/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright 2008 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
#pragma ident "%Z%%M% %I% %E% SMI"
/*
* This file only contains the transaction commit logic.
*/
#include <assert.h>
#include <alloca.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <strings.h>
#include <sys/sysmacros.h>
#include "configd.h"
#define INVALID_OBJ_ID ((uint32_t)-1)
#define INVALID_TYPE ((uint32_t)-1)
struct tx_cmd {
const struct rep_protocol_transaction_cmd *tx_cmd;
const char *tx_prop;
uint32_t *tx_values;
uint32_t tx_nvalues;
uint32_t tx_orig_value_id;
char tx_found;
char tx_processed;
char tx_bad;
};
static int
tx_cmd_compare(const void *key, const void *elem_arg)
{
const struct tx_cmd *elem = elem_arg;
return (strcmp((const char *)key, elem->tx_prop));
}
struct tx_commit_data {
uint32_t txc_pg_id;
uint32_t txc_gen;
uint32_t txc_oldgen;
short txc_backend;
backend_tx_t *txc_tx;
backend_query_t *txc_inserts;
size_t txc_count;
rep_protocol_responseid_t txc_result;
struct tx_cmd txc_cmds[1]; /* actually txc_count */
};
#define TX_COMMIT_DATA_SIZE(count) \
offsetof(struct tx_commit_data, txc_cmds[count])
/*ARGSUSED*/
static int
tx_check_genid(void *data_arg, int columns, char **vals, char **names)
{
tx_commit_data_t *data = data_arg;
assert(columns == 1);
if (atoi(vals[0]) != data->txc_oldgen)
data->txc_result = REP_PROTOCOL_FAIL_NOT_LATEST;
else
data->txc_result = REP_PROTOCOL_SUCCESS;
return (BACKEND_CALLBACK_CONTINUE);
}
/*
* tx_process_property() is called once for each property in current
* property group generation. Its purpose is threefold:
*
* 1. copy properties not mentioned in the transaction over unchanged.
* 2. mark DELETEd properties as seen (they will be left out of the new
* generation).
* 3. consistancy-check NEW, CLEAR, and REPLACE commands.
*
* Any consistancy problems set tx_bad, and seen properties are marked
* tx_found. These is used later, in tx_process_cmds().
*/
/*ARGSUSED*/
static int
tx_process_property(void *data_arg, int columns, char **vals, char **names)
{
tx_commit_data_t *data = data_arg;
struct tx_cmd *elem;
const char *prop_name = vals[0];
const char *prop_type = vals[1];
const char *lnk_val_id = vals[2];
char *endptr;
assert(columns == 3);
elem = bsearch(prop_name, data->txc_cmds, data->txc_count,
sizeof (*data->txc_cmds), tx_cmd_compare);
if (elem == NULL) {
backend_query_add(data->txc_inserts,
"INSERT INTO prop_lnk_tbl"
" (lnk_pg_id, lnk_gen_id, lnk_prop_name, lnk_prop_type,"
" lnk_val_id) "
"VALUES ( %d, %d, '%q', '%q', %Q );",
data->txc_pg_id, data->txc_gen, prop_name, prop_type,
lnk_val_id);
} else {
assert(!elem->tx_found);
elem->tx_found = 1;
if (lnk_val_id != NULL) {
errno = 0;
elem->tx_orig_value_id =
strtoul(lnk_val_id, &endptr, 10);
if (elem->tx_orig_value_id == 0 || *endptr != 0 ||
errno != 0) {
return (BACKEND_CALLBACK_ABORT);
}
} else {
elem->tx_orig_value_id = 0;
}
switch (elem->tx_cmd->rptc_action) {
case REP_PROTOCOL_TX_ENTRY_NEW:
elem->tx_bad = 1;
data->txc_result = REP_PROTOCOL_FAIL_EXISTS;
break;
case REP_PROTOCOL_TX_ENTRY_CLEAR:
if (REP_PROTOCOL_BASE_TYPE(elem->tx_cmd->rptc_type) !=
prop_type[0] &&
REP_PROTOCOL_SUBTYPE(elem->tx_cmd->rptc_type) !=
prop_type[1]) {
elem->tx_bad = 1;
data->txc_result =
REP_PROTOCOL_FAIL_TYPE_MISMATCH;
}
break;
case REP_PROTOCOL_TX_ENTRY_REPLACE:
break;
case REP_PROTOCOL_TX_ENTRY_DELETE:
elem->tx_processed = 1;
break;
default:
assert(0);
break;
}
}
return (BACKEND_CALLBACK_CONTINUE);
}
/*
* tx_process_cmds() finishes the job tx_process_property() started:
*
* 1. if tx_process_property() marked a command as bad, we skip it.
* 2. if a DELETE, REPLACE, or CLEAR operated on a non-existant property,
* we mark it as bad.
* 3. we complete the work of NEW, REPLACE, and CLEAR, by inserting the
* appropriate values into the database.
* 4. we delete all replaced data, if it is no longer referenced.
*
* Finally, we check all of the commands, and fail if anything was marked bad.
*/
static int
tx_process_cmds(tx_commit_data_t *data)
{
int idx;
int r;
int count = data->txc_count;
struct tx_cmd *elem;
uint32_t val_id = 0;
uint8_t type[3];
backend_query_t *q;
int do_delete;
/*
* For persistent pgs, we use backend_fail_if_seen to abort the
* deletion if there is a snapshot using our current state.
*
* All of the deletions in this function are safe, since
* rc_tx_commit() guarantees that all the data is in-cache.
*/
q = backend_query_alloc();
if (data->txc_backend != BACKEND_TYPE_NONPERSIST) {
backend_query_add(q,
"SELECT 1 FROM snaplevel_lnk_tbl "
" WHERE (snaplvl_pg_id = %d AND snaplvl_gen_id = %d); ",
data->txc_pg_id, data->txc_oldgen);
}
backend_query_add(q,
"DELETE FROM prop_lnk_tbl"
" WHERE (lnk_pg_id = %d AND lnk_gen_id = %d)",
data->txc_pg_id, data->txc_oldgen);
r = backend_tx_run(data->txc_tx, q, backend_fail_if_seen, NULL);
backend_query_free(q);
if (r == REP_PROTOCOL_SUCCESS)
do_delete = 1;
else if (r == REP_PROTOCOL_DONE)
do_delete = 0; /* old gen_id is in use */
else
return (r);
for (idx = 0; idx < count; idx++) {
elem = &data->txc_cmds[idx];
if (elem->tx_bad)
continue;
switch (elem->tx_cmd->rptc_action) {
case REP_PROTOCOL_TX_ENTRY_DELETE:
case REP_PROTOCOL_TX_ENTRY_REPLACE:
case REP_PROTOCOL_TX_ENTRY_CLEAR:
if (!elem->tx_found) {
elem->tx_bad = 1;
continue;
}
break;
case REP_PROTOCOL_TX_ENTRY_NEW:
break;
default:
assert(0);
break;
}
if (do_delete &&
elem->tx_cmd->rptc_action != REP_PROTOCOL_TX_ENTRY_NEW &&
elem->tx_orig_value_id != 0) {
/*
* delete the old values, if they are not in use
*/
q = backend_query_alloc();
backend_query_add(q,
"SELECT 1 FROM prop_lnk_tbl "
" WHERE (lnk_val_id = %d); "
"DELETE FROM value_tbl"
" WHERE (value_id = %d)",
elem->tx_orig_value_id, elem->tx_orig_value_id);
r = backend_tx_run(data->txc_tx, q,
backend_fail_if_seen, NULL);
backend_query_free(q);
if (r != REP_PROTOCOL_SUCCESS && r != REP_PROTOCOL_DONE)
return (r);
}
if (elem->tx_cmd->rptc_action == REP_PROTOCOL_TX_ENTRY_DELETE)
continue; /* no further work to do */
type[0] = REP_PROTOCOL_BASE_TYPE(elem->tx_cmd->rptc_type);
type[1] = REP_PROTOCOL_SUBTYPE(elem->tx_cmd->rptc_type);
type[2] = 0;
if (elem->tx_nvalues == 0) {
r = backend_tx_run_update(data->txc_tx,
"INSERT INTO prop_lnk_tbl"
" (lnk_pg_id, lnk_gen_id, "
" lnk_prop_name, lnk_prop_type, lnk_val_id) "
"VALUES ( %d, %d, '%q', '%q', NULL );",
data->txc_pg_id, data->txc_gen, elem->tx_prop,
type);
} else {
uint32_t *v, i = 0;
const char *str;
val_id = backend_new_id(data->txc_tx, BACKEND_ID_VALUE);
if (val_id == 0)
return (REP_PROTOCOL_FAIL_NO_RESOURCES);
r = backend_tx_run_update(data->txc_tx,
"INSERT INTO prop_lnk_tbl "
" (lnk_pg_id, lnk_gen_id, "
" lnk_prop_name, lnk_prop_type, lnk_val_id) "
"VALUES ( %d, %d, '%q', '%q', %d );",
data->txc_pg_id, data->txc_gen, elem->tx_prop,
type, val_id);
v = elem->tx_values;
for (i = 0; i < elem->tx_nvalues; i++) {
str = (const char *)&v[1];
/*
* Update values in backend, imposing
* ordering via the value_order column.
* This ordering is then used in subseqent
* value retrieval operations. We can
* safely assume that the repository schema
* has been upgraded (and hence has the
* value_order column in value_tbl), since
* it is upgraded as soon as the repository
* is writable.
*/
r = backend_tx_run_update(data->txc_tx,
"INSERT INTO value_tbl (value_id, "
"value_type, value_value, "
"value_order) VALUES (%d, '%c', "
"'%q', '%d');\n",
val_id, elem->tx_cmd->rptc_type,
str, i);
if (r != REP_PROTOCOL_SUCCESS)
break;
/*LINTED alignment*/
v = (uint32_t *)((caddr_t)str + TX_SIZE(*v));
}
}
if (r != REP_PROTOCOL_SUCCESS)
return (REP_PROTOCOL_FAIL_UNKNOWN);
elem->tx_processed = 1;
}
for (idx = 0; idx < count; idx++) {
elem = &data->txc_cmds[idx];
if (elem->tx_bad)
return (REP_PROTOCOL_FAIL_BAD_TX);
}
return (REP_PROTOCOL_SUCCESS);
}
static boolean_t
check_string(uintptr_t loc, uint32_t len, uint32_t sz)
{
const char *ptr = (const char *)loc;
if (len == 0 || len > sz || ptr[len - 1] != 0 || strlen(ptr) != len - 1)
return (0);
return (1);
}
static int
tx_check_and_setup(tx_commit_data_t *data, const void *cmds_arg,
uint32_t count)
{
const struct rep_protocol_transaction_cmd *cmds;
struct tx_cmd *cur;
struct tx_cmd *prev = NULL;
uintptr_t loc;
uint32_t sz, len;
int idx;
loc = (uintptr_t)cmds_arg;
for (idx = 0; idx < count; idx++) {
cur = &data->txc_cmds[idx];
cmds = (struct rep_protocol_transaction_cmd *)loc;
cur->tx_cmd = cmds;
sz = cmds->rptc_size;
loc += REP_PROTOCOL_TRANSACTION_CMD_MIN_SIZE;
sz -= REP_PROTOCOL_TRANSACTION_CMD_MIN_SIZE;
len = cmds->rptc_name_len;
if (len <= 1 || !check_string(loc, len, sz)) {
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
}
cur->tx_prop = (const char *)loc;
len = TX_SIZE(len);
loc += len;
sz -= len;
cur->tx_nvalues = 0;
cur->tx_values = (uint32_t *)loc;
while (sz > 0) {
if (sz < sizeof (uint32_t))
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
cur->tx_nvalues++;
len = *(uint32_t *)loc;
loc += sizeof (uint32_t);
sz -= sizeof (uint32_t);
if (!check_string(loc, len, sz))
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
/*
* XXX here, we should be checking that the values
* match the purported type
*/
len = TX_SIZE(len);
if (len > sz)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
loc += len;
sz -= len;
}
if (prev != NULL && strcmp(prev->tx_prop, cur->tx_prop) >= 0)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
prev = cur;
}
return (REP_PROTOCOL_SUCCESS);
}
/*
* Free the memory associated with a tx_commit_data structure.
*/
void
tx_commit_data_free(tx_commit_data_t *tx_data)
{
uu_free(tx_data);
}
/*
* Parse the data of a REP_PROTOCOL_PROPERTYGRP_TX_COMMIT message into a
* more useful form. The data in the message will be represented by a
* tx_commit_data_t structure which is allocated by this function. The
* address of the allocated structure is returned to *tx_data and must be
* freed by calling tx_commit_data_free().
*
* Parameters:
* cmds_arg Address of the commands in the
* REP_PROTOCOL_PROPERTYGRP_TX_COMMIT message.
*
* cmds_sz Number of message bytes at cmds_arg.
*
* tx_data Points to the place to receive the address of the
* allocated memory.
*
* Fails with
* _BAD_REQUEST
* _NO_RESOURCES
*/
int
tx_commit_data_new(const void *cmds_arg, size_t cmds_sz,
tx_commit_data_t **tx_data)
{
const struct rep_protocol_transaction_cmd *cmds;
tx_commit_data_t *data;
uintptr_t loc;
uint32_t count;
uint32_t sz;
int ret;
/*
* First, verify that the reported sizes make sense, and count
* the number of commands.
*/
count = 0;
loc = (uintptr_t)cmds_arg;
while (cmds_sz > 0) {
cmds = (struct rep_protocol_transaction_cmd *)loc;
if (cmds_sz <= REP_PROTOCOL_TRANSACTION_CMD_MIN_SIZE)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
sz = cmds->rptc_size;
if (sz <= REP_PROTOCOL_TRANSACTION_CMD_MIN_SIZE)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
sz = TX_SIZE(sz);
if (sz > cmds_sz)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
loc += sz;
cmds_sz -= sz;
count++;
}
data = uu_zalloc(TX_COMMIT_DATA_SIZE(count));
if (data == NULL)
return (REP_PROTOCOL_FAIL_NO_RESOURCES);
/*
* verify that everything looks okay, and set up our command
* datastructures.
*/
data->txc_count = count;
ret = tx_check_and_setup(data, cmds_arg, count);
if (ret == REP_PROTOCOL_SUCCESS) {
*tx_data = data;
} else {
*tx_data = NULL;
uu_free(data);
}
return (ret);
}
/*
* The following are a set of accessor functions to retrieve data from a
* tx_commit_data_t that has been allocated by tx_commit_data_new().
*/
/*
* Return the action of the transaction command whose command number is
* cmd_no. The action is placed at *action.
*
* Returns:
* _FAIL_BAD_REQUEST cmd_no is out of range.
*/
int
tx_cmd_action(tx_commit_data_t *tx_data, size_t cmd_no,
enum rep_protocol_transaction_action *action)
{
struct tx_cmd *cur;
assert(cmd_no < tx_data->txc_count);
if (cmd_no >= tx_data->txc_count)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
cur = &tx_data->txc_cmds[cmd_no];
*action = cur->tx_cmd->rptc_action;
return (REP_PROTOCOL_SUCCESS);
}
/*
* Return the number of transaction commands held in tx_data.
*/
size_t
tx_cmd_count(tx_commit_data_t *tx_data)
{
return (tx_data->txc_count);
}
/*
* Return the number of property values that are associated with the
* transaction command whose number is cmd_no. The number of values is
* returned to *nvalues.
*
* Returns:
* _FAIL_BAD_REQUEST cmd_no is out of range.
*/
int
tx_cmd_nvalues(tx_commit_data_t *tx_data, size_t cmd_no, uint32_t *nvalues)
{
struct tx_cmd *cur;
assert(cmd_no < tx_data->txc_count);
if (cmd_no >= tx_data->txc_count)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
cur = &tx_data->txc_cmds[cmd_no];
*nvalues = cur->tx_nvalues;
return (REP_PROTOCOL_SUCCESS);
}
/*
* Return a pointer to the property name of the command whose number is
* cmd_no. The property name pointer is returned to *pname.
*
* Returns:
* _FAIL_BAD_REQUEST cmd_no is out of range.
*/
int
tx_cmd_prop(tx_commit_data_t *tx_data, size_t cmd_no, const char **pname)
{
struct tx_cmd *cur;
assert(cmd_no < tx_data->txc_count);
if (cmd_no >= tx_data->txc_count)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
cur = &tx_data->txc_cmds[cmd_no];
*pname = cur->tx_prop;
return (REP_PROTOCOL_SUCCESS);
}
/*
* Return the property type of the property whose command number is
* cmd_no. The property type is returned to *ptype.
*
* Returns:
* _FAIL_BAD_REQUEST cmd_no is out of range.
*/
int
tx_cmd_prop_type(tx_commit_data_t *tx_data, size_t cmd_no, uint32_t *ptype)
{
struct tx_cmd *cur;
assert(cmd_no < tx_data->txc_count);
if (cmd_no >= tx_data->txc_count)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
cur = &tx_data->txc_cmds[cmd_no];
*ptype = cur->tx_cmd->rptc_type;
return (REP_PROTOCOL_SUCCESS);
}
/*
* This function is used to retrieve a property value from the transaction
* data. val_no specifies which value is to be retrieved from the
* transaction command whose number is cmd_no. A pointer to the specified
* value is placed in *val.
*
* Returns:
* _FAIL_BAD_REQUEST cmd_no or val_no is out of range.
*/
int
tx_cmd_value(tx_commit_data_t *tx_data, size_t cmd_no, uint32_t val_no,
const char **val)
{
const char *bp;
struct tx_cmd *cur;
uint32_t i;
uint32_t value_len;
assert(cmd_no < tx_data->txc_count);
if (cmd_no >= tx_data->txc_count)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
cur = &tx_data->txc_cmds[cmd_no];
assert(val_no < cur->tx_nvalues);
if (val_no >= cur->tx_nvalues)
return (REP_PROTOCOL_FAIL_BAD_REQUEST);
/* Find the correct value */
bp = (char *)cur->tx_values;
for (i = 0; i < val_no; i++) {
/* LINTED alignment */
value_len = *(uint32_t *)bp;
bp += sizeof (uint32_t) + TX_SIZE(value_len);
}
/* Bypass the count & return pointer to value. */
bp += sizeof (uint32_t);
*val = bp;
return (REP_PROTOCOL_SUCCESS);
}
int
object_tx_commit(rc_node_lookup_t *lp, tx_commit_data_t *data, uint32_t *gen)
{
uint32_t new_gen;
int ret;
rep_protocol_responseid_t r;
backend_tx_t *tx;
backend_query_t *q;
int backend = lp->rl_backend;
ret = backend_tx_begin(backend, &tx);
if (ret != REP_PROTOCOL_SUCCESS)
return (ret);
/* Make sure the pg is up-to-date. */
data->txc_oldgen = *gen;
data->txc_backend = backend;
data->txc_result = REP_PROTOCOL_FAIL_NOT_FOUND;
q = backend_query_alloc();
backend_query_add(q, "SELECT pg_gen_id FROM pg_tbl WHERE (pg_id = %d);",
lp->rl_main_id);
r = backend_tx_run(tx, q, tx_check_genid, data);
backend_query_free(q);
if (r != REP_PROTOCOL_SUCCESS ||
(r = data->txc_result) != REP_PROTOCOL_SUCCESS) {
backend_tx_rollback(tx);
goto end;
}
/* If the transaction is empty, cut out early. */
if (data->txc_count == 0) {
backend_tx_rollback(tx);
r = REP_PROTOCOL_DONE;
goto end;
}
new_gen = backend_new_id(tx, BACKEND_ID_GENERATION);
if (new_gen == 0) {
backend_tx_rollback(tx);
return (REP_PROTOCOL_FAIL_NO_RESOURCES);
}
data->txc_pg_id = lp->rl_main_id;
data->txc_gen = new_gen;
data->txc_tx = tx;
r = backend_tx_run_update(tx,
"UPDATE pg_tbl SET pg_gen_id = %d "
" WHERE (pg_id = %d AND pg_gen_id = %d);",
new_gen, lp->rl_main_id, *gen);
if (r != REP_PROTOCOL_SUCCESS) {
backend_tx_rollback(tx);
goto end;
}
q = backend_query_alloc();
backend_query_add(q,
"SELECT lnk_prop_name, lnk_prop_type, lnk_val_id "
"FROM prop_lnk_tbl "
"WHERE (lnk_pg_id = %d AND lnk_gen_id = %d)",
lp->rl_main_id, *gen);
data->txc_inserts = backend_query_alloc();
r = backend_tx_run(tx, q, tx_process_property, data);
backend_query_free(q);
if (r == REP_PROTOCOL_DONE)
r = REP_PROTOCOL_FAIL_UNKNOWN; /* corruption */
if (r != REP_PROTOCOL_SUCCESS ||
(r = data->txc_result) != REP_PROTOCOL_SUCCESS) {
backend_query_free(data->txc_inserts);
backend_tx_rollback(tx);
goto end;
}
r = backend_tx_run(tx, data->txc_inserts, NULL, NULL);
backend_query_free(data->txc_inserts);
if (r != REP_PROTOCOL_SUCCESS) {
backend_tx_rollback(tx);
goto end;
}
r = tx_process_cmds(data);
if (r != REP_PROTOCOL_SUCCESS) {
backend_tx_rollback(tx);
goto end;
}
r = backend_tx_commit(tx);
if (r == REP_PROTOCOL_SUCCESS)
*gen = new_gen;
end:
return (r);
}