void gtcm_cn_disc(omi_conn *cptr, omi_conn_ll *cll)
{
time_t end;
int i, nxact, nerrs;
int rc;
/* Cumulative statistics */
end = time((time_t *)0);
cll->st_cn.start += end - cptr->stats.start;
cll->st_cn.bytes_recv += cptr->stats.bytes_recv;
cll->st_cn.bytes_send += cptr->stats.bytes_send;
for (i = 0, nxact = 0; i < OMI_OP_MAX; i++)
{
cll->st_cn.xact[i] += cptr->stats.xact[i];
nxact += cptr->stats.xact[i];
}
for (i = 0, nerrs = 0; i < OMI_ER_MAX; i++)
{
cll->st_cn.errs[i] += cptr->stats.errs[i];
nerrs += cptr->stats.errs[i];
}
cll->stats.disc++;
if (cptr->state != OMI_ST_DISC)
cll->stats.clos++;
OMI_DBG_STMP;
OMI_DBG((omi_debug, "%s: connection %d to %s closed\n",
SRVR_NAME, cptr->stats.id, gtcm_hname(&cptr->stats.sin)));
OMI_DBG((omi_debug, "%s:\t%ld seconds connect time\n", SRVR_NAME, end - cptr->stats.start));
OMI_DBG((omi_debug, "%s:\t%d transactions\n", SRVR_NAME, nxact));
OMI_DBG((omi_debug, "%s:\t%d errors\n", SRVR_NAME, nerrs));
OMI_DBG((omi_debug, "%s:\t%d bytes recv'd\n", SRVR_NAME, cptr->stats.bytes_recv));
OMI_DBG((omi_debug, "%s:\t%d bytes sent\n", SRVR_NAME, cptr->stats.bytes_send));
#ifdef BSD_TCP
/* Close out the network connection */
CLOSEFILE_RESET(cptr->fd, rc); /* resets "cptr->fd" to FD_INVALID */
#endif /* defined(BSD_TCP) */
if (FD_INVALID != cptr->pklog)
CLOSEFILE_RESET(cptr->pklog, rc); /* resets "cptr->pklog" to FD_INVALID */
#ifdef GTCM_RC
if (cptr->of)
rc_oflow_fin(cptr->of);
#endif /* defined(GTCM_RC) */
/* Release locks held on this connection */
omi_prc_unla(cptr, cptr->buff, cptr->buff, cptr->buff);
/* Release the buffer and connection structure */
free(cptr->buff);
free(cptr);
return;
}
void cmj_incoming_call(struct NTD *tsk)
{
int rval, rc;
struct CLB *lnk;
struct sockaddr_in in;
GTM_SOCKLEN_TYPE sz = SIZEOF(in);
cmi_status_t status;
while ((-1 == (rval = ACCEPT(tsk->listen_fd, (struct sockaddr *)&in, (GTM_SOCKLEN_TYPE *)&sz))) && EINTR == errno)
;
while (rval >= 0)
{
status = cmj_setupfd(rval);
if (CMI_ERROR(status))
{
CLOSEFILE_RESET(rval, rc); /* resets "rval" to FD_INVALID */
return;
}
status = cmj_set_async(rval);
if (CMI_ERROR(status))
{
CLOSEFILE_RESET(rval, rc); /* resets "rval" to FD_INVALID */
return;
}
/* grab a clb off of the free list */
lnk = cmi_alloc_clb();
if (!lnk || !tsk->acc || !tsk->acc(lnk) || !tsk->crq)
{
/* no point if the callbacks are not in place */
cmi_free_clb(lnk);
CLOSEFILE_RESET(rval, rc); /* resets "rval" to FD_INVALID */
return;
}
if (rval > tsk->max_fd)
tsk->max_fd = rval;
lnk->mun = rval;
lnk->sta = CM_CLB_IDLE;
lnk->peer = in;
insqh(&lnk->cqe, &tsk->cqh);
lnk->ntd = tsk;
FD_SET(rval, &tsk->es);
/* setup for callback processing */
lnk->deferred_event = TRUE;
lnk->deferred_reason = CMI_REASON_CONNECT;
while ((-1 == (rval = ACCEPT(tsk->listen_fd, (struct sockaddr *)&in, (GTM_SOCKLEN_TYPE *)&sz))) && EINTR == errno)
;
}
}
void iott_close(io_desc *v, mval *pp)
{
d_tt_struct *ttptr;
params ch;
int status;
int p_offset;
boolean_t ch_set;
assert(v->type == tt);
if (v->state != dev_open)
return;
ESTABLISH_GTMIO_CH(&v->pair, ch_set);
iott_flush(v);
if (v->pair.in != v)
assert(v->pair.out == v);
ttptr = (d_tt_struct *)v->dev_sp;
if (v->pair.out != v)
assert(v->pair.in == v);
v->state = dev_closed;
resetterm(v);
p_offset = 0;
while (*(pp->str.addr + p_offset) != iop_eol)
{
if ((ch = *(pp->str.addr + p_offset++)) == iop_exception)
{
v->error_handler.len = *(pp->str.addr + p_offset);
v->error_handler.addr = (char *)(pp->str.addr + p_offset + 1);
s2pool(&v->error_handler);
}
p_offset += ((IOP_VAR_SIZE == io_params_size[ch]) ?
(unsigned char)*(pp->str.addr + p_offset) + 1 : io_params_size[ch]);
}
if (v == io_std_device.in || (v == io_std_device.out))
{
REVERT_GTMIO_CH(&v->pair, ch_set);
return;
}
CLOSEFILE_RESET(ttptr->fildes, status); /* resets "ttptr->fildes" to FD_INVALID */
if (0 != status)
{
assert(status == errno);
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8) ERR_SYSCALL, 5,
RTS_ERROR_LITERAL("iott_close(CLOSEFILE)"), CALLFROM, status);
}
if (ttptr->recall_buff.addr)
{
free(ttptr->recall_buff.addr);
ttptr->recall_buff.addr = NULL;
}
REVERT_GTMIO_CH(&v->pair, ch_set);
return;
}
int gtm_pipe(char *command, pipe_type pt)
{
int pfd[2], child, parent;
int dup2_res, rc;
pid_t child_pid;
parent = (int)pt;
child = 1 - parent;
if (0 > pipe(pfd))
{
PERROR("pipe : ");
return -2;
}
if (-1 == (child_pid = fork())) /* BYPASSOK: we exit immediately, no FORK_CLEAN needed */
{
PERROR("fork : ");
return -1;
} else if (0 == child_pid)
{ /* child process */
CLOSEFILE_RESET(pfd[parent], rc); /* resets "pfd[parent]" to FD_INVALID */
DUP2(pfd[child], child, dup2_res);
CLOSEFILE_RESET(pfd[child], rc); /* resets "pfd[child]" to FD_INVALID */
/* We should have used exec instead of SYSTEM. Earlier it was followed by exit(0), which calls exit_handler.
* So both child and parent will do exit handling. This can make ref_cnt < 0, or, it can release semaphores,
* which we should not release until parent exists. So we just call _exit(0). Add the do nothing if to
* keep compiler happy since exiting anyway.
*/
if (-1 == SYSTEM(command));
_exit(0); /* just exit from here */
} else
{ /* parent process */
pipe_child = child_pid;
CLOSEFILE_RESET(pfd[child], rc); /* resets "pfd[child]" to FD_INVALID */
return pfd[parent];
}
assert(FALSE);
/* It should never get here, just to keep compiler happy. */
return -3;
}
int gtm_pipe(char *command, pipe_type pt)
{
int pfd[2], child, parent;
int dup2_res, rc;
pid_t child_pid;
parent = (int)pt;
child = 1 - parent;
if (0 > pipe(pfd))
{
PERROR("pipe : ");
return -2;
}
FORK(child_pid);
if (-1 == child_pid)
{
PERROR("fork : ");
return -1;
} else if (0 == child_pid)
{ /* child process */
CLOSEFILE_RESET(pfd[parent], rc); /* resets "pfd[parent]" to FD_INVALID */
DUP2(pfd[child], child, dup2_res);
CLOSEFILE_RESET(pfd[child], rc); /* resets "pfd[child]" to FD_INVALID */
/* We should have used exec instead of SYSTEM. Earlier it was followed by exit(EXIT_SUCCESS), which calls
* exit_handler. So both child and parent will do exit handling. This can make ref_cnt < 0, or, it can release
* semaphores, which we should not release until parent exists. So we just call _exit(EXIT_SUCCESS). Add the do
* nothing if to keep compiler happy since exiting anyway.
*/
rc = SYSTEM(command);
_exit(EXIT_SUCCESS); /* just exit from here */
} else
{ /* parent process */
pipe_child = child_pid;
CLOSEFILE_RESET(pfd[child], rc); /* resets "pfd[child]" to FD_INVALID */
return pfd[parent];
}
}
void close_repl_logfiles()
{
int rc;
if (FD_INVALID != gtmsource_log_fd)
CLOSEFILE_RESET(gtmsource_log_fd, rc); /* resets "gtmsource_log_fd" to FD_INVALID */
if (NULL != gtmsource_log_fp)
FCLOSE(gtmsource_log_fp, rc);
if (FD_INVALID != gtmrecv_log_fd)
CLOSEFILE_RESET(gtmrecv_log_fd, rc); /* resets "gtmrecv_log_fd" to FD_INVALID */
if (NULL != gtmrecv_log_fp)
FCLOSE(gtmrecv_log_fp, rc);
if (FD_INVALID != updproc_log_fd)
{
assert(updproc_log_fd != updhelper_log_fd);
CLOSEFILE_RESET(updproc_log_fd, rc); /* resets "updproc_log_fd" to FD_INVALID */
}
if (NULL != updproc_log_fp)
FCLOSE(updproc_log_fp, rc);
if (FD_INVALID != updhelper_log_fd)
CLOSEFILE_RESET(updhelper_log_fd, rc); /* resets "updhelper_log_fd" to FD_INVALID */
if (NULL != updhelper_log_fp)
FCLOSE(updhelper_log_fp, rc);
}
short ionl_open(io_log_name *dev_name, mval *pp, int fd, mval *mspace, int4 timeout)
{
unsigned char ch;
io_desc *d_in, *d_out, *ioptr;
int p_offset, status;
p_offset = 0;
/* If UNIX, then /dev/null was actually opened by io_open_try so we have to close it
since we don't use the device, we just simulate it by doing nothing on writes except
maintaining the appropriate pointers. We test for fd >= 0 since the less than zero
values mean no device was opened.
*/
UNIX_ONLY(
if (0 <= fd)
CLOSEFILE_RESET(fd, status); /* resets "fd" to FD_INVALID */
);
void iopi_iocontrol(mstr *d)
{
char action[MAX_DEVCTL_LENGTH];
d_rm_struct *d_rm;
int rc;
d_rm = (d_rm_struct *) io_curr_device.out->dev_sp;
/* WRITE /EOF only applies to PIPE devices. Sequential file and FIFO devices should be closed with CLOSE.*/
if (!d_rm->pipe)
return;
/* we should not get here unless there is some string length after write / */
assert((int)d->len);
if (0 == d->len)
return;
lower_to_upper((uchar_ptr_t)&action[0], (uchar_ptr_t)d->addr, MIN(d->len, SIZEOF(action)));
if (0 == memcmp(&action[0], "EOF", MIN(d->len, SIZEOF(action))))
{ /* Implement the write /EOF action. Close the output stream to force any blocked output to complete.
* Doing a write /EOF closes the output file descriptor for the pipe device but does not close the
* device. Since the M program could attempt this command more than once, check if the file descriptor
* is already closed before the actual close.
* Ignore the /EOF action if the device is read-only as this is a nop
*/
if (d_rm->noread)
return;
if (FD_INVALID != d_rm->fildes)
{
/* The output will be flushed via iorm_flush() like in iorm_close.c. After this call returns,
* $X will be zero which will keep iorm_readfl() from attempting an iorm_wteol() in the fix mode
* after the file descriptor has been closed.
*/
iorm_flush(io_curr_device.in);
CLOSEFILE_RESET(d_rm->fildes, rc); /* resets "d_rm->fildes" to FD_INVALID */
}
} else
rts_error(VARLSTCNT(1) ERR_INVCTLMNE);
return;
}
uint4 iomt_reopen (io_desc *dv, unsigned short mode, int rewind)
{
uint4 status;
d_mt_struct *mt_ptr;
int res, fpos, rpos;
int4 mt_fileno, mt_blkno;
int rc;
error_def (ERR_MTIOERR);
mt_ptr = (d_mt_struct *) dv->dev_sp;
#ifdef DP
FPRINTF(stderr, ">> iomt_reopen\n");
FPRINTF(stderr, "iomt_wrtblk - \n");
FPRINTF(stderr, " - filepos %d\n", mt_ptr->filepos);
FPRINTF(stderr, " - recpos %d\n", mt_ptr->recpos);
#endif
if (!rewind)
{
fpos = mt_ptr->filepos;
rpos = mt_ptr->recpos;
}
CLOSEFILE_RESET(mt_ptr->access_id, rc); /* resets "mt_ptr->access_id" to FD_INVALID */
if (0 > rc)
{
PERROR("iomt_reopen");
rts_error (VARLSTCNT (5) ERR_MTIOERR, 2, dv->name->len, dv->name->dollar_io, errno);
}
#ifdef DP
FPRINTF(stderr, "Rewind: %d Drive going from ", rewind);
if (mt_ptr->mode == MT_M_WRITE)
FPRINTF(stderr, "WRITE ");
else
FPRINTF(stderr, "READ ");
if (mode == MT_M_WRITE)
FPRINTF(stderr, "to WRITE\n");
else
FPRINTF(stderr, "to READ\n");
#endif
if (mode == MT_M_WRITE)
{
mt_ptr->access_id = OPEN(dv->trans_name->dollar_io, O_WRONLY);
mt_ptr->mode = MT_M_WRITE;
} else
{
mt_ptr->access_id = OPEN(dv->trans_name->dollar_io, O_RDONLY | O_NDELAY);
mt_ptr->mode = MT_M_READ;
}
if (mt_ptr->access_id < 0)
{
status = MT_TAPERROR;
rts_error (VARLSTCNT (5) ERR_MTIOERR, 2, dv->name->len, dv->name->dollar_io, errno);
} else
status = SS_NORMAL;
#ifdef DP
FPRINTF(stderr, "iomt_wrtblk - \n");
FPRINTF(stderr, " - filepos %d\n", mt_ptr->filepos);
FPRINTF(stderr, " - recpos %d\n", mt_ptr->recpos);
FPRINTF(stderr, "<< iomt_reopen(%d)\n",mt_ptr->access_id);
#endif
return status;
}
void compile_source_file(unsigned short flen, char *faddr, boolean_t MFtIsReqd)
{
plength plen;
mval fstr, ret;
int i, rc;
unsigned char *p;
boolean_t wildcarded, dm_action;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
if (MAX_FBUFF < flen)
{
dec_err(VARLSTCNT(4) ERR_FILEPARSE, 2, flen, faddr);
TREF(dollar_zcstatus) = ERR_ERRORSUMMARY;
} else
{
object_file_des = FD_INVALID;
fstr.mvtype = MV_STR;
fstr.str.addr = faddr;
fstr.str.len = flen;
ESTABLISH(source_ch);
tt_so_do_once = FALSE;
zsrch_clr(STRM_COMP_SRC); /* Clear any existing search cache */
for (i = 0; ; i++)
{
plen.p.pint = op_fnzsearch(&fstr, STRM_COMP_SRC, 0, &ret);
if (!ret.str.len)
{
if (!i)
{
dec_err(VARLSTCNT(4) ERR_FILENOTFND, 2, fstr.str.len, fstr.str.addr);
TREF(dollar_zcstatus) = ERR_ERRORSUMMARY;
}
break;
}
assert(ret.mvtype == MV_STR);
assert(ret.str.len <= MAX_FBUFF);
source_name_len = ret.str.len;
memcpy(source_file_name, ret.str.addr, source_name_len);
source_file_name[source_name_len] = 0;
p = &source_file_name[plen.p.pblk.b_dir];
if ((plen.p.pblk.b_dir >= SIZEOF("/dev/") - 1) && !MEMCMP_LIT(source_file_name, "/dev/"))
tt_so_do_once = TRUE;
else if (MFtIsReqd && (plen.p.pblk.b_ext != 2 || ('M' != p[plen.p.pblk.b_name + 1]
&& 'm' != p[plen.p.pblk.b_name + 1])))
{ /* M filetype is required but not present */
dec_err(VARLSTCNT(4) ERR_FILEPARSE, 2, source_name_len, source_file_name);
TREF(dollar_zcstatus) = ERR_ERRORSUMMARY;
continue;
}
if (compiler_startup())
TREF(dollar_zcstatus) = ERR_ERRORSUMMARY;
if (FD_INVALID != object_file_des)
{
CLOSEFILE_RESET(object_file_des, rc); /* resets "object_file_des" to FD_INVALID */
if (-1 == rc)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(5) ERR_OBJFILERR, 2,
object_name_len, object_file_name, errno);
}
if (tt_so_do_once)
break;
}
REVERT;
}
}
/* Takes an entry from 'ipcs -am' and checks for its validity to be a GT.M replication segment.
* Returns TRUE if the shared memory segment is a valid GT.M replication segment
* (based on a check on some fields in the shared memory) else FALSE.
* If the segment belongs to GT.M, it returns the replication id of the segment
* by the second argument.
* Sets exit_stat to ERR_MUNOTALLSEC if appropriate.
*/
boolean_t validate_replpool_shm_entry(shm_parms *parm_buff, replpool_id_ptr_t replpool_id, int *exit_stat)
{
boolean_t remove_shmid, jnlpool_segment;
int fd;
repl_inst_hdr repl_instance;
sm_uc_ptr_t start_addr;
int save_errno, status, shmid;
struct shmid_ds shmstat;
char msgbuff[OUT_BUFF_SIZE], *instfilename;
if (NULL == parm_buff)
return FALSE;
/* Check for the bare minimum size of the replic shared segment that we expect */
/* if (parm_buff->sgmnt_siz < (SIZEOF(replpool_identifier) + MIN(MIN_JNLPOOL_SIZE, MIN_RECVPOOL_SIZE))) */
if (parm_buff->sgmnt_siz < MIN(MIN_JNLPOOL_SIZE, MIN_RECVPOOL_SIZE))
return FALSE;
if (IPC_PRIVATE != parm_buff->key)
return FALSE;
shmid = parm_buff->shmid;
/* we do not need to lock the shm for reading the rundown information as
* the other rundowns (if any) can also be allowed to share reading the
* same info concurrently.
*/
if (-1 == (sm_long_t)(start_addr = (sm_uc_ptr_t) do_shmat(shmid, 0, SHM_RND)))
return FALSE;
memcpy((void *)replpool_id, (void *)start_addr, SIZEOF(replpool_identifier));
instfilename = replpool_id->instfilename;
/* Even though we could be looking at a replication pool structure that has been created by an older version
* or newer version of GT.M, the format of the "replpool_identifier" structure is expected to be the same
* across all versions so we can safely dereference the "label" and "instfilename" fields in order to generate
* user-friendly error messages. Asserts for the layout are in "mu_rndwn_repl_instance" (not here) with a
* comment there as to why that location was chosen.
*/
if (memcmp(replpool_id->label, GDS_RPL_LABEL, GDS_LABEL_SZ - 1))
{
if (!memcmp(replpool_id->label, GDS_RPL_LABEL, GDS_LABEL_SZ - 3))
{
util_out_print("Cannot rundown replpool shmid = !UL as it has format !AD "
"created by !AD but this mupip is version and uses format !AD",
TRUE, shmid, GDS_LABEL_SZ - 1, replpool_id->label,
LEN_AND_STR(replpool_id->now_running), gtm_release_name_len, gtm_release_name,
GDS_LABEL_SZ - 1, GDS_RPL_LABEL);
*exit_stat = ERR_MUNOTALLSEC;
}
shmdt((void *)start_addr);
return FALSE;
}
assert(JNLPOOL_SEGMENT == replpool_id->pool_type || RECVPOOL_SEGMENT == replpool_id->pool_type);
if(JNLPOOL_SEGMENT != replpool_id->pool_type && RECVPOOL_SEGMENT != replpool_id->pool_type)
{
shmdt((void *)start_addr);
return FALSE;
}
jnlpool_segment = (JNLPOOL_SEGMENT == replpool_id->pool_type);
if (-1 == shmctl(shmid, IPC_STAT, &shmstat))
{
save_errno = errno;
assert(FALSE);/* we were able to attach to this shmid before so should be able to get stats on it */
util_out_print("!AD -> Error with shmctl for shmid = !UL",
TRUE, LEN_AND_STR(instfilename), shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
/* Check if instance filename reported in shared memory still exists. If not, clean this
* shared memory section without even invoking "mu_rndwn_repl_instance" as that expects
* the instance file to exist. Same case if shared memory points back to an instance file
* whose file header does not have this shmid.
*/
OPENFILE(instfilename, O_RDONLY, fd); /* check if we can open it */
msgbuff[0] = '\0';
remove_shmid = FALSE;
if (FD_INVALID == fd)
{
if (ENOENT == errno)
{
SNPRINTF(msgbuff, OUT_BUFF_SIZE, "File %s does not exist", instfilename);
if (1 < shmstat.shm_nattch)
{
PRINT_AND_SEND_REPLPOOL_FAILURE_MSG(msgbuff, replpool_id, shmid);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = TRUE;
} else
{ /* open() errored out e.g. due to file permissions. Log that */
save_errno = errno;
util_out_print("Cannot rundown replpool shmid !UL for instance file"
" !AD as open() on the file returned the following error",
TRUE, shmid, LEN_AND_STR(instfilename));
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
} else
{
LSEEKREAD(fd, 0, &repl_instance, SIZEOF(repl_inst_hdr), status);
if (0 != status)
{
save_errno = errno;
util_out_print("!AD -> Error with LSEEKREAD for shmid = !UL", TRUE,
LEN_AND_STR(instfilename), shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
if ((jnlpool_segment && (repl_instance.jnlpool_shmid != shmid))
|| (!jnlpool_segment && (repl_instance.recvpool_shmid != shmid)))
{
SNPRINTF(msgbuff, OUT_BUFF_SIZE, "%s SHMID (%d) in the instance file header does not match with the"
" one reported by \"ipcs\" command (%d)", jnlpool_segment ? "Journal Pool" : "Receive Pool",
jnlpool_segment ? repl_instance.jnlpool_shmid : repl_instance.recvpool_shmid, shmid);
if (1 < shmstat.shm_nattch)
{
PRINT_AND_SEND_REPLPOOL_FAILURE_MSG(msgbuff, replpool_id, shmid);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = TRUE;
}
CLOSEFILE_RESET(fd, status); /* resets "fd" to FD_INVALID */
}
shmdt((void *)start_addr);
if (remove_shmid)
{
assert('\0' != msgbuff[0]);
if (0 != shm_rmid(shmid))
{
save_errno = errno;
util_out_print("!AD -> Error removing shared memory for shmid = !UL",
TRUE, LEN_AND_STR(instfilename), shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
return FALSE;
}
PRINT_AND_SEND_SHMREMOVED_MSG(msgbuff, STRLEN(instfilename), instfilename, shmid);
*exit_stat = ERR_SHMREMOVED;
} else
*exit_stat = SS_NORMAL;
return TRUE;
}
void mubclnup(backup_reg_list *curr_ptr, clnup_stage stage)
{
sgmnt_addrs *csa;
backup_reg_list *ptr, *next;
uint4 status;
boolean_t had_lock;
unix_db_info *udi;
int rc;
assert(stage >= need_to_free_space && stage < num_of_clnup_stage);
free(stringpool.base);
switch(stage)
{
case need_to_rel_crit:
for (ptr = (backup_reg_list *)grlist; ptr != NULL && ptr != curr_ptr && ptr != (backup_reg_list *)halt_ptr;)
{
if (keep_going == ptr->not_this_time)
{
csa = &FILE_INFO(ptr->reg)->s_addrs;
DECR_INHIBIT_KILLS(csa->nl);
rel_crit(ptr->reg);
}
ptr = ptr->fPtr;
}
curr_ptr = (backup_reg_list *)halt_ptr;
/* Intentional Fall Through */
case need_to_del_tempfile:
for (ptr = (backup_reg_list *)grlist; ptr != NULL && ptr != curr_ptr;)
{
assert(3 == num_backup_proc_status); /* Ensure there are only 3 possible values for "ptr->not_this_time".
* The assert below and the following if check rely on this. */
assert((keep_going == ptr->not_this_time)
|| (give_up_before_create_tempfile == ptr->not_this_time)
|| (give_up_after_create_tempfile == ptr->not_this_time));
if (give_up_before_create_tempfile != ptr->not_this_time)
{
free(ptr->backup_hdr);
if (online)
{ /* Stop temporary file from growing if we made it active */
if (keep_going == ptr->not_this_time)
{
csa = &FILE_INFO(ptr->reg)->s_addrs;
csa->nl->nbb = BACKUP_NOT_IN_PROGRESS;
/* Make sure all running processes have a chance to see this backup
state change so they won't be trying to flush when we go to delete
the temporary files (mostly an issue on VMS).
This operation notifies other processes by:
1) Using a compswap lock with builtin memory barriers so other
processors know the memory state change.
2) Processes obtaining the lock after we release it will do their
own memory barrier operation and see the change.
3) By grabbing the lock, we are assured that anyone else getting the
lock after us will also be checking the errno flag AFTER getting the
lock (see backup_buffer_flush()) and see no flush is necessary.
*/
if (!(had_lock = shmpool_lock_held_by_us(ptr->reg)))
shmpool_lock_hdr(ptr->reg);
if (backup_interrupted && 0 == csa->shmpool_buffer->backup_errno)
/* Needs a non-zero value to stop the backup */
csa->shmpool_buffer->backup_errno = ERR_FORCEDHALT;
if (!had_lock)
shmpool_unlock_hdr(ptr->reg);
}
/* get rid of the temporary file */
if (ptr->backup_fd > 2)
{
CLOSEFILE_RESET(ptr->backup_fd, rc); /* resets "ptr" to FD_INVALID */
UNLINK(ptr->backup_tempfile);
}
} else /* defreeze the databases */
region_freeze(ptr->reg, FALSE, FALSE, FALSE);
}
ptr = ptr->fPtr;
}
/* Intentional fall through */
case need_to_free_space:
for (ptr = (backup_reg_list *)grlist; ptr != NULL;)
{
next = ptr->fPtr;
if (keep_going != ptr->not_this_time)
error_mupip = TRUE;
if (NULL != ptr->backup_file.addr)
free(ptr->backup_file.addr);
free(ptr);
ptr = next;
}
}
/* Release FTOK lock on the replication instance file if holding it */
assert((NULL == jnlpool.jnlpool_dummy_reg) || (NULL != mu_repl_inst_reg_list) || jnlpool_init_needed);
if ((NULL != mu_repl_inst_reg_list) && (NULL != jnlpool.jnlpool_dummy_reg) && jnlpool.jnlpool_dummy_reg->open)
{
udi = FILE_INFO(jnlpool.jnlpool_dummy_reg);
assert(NULL != udi);
if (NULL != udi)
{ /* See gv_rundown.c comment for why ftok_sem_release 2nd parameter is FALSE below */
if (udi->grabbed_ftok_sem)
ftok_sem_release(jnlpool.jnlpool_dummy_reg, FALSE, TRUE);
assert(!udi->grabbed_ftok_sem);
}
}
return;
}
void db_init_err_cleanup(boolean_t retry_dbinit)
{
unix_db_info *udi;
gd_segment *seg;
sgmnt_addrs *csa;
int rc, lcl_new_dbinit_ipc;
boolean_t ftok_counter_halted, access_counter_halted;
/* Here, we can not rely on the validity of csa->hdr because this function can be triggered anywhere in db_init().Because
* we don't have access to file header, we can not know if counters are disabled so we go by our best guess, not disabled,
* during cleanup.
*/
assert(NULL != db_init_region);
seg = db_init_region->dyn.addr;
udi = NULL;
if (NULL != seg->file_cntl)
udi = FILE_INFO(db_init_region);
if (NULL != udi)
{
if (FD_INVALID != udi->fd && !retry_dbinit)
CLOSEFILE_RESET(udi->fd, rc); /* resets "udi->fd" to FD_INVALID */
assert(FD_INVALID == udi->fd || retry_dbinit);
csa = &udi->s_addrs;
# ifdef _AIX
if ((NULL != csa->hdr) && (dba_mm == db_init_region->dyn.addr->acc_meth))
{
assert((NULL != csa->db_addrs[1]) && (csa->db_addrs[1] > csa->db_addrs[0]));
munmap((caddr_t)csa->db_addrs[0], (size_t)(csa->db_addrs[1] - csa->db_addrs[0]));
}
# endif
if (NULL != csa->jnl)
{
free(csa->jnl);
csa->jnl = NULL;
}
/* If shared memory is not available or if this is a VERMISMATCH error situation (where we do not know the exact
* position of csa->nl->ftok_counter_halted or if it even exists in the other version), we have to be pessimistic
* and assume the counters are halted. This avoids prematurely removing the semaphores.
*/
if ((NULL != csa->nl) && ((int)ERR_VERMISMATCH != SIGNAL))
{
ftok_counter_halted = csa->nl->ftok_counter_halted;
access_counter_halted = csa->nl->access_counter_halted;
shmdt((caddr_t)csa->nl);
csa->nl = (node_local_ptr_t)NULL;
} else
{
ftok_counter_halted = TRUE;
access_counter_halted = TRUE;
}
if (udi->shm_created && (INVALID_SHMID != udi->shmid))
{
shm_rmid(udi->shmid);
udi->shmid = INVALID_SHMID;
udi->shm_created = FALSE;
}
if (udi->sem_created && (INVALID_SEMID != udi->semid))
{
sem_rmid(udi->semid);
udi->semid = INVALID_SEMID;
udi->sem_created = FALSE;
udi->grabbed_access_sem = FALSE;
udi->counter_acc_incremented = FALSE;
}
if (udi->counter_acc_incremented && !access_counter_halted)
{
assert((INVALID_SEMID != udi->semid) && !db_init_region->read_only);
/* decrement the read-write sem */
do_semop(udi->semid, DB_COUNTER_SEM, -DB_COUNTER_SEM_INCR, SEM_UNDO | IPC_NOWAIT);
udi->counter_acc_incremented = FALSE;
}
if (udi->grabbed_access_sem)
{
do_semop(udi->semid, DB_CONTROL_SEM, -1, SEM_UNDO | IPC_NOWAIT); /* release the startup-shutdown sem */
udi->grabbed_access_sem = FALSE;
}
if (udi->grabbed_ftok_sem)
ftok_sem_release(db_init_region, udi->counter_ftok_incremented && !ftok_counter_halted, TRUE);
else if (udi->counter_ftok_incremented && !ftok_counter_halted)
do_semop(udi->ftok_semid, DB_COUNTER_SEM, -DB_COUNTER_SEM_INCR, SEM_UNDO | IPC_NOWAIT);
udi->counter_ftok_incremented = FALSE;
udi->grabbed_ftok_sem = FALSE;
if (!IS_GTCM_GNP_SERVER_IMAGE && !retry_dbinit) /* gtcm_gnp_server reuses file_cntl */
{
free(seg->file_cntl->file_info);
free(seg->file_cntl);
seg->file_cntl = NULL;
}
}
/* Enable interrupts in case we are here with intrpt_ok_state == INTRPT_IN_GVCST_INIT due to an rts error.
* Normally we would have the new state stored in "prev_intrpt_state" but that is not possible here because
* the corresponding DEFER_INTERRUPTS happened in gvcst_init.c (a different function) so we have an assert
* there that the previous state was INTRPT_OK_TO_INTERRUPT and use that instead of prev_intrpt_state here.
*/
if (!retry_dbinit)
ENABLE_INTERRUPTS(INTRPT_IN_GVCST_INIT, INTRPT_OK_TO_INTERRUPT);
}
int gtcm_bgn_net(omi_conn_ll *cll)
{
extern int4 omi_nxact, omi_nerrs, omi_brecv, omi_bsent;
omi_fd fd;
int i;
int save_errno;
int rc;
#ifdef NET_TCP
struct servent *se;
unsigned short port;
char port_buffer[NI_MAXSERV];
#endif /* defined(NET_TCP) */
#ifdef BSD_TCP
struct addrinfo *ai_ptr, hints;
const boolean_t reuseaddr = TRUE;
int errcode;
#else /* defined(BSD_TCP) */
#ifdef SYSV_TCP
struct t_bind *bind;
#endif /* defined(SYSV_TCP) */
#endif /* !defined(BSD_TCP) */
/* The linked list of connections */
cll->head = cll->tail = (omi_conn *)0;
/* The statistics */
cll->stats.conn = cll->stats.clos = cll->stats.disc = 0;
cll->st_cn.bytes_recv = 0;
cll->st_cn.bytes_send = 0;
cll->st_cn.start = 0;
for (i = 0; i < OMI_OP_MAX; i++)
cll->st_cn.xact[i] = 0;
for (i = 0; i < OMI_ER_MAX; i++)
cll->st_cn.errs[i] = 0;
omi_nxact = omi_nerrs = omi_brecv = omi_bsent = 0;
/* Fall back on a compile time constant */
if (!omi_service)
omi_service = SRVC_NAME;
#ifdef NET_TCP
/* NET_TCP is defined only when BSD_TCP is defined or SYSV_TCP is defined, but SYSV_TCP is never defined (a bug?)
* so we move the code of obtaining port information from service down to #ifdef BSD_TCP
*/
#ifdef SYSV_TCP
GTMASSERT;
#endif
#endif /* defined(NET_TCP) */
#ifdef BSD_TCP
/* Create a socket always tries IPv6 first */
SERVER_HINTS(hints, ((GTM_IPV6_SUPPORTED && !ipv4_only) ? AF_INET6 : AF_INET));
if ((fd = socket(hints.ai_family, SOCK_STREAM, 0)) < 0)
{
if ((fd = socket(AF_INET, SOCK_STREAM, 0)) < 0)
{
save_errno = errno;
return save_errno;
}
hints.ai_family = AF_INET;
}
/* Bind an address to the socket */
if (0 != (errcode = getaddrinfo(NULL, omi_service, &hints, &ai_ptr)))
{
RTS_ERROR_ADDRINFO(NULL, ERR_GETADDRINFO, errcode);
return errcode;
}
if (ISDIGIT_ASCII(*omi_service))
port = atoi(omi_service);
else
{
if (0 != (errcode = getnameinfo(ai_ptr->ai_addr, ai_ptr->ai_addrlen, NULL, 0, port_buffer,
NI_MAXSERV, NI_NUMERICSERV)))
{
assert(FALSE);
RTS_ERROR_ADDRINFO(NULL, ERR_GETNAMEINFO, errcode);
return errcode;
}
port = atoi(port_buffer);
}
/* Reuse a specified address */
if (port && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (const void *)&reuseaddr, SIZEOF(reuseaddr)) < 0)
{
save_errno = errno;
CLOSEFILE_RESET(fd, rc); /* resets "fd" to FD_INVALID */
return save_errno;
}
if (bind(fd, ai_ptr->ai_addr, ai_ptr->ai_addrlen) < 0)
{
save_errno = errno;
CLOSEFILE_RESET(fd, rc); /* resets "fd" to FD_INVALID */
return save_errno;
}
/* Initialize the listen queue */
if (listen(fd, 5) < 0)
{
save_errno = errno;
CLOSEFILE_RESET(fd, rc); /* resets "fd" to FD_INVALID */
return save_errno;
}
/* set up raw socket for use with pinging option */
if (ping_keepalive)
psock = init_ping();
/* Store the file descriptor away for use later */
cll->nve = fd;
OMI_DBG_STMP;
OMI_DBG((omi_debug, "%s: socket registered at port %d\n", SRVR_NAME, (int)port));
#ifdef GTCM_RC
OMI_DBG((omi_debug, "RC server ID %d, Process ID %d\n", rc_server_id, omi_pid));
#endif
if (authenticate)
OMI_DBG((omi_debug, "Password verification on OMI connections enabled.\n"));
if (!one_conn_per_inaddr)
OMI_DBG((omi_debug, "Multiple connections from the same internet address allowed.\n"));
if (psock > 0)
OMI_DBG((omi_debug, "Keepalive option (-ping) enabled.\n"));
return 0;
#else /* defined(BSD_TCP) */
#ifdef SYSV_TCP
GTMASSERT;
if ((fd = t_open(SYSV_TCP, O_RDWR, NULL)) < 0)
{
save_errno = errno;
return save_errno;
}
if (!(bind = (struct t_bind *)t_alloc(fd, T_BIND, T_ALL)))
{
save_errno = errno;
(void) t_close(fd);
return save_errno;
}
bind->qlen = 5;
bind->addr.len = 0;
bind->addr.buf = 0;
if (t_bind(fd, bind, bind) < 0)
{
save_errno = errno;
(void) t_free(bind, T_BIND);
(void) t_close(fd);
return save_errno;
}
/* Store the file descriptor away for use later */
cll->nve = fd;
OMI_DBG_STMP;
OMI_DBG((omi_debug, "%s: socket registered at port %d\n", SRVR_NAME, (int)port));
#ifdef GTCM_RC
OMI_DBG((omi_debug, "RC server ID %d\n", rc_server_id));
#endif
return 0;
#else /* defined(SYSV_TCP) */
cll->nve = FD_INVALID;
return -1;
#endif /* !defined(SYSV_TCP) */
#endif /* !defined(BSD_TCP) */
}
void iob_read(BFILE *bf, char *buf, int nbytes)
{
int4 nread, nreread;
int rc;
error_def(ERR_IOEOF);
if (bf->write_mode)
{
iob_flush(bf);
bf->write_mode = FALSE;
}
#ifdef DEBUG_IOB
PRINTF("iob_read:\tiob_read(%x, %x, %d), bf->remaining = %d\n",
bf, buf, nbytes, bf->remaining);
#endif
while (nbytes > bf->remaining)
{
/* copy bytes remaining in buffer */
memcpy(buf, bf->bptr, bf->remaining);
nbytes -= bf->remaining;
buf += bf->remaining;
/* refill */
DOREADRL(bf->fd, bf->buf, bf->bufsiz, nread);
#ifdef DEBUG_IOB
PRINTF("iob_read:\t\tread(%d, %x, %d) = %d\n", bf->fd, bf->buf,
bf->bufsiz, nread);
#endif
bf->bptr = bf->buf;
bf->remaining = nread;
if (nread == -1)
{
rts_error(VARLSTCNT(1) errno);
return;
}
else if (nread == 0 || nread % bf->blksiz)
{
CLOSEFILE_RESET(bf->fd, rc); /* resets "bf->fd" to FD_INVALID */
rts_error(VARLSTCNT(1) ERR_IOEOF);
/* if we continued from here, assume that this is a magnetic
tape and we have loaded the next volume. Re-open and
finish the read operation.
*/
while (FD_INVALID == (bf->fd = OPEN3(bf->path,bf->oflag,bf->mode)))
rts_error(VARLSTCNT(1) errno);
DOREADRL(bf->fd, bf->buf + nread, bf->bufsiz - nread, nreread);
#ifdef DEBUG_IOB
PRINTF("iob_read:\t\tread(%d, %x, %d) = %d\n", bf->fd, bf->buf,
bf->bufsiz, nread);
#endif
if (nreread < bf->bufsiz - nread)
{
rts_error(VARLSTCNT(1) errno);
return;
}
bf->remaining = nread;
}
}
memcpy(buf, bf->bptr, nbytes);
bf->bptr += nbytes;
bf->remaining -= nbytes;
}
int gtcm_cn_acpt(omi_conn_ll *cll, int now) /* now --> current time in seconds */
{
int i;
omi_conn *cptr;
omi_fd fd;
int rc;
char *tmp_time;
#ifdef BSD_TCP
GTM_SOCKLEN_TYPE sln;
struct sockaddr_storage sas;
int optsize;
const boolean_t keepalive = TRUE;
/* Accept the connection from the network layer */
sln = SIZEOF(sas);
if ((fd = accept(cll->nve, (struct sockaddr *)&sas, (GTM_SOCKLEN_TYPE *)&sln)) < 0)
return -1;
#endif /* defined(BSD_TCP) */
/* Build the client data structure */
if (!(cptr = (omi_conn *)malloc(SIZEOF(omi_conn))) || !(cptr->buff = (char *)malloc(OMI_BUFSIZ)))
{
if (cptr)
free(cptr);
CLOSEFILE_RESET(fd, rc); /* resets "fd" to FD_INVALID */
return -1;
}
/* Initialize the connection structure */
cptr->next = (omi_conn *)0;
cptr->fd = fd;
cptr->ping_cnt = 0;
cptr->timeout = now + conn_timeout;
cptr->bsiz = OMI_BUFSIZ;
cptr->bptr = cptr->buff;
cptr->xptr = (char *)0;
cptr->blen = 0;
cptr->exts = 0;
cptr->state = OMI_ST_DISC;
cptr->ga = (ga_struct *)0; /* struct gd_addr_struct */
cptr->of = (oof_struct *) malloc(SIZEOF(struct rc_oflow));
memset(cptr->of, 0, SIZEOF(struct rc_oflow));
cptr->pklog = FD_INVALID;
/* Initialize the statistics */
memcpy(&cptr->stats.sas, &sas, sln);
cptr->stats.ai.ai_addr = (struct sockaddr *)&cptr->stats.sas;
cptr->stats.ai.ai_addrlen = sln;
cptr->stats.bytes_recv = 0;
cptr->stats.bytes_send = 0;
cptr->stats.start = time((time_t *)0);
for (i = 0; i < OMI_OP_MAX; i++)
cptr->stats.xact[i] = 0;
for (i = 0; i < OMI_ER_MAX; i++)
cptr->stats.errs[i] = 0;
/* if we only allowing one connection per internet address, close any existing ones with the same addr. */
if (one_conn_per_inaddr)
{
omi_conn *this, *prev;
for (prev = NULL, this = cll->head; this; prev = this, this = this->next)
{
if (0 == memcmp((sockaddr_ptr)(&this->stats.sas), (sockaddr_ptr)&sas, sln))
{
if (cll->tail == this)
cll->tail = cptr;
if (prev)
prev->next = cptr;
else
cll->head = cptr;
cptr->next = this->next;
OMI_DBG_STMP;
OMI_DBG((omi_debug, "%s: dropping old connection to %s\n",
SRVR_NAME, gtcm_hname(&cptr->stats.ai)));
gtcm_cn_disc(this, cll);
break;
}
}
/* not found - add to the end of the list */
if (!this)
{
if (cll->tail)
{
cll->tail->next = cptr;
cll->tail = cptr;
} else
cll->head = cll->tail = cptr;
}
} else
{
/* Insert the client into the list of connections */
if (cll->tail)
{
cll->tail->next = cptr;
cll->tail = cptr;
} else
cll->head = cll->tail = cptr;
}
cptr->stats.id = ++cll->stats.conn;
DEBUG_ONLY(
if (omi_pklog)
{
int errno_save;
char pklog[1024];
(void)SPRINTF(pklog, "%s.%04d", omi_pklog, cptr->stats.id);
if (INV_FD_P((cptr->pklog = OPEN3(pklog, O_WRONLY|O_CREAT|O_APPEND|O_TRUNC, 0644))))
{
errno_save = errno;
OMI_DBG_STMP;
OMI_DBG((omi_debug, "%s: unable to open packet log \"%s\"\n\t%s\n",
SRVR_NAME, pklog, STRERROR(errno_save)));
}
}
)
int gtmrecv_end1(boolean_t auto_shutdown)
{
int4 strm_idx;
int exit_status, idx, status, save_errno;
int fclose_res, rc;
seq_num log_seqno, log_seqno1, jnlpool_seqno, jnlpool_strm_seqno[MAX_SUPPL_STRMS];
uint4 savepid;
exit_status = gtmrecv_end_helpers(TRUE);
exit_status = gtmrecv_endupd();
log_seqno = recvpool.recvpool_ctl->jnl_seqno;
log_seqno1 = recvpool.upd_proc_local->read_jnl_seqno;
strm_idx = recvpool.gtmrecv_local->strm_index;
/* Detach from receive pool */
recvpool.gtmrecv_local->shutdown = exit_status;
recvpool.gtmrecv_local->recv_serv_pid = 0;
if (0 > SHMDT(recvpool.recvpool_ctl))
repl_log(stderr, TRUE, TRUE, "Error detaching from Receive Pool : %s\n", REPL_STR_ERROR);
recvpool.recvpool_ctl = NULL;
assert((NULL != jnlpool_ctl) && (jnlpool_ctl == jnlpool.jnlpool_ctl));
if (NULL != jnlpool.jnlpool_ctl)
{ /* Reset fields that might have been initialized by the receiver server after connecting to the primary.
* It is ok not to hold the journal pool lock while updating jnlpool_ctl fields since this will be the
* only process updating those fields.
*/
jnlpool.jnlpool_ctl->primary_instname[0] = '\0';
jnlpool.jnlpool_ctl->gtmrecv_pid = 0;
jnlpool_seqno = jnlpool.jnlpool_ctl->jnl_seqno;
for (idx = 0; idx < MAX_SUPPL_STRMS; idx++)
jnlpool_strm_seqno[idx] = jnlpool.jnlpool_ctl->strm_seqno[idx];
/* Also take this opportunity to detach from the journal pool except in the auto_shutdown case. This is because
* the fields "jnlpool_ctl->repl_inst_filehdr->recvpool_semid" and "jnlpool_ctl->repl_inst_filehdr->recvpool_shmid"
* need to be reset by "gtmrecv_jnlpool_reset" (called from "gtmrecv_shutdown") which is invoked a little later.
*/
if (!auto_shutdown)
{
JNLPOOL_SHMDT(status, save_errno);
if (0 > status)
repl_log(stderr, TRUE, TRUE, "Error detaching from Journal Pool : %s\n", STRERROR(save_errno));
jnlpool.jnlpool_ctl = jnlpool_ctl = NULL;
jnlpool.repl_inst_filehdr = NULL;
jnlpool.gtmsrc_lcl_array = NULL;
jnlpool.gtmsource_local_array = NULL;
jnlpool.jnldata_base = NULL;
pool_init = FALSE;
}
} else
jnlpool_seqno = 0;
gtmrecv_free_msgbuff();
gtmrecv_free_filter_buff();
recvpool.recvpool_ctl = NULL;
/* Close the connection with the Receiver */
if (FD_INVALID != gtmrecv_listen_sock_fd)
CLOSEFILE_RESET(gtmrecv_listen_sock_fd, rc); /* resets "gtmrecv_listen_sock_fd" to FD_INVALID */
if (FD_INVALID != gtmrecv_sock_fd)
CLOSEFILE_RESET(gtmrecv_sock_fd, rc); /* resets "gtmrecv_sock_fd" to FD_INVALID */
repl_log(gtmrecv_log_fp, TRUE, FALSE, "REPL INFO - Current Jnlpool Seqno : %llu\n", jnlpool_seqno);
for (idx = 0; idx < MAX_SUPPL_STRMS; idx++)
{
if (jnlpool_strm_seqno[idx])
repl_log(gtmrecv_log_fp, TRUE, FALSE, "REPL INFO - Stream # %d : Current Jnlpool Stream Seqno : %llu\n",
idx, jnlpool_strm_seqno[idx]);
}
if (0 < strm_idx)
repl_log(gtmrecv_log_fp, TRUE, FALSE, "REPL INFO - Receiver server has Stream # %d\n", strm_idx);
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL INFO - Current Update process Read Seqno : %llu\n", log_seqno1);
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL INFO - Current Receive Pool Seqno : %llu\n", log_seqno);
/* If log_seqno/log_seqno1 is 0, then do not decrement it as that will be interpreted as a huge positive seqno. Keep it 0 */
if (log_seqno)
log_seqno--;
if (log_seqno1)
log_seqno1--;
repl_log(gtmrecv_log_fp, TRUE, FALSE, "REPL INFO - Last Recvd Seqno : %llu Jnl Total : %llu Msg Total : %llu\n",
log_seqno, repl_recv_data_processed, repl_recv_data_recvd);
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL INFO - Last Seqno processed by update process : %llu\n", log_seqno1);
gtm_event_log_close();
if (gtmrecv_filter & EXTERNAL_FILTER)
repl_stop_filter();
if (auto_shutdown)
return (exit_status);
else
gtmrecv_exit(exit_status - NORMAL_SHUTDOWN);
return -1; /* This will never get executed, added to make compiler happy */
}
int gtmrecv_poll_actions1(int *pending_data_len, int *buff_unprocessed, unsigned char *buffp)
{
static int report_cnt = 1;
static int next_report_at = 1;
static boolean_t send_xoff = FALSE;
static boolean_t xoff_sent = FALSE;
static seq_num send_seqno;
static boolean_t log_draining_msg = FALSE;
static boolean_t send_badtrans = FALSE;
static boolean_t send_cmp2uncmp = FALSE;
static boolean_t upd_shut_too_early_logged = FALSE;
static time_t last_reap_time = 0;
repl_msg_t xoff_msg;
repl_badtrans_msg_t bad_trans_msg;
boolean_t alert = FALSE, info = FALSE;
int return_status;
gd_region *region_top;
unsigned char *msg_ptr; /* needed for REPL_{SEND,RECV}_LOOP */
int tosend_len, sent_len, sent_this_iter; /* needed for REPL_SEND_LOOP */
int torecv_len, recvd_len, recvd_this_iter; /* needed for REPL_RECV_LOOP */
int status; /* needed for REPL_{SEND,RECV}_LOOP */
int temp_len, pending_msg_size;
int upd_start_status, upd_start_attempts;
int buffered_data_len;
int upd_exit_status;
seq_num temp_send_seqno;
boolean_t bad_trans_detected = FALSE, onln_rlbk_flg_set = FALSE;
uint4 jnl_status;
recvpool_ctl_ptr_t recvpool_ctl;
upd_proc_local_ptr_t upd_proc_local;
gtmrecv_local_ptr_t gtmrecv_local;
upd_helper_ctl_ptr_t upd_helper_ctl;
pid_t waitpid_res;
int4 msg_type, msg_len;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
recvpool_ctl = recvpool.recvpool_ctl;
upd_proc_local = recvpool.upd_proc_local;
gtmrecv_local = recvpool.gtmrecv_local;
upd_helper_ctl = recvpool.upd_helper_ctl;
jnl_status = 0;
if (SHUTDOWN == gtmrecv_local->shutdown)
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Shutdown signalled\n");
gtmrecv_end(); /* Won't return */
}
/* Reset report_cnt and next_report_at to 1 when a new upd proc is forked */
if ((1 == report_cnt) || (report_cnt == next_report_at))
{
/* A comment on the usage of NO_SHUTDOWN below for the alert variable. Since upd_proc_local->upd_proc_shutdown is
* a shared memory field (and could be concurrently changed by either the receiver server or the update process),
* we want to make sure it is the same value BEFORE and AFTER checking whether the update process is alive or not.
* If it is not NO_SHUTDOWN (i.e. is SHUTDOWN or NORMAL_SHUTDOWN or ABNORMAL_SHUTDOWN) it has shut down due to
* an external request so we do want to send out a false update-process-is-not-alive alert.
*/
if ((alert = ((NO_SHUTDOWN == upd_proc_local->upd_proc_shutdown) && (SRV_DEAD == is_updproc_alive())
&& (NO_SHUTDOWN == upd_proc_local->upd_proc_shutdown)))
|| (info = (((NORMAL_SHUTDOWN == upd_proc_local->upd_proc_shutdown)
|| (ABNORMAL_SHUTDOWN == upd_proc_local->upd_proc_shutdown)) && (SRV_DEAD == is_updproc_alive()))))
{
if (alert)
repl_log(gtmrecv_log_fp, TRUE, TRUE,
"ALERT : Receiver Server detected that Update Process is not ALIVE\n");
else
repl_log(gtmrecv_log_fp, TRUE, TRUE,
"INFO : Update process not running due to user initiated shutdown\n");
if (1 == report_cnt)
{
send_xoff = TRUE;
QWASSIGN(recvpool_ctl->old_jnl_seqno, recvpool_ctl->jnl_seqno);
QWASSIGNDW(recvpool_ctl->jnl_seqno, 0);
/* Even though we have identified that the update process is NOT alive, a waitpid on the update
* process PID is necessary so that the system doesn't leave any zombie process lying around.
* This is possible since any child process that dies without the parent doing a waitpid on it
* will be defunct unless the parent dies at which point the "init" process takes the role of
* the parent and invokes waitpid to remove the zombies.
* NOTE: It is possible that the update process was killed before the receiver server got a
* chance to record it's PID in the recvpool.upd_proc_local structure. In such a case, don't
* invoke waitpid as that will block us (receiver server) if this instance of the receiver
* server was started with helper processes.
*/
if (0 < upd_proc_local->upd_proc_pid)
{
WAITPID(upd_proc_local->upd_proc_pid, &upd_exit_status, 0, waitpid_res);
/* Since the update process as part of its shutdown does NOT reset the upd_proc_pid, reset
* it here ONLY if the update process was NOT kill -9ed. This is needed because receiver
* server as part of its shutdown relies on this field (upd_proc_pid) to determine if the
* update process was cleanly shutdown or was kill -9ed.
*/
if (!alert)
upd_proc_local->upd_proc_pid = 0;
}
upd_proc_local->bad_trans = FALSE; /* No point in doing bad transaction processing */
upd_proc_local->onln_rlbk_flg = FALSE; /* No point handling online rollback */
}
gtmrecv_wait_for_jnl_seqno = TRUE;
REPL_DPRINT1(
"gtmrecv_poll_actions : Setting gtmrecv_wait_for_jnl_seqno to TRUE because of upd crash/shutdown\n");
next_report_at *= GTMRECV_NEXT_REPORT_FACTOR;
report_cnt++;
}
} else
report_cnt++;
/* Check if REPL_CMP2UNCMP or REPL_BADTRANS message needs to be sent */
if (upd_proc_local->onln_rlbk_flg)
{ /* Update process detected an online rollback and is requesting us to restart the connection. But before that, send
* REPL_XOFF source side and drain the replication pipe
*/
onln_rlbk_flg_set = TRUE;
send_xoff = TRUE;
}
else if (!send_cmp2uncmp && gtmrecv_send_cmp2uncmp)
{
send_xoff = TRUE;
send_seqno = recvpool_ctl->jnl_seqno;
send_cmp2uncmp = TRUE;
} else if (!send_badtrans && upd_proc_local->bad_trans)
{
send_xoff = TRUE;
send_seqno = upd_proc_local->read_jnl_seqno;
send_badtrans = TRUE;
bad_trans_detected = TRUE;
} else if (!upd_proc_local->bad_trans && send_badtrans && 1 != report_cnt)
{
send_badtrans = FALSE;
bad_trans_detected = FALSE;
}
if (send_xoff && !xoff_sent)
{ /* Send XOFF_ACK_ME if the receiver has a connection to the source. Do not attempt to send it if we dont even
* know the endianness of the remote side. In that case, we are guaranteed no initial handshake occurred and
* so no point sending the XOFF too. This saves us lots of trouble in case of cross-endian replication connections.
*/
assert((FD_INVALID != gtmrecv_sock_fd) || repl_connection_reset);
if ((FD_INVALID != gtmrecv_sock_fd) && remote_side->endianness_known)
{
send_seqno = upd_proc_local->read_jnl_seqno;
if (!remote_side->cross_endian)
{
xoff_msg.type = REPL_XOFF_ACK_ME;
xoff_msg.len = MIN_REPL_MSGLEN;
memcpy((uchar_ptr_t)&xoff_msg.msg[0], (uchar_ptr_t)&send_seqno, SIZEOF(seq_num));
} else
{
xoff_msg.type = GTM_BYTESWAP_32(REPL_XOFF_ACK_ME);
xoff_msg.len = GTM_BYTESWAP_32(MIN_REPL_MSGLEN);
temp_send_seqno = GTM_BYTESWAP_64(send_seqno);
memcpy((uchar_ptr_t)&xoff_msg.msg[0], (uchar_ptr_t)&temp_send_seqno, SIZEOF(seq_num));
}
REPL_SEND_LOOP(gtmrecv_sock_fd, &xoff_msg, MIN_REPL_MSGLEN, FALSE, >mrecv_poll_immediate)
; /* Empty Body */
if (SS_NORMAL != status)
{
if (REPL_CONN_RESET(status) && EREPL_SEND == repl_errno)
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Connection reset while sending XOFF_ACK_ME. "
"Status = %d ; %s\n", status, STRERROR(status));
repl_close(>mrecv_sock_fd);
repl_connection_reset = TRUE;
xoff_sent = FALSE;
send_badtrans = FALSE;
} else if (EREPL_SEND == repl_errno)
rts_error(VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error sending XOFF msg due to BAD_TRANS or UPD crash/shutdown. "
"Error in send"), status);
else
{
assert(EREPL_SELECT == repl_errno);
rts_error(VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error sending XOFF msg due to BAD_TRANS or UPD crash/shutdown. "
"Error in select"), status);
}
} else
{
xoff_sent = TRUE;
log_draining_msg = TRUE;
}
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL_XOFF_ACK_ME sent due to upd shutdown/crash or bad trans "
"or ONLINE_ROLLBACK\n");
send_xoff = FALSE;
} else
{ /* Connection has been lost OR initial handshake needs to happen again, so no point sending XOFF/BADTRANS */
send_xoff = FALSE;
send_badtrans = FALSE;
}
}
/* Drain pipe */
if (xoff_sent)
{
if (log_draining_msg)
{ /* avoid multiple logs per instance */
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL INFO - Draining replication pipe due to %s\n",
send_cmp2uncmp ? "CMP2UNCMP" : (send_badtrans ? "BAD_TRANS" :
(onln_rlbk_flg_set ? "ONLINE_ROLLBACK" : "UPD shutdown/crash")));
log_draining_msg = FALSE;
}
if (0 != *buff_unprocessed)
{ /* Throw away the current contents of the buffer */
buffered_data_len = ((*pending_data_len <= *buff_unprocessed) ? *pending_data_len : *buff_unprocessed);
*buff_unprocessed -= buffered_data_len;
buffp += buffered_data_len;
*pending_data_len -= buffered_data_len;
REPL_DPRINT2("gtmrecv_poll_actions : (1) Throwing away %d bytes from old buffer while draining\n",
buffered_data_len);
assert(remote_side->endianness_known); /* only then is remote_side->cross_endian reliable */
while (REPL_MSG_HDRLEN <= *buff_unprocessed)
{
assert(0 == (((unsigned long)buffp) % REPL_MSG_ALIGN));
msg_len = ((repl_msg_ptr_t)buffp)->len;
msg_type = ((repl_msg_ptr_t)buffp)->type;
if (remote_side->cross_endian)
{
msg_len = GTM_BYTESWAP_32(msg_len);
msg_type = GTM_BYTESWAP_32(msg_type);
}
msg_type = (msg_type & REPL_TR_CMP_MSG_TYPE_MASK);
assert((REPL_TR_CMP_JNL_RECS == msg_type) || (0 == (msg_len % REPL_MSG_ALIGN)));
*pending_data_len = ROUND_UP2(msg_len, REPL_MSG_ALIGN);
buffered_data_len = ((*pending_data_len <= *buff_unprocessed) ?
*pending_data_len : *buff_unprocessed);
*buff_unprocessed -= buffered_data_len;
buffp += buffered_data_len;
*pending_data_len -= buffered_data_len;
REPL_DPRINT3("gtmrecv_poll_actions : (1) Throwing away message of "
"type %d and length %d from old buffer while draining\n", msg_type, buffered_data_len);
}
if (0 < *buff_unprocessed)
{
memmove((unsigned char *)gtmrecv_msgp, buffp, *buff_unprocessed);
REPL_DPRINT2("gtmrecv_poll_actions : Incomplete header of length %d while draining\n",
*buff_unprocessed);
}
}
status = SS_NORMAL;
if (0 != *buff_unprocessed || 0 == *pending_data_len)
{ /* Receive the header of a message */
assert(REPL_MSG_HDRLEN > *buff_unprocessed); /* so we dont pass negative length in REPL_RECV_LOOP */
REPL_RECV_LOOP(gtmrecv_sock_fd, ((unsigned char *)gtmrecv_msgp) + *buff_unprocessed,
(REPL_MSG_HDRLEN - *buff_unprocessed), FALSE, >mrecv_poll_interval)
; /* Empty Body */
if (SS_NORMAL == status)
{
assert(remote_side->endianness_known); /* only then is remote_side->cross_endian reliable */
if (!remote_side->cross_endian)
{
msg_len = gtmrecv_msgp->len;
msg_type = gtmrecv_msgp->type;
} else
{
msg_len = GTM_BYTESWAP_32(gtmrecv_msgp->len);
msg_type = GTM_BYTESWAP_32(gtmrecv_msgp->type);
}
msg_type = (msg_type & REPL_TR_CMP_MSG_TYPE_MASK);
assert((REPL_TR_CMP_JNL_RECS == msg_type) || (0 == (msg_len % REPL_MSG_ALIGN)));
msg_len = ROUND_UP2(msg_len, REPL_MSG_ALIGN);
REPL_DPRINT3("gtmrecv_poll_actions : Received message of type %d and length %d while draining\n",
msg_type, msg_len);
}
}
if ((SS_NORMAL == status) && (0 != *buff_unprocessed || 0 == *pending_data_len) && (REPL_XOFF_ACK == msg_type))
{ /* Receive the rest of the XOFF_ACK msg and signal the drain as complete */
REPL_RECV_LOOP(gtmrecv_sock_fd, gtmrecv_msgp, (MIN_REPL_MSGLEN - REPL_MSG_HDRLEN), FALSE,
>mrecv_poll_interval)
; /* Empty Body */
if (SS_NORMAL == status)
{
repl_log(gtmrecv_log_fp, TRUE, TRUE,
"REPL INFO - XOFF_ACK received. Drained replication pipe completely\n");
upd_shut_too_early_logged = FALSE;
xoff_sent = FALSE;
return_status = STOP_POLL;
}
} else if (SS_NORMAL == status)
{ /* Drain the rest of the message */
if (0 < *pending_data_len)
{
pending_msg_size = *pending_data_len;
REPL_DPRINT2("gtmrecv_poll_actions : (2) Throwing away %d bytes from pipe\n", pending_msg_size);
} else
{
pending_msg_size = msg_len - REPL_MSG_HDRLEN;
REPL_DPRINT3("gtmrecv_poll_actions : (2) Throwing away message of "
"type %d and length %d from pipe\n", msg_type, msg_len);
}
for ( ; SS_NORMAL == status && 0 < pending_msg_size; pending_msg_size -= gtmrecv_max_repl_msglen)
{
temp_len = (pending_msg_size < gtmrecv_max_repl_msglen)? pending_msg_size : gtmrecv_max_repl_msglen;
REPL_RECV_LOOP(gtmrecv_sock_fd, gtmrecv_msgp, temp_len, FALSE, >mrecv_poll_interval)
; /* Empty Body */
}
*buff_unprocessed = 0; *pending_data_len = 0;
if (SS_NORMAL == status && info && !upd_shut_too_early_logged)
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "ALERT : User initiated shutdown of Update Process done "
"when there was data in the replication pipe\n");
upd_shut_too_early_logged = TRUE;
}
return_status = CONTINUE_POLL;
}
if (SS_NORMAL != status)
{
if (EREPL_RECV == repl_errno)
{
if (REPL_CONN_RESET(status))
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Connection reset while receiving XOFF_ACK. "
"Status = %d ; %s\n", status, STRERROR(status));
repl_close(>mrecv_sock_fd);
repl_connection_reset = TRUE;
xoff_sent = FALSE;
send_badtrans = FALSE;
return_status = STOP_POLL;
} else
rts_error(VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error while draining replication pipe. Error in recv"), status);
} else
{
assert(EREPL_SELECT == repl_errno);
rts_error(VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error while draining replication pipe. Error in select"), status);
}
}
} else
return_status = STOP_POLL;
/* Like was done before for the XOFF_ACK_ME message, send a BADTRANS/CMP2UNCMP message only if we know
* the endianness of the other side. If not, no point in sending one anyways and saves us trouble in
* case of cross-endian replication connections.
*/
if ((STOP_POLL == return_status) && (send_badtrans || send_cmp2uncmp)
&& (FD_INVALID != gtmrecv_sock_fd) && remote_side->endianness_known)
{ /* Send REPL_BADTRANS or REPL_CMP2UNCMP message */
if (!remote_side->cross_endian)
{
bad_trans_msg.type = send_cmp2uncmp ? REPL_CMP2UNCMP : REPL_BADTRANS;
bad_trans_msg.len = MIN_REPL_MSGLEN;
bad_trans_msg.start_seqno = send_seqno;
} else
{
bad_trans_msg.type = send_cmp2uncmp ? GTM_BYTESWAP_32(REPL_CMP2UNCMP) : GTM_BYTESWAP_32(REPL_BADTRANS);
bad_trans_msg.len = GTM_BYTESWAP_32(MIN_REPL_MSGLEN);
bad_trans_msg.start_seqno = GTM_BYTESWAP_64(send_seqno);
}
REPL_SEND_LOOP(gtmrecv_sock_fd, &bad_trans_msg, bad_trans_msg.len, FALSE, >mrecv_poll_immediate)
; /* Empty Body */
if (SS_NORMAL == status)
{
if (send_cmp2uncmp)
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL_CMP2UNCMP message sent with seqno %llu\n", send_seqno);
else
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL_BADTRANS message sent with seqno %llu\n", send_seqno);
} else
{
if (REPL_CONN_RESET(status) && EREPL_SEND == repl_errno)
{
if (send_cmp2uncmp)
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Connection reset while sending REPL_CMP2UNCMP. "
"Status = %d ; %s\n", status, STRERROR(status));
} else
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Connection reset while sending REPL_BADTRANS. "
"Status = %d ; %s\n", status, STRERROR(status));
}
repl_close(>mrecv_sock_fd);
repl_connection_reset = TRUE;
return_status = STOP_POLL;
} else if (EREPL_SEND == repl_errno)
rts_error(VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error sending REPL_BADTRANS/REPL_CMP2UNCMP. Error in send"), status);
else
{
assert(EREPL_SELECT == repl_errno);
rts_error(VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error sending REPL_BADTRANS/REPL_CMP2UNCMP. Error in select"), status);
}
}
send_badtrans = FALSE;
if (send_cmp2uncmp)
{
REPL_DPRINT1("gtmrecv_poll_actions : Setting gtmrecv_wait_for_jnl_seqno to TRUE because this receiver"
"server requested a fall-back from compressed to uncompressed operation\n");
gtmrecv_wait_for_jnl_seqno = TRUE;/* set this to TRUE to break out and go back to a fresh "do_main_loop" */
gtmrecv_bad_trans_sent = TRUE;
gtmrecv_send_cmp2uncmp = FALSE;
send_cmp2uncmp = FALSE;
}
}
if ((upd_proc_local->bad_trans && bad_trans_detected) || onln_rlbk_flg_set
|| (UPDPROC_START == upd_proc_local->start_upd) && (1 != report_cnt))
{
if (UPDPROC_START == upd_proc_local->start_upd)
{
assert(is_updproc_alive() != SRV_ALIVE);
upd_proc_local->upd_proc_shutdown = NO_SHUTDOWN;
}
recvpool_ctl->wrapped = FALSE;
recvpool_ctl->write_wrap = recvpool_ctl->recvpool_size;
recvpool_ctl->write = 0;
/* Reset last_rcvd_histinfo, last_valid_histinfo etc. as they reflect context from unprocessed data
* in the receive pool and those are no longer valid because we have drained the receive pool.
*/
GTMRECV_CLEAR_CACHED_HISTINFO(recvpool.recvpool_ctl, jnlpool, jnlpool_ctl, INSERT_STRM_HISTINFO_FALSE);
if (UPDPROC_START == upd_proc_local->start_upd)
{
/* Attempt starting the update process */
for (upd_start_attempts = 0;
UPDPROC_START_ERR == (upd_start_status = gtmrecv_upd_proc_init(FALSE)) &&
GTMRECV_MAX_UPDSTART_ATTEMPTS > upd_start_attempts;
upd_start_attempts++)
{
if (EREPL_UPDSTART_SEMCTL == repl_errno || EREPL_UPDSTART_BADPATH == repl_errno)
{
gtmrecv_autoshutdown();
} else if (EREPL_UPDSTART_FORK == repl_errno)
{
/* Couldn't start up update now, can try later */
LONG_SLEEP(GTMRECV_WAIT_FOR_PROC_SLOTS);
continue;
} else if (EREPL_UPDSTART_EXEC == repl_errno)
{
/* In forked child, could not exec, should exit */
gtmrecv_exit(ABNORMAL_SHUTDOWN);
}
}
if (UPDPROC_STARTED == (upd_proc_local->start_upd = upd_start_status))
{
REPL_DPRINT1("gtmrecv_poll_actions : Setting gtmrecv_wait_for_jnl_seqno to TRUE because of "
"upd restart\n");
gtmrecv_wait_for_jnl_seqno = TRUE;
report_cnt = next_report_at = 1;
if (send_xoff && (FD_INVALID == gtmrecv_sock_fd))
{
/* Update start command was issued before connection was established,
* no point in sending XOFF. */
send_xoff = FALSE;
}
} else
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "%d failed attempts to fork update process. Try later\n",
upd_start_attempts);
}
} else
{
gtmrecv_wait_for_jnl_seqno = TRUE;/* set this to TRUE to break out and go back to a fresh "do_main_loop" */
if (onln_rlbk_flg_set)
{
assert(NULL != jnlpool_ctl);
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Closing connection due to ONLINE ROLLBACK\n");
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL INFO - Current Jnlpool Seqno : %llu\n",
jnlpool_ctl->jnl_seqno);
repl_log(gtmrecv_log_fp, TRUE, TRUE, "REPL INFO - Current Receive Pool Seqno : %llu\n",
recvpool_ctl->jnl_seqno);
repl_close(>mrecv_sock_fd);
repl_connection_reset = TRUE;
xoff_sent = FALSE;
send_badtrans = FALSE;
upd_proc_local->onln_rlbk_flg = FALSE;
/* Before restarting afresh, sync the online rollback cycles. This way any future grab_lock that
* we do after restarting should not realize an unhandled online rollback. For receiver, it is
* just syncing the journal pool cycles as the databases are not opened. But, to be safe, grab
* the lock and sync the cycles.
*/
GRAB_LOCK(jnlpool.jnlpool_dummy_reg, GRAB_LOCK_ONLY);
SYNC_ONLN_RLBK_CYCLES;
rel_lock(jnlpool.jnlpool_dummy_reg);
return_status = STOP_POLL;
recvpool_ctl->jnl_seqno = 0;
} else
{
REPL_DPRINT1("gtmrecv_poll_actions : Setting gtmrecv_wait_for_jnl_seqno to TRUE because bad trans"
"sent\n");
gtmrecv_bad_trans_sent = TRUE;
upd_proc_local->bad_trans = FALSE;
recvpool_ctl->jnl_seqno = upd_proc_local->read_jnl_seqno;
}
}
}
if ((0 == *pending_data_len) && (0 != gtmrecv_local->changelog))
{
if (gtmrecv_local->changelog & REPLIC_CHANGE_LOGINTERVAL)
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Changing log interval from %u to %u\n",
log_interval, gtmrecv_local->log_interval);
log_interval = gtmrecv_local->log_interval;
gtmrecv_reinit_logseqno(); /* will force a LOG on the first recv following the interval change */
}
if (gtmrecv_local->changelog & REPLIC_CHANGE_LOGFILE)
{
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Changing log file to %s\n", gtmrecv_local->log_file);
repl_log_init(REPL_GENERAL_LOG, >mrecv_log_fd, NULL, gtmrecv_local->log_file, NULL);
repl_log_fd2fp(>mrecv_log_fp, gtmrecv_log_fd);
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Change log to %s successful\n",gtmrecv_local->log_file);
}
upd_proc_local->changelog = gtmrecv_local->changelog; /* Pass changelog request to the update process */
/* NOTE: update process and receiver each ignore any setting specific to the other (REPLIC_CHANGE_UPD_LOGINTERVAL,
* REPLIC_CHANGE_LOGINTERVAL) */
gtmrecv_local->changelog = 0;
}
if (0 == *pending_data_len && !gtmrecv_logstats && gtmrecv_local->statslog)
{
gtmrecv_logstats = TRUE;
repl_log_init(REPL_STATISTICS_LOG, >mrecv_log_fd, >mrecv_statslog_fd, gtmrecv_local->log_file,
gtmrecv_local->statslog_file);
repl_log_fd2fp(>mrecv_statslog_fp, gtmrecv_statslog_fd);
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Starting stats log to %s\n", gtmrecv_local->statslog_file);
repl_log(gtmrecv_statslog_fp, TRUE, TRUE, "Begin statistics logging\n");
} else if (0 == *pending_data_len && gtmrecv_logstats && !gtmrecv_local->statslog)
{
gtmrecv_logstats = FALSE;
repl_log(gtmrecv_log_fp, TRUE, TRUE, "Stopping stats log\n");
/* Force all data out to the file before closing the file */
repl_log(gtmrecv_statslog_fp, TRUE, TRUE, "End statistics logging\n");
CLOSEFILE_RESET(gtmrecv_statslog_fd, status); /* resets "gtmrecv_statslog_fd" to FD_INVALID */
/* We need to FCLOSE because a later open() in repl_log_init() might return the same file descriptor as the one
* that we just closed. In that case, FCLOSE done in repl_log_fd2fp() affects the newly opened file and
* FDOPEN will fail returning NULL for the file pointer. So, we close both the file descriptor and file pointer.
* Note the same problem does not occur with GENERAL LOG because the current log is kept open while opening
* the new log and hence the new file descriptor will be different (we keep the old log file open in case there
* are errors during DUPing. In such a case, we do not switch the log file, but keep the current one).
* We can FCLOSE the old file pointer later in repl_log_fd2fp() */
FCLOSE(gtmrecv_statslog_fp, status);
gtmrecv_statslog_fp = NULL;
}
if (0 == *pending_data_len)
{
if (upd_helper_ctl->start_helpers)
{
gtmrecv_helpers_init(upd_helper_ctl->start_n_readers, upd_helper_ctl->start_n_writers);
upd_helper_ctl->start_helpers = FALSE;
}
if (HELPER_REAP_NONE != (status = upd_helper_ctl->reap_helpers) ||
(double)GTMRECV_REAP_HELPERS_INTERVAL <= difftime(gtmrecv_now, last_reap_time))
{
gtmrecv_reap_helpers(HELPER_REAP_WAIT == status);
last_reap_time = gtmrecv_now;
}
}
return (return_status);
}
cmi_status_t cmi_init(cmi_descriptor *tnd, unsigned char tnr,
void (*err)(struct NTD *, struct CLB *, cmi_reason_t reason),
void (*crq)(struct CLB *),
bool (*acc)(struct CLB *),
void (*urg)(struct CLB *, unsigned char data),
size_t pool_size,
size_t usr_size,
size_t mbl)
{
cmi_status_t status = SS_NORMAL;
char *envvar;
struct protoent *p;
unsigned short myport;
struct sockaddr_in in;
sigset_t oset;
int on = 1;
int rval, rc, save_errno;
status = cmj_netinit();
if (CMI_ERROR(status))
return status;
status = cmj_getsockaddr(tnd, &in);
if (CMI_ERROR(status))
return status;
ntd_root->pool_size = pool_size;
ntd_root->usr_size = usr_size;
ntd_root->mbl = mbl;
p = getprotobyname(GTCM_SERVER_PROTOCOL);
endprotoent();
if (!p)
return CMI_NETFAIL;
/* create the listening socket */
ntd_root->listen_fd = socket(AF_INET, SOCK_STREAM, p->p_proto);
if (FD_INVALID == ntd_root->listen_fd)
return errno;
/* make sure we can re-run quickly w/o reuse problems */
status = setsockopt(ntd_root->listen_fd, SOL_SOCKET, SO_REUSEADDR,
(void*)&on, SIZEOF(on));
if (-1 == status)
{
save_errno = errno;
CLOSEFILE_RESET(ntd_root->listen_fd, rc); /* resets "ntd_root->listen_fd" to FD_INVALID */
return save_errno;
}
status = bind(ntd_root->listen_fd, (struct sockaddr*)&in, SIZEOF(in));
if (-1 == status)
{
save_errno = errno;
CLOSEFILE_RESET(ntd_root->listen_fd, rc); /* resets "ntd_root->listen_fd" to FD_INVALID */
return save_errno;
}
status = cmj_setupfd(ntd_root->listen_fd);
if (CMI_ERROR(status))
{
CLOSEFILE_RESET(ntd_root->listen_fd, rc); /* resets "ntd_root->listen_fd" to FD_INVALID */
return status;
}
SIGPROCMASK(SIG_BLOCK, &ntd_root->mutex_set, &oset, rc);
rval = listen(ntd_root->listen_fd, MAX_CONN_IND);
if (-1 == rval)
{
save_errno = errno;
CLOSEFILE_RESET(ntd_root->listen_fd, rc); /* resets "ntd_root->listen_fd" to FD_INVALID */
SIGPROCMASK(SIG_SETMASK, &oset, NULL, rc);
return save_errno;
}
status = cmj_set_async(ntd_root->listen_fd);
if (CMI_ERROR(status))
{
CLOSEFILE_RESET(ntd_root->listen_fd, rc); /* resets "ntd_root->listen_fd" to FD_INVALID */
SIGPROCMASK(SIG_SETMASK, &oset, NULL, rc);
return status;
}
FD_SET(ntd_root->listen_fd, &ntd_root->rs);
FD_SET(ntd_root->listen_fd, &ntd_root->es);
ntd_root->err = err;
ntd_root->crq = crq;
ntd_root->acc = acc;
ntd_root->urg = urg;
if (ntd_root->listen_fd > ntd_root->max_fd)
ntd_root->max_fd = ntd_root->listen_fd;
cmj_housekeeping(); /* will establish listening pools */
SIGPROCMASK(SIG_SETMASK, &oset, NULL, rc);
return SS_NORMAL;
}
/* Initialize communication stuff */
int gtmrecv_comm_init(in_port_t port)
{
struct addrinfo *ai_ptr = NULL, hints;
const int enable_reuseaddr = 1;
struct linger disable_linger = {0, 0};
int rc;
int errcode;
char port_buffer[NI_MAXSERV];
int port_buffer_len;
int temp_sock_fd;
int af;
if (FD_INVALID != gtmrecv_listen_sock_fd) /* Initialization done already */
return (0);
/* Create the socket used for communicating with primary */
af = ((GTM_IPV6_SUPPORTED && !ipv4_only) ? AF_INET6 : AF_INET);
if (FD_INVALID == (temp_sock_fd = socket(af, SOCK_STREAM, IPPROTO_TCP)))
{
af = AF_INET;
if (FD_INVALID == (temp_sock_fd = socket(af, SOCK_STREAM, IPPROTO_TCP)))
{
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error with receiver server socket create"), ERRNO);
return (-1);
}
}
/* Make it known to the world that you are ready for a Source Server */
SERVER_HINTS(hints, af);
SPRINTF(port_buffer, "%hu", port);
if (0 != (errcode = getaddrinfo(NULL, port_buffer, &hints, &ai_ptr)))
{
CLOSEFILE(temp_sock_fd, rc);
RTS_ERROR_ADDRINFO_CTX(NULL, ERR_GETADDRINFO, errcode, "FAILED in obtaining IP address on receiver server.");
return -1;
}
gtmrecv_listen_sock_fd = temp_sock_fd;
if (0 > setsockopt(gtmrecv_listen_sock_fd, SOL_SOCKET, SO_LINGER, (const void *)&disable_linger, SIZEOF(disable_linger)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error with receiver server listen socket disable linger"), ERRNO);
if (0 > setsockopt(gtmrecv_listen_sock_fd, SOL_SOCKET, SO_REUSEADDR, (const void *)&enable_reuseaddr,
SIZEOF(enable_reuseaddr)))
{
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Error with receiver server listen socket enable reuseaddr"), ERRNO);
}
if (0 > BIND(gtmrecv_listen_sock_fd, ai_ptr->ai_addr, ai_ptr->ai_addrlen))
{
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Could not bind local address"), ERRNO);
CLOSEFILE_RESET(gtmrecv_listen_sock_fd, rc); /* resets "gtmrecv_listen_sock_fd" to FD_INVALID */
return (-1);
}
if (0 > listen(gtmrecv_listen_sock_fd, 5))
{
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_REPLCOMM, 0, ERR_TEXT, 2,
RTS_ERROR_LITERAL("Could not listen"), ERRNO);
CLOSEFILE_RESET(gtmrecv_listen_sock_fd, rc); /* resets "gtmrecv_listen_sock_fd" to FD_INVALID */
return (-1);
}
return (0);
}
int4 gds_rundown(void)
{
boolean_t canceled_dbsync_timer, canceled_flush_timer, ok_to_write_pfin;
boolean_t have_standalone_access, ipc_deleted, err_caught;
boolean_t is_cur_process_ss_initiator, remove_shm, vermismatch, we_are_last_user, we_are_last_writer, is_mm;
boolean_t unsafe_last_writer;
char time_str[CTIME_BEFORE_NL + 2]; /* for GET_CUR_TIME macro */
gd_region *reg;
int save_errno, status, rc;
int4 semval, ftok_semval, sopcnt, ftok_sopcnt;
short crash_count;
sm_long_t munmap_len;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
node_local_ptr_t cnl;
struct shmid_ds shm_buf;
struct sembuf sop[2], ftok_sop[2];
uint4 jnl_status;
unix_db_info *udi;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
shm_snapshot_t *ss_shm_ptr;
uint4 ss_pid, onln_rlbk_pid, holder_pid;
boolean_t was_crit;
boolean_t safe_mode; /* Do not flush or take down shared memory. */
boolean_t bypassed_ftok = FALSE, bypassed_access = FALSE, may_bypass_ftok, inst_is_frozen,
ftok_counter_halted,
access_counter_halted;
int secshrstat;
intrpt_state_t prev_intrpt_state;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
jnl_status = 0;
reg = gv_cur_region; /* Local copy */
/* early out for cluster regions
* to avoid tripping the assert below.
* Note:
* This early out is consistent with VMS. It has been
* noted that all of the gtcm assignments
* to gv_cur_region should use the TP_CHANGE_REG
* macro. This would also avoid the assert problem
* and should be done eventually.
*/
if (dba_cm == reg->dyn.addr->acc_meth)
return EXIT_NRM;
udi = FILE_INFO(reg);
csa = &udi->s_addrs;
csd = csa->hdr;
assert(csa == cs_addrs && csd == cs_data);
if ((reg->open) && (dba_usr == csd->acc_meth))
{
change_reg();
gvusr_rundown();
return EXIT_NRM;
}
/* If the process has standalone access, it has udi->grabbed_access_sem set to TRUE at this point. Note that down in a local
* variable as the udi->grabbed_access_sem is set to TRUE even for non-standalone access below and hence we can't rely on
* that later to determine if the process had standalone access or not when it entered this function. We need to guarantee
* that none else access database file header when semid/shmid fields are reset. We already have created ftok semaphore in
* db_init or, mu_rndwn_file and did not remove it. So just lock it. We do it in blocking mode.
*/
have_standalone_access = udi->grabbed_access_sem; /* process holds standalone access */
DEFER_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
ESTABLISH_NORET(gds_rundown_ch, err_caught);
if (err_caught)
{
REVERT;
WITH_CH(gds_rundown_ch, gds_rundown_err_cleanup(have_standalone_access), 0);
ENABLE_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
DEBUG_ONLY(ok_to_UNWIND_in_exit_handling = FALSE);
return EXIT_ERR;
}
assert(reg->open); /* if we failed to open, dbinit_ch should have taken care of proper clean up */
assert(!reg->opening); /* see comment above */
assert((dba_bg == csd->acc_meth) || (dba_mm == csd->acc_meth));
is_mm = (dba_bg != csd->acc_meth);
assert(!csa->hold_onto_crit || (csa->now_crit && jgbl.onlnrlbk));
/* If we are online rollback, we should already be holding crit and should release it only at the end of this module. This
* is usually done by noting down csa->now_crit in a local variable (was_crit) and using it whenever we are about to
* grab_crit. But, there are instances (like mupip_set_journal.c) where we grab_crit but invoke gds_rundown without any
* preceeding rel_crit. Such code relies on the fact that gds_rundown does rel_crit unconditionally (to get locks to a known
* state). So, augment csa->now_crit with jgbl.onlnrlbk to track if we can rel_crit unconditionally or not in gds_rundown.
*/
was_crit = (csa->now_crit && jgbl.onlnrlbk);
/* Cancel any pending flush timer for this region by this task */
canceled_flush_timer = FALSE;
canceled_dbsync_timer = FALSE;
CANCEL_DB_TIMERS(reg, csa, canceled_flush_timer, canceled_dbsync_timer);
we_are_last_user = FALSE;
inst_is_frozen = IS_REPL_INST_FROZEN && REPL_ALLOWED(csa->hdr);
if (!csa->persistent_freeze)
region_freeze(reg, FALSE, FALSE, FALSE);
if (!was_crit)
{
rel_crit(reg); /* get locks to known state */
mutex_cleanup(reg);
}
/* The only process that can invoke gds_rundown while holding access control semaphore is RECOVER/ROLLBACK. All the others
* (like MUPIP SET -FILE/MUPIP EXTEND would have invoked db_ipcs_reset() before invoking gds_rundown (from
* mupip_exit_handler). The only exception is when these processes encounter a terminate signal and they reach
* mupip_exit_handler while holding access control semaphore. Assert accordingly.
*/
assert(!have_standalone_access || mupip_jnl_recover || process_exiting);
/* If we have standalone access, then ensure that a concurrent online rollback cannot be running at the same time as it
* needs the access control lock as well. The only expection is we are online rollback and currently running down.
*/
cnl = csa->nl;
onln_rlbk_pid = cnl->onln_rlbk_pid;
assert(!have_standalone_access || mupip_jnl_recover || !onln_rlbk_pid || !is_proc_alive(onln_rlbk_pid, 0));
if (!have_standalone_access)
{
if (-1 == (ftok_semval = semctl(udi->ftok_semid, DB_COUNTER_SEM, GETVAL))) /* Check # of procs counted on FTOK */
{
save_errno = errno;
assert(FALSE);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get ftok_semval"), CALLFROM, errno);
}
may_bypass_ftok = CAN_BYPASS(ftok_semval, csd, inst_is_frozen); /* Do we need a blocking wait? */
/* We need to guarantee that no one else access database file header when semid/shmid fields are reset.
* We already have created ftok semaphore in db_init or mu_rndwn_file and did not remove it. So just lock it.
*/
if (!ftok_sem_lock(reg, may_bypass_ftok))
{
if (may_bypass_ftok)
{ /* We did a non-blocking wait. It's ok to proceed without locking */
bypassed_ftok = TRUE;
holder_pid = semctl(udi->ftok_semid, DB_CONTROL_SEM, GETPID);
if ((uint4)-1 == holder_pid)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get holder_pid"),
CALLFROM, errno);
if (!IS_GTM_IMAGE) /* MUMPS processes should not flood syslog with bypass messages. */
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_RESRCINTRLCKBYPAS, 10,
LEN_AND_STR(gtmImageNames[image_type].imageName), process_id, LEN_AND_LIT("FTOK"),
REG_LEN_STR(reg), DB_LEN_STR(reg), holder_pid);
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_TEXT, 2,
LEN_AND_LIT("FTOK bypassed at rundown"));
}
} else
{ /* We did a blocking wait but something bad happened. */
FTOK_TRACE(csa, csa->ti->curr_tn, ftok_ops_lock, process_id);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
}
}
sop[0].sem_num = DB_CONTROL_SEM; sop[0].sem_op = 0; /* Wait for 0 */
sop[1].sem_num = DB_CONTROL_SEM; sop[1].sem_op = 1; /* Lock */
sopcnt = 2;
sop[0].sem_flg = sop[1].sem_flg = SEM_UNDO | IPC_NOWAIT; /* Don't wait the first time thru */
SEMOP(udi->semid, sop, sopcnt, status, NO_WAIT);
if (0 != status)
{
save_errno = errno;
/* Check # of processes counted on access sem. */
if (-1 == (semval = semctl(udi->semid, DB_COUNTER_SEM, GETVAL)))
{
assert(FALSE);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get semval"), CALLFROM, errno);
}
bypassed_access = CAN_BYPASS(semval, csd, inst_is_frozen) || onln_rlbk_pid || csd->file_corrupt;
/* Before attempting again in the blocking mode, see if the holding process is an online rollback.
* If so, it is likely we won't get the access control semaphore anytime soon. In that case, we
* are better off skipping rundown and continuing with sanity cleanup and exit.
*/
holder_pid = semctl(udi->semid, DB_CONTROL_SEM, GETPID);
if ((uint4)-1 == holder_pid)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get holder_pid"), CALLFROM, errno);
if (!bypassed_access)
{ /* We couldn't get it in one shot-- see if we already have it */
if (holder_pid == process_id)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(5) MAKE_MSG_INFO(ERR_CRITSEMFAIL), 2, DB_LEN_STR(reg),
ERR_RNDWNSEMFAIL);
REVERT;
ENABLE_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
assert(FALSE);
return EXIT_ERR;
}
if (EAGAIN != save_errno)
{
assert(FALSE);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMOP on access control semaphore"),
CALLFROM, save_errno);
}
sop[0].sem_flg = sop[1].sem_flg = SEM_UNDO; /* Try again - blocking this time */
SEMOP(udi->semid, sop, 2, status, FORCED_WAIT);
if (-1 == status) /* We couldn't get it at all.. */
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMOP on access control semaphore"),
CALLFROM, errno);
} else if (!IS_GTM_IMAGE)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_RESRCINTRLCKBYPAS, 10,
LEN_AND_STR(gtmImageNames[image_type].imageName), process_id,
LEN_AND_LIT("access control"), REG_LEN_STR(reg), DB_LEN_STR(reg), holder_pid);
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_TEXT, 2,
LEN_AND_LIT("Access control bypassed at rundown"));
}
udi->grabbed_access_sem = !bypassed_access;
}
} /* else we we hold the access control semaphore and therefore have standalone access. We do not release it now - we
* release it later in mupip_exit_handler.c. Since we already hold the access control semaphore, we don't need the
* ftok semaphore and trying it could cause deadlock
*/
/* Note that in the case of online rollback, "udi->grabbed_access_sem" (and in turn "have_standalone_access") is TRUE.
* But there could be other processes still having the database open so we cannot safely reset the halted fields.
*/
if (have_standalone_access && !jgbl.onlnrlbk)
csd->ftok_counter_halted = csd->access_counter_halted = FALSE;
ftok_counter_halted = csd->ftok_counter_halted;
access_counter_halted = csd->access_counter_halted;
/* If we bypassed any of the semaphores, activate safe mode.
* Also, if the replication instance is frozen and this db has replication turned on (which means
* no flushes of dirty buffers to this db can happen while the instance is frozen) activate safe mode.
*/
ok_to_write_pfin = !(bypassed_access || bypassed_ftok || inst_is_frozen);
safe_mode = !ok_to_write_pfin || ftok_counter_halted || access_counter_halted;
/* At this point we are guaranteed no one else is doing a db_init/rundown as we hold the access control semaphore */
assert(csa->ref_cnt); /* decrement private ref_cnt before shared ref_cnt decrement. */
csa->ref_cnt--; /* Currently journaling logic in gds_rundown() in VMS relies on this order to detect last writer */
assert(!csa->ref_cnt);
--cnl->ref_cnt;
if (memcmp(cnl->now_running, gtm_release_name, gtm_release_name_len + 1))
{ /* VERMISMATCH condition. Possible only if DSE */
assert(dse_running);
vermismatch = TRUE;
} else
vermismatch = FALSE;
if (-1 == shmctl(udi->shmid, IPC_STAT, &shm_buf))
{
save_errno = errno;
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown shmctl"), CALLFROM, save_errno);
} else
we_are_last_user = (1 == shm_buf.shm_nattch) && !vermismatch && !safe_mode;
/* recover => one user except ONLINE ROLLBACK, or standalone with frozen instance */
assert(!have_standalone_access || we_are_last_user || jgbl.onlnrlbk || inst_is_frozen);
if (-1 == (semval = semctl(udi->semid, DB_COUNTER_SEM, GETVAL)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get semval"), CALLFROM, errno);
/* There's one writer left and I am it */
assert(reg->read_only || semval >= 0);
unsafe_last_writer = (DB_COUNTER_SEM_INCR == semval) && (FALSE == reg->read_only) && !vermismatch;
we_are_last_writer = unsafe_last_writer && !safe_mode;
assert(!we_are_last_writer || !safe_mode);
assert(!we_are_last_user || !safe_mode);
/* recover + R/W region => one writer except ONLINE ROLLBACK, or standalone with frozen instance, leading to safe_mode */
assert(!(have_standalone_access && !reg->read_only) || we_are_last_writer || jgbl.onlnrlbk || inst_is_frozen);
GTM_WHITE_BOX_TEST(WBTEST_ANTIFREEZE_JNLCLOSE, we_are_last_writer, 1); /* Assume we are the last writer to invoke wcs_flu */
if (!have_standalone_access && (-1 == (ftok_semval = semctl(udi->ftok_semid, DB_COUNTER_SEM, GETVAL))))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get ftok_semval"), CALLFROM, errno);
if (NULL != csa->ss_ctx)
ss_destroy_context(csa->ss_ctx);
/* SS_MULTI: If multiple snapshots are supported, then we have to run through each of the snapshots */
assert(1 == MAX_SNAPSHOTS);
ss_shm_ptr = (shm_snapshot_ptr_t)SS_GETSTARTPTR(csa);
ss_pid = ss_shm_ptr->ss_info.ss_pid;
is_cur_process_ss_initiator = (process_id == ss_pid);
if (ss_pid && (is_cur_process_ss_initiator || we_are_last_user))
{
/* Try getting snapshot crit latch. If we don't get latch, we won't hang for eternity and will skip
* doing the orphaned snapshot cleanup. It will be cleaned up eventually either by subsequent MUPIP
* INTEG or by a MUPIP RUNDOWN.
*/
if (ss_get_lock_nowait(reg) && (ss_pid == ss_shm_ptr->ss_info.ss_pid)
&& (is_cur_process_ss_initiator || !is_proc_alive(ss_pid, 0)))
{
ss_release(NULL);
ss_release_lock(reg);
}
}
/* If cnl->donotflush_dbjnl is set, it means mupip recover/rollback was interrupted and therefore we need not flush
* shared memory contents to disk as they might be in an inconsistent state. Moreover, any more flushing will only cause
* future rollback to undo more journal records (PBLKs). In this case, we will go ahead and remove shared memory (without
* flushing the contents) in this routine. A reissue of the recover/rollback command will restore the database to a
* consistent state.
*/
if (!cnl->donotflush_dbjnl && !reg->read_only && !vermismatch)
{ /* If we had an orphaned block and were interrupted, set wc_blocked so we can invoke wcs_recover. Do it ONLY
* if there is NO concurrent online rollback running (as we need crit to set wc_blocked)
*/
if (csa->wbuf_dqd && !is_mm)
{ /* If we had an orphaned block and were interrupted, mupip_exit_handler will invoke secshr_db_clnup which
* will clear this field and so we should never come to gds_rundown with a non-zero wbuf_dqd. The only
* exception is if we are recover/rollback in which case gds_rundown (from mur_close_files) is invoked
* BEFORE secshr_db_clnup in mur_close_files.
* Note: It is NOT possible for online rollback to reach here with wbuf_dqd being non-zero. This is because
* the moment we apply the first PBLK, we stop all interrupts and hence can never be interrupted in
* wcs_wtstart or wcs_get_space. Assert accordingly.
*/
assert(mupip_jnl_recover && !jgbl.onlnrlbk && !safe_mode);
if (!was_crit)
grab_crit(reg);
SET_TRACEABLE_VAR(cnl->wc_blocked, TRUE);
BG_TRACE_PRO_ANY(csa, wcb_gds_rundown);
send_msg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_WCBLOCKED, 6, LEN_AND_LIT("wcb_gds_rundown"),
process_id, &csa->ti->curr_tn, DB_LEN_STR(reg));
csa->wbuf_dqd = 0;
wcs_recover(reg);
BG_TRACE_PRO_ANY(csa, lost_block_recovery);
if (!was_crit)
rel_crit(reg);
}
if (JNL_ENABLED(csd) && IS_GTCM_GNP_SERVER_IMAGE)
originator_prc_vec = NULL;
/* If we are the last writing user, then everything must be flushed */
if (we_are_last_writer)
{ /* Time to flush out all of our buffers */
assert(!safe_mode);
if (is_mm)
{
MM_DBFILEXT_REMAP_IF_NEEDED(csa, reg);
cnl->remove_shm = TRUE;
}
if (cnl->wc_blocked && jgbl.onlnrlbk)
{ /* if the last update done by online rollback was not committed in the normal code-path but was
* completed by secshr_db_clnup, wc_blocked will be set to TRUE. But, since online rollback never
* invokes grab_crit (since csa->hold_onto_crit is set to TRUE), wcs_recover is never invoked. This
* could result in the last update never getting flushed to the disk and if online rollback happened
* to be the last writer then the shared memory will be flushed and removed and the last update will
* be lost. So, force wcs_recover if we find ourselves in such a situation. But, wc_blocked is
* possible only if phase1 or phase2 errors are induced using white box test cases
*/
assert(WB_COMMIT_ERR_ENABLED);
wcs_recover(reg);
}
/* Note WCSFLU_SYNC_EPOCH ensures the epoch is synced to the journal and indirectly
* also ensures that the db is fsynced. We don't want to use it in the calls to
* wcs_flu() from t_end() and tp_tend() since we can defer it to out-of-crit there.
* In this case, since we are running down, we don't have any such option.
*/
cnl->remove_shm = wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SYNC_EPOCH);
/* Since we_are_last_writer, we should be guaranteed that wcs_flu() did not change csd, (in
* case of MM for potential file extension), even if it did a grab_crit(). Therefore, make
* sure that's true.
*/
assert(csd == csa->hdr);
assert(0 == memcmp(csd->label, GDS_LABEL, GDS_LABEL_SZ - 1));
} else if (((canceled_flush_timer && (0 > cnl->wcs_timers)) || canceled_dbsync_timer) && !inst_is_frozen)
{ /* canceled pending db or jnl flush timers - flush database and journal buffers to disk */
if (!was_crit)
grab_crit(reg);
/* we need to sync the epoch as the fact that there is no active pending flush timer implies
* there will be noone else who will flush the dirty buffers and EPOCH to disk in a timely fashion
*/
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SYNC_EPOCH);
if (!was_crit)
rel_crit(reg);
assert((dba_mm == cs_data->acc_meth) || (csd == cs_data));
csd = cs_data; /* In case this is MM and wcs_flu() remapped an extended database, reset csd */
}
/* Do rundown journal processing after buffer flushes since they require jnl to be open */
if (JNL_ENABLED(csd))
{ /* the following tp_change_reg() is not needed due to the assert csa == cs_addrs at the beginning
* of gds_rundown(), but just to be safe. To be removed by 2002!! --- nars -- 2001/04/25.
*/
tp_change_reg(); /* call this because jnl_ensure_open checks cs_addrs rather than gv_cur_region */
jpc = csa->jnl;
jbp = jpc->jnl_buff;
if (jbp->fsync_in_prog_latch.u.parts.latch_pid == process_id)
{
assert(FALSE);
COMPSWAP_UNLOCK(&jbp->fsync_in_prog_latch, process_id, 0, LOCK_AVAILABLE, 0);
}
if (jbp->io_in_prog_latch.u.parts.latch_pid == process_id)
{
assert(FALSE);
COMPSWAP_UNLOCK(&jbp->io_in_prog_latch, process_id, 0, LOCK_AVAILABLE, 0);
}
if ((((NOJNL != jpc->channel) && !JNL_FILE_SWITCHED(jpc))
|| we_are_last_writer && (0 != cnl->jnl_file.u.inode)) && ok_to_write_pfin)
{ /* We need to close the journal file cleanly if we have the latest generation journal file open
* or if we are the last writer and the journal file is open in shared memory (not necessarily
* by ourselves e.g. the only process that opened the journal got shot abnormally)
* Note: we should not infer anything from the shared memory value of cnl->jnl_file.u.inode
* if we are not the last writer as it can be concurrently updated.
*/
if (!was_crit)
grab_crit(reg);
if (JNL_ENABLED(csd))
{
SET_GBL_JREC_TIME; /* jnl_ensure_open/jnl_put_jrt_pini/pfin/jnl_file_close all need it */
/* Before writing to jnlfile, adjust jgbl.gbl_jrec_time if needed to maintain time order
* of jnl records. This needs to be done BEFORE the jnl_ensure_open as that could write
* journal records (if it decides to switch to a new journal file).
*/
ADJUST_GBL_JREC_TIME(jgbl, jbp);
jnl_status = jnl_ensure_open();
if (0 == jnl_status)
{ /* If we_are_last_writer, we would have already done a wcs_flu() which would
* have written an epoch record and we are guaranteed no further updates
* since we are the last writer. So, just close the journal.
* If the freeaddr == post_epoch_freeaddr, wcs_flu may have skipped writing
* a pini, so allow for that.
*/
assert(!jbp->before_images || is_mm
|| !we_are_last_writer || (0 != jpc->pini_addr) || jgbl.mur_extract
|| (jpc->jnl_buff->freeaddr == jpc->jnl_buff->post_epoch_freeaddr));
/* If we haven't written a pini, let jnl_file_close write the pini/pfin. */
if (!jgbl.mur_extract && (0 != jpc->pini_addr))
jnl_put_jrt_pfin(csa);
/* If not the last writer and no pending flush timer left, do jnl flush now */
if (!we_are_last_writer && (0 > cnl->wcs_timers))
{
if (SS_NORMAL == (jnl_status = jnl_flush(reg)))
{
assert(jbp->freeaddr == jbp->dskaddr);
jnl_fsync(reg, jbp->dskaddr);
assert(jbp->fsync_dskaddr == jbp->dskaddr);
} else
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_JNLFLUSH, 2,
JNL_LEN_STR(csd), ERR_TEXT, 2,
RTS_ERROR_TEXT("Error with journal flush in gds_rundown"),
jnl_status);
assert(NOJNL == jpc->channel);/* jnl file lost has been triggered */
/* In this routine, all code that follows from here on does not
* assume anything about the journaling characteristics of this
* database so it is safe to continue execution even though
* journaling got closed in the middle.
*/
}
}
jnl_file_close(reg, we_are_last_writer, FALSE);
} else
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd),
DB_LEN_STR(reg));
}
if (!was_crit)
rel_crit(reg);
}
}
if (we_are_last_writer) /* Flush the fileheader last and harden the file to disk */
{
if (!was_crit)
grab_crit(reg); /* To satisfy crit requirement in fileheader_sync() */
memset(csd->machine_name, 0, MAX_MCNAMELEN); /* clear the machine_name field */
if (!have_standalone_access && we_are_last_user)
{ /* mupip_exit_handler will do this after mur_close_file */
csd->semid = INVALID_SEMID;
csd->shmid = INVALID_SHMID;
csd->gt_sem_ctime.ctime = 0;
csd->gt_shm_ctime.ctime = 0;
}
fileheader_sync(reg);
if (!was_crit)
rel_crit(reg);
if (!is_mm)
{
GTM_DB_FSYNC(csa, udi->fd, rc); /* Sync it all */
if (-1 == rc)
{
rts_error_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error during file sync at close"), errno);
}
} else
{ /* Now do final MM file sync before exit */
assert(csa->ti->total_blks == csa->total_blks);
#ifdef _AIX
GTM_DB_FSYNC(csa, udi->fd, rc);
if (-1 == rc)
#else
if (-1 == MSYNC((caddr_t)csa->db_addrs[0], (caddr_t)csa->db_addrs[1]))
#endif
{
rts_error_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error during file sync at close"), errno);
}
}
} else if (unsafe_last_writer && !cnl->lastwriterbypas_msg_issued)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_LASTWRITERBYPAS, 2, DB_LEN_STR(reg));
cnl->lastwriterbypas_msg_issued = TRUE;
}
} /* end if (!reg->read_only && !cnl->donotflush_dbjnl) */
/* We had canceled all db timers at start of rundown. In case as part of rundown (wcs_flu above), we had started
* any timers, cancel them BEFORE setting reg->open to FALSE (assert in wcs_clean_dbsync relies on this).
*/
CANCEL_DB_TIMERS(reg, csa, canceled_flush_timer, canceled_dbsync_timer);
if (reg->read_only && we_are_last_user && !have_standalone_access && cnl->remove_shm)
{ /* mupip_exit_handler will do this after mur_close_file */
db_ipcs.semid = INVALID_SEMID;
db_ipcs.shmid = INVALID_SHMID;
db_ipcs.gt_sem_ctime = 0;
db_ipcs.gt_shm_ctime = 0;
db_ipcs.fn_len = reg->dyn.addr->fname_len;
memcpy(db_ipcs.fn, reg->dyn.addr->fname, reg->dyn.addr->fname_len);
db_ipcs.fn[reg->dyn.addr->fname_len] = 0;
/* request gtmsecshr to flush. read_only cannot flush itself */
WAIT_FOR_REPL_INST_UNFREEZE_SAFE(csa);
if (!csa->read_only_fs)
{
secshrstat = send_mesg2gtmsecshr(FLUSH_DB_IPCS_INFO, 0, (char *)NULL, 0);
if (0 != secshrstat)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("gtmsecshr failed to update database file header"));
}
}
/* Done with file now, close it */
CLOSEFILE_RESET(udi->fd, rc); /* resets "udi->fd" to FD_INVALID */
if (-1 == rc)
{
rts_error_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error during file close"), errno);
}
/* Unmap storage if mm mode but only the part that is not the fileheader (so shows up in dumps) */
# if !defined(_AIX)
if (is_mm && (NULL != csa->db_addrs[0]))
{
assert(csa->db_addrs[1] > csa->db_addrs[0]);
munmap_len = (sm_long_t)(csa->db_addrs[1] - csa->db_addrs[0]);
if (0 < munmap_len)
munmap((caddr_t)(csa->db_addrs[0]), (size_t)(munmap_len));
}
# endif
/* Detach our shared memory while still under lock so reference counts will be correct for the next process to run down
* this region. In the process also get the remove_shm status from node_local before detaching.
* If cnl->donotflush_dbjnl is TRUE, it means we can safely remove shared memory without compromising data
* integrity as a reissue of recover will restore the database to a consistent state.
*/
remove_shm = !vermismatch && (cnl->remove_shm || cnl->donotflush_dbjnl);
/* We are done with online rollback on this region. Indicate to other processes by setting the onln_rlbk_pid to 0.
* Do it before releasing crit (t_end relies on this ordering when accessing cnl->onln_rlbk_pid).
*/
if (jgbl.onlnrlbk)
cnl->onln_rlbk_pid = 0;
rel_crit(reg); /* Since we are about to detach from the shared memory, release crit and reset onln_rlbk_pid */
/* If we had skipped flushing journal and database buffers due to a concurrent online rollback, increment the counter
* indicating that in the shared memory so that online rollback can report the # of such processes when it shuts down.
* The same thing is done for both FTOK and access control semaphores when there are too many MUMPS processes.
*/
if (safe_mode) /* indicates flushing was skipped */
{
if (bypassed_access)
cnl->dbrndwn_access_skip++; /* Access semaphore can be bypassed during online rollback */
if (bypassed_ftok)
cnl->dbrndwn_ftok_skip++;
}
if (jgbl.onlnrlbk)
csa->hold_onto_crit = FALSE;
GTM_WHITE_BOX_TEST(WBTEST_HOLD_SEM_BYPASS, cnl->wbox_test_seq_num, 0);
status = shmdt((caddr_t)cnl);
csa->nl = NULL; /* dereferencing nl after detach is not right, so we set it to NULL so that we can test before dereference*/
/* Note that although csa->nl is NULL, we use CSA_ARG(csa) below (not CSA_ARG(NULL)) to be consistent with similar
* usages before csa->nl became NULL. The "is_anticipatory_freeze_needed" function (which is in turn called by the
* CHECK_IF_FREEZE_ON_ERROR_NEEDED macro) does a check of csa->nl before dereferencing shared memory contents so
* we are safe passing "csa".
*/
if (-1 == status)
send_msg_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg), ERR_TEXT, 2,
LEN_AND_LIT("Error during shmdt"), errno);
REMOVE_CSA_FROM_CSADDRSLIST(csa); /* remove "csa" from list of open regions (cs_addrs_list) */
reg->open = FALSE;
/* If file is still not in good shape, die here and now before we get rid of our storage */
assertpro(0 == csa->wbuf_dqd);
ipc_deleted = FALSE;
/* If we are the very last user, remove shared storage id and the semaphores */
if (we_are_last_user)
{ /* remove shared storage, only if last writer to rundown did a successful wcs_flu() */
assert(!vermismatch);
if (remove_shm)
{
ipc_deleted = TRUE;
if (0 != shm_rmid(udi->shmid))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Unable to remove shared memory"));
/* Note that we no longer have a new shared memory. Currently only used/usable for standalone rollback. */
udi->new_shm = FALSE;
/* mupip recover/rollback don't release the semaphore here, but do it later in db_ipcs_reset (invoked from
* mur_close_files())
*/
if (!have_standalone_access)
{
if (0 != sem_rmid(udi->semid))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Unable to remove semaphore"));
udi->new_sem = FALSE; /* Note that we no longer have a new semaphore */
udi->grabbed_access_sem = FALSE;
udi->counter_acc_incremented = FALSE;
}
} else if (is_src_server || is_updproc)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
} else
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
} else
{
assert(!have_standalone_access || jgbl.onlnrlbk || safe_mode);
if (!jgbl.onlnrlbk && !have_standalone_access)
{ /* If we were writing, get rid of our writer access count semaphore */
if (!reg->read_only)
{
if (!access_counter_halted)
{
save_errno = do_semop(udi->semid, DB_COUNTER_SEM, -DB_COUNTER_SEM_INCR, SEM_UNDO);
if (0 != save_errno)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown access control semaphore decrement"),
CALLFROM, save_errno);
}
udi->counter_acc_incremented = FALSE;
}
assert(safe_mode || !bypassed_access);
/* Now remove the rundown lock */
if (!bypassed_access)
{
if (0 != (save_errno = do_semop(udi->semid, DB_CONTROL_SEM, -1, SEM_UNDO)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown access control semaphore release"),
CALLFROM, save_errno);
udi->grabbed_access_sem = FALSE;
}
} /* else access control semaphore will be released in db_ipcs_reset */
}
if (!have_standalone_access)
{
if (bypassed_ftok)
{
if (!ftok_counter_halted)
if (0 != (save_errno = do_semop(udi->ftok_semid, DB_COUNTER_SEM, -DB_COUNTER_SEM_INCR, SEM_UNDO)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
} else if (!ftok_sem_release(reg, !ftok_counter_halted, FALSE))
{
FTOK_TRACE(csa, csa->ti->curr_tn, ftok_ops_release, process_id);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
}
udi->grabbed_ftok_sem = FALSE;
udi->counter_ftok_incremented = FALSE;
}
ENABLE_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
if (!ipc_deleted)
{
GET_CUR_TIME(time_str);
if (is_src_server)
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Source server"), REG_LEN_STR(reg));
if (is_updproc)
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Update process"), REG_LEN_STR(reg));
if (mupip_jnl_recover && (!jgbl.onlnrlbk || !we_are_last_user))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Mupip journal process"), REG_LEN_STR(reg));
send_msg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Mupip journal process"), REG_LEN_STR(reg));
}
}
REVERT;
return EXIT_NRM;
}
