boolean_t gvcst_order(void)
{ /* See gvcst_query.c */
boolean_t found, is_hidden, sn_tpwrapped;
boolean_t est_first_pass;
gv_key save_currkey[DBKEYALLOC(MAX_KEY_SZ)];
int end, prev, oldend;
int save_dollar_tlevel;
DEBUG_ONLY(save_dollar_tlevel = dollar_tlevel);
found = gvcst_order2();
# ifdef UNIX
assert(save_dollar_tlevel == dollar_tlevel);
CHECK_HIDDEN_SUBSCRIPT_AND_RETURN(found, gv_altkey, is_hidden);
assert(found && is_hidden);
IF_SN_DISALLOWED_AND_NO_SPAN_IN_DB(return found);
SAVE_GV_CURRKEY_LAST_SUBSCRIPT(save_currkey, prev, oldend);
if (!dollar_tlevel)
{
sn_tpwrapped = TRUE;
op_tstart((IMPLICIT_TSTART), TRUE, &literal_batch, 0);
ESTABLISH_NORET(gvcst_order_ch, est_first_pass);
GVCST_ROOT_SEARCH_AND_PREP(est_first_pass);
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_order, (gtm_uint64_t) -1);
found = gvcst_order2();
} else
sn_tpwrapped = FALSE;
if (found)
{
CHECK_HIDDEN_SUBSCRIPT(gv_altkey, is_hidden);
if (is_hidden)
{ /* Replace last subscript to be the highest possible hidden subscript so another
* gvcst_order2 will give us the next non-hidden subscript.
*/
REPLACE_HIDDEN_SUB_TO_HIGHEST(gv_altkey, gv_currkey); /* uses gv_altkey to modify gv_currkey */
/* fix up since it should only be externally counted as one $order */
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_order, (gtm_uint64_t) -1);
found = gvcst_order2();
}
}
if (sn_tpwrapped)
{
op_tcommit();
REVERT; /* remove our condition handler */
}
RESTORE_GV_CURRKEY_LAST_SUBSCRIPT(save_currkey, prev, oldend);
assert(save_dollar_tlevel == dollar_tlevel);
# endif
return found;
}
boolean_t gvcst_query(void)
{ /* Similar to gvcst_order and gvcst_zprevious. In each case we skip over hidden subscripts as needed.
*
* 1 2 3 NULL <--- order/zprev...
* 1 2 3 NULL NULL
* 1 2 3 NULL NULL NULL <--- query from here...
* 1 2 3 NULL NULL NULL hidden
* 1 2 3 NULL NULL hidden
* 1 2 3 NULL hidden
* 1 2 3 hidden <--- ... skip this guy and go to bottom/top, respectively
* 1 2 3 7 <--- ... needs to end up here
*/
boolean_t found, is_hidden, sn_tpwrapped;
boolean_t est_first_pass;
gv_key save_currkey[DBKEYALLOC(MAX_KEY_SZ)];
int i;
int save_dollar_tlevel;
DEBUG_ONLY(save_dollar_tlevel = dollar_tlevel);
found = gvcst_query2();
# ifdef UNIX
assert(save_dollar_tlevel == dollar_tlevel);
CHECK_HIDDEN_SUBSCRIPT_AND_RETURN(found, gv_altkey, is_hidden);
IF_SN_DISALLOWED_AND_NO_SPAN_IN_DB(return found);
assert(found && is_hidden);
SAVE_GV_CURRKEY(save_currkey);
if (!dollar_tlevel)
{
sn_tpwrapped = TRUE;
op_tstart((IMPLICIT_TSTART), TRUE, &literal_batch, 0);
ESTABLISH_NORET(gvcst_query_ch, est_first_pass);
GVCST_ROOT_SEARCH_AND_PREP(est_first_pass);
} else
sn_tpwrapped = FALSE;
for (i = 0; i <= MAX_GVSUBSCRIPTS; i++)
{
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_query, (gtm_uint64_t) -1);
found = gvcst_query2();
CHECK_HIDDEN_SUBSCRIPT_AND_BREAK(found, gv_altkey, is_hidden);
assert(found && is_hidden);
/* Replace last subscript to be the highest possible hidden subscript so another
* gvcst_query2 will give us the next non-hidden subscript.
*/
REPLACE_HIDDEN_SUB_TO_HIGHEST(gv_altkey, gv_currkey); /* uses gv_altkey to modify gv_currkey */
}
if (sn_tpwrapped)
{
op_tcommit();
REVERT; /* remove our condition handler */
}
RESTORE_GV_CURRKEY(save_currkey);
assert(save_dollar_tlevel == dollar_tlevel);
# endif
return found;
}
boolean_t gvcst_queryget(mval *val)
{
bool found, is_hidden, is_dummy = FALSE, sn_tpwrapped;
boolean_t est_first_pass;
char save_currkey[SIZEOF(gv_key) + DBKEYSIZE(MAX_KEY_SZ)];
gv_key *save_gv_currkey;
int save_dollar_tlevel;
DEBUG_ONLY(save_dollar_tlevel = dollar_tlevel);
found = gvcst_queryget2(val, NULL);
# ifdef UNIX
assert(save_dollar_tlevel == dollar_tlevel);
CHECK_HIDDEN_SUBSCRIPT(gv_altkey, is_hidden);
if (found && IS_SN_DUMMY(val->str.len, val->str.addr))
is_dummy = TRUE;
if (!found || (!is_dummy && !is_hidden))
return found;
IF_SN_DISALLOWED_AND_NO_SPAN_IN_DB(return found);
SAVE_GV_CURRKEY;
if (!dollar_tlevel)
{
sn_tpwrapped = TRUE;
op_tstart((IMPLICIT_TSTART), TRUE, &literal_batch, 0);
ESTABLISH_NORET(gvcst_queryget_ch, est_first_pass);
GVCST_ROOT_SEARCH_AND_PREP(est_first_pass);
} else
sn_tpwrapped = FALSE;
found = gvcst_query();
COPY_KEY(gv_currkey, gv_altkey); /* set gv_currkey to gv_altkey */
found = gvcst_get(val);
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_get, (gtm_uint64_t) -1); /* only counted externally as one get */
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_query, (gtm_uint64_t) -1);
if (sn_tpwrapped)
{
op_tcommit();
REVERT; /* remove our condition handler */
}
RESTORE_GV_CURRKEY;
assert(save_dollar_tlevel == dollar_tlevel);
# endif
return found;
}
boolean_t gvcst_query2(void)
{
boolean_t found, two_histories;
enum cdb_sc status;
blk_hdr_ptr_t bp;
rec_hdr_ptr_t rp;
unsigned char *c1, *c2;
srch_blk_status *bh;
srch_hist *rt_history;
T_BEGIN_READ_NONTP_OR_TP(ERR_GVQUERYFAIL);
assert(t_tries < CDB_STAGNATE || cs_addrs->now_crit); /* we better hold crit in the final retry (TP & non-TP) */
for (;;)
{
two_histories = FALSE;
# if defined(DEBUG) && defined(UNIX)
if (gtm_white_box_test_case_enabled && (WBTEST_ANTIFREEZE_GVQUERYFAIL == gtm_white_box_test_case_number))
{
t_retry(cdb_sc_blknumerr);
continue;
}
# endif
if (cdb_sc_normal == (status = gvcst_search(gv_currkey, 0)))
{
found = TRUE;
bh = &gv_target->hist.h[0];
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
bp = (blk_hdr_ptr_t)bh->buffaddr;
if (rp >= (rec_hdr_ptr_t)CST_TOB(bp))
{
two_histories = TRUE;
rt_history = gv_target->alt_hist;
status = gvcst_rtsib(rt_history, 0);
if (cdb_sc_endtree == status) /* end of tree */
{
found = FALSE;
two_histories = FALSE; /* second history not valid */
} else if (cdb_sc_normal != status)
{
t_retry(status);
continue;
} else
{
bh = &rt_history->h[0];
if (cdb_sc_normal != (status = gvcst_search_blk(gv_currkey, bh)))
{
t_retry(status);
continue;
}
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
bp = (blk_hdr_ptr_t)bh->buffaddr;
}
}
if (found)
{ /* !found indicates that the end of tree has been reached (see call to
* gvcst_rtsib). If there is no more tree, don't bother doing expansion.
*/
status = gvcst_expand_curr_key(bh, gv_currkey, gv_altkey);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
}
if (!dollar_tlevel)
{
if ((trans_num)0 == t_end(&gv_target->hist, !two_histories ? NULL : rt_history, TN_NOT_SPECIFIED))
continue;
} else
{
status = tp_hist(!two_histories ? NULL : rt_history);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
}
assert(cs_data == cs_addrs->hdr);
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_query, 1);
if (found)
{
c1 = &gv_altkey->base[0];
c2 = &gv_currkey->base[0];
for ( ; *c2; )
{
if (*c2++ != *c1++)
break;
}
if (!*c2 && !*c1)
return TRUE;
}
return FALSE;
}
t_retry(status);
}
}
mint gvcst_data(void)
{
blk_hdr_ptr_t bp;
enum cdb_sc status;
mint val;
rec_hdr_ptr_t rp;
unsigned short rec_size;
srch_blk_status *bh;
srch_hist *rt_history;
sm_uc_ptr_t b_top;
assert((gv_target->root < cs_addrs->ti->total_blks) || (0 < dollar_tlevel));
T_BEGIN_READ_NONTP_OR_TP(ERR_GVDATAFAIL);
assert(t_tries < CDB_STAGNATE || cs_addrs->now_crit); /* we better hold crit in the final retry (TP & non-TP) */
for (;;)
{
rt_history = gv_target->alt_hist;
rt_history->h[0].blk_num = 0;
if ((status = gvcst_search(gv_currkey, NULL)) != cdb_sc_normal)
{
t_retry(status);
continue;
}
bh = gv_target->hist.h;
bp = (blk_hdr_ptr_t)bh->buffaddr;
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
b_top = bh->buffaddr + bp->bsiz;
val = 0;
if (gv_currkey->end + 1 == bh->curr_rec.match)
val = 1;
else if (bh->curr_rec.match >= gv_currkey->end)
val = 10;
if (1 == val || rp == (rec_hdr_ptr_t)b_top)
{
GET_USHORT(rec_size, &rp->rsiz);
if (rp == (rec_hdr_ptr_t)b_top || (sm_uc_ptr_t)rp + rec_size == b_top)
{
if (cdb_sc_endtree != (status = gvcst_rtsib(rt_history, 0)))
{
if ((cdb_sc_normal != status)
|| (cdb_sc_normal != (status = gvcst_search_blk(gv_currkey, rt_history->h))))
{
t_retry(status);
continue;
}
if (rt_history->h[0].curr_rec.match >= gv_currkey->end)
val += 10;
}
} else
{
if ((sm_uc_ptr_t)rp + rec_size > b_top)
{
t_retry(cdb_sc_rmisalign);
continue;
}
rp = (rec_hdr_ptr_t)((sm_uc_ptr_t)rp + rec_size);
if (rp->cmpc >= gv_currkey->end)
val += 10;
}
}
if (0 == dollar_tlevel)
{
if ((trans_num)0 == t_end(&gv_target->hist, 0 == rt_history->h[0].blk_num ? NULL : rt_history))
continue;
} else
{
status = tp_hist(0 == rt_history->h[0].blk_num ? NULL : rt_history);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
}
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_data, 1);
return val;
}
}
boolean_t gvcst_order2(void)
{
blk_hdr_ptr_t bp;
boolean_t found, two_histories;
enum cdb_sc status;
rec_hdr_ptr_t rp;
unsigned short rec_size;
srch_blk_status *bh;
srch_hist *rt_history;
sm_uc_ptr_t c1, c2, ctop, alt_top;
int tmp_cmpc;
T_BEGIN_READ_NONTP_OR_TP(ERR_GVORDERFAIL);
for (;;)
{
assert(t_tries < CDB_STAGNATE || cs_addrs->now_crit); /* we better hold crit in the final retry (TP & non-TP) */
two_histories = FALSE;
#if defined(DEBUG) && defined(UNIX)
if (gtm_white_box_test_case_enabled && (WBTEST_ANTIFREEZE_GVORDERFAIL == gtm_white_box_test_case_number))
{
status = cdb_sc_blknumerr;
t_retry(status);
continue;
}
#endif
if (cdb_sc_normal == (status = gvcst_search(gv_currkey, NULL)))
{
found = TRUE;
bh = gv_target->hist.h;
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
bp = (blk_hdr_ptr_t)bh->buffaddr;
if ((rec_hdr_ptr_t)CST_TOB(bp) <= rp)
{
two_histories = TRUE;
rt_history = gv_target->alt_hist;
status = gvcst_rtsib(rt_history, 0);
if (cdb_sc_normal == status)
{
bh = rt_history->h;
if (cdb_sc_normal != (status = gvcst_search_blk(gv_currkey, bh)))
{
t_retry(status);
continue;
}
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
bp = (blk_hdr_ptr_t)bh->buffaddr;
} else
{
if (cdb_sc_endtree == status)
{
found = FALSE;
two_histories = FALSE; /* second history not valid */
} else
{
t_retry(status);
continue;
}
}
}
if (found)
{
assert(gv_altkey->top == gv_currkey->top);
assert(gv_altkey->top == gv_keysize);
assert(gv_altkey->end < gv_altkey->top);
/* store new subscipt */
c1 = gv_altkey->base;
alt_top = gv_altkey->base + gv_altkey->top - 1;
/* Make alt_top one less than gv_altkey->top to allow double-null at end of a key-name */
/* 4/17/96
* HP compiler bug work-around. The original statement was
* c2 = (unsigned char *)CST_BOK(rp) + bh->curr_rec.match - rp->cmpc;
*
* ...but this was sometimes compiled incorrectly (the lower 4 bits
* of rp->cmpc, sign extended, were subtracted from bh->curr_rec.match).
* I separated out the subtraction of rp->cmpc.
*
* -VTF.
*/
c2 = (sm_uc_ptr_t)CST_BOK(rp) + bh->curr_rec.match;
memcpy(c1, gv_currkey->base, bh->curr_rec.match);
c1 += bh->curr_rec.match;
c2 -= EVAL_CMPC(rp);
GET_USHORT(rec_size, &rp->rsiz);
ctop = (sm_uc_ptr_t)rp + rec_size;
for (;;)
{
if (c2 >= ctop || c1 >= alt_top)
{
assert(CDB_STAGNATE > t_tries);
status = cdb_sc_rmisalign;
goto restart; /* goto needed because of nested FOR loop */
}
if (0 == (*c1++ = *c2++))
{
*c1 = 0;
break;
}
}
gv_altkey->end = c1 - gv_altkey->base;
assert(gv_altkey->end < gv_altkey->top);
}
if (!dollar_tlevel)
{
if ((trans_num)0 == t_end(&gv_target->hist, two_histories ? rt_history : NULL, TN_NOT_SPECIFIED))
continue;
} else
{
status = tp_hist(two_histories ? rt_history : NULL);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
}
assert(cs_data == cs_addrs->hdr);
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_order, 1);
return (found && (bh->curr_rec.match >= gv_currkey->prev));
}
restart: t_retry(status);
}
}
mint gvcst_data(void)
{
blk_hdr_ptr_t bp;
boolean_t do_rtsib;
enum cdb_sc status;
mint val;
rec_hdr_ptr_t rp;
unsigned short match, rsiz;
srch_blk_status *bh;
srch_hist *rt_history;
sm_uc_ptr_t b_top;
assert((gv_target->root < cs_addrs->ti->total_blks) || dollar_tlevel);
T_BEGIN_READ_NONTP_OR_TP(ERR_GVDATAFAIL);
assert(t_tries < CDB_STAGNATE || cs_addrs->now_crit); /* we better hold crit in the final retry (TP & non-TP) */
for (;;)
{
/* The following code is duplicated in gvcst_dataget. Any changes here might need to be reflected there as well */
rt_history = gv_target->alt_hist;
rt_history->h[0].blk_num = 0;
if (cdb_sc_normal != (status = gvcst_search(gv_currkey, NULL)))
{
t_retry(status);
continue;
}
bh = gv_target->hist.h;
bp = (blk_hdr_ptr_t)bh->buffaddr;
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
b_top = bh->buffaddr + bp->bsiz;
match = bh->curr_rec.match;
do_rtsib = FALSE;
if (gv_currkey->end + 1 == match)
{
val = 1;
GET_USHORT(rsiz, &rp->rsiz);
rp = (rec_hdr_ptr_t)((sm_uc_ptr_t)rp + rsiz);
if ((sm_uc_ptr_t)rp > b_top)
{
t_retry(cdb_sc_rmisalign);
continue;
} else if ((sm_uc_ptr_t)rp == b_top)
do_rtsib = TRUE;
else if (rp->cmpc >= gv_currkey->end)
val += 10;
} else if (match >= gv_currkey->end)
val = 10;
else
{
val = 0;
if (rp == (rec_hdr_ptr_t)b_top)
do_rtsib = TRUE;
}
if (do_rtsib && (cdb_sc_endtree != (status = gvcst_rtsib(rt_history, 0))))
{
if ((cdb_sc_normal != status) || (cdb_sc_normal != (status = gvcst_search_blk(gv_currkey, rt_history->h))))
{
t_retry(status);
continue;
}
if (rt_history->h[0].curr_rec.match >= gv_currkey->end)
{
assert(1 >= val);
val += 10;
}
}
if (!dollar_tlevel)
{
if ((trans_num)0 == t_end(&gv_target->hist, 0 == rt_history->h[0].blk_num ? NULL : rt_history,
TN_NOT_SPECIFIED))
continue;
} else
{
status = tp_hist(0 == rt_history->h[0].blk_num ? NULL : rt_history);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
}
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_data, 1);
return val;
}
}
boolean_t gvcst_queryget2(mval *val, unsigned char *sn_ptr)
{
blk_hdr_ptr_t bp;
boolean_t found, two_histories;
enum cdb_sc status;
int rsiz, key_size, data_len;
rec_hdr_ptr_t rp;
srch_blk_status *bh;
srch_hist *rt_history;
unsigned short temp_ushort;
int tmp_cmpc;
DEBUG_ONLY(unsigned char *save_strp = NULL);
T_BEGIN_READ_NONTP_OR_TP(ERR_GVQUERYGETFAIL);
assert((CDB_STAGNATE > t_tries) || cs_addrs->now_crit); /* we better hold crit in the final retry (TP & non-TP) */
for (;;)
{
two_histories = FALSE;
#if defined(DEBUG) && defined(UNIX)
if (gtm_white_box_test_case_enabled && (WBTEST_ANTIFREEZE_GVQUERYGETFAIL == gtm_white_box_test_case_number))
{
status = cdb_sc_blknumerr;
t_retry(status);
continue;
}
#endif
if (cdb_sc_normal == (status = gvcst_search(gv_currkey, 0)))
{
found = TRUE;
bh = &gv_target->hist.h[0];
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
bp = (blk_hdr_ptr_t)bh->buffaddr;
if (rp >= (rec_hdr_ptr_t)CST_TOB(bp))
{
two_histories = TRUE;
rt_history = gv_target->alt_hist;
status = gvcst_rtsib(rt_history, 0);
if (cdb_sc_endtree == status) /* end of tree */
{
found = FALSE;
two_histories = FALSE; /* second history not valid */
} else if (cdb_sc_normal != status)
{
t_retry(status);
continue;
} else
{
bh = &rt_history->h[0];
if (cdb_sc_normal != (status = gvcst_search_blk(gv_currkey, bh)))
{
t_retry(status);
continue;
}
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->curr_rec.offset);
bp = (blk_hdr_ptr_t)bh->buffaddr;
}
}
/* !found indicates that the end of tree has been reached (see call to
* gvcst_rtsib). If there is no more tree, don't bother doing expansion.
*/
if (found)
{
status = gvcst_expand_key((blk_hdr_ptr_t)bh->buffaddr, (int4)((sm_uc_ptr_t)rp - bh->buffaddr),
gv_altkey);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
key_size = gv_altkey->end + 1;
GET_RSIZ(rsiz, rp);
data_len = rsiz + EVAL_CMPC(rp) - SIZEOF(rec_hdr) - key_size;
if (data_len < 0 || (sm_uc_ptr_t)rp + rsiz > (sm_uc_ptr_t)bp + ((blk_hdr_ptr_t)bp)->bsiz)
{
assert(CDB_STAGNATE > t_tries);
t_retry(cdb_sc_rmisalign1);
continue;
}
ENSURE_STP_FREE_SPACE(data_len);
DEBUG_ONLY (
if (!save_strp)
save_strp = stringpool.free);
assert(stringpool.top - stringpool.free >= data_len);
memcpy(stringpool.free, (sm_uc_ptr_t)rp + rsiz - data_len, data_len);
/* Assumption: t_end/tp_hist will never cause stp_gcol() call BYPASSOK */
}
if (!dollar_tlevel)
{
if ((trans_num)0 == t_end(&gv_target->hist, !two_histories ? NULL : rt_history, TN_NOT_SPECIFIED))
continue;
} else
{
status = tp_hist(!two_histories ? NULL : rt_history);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
}
if (found)
{
DEBUG_ONLY(assert(save_strp == stringpool.free));
/* Process val first. Already copied to string pool. */
val->mvtype = MV_STR;
val->str.addr = (char *)stringpool.free;
val->str.len = data_len;
stringpool.free += data_len;
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_get, 1);
}
return found;
}
t_retry(status);
}
bool gvcst_zprevious(void)
{
static gv_key *zprev_temp_key;
static int4 zprev_temp_keysize = 0;
blk_hdr_ptr_t bp;
bool found, two_histories;
enum cdb_sc status;
rec_hdr_ptr_t rp;
unsigned char *c1, *c2, *ctop;
srch_blk_status *bh;
srch_hist *lft_history;
T_BEGIN_READ_NONTP_OR_TP(ERR_GVORDERFAIL);
for (;;)
{
assert(t_tries < CDB_STAGNATE || cs_addrs->now_crit); /* we better hold crit in the final retry (TP & non-TP) */
two_histories = FALSE;
if (cdb_sc_normal == (status = gvcst_search(gv_currkey, NULL)))
{
found = TRUE;
bh = gv_target->hist.h;
if (0 == bh->prev_rec.offset)
{
two_histories = TRUE;
lft_history = gv_target->alt_hist;
status = gvcst_lftsib(lft_history);
if (cdb_sc_normal == status)
{
bh = lft_history->h;
if (cdb_sc_normal != (status = gvcst_search_blk(gv_currkey, bh)))
{
t_retry(status);
continue;
}
} else if (cdb_sc_endtree == status)
{
found = FALSE;
two_histories = FALSE; /* second history not valid */
} else
{
t_retry(status);
continue;
}
}
if (found)
{ /* store new subscipt */
assert(gv_altkey->top == gv_currkey->top);
assert(gv_altkey->top == gv_keysize);
assert(gv_currkey->end < gv_currkey->top);
rp = (rec_hdr_ptr_t)(bh->buffaddr + bh->prev_rec.offset);
bp = (blk_hdr_ptr_t)bh->buffaddr;
c1 = gv_altkey->base;
memcpy(c1, gv_currkey->base, bh->prev_rec.match);
c1 += bh->prev_rec.match;
assert(zprev_temp_keysize <= gv_keysize);
if (zprev_temp_keysize < gv_keysize)
{
zprev_temp_keysize = gv_keysize;
GVKEY_INIT(zprev_temp_key, zprev_temp_keysize);
}
assert(zprev_temp_key->top >= gv_currkey->top);
if (cdb_sc_normal != (status = gvcst_expand_key((blk_hdr_ptr_t)bh->buffaddr, bh->prev_rec.offset,
zprev_temp_key)))
{
t_retry(status);
continue;
}
if ((zprev_temp_key->end < gv_currkey->end) && (zprev_temp_key->end <= gv_currkey->prev))
found = FALSE;
else
{
c2 = zprev_temp_key->base + bh->prev_rec.match;
ctop = zprev_temp_key->base + zprev_temp_key->end;
for (;;)
{
if (c2 >= ctop)
{
assert(CDB_STAGNATE > t_tries);
status = cdb_sc_rmisalign;
goto restart; /* goto needed because of nested FOR loop */
}
if (0 == (*c1++ = *c2++))
{
*c1 = 0;
break;
}
}
}
gv_altkey->end = c1 - gv_altkey->base;
assert(gv_altkey->end < gv_altkey->top);
}
if (!dollar_tlevel)
{
if ((trans_num)0 == t_end(&gv_target->hist, two_histories ? lft_history : NULL, TN_NOT_SPECIFIED))
continue;
} else
{
status = tp_hist(two_histories ? lft_history : NULL);
if (cdb_sc_normal != status)
{
t_retry(status);
continue;
}
}
assert(cs_data == cs_addrs->hdr);
INCR_GVSTATS_COUNTER(cs_addrs, cs_addrs->nl, n_zprev, 1);
return (found && (bh->prev_rec.match >= gv_currkey->prev));
}
restart: t_retry(status);
}
}
uint4 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;
unsigned char hdr_buff[REAL_JNL_HDR_LEN + MAX_IO_BLOCK_SIZE];
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, warn_blocks, result;
gtm_uint64_t avail_blocks;
uint4 aligned_tot_jrec_size, count;
uint4 jnl_fs_block_size, read_write_size;
DCL_THREADGBL_ACCESS;
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(&FILE_INFO(jpc->region)->s_addrs == csa);
assert(csa->jnl_state == csd->jnl_state);
assertpro(JNL_ENABLED(csa) && (NOJNL != jpc->channel) && (!JNL_FILE_SWITCHED(jpc)));
/* crit and messing with the journal file - how could it have vanished? */
if (!csd->jnl_deq || (csd->jnl_alq + csd->jnl_deq > csd->autoswitchlimit))
{
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)))
{
warn_blocks = (csd->jnl_alq + csd->jnl_deq > csd->autoswitchlimit)
? ((csd->jnl_deq > csd->autoswitchlimit) ? csd->jnl_deq : csd->autoswitchlimit)
: new_blocks;
if ((warn_blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
if (new_blocks > avail_blocks)
{ /* If we cannot satisfy the request, it is an error, unless the anticipatory freeze
* scheme is in effect in which case, we will assume space is available even if
* it is not and go ahead with writes to the disk. If the writes fail with ENOSPC
* we will freeze the instance and wait for space to become available and keep
* retrying the writes. Therefore, we make the NOSPACEEXT a warning in this case.
*/
SETUP_THREADGBL_ACCESS;
if (!ANTICIPATORY_FREEZE_ENABLED(csa))
{
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(6) MAKE_MSG_WARNING(ERR_NOSPACEEXT), 4,
JNL_LEN_STR(csd), new_blocks, avail_blocks);
} else
send_msg(VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, JNL_LEN_STR(csd), (avail_blocks - warn_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. Set a global variable in_jnl_file_autoswitch
* so jnl_write can know not to do the padding check. But because this is a global
* variable, we also need to make sure it is reset in case of errors during the
* autoswitch (or else calls to jnl_write after we are out of the autoswitch logic
* will continue to incorrectly not do the padding check. Hence a condition handler.
*/
assert(!in_jnl_file_autoswitch);
in_jnl_file_autoswitch = TRUE;
/* Also make sure time is not changed. This way if "jnl_write" as part of writing a
* journal record invokes jnl_file_extend, when the autoswitch is done and writing
* of the parent jnl_write resumes, we want it to continue with the same timestamp
* and not have to reset its time (non-trivial task) to reflect any changes since then.
*/
assert(!jgbl.save_dont_reset_gbl_jrec_time);
jgbl.save_dont_reset_gbl_jrec_time = jgbl.dont_reset_gbl_jrec_time;
jgbl.dont_reset_gbl_jrec_time = TRUE;
/* Establish a condition handler so we reset a few global variables that have
* temporarily been modified in case of errors inside wcs_flu/jnl_file_close.
*/
ESTABLISH_RET(jnl_file_autoswitch_ch, EXIT_ERR);
/* It is possible we still have not written a PINI record in this journal file
* (e.g. mupip extend saw the need to do jnl_file_extend inside jnl_write while
* trying to write a PINI record). Write a PINI record in that case before closing
* the journal file that way the EOF record will have a non-zero pini_addr.
*/
if (0 == jpc->pini_addr)
jnl_put_jrt_pini(csa);
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SPEEDUP_NOBEFORE);
jnl_file_close(gv_cur_region, TRUE, TRUE);
REVERT;
in_jnl_file_autoswitch = FALSE;
jgbl.dont_reset_gbl_jrec_time = jgbl.save_dont_reset_gbl_jrec_time;
DEBUG_ONLY(jgbl.save_dont_reset_gbl_jrec_time = FALSE);
assert((dba_mm == cs_data->acc_meth) || (csd == cs_data));
csd = cs_data; /* In MM, wcs_flu() can remap an extended DB, so reset csd to be sure */
} else
{
if (SS_NORMAL != jpc->status)
rts_error(VARLSTCNT(7) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region),
jpc->status);
else
rts_error(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
}
assert(!jgbl.forw_phase_recovery || (NULL != jgbl.mur_pini_addr_reset_fnptr));
assert(jgbl.forw_phase_recovery || (NULL == jgbl.mur_pini_addr_reset_fnptr));
if (NULL != jgbl.mur_pini_addr_reset_fnptr)
(*jgbl.mur_pini_addr_reset_fnptr)(csa);
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->jnl_checksum = jnl_info.checksum;
csd->jnl_eovtn = csd->trans_hist.curr_tn;
send_msg(VARLSTCNT(4) ERR_NEWJNLFILECREAT, 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;
fc->op_len = SGMNT_HDR_LEN;
fc->op_pos = 1;
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)
{
if (jpc->status)
rts_error(VARLSTCNT(7) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region),
jpc->status);
else
rts_error(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
}
assert(jb->filesize == csd->jnl_alq);
if (csd->jnl_alq + csd->jnl_deq <= csd->autoswitchlimit)
{
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_JNLNOCREATE, 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 */
jnl_fs_block_size = jb->fs_block_size;
header = (jnl_file_header *)(ROUND_UP2((uintszofptr_t)hdr_buff, jnl_fs_block_size));
read_write_size = ROUND_UP2(REAL_JNL_HDR_LEN, jnl_fs_block_size);
assert((unsigned char *)header + read_write_size <= ARRAYTOP(hdr_buff));
DO_FILE_READ(jpc->channel, 0, header, read_write_size, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
{
assert(FALSE);
rts_error(VARLSTCNT(5) ERR_JNLRDERR, 2, JNL_LEN_STR(csd), jpc->status);
}
assert((header->virtual_size + new_blocks) == new_alq);
jb->filesize = new_alq; /* Actually this is virtual file size blocks */
header->virtual_size = new_alq;
JNL_DO_FILE_WRITE(csa, csd->jnl_file_name, jpc->channel, 0,
header, read_write_size, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
{
assert(FALSE);
rts_error(VARLSTCNT(5) ERR_JNLWRERR, 2, JNL_LEN_STR(csd), jpc->status);
}
}
if (0 >= new_blocks)
break;
}
if (0 < new_blocks)
{
INCR_GVSTATS_COUNTER(csa, csa->nl, n_jnl_extends, 1);
return EXIT_NRM;
}
jpc->status = ERR_JNLREADEOF;
jnl_file_lost(jpc, ERR_JNLEXTEND);
return EXIT_ERR;
}