/* Waits for a concurrently running write (of a global buffer to disk) to complete.
*
* Returns TRUE if write completes within timeout of approx. 1 minute.
* Returns FALSE otherwise.
*/
boolean_t wcs_write_in_progress_wait(node_local_ptr_t cnl, cache_rec_ptr_t cr, wbtest_code_t wbox_test_code)
{
uint4 lcnt;
int4 n;
for (lcnt = 1; ; lcnt++)
{ /* the design here is that either this process owns the block, or the writer does.
* if the writer does, it must be allowed to finish its write; then it will release the block
* and the next LOCK will establish ownership
*/
LOCK_BUFF_FOR_UPDATE(cr, n, &cnl->db_latch);
/* This destroys evidence of writer ownership, but this is really a test that
* there was no prior owner. It will only be true if the writer has cleared it.
*/
if (OWN_BUFF(n))
break;
else
{
GTM_WHITE_BOX_TEST(wbox_test_code, lcnt, (2 * BUF_OWNER_STUCK));
/* We have noticed the below assert to fail occasionally on some platforms
* We suspect it is because of waiting for another writer that is in jnl_fsync
* (as part of flushing a global buffer) which takes more than a minute to finish.
* To avoid false failures (where the other writer finishes its job in a little over
* a minute) we wait for twice the time in the debug version.
*/
DEBUG_ONLY(
if ((BUF_OWNER_STUCK == lcnt) && cr->epid)
GET_C_STACK_FROM_SCRIPT("WRITEWAITPID", process_id, cr->epid, ONCE);
)
if (BUF_OWNER_STUCK DEBUG_ONLY( * 2) < lcnt)
{ /* sick of waiting */
if (0 == cr->dirty)
{ /* someone dropped something; assume it was the writer and go on */
LOCK_NEW_BUFF_FOR_UPDATE(cr);
break;
} else
{
if (cr->epid)
{
#ifdef DEBUG
GET_C_STACK_FROM_SCRIPT("WRITEWAITPID", process_id, cr->epid, TWICE);
send_msg(VARLSTCNT(8) ERR_WRITEWAITPID, 6, process_id, TWICE, \
cr->epid, cr->blk, DB_LEN_STR(gv_cur_region));
#else
GET_C_STACK_FROM_SCRIPT("WRITEWAITPID", process_id, cr->epid, ONCE);
send_msg(VARLSTCNT(8) ERR_WRITEWAITPID, 6, process_id, ONCE, \
cr->epid, cr->blk, DB_LEN_STR(gv_cur_region));
#endif
}
return FALSE;
}
}
if (WRITER_STILL_OWNS_BUFF(cr, n))
wcs_sleep(lcnt);
}
} /* end of for loop to control buffer */
void 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;
error_def(ERR_REPLJNLCLOSED);
error_def(ERR_JNLCLOSED);
switch(jpc->region->dyn.addr->acc_meth)
{
case dba_mm:
case dba_bg:
csa = &FILE_INFO(jpc->region)->s_addrs;
break;
default:
GTMASSERT;
}
#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);
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(VARLSTCNT(8) ERR_REPLJNLCLOSED, 6, DB_LEN_STR(jpc->region), ®_seqno, ®_seqno, &jnlseqno, &jnlseqno);
} else
send_msg(VARLSTCNT(5) ERR_JNLCLOSED, 3, DB_LEN_STR(jpc->region), &csa->ti->curr_tn);
#ifdef VMS
assert(0 != csa->jnl->jnllsb->lockid);
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 (SS$_NORMAL == status)
status = gtm_deq(csa->jnl->jnllsb->lockid, NULL, PSL$C_USER, 0);
if (SS$_NORMAL != status)
GTMASSERT;
# else
jnl_file_close(jpc->region, FALSE, FALSE);
#endif
}
/* input parameter "command_name" is a string that is either "MUPIP REORG UPGRADE/DOWNGRADE" or "MUPIP SET VERSION" */
int4 desired_db_format_set(gd_region *reg, enum db_ver new_db_format, char *command_name)
{
boolean_t was_crit;
char *db_fmt_str;
char *wcblocked_ptr;
int4 status;
uint4 jnl_status;
inctn_opcode_t save_inctn_opcode;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
trans_num curr_tn;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
assert(reg->open);
csa = &FILE_INFO(reg)->s_addrs;
csd = csa->hdr;
GTMCRYPT_ONLY(
/* We don't allow databases to be encrypted if the version is V4 */
if (csd->is_encrypted && (GDSV4 == new_db_format))
{
gtm_putmsg(VARLSTCNT(4) ERR_CRYPTNOV4, 2, DB_LEN_STR(reg));
return ERR_CRYPTNOV4;
}
)
boolean_t mur_report_error(jnl_ctl_list *jctl, enum mur_error code)
{
error_def(ERR_BOVTMGTEOVTM);
error_def(ERR_DUPTOKEN);
error_def(ERR_JNLBADRECFMT);
error_def(ERR_PREVJNLNOEOF);
error_def(ERR_UNKNOWNRECTYPE);
switch (code)
{
default:
assert(FALSE);
break;
case MUR_DUPTOKEN:
assert(FALSE);
gtm_putmsg(VARLSTCNT(7) ERR_DUPTOKEN, 5, &((struct_jrec_tcom *)jctl->reg_ctl->mur_desc->jnlrec)->token_seq.token,
jctl->jnl_fn_len, jctl->jnl_fn, DB_LEN_STR(jctl->reg_ctl->gd));
break;
case MUR_PREVJNLNOEOF:
gtm_putmsg(VARLSTCNT(4) ERR_PREVJNLNOEOF, 2, jctl->jnl_fn_len, jctl->jnl_fn);
break;
case MUR_JNLBADRECFMT:
gtm_putmsg(VARLSTCNT(5) ERR_JNLBADRECFMT, 3, jctl->jnl_fn_len, jctl->jnl_fn, jctl->rec_offset);
break;
case MUR_BOVTMGTEOVTM:
gtm_putmsg(VARLSTCNT(6) ERR_BOVTMGTEOVTM, 4, jctl->jnl_fn_len, jctl->jnl_fn,
&jctl->jfh->bov_timestamp, &jctl->jfh->eov_timestamp);
break;
}
return MUR_WITHIN_ERROR_LIMIT(murgbl.err_cnt, mur_options.error_limit); /* side-effect : increments murgbl.err_cnt */
}
boolean_t mur_report_error(enum mur_error code)
{
error_def(ERR_UNKNOWNRECTYPE);
error_def(ERR_DUPTOKEN);
error_def(ERR_PREVJNLNOEOF);
error_def(ERR_JNLBADRECFMT);
switch (code)
{
default:
assert(FALSE);
break;
case MUR_DUPTOKEN:
assert(FALSE);
gtm_putmsg(VARLSTCNT(7) ERR_DUPTOKEN, 5, &((struct_jrec_tcom *)mur_rab.jnlrec)->token_seq.token,
mur_jctl->jnl_fn_len, mur_jctl->jnl_fn, DB_LEN_STR(mur_ctl[mur_regno].gd));
break;
case MUR_PREVJNLNOEOF:
gtm_putmsg(VARLSTCNT(4) ERR_PREVJNLNOEOF, 2, mur_jctl->jnl_fn_len, mur_jctl->jnl_fn);
break;
case MUR_JNLBADRECFMT:
gtm_putmsg(VARLSTCNT(5) ERR_JNLBADRECFMT, 3, mur_jctl->jnl_fn_len, mur_jctl->jnl_fn, mur_jctl->rec_offset);
break;
case MUR_BOVTMGTEOVTM:
break;
}
return MUR_WITHIN_ERROR_LIMIT(murgbl.err_cnt, mur_options.error_limit); /* side-effect : increments murgbl.err_cnt */
}
int main(int argc, char *argv[])
{
DCL_THREADGBL_ACCESS;
GTM_THREADGBL_INIT;
set_blocksig();
gtm_imagetype_init(DSE_IMAGE);
gtm_wcswidth_fnptr = gtm_wcswidth;
gtm_env_init(); /* read in all environment variables */
licensed = TRUE;
TREF(transform) = TRUE;
op_open_ptr = op_open;
patch_curr_blk = get_dir_root();
err_init(util_base_ch);
GTM_ICU_INIT_IF_NEEDED; /* Note: should be invoked after err_init (since it may error out) and before CLI parsing */
sig_init(generic_signal_handler, dse_ctrlc_handler, suspsigs_handler);
atexit(util_exit_handler);
SET_LATCH_GLOBAL(&defer_latch, LOCK_AVAILABLE);
get_page_size();
stp_init(STP_INITSIZE);
rts_stringpool = stringpool;
getjobname();
INVOKE_INIT_SECSHR_ADDRS;
getzdir();
prealloc_gt_timers();
initialize_pattern_table();
gvinit();
region_init(FALSE);
INIT_GBL_ROOT(); /* Needed for GVT initialization */
getjobnum();
util_out_print("!/File !_!AD", TRUE, DB_LEN_STR(gv_cur_region));
util_out_print("Region!_!AD!/", TRUE, REG_LEN_STR(gv_cur_region));
cli_lex_setup(argc, argv);
CREATE_DUMMY_GBLDIR(gd_header, original_header, gv_cur_region, gd_map, gd_map_top);
gtm_chk_dist(argv[0]);
# ifdef DEBUG
if ((gtm_white_box_test_case_enabled && (WBTEST_SEMTOOLONG_STACK_TRACE == gtm_white_box_test_case_number) ))
{
sgmnt_addrs * csa;
node_local_ptr_t cnl;
csa = &FILE_INFO(gv_cur_region)->s_addrs;
cnl = csa->nl;
cnl->wbox_test_seq_num = 1; /*Signal the first step and wait here*/
while (2 != cnl->wbox_test_seq_num) /*Wait for another process to get hold of the semaphore and signal next step*/
LONG_SLEEP(10);
}
# endif
if (argc < 2)
display_prompt();
io_init(TRUE);
while (1)
{
if (!dse_process(argc))
break;
display_prompt();
}
dse_exit();
REVERT;
}
uint4 mupip_set_journal_newstate(set_jnl_options *jnl_options, jnl_create_info *jnl_info, mu_set_rlist *rptr)
{
enum jnl_state_codes jnl_curr_state;
enum repl_state_codes repl_curr_state;
boolean_t current_image;
enum db_acc_method acc_meth;
error_def(ERR_REPLNOBEFORE);
error_def(ERR_REPLJNLCNFLCT);
error_def(ERR_JNLDISABLE);
error_def(ERR_MMBEFOREJNL);
error_def(ERR_MMNOBFORRPL);
jnl_curr_state = (enum jnl_state_codes)rptr->sd->jnl_state;
repl_curr_state = (enum repl_state_codes)rptr->sd->repl_state;
acc_meth = rptr->sd->acc_meth;
current_image = rptr->sd->jnl_before_image;
if (CLI_ABSENT == jnl_options->cli_journal)
rptr->jnl_new_state = jnl_curr_state;
else if ((CLI_NEGATED == jnl_options->cli_journal) || (CLI_NEGATED == jnl_options->cli_enable))
rptr->jnl_new_state = jnl_notallowed; /* DISABLE specified */
else if ((jnl_notallowed != jnl_curr_state) || (CLI_PRESENT == jnl_options->cli_enable))
{ /* journaling is already ENABLED or ENABLE is explicitly specified */
if (CLI_NEGATED == jnl_options->cli_on) /* OFF specified */
rptr->jnl_new_state = jnl_closed;
else if (repl_curr_state == repl_was_open && CLI_PRESENT != jnl_options->cli_replic_on)
{ /* Journaling was turned OFF by jnl_file_lost(). Do not allow turning journaling ON without also
turning replication ON */
gtm_putmsg(VARLSTCNT(10) ERR_REPLJNLCNFLCT, 8, LEN_AND_STR(jnl_state_lit[jnl_open]),
DB_LEN_STR(gv_cur_region), LEN_AND_STR(repl_state_lit[repl_closed]),
LEN_AND_STR(jnl_state_lit[jnl_open]));
return EXIT_WRN;
}
else /* ON explicitly specified or present by default */
rptr->jnl_new_state = jnl_open;
} else /* jnl_notallowed == jnl_curr_state && CLI_ABSENT == jnl_options->cli_enable */
{
if (CLI_PRESENT != jnl_options->cli_replic_on)
{
gtm_putmsg(VARLSTCNT(4) ERR_JNLDISABLE, 2, DB_LEN_STR(gv_cur_region));
return EXIT_WRN;
} else
rptr->jnl_new_state = jnl_open; /* turn journaling on for REPLICATION=ON */
}
VMS_ONLY(rptr->before_images = (jnl_options->image_type_specified ? jnl_info->before_images : current_image);)
void set_enospc_flags(gd_addr *addr_ptr, char enospc_enable_list[], boolean_t ok_to_interrupt)
{
gd_region *r_local, *r_top;
int i;
sgmnt_addrs *csa;
const char *syslog_msg;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
for (r_local = addr_ptr->regions, r_top = r_local + addr_ptr->n_regions, i = 0;
r_local < r_top; r_local++, i++)
{
if (!r_local->open || r_local->was_open)
continue;
if ((dba_bg != r_local->dyn.addr->acc_meth) && (dba_mm != r_local->dyn.addr->acc_meth))
continue;
csa = &FILE_INFO(r_local)->s_addrs;
if (ANTICIPATORY_FREEZE_ENABLED(csa))
{
switch(enospc_enable_list[i])
{
case NONE:
syslog_msg = "Turning off fake ENOSPC for both database and journal file.";
csa->nl->fake_db_enospc = FALSE;
csa->nl->fake_jnl_enospc = FALSE;
break;
case DB_ON:
syslog_msg = "Turning on fake ENOSPC only for database file.";
csa->nl->fake_db_enospc = TRUE;
csa->nl->fake_jnl_enospc = FALSE;
break;
case JNL_ON:
syslog_msg = "Turning on fake ENOSPC only for journal file.";
csa->nl->fake_db_enospc = FALSE;
csa->nl->fake_jnl_enospc = TRUE;
break;
case DB_AND_JNL_ON:
syslog_msg = "Turning on fake ENOSPC for both database and journal file.";
csa->nl->fake_db_enospc = TRUE;
csa->nl->fake_jnl_enospc = TRUE;
break;
default:
assert(FALSE);
}
if (ok_to_interrupt)
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(8) ERR_TEXT, 2, DB_LEN_STR(r_local), ERR_TEXT, 2,
LEN_AND_STR(syslog_msg));
}
}
}
STATICFNDEF void mu_rndwn_all_helper(shm_parms *parm_buff, char *fname, int *exit_status, int *tmp_exit_status)
{
replpool_identifier replpool_id;
boolean_t ret_status, jnlpool_sem_created;
unsigned char ipcs_buff[MAX_IPCS_ID_BUF], *ipcs_ptr;
ESTABLISH(mu_rndwn_all_helper_ch);
if (validate_db_shm_entry(parm_buff, fname, tmp_exit_status))
{
if (SS_NORMAL == *tmp_exit_status)
{ /* shm still exists */
mu_gv_cur_reg_init();
gv_cur_region->dyn.addr->fname_len = strlen(fname);
STRNCPY_STR(gv_cur_region->dyn.addr->fname, fname, gv_cur_region->dyn.addr->fname_len);
if (mu_rndwn_file(gv_cur_region, FALSE))
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_MUFILRNDWNSUC, 2, DB_LEN_STR(gv_cur_region));
else
{ /* Save semid so that it will not be removed by mu_rndwn_sem_all() */
add_to_semids_list(FILE_INFO(gv_cur_region)->semid);
*exit_status = ERR_MUNOTALLSEC;
}
mu_gv_cur_reg_free();
} else
{ /* shm has been cleaned up by "validate_db_shm_entry" so no need of any more cleanup here */
assert(ERR_SHMREMOVED == *tmp_exit_status);
*tmp_exit_status = SS_NORMAL; /* reset tmp_exit_status for below logic to treat this as normal */
}
} else if ((SS_NORMAL == *tmp_exit_status)
&& validate_replpool_shm_entry(parm_buff, (replpool_id_ptr_t)&replpool_id, tmp_exit_status))
{
if (SS_NORMAL == *tmp_exit_status)
{
assert(JNLPOOL_SEGMENT == replpool_id.pool_type || RECVPOOL_SEGMENT == replpool_id.pool_type);
ret_status = mu_rndwn_repl_instance(&replpool_id, TRUE, FALSE, &jnlpool_sem_created);
ipcs_ptr = i2asc((uchar_ptr_t)ipcs_buff, parm_buff->shmid);
*ipcs_ptr = '\0';
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(6) (JNLPOOL_SEGMENT == replpool_id.pool_type) ?
(ret_status ? ERR_MUJPOOLRNDWNSUC : ERR_MUJPOOLRNDWNFL) :
(ret_status ? ERR_MURPOOLRNDWNSUC : ERR_MURPOOLRNDWNFL),
4, LEN_AND_STR(ipcs_buff), LEN_AND_STR(replpool_id.instfilename));
if (!ret_status)
*exit_status = ERR_MUNOTALLSEC;
} else
{ /* shm has been cleaned up by "validate_replpool_shm_entry" so no need of any more cleanup here */
assert(ERR_SHMREMOVED == *tmp_exit_status);
*tmp_exit_status = SS_NORMAL; /* reset tmp_exit_status for below logic to treat this as normal */
}
}
REVERT;
}
/* This routine returns whether the free_blocks counter in the file-header is ok (TRUE) or not (FALSE).
* If not, it corrects it. This assumes cs_addrs, cs_data and gv_cur_region to point to the region of interest.
* It also assumes that the master-map is correct and finds out non-full local bitmaps and counts the number of
* free blocks in each of them and sums them up to determine the perceived correct free_blocks count.
* The reason why this is ok is that even if the master-map incorrectly reports a local bitmap as full, our new free_blocks
* count will effectively make the free space in that local-bitmap invisible and make a gdsfilext necessary and valid.
* A later mupip integ will scavenge that invisible space for us. The worst that can therefore happen is that we will transiently
* not be using up existing space. But we will always ensure that the free_blocks counter goes in sync with the master-map.
*/
boolean_t is_free_blks_ctr_ok(void)
{
boolean_t blk_used;
block_id bml, free_bit, free_bml, maxbitsthismap;
cache_rec_ptr_t cr;
int cycle;
sm_uc_ptr_t bmp;
unsigned int local_maps, total_blks, free_blocks;
error_def(ERR_DBBADFREEBLKCTR);
assert(&FILE_INFO(gv_cur_region)->s_addrs == cs_addrs && cs_addrs->hdr == cs_data && cs_addrs->now_crit);
total_blks = (dba_mm == cs_data->acc_meth) ? cs_addrs->total_blks : cs_addrs->ti->total_blks;
local_maps = DIVIDE_ROUND_UP(total_blks, BLKS_PER_LMAP);
for (free_blocks = 0, free_bml = 0; free_bml < local_maps; free_bml++)
{
bml = bmm_find_free((uint4)free_bml, (sm_uc_ptr_t)MM_ADDR(cs_data), local_maps);
if (bml < free_bml)
break;
free_bml = bml;
bml *= BLKS_PER_LMAP;
if (!(bmp = t_qread(bml, (sm_int_ptr_t)&cycle, &cr))
|| (BM_SIZE(BLKS_PER_LMAP) != ((blk_hdr_ptr_t)bmp)->bsiz)
|| (LCL_MAP_LEVL != ((blk_hdr_ptr_t)bmp)->levl))
{
assert(FALSE); /* In pro, we will simply skip counting this local bitmap. */
continue;
}
assert(free_bml <= (local_maps - 1));
maxbitsthismap = (free_bml != (local_maps - 1)) ? BLKS_PER_LMAP : total_blks - bml;
for (free_bit = 0; free_bit < maxbitsthismap; free_bit++)
{
free_bit = bm_find_blk(free_bit, (sm_uc_ptr_t)bmp + sizeof(blk_hdr), maxbitsthismap, &blk_used);
assert(NO_FREE_SPACE <= free_bit);
if (0 > free_bit)
break;
free_blocks++;
}
}
assert(cs_addrs->ti->free_blocks == free_blocks);
if (cs_addrs->ti->free_blocks != free_blocks)
{
send_msg(VARLSTCNT(6) ERR_DBBADFREEBLKCTR, 4, DB_LEN_STR(gv_cur_region), cs_addrs->ti->free_blocks, free_blocks);
cs_addrs->ti->free_blocks = free_blocks;
return FALSE;
}
return TRUE;
}
bool gtcmtr_kill(void)
{
cm_region_list *reg_ref;
unsigned char *ptr, regnum;
unsigned short len;
static readonly gds_file_id file;
error_def(ERR_DBPRIVERR);
ptr = curr_entry->clb_ptr->mbf;
assert(*ptr == CMMS_Q_KILL);
ptr++;
GET_SHORT(len, ptr);
ptr += sizeof(unsigned short);
regnum = *ptr++;
reg_ref = gtcm_find_region(curr_entry,regnum);
len--; /* subtract size of regnum */
CM_GET_GVCURRKEY(ptr, len);
gtcm_bind_name(reg_ref->reghead, TRUE);
if (gv_cur_region->read_only)
rts_error(VARLSTCNT(4) ERR_DBPRIVERR, 2, DB_LEN_STR(gv_cur_region));
if (JNL_ALLOWED(cs_addrs))
{ /* we need to copy client's specific prc_vec into the global variable in order that the gvcst* routines
* do the right job. actually we need to do this only if JNL_ENABLED(cs_addrs), but since it is not
* easy to re-execute the following two assignments in case gvcst_kill()'s call to t_end() encounters a
* cdb_sc_jnlstatemod retry code, we choose the easier approach of executing the following segment
* if JNL_ALLOWED(cs_addrs) is TRUE instead of checking for JNL_ENABLED(cs_addrs) to be TRUE.
* this approach has the overhead that we will be doing the following assignments even though JNL_ENABLED
* might not be TRUE but since the following two are just pointer copies, it is not considered a big overhead.
* this approach ensures that the jnl_put_jrt_pini() gets the appropriate prc_vec for writing into the
* journal record in case JNL_ENABLED turns out to be TRUE in t_end() time.
* note that the value of JNL_ALLOWED(cs_addrs) cannot be changed on the fly without obtaining standalone access
* and hence the correctness of prc_vec whenever it turns out necessary is guaranteed.
*/
originator_prc_vec = curr_entry->pvec;
cs_addrs->jnl->pini_addr = reg_ref->pini_addr;
}
if (gv_target->root)
gvcst_kill(TRUE);
if (JNL_ALLOWED(cs_addrs))
reg_ref->pini_addr = cs_addrs->jnl->pini_addr; /* In case journal switch occurred */
ptr = curr_entry->clb_ptr->mbf;
*ptr++ = CMMS_R_KILL;
curr_entry->clb_ptr->cbl = S_HDRSIZE;
return TRUE;
}
uint4 set_jnl_file_close(set_jnl_file_close_opcode_t set_jnl_file_close_opcode)
{
uint4 jnl_status = 0;
cs_addrs = &FILE_INFO(gv_cur_region)->s_addrs;
jnl_status = jnl_ensure_open();
if (0 == jnl_status)
{
if (0 == cs_addrs->jnl->pini_addr)
jnl_put_jrt_pini(cs_addrs);
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH);
jnl_put_jrt_pfin(cs_addrs);
jnl_file_close(gv_cur_region, TRUE, TRUE);
} else
gtm_putmsg(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(cs_addrs->hdr),
DB_LEN_STR(gv_cur_region));
return jnl_status;
}
void op_gvkill(void)
{ gd_region *reg;
error_def(ERR_DBPRIVERR);
if (gv_cur_region->read_only)
rts_error(VARLSTCNT(4) ERR_DBPRIVERR, 2, DB_LEN_STR(gv_cur_region));
if (gv_curr_subsc_null && gv_cur_region->null_subs == FALSE)
sgnl_gvnulsubsc();
if (gv_cur_region->dyn.addr->acc_meth == dba_bg || gv_cur_region->dyn.addr->acc_meth == dba_mm)
{
if (gv_target->root)
{ gvcst_kill(TRUE);
}
}
else if (gv_cur_region->dyn.addr->acc_meth == dba_cm)
{ gvcmx_kill(TRUE);
}else
{ gvusr_kill(TRUE);
}
if (gv_cur_region->dyn.addr->repl_list)
{
gv_replication_error = gv_replopen_error;
gv_replopen_error = FALSE;
reg = gv_cur_region;
while (gv_cur_region = gv_cur_region->dyn.addr->repl_list) /* set replicated segments */
{
if (gv_cur_region->open)
{
change_reg();
kill_var();
}
else
gv_replication_error = TRUE;
}
gv_cur_region = reg;
change_reg();
if (gv_replication_error)
sgnl_gvreplerr();
}
}
uint4 jnl_file_open_switch(gd_region *reg, uint4 sts)
{
sgmnt_addrs *csa;
jnl_private_control *jpc;
jnl_create_info create;
char prev_jnl_fn[JNL_NAME_SIZE];
csa = &FILE_INFO(reg)->s_addrs;
jpc = csa->jnl;
assert((ERR_JNLFILOPN != sts) && (NOJNL != jpc->channel) || (ERR_JNLFILOPN == sts) && (NOJNL == jpc->channel));
if ((ERR_JNLFILOPN != sts) && (NOJNL != jpc->channel))
F_CLOSE(jpc->channel);
jpc->channel = NOJNL;
jnl_send_oper(jpc, sts);
/* attempt to create a new journal file */
memset(&create, 0, sizeof(create));
create.status = create.status2 = SS_NORMAL;
create.prev_jnl = &prev_jnl_fn[0];
set_jnl_info(reg, &create);
create.no_prev_link = TRUE;
create.no_rename = FALSE;
if (!jgbl.forw_phase_recovery)
JNL_SHORT_TIME(jgbl.gbl_jrec_time); /* needed for cre_jnl_file() */
/* else mur_output_record() would have already set jgbl.gbl_jrec_time */
assert(jgbl.gbl_jrec_time);
if (EXIT_NRM != cre_jnl_file(&create))
{
jpc->status = create.status;
jpc->status2 = create.status2;
return ERR_JNLINVALID;
} else
{
jpc->status = SS_NORMAL;
sts = 0;
}
send_msg(VARLSTCNT(6) ERR_PREVJNLLINKCUT, 4, JNL_LEN_STR(csa->hdr), DB_LEN_STR(reg));
assert(csa->hdr->jnl_file_len == create.jnl_len);
assert(0 == memcmp(csa->hdr->jnl_file_name, create.jnl, create.jnl_len));
return sts;
}
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 op_gvkill(void)
{
gd_region *reg;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
if (gv_cur_region->read_only)
rts_error(VARLSTCNT(4) ERR_DBPRIVERR, 2, DB_LEN_STR(gv_cur_region));
if (TREF(gv_last_subsc_null) && NEVER == gv_cur_region->null_subs)
sgnl_gvnulsubsc();
if (gv_cur_region->dyn.addr->acc_meth == dba_bg || gv_cur_region->dyn.addr->acc_meth == dba_mm)
{
if (IS_OK_TO_INVOKE_GVCST_KILL(gv_target))
gvcst_kill(TRUE);
} else if (gv_cur_region->dyn.addr->acc_meth == dba_cm)
gvcmx_kill(TRUE);
else
gvusr_kill(TRUE);
if (gv_cur_region->dyn.addr->repl_list)
{
gv_replication_error = gv_replopen_error;
gv_replopen_error = FALSE;
reg = gv_cur_region;
while (gv_cur_region = gv_cur_region->dyn.addr->repl_list) /* set replicated segments */
{
if (gv_cur_region->open)
{
change_reg();
kill_var();
} else
gv_replication_error = TRUE;
}
gv_cur_region = reg;
change_reg();
if (gv_replication_error)
sgnl_gvreplerr();
}
}
block_id dse_getblk(char *element, boolean_t nobml, boolean_t carry_curr)
{
block_id blk;
if (!cli_get_hex(element, (uint4 *)&blk))
blk = patch_curr_blk;
else
CLEAR_DSE_COMPRESS_KEY;
if ((blk < 0) || (blk >= cs_addrs->ti->total_blks))
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_BLKINVALID, 4, blk, DB_LEN_STR(gv_cur_region),
cs_addrs->ti->total_blks);
return BADDSEBLK;
}
if (nobml && IS_BITMAP_BLK(blk))
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_CANTBITMAP);
return BADDSEBLK;
}
if (carry_curr)
patch_curr_blk = blk;
return blk;
}
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 lke_show(void)
{
bool locks, all = TRUE, wait = TRUE, interactive = FALSE, match = FALSE, memory = TRUE, nocrit = TRUE;
boolean_t exact = FALSE, was_crit;
int4 pid;
size_t ls_len;
int n;
char regbuf[MAX_RN_LEN], nodebuf[32], one_lockbuf[MAX_KEY_SZ];
mlk_ctldata_ptr_t ctl;
mstr reg, node, one_lock;
int shr_sub_len = 0;
float ls_free = 0; /* Free space in bottleneck subspace */
/* Get all command parameters */
reg.addr = regbuf;
reg.len = SIZEOF(regbuf);
node.addr = nodebuf;
node.len = SIZEOF(nodebuf);
one_lock.addr = one_lockbuf;
one_lock.len = SIZEOF(one_lockbuf);
if (lke_getcli(&all, &wait, &interactive, &pid, ®, &node, &one_lock, &memory, &nocrit, &exact) == 0)
return;
/* Search all regions specified on the command line */
for (gv_cur_region = gd_header->regions, n = 0; n != gd_header->n_regions; ++gv_cur_region, ++n)
{
/* If region matches and is open */
if ((reg.len == 0 ||
gv_cur_region->rname_len == reg.len && memcmp(gv_cur_region->rname, reg.addr, reg.len) == 0) &&
gv_cur_region->open)
{
match = TRUE;
util_out_print("!/!AD!/", NOFLUSH, REG_LEN_STR(gv_cur_region));
/* If distributed database, the region is located on another node */
if (gv_cur_region->dyn.addr->acc_meth == dba_cm)
{
# if defined(LKE_WORKS_OK_WITH_CM)
/* Obtain lock info from the remote node */
locks = gtcmtr_lke_showreq(gv_cur_region->dyn.addr->cm_blk, gv_cur_region->cmx_regnum,
all, wait, pid, &node);
# else
gtm_putmsg(VARLSTCNT(10) ERR_UNIMPLOP, 0, ERR_TEXT, 2,
LEN_AND_LIT("GT.CM region - locks must be displayed on the local node"),
ERR_TEXT, 2, REG_LEN_STR(gv_cur_region));
continue;
# endif
} else if (gv_cur_region->dyn.addr->acc_meth == dba_bg || gv_cur_region->dyn.addr->acc_meth == dba_mm)
{ /* Local region */
cs_addrs = &FILE_INFO(gv_cur_region)->s_addrs;
ls_len = (size_t)(cs_addrs->lock_addrs[1] - cs_addrs->lock_addrs[0]);
ctl = (mlk_ctldata_ptr_t)malloc(ls_len);
/* Prevent any modification of the lock space while we make a local copy of it */
if (cs_addrs->critical != NULL)
crash_count = cs_addrs->critical->crashcnt;
was_crit = cs_addrs->now_crit;
if (!nocrit && !was_crit)
grab_crit(gv_cur_region);
longcpy((uchar_ptr_t)ctl, (uchar_ptr_t)cs_addrs->lock_addrs[0], ls_len);
assert((ctl->max_blkcnt > 0) && (ctl->max_prccnt > 0) && ((ctl->subtop - ctl->subbase) > 0));
if (!nocrit && !was_crit)
rel_crit(gv_cur_region);
shr_sub_len = 0;
locks = ctl->blkroot == 0 ?
FALSE:
lke_showtree(NULL, (mlk_shrblk_ptr_t)R2A(ctl->blkroot), all, wait, pid,
one_lock, memory, &shr_sub_len);
/* lock space usage consists of: control_block + nodes(locks) + processes + substrings */
/* any of those subspaces can be bottleneck.
* Therefore we will report the subspace which is running out.
*/
ls_free = MIN(((float)ctl->blkcnt) / ctl->max_blkcnt, ((float)ctl->prccnt) / ctl->max_prccnt);
ls_free = MIN(1-(((float)shr_sub_len) / (ctl->subtop - ctl->subbase)), ls_free);
ls_free *= 100; /* Scale to [0-100] range. (couldn't do this inside util_out_print) */
if (ls_free < 1) /* No memory? Notify user. */
gtm_putmsg(VARLSTCNT(4) ERR_LOCKSPACEFULL, 2, DB_LEN_STR(gv_cur_region));
if (ls_free < 1 || memory)
{
if (ctl->subtop > ctl->subfree)
gtm_putmsg(VARLSTCNT(10) ERR_LOCKSPACEINFO, 8, REG_LEN_STR(gv_cur_region),
(ctl->max_prccnt - ctl->prccnt), ctl->max_prccnt,
(ctl->max_blkcnt - ctl->blkcnt), ctl->max_blkcnt, LEN_AND_LIT(" not "));
else
gtm_putmsg(VARLSTCNT(10) ERR_LOCKSPACEINFO, 8, REG_LEN_STR(gv_cur_region),
(ctl->max_prccnt - ctl->prccnt), ctl->max_prccnt,
(ctl->max_blkcnt - ctl->blkcnt), ctl->max_blkcnt, LEN_AND_LIT(" "));
}
free(ctl);
} else
{
gtm_putmsg(VARLSTCNT(2) ERR_BADREGION, 0);
locks = TRUE;
}
if (!locks)
{
gtm_putmsg(VARLSTCNT(4) ERR_NOLOCKMATCH, 2, REG_LEN_STR(gv_cur_region));
}
assert((ls_free <= 100) && (ls_free >= 0));
gtm_putmsg(VARLSTCNT(4) ERR_LOCKSPACEUSE, 2, ((int)ls_free),
cs_addrs->hdr->lock_space_size/OS_PAGELET_SIZE);
}
}
if (!match && reg.len != 0)
rts_error(VARLSTCNT(4) ERR_NOREGION, 2, reg.len, reg.addr);
}
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;
}
void db_init(gd_region *reg, sgmnt_data_ptr_t tsd)
{
static boolean_t mutex_init_done = FALSE;
boolean_t is_bg, read_only;
char machine_name[MAX_MCNAMELEN];
file_control *fc;
int gethostname_res, stat_res, mm_prot;
int4 status, semval, dblksize, fbwsize;
sm_long_t status_l;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
struct sembuf sop[3];
struct stat stat_buf;
union semun semarg;
struct semid_ds semstat;
struct shmid_ds shmstat;
struct statvfs dbvfs;
uint4 sopcnt;
unix_db_info *udi;
#ifdef periodic_timer_removed
void periodic_flush_check();
#endif
error_def(ERR_CLSTCONFLICT);
error_def(ERR_CRITSEMFAIL);
error_def(ERR_DBNAMEMISMATCH);
error_def(ERR_DBIDMISMATCH);
error_def(ERR_NLMISMATCHCALC);
error_def(ERR_REQRUNDOWN);
error_def(ERR_SYSCALL);
assert(tsd->acc_meth == dba_bg || tsd->acc_meth == dba_mm);
is_bg = (dba_bg == tsd->acc_meth);
read_only = reg->read_only;
new_dbinit_ipc = FALSE; /* we did not create a new ipc resource */
udi = FILE_INFO(reg);
memset(machine_name, 0, sizeof(machine_name));
if (GETHOSTNAME(machine_name, MAX_MCNAMELEN, gethostname_res))
rts_error(VARLSTCNT(5) ERR_TEXT, 2, LEN_AND_LIT("Unable to get the hostname"), errno);
assert(strlen(machine_name) < MAX_MCNAMELEN);
csa = &udi->s_addrs;
csa->db_addrs[0] = csa->db_addrs[1] = csa->lock_addrs[0] = NULL; /* to help in dbinit_ch and gds_rundown */
reg->opening = TRUE;
/*
* Create ftok semaphore for this region.
* We do not want to make ftok counter semaphore to be 2 for on mupip journal recover process.
*/
if (!ftok_sem_get(reg, !mupip_jnl_recover, GTM_ID, FALSE))
rts_error(VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
/*
* At this point we have ftok_semid sempahore based on ftok key.
* Any ftok conflicted region will block at this point.
* Say, a.dat and b.dat both has same ftok and we have process A to access a.dat and
* process B to access b.dat. In this case only one can continue to do db_init()
*/
fc = reg->dyn.addr->file_cntl;
fc->file_type = reg->dyn.addr->acc_meth;
fc->op = FC_READ;
fc->op_buff = (sm_uc_ptr_t)tsd;
fc->op_len = sizeof(*tsd);
fc->op_pos = 1;
dbfilop(fc); /* Read file header */
udi->shmid = tsd->shmid;
udi->semid = tsd->semid;
udi->sem_ctime = tsd->sem_ctime.ctime;
udi->shm_ctime = tsd->shm_ctime.ctime;
dbsecspc(reg, tsd); /* Find db segment size */
if (!mupip_jnl_recover)
{
if (INVALID_SEMID == udi->semid)
{
if (0 != udi->sem_ctime || INVALID_SHMID != udi->shmid || 0 != udi->shm_ctime)
/* We must have somthing wrong in protocol or, code, if this happens */
GTMASSERT;
/*
* Create new semaphore using IPC_PRIVATE. System guarantees a unique id.
*/
if (-1 == (udi->semid = semget(IPC_PRIVATE, FTOK_SEM_PER_ID, RWDALL | IPC_CREAT)))
{
udi->semid = INVALID_SEMID;
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control semget"), errno);
}
udi->shmid = INVALID_SHMID; /* reset shmid so dbinit_ch does not get confused in case we go there */
new_dbinit_ipc = TRUE;
tsd->semid = udi->semid;
semarg.val = GTM_ID;
/*
* Following will set semaphore number 2 (=FTOK_SEM_PER_ID - 1) value as GTM_ID.
* In case we have orphaned semaphore for some reason, mupip rundown will be
* able to identify GTM semaphores from the value and can remove.
*/
if (-1 == semctl(udi->semid, FTOK_SEM_PER_ID - 1, SETVAL, semarg))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control semctl SETVAL"), errno);
/*
* Warning: We must read the sem_ctime using IPC_STAT after SETVAL, which changes it.
* We must NOT do any more SETVAL after this. Our design is to use
* sem_ctime as creation time of semaphore.
*/
semarg.buf = &semstat;
if (-1 == semctl(udi->semid, FTOK_SEM_PER_ID - 1, IPC_STAT, semarg))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control semctl IPC_STAT"), errno);
tsd->sem_ctime.ctime = udi->sem_ctime = semarg.buf->sem_ctime;
} else
{
if (INVALID_SHMID == udi->shmid)
/* if mu_rndwn_file gets standalone access of this region and
* somehow mupip process crashes, we can have semid != -1 but shmid == -1
*/
rts_error(VARLSTCNT(10) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name),
ERR_TEXT, 2, LEN_AND_LIT("semid is valid but shmid is invalid"));
semarg.buf = &semstat;
if (-1 == semctl(udi->semid, 0, IPC_STAT, semarg))
/* file header has valid semid but semaphore does not exists */
rts_error(VARLSTCNT(6) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name));
else if (semarg.buf->sem_ctime != tsd->sem_ctime.ctime)
rts_error(VARLSTCNT(10) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name),
ERR_TEXT, 2, LEN_AND_LIT("sem_ctime does not match"));
if (-1 == shmctl(udi->shmid, IPC_STAT, &shmstat))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control shmctl"), errno);
else if (shmstat.shm_ctime != tsd->shm_ctime.ctime)
rts_error(VARLSTCNT(10) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name),
ERR_TEXT, 2, LEN_AND_LIT("shm_ctime does not match"));
}
/* We already have ftok semaphore of this region, so just plainly do semaphore operation */
/* This is the database access control semaphore for any region */
sop[0].sem_num = 0; sop[0].sem_op = 0; /* Wait for 0 */
sop[1].sem_num = 0; sop[1].sem_op = 1; /* Lock */
sopcnt = 2;
if (!read_only)
{
sop[2].sem_num = 1; sop[2].sem_op = 1; /* increment r/w access counter */
sopcnt = 3;
}
sop[0].sem_flg = sop[1].sem_flg = sop[2].sem_flg = SEM_UNDO | IPC_NOWAIT;
SEMOP(udi->semid, sop, sopcnt, status);
if (-1 == status)
{
errno_save = errno;
gtm_putmsg(VARLSTCNT(4) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg));
rts_error(VARLSTCNT(8) ERR_SYSCALL, 5, RTS_ERROR_LITERAL("semop()"), CALLFROM, errno_save);
}
} else /* for mupip_jnl_recover we were already in mu_rndwn_file and got "semid" semaphore */
{
if (INVALID_SEMID == udi->semid || 0 == udi->sem_ctime)
/* make sure mu_rndwn_file() has reset created semaphore for standalone access */
GTMASSERT;
if (INVALID_SHMID != udi->shmid || 0 != udi->shm_ctime)
/* make sure mu_rndwn_file() has reset shared memory */
GTMASSERT;
udi->shmid = INVALID_SHMID; /* reset shmid so dbinit_ch does not get confused in case we go there */
new_dbinit_ipc = TRUE;
}
sem_incremented = TRUE;
if (new_dbinit_ipc)
{
/* Create new shared memory using IPC_PRIVATE. System guarantees a unique id */
#ifdef __MVS__
if (-1 == (status_l = udi->shmid = shmget(IPC_PRIVATE, ROUND_UP(reg->sec_size, MEGA_BOUND),
__IPC_MEGA | IPC_CREAT | RWDALL)))
#else
if (-1 == (status_l = udi->shmid = shmget(IPC_PRIVATE, reg->sec_size, RWDALL | IPC_CREAT)))
#endif
{
udi->shmid = status_l = INVALID_SHMID;
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database shmget"), errno);
}
tsd->shmid = udi->shmid;
if (-1 == shmctl(udi->shmid, IPC_STAT, &shmstat))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control shmctl"), errno);
tsd->shm_ctime.ctime = udi->shm_ctime = shmstat.shm_ctime;
}
#ifdef DEBUG_DB64
status_l = (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)do_shmat(udi->shmid, next_smseg, SHM_RND));
next_smseg = (sm_uc_ptr_t)ROUND_UP((sm_long_t)(next_smseg + reg->sec_size), SHMAT_ADDR_INCS);
#else
status_l = (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)do_shmat(udi->shmid, 0, SHM_RND));
#endif
if (-1 == status_l)
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error attaching to database shared memory"), errno);
}
csa->nl = (node_local_ptr_t)csa->db_addrs[0];
csa->critical = (mutex_struct_ptr_t)(csa->db_addrs[0] + NODE_LOCAL_SIZE);
assert(((int)csa->critical & 0xf) == 0); /* critical should be 16-byte aligned */
#ifdef CACHELINE_SIZE
assert(0 == ((int)csa->critical & (CACHELINE_SIZE - 1)));
#endif
/* Note: Here we check jnl_sate from database file and its value cannot change without standalone access.
* The jnl_buff buffer should be initialized irrespective of read/write process */
JNL_INIT(csa, reg, tsd);
csa->backup_buffer = (backup_buff_ptr_t)(csa->db_addrs[0] + NODE_LOCAL_SPACE + JNL_SHARE_SIZE(tsd));
csa->lock_addrs[0] = (sm_uc_ptr_t)csa->backup_buffer + BACKUP_BUFFER_SIZE + 1;
csa->lock_addrs[1] = csa->lock_addrs[0] + LOCK_SPACE_SIZE(tsd) - 1;
csa->total_blks = tsd->trans_hist.total_blks; /* For test to see if file has extended */
if (new_dbinit_ipc)
{
memset(csa->nl, 0, sizeof(*csa->nl)); /* We allocated shared storage -- we have to init it */
if (JNL_ALLOWED(csa))
{ /* initialize jb->cycle to a value different from initial value of jpc->cycle (0). although this is not
* necessary right now, in the future, the plan is to change jnl_ensure_open() to only do a cycle mismatch
* check in order to determine whether to call jnl_file_open() or not. this is in preparation for that.
*/
csa->jnl->jnl_buff->cycle = 1;
}
}
if (is_bg)
csd = csa->hdr = (sgmnt_data_ptr_t)(csa->lock_addrs[1] + 1 + CACHE_CONTROL_SIZE(tsd));
else
{
csa->acc_meth.mm.mmblk_state = (mmblk_que_heads_ptr_t)(csa->lock_addrs[1] + 1);
FSTAT_FILE(udi->fd, &stat_buf, stat_res);
if (-1 == stat_res)
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), errno);
mm_prot = read_only ? PROT_READ : (PROT_READ | PROT_WRITE);
#ifdef DEBUG_DB64
if (-1 == (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)mmap((caddr_t)get_mmseg((size_t)stat_buf.st_size),
(size_t)stat_buf.st_size,
mm_prot,
GTM_MM_FLAGS, udi->fd, (off_t)0)))
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), errno);
put_mmseg((caddr_t)(csa->db_addrs[0]), (size_t)stat_buf.st_size);
#else
if (-1 == (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)mmap((caddr_t)NULL,
(size_t)stat_buf.st_size,
mm_prot,
GTM_MM_FLAGS, udi->fd, (off_t)0)))
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), errno);
#endif
csa->db_addrs[1] = csa->db_addrs[0] + stat_buf.st_size - 1;
csd = csa->hdr = (sgmnt_data_ptr_t)csa->db_addrs[0];
}
if (!csa->nl->glob_sec_init)
{
assert(new_dbinit_ipc);
if (is_bg)
*csd = *tsd;
if (csd->machine_name[0]) /* crash occured */
{
if (0 != memcmp(csd->machine_name, machine_name, MAX_MCNAMELEN)) /* crashed on some other node */
rts_error(VARLSTCNT(6) ERR_CLSTCONFLICT, 4, DB_LEN_STR(reg), LEN_AND_STR(csd->machine_name));
else
rts_error(VARLSTCNT(6) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(csd->machine_name));
}
if (is_bg)
{
bt_malloc(csa);
csa->nl->cache_off = -CACHE_CONTROL_SIZE(tsd);
db_csh_ini(csa);
}
db_csh_ref(csa);
strcpy(csa->nl->machine_name, machine_name); /* machine name */
assert(MAX_REL_NAME > gtm_release_name_len);
memcpy(csa->nl->now_running, gtm_release_name, gtm_release_name_len + 1); /* GT.M release name */
memcpy(csa->nl->label, GDS_LABEL, GDS_LABEL_SZ - 1); /* GDS label */
memcpy(csa->nl->fname, reg->dyn.addr->fname, reg->dyn.addr->fname_len); /* database filename */
csa->nl->creation_date_time = csd->creation.date_time;
csa->nl->highest_lbm_blk_changed = -1;
csa->nl->wcs_timers = -1;
csa->nl->nbb = BACKUP_NOT_IN_PROGRESS;
csa->nl->unique_id.uid = FILE_INFO(reg)->fileid; /* save what file we initialized this storage for */
/* save pointers in csa to access shared memory */
csa->nl->critical = (sm_off_t)((sm_uc_ptr_t)csa->critical - (sm_uc_ptr_t)csa->nl);
if (JNL_ALLOWED(csa))
csa->nl->jnl_buff = (sm_off_t)((sm_uc_ptr_t)csa->jnl->jnl_buff - (sm_uc_ptr_t)csa->nl);
csa->nl->backup_buffer = (sm_off_t)((sm_uc_ptr_t)csa->backup_buffer - (sm_uc_ptr_t)csa->nl);
csa->nl->hdr = (sm_off_t)((sm_uc_ptr_t)csd - (sm_uc_ptr_t)csa->nl);
csa->nl->lock_addrs = (sm_off_t)((sm_uc_ptr_t)csa->lock_addrs[0] - (sm_uc_ptr_t)csa->nl);
if (!read_only || is_bg)
{
csd->trans_hist.early_tn = csd->trans_hist.curr_tn;
csd->max_update_array_size = csd->max_non_bm_update_array_size
= ROUND_UP2(MAX_NON_BITMAP_UPDATE_ARRAY_SIZE(csd), UPDATE_ARRAY_ALIGN_SIZE);
csd->max_update_array_size += ROUND_UP2(MAX_BITMAP_UPDATE_ARRAY_SIZE, UPDATE_ARRAY_ALIGN_SIZE);
/* add current db_csh counters into the cumulative counters and reset the current counters */
#define TAB_DB_CSH_ACCT_REC(COUNTER, DUMMY1, DUMMY2) \
csd->COUNTER.cumul_count += csd->COUNTER.curr_count; \
csd->COUNTER.curr_count = 0;
#include "tab_db_csh_acct_rec.h"
#undef TAB_DB_CSH_ACCT_REC
}
if (!read_only)
{
if (is_bg)
{
assert(memcmp(csd, GDS_LABEL, GDS_LABEL_SZ - 1) == 0);
LSEEKWRITE(udi->fd, (off_t)0, (sm_uc_ptr_t)csd, sizeof(sgmnt_data), errno_save);
if (0 != errno_save)
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database write"), errno_save);
}
}
}
reg->dyn.addr->ext_blk_count = csd->extension_size;
mlk_shr_init(csa->lock_addrs[0], csd->lock_space_size, csa, (FALSE == read_only));
DEBUG_ONLY(locknl = csa->nl;) /* for DEBUG_ONLY LOCK_HIST macro */
gd_region *dbfilopn (gd_region *reg)
{
unix_db_info *udi;
parse_blk pblk;
mstr file;
char *fnptr, fbuff[MAX_FBUFF + 1];
struct stat buf;
gd_region *prev_reg;
gd_segment *seg;
int status;
bool raw;
int stat_res;
seg = reg->dyn.addr;
assert(seg->acc_meth == dba_bg || seg->acc_meth == dba_mm);
if (NULL == seg->file_cntl)
{
seg->file_cntl = (file_control *)malloc(sizeof(*seg->file_cntl));
memset(seg->file_cntl, 0, sizeof(*seg->file_cntl));
}
if (NULL == seg->file_cntl->file_info)
{
seg->file_cntl->file_info = (void *)malloc(sizeof(unix_db_info));
memset(seg->file_cntl->file_info, 0, sizeof(unix_db_info));
}
file.addr = (char *)seg->fname;
file.len = seg->fname_len;
memset(&pblk, 0, sizeof(pblk));
pblk.buffer = fbuff;
pblk.buff_size = MAX_FBUFF;
pblk.fop = (F_SYNTAXO | F_PARNODE);
memcpy(fbuff,file.addr,file.len);
*(fbuff + file.len) = '\0';
if (is_raw_dev(fbuff))
{
raw = TRUE;
pblk.def1_buf = DEF_NODBEXT;
pblk.def1_size = sizeof(DEF_NODBEXT) - 1;
} else
{
raw = FALSE;
pblk.def1_buf = DEF_DBEXT;
pblk.def1_size = sizeof(DEF_DBEXT) - 1;
}
status = parse_file(&file, &pblk);
if (!(status & 1))
{
if (GTCM_GNP_SERVER_IMAGE != image_type)
{
free(seg->file_cntl->file_info);
free(seg->file_cntl);
seg->file_cntl = 0;
}
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), status);
}
assert(pblk.b_esl < sizeof(seg->fname));
memcpy(seg->fname, pblk.buffer, pblk.b_esl);
pblk.buffer[pblk.b_esl] = 0;
seg->fname[pblk.b_esl] = 0;
seg->fname_len = pblk.b_esl;
if (pblk.fnb & F_HAS_NODE)
{ /* Remote node specification given */
assert(pblk.b_node && pblk.l_node[pblk.b_node - 1] == ':');
gvcmy_open(reg, &pblk);
return (gd_region *)-1;
}
fnptr = (char *)seg->fname + pblk.b_node;
udi = FILE_INFO(reg);
udi->raw = raw;
udi->fn = (char *)fnptr;
OPENFILE(fnptr, O_RDWR, udi->fd);
udi->ftok_semid = INVALID_SEMID;
udi->semid = INVALID_SEMID;
udi->shmid = INVALID_SHMID;
udi->sem_ctime = 0;
udi->shm_ctime = 0;
reg->read_only = FALSE; /* maintain csa->read_write simultaneously */
udi->s_addrs.read_write = TRUE; /* maintain reg->read_only simultaneously */
if (udi->fd == -1)
{
OPENFILE(fnptr, O_RDONLY, udi->fd);
if (udi->fd == -1)
{
errno_save = errno;
if (GTCM_GNP_SERVER_IMAGE != image_type)
{
free(seg->file_cntl->file_info);
free(seg->file_cntl);
seg->file_cntl = 0;
}
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), errno_save);
}
reg->read_only = TRUE; /* maintain csa->read_write simultaneously */
udi->s_addrs.read_write = FALSE; /* maintain reg->read_only simultaneously */
}
STAT_FILE(fnptr, &buf, stat_res);
set_gdid_from_stat(&udi->fileid, &buf);
if (prev_reg = gv_match(reg))
{
close(udi->fd);
free(seg->file_cntl->file_info);
free(seg->file_cntl);
seg->file_cntl = 0;
return prev_reg;
}
return reg;
}
boolean_t mu_truncate(int4 truncate_percent)
{
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
int num_local_maps;
int lmap_num, lmap_blk_num;
int bml_status, sigkill;
int save_errno;
int ftrunc_status;
uint4 jnl_status;
uint4 old_total, new_total;
uint4 old_free, new_free;
uint4 end_blocks;
int4 blks_in_lmap, blk;
gtm_uint64_t before_trunc_file_size;
off_t trunc_file_size;
off_t padding;
uchar_ptr_t lmap_addr;
boolean_t was_crit;
uint4 found_busy_blk;
srch_blk_status bmphist;
srch_blk_status *blkhist;
srch_hist alt_hist;
trans_num curr_tn;
blk_hdr_ptr_t lmap_blk_hdr;
block_id *blkid_ptr;
unix_db_info *udi;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
char *err_msg;
intrpt_state_t prev_intrpt_state;
off_t offset;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
csa = cs_addrs;
csd = cs_data;
if (dba_mm == csd->acc_meth)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCNOTBG, 2, REG_LEN_STR(gv_cur_region));
return TRUE;
}
if ((GDSVCURR != csd->desired_db_format) || (csd->blks_to_upgrd != 0))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCNOV4, 2, REG_LEN_STR(gv_cur_region));
return TRUE;
}
if (csa->ti->free_blocks < (truncate_percent * csa->ti->total_blks / 100))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_MUTRUNCNOSPACE, 3, REG_LEN_STR(gv_cur_region), truncate_percent);
return TRUE;
}
/* already checked for parallel truncates on this region --- see mupip_reorg.c */
gv_target = NULL;
assert(csa->nl->trunc_pid == process_id);
assert(dba_mm != csd->acc_meth);
old_total = csa->ti->total_blks;
old_free = csa->ti->free_blocks;
sigkill = 0;
found_busy_blk = 0;
memset(&alt_hist, 0, SIZEOF(alt_hist)); /* null-initialize history */
assert(csd->bplmap == BLKS_PER_LMAP);
end_blocks = old_total % BLKS_PER_LMAP; /* blocks in the last lmap (first one we start scanning) */
if (0 == end_blocks)
end_blocks = BLKS_PER_LMAP;
num_local_maps = DIVIDE_ROUND_UP(old_total, BLKS_PER_LMAP);
/* ======================================== PHASE 1 ======================================== */
for (lmap_num = num_local_maps - 1; (lmap_num > 0 && !found_busy_blk); lmap_num--)
{
if (mu_ctrly_occurred || mu_ctrlc_occurred)
return TRUE;
assert(csa->ti->total_blks >= old_total); /* otherwise, a concurrent truncate happened... */
if (csa->ti->total_blks != old_total) /* Extend (likely called by mupip extend) -- don't truncate */
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_MUTRUNCNOSPACE, 3, REG_LEN_STR(gv_cur_region),
truncate_percent);
return TRUE;
}
lmap_blk_num = lmap_num * BLKS_PER_LMAP;
if (csa->nl->highest_lbm_with_busy_blk >= lmap_blk_num)
{
found_busy_blk = lmap_blk_num;
break;
}
blks_in_lmap = (lmap_num == num_local_maps - 1) ? end_blocks : BLKS_PER_LMAP;
/* Loop through non-bitmap blocks of this lmap, do recycled2free */
DBGEHND((stdout, "DBG:: lmap_num = [%lu], lmap_blk_num = [%lu], blks_in_lmap = [%lu]\n",
lmap_num, lmap_blk_num, blks_in_lmap));
for (blk = 1; blk < blks_in_lmap && blk != -1 && !found_busy_blk;)
{
t_begin(ERR_MUTRUNCFAIL, UPDTRNS_DB_UPDATED_MASK);
for (;;) /* retry loop for recycled to free transactions */
{
curr_tn = csd->trans_hist.curr_tn;
/* Read the nth local bitmap into memory */
bmphist.blk_num = lmap_blk_num;
bmphist.buffaddr = t_qread(bmphist.blk_num, &bmphist.cycle, &bmphist.cr);
lmap_blk_hdr = (blk_hdr_ptr_t)bmphist.buffaddr;
if (!(bmphist.buffaddr) || (BM_SIZE(BLKS_PER_LMAP) != lmap_blk_hdr->bsiz))
{ /* Could not read the block successfully. Retry. */
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
lmap_addr = bmphist.buffaddr + SIZEOF(blk_hdr);
/* starting from the hint (blk itself), find the first busy or recycled block */
blk = bml_find_busy_recycled(blk, lmap_addr, blks_in_lmap, &bml_status);
assert(blk < BLKS_PER_LMAP);
if (blk == -1 || blk >= blks_in_lmap)
{ /* done with this lmap, continue to next */
t_abort(gv_cur_region, csa);
break;
}
else if (BLK_BUSY == bml_status || csa->nl->highest_lbm_with_busy_blk >= lmap_blk_num)
{ /* stop processing blocks... skip ahead to phase 2 */
found_busy_blk = lmap_blk_num;
t_abort(gv_cur_region, csa);
break;
}
else if (BLK_RECYCLED == bml_status)
{ /* Write PBLK records for recycled blocks only if before_image journaling is
* enabled. t_end() takes care of checking if journaling is enabled and
* writing PBLK record. We have to at least mark the recycled block as free.
*/
RESET_UPDATE_ARRAY;
update_trans = UPDTRNS_DB_UPDATED_MASK;
*((block_id *)update_array_ptr) = blk;
update_array_ptr += SIZEOF(block_id);
*(int *)update_array_ptr = 0;
alt_hist.h[1].blk_num = 0;
alt_hist.h[0].level = 0;
alt_hist.h[0].cse = NULL;
alt_hist.h[0].tn = curr_tn;
alt_hist.h[0].blk_num = lmap_blk_num + blk;
alt_hist.h[0].buffaddr = t_qread(alt_hist.h[0].blk_num,
&alt_hist.h[0].cycle, &alt_hist.h[0].cr);
if (!alt_hist.h[0].buffaddr)
{
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
if (!t_recycled2free(&alt_hist.h[0]))
{
t_retry(cdb_sc_lostbmlcr);
continue;
}
t_write_map(&bmphist, (unsigned char *)update_array, curr_tn, 0);
/* Set the opcode for INCTN record written by t_end() */
inctn_opcode = inctn_blkmarkfree;
if ((trans_num)0 == t_end(&alt_hist, NULL, TN_NOT_SPECIFIED))
continue;
/* block processed, scan from the next one */
blk++;
break;
} else
{
assert(t_tries < CDB_STAGNATE);
t_retry(cdb_sc_badbitmap);
continue;
}
} /* END recycled2free retry loop */
} /* END scanning blocks of this particular lmap */
/* Write PBLK for the bitmap block, in case it hasn't been written i.e. t_end() was never called above */
/* Do a transaction that just increments the bitmap block's tn so that t_end() can do its thing */
DBGEHND((stdout, "DBG:: bitmap block inctn -- lmap_blk_num = [%lu]\n", lmap_blk_num));
t_begin(ERR_MUTRUNCFAIL, UPDTRNS_DB_UPDATED_MASK);
for (;;)
{
RESET_UPDATE_ARRAY;
BLK_ADDR(blkid_ptr, SIZEOF(block_id), block_id);
*blkid_ptr = 0;
update_trans = UPDTRNS_DB_UPDATED_MASK;
inctn_opcode = inctn_mu_reorg; /* inctn_mu_truncate */
curr_tn = csd->trans_hist.curr_tn;
blkhist = &alt_hist.h[0];
blkhist->blk_num = lmap_blk_num;
blkhist->tn = curr_tn;
blkhist->cse = NULL; /* start afresh (do not use value from previous retry) */
/* Read the nth local bitmap into memory */
blkhist->buffaddr = t_qread(lmap_blk_num, (sm_int_ptr_t)&blkhist->cycle, &blkhist->cr);
lmap_blk_hdr = (blk_hdr_ptr_t)blkhist->buffaddr;
if (!(blkhist->buffaddr) || (BM_SIZE(BLKS_PER_LMAP) != lmap_blk_hdr->bsiz))
{ /* Could not read the block successfully. Retry. */
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
t_write_map(blkhist, (unsigned char *)blkid_ptr, curr_tn, 0);
blkhist->blk_num = 0; /* create empty history for bitmap block */
if ((trans_num)0 == t_end(&alt_hist, NULL, TN_NOT_SPECIFIED))
continue;
break;
}
} /* END scanning lmaps */
/* ======================================== PHASE 2 ======================================== */
assert(!csa->now_crit);
for (;;)
{ /* wait for FREEZE, we don't want to truncate a frozen database */
grab_crit(gv_cur_region);
if (FROZEN_CHILLED(cs_data))
DO_CHILLED_AUTORELEASE(csa, 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;
}
}
assert(csa->nl->trunc_pid == process_id);
/* Flush pending updates to disk. If this is not done, old updates can be flushed AFTER ftruncate, extending the file. */
if (!wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_MSYNC_DB))
{
assert(FALSE);
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_BUFFLUFAILED, 4, LEN_AND_LIT("MUPIP REORG TRUNCATE"),
DB_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return FALSE;
}
csa->nl->highest_lbm_with_busy_blk = MAX(found_busy_blk, csa->nl->highest_lbm_with_busy_blk);
assert(IS_BITMAP_BLK(csa->nl->highest_lbm_with_busy_blk));
new_total = MIN(old_total, csa->nl->highest_lbm_with_busy_blk + BLKS_PER_LMAP);
if (mu_ctrly_occurred || mu_ctrlc_occurred)
{
rel_crit(gv_cur_region);
return TRUE;
} else if (csa->ti->total_blks != old_total || new_total == old_total)
{
assert(csa->ti->total_blks >= old_total); /* Better have been an extend, not a truncate... */
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_MUTRUNCNOSPACE, 3, REG_LEN_STR(gv_cur_region), truncate_percent);
rel_crit(gv_cur_region);
return TRUE;
} else if (GDSVCURR != csd->desired_db_format || csd->blks_to_upgrd != 0 || !csd->fully_upgraded)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCNOV4, 2, REG_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return TRUE;
} else if (SNAPSHOTS_IN_PROG(csa->nl))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCSSINPROG, 2, REG_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return TRUE;
} else if (BACKUP_NOT_IN_PROGRESS != cs_addrs->nl->nbb)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCBACKINPROG, 2, REG_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return TRUE;
}
DEFER_INTERRUPTS(INTRPT_IN_TRUNC, prev_intrpt_state);
if (JNL_ENABLED(csa))
{ /* Write JRT_TRUNC and INCTN records */
if (!jgbl.dont_reset_gbl_jrec_time)
SET_GBL_JREC_TIME; /* needed before jnl_ensure_open as that can write jnl records */
jpc = csa->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, csa);
if (SS_NORMAL != jnl_status)
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
else
{
if (0 == jpc->pini_addr)
jnl_put_jrt_pini(csa);
jnl_write_trunc_rec(csa, old_total, csa->ti->free_blocks, new_total);
inctn_opcode = inctn_mu_reorg;
jnl_write_inctn_rec(csa);
jnl_status = jnl_flush(gv_cur_region);
if (SS_NORMAL != jnl_status)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_JNLFLUSH, 2, JNL_LEN_STR(csd),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error with journal flush during mu_truncate"),
jnl_status);
assert(NOJNL == jpc->channel); /* jnl file lost has been triggered */
}
}
}
/* Good to go ahead and REALLY truncate (reduce total_blks, clear cache_array, FTRUNCATE) */
curr_tn = csa->ti->curr_tn;
CHECK_TN(csa, csd, curr_tn);
udi = FILE_INFO(gv_cur_region);
/* Information used by recover_truncate to check if the file size and csa->ti->total_blks are INCONSISTENT */
trunc_file_size = BLK_ZERO_OFF(csd->start_vbn) + ((off_t)csd->blk_size * (new_total + 1));
csd->after_trunc_total_blks = new_total;
csd->before_trunc_free_blocks = csa->ti->free_blocks;
csd->before_trunc_total_blks = old_total; /* Flags interrupted truncate for recover_truncate */
/* file size and total blocks: INCONSISTENT */
csa->ti->total_blks = new_total;
/* past the point of no return -- shared memory intact */
assert(csa->ti->free_blocks >= DELTA_FREE_BLOCKS(old_total, new_total));
csa->ti->free_blocks -= DELTA_FREE_BLOCKS(old_total, new_total);
new_free = csa->ti->free_blocks;
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_1); /* 55 : Issue a kill -9 before 1st fsync */
fileheader_sync(gv_cur_region);
DB_FSYNC(gv_cur_region, udi, csa, db_fsync_in_prog, save_errno);
CHECK_DBSYNC(gv_cur_region, save_errno);
/* past the point of no return -- shared memory deleted */
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_2); /* 56 : Issue a kill -9 after 1st fsync */
clear_cache_array(csa, csd, gv_cur_region, new_total, old_total);
offset = (off_t)BLK_ZERO_OFF(csd->start_vbn) + (off_t)new_total * csd->blk_size;
save_errno = db_write_eof_block(udi, udi->fd, csd->blk_size, offset, &(TREF(dio_buff)));
if (0 != save_errno)
{
err_msg = (char *)STRERROR(errno);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_MUTRUNCERROR, 4, REG_LEN_STR(gv_cur_region), LEN_AND_STR(err_msg));
return FALSE;
}
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_3); /* 57 : Issue a kill -9 after reducing csa->ti->total_blks, before FTRUNCATE */
/* Execute an ftruncate() and truncate the DB file
* ftruncate() is a SYSTEM CALL on almost all platforms (except SunOS)
* It ignores kill -9 signal till its operation is completed.
* So we can safely assume that the result of ftruncate() will be complete.
*/
FTRUNCATE(FILE_INFO(gv_cur_region)->fd, trunc_file_size, ftrunc_status);
if (0 != ftrunc_status)
{
err_msg = (char *)STRERROR(errno);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_MUTRUNCERROR, 4, REG_LEN_STR(gv_cur_region), LEN_AND_STR(err_msg));
/* should go through recover_truncate now, which will again try to FTRUNCATE */
return FALSE;
}
/* file size and total blocks: CONSISTENT (shrunk) */
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_4); /* 58 : Issue a kill -9 after FTRUNCATE, before 2nd fsync */
csa->nl->root_search_cycle++; /* Force concurrent processes to restart in t_end/tp_tend to make sure no one
* tries to commit updates past the end of the file. Bitmap validations together
* with highest_lbm_with_busy_blk should actually be sufficient, so this is
* just to be safe.
*/
csd->before_trunc_total_blks = 0; /* indicate CONSISTENT */
/* Increment TN */
assert(csa->ti->early_tn == csa->ti->curr_tn);
csd->trans_hist.early_tn = csd->trans_hist.curr_tn + 1;
INCREMENT_CURR_TN(csd);
fileheader_sync(gv_cur_region);
DB_FSYNC(gv_cur_region, udi, csa, db_fsync_in_prog, save_errno);
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_5); /* 58 : Issue a kill -9 after after 2nd fsync */
CHECK_DBSYNC(gv_cur_region, save_errno);
ENABLE_INTERRUPTS(INTRPT_IN_TRUNC, prev_intrpt_state);
curr_tn = csa->ti->curr_tn;
rel_crit(gv_cur_region);
send_msg_csa(CSA_ARG(csa) VARLSTCNT(7) ERR_MUTRUNCSUCCESS, 5, DB_LEN_STR(gv_cur_region), old_total, new_total, &curr_tn);
util_out_print("Truncated region: !AD. Reduced total blocks from [!UL] to [!UL]. Reduced free blocks from [!UL] to [!UL].",
FLUSH, REG_LEN_STR(gv_cur_region), old_total, new_total, old_free, new_free);
return TRUE;
} /* END of mu_truncate() */
void dse_crit(void)
{
int util_len, dse_crit_count;
char util_buff[MAX_UTIL_LEN];
boolean_t crash = FALSE, cycle = FALSE, owner = FALSE;
gd_region *save_region, *r_local, *r_top;
crash = ((cli_present("CRASH") == CLI_PRESENT) || (cli_present("RESET") == CLI_PRESENT));
cycle = (CLI_PRESENT == cli_present("CYCLE"));
if (cli_present("SEIZE") == CLI_PRESENT || cycle)
{
if (gv_cur_region->read_only && !cycle)
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
if (cs_addrs->now_crit)
{
util_out_print("!/Write critical section already seized.!/", TRUE);
return;
}
crash_count = cs_addrs->critical->crashcnt;
grab_crit_encr_cycle_sync(gv_cur_region);
cs_addrs->hold_onto_crit = TRUE; /* need to do this AFTER grab_crit */
cs_addrs->dse_crit_seize_done = TRUE;
util_out_print("!/Seized write critical section.!/", TRUE);
if (!cycle)
return;
}
if (cli_present("RELEASE") == CLI_PRESENT || cycle)
{
if (gv_cur_region->read_only && !cycle)
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
if (!cs_addrs->now_crit)
{
util_out_print("!/Critical section already released.!/", TRUE);
return;
}
crash_count = cs_addrs->critical->crashcnt;
if (cs_addrs->now_crit)
{ /* user wants crit to be released unconditionally so "was_crit" not checked like everywhere else */
assert(cs_addrs->hold_onto_crit && cs_addrs->dse_crit_seize_done);
cs_addrs->dse_crit_seize_done = FALSE;
cs_addrs->hold_onto_crit = FALSE; /* need to do this before the rel_crit */
rel_crit(gv_cur_region);
util_out_print("!/Released write critical section.!/", TRUE);
}
# ifdef DEBUG
else
assert(!cs_addrs->hold_onto_crit && !cs_addrs->dse_crit_seize_done);
# endif
return;
}
if (cli_present("INIT") == CLI_PRESENT)
{
if (gv_cur_region->read_only)
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
cs_addrs->hdr->image_count = 0;
UNIX_ONLY(gtm_mutex_init(gv_cur_region, NUM_CRIT_ENTRY(cs_addrs->hdr), crash));
VMS_ONLY(mutex_init(cs_addrs->critical, NUM_CRIT_ENTRY(cs_addrs->hdr), crash));
cs_addrs->nl->in_crit = 0;
cs_addrs->now_crit = FALSE;
util_out_print("!/Reinitialized critical section.!/", TRUE);
return;
}
if (cli_present("REMOVE") == CLI_PRESENT)
{
if (gv_cur_region->read_only)
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
if (cs_addrs->nl->in_crit == 0)
{
util_out_print("!/The write critical section is unowned!/", TRUE);
return;
}
UNIX_ONLY(assert(LOCK_AVAILABLE != cs_addrs->critical->semaphore.u.parts.latch_pid);)
VMS_ONLY(assert(cs_addrs->critical->semaphore >= 0);)
void dse_chng_bhead(void)
{
blk_hdr new_hdr;
blk_segment *bs1, *bs_ptr;
block_id blk;
boolean_t chng_blk, ismap, was_hold_onto_crit;
int4 blk_seg_cnt, blk_size; /* needed for BLK_INIT,BLK_SEG and BLK_FINI macros */
int4 x;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
srch_blk_status blkhist;
trans_num tn;
uint4 mapsize;
csa = cs_addrs;
if (gv_cur_region->read_only)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
chng_blk = FALSE;
if (BADDSEBLK == (blk = dse_getblk("BLOCK", DSEBMLOK, DSEBLKCUR))) /* WARNING: assignment */
return;
csd = csa->hdr;
assert(csd == cs_data);
blk_size = csd->blk_size;
ismap = IS_BITMAP_BLK(blk);
mapsize = BM_SIZE(csd->bplmap);
t_begin_crit(ERR_DSEFAIL);
blkhist.blk_num = blk;
if (!(blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(1) ERR_DSEBLKRDFAIL);
new_hdr = *(blk_hdr_ptr_t)blkhist.buffaddr;
if (CLI_PRESENT == cli_present("LEVEL"))
{
if (!cli_get_hex("LEVEL", (uint4 *)&x))
{
t_abort(gv_cur_region, csa);
return;
}
if (ismap && (unsigned char)x != LCL_MAP_LEVL)
{
util_out_print("Error: invalid level for a bit map block.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
if (!ismap && (x < 0 || x > MAX_BT_DEPTH + 1))
{
util_out_print("Error: invalid level.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
new_hdr.levl = (unsigned char)x;
chng_blk = TRUE;
if (new_hdr.bsiz < SIZEOF(blk_hdr))
new_hdr.bsiz = SIZEOF(blk_hdr);
if (new_hdr.bsiz > blk_size)
new_hdr.bsiz = blk_size;
}
if (CLI_PRESENT == cli_present("BSIZ"))
{
if (!cli_get_hex("BSIZ", (uint4 *)&x))
{
t_abort(gv_cur_region, csa);
return;
}
if (ismap && x != mapsize)
{
util_out_print("Error: invalid bsiz.", TRUE);
t_abort(gv_cur_region, csa);
return;
} else if (x < SIZEOF(blk_hdr) || x > blk_size)
{
util_out_print("Error: invalid bsiz.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
chng_blk = TRUE;
new_hdr.bsiz = x;
}
if (!chng_blk)
t_abort(gv_cur_region, csa);
else
{
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blkhist.buffaddr + SIZEOF(new_hdr), new_hdr.bsiz - SIZEOF(new_hdr));
if (!BLK_FINI(bs_ptr, bs1))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_AIMGBLKFAIL, 3, blk, DB_LEN_STR(gv_cur_region));
t_abort(gv_cur_region, csa);
return;
}
t_write(&blkhist, (unsigned char *)bs1, 0, 0, new_hdr.levl, TRUE, FALSE, GDS_WRITE_KILLTN);
BUILD_AIMG_IF_JNL_ENABLED(csd, csa->ti->curr_tn);
t_end(&dummy_hist, NULL, TN_NOT_SPECIFIED);
}
if (CLI_PRESENT == cli_present("TN"))
{
if (!cli_get_hex64("TN", &tn))
return;
t_begin_crit(ERR_DSEFAIL);
CHECK_TN(csa, csd, csd->trans_hist.curr_tn); /* can issue rts_error TNTOOLARGE */
assert(csa->ti->early_tn == csa->ti->curr_tn);
if (NULL == (blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(1) ERR_DSEBLKRDFAIL);
t_abort(gv_cur_region, csa);
return;
}
if (new_hdr.bsiz < SIZEOF(blk_hdr))
new_hdr.bsiz = SIZEOF(blk_hdr);
if (new_hdr.bsiz > blk_size)
new_hdr.bsiz = blk_size;
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blkhist.buffaddr + SIZEOF(new_hdr), new_hdr.bsiz - SIZEOF(new_hdr));
BLK_FINI(bs_ptr, bs1);
t_write(&blkhist, (unsigned char *)bs1, 0, 0,
((blk_hdr_ptr_t)blkhist.buffaddr)->levl, TRUE, FALSE, GDS_WRITE_KILLTN);
/* Pass the desired tn as argument to bg_update/mm_update below */
BUILD_AIMG_IF_JNL_ENABLED_AND_T_END_WITH_EFFECTIVE_TN(csa, csd, tn, &dummy_hist);
}
return;
}
int jnl_file_extend(jnl_private_control *jpc, uint4 total_jnl_rec_size)
{
file_control *fc;
boolean_t need_extend;
jnl_buffer_ptr_t jb;
jnl_create_info jnl_info;
jnl_file_header header;
uint4 new_alq;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
char prev_jnl_fn[JNL_NAME_SIZE];
uint4 jnl_status = 0, status;
int new_blocks, result;
GTM_BAVAIL_TYPE avail_blocks;
uint4 aligned_tot_jrec_size, count;
switch(jpc->region->dyn.addr->acc_meth)
{
case dba_mm:
case dba_bg:
csa = &FILE_INFO(jpc->region)->s_addrs;
break;
default:
GTMASSERT;
}
csd = csa->hdr;
assert(csa == cs_addrs && csd == cs_data);
assert(csa->now_crit || (csd->clustered && (CCST_CLOSED == csa->nl->ccp_state)));
assert(jpc->region == gv_cur_region);
assert(csa->jnl_state == csd->jnl_state);
if (!JNL_ENABLED(csa) || (NOJNL == jpc->channel) || (JNL_FILE_SWITCHED(jpc)))
GTMASSERT; /* crit and messing with the journal file - how could it have vanished? */
if (!csd->jnl_deq)
{
assert(DIVIDE_ROUND_UP(total_jnl_rec_size, DISK_BLOCK_SIZE) <= csd->jnl_alq);
assert(csd->jnl_alq == csd->autoswitchlimit);
new_blocks = csd->jnl_alq;
} else
/* May cause extension of csd->jnl_deq * n blocks where n > 0 */
new_blocks = ROUND_UP(DIVIDE_ROUND_UP(total_jnl_rec_size, DISK_BLOCK_SIZE), csd->jnl_deq);
jpc->status = SS_NORMAL;
jb = jpc->jnl_buff;
assert(0 <= new_blocks);
DEBUG_ONLY(count = 0);
for (need_extend = (0 != new_blocks); need_extend; )
{
DEBUG_ONLY(count++);
/* usually we will do the loop just once where we do the file extension.
* rarely we might need to do an autoswitch instead after which again rarely
* we might need to do an extension on the new journal to fit in the transaction's journal requirements.
* therefore we should do this loop a maximum of twice. hence the assert below.
*/
assert(count <= 2);
need_extend = FALSE;
if (SS_NORMAL == (status = disk_block_available(jpc->channel, &avail_blocks, TRUE)))
{
if ((new_blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
if (new_blocks > avail_blocks)
{ /* if we cannot satisfy the request, it is an error */
send_msg(VARLSTCNT(6) ERR_NOSPACEEXT, 4, JNL_LEN_STR(csd),
new_blocks, avail_blocks);
new_blocks = 0;
jpc->status = SS_NORMAL;
break;
} else
send_msg(VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, JNL_LEN_STR(csd),
(avail_blocks - new_blocks));
}
} else
send_msg(VARLSTCNT(5) ERR_JNLFILEXTERR, 2, JNL_LEN_STR(csd), status);
new_alq = jb->filesize + new_blocks;
/* ensure current journal file size is well within autoswitchlimit --> design constraint */
assert(csd->autoswitchlimit >= jb->filesize);
if (csd->autoswitchlimit < (jb->filesize + (EXTEND_WARNING_FACTOR * new_blocks))) /* close to max */
send_msg(VARLSTCNT(5) ERR_JNLSPACELOW, 3, JNL_LEN_STR(csd), csd->autoswitchlimit - jb->filesize);
if (csd->autoswitchlimit < new_alq)
{ /* Reached max, need to autoswitch */
/* Ensure new journal file can hold the entire current transaction's journal record requirements */
assert(csd->autoswitchlimit >= MAX_REQD_JNL_FILE_SIZE(total_jnl_rec_size));
memset(&jnl_info, 0, sizeof(jnl_info));
jnl_info.prev_jnl = &prev_jnl_fn[0];
set_jnl_info(gv_cur_region, &jnl_info);
assert(JNL_ENABLED(csa) && (NOJNL != jpc->channel) && !(JNL_FILE_SWITCHED(jpc)));
jnl_status = jnl_ensure_open();
if (0 == jnl_status)
{ /* flush the cache and jnl-buffer-contents to current journal file before
* switching to a new journal. */
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH);
jnl_file_close(gv_cur_region, TRUE, TRUE);
} else
rts_error(VARLSTCNT(7) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region), jpc->status);
assert(!jgbl.forw_phase_recovery || (NULL != jgbl.mur_pini_addr_reset_fnptr));
if (jgbl.forw_phase_recovery && (NULL != jgbl.mur_pini_addr_reset_fnptr))
(*jgbl.mur_pini_addr_reset_fnptr)();
assert(!jnl_info.no_rename);
assert(!jnl_info.no_prev_link);
if (EXIT_NRM == cre_jnl_file(&jnl_info))
{
assert(0 == memcmp(csd->jnl_file_name, jnl_info.jnl, jnl_info.jnl_len));
assert(csd->jnl_file_name[jnl_info.jnl_len] == '\0');
assert(csd->jnl_file_len == jnl_info.jnl_len);
assert(csd->jnl_buffer_size == jnl_info.buffer);
assert(csd->jnl_alq == jnl_info.alloc);
assert(csd->jnl_deq == jnl_info.extend);
assert(csd->jnl_before_image == jnl_info.before_images);
csd->trans_hist.header_open_tn = jnl_info.tn; /* needed for successful jnl_file_open() */
send_msg(VARLSTCNT(4) ERR_NEWJNLFILECREATE, 2, JNL_LEN_STR(csd));
fc = gv_cur_region->dyn.addr->file_cntl;
fc->op = FC_WRITE;
fc->op_buff = (sm_uc_ptr_t)csd;
status = dbfilop(fc);
if (SS_NORMAL != status)
send_msg(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), status);
assert(JNL_ENABLED(csa));
/* call jnl_ensure_open instead of jnl_file_open to make sure jpc->pini_addr is set to 0 */
jnl_status = jnl_ensure_open(); /* sets jpc->status */
if (0 != jnl_status)
rts_error(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
assert(jb->filesize == csd->jnl_alq);
aligned_tot_jrec_size = ALIGNED_ROUND_UP(MAX_REQD_JNL_FILE_SIZE(total_jnl_rec_size),
csd->jnl_alq, csd->jnl_deq);
if (aligned_tot_jrec_size > csd->jnl_alq)
{ /* need to extend more than initial allocation in the new journal file
* to accommodate the current transaction.
*/
new_blocks = aligned_tot_jrec_size - csd->jnl_alq;
assert(new_blocks);
assert(0 == new_blocks % csd->jnl_deq);
need_extend = TRUE;
}
} else
{
send_msg(VARLSTCNT(4) ERR_JNLCREATERR, 2, JNL_LEN_STR(csd));
jpc->status = ERR_JNLNOCREATE;
new_blocks = -1;
}
} else
{
assert(!need_extend); /* ensure we won't go through the for loop again */
/* Virtually extend currently used journal file */
jb->filesize = new_alq; /* Actually this is virtual file size blocks */
DO_FILE_READ(jpc->channel, 0, &header, JNL_HDR_LEN, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
rts_error(VARLSTCNT(5) ERR_JNLRDERR, 2, JNL_LEN_STR(csd), jpc->status);
assert((header.virtual_size + new_blocks) == new_alq);
header.virtual_size = new_alq;
DO_FILE_WRITE(jpc->channel, 0, &header, JNL_HDR_LEN, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
rts_error(VARLSTCNT(5) ERR_JNLWRERR, 2, JNL_LEN_STR(csd), jpc->status);
}
if (0 >= new_blocks)
break;
}
if (0 >= new_blocks)
{
jpc->status = ERR_JNLREADEOF;
jnl_file_lost(jpc, ERR_JNLEXTEND);
new_blocks = -1;
}
return new_blocks;
}
/* This function is called primarily to append a new histinfo record to the replication instance file by one of the following
* 1) MUPIP REPLIC -SOURCE -START -ROOTPRIMARY command (after forking the child source server) if it created the journal pool.
* 2) MUPIP REPLIC -SOURCE -ACTIVATE -ROOTPRIMARY command if this is a propagating primary to root primary transition.
* In addition, this function also initializes the "lms_group_info" field in the instance file (from the "inst_info" field)
* if the current value is NULL.
*/
void gtmsource_rootprimary_init(seq_num start_seqno)
{
unix_db_info *udi;
repl_histinfo histinfo;
boolean_t was_crit, switch_jnl;
gd_region *reg, *region_top;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
uint4 jnl_status;
udi = FILE_INFO(jnlpool.jnlpool_dummy_reg);
assert(NULL != jnlpool.repl_inst_filehdr);
/* Update journal pool fields to reflect this is a root primary startup and updates are enabled */
assert(!udi->s_addrs.hold_onto_crit || jgbl.onlnrlbk);
was_crit = udi->s_addrs.now_crit;
if (!was_crit)
grab_lock(jnlpool.jnlpool_dummy_reg, TRUE, ASSERT_NO_ONLINE_ROLLBACK);
jnlpool.repl_inst_filehdr->root_primary_cycle++;
/* If this instance is transitioning from a non-rootprimary to rootprimary, switch journal files.
* This helps with maintaining accurate value of csd->zqgblmod_tn when the former primary connects
* to the current primary through a fetchresync-rollback or receiver-server-autorollback..
*/
switch_jnl = (!jnlpool.repl_inst_filehdr->was_rootprimary && (0 < jnlpool.repl_inst_filehdr->num_histinfo));
jnlpool.repl_inst_filehdr->was_rootprimary = TRUE;
assert(start_seqno >= jnlpool.jnlpool_ctl->start_jnl_seqno);
assert(start_seqno == jnlpool.jnlpool_ctl->jnl_seqno);
jnlpool.repl_inst_filehdr->jnl_seqno = start_seqno;
assert(jgbl.onlnrlbk || jnlpool.jnlpool_ctl->upd_disabled);
if (!jgbl.onlnrlbk)
jnlpool.jnlpool_ctl->upd_disabled = FALSE;
if (IS_REPL_INST_UUID_NULL(jnlpool.repl_inst_filehdr->lms_group_info))
{ /* This is the first time this instance is being brought up either as a root primary or as a propagating
* primary. Initialize the "lms_group_info" fields in the instance file header in journal pool shared memory.
* They will be flushed to the instance file as part of the "repl_inst_histinfo_add -> repl_inst_flush_filehdr"
* function invocation below.
*/
assert('\0' == jnlpool.repl_inst_filehdr->lms_group_info.created_nodename[0]);
assert('\0' == jnlpool.repl_inst_filehdr->lms_group_info.this_instname[0]);
assert(!jnlpool.repl_inst_filehdr->lms_group_info.creator_pid);
jnlpool.repl_inst_filehdr->lms_group_info = jnlpool.repl_inst_filehdr->inst_info;
assert('\0' != jnlpool.repl_inst_filehdr->lms_group_info.created_nodename[0]);
DBG_CHECK_CREATED_NODENAME(jnlpool.repl_inst_filehdr->lms_group_info.created_nodename);
assert('\0' != jnlpool.repl_inst_filehdr->lms_group_info.this_instname[0]);
assert(jnlpool.repl_inst_filehdr->lms_group_info.created_time);
assert(jnlpool.repl_inst_filehdr->lms_group_info.creator_pid);
}
/* Initialize histinfo fields */
memcpy(histinfo.root_primary_instname, jnlpool.repl_inst_filehdr->inst_info.this_instname, MAX_INSTNAME_LEN - 1);
histinfo.root_primary_instname[MAX_INSTNAME_LEN - 1] = '\0';
assert('\0' != histinfo.root_primary_instname[0]);
histinfo.start_seqno = start_seqno;
assert(jnlpool.jnlpool_ctl->strm_seqno[0] == jnlpool.repl_inst_filehdr->strm_seqno[0]);
assert(jnlpool.repl_inst_filehdr->is_supplementary || (0 == jnlpool.jnlpool_ctl->strm_seqno[0]));
histinfo.strm_seqno = (!jnlpool.repl_inst_filehdr->is_supplementary) ? 0 : jnlpool.jnlpool_ctl->strm_seqno[0];
histinfo.root_primary_cycle = jnlpool.repl_inst_filehdr->root_primary_cycle;
assert(process_id == getpid());
histinfo.creator_pid = process_id;
JNL_SHORT_TIME(histinfo.created_time);
histinfo.strm_index = 0;
histinfo.history_type = HISTINFO_TYPE_NORMAL;
NULL_INITIALIZE_REPL_INST_UUID(histinfo.lms_group);
/* The following fields will be initialized in the "repl_inst_histinfo_add" function call below.
* histinfo.histinfo_num
* histinfo.prev_histinfo_num
* histinfo.last_histinfo_num[]
*/
/* Add the histinfo record to the instance file and flush the changes in the journal pool to the file header */
repl_inst_histinfo_add(&histinfo);
if (!was_crit)
rel_lock(jnlpool.jnlpool_dummy_reg);
if (switch_jnl)
{
SET_GBL_JREC_TIME; /* jnl_ensure_open/jnl_file_extend and its callees assume jgbl.gbl_jrec_time is set */
for (reg = gd_header->regions, region_top = gd_header->regions + gd_header->n_regions; reg < region_top; reg++)
{
gv_cur_region = reg;
change_reg(); /* sets cs_addrs/cs_data (needed by jnl_ensure_open) */
if (!JNL_ENABLED(cs_addrs))
continue;
grab_crit(gv_cur_region);
jpc = cs_addrs->jnl;
/* 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)
*/
jbp = jpc->jnl_buff;
ADJUST_GBL_JREC_TIME(jgbl, jbp);
jnl_status = jnl_ensure_open();
if (0 == jnl_status)
{
if (EXIT_ERR == SWITCH_JNL_FILE(jpc))
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(4) ERR_JNLEXTEND, 2, JNL_LEN_STR(cs_data));
} else
{
if (SS_NORMAL != jpc->status)
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(7) jnl_status, 4, JNL_LEN_STR(cs_data),
DB_LEN_STR(gv_cur_region), jpc->status);
else
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(cs_data),
DB_LEN_STR(gv_cur_region));
}
rel_crit(gv_cur_region);
}
}
}
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;
}
trans_num gvcst_bmp_mark_free(kill_set *ks)
{
block_id bit_map, next_bm, *updptr;
blk_ident *blk, *blk_top, *nextblk;
trans_num ctn, start_db_fmt_tn;
unsigned int len;
# if defined(UNIX) && defined(DEBUG)
unsigned int lcl_t_tries;
# endif
int4 blk_prev_version;
srch_hist alt_hist;
trans_num ret_tn = 0;
boolean_t visit_blks;
srch_blk_status bmphist;
cache_rec_ptr_t cr;
enum db_ver ondsk_blkver;
enum cdb_sc status;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
TREF(in_gvcst_bmp_mark_free) = TRUE;
assert(inctn_bmp_mark_free_gtm == inctn_opcode || inctn_bmp_mark_free_mu_reorg == inctn_opcode);
/* Note down the desired_db_format_tn before you start relying on cs_data->fully_upgraded.
* If the db is fully_upgraded, take the optimal path that does not need to read each block being freed.
* But in order to detect concurrent desired_db_format changes, note down the tn (when the last format change occurred)
* before the fully_upgraded check and after having noted down the database current_tn.
* If they are the same, then we are guaranteed no concurrent desired_db_format change occurred.
* If they are not, then fall through to the non-optimal path where each to-be-killed block has to be visited.
* The reason we need to visit every block in case desired_db_format changes is to take care of the case where
* MUPIP REORG DOWNGRADE concurrently changes a block that we are about to free.
*/
start_db_fmt_tn = cs_data->desired_db_format_tn;
visit_blks = (!cs_data->fully_upgraded); /* Local evaluation */
assert(!visit_blks || (visit_blks && dba_bg == cs_addrs->hdr->acc_meth)); /* must have blks_to_upgrd == 0 for non-BG */
assert(!dollar_tlevel); /* Should NOT be in TP now */
blk = &ks->blk[0];
blk_top = &ks->blk[ks->used];
if (!visit_blks)
{ /* Database has been completely upgraded. Free all blocks in one bitmap as part of one transaction. */
assert(cs_data->db_got_to_v5_once); /* assert all V4 fmt blocks (including RECYCLED) have space for V5 upgrade */
inctn_detail.blknum_struct.blknum = 0; /* to indicate no adjustment to "blks_to_upgrd" necessary */
/* If any of the mini transaction below restarts because of an online rollback, we don't want the application
* refresh to happen (like $ZONLNRLBK++ or rts_error(DBROLLEDBACK). This is because, although we are currently in
* non-tp (dollar_tleve = 0), we could actually be in a TP transaction and have actually faked dollar_tlevel. In
* such a case, we should NOT * be issuing a DBROLLEDBACK error as TP transactions are supposed to just restart in
* case of an online rollback. So, set the global variable that gtm_onln_rlbk_clnup can check and skip doing the
* application refresh, but will reset the clues. The next update will see the cycle mismatch and will accordingly
* take the right action.
*/
for ( ; blk < blk_top; blk = nextblk)
{
if (0 != blk->flag)
{
nextblk = blk + 1;
continue;
}
assert(0 < blk->block);
assert((int4)blk->block < cs_addrs->ti->total_blks);
bit_map = ROUND_DOWN2((int)blk->block, BLKS_PER_LMAP);
next_bm = bit_map + BLKS_PER_LMAP;
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
/* Scan for the next local bitmap */
updptr = (block_id *)update_array_ptr;
for (nextblk = blk;
(0 == nextblk->flag) && (nextblk < blk_top) && ((block_id)nextblk->block < next_bm);
++nextblk)
{
assert((block_id)nextblk->block - bit_map);
*updptr++ = (block_id)nextblk->block - bit_map;
}
len = (unsigned int)((char *)nextblk - (char *)blk);
update_array_ptr = (char *)updptr;
alt_hist.h[0].blk_num = 0; /* need for calls to T_END for bitmaps */
alt_hist.h[0].blk_target = NULL; /* need to initialize for calls to T_END */
/* the following assumes SIZEOF(blk_ident) == SIZEOF(int) */
assert(SIZEOF(blk_ident) == SIZEOF(int));
*(int *)update_array_ptr = 0;
t_begin(ERR_GVKILLFAIL, UPDTRNS_DB_UPDATED_MASK);
for (;;)
{
ctn = cs_addrs->ti->curr_tn;
/* Need a read fence before reading fields from cs_data as we are reading outside
* of crit and relying on this value to detect desired db format state change.
*/
SHM_READ_MEMORY_BARRIER;
if (start_db_fmt_tn != cs_data->desired_db_format_tn)
{ /* Concurrent db format change has occurred. Need to visit every block to be killed
* to determine its block format. Fall through to the non-optimal path below
*/
ret_tn = 0;
break;
}
bmphist.blk_num = bit_map;
if (NULL == (bmphist.buffaddr = t_qread(bmphist.blk_num, (sm_int_ptr_t)&bmphist.cycle,
&bmphist.cr)))
{
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
t_write_map(&bmphist, (uchar_ptr_t)update_array, ctn, -(int4)(nextblk - blk));
UNIX_ONLY(DEBUG_ONLY(lcl_t_tries = t_tries));
if ((trans_num)0 == (ret_tn = t_end(&alt_hist, NULL, TN_NOT_SPECIFIED)))
{
# ifdef UNIX
assert((CDB_STAGNATE == t_tries) || (lcl_t_tries == t_tries - 1));
status = LAST_RESTART_CODE;
if ((cdb_sc_onln_rlbk1 == status) || (cdb_sc_onln_rlbk2 == status)
|| TREF(rlbk_during_redo_root))
{ /* t_end restarted due to online rollback. Discard bitmap free-up and return control
* to the application. But, before that reset only_reset_clues_if_onln_rlbk to FALSE
*/
TREF(in_gvcst_bmp_mark_free) = FALSE;
send_msg(VARLSTCNT(6) ERR_IGNBMPMRKFREE, 4, REG_LEN_STR(gv_cur_region),
DB_LEN_STR(gv_cur_region));
t_abort(gv_cur_region, cs_addrs);
return ret_tn; /* actually 0 */
}
# endif
continue;
}
break;
}
if (0 == ret_tn) /* db format change occurred. Fall through to below for loop to visit each block */
{
/* Abort any active transaction to get rid of lingering Non-TP artifacts */
t_abort(gv_cur_region, cs_addrs);
break;
}
}
} /* for all blocks in the kill_set */
bool gtcmtr_increment(void)
{
cm_region_list *reg_ref;
mval incr_delta, post_incr;
unsigned char buff[MAX_ZWR_KEY_SZ], *end;
unsigned char *ptr, regnum;
short n;
unsigned short top, len, temp_short;
static readonly gds_file_id file;
error_def(ERR_KEY2BIG);
error_def(ERR_GVIS);
error_def(ERR_DBPRIVERR);
ptr = curr_entry->clb_ptr->mbf;
assert(*ptr == CMMS_Q_INCREMENT);
ptr++;
GET_USHORT(len, ptr);
ptr += SIZEOF(unsigned short);
regnum = *ptr++;
reg_ref = gtcm_find_region(curr_entry,regnum);
len--; /* subtract size of regnum */
CM_GET_GVCURRKEY(ptr, len);
gtcm_bind_name(reg_ref->reghead, TRUE);
if (gv_cur_region->read_only)
rts_error(VARLSTCNT(4) ERR_DBPRIVERR, 2, DB_LEN_STR(gv_cur_region));
if (JNL_ALLOWED(cs_addrs))
{ /* we need to copy client's specific prc_vec into the global variable in order that the gvcst* routines
* do the right job. actually we need to do this only if JNL_ENABLED(cs_addrs), but since it is not
* easy to re-execute the following two assignments in case gvcst_incr's call to t_end encounters a
* cdb_sc_jnlstatemod retry code, we choose the easier approach of executing the following segment
* if JNL_ALLOWED(cs_addrs) is TRUE instead of checking for JNL_ENABLED(cs_addrs) to be TRUE.
* this approach has the overhead that we will be doing the following assignments even though JNL_ENABLED
* might not be TRUE but since the following two are just pointer copies, it is not considered a big overhead.
* this approach ensures that the jnl_put_jrt_pini gets the appropriate prc_vec for writing into the
* journal record in case JNL_ENABLED turns out to be TRUE in t_end time.
* note that the value of JNL_ALLOWED(cs_addrs) cannot be changed on the fly without obtaining standalone access
* and hence the correctness of prc_vec (whenever it turns out necessary) is guaranteed.
*/
originator_prc_vec = curr_entry->pvec;
cs_addrs->jnl->pini_addr = reg_ref->pini_addr;
}
GET_USHORT(len, ptr);
ptr += SIZEOF(unsigned short);
incr_delta.mvtype = MV_STR;
incr_delta.str.len = len;
incr_delta.str.addr = (char *)ptr;
if ((n = gv_currkey->end + 1) > gv_cur_region->max_key_size)
{
if ((end = format_targ_key(&buff[0], MAX_ZWR_KEY_SZ, gv_currkey, TRUE)) == 0)
end = &buff[MAX_ZWR_KEY_SZ - 1];
rts_error(VARLSTCNT(11) ERR_KEY2BIG, 4, n, (int4)gv_cur_region->max_key_size,
REG_LEN_STR(gv_cur_region), 0, ERR_GVIS, 2, end - buff, buff);
}
MV_FORCE_NUMD(&incr_delta);
gvcst_incr(&incr_delta, &post_incr);
if (JNL_ALLOWED(cs_addrs))
reg_ref->pini_addr = cs_addrs->jnl->pini_addr; /* In case journal switch occurred */
ptr = curr_entry->clb_ptr->mbf;
if (MV_DEFINED(&post_incr))
{
temp_short = (unsigned short)post_incr.str.len;
assert((int4)temp_short == post_incr.str.len); /* ushort <- int4 assignment lossy? */
if (curr_entry->clb_ptr->mbl < 1 + /* msg header */
SIZEOF(temp_short) + /* size of length of $INCR return value */
temp_short) /* length of $INCR return value */
{ /* resize buffer */
cmi_realloc_mbf(curr_entry->clb_ptr, 1 + SIZEOF(temp_short) + temp_short);
ptr = curr_entry->clb_ptr->mbf;
}
*ptr++ = CMMS_R_INCREMENT;
PUT_USHORT(ptr, temp_short);
ptr += SIZEOF(unsigned short);
memcpy(ptr, post_incr.str.addr, temp_short);
ptr += temp_short;
} else
