int updhelper_reader(void)
{
uint4 pre_read_offset;
int lcnt;
boolean_t continue_reading;
call_on_signal = updhelper_reader_sigstop;
updhelper_init(UPD_HELPER_READER);
repl_log(updhelper_log_fp, TRUE, TRUE, "Helper reader started. PID %d [0x%X]\n", process_id, process_id);
GVKEY_INIT(gv_currkey, gv_keysize);
GVKEY_INIT(gv_altkey, gv_keysize);
last_pre_read_offset = 0;
continue_reading = TRUE;
do
{
REPL_DPRINT1("Will wait\n");
for (lcnt = 1; continue_reading && last_pre_read_offset ==
(pre_read_offset = recvpool.upd_helper_ctl->pre_read_offset); )
{
SHORT_SLEEP(lcnt);
if (lcnt <= MAX_LCNT)
lcnt++;
continue_reading = NO_SHUTDOWN == helper_entry->helper_shutdown;
}
last_pre_read_offset = pre_read_offset;
} while (continue_reading && updproc_preread());
updhelper_reader_end();
return SS_NORMAL;
}
/* Hibernate by starting a timer and waiting for it or any other timer to pop. */
void hiber_start_wait_any(uint4 hiber)
{
sigset_t savemask;
if (1000 > hiber)
{
SHORT_SLEEP(hiber); /* note: some platforms call hiber_start */
return;
}
if (1 <= timer_stack_count) /* timer services are unavailable from within a timer handler */
GTMASSERT;
sigprocmask(SIG_BLOCK, &blockalrm, &savemask); /* block SIGALRM signal and set new timer */
start_timer_int((TID)hiber_start_wait_any, hiber, NULL, 0, NULL);
sigsuspend(&savemask); /* unblock SIGALRM and wait for timer interrupt */
cancel_timer((TID)hiber_start_wait_any); /* cancel timer block before reenabling */
sigprocmask(SIG_SETMASK, &savemask, NULL); /* reset signal handlers */
}
int gtmrecv_endupd(void)
{
pid_t savepid;
int exit_status;
pid_t waitpid_res;
repl_log(stdout, TRUE, TRUE, "Initiating shut down of Update Process\n");
recvpool.upd_proc_local->upd_proc_shutdown = SHUTDOWN;
/* Wait for update process to shut down */
while((SHUTDOWN == recvpool.upd_proc_local->upd_proc_shutdown)
&& (0 < (savepid = (pid_t)recvpool.upd_proc_local->upd_proc_pid)) && is_proc_alive(savepid, 0))
{
SHORT_SLEEP(GTMRECV_WAIT_FOR_UPD_SHUTDOWN);
WAITPID(savepid, &exit_status, WNOHANG, waitpid_res); /* Release defunct update process if dead */
}
exit_status = recvpool.upd_proc_local->upd_proc_shutdown;
if (SHUTDOWN == exit_status)
{
if (0 == savepid) /* No Update Process */
exit_status = NORMAL_SHUTDOWN;
else /* Update Process Crashed */
{
repl_log(stderr, TRUE, TRUE, "Update Process exited abnormally, INTEGRITY CHECK might be warranted\n");
exit_status = ABNORMAL_SHUTDOWN;
}
}
/* Wait for the Update Process to detach */
if (0 == grab_sem(RECV, UPD_PROC_COUNT_SEM))
{
if(0 != (errno = rel_sem(RECV, UPD_PROC_COUNT_SEM)))
repl_log(stderr, TRUE, TRUE, "Error releasing the Update Process Count semaphore : %s\n", REPL_SEM_ERROR);
repl_log(stdout, TRUE, TRUE, "Update Process exited\n");
} else
{
repl_log(stderr, TRUE, TRUE, "Error in update proc count semaphore : %s\n", REPL_SEM_ERROR);
exit_status = ABNORMAL_SHUTDOWN;
}
return (exit_status);
}
int gtmrecv_start_helpers(int n_readers, int n_writers)
{ /* Set flag in recvpool telling the receiver server to start n_readers and n_writers helper processes.
* Wait for receiver server to complete the process - completed successfully, or terminated with error
*/
upd_helper_ctl_ptr_t upd_helper_ctl;
upd_helper_entry_ptr_t helper, helper_top;
char err_str[BUFSIZ];
int avail_slots, started_readers, started_writers;
error_def(ERR_REPLERR);
error_def(ERR_REPLINFO);
error_def(ERR_REPLWARN);
assert(0 != n_readers || 0 != n_writers);
upd_helper_ctl = recvpool.upd_helper_ctl;
/* let's clean up dead helpers first so we get an accurate count of available slots */
upd_helper_ctl->reap_helpers = HELPER_REAP_NOWAIT;
while (HELPER_REAP_NONE != upd_helper_ctl->reap_helpers && SRV_ALIVE == is_recv_srv_alive())
SHORT_SLEEP(GTMRECV_WAIT_FOR_UPD_SHUTDOWN);
upd_helper_ctl->reap_helpers = HELPER_REAP_NONE; /* just in case recvr died */
/* count available slots so receiver doesn't have to */
for (avail_slots = 0, helper = upd_helper_ctl->helper_list, helper_top = helper + MAX_UPD_HELPERS;
helper < helper_top; helper++)
{
if (0 == helper->helper_pid)
{
avail_slots++;
helper->helper_pid_prev = 0; /* force out abnormally terminated helpers as well */
helper->helper_shutdown = NO_SHUTDOWN; /* clean state */
}
}
if (avail_slots < n_readers + n_writers)
{
SNPRINTF(err_str, SIZEOF(err_str),
"%d helpers will exceed the maximum allowed (%d), limit the helpers to %d\n",
n_readers + n_writers, MAX_UPD_HELPERS, avail_slots);
gtm_putmsg(VARLSTCNT(4) ERR_REPLERR, 2, LEN_AND_STR(err_str));
return ABNORMAL_SHUTDOWN;
}
upd_helper_ctl->start_n_readers = n_readers;
upd_helper_ctl->start_n_writers = n_writers;
SHM_WRITE_MEMORY_BARRIER;
upd_helper_ctl->start_helpers = TRUE; /* hey receiver, let's go, start 'em up */
while (upd_helper_ctl->start_helpers && SRV_ALIVE == is_recv_srv_alive())
SHORT_SLEEP(GTMRECV_WAIT_FOR_SRV_START);
if (!upd_helper_ctl->start_helpers)
{
started_readers = upd_helper_ctl->start_n_readers;
started_writers = upd_helper_ctl->start_n_writers;
SNPRINTF(err_str, SIZEOF(err_str), "%s %d out of %d readers and %d out of %d writers started",
((started_readers + started_writers) == (n_readers + n_writers)) ? "All" : "Only",
started_readers, n_readers, started_writers, n_writers);
if ((started_readers + started_writers) == (n_readers + n_writers))
{
gtm_putmsg(VARLSTCNT(4) ERR_REPLINFO, 2, LEN_AND_STR(err_str));
return NORMAL_SHUTDOWN;
}
gtm_putmsg(VARLSTCNT(4) ERR_REPLWARN, 2, LEN_AND_STR(err_str));
return ABNORMAL_SHUTDOWN;
}
gtm_putmsg(VARLSTCNT(4) ERR_REPLERR, 2,
LEN_AND_LIT("Receiver server is not alive to start helpers. Start receiver server first"));
return ABNORMAL_SHUTDOWN;
}
int gtmrecv_upd_proc_init(boolean_t fresh_start)
{
/* Update Process initialization */
mstr upd_proc_log_cmd, upd_proc_trans_cmd;
char upd_proc_cmd[UPDPROC_CMD_MAXLEN];
int status;
int upd_status, save_upd_status;
#ifdef UNIX
pid_t upd_pid, waitpid_res;
#elif defined(VMS)
uint4 upd_pid;
uint4 cmd_channel;
$DESCRIPTOR(cmd_desc, UPDPROC_CMD_STR);
#endif
/* Check if the update process is alive */
if ((upd_status = is_updproc_alive()) == SRV_ERR)
{
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Receive pool semctl failure"), REPL_SEM_ERRNO);
repl_errno = EREPL_UPDSTART_SEMCTL;
return(UPDPROC_START_ERR);
} else if (upd_status == SRV_ALIVE && !fresh_start)
{
gtm_putmsg(VARLSTCNT(4) ERR_TEXT, 2, RTS_ERROR_LITERAL("Update process already exists. Not starting it"));
return(UPDPROC_EXISTS);
} else if (upd_status == SRV_ALIVE)
{
gtm_putmsg(VARLSTCNT(6) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Update process already exists. Please kill it before a fresh start"));
return(UPDPROC_EXISTS);
}
save_upd_status = recvpool.upd_proc_local->upd_proc_shutdown;
recvpool.upd_proc_local->upd_proc_shutdown = NO_SHUTDOWN;
#ifdef UNIX
if (0 > (upd_pid = fork())) /* BYPASSOK: we exec immediately, no FORK_CLEAN needed */
{
recvpool.upd_proc_local->upd_proc_shutdown = save_upd_status;
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Could not fork update process"), errno);
repl_errno = EREPL_UPDSTART_FORK;
return(UPDPROC_START_ERR);
}
if (0 == upd_pid)
{
/* Update Process */
upd_proc_log_cmd.len = SIZEOF(UPDPROC_CMD) - 1;
upd_proc_log_cmd.addr = UPDPROC_CMD;
status = TRANS_LOG_NAME(&upd_proc_log_cmd, &upd_proc_trans_cmd, upd_proc_cmd, SIZEOF(upd_proc_cmd),
dont_sendmsg_on_log2long);
if (status != SS_NORMAL)
{
gtm_putmsg(VARLSTCNT(6) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Could not find path of Update Process. Check value of $gtm_dist"));
if (SS_LOG2LONG == status)
gtm_putmsg(VARLSTCNT(5) ERR_LOGTOOLONG, 3, LEN_AND_LIT(UPDPROC_CMD), SIZEOF(upd_proc_cmd) - 1);
repl_errno = EREPL_UPDSTART_BADPATH;
return(UPDPROC_START_ERR);
}
upd_proc_cmd[upd_proc_trans_cmd.len] = '\0';
if (EXECL(upd_proc_cmd, upd_proc_cmd, UPDPROC_CMD_ARG1, UPDPROC_CMD_ARG2, NULL) < 0)
{
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Could not exec Update Process"), errno);
repl_errno = EREPL_UPDSTART_EXEC;
return(UPDPROC_START_ERR);
}
}
#elif defined(VMS)
/* Create detached server and write startup commands to it */
status = repl_create_server(&cmd_desc, "GTMU", "", &cmd_channel, &upd_pid, ERR_RECVPOOLSETUP);
if (SS_NORMAL != status)
{
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Unable to spawn Update process"), status);
recvpool.upd_proc_local->upd_proc_shutdown = save_upd_status;
repl_errno = EREPL_UPDSTART_FORK;
return(UPDPROC_START_ERR);
}
#endif
if (recvpool.upd_proc_local->upd_proc_pid)
recvpool.upd_proc_local->upd_proc_pid_prev = recvpool.upd_proc_local->upd_proc_pid;
else
recvpool.upd_proc_local->upd_proc_pid_prev = upd_pid;
recvpool.upd_proc_local->upd_proc_pid = upd_pid;
/* Receiver Server; wait for the update process to startup */
REPL_DPRINT2("Waiting for update process %d to startup\n", upd_pid);
while (get_sem_info(RECV, UPD_PROC_COUNT_SEM, SEM_INFO_VAL) == 0 && is_proc_alive(upd_pid, 0))
{
/* To take care of reassignment of PIDs, the while condition should be && with the
* condition (PPID of pid == process_id)
*/
REPL_DPRINT2("Waiting for update process %d to startup\n", upd_pid);
UNIX_ONLY(WAITPID(upd_pid, &status, WNOHANG, waitpid_res);) /* Release defunct update process if dead */
SHORT_SLEEP(GTMRECV_WAIT_FOR_SRV_START);
}
short iotcp_open(io_log_name *dev, mval *pp, int file_des, mval *mspace, int4 timeout)
{
boolean_t no_time_left = FALSE, timed;
char addr[SA_MAXLEN+1], *errptr, sockaddr[SA_MAXLEN+1],
temp_addr[SA_MAXLEN+1], temp_ch;
unsigned char ch, len;
int4 length, width;
unsigned short port;
int4 errlen, msec_timeout;
int ii, status, size,
on = 1,
p_offset = 0,
temp_1 = -2;
TID timer_id;
ABS_TIME cur_time, end_time, time_for_read, save_time_for_read;
d_tcp_struct *tcpptr, newtcp;
io_desc *ioptr;
struct sockaddr_in peer; /* socket address + port */
fd_set tcp_fd;
int lsock;
error_def(ERR_DEVPARMNEG);
error_def(ERR_INVADDRSPEC);
error_def(ERR_INVPORTSPEC);
error_def(ERR_IPADDRREQ);
error_def(ERR_OPENCONN);
error_def(ERR_SOCKACPT);
error_def(ERR_SOCKINIT);
error_def(ERR_SOCKPARMREQ);
error_def(ERR_SOCKWAIT);
error_def(ERR_TEXT);
#ifdef DEBUG_TCP
PRINTF("iotcp_open.c >>> tt = %d\n", t);
#endif
ioptr = dev->iod;
assert((params) *(pp->str.addr + p_offset) < (unsigned char)n_iops);
assert(0 != ioptr);
assert(ioptr->state >= 0 && ioptr->state < n_io_dev_states);
assert(tcp == ioptr->type);
if (dev_never_opened == ioptr->state)
{
ioptr->dev_sp = (void *)malloc(sizeof(d_tcp_struct));
memset(ioptr->dev_sp, 0, sizeof(d_tcp_struct));
}
tcpptr = (d_tcp_struct *)ioptr->dev_sp;
if (dev_never_opened == ioptr->state)
{
ioptr->state = dev_closed;
ioptr->width = TCPDEF_WIDTH;
ioptr->length = TCPDEF_LENGTH;
ioptr->wrap = TRUE;
if (-1 == iotcp_fillroutine())
assert(FALSE);
}
ioptr->dollar.zeof = FALSE;
newtcp = *tcpptr;
memcpy(newtcp.dollar_device, LITZERO, sizeof(LITZERO));
newtcp.passive = FALSE;
while (iop_eol != *(pp->str.addr + p_offset))
{
switch (ch = *(pp->str.addr + p_offset++))
{
case iop_width:
GET_LONG(width, pp->str.addr + p_offset);
if (0 == width)
newtcp.width = TCPDEF_WIDTH;
else if (width > 0)
newtcp.width = width;
else
rts_error(VARLSTCNT(1) ERR_DEVPARMNEG);
break;
case iop_length:
GET_LONG(length, pp->str.addr + p_offset);
if (0 == length)
newtcp.length = TCPDEF_LENGTH;
else if (length > 0)
newtcp.length = length;
else
rts_error(VARLSTCNT(1) ERR_DEVPARMNEG);
break;
case iop_listen:
newtcp.passive = TRUE;
break;
case iop_socket:
len = *(pp->str.addr + p_offset);
memset(sockaddr, 0, SA_MAXLEN+1);
memcpy(sockaddr, pp->str.addr + p_offset + 1, (len <= SA_MAXLEN) ? len : SA_MAXLEN);
*temp_addr = '\0';
*addr = '\0';
port = 0;
if (SSCANF(sockaddr, "%[^,], %hu", temp_addr, &port) < 2)
{
newtcp.sin.sin_addr.s_addr = INADDR_ANY;
if (SSCANF(sockaddr, ",%hu", &port) < 1)
{
rts_error(VARLSTCNT(1) ERR_INVPORTSPEC);
return FALSE;
}
} else
{
ii = 0;
temp_ch = temp_addr[0];
while(ISDIGIT(temp_ch) || ('.' == temp_ch))
{
ii++;
temp_ch = temp_addr[ii];
}
if ('\0' != temp_ch)
SPRINTF(addr, "%s", iotcp_name2ip(temp_addr));
else
SPRINTF(addr, "%s", temp_addr);
if ((in_addr_t)-1 == (newtcp.sin.sin_addr.s_addr = tcp_routines.aa_inet_addr(addr)))
{
rts_error(VARLSTCNT(1) ERR_INVADDRSPEC);
return FALSE;
}
}
newtcp.sin.sin_port = GTM_HTONS(port);
newtcp.sin.sin_family = AF_INET;
break;
case iop_exception:
ioptr->error_handler.len = *(pp->str.addr + p_offset);
ioptr->error_handler.addr = (char *)(pp->str.addr + p_offset + 1);
s2pool(&ioptr->error_handler);
break;
default:
break;
}
p_offset += ((IOP_VAR_SIZE == io_params_size[ch]) ?
(unsigned char)*(pp->str.addr + p_offset) + 1 : io_params_size[ch]);
}
if ((0 == newtcp.sin.sin_port) && (0 == newtcp.sin.sin_addr.s_addr))
{
rts_error(VARLSTCNT(1) ERR_SOCKPARMREQ);
return FALSE;
}
/* active connection must have a complete address specification */
if ((INADDR_ANY == newtcp.sin.sin_addr.s_addr) && !newtcp.passive)
{
rts_error(VARLSTCNT(1) ERR_IPADDRREQ);
return FALSE;
}
if (dev_closed == ioptr->state)
{
if (newtcp.passive) /* passive connection */
{
/* no listening socket for this addr? make one. */
memcpy(ioptr->dev_sp, &newtcp, sizeof(d_tcp_struct));
if (!(lsock = iotcp_getlsock(dev)))
return FALSE; /* could not create listening socket */
timer_id = (TID)iotcp_open;
out_of_time = FALSE;
time_for_read.at_sec = ((NO_M_TIMEOUT == timeout) ? 0 : 1);
time_for_read.at_usec = 0;
if (NO_M_TIMEOUT == timeout)
{
timed = FALSE;
msec_timeout = NO_M_TIMEOUT;
} else
{
timed = TRUE;
msec_timeout = timeout2msec(timeout);
if (msec_timeout > 0)
{ /* there is time to wait */
sys_get_curr_time(&cur_time);
add_int_to_abs_time(&cur_time, msec_timeout, &end_time);
start_timer(timer_id, msec_timeout, wake_alarm, 0, NULL);
} else
out_of_time = TRUE;
}
for (status = 0; 0 == status; )
{
FD_ZERO(&tcp_fd);
FD_SET(lsock, &tcp_fd);
/*
* Note: the check for EINTR from the select below should remain, as aa_select is a
* function, and not all callers of aa_select behave the same when EINTR is returned.
*/
save_time_for_read = time_for_read;
status = tcp_routines.aa_select(lsock + 1, (void *)&tcp_fd, (void *)0, (void *)0, &time_for_read);
time_for_read = save_time_for_read;
if (0 > status)
{
if (EINTR == errno && FALSE == out_of_time)
/* interrupted by a signal which is not OUR timer */
status = 0;
else
break;
}
if (outofband)
break;
if (timed)
{
if (msec_timeout > 0)
{
sys_get_curr_time(&cur_time);
cur_time = sub_abs_time(&end_time, &cur_time);
if (cur_time.at_sec <= 0)
{
out_of_time = TRUE;
cancel_timer(timer_id);
break;
}
} else
break;
}
#ifdef __linux__
time_for_read.at_sec = ((NO_M_TIMEOUT == timeout) ? 0 : 1);
time_for_read.at_usec = 0;
#endif
}
if (timed)
{
if (0 != msec_timeout)
{
cancel_timer(timer_id);
if (out_of_time || outofband)
return FALSE;
/*if (outofband)
outofband_action(FALSE);*/
}
}
if (0 > status)
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
iotcp_rmlsock((io_desc *)dev->iod);
rts_error(VARLSTCNT(6) ERR_SOCKWAIT, 0, ERR_TEXT, 2, errlen, errptr);
return FALSE;
}
size = sizeof(struct sockaddr_in);
status = tcp_routines.aa_accept(lsock, &peer, &size);
if (-1 == status)
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
iotcp_rmlsock((io_desc *)dev->iod);
rts_error(VARLSTCNT(6) ERR_SOCKACPT, 0, ERR_TEXT, 2, errlen, errptr);
return FALSE;
}
SPRINTF(newtcp.saddr, "%s,%d", tcp_routines.aa_inet_ntoa(peer.sin_addr),
GTM_NTOHS(newtcp.sin.sin_port));
newtcp.socket = status;
} else /* active connection */
{
if (NO_M_TIMEOUT != timeout)
{
msec_timeout = timeout2msec(timeout);
sys_get_curr_time(&cur_time);
add_int_to_abs_time(&cur_time, msec_timeout, &end_time);
}
no_time_left = FALSE;
temp_1 = 1;
do
{
if(1 != temp_1)
tcp_routines.aa_close(newtcp.socket);
newtcp.socket = tcp_routines.aa_socket(AF_INET, SOCK_STREAM, 0);
if (-1 == newtcp.socket)
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
rts_error(VARLSTCNT(5) ERR_SOCKINIT, 3, errno, errlen, errptr);
return FALSE;
}
/* allow multiple connections to the same IP address */
if (-1 == tcp_routines.aa_setsockopt(newtcp.socket, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)))
{
(void)tcp_routines.aa_close(newtcp.socket);
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
rts_error(VARLSTCNT(5) ERR_SOCKINIT, 3, errno, errlen, errptr);
return FALSE;
}
size=sizeof(newtcp.bufsiz);
if (-1 == tcp_routines.aa_getsockopt(newtcp.socket, SOL_SOCKET, SO_RCVBUF, &newtcp.bufsiz, &size))
{
(void)tcp_routines.aa_close(newtcp.socket);
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
rts_error(VARLSTCNT(5) ERR_SOCKINIT, 3, errno, errlen, errptr);
return FALSE;
}
/*
* Note: the check for EINTR from the connect need not be converted to an EINTR wrapper macro,
* since the connect is not retried on EINTR.
*/
temp_1 = tcp_routines.aa_connect(newtcp.socket, (struct sockaddr *)&newtcp.sin, sizeof(newtcp.sin));
if ((temp_1 < 0) && (ECONNREFUSED != errno) && (EINTR != errno))
{
(void)tcp_routines.aa_close(newtcp.socket);
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
rts_error(VARLSTCNT(6) ERR_OPENCONN, 0, ERR_TEXT, 2, errlen, errptr);
return FALSE;
}
if ((temp_1 < 0) && (EINTR == errno))
{
(void)tcp_routines.aa_close(newtcp.socket);
return FALSE;
}
if ((temp_1 < 0) && (NO_M_TIMEOUT != timeout))
{
sys_get_curr_time(&cur_time);
cur_time = sub_abs_time(&end_time, &cur_time);
if (cur_time.at_sec <= 0)
no_time_left = TRUE;
}
SHORT_SLEEP(100); /* Sleep for 100 ms */
}
while ((TRUE != no_time_left) && (temp_1 < 0));
if (temp_1 < 0) /* out of time */
{
tcp_routines.aa_close(newtcp.socket);
return FALSE;
}
#ifdef ntohs /* if it's a macro, use it instead of tcp_routines.aa_ntohs */
SPRINTF(newtcp.saddr, "%s,%d", tcp_routines.aa_inet_ntoa(newtcp.sin.sin_addr),
ntohs(newtcp.sin.sin_port));
#else
SPRINTF(newtcp.saddr, "%s,%d", tcp_routines.aa_inet_ntoa(newtcp.sin.sin_addr),
tcp_routines.aa_ntohs(newtcp.sin.sin_port));
#endif
}
memcpy(ioptr->dev_sp, &newtcp, sizeof(d_tcp_struct));
ioptr->state = dev_open;
}
#ifdef DEBUG_TCP
PRINTF("%s (%d) <<<\n", __FILE__, tcpptr->socket);
#endif
return TRUE;
}
int gtmsource()
{
int status, log_init_status, waitpid_res, save_errno;
char print_msg[1024], tmpmsg[1024];
gd_region *reg, *region_top;
sgmnt_addrs *csa, *repl_csa;
boolean_t all_files_open, isalive;
pid_t pid, ppid, procgp;
seq_num read_jnl_seqno, jnl_seqno;
unix_db_info *udi;
gtmsource_local_ptr_t gtmsource_local;
boolean_t this_side_std_null_coll;
int null_fd, rc;
memset((uchar_ptr_t)&jnlpool, 0, SIZEOF(jnlpool_addrs));
call_on_signal = gtmsource_sigstop;
ESTABLISH_RET(gtmsource_ch, SS_NORMAL);
if (-1 == gtmsource_get_opt())
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_MUPCLIERR);
if (gtmsource_options.shut_down)
{ /* Wait till shutdown time nears even before going to "jnlpool_init". This is because the latter will return
* with the ftok semaphore and access semaphore held and we do not want to be holding those locks (while
* waiting for the user specified timeout to expire) as that will affect new GTM processes and/or other
* MUPIP REPLIC commands that need these locks for their function.
*/
if (0 < gtmsource_options.shutdown_time)
{
repl_log(stdout, TRUE, TRUE, "Waiting for %d seconds before signalling shutdown\n",
gtmsource_options.shutdown_time);
LONG_SLEEP(gtmsource_options.shutdown_time);
} else
repl_log(stdout, TRUE, TRUE, "Signalling shutdown immediate\n");
} else if (gtmsource_options.start)
{
repl_log(stdout, TRUE, TRUE, "Initiating START of source server for secondary instance [%s]\n",
gtmsource_options.secondary_instname);
}
if (gtmsource_options.activate && (ROOTPRIMARY_SPECIFIED == gtmsource_options.rootprimary))
{ /* MUPIP REPLIC -SOURCE -ACTIVATE -UPDOK has been specified. We need to open the gld and db regions now
* in case this is a secondary -> primary transition. This is so we can later switch journal files in all
* journaled regions when the transition actually happens inside "gtmsource_rootprimary_init". But since
* we have not yet done a "jnlpool_init", we dont know if updates are disabled in it or not. Although we
* need to do the gld/db open only if updates are currently disabled in the jnlpool, we do this always
* because once we do a jnlpool_init, we will come back with the ftok on the jnlpool held and that has
* issues with later db open since we will try to hold the db ftok as part of db open and the ftok logic
* currently has assumptions that a process holds only one ftok at any point in time.
*/
assert(NULL == gd_header);
gvinit();
all_files_open = region_init(FALSE);
if (!all_files_open)
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_NOTALLDBOPN);
gtmsource_exit(ABNORMAL_SHUTDOWN);
}
}
jnlpool_init(GTMSOURCE, gtmsource_options.start, &is_jnlpool_creator);
/* is_jnlpool_creator == TRUE ==> this process created the journal pool
* is_jnlpool_creator == FALSE ==> journal pool already existed and this process simply attached to it.
*/
if (gtmsource_options.shut_down)
gtmsource_exit(gtmsource_shutdown(FALSE, NORMAL_SHUTDOWN) - NORMAL_SHUTDOWN);
else if (gtmsource_options.activate)
gtmsource_exit(gtmsource_mode_change(GTMSOURCE_MODE_ACTIVE_REQUESTED) - NORMAL_SHUTDOWN);
else if (gtmsource_options.deactivate)
gtmsource_exit(gtmsource_mode_change(GTMSOURCE_MODE_PASSIVE_REQUESTED) - NORMAL_SHUTDOWN);
else if (gtmsource_options.checkhealth)
gtmsource_exit(gtmsource_checkhealth() - NORMAL_SHUTDOWN);
else if (gtmsource_options.changelog)
gtmsource_exit(gtmsource_changelog() - NORMAL_SHUTDOWN);
else if (gtmsource_options.showbacklog)
gtmsource_exit(gtmsource_showbacklog() - NORMAL_SHUTDOWN);
else if (gtmsource_options.stopsourcefilter)
gtmsource_exit(gtmsource_stopfilter() - NORMAL_SHUTDOWN);
else if (gtmsource_options.jnlpool)
gtmsource_exit(gtmsource_jnlpool() - NORMAL_SHUTDOWN);
else if (gtmsource_options.losttncomplete)
gtmsource_exit(gtmsource_losttncomplete() - NORMAL_SHUTDOWN);
else if (gtmsource_options.needrestart)
gtmsource_exit(gtmsource_needrestart() - NORMAL_SHUTDOWN);
else if (gtmsource_options.showfreeze)
gtmsource_exit(gtmsource_showfreeze() - NORMAL_SHUTDOWN);
else if (gtmsource_options.setfreeze)
gtmsource_exit(gtmsource_setfreeze() - NORMAL_SHUTDOWN);
else if (!gtmsource_options.start)
{
assert(CLI_PRESENT == cli_present("STATSLOG"));
gtmsource_exit(gtmsource_statslog() - NORMAL_SHUTDOWN);
}
assert(gtmsource_options.start);
# ifndef REPL_DEBUG_NOBACKGROUND
/* Set "child_server_running" to FALSE before forking off child. Wait for it to be set to TRUE by the child. */
gtmsource_local = jnlpool.gtmsource_local;
gtmsource_local->child_server_running = FALSE;
FORK(pid);
if (0 > pid)
{
save_errno = errno;
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_JNLPOOLSETUP, 0,
ERR_TEXT, 2, RTS_ERROR_LITERAL("Could not fork source server"), save_errno);
} else if (0 < pid)
{ /* Parent. Wait until child sets "child_server_running" to FALSE. That is an indication that the child
* source server has completed its initialization phase and is all set so the parent command can return.
*/
while (isalive = is_proc_alive(pid, 0)) /* note : intended assignment */
{
if (gtmsource_local->child_server_running)
break;
/* To take care of reassignment of PIDs, the while condition should be && with the condition
* (PPID of pid == process_id)
*/
SHORT_SLEEP(GTMSOURCE_WAIT_FOR_SRV_START);
WAITPID(pid, &status, WNOHANG, waitpid_res); /* Release defunct child if dead */
}
if (isalive)
{ /* Child process is alive and started with no issues */
if (0 != (save_errno = rel_sem(SOURCE, JNL_POOL_ACCESS_SEM)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_JNLPOOLSETUP, 0,
ERR_TEXT, 2, RTS_ERROR_LITERAL("Error in rel_sem"), save_errno);
ftok_sem_release(jnlpool.jnlpool_dummy_reg, TRUE, TRUE);
} else
{ /* Child source server process errored out at startup and is no longer alive.
* If we were the one who created the journal pool, let us clean it up.
*/
repl_log(stdout, TRUE, TRUE, "Source server startup failed. See source server log file\n");
if (is_jnlpool_creator)
status = gtmsource_shutdown(TRUE, NORMAL_SHUTDOWN);
}
/* If the parent is killed (or crashes) between the fork and exit, checkhealth may not detect that startup
* is in progress - parent forks and dies, the system will release sem 0 and 1, checkhealth might test the
* value of sem 1 before the child grabs sem 1.
*/
gtmsource_exit(isalive ? SRV_ALIVE : SRV_ERR);
}
/* Point stdin to /dev/null */
OPENFILE("/dev/null", O_RDONLY, null_fd);
if (0 > null_fd)
rts_error_csa(CSA_ARG(NULL) ERR_REPLERR, RTS_ERROR_LITERAL("Failed to open /dev/null for read"), errno, 0);
FCNTL3(null_fd, F_DUPFD, 0, rc);
if (0 > rc)
rts_error_csa(CSA_ARG(NULL) ERR_REPLERR, RTS_ERROR_LITERAL("Failed to set stdin to /dev/null"), errno, 0);
CLOSEFILE(null_fd, rc);
if (0 > rc)
rts_error_csa(CSA_ARG(NULL) ERR_REPLERR, RTS_ERROR_LITERAL("Failed to close /dev/null"), errno, 0);
/* The parent process (source server startup command) will be holding the ftok semaphore and jnlpool access semaphore
* at this point. The variables that indicate this would have been copied over to the child during the fork. This will
* make the child think it is actually holding them as well when actually it is not. Reset those variables in the child
* to ensure they do not misrepresent the holder of those semaphores.
*/
ftok_sem_reg = NULL;
udi = FILE_INFO(jnlpool.jnlpool_dummy_reg);
assert(udi->grabbed_ftok_sem);
udi->grabbed_ftok_sem = FALSE;
assert(holds_sem[SOURCE][JNL_POOL_ACCESS_SEM]);
holds_sem[SOURCE][JNL_POOL_ACCESS_SEM] = FALSE;
assert(!holds_sem[SOURCE][SRC_SERV_COUNT_SEM]);
/* Start child source server initialization */
is_src_server = TRUE;
OPERATOR_LOG_MSG;
process_id = getpid();
/* Reinvoke secshr related initialization with the child's pid */
INVOKE_INIT_SECSHR_ADDRS;
/* Initialize mutex socket, memory semaphore etc. before any "grab_lock" is done by this process on the journal pool.
* Note that the initialization would already have been done by the parent receiver startup command but we need to
* redo the initialization with the child process id.
*/
assert(mutex_per_process_init_pid && (mutex_per_process_init_pid != process_id));
mutex_per_process_init();
START_HEARTBEAT_IF_NEEDED;
ppid = getppid();
log_init_status = repl_log_init(REPL_GENERAL_LOG, >msource_log_fd, gtmsource_options.log_file);
assert(SS_NORMAL == log_init_status);
repl_log_fd2fp(>msource_log_fp, gtmsource_log_fd);
if (-1 == (procgp = setsid()))
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_JNLPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Source server error in setsid"), errno);
# endif /* REPL_DEBUG_NOBACKGROUND */
if (ZLIB_CMPLVL_NONE != gtm_zlib_cmp_level)
gtm_zlib_init(); /* Open zlib shared library for compression/decompression */
REPL_DPRINT1("Setting up regions\n");
gvinit();
/* We use the same code dse uses to open all regions but we must make sure they are all open before proceeding. */
all_files_open = region_init(FALSE);
if (!all_files_open)
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_NOTALLDBOPN);
gtmsource_exit(ABNORMAL_SHUTDOWN);
}
/* Determine primary side null subscripts collation order */
/* Also check whether all regions have same null collation order */
this_side_std_null_coll = -1;
for (reg = gd_header->regions, region_top = gd_header->regions + gd_header->n_regions; reg < region_top; reg++)
{
csa = &FILE_INFO(reg)->s_addrs;
if (this_side_std_null_coll != csa->hdr->std_null_coll)
{
if (-1 == this_side_std_null_coll)
this_side_std_null_coll = csa->hdr->std_null_coll;
else
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_NULLCOLLDIFF);
gtmsource_exit(ABNORMAL_SHUTDOWN);
}
}
if (!REPL_ALLOWED(csa) && JNL_ALLOWED(csa))
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_REPLOFFJNLON, 2, DB_LEN_STR(reg));
gtmsource_exit(ABNORMAL_SHUTDOWN);
}
if (reg->read_only && REPL_ALLOWED(csa))
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_JNLPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Source Server does not have write permissions to one or "
"more database files that are replicated"));
gtmsource_exit(ABNORMAL_SHUTDOWN);
}
}
/* Initialize source server alive/dead state related fields in "gtmsource_local" before the ftok semaphore is released */
gtmsource_local->gtmsource_pid = process_id;
gtmsource_local->gtmsource_state = GTMSOURCE_START;
if (is_jnlpool_creator)
{
DEBUG_ONLY(jnlpool.jnlpool_ctl->jnlpool_creator_pid = process_id);
gtmsource_seqno_init(this_side_std_null_coll);
if (ROOTPRIMARY_SPECIFIED == gtmsource_options.rootprimary)
{ /* Created the journal pool as a root primary. Append a history record to the replication instance file.
* Invoke the function "gtmsource_rootprimary_init" to do that.
*/
gtmsource_rootprimary_init(jnlpool.jnlpool_ctl->jnl_seqno);
}
}
/* after this point we can no longer have the case where all the regions are unreplicated/non-journaled. */
# ifndef REPL_DEBUG_NOBACKGROUND
/* It is necessary for every process that is using the ftok semaphore to increment the counter by 1. This is used
* by the last process that shuts down to delete the ftok semaphore when it notices the counter to be 0.
* Note that the parent source server startup command would have done an increment of the ftok counter semaphore
* for the replication instance file. But the source server process (the child) that comes here would not have done
* that. Do that while the parent is still holding on to the ftok semaphore waiting for our okay.
*/
if (!ftok_sem_incrcnt(jnlpool.jnlpool_dummy_reg))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_JNLPOOLSETUP);
/* Increment the source server count semaphore */
status = incr_sem(SOURCE, SRC_SERV_COUNT_SEM);
if (0 != status)
{
save_errno = errno;
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_JNLPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Counter semaphore increment failure in child source server"), save_errno);
}
# else
if (0 != (save_errno = rel_sem_immediate(SOURCE, JNL_POOL_ACCESS_SEM)))
{
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_JNLPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error in rel_sem_immediate"), save_errno);
}
# endif /* REPL_DEBUG_NOBACKGROUND */
gtmsource_srv_count++;
gtmsource_local->child_server_running = TRUE; /* At this point, the parent startup command will stop waiting for child */
gtm_event_log_init();
/* Log source server startup command line first */
SPRINTF(tmpmsg, "%s %s\n", cli_lex_in_ptr->argv[0], cli_lex_in_ptr->in_str);
repl_log(gtmsource_log_fp, TRUE, TRUE, tmpmsg);
SPRINTF(tmpmsg, "GTM Replication Source Server with Pid [%d] started for Secondary Instance [%s]",
process_id, gtmsource_local->secondary_instname);
sgtm_putmsg(print_msg, VARLSTCNT(4) ERR_REPLINFO, 2, LEN_AND_STR(tmpmsg));
repl_log(gtmsource_log_fp, TRUE, TRUE, print_msg);
if (is_jnlpool_creator)
{
repl_log(gtmsource_log_fp, TRUE, TRUE, "Created jnlpool with shmid = [%d] and semid = [%d]\n",
jnlpool.repl_inst_filehdr->jnlpool_shmid, jnlpool.repl_inst_filehdr->jnlpool_semid);
} else
repl_log(gtmsource_log_fp, TRUE, TRUE, "Attached to existing jnlpool with shmid = [%d] and semid = [%d]\n",
jnlpool.repl_inst_filehdr->jnlpool_shmid, jnlpool.repl_inst_filehdr->jnlpool_semid);
gtm_event_log(GTM_EVENT_LOG_ARGC, "MUPIP", "REPLINFO", print_msg);
# ifdef GTM_TLS
if (REPL_TLS_REQUESTED)
{
repl_do_tls_init(gtmsource_log_fp);
assert(REPL_TLS_REQUESTED || PLAINTEXT_FALLBACK);
}
# endif
if (jnlpool.jnlpool_ctl->freeze)
{
last_seen_freeze_flag = jnlpool.jnlpool_ctl->freeze;
sgtm_putmsg(print_msg, VARLSTCNT(3) ERR_REPLINSTFROZEN, 1, jnlpool.repl_inst_filehdr->inst_info.this_instname);
repl_log(gtmsource_log_fp, TRUE, FALSE, print_msg);
sgtm_putmsg(print_msg, VARLSTCNT(3) ERR_REPLINSTFREEZECOMMENT, 1, jnlpool.jnlpool_ctl->freeze_comment);
repl_log(gtmsource_log_fp, TRUE, TRUE, print_msg);
}
gtmsource_local->jnlfileonly = gtmsource_options.jnlfileonly;
do
{ /* If mode is passive, go to sleep. Wakeup every now and then and check to see if I have to become active. */
gtmsource_state = gtmsource_local->gtmsource_state = GTMSOURCE_START;
if ((gtmsource_local->mode == GTMSOURCE_MODE_PASSIVE) && (gtmsource_local->shutdown == NO_SHUTDOWN))
{
gtmsource_poll_actions(FALSE);
SHORT_SLEEP(GTMSOURCE_WAIT_FOR_MODE_CHANGE);
continue;
}
if (GTMSOURCE_MODE_PASSIVE == gtmsource_local->mode)
{ /* Shutdown initiated */
assert(gtmsource_local->shutdown == SHUTDOWN);
sgtm_putmsg(print_msg, VARLSTCNT(4) ERR_REPLINFO, 2,
RTS_ERROR_LITERAL("GTM Replication Source Server Shutdown signalled"));
repl_log(gtmsource_log_fp, TRUE, TRUE, print_msg);
gtm_event_log(GTM_EVENT_LOG_ARGC, "MUPIP", "REPLINFO", print_msg);
break;
}
gtmsource_poll_actions(FALSE);
if (GTMSOURCE_CHANGING_MODE == gtmsource_state)
continue;
if (GTMSOURCE_MODE_ACTIVE_REQUESTED == gtmsource_local->mode)
gtmsource_local->mode = GTMSOURCE_MODE_ACTIVE;
SPRINTF(tmpmsg, "GTM Replication Source Server now in ACTIVE mode using port %d", gtmsource_local->secondary_port);
sgtm_putmsg(print_msg, VARLSTCNT(4) ERR_REPLINFO, 2, LEN_AND_STR(tmpmsg));
repl_log(gtmsource_log_fp, TRUE, TRUE, print_msg);
gtm_event_log(GTM_EVENT_LOG_ARGC, "MUPIP", "REPLINFO", print_msg);
DEBUG_ONLY(repl_csa = &FILE_INFO(jnlpool.jnlpool_dummy_reg)->s_addrs;)
assert(!repl_csa->hold_onto_crit); /* so it is ok to invoke "grab_lock" and "rel_lock" unconditionally */
grab_lock(jnlpool.jnlpool_dummy_reg, TRUE, HANDLE_CONCUR_ONLINE_ROLLBACK);
if (GTMSOURCE_HANDLE_ONLN_RLBK == gtmsource_state)
{
repl_log(gtmsource_log_fp, TRUE, TRUE, "Starting afresh due to ONLINE ROLLBACK\n");
repl_log(gtmsource_log_fp, TRUE, TRUE, "REPL INFO - Current Jnlpool Seqno : %llu\n",
jnlpool.jnlpool_ctl->jnl_seqno);
continue;
}
QWASSIGN(gtmsource_local->read_addr, jnlpool.jnlpool_ctl->write_addr);
gtmsource_local->read = jnlpool.jnlpool_ctl->write;
gtmsource_local->read_state = gtmsource_local->jnlfileonly ? READ_FILE : READ_POOL;
read_jnl_seqno = gtmsource_local->read_jnl_seqno;
assert(read_jnl_seqno <= jnlpool.jnlpool_ctl->jnl_seqno);
if (read_jnl_seqno < jnlpool.jnlpool_ctl->jnl_seqno)
{
gtmsource_local->read_state = READ_FILE;
QWASSIGN(gtmsource_save_read_jnl_seqno, jnlpool.jnlpool_ctl->jnl_seqno);
gtmsource_pool2file_transition = TRUE; /* so that we read the latest gener jnl files */
}
rel_lock(jnlpool.jnlpool_dummy_reg);
if (SS_NORMAL != (status = gtmsource_alloc_tcombuff()))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error allocating initial tcom buffer space. Malloc error"), status);
gtmsource_filter = NO_FILTER;
if ('\0' != gtmsource_local->filter_cmd[0])
{
if (SS_NORMAL == (status = repl_filter_init(gtmsource_local->filter_cmd)))
gtmsource_filter |= EXTERNAL_FILTER;
else
gtmsource_exit(ABNORMAL_SHUTDOWN);
}
gtmsource_process();
/* gtmsource_process returns only when mode needs to be changed to PASSIVE */
assert(gtmsource_state == GTMSOURCE_CHANGING_MODE);
gtmsource_ctl_close();
gtmsource_free_msgbuff();
gtmsource_free_tcombuff();
gtmsource_free_filter_buff();
gtmsource_stop_heartbeat();
if (FD_INVALID != gtmsource_sock_fd)
repl_close(>msource_sock_fd);
if (gtmsource_filter & EXTERNAL_FILTER)
repl_stop_filter();
} while (TRUE);
int gtmrecv_shutdown(boolean_t auto_shutdown, int exit_status)
{
uint4 savepid;
boolean_t shut_upd_too = FALSE;
int status;
unix_db_info *udi;
error_def(ERR_RECVPOOLSETUP);
error_def(ERR_TEXT);
repl_log(stdout, TRUE, TRUE, "Initiating shut down\n");
call_on_signal = NULL; /* So we don't reenter on error */
/* assert that auto shutdown should be invoked only if the current process is a receiver server */
assert(!auto_shutdown || gtmrecv_srv_count);
if (auto_shutdown)
{ /* grab the ftok semaphore and recvpool access control lock IN THAT ORDER (to avoid deadlocks) */
repl_inst_ftok_sem_lock();
status = grab_sem(RECV, RECV_POOL_ACCESS_SEM);
if (0 > status)
{
repl_log(stderr, TRUE, TRUE,
"Error grabbing receive pool control semaphore : %s. Shutdown not complete\n", REPL_SEM_ERROR);
return (ABNORMAL_SHUTDOWN);
}
} else
{ /* ftok semaphore and recvpool access semaphore should already be held from the previous call to "recvpool_init" */
DEBUG_ONLY(udi = (unix_db_info *)FILE_INFO(recvpool.recvpool_dummy_reg);)
assert(udi->grabbed_ftok_sem);
assert(holds_sem[RECV][RECV_POOL_ACCESS_SEM]);
/* We do not want to hold the options semaphore to avoid deadlocks with receiver server startup (C9F12-002766) */
assert(!holds_sem[RECV][RECV_SERV_OPTIONS_SEM]);
recvpool.gtmrecv_local->shutdown = SHUTDOWN;
/* Wait for receiver server to die. But release ftok semaphore and recvpool access control semaphore before
* waiting as the concurrently running receiver server might need these (e.g. if it is about to call the
* function "repl_inst_was_rootprimary").
*/
if (0 != rel_sem(RECV, RECV_POOL_ACCESS_SEM))
gtm_putmsg(VARLSTCNT(7) ERR_TEXT, 2, RTS_ERROR_LITERAL("Error in receiver server shutdown rel_sem"),
REPL_SEM_ERRNO);
repl_inst_ftok_sem_release();
/* Wait for receiver server to shut down */
while((SHUTDOWN == recvpool.gtmrecv_local->shutdown)
&& (0 < (savepid = recvpool.gtmrecv_local->recv_serv_pid))
&& is_proc_alive(savepid, 0))
SHORT_SLEEP(GTMRECV_WAIT_FOR_SHUTDOWN);
/* (Re)Grab the ftok semaphore and recvpool access control semaphore IN THAT ORDER (to avoid deadlocks) */
repl_inst_ftok_sem_lock();
status = grab_sem(RECV, RECV_POOL_ACCESS_SEM);
if (0 > status)
{
repl_log(stderr, TRUE, TRUE,
"Error regrabbing receive pool control semaphore : %s. Shutdown not complete\n", REPL_SEM_ERROR);
return (ABNORMAL_SHUTDOWN);
}
exit_status = recvpool.gtmrecv_local->shutdown;
if (SHUTDOWN == exit_status)
{
if (0 == savepid) /* No Receiver Process */
exit_status = NORMAL_SHUTDOWN;
else /* Receiver Server Crashed */
{
repl_log(stderr, FALSE, TRUE, "Receiver Server exited abnormally\n");
exit_status = ABNORMAL_SHUTDOWN;
shut_upd_too = TRUE;
}
}
}
static int helper_init(upd_helper_entry_ptr_t helper, recvpool_user helper_type)
{
int save_errno, save_shutdown;
char helper_cmd[UPDHELPER_CMD_MAXLEN];
int status;
int4 i4status;
mstr helper_log_cmd, helper_trans_cmd;
upd_helper_ctl_ptr_t upd_helper_ctl;
#ifdef UNIX
pid_t helper_pid, waitpid_res;
#elif defined(VMS)
uint4 helper_pid, cmd_channel;
char mbx_suffix[2 + 1]; /* hex representation of numbers 0 through MAX_UPD_HELPERS-1, +1 for '\0' */
$DESCRIPTOR(cmd_desc_reader, UPDHELPER_READER_CMD_STR);
$DESCRIPTOR(cmd_desc_writer, UPDHELPER_WRITER_CMD_STR);
#endif
upd_helper_ctl = recvpool.upd_helper_ctl;
save_shutdown = helper->helper_shutdown;
helper->helper_shutdown = NO_SHUTDOWN;
#ifdef UNIX
if (0 > (helper_pid = fork())) /* BYPASSOK: we exec immediately, no FORK_CLEAN needed */
{
save_errno = errno;
helper->helper_shutdown = save_shutdown;
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
LEN_AND_LIT("Could not fork update process"), save_errno);
repl_errno = EREPL_UPDSTART_FORK;
return UPDPROC_START_ERR;
}
if (0 == helper_pid)
{ /* helper */
getjobnum();
helper->helper_pid_prev = process_id; /* identify owner of slot */
helper_log_cmd.len = STR_LIT_LEN(UPDHELPER_CMD);
helper_log_cmd.addr = UPDHELPER_CMD;
if (SS_NORMAL != (i4status = TRANS_LOG_NAME(&helper_log_cmd, &helper_trans_cmd, helper_cmd, SIZEOF(helper_cmd),
dont_sendmsg_on_log2long)))
{
helper->helper_shutdown = save_shutdown;
gtm_putmsg(VARLSTCNT(6) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
LEN_AND_LIT("Could not find path of Helper Process. Check value of $gtm_dist"));
if (SS_LOG2LONG == i4status)
gtm_putmsg(VARLSTCNT(5) ERR_LOGTOOLONG, 3, LEN_AND_LIT(UPDHELPER_CMD), SIZEOF(helper_cmd) - 1);
repl_errno = EREPL_UPDSTART_BADPATH;
return UPDPROC_START_ERR;
}
helper_cmd[helper_trans_cmd.len] = '\0';
if (-1 == EXECL(helper_cmd, helper_cmd, UPDHELPER_CMD_ARG1, UPDHELPER_CMD_ARG2,
(UPD_HELPER_READER == helper_type) ? UPDHELPER_READER_CMD_ARG3 : UPDHELPER_WRITER_CMD_ARG3, NULL))
{
save_errno = errno;
helper->helper_shutdown = save_shutdown;
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
LEN_AND_LIT("Could not exec Helper Process"), save_errno);
repl_errno = EREPL_UPDSTART_EXEC;
return UPDPROC_START_ERR;
}
}
#elif defined(VMS)
/* Create detached server and write startup commands to it */
i2hex(helper - upd_helper_ctl->helper_list, LIT_AND_LEN(mbx_suffix));
mbx_suffix[SIZEOF(mbx_suffix) - 1] = '\0';
/* A mailbox is created per helper, and the mailbox name is assigned to a logical. This logical will persist until the
* helper terminates. So, we need to assign a unique logical per helper. Hence the suffix. */
if (SS_NORMAL != (status = repl_create_server((UPD_HELPER_READER == helper_type) ? &cmd_desc_reader : &cmd_desc_writer,
UPDHELPER_MBX_PREFIX, mbx_suffix, &cmd_channel, &helper->helper_pid_prev,
ERR_RECVPOOLSETUP)))
{
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2, LEN_AND_LIT("Unable to spawn Helper process"), status);
helper->helper_shutdown = save_shutdown;
repl_errno = EREPL_UPDSTART_FORK;
return UPDPROC_START_ERR;
}
helper_pid = helper->helper_pid_prev;
#endif
/* Wait for helper to startup */
while (helper_pid != helper->helper_pid && is_proc_alive(helper_pid, 0))
{
SHORT_SLEEP(GTMRECV_WAIT_FOR_SRV_START);
UNIX_ONLY(WAITPID(helper_pid, &status, WNOHANG, waitpid_res);) /* Release defunct helper process if dead */
}
/* The helper has now gone far enough in the initialization, or died before initialization. Consider startup completed. */
#if defined(VMS)
/* Deassign the send-cmd mailbox channel */
if (SS_NORMAL != (status = sys$dassgn(cmd_channel)))
{
gtm_putmsg(VARLSTCNT(7) ERR_RECVPOOLSETUP, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Unable to close upd-send-cmd mbox channel"), status);
helper->helper_shutdown = save_shutdown;
repl_errno = EREPL_UPDSTART_BADPATH; /* Just to make an auto-shutdown */
return UPDPROC_START_ERR;
}
#endif
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Helper %s started. PID %d [0x%X]\n",
(UPD_HELPER_READER == helper_type) ? "reader" : "writer", helper_pid, helper_pid);
return UPDPROC_STARTED;
}
int tcp_open(char *host, unsigned short port, int4 timeout, boolean_t passive) /* host needs to be NULL terminated */
{
boolean_t no_time_left = FALSE, error_given = FALSE;
char temp_addr[SA_MAXLEN + 1], addr[SA_MAXLEN + 1];
char *from, *to, *errptr, *temp_ch;
int match, sock, sendbufsize, size, ii, on = 1, temp_1 = -2;
int4 errlen, rv, msec_timeout;
struct sockaddr_in sin;
in_addr_t temp_sin_addr;
char msg_buffer[1024];
mstr msg_string;
ABS_TIME cur_time, end_time;
fd_set tcp_fd;
struct sockaddr_in peer;
error_def(ERR_INVADDRSPEC);
temp_sin_addr = 0;
msg_string.len = sizeof(msg_buffer);
msg_string.addr = msg_buffer;
memset((char *)&sin, 0, sizeof(struct sockaddr_in));
/* ============================= initialize structures ============================== */
if (NULL != host)
{
temp_ch = host;
while(ISDIGIT(*temp_ch) || ('.' == *temp_ch))
temp_ch++;
if ('\0' != *temp_ch)
SPRINTF(addr, "%s", iotcp_name2ip(host));
else
SPRINTF(addr, "%s", host);
if (-1 == (temp_sin_addr = tcp_routines.aa_inet_addr(addr)))
{
gtm_getmsg(ERR_INVADDRSPEC, &msg_string);
util_out_print(msg_string.addr, TRUE, ERR_INVADDRSPEC);
return -1;
}
}
if (passive)
/* We can only listen on our own system */
sin.sin_addr.s_addr = INADDR_ANY;
else
{
if (0 == temp_sin_addr)
{ /* If no address was specified */
util_out_print("An address has to be specified for an active connection.", TRUE);
return -1;
}
/* Set where to send the connection attempt */
sin.sin_addr.s_addr = temp_sin_addr;
}
sin.sin_port = GTM_HTONS(port);
sin.sin_family = AF_INET;
/* ============================== do the connection ============================== */
if (passive)
{
struct timeval utimeout, save_utimeout;
int lsock;
lsock = tcp_routines.aa_socket(AF_INET, SOCK_STREAM, 0);
if (-1 == lsock)
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
util_out_print(errptr, TRUE, errno);
return -1;
}
/* allow multiple connections to the same IP address */
if (-1 == tcp_routines.aa_setsockopt(lsock, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)))
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
(void)tcp_routines.aa_close(lsock);
util_out_print(errptr, TRUE, errno);
return -1;
}
if (-1 == tcp_routines.aa_bind(lsock, (struct sockaddr *)&sin, sizeof(struct sockaddr)))
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
(void)tcp_routines.aa_close(lsock);
util_out_print(errptr, TRUE, errno);
return -1;
}
/* establish a queue of length MAX_CONN_PENDING for incoming connections */
if (-1 == tcp_routines.aa_listen(lsock, MAX_CONN_PENDING))
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
(void)tcp_routines.aa_close(lsock);
util_out_print(errptr, TRUE, errno);
return -1;
}
if (NO_M_TIMEOUT != timeout)
{
msec_timeout = timeout2msec(timeout);
sys_get_curr_time(&cur_time);
add_int_to_abs_time(&cur_time, msec_timeout, &end_time);
utimeout.tv_sec = timeout;
utimeout.tv_usec = 0;
}
while(1)
{
while(1)
{
/*
* the check for EINTR below is valid and should not be converted to an EINTR
* wrapper macro, since it might be a timeout.
*/
FD_ZERO(&tcp_fd);
FD_SET(lsock, &tcp_fd);
save_utimeout = utimeout;
rv = tcp_routines.aa_select(lsock + 1, (void *)&tcp_fd, (void *)0, (void *)0,
(NO_M_TIMEOUT == timeout) ? (struct timeval *)0 : &utimeout);
utimeout = save_utimeout;
if ((0 <= rv) || (EINTR != errno))
break;
if (NO_M_TIMEOUT != timeout)
{
sys_get_curr_time(&cur_time);
cur_time = sub_abs_time(&end_time, &cur_time);
if (0 >= (utimeout.tv_sec = cur_time.at_sec))
{
rv = 0;
break;
}
}
}
if (0 == rv)
{
util_out_print("Listening timed out.\n", TRUE);
(void)tcp_routines.aa_close(lsock);
return -1;
} else if (0 > rv)
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
util_out_print(errptr, TRUE, errno);
(void)tcp_routines.aa_close(lsock);
return -1;
}
size = sizeof(struct sockaddr_in);
sock = tcp_routines.aa_accept(lsock, &peer, &size);
if (-1 == sock)
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
util_out_print(errptr, TRUE, errno);
(void)tcp_routines.aa_close(lsock);
return -1;
}
SPRINTF(&temp_addr[0], "%s", tcp_routines.aa_inet_ntoa(peer.sin_addr));
#ifdef DEBUG_ONLINE
PRINTF("Connection is from : %s\n", &temp_addr[0]);
#endif
/* Check if connection is from whom we want it to be from. Note that this is not a robust check
(potential for multiple IP addrs for a resolved name but workarounds for this exist so not a lot
of effort has been expended here at this time. Especially since the check is easily spoofed with
raw sockets anyway. It is more for the accidental "oops" type check than serious security..
*/
if ((0 == temp_sin_addr) || (0 == memcmp(&addr[0], &temp_addr[0], strlen(addr))))
break;
else
{ /* Connection not from expected host */
(void)tcp_routines.aa_close(sock);
if (NO_M_TIMEOUT != timeout)
{
sys_get_curr_time(&cur_time);
cur_time = sub_abs_time(&end_time, &cur_time);
utimeout.tv_sec = ((cur_time.at_sec > 0) ? cur_time.at_sec : 0);
}
if (!error_given)
{
util_out_print("Connection from !AD rejected and ignored", TRUE,
LEN_AND_STR(&temp_addr[0]));
error_given = TRUE;
}
}
}
(void)tcp_routines.aa_close(lsock);
} else
{
if (NO_M_TIMEOUT != timeout)
{
msec_timeout = timeout2msec(timeout);
sys_get_curr_time(&cur_time);
add_int_to_abs_time(&cur_time, msec_timeout, &end_time);
}
no_time_left = FALSE;
temp_1 = 1;
do
{
if (1 != temp_1)
tcp_routines.aa_close(sock);
sock = tcp_routines.aa_socket(AF_INET, SOCK_STREAM, 0);
if (-1 == sock)
{
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
util_out_print(errptr, TRUE, errno);
return -1;
}
/* allow multiple connections to the same IP address */
if (-1 == tcp_routines.aa_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)))
{
(void)tcp_routines.aa_close(sock);
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
util_out_print(errptr, TRUE, errno);
return -1;
}
temp_1 = tcp_routines.aa_connect(sock, (struct sockaddr *)(&sin), sizeof(sin));
/*
* the check for EINTR below is valid and should not be converted to an EINTR
* wrapper macro, because other error conditions are checked, and a retry is not
* immediately performed.
*/
if ((0 > temp_1) && (ECONNREFUSED != errno) && (EINTR != errno))
{
(void)tcp_routines.aa_close(sock);
errptr = (char *)STRERROR(errno);
errlen = strlen(errptr);
util_out_print(errptr, TRUE, errno);
return -1;
}
if ((0 > temp_1) && (EINTR == errno))
{
(void)tcp_routines.aa_close(sock);
util_out_print("Interrupted.", TRUE);
return -1;
}
if ((temp_1 < 0) && (NO_M_TIMEOUT != timeout))
{
sys_get_curr_time(&cur_time);
cur_time = sub_abs_time(&end_time, &cur_time);
if (cur_time.at_sec <= 0)
no_time_left = TRUE;
}
SHORT_SLEEP(NAP_LENGTH); /* Sleep for NAP_LENGTH ms */
} while ((FALSE == no_time_left) && (0 > temp_1));
if (0 > temp_1) /* out of time */
{
tcp_routines.aa_close(sock);
util_out_print("Connection timed out.", TRUE);
return -1;
}
}
return sock;
}
static int helper_init(upd_helper_entry_ptr_t helper, recvpool_user helper_type)
{
int save_errno, save_shutdown;
char helper_cmd[UPDHELPER_CMD_MAXLEN];
int helper_cmd_len;
int status;
int4 i4status;
pid_t helper_pid, waitpid_res;
save_shutdown = helper->helper_shutdown;
helper->helper_shutdown = NO_SHUTDOWN;
if (!gtm_dist_ok_to_use)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_GTMDISTUNVERIF, 4, STRLEN(gtm_dist), gtm_dist,
gtmImageNames[image_type].imageNameLen, gtmImageNames[image_type].imageName);
if (WBTEST_ENABLED(WBTEST_MAXGTMDIST_HELPER_PROCESS))
{
memset(gtm_dist, 'a', GTM_PATH_MAX-2);
gtm_dist[GTM_PATH_MAX-1] = '\0';
}
helper_cmd_len = SNPRINTF(helper_cmd, UPDHELPER_CMD_MAXLEN, UPDHELPER_CMD, gtm_dist);
if ((-1 == helper_cmd_len) || (UPDHELPER_CMD_MAXLEN <= helper_cmd_len))
{
helper->helper_shutdown = save_shutdown;
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_HLPPROC, 0, ERR_TEXT, 2,
LEN_AND_LIT("Could not find path of Helper Process. Check value of $gtm_dist"));
repl_errno = EREPL_UPDSTART_BADPATH;
return UPDPROC_START_ERR ;
}
FORK(helper_pid);
if (0 > helper_pid)
{
save_errno = errno;
helper->helper_shutdown = save_shutdown;
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_HLPPROC, 0, ERR_TEXT, 2,
LEN_AND_LIT("Could not fork Helper process"), save_errno);
repl_errno = EREPL_UPDSTART_FORK;
return UPDPROC_START_ERR;
}
if (0 == helper_pid)
{ /* helper */
getjobnum();
helper->helper_pid_prev = process_id; /* identify owner of slot */
if (WBTEST_ENABLED(WBTEST_BADEXEC_HELPER_PROCESS))
STRCPY(helper_cmd, "ersatz");
if (-1 == EXECL(helper_cmd, helper_cmd, UPDHELPER_CMD_ARG1, UPDHELPER_CMD_ARG2,
(UPD_HELPER_READER == helper_type) ? UPDHELPER_READER_CMD_ARG3 : UPDHELPER_WRITER_CMD_ARG3, NULL))
{
save_errno = errno;
helper->helper_shutdown = save_shutdown;
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_HLPPROC, 0, ERR_TEXT, 2,
LEN_AND_LIT("Could not exec Helper Process"), save_errno);
repl_errno = EREPL_UPDSTART_EXEC;
UNDERSCORE_EXIT(UPDPROC_START_ERR);
}
}
/* Wait for helper to startup */
while (helper_pid != helper->helper_pid && is_proc_alive(helper_pid, 0))
{
SHORT_SLEEP(GTMRECV_WAIT_FOR_SRV_START);
UNIX_ONLY(WAITPID(helper_pid, &status, WNOHANG, waitpid_res);) /* Release defunct helper process if dead */
}
/* The helper has now gone far enough in the initialization, or died before initialization. Consider startup completed. */
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Helper %s started. PID %d [0x%X]\n",
(UPD_HELPER_READER == helper_type) ? "reader" : "writer", helper_pid, helper_pid);
return UPDPROC_STARTED;
}
