoint(); if (status == 0) return (DB_STORAGE_LIMIT); return (DB_SUCCESS); } /* Like db_checkpoint_aux except only stops on LIMIT errors */ static db_status db_checkpoint_aux_cont(db_table_desc *current) { db_status status = db_checkpoint_aux(current); if (status == DB_STORAGE_LIMIT || status == DB_MEMORY_LIMIT) return (status); else return (DB_SUCCESS); } db_status db_dictionary::db_checkpoint(char *tab) { db_table_desc *tbl; db_status ret; bool_t init; READLOCK(this, DB_LOCK_ERROR, "r db_dictionary::db_checkpoint"); init = initialized; READUNLOCK(this, DB_LOCK_ERROR, "ru db_dictionary::db_checkpoint"); if (!init) return (DB_BADDICTIONARY); checkpoint(); // checkpoint dictionary first READLOCK(this, DB_LOCK_ERROR, "r db_dictionary::db_checkpoint"); if (tab == NULL) { ret = enumerate_dictionary(dictionary, &db_checkpoint_aux_cont); READUNLOCK(this, ret, "ru db_dictionary::db_checkpoint"); return (ret); } if ((tbl = find_table_desc(tab)) == NULL) { READUNLOCK(this, DB_LOCK_ERROR, "ru db_dictionary::db_checkpoint"); return (DB_BADTABLE); } ret = db_checkpoint_aux(tbl); READUNLOCK(this, ret, "ru db_dictionary::db_checkpoint"); return (ret); } /* *********************** db_standby **************************** */ /* Deal with list of tables that need to be 'closed' */ #define OPENED_DBS_CHUNK 12 static db **db_standby_list; static uint_t db_standby_size = 0; static uint_t db_standby_count = 0; DECLMUTEXLOCK(db_standby_list); /* * Returns 1 if all databases on the list could be closed, 0 * otherwise. */ static int close_standby_list() { db *database; int i, ret; const char *myself = "close_standby_list"; MUTEXLOCK(db_standby_list, "close_standby_list"); if (db_standby_count == 0) { MUTEXUNLOCK(db_standby_list, "close_standby_list"); return (1); } for (i = 0, ret = 0; i < db_standby_size; i++) { if ((database = db_standby_list[i])) { /* * In order to avoid a potential dead-lock, we * check to see if close_log() would be able to * lock the db; if not, just skip the db. */ int lockok; TRYWRITELOCK(database, lockok, "try w db_dictionary::close_standby_list"); if (lockok == 0) { database->close_log(1); db_standby_list[i] = (db*)NULL; --db_standby_count; WRITEUNLOCK(database, db_standby_count == 0, "db_dictionary::close_standby_list"); if (db_standby_count == 0) { ret = 1; break; } } else if (lockok != EBUSY) { logmsg(MSG_NOTIMECHECK, LOG_INFO, "%s: try-lock error %d", myself, lockok); } /* else it's EBUSY; skip to the next one */ } } MUTEXUNLOCK(db_standby_list, "close_standby_list"); return (ret); } /* * Add given database to list of databases that have been opened for updates. * If size of list exceeds maximum, close opened databases first. */ int add_to_standby_list(db* database) { int i; const char *myself = "add_to_standby_list"; MUTEXLOCK(db_standby_list, "add_to_standby_list"); if (database == 0) { MUTEXUNLOCK(db_standby_list, "add_to_standby_list"); return (1); } /* Try to keep the list below OPENED_DBS_CHUNK */ if (db_standby_count >= OPENED_DBS_CHUNK) { MUTEXUNLOCK(db_standby_list, "add_to_standby_list"); close_standby_list(); MUTEXLOCK(db_standby_list, "add_to_standby_list"); } if (db_standby_count >= db_standby_size) { db **ndsl = (db **)realloc(db_standby_list, (db_standby_size+OPENED_DBS_CHUNK) * sizeof (ndsl[0])); if (ndsl == 0) { MUTEXUNLOCK(db_standby_list, "add_to_standby_list"); logmsg(MSG_NOMEM, LOG_ERR, "%s: realloc(%d) => NULL", myself, (db_standby_size+OPENED_DBS_CHUNK) * sizeof (ndsl[0])); return (0); } db_standby_list = ndsl; for (i = db_standby_size; i < db_standby_size+OPENED_DBS_CHUNK; i++) db_standby_list[i] = 0; db_standby_size += OPENED_DBS_CHUNK; } for (i = 0; i < db_standby_size; i++) { if (db_standby_list[i] == (db*)NULL) { db_standby_list[i] = database; ++db_standby_count; MUTEXUNLOCK(db_standby_list, "add_to_standby_list"); return (1); } } MUTEXUNLOCK(db_standby_list, "add_to_standby_list"); return (0); } int remove_from_standby_list(db* database) { int i; MUTEXLOCK(db_standby_list, "remove_from_standby_list"); if (database == 0) { MUTEXUNLOCK(db_standby_list, "remove_from_standby_list"); return (1); } for (i = 0; i < db_standby_size; i++) { if ((database == db_standby_list[i])) { db_standby_list[i] = (db*)NULL; --db_standby_count; MUTEXUNLOCK(db_standby_list, "remove_from_standby_list"); return (1); } } MUTEXUNLOCK(db_standby_list, "remove_from_standby_list"); return (0); } /* Release space for copied dictionary */ static void db_release_dictionary(db_dict_desc_p d) { int i; if (d != NULL) { for (i = 0; i < d->tables.tables_len; i++) { db_table_desc_p n, t = d->tables.tables_val[i]; while (t != NULL) { n = t->next; delete_table_desc(t); t = n; } } delete d; } return; } /* * Make a copy of the dictionary */ db_dict_desc_p db_dictionary::db_copy_dictionary(void) { db_dict_desc_p tmp; int i, ok = 1, count = 0; WRITELOCK(this, NULL, "db_dictionary::db_copy_dictionary w"); if (dictionary == NULL) { WRITEUNLOCK(this, NULL, "db_dictionary::db_copy_dictionary wu"); return (NULL); } tmp = new db_dict_desc; if (tmp == NULL) { WRITEUNLOCK(this, NULL, "db_dictionary::db_copy_dictionary wu: no memory"); return (NULL); } tmp->tables.tables_val = (db_table_desc_p *)calloc( tmp->tables.tables_len, sizeof (tmp->tables.tables_val[0])); if (tmp->tables.tables_val == NULL) { delete tmp; WRITEUNLOCK(this, NULL, "db_dictionary::db_copy_dictionary wu: no memory"); return (NULL); } tmp->impl_vers = dictionary->impl_vers; tmp->tables.tables_len = 0; tmp->count = 0; /* For each table ... */ for (i = 0; ok && i < dictionary->tables.tables_len; i++) { db_table_desc_p tbl = NULL, t = dictionary->tables.tables_val[i]; /* ... and each bucket in the chain ... */ while (ok && t != NULL) { db_table_desc_p n, savenext = t->next; t->next = NULL; if (db_clone_bucket(t, &n)) { if (tbl != NULL) { tbl->next = n; } else { tmp->tables.tables_val[i] = n; } tbl = n; tmp->count++; } else { ok = 0; } t->next = savenext; } tmp->tables.tables_len++; } if (ok) { #ifdef NISDB_LDAP_DEBUG if ((tmp->tables.tables_len != dictionary->tables.tables_len) || (tmp->count != dictionary->count)) abort(); #endif /* NISDB_LDAP_DEBUG */ } else { db_release_dictionary(tmp); tmp = NULL; } return (tmp); } /* * Set deferred commit mode. To do this, we make a copy of the table * (structures and data), and put that on the deferred dictionary list. * This list is used for lookups during a resync, so clients continue * to see the pre-resync data. Meanwhile, any changes (including table * deletes) are done to the (temporarily hidden to clients) table in * the normal dictionary. */ db_status db_dictionary::defer(char *table) { db_status ret = DB_SUCCESS; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::defer"); db_table_desc *tbl = find_table_desc(table); int res; const char *myself = "db_dictionary::defer"; if (tbl != NULL) { db_table_desc *clone, *savenext = tbl->next; /* * Only want to clone one db_table_desc, so temporarily * unlink the tail. */ tbl->next = NULL; res = db_clone_bucket(tbl, &clone); /* Restore link to tail */ tbl->next = savenext; if (res == 1) { db_status stat; if (deferred.dictionary == NULL) { deferred.dictionary = new db_dict_desc; if (deferred.dictionary == NULL) { WRITEUNLOCK(this, DB_MEMORY_LIMIT, "wu db_dictionary::defer"); return (DB_MEMORY_LIMIT); } deferred.dictionary->tables.tables_len = 0; deferred.dictionary->tables.tables_val = NULL; deferred.dictionary->count = 0; deferred.dictionary->impl_vers = DB_CURRENT_VERSION; } ret = DB_SUCCESS; /* Initialize and load the database for the clone */ if (clone->database == 0) { clone->database = new db(table); if (clone->database != 0) { if (clone->database->load()) { logmsg(MSG_NOTIMECHECK, #ifdef NISDB_LDAP_DEBUG LOG_WARNING, #else LOG_INFO, #endif /* NISDB_LDAP_DEBUG */ "%s: Clone DB for \"%s\" loaded", myself, NIL(table)); } else { logmsg(MSG_NOTIMECHECK, LOG_ERR, "%s: Error loading clone DB for \"%s\"", myself, NIL(table)); delete clone->database; clone->database = 0; ret = DB_INTERNAL_ERROR; } } else { logmsg(MSG_NOTIMECHECK, LOG_ERR, "%s: Unable to clone DB for \"%s\"", myself, NIL(table)); ret = DB_MEMORY_LIMIT; } } if (clone->database != 0) { clone->database->markDeferred(); stat = add_to_dictionary(deferred.dictionary, clone); ret = stat; if (stat != DB_SUCCESS) { delete clone->database; clone->database = 0; delete clone; if (stat == DB_NOTUNIQUE) { /* Already deferred */ ret = DB_SUCCESS; } } } else { delete clone; /* Return value already set above */ } } else { ret = DB_INTERNAL_ERROR; } } else { ret = DB_NOTFOUND; } WRITEUNLOCK(this, ret, "wu db_dictionary::defer"); return (ret); } /* * Unset deferred commit mode and roll back changes; doesn't recover the * disk data, but that's OK, since we only want to be able to continue * serving the table until we can try a full dump again. * * The rollback is done by removing (and deleting) the updated table from * the dictionary, and then moving the saved table from the deferred * dictionary list to the actual one. */ db_status db_dictionary::rollback(char *table) { db_status ret = DB_SUCCESS; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::rollback"); db_table_desc *old = search_dictionary(deferred.dictionary, table); db_table_desc *upd = search_dictionary(dictionary, table); if (old == NULL) { WRITEUNLOCK(this, DB_NOTFOUND, "wu db_dictionary::rollback"); return (DB_NOTFOUND); } /* * Remove old incarnation from deferred dictionary. We already hold * a pointer ('old') to it, so don't delete. */ ret = remove_from_dictionary(deferred.dictionary, table, FALSE); if (ret != DB_SUCCESS) { #ifdef NISDB_LDAP_DEBUG abort(); #endif /* NISDB_LDAP_DEBUG */ WRITEUNLOCK(this, ret, "wu db_dictionary::rollback"); return (ret); } if (old->database != 0) old->database->unmarkDeferred(); /* * Remove updated incarnation from dictionary. If 'upd' is NULL, * the table has been removed while we were in deferred mode, and * that's OK; we just need to retain the old incarnation. */ if (upd != NULL) { ret = remove_from_dictionary(dictionary, table, FALSE); if (ret != DB_SUCCESS) { #ifdef NISDB_LDAP_DEBUG abort(); #endif /* NISDB_LDAP_DEBUG */ /* * Cut our losses; delete old incarnation, and leave * updated one in place. */ delete_table_desc(old); WRITEUNLOCK(this, ret, "wu db_dictionary::rollback"); return (ret); } /* Throw away updates */ delete_table_desc(upd); } /* (Re-)insert old incarnation in the dictionary */ ret = add_to_dictionary(dictionary, old); if (ret != DB_SUCCESS) { #ifdef NISDB_LDAP_DEBUG abort(); #endif /* NISDB_LDAP_DEBUG */ /* At least avoid memory leak */ delete_table_desc(old); syslog(LOG_ERR, "db_dictionary::rollback: rollback error %d for \"%s\"", ret, table); } WRITEUNLOCK(this, ret, "wu db_dictionary::rollback"); return (ret); } /* * Commit changes. Done by simply removing and deleting the pre-resync * data from the deferred dictionary. */ db_status db_dictionary::commit(char *table) { db_status ret = DB_SUCCESS; WRITELOCK(this, DB_LOCK_ERROR, "w db_dictionary::commit"); db_table_desc *old = search_dictionary(deferred.dictionary, table); if (old == NULL) { /* Fine (we hope); nothing to do */ WRITEUNLOCK(this, ret, "wu db_dictionary::commit"); return (DB_SUCCESS); } ret = remove_from_dictionary(deferred.dictionary, table, FALSE); if (ret == DB_SUCCESS) delete_table_desc(old); #ifdef NISDB_LDAP_DEBUG else abort(); #endif /* NISDB_LDAP_DEBUG */ WRITEUNLOCK(this, ret, "wu db_dictionary::commit"); return (ret); } /* * The noWriteThrough flag is used to prevent modifies/updates to LDAP * while we're incorporating log data into the in-memory tables. */ void db_dictionary::setNoWriteThrough(void) { ASSERTWHELD(this->dict); noWriteThrough.flag++; } void db_dictionary::clearNoWriteThrough(void) { ASSERTWHELD(this->dict); if (noWriteThrough.flag > 0) noWriteThrough.flag--; #ifdef NISDB_LDAP_DEBUG else abort(); #endif /* NISDB_LDAP_DEBUG */ }