void iosocket_destroy (io_desc *ciod)
{
io_log_name **lpp, *lp; /* logical name pointers */
d_socket_struct *dsocketptr;
assertpro(ciod->type == gtmsocket);
assertpro(ciod->state == dev_closed);
dsocketptr = (d_socket_struct *) ciod->dev_sp;
assertpro(dsocketptr != NULL);
for (lpp = &io_root_log_name, lp = *lpp; lp; lp = *lpp)
{
if (lp->iod->pair.in == ciod)
{
/* The only device that may be "split" is the principal device. Since it is permanently open,
* it will never get here.
*/
assert(lp->iod == ciod);
assert(lp->iod->pair.out == ciod);
*lpp = (*lpp)->next;
free(lp);
}
else
lpp = &lp->next;
}
free(dsocketptr);
free(ciod);
}
/* Halt the process similar to op_halt but allow a return code to be specified. If no return code
* is specified, return code 0 is used as a default (making it identical to op_halt).
*/
int m_zhalt(void)
{
triple *triptr;
oprtype ot;
int status;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
/* Let m_halt() handle the case of the missing return code */
if ((TK_SPACE == TREF(window_token)) || (TK_EOL == TREF(window_token)))
return m_halt();
switch (status = expr(&ot, MUMPS_NUM)) /* NOTE assignment */
{
case EXPR_FAIL:
return FALSE;
case EXPR_GOOD:
triptr = newtriple(OC_ZHALT);
triptr->operand[0] = ot;
return TRUE;
case EXPR_INDR:
make_commarg(&ot, indir_zhalt);
return TRUE;
default:
assertpro(FALSE);
}
return FALSE; /* This should never get executed, added to make compiler happy */
}
int fd_ioready(int sock_fd, int poll_direction, int timeout)
{
int save_errno, status, EAGAIN_cnt = 0;
# ifdef USE_POLL
struct pollfd fds;
# else
fd_set fds, *readfds, *writefds;
struct timeval timeout_spec;
# endif
assert(timeout < MILLISECS_IN_SEC);
SELECT_ONLY(timeout = timeout * 1000); /* Convert to microseconds (~ 1sec) */
assert((timeout >= 0) && (timeout < POLL_ONLY(MILLISECS_IN_SEC) SELECT_ONLY(MICROSEC_IN_SEC)));
# ifdef USE_POLL
fds.fd = sock_fd;
fds.events = (REPL_POLLIN == poll_direction) ? POLLIN : POLLOUT;
# else
readfds = writefds = NULL;
timeout_spec.tv_sec = 0;
timeout_spec.tv_usec = timeout;
assertpro(FD_SETSIZE > sock_fd);
FD_ZERO(&fds);
FD_SET(sock_fd, &fds);
writefds = (REPL_POLLOUT == poll_direction) ? &fds : NULL;
readfds = (REPL_POLLIN == poll_direction) ? &fds : NULL;
# endif
POLL_ONLY(while (-1 == (status = poll(&fds, 1, timeout))))
SELECT_ONLY(while (-1 == (status = select(sock_fd + 1, readfds, writefds, NULL, &timeout_spec))))
{
save_errno = ERRNO;
if (EINTR == save_errno)
{ /* Give it another shot. But, halve the timeout so we don't keep doing this forever. */
timeout = timeout >> 1;
} else if (EAGAIN == save_errno)
static char *encode_os()
{
unsigned char *p;
int count, osidx;
count = 0;
p = (unsigned char *)gtm_release_name;
/* third arg in release name string */
while (*p && count < 2)
{
if (*p == ' ')
count++;
p++;
}
if (count == 2)
{
for (osidx = 0; osidx < SIZEOF(gtcm_proto_os_info)/SIZEOF(gtcm_proto_os_info_t) - 1; osidx++)
{
if (0 == memcmp(p, gtcm_proto_os_info[osidx].os_in_rel_str,
gtcm_proto_os_info[osidx].size_of_os_in_rel_str))
return gtcm_proto_os_info[osidx].proto_os;
}
}
assertpro(FALSE && "os match");
return NULL; /* Added to make compiler happy and not throw warning */
}
static char *encode_cpu()
{
unsigned char *p;
int count, cpuidx;
count = 0;
p = (unsigned char *)gtm_release_name;
/* fourth arg in release name string */
while (*p && count < 3)
{
if (*p == ' ')
count++;
p++;
}
if (count == 3)
{
for (cpuidx = 0; cpuidx < SIZEOF(gtcm_proto_cpu_info)/SIZEOF(gtcm_proto_cpu_info_t) - 1; cpuidx++)
{
if (0 == memcmp(p, gtcm_proto_cpu_info[cpuidx].cpu_in_rel_str,
gtcm_proto_cpu_info[cpuidx].size_of_cpu_in_rel_str))
return gtcm_proto_cpu_info[cpuidx].proto_cpu;
}
}
assertpro(FALSE && "cpu match");
return NULL; /* Added to make compiler happy and not throw warning */
}
void mubexpfilnam(char *dirname, unsigned int dirlen, backup_reg_list *list)
{
char *c1;
mstr file;
char tmp_mstr_addr[MAX_FN_LEN];
file.len = MAX_FN_LEN;
file.addr = tmp_mstr_addr;
if (list != mu_repl_inst_reg_list)
{ /* Database region */
if (!mupfndfil(list->reg, &file, LOG_ERROR_TRUE))
{
util_out_print("Backup not finished because of the above error.", TRUE);
error_mupip = TRUE;
return;
}
} else
{ /* Replication instance region */
if (!repl_inst_get_name(file.addr, (unsigned int *)&file.len, MAX_FN_LEN, issue_rts_error))
assertpro(FALSE); /* rts_error should have been issued by repl_inst_get_name */
}
for (c1 = file.addr + file.len; (*c1 != '/') && (c1 != file.addr); c1--)
;
list->backup_file.len = INTCAST(dirlen + (file.len - (c1 - file.addr)));
list->backup_file.addr = (char *)malloc(list->backup_file.len + 1);
memcpy(list->backup_file.addr, dirname, dirlen);
memcpy(list->backup_file.addr + dirlen, c1, (file.len - (c1 - file.addr)));
list->backup_file.addr[list->backup_file.len] = '\0';
return;
}
unsigned char i386_reg(unsigned char vax_reg)
{
unsigned char reg;
switch (vax_reg & 0xf) /* mask out VAX register mode field */
{
case 0:
reg = I386_REG_EAX;
break;
case 1:
reg = I386_REG_EDX;
break;
case 8:
reg = I386_REG_ESI;
break;
case 9:
reg = I386_REG_EDI;
break;
case 11:
reg = I386_REG_EBX;
break;
default:
assertpro(FALSE && (vax_reg & 0xf));
break;
}
return reg;
}
/* This routine finds the position of a region in the global directory which we use to index arrays of region information */
int find_reg_hash_idx(gd_region *reg)
{
gd_region *regl;
int index;
for (index = gd_header->n_regions-1, regl = gd_header->regions + index; reg != regl; regl--, index--)
assertpro(0 <= index);
return index;
}
/* ------------------------------------------------------------------
* Perform action corresponding to the first async event that
* was logged.
* ------------------------------------------------------------------
*/
void async_action(bool lnfetch_or_start)
{
/* Double-check that we should be here: */
assert(0 < num_deferred);
switch(first_event)
{
case (outofband_event):
/* This function can be invoked only by a op_*intrrpt* transfer table function. Those transfer table
* functions should be active only for a short duration between the occurrence of an outofband event
* and the handling of it at a logical boundary (next M-line). We dont expect to be running with
* those transfer table functions for more than one M-line. If "outofband" is set to 0, the call to
* "outofband_action" below will do nothing and we will end up running with the op_*intrrpt* transfer
* table functions indefinitely. In this case M-FOR loops are known to return incorrect results which
* might lead to application integrity issues. It is therefore considered safer to GTMASSERT as we
* will at least have the core for analysis.
*/
assertpro(0 != outofband);
outofband_action(lnfetch_or_start);
break;
case (tt_write_error_event):
# ifdef UNIX
xfer_reset_if_setter(tt_write_error_event);
iott_wrterr();
# endif
/* VMS tt error processing is done in op_*intrrpt */
break;
case (network_error_event):
/* -------------------------------------------------------
* Network error not implemented here yet. Need to move
* from mdb_condition_handler after review.
* -------------------------------------------------------
*/
case (zstp_or_zbrk_event):
/* -------------------------------------------------------
* ZStep/Zbreak events not implemented here yet. Need to
* move here after review.
* -------------------------------------------------------
*/
default:
assertpro(FALSE); /* see above assertpro() for comment as to why this is needed */
}
}
int trigger_fill_xecute_buffer(gv_trigger_t *trigdsc)
{
int src_fetch_status;
assert(!dollar_tlevel || (tstart_trigger_depth <= gtm_trigger_depth));
/* We have 3 cases to consider - all of which REQUIRE a TP fence to already be in effect. The reason for this is, if we
* detect a restartable condition, we are going to cause this region's triggers to be unloaded which destroys the block
* our parameter is pointing to so the restart logic MUST take place outside of this routine.
*
* 1. We have an active transaction due to an IMPLICIT TSTART done by trigger handling but the trigger level has not yet
* been created. We don't need another TP wrapper in this case but we do need a condition handler to trap the thrown
* retry to again prevent C stack unwind and return to the caller in the same shape that gtm_trigger would return.
* 2. We have an active transaction due to an EXPLICIT M-code TSTART command. For this case, the trigger loads proceed
* as normal with restarts handled in the regular automatic fashion. Note this case also covers the tp restarts done
* by both the update process and mupip recover forward since those functions have their own way of intercepting and
* dealing with restarts. To cover those cases, tp->implicit_tstart can be TRUE but tp_implicit_trigger MUST be
* FALSE.
* 3. We have an active transaction due to an IMPLICIT TSTART done by trigger handling and one or more triggers are
* running. This becomes like case 2 since the restart will be handled by gtm_trigger and the proper thing will
* be done.
*
* An extra note about case 3. Case 3 can be the identified case if in a nested trigger we are in trigger-no-mans-land
* with a base frame for the nested trigger (having driven one of a set of parallel nested triggers) but no actual trigger
* execution frame yet exists. This is really a case 1 situation with a nested trigger but it turns out that dealing with
* like case 3 does the right thing because if/when mdb_condition_handler catches a thrown TPRETRY error, mdb_condition
* handler will peal the nested trigger frame off before doing the restart which works for us and avoids issues of
* multi-level implicit restarts we would otherwise have to handle.
*
* Note, this routine is for loading trigger source when the trigger is to be driven. The trigger_source_read_andor_verify()
* routine should be used when fetching trigger source for reasons other than driving the triggers. This routine is lighter
* weight but has a dependence on the restartability of the trigger-drive logic for getting the triggers reloaded as
* necessary.
*/
assertpro(0 < dollar_tlevel);
if (!tp_pointer->implicit_trigger /* Case 2 */
|| (tp_pointer->implicit_tstart && tp_pointer->implicit_trigger
&& (tstart_trigger_depth != gtm_trigger_depth))) /* Case 3 */
{ /* Test for Case 3/4 where we get to do very little: */
DBGTRIGR((stderr, "trigger_fill_xecute_buffer: Case 2/3\n"));
assert((!tp_pointer->implicit_trigger) || (0 < gtm_trigger_depth));
trigger_fill_xecute_buffer_read_trigger_source(trigdsc);
} else
{ /* Test for Case 1 where we only need a condition handler */
DBGTRIGR((stderr, "trigger_fill_xecute_buffer: Case 1\n"));
assert(tp_pointer->implicit_tstart && tp_pointer->implicit_trigger);
assert(tstart_trigger_depth == gtm_trigger_depth);
ESTABLISH_RET(trigger_fill_xecute_buffer_ch, SIGNAL);
trigger_fill_xecute_buffer_read_trigger_source(trigdsc);
REVERT;
}
/* return our bounty to caller */
trigdsc->xecute_str.mvtype = MV_STR;
return 0; /* Could return ERR_TPRETRY if return is via our condition handler */
}
boolean_t op_gvqueryget(mval *key, mval *val)
{
boolean_t gotit;
gv_key *save_key;
gvnh_reg_t *gvnh_reg;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
if (TREF(gv_last_subsc_null) && NEVER == gv_cur_region->null_subs)
sgnl_gvnulsubsc();
switch (REG_ACC_METH(gv_cur_region))
{
case dba_bg:
case dba_mm:
gvnh_reg = TREF(gd_targ_gvnh_reg);
if (NULL == gvnh_reg)
gotit = ((0 != gv_target->root) ? gvcst_queryget(val) : FALSE); /* global does not exist if root is 0 */
else
INVOKE_GVCST_SPR_XXX(gvnh_reg, gotit = gvcst_spr_queryget(val));
break;
case dba_cm:
gotit = gvcmx_query(val);
break;
case dba_usr:
save_key = gv_currkey;
gv_currkey = gv_altkey;
/* We rely on the fact that *gv_altkey area is not modified by gvusr_queryget, and don't change gv_altkey.
* If and when *gv_altkey area is modified by gvusr_queryget, we have to set up a spare key area
* (apart from gv_altkey and gv_currkey), and make gv_altkey point the spare area before calling gvusr_queryget */
memcpy(gv_currkey, save_key, SIZEOF(*save_key) + save_key->end);
gotit = gvusr_queryget(val);
gv_altkey = gv_currkey;
gv_currkey = save_key;
break;
default:
assertpro(FALSE && REG_ACC_METH(gv_cur_region));
}
if (gotit)
{
key->mvtype = MV_STR;
key->str.addr = (char *)gv_altkey->base;
key->str.len = gv_altkey->end + 1;
s2pool(&key->str);
} else
{
*key = literal_null;
*val = literal_null;
}
return gotit;
}
void print_destination_operand()
{
switch(instruction.destination_operand_class)
{
case undefined_class :
assertpro(FALSE);
break;
case register_class :
assert(instruction.destination_operand_reg != NULL);
*obpt++ = '%';
*obpt++ = instruction.reg_prefix;
SET_OBPT_STR(instruction.destination_operand_reg, STRLEN(instruction.destination_operand_reg));
break;
case memory_class :
assert(instruction.source_operand_class != memory_class);
if (instruction.destination_operand_reg != NULL)
{
SET_OBPT_INT4(instruction.offset);
*obpt++ = '(';
*obpt++ = '%';
*obpt++ = instruction.reg_prefix;
SET_OBPT_STR(instruction.destination_operand_reg, STRLEN(instruction.destination_operand_reg));
*obpt++ = ')';
} else
{
SET_OBPT_INT4(instruction.offset);
}
break;
case immediate_class :
*obpt++ = '0';
*obpt++ = 'x';
SET_OBPT_INT8(instruction.immediate);
break;
default :
assertpro(FALSE);
}
}
void print_instruction()
{
list_chkpage();
obpt = &outbuf[0];
memset(obpt, SP, ASM_OUT_BUFF);
obpt += 10;
i2hex((curr_addr - PTEXT_OFFSET), obpt, 8);
curr_addr += (instidx - prev_idx);
obpt += 10;
for( ; prev_idx < instidx; prev_idx++)
{
i2hex(code_buf[prev_idx], obpt, 2);
obpt += 2;
}
obpt += 10;
*obpt++ = '\n';
*obpt++ = '\t';
*obpt++ = '\t';
*obpt++ = '\t';
*obpt++ = '\t';
*obpt++ = '\t';
*obpt++ = '\t';
assert( instruction.opcode_mnemonic != NULL );
SET_OBPT_STR(instruction.opcode_mnemonic, STRLEN(instruction.opcode_mnemonic));
*obpt++ = instruction.opcode_suffix;
*obpt++ = '\t';
instruction.num_operands = (instruction.num_operands > grp_prefix)
? (instruction.num_operands - grp_prefix) : instruction.num_operands;
switch (instruction.num_operands)
{
case 0 :
break;
case 1 :
/* single operand assumed to be in the source operand only.. */
assert(instruction.destination_operand_class == undefined_class);
print_source_operand();
break;
case 2 :
print_source_operand();
*obpt++ = ',';
print_destination_operand();
break;
default :
assertpro(FALSE);
}
/* Now reset the instruction structure */
emit_eoi();
reset_instruction();
}
int gtmrecv_helpers_init(int n_readers, int n_writers)
{ /* Receiver server interface to start n_readers and n_writers helper processes */
upd_helper_ctl_ptr_t upd_helper_ctl;
upd_helper_entry_ptr_t helper, helper_top;
int reader_count, writer_count, error_count, avail_slots, status;
assert(0 != n_readers || 0 != n_writers);
upd_helper_ctl = recvpool.upd_helper_ctl;
for (avail_slots = 0, helper = upd_helper_ctl->helper_list, helper_top = helper + MAX_UPD_HELPERS;
helper < helper_top; helper++)
{
if (0 == helper->helper_pid)
avail_slots++;
}
if (n_readers + n_writers > avail_slots)
{ /* adjust reader/writer count for available slots according to the percentage specified by user */
n_writers = (int)(((float)n_writers/(n_readers + n_writers)) * (float)avail_slots); /* may round down */
n_readers = avail_slots - n_writers; /* preference to readers, writer count may round down */
}
/* Start helpers, readers first */
for (helper = upd_helper_ctl->helper_list, helper_top = helper + MAX_UPD_HELPERS,
reader_count = 0, writer_count = 0, error_count = 0;
(reader_count + writer_count + error_count) < (n_readers + n_writers); )
{
for (; 0 != helper->helper_pid && helper < helper_top; helper++) /* find next vacant slot */
;
assertpro(helper != helper_top);
status = helper_init(helper, ((reader_count + error_count) < n_readers) ? UPD_HELPER_READER : UPD_HELPER_WRITER);
if (UPDPROC_STARTED == status)
{
if ((reader_count + error_count) < n_readers)
reader_count++;
else
writer_count++;
} else /* UPDPROC_START_ERR == status */
{
if ((EREPL_UPDSTART_BADPATH == repl_errno) /* receiver server lost gtm_dist environment, bad situation */
|| (EREPL_UPDSTART_EXEC == repl_errno)) /* in forked child, could not exec, should exit */
gtmrecv_exit(ABNORMAL_SHUTDOWN);
error_count++;
}
}
upd_helper_ctl->start_n_readers = reader_count;
upd_helper_ctl->start_n_writers = writer_count;
SHM_WRITE_MEMORY_BARRIER;
upd_helper_ctl->start_helpers = FALSE;
return ((0 == error_count) ? NORMAL_SHUTDOWN : ABNORMAL_SHUTDOWN);
}
void set_memory_reg()
{
instruction.reg_prefix = 'r';
if (instruction.source_operand_class == memory_class)
instruction.source_operand_reg = (char *)register_list[modrm_byte.modrm.r_m + 8 * rex_prefix.Base];
else if (instruction.destination_operand_class == memory_class)
instruction.destination_operand_reg = (char *)register_list[modrm_byte.modrm.r_m + 8 * rex_prefix.Base];
/* Printing of RIP has to be handled differently */
if (instruction.reg_rip)
if (instruction.source_operand_class == memory_class)
instruction.source_operand_reg = (char *)register_list[REG_RIP];
else if (instruction.destination_operand_class == memory_class)
instruction.destination_operand_reg = (char *)register_list[REG_RIP];
else
assertpro(FALSE);
}
/* The routine that does the actual work of determining the length and responding appropriately in the event an invalid
* UTF8 character is detected.
*/
STATICFNDEF int utf8_len_real(utf8_err_type err_type, mstr* str)
{
int charlen, bytelen;
char *ptrtop, *ptr;
boolean_t err_raised;
assert(gtm_utf8_mode);
ptr = str->addr;
ptrtop = ptr + str->len;
charlen = 0;
err_raised = FALSE;
if (!badchar_inhibit)
{
for (; ptr < ptrtop; charlen++, ptr += bytelen)
{
if (!UTF8_VALID(ptr, ptrtop, bytelen))
{
switch(err_type)
{
case err_rts:
UTF8_BADCHAR(0, ptr, ptrtop, 0, NULL);
break; /* Never get here but keeps compiler happy */
case err_stx:
UTF8_BADCHAR_STX(0, ptr, ptrtop, 0, NULL);
return -1;
case err_dec:
if (!err_raised)
{
UTF8_BADCHAR_DEC(0, ptr, ptrtop, 0, NULL);
err_raised = TRUE;
}
bytelen = 1; /* Assume only one char is broken */
break;
default:
assertpro(FALSE /* Invalid error type */);
}
}
}
} else
{
for (; ptr < ptrtop; charlen++)
ptr = (char *)UTF8_MBNEXT(ptr, ptrtop);
}
assert(ptr == ptrtop);
str->char_len = charlen;
return charlen;
}
void op_gvincr(mval *increment, mval *result)
{
unsigned char buff[MAX_ZWR_KEY_SZ], *end;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
/* If specified var name is global ^%Y*, the name is illegal to use in a SET or KILL command, only GETs are allowed */
if ((RESERVED_NAMESPACE_LEN <= gv_currkey->end) && (0 == MEMCMP_LIT(gv_currkey->base, RESERVED_NAMESPACE)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_PCTYRESERVED);
if (gv_cur_region->read_only)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_DBPRIVERR, 2, DB_LEN_STR(gv_cur_region));
if ((TREF(gv_last_subsc_null) || TREF(gv_some_subsc_null)) && (ALWAYS != gv_cur_region->null_subs))
sgnl_gvnulsubsc();
assert(gv_currkey->end + 1 <= gv_cur_region->max_key_size);
MV_FORCE_NUM(increment);
switch (gv_cur_region->dyn.addr->acc_meth)
{
case dba_bg:
case dba_mm:
gvcst_incr(increment, result);
break;
case dba_cm:
gvcmx_increment(increment, result);
break;
case dba_usr:
/* $INCR not supported for DDP/USR access method */
if (0 == (end = format_targ_key(buff, MAX_ZWR_KEY_SZ, gv_currkey, TRUE)))
end = &buff[MAX_ZWR_KEY_SZ - 1];
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(10) ERR_UNIMPLOP, 0,
ERR_TEXT, 2, LEN_AND_LIT("GTCM DDP server does not support $INCREMENT"),
ERR_GVIS, 2, end - buff, buff,
ERR_TEXT, 2, REG_LEN_STR(gv_cur_region));
break;
default:
assertpro(FALSE);
}
assert(MV_DEFINED(result));
}
void job_addr(mstr *rtn, mstr *label, int4 offset, char **hdr, char **labaddr)
{
rhdtyp *rt_hdr;
int4 *lp;
mval rt;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
if (NULL == (rt_hdr = find_rtn_hdr(rtn)))
{
rt.mvtype = MV_STR;
rt.str = *rtn;
op_zlink(&rt, NULL);
assertpro(NULL != (rt_hdr = find_rtn_hdr(rtn)));
}
lp = NULL;
if ((rt_hdr->compiler_qlf & CQ_LINE_ENTRY) || (0 == offset))
/* Label offset with routine compiled with NOLINE_ENTRY should cause error. */
lp = find_line_addr(rt_hdr, label, offset, NULL);
if (!lp)
rts_error(VARLSTCNT(1) ERR_JOBLABOFF);
/* Set the pointer to address / offset for line number entry storage in lab_proxy. */
USHBIN_ONLY((TREF(lab_proxy)).lnr_adr = lp;)
int x86_64_arg_reg(int indx)
{
switch(indx)
{
case 0:
return I386_REG_RDI;
case 1:
return I386_REG_RSI;
case 2:
return I386_REG_RDX;
case 3:
return I386_REG_RCX;
case 4:
return I386_REG_R8;
case 5:
return I386_REG_R9;
default:
assertpro(FALSE);
break ;
}
/* Control will never reach here */
return -1 ;
}
void op_zut(mval *s)
{
struct timeval tv;
gtm_int8 microseconds, msectmp;
int numdigs;
int4 pwr;
assertpro(-1 != gettimeofday(&tv, NULL));
microseconds = (1LL * MICROSEC_IN_SEC * tv.tv_sec) + tv.tv_usec;
if ((microseconds < 0) && (microseconds > E_18))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_WEIRDSYSTIME);
if (microseconds < E_6)
{
s->m[1] = ((int4)microseconds * 1000);
s->mvtype = MV_INT | MV_NM;
} else
{
msectmp = microseconds;
/* Count the number of digits */
for (numdigs = 0; msectmp; numdigs++, msectmp /= 10);
if (numdigs <= NUM_DEC_DG_1L)
{
s->m[0] = 0;
s->m[1] = (int4)microseconds * ten_pwr[NUM_DEC_DG_1L - numdigs];
} else
{
pwr = ten_pwr[numdigs - NUM_DEC_DG_1L];
s->m[0] = (microseconds % pwr) * ten_pwr[NUM_DEC_DG_2L - numdigs];
s->m[1] = microseconds / pwr;
}
s->mvtype = MV_NM;
s->e = MV_XBIAS + numdigs;
}
s->sgn = 0;
return;
}
void tp_unwind(uint4 newlevel, enum tp_unwind_invocation invocation_type, int *tprestart_rc)
{
mlk_pvtblk **prior, *mlkp;
mlk_tp *oldlock, *nextlock;
int tl;
lv_val *save_lv, *curr_lv, *lv;
tp_var *restore_ent;
mv_stent *mvc;
boolean_t restore_lv, rollback_locks;
lvscan_blk *lvscan, *lvscan_next, first_lvscan;
int elemindx, rc;
lvTree *lvt_child;
/* We are about to clean up structures. Defer MUPIP STOP/signal handling until function end. */
DEFER_INTERRUPTS(INTRPT_IN_TP_UNWIND);
/* Unwind the requested TP levels */
# if defined(DEBUG_REFCNT) || defined(DEBUG_ERRHND)
DBGFPF((stderr, "\ntp_unwind: Beginning TP unwind process\n"));
# endif
restore_lv = (RESTART_INVOCATION == invocation_type);
lvscan = &first_lvscan;
lvscan->next = NULL;
lvscan->elemcnt = 0;
assert((tp_sp <= tpstackbase) && (tp_sp > tpstacktop));
assert((tp_pointer <= (tp_frame *)tpstackbase) && (tp_pointer > (tp_frame *)tpstacktop));
for (tl = dollar_tlevel; tl > newlevel; --tl)
{
DBGRFCT((stderr, "\ntp_unwind: Unwinding level %d -- tp_pointer: 0x"lvaddr"\n", tl, tp_pointer));
assertpro(NULL != tp_pointer);
for (restore_ent = tp_pointer->vars; NULL != restore_ent; restore_ent = tp_pointer->vars)
{
/*********************************************************************************/
/* TP_VAR_CLONE sets the var_cloned flag, showing that the tree has been cloned */
/* If var_cloned is not set, it shows that curr_lv and save_lv are still sharing */
/* the tree, so it should not be killed. */
/*********************************************************************************/
curr_lv = restore_ent->current_value;
save_lv = restore_ent->save_value;
assert(curr_lv);
assert(save_lv);
assert(LV_IS_BASE_VAR(curr_lv));
assert(LV_IS_BASE_VAR(save_lv));
assert(0 < curr_lv->stats.trefcnt);
assert(curr_lv->tp_var);
assert(curr_lv->tp_var == restore_ent);
/* In order to restart sub-transactions, this would have to maintain
* the chain that currently is not built by op_tstart()
*/
if (restore_lv)
{
rc = tp_unwind_restlv(curr_lv, save_lv, restore_ent, NULL, tprestart_rc);
# ifdef GTM_TRIGGER
if (0 != rc)
{
dollar_tlevel = tl; /* Record fact if we unwound some tp_frames */
ENABLE_INTERRUPTS(INTRPT_IN_TP_UNWIND); /* drive any MUPIP STOP/signals deferred
* while in this function */
TPUNWND_WBOX_TEST; /* Debug-only wbox-test to simulate SIGTERM */
INVOKE_RESTART;
}
# endif
} else if (restore_ent->var_cloned)
{ /* curr_lv has been cloned.
* Note: LV_CHILD(save_lv) can be non-NULL only if restore_ent->var_cloned is TRUE
*/
DBGRFCT((stderr, "\ntp_unwind: Not restoring curr_lv and is cloned\n"));
lvt_child = LV_GET_CHILD(save_lv);
if (NULL != lvt_child)
{ /* If subtree exists, we have to blow away the cloned tree */
DBGRFCT((stderr, "\ntp_unwind: save_lv has children\n"));
assert(save_lv->tp_var);
DBGRFCT((stderr,"\ntp_unwind: For lv_val 0x"lvaddr": Deleting saved lv_val 0x"lvaddr"\n",
curr_lv, save_lv));
assert(LVT_PARENT(lvt_child) == (lvTreeNode *)save_lv);
lv_kill(save_lv, DOTPSAVE_FALSE, DO_SUBTREE_TRUE);
}
restore_ent->var_cloned = FALSE;
} else
{ /* If not cloned, we still have to reduce the reference counts of any
* container vars in the untouched tree that were added to keep anything
* they referenced from disappearing.
*/
DBGRFCT((stderr, "\ntp_unwind: Not restoring curr_lv and is NOT cloned\n"));
lvt_child = LV_GET_CHILD(curr_lv);
if (NULL != lvt_child)
{
DBGRFCT((stderr, "\ntp_unwind: curr_lv has children and so reducing ref counts\n"));
TPUNWND_CNTNRS_IN_TREE(curr_lv);
}
}
LV_FREESLOT(save_lv);
/* Not easy to predict what the trefcnt will be except that it should be greater than zero. In
* most cases, it will have its own hash table ref plus the extras we added but it is also
* possible that the entry has been kill *'d in which case the ONLY ref that will be left is
* our own increment but there is no [quick] way to distinguish this case so we just
* test for > 0.
*/
assert(0 < curr_lv->stats.trefcnt);
assert(0 < curr_lv->stats.crefcnt);
DECR_CREFCNT(curr_lv); /* Remove the copy refcnt we added in in op_tstart() or lv_newname() */
DECR_BASE_REF_NOSYM(curr_lv, FALSE);
curr_lv->tp_var = NULL;
tp_pointer->vars = restore_ent->next;
free(restore_ent);
}
if ((tp_pointer->fp == frame_pointer) && (MVST_TPHOLD == mv_chain->mv_st_type)
&& (msp == (unsigned char *)mv_chain))
POP_MV_STENT();
if (NULL == tp_pointer->old_tp_frame)
tp_sp = tpstackbase;
else
tp_sp = (unsigned char *)tp_pointer->old_tp_frame;
if (tp_sp > tpstackbase)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_STACKUNDERFLO);
if (tp_pointer->tp_save_all_flg)
--tp_pointer->sym->tp_save_all;
if ((NULL != (tp_pointer = tp_pointer->old_tp_frame)) /* Note assignment */
&& ((tp_pointer < (tp_frame *)tp_sp) || (tp_pointer > (tp_frame *)tpstackbase)
|| (tp_pointer < (tp_frame *)tpstacktop)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_STACKUNDERFLO);
}
if ((0 != newlevel) && restore_lv)
{ /* Restore current context (without releasing) */
assertpro(NULL != tp_pointer);
DBGRFCT((stderr, "\n\n** tp_unwind: Newlevel (%d) != 0 loop processing\n", newlevel));
for (restore_ent = tp_pointer->vars; NULL != restore_ent; restore_ent = restore_ent->next)
{
curr_lv = restore_ent->current_value;
save_lv = restore_ent->save_value;
assert(curr_lv);
assert(save_lv);
assert(LV_IS_BASE_VAR(curr_lv));
assert(LV_IS_BASE_VAR(save_lv));
assert(curr_lv->tp_var);
assert(curr_lv->tp_var == restore_ent);
assert(0 < curr_lv->stats.trefcnt);
rc = tp_unwind_restlv(curr_lv, save_lv, restore_ent, &lvscan, tprestart_rc);
# ifdef GTM_TRIGGER
if (0 != rc)
{
dollar_tlevel = tl; /* Record fact if we unwound some levels */
ENABLE_INTERRUPTS(INTRPT_IN_TP_UNWIND); /* drive any MUPIP STOP/signals deferred while
* in this function */
TPUNWND_WBOX_TEST; /* Debug-only wbox-test to simulate SIGTERM */
INVOKE_RESTART;
}
# endif
assert(0 < curr_lv->stats.trefcnt); /* Should have its own hash table ref plus the extras we added */
assert(0 < curr_lv->stats.crefcnt);
}
/* If we have any lv_vals queued up to be scanned for container vars, do that now */
DBGRFCT((stderr, "\ntp_unwind: Starting deferred rescan of lv trees needing refcnt processing\n"));
while (0 < lvscan->elemcnt)
{
assert(ARY_SCNCNTNR_DIM >= lvscan->elemcnt);
for (elemindx = 0; lvscan->elemcnt > elemindx; ++elemindx)
{
lv = lvscan->ary_scncntnr[elemindx];
DBGRFCT((stderr, "\n**tp_unwind_process_lvscan_array: Deferred processing lv 0x"lvaddr"\n", lv));
assert(LV_IS_BASE_VAR(lv));
/* This is the final level being restored so redo the counters on these vars */
TPREST_CNTNRS_IN_TREE(lv);
}
/* If we allocated any secondary blocks, we are done with them now so release them. Only the
* very last block on the chain is the original block that was automatically allocated which
* should not be freed in this fashion.
*/
lvscan_next = lvscan->next;
if (NULL != lvscan_next)
{ /* There is another block on the chain so this one can be freed */
free(lvscan);
DBGRFCT((stderr, "\ntp_unwind_process_lvscan_array: Freeing lvscan array\n"));
lvscan = lvscan_next;
} else
{ /* Since this is the original block allocated on the C stack which we may reuse,
* zero the element count.
*/
lvscan->elemcnt = 0;
DBGRFCT((stderr, "\ntp_unwind_process_lvscan_array: Setting elemcnt to 0 in original "
"lvscan block\n"));
assert(lvscan == &first_lvscan);
}
}
}
assert(0 == lvscan->elemcnt); /* verify no elements queued that were not scanned */
rollback_locks = (COMMIT_INVOCATION != invocation_type);
for (prior = &mlk_pvt_root, mlkp = *prior; NULL != mlkp; mlkp = *prior)
{
if (mlkp->granted)
{ /* This was a pre-existing lock */
for (oldlock = mlkp->tp; (NULL != oldlock) && ((int)oldlock->tplevel > newlevel); oldlock = nextlock)
{ /* Remove references to the lock from levels being unwound */
nextlock = oldlock->next;
free(oldlock);
}
if (rollback_locks)
{
if (NULL == oldlock)
{ /* Lock did not exist at the tp level being unwound to */
mlk_unlock(mlkp);
mlk_pvtblk_delete(prior);
continue;
} else
{ /* Lock still exists but restore lock state as it was when the transaction started. */
mlkp->level = oldlock->level;
mlkp->zalloc = oldlock->zalloc;
}
}
if ((NULL != oldlock) && (oldlock->tplevel == newlevel))
{ /* Remove lock reference from level being unwound to,
* now that any {level,zalloc} state information has been restored.
*/
assert((NULL == oldlock->next) || (oldlock->next->tplevel < newlevel));
mlkp->tp = oldlock->next; /* update root reference pointer */
free(oldlock);
} else
mlkp->tp = oldlock; /* update root reference pointer */
prior = &mlkp->next;
} else
mlk_pvtblk_delete(prior);
}
DBGRFCT((stderr, "tp_unwind: Processing complete\n"));
dollar_tlevel = newlevel;
ENABLE_INTERRUPTS(INTRPT_IN_TP_UNWIND); /* check if any MUPIP STOP/signals were deferred while in this function */
}
/* Restore given local variable from supplied TP restore entry into given symval. Note lvscan_anchor will only be non-NULL
* for the final level we are restoring (but not unwinding). We don't need to restore counters for any vars except the
* very last level.
*
* The return code is only used when unrolling the M stack runs into a trigger base frame which must be unrolled
* by gtm_trigger. A non-zero return code signals to tp_unwind() that it needs to rethrow the tprestart error.
*/
int tp_unwind_restlv(lv_val *curr_lv, lv_val *save_lv, tp_var *restore_ent, lvscan_blk **lvscan_anchor, int *tprestart_rc)
{
ht_ent_mname *tabent;
lv_val *inuse_lv;
int elemindx;
mv_stent *mvc;
lvscan_blk *lvscan, *newlvscan;
lvTree *lvt_child;
boolean_t var_cloned;
assert(curr_lv);
assert(LV_IS_BASE_VAR(curr_lv));
assert(curr_lv->tp_var);
DBGRFCT((stderr, "\ntp_unwind_restlv: Entered for varname: '%.*s' curr_lv: 0x"lvaddr" save_lv: 0x"lvaddr"\n",
restore_ent->key.var_name.len, restore_ent->key.var_name.addr, curr_lv, save_lv));
DBGRFCT((stderr, "tp_unwind_restlv: tp_pointer/current: fp: 0x"lvaddr"/0x"lvaddr" mvc: 0x"lvaddr"/0x"lvaddr
" symval: 0x"lvaddr"/0x"lvaddr"\n",
tp_pointer->fp, frame_pointer, tp_pointer->mvc, mv_chain, tp_pointer->sym, curr_symval));
/* First get the stack in the position where we can actually process this entry. Need to make sure we are processing
* the symbol table we need to be processing so unwind enough stuff to get there.
*/
if (curr_symval != tp_pointer->sym)
{ /* Unwind as many stackframes as are necessary up to the max */
while((curr_symval != tp_pointer->sym) && (frame_pointer < tp_pointer->fp))
{
# ifdef GTM_TRIGGER
if (SFT_TRIGR & frame_pointer->type)
{ /* We have encountered a trigger base frame. We cannot unroll it because there are C frames
* associated with it so we must interrupt this tp_restart and return to gtm_trigger() so
* it can unroll the base frame and rethrow the error to properly unroll the C stack.
*/
*tprestart_rc = ERR_TPRETRY;
tprestart_state = TPRESTART_STATE_TPUNW;
DBGTRIGR((stderr, "tp_unwind: Encountered trigger base frame during M-stack unwind - "
"rethrowing\n"));
return -1;
}
# endif
op_unwind();
}
if (curr_symval != tp_pointer->sym)
{ /* Unwind as many mv_stents as are necessary up to the max */
mvc = mv_chain;
while((curr_symval != tp_pointer->sym) && (mvc < tp_pointer->mvc))
{
unw_mv_ent(mvc);
mvc = (mv_stent *)(mvc->mv_st_next + (char *)mvc);
}
mv_chain = mvc;
/* Final check */
assertpro(curr_symval == tp_pointer->sym);
}
}
var_cloned = curr_lv->tp_var->var_cloned;
if (var_cloned)
{ /* Var/tree has been copied (and modified) -- see about restoring it */
DBGRFCT((stderr, "\ntp_unwind_restlv: curr_lv was modified and cloned -- needs restoration\n"));
if (NULL != restore_ent->key.var_name.addr)
{ /* Restore data into a named variable (hash table entry)
* Step 1 -- find its hash table address to see what lv_val is there now.
*/
tabent = lookup_hashtab_mname(&((tp_pointer->sym)->h_symtab), &restore_ent->key);
assert(tabent);
/* Step 2 -- If lv_val is NOT the same as it was, then we must replace the lv_val
* currently in use. Decrement its use count (which will delete it and the tree if
* it is no longer used) and replace with desired previous lv_val whose use count
* was incremented when it was saved.
*/
if (curr_lv != (inuse_lv = (lv_val *)tabent->value)) /* Note assignment */
{
if (inuse_lv)
DECR_BASE_REF_RQ(tabent, inuse_lv, FALSE);
DBGRFCT((stderr, "tp_unwind: hte 0x"lvaddr" being reset from 0x"lvaddr" to 0x"lvaddr"\n",
tabent, tabent->value, curr_lv));
tabent->value = (void *)curr_lv;
INCR_TREFCNT(curr_lv); /* Back in the hash table, bump its reference */
}
} /* Else, if restoring orphaned data, just prune the old var and copy in the saved tree (if one existed) */
/* Step 3 -- We have the correct lv_val in the hash table now but it has the wrong value.
* Get rid of its current tree if any.
*/
if (lvt_child = LV_GET_CHILD(curr_lv)) /* Note assignment */
{
DBGRFCT((stderr, "\ntp_unwind_restlv: Killing children of curr_lv 0x"lvaddr"\n", curr_lv));
assert((lvTreeNode *)curr_lv == LVT_PARENT(lvt_child));
LV_CHILD(curr_lv) = NULL; /* prevent recursion due to alias containers */
lv_killarray(lvt_child, FALSE);
}
/* Step 4: Copy in the needed fields from the saved flavor lv_val back to curr_lv.
* Preserve the ref counts of the current var since the copy's ref counts have not been kept up to date.
*/
DBGRFCT((stderr, "\ntp_unwind_restlv: Restoring value of lv 0x"lvaddr" back into lv 0x"lvaddr"\n",
save_lv, curr_lv));
/* The following is optimized to do the initialization of just the needed structure members. For that it assumes a
* particular "lv_val" structure layout. The assumed layout is asserted so any changes to the layout will
* automatically show an issue here and cause the below initialization to be accordingly reworked.
*/
assert(0 == OFFSETOF(lv_val, v));
assert(OFFSETOF(lv_val, v) + SIZEOF(curr_lv->v) == OFFSETOF(lv_val, ptrs));
assert(OFFSETOF(lv_val, ptrs) + SIZEOF(curr_lv->ptrs) == OFFSETOF(lv_val, stats));
assert(OFFSETOF(lv_val, stats) + SIZEOF(curr_lv->stats) == OFFSETOF(lv_val, has_aliascont));
assert(OFFSETOF(lv_val, has_aliascont) + SIZEOF(curr_lv->has_aliascont) == OFFSETOF(lv_val, lvmon_mark));
assert(OFFSETOF(lv_val, lvmon_mark) + SIZEOF(curr_lv->lvmon_mark) == OFFSETOF(lv_val, tp_var));
assert(OFFSETOF(lv_val, tp_var) + SIZEOF(curr_lv->tp_var) == SIZEOF(lv_val));
/* save_lv -> curr_lv Copy begin */
curr_lv->v = save_lv->v;
curr_lv->ptrs = save_lv->ptrs;
assert(0 < curr_lv->stats.trefcnt); /* No need to copy "stats" as curr_lv is more uptodate */
assert(0 < curr_lv->stats.crefcnt);
assert(8 == (OFFSETOF(lv_val, tp_var) - OFFSETOF(lv_val, has_aliascont)));
curr_lv->has_aliascont = save_lv->has_aliascont;
DBGALS_ONLY(curr_lv->lvmon_mark = save_lv->has_aliascont);
assert(save_lv->tp_var == curr_lv->tp_var); /* no need to copy this field */
/* save_lv -> curr_lv Copy done */
/* Some fixup may need to be done if the variable was cloned (and thus moved around) */
curr_lv->tp_var->var_cloned = FALSE;
if (lvt_child = LV_GET_CHILD(curr_lv))
{ /* Some pointer fix up needs to be done since the owner of the restored tree changed */
assert(LVT_PARENT(lvt_child) == ((lvTreeNode *)curr_lv->tp_var->save_value));
LV_CHILD(save_lv) = NULL; /* now that curr_lv->tp_var->var_cloned has been reset */
LVT_PARENT(lvt_child) = (lvTreeNode *)curr_lv;
if (curr_lv->has_aliascont && (NULL != lvscan_anchor))
{ /* Some ref counts need to be restored for arrays this tree points to -- but only if the
* array contains pointers (alias containers).
*/
DBGRFCT((stderr, "\ntp_unwind_restlv: Putting lv 0x:"lvaddr" on the lvscan list\n", curr_lv));
/* This array needs to have container pointer target reference counts reestablished. Record
* the lv so this can happen after all vars are restored.
*/
lvscan = *lvscan_anchor;
elemindx = lvscan->elemcnt++; /* Note post increment so elemindx has minus-one value */
if (ARY_SCNCNTNR_MAX < elemindx)
{ /* Present block is full so allocate a new one and chain it on */
lvscan->elemcnt--; /* New element ended up not being in that block.. */
newlvscan = (lvscan_blk *)malloc(SIZEOF(lvscan_blk));
newlvscan->next = lvscan;
newlvscan->elemcnt = 1; /* Going to use first one *now* */
elemindx = 0;
*lvscan_anchor = newlvscan;
lvscan = newlvscan;
}
assert((ARY_SCNCNTNR_MAX >= elemindx) && (0 <= elemindx));
lvscan->ary_scncntnr[elemindx] = curr_lv;
}
}
} else
{
DBGRFCT((stderr, "\ntp_unwind_restlv: curr_lv was NOT modified or cloned\n"));
assert(NULL == LV_CHILD(save_lv));
assert(!save_lv->tp_var->var_cloned);
/* We know that the subscript array underneath curr_lv did not change since saving it into save_lv. But the
* unsubscripted lv could have changed (have no way of checking if that is the case) so restore it unconditionally.
*/
curr_lv->v = save_lv->v;
/* No need to copy "save_lv->ptrs" as "ptrs" contains 2 fields both of which are already correct in "curr_lv" */
assert(save_lv->ptrs.val_ent.parent.sym == curr_lv->ptrs.val_ent.parent.sym);
assert(NULL == save_lv->ptrs.val_ent.children);
/* No need to copy "save_lv->stats" as "curr_lv->stats" is more uptodate */
assert(save_lv->has_aliascont == curr_lv->has_aliascont); /* no need to copy this field */
assert(save_lv->lvmon_mark == curr_lv->lvmon_mark); /* no need to copy this field */
assert(save_lv->tp_var == curr_lv->tp_var); /* no need to copy this field */
}
if (NULL == lvscan_anchor)
/* Means this is completely unwinding a nested level so we need to reset the tstartcycle in this
* lvval so it gets handled correctly when this lv is encountered again after the restart completes.
*/
curr_lv->stats.tstartcycle = 0;
return 0;
}
void bx_boollit_tail(triple *t, boolean_t jmp_type_one, boolean_t jmp_to_next, boolean_t sense, oprtype *addr)
/* search the Boolean in t (recursively) for literal leaves; the logic is similar to bx_tail
* the rest of the arguments parallel those in bx_boolop and used primarily handling basic Boolean operations (ON, NOR, AND, NAND)
* to get the jump target and sense right for the left-hand operand of the operation
* jmp_type_one gives the sense of the jump associated with the first operand
* jmp_to_next gives whether we need a second jump to complete the operation
* sense gives the sense of the requested operation
* *addr points the operand for the jump and is eventually used by logic back in the invocation stack to fill in a target location
*/
{
boolean_t sin[ARRAYSIZE(t->operand)], tv[ARRAYSIZE(t->operand)];
int com, comval, dummy, j, neg, num, tvr;
mval *mv, *v[ARRAYSIZE(t->operand)];
opctype c;
oprtype *i, *p;
triple *cob[ARRAYSIZE(t->operand)], *ref0, *tl[ARRAYSIZE(t->operand)];
assert(OCT_BOOL & oc_tab[t->opcode].octype);
assert(TRIP_REF == t->operand[0].oprclass);
assert((OC_COBOOL != t->opcode) && (OC_COM != t->opcode) || (TRIP_REF == t->operand[1].oprclass));
for (i = t->operand, j = 0; i < ARRAYTOP(t->operand); i++, j++)
{ /* checkout an operand to see if we can simplify it */
p = i;
com = 0;
for (tl[j] = i->oprval.tref; OCT_UNARY & oc_tab[(c = tl[j]->opcode)].octype; tl[j] = p->oprval.tref)
{ /* find the real object of affection; WARNING assignment above */
assert((TRIP_REF == tl[j]->operand[0].oprclass) && (NO_REF == tl[j]->operand[1].oprclass));
com ^= (OC_COM == c); /* if we make a recursive call below, COM matters, but NEG and FORCENUM don't */
p = &tl[j]->operand[0];
}
if (OCT_ARITH & oc_tab[c].octype)
ex_tail(p); /* chained arithmetic */
else if (OCT_BOOL & oc_tab[c].octype)
{ /* recursively check an operand */
sin[j] = sense;
p = addr;
if (!j && !(OCT_REL & oc_tab[t->opcode].octype))
{ /* left hand operand of parent */
sin[j] = jmp_type_one;
if (jmp_to_next)
{ /* left operands need extra attention to decide between jump next or to the end */
p = (oprtype *)mcalloc(SIZEOF(oprtype));
*p = put_tjmp(t);
}
}
bx_boollit(tl[j], sin[j] ^ com, p);
}
if ((OC_JMPTRUE != tl[j]->opcode) && (OC_JMPFALSE != tl[j]->opcode) && (OC_LIT != tl[j]->opcode))
return; /* this operation doesn't qualify */
com = comval = neg = num = 0;
cob[j] = NULL;
for (ref0 = i->oprval.tref; OCT_UNARY & oc_tab[(c = ref0->opcode)].octype; ref0 = ref0->operand[0].oprval.tref)
{ /* we may be able to clean up this operand; WARNING assignment above */
assert((TRIP_REF == ref0->operand[0].oprclass) && (NO_REF == ref0->operand[1].oprclass));
num += (OC_FORCENUM == c);
com += (OC_COM == c);
if (!com) /* "outside" com renders neg mute */
neg ^= (OC_NEG == c);
if (!comval && (NULL == cob[j]))
{
if (comval = (OC_COMVAL == c)) /* WARNING assignment */
{
if (ref0 != t->operand[j].oprval.tref)
dqdel(t->operand[j].oprval.tref, exorder);
t->operand[j].oprval.tref = tl[j]; /* need mval: no COBOOL needed */
}
else if (OC_COBOOL == c)
{ /* the operand needs a COBOOL in case its operator remains unresolved */
cob[j] = t->operand[j].oprval.tref;
if (ref0 == cob[j])
continue; /* already where it belongs */
cob[j]->opcode = OC_COBOOL;
cob[j]->operand[0].oprval.tref = tl[j];
} else if (ref0 == t->operand[j].oprval.tref)
continue;
}
dqdel(ref0, exorder);
}
assert(ref0 == tl[j]);
if (!comval && (NULL == cob[j]) && (tl[j] != t->operand[j].oprval.tref))
{ /* left room for a COBOOL, but there's no need */
dqdel(t->operand[j].oprval.tref, exorder);
t->operand[j].oprval.tref = tl[j];
}
if ((OC_JMPTRUE == ref0->opcode) || (OC_JMPFALSE == ref0->opcode))
{ /* switch to a literal representation of TRUE / FALSE */
assert(INDR_REF == ref0->operand[0].oprclass);
ref0->operand[1] = ref0->operand[0]; /* track info as we switch opcode */
PUT_LITERAL_TRUTH((sin[j] ? OC_JMPFALSE : OC_JMPTRUE) == ref0->opcode, ref0);
ref0->opcode = OC_LIT;
com = 0; /* already accounted for by sin */
}
assert((OC_LIT == ref0->opcode) && (MLIT_REF == ref0->operand[0].oprclass));
v[j] = &ref0->operand[0].oprval.mlit->v;
if (com)
{ /* any complement reduces the literal value to [unsigned] 1 or 0 */
unuse_literal(v[j]);
tv[j] = (0 == v[j]->m[1]);
assert(ref0 == tl[j]);
PUT_LITERAL_TRUTH(tv[j], ref0);
v[j] = &ref0->operand[0].oprval.mlit->v;
num = 0; /* any complement trumps num */
}
if (neg || num)
{ /* get literal into uniform state */
unuse_literal(v[j]);
mv = (mval *)mcalloc(SIZEOF(mval));
*mv = *v[j];
if (neg)
{
if (MV_INT & mv->mvtype)
{
if (0 != mv->m[1])
mv->m[1] = -mv->m[1];
else
mv->sgn = 0;
} else if (MV_NM & mv->mvtype)
mv->sgn = !mv->sgn;
} else
s2n(mv);
n2s(mv);
v[j] = mv;
assert(ref0 == tl[j]);
put_lit_s(v[j], ref0);
}
}
assert(tl[0] != tl[1]); /* start processing a live one */
for (tvr = j, j = 0; j < tvr; j++)
{ /* both arguments are literals, so do the operation at compile time */
if (NULL != cob[j])
dqdel(cob[j], exorder);
v[j] = &tl[j]->operand[0].oprval.mlit->v;
tv[j] = (0 != v[j]->m[1]);
unuse_literal(v[j]);
tl[j]->opcode = OC_NOOP;
tl[j]->operand[0].oprclass = NO_REF;
}
t->operand[1].oprclass = NO_REF;
switch (c = t->opcode) /* WARNING assignment */
{ /* optimize the Boolean operations here */
case OC_NAND:
case OC_AND:
tvr = (tv[0] && tv[1]);
break;
case OC_NOR:
case OC_OR:
tvr = (tv[0] || tv[1]);
break;
case OC_NCONTAIN:
case OC_CONTAIN:
tvr = 1;
(void)matchc(v[1]->str.len, (unsigned char *)v[1]->str.addr, v[0]->str.len,
(unsigned char *)v[0]->str.addr, &dummy, &tvr);
tvr ^= 1;
break;
case OC_NEQU:
case OC_EQU:
tvr = is_equ(v[0], v[1]);
break;
case OC_NFOLLOW:
case OC_FOLLOW:
tvr = 0 < memvcmp(v[0]->str.addr, v[0]->str.len, v[1]->str.addr, v[1]->str.len);
break;
case OC_NGT:
case OC_GT:
tvr = 0 < numcmp(v[0], v[1]);
break;
case OC_NLT:
case OC_LT:
tvr = 0 > numcmp(v[0], v[1]);
break;
case OC_NPATTERN:
case OC_PATTERN:
tvr = !(*(uint4 *)v[1]->str.addr) ? do_pattern(v[0], v[1]) : do_patfixed(v[0], v[1]);
break;
case OC_NSORTS_AFTER:
case OC_SORTS_AFTER:
tvr = 0 < sorts_after(v[0], v[1]);
break;
default:
assertpro(FALSE);
}
tvr ^= !sense;
t->operand[0] = put_indr(addr);
t->opcode = tvr ? OC_JMPFALSE : OC_JMPTRUE;
return;
}
uint4 gdsfilext(uint4 blocks, uint4 filesize, boolean_t trans_in_prog)
{
sm_uc_ptr_t old_base[2], mmap_retaddr;
boolean_t was_crit, is_mm;
int result, save_errno, status;
DEBUG_ONLY(int first_save_errno);
uint4 new_bit_maps, bplmap, map, new_blocks, new_total, max_tot_blks, old_total;
uint4 jnl_status;
gtm_uint64_t avail_blocks, mmap_sz;
off_t new_eof, new_size;
trans_num curr_tn;
unix_db_info *udi;
inctn_opcode_t save_inctn_opcode;
int4 prev_extend_blks_to_upgrd;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
cache_rec_ptr_t cr;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(!IS_DSE_IMAGE);
assert((cs_addrs->nl == NULL) || (process_id != cs_addrs->nl->trunc_pid)); /* mu_truncate shouldn't extend file... */
assert(!process_exiting);
DEBUG_ONLY(old_base[0] = old_base[1] = NULL);
assert(!gv_cur_region->read_only);
udi = FILE_INFO(gv_cur_region);
is_mm = (dba_mm == cs_addrs->hdr->acc_meth);
# if !defined(MM_FILE_EXT_OK)
if (!udi->grabbed_access_sem && is_mm)
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not allowed ? */
# endif
/* Both blocks and total blocks are unsigned ints so make sure we aren't asking for huge numbers that will
overflow and end up doing silly things.
*/
assert((blocks <= (MAXTOTALBLKS(cs_data) - cs_data->trans_hist.total_blks)) || WBTEST_ENABLED(WBTEST_FILE_EXTEND_ERROR));
# if defined(__sun) || defined(__hpux)
cs_data->defer_allocate = TRUE;
# endif
if (!blocks && (cs_data->defer_allocate || (TRANS_IN_PROG_TRUE == trans_in_prog)))
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not enabled ? */
bplmap = cs_data->bplmap;
/* New total of non-bitmap blocks will be number of current, non-bitmap blocks, plus new blocks desired
* There are (bplmap - 1) non-bitmap blocks per bitmap, so add (bplmap - 2) to number of non-bitmap blocks
* and divide by (bplmap - 1) to get total number of bitmaps for expanded database. (must round up in this
* manner as every non-bitmap block must have an associated bitmap)
* Current number of bitmaps is (total number of current blocks + bplmap - 1) / bplmap.
* Subtract current number of bitmaps from number needed for expanded database to get number of new bitmaps needed.
*/
new_bit_maps = DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap) + blocks, bplmap - 1)
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap);
new_blocks = blocks + new_bit_maps;
assert((0 < (int)new_blocks) || (!cs_data->defer_allocate && (0 == new_blocks)));
if (new_blocks + cs_data->trans_hist.total_blks > MAXTOTALBLKS(cs_data))
{
assert(WBTEST_ENABLED(WBTEST_FILE_EXTEND_ERROR));
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(1) ERR_TOTALBLKMAX);
return (uint4)(NO_FREE_SPACE);
}
if (0 != (save_errno = disk_block_available(udi->fd, &avail_blocks, FALSE)))
{
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
} else
{
if (!(gtmDebugLevel & GDL_IgnoreAvailSpace))
{ /* Bypass this space check if debug flag above is on. Allows us to create a large sparce DB
* in space it could never fit it if wasn't sparse. Needed for some tests.
*/
avail_blocks = avail_blocks / (cs_data->blk_size / DISK_BLOCK_SIZE);
if ((blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
if (blocks > (uint4)avail_blocks)
{
if (!INST_FREEZE_ON_NOSPC_ENABLED(cs_addrs))
return (uint4)(NO_FREE_SPACE);
else
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(6) MAKE_MSG_WARNING(ERR_NOSPACEEXT), 4,
DB_LEN_STR(gv_cur_region), new_blocks, (uint4)avail_blocks);
} else
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, DB_LEN_STR(gv_cur_region),
(uint4)(avail_blocks - ((new_blocks <= avail_blocks) ? new_blocks : 0)));
}
}
}
# ifdef DEBUG
if (WBTEST_ENABLED(WBTEST_MM_CONCURRENT_FILE_EXTEND) && dollar_tlevel && !MEMCMP_LIT(gv_cur_region->rname, "DEFAULT"))
{
SYSTEM("$gtm_dist/mumps -run $gtm_wbox_mrtn");
assert(1 == cs_addrs->nl->wbox_test_seq_num); /* should have been set by mubfilcpy */
cs_addrs->nl->wbox_test_seq_num = 2; /* signal mupip backup to stop sleeping in mubfilcpy */
}
# endif
/* From here on, we need to use GDSFILEXT_CLNUP before returning to the caller */
was_crit = cs_addrs->now_crit;
assert(!cs_addrs->hold_onto_crit || was_crit);
/* If we are coming from mupip_extend (which gets crit itself) we better have waited for any unfreezes to occur.
* If we are coming from online rollback (when that feature is available), we will come in holding crit and in
* the final retry. In that case too, we expect to have waited for unfreezes to occur in the caller itself.
* Therefore if we are coming in holding crit from MUPIP, we expect the db to be unfrozen so no need to wait for
* freeze.
* If we are coming from GT.M and final retry (in which case we come in holding crit) we expect to have waited
* for any unfreezes (by invoking tp_crit_all_regions) to occur (TP or non-TP) before coming into this
* function. However, there is one exception. In the final retry, if tp_crit_all_regions notices that
* at least one of the participating regions did ONLY READs, it will not wait for any freeze on THAT region
* to complete before grabbing crit. Later, in the final retry, if THAT region did an update which caused
* op_tcommit to invoke bm_getfree->gdsfilext, then we would have come here with a frozen region on which
* we hold crit.
*/
assert(!was_crit || !FROZEN_HARD(cs_data) || (dollar_tlevel && (CDB_STAGNATE <= t_tries)));
/*
* If we are in the final retry and already hold crit, it is possible that csa->nl->wc_blocked is also set to TRUE
* (by a concurrent process in phase2 which encountered an error in the midst of commit and secshr_db_clnup
* finished the job for it). In this case we do NOT want to invoke wcs_recover as that will update the "bt"
* transaction numbers without correspondingly updating the history transaction numbers (effectively causing
* a cdb_sc_blkmod type of restart). Therefore do NOT call grab_crit (which unconditionally invokes wcs_recover)
* if we already hold crit.
*/
if (!was_crit)
{
for ( ; ; )
{
grab_crit(gv_cur_region);
if (FROZEN_CHILLED(cs_data))
DO_CHILLED_AUTORELEASE(cs_addrs, cs_data);
if (!FROZEN(cs_data) && !IS_REPL_INST_FROZEN)
break;
rel_crit(gv_cur_region);
while (FROZEN(cs_data) || IS_REPL_INST_FROZEN)
{
hiber_start(1000);
if (FROZEN_CHILLED(cs_data) && CHILLED_AUTORELEASE(cs_data))
break;
}
}
} else if (FROZEN_HARD(cs_data) && dollar_tlevel)
{ /* We don't want to continue with file extension as explained above. Hence return with an error code which
* op_tcommit will recognize (as a cdb_sc_needcrit/cdb_sc_instancefreeze type of restart) and restart accordingly.
*/
assert(CDB_STAGNATE <= t_tries);
GDSFILEXT_CLNUP;
return (uint4)FINAL_RETRY_FREEZE_PROG;
} else
WAIT_FOR_REGION_TO_UNCHILL(cs_addrs, cs_data);
if (IS_REPL_INST_FROZEN && trans_in_prog)
{
assert(CDB_STAGNATE <= t_tries);
GDSFILEXT_CLNUP;
return (uint4)FINAL_RETRY_INST_FREEZE;
}
assert(cs_addrs->ti->total_blks == cs_data->trans_hist.total_blks);
old_total = cs_data->trans_hist.total_blks;
if (old_total != filesize)
{ /* Somebody else has already extended it, since we are in crit, this is trust-worthy. However, in case of MM,
* we still need to remap the database
*/
assert((old_total > filesize) || !is_mm);
/* For BG, someone else could have truncated or extended - we have no idea */
GDSFILEXT_CLNUP;
return (SS_NORMAL);
}
if (trans_in_prog && SUSPICIOUS_EXTEND)
{
if (!was_crit)
{
GDSFILEXT_CLNUP;
return (uint4)(EXTEND_SUSPECT);
}
/* If free_blocks counter is not ok, then correct it. Do the check again. If still fails, then it means we held
* crit through bm_getfree into gdsfilext and still didn't get it right.
*/
assertpro(!is_free_blks_ctr_ok() && !SUSPICIOUS_EXTEND);
}
if (JNL_ENABLED(cs_data))
{
if (!jgbl.dont_reset_gbl_jrec_time)
SET_GBL_JREC_TIME; /* needed before jnl_ensure_open as that can write jnl records */
jpc = cs_addrs->jnl;
jbp = jpc->jnl_buff;
/* 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(gv_cur_region, cs_addrs);
if (jnl_status)
{
GDSFILEXT_CLNUP;
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(cs_data), DB_LEN_STR(gv_cur_region));
return (uint4)(NO_FREE_SPACE); /* should have better return status */
}
}
if (is_mm)
{
cs_addrs->nl->mm_extender_pid = process_id;
status = wcs_wtstart(gv_cur_region, 0, NULL, NULL);
cs_addrs->nl->mm_extender_pid = 0;
assertpro(SS_NORMAL == status);
old_base[0] = cs_addrs->db_addrs[0];
old_base[1] = cs_addrs->db_addrs[1];
cs_addrs->db_addrs[0] = NULL; /* don't rely on it until the mmap below */
# ifdef _AIX
status = shmdt(old_base[0] - BLK_ZERO_OFF(cs_data->start_vbn));
# else
status = munmap((caddr_t)old_base[0], (size_t)(old_base[1] - old_base[0]));
# endif
if (0 != status)
{
save_errno = errno;
GDSFILEXT_CLNUP;
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(12) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region),
ERR_SYSCALL, 5, LEN_AND_STR(MEM_UNMAP_SYSCALL), CALLFROM, save_errno);
return (uint4)(NO_FREE_SPACE);
}
} else
{ /* Due to concurrency issues, it is possible some process had issued a disk read of the GDS block# corresponding
* to "old_total" right after a truncate wrote a GDS-block of zeros on disk (to signal end of the db file).
* If so, the global buffer containing this block needs to be invalidated now as part of the extend. If not, it is
* possible the EOF block on disk is now going to be overwritten by a properly initialized bitmap block (as part
* of the gdsfilext below) while the global buffer continues to have an incorrect copy of that bitmap block and
* this in turn would cause XXXX failures due to a bad bitmap block in shared memory. (GTM-7519)
*/
cr = db_csh_get((block_id)old_total);
if ((NULL != cr) && ((cache_rec_ptr_t)CR_NOTVALID != cr))
{
assert((0 == cr->dirty) && (0 == cr->bt_index) && !cr->stopped);
cr->cycle++;
cr->blk = CR_BLKEMPTY;
}
}
CHECK_TN(cs_addrs, cs_data, cs_data->trans_hist.curr_tn); /* can issue rts_error TNTOOLARGE */
new_total = old_total + new_blocks;
new_eof = BLK_ZERO_OFF(cs_data->start_vbn) + ((off_t)new_total * cs_data->blk_size);
# if !defined(__sun) && !defined(__hpux)
if (!cs_data->defer_allocate)
{
new_size = new_eof + cs_data->blk_size;
save_errno = posix_fallocate(udi->fd, 0, new_size);
DEBUG_ONLY(first_save_errno = save_errno);
if ((ENOSPC == save_errno) && IS_GTM_IMAGE)
save_errno = extend_wait_for_fallocate(udi, new_size);
if (0 != save_errno)
{
GDSFILEXT_CLNUP;
assert(ENOSPC == save_errno);
if (ENOSPC != save_errno)
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_PREALLOCATEFAIL, 2, DB_LEN_STR(gv_cur_region),
save_errno);
return (uint4)(NO_FREE_SPACE);
}
}
# endif
save_errno = db_write_eof_block(udi, udi->fd, cs_data->blk_size, new_eof, &(TREF(dio_buff)));
if ((ENOSPC == save_errno) && IS_GTM_IMAGE)
save_errno = extend_wait_for_write(udi, cs_data->blk_size, new_eof);
if (0 != save_errno)
{
GDSFILEXT_CLNUP;
if (ENOSPC != save_errno)
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
return (uint4)(NO_FREE_SPACE);
}
if (WBTEST_ENABLED(WBTEST_FILE_EXTEND_INTERRUPT_1))
{
LONG_SLEEP(600);
assert(FALSE);
}
/* Ensure the EOF and metadata get to disk BEFORE any bitmap writes. Otherwise, the file size could no longer reflect
* a proper extent and subsequent invocations of gdsfilext could corrupt the database.
*/
if (!IS_STATSDB_CSA(cs_addrs))
{
GTM_DB_FSYNC(cs_addrs, udi->fd, status);
assert(0 == status);
if (0 != status)
{
GDSFILEXT_CLNUP;
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(8) ERR_DBFILERR, 5,
RTS_ERROR_LITERAL("fsync1()"), CALLFROM, status);
return (uint4)(NO_FREE_SPACE);
}
}
if (WBTEST_ENABLED(WBTEST_FILE_EXTEND_INTERRUPT_2))
{
LONG_SLEEP(600);
assert(FALSE); /* Should be killed before that */
}
DEBUG_ONLY(prev_extend_blks_to_upgrd = cs_data->blks_to_upgrd;)
void view_arg_convert(viewtab_entry *vtp, int vtp_parm, mval *parm, viewparm *parmblk, boolean_t is_dollar_view)
{
static int4 first_time = TRUE;
char *cptr;
char *strtokptr;
gd_binding *gd_map;
gd_region *gd_reg_start, *r_ptr, *r_top;
gvnh_reg_t *gvnh_reg;
gvnh_spanreg_t *gvspan;
gv_namehead *tmp_gvt;
ht_ent_mname *tabent;
int n, reg_index;
mident_fixed lcl_buff;
mname_entry gvent, lvent;
mstr namestr, tmpstr;
unsigned char *c, *c_top, *dst, *dst_top, global_names[1024], *nextsrc, *src, *src_top, stashed, y;
switch (vtp_parm)
{
case VTP_NULL:
if (parm != 0)
rts_error_csa(CSA_ARG(NULL)
VARLSTCNT(4) ERR_VIEWARGCNT, 2, strlen((const char *)vtp->keyword), vtp->keyword);
break;
case (VTP_NULL | VTP_VALUE):
if (NULL == parm)
{
parmblk->value = (mval *)&literal_one;
break;
}
/* caution: fall through */
case VTP_VALUE:
if (NULL == parm)
rts_error_csa(CSA_ARG(NULL)
VARLSTCNT(4) ERR_VIEWARGCNT, 2, strlen((const char *)vtp->keyword), vtp->keyword);
parmblk->value = parm;
break;
case (VTP_NULL | VTP_DBREGION):
if (!is_dollar_view && ((NULL == parm) || ((1 == parm->str.len) && ('*' == *parm->str.addr))))
{
parmblk->gv_ptr = NULL;
break;
}
/* caution: fall through */
case VTP_DBREGION:
if (NULL == parm)
rts_error_csa(CSA_ARG(NULL)
VARLSTCNT(4) ERR_VIEWARGCNT, 2, strlen((const char *)vtp->keyword), vtp->keyword);
if (!gd_header) /* IF GD_HEADER ==0 THEN OPEN GBLDIR */
gvinit();
r_ptr = gd_header->regions;
if (!parm->str.len && vtp->keycode == VTK_GVNEXT) /* "" => 1st region */
parmblk->gv_ptr = r_ptr;
else
{
for (cptr = parm->str.addr, n = 0; n < parm->str.len; cptr++, n++)
lcl_buff.c[n] = TOUPPER(*cptr); /* Region names are upper-case ASCII */
namestr.len = n;
namestr.addr = &lcl_buff.c[0];
for (r_top = r_ptr + gd_header->n_regions; ; r_ptr++)
{
if (r_ptr >= r_top)
{
format2zwr((sm_uc_ptr_t)parm->str.addr, parm->str.len, global_names, &n);
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_NOREGION,2, n, global_names);
}
tmpstr.len = r_ptr->rname_len;
tmpstr.addr = (char *)r_ptr->rname;
MSTR_CMP(tmpstr, namestr, n);
if (0 == n)
break;
}
parmblk->gv_ptr = r_ptr;
}
break;
case VTP_DBKEY:
if (NULL == parm)
rts_error_csa(CSA_ARG(NULL)
VARLSTCNT(4) ERR_VIEWARGCNT, 2, strlen((const char *)vtp->keyword), vtp->keyword);
if (!parm->str.len)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_NOTGBL, 2, parm->str.len, NULL);
if (!gd_header) /* IF GD_HEADER ==0 THEN OPEN GBLDIR */
gvinit();
c = (unsigned char *)parm->str.addr;
if ('^' != *c)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_NOTGBL, 2, parm->str.len, c);
c_top = c + parm->str.len;
c++; /* skip initial '^' */
parmblk->str.addr = (char *)c;
for ( ; (c < c_top) && ('(' != *c); c++)
;
parmblk->str.len = (char *)c - parmblk->str.addr;
if (MAX_MIDENT_LEN < parmblk->str.len)
parmblk->str.len = MAX_MIDENT_LEN;
if (!valid_mname(&parmblk->str))
{
format2zwr((sm_uc_ptr_t)parm->str.addr, parm->str.len, global_names, &n);
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWGVN, 2, n, global_names);
}
break;
case VTP_RTNAME:
if (NULL == parm)
rts_error_csa(CSA_ARG(NULL)
VARLSTCNT(4) ERR_VIEWARGCNT, 2, strlen((const char *)vtp->keyword), vtp->keyword);
memset(&parmblk->ident.c[0], 0, SIZEOF(parmblk->ident));
if (parm->str.len > 0)
memcpy(&parmblk->ident.c[0], parm->str.addr,
(parm->str.len <= MAX_MIDENT_LEN ? parm->str.len : MAX_MIDENT_LEN));
break;
case VTP_NULL | VTP_DBKEYLIST:
if (NULL == parm || 0 == parm->str.len)
{
parmblk->ni_list.gvnh_list = NULL;
parmblk->ni_list.type = NOISOLATION_NULL;
break;
}
/* caution : explicit fall through */
case VTP_DBKEYLIST:
if (NULL == parm)
rts_error_csa(CSA_ARG(NULL)
VARLSTCNT(4) ERR_VIEWARGCNT, 2, strlen((const char *)vtp->keyword), vtp->keyword);
if (!gd_header)
gvinit();
if (first_time)
{
noisolation_buddy_list = (buddy_list *)malloc(SIZEOF(buddy_list));
initialize_list(noisolation_buddy_list, SIZEOF(noisolation_element), NOISOLATION_INIT_ALLOC);
gvt_pending_buddy_list = (buddy_list *)malloc(SIZEOF(buddy_list));
initialize_list(gvt_pending_buddy_list, SIZEOF(gvt_container), NOISOLATION_INIT_ALLOC);
first_time = FALSE;
}
assertpro(SIZEOF(global_names) > parm->str.len);
tmpstr.len = parm->str.len; /* we need to change len and should not change parm->str, so take a copy */
tmpstr.addr = parm->str.addr;
if (0 != tmpstr.len)
{
switch (*tmpstr.addr)
{
case '+' :
parmblk->ni_list.type = NOISOLATION_PLUS;
tmpstr.addr++;
tmpstr.len--;
break;
case '-' :
parmblk->ni_list.type = NOISOLATION_MINUS;
tmpstr.addr++;
tmpstr.len--;
break;
default :
parmblk->ni_list.type = NOISOLATION_NULL;
break;
}
if (!tmpstr.len)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWGVN, 2, tmpstr.len, NULL);
memcpy(global_names, tmpstr.addr, tmpstr.len);
global_names[tmpstr.len] = '\0';
src = (unsigned char *)STRTOK_R((char *)global_names, ",", &strtokptr);
REINITIALIZE_LIST(noisolation_buddy_list); /* reinitialize the noisolation buddy_list */
parmblk->ni_list.gvnh_list = NULL;
for ( ; src < &global_names[tmpstr.len + 1]; src = nextsrc)
{
nextsrc = (unsigned char *)STRTOK_R(NULL, ",", &strtokptr);
if (NULL == nextsrc)
nextsrc = &global_names[tmpstr.len + 1];
if (nextsrc - src >= 2 && '^' == *src)
{
namestr.addr = (char *)src + 1; /* skip initial '^' */
namestr.len = INTCAST(nextsrc - src - 2); /* don't count initial ^ and trailing 0 */
if (namestr.len > MAX_MIDENT_LEN)
namestr.len = MAX_MIDENT_LEN;
if (valid_mname(&namestr))
{
memcpy(&lcl_buff.c[0], namestr.addr, namestr.len);
gvent.var_name.len = namestr.len;
} else
{
memcpy(&lcl_buff.c[0], src, nextsrc - src - 1);
format2zwr((sm_uc_ptr_t)&lcl_buff.c, nextsrc - src - 1, global_names, &n);
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWGVN, 2, n, global_names);
}
} else
{
memcpy(&lcl_buff.c[0], src, nextsrc - src - 1);
format2zwr((sm_uc_ptr_t)&lcl_buff.c, nextsrc - src - 1, global_names, &n);
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWGVN, 2, n, global_names);
}
tmp_gvt = NULL;
gvent.var_name.addr = &lcl_buff.c[0];
COMPUTE_HASH_MNAME(&gvent);
if (NULL != (tabent = lookup_hashtab_mname(gd_header->tab_ptr, &gvent)))
{
gvnh_reg = (gvnh_reg_t *)tabent->value;
assert(NULL != gvnh_reg);
tmp_gvt = gvnh_reg->gvt;
} else
{
gd_map = gv_srch_map(gd_header, gvent.var_name.addr, gvent.var_name.len,
SKIP_BASEDB_OPEN_FALSE);
r_ptr = gd_map->reg.addr;
tmp_gvt = (gv_namehead *)targ_alloc(r_ptr->max_key_size, &gvent, r_ptr);
GVNH_REG_INIT(gd_header, gd_header->tab_ptr, gd_map, tmp_gvt,
r_ptr, gvnh_reg, tabent);
/* In case of a global spanning multiple regions, the gvt pointer corresponding to
* the region where the unsubscripted global reference maps to is stored in TWO
* locations (one in gvnh_reg->gvspan->gvt_array[index] and one in gvnh_reg->gvt.
* So pass in both these pointer addresses to be stored in the pending list in
* case this gvt gets reallocated (due to different keysizes between gld and db).
*/
if (NULL == (gvspan = gvnh_reg->gvspan))
{
ADD_TO_GVT_PENDING_LIST_IF_REG_NOT_OPEN(r_ptr, &gvnh_reg->gvt, NULL);
} else
{
gd_reg_start = &gd_header->regions[0];
GET_REG_INDEX(gd_header, gd_reg_start, r_ptr, reg_index);
/* the above sets "reg_index" */
assert(reg_index >= gvspan->min_reg_index);
assert(reg_index <= gvspan->max_reg_index);
reg_index -= gvspan->min_reg_index;
ADD_TO_GVT_PENDING_LIST_IF_REG_NOT_OPEN(r_ptr,
&gvspan->gvt_array[reg_index], &gvnh_reg->gvt);
}
}
ADD_GVT_TO_VIEW_NOISOLATION_LIST(tmp_gvt, parmblk);
if (!is_dollar_view && (NULL != gvnh_reg->gvspan))
{ /* Global spans multiple regions. Make sure gv_targets corresponding to ALL
* spanned regions are allocated so NOISOLATION status can be set in all of
* them even if the corresponding regions are not open yet. Do this only for
* VIEW "NOISOLATION" commands which change the noisolation characteristic.
* $VIEW("NOISOLATION") only examines the characteristics and so no need to
* allocate all the gv-targets in that case. Just one is enough.
*/
gvnh_spanreg_subs_gvt_init(gvnh_reg, gd_header, parmblk);
}
}
} else
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWGVN, 2, tmpstr.len, tmpstr.addr);
break;
case VTP_LVN:
if (NULL == parm)
rts_error_csa(CSA_ARG(NULL)
VARLSTCNT(4) ERR_VIEWARGCNT, 2, strlen((const char *)vtp->keyword), vtp->keyword);
if (0 < parm->str.len)
{
lvent.var_name.addr = parm->str.addr;
lvent.var_name.len = parm->str.len;
if (lvent.var_name.len > MAX_MIDENT_LEN)
lvent.var_name.len = MAX_MIDENT_LEN;
if (!valid_mname(&lvent.var_name))
{
format2zwr((sm_uc_ptr_t)parm->str.addr, parm->str.len, global_names, &n);
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWLVN, 2, n, global_names);
}
} else
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWLVN, 2, parm->str.len, parm->str.addr);
/* Now look up the name.. */
COMPUTE_HASH_MNAME(&lvent);
if ((tabent = lookup_hashtab_mname(&curr_symval->h_symtab, &lvent)) && (NULL != tabent->value))
parmblk->value = (mval *)tabent->value; /* Return lv_val ptr */
else
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_VIEWLVN, 2, parm->str.len, parm->str.addr);
break;
default:
assertpro(FALSE && vtp_parm);
}
}
uint4 jnl_file_lost(jnl_private_control *jpc, uint4 jnl_stat)
{ /* Notify operator and terminate journaling */
unsigned int status;
sgmnt_addrs *csa;
seq_num reg_seqno, jnlseqno;
boolean_t was_lockid = FALSE, instfreeze_environ;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
switch(jpc->region->dyn.addr->acc_meth)
{
case dba_mm:
case dba_bg:
csa = &FILE_INFO(jpc->region)->s_addrs;
break;
default:
assertpro(FALSE && jpc->region->dyn.addr->acc_meth);
}
# ifdef VMS
/* The following assert has been removed as it could be FALSE if the caller is "jnl_file_extend"
* assert(0 != memcmp(csa->nl->jnl_file.jnl_file_id.fid, zero_fid, SIZEOF(zero_fid)));
*/
# endif
assert(csa->now_crit);
/* We issue an rts_error (instead of shutting off journaling) in the following cases : {BYPASSOK}
* 1) $gtm_error_on_jnl_file_lost is set to issue runtime error (if not already issued) in case of journaling issues.
* 2) The process has the given message set in $gtm_custom_errors (indicative of instance freeze on error setup)
* in which case the goal is to never shut-off journaling
*/
UNIX_ONLY(assert(jnlpool.jnlpool_ctl == jnlpool_ctl));
UNIX_ONLY(instfreeze_environ = INST_FREEZE_ON_MSG_ENABLED(csa, jnl_stat));
VMS_ONLY(instfreeze_environ = FALSE);
if ((JNL_FILE_LOST_ERRORS == TREF(error_on_jnl_file_lost)) || instfreeze_environ)
{
VMS_ONLY(assert(FALSE)); /* Not fully implemented / supported on VMS. */
if (!process_exiting || instfreeze_environ || !csa->jnl->error_reported)
{
csa->jnl->error_reported = TRUE;
in_wcs_recover = FALSE; /* in case we're called in wcs_recover() */
if (SS_NORMAL != jpc->status)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(7) jnl_stat, 4, JNL_LEN_STR(csa->hdr),
DB_LEN_STR(gv_cur_region), jpc->status);
else
rts_error_csa(CSA_ARG(csa) VARLSTCNT(6) jnl_stat, 4, JNL_LEN_STR(csa->hdr),
DB_LEN_STR(gv_cur_region));
}
return jnl_stat;
}
if (0 != jnl_stat)
jnl_send_oper(jpc, jnl_stat);
csa->hdr->jnl_state = jnl_closed;
jpc->jnl_buff->cycle++; /* increment shared cycle so all future callers of jnl_ensure_open recognize journal switch */
assert(jpc->cycle < jpc->jnl_buff->cycle);
if (REPL_ENABLED(csa->hdr))
{
csa->hdr->repl_state = repl_was_open;
reg_seqno = csa->hdr->reg_seqno;
jnlseqno = (NULL != jnlpool.jnlpool_ctl) ? jnlpool.jnlpool_ctl->jnl_seqno : MAX_SEQNO;
send_msg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_REPLJNLCLOSED, 6, DB_LEN_STR(jpc->region), ®_seqno, ®_seqno,
&jnlseqno, &jnlseqno);
} else
send_msg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_JNLCLOSED, 3, DB_LEN_STR(jpc->region), &csa->ti->curr_tn);
#ifdef VMS
/* We can get a jnl_file_lost before the file is even created, so locking is done only if the lock exist */
if (0 != csa->jnl->jnllsb->lockid)
{
was_lockid = TRUE;
status = gtm_enqw(EFN$C_ENF, LCK$K_EXMODE, csa->jnl->jnllsb, LCK$M_CONVERT | LCK$M_NODLCKBLK,
NULL, 0, NULL, 0, NULL, PSL$C_USER, 0);
if (SS$_NORMAL == status)
status = csa->jnl->jnllsb->cond;
}
jnl_file_close(jpc->region, FALSE, FALSE);
if (was_lockid)
{
if (SS$_NORMAL == status)
status = gtm_deq(csa->jnl->jnllsb->lockid, NULL, PSL$C_USER, 0);
assertpro(SS$_NORMAL == status);
}
# else
jnl_file_close(jpc->region, FALSE, FALSE);
#endif
return EXIT_NRM;
}
/*
* This will rundown a replication instance journal (and receiver) pool.
* Input Parameter:
* replpool_id of the instance. Instance file name must be null terminated in replpool_id.
* Returns :
* TRUE, if successful.
* FALSE, otherwise.
*/
boolean_t mu_rndwn_repl_instance(replpool_identifier *replpool_id, boolean_t immediate, boolean_t rndwn_both_pools,
boolean_t *jnlpool_sem_created)
{
boolean_t jnlpool_stat = SS_NORMAL, recvpool_stat = SS_NORMAL, decr_cnt, sem_created = FALSE, ipc_rmvd;
char *instfilename;
unsigned char ipcs_buff[MAX_IPCS_ID_BUF], *ipcs_ptr;
gd_region *r_save;
repl_inst_hdr repl_instance;
static gd_region *reg = NULL;
struct semid_ds semstat;
struct shmid_ds shmstat;
unix_db_info *udi;
int save_errno, sem_id, shm_id, status;
sgmnt_addrs *repl_csa;
boolean_t was_crit;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
if (NULL == reg)
{
r_save = gv_cur_region;
mu_gv_cur_reg_init();
reg = gv_cur_region;
gv_cur_region = r_save;
}
*jnlpool_sem_created = FALSE;
/* Assert that the layout of replpool_identifier is identical for all versions going forward as the function
* "validate_replpool_shm_entry" (used by the argumentless mupip rundown aka "mupip rundown") relies on this.
* This assert is placed here (instead of there) because the automated tests exercise this logic much more
* than the argumentless code. If any of these asserts fail, "validate_replpool_shm_entry" needs to change
* to handle the old and new layouts.
*
* Structure ----> replpool_identifier <---- size 312 [0x0138]
*
* offset = 0000 [0x0000] size = 0012 [0x000c] ----> replpool_identifier.label
* offset = 0012 [0x000c] size = 0001 [0x0001] ----> replpool_identifier.pool_type
* offset = 0013 [0x000d] size = 0036 [0x0024] ----> replpool_identifier.now_running
* offset = 0052 [0x0034] size = 0004 [0x0004] ----> replpool_identifier.repl_pool_key_filler
* offset = 0056 [0x0038] size = 0256 [0x0100] ----> replpool_identifier.instfilename
*/
assert(0 == OFFSETOF(replpool_identifier, label[0]));
assert(12 == SIZEOF(((replpool_identifier *)NULL)->label));
assert(12 == OFFSETOF(replpool_identifier, pool_type));
assert(1 == SIZEOF(((replpool_identifier *)NULL)->pool_type));
assert(13 == OFFSETOF(replpool_identifier, now_running[0]));
assert(36 == SIZEOF(((replpool_identifier *)NULL)->now_running));
assert(56 == OFFSETOF(replpool_identifier, instfilename[0]));
assert(256 == SIZEOF(((replpool_identifier *)NULL)->instfilename));
/* End asserts */
jnlpool.jnlpool_dummy_reg = reg;
recvpool.recvpool_dummy_reg = reg;
instfilename = replpool_id->instfilename;
reg->dyn.addr->fname_len = strlen(instfilename);
assert(0 == instfilename[reg->dyn.addr->fname_len]);
memcpy((char *)reg->dyn.addr->fname, instfilename, reg->dyn.addr->fname_len + 1);
udi = FILE_INFO(reg);
udi->fn = (char *)reg->dyn.addr->fname;
/* Lock replication instance using ftok semaphore so that no other replication process can startup until we are done with
* rundown
*/
if (!ftok_sem_get(reg, TRUE, REPLPOOL_ID, immediate))
return FALSE;
ESTABLISH_RET(mu_rndwn_repl_instance_ch, FALSE);
repl_inst_read(instfilename, (off_t)0, (sm_uc_ptr_t)&repl_instance, SIZEOF(repl_inst_hdr));
assert(rndwn_both_pools || JNLPOOL_SEGMENT == replpool_id->pool_type || RECVPOOL_SEGMENT == replpool_id->pool_type);
if (rndwn_both_pools || (JNLPOOL_SEGMENT == replpool_id->pool_type))
{ /* --------------------------
* First rundown Journal pool
* --------------------------
*/
shm_id = repl_instance.jnlpool_shmid;
if (SS_NORMAL == (jnlpool_stat = mu_replpool_grab_sem(&repl_instance, JNLPOOL_SEGMENT, &sem_created, immediate)))
{
/* Got JNL_POOL_ACCESS_SEM and incremented SRC_SRV_COUNT_SEM */
assert(holds_sem[SOURCE][JNL_POOL_ACCESS_SEM]);
assert(holds_sem[SOURCE][SRC_SERV_COUNT_SEM]);
sem_id = repl_instance.jnlpool_semid;
if ((INVALID_SHMID == shm_id) || (-1 == shmctl(shm_id, IPC_STAT, &shmstat))
|| (shmstat.shm_ctime != repl_instance.jnlpool_shmid_ctime))
{
repl_instance.jnlpool_shmid = shm_id = INVALID_SHMID;
repl_instance.jnlpool_shmid_ctime = 0;
}
assert((INVALID_SHMID != shm_id) || ((NULL == jnlpool.jnlpool_ctl) && (NULL == jnlpool_ctl)));
ipc_rmvd = TRUE;
if (INVALID_SHMID != shm_id)
{
replpool_id->pool_type = JNLPOOL_SEGMENT;
jnlpool_stat = mu_rndwn_replpool(replpool_id, &repl_instance, shm_id, &ipc_rmvd);
ipcs_ptr = i2asc((uchar_ptr_t)ipcs_buff, shm_id);
*ipcs_ptr = '\0';
if (rndwn_both_pools && ((SS_NORMAL != jnlpool_stat) || ipc_rmvd))
gtm_putmsg(VARLSTCNT(6) (jnlpool_stat ? ERR_MUJPOOLRNDWNFL : ERR_MUJPOOLRNDWNSUC),
4, LEN_AND_STR(ipcs_buff), LEN_AND_STR(instfilename));
}
assert(ipc_rmvd || (NULL != jnlpool_ctl));
assert((NULL == jnlpool.jnlpool_ctl) || (SS_NORMAL == jnlpool_stat) || jgbl.onlnrlbk);
assert((INVALID_SHMID != repl_instance.jnlpool_shmid) || (0 == repl_instance.jnlpool_shmid_ctime));
assert((INVALID_SHMID == repl_instance.jnlpool_shmid) || (0 != repl_instance.jnlpool_shmid_ctime));
assert(INVALID_SEMID != sem_id);
if (!mur_options.rollback)
{ /* Invoked by MUPIP RUNDOWN in which case the semaphores needs to be removed. But, remove the
* semaphore ONLY if we created it here OR the journal pool was successfully removed.
*/
if (NULL == jnlpool_ctl)
{
if (((sem_created || (SS_NORMAL == jnlpool_stat))
&& (SS_NORMAL == mu_replpool_release_sem(&repl_instance, JNLPOOL_SEGMENT, TRUE))))
{ /* Now that semaphores are removed, reset fields in file header */
if (!sem_created)
{ /* If sem_id was created by mu_replpool_grab_sem then do NOT report the
* MURPOOLRNDWNSUC message as it indicates that the semaphore was orphaned
* and we removed it when in fact there was no orphaned semaphore and we
* created it as part of mu_replpool_grab_sem to get standalone access to
* rundown the receiver pool (which may or may not exist)
*/
ipcs_ptr = i2asc((uchar_ptr_t)ipcs_buff, sem_id);
*ipcs_ptr = '\0';
gtm_putmsg(VARLSTCNT(9) ERR_MUJPOOLRNDWNSUC, 4, LEN_AND_STR(ipcs_buff),
LEN_AND_STR(instfilename), ERR_SEMREMOVED, 1, sem_id);
}
repl_inst_jnlpool_reset();
}
} else
{ /* Anticipatory Freeze scheme is turned ON. So, release just the JNL_POOL_ACCESS_SEM. The
* semaphore will be released/removed in the caller (mupip_rundown)
*/
assert(ANTICIPATORY_FREEZE_AVAILABLE);
assertpro(SS_NORMAL == (status = rel_sem(SOURCE, JNL_POOL_ACCESS_SEM)));
assert(!holds_sem[SOURCE][JNL_POOL_ACCESS_SEM]);
/* Since we are not resetting the semaphore IDs in the file header, we need to write out
* the semaphore IDs in the instance file (if we created them).
*/
if (sem_created)
repl_inst_write(instfilename, (off_t)0, (sm_uc_ptr_t)&repl_instance,
SIZEOF(repl_inst_hdr));
}
/* If semaphore is not created and the journal pool rundown failed (due to attached processes),
* rundown process continues to holds the journal pool access control semaphore. This way, we hold
* the semaphore on behalf of the source server (now no longer alive) to prevent mu_rndwn_sem_all
* (invoked later) from cleaning up this orphaned semaphore (which causes REPLREQROLLBACK if the
* source server is restarted). But, since the semaphore is not released (until the rundown process
* dies), holds_sem[SOURCE][JNL_POOL_ACCESS_SEM] continues to remain TRUE. This causes asserts in
* ftok_sem_get if mu_rndwn_repl_instance is invoked for a different journal/receive pool. To
* workaround it, set holds_sem[SOURCE][JNL_POOL_ACCESS_SEM] to FALSE. This is an interim solution
* until we record such semaphores in an ignore-list (or some such) and change mu_rndwn_sem_all to
* skip the ones that are present in the ignore list.
*/
holds_sem[SOURCE][JNL_POOL_ACCESS_SEM] = FALSE;
}
} else if (rndwn_both_pools && (INVALID_SHMID != shm_id))
{
ipcs_ptr = i2asc((uchar_ptr_t)ipcs_buff, shm_id);
*ipcs_ptr = '\0';
if (rndwn_both_pools)
gtm_putmsg(VARLSTCNT(6) ERR_MUJPOOLRNDWNFL, 4, LEN_AND_STR(ipcs_buff),
LEN_AND_STR(instfilename));
}
*jnlpool_sem_created = sem_created;
}
if (((SS_NORMAL == jnlpool_stat) || !jgbl.mur_rollback) &&
(rndwn_both_pools || (RECVPOOL_SEGMENT == replpool_id->pool_type)))
{ /* --------------------------
* Now rundown Receivpool
* --------------------------
* Note: RECVPOOL is rundown ONLY if the JNLPOOL rundown was successful. This way, we don't end up
* creating new semaphores for the RECVPOOL if ROLLBACK is not going to start anyways because of the failed
* JNLPOOL rundown. The only exception is MUPIP RUNDOWN command in which case we try running down the
* RECVPOOL even if the JNLPOOL rundown failed.
*/
shm_id = repl_instance.recvpool_shmid;
if (SS_NORMAL == (recvpool_stat = mu_replpool_grab_sem(&repl_instance, RECVPOOL_SEGMENT, &sem_created, immediate)))
{
sem_id = repl_instance.recvpool_semid;
if ((INVALID_SHMID == shm_id) || (-1 == shmctl(shm_id, IPC_STAT, &shmstat))
|| (shmstat.shm_ctime != repl_instance.recvpool_shmid_ctime))
{
repl_instance.recvpool_shmid = shm_id = INVALID_SHMID;
repl_instance.recvpool_shmid_ctime = 0;
}
ipc_rmvd = TRUE;
if (INVALID_SHMID != shm_id)
{
replpool_id->pool_type = RECVPOOL_SEGMENT;
recvpool_stat = mu_rndwn_replpool(replpool_id, &repl_instance, shm_id, &ipc_rmvd);
ipcs_ptr = i2asc((uchar_ptr_t)ipcs_buff, shm_id);
*ipcs_ptr = '\0';
if (rndwn_both_pools && ((SS_NORMAL != recvpool_stat) || ipc_rmvd))
gtm_putmsg(VARLSTCNT(6) (recvpool_stat ? ERR_MURPOOLRNDWNFL : ERR_MURPOOLRNDWNSUC),
4, LEN_AND_STR(ipcs_buff), LEN_AND_STR(instfilename));
}
assert((TRUE == ipc_rmvd) || (SS_NORMAL != recvpool_stat) || jgbl.onlnrlbk);
assert((INVALID_SHMID != repl_instance.recvpool_shmid) || (0 == repl_instance.recvpool_shmid_ctime));
assert((INVALID_SHMID == repl_instance.recvpool_shmid) || (0 != repl_instance.recvpool_shmid_ctime));
assert(INVALID_SEMID != sem_id);
if (!mur_options.rollback)
{ /* Invoked by MUPIP RUNDOWN in which case the semaphores needs to be removed. But, remove the
* semaphore ONLY if we created it here OR the receive pool was successfully removed.
*/
if ((sem_created || (SS_NORMAL == recvpool_stat))
&& (SS_NORMAL == mu_replpool_release_sem(&repl_instance, RECVPOOL_SEGMENT, TRUE)))
{ /* Now that semaphores are removed, reset fields in file header */
if (!sem_created)
{ /* if sem_id was "created" by mu_replpool_grab_sem then do NOT report the
* MURPOOLRNDWNSUC message as it indicates that the semaphore was orphaned and we
* removed it when in fact there was no orphaned semaphore and we "created" it as
* part of mu_replpool_grab_sem to get standalone access to rundown the receiver
* pool (which may or may not exist)
*/
ipcs_ptr = i2asc((uchar_ptr_t)ipcs_buff, sem_id);
*ipcs_ptr = '\0';
gtm_putmsg(VARLSTCNT(9) ERR_MURPOOLRNDWNSUC, 4, LEN_AND_STR(ipcs_buff),
LEN_AND_STR(instfilename), ERR_SEMREMOVED, 1, sem_id);
}
if (NULL != jnlpool_ctl)
{ /* Journal pool is not yet removed. So, grab lock before resetting semid/shmid
* fields in the file header as the function expects the caller to hold crit
* if the journal pool is available
*/
repl_csa = &FILE_INFO(jnlpool.jnlpool_dummy_reg)->s_addrs;
assert(!repl_csa->now_crit);
assert(!repl_csa->hold_onto_crit);
was_crit = repl_csa->now_crit;
/* Since we do grab_lock, below, we need to do a per-process initialization. Also,
* start heartbeat so that grab_lock can issue MUTEXLCKALERT and get C-stacks if
* waiting for crit
*/
START_HEARTBEAT_IF_NEEDED;
mutex_per_process_init();
if (!was_crit)
grab_lock(jnlpool.jnlpool_dummy_reg, TRUE, GRAB_LOCK_ONLY);
}
repl_inst_recvpool_reset();
if ((NULL != jnlpool_ctl) && !was_crit)
rel_lock(jnlpool.jnlpool_dummy_reg);
}
/* If semaphore is not created and the receive pool rundown failed (due to attached processes),
* rundown process continues to holds the receive pool access control semaphore. This way, we hold
* the semaphore on behalf of the receiver server (now no longer alive) to prevent mu_rndwn_sem_all
* (invoked later) from cleaning up this orphaned semaphore (which causes REPLREQROLLBACK if the
* receiver is restarted). But, since the semaphore is not released (until the rundown process
* dies), holds_sem[RECV][RECV_POOL_ACCESS_SEM] continues to remain TRUE. This causes asserts in
* ftok_sem_get if mu_rndwn_repl_instance is invoked for a different journal/receive pool. To
* workaround it, set holds_sem[SOURCE][RECV_POOL_ACCESS_SEM] to FALSE. This is an interim solution
* until we record such semaphores in an ignore-list (or some such) and change mu_rndwn_sem_all to
* skip the ones that are present in the ignore list.
*/
assert((sem_created || (SS_NORMAL == recvpool_stat)) || holds_sem[RECV][RECV_POOL_ACCESS_SEM]);
DEBUG_ONLY(set_sem_set_recvr(sem_id));
}
} else if (rndwn_both_pools && (INVALID_SHMID != shm_id))
{
ipcs_ptr = i2asc((uchar_ptr_t)ipcs_buff, shm_id);
*ipcs_ptr = '\0';
if (rndwn_both_pools)
gtm_putmsg(VARLSTCNT(6) ERR_MURPOOLRNDWNFL, 4, LEN_AND_STR(ipcs_buff),
LEN_AND_STR(instfilename));
}
}
assert(jgbl.onlnrlbk || ANTICIPATORY_FREEZE_AVAILABLE || (NULL == jnlpool.repl_inst_filehdr));
if (mur_options.rollback && (SS_NORMAL == jnlpool_stat) && (SS_NORMAL == recvpool_stat))
{
assert(jgbl.onlnrlbk || ANTICIPATORY_FREEZE_AVAILABLE || ((INVALID_SHMID == repl_instance.jnlpool_shmid)
&& (INVALID_SHMID == repl_instance.recvpool_shmid)));
/* Initialize jnlpool.repl_inst_filehdr as it is used later by gtmrecv_fetchresync() */
decr_cnt = FALSE;
if (NULL == jnlpool.repl_inst_filehdr)
{ /* Possible if there is NO journal pool in the first place. In this case, malloc the structure here and
* copy the file header from repl_instance structure.
*/
jnlpool.repl_inst_filehdr = (repl_inst_hdr_ptr_t)malloc(SIZEOF(repl_inst_hdr));
memcpy(jnlpool.repl_inst_filehdr, &repl_instance, SIZEOF(repl_inst_hdr));
} else
{
assert(repl_instance.jnlpool_semid == jnlpool.repl_inst_filehdr->jnlpool_semid);
assert(repl_instance.jnlpool_semid_ctime == jnlpool.repl_inst_filehdr->jnlpool_semid_ctime);
assert(repl_instance.jnlpool_shmid == jnlpool.repl_inst_filehdr->jnlpool_shmid);
assert(repl_instance.jnlpool_shmid_ctime == jnlpool.repl_inst_filehdr->jnlpool_shmid_ctime);
/* If the ONLINE ROLLBACK command is run on the primary when the source server is up and running,
* jnlpool.repl_inst_filehdr->recvpool_semid will be INVALID because there is NO receiver server
* running. However, ROLLBACK creates semaphores for both journal pool and receive pool and writes
* it to the instance file header. Copy this information to the file header copy in the jnlpool
* as well
*/
jnlpool.repl_inst_filehdr->recvpool_semid = repl_instance.recvpool_semid;
jnlpool.repl_inst_filehdr->recvpool_semid_ctime = repl_instance.recvpool_semid_ctime;
}
/* Flush changes to the replication instance file header to disk */
repl_inst_write(instfilename, (off_t)0, (sm_uc_ptr_t)&repl_instance, SIZEOF(repl_inst_hdr));
} else /* for MUPIP RUNDOWN, semid fields in the file header are reset and is written in mu_replpool_release_sem() above */
decr_cnt = (NULL == jnlpool_ctl); /* for anticipatory freeze, mupip_rundown releases the semaphore */
REVERT;
/* Release replication instance ftok semaphore lock */
if (!ftok_sem_release(reg, decr_cnt, immediate)) /* Do not decrement the counter if ROLLBACK */
return FALSE;
return ((SS_NORMAL == jnlpool_stat) && (SS_NORMAL == recvpool_stat));
}
void ojparams (char *p, job_params_type *job_params)
{
unsigned char ch;
int4 status;
mstr_len_t handle_len;
/* Initializations */
job_params->baspri = 0;
job_params->input.len = 0;
job_params->output.len = 0;
job_params->error.len = 0;
job_params->gbldir.len = 0;
job_params->startup.len = 0;
job_params->directory.len = 0;
job_params->directory.addr = 0;
job_params->cmdline.len = 0;
job_params->cmdline.addr = 0;
job_params->passcurlvn = FALSE;
/* Process parameter list */
while (*p != jp_eol)
{
switch (ch = *p++)
{
case jp_default:
if (*p != 0)
{
job_params->directory.len = (int)((unsigned char) *p);
job_params->directory.addr = (p + 1);
}
break;
case jp_error:
if (*p != 0)
{
job_params->error.len = (int)((unsigned char) *p);
job_params->error.addr = (p + 1);
}
break;
case jp_gbldir:
if (*p != 0)
{
job_params->gbldir.len = (int)((unsigned char) *p);
job_params->gbldir.addr = (p + 1);
}
break;
case jp_input:
if (*p != 0)
{
job_params->input.len = (int)((unsigned char) *p);
job_params->input.addr = p + 1;
}
break;
case jp_output:
if (*p != 0)
{
job_params->output.len = (int)((unsigned char) *p);
job_params->output.addr = p + 1;
}
break;
case jp_priority:
job_params->baspri = (int4)(*((int4 *)p));
break;
case jp_startup:
if (*p != 0)
{
job_params->startup.len = (int)((unsigned char) *p);
job_params->startup.addr = p + 1;
}
break;
case jp_cmdline:
if(*p != 0)
{
job_params->cmdline.len = (int)((unsigned char) *p);
job_params->cmdline.addr = p + 1;
}
break;
case jp_passcurlvn:
job_params->passcurlvn = TRUE;
break;
case jp_account:
case jp_detached:
case jp_image:
case jp_logfile:
case jp_noaccount:
case jp_nodetached:
case jp_noswapping:
case jp_process_name:
case jp_schedule:
case jp_swapping:
break;
default:
assertpro(ch != ch);
}
switch (job_param_datatypes[ch])
{
case jpdt_nul:
break;
case jpdt_num:
p += SIZEOF(int4);
break;
case jpdt_str:
p += ((int)((unsigned char)*p)) + 1;
break;
default:
assertpro((jpdt_nul == job_param_datatypes[ch])
|| (jpdt_num == job_param_datatypes[ch])
|| (jpdt_str == job_param_datatypes[ch]));
}
}
/* Defaults and Checks */
/*
* Input file
*/
if (job_params->input.len == 0)
{
job_params->input.len = STRLEN(definput);
job_params->input.addr = definput;
}
else if (IS_JOB_SOCKET(job_params->input.addr, job_params->input.len))
{
handle_len = JOB_SOCKET_HANDLE_LEN(job_params->input.len);
if ((NULL == socket_pool) || (-1 == iosocket_handle(JOB_SOCKET_HANDLE(job_params->input.addr),
&handle_len, FALSE, socket_pool)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 5, "INPUT",
job_params->input.len, job_params->input.addr);
}
else
if (!(status = ojchkfs (job_params->input.addr,
job_params->input.len, TRUE)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 5, "INPUT",
job_params->input.len, job_params->input.addr);
/*
* Output file
*/
if (job_params->output.len == 0)
{
if (!defoutbuf)
defoutbuf = malloc(MAX_FILSPC_LEN);
memcpy (&defoutbuf[0], job_params->routine.addr,
job_params->routine.len);
memcpy (&defoutbuf[job_params->routine.len],
defoutext.addr, defoutext.len);
if (*defoutbuf == '%')
*defoutbuf = '_';
job_params->output.len = job_params->routine.len
+ defoutext.len;
job_params->output.addr = &defoutbuf[0];
}
else if (IS_JOB_SOCKET(job_params->output.addr, job_params->output.len))
{
handle_len = JOB_SOCKET_HANDLE_LEN(job_params->output.len);
if ((NULL == socket_pool) || (-1 == iosocket_handle(JOB_SOCKET_HANDLE(job_params->output.addr),
&handle_len, FALSE, socket_pool)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 5, "OUTPUT",
job_params->output.len, job_params->output.addr);
}
else
if (!(status = ojchkfs (job_params->output.addr,
job_params->output.len, FALSE)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 6,
"OUTPUT", job_params->output.len,
job_params->output.addr);
/*
* Error file
*/
if (job_params->error.len == 0)
{
if (!deferrbuf)
deferrbuf = malloc(MAX_FILSPC_LEN);
memcpy (&deferrbuf[0], job_params->routine.addr,
job_params->routine.len);
memcpy (&deferrbuf[job_params->routine.len],
deferrext.addr, deferrext.len);
if (*deferrbuf == '%')
*deferrbuf = '_';
job_params->error.len = job_params->routine.len
+ deferrext.len;
job_params->error.addr = &deferrbuf[0];
}
else if (IS_JOB_SOCKET(job_params->error.addr, job_params->error.len))
{
handle_len = JOB_SOCKET_HANDLE_LEN(job_params->error.len);
if ((NULL == socket_pool) || (-1 == iosocket_handle(JOB_SOCKET_HANDLE(job_params->error.addr),
&handle_len, FALSE, socket_pool)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 5, "ERROR",
job_params->error.len, job_params->error.addr);
}
else
if (!(status = ojchkfs (job_params->error.addr,
job_params->error.len, FALSE)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 5, "ERROR",
job_params->error.len,
job_params->error.addr);
/*
* Global Directory
*/
if (job_params->gbldir.len)
if (!(status = ojchkfs (job_params->gbldir.addr,
job_params->gbldir.len, FALSE)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 6, "GBLDIR",
job_params->gbldir.len, job_params->gbldir.addr);
/*
* Startup
*/
if (job_params->startup.len)
if (!(status = ojchkfs (job_params->startup.addr,
job_params->startup.len, TRUE)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 7, "STARTUP",
job_params->startup.len, job_params->startup.addr);
/*
* Default Directory
*/
if (job_params->directory.len)
if (!(status = ojchkfs (job_params->directory.addr,
job_params->directory.len, FALSE)))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_PARFILSPC, 4, 7, "DEFAULT",
job_params->directory.len, job_params->directory.addr);
}
/* ------------------------------------------------------------------
* Reset transfer table to normal settings.
*
* - Intent: Put back all state that was or could have been changed
* due to prior deferral(s).
* - Would be easier to implement this assumption if this routine
* were changed to delegate responsibility as does the
* corresponding set routine.
* - Note that all events are reenabled before user's handler
* would be executed (assuming one is appropriate for this event
* and has been specified)
* => It's possible to have handler-in-handler execution.
* => If no handler executed, would lose other deferred events due
* to reset of all pending.
* - If M profiling is active, some entries should be set to the
* op_mprof* routines.
* - Return value indicates whether reset type matches set type.
* If it does not, this indicates an "abnormal" path.
* - Should still reset the table in this case.
* - BUT: Consider also calling a reset routine for all setters
* that have been logged, to allow them to reset themselves,
* (for example, to reset TP timer & flags, or anything else
* that could cause unintended effects if left set after
* deferred events have been cleared).
* - May need to update behavior to ensure it doesn't miss a
* critical event between registration of first event
* and clearing of all events. This seems problematic only if
* the following are true:
* - Two events are deferred at one time (call them A and B).
* - An M exception handler (ZTRAP or device) is required to
* execute due to B and perform action X.
* - Either no handler executes due to A, or the handler that
* does execute does not perform action X in response to B
* (this includes the possibility of performing X but not
* as needed by B, e.g. perhaps it should happen for both
* A and B but only happens for A).
* Seems like most or all of these can be addressed by carefully
* specifying coding requirements on M handlers.
* ------------------------------------------------------------------
*/
boolean_t xfer_reset_handlers(int4 event_type)
{
int4 e_type;
boolean_t reset_type_is_set_type;
int4 status;
int e, ei, e_tot = 0;
/* ------------------------------------------------------------------
* Note: If reset routine can preempt path from handler to
* set routine (e.g. clearing event before acting on it),
* these assertions can fail.
* Should not happen in current design.
* ------------------------------------------------------------------
*/
assert(0 < num_deferred);
assert(0 < xfer_table_events[event_type]);
if (is_tracing_on)
{
FIX_XFER_ENTRY(xf_linefetch, op_mproflinefetch);
FIX_XFER_ENTRY(xf_linestart, op_mproflinestart);
FIX_XFER_ENTRY(xf_forchk1, op_mprofforchk1);
} else
{
FIX_XFER_ENTRY(xf_linefetch, op_linefetch);
FIX_XFER_ENTRY(xf_linestart, op_linestart);
FIX_XFER_ENTRY(xf_forchk1, op_forchk1);
}
FIX_XFER_ENTRY(xf_forloop, op_forloop);
FIX_XFER_ENTRY(xf_zbfetch, op_zbfetch);
FIX_XFER_ENTRY(xf_zbstart, op_zbstart);
FIX_XFER_ENTRY(xf_ret, opp_ret);
FIX_XFER_ENTRY(xf_retarg, op_retarg);
DBGDFRDEVNT((stderr, "xfer_reset_handlers: Reset xfer_table for event type %d.\n", event_type));
reset_type_is_set_type = (event_type == first_event);
# ifdef DEBUG
if (!reset_type_is_set_type)
rts_error(VARLSTCNT(4) ERR_DEFEREVENT, 2, event_type, first_event);
# endif
# ifdef DEBUG_DEFERRED_EVENT
/* Note: concurrent modification of array elements means events that occur during this section will
* cause inconsistent totals.
*/
for (ei = no_event; ei < DEFERRED_EVENTS; ei++)
e_tot += xfer_table_events[ei];
if (1 < e_tot)
{
DBGDFRDEVNT((stderr, "xfer_reset_handlers: Event Log:\n"));
for (ei=no_event; ei<DEFERRED_EVENTS; ei++)
DBGDFRDEVNT((stderr, "xfer_reset_handlers: Event type %d: count was %d.\n",
ei, xfer_table_events[ei]));
}
# endif
/* -------------------------------------------------------------------------
* Kluge(?): set all locations to nonzero value to
* prevent interleaving with reset activities.
*
* Would be better to aswp with 0:
* - Won't lose any new events that way.
* -------------------------------------------------------------------------
*/
for (e_type = 1; DEFERRED_EVENTS > e_type; e_type++)
{
xfer_table_events[e_type] = 1;
}
/* -------------------------------------------------------------------------
* Reset external event modules that need it.
* (Should do this in a more modular fashion.)
* None
* -------------------------------------------------------------------------
*/
/* --------------------------------------------
* Reset private variables.
* --------------------------------------------
*/
first_event = no_event;
num_deferred = 0;
ctrap_action_is = 0;
outofband = 0;
# ifdef VMS
status = sys$clref(efn_outofband);
assert(SS$_WASSET == status);
assertpro((SS$_WASSET == status) || (SS$_WASCLR == status));
# endif
/* ******************************************************************
* There is a race here:
* If a new event interrupts after previous line and before
* corresponding assignment in next loop, it will be missed.
* For most events, we're going to an M handler anyway, so it won't
* matter (assuming the handler would handle all pending events).
* But if not going to an M handler (e.g. if resetting zbreak/zstep),
* could miss another event.
*
* Better (to avoid missing any events):
* aswp xfer_table_events elements (as described above), and
* check here if still zero. If not, must have missed that event
* since aswp, possibly before num_deferred was reset => never set
* xfer_table => should do that now.
* If more than one is nonzero, choose first arbitrarily
* unless first_event is now set -- unless it is, we've lost track of
* which event was first.
* ******************************************************************
*/
/* Clear to allow new events to be reset only after we're all done. */
for (e_type = 1; DEFERRED_EVENTS > e_type; e_type++)
xfer_table_events[e_type] = FALSE;
return reset_type_is_set_type;
}
int mur_forward_multi_proc(reg_ctl_list *rctl)
{
boolean_t multi_proc, this_reg_stuck, release_latch, ok_to_play;
boolean_t cancelled_dbsync_timer, cancelled_timer;
reg_ctl_list *rctl_top, *prev_rctl;
jnl_ctl_list *jctl;
gd_region *reg;
sgmnt_addrs *csa;
seq_num rec_token_seq;
jnl_tm_t rec_time;
enum broken_type recstat;
jnl_record *rec;
enum jnl_record_type rectype;
char errstr[256];
int i, rctl_index, save_errno, num_procs_stuck, num_reg_stuck;
uint4 status, regcnt_stuck, num_partners, start_hrtbt_cntr;
forw_multi_struct *forw_multi;
shm_forw_multi_t *sfm;
multi_struct *multi;
jnl_tm_t adjusted_resolve_time;
shm_reg_ctl_t *shm_rctl_start, *shm_rctl, *first_shm_rctl;
size_t shm_size, reccnt, copy_size;
int4 *size_ptr;
char *shmPtr; /* not using "shm_ptr" since it is already used in an AIX include file */
int shmid;
multi_proc_shm_hdr_t *mp_hdr; /* Pointer to "multi_proc_shm_hdr_t" structure in shared memory */
status = 0;
/* Although we made sure the # of tasks is the same as the # of processes forked off (in the "gtm_multi_proc"
* invocation in "mur_forward"), it is possible one of the forked process finishes one invocation of
* "mur_forward_multi_proc" before even another forked process gets assigned one task in "gtm_multi_proc_helper".
* In this case, we would be invoked more than once. But the first invocation would have done all the needed stuff
* so return for later invocations.
*/
if (mur_forward_multi_proc_done)
return 0;
mur_forward_multi_proc_done = TRUE;
/* Note: "rctl" is unused. But cannot avoid passing it since "gtm_multi_proc" expects something */
prev_rctl = NULL;
rctl_start = NULL;
adjusted_resolve_time = murgbl.adjusted_resolve_time;
assert(0 == murgbl.regcnt_remaining);
multi_proc = multi_proc_in_use; /* cache value in "local" to speed up access inside loops below */
if (multi_proc)
{
mp_hdr = multi_proc_shm_hdr;
shm_rctl_start = mur_shm_hdr->shm_rctl_start;
if (jgbl.onlnrlbk)
{
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
assert(rctl->csa->hold_onto_crit); /* would have been set in parent process */
rctl->csa->hold_onto_crit = FALSE; /* reset since we dont own this region */
assert(rctl->csa->now_crit); /* would have been set in parent process */
rctl->csa->now_crit = FALSE; /* reset since we dont own this region */
}
}
START_HEARTBEAT_IF_NEEDED; /* heartbeat timer needed later (in case not already started by "gtm_multi_proc") */
}
first_shm_rctl = NULL;
/* Phase1 of forward recovery starts */
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
/* Check if "rctl" is available for us or if some other concurrent process has taken it */
if (multi_proc)
{
rctl_index = rctl - &mur_ctl[0];
shm_rctl = &shm_rctl_start[rctl_index];
if (shm_rctl->owning_pid)
{
assert(process_id != shm_rctl->owning_pid);
continue;
}
GRAB_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
assert(release_latch);
for ( ; rctl < rctl_top; rctl++, shm_rctl++)
{
if (shm_rctl->owning_pid)
{
assert(process_id != shm_rctl->owning_pid);
continue;
}
shm_rctl->owning_pid = process_id; /* Declare ownership */
rctl->this_pid_is_owner = TRUE;
if (jgbl.onlnrlbk)
{ /* This is an online rollback and crit was grabbed on all regions by the parent rollback
* process. But this child process now owns this region and does the actual rollback on
* this region so borrow crit for the duration of this child process.
*/
csa = rctl->csa;
csa->hold_onto_crit = TRUE;
csa->now_crit = TRUE;
assert(csa->nl->in_crit == mp_hdr->parent_pid);
csa->nl->in_crit = process_id;
assert(csa->nl->onln_rlbk_pid == mp_hdr->parent_pid);
csa->nl->onln_rlbk_pid = process_id;
}
if (NULL == first_shm_rctl)
first_shm_rctl = shm_rctl;
break;
}
REL_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
if (rctl >= rctl_top)
{
assert(rctl == rctl_top);
break;
}
/* Set key to print this rctl'ss region-name as prefix in case this forked off process prints any output */
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key);
# ifdef MUR_DEBUG
fprintf(stderr, "pid = %d : Owns region %s\n", process_id, multi_proc_key);
# endif
} else
rctl->this_pid_is_owner = TRUE;
if (mur_options.forward)
{
assert(NULL == rctl->jctl_turn_around);
jctl = rctl->jctl = rctl->jctl_head;
assert(jctl->reg_ctl == rctl);
jctl->rec_offset = JNL_HDR_LEN;
jnl_fence_ctl.fence_list = JNL_FENCE_LIST_END; /* initialized to reflect journaling is not enabled */
if (mur_options.rollback)
jgbl.mur_jrec_seqno = jctl->jfh->start_seqno;
} else
{
jctl = rctl->jctl = (NULL == rctl->jctl_turn_around) ? rctl->jctl_head : rctl->jctl_turn_around;
assert(jctl->reg_ctl == rctl);
jctl->rec_offset = jctl->turn_around_offset;
jgbl.mur_jrec_seqno = jctl->turn_around_seqno;
assert((NULL != rctl->jctl_turn_around) || (0 == jctl->rec_offset));
}
if (mur_options.rollback)
{
if (murgbl.consist_jnl_seqno < jgbl.mur_jrec_seqno)
{
/* Assert that murgbl.losttn_seqno is never lesser than jgbl.mur_jrec_seqno (the turnaround
* point seqno) as this is what murgbl.consist_jnl_seqno is going to be set to and will
* eventually be the post-rollback seqno. If this condition is violated, the result of the
* recovery is a compromised database (the file header will indicate a Region Seqno which
* is not necessarily correct since seqnos prior to it might be absent in the database).
* Therefore, this is an out-of-design situation with respect to rollback and so stop it.
*/
assert(murgbl.losttn_seqno >= jgbl.mur_jrec_seqno);
murgbl.consist_jnl_seqno = jgbl.mur_jrec_seqno;
}
assert(murgbl.consist_jnl_seqno <= murgbl.losttn_seqno);
}
if (mur_options.update || mur_options.extr[GOOD_TN])
{
reg = rctl->gd;
gv_cur_region = reg;
tp_change_reg(); /* note : sets cs_addrs to non-NULL value even if gv_cur_region->open is FALSE
* (cs_data could still be NULL). */
rctl->csa = cs_addrs;
cs_addrs->miscptr = (void *)rctl;
rctl->csd = cs_data;
rctl->sgm_info_ptr = cs_addrs->sgm_info_ptr;
assert(!reg->open || (NULL != cs_addrs->dir_tree));
gv_target = cs_addrs->dir_tree;
}
jctl->after_end_of_data = FALSE;
status = mur_next(jctl, jctl->rec_offset);
assert(ERR_JNLREADEOF != status); /* cannot get EOF at start of forward processing */
if (SS_NORMAL != status)
goto finish;
PRINT_VERBOSE_STAT(jctl, "mur_forward:at the start");
rctl->process_losttn = FALSE;
/* Any multi-region TP transaction will be processed as multiple single-region TP transactions up
* until the tp-resolve-time is reached. From then on, they will be treated as one multi-region TP
* transaction. This is needed for proper lost-tn determination (any multi-region transaction that
* gets played in a region AFTER it has already encountered a broken tn should treat this as a lost tn).
*/
do
{
if (multi_proc && IS_FORCED_MULTI_PROC_EXIT(mp_hdr))
{ /* We are at a logical point. So exit if signaled by parent */
status = ERR_FORCEDHALT;
goto finish;
}
assert(jctl == rctl->jctl);
rec = rctl->mur_desc->jnlrec;
rec_time = rec->prefix.time;
if (rec_time > mur_options.before_time)
break; /* Records after -BEFORE_TIME do not go to extract or losttrans or brkntrans files */
if (rec_time < mur_options.after_time)
{
status = mur_next_rec(&jctl);
continue; /* Records before -AFTER_TIME do not go to extract or losttrans or brkntrans files */
}
if (rec_time >= adjusted_resolve_time)
break; /* Records after this adjusted resolve_time will be processed below in phase2 */
/* Note: Since we do hashtable token processing only for records from tp_resolve_time onwards,
* it is possible that if we encounter any broken transactions here we wont know they are broken
* but will play them as is. That is unavoidable. Specify -SINCE_TIME (for -BACKWARD rollback/recover)
* and -VERIFY (for -FORWARD rollback/recover) to control tp_resolve_time (and in turn more
* effective broken tn determination).
*/
status = mur_forward_play_cur_jrec(rctl);
if (SS_NORMAL != status)
break;
status = mur_next_rec(&jctl);
} while (SS_NORMAL == status);
CHECK_IF_EOF_REACHED(rctl, status); /* sets rctl->forw_eof_seen if needed; resets "status" to SS_NORMAL */
if (SS_NORMAL != status)
{ /* ERR_FILENOTCREATE is possible from "mur_cre_file_extfmt" OR ERR_FORCEDHALT is possible
* from "mur_forward_play_cur_jrec". No other errors are known to occur here. Assert accordingly.
*/
assert((ERR_FILENOTCREATE == status) || (ERR_FORCEDHALT == status));
goto finish;
}
if (rctl->forw_eof_seen)
{
PRINT_VERBOSE_STAT(jctl, "mur_forward:Reached EOF before tp_resolve_time");
continue; /* Reached EOF before even getting to tp_resolve_time.
* Do not even consider region for next processing loop */
}
rctl->last_tn = 0;
murgbl.regcnt_remaining++; /* # of regions participating in recovery at this point */
if (NULL == rctl_start)
rctl_start = rctl;
if (NULL != prev_rctl)
{
prev_rctl->next_rctl = rctl;
rctl->prev_rctl = prev_rctl;
}
prev_rctl = rctl;
assert(murgbl.ok_to_update_db || !rctl->db_updated);
PRINT_VERBOSE_STAT(jctl, "mur_forward:at tp_resolve_time");
}
if (multi_proc)
multi_proc_key = NULL; /* reset key until it can be set to rctl's region-name again */
/* Note that it is possible for rctl_start to be NULL at this point. That is there is no journal record in any region
* AFTER the calculated tp-resolve-time. This is possible if for example -AFTER_TIME was used and has a time later
* than any journal record in all journal files. If rctl_start is NULL, prev_rctl should also be NULL and vice versa.
*/
if (NULL != rctl_start)
{
assert(NULL != prev_rctl);
prev_rctl->next_rctl = rctl_start;
rctl_start->prev_rctl = prev_rctl;
}
rctl = rctl_start;
regcnt_stuck = 0; /* # of regions we are stuck in waiting for other regions to resolve a multi-region TP transaction */
assert((NULL == rctl) || (NULL == rctl->forw_multi));
gv_cur_region = NULL; /* clear out any previous value to ensure gv_cur_region/cs_addrs/cs_data
* all get set in sync by the MUR_CHANGE_REG macro below.
*/
/* Phase2 of forward recovery starts */
while (NULL != rctl)
{ /* while there is at least one region remaining with unprocessed journal records */
assert(NULL != rctl_start);
assert(0 < murgbl.regcnt_remaining);
if (NULL != rctl->forw_multi)
{ /* This region's current journal record is part of a TP transaction waiting for other regions */
regcnt_stuck++;
assert(regcnt_stuck <= murgbl.regcnt_remaining);
if (regcnt_stuck == murgbl.regcnt_remaining)
{
assertpro(multi_proc_in_use); /* Else : Out-of-design situation. Stuck in ALL regions. */
/* Check one last time if all regions are stuck waiting for another process to resolve the
* multi-region TP transaction. If so, wait in a sleep loop. If not, we can proceed.
*/
rctl = rctl_start;
start_hrtbt_cntr = heartbeat_counter;
do
{
if (IS_FORCED_MULTI_PROC_EXIT(mp_hdr))
{ /* We are at a logical point. So exit if signaled by parent */
status = ERR_FORCEDHALT;
goto finish;
}
forw_multi = rctl->forw_multi;
assert(NULL != forw_multi);
sfm = forw_multi->shm_forw_multi;
assert(NULL != sfm);
assert(sfm->num_reg_seen_forward <= sfm->num_reg_seen_backward);
# ifdef MUR_DEBUG
fprintf(stderr, "Pid = %d : Line %d : token = %llu : forward = %d : backward = %d\n",
process_id, __LINE__, (long long int)sfm->token,
sfm->num_reg_seen_forward, sfm->num_reg_seen_backward);
# endif
if (sfm->num_reg_seen_forward == sfm->num_reg_seen_backward)
{ /* We are no longer stuck in this region */
assert(!forw_multi->no_longer_stuck);
forw_multi->no_longer_stuck = TRUE;
break;
}
rctl = rctl->next_rctl; /* Move on to the next available region */
assert(NULL != rctl);
if (rctl == rctl_start)
{ /* We went through all regions once and are still stuck.
* Sleep until at leat TWO heartbeats have elapsed after which check for deadlock.
* Do this only in the child process that owns the FIRST region in the region list.
* This way we dont have contention for the GRAB_MULTI_PROC_LATCH from
* all children at more or less the same time.
*/
if ((rctl == mur_ctl) && (heartbeat_counter > (start_hrtbt_cntr + 2)))
{ /* Check if all processes are stuck for a while. If so assertpro */
GRAB_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
assert(release_latch);
shm_rctl_start = mur_shm_hdr->shm_rctl_start;
num_reg_stuck = 0;
for (i = 0; i < murgbl.reg_total; i++)
{
shm_rctl = &shm_rctl_start[i];
sfm = shm_rctl->shm_forw_multi;
if (NULL != sfm)
{
if (sfm->num_reg_seen_forward != sfm->num_reg_seen_backward)
num_reg_stuck++;
}
}
REL_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
/* If everyone is stuck at this point, it is an out-of-design situation */
assertpro(num_reg_stuck < murgbl.reg_total);
start_hrtbt_cntr = heartbeat_counter;
} else
{ /* Sleep and recheck if any region we are stuck in got resolved.
* To minimize time spent sleeping, we just yield our timeslice.
*/
rel_quant();
continue;
}
}
} while (TRUE);
} else
{
rctl = rctl->next_rctl; /* Move on to the next available region */
assert(NULL != rctl);
continue;
}
}
regcnt_stuck = 0; /* restart the counter now that we found at least one non-stuck region */
MUR_CHANGE_REG(rctl);
jctl = rctl->jctl;
this_reg_stuck = FALSE;
for ( status = SS_NORMAL; SS_NORMAL == status; )
{
if (multi_proc && IS_FORCED_MULTI_PROC_EXIT(mp_hdr))
{ /* We are at a logical point. So exit if signaled by parent */
status = ERR_FORCEDHALT;
goto finish;
}
assert(jctl == rctl->jctl);
rec = rctl->mur_desc->jnlrec;
rec_time = rec->prefix.time;
if (rec_time > mur_options.before_time)
break; /* Records after -BEFORE_TIME do not go to extract or losttrans or brkntrans files */
assert((rec_time >= adjusted_resolve_time) || (mur_options.notncheck && !mur_options.verify));
assert((0 == mur_options.after_time) || (mur_options.forward && !rctl->db_updated));
if (rec_time < mur_options.after_time)
{
status = mur_next_rec(&jctl);
continue; /* Records before -AFTER_TIME do not go to extract or losttrans or brkntrans files */
}
/* Check if current journal record can be played right away or need to wait for corresponding journal
* records from other participating TP regions to be reached. A non-TP or ZTP transaction can be played
* without issues (i.e. has no dependencies with any other regions). A single-region TP transaction too
* falls in the same category. A multi-region TP transaction needs to wait until all participating regions
* have played all journal records BEFORE this TP in order to ensure recover plays records in the exact
* same order that GT.M performed them in.
*/
/* If FENCE_NONE is specified, we would not have maintained any multi hashtable in mur_back_process for
* broken transaction processing. So we process multi-region TP transactions as multiple single-region
* TP transactions in forward phase.
*/
if (FENCE_NONE != mur_options.fences)
{
rectype = (enum jnl_record_type)rec->prefix.jrec_type;
if (IS_TP(rectype) && IS_TUPD(rectype))
{
assert(IS_SET_KILL_ZKILL_ZTWORM_LGTRIG_ZTRIG(rectype));
assert(&rec->jrec_set_kill.num_participants == &rec->jrec_ztworm.num_participants);
assert(&rec->jrec_set_kill.num_participants == &rec->jrec_lgtrig.num_participants);
num_partners = rec->jrec_set_kill.num_participants;
assert(0 < num_partners);
if (1 < num_partners)
{
this_reg_stuck = TRUE;
assert(&rec->jrec_set_kill.update_num == &rec->jrec_ztworm.update_num);
assert(&rec->jrec_set_kill.update_num == &rec->jrec_lgtrig.update_num);
}
}
}
if (this_reg_stuck)
{
rec_token_seq = GET_JNL_SEQNO(rec);
MUR_FORW_TOKEN_LOOKUP(forw_multi, rec_token_seq, rec_time);
if (NULL != forw_multi)
{ /* This token has already been seen in another region in forward processing.
* Add current region as well. If all regions have been resolved, then play
* the entire transaction maintaining the exact same order of updates within.
*/
if (!forw_multi->no_longer_stuck)
MUR_FORW_TOKEN_ONE_MORE_REG(forw_multi, rctl);
} else
{ /* First time we are seeing this token in forward processing. Check if this
* has already been determined to be a broken transaction.
*/
recstat = GOOD_TN;
multi = NULL;
if (IS_REC_POSSIBLY_BROKEN(rec_time, rec_token_seq))
{
multi = MUR_TOKEN_LOOKUP(rec_token_seq, rec_time, TPFENCE);
if ((NULL != multi) && (0 < multi->partner))
recstat = BROKEN_TN;
}
MUR_FORW_TOKEN_ADD(forw_multi, rec_token_seq, rec_time, rctl, num_partners,
recstat, multi);
}
/* Check that "tabent" field has been initialized above (by either the MUR_FORW_TOKEN_LOOKUP
* or MUR_FORW_TOKEN_ADD macros). This is relied upon by "mur_forward_play_multireg_tp" below.
*/
assert(NULL != forw_multi->u.tabent);
assert(forw_multi->num_reg_seen_forward <= forw_multi->num_reg_seen_backward);
if (multi_proc)
{
sfm = forw_multi->shm_forw_multi;
ok_to_play = (NULL == sfm) || (sfm->num_reg_seen_forward == sfm->num_reg_seen_backward);
} else
ok_to_play = (forw_multi->num_reg_seen_forward == forw_multi->num_reg_seen_backward);
assert(ok_to_play || !forw_multi->no_longer_stuck);
if (ok_to_play )
{ /* We have enough information to proceed with playing this multi-region TP in
* forward processing (even if we might not have seen all needed regions). Now play it.
* Note that the TP could be BROKEN_TN or GOOD_TN. The callee handles it.
*/
assert(forw_multi == rctl->forw_multi);
status = mur_forward_play_multireg_tp(forw_multi, rctl);
this_reg_stuck = FALSE;
/* Note that as part of playing the TP transaction, we could have reached
* the EOF of rctl. In this case, we need to break out of the loop.
*/
if ((SS_NORMAL != status) || rctl->forw_eof_seen)
break;
assert(NULL == rctl->forw_multi);
assert(!dollar_tlevel);
jctl = rctl->jctl; /* In case the first record after the most recently processed
* TP transaction is in the next generation journal file */
continue;
}
break;
} else
{
status = mur_forward_play_cur_jrec(rctl);
if (SS_NORMAL != status)
break;
}
assert(!this_reg_stuck);
status = mur_next_rec(&jctl);
}
assert((NULL == rctl->forw_multi) || this_reg_stuck);
assert((NULL != rctl->forw_multi) || !this_reg_stuck);
if (!this_reg_stuck)
{ /* We are not stuck in this region (to resolve a multi-region TP).
* This means we are done processing all the records of this region.
*/
assert(NULL == rctl->forw_multi);
if (!rctl->forw_eof_seen)
{
CHECK_IF_EOF_REACHED(rctl, status);
/* sets rctl->forw_eof_seen if needed; resets "status" to SS_NORMAL */
if (SS_NORMAL != status)
{
assert(ERR_FILENOTCREATE == status);
goto finish;
}
assert(!dollar_tlevel);
DELETE_RCTL_FROM_UNPROCESSED_LIST(rctl); /* since all of its records should have been processed */
} else
{ /* EOF was seen in rctl inside "mur_forward_play_multireg_tp" and it was removed
* from the unprocessed list of rctls. At the time rctl was removed, its "next_rctl"
* field could have been pointing to another <rctl> that has since then also been
* removed inside the same function. Therefore the "next_rctl" field is not reliable
* in this case but instead we should rely on the global variable "rctl_start" which
* points to the list of unprocessed rctls. Set "next_rctl" accordingly.
*/
rctl->next_rctl = rctl_start;
if (ERR_JNLREADEOF == status)
status = SS_NORMAL;
}
assert(rctl->deleted_from_unprocessed_list);
}
assert(SS_NORMAL == status);
assert(!this_reg_stuck || !rctl->forw_eof_seen);
assert((NULL == rctl->next_rctl) || (NULL != rctl_start));
assert((NULL == rctl->next_rctl) || (0 < murgbl.regcnt_remaining));
rctl = rctl->next_rctl; /* Note : even though "rctl" could have been deleted from the doubly linked list above,
* rctl->next_rctl is not touched so we can still use it to get to the next element. */
}
assert(0 == murgbl.regcnt_remaining);
jgbl.mur_pini_addr_reset_fnptr = NULL; /* No more simulation of GT.M activity for any region */
prc_vec = murgbl.prc_vec; /* Use process-vector of MUPIP RECOVER (not any simulating GT.M process) now onwards */
assert(0 == dollar_tlevel);
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
if (!rctl->this_pid_is_owner)
{
assert(multi_proc_in_use);
continue; /* in a parallel processing environment, process only regions we own */
}
if (multi_proc)
{ /* Set key to print this rctl's region-name as prefix in case this forked off process prints any output */
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key);
}
PRINT_VERBOSE_STAT(rctl->jctl, "mur_forward:at the end");
assert(!mur_options.rollback || (0 != murgbl.consist_jnl_seqno));
assert(mur_options.rollback || (0 == murgbl.consist_jnl_seqno));
assert(!dollar_tlevel); /* In case it applied a broken TUPD */
assert(murgbl.ok_to_update_db || !rctl->db_updated);
rctl->mur_plst = NULL; /* reset now that simulation of GT.M updates is done */
/* Ensure mur_block_count_correct is called if updates allowed */
if (murgbl.ok_to_update_db && (SS_NORMAL != mur_block_count_correct(rctl)))
{
gtm_putmsg_csa(CSA_ARG(rctl->csa) VARLSTCNT(4) ERR_BLKCNTEDITFAIL, 2, DB_LEN_STR(rctl->gd));
murgbl.wrn_count++;
}
}
finish:
if (multi_proc)
multi_proc_key = NULL; /* reset key until it can be set to rctl's region-name again */
if ((SS_NORMAL == status) && mur_options.show)
mur_output_show();
if (NULL != first_shm_rctl)
{ /* Transfer needed process-private information to shared memory so parent process can later inherit this. */
first_shm_rctl->err_cnt = murgbl.err_cnt;
first_shm_rctl->wrn_count = murgbl.wrn_count;
first_shm_rctl->consist_jnl_seqno = murgbl.consist_jnl_seqno;
/* If extract files were created by this process for one or more regions, then copy that information to
* shared memory so parent process can use this information to do a merge sort.
*/
shm_rctl = mur_shm_hdr->shm_rctl_start;
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++, shm_rctl++)
{
assert(multi_proc_in_use);
if (!rctl->this_pid_is_owner)
continue; /* in a parallel processing environment, process only regions we own */
/* Cancel any flush/dbsync timers by this child process for this region. This is because the
* child is not going to go through exit handling code (no gds_rundown etc.). And we need to
* clear up csa->nl->wcs_timers. (normally done by gds_rundown).
*/
if (NULL != rctl->csa) /* rctl->csa can be NULL in case of "mupip journal -extract" etc. */
CANCEL_DB_TIMERS(rctl->gd, rctl->csa, cancelled_timer, cancelled_dbsync_timer);
reccnt = 0;
for (size_ptr = &rctl->jnlext_multi_list_size[0], recstat = 0;
recstat < TOT_EXTR_TYPES;
recstat++, size_ptr++)
{ /* Assert "extr_file_created" information is in sync between rctl and shm_rctl.
* This was done at the end of "mur_cre_file_extfmt".
*/
assert(shm_rctl->extr_file_created[recstat] == rctl->extr_file_created[recstat]);
/* Assert that if *size_ptr is non-zero, then we better have created an extract file.
* Note that the converse is not true. It is possible we created a file for example to
* write an INCTN record but decided to not write anything because it was not a -detail
* type of extract. So *sizeptr could be 0 even though we created the extract file.
*/
assert(!*size_ptr || rctl->extr_file_created[recstat]);
shm_rctl->jnlext_list_size[recstat] = *size_ptr;
reccnt += *size_ptr;
}
assert(INVALID_SHMID == shm_rctl->jnlext_shmid);
shm_size = reccnt * SIZEOF(jnlext_multi_t);
/* If we are quitting because of an abnormal status OR a forced signal to terminate
* OR if the parent is dead (kill -9) dont bother creating shmid to communicate back with parent.
*/
if (mp_hdr->parent_pid != getppid())
{
SET_FORCED_MULTI_PROC_EXIT; /* Also signal sibling children to stop processing */
if (SS_NORMAL != status)
status = ERR_FORCEDHALT;
}
if ((SS_NORMAL == status) && shm_size)
{
shmid = shmget(IPC_PRIVATE, shm_size, 0600 | IPC_CREAT);
if (-1 == shmid)
{
save_errno = errno;
SNPRINTF(errstr, SIZEOF(errstr),
"shmget() : shmsize=0x%llx", shm_size);
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key); /* to print region name prefix */
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8)
ERR_SYSCALL, 5, LEN_AND_STR(errstr), CALLFROM, save_errno);
}
shmPtr = (char *)do_shmat(shmid, 0, 0);
if (-1 == (sm_long_t)shmPtr)
{
save_errno = errno;
SNPRINTF(errstr, SIZEOF(errstr),
"shmat() : shmid=%d shmsize=0x%llx", shmid, shm_size);
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key); /* to print region name prefix */
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8)
ERR_SYSCALL, 5, LEN_AND_STR(errstr), CALLFROM, save_errno);
}
shm_rctl->jnlext_shmid = shmid;
shm_rctl->jnlext_shm_size = shm_size;
for (size_ptr = &rctl->jnlext_multi_list_size[0], recstat = 0;
recstat < TOT_EXTR_TYPES;
recstat++, size_ptr++)
{
shm_size = *size_ptr;
if (shm_size)
{
copy_size = copy_list_to_buf(rctl->jnlext_multi_list[recstat],
(int4)shm_size, shmPtr);
assert(copy_size == (shm_size * SIZEOF(jnlext_multi_t)));
shmPtr += copy_size;
}
}
}
}
}
mur_close_file_extfmt(IN_MUR_CLOSE_FILES_FALSE); /* Need to flush buffered extract/losttrans/brokentrans files */
return (int)status;
}
