i8_p que_obj_tail(que_obj_p p_obj)
{
pos_t tail = p_obj->tail - 1;
if (!que_obj_empty(p_obj)) {
if (tail < 0) {
tail = p_obj->blk_num - 1;
}
return BLK_ADDR(tail, p_obj);
} else {
return NULL;
}
}
i8_t que_obj_pop(que_obj_p p_obj)
{
elem_p p_elem = NULL;
if (p_obj->front != p_obj->tail) {
// not empty
p_elem = (elem_p)BLK_ADDR(p_obj->front, p_obj);
p_elem->data_len = 0;
p_elem->status = 0x00;
p_obj->front = (p_obj->front+1)%(p_obj->blk_num);
}
return F1G_OK;
}
/***********************************************************************************************
Input Parameters:
cur_level: Working block's level
d_max_fill: Database fill factor
i_max_fill: Index fill factor
Output Parameters:
blks_created: how many new blocks are created
lvls_increased : How much level is increased
Input/Output Parameters:
gv_target: History of working block
Here it is assumed that i_max_fill or, d_max_fill is strictly less than block size.
Returns:
cdb_sc_normal: if successful
cdb_sc status otherwise
************************************************************************************************/
enum cdb_sc mu_split(int cur_level, int i_max_fill, int d_max_fill, int *blks_created, int *lvls_increased)
{
boolean_t first_copy, new_rtblk_star_only, create_root = FALSE, split_required, insert_in_left;
unsigned char curr_prev_key[MAX_KEY_SZ+1], new_blk1_last_key[MAX_KEY_SZ+1];
unsigned short temp_ushort;
int rec_size, new_ins_keycmpc, tkeycmpc, new_ances_currkeycmpc, old_ances_currkeycmpc;
int tmp_cmpc;
block_index left_index, right_index;
block_offset ins_off, ins_off2;
int level;
int new_ins_keysz, new_ances_currkeysz, new_blk1_last_keysz, newblk2_first_keysz, next_gv_currkeysz;
int old_ances_currkeylen, new_ins_keylen, new_ances_currkeylen, tkeylen, newblk2_first_keylen;
int old_blk1_last_rec_size, old_blk1_sz, save_blk_piece_len, old_right_piece_len;
int delta, max_fill;
enum cdb_sc status;
int blk_seg_cnt, blk_size, new_leftblk_top_off;
block_id allocation_clue;
sm_uc_ptr_t rPtr1, rPtr2, rec_base, key_base, next_gv_currkey,
bn_ptr1, bn_ptr2, save_blk_piece,
old_blk_after_currec, ances_currkey,
old_blk1_base,
new_blk1_top, new_blk2_top,
new_blk2_frec_base, new_blk2_rem,
newblk2_first_key, new_ins_key;
blk_segment *bs_ptr1, *bs_ptr2;
cw_set_element *cse;
rec_hdr_ptr_t star_rec_hdr, new_rec_hdr1a, new_rec_hdr1b, new_rec_hdr2, root_hdr;
blk_hdr_ptr_t blk_hdr_ptr;
blk_size = cs_data->blk_size;
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
BLK_ADDR(star_rec_hdr, SIZEOF(rec_hdr), rec_hdr);
star_rec_hdr->rsiz = BSTAR_REC_SIZE;
SET_CMPC(star_rec_hdr, 0);
level = cur_level;
max_fill = (0 == level)? d_max_fill : i_max_fill;
/* -------------------
* Split working block.
* -------------------
* new_blk1_last_key = last key of the new working block after split
* new_blk1_last_keysz = size of new_blk1_last_key
* old_blk1_last_rec_size = last record size of the new working block after split (for old block)
* new_blk2_frec_base = base of first record of right block created after split
* newblk2_first_key = first key of new block created after split
* newblk2_first_keysz = size of newblk2_first_key
* new_blk2_rem = pointer to new block to be created after split exclude 1st record header + key
*/
blk_hdr_ptr = (blk_hdr_ptr_t)(gv_target->hist.h[level].buffaddr);
old_blk1_base = (sm_uc_ptr_t)blk_hdr_ptr;
old_blk1_sz = blk_hdr_ptr->bsiz;
new_blk2_top = old_blk1_base + old_blk1_sz;
if (cdb_sc_normal != (status = locate_block_split_point (old_blk1_base, level, old_blk1_sz, max_fill,
&old_blk1_last_rec_size, new_blk1_last_key, &new_blk1_last_keysz, &new_leftblk_top_off)))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
if (new_leftblk_top_off + BSTAR_REC_SIZE >= old_blk1_sz)
/* Avoid split to create a small right sibling. Note this should not happen often when tolerance is high */
return cdb_sc_oprnotneeded;
old_right_piece_len = old_blk1_sz - new_leftblk_top_off;
new_blk2_frec_base = old_blk1_base + new_leftblk_top_off;
BLK_ADDR(newblk2_first_key, gv_cur_region->max_rec_size + 1, unsigned char);
READ_RECORD(level, new_blk2_frec_base, tkeycmpc, rec_size, newblk2_first_key, newblk2_first_keylen, status);
if (cdb_sc_normal != status) /* restart for cdb_sc_starrecord too, because we eliminated the possibility already */
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
memcpy(newblk2_first_key, &new_blk1_last_key[0], tkeycmpc); /* copy the compressed key piece */
new_blk2_rem = new_blk2_frec_base + SIZEOF(rec_hdr) + newblk2_first_keylen;
newblk2_first_keysz = newblk2_first_keylen + tkeycmpc;
/* gv_currkey_next_reorg will be saved for next iteration in mu_reorg */
next_gv_currkey = newblk2_first_key;
next_gv_currkeysz = newblk2_first_keysz;
BLK_ADDR(new_rec_hdr1b, SIZEOF(rec_hdr), rec_hdr);
new_rec_hdr1b->rsiz = rec_size + tkeycmpc;
SET_CMPC(new_rec_hdr1b, 0);
/* Create new split piece, we already know that this will not be *-rec only.
* Note that this has to be done BEFORE modifying working block as building this buffer relies on the
* working block to be pinned which is possible only if this cw-set-element is created ahead of that
* of the working block (since order in which blocks are built is the order in which cses are created).
*/
BLK_INIT(bs_ptr2, bs_ptr1);
BLK_SEG(bs_ptr2, (sm_uc_ptr_t)new_rec_hdr1b, SIZEOF(rec_hdr));
BLK_SEG(bs_ptr2, newblk2_first_key, newblk2_first_keysz);
BLK_SEG(bs_ptr2, new_blk2_rem, new_blk2_top - new_blk2_rem);
if (!BLK_FINI(bs_ptr2, bs_ptr1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
allocation_clue = ALLOCATION_CLUE(cs_data->trans_hist.total_blks);
right_index = t_create(allocation_clue++, (unsigned char *)bs_ptr1, 0, 0, level);
(*blks_created)++;
/* Modify working block removing split piece */
BLK_INIT(bs_ptr2, bs_ptr1);
if (0 == level)
{
BLK_SEG(bs_ptr2, old_blk1_base + SIZEOF(blk_hdr), new_leftblk_top_off - SIZEOF(blk_hdr));
}
else
{
BLK_SEG(bs_ptr2, old_blk1_base + SIZEOF(blk_hdr),
new_leftblk_top_off - SIZEOF(blk_hdr) - old_blk1_last_rec_size);
BLK_SEG(bs_ptr2, (sm_uc_ptr_t)star_rec_hdr, SIZEOF(rec_hdr) );
BLK_ADDR(bn_ptr1, SIZEOF(block_id), unsigned char);
memcpy(bn_ptr1, old_blk1_base + new_leftblk_top_off - SIZEOF(block_id), SIZEOF(block_id));
BLK_SEG(bs_ptr2, bn_ptr1, SIZEOF(block_id));
}
if ( !BLK_FINI(bs_ptr2, bs_ptr1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
t_write(&gv_target->hist.h[level], (unsigned char *)bs_ptr1, 0, 0, level, FALSE, TRUE, GDS_WRITE_KILLTN);
/*
----------------------------------------------------------------------------
Modify ancestor block for the split in current level.
new_ins_key = new key to be inserted in parent because of split in child
new_ins_key will be inserted after gv_target->hist.h[level].prev_rec and
before gv_target->hist.h[level].curr_rec
new_ins_keysz = size of new_ins_key
Note: A restriction of the algorithm is to have current key and new_ins_key
in the same block, either left or, new right block
----------------------------------------------------------------------------
*/
BLK_ADDR(new_ins_key, new_blk1_last_keysz, unsigned char);
memcpy(new_ins_key, &new_blk1_last_key[0], new_blk1_last_keysz);
new_ins_keysz = new_blk1_last_keysz;
for(;;) /* ========== loop through ancestors as necessary ======= */
{
level ++;
max_fill = i_max_fill;
/*
old_blk_after_currec = remaining of current block after currec
ances_currkey = old real value of currkey in ancestor block
*/
blk_hdr_ptr = (blk_hdr_ptr_t)(gv_target->hist.h[level].buffaddr);
old_blk1_base = (sm_uc_ptr_t)blk_hdr_ptr;
old_blk1_sz = blk_hdr_ptr->bsiz;
new_blk2_top = old_blk1_base + old_blk1_sz;
rec_base = old_blk1_base + gv_target->hist.h[level].curr_rec.offset;
GET_RSIZ(rec_size, rec_base);
old_blk_after_currec = rec_base + rec_size;
old_ances_currkeycmpc = EVAL_CMPC((rec_hdr_ptr_t)rec_base);
old_ances_currkeylen = rec_size - BSTAR_REC_SIZE;
if (INVALID_RECORD(level, rec_size, old_ances_currkeylen, old_ances_currkeycmpc))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
if (0 == old_ances_currkeylen)
{
if (0 != old_ances_currkeycmpc)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
new_ances_currkeycmpc = new_ances_currkeylen = 0;
}
else
{
BLK_ADDR(ances_currkey, gv_cur_region->max_rec_size + 1, unsigned char);
key_base = rec_base + SIZEOF(rec_hdr);
}
new_ances_currkeysz = old_ances_currkeycmpc + old_ances_currkeylen;
if (SIZEOF(blk_hdr) != gv_target->hist.h[level].curr_rec.offset) /* cur_rec is not first key */
{
if (cdb_sc_normal != (status = gvcst_expand_any_key(old_blk1_base,
old_blk1_base + gv_target->hist.h[level].curr_rec.offset,
&curr_prev_key[0], &rec_size, &tkeylen, &tkeycmpc, NULL)))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
if (old_ances_currkeycmpc)
memcpy(ances_currkey, &curr_prev_key[0], old_ances_currkeycmpc);
}
if (old_ances_currkeylen)
{
memcpy(ances_currkey + old_ances_currkeycmpc, key_base, old_ances_currkeylen);
GET_CMPC(new_ances_currkeycmpc, new_ins_key, ances_currkey);
new_ances_currkeylen = new_ances_currkeysz - new_ances_currkeycmpc;
}
if (SIZEOF(blk_hdr) != gv_target->hist.h[level].curr_rec.offset)
{
/* new_ins_key will be inseted after curr_prev_key */
GET_CMPC(new_ins_keycmpc, &curr_prev_key[0], new_ins_key);
}
else
new_ins_keycmpc = 0; /* new_ins_key will be the 1st key */
new_ins_keylen = new_ins_keysz - new_ins_keycmpc ;
delta = BSTAR_REC_SIZE + new_ins_keylen - old_ances_currkeylen + new_ances_currkeylen;
if (old_blk1_sz + delta > blk_size - cs_data->reserved_bytes) /* split required */
{
split_required = TRUE;
if (level == gv_target->hist.depth)
{
create_root = TRUE;
if (MAX_BT_DEPTH - 1 <= level) /* maximum level reached */
return cdb_sc_maxlvl;
}
if (max_fill + BSTAR_REC_SIZE > old_blk1_sz)
{
if (SIZEOF(blk_hdr) + BSTAR_REC_SIZE == old_blk1_sz)
return cdb_sc_oprnotneeded; /* Improve code to avoid this */
max_fill = old_blk1_sz - BSTAR_REC_SIZE;
}
status = locate_block_split_point(old_blk1_base, level, old_blk1_sz, max_fill,
&old_blk1_last_rec_size, new_blk1_last_key, &new_blk1_last_keysz, &new_leftblk_top_off);
if (cdb_sc_normal != status || new_leftblk_top_off >= old_blk1_sz
|| 0 == new_blk1_last_keysz)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
assert(BSTAR_REC_SIZE != old_blk1_last_rec_size);
old_right_piece_len = old_blk1_sz - new_leftblk_top_off;
new_blk2_frec_base = new_blk1_top = old_blk1_base + new_leftblk_top_off;
if (BSTAR_REC_SIZE == old_right_piece_len)
new_rtblk_star_only = TRUE;
else
new_rtblk_star_only = FALSE;
if (new_leftblk_top_off == gv_target->hist.h[level].curr_rec.offset)
{
/* inserted key will be the first record of new right block */
new_ins_keylen = new_ins_keysz;
new_ins_keycmpc = 0;
}
else
/* process 1st record of new right block */
{
BLK_ADDR(newblk2_first_key, gv_cur_region->max_rec_size + 1, unsigned char);
READ_RECORD(level, new_blk2_frec_base, tkeycmpc, rec_size,
newblk2_first_key, newblk2_first_keylen, status);
if (cdb_sc_normal == status)
{
memcpy(newblk2_first_key, &new_blk1_last_key[0], tkeycmpc); /* compressed piece */
new_blk2_rem = new_blk2_frec_base + SIZEOF(rec_hdr) + newblk2_first_keylen;
newblk2_first_keysz = newblk2_first_keylen + tkeycmpc;
BLK_ADDR(new_rec_hdr2, SIZEOF(rec_hdr), rec_hdr);
new_rec_hdr2->rsiz = newblk2_first_keysz + BSTAR_REC_SIZE;
SET_CMPC(new_rec_hdr2, 0);
}
else if (cdb_sc_starrecord != status || !new_rtblk_star_only)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
}
/* else gv_target->hist.h[level].curr_rec will be newblk2_first_key */
if (new_leftblk_top_off > gv_target->hist.h[level].curr_rec.offset +
old_ances_currkeylen + BSTAR_REC_SIZE)
{
/* in this case prev_rec (if exists), new key and curr_rec should go into left block */
if (new_leftblk_top_off + delta - old_blk1_last_rec_size + BSTAR_REC_SIZE
<= blk_size - cs_data->reserved_bytes)
insert_in_left = TRUE;
else
{
/* cannot handle it now */
return cdb_sc_oprnotneeded;
}
}
else if (new_leftblk_top_off < gv_target->hist.h[level].curr_rec.offset +
old_ances_currkeylen + BSTAR_REC_SIZE)
{
/* if gv_target->hist.h[level].curr_rec is the first key in old_blk1
then in new right block,
new_ins_key will be the 1st record key and
curr_rec will be 2nd record and
there will be no prev_rec in right block.
Else (if curr_rec is not first key)
there will be some records before new_ins_key, at least prev_rec */
delta = (int)(BSTAR_REC_SIZE + new_ins_keylen
- old_ances_currkeylen + new_ances_currkeylen
+ ((0 == new_ins_keycmpc) ? 0 : (EVAL_CMPC((rec_hdr_ptr_t)new_blk2_frec_base))));
if (SIZEOF(blk_hdr) + old_right_piece_len + delta <= blk_size - cs_data->reserved_bytes)
{
insert_in_left = FALSE;
if (new_leftblk_top_off + BSTAR_REC_SIZE >= old_blk1_sz)
{
/* cannot handle it now */
return cdb_sc_oprnotneeded;
}
}
else
{
/* cannot handle it now */
return cdb_sc_oprnotneeded;
}
}
else
{
/* in this case prev_rec (if exists), new key and curr_rec should go into left block
and curr_rec will be the last record (*-key) of left new block */
delta = BSTAR_REC_SIZE + new_ins_keylen;
if (new_leftblk_top_off + delta <= blk_size - cs_data->reserved_bytes)
insert_in_left = TRUE;
else
{
/* cannot handle it now */
return cdb_sc_oprnotneeded;
}
}
} /* end if split required */
else
enum cdb_sc gvcst_kill_blk(srch_blk_status *blkhist,
char level,
gv_key *search_key,
srch_rec_status low,
srch_rec_status high,
boolean_t right_extra,
cw_set_element **cseptr)
{
typedef sm_uc_ptr_t bytptr;
unsigned short temp_ushort;
int4 temp_long;
int tmp_cmpc;
int blk_size, blk_seg_cnt, lmatch, rmatch, targ_len, prev_len, targ_base, next_rec_shrink,
temp_int, blkseglen;
bool kill_root, first_copy;
blk_hdr_ptr_t old_blk_hdr;
rec_hdr_ptr_t left_ptr; /*pointer to record before first record to delete*/
rec_hdr_ptr_t del_ptr; /*pointer to first record to delete*/
rec_hdr_ptr_t right_ptr; /*pointer to record after last record to delete*/
rec_hdr_ptr_t right_prev_ptr;
rec_hdr_ptr_t rp, rp1; /*scratch record pointer*/
rec_hdr_ptr_t first_in_blk, top_of_block, new_rec_hdr, star_rec_hdr;
blk_segment *bs1, *bs_ptr;
block_index new_block_index;
unsigned char *skb;
static readonly block_id zeroes = 0;
cw_set_element *cse, *old_cse;
bytptr curr, prev, right_bytptr;
off_chain chain1, curr_chain, prev_chain;
block_id blk;
sm_uc_ptr_t buffer;
srch_blk_status *t1;
*cseptr = NULL;
if (low.offset == high.offset)
return cdb_sc_normal;
blk = blkhist->blk_num;
if (dollar_tlevel)
{
PUT_LONG(&chain1, blk);
if ((1 == chain1.flag) && ((int)chain1.cw_index >= sgm_info_ptr->cw_set_depth))
{
assert(sgm_info_ptr->tp_csa == cs_addrs);
assert(FALSE == cs_addrs->now_crit);
return cdb_sc_blknumerr;
}
}
buffer = blkhist->buffaddr;
old_blk_hdr = (blk_hdr_ptr_t)buffer;
kill_root = FALSE;
blk_size = cs_data->blk_size;
first_in_blk = (rec_hdr_ptr_t)((bytptr)old_blk_hdr + SIZEOF(blk_hdr));
top_of_block = (rec_hdr_ptr_t)((bytptr)old_blk_hdr + old_blk_hdr->bsiz);
left_ptr = (rec_hdr_ptr_t)((bytptr)old_blk_hdr + low.offset);
right_ptr = (rec_hdr_ptr_t)((bytptr)old_blk_hdr + high.offset);
if (right_extra && right_ptr < top_of_block)
{
right_prev_ptr = right_ptr;
GET_USHORT(temp_ushort, &right_ptr->rsiz);
right_ptr = (rec_hdr_ptr_t)((bytptr)right_ptr + temp_ushort);
}
if ((bytptr)left_ptr < (bytptr)old_blk_hdr ||
(bytptr)right_ptr > (bytptr)top_of_block ||
(bytptr)left_ptr >= (bytptr)right_ptr)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_rmisalign;
}
if ((bytptr)left_ptr == (bytptr)old_blk_hdr)
{
if ((bytptr)right_ptr == (bytptr)top_of_block)
{
if ((bytptr)first_in_blk == (bytptr)top_of_block)
{
if (0 != level)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_rmisalign;
}
return cdb_sc_normal;
}
if (!gv_target->hist.h[level + 1].blk_num)
kill_root = TRUE;
else
{ /* We are about to free up the contents of this entire block. If this block corresponded to
* a global that has NOISOLATION turned on and has a non-zero recompute list (i.e. some SETs
* already happened in this same TP transaction), make sure we disable the NOISOLATION
* optimization in this case as that is applicable only if one or more SETs happened in this
* data block and NOT if a KILL happens. Usually this is done by a t_write(GDS_WRITE_KILLTN)
* call but since in this case the entire block is being freed, "t_write" wont be invoked
* so we need to explicitly set GDS_WRITE_KILLTN like t_write would have (GTM-8269).
* Note: blkhist->first_tp_srch_status is not reliable outside of TP. Thankfully the recompute
* list is also maintained only in case of TP so a check of dollar_tlevel is enough to
* dereference both "first_tp_srch_status" and "recompute_list_head".
*/
if (dollar_tlevel)
{
t1 = blkhist->first_tp_srch_status ? blkhist->first_tp_srch_status : blkhist;
cse = t1->cse;
if ((NULL != cse) && cse->recompute_list_head)
cse->write_type |= GDS_WRITE_KILLTN;
}
return cdb_sc_delete_parent;
}
}
del_ptr = first_in_blk;
} else
{
GET_USHORT(temp_ushort, &left_ptr->rsiz);
del_ptr = (rec_hdr_ptr_t)((bytptr)left_ptr + temp_ushort);
if ((bytptr)del_ptr <= (bytptr)(left_ptr + 1) || (bytptr)del_ptr > (bytptr)right_ptr)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_rmisalign;
}
}
if ((bytptr)del_ptr == (bytptr)right_ptr)
return cdb_sc_normal;
lmatch = low.match;
rmatch = high.match;
if (level)
{
for (rp = del_ptr ; rp < right_ptr ; rp = rp1)
{
GET_USHORT(temp_ushort, &rp->rsiz);
rp1 = (rec_hdr_ptr_t)((bytptr)rp + temp_ushort);
if (((bytptr)rp1 < (bytptr)(rp + 1) + SIZEOF(block_id)) ||
((bytptr)rp1 < buffer) || ((bytptr)rp1 > (buffer + blk_size)))
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_rmisalign;
}
GET_LONG(temp_long, ((bytptr)rp1 - SIZEOF(block_id)));
if (dollar_tlevel)
{
chain1 = *(off_chain *)&temp_long;
if ((1 == chain1.flag) && ((int)chain1.cw_index >= sgm_info_ptr->cw_set_depth))
{
assert(sgm_info_ptr->tp_csa == cs_addrs);
assert(FALSE == cs_addrs->now_crit);
return cdb_sc_blknumerr;
}
}
gvcst_delete_blk(temp_long, level - 1, FALSE);
}
}
if (kill_root)
{ /* create an empty data block */
BLK_INIT(bs_ptr, bs1);
if (!BLK_FINI(bs_ptr, bs1))
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_mkblk;
}
new_block_index = t_create(blk, (uchar_ptr_t)bs1, 0, 0, 0);
/* create index block */
BLK_ADDR(new_rec_hdr, SIZEOF(rec_hdr), rec_hdr);
new_rec_hdr->rsiz = SIZEOF(rec_hdr) + SIZEOF(block_id);
SET_CMPC(new_rec_hdr, 0);
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, (bytptr)new_rec_hdr, SIZEOF(rec_hdr));
BLK_SEG(bs_ptr, (bytptr)&zeroes, SIZEOF(block_id));
if (!BLK_FINI(bs_ptr, bs1))
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_mkblk;
}
cse = t_write(blkhist, (unsigned char *)bs1, SIZEOF(blk_hdr) + SIZEOF(rec_hdr), new_block_index, 1,
TRUE, FALSE, GDS_WRITE_KILLTN);
assert(!dollar_tlevel || !cse->high_tlevel);
*cseptr = cse;
if (NULL != cse)
cse->first_off = 0;
return cdb_sc_normal;
}
next_rec_shrink = (int)(old_blk_hdr->bsiz + ((bytptr)del_ptr - (bytptr)right_ptr));
if (SIZEOF(blk_hdr) >= next_rec_shrink)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_rmisalign;
}
if ((bytptr)right_ptr == (bytptr)top_of_block)
{
if (level)
{
GET_USHORT(temp_ushort, &left_ptr->rsiz);
next_rec_shrink += SIZEOF(rec_hdr) + SIZEOF(block_id) - temp_ushort;
}
} else
{
targ_base = (rmatch < lmatch) ? rmatch : lmatch;
prev_len = 0;
if (right_extra)
{
EVAL_CMPC2(right_prev_ptr, tmp_cmpc);
targ_len = tmp_cmpc - targ_base;
if (targ_len < 0)
targ_len = 0;
temp_int = tmp_cmpc - EVAL_CMPC(right_ptr);
if (0 >= temp_int)
prev_len = - temp_int;
else
{
if (temp_int < targ_len)
targ_len -= temp_int;
else
targ_len = 0;
}
} else
{
targ_len = EVAL_CMPC(right_ptr) - targ_base;
if (targ_len < 0)
targ_len = 0;
}
next_rec_shrink += targ_len + prev_len;
}
BLK_INIT(bs_ptr, bs1);
first_copy = TRUE;
blkseglen = (int)((bytptr)del_ptr - (bytptr)first_in_blk);
if (0 < blkseglen)
{
if (((bytptr)right_ptr != (bytptr)top_of_block) || (0 == level))
{
BLK_SEG(bs_ptr, (bytptr)first_in_blk, blkseglen);
first_copy = FALSE;
} else
{
blkseglen = (int)((bytptr)left_ptr - (bytptr)first_in_blk);
if (0 < blkseglen)
{
BLK_SEG(bs_ptr, (bytptr)first_in_blk, blkseglen);
first_copy = FALSE;
}
BLK_ADDR(star_rec_hdr, SIZEOF(rec_hdr), rec_hdr);
SET_CMPC(star_rec_hdr, 0);
star_rec_hdr->rsiz = (unsigned short)(SIZEOF(rec_hdr) + SIZEOF(block_id));
BLK_SEG(bs_ptr, (bytptr)star_rec_hdr, SIZEOF(rec_hdr));
GET_USHORT(temp_ushort, &left_ptr->rsiz);
BLK_SEG(bs_ptr, ((bytptr)left_ptr + temp_ushort - SIZEOF(block_id)), SIZEOF(block_id));
}
}
blkseglen = (int)((bytptr)top_of_block - (bytptr)right_ptr);
assert(0 <= blkseglen);
if (0 != blkseglen)
{
next_rec_shrink = targ_len + prev_len;
if (0 >= next_rec_shrink)
{
BLK_SEG(bs_ptr, (bytptr)right_ptr, blkseglen);
} else
{
BLK_ADDR(new_rec_hdr, SIZEOF(rec_hdr), rec_hdr);
SET_CMPC(new_rec_hdr, EVAL_CMPC(right_ptr) - next_rec_shrink);
GET_USHORT(temp_ushort, &right_ptr->rsiz);
new_rec_hdr->rsiz = temp_ushort + next_rec_shrink;
BLK_SEG(bs_ptr, (bytptr)new_rec_hdr, SIZEOF(rec_hdr));
if (targ_len)
{
BLK_ADDR(skb, targ_len, unsigned char);
memcpy(skb, &search_key->base[targ_base], targ_len);
BLK_SEG(bs_ptr, skb, targ_len);
}
if (prev_len)
BLK_SEG(bs_ptr, (bytptr)(right_prev_ptr + 1) , prev_len);
right_bytptr = (bytptr)(right_ptr + 1);
blkseglen = (int)((bytptr)top_of_block - right_bytptr);
if (0 < blkseglen)
{
BLK_SEG(bs_ptr, right_bytptr, blkseglen);
} else
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_rmisalign;
}
}
}
if (!BLK_FINI(bs_ptr, bs1))
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_mkblk;
}
cse = t_write(blkhist, (unsigned char *)bs1, 0, 0, level, first_copy, TRUE, GDS_WRITE_KILLTN);
assert(!dollar_tlevel || !cse->high_tlevel);
*cseptr = cse;
if (horiz_growth)
{
old_cse = cse->low_tlevel;
assert(old_cse && old_cse->done);
assert(2 == (SIZEOF(old_cse->undo_offset) / SIZEOF(old_cse->undo_offset[0])));
assert(2 == (SIZEOF(old_cse->undo_next_off) / SIZEOF(old_cse->undo_next_off[0])));
assert(!old_cse->undo_next_off[0] && !old_cse->undo_offset[0]);
assert(!old_cse->undo_next_off[1] && !old_cse->undo_offset[1]);
}
if ((NULL != cse) && (0 != cse->first_off))
{ /* fix up chains in the block to account for deleted records */
prev = NULL;
curr = buffer + cse->first_off;
GET_LONGP(&curr_chain, curr);
while (curr < (bytptr)del_ptr)
{ /* follow chain to first deleted record */
if (0 == curr_chain.next_off)
break;
if (right_ptr == top_of_block && (bytptr)del_ptr - curr == SIZEOF(off_chain))
break; /* special case described below: stop just before the first deleted record */
prev = curr;
curr += curr_chain.next_off;
GET_LONGP(&curr_chain, curr);
}
if (right_ptr == top_of_block && (bytptr)del_ptr - curr == SIZEOF(off_chain))
{
/* if the right side of the block is gone and our last chain is in the last record,
* terminate the chain and adjust the previous entry to point at the new *-key
* NOTE: this assumes there's NEVER a TP delete of records in the GVT
*/
assert(0 != level);
/* store next_off in old_cse before actually changing it in the buffer(for rolling back) */
if (horiz_growth)
{
old_cse->undo_next_off[0] = curr_chain.next_off;
old_cse->undo_offset[0] = (block_offset)(curr - buffer);
assert(old_cse->undo_offset[0]);
}
curr_chain.next_off = 0;
GET_LONGP(curr, &curr_chain);
if (NULL != prev)
{ /* adjust previous chain next_off to reflect the fact that the record it refers to is now a *-key */
GET_LONGP(&prev_chain, prev);
/* store next_off in old_cse before actually changing it in the buffer(for rolling back) */
if (horiz_growth)
{
old_cse->undo_next_off[1] = prev_chain.next_off;
old_cse->undo_offset[1] = (block_offset)(prev - buffer);
assert(old_cse->undo_offset[1]);
}
prev_chain.next_off = (unsigned int)((bytptr)left_ptr - prev + (unsigned int)(SIZEOF(rec_hdr)));
GET_LONGP(prev, &prev_chain);
} else /* it's the first (and only) one */
cse->first_off = (block_offset)((bytptr)left_ptr - buffer + SIZEOF(rec_hdr));
} else if (curr >= (bytptr)del_ptr)
{ /* may be more records on the right that aren't deleted */
while (curr < (bytptr)right_ptr)
{ /* follow chain past last deleted record */
if (0 == curr_chain.next_off)
break;
curr += curr_chain.next_off;
GET_LONGP(&curr_chain, curr);
}
/* prev : ptr to chain record immediately preceding the deleted area,
* or 0 if none.
*
* curr : ptr to chain record immediately following the deleted area,
* or to last chain record.
*/
if (curr < (bytptr)right_ptr)
{ /* the former end of the chain is going, going, gone */
if (NULL != prev)
{ /* terminate the chain before the delete */
GET_LONGP(&prev_chain, prev);
/* store next_off in old_cse before actually changing it in the buffer(for rolling back) */
if (horiz_growth)
{
old_cse->undo_next_off[0] = prev_chain.next_off;
old_cse->undo_offset[0] = (block_offset)(prev - buffer);
assert(old_cse->undo_offset[0]);
}
prev_chain.next_off = 0;
GET_LONGP(prev, &prev_chain);
} else
cse->first_off = 0; /* the whole chain is gone */
} else
{ /* stitch up the left and right to account for the hole in the middle */
/* next_rec_shrink is the change in record size due to the new compression count */
if (NULL != prev)
{
GET_LONGP(&prev_chain, prev);
/* ??? new compression may be less (ie +) so why are negative shrinks ignored? */
/* store next_off in old_cse before actually changing it in the buffer(for rolling back) */
if (horiz_growth)
{
old_cse->undo_next_off[0] = prev_chain.next_off;
old_cse->undo_offset[0] = (block_offset)(prev - buffer);
assert(old_cse->undo_offset[0]);
}
prev_chain.next_off = (unsigned int)(curr - prev - ((bytptr)right_ptr - (bytptr)del_ptr)
+ (next_rec_shrink > 0 ? next_rec_shrink : 0));
GET_LONGP(prev, &prev_chain);
} else /* curr remains first: adjust the head */
cse->first_off = (block_offset)(curr - buffer - ((bytptr)right_ptr - (bytptr)del_ptr)
+ (next_rec_shrink > 0 ? next_rec_shrink : 0));
}
}
}
horiz_growth = FALSE;
return cdb_sc_normal;
}
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() */
/*************************************************************************************************
Input Parameters:
gv_target: working block's history
level : Level of working block and its right sibling
d_blk_fill_size : Maximum fill allowed in a data block
i_blk_fill_size : Maximum fill allowed in an index block
Output Parameters:
kill_set_ptr : List of blocks to be freed from LBM (already killed in mu_clsce)
remove_rtsib : if right sibling was completely merged with working
Returns:
cdb_sc_normal on success
Other wise error status
*************************************************************************************************/
enum cdb_sc mu_clsce(int level, int i_max_fill, int d_max_fill, kill_set *kill_set_ptr,
boolean_t *remove_rtsib)
{
boolean_t complete_merge = FALSE,
old_ref_star_only = FALSE,
new_rtsib_star_only = FALSE,
star_only_merge = FALSE,
blk2_ances_star_only = FALSE,
delete_all_blk2_ances = TRUE,
levelp_next_is_star, forward_process;
unsigned char oldblk1_prev_key[MAX_KEY_SZ+1],
old_levelp_cur_prev_key[MAX_KEY_SZ+1],
old_levelp_cur_key[MAX_KEY_SZ+1]; /* keys in private memory */
unsigned short temp_ushort;
int new_levelp_cur_cmpc, new_levelp_cur_next_cmpc, tkeycmpc,
oldblk1_last_cmpc, newblk1_mid_cmpc, newblk1_last_cmpc;
int levelp, level2;
int old_blk1_sz, old_blk2_sz;
int old_levelp_cur_prev_keysz,
old_levelp_cur_keysz,
old_levelp_cur_next_keysz,
newblk1_last_keysz,
newblk2_first_keysz,
new_blk2_ances_first_keysz;
int old_levelp_cur_keylen,
new_levelp_cur_keylen,
old_levelp_cur_next_keylen,
new_levelp_cur_next_keylen,
oldblk1_last_keylen,
newblk1_last_keylen,
newblk2_first_keylen;
int rec_size, piece_len, tkeylen, old_levelp_rec_offset;
int blk_seg_cnt, blk_size;
uint4 save_t_err;
enum cdb_sc status;
sm_uc_ptr_t oldblk1_last_key, old_levelp_cur_next_key,
newblk1_last_key, newblk2_first_key, new_blk2_ances_first_key; /* shared memory keys */
sm_uc_ptr_t rec_base, old_levelp_blk_base,
bn_ptr1, bn_ptr2, blk2_ances_remain, old_blk1_base, old_blk2_base,
new_blk1_top, new_blk2_first_rec_base, new_blk2_remain; /* shared memory pointers */
sm_uc_ptr_t rPtr1, rPtr2;
rec_hdr_ptr_t star_rec_hdr, old_last_rec_hdr1, new_rec_hdr1, new_rec_hdr2,
blk2_ances_hdr, new_levelp_cur_hdr, new_levelp_cur_next_hdr;
blk_segment *bs_ptr1, *bs_ptr2;
srch_hist *blk1ptr, *blk2ptr; /* blk2ptr is for right sibling's hist from a minimum sub-tree containing both blocks */
error_def(ERR_GVKILLFAIL);
blk_size = cs_data->blk_size;
assert(update_array != NULL);
update_array_ptr = update_array;
blk1ptr = &(gv_target->hist);
blk2ptr = gv_target->alt_hist;
old_blk1_base = blk1ptr->h[level].buffaddr;
old_blk2_base = blk2ptr->h[level].buffaddr;
old_blk1_sz = ((blk_hdr_ptr_t)old_blk1_base)->bsiz;
old_blk2_sz = ((blk_hdr_ptr_t)old_blk2_base)->bsiz;
if (0 != level && sizeof(blk_hdr) + BSTAR_REC_SIZE == old_blk1_sz)
old_ref_star_only = TRUE;
/* Search an ancestor block at levelp >= level+1,
which has a real key value corresponding to the working block.
This key value will be changed after coalesce. */
levelp = level;
do
{
if (++levelp > blk1ptr->depth || levelp > blk2ptr->depth)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
old_levelp_blk_base = blk1ptr->h[levelp].buffaddr;
old_levelp_rec_offset = blk1ptr->h[levelp].curr_rec.offset;
rec_base = old_levelp_blk_base + old_levelp_rec_offset;
GET_RSIZ(rec_size, rec_base);
} while (BSTAR_REC_SIZE == rec_size); /* search ancestors to get a real value */
/*
old_levelp_cur_prev_key = real value of the key before the curr_key at levelp
old_levelp_cur_prev_keysz = uncompressed size of the key
Note: we may not have a previous key (old_levelp_cur_prev_keysz = 0)
*/
if (sizeof(blk_hdr) == old_levelp_rec_offset)
old_levelp_cur_prev_keysz = 0;
else
{
if (cdb_sc_normal != (status = gvcst_expand_any_key (old_levelp_blk_base, rec_base,
&old_levelp_cur_prev_key[0], &rec_size, &tkeylen, &tkeycmpc, NULL)))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
old_levelp_cur_prev_keysz = tkeylen + tkeycmpc;
}
/*
old_levelp_cur_key = real value of the curr_key at levelp
old_levelp_cur_keysz = uncompressed size of the key
old_levelp_cur_keylen = compressed size of the key
*/
READ_RECORD(levelp, rec_base, tkeycmpc, rec_size, &old_levelp_cur_key[0], old_levelp_cur_keylen, status);
if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
if (old_levelp_cur_prev_keysz)
memcpy(&old_levelp_cur_key[0], &old_levelp_cur_prev_key[0], tkeycmpc);
rec_base += rec_size;
old_levelp_cur_keysz = old_levelp_cur_keylen + tkeycmpc;
/*
old_levelp_cur_next_key = uncompressed value of the next right key of old_levelp_cur_key
old_levelp_cur_next_keysz = uncomressed size of the key
old_levelp_cur_next_keylen = comressed size of the key
Note: we may not have a next key (old_levelp_cur_next_keysz = 0)
*/
BLK_ADDR(old_levelp_cur_next_key, gv_cur_region->max_key_size + 1, unsigned char);
READ_RECORD(levelp, rec_base, tkeycmpc, rec_size, old_levelp_cur_next_key, old_levelp_cur_next_keylen, status);
if (cdb_sc_starrecord == status)
levelp_next_is_star = TRUE;
else if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
else
{
memcpy(old_levelp_cur_next_key, &old_levelp_cur_key[0], tkeycmpc);
old_levelp_cur_next_keysz = old_levelp_cur_next_keylen + tkeycmpc;
levelp_next_is_star = FALSE;
}
/*
Now process the actual working block at current level
oldblk1_last_key = real value of last key of the working block
For index block decompress *-key
oldblk1_last_keylen = compressed size of the last key
oldblk1_last_cmpc = compression count of last key of working block
old_last_rec_hdr1 = New working index block's last record header
*/
BLK_ADDR(oldblk1_last_key, gv_cur_region->max_key_size + 1, unsigned char);
if (0 == level) /* data block */
{
if (cdb_sc_normal != (status = gvcst_expand_any_key (old_blk1_base, old_blk1_base + old_blk1_sz,
oldblk1_last_key, &rec_size, &oldblk1_last_keylen, &oldblk1_last_cmpc, NULL)))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
rec_base = old_blk1_base + old_blk1_sz;
}
else /* Index blocks */
{
/* Since we will join this working block with the right sibling,
we need to remove the *-key at the end of working block
and replace with actual key value (with required compression).
We will get the real value of *-rec from its ancestor at levelp */
memcpy (oldblk1_last_key, &old_levelp_cur_key[0], old_levelp_cur_keysz);
if (!old_ref_star_only) /* if the index block is not a *-key only block) */
{
if (cdb_sc_normal != (status = gvcst_expand_any_key (old_blk1_base,
old_blk1_base + old_blk1_sz - BSTAR_REC_SIZE, &oldblk1_prev_key[0],
&rec_size, &tkeylen, &tkeycmpc, NULL)))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
GET_CMPC(oldblk1_last_cmpc, &oldblk1_prev_key[0], &old_levelp_cur_key[0]);
oldblk1_last_keylen = old_levelp_cur_keysz - oldblk1_last_cmpc;
}
else /* working block has a *-key record only */
{
/* get key value from ancestor blocks key */
oldblk1_last_keylen = old_levelp_cur_keysz;
oldblk1_last_cmpc = 0;
}
BLK_ADDR(old_last_rec_hdr1, sizeof(rec_hdr), rec_hdr);
old_last_rec_hdr1->rsiz = BSTAR_REC_SIZE + oldblk1_last_keylen;
old_last_rec_hdr1->cmpc = oldblk1_last_cmpc;
}
/*
newblk1_last_key = new working blocks final appended key
newblk1_mid_cmpc = new working blocks firstly appended key's cmpc
newblk1_last_keysz = new working blocks lastly appended key's size
star_only_merge = TRUE, we can append only a *-key record into the working block
(decompressing current *-key)
complete_merge = TRUE, rtsib can be completely merged with working block
piece_len = Size of data from old rtsibling to be merged into working block (includes rec_hdr size)
*/
BLK_ADDR(newblk1_last_key, gv_cur_region->max_key_size + 1, unsigned char);
rec_base = old_blk2_base + sizeof(blk_hdr);
READ_RECORD(level, rec_base, newblk1_last_cmpc, rec_size, newblk1_last_key, newblk1_last_keylen, status);
if (cdb_sc_starrecord == status) /* rtsib index block has *-record only */
{
if (old_blk1_sz + oldblk1_last_keylen + BSTAR_REC_SIZE > i_max_fill ) /* cannot fit even one record */
return cdb_sc_oprnotneeded;
star_only_merge = TRUE;
complete_merge = TRUE;
rec_base = old_blk2_base + sizeof(blk_hdr) + BSTAR_REC_SIZE;
}
else if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);;
return cdb_sc_blkmod;
}
else /* for both data and non-* index block */
{
newblk1_last_keysz = newblk1_last_keylen; /* first key has uncompressed real value */
GET_CMPC(newblk1_mid_cmpc, oldblk1_last_key, newblk1_last_key);
piece_len = rec_size - newblk1_mid_cmpc;
if (level == 0) /* data block */
{
if (old_blk1_sz + piece_len > d_max_fill ) /* cannot fit even one record */
return cdb_sc_oprnotneeded;
}
else /* else an index block */
{
if (old_blk1_sz + oldblk1_last_keylen + BSTAR_REC_SIZE > i_max_fill ) /* cannot fit even one record */
return cdb_sc_oprnotneeded;
if (old_blk1_sz + oldblk1_last_keylen + piece_len + BSTAR_REC_SIZE > i_max_fill )
star_only_merge = TRUE; /* can fit only a *-record */
}
rec_base += rec_size;
}
/* new_blk2_first_rec_base and new_blk1_top is set with final value for star_only_merge for index block */
new_blk2_first_rec_base = new_blk1_top = rec_base;
if (!star_only_merge)
{
BLK_ADDR(new_rec_hdr1, sizeof(rec_hdr), rec_hdr);
new_rec_hdr1->rsiz = piece_len;
new_rec_hdr1->cmpc = newblk1_mid_cmpc;
}
/* else only new_blk1_last_key will be appeneded in working block */
/* find a piece of the right sibling to be copied into the working block.
Note: rec_base points to 2nd record of old rtsib */
if (0 == level) /* if data block */
{
complete_merge = TRUE;
while (rec_base < old_blk2_base + old_blk2_sz)
{
GET_RSIZ(rec_size, rec_base);
if (old_blk1_sz + piece_len + rec_size > d_max_fill )
{
complete_merge = FALSE;
break;
}
READ_RECORD(level, rec_base, newblk1_last_cmpc, rec_size, newblk1_last_key, newblk1_last_keylen, status);
if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);;
return cdb_sc_blkmod;
}
newblk1_last_keysz = newblk1_last_keylen + newblk1_last_cmpc;
rec_base += rec_size;
piece_len += rec_size;
}/* end of "while" loop */
new_blk1_top = new_blk2_first_rec_base = rec_base;
}
else /* index block */
{
if (!star_only_merge)
{
/* we know we can fit more record in working block and rtsibling has more records */
complete_merge = TRUE;
while (rec_base < old_blk2_base + old_blk2_sz)
{
GET_RSIZ(rec_size, rec_base);
if (BSTAR_REC_SIZE == rec_size)
{
rec_base += rec_size;
piece_len += rec_size;
break; /* already we know we can fit this *-record in working block */
}
READ_RECORD(level, rec_base, newblk1_last_cmpc, rec_size,
newblk1_last_key, newblk1_last_keylen, status);
if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);;
return cdb_sc_blkmod;
}
newblk1_last_keysz = newblk1_last_keylen + newblk1_last_cmpc;
rec_base += rec_size;
piece_len += rec_size;
if (old_blk1_sz + oldblk1_last_keylen + piece_len + BSTAR_REC_SIZE > i_max_fill )
{
complete_merge = FALSE;
break;
}
}/* end of "while" loop */
new_blk1_top = new_blk2_first_rec_base = rec_base;
} /* end else *-only merge */
} /* end else index block */
if (!complete_merge)
{
/*
Adjust new right sibling's buffer
if new_rtsib_star_only == TRUE then
new right sibling will have a *-key record only
else
new_blk2_remain = base pointer of buffer including 1st record but exclude rec_header and key
new_blk2_first_keysz = size of new rtsib block's first key
*/
BLK_ADDR(newblk2_first_key, gv_cur_region->max_key_size + 1, unsigned char);
READ_RECORD(level, new_blk2_first_rec_base, tkeycmpc, rec_size,
newblk2_first_key, newblk2_first_keylen, status);
if (cdb_sc_starrecord == status) /* new rtsib will have a *-record only */
new_rtsib_star_only = TRUE;
else if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);;
return cdb_sc_blkmod;
}
else
{
memcpy(newblk2_first_key, newblk1_last_key, tkeycmpc); /* copy the compressed piece */
newblk2_first_keysz = newblk2_first_keylen + tkeycmpc;
new_blk2_remain = new_blk2_first_rec_base + sizeof(rec_hdr) + newblk2_first_keylen;
BLK_ADDR(new_rec_hdr2, sizeof(rec_hdr), rec_hdr);
new_rec_hdr2->rsiz = rec_size + tkeycmpc;
new_rec_hdr2->cmpc = 0;
}
}
block_id bm_getfree(block_id orig_hint, boolean_t *blk_used, unsigned int cw_work, cw_set_element *cs, int *cw_depth_ptr)
{
cw_set_element *cs1;
sm_uc_ptr_t bmp;
block_id bml, hint, hint_cycled, hint_limit;
block_id_ptr_t b_ptr;
int cw_set_top, depth, lcnt;
unsigned int local_maps, map_size, n_decrements = 0, total_blks;
trans_num ctn;
int4 free_bit, offset;
uint4 space_needed;
uint4 status;
srch_blk_status blkhist;
total_blks = (dba_mm == cs_data->acc_meth) ? cs_addrs->total_blks : cs_addrs->ti->total_blks;
if (orig_hint >= total_blks) /* for TP, hint can be > total_blks */
orig_hint = 1;
hint = orig_hint;
hint_cycled = DIVIDE_ROUND_UP(total_blks, BLKS_PER_LMAP);
hint_limit = DIVIDE_ROUND_DOWN(orig_hint, BLKS_PER_LMAP);
local_maps = hint_cycled + 2; /* for (up to) 2 wraps */
for (lcnt = 0; lcnt <= local_maps; lcnt++)
{
bml = bmm_find_free(hint / BLKS_PER_LMAP, (sm_uc_ptr_t)MM_ADDR(cs_data), local_maps);
if ((NO_FREE_SPACE == bml) || (bml >= hint_cycled))
{ /* if no free space or might have looped to original map, extend */
if ((NO_FREE_SPACE != bml) && (hint_limit < hint_cycled))
{
hint_cycled = hint_limit;
hint = 1;
continue;
}
if (SS_NORMAL != (status = gdsfilext(cs_data->extension_size, total_blks)))
return (status);
if (dba_mm == cs_data->acc_meth)
return (FILE_EXTENDED);
hint = total_blks;
total_blks = cs_addrs->ti->total_blks;
hint_cycled = DIVIDE_ROUND_UP(total_blks, BLKS_PER_LMAP);
local_maps = hint_cycled + 2; /* for (up to) 2 wraps */
/*
* note that you can make an optimization of not going back over the whole database and going over
* only the extended section. but since it is very unlikely that a free block won't be found
* in the extended section and the fact that we are starting from the extended section in either
* approach and the fact that we have a GTMASSERT to check that we don't have a lot of
* free blocks while doing an extend and the fact that it is very easy to make the change to do
* a full-pass, the full-pass solution is currently being implemented
*/
lcnt = -1; /* allow it one extra pass to ensure that it can take advantage of the entension */
n_decrements++; /* used only for debugging purposes */
continue;
}
bml *= BLKS_PER_LMAP;
if (ROUND_DOWN2(hint, BLKS_PER_LMAP) != bml)
{ /* not within requested map */
if ((bml < hint) && (hint_cycled)) /* wrap? - second one should force an extend for sure */
hint_cycled = (hint_limit < hint_cycled) ? hint_limit: 0;
hint = bml + 1; /* start at beginning */
}
if (ROUND_DOWN2(total_blks, BLKS_PER_LMAP) == bml)
map_size = (total_blks - bml);
else
map_size = BLKS_PER_LMAP;
if (0 != dollar_tlevel)
{
depth = cw_work;
cw_set_top = *cw_depth_ptr;
if (depth < cw_set_top)
tp_get_cw(cs, cw_work, &cs1);
for (; depth < cw_set_top; depth++, cs1 = cs1->next_cw_set)
{ /* do tp front to back because list is more efficient than tp_get_cw and forward pointers exist */
if (bml == cs1->blk)
{
TRAVERSE_TO_LATEST_CSE(cs1);
break;
}
}
if (depth >= cw_set_top)
{
assert(cw_set_top == depth);
depth = 0;
}
} else
{
for (depth = *cw_depth_ptr - 1; depth >= cw_work; depth--)
{ /* do non-tp back to front, because of adjacency */
if (bml == (cs + depth)->blk)
{
cs1 = cs + depth;
break;
}
}
if (depth < cw_work)
{
assert(cw_work - 1 == depth);
depth = 0;
}
}
if (0 == depth)
{
ctn = cs_addrs->ti->curr_tn;
if (!(bmp = t_qread(bml, (sm_int_ptr_t)&blkhist.cycle, &blkhist.cr)))
return MAP_RD_FAIL;
if ((BM_SIZE(BLKS_PER_LMAP) != ((blk_hdr_ptr_t)bmp)->bsiz) || (LCL_MAP_LEVL != ((blk_hdr_ptr_t)bmp)->levl))
{
assert(CDB_STAGNATE > t_tries);
rdfail_detail = cdb_sc_badbitmap;
return MAP_RD_FAIL;
}
offset = 0;
} else
{
bmp = cs1->old_block;
b_ptr = (block_id_ptr_t)(cs1->upd_addr);
b_ptr += cs1->reference_cnt - 1;
offset = *b_ptr + 1;
}
if (offset < map_size)
{
free_bit = bm_find_blk(offset, (sm_uc_ptr_t)bmp + sizeof(blk_hdr), map_size, blk_used);
if (MAP_RD_FAIL == free_bit)
return MAP_RD_FAIL;
} else
free_bit = NO_FREE_SPACE;
if (NO_FREE_SPACE != free_bit)
break;
if ((hint = bml + BLKS_PER_LMAP) >= total_blks) /* if map is full, start at 1st blk in next map */
{ /* wrap - second one should force an extend for sure */
hint = 1;
if (hint_cycled)
hint_cycled = (hint_limit < hint_cycled) ? hint_limit: 0;
}
if ((0 == depth) && (FALSE != cs_addrs->now_crit)) /* if it's from the cw_set, its state is murky */
bit_clear(bml / BLKS_PER_LMAP, MM_ADDR(cs_data)); /* if crit, repair master map error */
}
/* If not in the final retry, it is possible that free_bit is >= map_size (e.g. if bitmap block gets recycled). */
if (map_size <= (uint4)free_bit && CDB_STAGNATE <= t_tries)
{ /* bad free bit */
assert((NO_FREE_SPACE == free_bit) && (lcnt > local_maps)); /* All maps full, should have extended */
GTMASSERT;
}
if (0 != depth)
{
b_ptr = (block_id_ptr_t)(cs1->upd_addr);
b_ptr += cs1->reference_cnt++;
*b_ptr = free_bit;
} else
{
space_needed = (BLKS_PER_LMAP + 1) * sizeof(block_id);
if (dollar_tlevel)
{
ENSURE_UPDATE_ARRAY_SPACE(space_needed); /* have brackets for "if" for macros */
}
BLK_ADDR(b_ptr, space_needed, block_id);
memset(b_ptr, 0, space_needed);
*b_ptr = free_bit;
blkhist.blk_num = bml;
blkhist.buffaddr = bmp; /* cycle and cr have already been assigned from t_qread */
t_write_map(&blkhist, (uchar_ptr_t)b_ptr, ctn, 1); /* last parameter 1 is what cs->reference_cnt gets set to */
}
return bml + free_bit;
}
void mu_reorg_upgrd_dwngrd(void)
{
blk_hdr new_hdr;
blk_segment *bs1, *bs_ptr;
block_id *blkid_ptr, curblk, curbmp, start_blk, stop_blk, start_bmp, last_bmp;
block_id startblk_input, stopblk_input;
boolean_t upgrade, downgrade, safejnl, nosafejnl, region, first_reorg_in_this_db_fmt, reorg_entiredb;
boolean_t startblk_specified, stopblk_specified, set_fully_upgraded, db_got_to_v5_once, mark_blk_free;
cache_rec_ptr_t cr;
char *bml_lcl_buff = NULL, *command, *reorg_command;
sm_uc_ptr_t bptr = NULL;
cw_set_element *cse;
enum cdb_sc cdb_status;
enum db_ver new_db_format, ondsk_blkver;
gd_region *reg;
int cycle;
int4 blk_seg_cnt, blk_size; /* needed for BLK_INIT,BLK_SEG and BLK_FINI macros */
int4 blocks_left, expected_blks2upgrd, actual_blks2upgrd, total_blks, free_blks;
int4 status, status1, mapsize, lcnt, bml_status;
reorg_stats_t reorg_stats;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
sm_uc_ptr_t blkBase, bml_sm_buff; /* shared memory pointer to the bitmap global buffer */
srch_hist alt_hist;
srch_blk_status *blkhist, bmlhist;
tp_region *rptr;
trans_num curr_tn;
unsigned char save_cw_set_depth;
uint4 lcl_update_trans;
region = (CLI_PRESENT == cli_present("REGION"));
upgrade = (CLI_PRESENT == cli_present("UPGRADE"));
downgrade = (CLI_PRESENT == cli_present("DOWNGRADE"));
assert(upgrade && !downgrade || !upgrade && downgrade);
command = upgrade ? "UPGRADE" : "DOWNGRADE";
reorg_command = upgrade ? "MUPIP REORG UPGRADE" : "MUPIP REORG DOWNGRADE";
reorg_entiredb = TRUE; /* unless STARTBLK or STOPBLK is specified we are going to {up,down}grade the entire database */
startblk_specified = FALSE;
assert(SIZEOF(block_id) == SIZEOF(uint4));
if ((CLI_PRESENT == cli_present("STARTBLK")) && (cli_get_hex("STARTBLK", (uint4 *)&startblk_input)))
{
reorg_entiredb = FALSE;
startblk_specified = TRUE;
}
stopblk_specified = FALSE;
assert(SIZEOF(block_id) == SIZEOF(uint4));
if ((CLI_PRESENT == cli_present("STOPBLK")) && (cli_get_hex("STOPBLK", (uint4 *)&stopblk_input)))
{
reorg_entiredb = FALSE;
stopblk_specified = TRUE;
}
mu_reorg_upgrd_dwngrd_in_prog = TRUE;
mu_reorg_nosafejnl = (CLI_NEGATED == cli_present("SAFEJNL")) ? TRUE : FALSE;
assert(region);
status = SS_NORMAL;
error_mupip = FALSE;
gvinit(); /* initialize gd_header (needed by the later call to mu_getlst) */
mu_getlst("REG_NAME", SIZEOF(tp_region)); /* get the parameter corresponding to REGION qualifier */
if (error_mupip)
{
util_out_print("!/MUPIP REORG !AD cannot proceed with above errors!/", TRUE, LEN_AND_STR(command));
mupip_exit(ERR_MUNOACTION);
}
assert(DBKEYSIZE(MAX_KEY_SZ) == gv_keysize); /* no need to invoke GVKEYSIZE_INIT_IF_NEEDED macro */
gv_target = targ_alloc(gv_keysize, NULL, NULL); /* t_begin needs this initialized */
gv_target_list = NULL;
memset(&alt_hist, 0, SIZEOF(alt_hist)); /* null-initialize history */
blkhist = &alt_hist.h[0];
for (rptr = grlist; NULL != rptr; rptr = rptr->fPtr)
{
if (mu_ctrly_occurred || mu_ctrlc_occurred)
break;
reg = rptr->reg;
util_out_print("!/Region !AD : MUPIP REORG !AD started", TRUE, REG_LEN_STR(reg), LEN_AND_STR(command));
if (reg_cmcheck(reg))
{
util_out_print("Region !AD : MUPIP REORG !AD cannot run across network",
TRUE, REG_LEN_STR(reg), LEN_AND_STR(command));
status = ERR_MUNOFINISH;
continue;
}
mu_reorg_process = TRUE; /* gvcst_init will use this value to use gtm_poollimit settings. */
gvcst_init(reg);
mu_reorg_process = FALSE;
assert(update_array != NULL);
/* access method stored in global directory and database file header might be different in which case
* the database setting prevails. therefore, the access method check can be done only after opening
* the database (i.e. after the gvcst_init)
*/
if (dba_bg != REG_ACC_METH(reg))
{
util_out_print("Region !AD : MUPIP REORG !AD cannot continue as access method is not BG",
TRUE, REG_LEN_STR(reg), LEN_AND_STR(command));
status = ERR_MUNOFINISH;
continue;
}
/* The mu_getlst call above uses insert_region to create the grlist, which ensures that duplicate regions mapping to
* the same db file correspond to only one grlist entry.
*/
assert(FALSE == reg->was_open);
TP_CHANGE_REG(reg); /* sets gv_cur_region, cs_addrs, cs_data */
csa = cs_addrs;
csd = cs_data;
blk_size = csd->blk_size; /* "blk_size" is used by the BLK_FINI macro */
if (reg->read_only)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(reg));
status = ERR_MUNOFINISH;
continue;
}
assert(GDSVCURR == GDSV6); /* so we trip this assert in case GDSVCURR changes without a change to this module */
new_db_format = (upgrade ? GDSV6 : GDSV4);
grab_crit(reg);
curr_tn = csd->trans_hist.curr_tn;
/* set the desired db format in the file header to the appropriate version, increment transaction number */
status1 = desired_db_format_set(reg, new_db_format, reorg_command);
assert(csa->now_crit); /* desired_db_format_set() should not have released crit */
first_reorg_in_this_db_fmt = TRUE; /* with the current desired_db_format, this is the first reorg */
if (SS_NORMAL != status1)
{ /* "desired_db_format_set" would have printed appropriate error messages */
if (ERR_MUNOACTION != status1)
{ /* real error occurred while setting the db format. skip to next region */
status = ERR_MUNOFINISH;
rel_crit(reg);
continue;
}
util_out_print("Region !AD : Desired DB Format remains at !AD after !AD", TRUE, REG_LEN_STR(reg),
LEN_AND_STR(gtm_dbversion_table[new_db_format]), LEN_AND_STR(reorg_command));
if (csd->reorg_db_fmt_start_tn == csd->desired_db_format_tn)
first_reorg_in_this_db_fmt = FALSE;
} else
util_out_print("Region !AD : Desired DB Format set to !AD by !AD", TRUE, REG_LEN_STR(reg),
LEN_AND_STR(gtm_dbversion_table[new_db_format]), LEN_AND_STR(reorg_command));
assert(dba_bg == csd->acc_meth);
/* Check blks_to_upgrd counter to see if upgrade/downgrade is complete */
total_blks = csd->trans_hist.total_blks;
free_blks = csd->trans_hist.free_blocks;
actual_blks2upgrd = csd->blks_to_upgrd;
/* If MUPIP REORG UPGRADE and there is no block to upgrade in the database as indicated by BOTH
* "csd->blks_to_upgrd" and "csd->fully_upgraded", then we can skip processing.
* If MUPIP REORG UPGRADE and all non-free blocks need to be upgraded then again we can skip processing.
*/
if ((upgrade && (0 == actual_blks2upgrd) && csd->fully_upgraded)
|| (!upgrade && ((total_blks - free_blks) == actual_blks2upgrd)))
{
util_out_print("Region !AD : Blocks to Upgrade counter indicates no action needed for MUPIP REORG !AD",
TRUE, REG_LEN_STR(reg), LEN_AND_STR(command));
util_out_print("Region !AD : Total Blocks = [0x!XL] : Free Blocks = [0x!XL] : "
"Blocks to upgrade = [0x!XL]",
TRUE, REG_LEN_STR(reg), total_blks, free_blks, actual_blks2upgrd);
util_out_print("Region !AD : MUPIP REORG !AD finished!/", TRUE, REG_LEN_STR(reg), LEN_AND_STR(command));
rel_crit(reg);
continue;
}
stop_blk = total_blks;
if (stopblk_specified && stopblk_input <= stop_blk)
stop_blk = stopblk_input;
if (first_reorg_in_this_db_fmt)
{ /* Note down reorg start tn (in case we are interrupted, future reorg will know to resume) */
csd->reorg_db_fmt_start_tn = csd->desired_db_format_tn;
csd->reorg_upgrd_dwngrd_restart_block = 0;
start_blk = (startblk_specified ? startblk_input : 0);
} else
{ /* Either a concurrent MUPIP REORG of the same type ({up,down}grade) is currently running
* or a previously running REORG of the same type was interrupted (Ctrl-Ced).
* In either case resume processing from whatever restart block number is stored in fileheader
* the only exception is if "STARTBLK" was specified in the input in which use that unconditionally.
*/
start_blk = (startblk_specified ? startblk_input : csd->reorg_upgrd_dwngrd_restart_block);
}
if (start_blk > stop_blk)
start_blk = stop_blk;
mu_reorg_upgrd_dwngrd_start_tn = csd->reorg_db_fmt_start_tn;
/* Before releasing crit, flush the file-header and dirty buffers in cache to disk. This is because we are now
* going to read each GDS block directly from disk to determine if it needs to be upgraded/downgraded or not.
*/
if (!wcs_flu(WCSFLU_FLUSH_HDR)) /* wcs_flu assumes gv_cur_region is set (which it is in this routine) */
{
rel_crit(reg);
gtm_putmsg_csa(CSA_ARG(csa)
VARLSTCNT(6) ERR_BUFFLUFAILED, 4, LEN_AND_LIT("MUPIP REORG UPGRADE/DOWNGRADE"), DB_LEN_STR(reg));
status = ERR_MUNOFINISH;
continue;
}
rel_crit(reg);
/* Loop through entire database one GDS block at a time and upgrade/downgrade each of them */
status1 = SS_NORMAL;
start_bmp = ROUND_DOWN2(start_blk, BLKS_PER_LMAP);
last_bmp = ROUND_DOWN2(stop_blk - 1, BLKS_PER_LMAP);
curblk = start_blk; /* curblk is the block to be upgraded/downgraded */
util_out_print("Region !AD : Started processing from block number [0x!XL]", TRUE, REG_LEN_STR(reg), curblk);
if (NULL != bptr)
{ /* malloc/free "bptr" for each region as GDS block-size can be different */
free(bptr);
bptr = NULL;
}
memset(&reorg_stats, 0, SIZEOF(reorg_stats)); /* initialize statistics for this region */
for (curbmp = start_bmp; curbmp <= last_bmp; curbmp += BLKS_PER_LMAP)
{
if (mu_ctrly_occurred || mu_ctrlc_occurred)
{
status1 = ERR_MUNOFINISH;
break;
}
/* --------------------------------------------------------------
* Read in current bitmap block
* --------------------------------------------------------------
*/
assert(!csa->now_crit);
bml_sm_buff = t_qread(curbmp, (sm_int_ptr_t)&cycle, &cr); /* bring block into the cache outside of crit */
reorg_stats.blks_read_from_disk_bmp++;
grab_crit_encr_cycle_sync(reg); /* needed so t_qread does not return NULL below */
if (mu_reorg_upgrd_dwngrd_start_tn != csd->desired_db_format_tn)
{ /* csd->desired_db_format changed since reorg started. discontinue the reorg */
/* see later comment on "csd->reorg_upgrd_dwngrd_restart_block" for why the assignment
* of this field should be done only if a db format change did not occur.
*/
rel_crit(reg);
status1 = ERR_MUNOFINISH;
/* This "start_tn" check is redone after the for-loop and an error message is printed there */
break;
} else if (reorg_entiredb)
{ /* Change "csd->reorg_upgrd_dwngrd_restart_block" only if STARTBLK or STOPBLK was NOT specified */
assert(csd->reorg_upgrd_dwngrd_restart_block <= MAX(start_blk, curbmp));
csd->reorg_upgrd_dwngrd_restart_block = curbmp; /* previous blocks have been upgraded/downgraded */
}
/* Check blks_to_upgrd counter to see if upgrade/downgrade is complete.
* Repeat check done a few steps earlier outside of this for loop.
*/
total_blks = csd->trans_hist.total_blks;
free_blks = csd->trans_hist.free_blocks;
actual_blks2upgrd = csd->blks_to_upgrd;
if ((upgrade && (0 == actual_blks2upgrd) && csd->fully_upgraded)
|| (!upgrade && ((total_blks - free_blks) == actual_blks2upgrd)))
{
rel_crit(reg);
break;
}
bml_sm_buff = t_qread(curbmp, (sm_int_ptr_t)&cycle, &cr); /* now that in crit, note down stable buffer */
if (NULL == bml_sm_buff)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(1) ERR_DSEBLKRDFAIL);
ondsk_blkver = cr->ondsk_blkver; /* note down db fmt on disk for bitmap block */
/* Take a copy of the shared memory bitmap buffer into process-private memory before releasing crit.
* We are interested in those blocks that are currently marked as USED in the bitmap.
* It is possible that once we release crit, concurrent updates change the bitmap state of those blocks.
* In that case, those updates will take care of doing the upgrade/downgrade of those blocks in the
* format currently set in csd->desired_db_format i.e. accomplishing MUPIP REORG UPGRADE/DOWNGRADE's job.
* If the desired_db_format changes concurrently, we will stop doing REORG UPGRADE/DOWNGRADE processing.
*/
if (NULL == bml_lcl_buff)
bml_lcl_buff = malloc(BM_SIZE(BLKS_PER_LMAP));
memcpy(bml_lcl_buff, (blk_hdr_ptr_t)bml_sm_buff, BM_SIZE(BLKS_PER_LMAP));
if (FALSE == cert_blk(reg, curbmp, (blk_hdr_ptr_t)bml_lcl_buff, 0, FALSE))
{ /* certify the block while holding crit as cert_blk uses fields from file-header (shared memory) */
assert(FALSE); /* in pro, skip ugprading/downgarding all blks in this unreliable local bitmap */
rel_crit(reg);
util_out_print("Region !AD : Bitmap Block [0x!XL] has integrity errors. Skipping this bitmap.",
TRUE, REG_LEN_STR(reg), curbmp);
status1 = ERR_MUNOFINISH;
continue;
}
rel_crit(reg);
/* ------------------------------------------------------------------------
* Upgrade/Downgrade all BUSY blocks in the current bitmap
* ------------------------------------------------------------------------
*/
curblk = (curbmp == start_bmp) ? start_blk : curbmp;
mapsize = (curbmp == last_bmp) ? (stop_blk - curbmp) : BLKS_PER_LMAP;
assert(0 != mapsize);
assert(mapsize <= BLKS_PER_LMAP);
db_got_to_v5_once = csd->db_got_to_v5_once;
for (lcnt = curblk - curbmp; lcnt < mapsize; lcnt++, curblk++)
{
if (mu_ctrly_occurred || mu_ctrlc_occurred)
{
status1 = ERR_MUNOFINISH;
goto stop_reorg_on_this_reg; /* goto needed because of nested FOR Loop */
}
GET_BM_STATUS(bml_lcl_buff, lcnt, bml_status);
assert(BLK_MAPINVALID != bml_status); /* cert_blk ran clean so we dont expect invalid entries */
if (BLK_FREE == bml_status)
{
reorg_stats.blks_skipped_free++;
continue;
}
/* MUPIP REORG UPGRADE/DOWNGRADE will convert USED & RECYCLED blocks */
if (db_got_to_v5_once || (BLK_RECYCLED != bml_status))
{ /* Do NOT read recycled V4 block from disk unless it is guaranteed NOT to be too full */
if (lcnt)
{ /* non-bitmap block */
/* read in block from disk into private buffer. dont pollute the cache yet */
if (NULL == bptr)
bptr = (sm_uc_ptr_t)malloc(blk_size);
status1 = dsk_read(curblk, bptr, &ondsk_blkver, FALSE);
/* dsk_read on curblk could return an error (DYNUPGRDFAIL) if curblk needs to be
* upgraded and if its block size was too big to allow the extra block-header space
* requirements for a dynamic upgrade. a MUPIP REORG DOWNGRADE should not error out
* in that case as the block is already in the downgraded format.
*/
if (SS_NORMAL != status1)
{
if (!upgrade && (ERR_DYNUPGRDFAIL == status1))
{
assert(GDSV4 == new_db_format);
ondsk_blkver = new_db_format;
} else
{
gtm_putmsg_csa(CSA_ARG(csa)
VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), status1);
util_out_print("Region !AD : Error occurred while reading block "
"[0x!XL]", TRUE, REG_LEN_STR(reg), curblk);
status1 = ERR_MUNOFINISH;
goto stop_reorg_on_this_reg;/* goto needed due to nested FOR Loop */
}
}
reorg_stats.blks_read_from_disk_nonbmp++;
} /* else bitmap block has been read in crit earlier and ondsk_blkver appropriately set */
if (new_db_format == ondsk_blkver)
{
assert((SS_NORMAL == status1) || (!upgrade && (ERR_DYNUPGRDFAIL == status1)));
status1 = SS_NORMAL; /* treat DYNUPGRDFAIL as no error in case of downgrade */
reorg_stats.blks_skipped_newfmtindisk++;
continue; /* current disk version is identical to what is desired */
}
assert(SS_NORMAL == status1);
}
/* Begin non-TP transaction to upgrade/downgrade the block.
* The way we do that is by updating the block using a null update array.
* Any update to a block will trigger an automatic upgrade/downgrade of the block based on
* the current fileheader desired_db_format setting and we use that here.
*/
t_begin(ERR_MUREORGFAIL, UPDTRNS_DB_UPDATED_MASK);
for (; ;)
{
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
curr_tn = csd->trans_hist.curr_tn;
db_got_to_v5_once = csd->db_got_to_v5_once;
if (db_got_to_v5_once || (BLK_RECYCLED != bml_status))
{
blkhist->cse = NULL; /* start afresh (do not use value from previous retry) */
blkBase = t_qread(curblk, (sm_int_ptr_t)&blkhist->cycle, &blkhist->cr);
if (NULL == blkBase)
{
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
blkhist->blk_num = curblk;
blkhist->buffaddr = blkBase;
ondsk_blkver = blkhist->cr->ondsk_blkver;
new_hdr = *(blk_hdr_ptr_t)blkBase;
mu_reorg_upgrd_dwngrd_blktn = new_hdr.tn;
mark_blk_free = FALSE;
inctn_opcode = upgrade ? inctn_blkupgrd : inctn_blkdwngrd;
} else
{
mark_blk_free = TRUE;
inctn_opcode = inctn_blkmarkfree;
}
inctn_detail.blknum_struct.blknum = curblk;
/* t_end assumes that the history it is passed does not contain a bitmap block.
* for bitmap block, the history validation information is passed through cse instead.
* therefore we need to handle bitmap and non-bitmap cases separately.
*/
if (!lcnt)
{ /* Means a bitmap block.
* At this point we can do a "new_db_format != ondsk_blkver" check to determine
* if the block got converted since we did the dsk_read (see the non-bitmap case
* for a similar check done there), but in that case we will have a transaction
* which has read 1 bitmap block and is updating no block. "t_end" currently cannot
* handle this case as it expects any bitmap block that needs validation to also
* have a corresponding cse which will hold its history. Hence we avoid doing the
* new_db_format check. The only disadvantage of this is that we will end up
* modifying the bitmap block as part of this transaction (in an attempt to convert
* its ondsk_blkver) even though it is already in the right format. Since this
* overhead is going to be one per bitmap block and since the block is in the cache
* at this point, we should not lose much.
*/
assert(!mark_blk_free);
BLK_ADDR(blkid_ptr, SIZEOF(block_id), block_id);
*blkid_ptr = 0;
t_write_map(blkhist, (unsigned char *)blkid_ptr, curr_tn, 0);
assert(&alt_hist.h[0] == blkhist);
alt_hist.h[0].blk_num = 0; /* create empty history for bitmap block */
assert(update_trans);
} else
{ /* non-bitmap block. fill in history for validation in t_end */
assert(curblk); /* we should never come here for block 0 (bitmap) */
if (!mark_blk_free)
{
assert(blkhist->blk_num == curblk);
assert(blkhist->buffaddr == blkBase);
blkhist->tn = curr_tn;
alt_hist.h[1].blk_num = 0;
}
/* Also need to pass the bitmap as history to detect if any concurrent M-kill
* is freeing up the same USED block that we are trying to convert OR if any
* concurrent M-set is reusing the same RECYCLED block that we are trying to
* convert. Because of t_end currently not being able to validate a bitmap
* without that simultaneously having a cse, we need to create a cse for the
* bitmap that is used only for bitmap history validation, but should not be
* used to update the contents of the bitmap block in bg_update.
*/
bmlhist.buffaddr = t_qread(curbmp, (sm_int_ptr_t)&bmlhist.cycle, &bmlhist.cr);
if (NULL == bmlhist.buffaddr)
{
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
bmlhist.blk_num = curbmp;
bmlhist.tn = curr_tn;
GET_BM_STATUS(bmlhist.buffaddr, lcnt, bml_status);
if (BLK_MAPINVALID == bml_status)
{
t_retry(cdb_sc_lostbmlcr);
continue;
}
if (!mark_blk_free)
{
if ((new_db_format != ondsk_blkver) && (BLK_FREE != bml_status))
{ /* block still needs to be converted. create cse */
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blkBase + 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)blkBase)->levl, FALSE,
FALSE, GDS_WRITE_PLAIN);
/* The directory tree status for now is only used to determine
* whether writing the block to snapshot file (see t_end_sysops.c).
* For reorg upgrade/downgrade process, the block is updated in a
* sequential way without changing the gv_target. In this case, we
* assume the block is in directory tree so as to have it written to
* the snapshot file.
*/
BIT_SET_DIR_TREE(cw_set[cw_set_depth-1].blk_prior_state);
/* reset update_trans in case previous retry had set it to 0 */
update_trans = UPDTRNS_DB_UPDATED_MASK;
if (BLK_RECYCLED == bml_status)
{ /* If block that we are upgarding is RECYCLED, indicate to
* bg_update that blks_to_upgrd counter should NOT be
* touched in this case by setting "mode" to a special value
*/
assert(cw_set[cw_set_depth-1].mode == gds_t_write);
cw_set[cw_set_depth-1].mode = gds_t_write_recycled;
/* we SET block as NOT RECYCLED, otherwise, the mm_update()
* or bg_update_phase2 may skip writing it to snapshot file
* when its level is 0
*/
BIT_CLEAR_RECYCLED(cw_set[cw_set_depth-1].blk_prior_state);
}
} else
{ /* Block got converted by another process since we did the dsk_read.
* or this block became marked free in the bitmap.
* No need to update this block. just call t_end for validation of
* both the non-bitmap block as well as the bitmap block.
* Note down that this transaction is no longer updating any blocks.
*/
update_trans = 0;
}
/* Need to put bit maps on the end of the cw set for concurrency checking.
* We want to simulate t_write_map, except we want to update "cw_map_depth"
* instead of "cw_set_depth". Hence the save and restore logic below.
* This part of the code is similar to the one in mu_swap_blk.c
*/
save_cw_set_depth = cw_set_depth;
assert(!cw_map_depth);
t_write_map(&bmlhist, NULL, curr_tn, 0); /* will increment cw_set_depth */
cw_map_depth = cw_set_depth; /* set cw_map_depth to latest cw_set_depth */
cw_set_depth = save_cw_set_depth;/* restore cw_set_depth */
/* t_write_map simulation end */
} else
{
if (BLK_RECYCLED != bml_status)
{ /* Block was RECYCLED at beginning but no longer so. Retry */
t_retry(cdb_sc_bmlmod);
continue;
}
/* Mark recycled block as FREE in bitmap */
assert(lcnt == (curblk - curbmp));
assert(update_array_ptr == update_array);
*((block_id *)update_array_ptr) = lcnt;
update_array_ptr += SIZEOF(block_id);
/* the following assumes SIZEOF(block_id) == SIZEOF(int) */
assert(SIZEOF(block_id) == SIZEOF(int));
*(int *)update_array_ptr = 0;
t_write_map(&bmlhist, (unsigned char *)update_array, curr_tn, 0);
update_trans = UPDTRNS_DB_UPDATED_MASK;
}
}
assert(SIZEOF(lcl_update_trans) == SIZEOF(update_trans));
lcl_update_trans = update_trans; /* take a copy before t_end modifies it */
if ((trans_num)0 != t_end(&alt_hist, NULL, TN_NOT_SPECIFIED))
{ /* In case this is MM and t_end() remapped an extended database, reset csd */
assert(csd == cs_data);
if (!lcl_update_trans)
{
assert(lcnt);
assert(!mark_blk_free);
assert((new_db_format == ondsk_blkver) || (BLK_BUSY != bml_status));
if (BLK_BUSY != bml_status)
reorg_stats.blks_skipped_free++;
else
reorg_stats.blks_skipped_newfmtincache++;
} else if (!lcnt)
reorg_stats.blks_converted_bmp++;
else
reorg_stats.blks_converted_nonbmp++;
break;
}
assert(csd == cs_data);
}
}
}
stop_reorg_on_this_reg:
/* even though ctrl-c occurred, update file-header fields to store reorg's progress before exiting */
grab_crit(reg);
blocks_left = 0;
assert(csd->trans_hist.total_blks >= csd->blks_to_upgrd);
actual_blks2upgrd = csd->blks_to_upgrd;
total_blks = csd->trans_hist.total_blks;
free_blks = csd->trans_hist.free_blocks;
/* Care should be taken not to set "csd->reorg_upgrd_dwngrd_restart_block" in case of a concurrent db fmt
* change. This is because let us say we are doing REORG UPGRADE. A concurrent REORG DOWNGRADE would
* have reset "csd->reorg_upgrd_dwngrd_restart_block" field to 0 and if that reorg is interrupted by a
* Ctrl-C (before this reorg came here) it would have updated "csd->reorg_upgrd_dwngrd_restart_block" to
* a non-zero value indicating how many blocks from 0 have been downgraded. We should not reset this
* field to "curblk" as it will be mis-interpreted as the number of blocks that have been DOWNgraded.
*/
set_fully_upgraded = FALSE;
if (mu_reorg_upgrd_dwngrd_start_tn != csd->desired_db_format_tn)
{ /* csd->desired_db_format changed since reorg started. discontinue the reorg */
util_out_print("Region !AD : Desired DB Format changed during REORG. Stopping REORG.",
TRUE, REG_LEN_STR(reg));
status1 = ERR_MUNOFINISH;
} else if (reorg_entiredb)
{ /* Change "csd->reorg_upgrd_dwngrd_restart_block" only if STARTBLK or STOPBLK was NOT specified */
assert(csd->reorg_upgrd_dwngrd_restart_block <= curblk);
csd->reorg_upgrd_dwngrd_restart_block = curblk; /* blocks lesser than this have been upgraded/downgraded */
expected_blks2upgrd = upgrade ? 0 : (total_blks - free_blks);
blocks_left = upgrade ? actual_blks2upgrd : (expected_blks2upgrd - actual_blks2upgrd);
/* If this reorg command went through all blocks in the database, then it should have
* correctly concluded at this point whether the reorg is complete or not.
* If this reorg command started from where a previous incomplete reorg left
* (i.e. first_reorg_in_this_db_fmt is FALSE), it cannot determine if the initial
* GDS blocks that it skipped are completely {up,down}graded or not.
*/
assert((0 == blocks_left) || (SS_NORMAL != status1) || !first_reorg_in_this_db_fmt);
/* If this is a MUPIP REORG UPGRADE that did go through every block in the database (indicated by
* "reorg_entiredb" && "first_reorg_in_this_db_fmt") and the current count of "blks_to_upgrd" is
* 0 in the file-header and the desired_db_format did not change since the start of the REORG,
* we can be sure that the entire database has been upgraded. Set "csd->fully_upgraded" to TRUE.
*/
if ((SS_NORMAL == status1) && first_reorg_in_this_db_fmt && upgrade && (0 == actual_blks2upgrd))
{
csd->fully_upgraded = TRUE;
csd->db_got_to_v5_once = TRUE;
set_fully_upgraded = TRUE;
}
/* flush all changes noted down in the file-header */
if (!wcs_flu(WCSFLU_FLUSH_HDR)) /* wcs_flu assumes gv_cur_region is set (which it is in this routine) */
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_BUFFLUFAILED, 4,
LEN_AND_LIT("MUPIP REORG UPGRADE/DOWNGRADE"), DB_LEN_STR(reg));
status = ERR_MUNOFINISH;
rel_crit(reg);
continue;
}
}
curr_tn = csd->trans_hist.curr_tn;
rel_crit(reg);
util_out_print("Region !AD : Stopped processing at block number [0x!XL]", TRUE, REG_LEN_STR(reg), curblk);
/* Print statistics */
util_out_print("Region !AD : Statistics : Blocks Read From Disk (Bitmap) : 0x!XL",
TRUE, REG_LEN_STR(reg), reorg_stats.blks_read_from_disk_bmp);
util_out_print("Region !AD : Statistics : Blocks Skipped (Free) : 0x!XL",
TRUE, REG_LEN_STR(reg), reorg_stats.blks_skipped_free);
util_out_print("Region !AD : Statistics : Blocks Read From Disk (Non-Bitmap) : 0x!XL",
TRUE, REG_LEN_STR(reg), reorg_stats.blks_read_from_disk_nonbmp);
util_out_print("Region !AD : Statistics : Blocks Skipped (new fmt in disk) : 0x!XL",
TRUE, REG_LEN_STR(reg), reorg_stats.blks_skipped_newfmtindisk);
util_out_print("Region !AD : Statistics : Blocks Skipped (new fmt in cache) : 0x!XL",
TRUE, REG_LEN_STR(reg), reorg_stats.blks_skipped_newfmtincache);
util_out_print("Region !AD : Statistics : Blocks Converted (Bitmap) : 0x!XL",
TRUE, REG_LEN_STR(reg), reorg_stats.blks_converted_bmp);
util_out_print("Region !AD : Statistics : Blocks Converted (Non-Bitmap) : 0x!XL",
TRUE, REG_LEN_STR(reg), reorg_stats.blks_converted_nonbmp);
if (reorg_entiredb && (SS_NORMAL == status1) && (0 != blocks_left))
{ /* file-header counter does not match what reorg on the entire database expected to see */
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBBTUWRNG, 2, expected_blks2upgrd, actual_blks2upgrd);
util_out_print("Region !AD : Run MUPIP INTEG (without FAST qualifier) to fix the counter",
TRUE, REG_LEN_STR(reg));
status1 = ERR_MUNOFINISH;
} else
util_out_print("Region !AD : Total Blocks = [0x!XL] : Free Blocks = [0x!XL] : "
"Blocks to upgrade = [0x!XL]",
TRUE, REG_LEN_STR(reg), total_blks, free_blks, actual_blks2upgrd);
/* Issue success or failure message for this region */
if (SS_NORMAL == status1)
{ /* issue success only if REORG did not encounter any error in its processing */
if (set_fully_upgraded)
util_out_print("Region !AD : Database is now FULLY UPGRADED", TRUE, REG_LEN_STR(reg));
util_out_print("Region !AD : MUPIP REORG !AD finished!/", TRUE, REG_LEN_STR(reg), LEN_AND_STR(command));
send_msg_csa(CSA_ARG(csa) VARLSTCNT(7) ERR_MUREUPDWNGRDEND, 5, REG_LEN_STR(reg),
process_id, process_id, &curr_tn);
} else
{
assert(ERR_MUNOFINISH == status1);
assert((SS_NORMAL == status) || (ERR_MUNOFINISH == status));
util_out_print("Region !AD : MUPIP REORG !AD incomplete. See above messages.!/",
TRUE, REG_LEN_STR(reg), LEN_AND_STR(command));
status = status1;
}
}
if (NULL != bptr)
free(bptr);
if (NULL != bml_lcl_buff)
free(bml_lcl_buff);
if (mu_ctrly_occurred || mu_ctrlc_occurred)
{
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_REORGCTRLY);
status = ERR_MUNOFINISH;
}
mupip_exit(status);
}
i8_p que_obj_next_freeblk(que_obj_p p_obj)
{
return BLK_ADDR(p_obj->tail, p_obj);
}
void dse_chng_bhead(void)
{
block_id blk;
block_id *blkid_ptr;
sgm_info *dummysi = NULL;
int4 x;
cache_rec_ptr_t cr;
uchar_ptr_t bp;
sm_uc_ptr_t blkBase;
blk_hdr new_hdr;
blk_segment *bs1, *bs_ptr;
cw_set_element *cse;
int4 blk_seg_cnt, blk_size; /* needed for BLK_INIT,BLK_SEG and BLK_FINI macros */
bool ismap;
bool chng_blk;
uint4 mapsize;
uint4 jnl_status;
error_def(ERR_DSEBLKRDFAIL);
error_def(ERR_DSEFAIL);
error_def(ERR_DBRDONLY);
if (gv_cur_region->read_only)
rts_error(VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
assert(update_array);
/* reset new block mechanism */
update_array_ptr = update_array;
chng_blk = FALSE;
if (cli_present("BLOCK") == CLI_PRESENT)
{
if (!cli_get_hex("BLOCK",&blk))
return;
if (blk < 0 || blk > cs_addrs->ti->total_blks)
{ util_out_print("Error: invalid block number.",TRUE);
return;
}
patch_curr_blk = blk;
}
blk_size = cs_addrs->hdr->blk_size;
ismap = (patch_curr_blk / cs_addrs->hdr->bplmap * cs_addrs->hdr->bplmap == patch_curr_blk);
mapsize = BM_SIZE(cs_addrs->hdr->bplmap);
t_begin_crit (ERR_DSEFAIL);
if (!(bp = t_qread (patch_curr_blk,&dummy_hist.h[0].cycle,&dummy_hist.h[0].cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
new_hdr = *(blk_hdr_ptr_t)bp;
if (cli_present("LEVEL") == CLI_PRESENT)
{
if (!cli_get_num("LEVEL",&x))
{
t_abort(gv_cur_region, cs_addrs);
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, cs_addrs);
return;
}
if (!ismap && (x < 0 || x > MAX_BT_DEPTH + 1))
{
util_out_print("Error: invalid level.",TRUE);
t_abort(gv_cur_region, cs_addrs);
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("BSIZ") == CLI_PRESENT)
{
if (!cli_get_hex("BSIZ",&x))
{
t_abort(gv_cur_region, cs_addrs);
return;
}
if (ismap && x != mapsize)
{
util_out_print("Error: invalid bsiz.",TRUE);
t_abort(gv_cur_region, cs_addrs);
return;
}
else if (x < sizeof(blk_hdr) || x > blk_size)
{
util_out_print("Error: invalid bsiz.",TRUE);
t_abort(gv_cur_region, cs_addrs);
return;
}
chng_blk = TRUE;
new_hdr.bsiz = x;
}
if (!chng_blk)
t_abort(gv_cur_region, cs_addrs);
else
{
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, bp + sizeof(new_hdr), new_hdr.bsiz - sizeof(new_hdr));
if (!BLK_FINI(bs_ptr, bs1))
{
util_out_print("Error: bad block build.",TRUE);
t_abort(gv_cur_region, cs_addrs);
return;
}
t_write (patch_curr_blk, (unsigned char *)bs1, 0, 0, bp, new_hdr.levl, TRUE, FALSE);
BUILD_AIMG_IF_JNL_ENABLED(cs_addrs, cs_data, non_tp_jfb_buff_ptr, cse);
t_end(&dummy_hist, 0);
}
if (cli_present("TN") == CLI_PRESENT)
{
if (!cli_get_hex("TN",&x))
return;
t_begin_crit(ERR_DSEFAIL);
assert(cs_addrs->ti->early_tn == cs_addrs->ti->curr_tn);
cs_addrs->ti->early_tn++;
blkBase = t_qread(patch_curr_blk, &dummy_hist.h[0].cycle, &dummy_hist.h[0].cr);
if (NULL == blkBase)
{
rel_crit(gv_cur_region);
util_out_print("Error: Unable to read buffer.", TRUE);
t_abort(gv_cur_region, cs_addrs);
return;
}
/* Create a null update array for a block */
if (ismap)
{
BLK_ADDR(blkid_ptr, sizeof(block_id), block_id);
*blkid_ptr = 0;
t_write_map(patch_curr_blk, blkBase, (unsigned char *)blkid_ptr, cs_addrs->ti->curr_tn);
cr_array_index = 0;
block_saved = FALSE;
} else
{
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, bp + sizeof(new_hdr), new_hdr.bsiz - sizeof(new_hdr));
BLK_FINI(bs_ptr, bs1);
t_write(patch_curr_blk, (unsigned char *)bs1, 0, 0, blkBase,
((blk_hdr_ptr_t)blkBase)->levl, TRUE, FALSE);
cr_array_index = 0;
block_saved = FALSE;
if (JNL_ENABLED(cs_data))
{
JNL_SHORT_TIME(jgbl.gbl_jrec_time); /* needed for jnl_put_jrt_pini() and jnl_write_aimg_rec() */
jnl_status = jnl_ensure_open();
if (0 == jnl_status)
{
cse = (cw_set_element *)(&cw_set[0]);
cse->new_buff = non_tp_jfb_buff_ptr;
gvcst_blk_build(cse, (uchar_ptr_t)cse->new_buff, x);
cse->done = TRUE;
if (0 == cs_addrs->jnl->pini_addr)
jnl_put_jrt_pini(cs_addrs);
jnl_write_aimg_rec(cs_addrs, cse->blk, (blk_hdr_ptr_t)cse->new_buff);
} else
rts_error(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(cs_data), DB_LEN_STR(gv_cur_region));
}
}
/* Pass the desired tn "x" as argument to bg_update or mm_update */
if (dba_bg == cs_addrs->hdr->acc_meth)
bg_update(cw_set, cw_set + cw_set_depth, cs_addrs->ti->curr_tn, x, dummysi);
else
mm_update(cw_set, cw_set + cw_set_depth, cs_addrs->ti->curr_tn, x, dummysi);
cs_addrs->ti->curr_tn++;
assert(cs_addrs->ti->early_tn == cs_addrs->ti->curr_tn);
/* the following code is analogous to that in t_end and should be maintained in a similar fashion */
while (cr_array_index)
cr_array[--cr_array_index]->in_cw_set = FALSE;
rel_crit(gv_cur_region);
if (block_saved)
backup_buffer_flush(gv_cur_region);
UNIX_ONLY(
if (unhandled_stale_timer_pop)
process_deferred_stale();
)
wcs_timer_start(gv_cur_region, TRUE);
}
