postgresql/src/backend/commands/user.c

1487 lines
37 KiB
C
Raw Normal View History

/*-------------------------------------------------------------------------
*
* user.c
* Commands for manipulating users and groups.
*
2002-06-20 22:29:54 +02:00
* Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
2002-09-14 15:46:24 +02:00
* $Header: /cvsroot/pgsql/src/backend/commands/user.c,v 1.112 2002/09/14 13:46:23 petere Exp $
*
*-------------------------------------------------------------------------
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
#include "postgres.h"
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <unistd.h>
1999-07-16 07:00:38 +02:00
#include "access/heapam.h"
#include "catalog/catname.h"
#include "catalog/indexing.h"
#include "catalog/pg_database.h"
#include "catalog/pg_group.h"
#include "catalog/pg_shadow.h"
#include "catalog/pg_type.h"
1999-07-16 07:00:38 +02:00
#include "commands/user.h"
#include "libpq/crypt.h"
1999-07-16 07:00:38 +02:00
#include "miscadmin.h"
#include "storage/pmsignal.h"
#include "utils/acl.h"
#include "utils/array.h"
#include "utils/builtins.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
#include "utils/lsyscache.h"
#include "utils/syscache.h"
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
extern bool Password_encryption;
1998-12-14 07:50:32 +01:00
static void CheckPgUserAclNotNull(void);
static void UpdateGroupMembership(Relation group_rel, HeapTuple group_tuple,
2002-09-04 22:31:48 +02:00
List *members);
static IdList *IdListToArray(List *members);
static List *IdArrayToList(IdList *oldarray);
/*
* fputs_quote
*
* Outputs string in quotes, with double-quotes duplicated.
* We could use quote_ident(), but that expects a TEXT argument.
*/
2002-09-04 22:31:48 +02:00
static void
fputs_quote(char *str, FILE *fp)
{
fputc('"', fp);
while (*str)
{
fputc(*str, fp);
if (*str == '"')
fputc('"', fp);
str++;
}
fputc('"', fp);
}
/*
* group_getfilename --- get full pathname of group file
*
* Note that result string is palloc'd, and should be freed by the caller.
*/
char *
group_getfilename(void)
{
int bufsize;
char *pfnam;
bufsize = strlen(DataDir) + strlen("/global/") +
2002-09-04 22:31:48 +02:00
strlen(USER_GROUP_FILE) + 1;
pfnam = (char *) palloc(bufsize);
snprintf(pfnam, bufsize, "%s/global/%s", DataDir, USER_GROUP_FILE);
return pfnam;
}
/*
* Get full pathname of password file.
* Note that result string is palloc'd, and should be freed by the caller.
*/
char *
user_getfilename(void)
{
int bufsize;
char *pfnam;
bufsize = strlen(DataDir) + strlen("/global/") +
2002-09-04 22:31:48 +02:00
strlen(PWD_FILE) + 1;
pfnam = (char *) palloc(bufsize);
snprintf(pfnam, bufsize, "%s/global/%s", DataDir, PWD_FILE);
return pfnam;
}
/*
* write_group_file for trigger update_pg_pwd_and_pg_group
*/
static void
write_group_file(Relation urel, Relation grel)
{
char *filename,
*tempname;
int bufsize;
FILE *fp;
mode_t oumask;
HeapScanDesc scan;
HeapTuple tuple;
TupleDesc dsc = RelationGetDescr(grel);
/*
* Create a temporary filename to be renamed later. This prevents the
2002-09-04 22:31:48 +02:00
* backend from clobbering the pg_group file while the postmaster
* might be reading from it.
*/
filename = group_getfilename();
bufsize = strlen(filename) + 12;
tempname = (char *) palloc(bufsize);
snprintf(tempname, bufsize, "%s.%d", filename, MyProcPid);
oumask = umask((mode_t) 077);
fp = AllocateFile(tempname, "w");
umask(oumask);
if (fp == NULL)
elog(ERROR, "write_group_file: unable to write %s: %m", tempname);
/* read table */
scan = heap_beginscan(grel, SnapshotSelf, 0, NULL);
while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
2002-09-04 22:31:48 +02:00
Datum datum,
grolist_datum;
bool isnull;
char *groname;
IdList *grolist_p;
AclId *aidp;
2002-09-04 22:31:48 +02:00
int i,
j,
num;
2002-09-04 22:31:48 +02:00
char *usename;
bool first_user = true;
datum = heap_getattr(tuple, Anum_pg_group_groname, dsc, &isnull);
/* ignore NULL groupnames --- shouldn't happen */
if (isnull)
continue;
groname = NameStr(*DatumGetName(datum));
/*
* Check for illegal characters in the group name.
*/
i = strcspn(groname, "\n");
if (groname[i] != '\0')
{
elog(LOG, "Invalid group name '%s'", groname);
continue;
}
grolist_datum = heap_getattr(tuple, Anum_pg_group_grolist, dsc, &isnull);
/* Ignore NULL group lists */
if (isnull)
continue;
/* be sure the IdList is not toasted */
grolist_p = DatumGetIdListP(grolist_datum);
/* scan grolist */
num = IDLIST_NUM(grolist_p);
aidp = IDLIST_DAT(grolist_p);
for (i = 0; i < num; ++i)
{
tuple = SearchSysCache(SHADOWSYSID,
PointerGetDatum(aidp[i]),
0, 0, 0);
if (HeapTupleIsValid(tuple))
{
usename = NameStr(((Form_pg_shadow) GETSTRUCT(tuple))->usename);
/*
* Check for illegal characters in the user name.
*/
j = strcspn(usename, "\n");
if (usename[j] != '\0')
{
elog(LOG, "Invalid user name '%s'", usename);
continue;
}
2002-09-04 22:31:48 +02:00
/*
* File format is: "dbname" "user1" "user2" "user3"
*/
if (first_user)
{
fputs_quote(groname, fp);
fputs("\t", fp);
}
else
fputs(" ", fp);
first_user = false;
fputs_quote(usename, fp);
ReleaseSysCache(tuple);
}
}
if (!first_user)
fputs("\n", fp);
/* if IdList was toasted, free detoasted copy */
if ((Pointer) grolist_p != DatumGetPointer(grolist_datum))
pfree(grolist_p);
}
heap_endscan(scan);
fflush(fp);
if (ferror(fp))
elog(ERROR, "%s: %m", tempname);
FreeFile(fp);
/*
* Rename the temp file to its final name, deleting the old pg_pwd. We
* expect that rename(2) is an atomic action.
*/
if (rename(tempname, filename))
elog(ERROR, "rename %s to %s: %m", tempname, filename);
pfree((void *) tempname);
pfree((void *) filename);
}
/*
* write_password_file for trigger update_pg_pwd_and_pg_group
*
* copy the modified contents of pg_shadow to a file used by the postmaster
* for user authentication. The file is stored as $PGDATA/global/pg_pwd.
*
* This function set is both a trigger function for direct updates to pg_shadow
* as well as being called directly from create/alter/drop user.
*
* We raise an error to force transaction rollback if we detect an illegal
* username or password --- illegal being defined as values that would
* mess up the pg_pwd parser.
*/
static void
write_user_file(Relation urel)
{
1999-05-25 18:15:34 +02:00
char *filename,
*tempname;
int bufsize;
FILE *fp;
mode_t oumask;
HeapScanDesc scan;
HeapTuple tuple;
TupleDesc dsc = RelationGetDescr(urel);
/*
* Create a temporary filename to be renamed later. This prevents the
* backend from clobbering the pg_pwd file while the postmaster might
* be reading from it.
*/
filename = user_getfilename();
1998-12-14 07:50:32 +01:00
bufsize = strlen(filename) + 12;
tempname = (char *) palloc(bufsize);
1998-12-14 07:50:32 +01:00
snprintf(tempname, bufsize, "%s.%d", filename, MyProcPid);
oumask = umask((mode_t) 077);
fp = AllocateFile(tempname, "w");
umask(oumask);
if (fp == NULL)
elog(ERROR, "write_password_file: unable to write %s: %m", tempname);
/* read table */
scan = heap_beginscan(urel, SnapshotSelf, 0, NULL);
while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
Datum datum;
bool isnull;
char *usename,
*passwd,
*valuntil;
int i;
datum = heap_getattr(tuple, Anum_pg_shadow_usename, dsc, &isnull);
/* ignore NULL usernames (shouldn't happen) */
if (isnull)
continue;
usename = NameStr(*DatumGetName(datum));
datum = heap_getattr(tuple, Anum_pg_shadow_passwd, dsc, &isnull);
/*
* It can be argued that people having a null password shouldn't
* be allowed to connect under password authentication, because
* they need to have a password set up first. If you think
* assuming an empty password in that case is better, change this
* logic to look something like the code for valuntil.
*/
if (isnull)
continue;
passwd = DatumGetCString(DirectFunctionCall1(textout, datum));
datum = heap_getattr(tuple, Anum_pg_shadow_valuntil, dsc, &isnull);
if (isnull)
valuntil = pstrdup("");
else
valuntil = DatumGetCString(DirectFunctionCall1(nabstimeout, datum));
/*
* Check for illegal characters in the username and password.
*/
i = strcspn(usename, "\n");
if (usename[i] != '\0')
elog(ERROR, "Invalid user name '%s'", usename);
i = strcspn(passwd, "\n");
if (passwd[i] != '\0')
elog(ERROR, "Invalid user password '%s'", passwd);
/*
* The extra columns we emit here are not really necessary. To
* remove them, the parser in backend/libpq/crypt.c would need to
* be adjusted.
*/
fputs_quote(usename, fp);
fputs(" ", fp);
fputs_quote(passwd, fp);
fputs(" ", fp);
fputs_quote(valuntil, fp);
fputs("\n", fp);
pfree(passwd);
pfree(valuntil);
}
heap_endscan(scan);
fflush(fp);
if (ferror(fp))
elog(ERROR, "%s: %m", tempname);
FreeFile(fp);
/*
* Rename the temp file to its final name, deleting the old pg_pwd. We
* expect that rename(2) is an atomic action.
*/
if (rename(tempname, filename))
elog(ERROR, "rename %s to %s: %m", tempname, filename);
pfree((void *) tempname);
pfree((void *) filename);
}
/* This is the wrapper for triggers. */
Datum
update_pg_pwd_and_pg_group(PG_FUNCTION_ARGS)
{
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
/*
* ExclusiveLock ensures no one modifies pg_shadow while we read it,
* and that only one backend rewrites the flat file at a time. It's
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
* OK to allow normal reads of pg_shadow in parallel, however.
*/
Relation urel = heap_openr(ShadowRelationName, ExclusiveLock);
Relation grel = heap_openr(GroupRelationName, ExclusiveLock);
write_user_file(urel);
write_group_file(urel, grel);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
/* OK to release lock, since we did not modify the relation */
heap_close(grel, ExclusiveLock);
heap_close(urel, ExclusiveLock);
/*
* Signal the postmaster to reload its password & group-file cache.
*/
SendPostmasterSignal(PMSIGNAL_PASSWORD_CHANGE);
return PointerGetDatum(NULL);
}
/*
* CREATE USER
*/
void
CreateUser(CreateUserStmt *stmt)
{
1999-05-25 18:15:34 +02:00
Relation pg_shadow_rel;
TupleDesc pg_shadow_dsc;
HeapScanDesc scan;
HeapTuple tuple;
Datum new_record[Natts_pg_shadow];
char new_record_nulls[Natts_pg_shadow];
bool user_exists = false,
sysid_exists = false,
havesysid = false;
int max_id;
List *item,
*option;
char *password = NULL; /* PostgreSQL user password */
bool encrypt_password = Password_encryption; /* encrypt password? */
char encrypted_password[MD5_PASSWD_LEN + 1];
int sysid = 0; /* PgSQL system id (valid if havesysid) */
bool createdb = false; /* Can the user create databases? */
bool createuser = false; /* Can this user create users? */
List *groupElts = NIL; /* The groups the user is a member of */
char *validUntil = NULL; /* The time the login is valid
* until */
DefElem *dpassword = NULL;
DefElem *dsysid = NULL;
DefElem *dcreatedb = NULL;
DefElem *dcreateuser = NULL;
DefElem *dgroupElts = NULL;
DefElem *dvalidUntil = NULL;
/* Extract options from the statement node tree */
foreach(option, stmt->options)
{
DefElem *defel = (DefElem *) lfirst(option);
if (strcmp(defel->defname, "password") == 0 ||
strcmp(defel->defname, "encryptedPassword") == 0 ||
strcmp(defel->defname, "unencryptedPassword") == 0)
{
if (dpassword)
elog(ERROR, "CREATE USER: conflicting options");
dpassword = defel;
if (strcmp(defel->defname, "encryptedPassword") == 0)
encrypt_password = true;
else if (strcmp(defel->defname, "unencryptedPassword") == 0)
encrypt_password = false;
}
else if (strcmp(defel->defname, "sysid") == 0)
{
if (dsysid)
elog(ERROR, "CREATE USER: conflicting options");
dsysid = defel;
}
else if (strcmp(defel->defname, "createdb") == 0)
{
if (dcreatedb)
elog(ERROR, "CREATE USER: conflicting options");
dcreatedb = defel;
}
else if (strcmp(defel->defname, "createuser") == 0)
{
if (dcreateuser)
elog(ERROR, "CREATE USER: conflicting options");
dcreateuser = defel;
}
else if (strcmp(defel->defname, "groupElts") == 0)
{
if (dgroupElts)
elog(ERROR, "CREATE USER: conflicting options");
dgroupElts = defel;
}
else if (strcmp(defel->defname, "validUntil") == 0)
{
if (dvalidUntil)
elog(ERROR, "CREATE USER: conflicting options");
dvalidUntil = defel;
}
else
elog(ERROR, "CREATE USER: option \"%s\" not recognized",
defel->defname);
}
if (dcreatedb)
createdb = intVal(dcreatedb->arg) != 0;
if (dcreateuser)
createuser = intVal(dcreateuser->arg) != 0;
if (dsysid)
{
sysid = intVal(dsysid->arg);
if (sysid <= 0)
elog(ERROR, "user id must be positive");
havesysid = true;
}
if (dvalidUntil)
validUntil = strVal(dvalidUntil->arg);
if (dpassword)
password = strVal(dpassword->arg);
if (dgroupElts)
groupElts = (List *) dgroupElts->arg;
/* Check some permissions first */
if (password)
CheckPgUserAclNotNull();
if (!superuser())
elog(ERROR, "CREATE USER: permission denied");
if (strcmp(stmt->user, "public") == 0)
elog(ERROR, "CREATE USER: user name \"%s\" is reserved",
stmt->user);
/*
* Scan the pg_shadow relation to be certain the user or id doesn't
* already exist. Note we secure exclusive lock, because we also need
* to be sure of what the next usesysid should be, and we need to
* protect our update of the flat password file.
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_shadow_rel = heap_openr(ShadowRelationName, ExclusiveLock);
1998-09-01 05:29:17 +02:00
pg_shadow_dsc = RelationGetDescr(pg_shadow_rel);
scan = heap_beginscan(pg_shadow_rel, SnapshotNow, 0, NULL);
max_id = 99; /* start auto-assigned ids at 100 */
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
while (!user_exists && !sysid_exists &&
(tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
Form_pg_shadow shadow_form = (Form_pg_shadow) GETSTRUCT(tuple);
int32 this_sysid;
user_exists = (strcmp(NameStr(shadow_form->usename), stmt->user) == 0);
this_sysid = shadow_form->usesysid;
if (havesysid) /* customized id wanted */
sysid_exists = (this_sysid == sysid);
else
{
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
/* pick 1 + max */
if (this_sysid > max_id)
max_id = this_sysid;
}
}
heap_endscan(scan);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
if (user_exists)
elog(ERROR, "CREATE USER: user name \"%s\" already exists",
stmt->user);
if (sysid_exists)
elog(ERROR, "CREATE USER: sysid %d is already assigned", sysid);
/* If no sysid given, use max existing id + 1 */
if (!havesysid)
sysid = max_id + 1;
/*
* Build a tuple to insert
*/
MemSet(new_record, 0, sizeof(new_record));
MemSet(new_record_nulls, ' ', sizeof(new_record_nulls));
new_record[Anum_pg_shadow_usename - 1] =
DirectFunctionCall1(namein, CStringGetDatum(stmt->user));
new_record[Anum_pg_shadow_usesysid - 1] = Int32GetDatum(sysid);
AssertState(BoolIsValid(createdb));
new_record[Anum_pg_shadow_usecreatedb - 1] = BoolGetDatum(createdb);
AssertState(BoolIsValid(createuser));
new_record[Anum_pg_shadow_usesuper - 1] = BoolGetDatum(createuser);
/* superuser gets catupd right by default */
new_record[Anum_pg_shadow_usecatupd - 1] = BoolGetDatum(createuser);
if (password)
{
if (!encrypt_password || isMD5(password))
new_record[Anum_pg_shadow_passwd - 1] =
DirectFunctionCall1(textin, CStringGetDatum(password));
else
{
2001-08-17 04:59:20 +02:00
if (!EncryptMD5(password, stmt->user, strlen(stmt->user),
encrypted_password))
elog(ERROR, "CREATE USER: password encryption failed");
new_record[Anum_pg_shadow_passwd - 1] =
DirectFunctionCall1(textin, CStringGetDatum(encrypted_password));
}
}
else
new_record_nulls[Anum_pg_shadow_passwd - 1] = 'n';
if (validUntil)
new_record[Anum_pg_shadow_valuntil - 1] =
DirectFunctionCall1(nabstimein, CStringGetDatum(validUntil));
else
new_record_nulls[Anum_pg_shadow_valuntil - 1] = 'n';
new_record_nulls[Anum_pg_shadow_useconfig - 1] = 'n';
tuple = heap_formtuple(pg_shadow_dsc, new_record, new_record_nulls);
/*
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
* Insert new record in the pg_shadow table
*/
simple_heap_insert(pg_shadow_rel, tuple);
/* Update indexes */
CatalogUpdateIndexes(pg_shadow_rel, tuple);
/*
* Add the user to the groups specified. We'll just call the below
* AlterGroup for this.
*/
foreach(item, groupElts)
{
AlterGroupStmt ags;
ags.name = strVal(lfirst(item)); /* the group name to add
* this in */
ags.action = +1;
ags.listUsers = makeList1(makeInteger(sysid));
AlterGroup(&ags, "CREATE USER");
}
/*
* Now we can clean up; but keep lock until commit.
*/
heap_close(pg_shadow_rel, NoLock);
/*
* Write the updated pg_shadow and pg_group data to the flat file.
*/
update_pg_pwd_and_pg_group(NULL);
}
/*
* ALTER USER
*/
void
AlterUser(AlterUserStmt *stmt)
{
Datum new_record[Natts_pg_shadow];
char new_record_nulls[Natts_pg_shadow];
char new_record_repl[Natts_pg_shadow];
Relation pg_shadow_rel;
TupleDesc pg_shadow_dsc;
HeapTuple tuple,
new_tuple;
List *option;
char *password = NULL; /* PostgreSQL user password */
bool encrypt_password = Password_encryption; /* encrypt password? */
char encrypted_password[MD5_PASSWD_LEN + 1];
int createdb = -1; /* Can the user create databases? */
int createuser = -1; /* Can this user create users? */
char *validUntil = NULL; /* The time the login is valid
* until */
DefElem *dpassword = NULL;
DefElem *dcreatedb = NULL;
DefElem *dcreateuser = NULL;
DefElem *dvalidUntil = NULL;
/* Extract options from the statement node tree */
foreach(option, stmt->options)
{
DefElem *defel = (DefElem *) lfirst(option);
if (strcmp(defel->defname, "password") == 0 ||
strcmp(defel->defname, "encryptedPassword") == 0 ||
strcmp(defel->defname, "unencryptedPassword") == 0)
{
if (dpassword)
elog(ERROR, "ALTER USER: conflicting options");
dpassword = defel;
if (strcmp(defel->defname, "encryptedPassword") == 0)
encrypt_password = true;
else if (strcmp(defel->defname, "unencryptedPassword") == 0)
encrypt_password = false;
}
else if (strcmp(defel->defname, "createdb") == 0)
{
if (dcreatedb)
elog(ERROR, "ALTER USER: conflicting options");
dcreatedb = defel;
}
else if (strcmp(defel->defname, "createuser") == 0)
{
if (dcreateuser)
elog(ERROR, "ALTER USER: conflicting options");
dcreateuser = defel;
}
else if (strcmp(defel->defname, "validUntil") == 0)
{
if (dvalidUntil)
elog(ERROR, "ALTER USER: conflicting options");
dvalidUntil = defel;
}
else
elog(ERROR, "ALTER USER: option \"%s\" not recognized",
defel->defname);
}
if (dcreatedb)
createdb = intVal(dcreatedb->arg);
if (dcreateuser)
createuser = intVal(dcreateuser->arg);
if (dvalidUntil)
validUntil = strVal(dvalidUntil->arg);
if (dpassword)
password = strVal(dpassword->arg);
if (password)
CheckPgUserAclNotNull();
/* must be superuser or just want to change your own password */
if (!superuser() &&
!(createdb < 0 &&
createuser < 0 &&
!validUntil &&
password &&
strcmp(GetUserNameFromId(GetUserId()), stmt->user) == 0))
elog(ERROR, "ALTER USER: permission denied");
/* changes to the flat password file cannot be rolled back */
if (IsTransactionBlock() && password)
elog(NOTICE, "ALTER USER: password changes cannot be rolled back");
/*
* Scan the pg_shadow relation to be certain the user exists. Note we
* secure exclusive lock to protect our update of the flat password
* file.
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_shadow_rel = heap_openr(ShadowRelationName, ExclusiveLock);
1998-09-01 05:29:17 +02:00
pg_shadow_dsc = RelationGetDescr(pg_shadow_rel);
tuple = SearchSysCache(SHADOWNAME,
PointerGetDatum(stmt->user),
0, 0, 0);
if (!HeapTupleIsValid(tuple))
elog(ERROR, "ALTER USER: user \"%s\" does not exist", stmt->user);
/*
* Build an updated tuple, perusing the information just obtained
*/
MemSet(new_record, 0, sizeof(new_record));
MemSet(new_record_nulls, ' ', sizeof(new_record_nulls));
MemSet(new_record_repl, ' ', sizeof(new_record_repl));
new_record[Anum_pg_shadow_usename - 1] = DirectFunctionCall1(namein,
2001-03-22 05:01:46 +01:00
CStringGetDatum(stmt->user));
new_record_repl[Anum_pg_shadow_usename - 1] = 'r';
/* createdb */
if (createdb >= 0)
{
new_record[Anum_pg_shadow_usecreatedb - 1] = BoolGetDatum(createdb > 0);
new_record_repl[Anum_pg_shadow_usecreatedb - 1] = 'r';
}
/*
* createuser (superuser) and catupd
*
* XXX It's rather unclear how to handle catupd. It's probably best to
* keep it equal to the superuser status, otherwise you could end up
* with a situation where no existing superuser can alter the
* catalogs, including pg_shadow!
*/
if (createuser >= 0)
{
new_record[Anum_pg_shadow_usesuper - 1] = BoolGetDatum(createuser > 0);
new_record_repl[Anum_pg_shadow_usesuper - 1] = 'r';
new_record[Anum_pg_shadow_usecatupd - 1] = BoolGetDatum(createuser > 0);
new_record_repl[Anum_pg_shadow_usecatupd - 1] = 'r';
}
/* password */
if (password)
{
if (!encrypt_password || isMD5(password))
new_record[Anum_pg_shadow_passwd - 1] =
DirectFunctionCall1(textin, CStringGetDatum(password));
else
{
2001-08-17 04:59:20 +02:00
if (!EncryptMD5(password, stmt->user, strlen(stmt->user),
encrypted_password))
elog(ERROR, "CREATE USER: password encryption failed");
new_record[Anum_pg_shadow_passwd - 1] =
DirectFunctionCall1(textin, CStringGetDatum(encrypted_password));
}
new_record_repl[Anum_pg_shadow_passwd - 1] = 'r';
}
/* valid until */
if (validUntil)
{
new_record[Anum_pg_shadow_valuntil - 1] =
DirectFunctionCall1(nabstimein, CStringGetDatum(validUntil));
new_record_repl[Anum_pg_shadow_valuntil - 1] = 'r';
}
new_tuple = heap_modifytuple(tuple, pg_shadow_rel, new_record,
new_record_nulls, new_record_repl);
simple_heap_update(pg_shadow_rel, &tuple->t_self, new_tuple);
/* Update indexes */
CatalogUpdateIndexes(pg_shadow_rel, new_tuple);
ReleaseSysCache(tuple);
heap_freetuple(new_tuple);
/*
* Now we can clean up.
*/
heap_close(pg_shadow_rel, NoLock);
/*
* Write the updated pg_shadow and pg_group data to the flat file.
*/
update_pg_pwd_and_pg_group(NULL);
}
/*
* ALTER USER ... SET
*/
void
AlterUserSet(AlterUserSetStmt *stmt)
{
char *valuestr;
HeapTuple oldtuple,
newtuple;
Relation rel;
Datum repl_val[Natts_pg_shadow];
char repl_null[Natts_pg_shadow];
char repl_repl[Natts_pg_shadow];
int i;
valuestr = flatten_set_variable_args(stmt->variable, stmt->value);
/*
2002-09-04 22:31:48 +02:00
* RowExclusiveLock is sufficient, because we don't need to update the
* flat password file.
*/
rel = heap_openr(ShadowRelationName, RowExclusiveLock);
oldtuple = SearchSysCache(SHADOWNAME,
PointerGetDatum(stmt->user),
0, 0, 0);
if (!HeapTupleIsValid(oldtuple))
elog(ERROR, "user \"%s\" does not exist", stmt->user);
if (!(superuser()
2002-09-04 22:31:48 +02:00
|| ((Form_pg_shadow) GETSTRUCT(oldtuple))->usesysid == GetUserId()))
elog(ERROR, "permission denied");
for (i = 0; i < Natts_pg_shadow; i++)
repl_repl[i] = ' ';
2002-09-04 22:31:48 +02:00
repl_repl[Anum_pg_shadow_useconfig - 1] = 'r';
if (strcmp(stmt->variable, "all") == 0 && valuestr == NULL)
/* RESET ALL */
2002-09-04 22:31:48 +02:00
repl_null[Anum_pg_shadow_useconfig - 1] = 'n';
else
{
2002-09-04 22:31:48 +02:00
Datum datum;
bool isnull;
ArrayType *array;
2002-09-04 22:31:48 +02:00
repl_null[Anum_pg_shadow_useconfig - 1] = ' ';
datum = SysCacheGetAttr(SHADOWNAME, oldtuple,
Anum_pg_shadow_useconfig, &isnull);
array = isnull ? ((ArrayType *) NULL) : DatumGetArrayTypeP(datum);
if (valuestr)
array = GUCArrayAdd(array, stmt->variable, valuestr);
else
array = GUCArrayDelete(array, stmt->variable);
2002-09-04 22:31:48 +02:00
repl_val[Anum_pg_shadow_useconfig - 1] = PointerGetDatum(array);
}
newtuple = heap_modifytuple(oldtuple, rel, repl_val, repl_null, repl_repl);
simple_heap_update(rel, &oldtuple->t_self, newtuple);
CatalogUpdateIndexes(rel, newtuple);
ReleaseSysCache(oldtuple);
heap_close(rel, RowExclusiveLock);
}
/*
* DROP USER
*/
void
DropUser(DropUserStmt *stmt)
{
Relation pg_shadow_rel;
TupleDesc pg_shadow_dsc;
List *item;
if (!superuser())
elog(ERROR, "DROP USER: permission denied");
if (IsTransactionBlock())
elog(NOTICE, "DROP USER cannot be rolled back completely");
/*
* Scan the pg_shadow relation to find the usesysid of the user to be
* deleted. Note we secure exclusive lock, because we need to protect
* our update of the flat password file.
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_shadow_rel = heap_openr(ShadowRelationName, ExclusiveLock);
pg_shadow_dsc = RelationGetDescr(pg_shadow_rel);
foreach(item, stmt->users)
{
const char *user = strVal(lfirst(item));
HeapTuple tuple,
tmp_tuple;
Relation pg_rel;
TupleDesc pg_dsc;
ScanKeyData scankey;
HeapScanDesc scan;
int32 usesysid;
tuple = SearchSysCache(SHADOWNAME,
PointerGetDatum(user),
0, 0, 0);
if (!HeapTupleIsValid(tuple))
2002-09-14 15:46:24 +02:00
{
if (length(stmt->users) > 1)
elog(ERROR, "DROP USER: user \"%s\" does not exist (no users removed)", user);
else
elog(ERROR, "DROP USER: user \"%s\" does not exist", user);
}
usesysid = ((Form_pg_shadow) GETSTRUCT(tuple))->usesysid;
if (usesysid == GetUserId())
elog(ERROR, "current user cannot be dropped");
if (usesysid == GetSessionUserId())
elog(ERROR, "session user cannot be dropped");
/*
* Check if user still owns a database. If so, error out.
*
* (It used to be that this function would drop the database
* automatically. This is not only very dangerous for people that
* don't read the manual, it doesn't seem to be the behaviour one
* would expect either.) -- petere 2000/01/14)
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_rel = heap_openr(DatabaseRelationName, AccessShareLock);
pg_dsc = RelationGetDescr(pg_rel);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
ScanKeyEntryInitialize(&scankey, 0x0,
Anum_pg_database_datdba, F_INT4EQ,
Int32GetDatum(usesysid));
scan = heap_beginscan(pg_rel, SnapshotNow, 1, &scankey);
if ((tmp_tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
char *dbname;
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
dbname = NameStr(((Form_pg_database) GETSTRUCT(tmp_tuple))->datname);
elog(ERROR, "DROP USER: user \"%s\" owns database \"%s\", cannot be removed%s",
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
user, dbname,
(length(stmt->users) > 1) ? " (no users removed)" : "");
}
heap_endscan(scan);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
heap_close(pg_rel, AccessShareLock);
/*
* Somehow we'd have to check for tables, views, etc. owned by the
* user as well, but those could be spread out over all sorts of
* databases which we don't have access to (easily).
*/
/*
* Remove the user from the pg_shadow table
*/
simple_heap_delete(pg_shadow_rel, &tuple->t_self);
ReleaseSysCache(tuple);
/*
* Remove user from groups
*
* try calling alter group drop user for every group
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_rel = heap_openr(GroupRelationName, ExclusiveLock);
pg_dsc = RelationGetDescr(pg_rel);
scan = heap_beginscan(pg_rel, SnapshotNow, 0, NULL);
while ((tmp_tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
AlterGroupStmt ags;
/* the group name from which to try to drop the user: */
ags.name = pstrdup(NameStr(((Form_pg_group) GETSTRUCT(tmp_tuple))->groname));
ags.action = -1;
ags.listUsers = makeList1(makeInteger(usesysid));
AlterGroup(&ags, "DROP USER");
}
heap_endscan(scan);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
heap_close(pg_rel, ExclusiveLock);
2001-03-22 05:01:46 +01:00
/*
* Advance command counter so that later iterations of this loop
* will see the changes already made. This is essential if, for
* example, we are trying to drop two users who are members of the
* same group --- the AlterGroup for the second user had better
* see the tuple updated from the first one.
*/
CommandCounterIncrement();
}
/*
* Now we can clean up.
*/
heap_close(pg_shadow_rel, NoLock);
/*
* Write the updated pg_shadow and pg_group data to the flat file.
*/
update_pg_pwd_and_pg_group(NULL);
}
/*
* CheckPgUserAclNotNull
*
* check to see if there is an ACL on pg_shadow
*/
static void
CheckPgUserAclNotNull(void)
{
HeapTuple htup;
htup = SearchSysCache(RELOID,
ObjectIdGetDatum(RelOid_pg_shadow),
0, 0, 0);
if (!HeapTupleIsValid(htup))
elog(ERROR, "CheckPgUserAclNotNull: \"%s\" not found",
ShadowRelationName);
if (heap_attisnull(htup, Anum_pg_class_relacl))
elog(ERROR,
"To use passwords, you have to revoke permissions on %s "
"so normal users cannot read the passwords. "
"Try 'REVOKE ALL ON \"%s\" FROM PUBLIC'.",
ShadowRelationName, ShadowRelationName);
ReleaseSysCache(htup);
}
/*
* CREATE GROUP
*/
void
CreateGroup(CreateGroupStmt *stmt)
{
Relation pg_group_rel;
HeapScanDesc scan;
HeapTuple tuple;
TupleDesc pg_group_dsc;
bool group_exists = false,
sysid_exists = false,
havesysid = false;
int max_id;
Datum new_record[Natts_pg_group];
char new_record_nulls[Natts_pg_group];
List *item,
*option,
*newlist = NIL;
2002-04-28 02:36:38 +02:00
IdList *grolist;
int sysid = 0;
List *userElts = NIL;
DefElem *dsysid = NULL;
DefElem *duserElts = NULL;
foreach(option, stmt->options)
{
DefElem *defel = (DefElem *) lfirst(option);
if (strcmp(defel->defname, "sysid") == 0)
{
if (dsysid)
elog(ERROR, "CREATE GROUP: conflicting options");
dsysid = defel;
}
else if (strcmp(defel->defname, "userElts") == 0)
{
if (duserElts)
elog(ERROR, "CREATE GROUP: conflicting options");
duserElts = defel;
}
else
elog(ERROR, "CREATE GROUP: option \"%s\" not recognized",
defel->defname);
}
if (dsysid)
{
sysid = intVal(dsysid->arg);
if (sysid <= 0)
elog(ERROR, "group id must be positive");
havesysid = true;
}
if (duserElts)
userElts = (List *) duserElts->arg;
/*
* Make sure the user can do this.
*/
if (!superuser())
elog(ERROR, "CREATE GROUP: permission denied");
if (strcmp(stmt->name, "public") == 0)
elog(ERROR, "CREATE GROUP: group name \"%s\" is reserved",
stmt->name);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_group_rel = heap_openr(GroupRelationName, ExclusiveLock);
pg_group_dsc = RelationGetDescr(pg_group_rel);
scan = heap_beginscan(pg_group_rel, SnapshotNow, 0, NULL);
max_id = 99; /* start auto-assigned ids at 100 */
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
while (!group_exists && !sysid_exists &&
(tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
Form_pg_group group_form = (Form_pg_group) GETSTRUCT(tuple);
int32 this_sysid;
group_exists = (strcmp(NameStr(group_form->groname), stmt->name) == 0);
this_sysid = group_form->grosysid;
if (havesysid) /* customized id wanted */
sysid_exists = (this_sysid == sysid);
else
{
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
/* pick 1 + max */
if (this_sysid > max_id)
max_id = this_sysid;
}
}
heap_endscan(scan);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
if (group_exists)
elog(ERROR, "CREATE GROUP: group name \"%s\" already exists",
stmt->name);
if (sysid_exists)
elog(ERROR, "CREATE GROUP: group sysid %d is already assigned",
sysid);
if (!havesysid)
sysid = max_id + 1;
/*
* Translate the given user names to ids
*/
foreach(item, userElts)
{
const char *groupuser = strVal(lfirst(item));
int32 userid = get_usesysid(groupuser);
if (!intMember(userid, newlist))
newlist = lappendi(newlist, userid);
}
/* build an array to insert */
if (newlist)
2002-04-28 02:36:38 +02:00
grolist = IdListToArray(newlist);
else
2002-04-28 02:36:38 +02:00
grolist = NULL;
/*
* Form a tuple to insert
*/
new_record[Anum_pg_group_groname - 1] =
DirectFunctionCall1(namein, CStringGetDatum(stmt->name));
new_record[Anum_pg_group_grosysid - 1] = Int32GetDatum(sysid);
2002-04-28 02:36:38 +02:00
new_record[Anum_pg_group_grolist - 1] = PointerGetDatum(grolist);
new_record_nulls[Anum_pg_group_groname - 1] = ' ';
new_record_nulls[Anum_pg_group_grosysid - 1] = ' ';
2002-04-28 02:36:38 +02:00
new_record_nulls[Anum_pg_group_grolist - 1] = grolist ? ' ' : 'n';
tuple = heap_formtuple(pg_group_dsc, new_record, new_record_nulls);
/*
* Insert a new record in the pg_group table
*/
simple_heap_insert(pg_group_rel, tuple);
/* Update indexes */
CatalogUpdateIndexes(pg_group_rel, tuple);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
heap_close(pg_group_rel, NoLock);
/*
* Write the updated pg_shadow and pg_group data to the flat file.
*/
update_pg_pwd_and_pg_group(NULL);
}
/*
* ALTER GROUP
*/
void
AlterGroup(AlterGroupStmt *stmt, const char *tag)
{
Relation pg_group_rel;
TupleDesc pg_group_dsc;
HeapTuple group_tuple;
IdList *oldarray;
Datum datum;
bool null;
List *newlist,
*item;
/*
* Make sure the user can do this.
*/
if (!superuser())
elog(ERROR, "%s: permission denied", tag);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_group_rel = heap_openr(GroupRelationName, ExclusiveLock);
pg_group_dsc = RelationGetDescr(pg_group_rel);
/*
* Fetch existing tuple for group.
*/
group_tuple = SearchSysCache(GRONAME,
PointerGetDatum(stmt->name),
0, 0, 0);
if (!HeapTupleIsValid(group_tuple))
elog(ERROR, "%s: group \"%s\" does not exist", tag, stmt->name);
/* Fetch old group membership. */
datum = heap_getattr(group_tuple, Anum_pg_group_grolist,
pg_group_dsc, &null);
oldarray = null ? ((IdList *) NULL) : DatumGetIdListP(datum);
/* initialize list with old array contents */
newlist = IdArrayToList(oldarray);
/*
* Now decide what to do.
*/
AssertState(stmt->action == +1 || stmt->action == -1);
if (stmt->action == +1) /* add users, might also be invoked by
* create user */
{
/*
2002-09-04 22:31:48 +02:00
* convert the to be added usernames to sysids and add them to the
* list
*/
foreach(item, stmt->listUsers)
{
2002-09-04 22:31:48 +02:00
int32 sysid;
if (strcmp(tag, "ALTER GROUP") == 0)
{
/* Get the uid of the proposed user to add. */
sysid = get_usesysid(strVal(lfirst(item)));
}
else if (strcmp(tag, "CREATE USER") == 0)
{
/*
* in this case we already know the uid and it wouldn't be
* in the cache anyway yet
*/
sysid = intVal(lfirst(item));
}
else
{
elog(ERROR, "AlterGroup: unknown tag %s", tag);
sysid = 0; /* keep compiler quiet */
}
if (!intMember(sysid, newlist))
newlist = lappendi(newlist, sysid);
else
2002-09-04 22:31:48 +02:00
/*
* we silently assume here that this error will only come
* up in a ALTER GROUP statement
*/
elog(WARNING, "%s: user \"%s\" is already in group \"%s\"",
tag, strVal(lfirst(item)), stmt->name);
}
/* Do the update */
UpdateGroupMembership(pg_group_rel, group_tuple, newlist);
} /* endif alter group add user */
else if (stmt->action == -1) /* drop users from group */
{
bool is_dropuser = strcmp(tag, "DROP USER") == 0;
if (newlist == NIL)
{
if (!is_dropuser)
elog(WARNING, "ALTER GROUP: group \"%s\" does not have any members", stmt->name);
}
else
{
/*
2002-09-04 22:31:48 +02:00
* convert the to be dropped usernames to sysids and remove
* them from the list
*/
foreach(item, stmt->listUsers)
{
int32 sysid;
if (!is_dropuser)
{
/* Get the uid of the proposed user to drop. */
sysid = get_usesysid(strVal(lfirst(item)));
}
else
{
/* for dropuser we already know the uid */
sysid = intVal(lfirst(item));
}
if (intMember(sysid, newlist))
newlist = lremovei(sysid, newlist);
else if (!is_dropuser)
elog(WARNING, "ALTER GROUP: user \"%s\" is not in group \"%s\"", strVal(lfirst(item)), stmt->name);
}
/* Do the update */
UpdateGroupMembership(pg_group_rel, group_tuple, newlist);
} /* endif group not null */
} /* endif alter group drop user */
ReleaseSysCache(group_tuple);
/*
* Write the updated pg_shadow and pg_group data to the flat files.
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
heap_close(pg_group_rel, NoLock);
/*
* Write the updated pg_shadow and pg_group data to the flat file.
*/
update_pg_pwd_and_pg_group(NULL);
}
/*
* Subroutine for AlterGroup: given a pg_group tuple and a desired new
* membership (expressed as an integer list), form and write an updated tuple.
* The pg_group relation must be open and locked already.
*/
static void
UpdateGroupMembership(Relation group_rel, HeapTuple group_tuple,
List *members)
{
IdList *newarray;
Datum new_record[Natts_pg_group];
char new_record_nulls[Natts_pg_group];
char new_record_repl[Natts_pg_group];
HeapTuple tuple;
newarray = IdListToArray(members);
/*
* Form an updated tuple with the new array and write it back.
*/
MemSet(new_record, 0, sizeof(new_record));
MemSet(new_record_nulls, ' ', sizeof(new_record_nulls));
MemSet(new_record_repl, ' ', sizeof(new_record_repl));
new_record[Anum_pg_group_grolist - 1] = PointerGetDatum(newarray);
new_record_repl[Anum_pg_group_grolist - 1] = 'r';
tuple = heap_modifytuple(group_tuple, group_rel,
2002-09-04 22:31:48 +02:00
new_record, new_record_nulls, new_record_repl);
simple_heap_update(group_rel, &group_tuple->t_self, tuple);
/* Update indexes */
CatalogUpdateIndexes(group_rel, tuple);
}
/*
* Convert an integer list of sysids to an array.
*/
static IdList *
IdListToArray(List *members)
{
int nmembers = length(members);
IdList *newarray;
List *item;
int i;
newarray = palloc(ARR_OVERHEAD(1) + nmembers * sizeof(int32));
newarray->size = ARR_OVERHEAD(1) + nmembers * sizeof(int32);
newarray->flags = 0;
newarray->elemtype = INT4OID;
ARR_NDIM(newarray) = 1; /* one dimensional array */
ARR_LBOUND(newarray)[0] = 1; /* axis starts at one */
2002-09-04 22:31:48 +02:00
ARR_DIMS(newarray)[0] = nmembers; /* axis is this long */
i = 0;
foreach(item, members)
((int *) ARR_DATA_PTR(newarray))[i++] = lfirsti(item);
return newarray;
}
/*
* Convert an array of sysids to an integer list.
*/
static List *
IdArrayToList(IdList *oldarray)
{
List *newlist = NIL;
int hibound,
i;
if (oldarray == NULL)
return NIL;
Assert(ARR_NDIM(oldarray) == 1);
Assert(ARR_ELEMTYPE(oldarray) == INT4OID);
hibound = ARR_DIMS(oldarray)[0];
for (i = 0; i < hibound; i++)
{
int32 sysid;
sysid = ((int32 *) ARR_DATA_PTR(oldarray))[i];
/* filter out any duplicates --- probably a waste of time */
if (!intMember(sysid, newlist))
newlist = lappendi(newlist, sysid);
}
return newlist;
}
/*
* DROP GROUP
*/
void
DropGroup(DropGroupStmt *stmt)
{
Relation pg_group_rel;
HeapTuple tuple;
/*
* Make sure the user can do this.
*/
if (!superuser())
elog(ERROR, "DROP GROUP: permission denied");
/*
* Drop the group.
*/
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
pg_group_rel = heap_openr(GroupRelationName, ExclusiveLock);
tuple = SearchSysCacheCopy(GRONAME,
PointerGetDatum(stmt->name),
0, 0, 0);
if (!HeapTupleIsValid(tuple))
elog(ERROR, "DROP GROUP: group \"%s\" does not exist", stmt->name);
simple_heap_delete(pg_group_rel, &tuple->t_self);
Clean up various to-do items associated with system indexes: pg_database now has unique indexes on oid and on datname. pg_shadow now has unique indexes on usename and on usesysid. pg_am now has unique index on oid. pg_opclass now has unique index on oid. pg_amproc now has unique index on amid+amopclaid+amprocnum. Remove pg_rewrite's unnecessary index on oid, delete unused RULEOID syscache. Remove index on pg_listener and associated syscache for performance reasons (caching rows that are certain to change before you need 'em again is rather pointless). Change pg_attrdef's nonunique index on adrelid into a unique index on adrelid+adnum. Fix various incorrect settings of pg_class.relisshared, make that the primary reference point for whether a relation is shared or not. IsSharedSystemRelationName() is now only consulted to initialize relisshared during initial creation of tables and indexes. In theory we might now support shared user relations, though it's not clear how one would get entries for them into pg_class &etc of multiple databases. Fix recently reported bug that pg_attribute rows created for an index all have the same OID. (Proof that non-unique OID doesn't matter unless it's actually used to do lookups ;-)) There's no need to treat pg_trigger, pg_attrdef, pg_relcheck as bootstrap relations. Convert them into plain system catalogs without hardwired entries in pg_class and friends. Unify global.bki and template1.bki into a single init script postgres.bki, since the alleged distinction between them was misleading and pointless. Not to mention that it didn't work for setting up indexes on shared system relations. Rationalize locking of pg_shadow, pg_group, pg_attrdef (no need to use AccessExclusiveLock where ExclusiveLock or even RowExclusiveLock will do). Also, hold locks until transaction commit where necessary.
2001-06-12 07:55:50 +02:00
heap_close(pg_group_rel, NoLock);
/*
* Write the updated pg_shadow and pg_group data to the flat file.
*/
update_pg_pwd_and_pg_group(NULL);
}