/*	$Id: krb_dbm.c,v 1.1 1995/12/14 06:52:36 tholo Exp $	*/

/*-
 * Copyright (C) 1989 by the Massachusetts Institute of Technology
 *
 * Export of this software from the United States of America is assumed
 * to require a specific license from the United States Government.
 * It is the responsibility of any person or organization contemplating
 * export to obtain such a license before exporting.
 *
 * WITHIN THAT CONSTRAINT, permission to use, copy, modify, and
 * distribute this software and its documentation for any purpose and
 * without fee is hereby granted, provided that the above copyright
 * notice appear in all copies and that both that copyright notice and
 * this permission notice appear in supporting documentation, and that
 * the name of M.I.T. not be used in advertising or publicity pertaining
 * to distribution of the software without specific, written prior
 * permission.  M.I.T. makes no representations about the suitability of
 * this software for any purpose.  It is provided "as is" without express
 * or implied warranty.
 */

#include "kdb_locl.h"

#include <ndbm.h>

#define KERB_DB_MAX_RETRY 5

#ifdef DEBUG
extern int debug;
extern long kerb_debug;
extern char *progname;
#endif

static  init = 0;
static char default_db_name[] = DBM_FILE;
static char *current_db_name = default_db_name;

static struct timeval timestamp;/* current time of request */
static int non_blocking = 0;

/*
 * This module contains all of the code which directly interfaces to
 * the underlying representation of the Kerberos database; this
 * implementation uses a DBM or NDBM indexed "file" (actually
 * implemented as two separate files) to store the relations, plus a
 * third file as a semaphore to allow the database to be replaced out
 * from underneath the KDC server.
 */

/*
 * Locking:
 * 
 * There are two distinct locking protocols used.  One is designed to
 * lock against processes (the admin_server, for one) which make
 * incremental changes to the database; the other is designed to lock
 * against utilities (kdb_util, kpropd) which replace the entire
 * database in one fell swoop.
 *
 * The first locking protocol is implemented using flock() in the 
 * krb_dbl_lock() and krb_dbl_unlock routines.
 *
 * The second locking protocol is necessary because DBM "files" are
 * actually implemented as two separate files, and it is impossible to
 * atomically rename two files simultaneously.  It assumes that the
 * database is replaced only very infrequently in comparison to the time
 * needed to do a database read operation.
 *
 * A third file is used as a "version" semaphore; the modification
 * time of this file is the "version number" of the database.
 * At the start of a read operation, the reader checks the version
 * number; at the end of the read operation, it checks again.  If the
 * version number changed, or if the semaphore was nonexistant at
 * either time, the reader sleeps for a second to let things
 * stabilize, and then tries again; if it does not succeed after
 * KERB_DB_MAX_RETRY attempts, it gives up.
 * 
 * On update, the semaphore file is deleted (if it exists) before any
 * update takes place; at the end of the update, it is replaced, with
 * a version number strictly greater than the version number which
 * existed at the start of the update.
 * 
 * If the system crashes in the middle of an update, the semaphore
 * file is not automatically created on reboot; this is a feature, not
 * a bug, since the database may be inconsistant.  Note that the
 * absence of a semaphore file does not prevent another _update_ from
 * taking place later.  Database replacements take place automatically
 * only on slave servers; a crash in the middle of an update will be
 * fixed by the next slave propagation.  A crash in the middle of an
 * update on the master would be somewhat more serious, but this would
 * likely be noticed by an administrator, who could fix the problem and
 * retry the operation.
 */

/* Macros to convert ndbm names to dbm names.
 * Note that dbm_nextkey() cannot be simply converted using a macro, since
 * it is invoked giving the database, and nextkey() needs the previous key.
 *
 * Instead, all routines call "dbm_next" instead.
 */

#define dbm_next(db,key) dbm_nextkey(db)

static char *gen_dbsuffix __P((char *db_name, char *sfx));
static void decode_princ_key __P((datum *key, char *name, char *instance));
static void encode_princ_contents __P((datum *contents, Principal *principal));
static void decode_princ_contents __P((datum *contents, Principal *principal));
static void encode_princ_key __P((datum *key, char *name, char *instance));
static int kerb_dbl_init __P((void));
static void kerb_dbl_fini __P((void));
static int kerb_dbl_lock __P((int mode));
static void kerb_dbl_unlock __P((void));
static time_t kerb_start_update __P((char *db_name));
static int kerb_end_update __P((char *db_name, time_t age));
static time_t kerb_start_read __P((void));
static int kerb_end_read __P((time_t age));

/*
 * Utility routine: generate name of database file.
 */

static char *
gen_dbsuffix(db_name, sfx)
     char *db_name;
     char *sfx;
{
    char *dbsuffix;
    
    if (sfx == NULL)
	sfx = ".ok";

    dbsuffix = malloc (strlen(db_name) + strlen(sfx) + 1);
    strcpy(dbsuffix, db_name);
    strcat(dbsuffix, sfx);
    return dbsuffix;
}

static void
decode_princ_key(key, name, instance)
     datum *key;
     char *name;
     char *instance;
{
    strncpy(name, key->dptr, ANAME_SZ);
    strncpy(instance, key->dptr + ANAME_SZ, INST_SZ);
    name[ANAME_SZ - 1] = '\0';
    instance[INST_SZ - 1] = '\0';
}

static void
encode_princ_contents(contents, principal)
     datum *contents;
     Principal *principal;
{
    contents->dsize = sizeof(*principal);
    contents->dptr = (char *) principal;
}

static void
decode_princ_contents(contents, principal)
     datum *contents;
     Principal *principal;
{
    bcopy(contents->dptr, (char *) principal, sizeof(*principal));
}

static void
encode_princ_key(key, name, instance)
     datum *key;
     char *name;
     char *instance;
{
    static char keystring[ANAME_SZ + INST_SZ];

    bzero(keystring, ANAME_SZ + INST_SZ);
    strncpy(keystring, name, ANAME_SZ);
    strncpy(&keystring[ANAME_SZ], instance, INST_SZ);
    key->dptr = keystring;
    key->dsize = ANAME_SZ + INST_SZ;
}

static int dblfd = -1;		/* db LOCK fd */
static int mylock = 0;
static int inited = 0;

static int
kerb_dbl_init()
{
    if (!inited) {
	char *filename = gen_dbsuffix (current_db_name, ".ok");
	if ((dblfd = open(filename, O_RDWR)) < 0) {
	    fprintf(stderr, "kerb_dbl_init: couldn't open %s\n", filename);
	    fflush(stderr);
	    perror("open");
	    exit(1);
	}
	free(filename);
	inited++;
    }
    return (0);
}

static void
kerb_dbl_fini()
{
    close(dblfd);
    dblfd = -1;
    inited = 0;
    mylock = 0;
}

static int
kerb_dbl_lock(mode)
     int mode;
{
    int flock_mode;
    
    if (!inited)
	kerb_dbl_init();
    if (mylock) {		/* Detect lock call when lock already
				 * locked */
	fprintf(stderr, "Kerberos locking error (mylock)\n");
	fflush(stderr);
	exit(1);
    }
    switch (mode) {
    case KERB_DBL_EXCLUSIVE:
	flock_mode = LOCK_EX;
	break;
    case KERB_DBL_SHARED:
	flock_mode = LOCK_SH;
	break;
    default:
	fprintf(stderr, "invalid lock mode %d\n", mode);
	abort();
    }
    if (non_blocking)
	flock_mode |= LOCK_NB;
    
    if (flock(dblfd, flock_mode) < 0) 
	return errno;
    mylock++;
    return 0;
}

static void
kerb_dbl_unlock()
{
    if (!mylock) {		/* lock already unlocked */
	fprintf(stderr, "Kerberos database lock not locked when unlocking.\n");
	fflush(stderr);
	exit(1);
    }
    if (flock(dblfd, LOCK_UN) < 0) {
	fprintf(stderr, "Kerberos database lock error. (unlocking)\n");
	fflush(stderr);
	perror("flock");
	exit(1);
    }
    mylock = 0;
}

int
kerb_db_set_lockmode(mode)
     int mode;
{
    int old = non_blocking;
    non_blocking = mode;
    return old;
}

/*
 * initialization for data base routines.
 */

int
kerb_db_init()
{
    init = 1;
    return (0);
}

/*
 * gracefully shut down database--must be called by ANY program that does
 * a kerb_db_init 
 */

void
kerb_db_fini()
{
}

/*
 * Set the "name" of the current database to some alternate value.
 *
 * Passing a null pointer as "name" will set back to the default.
 * If the alternate database doesn't exist, nothing is changed.
 */

int
kerb_db_set_name(name)
     char *name;
{
    DBM *db;

    if (name == NULL)
	name = default_db_name;
    db = dbm_open(name, 0, 0);
    if (db == NULL)
	return errno;
    dbm_close(db);
    kerb_dbl_fini();
    current_db_name = name;
    return 0;
}

/*
 * Return the last modification time of the database.
 */

time_t
kerb_get_db_age()
{
    struct stat st;
    char *okname;
    time_t age;
    
    okname = gen_dbsuffix(current_db_name, ".ok");

    if (stat (okname, &st) < 0)
	age = 0;
    else
	age = st.st_mtime;

    free (okname);
    return age;
}

/*
 * Remove the semaphore file; indicates that database is currently
 * under renovation.
 *
 * This is only for use when moving the database out from underneath
 * the server (for example, during slave updates).
 */

static time_t
kerb_start_update(db_name)
     char *db_name;
{
    char *okname = gen_dbsuffix(db_name, ".ok");
    time_t age = kerb_get_db_age();
    
    if (unlink(okname) < 0
	&& errno != ENOENT) {
	    age = -1;
    }
    free (okname);
    return age;
}

static int
kerb_end_update(db_name, age)
     char *db_name;
     time_t age;
{
    int fd;
    int retval = 0;
    char *new_okname = gen_dbsuffix(db_name, ".ok#");
    char *okname = gen_dbsuffix(db_name, ".ok");
    
    fd = open (new_okname, O_CREAT|O_RDWR|O_TRUNC, 0600);
    if (fd < 0)
	retval = errno;
    else {
	struct stat st;
	struct utimbuf tv;
	/* make sure that semaphore is "after" previous value. */
	if (fstat (fd, &st) == 0
	    && st.st_mtime <= age) {
	    tv.actime = st.st_atime;
	    tv.modtime = age;
	    /* set times.. */
	    utime (new_okname, &tv);
	    fsync(fd);
	}
	close(fd);
	if (rename (new_okname, okname) < 0)
	    retval = errno;
    }

    free (new_okname);
    free (okname);

    return retval;
}

static time_t
kerb_start_read()
{
    return kerb_get_db_age();
}

static int
kerb_end_read(age)
     time_t age;
{
    if (kerb_get_db_age() != age || age == -1) {
	return -1;
    }
    return 0;
}

/*
 * Create the database, assuming it's not there.
 */

int
kerb_db_create(db_name)
     char *db_name;
{
    char *okname = gen_dbsuffix(db_name, ".ok");
    int fd;
    register int ret = 0;
    DBM *db;

    db = dbm_open(db_name, O_RDWR|O_CREAT|O_EXCL, 0600);
    if (db == NULL)
	ret = errno;
    else
	dbm_close(db);
    if (ret == 0) {
	fd = open (okname, O_CREAT|O_RDWR|O_TRUNC, 0600);
	if (fd < 0)
	    ret = errno;
	close(fd);
    }
    return ret;
}

/*
 * "Atomically" rename the database in a way that locks out read
 * access in the middle of the rename.
 *
 * Not perfect; if we crash in the middle of an update, we don't
 * necessarily know to complete the transaction the rename, but...
 */

int
kerb_db_rename(from, to)
     char *from;
     char *to;
{
#if defined(__NetBSD__) || defined(__FreeBSD__)
    char *fromdb = gen_dbsuffix (from, ".db");
    char *todb = gen_dbsuffix (from, ".db");
#else
    char *fromdir = gen_dbsuffix (from, ".dir");
    char *todir = gen_dbsuffix (to, ".dir");
    char *frompag = gen_dbsuffix (from , ".pag");
    char *topag = gen_dbsuffix (to, ".pag");
#endif
    char *fromok = gen_dbsuffix(from, ".ok");
    long trans = kerb_start_update(to);
    int ok = 0;
    
#if defined(__NetBSD__) || defined(__FreeBSD__)
    if (rename (fromdb, todb) == 0) {
	(void) unlink (fromdb);
	ok = 1;
    }
    free (fromdb);
    free (todb);
#else
    if ((rename (fromdir, todir) == 0)
	&& (rename (frompag, topag) == 0)) {
	(void) unlink (fromok);
	ok = 1;
    }
    free (fromdir);
    free (todir);
    free (frompag);
    free (topag);
#endif
    free (fromok);
    if (ok)
	return kerb_end_update(to, trans);
    else
	return -1;
}

/*
 * look up a principal in the data base returns number of principals
 * found , and whether there were more than requested. 
 */

int
kerb_db_get_principal(name, inst, principal, max, more)
     char *name;		/* could have wild card */
     char *inst;		/* could have wild card */
     Principal *principal;	/* max number of name structs to return */
     unsigned int max;		/* where there more than 'max' tuples? */
     int *more;
{
    int     found = 0, code;
    extern int errorproc();
    int     wildp, wildi;
    datum   key, contents;
    char    testname[ANAME_SZ], testinst[INST_SZ];
    u_long trans;
    int try;
    DBM    *db;

    if (!init)
	kerb_db_init();		/* initialize database routines */

    for (try = 0; try < KERB_DB_MAX_RETRY; try++) {
	trans = kerb_start_read();

	if ((code = kerb_dbl_lock(KERB_DBL_SHARED)) != 0)
	    return -1;

	db = dbm_open(current_db_name, O_RDONLY, 0600);

	*more = 0;

#ifdef DEBUG
	if (kerb_debug & 2)
	    fprintf(stderr,
		    "%s: db_get_principal for %s %s max = %d",
		    progname, name, inst, max);
#endif

	wildp = !strcmp(name, "*");
	wildi = !strcmp(inst, "*");

	if (!wildi && !wildp) {	/* nothing's wild */
	    encode_princ_key(&key, name, inst);
	    contents = dbm_fetch(db, key);
	    if (contents.dptr == NULL) {
		found = 0;
		goto done;
	    }
	    decode_princ_contents(&contents, principal);
#ifdef DEBUG
	    if (kerb_debug & 1) {
		fprintf(stderr, "\t found %s %s p_n length %d t_n length %d\n",
			principal->name, principal->instance,
			strlen(principal->name),
			strlen(principal->instance));
	    }
#endif
	    found = 1;
	    goto done;
	}
	/* process wild cards by looping through entire database */

	for (key = dbm_firstkey(db); key.dptr != NULL;
	     key = dbm_next(db, key)) {
	    decode_princ_key(&key, testname, testinst);
	    if ((wildp || !strcmp(testname, name)) &&
		(wildi || !strcmp(testinst, inst))) { /* have a match */
		if (found >= max) {
		    *more = 1;
		    goto done;
		} else {
		    found++;
		    contents = dbm_fetch(db, key);
		    decode_princ_contents(&contents, principal);
#ifdef DEBUG
		    if (kerb_debug & 1) {
			fprintf(stderr,
				"\tfound %s %s p_n length %d t_n length %d\n",
				principal->name, principal->instance,
				strlen(principal->name),
				strlen(principal->instance));
		    }
#endif
		    principal++; /* point to next */
		}
	    }
	}

    done:
	kerb_dbl_unlock();	/* unlock read lock */
	dbm_close(db);
	if (kerb_end_read(trans) == 0)
	    break;
	found = -1;
	if (!non_blocking)
	    sleep(1);
    }
    return (found);
}

/*
 * Update a name in the data base.  Returns number of names
 * successfully updated.
 */

int
kerb_db_put_principal(principal, max)
     Principal *principal;	/* number of principal structs to */
     unsigned int max;		/* update */

{
    int     found = 0, code;
    u_long  i;
    extern int errorproc();
    datum   key, contents;
    DBM    *db;

    gettimeofday(&timestamp, NULL);

    if (!init)
	kerb_db_init();

    if ((code = kerb_dbl_lock(KERB_DBL_EXCLUSIVE)) != 0)
	return -1;

    db = dbm_open(current_db_name, O_RDWR, 0600);

#ifdef DEBUG
    if (kerb_debug & 2)
	fprintf(stderr, "%s: kerb_db_put_principal  max = %d",
	    progname, max);
#endif

    /* for each one, stuff temps, and do replace/append */
    for (i = 0; i < max; i++) {
	encode_princ_contents(&contents, principal);
	encode_princ_key(&key, principal->name, principal->instance);
	dbm_store(db, key, contents, DBM_REPLACE);
#ifdef DEBUG
	if (kerb_debug & 1) {
	    fprintf(stderr, "\n put %s %s\n",
		principal->name, principal->instance);
	}
#endif
	found++;
	principal++;		/* bump to next struct			   */
    }

    dbm_close(db);
    kerb_dbl_unlock();		/* unlock database */
    return (found);
}

void
kerb_db_get_stat(s)
     DB_stat *s;
{
    gettimeofday(&timestamp, NULL);


    s->cpu = 0;
    s->elapsed = 0;
    s->dio = 0;
    s->pfault = 0;
    s->t_stamp = timestamp.tv_sec;
    s->n_retrieve = 0;
    s->n_replace = 0;
    s->n_append = 0;
    s->n_get_stat = 0;
    s->n_put_stat = 0;
    /* update local copy too */
}

void
kerb_db_put_stat(s)
     DB_stat *s;
{
}

void
delta_stat(a, b, c)
     DB_stat *a, *b, *c;
{
    /* c = a - b then b = a for the next time */

    c->cpu = a->cpu - b->cpu;
    c->elapsed = a->elapsed - b->elapsed;
    c->dio = a->dio - b->dio;
    c->pfault = a->pfault - b->pfault;
    c->t_stamp = a->t_stamp - b->t_stamp;
    c->n_retrieve = a->n_retrieve - b->n_retrieve;
    c->n_replace = a->n_replace - b->n_replace;
    c->n_append = a->n_append - b->n_append;
    c->n_get_stat = a->n_get_stat - b->n_get_stat;
    c->n_put_stat = a->n_put_stat - b->n_put_stat;

    bcopy(a, b, sizeof(DB_stat));
    return;
}

/*
 * look up a dba in the data base returns number of dbas found , and
 * whether there were more than requested. 
 */

int
kerb_db_get_dba(dba_name, dba_inst, dba, max, more)
     char *dba_name;		/* could have wild card */
     char *dba_inst;		/* could have wild card */
     Dba *dba;			/* max number of name structs to return */
     unsigned int max;		/* where there more than 'max' tuples? */
     int *more;
{
    *more = 0;
    return (0);
}

int
kerb_db_iterate (func, arg)
     int (*func)();
     char *arg;			/* void *, really */
{
    datum key, contents;
    Principal *principal;
    int code;
    DBM *db;
    
    kerb_db_init();		/* initialize and open the database */
    if ((code = kerb_dbl_lock(KERB_DBL_SHARED)) != 0)
	return code;

    db = dbm_open(current_db_name, O_RDONLY, 0600);

    for (key = dbm_firstkey (db); key.dptr != NULL; key = dbm_next(db, key)) {
	contents = dbm_fetch (db, key);
	/* XXX may not be properly aligned */
	principal = (Principal *) contents.dptr;
	if ((code = (*func)(arg, principal)) != 0)
	    return code;
    }
    dbm_close(db);
    kerb_dbl_unlock();
    return 0;
}