/*
get_lmap.c:
Reads local bit map and returns buffer address,
two bit local bit-map value corresponding to the block, cycle and cr
Input Parameter:
blk: block id of the block whose bit map this routine is to fetch
Output Parameter:
bits: two bit local bit map
cycle: Cycle value found in t_qread
cr: Cache Record value found in t_qread
Returns:
buffer address of local bitmap block
Null: if t_qread fails
*/
sm_uc_ptr_t get_lmap (block_id blk, unsigned char *bits, sm_int_ptr_t cycle, cache_rec_ptr_ptr_t cr)
{
sm_uc_ptr_t ptr, bp;
block_id index, offset;
error_def(ERR_DSEBLKRDFAIL);
index = ROUND_DOWN2(blk, BLKS_PER_LMAP);
offset = blk - index;
bp = t_qread (index, cycle, cr);
if (bp)
{
ptr = bp + SIZEOF(blk_hdr) + (offset * BML_BITS_PER_BLK) / 8;
*bits = *ptr;
switch (blk % (8 / BML_BITS_PER_BLK))
{ case 0: break;
case 1:
*bits = *bits >> BML_BITS_PER_BLK;
break;
case 2:
*bits = *bits >> 2 * BML_BITS_PER_BLK;
break;
case 3:
*bits = *bits >> 3 * BML_BITS_PER_BLK;
break;
}
*bits = *bits & 3;
}
return bp;
}
void dse_integ(void)
{
block_id blk;
char util_buff[MAX_UTIL_LEN];
sm_uc_ptr_t bp;
int4 dummy_int, nocrit_present;
cache_rec_ptr_t dummy_cr;
int util_len;
bool was_crit;
error_def(ERR_DSEBLKRDFAIL);
if (CLI_PRESENT == cli_present("BLOCK"))
{
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;
}
memcpy(util_buff, "!/Checking integrity of block ", 30);
util_len = 30;
util_len += i2hex_nofill(patch_curr_blk, (uchar_ptr_t)&util_buff[util_len], 8);
memcpy(&util_buff[util_len], ":", 1);
util_len += 1;
util_buff[util_len] = 0;
util_out_print(util_buff, TRUE);
was_crit = cs_addrs->now_crit;
nocrit_present = (CLI_NEGATED == cli_present("CRIT"));
if (!was_crit)
{
if (nocrit_present)
cs_addrs->now_crit = TRUE;
else
grab_crit(gv_cur_region);
}
if (!(bp = t_qread(patch_curr_blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (TRUE == cert_blk(gv_cur_region, patch_curr_blk, (blk_hdr_ptr_t)bp, 0))
util_out_print("!/ No errors detected.!/", TRUE);
else
util_out_print(NULL, TRUE);
if (!was_crit)
{
if (nocrit_present)
cs_addrs->now_crit = FALSE;
else
rel_crit(gv_cur_region);
}
return;
}
/* This routine returns whether the free_blocks counter in the file-header is ok (TRUE) or not (FALSE).
* If not, it corrects it. This assumes cs_addrs, cs_data and gv_cur_region to point to the region of interest.
* It also assumes that the master-map is correct and finds out non-full local bitmaps and counts the number of
* free blocks in each of them and sums them up to determine the perceived correct free_blocks count.
* The reason why this is ok is that even if the master-map incorrectly reports a local bitmap as full, our new free_blocks
* count will effectively make the free space in that local-bitmap invisible and make a gdsfilext necessary and valid.
* A later mupip integ will scavenge that invisible space for us. The worst that can therefore happen is that we will transiently
* not be using up existing space. But we will always ensure that the free_blocks counter goes in sync with the master-map.
*/
boolean_t is_free_blks_ctr_ok(void)
{
boolean_t blk_used;
block_id bml, free_bit, free_bml, maxbitsthismap;
cache_rec_ptr_t cr;
int cycle;
sm_uc_ptr_t bmp;
unsigned int local_maps, total_blks, free_blocks;
error_def(ERR_DBBADFREEBLKCTR);
assert(&FILE_INFO(gv_cur_region)->s_addrs == cs_addrs && cs_addrs->hdr == cs_data && cs_addrs->now_crit);
total_blks = (dba_mm == cs_data->acc_meth) ? cs_addrs->total_blks : cs_addrs->ti->total_blks;
local_maps = DIVIDE_ROUND_UP(total_blks, BLKS_PER_LMAP);
for (free_blocks = 0, free_bml = 0; free_bml < local_maps; free_bml++)
{
bml = bmm_find_free((uint4)free_bml, (sm_uc_ptr_t)MM_ADDR(cs_data), local_maps);
if (bml < free_bml)
break;
free_bml = bml;
bml *= BLKS_PER_LMAP;
if (!(bmp = t_qread(bml, (sm_int_ptr_t)&cycle, &cr))
|| (BM_SIZE(BLKS_PER_LMAP) != ((blk_hdr_ptr_t)bmp)->bsiz)
|| (LCL_MAP_LEVL != ((blk_hdr_ptr_t)bmp)->levl))
{
assert(FALSE); /* In pro, we will simply skip counting this local bitmap. */
continue;
}
assert(free_bml <= (local_maps - 1));
maxbitsthismap = (free_bml != (local_maps - 1)) ? BLKS_PER_LMAP : total_blks - bml;
for (free_bit = 0; free_bit < maxbitsthismap; free_bit++)
{
free_bit = bm_find_blk(free_bit, (sm_uc_ptr_t)bmp + sizeof(blk_hdr), maxbitsthismap, &blk_used);
assert(NO_FREE_SPACE <= free_bit);
if (0 > free_bit)
break;
free_blocks++;
}
}
assert(cs_addrs->ti->free_blocks == free_blocks);
if (cs_addrs->ti->free_blocks != free_blocks)
{
send_msg(VARLSTCNT(6) ERR_DBBADFREEBLKCTR, 4, DB_LEN_STR(gv_cur_region), cs_addrs->ti->free_blocks, free_blocks);
cs_addrs->ti->free_blocks = free_blocks;
return FALSE;
}
return TRUE;
}
void bm_setmap(block_id bml, block_id blk, int4 busy)
{
sm_uc_ptr_t bmp;
trans_num ctn;
srch_hist alt_hist;
srch_blk_status blkhist; /* block-history to fill in for t_write_map which uses "blk_num", "buffaddr", "cr", "cycle" */
cw_set_element *cse;
int lbm_status; /* local bitmap status of input "blk" i.e. BUSY or FREE or RECYCLED */
int4 reference_cnt;
uint4 bitnum;
error_def(ERR_DSEFAIL);
t_begin_crit(ERR_DSEFAIL);
ctn = cs_addrs->ti->curr_tn;
if (!(bmp = t_qread(bml, &blkhist.cycle, &blkhist.cr)))
t_retry((enum cdb_sc)rdfail_detail);
blkhist.blk_num = bml;
blkhist.buffaddr = bmp;
alt_hist.h[0].blk_num = 0; /* Need for calls to T_END for bitmaps */
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
bitnum = blk - bml;
/* Find out current status in order to determine if there is going to be a state transition */
assert(ROUND_DOWN2(blk, cs_data->bplmap) == bml);
GET_BM_STATUS(bmp, bitnum, lbm_status);
switch(lbm_status)
{
case BLK_BUSY:
reference_cnt = busy ? 0 : -1;
break;
case BLK_FREE:
case BLK_MAPINVALID:
case BLK_RECYCLED:
assert(BLK_MAPINVALID != lbm_status);
reference_cnt = busy ? 1 : 0;
break;
default:
assert(FALSE);
break;
}
if (reference_cnt)
{ /* Initialize update array with non-zero bitnum only if reference_cnt is non-zero. */
assert(bitnum);
*((block_id_ptr_t)update_array_ptr) = bitnum;
update_array_ptr += sizeof(block_id);
}
/* Terminate update array unconditionally with zero bitnum. */
*((block_id_ptr_t)update_array_ptr) = 0;
update_array_ptr += sizeof(block_id);
t_write_map(&blkhist, (uchar_ptr_t)update_array, ctn, reference_cnt);
if (JNL_ENABLED(cs_data))
{
cse = (cw_set_element *)(&cw_set[0]);
cse->new_buff = non_tp_jfb_buff_ptr;
memcpy(non_tp_jfb_buff_ptr, bmp, ((blk_hdr_ptr_t)bmp)->bsiz);
gvcst_map_build((uint4 *)cse->upd_addr, (uchar_ptr_t)cse->new_buff, cse, cs_addrs->ti->curr_tn);
cse->done = TRUE;
}
/* Call t_end till it succeeds or aborts (error will be reported) */
while ((trans_num)0 == t_end(&alt_hist, 0))
;
return;
}
boolean_t dse_b_dmp(void)
{
int4 util_len, head, lmap_num, iter1, iter2, mapsize, bplmap, nocrit_present, dummy_int, count;
unsigned char util_buff[MAX_UTIL_LEN], mask;
boolean_t free, was_crit, invalid_bitmap = FALSE;
block_id blk;
sm_uc_ptr_t bp, b_top, rp, mb, dump_record(sm_uc_ptr_t rp, block_id blk, sm_uc_ptr_t bp, sm_uc_ptr_t b_top);
cache_rec_ptr_t dummy_cr;
error_def(ERR_DSEBLKRDFAIL);
error_def(ERR_CTRLC);
error_def(ERR_BITMAPSBAD);
head = cli_present("HEADER");
if (CLI_PRESENT == cli_present("BLOCK"))
{
if (!cli_get_hex("BLOCK", &blk))
return FALSE;
if (blk < 0 || blk >= cs_addrs->ti->total_blks)
{
util_out_print("Error: invalid block number.", TRUE);
return FALSE;
}
patch_curr_blk = blk;
} else
blk = patch_curr_blk;
if (CLI_PRESENT == cli_present("COUNT"))
{
if (!cli_get_hex("COUNT", &count))
return FALSE;
if (count < 1)
return FALSE;
} else
count = 1;
util_out_print(0, TRUE);
bplmap = cs_addrs->hdr->bplmap;
mapsize = BM_SIZE(bplmap);
patch_rec_counter = 1;
was_crit = cs_addrs->now_crit;
nocrit_present = (CLI_NEGATED == cli_present("CRIT"));
if (!was_crit)
{
if (nocrit_present)
cs_addrs->now_crit = TRUE;
else
grab_crit(gv_cur_region);
}
for ( ; ; )
{
if (blk / bplmap * bplmap != blk)
{
if(!(bp = t_qread(blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (((blk_hdr_ptr_t) bp)->levl && patch_is_fdmp)
{
util_out_print("Error: cannot perform GLO/ZWR dump on index block.", TRUE);
if (!was_crit)
{
if (nocrit_present)
cs_addrs->now_crit = FALSE;
else
rel_crit(gv_cur_region);
}
return FALSE;
}
if (((blk_hdr_ptr_t) bp)->bsiz > cs_addrs->hdr->blk_size)
b_top = bp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t) bp)->bsiz < sizeof(blk_hdr))
b_top = bp + sizeof(blk_hdr);
else
b_top = bp + ((blk_hdr_ptr_t) bp)->bsiz;
if (CLI_NEGATED != head && !patch_is_fdmp)
{ memcpy(util_buff, "Block ", 6);
util_len = 6;
util_len += i2hex_nofill(blk, &util_buff[util_len], 8);
memcpy(&util_buff[util_len], " Size ", 8);
util_len += 8;
util_len += i2hex_nofill(((blk_hdr_ptr_t)bp)->bsiz, &util_buff[util_len], 4);
memcpy(&util_buff[util_len], " Level !UL TN ", 18);
util_len += 18;
util_len += i2hex_nofill(((blk_hdr_ptr_t)bp)->tn, &util_buff[util_len], 8);
memcpy(&util_buff[util_len], "!/", 2);
util_len += 2;
util_buff[util_len] = 0;
util_out_print(util_buff, TRUE, ((blk_hdr_ptr_t) bp)->levl );
}
rp = bp + sizeof(blk_hdr);
if (CLI_PRESENT != head && (!patch_is_fdmp || ((blk_hdr_ptr_t) bp)->levl == 0))
{
while (!util_interrupt && (rp = dump_record(rp, blk, bp, b_top)))
patch_rec_counter += 1;
}
if (util_interrupt)
{
if (!was_crit)
rel_crit(gv_cur_region);
rts_error(VARLSTCNT(1) ERR_CTRLC);
break;
}
if (CLI_NEGATED == head)
util_out_print(0, TRUE);
} else if (!patch_is_fdmp)
{
if(!(bp = t_qread(blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (CLI_NEGATED != head)
{
if (bplmap == 0)
{
memcpy(util_buff, "Block ", 6);
util_len = 6;
util_len += i2hex_nofill(blk, &util_buff[util_len], 8);
memcpy(&util_buff[util_len], " Size ", 8);
util_len += 8;
util_len += i2hex_nofill(mapsize, &util_buff[util_len], 4);
memcpy(&util_buff[util_len], " Master Status: Cannot Determine (bplmap == 0)!/", 50);
util_len += 50;
util_buff[util_len] = 0;
util_out_print(util_buff, TRUE );
} else
{
mb = cs_addrs->bmm + blk / (8 * bplmap);
lmap_num = blk / bplmap;
mask = 1 << ( lmap_num - lmap_num / 8 * 8);
free = mask & *mb;
memcpy(util_buff, "Block ", 6);
util_len = 6;
util_len += i2hex_nofill(blk, &util_buff[util_len], 8);
memcpy(&util_buff[util_len], " Size ", 7);
util_len += 7;
util_len += i2hex_nofill(((blk_hdr_ptr_t)bp)->bsiz, &util_buff[util_len], 4);
memcpy(&util_buff[util_len], " Level !SB TN ", 16);
util_len += 16;
util_len += i2hex_nofill(((blk_hdr_ptr_t)bp)->tn, &util_buff[util_len], 8);
util_buff[util_len] = 0;
util_out_print(util_buff, FALSE, ((blk_hdr_ptr_t) bp)->levl );
util_len = 0;
memcpy(&util_buff[util_len], " Master Status: !AD!/",23);
util_len = 23;
util_buff[util_len] = 0;
util_out_print(util_buff, TRUE, free ? 10 : 4, free ? "Free Space" : "Full");
}
}
if (CLI_PRESENT != head)
{
util_out_print (" !_Low order High order", TRUE);
lmap_num = 0;
while (lmap_num < bplmap)
{ memcpy(util_buff, "Block ", 6);
util_len = 6;
i2hex_blkfill(blk + lmap_num, &util_buff[util_len], 8);
util_len += 8;
memcpy(&util_buff[util_len], ":!_| ", 6);
util_len += 6;
util_buff[util_len] = 0;
util_out_print (util_buff, FALSE);
for (iter1 = 0; iter1 < 4; iter1++)
{
for (iter2 = 0; iter2 < 8; iter2++)
{
mask = dse_lm_blk_free(lmap_num * BML_BITS_PER_BLK, bp + sizeof(blk_hdr));
if (!mask)
util_out_print ("!AD", FALSE, 1, BUSY_CHAR);
else if (BLK_FREE == mask)
util_out_print ("!AD", FALSE, 1, FREE_CHAR);
else if (BLK_RECYCLED == mask)
util_out_print ("!AD", FALSE, 1, REUSABLE_CHAR);
else {
invalid_bitmap = TRUE;
util_out_print ("!AD", FALSE, 1, CORRUPT_CHAR);
}
if (++lmap_num >= bplmap)
break;
}
util_out_print (" ", FALSE);
if (lmap_num >= bplmap)
break;
}
util_out_print ("|", TRUE);
if (util_interrupt)
{
if (!was_crit)
rel_crit(gv_cur_region);
rts_error(VARLSTCNT(1) ERR_CTRLC);
}
}
util_out_print("!/'!AD' == BUSY '!AD' == FREE '!AD' == REUSABLE '!AD' == CORRUPT!/",
TRUE,1, BUSY_CHAR, 1, FREE_CHAR, 1, REUSABLE_CHAR, 1, CORRUPT_CHAR);
if (invalid_bitmap)
rts_error(VARLSTCNT(1) ERR_BITMAPSBAD);
}
}
count--;
if (count <= 0 || util_interrupt)
break;
blk++;
if (blk >= cs_addrs->ti->total_blks)
blk = 0;
}
patch_curr_blk = blk;
if (!was_crit)
{
if (nocrit_present)
cs_addrs->now_crit = FALSE;
else
rel_crit(gv_cur_region);
}
return TRUE;
}
int dse_ksrch(block_id srch,
block_id_ptr_t pp,
int4 *off,
char *targ_key,
int targ_len)
{
cache_rec_ptr_t dummy_cr;
int rsize, tmp_cmpc;
int4 cmp, dummy_int;
ssize_t size;
sm_uc_ptr_t blk_id, bp, b_top, key_top, rp, r_top;
unsigned short cc, dummy_short;
if(!(bp = t_qread(srch, &dummy_int, &dummy_cr)))
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (((blk_hdr_ptr_t) bp)->bsiz > cs_addrs->hdr->blk_size)
b_top = bp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t) bp)->bsiz < SIZEOF(blk_hdr))
b_top = bp + SIZEOF(blk_hdr);
else
b_top = bp + ((blk_hdr_ptr_t) bp)->bsiz;
CLEAR_DSE_COMPRESS_KEY;
*off = 0;
for (rp = bp + SIZEOF(blk_hdr); rp < b_top; rp = r_top)
{
*off = (int4)(rp - bp);
GET_SHORT(dummy_short, &((rec_hdr_ptr_t)rp)->rsiz);
rsize = dummy_short;
if (rsize < SIZEOF(rec_hdr))
r_top = rp + SIZEOF(rec_hdr);
else
r_top = rp + rsize;
if (r_top > b_top)
r_top = b_top;
if (r_top - rp < (((blk_hdr_ptr_t)bp)->levl ? SIZEOF(block_id) : MIN_DATA_SIZE) + SIZEOF(rec_hdr))
{
*pp = 0;
break;
}
for (key_top = rp + SIZEOF(rec_hdr); key_top < r_top ; )
if (!*key_top++ && !*key_top++)
break;
if (((blk_hdr_ptr_t)bp)->levl && key_top > (blk_id = r_top - SIZEOF(block_id)))
key_top = blk_id;
if (EVAL_CMPC((rec_hdr_ptr_t)rp) > patch_comp_count)
cc = patch_comp_count;
else
cc = EVAL_CMPC((rec_hdr_ptr_t)rp);
size = (ssize_t)(key_top - rp - SIZEOF(rec_hdr));
if (size > MAX_KEY_SZ - cc)
size = MAX_KEY_SZ - cc;
if (size < 0)
size = 0;
memcpy(&patch_comp_key[cc], rp + SIZEOF(rec_hdr), size);
patch_comp_count = (int)(cc + size);
GET_LONGP(pp, key_top);
cmp = memvcmp(targ_key, targ_len, &patch_comp_key[0], patch_comp_count);
if (0 > cmp)
break;
if (!cmp)
{
if (0 != ((blk_hdr_ptr_t)bp)->levl)
break;
if (patch_find_root_search)
{
for (key_top = rp + SIZEOF(rec_hdr); key_top < r_top; )
if (!*key_top++ && !*key_top++)
break;
GET_LONG(ksrch_root, key_top);
}
return TRUE;
}
}
patch_path_count++;
if (((blk_hdr_ptr_t) bp)->levl && *pp > 0 && *pp < cs_addrs->ti->total_blks && (*pp % cs_addrs->hdr->bplmap)
&& dse_ksrch(*pp, pp + 1, off + 1, targ_key, targ_len))
return TRUE;
return FALSE;
}
/* Performs a random traversal for the sampling methods */
enum cdb_sc mu_size_rand_traverse(double *r, double *a)
{
sm_uc_ptr_t pVal, pTop, pRec, pBlkBase;
block_id nBlkId;
block_id valBlk[MAX_RECS_PER_BLK]; /* valBlk[j] := value in j-th record of current block */
boolean_t is_mm;
cache_rec_ptr_t cr;
enum cdb_sc status;
int cycle;
int4 random;
int4 rCnt; /* number of entries in valBlk */
register gv_namehead *pTarg;
register srch_blk_status *pCurr;
register srch_hist *pTargHist;
trans_num tn;
unsigned char nLevl;
unsigned short nRecLen;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
is_mm = (dba_mm == cs_data->acc_meth);
pTarg = gv_target;
pTargHist = &gv_target->hist;
/* The following largely mimics gvcst_search/gvcst_search_blk */
nBlkId = pTarg->root;
tn = cs_addrs->ti->curr_tn;
if (NULL == (pBlkBase = t_qread(nBlkId, (sm_int_ptr_t)&cycle, &cr)))
return (enum cdb_sc)rdfail_detail;
nLevl = ((blk_hdr_ptr_t)pBlkBase)->levl;
if (MAX_BT_DEPTH < (int)nLevl)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_maxlvl;
}
if (0 == (int)nLevl)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_badlvl;
}
pTargHist->depth = (int)nLevl;
pCurr = &pTargHist->h[nLevl];
(pCurr + 1)->blk_num = 0;
pCurr->tn = tn;
pCurr->cycle = cycle;
pCurr->cr = cr;
for (;;)
{
assert(pCurr->level == nLevl);
pCurr->cse = NULL;
pCurr->blk_num = nBlkId;
pCurr->buffaddr = pBlkBase;
BLK_LOOP(rCnt, pRec, pBlkBase, pTop, nRecLen)
{ /* enumerate records in block */
GET_AND_CHECK_RECLEN(status, nRecLen, pRec, pTop, nBlkId);
if (cdb_sc_normal != status)
{
assert(CDB_STAGNATE > t_tries);
return status;
}
valBlk[rCnt] = nBlkId;
CHECK_ADJACENCY(nBlkId, nLevl, a[nLevl]);
}
r[nLevl] = rCnt;
/* randomly select next block */
random = (int4)(rCnt * drand48());
random = random & 0x7fffffff; /* to make sure that the sign bit(msb) is off */
nBlkId = valBlk[random];
if (is_mm && (nBlkId > cs_addrs->total_blks))
{
if (cs_addrs->total_blks < cs_addrs->ti->total_blks)
return cdb_sc_helpedout;
else
return cdb_sc_blknumerr;
}
--pCurr; --nLevl;
if (nLevl < 1)
break;
pCurr->tn = cs_addrs->ti->curr_tn;
if (NULL == (pBlkBase = t_qread(nBlkId, (sm_int_ptr_t)&pCurr->cycle, &pCurr->cr)))
return (enum cdb_sc)rdfail_detail;
if (((blk_hdr_ptr_t)pBlkBase)->levl != nLevl)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_badlvl;
}
}
/* Finds a free block and adds information to update array and cw_set */
block_id swap_root_or_directory_block(int parent_blk_lvl, int child_blk_lvl, srch_hist *dir_hist_ptr, block_id child_blk_id,
sm_uc_ptr_t child_blk_ptr, kill_set *kill_set_list, trans_num curr_tn)
{
sgmnt_data_ptr_t csd;
sgmnt_addrs *csa;
node_local_ptr_t cnl;
srch_blk_status bmlhist, freeblkhist;
block_id hint_blk_num, free_blk_id, parent_blk_id;
boolean_t free_blk_recycled;
int4 master_bit, num_local_maps, free_bit, hint_bit, maxbitsthismap;
uint4 total_blks;
int blk_seg_cnt, blk_size;
sm_uc_ptr_t parent_blk_ptr, bn_ptr, saved_blk;
blk_segment *bs1, *bs_ptr;
int parent_blk_size, child_blk_size, bsiz;
int rec_size1, curr_offset, bpntr_end, hdr_len;
int tmp_cmpc;
cw_set_element *tmpcse;
jnl_buffer_ptr_t jbbp; /* jbbp is non-NULL only if before-image journaling */
unsigned short temp_ushort;
unsigned long temp_long;
unsigned char save_cw_set_depth;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
csd = cs_data;
csa = cs_addrs;
cnl = csa->nl;
blk_size = csd->blk_size;
/* Find a free/recycled block for new block location. */
hint_blk_num = 0;
total_blks = csa->ti->total_blks;
num_local_maps = DIVIDE_ROUND_UP(total_blks, BLKS_PER_LMAP);
master_bit = bmm_find_free((hint_blk_num / BLKS_PER_LMAP), csa->bmm, num_local_maps);
if ((NO_FREE_SPACE == master_bit))
{
t_abort(gv_cur_region, csa);
return ABORT_SWAP;
}
bmlhist.blk_num = (block_id)master_bit * BLKS_PER_LMAP;
if (NULL == (bmlhist.buffaddr = t_qread(bmlhist.blk_num, (sm_int_ptr_t)&bmlhist.cycle, &bmlhist.cr)))
{
assert(t_tries < CDB_STAGNATE);
t_retry((enum cdb_sc)rdfail_detail);
return RETRY_SWAP;
}
hint_bit = 0;
maxbitsthismap = (master_bit != (num_local_maps - 1)) ? BLKS_PER_LMAP : total_blks - bmlhist.blk_num;
free_bit = bm_find_blk(hint_bit, bmlhist.buffaddr + SIZEOF(blk_hdr), maxbitsthismap, &free_blk_recycled);
free_blk_id = bmlhist.blk_num + free_bit;
if (DIR_ROOT >= free_blk_id)
{ /* Bitmap block 0 and directory tree root block 1 should always be marked busy. */
assert(t_tries < CDB_STAGNATE);
t_retry(cdb_sc_badbitmap);
return RETRY_SWAP;
}
if (child_blk_id <= free_blk_id)
{ /* stop swapping root or DT blocks once the database is truncated well enough. A good heuristic for this is to check
* if the block is to be swapped into a higher block number and if so do not swap
*/
t_abort(gv_cur_region, csa);
return ABORT_SWAP;
}
/* ====== begin update array ======
* Four blocks get changed.
* 1. Free block becomes busy and gains the contents of child (root block/directory tree block)
* 2. Parent block in directory tree remains busy, but points to new root block location.
* 3. Free block's corresponding bitmap reflects above change.
* 4. Child block gets marked recycled in bitmap. (GVCST_BMP_MARK_FREE)
*/
parent_blk_ptr = dir_hist_ptr->h[parent_blk_lvl].buffaddr; /* parent_blk_lvl is 0 iff we're moving a gvt root block */
parent_blk_id = dir_hist_ptr->h[parent_blk_lvl].blk_num;
CHECK_AND_RESET_UPDATE_ARRAY;
if (free_blk_recycled)
{ /* Otherwise, it's a completely free block, in which case no need to read. */
freeblkhist.blk_num = (block_id)free_blk_id;
if (NULL == (freeblkhist.buffaddr = t_qread(free_blk_id, (sm_int_ptr_t)&freeblkhist.cycle, &freeblkhist.cr)))
{
assert(t_tries < CDB_STAGNATE);
t_retry((enum cdb_sc)rdfail_detail);
return RETRY_SWAP;
}
}
child_blk_size = ((blk_hdr_ptr_t)child_blk_ptr)->bsiz;
BLK_INIT(bs_ptr, bs1);
BLK_ADDR(saved_blk, child_blk_size, unsigned char);
memcpy(saved_blk, child_blk_ptr, child_blk_size);
BLK_SEG(bs_ptr, saved_blk + SIZEOF(blk_hdr), child_blk_size - SIZEOF(blk_hdr));
assert(blk_seg_cnt == child_blk_size);
if (!BLK_FINI(bs_ptr, bs1))
{
assert(t_tries < CDB_STAGNATE);
t_retry(cdb_sc_blkmod);
return RETRY_SWAP;
}
tmpcse = &cw_set[cw_set_depth];
(free_blk_recycled) ? BIT_SET_RECYCLED_AND_CLEAR_FREE(tmpcse->blk_prior_state)
: BIT_CLEAR_RECYCLED_AND_SET_FREE(tmpcse->blk_prior_state);
t_create(free_blk_id, (unsigned char *)bs1, 0, 0, child_blk_lvl);
tmpcse->mode = gds_t_acquired;
if (!free_blk_recycled || !cs_data->db_got_to_v5_once)
tmpcse->old_block = NULL;
else
{
tmpcse->old_block = freeblkhist.buffaddr;
tmpcse->cr = freeblkhist.cr;
tmpcse->cycle = freeblkhist.cycle;
jbbp = (JNL_ENABLED(csa) && csa->jnl_before_image) ? csa->jnl->jnl_buff : NULL;
if ((NULL != jbbp) && (((blk_hdr_ptr_t)tmpcse->old_block)->tn < jbbp->epoch_tn))
{
bsiz = ((blk_hdr_ptr_t)(tmpcse->old_block))->bsiz;
if (bsiz > blk_size)
{
assert(CDB_STAGNATE > t_tries);
t_retry(cdb_sc_lostbmlcr);
return RETRY_SWAP;
}
JNL_GET_CHECKSUM_ACQUIRED_BLK(tmpcse, csd, csa, tmpcse->old_block, bsiz);
}
}
/* 2. Parent block in directory tree remains busy, but points to new child block location. */
curr_offset = dir_hist_ptr->h[parent_blk_lvl].curr_rec.offset;
parent_blk_size = ((blk_hdr_ptr_t)parent_blk_ptr)->bsiz;
GET_RSIZ(rec_size1, (parent_blk_ptr + curr_offset));
if ((parent_blk_size < rec_size1 + curr_offset) || (BSTAR_REC_SIZE > rec_size1))
{
assert(t_tries < CDB_STAGNATE);
t_retry(cdb_sc_blkmod);
return RETRY_SWAP;
}
BLK_INIT(bs_ptr, bs1);
if (0 == parent_blk_lvl)
/* There can be collation stuff in the record value after the block pointer. See gvcst_root_search. */
hdr_len = SIZEOF(rec_hdr) + gv_altkey->end + 1 - EVAL_CMPC((rec_hdr_ptr_t)(parent_blk_ptr + curr_offset));
else
hdr_len = rec_size1 - SIZEOF(block_id);
bpntr_end = curr_offset + hdr_len + SIZEOF(block_id);
BLK_SEG(bs_ptr, parent_blk_ptr + SIZEOF(blk_hdr), curr_offset + hdr_len - SIZEOF(blk_hdr));
BLK_ADDR(bn_ptr, SIZEOF(block_id), unsigned char);
PUT_LONG(bn_ptr, free_blk_id);
BLK_SEG(bs_ptr, bn_ptr, SIZEOF(block_id));
BLK_SEG(bs_ptr, parent_blk_ptr + bpntr_end, parent_blk_size - bpntr_end);
assert(blk_seg_cnt == parent_blk_size);
if (!BLK_FINI(bs_ptr, bs1))
{
assert(t_tries < CDB_STAGNATE);
t_retry(cdb_sc_blkmod);
return RETRY_SWAP;
}
t_write(&dir_hist_ptr->h[parent_blk_lvl], (unsigned char *)bs1, 0, 0, parent_blk_lvl, FALSE, TRUE, GDS_WRITE_KILLTN);
/* To indicate later snapshot file writing process during fast_integ not to skip writing the block to snapshot file */
BIT_SET_DIR_TREE(cw_set[cw_set_depth-1].blk_prior_state);
/* 3. Free block's corresponding bitmap reflects above change. */
PUT_LONG(update_array_ptr, free_bit);
save_cw_set_depth = cw_set_depth; /* Bit maps go on end of cw_set (more fake acquired) */
assert(!cw_map_depth);
t_write_map(&bmlhist, (uchar_ptr_t)update_array_ptr, curr_tn, 1);
cw_map_depth = cw_set_depth;
cw_set_depth = save_cw_set_depth;
update_array_ptr += SIZEOF(block_id);
temp_long = 0;
PUT_LONG(update_array_ptr, temp_long);
update_array_ptr += SIZEOF(block_id);
assert(1 == cw_set[cw_map_depth - 1].reference_cnt);
/* 4. Child block gets marked recycled in bitmap. (GVCST_BMP_MARK_FREE) */
kill_set_list->blk[kill_set_list->used].flag = 0;
kill_set_list->blk[kill_set_list->used].level = 0;
kill_set_list->blk[kill_set_list->used++].block = child_blk_id;
return free_blk_id;
}
void dse_range(void)
{
char lower[256], targ_key[256], upper[256], util_buff[MAX_UTIL_LEN];
block_id from, to, blk, blk_child;
sm_uc_ptr_t bp, b_top, key_bot, key_top, key_top1, rp, r_top;
char level;
int4 dummy_int, nocrit_present;
cache_rec_ptr_t dummy_cr;
short int rsize, size, size1;
int cnt, dummy, lower_len, util_len, upper_len;
boolean_t busy_matters, free, got_lonely_star, index, low, lost, star, up, was_crit, was_hold_onto_crit;
if (cli_present("FROM") == CLI_PRESENT)
{
if (!cli_get_hex("FROM", (uint4 *)&from))
return;
if (from < 0 || from > cs_addrs->ti->total_blks
|| !(from % cs_addrs->hdr->bplmap))
{
util_out_print("Error: invalid block number.", TRUE);
return;
}
}
else
from = 1;
if (cli_present("TO") == CLI_PRESENT)
{
if(!cli_get_hex("TO", (uint4 *)&to))
return;
if (to < 0 || to > cs_addrs->ti->total_blks
|| !(to % cs_addrs->hdr->bplmap))
{
util_out_print("Error: invalid block number.", TRUE);
return;
}
}
else
to = cs_addrs->ti->total_blks - 1;
if (low = (cli_present("LOWER") == CLI_PRESENT))
{
if (!dse_getki(&lower[0], &lower_len, LIT_AND_LEN("LOWER")))
return;
}
if (up = (cli_present("UPPER") == CLI_PRESENT))
{
if (!dse_getki(&upper[0], &upper_len, LIT_AND_LEN("UPPER")))
return;
}
star = (cli_present("STAR") == CLI_PRESENT);
if (!low && !up && !star)
{
util_out_print("Must specify star, or a lower or upper key limit.", TRUE);
return;
}
index = (cli_present("INDEX") == CLI_PRESENT);
lost = (cli_present("LOST") == CLI_PRESENT);
dummy = cli_present("BUSY");
if (dummy == CLI_PRESENT)
{
busy_matters = TRUE;
free = FALSE;
}
else if (dummy == CLI_NEGATED)
busy_matters = free = TRUE;
else
busy_matters = free = FALSE;
patch_path[0] = get_dir_root();
cnt = 0;
was_crit = cs_addrs->now_crit;
nocrit_present = (CLI_NEGATED == cli_present("CRIT"));
DSE_GRAB_CRIT_AS_APPROPRIATE(was_crit, was_hold_onto_crit, nocrit_present, cs_addrs, gv_cur_region);
for (blk = from; blk <= to ;blk++)
{
if (util_interrupt)
{
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, was_hold_onto_crit, nocrit_present, cs_addrs, gv_cur_region);
rts_error(VARLSTCNT(1) ERR_CTRLC);
break;
}
if (!(blk % cs_addrs->hdr->bplmap))
continue;
if (!(bp = t_qread(blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
level = ((blk_hdr_ptr_t)bp)->levl;
if (index && (level == 0))
continue;
if (busy_matters && (free != dse_is_blk_free(blk, &dummy_int, &dummy_cr)))
continue;
if (((blk_hdr_ptr_t) bp)->bsiz > cs_addrs->hdr->blk_size)
b_top = bp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t) bp)->bsiz < SIZEOF(blk_hdr))
b_top = bp + SIZEOF(blk_hdr);
else
b_top = bp + ((blk_hdr_ptr_t) bp)->bsiz;
rp = bp + SIZEOF(blk_hdr);
GET_SHORT(rsize, &((rec_hdr_ptr_t) rp)->rsiz);
if (rsize < SIZEOF(rec_hdr))
r_top = rp + SIZEOF(rec_hdr);
else
r_top = rp + rsize;
if (r_top >= b_top)
r_top = b_top;
got_lonely_star = FALSE;
if (((blk_hdr_ptr_t) bp)->levl)
{
key_top = r_top - SIZEOF(block_id);
if (star && (r_top == b_top))
got_lonely_star = TRUE;
} else
{
if (!up && !low)
continue;
for (key_top = rp + SIZEOF(rec_hdr); key_top < r_top ; )
if (!*key_top++ && !*key_top++)
break;
}
if (!got_lonely_star)
{
key_bot = rp + SIZEOF(rec_hdr);
size = key_top - key_bot;
if (size <= 0)
continue;
if (size > SIZEOF(targ_key))
size = SIZEOF(targ_key);
if (lost)
{
for (key_top1 = rp + SIZEOF(rec_hdr); key_top1 < r_top ; )
if (!*key_top1++)
break;
size1 = key_top1 - rp - SIZEOF(rec_hdr);
if (size1 > SIZEOF(targ_key))
size1 = SIZEOF(targ_key);
patch_find_root_search = TRUE;
patch_path_count = 1;
patch_find_blk = blk;
if (dse_is_blk_in(rp, r_top, size1))
continue;
}
if (low && memcmp(lower, key_bot, MIN(lower_len, size)) > 0)
continue;
if (up && memcmp(upper, key_bot, MIN(upper_len, size)) < 0)
continue;
} else
{
got_lonely_star = FALSE;
if (lost)
{
blk_child = *(block_id_ptr_t)key_top;
if (!(bp = t_qread(blk_child, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (((blk_hdr_ptr_t) bp)->bsiz > cs_addrs->hdr->blk_size)
b_top = bp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t) bp)->bsiz < SIZEOF(blk_hdr))
b_top = bp + SIZEOF(blk_hdr);
else
b_top = bp + ((blk_hdr_ptr_t) bp)->bsiz;
rp = bp + SIZEOF(blk_hdr);
GET_SHORT(rsize, &((rec_hdr_ptr_t) rp)->rsiz);
if (rsize < SIZEOF(rec_hdr))
r_top = rp + SIZEOF(rec_hdr);
else
r_top = rp + rsize;
if (r_top >= b_top)
r_top = b_top;
if (((blk_hdr_ptr_t) bp)->levl)
key_top = r_top - SIZEOF(block_id);
for (key_top1 = rp + SIZEOF(rec_hdr); key_top1 < r_top ; )
if (!*key_top1++)
break;
size1 = key_top1 - rp - SIZEOF(rec_hdr);
if (size1 > 0)
{
if (size1 > SIZEOF(targ_key))
size1 = SIZEOF(targ_key);
patch_find_root_search = TRUE;
patch_path_count = 1;
patch_find_blk = blk;
if (dse_is_blk_in(rp, r_top, size1))
continue;
}
}
}
if (!cnt++)
util_out_print("!/Blocks in the specified key range:", TRUE);
util_out_print("Block: !8XL Level: !2UL", TRUE, blk, level);
}
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, was_hold_onto_crit, nocrit_present, cs_addrs, gv_cur_region);
if (cnt)
util_out_print("Found !UL blocks", TRUE, cnt);
else
util_out_print("None found.", TRUE);
return;
}
void dse_exhaus(int4 pp, int4 op)
{
sm_uc_ptr_t bp, np, b_top, rp, r_top, nrp, nr_top, ptr;
char util_buff[MAX_UTIL_LEN];
block_id last;
short temp_short;
int count, util_len;
int4 dummy_int;
cache_rec_ptr_t dummy_cr;
global_dir_path *d_ptr, *temp;
error_def(ERR_DSEBLKRDFAIL);
error_def(ERR_CTRLC);
last = 0;
patch_path_count++;
if(!(bp = t_qread(patch_path[pp - 1],&dummy_int,&dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (((blk_hdr_ptr_t) bp)->bsiz > cs_addrs->hdr->blk_size)
b_top = bp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t) bp)->bsiz < sizeof(blk_hdr))
b_top = bp + sizeof(blk_hdr);
else
b_top = bp + ((blk_hdr_ptr_t) bp)->bsiz;
for (rp = bp + sizeof(blk_hdr); rp < b_top ;rp = r_top)
{
if (util_interrupt)
{
rts_error(VARLSTCNT(1) ERR_CTRLC);
break;
}
if (!(np = t_qread(patch_path[pp - 1],&dummy_int,&dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (np != bp)
{
b_top = np + (b_top - bp);
rp = np + (rp - bp);
r_top = np + (r_top - bp);
bp = np;
}
GET_SHORT(temp_short,&((rec_hdr_ptr_t) rp)->rsiz);
r_top = rp + temp_short;
if (r_top > b_top)
r_top = b_top;
if (r_top - rp < sizeof(block_id))
break;
if (((blk_hdr_ptr_t)bp)->levl)
GET_LONG(patch_path[pp],(r_top - sizeof(block_id)));
else
{
for (ptr = rp + sizeof(rec_hdr); ; )
{
if (*ptr++ == 0 && *ptr++ ==0)
break;
}
GET_LONG(patch_path[pp],ptr);
}
patch_offset[op] = rp - bp;
if (patch_path[pp] == patch_find_blk)
{
if (!patch_exh_found)
{
if (patch_find_sibs)
util_out_print("!/ Left siblings Right siblings",TRUE);
patch_exh_found = TRUE;
}
if (patch_find_sibs)
{
patch_left_sib = last;
if (r_top < b_top)
{
nrp = r_top;
GET_SHORT(temp_short,&((rec_hdr_ptr_t) rp)->rsiz);
nr_top = nrp + temp_short;
if (nr_top > b_top)
nr_top = b_top;
if (nr_top - nrp >= sizeof(block_id))
{
if (((blk_hdr_ptr_t)bp)->levl)
GET_LONG(patch_right_sib,(nr_top - sizeof(block_id)));
else
{
for (ptr = rp + sizeof(rec_hdr); ;)
{
if (*ptr++ == 0 && *ptr++ == 0)
break;
}
GET_LONG(patch_right_sib,ptr);
}
}
}
else
patch_right_sib = 0;
if (patch_left_sib)
{
memcpy(util_buff," ",1);
util_len = 1;
util_len += i2hex_nofill(patch_left_sib,(uchar_ptr_t)&util_buff[util_len],8);
memcpy(&util_buff[util_len]," ",1);
util_len += 1;
util_buff[util_len] = 0;
util_out_print(util_buff,FALSE);
}else
util_out_print(" none ",FALSE);
if (patch_right_sib)
{
memcpy(util_buff," ",1);
util_len = 1;
util_len += i2hex_nofill(patch_right_sib,(uchar_ptr_t)&util_buff[util_len],8);
util_buff[util_len] = 0;
util_out_print(util_buff,FALSE);
}
else
util_out_print(" none",TRUE);
}
else /* !patch_find_sibs */
{
patch_path_count--;
util_out_print(" Directory path!/ Path--blk:off",TRUE);
if (!patch_find_root_search)
{
d_ptr = global_roots_head->link->dir_path;
while(d_ptr)
{
memcpy(util_buff," ",1);
util_len = 1;
util_len += i2hex_nofill(d_ptr->block,(uchar_ptr_t)&util_buff[util_len],8);
memcpy(&util_buff[util_len],":",1);
util_len += 1;
util_len += i2hex_nofill(d_ptr->offset,(uchar_ptr_t)&util_buff[util_len],4);
util_buff[util_len] = 0;
util_out_print(util_buff,FALSE);
temp = d_ptr;
d_ptr = d_ptr->next;
free(temp);
}
global_roots_head->link->dir_path = 0;
util_out_print("!/!/ Global paths!/ Path--blk:off",TRUE);
}
for (count = 0; count < patch_path_count ;count++)
{
memcpy(util_buff," ",1);
util_len = 1;
util_len += i2hex_nofill(patch_path[count],(uchar_ptr_t)&util_buff[util_len],8);
memcpy(&util_buff[util_len],":",1);
util_len += 1;
util_len += i2hex_nofill(patch_offset[count],(uchar_ptr_t)&util_buff[util_len],4);
util_buff[util_len] = 0;
util_out_print(util_buff,FALSE);
}
memcpy(util_buff," ",1);
util_len = 1;
util_len += i2hex_nofill(patch_path[count],(uchar_ptr_t)&util_buff[util_len],8);
util_buff[util_len] = 0;
util_out_print(util_buff,TRUE);
patch_path_count++;
}
}
if (patch_path[pp] > 0 && patch_path[pp] < cs_addrs->ti->total_blks
&& (patch_path[pp] % cs_addrs->hdr->bplmap))
if (((blk_hdr_ptr_t) bp)->levl > 1)
dse_exhaus(pp + 1,op + 1);
else if (((blk_hdr_ptr_t)bp)->levl == 1 && patch_find_root_search)
dse_find_roots(patch_path[pp]);
last = patch_path[pp];
}
patch_path_count--;
return;
}
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);
}
/******************************************************************************************
Input Parameters:
level: level of working block
dest_blk_id: last destination used for swap
Output Parameters:
kill_set_ptr: Kill set to be freed
*exclude_glist_ptr: List of globals not to be moved for a swap destination
Input/Output Parameters:
gv_target : as working block's history
reorg_gv_target->hist : as desitnitions block's history
******************************************************************************************/
enum cdb_sc mu_swap_blk(int level, block_id *pdest_blk_id, kill_set *kill_set_ptr, glist *exclude_glist_ptr)
{
unsigned char x_blk_lmap;
unsigned short temp_ushort;
int rec_size1, rec_size2;
int wlevel, nslevel, dest_blk_level;
int piece_len1, piece_len2, first_offset, second_offset,
work_blk_size, work_parent_size, dest_blk_size, dest_parent_size;
int dest_child_cycle;
int blk_seg_cnt, blk_size;
trans_num ctn;
int key_len, key_len_dir;
block_id dest_blk_id, work_blk_id, child1, child2;
enum cdb_sc status;
srch_hist *dest_hist_ptr, *dir_hist_ptr;
cache_rec_ptr_t dest_child_cr;
blk_segment *bs1, *bs_ptr;
sm_uc_ptr_t saved_blk, work_blk_ptr, work_parent_ptr, dest_parent_ptr, dest_blk_ptr,
bn_ptr, bmp_buff, tblk_ptr, rec_base, rPtr1;
boolean_t gbl_target_was_set, blk_was_free, deleted;
gv_namehead *save_targ;
srch_blk_status bmlhist, destblkhist, *hist_ptr;
unsigned char save_cw_set_depth;
cw_set_element *tmpcse;
jnl_buffer_ptr_t jbbp; /* jbbp is non-NULL only if before-image journaling */
unsigned int bsiz;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
dest_blk_id = *pdest_blk_id;
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
if (NULL == TREF(gv_reorgkey))
GVKEY_INIT(TREF(gv_reorgkey), DBKEYSIZE(MAX_KEY_SZ));
dest_hist_ptr = &(reorg_gv_target->hist);
dir_hist_ptr = reorg_gv_target->alt_hist;
blk_size = cs_data->blk_size;
work_parent_ptr = gv_target->hist.h[level+1].buffaddr;
work_parent_size = ((blk_hdr_ptr_t)work_parent_ptr)->bsiz;
work_blk_ptr = gv_target->hist.h[level].buffaddr;
work_blk_size = ((blk_hdr_ptr_t)work_blk_ptr)->bsiz;
work_blk_id = gv_target->hist.h[level].blk_num;
if (blk_size < work_blk_size)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
cws_reorg_remove_index = 0;
/*===== Infinite loop to find the destination block =====*/
for ( ; ; )
{
blk_was_free = FALSE;
INCR_BLK_NUM(dest_blk_id);
/* A Pre-order traversal should not cause a child block to go to its parent.
* However, in case it happens because already the organization was like that or for any other reason, skip swap.
* If we decide to swap, code below should be changed to take care of the special case.
* Still a grand-child can go to its grand-parent. This is rare and following code can handle it.
*/
if (dest_blk_id == gv_target->hist.h[level+1].blk_num)
continue;
if (cs_data->trans_hist.total_blks <= dest_blk_id || dest_blk_id == work_blk_id)
{
*pdest_blk_id = dest_blk_id;
return cdb_sc_oprnotneeded;
}
ctn = cs_addrs->ti->curr_tn;
/* We need to save the block numbers that were NEWLY ADDED (since entering this function "mu_swap_blk")
* through the CWS_INSERT macro (in db_csh_get/db_csh_getn which can be called by t_qread or gvcst_search below).
* This is so that we can delete these blocks from the "cw_stagnate" hashtable in case we determine the need to
* choose a different "dest_blk_id" in this for loop (i.e. come to the next iteration). If these blocks are not
* deleted, then the hashtable will keep growing (a good example will be if -EXCLUDE qualifier is specified and
* a lot of prospective dest_blk_ids get skipped because they contain EXCLUDEd global variables) and very soon
* the hashtable will contain more entries than there are global buffers and at that point db_csh_getn will not
* be able to get a free global buffer for a new block (since it checks the "cw_stagnate" hashtable before reusing
* a buffer in case of MUPIP REORG). To delete these previous iteration blocks, we use the "cws_reorg_remove_array"
* variable. This array should have enough entries to accommodate the maximum number of blocks that can be t_qread
* in one iteration down below. And that number is the sum of
* + MAX_BT_DEPTH : for the t_qread while loop down the tree done below
* + 2 * MAX_BT_DEPTH : for the two calls to gvcst_search done below
* + 2 : 1 for the t_qread of dest_blk_id and 1 more for the t_qread of a
* bitmap block done inside the call to get_lmap below
* = 3 * MAX_BT_DEPTH + 2
* To be safe, we give a buffer of MAX_BT_DEPTH elements i.e. (4 * MAX_BT_DEPTH) + 2.
* This is defined in the macro CWS_REMOVE_ARRAYSIZE in cws_insert.h
*/
/* reset whatever blocks the previous iteration of this for loop had filled in the cw_stagnate hashtable */
for ( ; cws_reorg_remove_index > 0; cws_reorg_remove_index--)
{
deleted = delete_hashtab_int4(&cw_stagnate, (uint4 *)&cws_reorg_remove_array[cws_reorg_remove_index]);
assert(deleted);
}
/* read corresponding bitmap block before attempting to read destination block.
* if bitmap indicates block is free, we will not read the destination block
*/
bmp_buff = get_lmap(dest_blk_id, &x_blk_lmap, (sm_int_ptr_t)&bmlhist.cycle, &bmlhist.cr);
if (!bmp_buff || BLK_MAPINVALID == x_blk_lmap ||
((blk_hdr_ptr_t)bmp_buff)->bsiz != BM_SIZE(BLKS_PER_LMAP) ||
((blk_hdr_ptr_t)bmp_buff)->levl != LCL_MAP_LEVL)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_badbitmap;
}
if (BLK_FREE != x_blk_lmap)
{ /* x_blk_lmap is either BLK_BUSY or BLK_RECYCLED. In either case, we need to read destination block
* in case we later detect that the before-image needs to be written.
*/
if (!(dest_blk_ptr = t_qread(dest_blk_id, (sm_int_ptr_t)&destblkhist.cycle, &destblkhist.cr)))
{
assert(t_tries < CDB_STAGNATE);
return (enum cdb_sc)rdfail_detail;
}
destblkhist.blk_num = dest_blk_id;
destblkhist.buffaddr = dest_blk_ptr;
destblkhist.level = dest_blk_level = ((blk_hdr_ptr_t)dest_blk_ptr)->levl;
}
if (BLK_BUSY != x_blk_lmap)
{ /* x_blk_map is either BLK_FREE or BLK_RECYCLED both of which mean the block is not used in the bitmap */
blk_was_free = TRUE;
break;
}
/* dest_blk_id might contain a *-record only.
* So follow the pointer to go to the data/index block, which has a non-* key to search.
*/
nslevel = dest_blk_level;
if (MAX_BT_DEPTH <= nslevel)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_maxlvl;
}
rec_base = dest_blk_ptr + SIZEOF(blk_hdr);
GET_RSIZ(rec_size1, rec_base);
tblk_ptr = dest_blk_ptr;
while ((BSTAR_REC_SIZE == rec_size1) && (0 != nslevel))
{
GET_LONG(child1, (rec_base + SIZEOF(rec_hdr)));
if (0 == child1 || child1 > cs_data->trans_hist.total_blks - 1)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_rdfail;
}
if (!(tblk_ptr = t_qread(child1, (sm_int_ptr_t)&dest_child_cycle, &dest_child_cr)))
{
assert(t_tries < CDB_STAGNATE);
return (enum cdb_sc)rdfail_detail;
}
/* leaf of a killed GVT can have block header only. Skip those blocks */
if (SIZEOF(blk_hdr) >= ((blk_hdr_ptr_t)tblk_ptr)->bsiz)
break;
nslevel--;
rec_base = tblk_ptr + SIZEOF(blk_hdr);
GET_RSIZ(rec_size1, rec_base);
}
/* leaf of a killed GVT can have block header only. Skip those blocks */
if (SIZEOF(blk_hdr) >= ((blk_hdr_ptr_t)tblk_ptr)->bsiz)
continue;
/* get length of global variable name (do not read subscript) for dest_blk_id */
GET_GBLNAME_LEN(key_len_dir, rec_base + SIZEOF(rec_hdr));
/* key_len = length of 1st key value (including subscript) for dest_blk_id */
GET_KEY_LEN(key_len, rec_base + SIZEOF(rec_hdr));
if ((1 >= key_len_dir || MAX_MIDENT_LEN + 1 < key_len_dir) || (2 >= key_len || MAX_KEY_SZ < key_len))
{ /* Earlier used to restart here always. But dest_blk_id can be a block,
* which is just killed and still marked busy. Skip it, if we are in last retry.
*/
if (CDB_STAGNATE <= t_tries)
continue;
else
return cdb_sc_blkmod;
}
memcpy(&((TREF(gv_reorgkey))->base[0]), rec_base + SIZEOF(rec_hdr), key_len_dir);
(TREF(gv_reorgkey))->base[key_len_dir] = 0;
(TREF(gv_reorgkey))->end = key_len_dir;
if (exclude_glist_ptr->next)
{ /* exclude blocks for globals in the list of EXCLUDE option */
if (in_exclude_list(&((TREF(gv_reorgkey))->base[0]), key_len_dir - 1, exclude_glist_ptr))
continue;
}
save_targ = gv_target;
if (INVALID_GV_TARGET != reset_gv_target)
gbl_target_was_set = TRUE;
else
{
gbl_target_was_set = FALSE;
reset_gv_target = save_targ;
}
gv_target = reorg_gv_target;
gv_target->root = cs_addrs->dir_tree->root;
gv_target->clue.end = 0;
/* assign Directory tree path to find dest_blk_id in dir_hist_ptr */
status = gvcst_search(TREF(gv_reorgkey), dir_hist_ptr);
if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);
RESET_GV_TARGET_LCL_AND_CLR_GBL(save_targ, DO_GVT_GVKEY_CHECK);
return status;
}
if (dir_hist_ptr->h[0].curr_rec.match != (TREF(gv_reorgkey))->end + 1)
{ /* may be in a kill_set of another process */
RESET_GV_TARGET_LCL_AND_CLR_GBL(save_targ, DO_GVT_GVKEY_CHECK);
continue;
}
for (wlevel = 0; wlevel <= dir_hist_ptr->depth &&
dir_hist_ptr->h[wlevel].blk_num != dest_blk_id; wlevel++);
if (dir_hist_ptr->h[wlevel].blk_num == dest_blk_id)
{ /* do not swap a dir_tree block */
RESET_GV_TARGET_LCL_AND_CLR_GBL(save_targ, DO_GVT_GVKEY_CHECK);
continue;
}
/* gv_reorgkey will now have the first key from dest_blk_id,
* or, from a descendant of dest_blk_id (in case it had a *-key only).
*/
memcpy(&((TREF(gv_reorgkey))->base[0]), rec_base + SIZEOF(rec_hdr), key_len);
(TREF(gv_reorgkey))->end = key_len - 1;
GET_KEY_LEN(key_len_dir, dir_hist_ptr->h[0].buffaddr + dir_hist_ptr->h[0].curr_rec.offset + SIZEOF(rec_hdr));
/* Get root of GVT for dest_blk_id */
GET_LONG(gv_target->root,
dir_hist_ptr->h[0].buffaddr + dir_hist_ptr->h[0].curr_rec.offset + SIZEOF(rec_hdr) + key_len_dir);
if ((0 == gv_target->root) || (gv_target->root > (cs_data->trans_hist.total_blks - 1)))
{
assert(t_tries < CDB_STAGNATE);
RESET_GV_TARGET_LCL_AND_CLR_GBL(save_targ, DO_GVT_GVKEY_CHECK);
return cdb_sc_blkmod;
}
/* Assign Global Variable Tree path to find dest_blk_id in dest_hist_ptr */
gv_target->clue.end = 0;
status = gvcst_search(TREF(gv_reorgkey), dest_hist_ptr);
RESET_GV_TARGET_LCL_AND_CLR_GBL(save_targ, DO_GVT_GVKEY_CHECK);
if (dest_blk_level >= dest_hist_ptr->depth || /* do not swap in root level */
dest_hist_ptr->h[dest_blk_level].blk_num != dest_blk_id) /* must be in a kill set of another process. */
continue;
if ((cdb_sc_normal != status) || (dest_hist_ptr->h[nslevel].curr_rec.match != ((TREF(gv_reorgkey))->end + 1)))
{
assert(t_tries < CDB_STAGNATE);
return (cdb_sc_normal != status ? status : cdb_sc_blkmod);
}
for (wlevel = nslevel; wlevel <= dest_blk_level; wlevel++)
dest_hist_ptr->h[wlevel].tn = ctn;
dest_blk_ptr = dest_hist_ptr->h[dest_blk_level].buffaddr;
dest_blk_size = ((blk_hdr_ptr_t)dest_blk_ptr)->bsiz;
dest_parent_ptr = dest_hist_ptr->h[dest_blk_level+1].buffaddr;
dest_parent_size = ((blk_hdr_ptr_t)dest_parent_ptr)->bsiz;
break;
}
/*===== End of infinite loop to find the destination block =====*/
/*-----------------------------------------------------
Now modify blocks for swapping. Maximum of 4 blocks.
-----------------------------------------------------*/
if (!blk_was_free)
{ /* 1: dest_blk_id into work_blk_id */
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, dest_blk_ptr + SIZEOF(blk_hdr), dest_blk_size - SIZEOF(blk_hdr));
if (!BLK_FINI (bs_ptr,bs1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
assert(gv_target->hist.h[level].blk_num == work_blk_id);
assert(gv_target->hist.h[level].buffaddr == work_blk_ptr);
t_write(&gv_target->hist.h[level], (unsigned char *)bs1, 0, 0, dest_blk_level, TRUE, TRUE, GDS_WRITE_KILLTN);
}
/* 2: work_blk_id into dest_blk_id */
if (!blk_was_free && work_blk_id == dest_hist_ptr->h[dest_blk_level+1].blk_num)
{ /* work_blk_id will be swapped with its child.
* This is the only vertical swap. Here working block goes to its child.
* Working block cannot goto its parent because of traversal
*/
if (dest_blk_level + 1 != level || dest_parent_size != work_blk_size)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
BLK_INIT(bs_ptr, bs1);
BLK_ADDR(saved_blk, dest_parent_size, unsigned char);
memcpy(saved_blk, dest_parent_ptr, dest_parent_size);
first_offset = dest_hist_ptr->h[dest_blk_level+1].curr_rec.offset;
GET_RSIZ(rec_size1, saved_blk + first_offset);
if (work_blk_size < first_offset + rec_size1)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
piece_len1 = first_offset + rec_size1;
BLK_SEG(bs_ptr, saved_blk + SIZEOF(blk_hdr), piece_len1 - SIZEOF(block_id) - SIZEOF(blk_hdr));
BLK_ADDR(bn_ptr, SIZEOF(block_id), unsigned char);
PUT_LONG(bn_ptr, work_blk_id); /* since work_blk_id will now be the child of dest_blk_id */
BLK_SEG(bs_ptr, bn_ptr, SIZEOF(block_id));
BLK_SEG(bs_ptr, saved_blk + piece_len1, dest_parent_size - piece_len1);
if (!BLK_FINI(bs_ptr, bs1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
assert(dest_blk_id == dest_hist_ptr->h[dest_blk_level].blk_num);
assert(dest_blk_ptr == dest_hist_ptr->h[dest_blk_level].buffaddr);
t_write(&dest_hist_ptr->h[dest_blk_level], (unsigned char *)bs1, 0, 0, level, TRUE, TRUE, GDS_WRITE_KILLTN);
} else /* free block or, when working block does not move vertically (swap with parent/child) */
{
BLK_INIT(bs_ptr, bs1);
BLK_ADDR(saved_blk, work_blk_size, unsigned char);
memcpy(saved_blk, work_blk_ptr, work_blk_size);
BLK_SEG(bs_ptr, saved_blk + SIZEOF(blk_hdr), work_blk_size - SIZEOF(blk_hdr));
if (!BLK_FINI(bs_ptr, bs1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
if (blk_was_free)
{
tmpcse = &cw_set[cw_set_depth];
t_create(dest_blk_id, (unsigned char *)bs1, 0, 0, level);
/* Although we invoked t_create, we do not want t_end to allocate the block (i.e. change mode
* from gds_t_create to gds_t_acquired). Instead we do that and a little more (that t_end does) all here.
*/
assert(dest_blk_id == tmpcse->blk);
tmpcse->mode = gds_t_acquired;
/* If snapshots are in progress, we might want to read the before images of the FREE blocks also.
* Since mu_swap_blk mimics a small part of t_end, it sets cse->mode to gds_t_acquired and hence
* will not read the before images of the FREE blocks in t_end. To workaround this, set
* cse->was_free to TRUE so that in t_end, this condition can be used to read the before images of
* the FREE blocks if needed.
*/
(BLK_FREE == x_blk_lmap) ? SET_FREE(tmpcse) : SET_NFREE(tmpcse);
/* No need to write before-image in case the block is FREE. In case the database had never been fully
* upgraded from V4 to V5 format (after the MUPIP UPGRADE), all RECYCLED blocks can basically be considered
* FREE (i.e. no need to write before-images since backward journal recovery will never be expected
* to take the database to a point BEFORE the mupip upgrade).
*/
if ((BLK_FREE == x_blk_lmap) || !cs_data->db_got_to_v5_once)
tmpcse->old_block = NULL;
else
{ /* Destination is a recycled block that needs a before image */
tmpcse->old_block = destblkhist.buffaddr;
/* Record cr,cycle. This is used later in t_end to determine if checksums need to be recomputed */
tmpcse->cr = destblkhist.cr;
tmpcse->cycle = destblkhist.cycle;
jbbp = (JNL_ENABLED(cs_addrs) && cs_addrs->jnl_before_image) ? cs_addrs->jnl->jnl_buff : NULL;
if ((NULL != jbbp) && (((blk_hdr_ptr_t)tmpcse->old_block)->tn < jbbp->epoch_tn))
{ /* Compute CHECKSUM for writing PBLK record before getting crit.
* It is possible that we are reading a block that is actually marked free in
* the bitmap (due to concurrency issues at this point). Therefore we might be
* actually reading uninitialized block headers and in turn a bad value of
* "old_block->bsiz". Restart if we ever access a buffer whose size is greater
* than the db block size.
*/
bsiz = ((blk_hdr_ptr_t)(tmpcse->old_block))->bsiz;
if (bsiz > blk_size)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_lostbmlcr;
}
JNL_GET_CHECKSUM_ACQUIRED_BLK(tmpcse, cs_data, cs_addrs, tmpcse->old_block, bsiz);
}
}
assert(GDSVCURR == tmpcse->ondsk_blkver); /* should have been set by t_create above */
} else
{
hist_ptr = &dest_hist_ptr->h[dest_blk_level];
assert(dest_blk_id == hist_ptr->blk_num);
assert(dest_blk_ptr == hist_ptr->buffaddr);
t_write(hist_ptr, (unsigned char *)bs1, 0, 0, level, TRUE, TRUE, GDS_WRITE_KILLTN);
}
}
if (!blk_was_free)
{ /* 3: Parent of destination block (may be parent of working block too) */
if (gv_target->hist.h[level+1].blk_num == dest_hist_ptr->h[dest_blk_level+1].blk_num)
{ /* dest parent == work_blk parent */
BLK_INIT(bs_ptr, bs1);
/* Interchange pointer to dest_blk_id and work_blk_id */
if (level != dest_blk_level ||
gv_target->hist.h[level+1].curr_rec.offset == dest_hist_ptr->h[level+1].curr_rec.offset)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
if (gv_target->hist.h[level+1].curr_rec.offset < dest_hist_ptr->h[level+1].curr_rec.offset)
{
first_offset = gv_target->hist.h[level+1].curr_rec.offset;
second_offset = dest_hist_ptr->h[level+1].curr_rec.offset;
} else
{
first_offset = dest_hist_ptr->h[level+1].curr_rec.offset;
second_offset = gv_target->hist.h[level+1].curr_rec.offset;
}
GET_RSIZ(rec_size1, dest_parent_ptr + first_offset);
GET_RSIZ(rec_size2, dest_parent_ptr + second_offset);
if (dest_parent_size < first_offset + rec_size1 ||
dest_parent_size < second_offset + rec_size2 ||
BSTAR_REC_SIZE >= rec_size1 || BSTAR_REC_SIZE > rec_size2)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
piece_len1 = first_offset + rec_size1 - SIZEOF(block_id);
piece_len2 = second_offset + rec_size2 - SIZEOF(block_id);
GET_LONG(child1, dest_parent_ptr + piece_len1);
GET_LONG(child2, dest_parent_ptr + piece_len2);
BLK_SEG(bs_ptr, dest_parent_ptr + SIZEOF(blk_hdr), piece_len1 - SIZEOF(blk_hdr));
BLK_ADDR(bn_ptr, SIZEOF(block_id), unsigned char);
PUT_LONG(bn_ptr, child2);
BLK_SEG(bs_ptr, bn_ptr, SIZEOF(block_id));
BLK_SEG(bs_ptr, dest_parent_ptr + first_offset + rec_size1,
second_offset + rec_size2 - SIZEOF(block_id) - first_offset - rec_size1);
BLK_ADDR(bn_ptr, SIZEOF(block_id), unsigned char);
PUT_LONG(bn_ptr, child1);
BLK_SEG(bs_ptr, bn_ptr, SIZEOF(block_id));
BLK_SEG(bs_ptr, dest_parent_ptr + second_offset + rec_size2,
dest_parent_size - second_offset - rec_size2);
if (!BLK_FINI(bs_ptr,bs1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
assert(level == dest_blk_level);
assert(dest_parent_ptr == dest_hist_ptr->h[level+1].buffaddr);
t_write(&dest_hist_ptr->h[level+1], (unsigned char *)bs1, 0, 0, level+1, FALSE, TRUE, GDS_WRITE_KILLTN);
} else if (work_blk_id != dest_hist_ptr->h[dest_blk_level+1].blk_num)
{ /* Destination block moved in the position of working block.
* So destination block's parent's pointer should be changed to work_blk_id
*/
BLK_INIT(bs_ptr, bs1);
GET_RSIZ(rec_size1, dest_parent_ptr + dest_hist_ptr->h[dest_blk_level+1].curr_rec.offset);
if (dest_parent_size < rec_size1 + dest_hist_ptr->h[dest_blk_level+1].curr_rec.offset ||
BSTAR_REC_SIZE > rec_size1)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
BLK_SEG (bs_ptr, dest_parent_ptr + SIZEOF(blk_hdr),
dest_hist_ptr->h[dest_blk_level+1].curr_rec.offset + rec_size1 - SIZEOF(blk_hdr) - SIZEOF(block_id));
BLK_ADDR(bn_ptr, SIZEOF(block_id), unsigned char);
PUT_LONG(bn_ptr, work_blk_id);
BLK_SEG(bs_ptr, bn_ptr, SIZEOF(block_id));
BLK_SEG(bs_ptr, dest_parent_ptr + dest_hist_ptr->h[dest_blk_level+1].curr_rec.offset + rec_size1,
dest_parent_size - dest_hist_ptr->h[dest_blk_level+1].curr_rec.offset - rec_size1);
if (!BLK_FINI(bs_ptr,bs1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
assert(dest_parent_ptr == dest_hist_ptr->h[dest_blk_level+1].buffaddr);
t_write(&dest_hist_ptr->h[dest_blk_level+1], (unsigned char *)bs1, 0, 0, dest_blk_level+1,
FALSE, TRUE, GDS_WRITE_KILLTN);
}
}
/* 4: Parent of working block, if different than destination's parent or, destination was a free block */
if (blk_was_free || gv_target->hist.h[level+1].blk_num != dest_hist_ptr->h[dest_blk_level+1].blk_num)
{ /* Parent block of working blk should correctly point the working block. Working block went to dest_blk_id */
GET_RSIZ(rec_size1, (work_parent_ptr + gv_target->hist.h[level+1].curr_rec.offset));
if (work_parent_size < rec_size1 + gv_target->hist.h[level+1].curr_rec.offset || BSTAR_REC_SIZE > rec_size1)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, work_parent_ptr + SIZEOF(blk_hdr),
gv_target->hist.h[level+1].curr_rec.offset + rec_size1 - SIZEOF(blk_hdr) - SIZEOF(block_id));
BLK_ADDR(bn_ptr, SIZEOF(block_id), unsigned char);
PUT_LONG(bn_ptr, dest_blk_id);
BLK_SEG(bs_ptr, bn_ptr, SIZEOF(block_id));
BLK_SEG(bs_ptr, work_parent_ptr + gv_target->hist.h[level+1].curr_rec.offset + rec_size1,
work_parent_size - gv_target->hist.h[level+1].curr_rec.offset - rec_size1);
if (!BLK_FINI(bs_ptr, bs1))
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_blkmod;
}
assert(gv_target->hist.h[level+1].buffaddr == work_parent_ptr);
t_write(&gv_target->hist.h[level+1], (unsigned char *)bs1, 0, 0, level+1, FALSE, TRUE, GDS_WRITE_KILLTN);
}
/* else already taken care of, when dest_blk_id moved */
if (blk_was_free)
{ /* A free/recycled block will become busy block.
* So the local bitmap must be updated.
* Local bit map block will be added in the list of update arrray for concurrency check and
* also the cw_set element will be created to mark the free/recycled block as free.
* kill_set_ptr will save the block which will become free.
*/
child1 = ROUND_DOWN2(dest_blk_id, BLKS_PER_LMAP); /* bit map block */
bmlhist.buffaddr = bmp_buff;
bmlhist.blk_num = child1;
child1 = dest_blk_id - child1;
assert(child1);
PUT_LONG(update_array_ptr, child1);
/* 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 (for "cw_set_depth") below.
*/
save_cw_set_depth = cw_set_depth;
assert(!cw_map_depth);
t_write_map(&bmlhist, (uchar_ptr_t)update_array_ptr, ctn, 1); /* will increment cw_set_depth */
cw_map_depth = cw_set_depth; /* set cw_map_depth to the latest cw_set_depth */
cw_set_depth = save_cw_set_depth; /* restore cw_set_depth */
/* t_write_map simulation end */
update_array_ptr += SIZEOF(block_id);
child1 = 0;
PUT_LONG(update_array_ptr, child1);
update_array_ptr += SIZEOF(block_id);
assert(1 == cw_set[cw_map_depth - 1].reference_cnt); /* 1 free block is now becoming BLK_USED in the bitmap */
/* working block will be removed */
kill_set_ptr->blk[kill_set_ptr->used].flag = 0;
kill_set_ptr->blk[kill_set_ptr->used].level = 0;
kill_set_ptr->blk[kill_set_ptr->used++].block = work_blk_id;
}
*pdest_blk_id = dest_blk_id;
return cdb_sc_normal;
}
void dse_maps(void)
{
block_id blk, bml_blk;
blk_segment *bs1, *bs_ptr;
int4 blk_seg_cnt, blk_size; /* needed for BLK_INIT, BLK_SEG and BLK_FINI macros */
sm_uc_ptr_t bp;
char util_buff[MAX_UTIL_LEN];
int4 bml_size, bml_list_size, blk_index, bml_index;
int4 total_blks, blks_in_bitmap;
int4 bplmap, dummy_int;
unsigned char *bml_list;
cache_rec_ptr_t cr, dummy_cr;
bt_rec_ptr_t btr;
int util_len;
uchar_ptr_t blk_ptr;
boolean_t was_crit;
uint4 jnl_status;
srch_blk_status blkhist;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
if (CLI_PRESENT == cli_present("BUSY") || CLI_PRESENT == cli_present("FREE") ||
CLI_PRESENT == cli_present("MASTER") || CLI_PRESENT == cli_present("RESTORE_ALL"))
{
if (gv_cur_region->read_only)
rts_error(VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
}
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
csa = cs_addrs;
assert(&FILE_INFO(gv_cur_region)->s_addrs == csa);
was_crit = csa->now_crit;
if (csa->critical)
crash_count = csa->critical->crashcnt;
csd = csa->hdr;
bplmap = csd->bplmap;
if (CLI_PRESENT == cli_present("BLOCK"))
{
if (!cli_get_hex("BLOCK", (uint4 *)&blk))
return;
if (blk < 0 || blk >= csa->ti->total_blks)
{
util_out_print("Error: invalid block number.", TRUE);
return;
}
patch_curr_blk = blk;
}
else
blk = patch_curr_blk;
if (CLI_PRESENT == cli_present("FREE"))
{
if (0 == bplmap)
{
util_out_print("Cannot perform map updates: bplmap field of file header is zero.", TRUE);
return;
}
if (blk / bplmap * bplmap == blk)
{
util_out_print("Cannot perform action on a map block.", TRUE);
return;
}
bml_blk = blk / bplmap * bplmap;
bm_setmap(bml_blk, blk, FALSE);
return;
}
if (CLI_PRESENT == cli_present("BUSY"))
{
if (0 == bplmap)
{
util_out_print("Cannot perform map updates: bplmap field of file header is zero.", TRUE);
return;
}
if (blk / bplmap * bplmap == blk)
{
util_out_print("Cannot perform action on a map block.", TRUE);
return;
}
bml_blk = blk / bplmap * bplmap;
bm_setmap(bml_blk, blk, TRUE);
return;
}
blk_size = csd->blk_size;
if (CLI_PRESENT == cli_present("MASTER"))
{
if (0 == bplmap)
{
util_out_print("Cannot perform maps updates: bplmap field of file header is zero.", TRUE);
return;
}
if (!was_crit)
grab_crit(gv_cur_region);
bml_blk = blk / bplmap * bplmap;
if (dba_mm == csd->acc_meth)
bp = MM_BASE_ADDR(csa) + (off_t)bml_blk * blk_size;
else
{
assert(dba_bg == csd->acc_meth);
if (!(bp = t_qread(bml_blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
}
if ((csa->ti->total_blks / bplmap) * bplmap == bml_blk)
total_blks = (csa->ti->total_blks - bml_blk);
else
total_blks = bplmap;
if (NO_FREE_SPACE == bml_find_free(0, bp + SIZEOF(blk_hdr), total_blks))
bit_clear(bml_blk / bplmap, csa->bmm);
else
bit_set(bml_blk / bplmap, csa->bmm);
if (bml_blk > csa->nl->highest_lbm_blk_changed)
csa->nl->highest_lbm_blk_changed = bml_blk;
if (!was_crit)
rel_crit(gv_cur_region);
return;
}
if (CLI_PRESENT == cli_present("RESTORE_ALL"))
{
if (0 == bplmap)
{
util_out_print("Cannot perform maps updates: bplmap field of file header is zero.", TRUE);
return;
}
total_blks = csa->ti->total_blks;
assert(ROUND_DOWN2(blk_size, 2 * SIZEOF(int4)) == blk_size);
bml_size = BM_SIZE(bplmap);
bml_list_size = (total_blks + bplmap - 1) / bplmap * bml_size;
bml_list = (unsigned char *)malloc(bml_list_size);
for (blk_index = 0, bml_index = 0; blk_index < total_blks; blk_index += bplmap, bml_index++)
bml_newmap((blk_hdr_ptr_t)(bml_list + bml_index * bml_size), bml_size, csa->ti->curr_tn);
if (!was_crit)
{
grab_crit(gv_cur_region);
csa->hold_onto_crit = TRUE; /* need to do this AFTER grab_crit */
}
blk = get_dir_root();
assert(blk < bplmap);
csa->ti->free_blocks = total_blks - DIVIDE_ROUND_UP(total_blks, bplmap);
bml_busy(blk, bml_list + SIZEOF(blk_hdr));
csa->ti->free_blocks = csa->ti->free_blocks - 1;
dse_m_rest(blk, bml_list, bml_size, &csa->ti->free_blocks, TRUE);
for (blk_index = 0, bml_index = 0; blk_index < total_blks; blk_index += bplmap, bml_index++)
{
t_begin_crit(ERR_DSEFAIL);
CHECK_TN(csa, csd, csd->trans_hist.curr_tn); /* can issue rts_error TNTOOLARGE */
CWS_RESET;
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
assert(csa->ti->early_tn == csa->ti->curr_tn);
blk_ptr = bml_list + bml_index * bml_size;
blkhist.blk_num = blk_index;
if (!(blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blk_ptr + SIZEOF(blk_hdr), bml_size - SIZEOF(blk_hdr));
BLK_FINI(bs_ptr, bs1);
t_write(&blkhist, (unsigned char *)bs1, 0, 0, LCL_MAP_LEVL, TRUE, FALSE, GDS_WRITE_KILLTN);
BUILD_AIMG_IF_JNL_ENABLED(csd, csa->ti->curr_tn);
t_end(&dummy_hist, NULL, csa->ti->curr_tn);
}
/* Fill in master map */
for (blk_index = 0, bml_index = 0; blk_index < total_blks; blk_index += bplmap, bml_index++)
{
blks_in_bitmap = (blk_index + bplmap <= total_blks) ? bplmap : total_blks - blk_index;
assert(1 < blks_in_bitmap); /* the last valid block in the database should never be a bitmap block */
if (NO_FREE_SPACE != bml_find_free(0, (bml_list + bml_index * bml_size) + SIZEOF(blk_hdr), blks_in_bitmap))
bit_set(blk_index / bplmap, csa->bmm);
else
bit_clear(blk_index / bplmap, csa->bmm);
if (blk_index > csa->nl->highest_lbm_blk_changed)
csa->nl->highest_lbm_blk_changed = blk_index;
}
if (!was_crit)
{
csa->hold_onto_crit = FALSE; /* need to do this before the rel_crit */
rel_crit(gv_cur_region);
}
if (unhandled_stale_timer_pop)
process_deferred_stale();
free(bml_list);
csd->kill_in_prog = csd->abandoned_kills = 0;
return;
}
MEMCPY_LIT(util_buff, "!/Block ");
util_len = SIZEOF("!/Block ") - 1;
util_len += i2hex_nofill(blk, (uchar_ptr_t)&util_buff[util_len], 8);
memcpy(&util_buff[util_len], " is marked !AD in its local bit map.!/",
SIZEOF(" is marked !AD in its local bit map.!/") - 1);
util_len += SIZEOF(" is marked !AD in its local bit map.!/") - 1;
util_buff[util_len] = 0;
if (!was_crit)
grab_crit(gv_cur_region);
util_out_print(util_buff, TRUE, 4, dse_is_blk_free(blk, &dummy_int, &dummy_cr) ? "free" : "busy");
if (!was_crit)
rel_crit(gv_cur_region);
return;
}
void dse_chng_bhead(void)
{
block_id blk;
int4 x;
trans_num tn;
cache_rec_ptr_t cr;
blk_hdr new_hdr;
blk_segment *bs1, *bs_ptr;
int4 blk_seg_cnt, blk_size; /* needed for BLK_INIT,BLK_SEG and BLK_FINI macros */
boolean_t ismap;
boolean_t chng_blk;
boolean_t was_crit;
boolean_t was_hold_onto_crit;
uint4 mapsize;
srch_blk_status blkhist;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
# ifdef GTM_CRYPT
int req_enc_blk_size;
int crypt_status;
blk_hdr_ptr_t bp, save_bp, save_old_block;
# endif
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));
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
chng_blk = FALSE;
csa = cs_addrs;
if (cli_present("BLOCK") == CLI_PRESENT)
{
if (!cli_get_hex("BLOCK", (uint4 *)&blk))
return;
if (blk < 0 || blk > csa->ti->total_blks)
{
util_out_print("Error: invalid block number.", TRUE);
return;
}
patch_curr_blk = blk;
}
csd = csa->hdr;
assert(csd == cs_data);
blk_size = csd->blk_size;
ismap = (patch_curr_blk / csd->bplmap * csd->bplmap == patch_curr_blk);
mapsize = BM_SIZE(csd->bplmap);
t_begin_crit(ERR_DSEFAIL);
blkhist.blk_num = patch_curr_blk;
if (!(blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
new_hdr = *(blk_hdr_ptr_t)blkhist.buffaddr;
if (cli_present("LEVEL") == CLI_PRESENT)
{
if (!cli_get_hex("LEVEL", (uint4 *)&x))
{
t_abort(gv_cur_region, csa);
return;
}
if (ismap && (unsigned char)x != LCL_MAP_LEVL)
{
util_out_print("Error: invalid level for a bit map block.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
if (!ismap && (x < 0 || x > MAX_BT_DEPTH + 1))
{
util_out_print("Error: invalid level.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
new_hdr.levl = (unsigned char)x;
chng_blk = TRUE;
if (new_hdr.bsiz < SIZEOF(blk_hdr))
new_hdr.bsiz = SIZEOF(blk_hdr);
if (new_hdr.bsiz > blk_size)
new_hdr.bsiz = blk_size;
}
if (cli_present("BSIZ") == CLI_PRESENT)
{
if (!cli_get_hex("BSIZ", (uint4 *)&x))
{
t_abort(gv_cur_region, csa);
return;
}
if (ismap && x != mapsize)
{
util_out_print("Error: invalid bsiz.", TRUE);
t_abort(gv_cur_region, csa);
return;
} else if (x < SIZEOF(blk_hdr) || x > blk_size)
{
util_out_print("Error: invalid bsiz.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
chng_blk = TRUE;
new_hdr.bsiz = x;
}
if (!chng_blk)
t_abort(gv_cur_region, csa);
else
{
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blkhist.buffaddr + SIZEOF(new_hdr), new_hdr.bsiz - SIZEOF(new_hdr));
if (!BLK_FINI(bs_ptr, bs1))
{
util_out_print("Error: bad block build.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
t_write(&blkhist, (unsigned char *)bs1, 0, 0, new_hdr.levl, TRUE, FALSE, GDS_WRITE_KILLTN);
BUILD_AIMG_IF_JNL_ENABLED(csd, non_tp_jfb_buff_ptr, csa->ti->curr_tn);
t_end(&dummy_hist, NULL, TN_NOT_SPECIFIED);
}
if (cli_present("TN") == CLI_PRESENT)
{
if (!cli_get_hex64("TN", &tn))
return;
was_crit = csa->now_crit;
t_begin_crit(ERR_DSEFAIL);
CHECK_TN(csa, csd, csd->trans_hist.curr_tn); /* can issue rts_error TNTOOLARGE */
assert(csa->ti->early_tn == csa->ti->curr_tn);
if (NULL == (blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
{
util_out_print("Error: Unable to read buffer.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
if (new_hdr.bsiz < SIZEOF(blk_hdr))
new_hdr.bsiz = SIZEOF(blk_hdr);
if (new_hdr.bsiz > blk_size)
new_hdr.bsiz = blk_size;
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blkhist.buffaddr + SIZEOF(new_hdr), new_hdr.bsiz - SIZEOF(new_hdr));
BLK_FINI(bs_ptr, bs1);
t_write(&blkhist, (unsigned char *)bs1, 0, 0,
((blk_hdr_ptr_t)blkhist.buffaddr)->levl, TRUE, FALSE, GDS_WRITE_KILLTN);
/* Pass the desired tn as argument to bg_update/mm_update below */
BUILD_AIMG_IF_JNL_ENABLED(csd, non_tp_jfb_buff_ptr, tn);
was_hold_onto_crit = csa->hold_onto_crit;
csa->hold_onto_crit = TRUE;
t_end(&dummy_hist, NULL, tn);
# ifdef GTM_CRYPT
if (csd->is_encrypted && (tn < csa->ti->curr_tn))
{ /* BG and db encryption is enabled and the DSE update caused the block-header to potentially have a tn
* that is LESS than what it had before. At this point, the global buffer (corresponding to blkhist.blk_num)
* reflects the contents of the block AFTER the dse update (bg_update would have touched this) whereas
* the corresponding encryption global buffer reflects the contents of the block BEFORE the update.
* Normally wcs_wtstart takes care of propagating the tn update from the regular global buffer to the
* corresponding encryption buffer. But if before it gets a chance, let us say a process goes to t_end
* as part of a subsequent transaction and updates this same block. Since the blk-hdr-tn potentially
* decreased, it is possible that the PBLK writing check (comparing blk-hdr-tn with the epoch_tn) decides
* to write a PBLK for this block (even though a PBLK was already written for this block as part of a
* previous DSE CHANGE -BL -TN in the same epoch). In this case, since the db is encrypted, the logic
* will assume there were no updates to this block since the last time wcs_wtstart updated the encryption
* buffer and therefore use that to write the pblk, which is incorrect since it does not yet contain the
* tn update. The consequence of this is would be writing an older before-image PBLK) record to the
* journal file. To prevent this situation, we update the encryption buffer here (before releasing crit)
* using logic like that in wcs_wtstart to ensure it is in sync with the regular global buffer.
* Note:
* Although we use cw_set[0] to access the global buffer corresponding to the block number being updated,
* cw_set_depth at this point is 0 because t_end resets it. This is considered safe since cw_set is a
* static array (as opposed to malloc'ed memory) and hence is always available and valid until it gets
* overwritten by subsequent updates.
*/
bp = (blk_hdr_ptr_t)GDS_ANY_REL2ABS(csa, cw_set[0].cr->buffaddr);
DBG_ENSURE_PTR_IS_VALID_GLOBUFF(csa, csd, (sm_uc_ptr_t)bp);
save_bp = (blk_hdr_ptr_t)GDS_ANY_ENCRYPTGLOBUF(bp, csa);
DBG_ENSURE_PTR_IS_VALID_ENCTWINGLOBUFF(csa, csd, (sm_uc_ptr_t)save_bp);
assert((bp->bsiz <= csd->blk_size) && (bp->bsiz >= SIZEOF(*bp)));
req_enc_blk_size = MIN(csd->blk_size, bp->bsiz) - SIZEOF(*bp);
if (BLK_NEEDS_ENCRYPTION(bp->levl, req_enc_blk_size))
{
ASSERT_ENCRYPTION_INITIALIZED;
memcpy(save_bp, bp, SIZEOF(blk_hdr));
GTMCRYPT_ENCODE_FAST(csa->encr_key_handle, (char *)(bp + 1), req_enc_blk_size,
(char *)(save_bp + 1), crypt_status);
if (0 != crypt_status)
GC_GTM_PUTMSG(crypt_status, gv_cur_region->dyn.addr->fname);
} else
memcpy(save_bp, bp, bp->bsiz);
}
# endif
if (!was_hold_onto_crit)
csa->hold_onto_crit = FALSE;
if (!was_crit)
rel_crit(gv_cur_region);
if (unhandled_stale_timer_pop)
process_deferred_stale();
}
return;
}
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);
}
void dse_rmrec(void)
{
block_id blk;
blk_segment *bs1, *bs_ptr;
int4 blk_seg_cnt, blk_size, count;
sm_uc_ptr_t bp;
uchar_ptr_t lbp, b_top, rp, r_top, key_top, rp_base;
char cc, comp_key[256], cc_base;
short int size, i, rsize;
cw_set_element *cse;
error_def(ERR_DSEFAIL);
error_def(ERR_DSEBLKRDFAIL);
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;
if (cli_present("BLOCK") == CLI_PRESENT)
{
if(!cli_get_hex("BLOCK", &blk))
return;
if (blk < 0 || blk >= cs_addrs->ti->total_blks || !(blk % cs_addrs->hdr->bplmap))
{
util_out_print("Error: invalid block number.", TRUE);
return;
}
patch_curr_blk = blk;
}
if (cli_present("COUNT") == CLI_PRESENT)
{
if (!cli_get_hex("COUNT", &count) || count < 1)
return;
} else
count = 1;
t_begin_crit(ERR_DSEFAIL);
blk_size = cs_addrs->hdr->blk_size;
if(!(bp = t_qread(patch_curr_blk, &dummy_hist.h[0].cycle, &dummy_hist.h[0].cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
lbp = (uchar_ptr_t)malloc(blk_size);
memcpy(lbp, bp, blk_size);
if (((blk_hdr_ptr_t)lbp)->bsiz > cs_addrs->hdr->blk_size)
b_top = lbp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t)lbp)->bsiz < sizeof(blk_hdr))
b_top = lbp + sizeof(blk_hdr);
else
b_top = lbp + ((blk_hdr_ptr_t)lbp)->bsiz;
if (cli_present("RECORD") == CLI_PRESENT)
{
if (!(rp = rp_base = skan_rnum(lbp, FALSE)))
{
free(lbp);
t_abort(gv_cur_region, cs_addrs);
return;
}
} else if (!(rp = rp_base = skan_offset(lbp, FALSE)))
{
free(lbp);
t_abort(gv_cur_region, cs_addrs);
return;
}
memcpy(&comp_key[0], &patch_comp_key[0], sizeof(patch_comp_key));
cc_base = patch_comp_count;
for ( ; ; )
{
GET_SHORT(rsize, &((rec_hdr_ptr_t)rp)->rsiz);
if (rsize < sizeof(rec_hdr))
r_top = rp + sizeof(rec_hdr);
else
r_top = rp + rsize;
if (r_top >= b_top)
{
if (count)
{ if (((blk_hdr_ptr_t) lbp)->levl)
util_out_print("Warning: removed a star record from the end of this block.", TRUE);
((blk_hdr_ptr_t)lbp)->bsiz = rp_base - lbp;
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, (uchar_ptr_t)lbp + sizeof(blk_hdr),
(int)((blk_hdr_ptr_t)lbp)->bsiz - sizeof(blk_hdr));
if (!BLK_FINI(bs_ptr, bs1))
{
util_out_print("Error: bad blk build.",TRUE);
free(lbp);
t_abort(gv_cur_region, cs_addrs);
return;
}
t_write(patch_curr_blk, (unsigned char *)bs1, 0, 0, bp, ((blk_hdr_ptr_t)lbp)->levl, TRUE, FALSE);
BUILD_AIMG_IF_JNL_ENABLED(cs_addrs, cs_data, non_tp_jfb_buff_ptr, cse);
t_end(&dummy_hist, 0);
free(lbp);
return;
}
r_top = b_top;
}
if (((blk_hdr_ptr_t)lbp)->levl)
key_top = r_top - sizeof(block_id);
else
{
for (key_top = rp + sizeof(rec_hdr); key_top < r_top; )
if (!*key_top++ && !*key_top++)
break;
}
if (((rec_hdr_ptr_t)rp)->cmpc > patch_comp_count)
cc = patch_comp_count;
else
cc = ((rec_hdr_ptr_t)rp)->cmpc;
size = key_top - rp - sizeof(rec_hdr);
if (size < 0)
size = 0;
else if (size > sizeof(patch_comp_key) - 2)
size = sizeof(patch_comp_key) - 2;
memcpy(&patch_comp_key[cc], rp + sizeof(rec_hdr), size);
patch_comp_count = cc + size;
if (--count >= 0)
{
rp = r_top;
continue;
}
size = (patch_comp_count < cc_base) ? patch_comp_count : cc_base;
for (i = 0; i < size && patch_comp_key[i] == comp_key[i]; i++)
;
((rec_hdr_ptr_t)rp_base)->cmpc = i;
rsize = r_top - key_top + sizeof(rec_hdr) + patch_comp_count - i;
PUT_SHORT(&((rec_hdr_ptr_t)rp_base)->rsiz, rsize);
memcpy(rp_base + sizeof(rec_hdr), &patch_comp_key[i], patch_comp_count - i);
memcpy(rp_base + sizeof(rec_hdr) + patch_comp_count - i, key_top, b_top - key_top);
((blk_hdr_ptr_t)lbp)->bsiz = rp_base + rsize - lbp + b_top - r_top;
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, (uchar_ptr_t)lbp + sizeof(blk_hdr), ((blk_hdr_ptr_t)lbp)->bsiz - sizeof(blk_hdr));
if (!BLK_FINI(bs_ptr, bs1))
{
util_out_print("Error: bad blk build.", TRUE);
free(lbp);
t_abort(gv_cur_region, cs_addrs);
return;
}
t_write(patch_curr_blk, (unsigned char *)bs1, 0, 0, bp, ((blk_hdr_ptr_t)lbp)->levl, TRUE, FALSE);
BUILD_AIMG_IF_JNL_ENABLED(cs_addrs, cs_data, non_tp_jfb_buff_ptr, cse);
t_end(&dummy_hist, 0);
free(lbp);
return;
}
}
trans_num gvcst_bmp_mark_free(kill_set *ks)
{
block_id bit_map, next_bm, *updptr;
blk_ident *blk, *blk_top, *nextblk;
trans_num ctn, start_db_fmt_tn;
unsigned int len;
int4 blk_prev_version;
srch_hist alt_hist;
trans_num ret_tn = 0;
boolean_t visit_blks;
srch_blk_status bmphist;
cache_rec_ptr_t cr;
enum db_ver ondsk_blkver;
error_def(ERR_GVKILLFAIL);
assert(inctn_bmp_mark_free_gtm == inctn_opcode || inctn_bmp_mark_free_mu_reorg == inctn_opcode);
/* Note down the desired_db_format_tn before you start relying on cs_data->fully_upgraded.
* If the db is fully_upgraded, take the optimal path that does not need to read each block being freed.
* But in order to detect concurrent desired_db_format changes, note down the tn (when the last format change occurred)
* before the fully_upgraded check and after having noted down the database current_tn.
* If they are the same, then we are guaranteed no concurrent desired_db_format change occurred.
* If they are not, then fall through to the non-optimal path where each to-be-killed block has to be visited.
* The reason we need to visit every block in case desired_db_format changes is to take care of the case where
* MUPIP REORG DOWNGRADE concurrently changes a block that we are about to free.
*/
start_db_fmt_tn = cs_data->desired_db_format_tn;
visit_blks = (!cs_data->fully_upgraded); /* Local evaluation */
assert(!visit_blks || (visit_blks && dba_bg == cs_addrs->hdr->acc_meth)); /* must have blks_to_upgrd == 0 for non-BG */
assert(!dollar_tlevel); /* Should NOT be in TP now */
blk = &ks->blk[0];
blk_top = &ks->blk[ks->used];
if (!visit_blks)
{ /* Database has been completely upgraded. Free all blocks in one bitmap as part of one transaction. */
assert(cs_data->db_got_to_v5_once); /* assert all V4 fmt blocks (including RECYCLED) have space for V5 upgrade */
inctn_detail.blknum_struct.blknum = 0; /* to indicate no adjustment to "blks_to_upgrd" necessary */
for ( ; blk < blk_top; blk = nextblk)
{
if (0 != blk->flag)
{
nextblk = blk + 1;
continue;
}
assert(0 < blk->block);
assert((int4)blk->block < cs_addrs->ti->total_blks);
bit_map = ROUND_DOWN2((int)blk->block, BLKS_PER_LMAP);
next_bm = bit_map + BLKS_PER_LMAP;
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
/* Scan for the next local bitmap */
updptr = (block_id *)update_array_ptr;
for (nextblk = blk;
(0 == nextblk->flag) && (nextblk < blk_top) && ((block_id)nextblk->block < next_bm);
++nextblk)
{
assert((block_id)nextblk->block - bit_map);
*updptr++ = (block_id)nextblk->block - bit_map;
}
len = (unsigned int)((char *)nextblk - (char *)blk);
update_array_ptr = (char *)updptr;
alt_hist.h[0].blk_num = 0; /* need for calls to T_END for bitmaps */
/* the following assumes SIZEOF(blk_ident) == SIZEOF(int) */
assert(SIZEOF(blk_ident) == SIZEOF(int));
*(int *)update_array_ptr = 0;
t_begin(ERR_GVKILLFAIL, UPDTRNS_DB_UPDATED_MASK);
for (;;)
{
ctn = cs_addrs->ti->curr_tn;
/* Need a read fence before reading fields from cs_data as we are reading outside
* of crit and relying on this value to detect desired db format state change.
*/
SHM_READ_MEMORY_BARRIER;
if (start_db_fmt_tn != cs_data->desired_db_format_tn)
{ /* Concurrent db format change has occurred. Need to visit every block to be killed
* to determine its block format. Fall through to the non-optimal path below
*/
ret_tn = 0;
break;
}
bmphist.blk_num = bit_map;
if (NULL == (bmphist.buffaddr = t_qread(bmphist.blk_num, (sm_int_ptr_t)&bmphist.cycle,
&bmphist.cr)))
{
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
t_write_map(&bmphist, (uchar_ptr_t)update_array, ctn, -(int4)(nextblk - blk));
if ((trans_num)0 == (ret_tn = t_end(&alt_hist, NULL, TN_NOT_SPECIFIED)))
continue;
break;
}
if (0 == ret_tn) /* db format change occurred. Fall through to below for loop to visit each block */
break;
}
} /* for all blocks in the kill_set */
boolean_t dse_r_dmp(void)
{
block_id blk;
sm_uc_ptr_t bp, b_top, rp;
int4 count;
int4 dummy_int;
cache_rec_ptr_t dummy_cr;
short record, size;
boolean_t was_crit;
int4 nocrit_present;
error_def(ERR_DSEBLKRDFAIL);
error_def(ERR_CTRLC);
if (cli_present("BLOCK") == CLI_PRESENT)
{
uint4 tmp_blk;
if(!cli_get_hex("BLOCK", &tmp_blk))
return FALSE;
blk = (block_id)tmp_blk;
if (blk < 0 || blk >= cs_addrs->ti->total_blks || !(blk % cs_addrs->hdr->bplmap))
{
util_out_print("Error: invalid block number.", TRUE);
return FALSE;
}
patch_curr_blk = blk;
}
if (cli_present("COUNT") == CLI_PRESENT)
{
if (!cli_get_hex("COUNT", (uint4 *)&count))
return FALSE;
} else
count = 1;
was_crit = cs_addrs->now_crit;
nocrit_present = (CLI_NEGATED == cli_present("CRIT"));
DSE_GRAB_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
if (!(bp = t_qread(patch_curr_blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (((blk_hdr_ptr_t) bp)->bsiz > cs_addrs->hdr->blk_size)
b_top = bp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t) bp)->bsiz < SIZEOF(blk_hdr))
b_top = bp + SIZEOF(blk_hdr);
else
b_top = bp + ((blk_hdr_ptr_t) bp)->bsiz;
if (((blk_hdr_ptr_t) bp)->levl && patch_is_fdmp)
{
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
util_out_print("Error: cannot perform GLO/ZWR dump on index block.", TRUE);
return FALSE;
}
if (cli_present("RECORD") == CLI_PRESENT)
{
if (!(rp = skan_rnum (bp, FALSE)))
{
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
return FALSE;
}
} else if (!(rp = skan_offset (bp, FALSE)))
{
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
return FALSE;
}
util_out_print(0, TRUE);
for ( ; 0 < count; count--)
{
if (util_interrupt || !(rp = dump_record(rp, patch_curr_blk, bp, b_top)))
break;
patch_rec_counter += 1;
}
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
if (util_interrupt)
rts_error(VARLSTCNT(1) ERR_CTRLC);
else if (cli_present("HEADER") == CLI_NEGATED)
util_out_print(0, TRUE);
return TRUE;
}
void dse_chng_bhead(void)
{
blk_hdr new_hdr;
blk_segment *bs1, *bs_ptr;
block_id blk;
boolean_t chng_blk, ismap, was_hold_onto_crit;
int4 blk_seg_cnt, blk_size; /* needed for BLK_INIT,BLK_SEG and BLK_FINI macros */
int4 x;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
srch_blk_status blkhist;
trans_num tn;
uint4 mapsize;
csa = cs_addrs;
if (gv_cur_region->read_only)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
chng_blk = FALSE;
if (BADDSEBLK == (blk = dse_getblk("BLOCK", DSEBMLOK, DSEBLKCUR))) /* WARNING: assignment */
return;
csd = csa->hdr;
assert(csd == cs_data);
blk_size = csd->blk_size;
ismap = IS_BITMAP_BLK(blk);
mapsize = BM_SIZE(csd->bplmap);
t_begin_crit(ERR_DSEFAIL);
blkhist.blk_num = blk;
if (!(blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(1) ERR_DSEBLKRDFAIL);
new_hdr = *(blk_hdr_ptr_t)blkhist.buffaddr;
if (CLI_PRESENT == cli_present("LEVEL"))
{
if (!cli_get_hex("LEVEL", (uint4 *)&x))
{
t_abort(gv_cur_region, csa);
return;
}
if (ismap && (unsigned char)x != LCL_MAP_LEVL)
{
util_out_print("Error: invalid level for a bit map block.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
if (!ismap && (x < 0 || x > MAX_BT_DEPTH + 1))
{
util_out_print("Error: invalid level.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
new_hdr.levl = (unsigned char)x;
chng_blk = TRUE;
if (new_hdr.bsiz < SIZEOF(blk_hdr))
new_hdr.bsiz = SIZEOF(blk_hdr);
if (new_hdr.bsiz > blk_size)
new_hdr.bsiz = blk_size;
}
if (CLI_PRESENT == cli_present("BSIZ"))
{
if (!cli_get_hex("BSIZ", (uint4 *)&x))
{
t_abort(gv_cur_region, csa);
return;
}
if (ismap && x != mapsize)
{
util_out_print("Error: invalid bsiz.", TRUE);
t_abort(gv_cur_region, csa);
return;
} else if (x < SIZEOF(blk_hdr) || x > blk_size)
{
util_out_print("Error: invalid bsiz.", TRUE);
t_abort(gv_cur_region, csa);
return;
}
chng_blk = TRUE;
new_hdr.bsiz = x;
}
if (!chng_blk)
t_abort(gv_cur_region, csa);
else
{
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blkhist.buffaddr + SIZEOF(new_hdr), new_hdr.bsiz - SIZEOF(new_hdr));
if (!BLK_FINI(bs_ptr, bs1))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_AIMGBLKFAIL, 3, blk, DB_LEN_STR(gv_cur_region));
t_abort(gv_cur_region, csa);
return;
}
t_write(&blkhist, (unsigned char *)bs1, 0, 0, new_hdr.levl, TRUE, FALSE, GDS_WRITE_KILLTN);
BUILD_AIMG_IF_JNL_ENABLED(csd, csa->ti->curr_tn);
t_end(&dummy_hist, NULL, TN_NOT_SPECIFIED);
}
if (CLI_PRESENT == cli_present("TN"))
{
if (!cli_get_hex64("TN", &tn))
return;
t_begin_crit(ERR_DSEFAIL);
CHECK_TN(csa, csd, csd->trans_hist.curr_tn); /* can issue rts_error TNTOOLARGE */
assert(csa->ti->early_tn == csa->ti->curr_tn);
if (NULL == (blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(1) ERR_DSEBLKRDFAIL);
t_abort(gv_cur_region, csa);
return;
}
if (new_hdr.bsiz < SIZEOF(blk_hdr))
new_hdr.bsiz = SIZEOF(blk_hdr);
if (new_hdr.bsiz > blk_size)
new_hdr.bsiz = blk_size;
BLK_INIT(bs_ptr, bs1);
BLK_SEG(bs_ptr, blkhist.buffaddr + SIZEOF(new_hdr), new_hdr.bsiz - SIZEOF(new_hdr));
BLK_FINI(bs_ptr, bs1);
t_write(&blkhist, (unsigned char *)bs1, 0, 0,
((blk_hdr_ptr_t)blkhist.buffaddr)->levl, TRUE, FALSE, GDS_WRITE_KILLTN);
/* Pass the desired tn as argument to bg_update/mm_update below */
BUILD_AIMG_IF_JNL_ENABLED_AND_T_END_WITH_EFFECTIVE_TN(csa, csd, tn, &dummy_hist);
}
return;
}
void dse_f_free(void)
{
block_id blk;
bool in_last_bmap;
char util_buff[MAX_UTIL_LEN];
sm_uc_ptr_t lmap_base;
int4 bplmap, total_blks;
int4 util_len, master_bit, lmap_bit, hint_over_bplmap, hint_mod_bplmap;
boolean_t was_crit;
int4 dummy_int, nocrit_present;
cache_rec_ptr_t dummy_cr;
error_def(ERR_DSEBLKRDFAIL);
if (cs_addrs->hdr->bplmap == 0)
{ util_out_print("Cannot perform free block search: bplmap field of file header is zero.", TRUE);
return;
}
bplmap = cs_addrs->hdr->bplmap;
if(!cli_get_hex("HINT", (uint4 *)&blk))
return;
if (blk < 0 || blk >= cs_addrs->ti->total_blks || (blk / bplmap * bplmap == blk))
{ util_out_print("Error: invalid block number.", TRUE);
return;
}
hint_over_bplmap = blk / bplmap;
master_bit = bmm_find_free(hint_over_bplmap, cs_addrs->bmm,
(cs_addrs->ti->total_blks + bplmap - 1)/ bplmap);
if (master_bit == -1)
{ util_out_print("Error: database full.", TRUE);
return;
}
in_last_bmap = (master_bit == (cs_addrs->ti->total_blks / bplmap));
was_crit = cs_addrs->now_crit;
nocrit_present = (CLI_NEGATED == cli_present("CRIT"));
DSE_GRAB_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
if(!(lmap_base = t_qread(master_bit * bplmap, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (master_bit == hint_over_bplmap)
hint_mod_bplmap = blk - blk / bplmap * bplmap;
else
hint_mod_bplmap = 0;
if (in_last_bmap)
total_blks = (cs_addrs->ti->total_blks - master_bit);
else
total_blks = bplmap;
lmap_bit = bml_find_free(hint_mod_bplmap, lmap_base + SIZEOF(blk_hdr), total_blks);
if (lmap_bit == -1)
{ memcpy(util_buff, "Error: bit map in block ", 24);
util_len = 24;
util_len += i2hex_nofill(master_bit * bplmap, (uchar_ptr_t)&util_buff[util_len], 8);
memcpy(&util_buff[util_len], " incorrectly marked free in master map.", 39);
util_len += 39;
util_buff[util_len] = 0;
util_out_print(util_buff, TRUE);
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
return;
}
memcpy(util_buff, "!/Next free block is ", 21);
util_len = 21;
util_len += i2hex_nofill(master_bit * bplmap + lmap_bit, (uchar_ptr_t)&util_buff[util_len], 8);
memcpy(&util_buff[util_len], ".!/", 3);
util_len += 3;
util_buff[util_len] = 0;
util_out_print(util_buff, TRUE);
DSE_REL_CRIT_AS_APPROPRIATE(was_crit, nocrit_present, cs_addrs, gv_cur_region);
return;
}
/*
* Performs a random traversal for the sampling methods
*/
enum cdb_sc rand_traverse(double *r)
{
sm_uc_ptr_t pVal, pTop, pRec, pBlkBase;
register gv_namehead *pTarg;
register srch_blk_status *pCurr;
register srch_hist *pTargHist;
block_id nBlkId;
block_id valBlk[MAX_RECS_PER_BLK]; /* valBlk[j] := value in j-th record of current block */
unsigned char nLevl;
cache_rec_ptr_t cr;
int cycle;
trans_num tn;
sm_uc_ptr_t buffaddr;
unsigned short nRecLen;
uint4 tmp;
boolean_t is_mm;
int4 random;
int4 rCnt; /* number of entries in valBlk */
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
is_mm = (dba_mm == cs_data->acc_meth);
pTarg = gv_target;
pTargHist = &gv_target->hist;
/* The following largely mimics gvcst_search/gvcst_search_blk */
nBlkId = pTarg->root;
tn = cs_addrs->ti->curr_tn;
if (NULL == (pBlkBase = t_qread(nBlkId, (sm_int_ptr_t)&cycle, &cr)))
return (enum cdb_sc)rdfail_detail;
nLevl = ((blk_hdr_ptr_t)pBlkBase)->levl;
if (MAX_BT_DEPTH < (int)nLevl)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_maxlvl;
}
if (0 == (int)nLevl)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_badlvl;
}
pTargHist->depth = (int)nLevl;
pCurr = &pTargHist->h[nLevl];
(pCurr + 1)->blk_num = 0;
pCurr->tn = tn;
pCurr->cycle = cycle;
pCurr->cr = cr;
for (;;)
{
assert(pCurr->level == nLevl);
pCurr->cse = NULL;
pCurr->blk_num = nBlkId;
pCurr->buffaddr = pBlkBase;
for ( rCnt = 0, pRec = pBlkBase + SIZEOF(blk_hdr), pTop = pBlkBase + ((blk_hdr_ptr_t)pBlkBase)->bsiz;
pRec != pTop && rCnt < MAX_RECS_PER_BLK;
rCnt++, pRec += nRecLen )
{ /* enumerate records in block */
GET_USHORT(nRecLen, &((rec_hdr_ptr_t)pRec)->rsiz);
pVal = pRec + nRecLen - SIZEOF(block_id);
if (nRecLen == 0)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_badoffset;
}
if (pRec + nRecLen > pTop)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_blklenerr;
}
GET_LONG(tmp, pVal);
valBlk[rCnt] = tmp;
}
r[nLevl] = rCnt;
/* randomly select next block */
random = (int4)(rCnt * drand48());
random = random & 0x7fffffff; /* to make sure that the sign bit(msb) is off */
nBlkId = valBlk[random];
if (is_mm && (nBlkId > cs_addrs->total_blks))
{
if (cs_addrs->total_blks < cs_addrs->ti->total_blks)
return cdb_sc_helpedout;
else
return cdb_sc_blknumerr;
}
--pCurr; --nLevl;
if (nLevl < 1)
break;
pCurr->tn = cs_addrs->ti->curr_tn;
if (NULL == (pBlkBase = t_qread(nBlkId, (sm_int_ptr_t)&pCurr->cycle, &pCurr->cr)))
return (enum cdb_sc)rdfail_detail;
if (((blk_hdr_ptr_t)pBlkBase)->levl != nLevl)
{
assert(CDB_STAGNATE > t_tries);
return cdb_sc_badlvl;
}
}
return cdb_sc_normal;
}
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;
}
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 Parameter:
blk_base = Block's base which has the key
rec_top = record top of the record which will be expanded
Output Parameter:
expanded_key = expanded key
rec_size = last record size whic has the key
keylen = key size
keycmpc = key compression cound
hist_ptr = history of blocks read, while expanding a *-key
History excludes the working block from which key is expanded and
includes the blocks read below the current block to expand a *-key
NOTE: hist_ptr.depth will be unchanged
Return:
cdb_sc_normal on success
failure code on concurrency failure
*******************************************************************************************/
enum cdb_sc gvcst_expand_any_key (sm_uc_ptr_t blk_base, sm_uc_ptr_t rec_top, sm_uc_ptr_t expanded_key,
int *rec_size, int *keylen, int *keycmpc, srch_hist *hist_ptr)
{
enum cdb_sc status;
unsigned char expanded_star_key[MAX_KEY_SZ];
unsigned short temp_ushort;
int cur_level;
int star_keycmpc;
int star_keylen;
int star_rec_size;
int tblk_size;
block_id tblk_num;
sm_uc_ptr_t rPtr1, rPtr2, curptr;
cur_level = ((blk_hdr_ptr_t)blk_base)->levl;
curptr = blk_base + sizeof(blk_hdr);
*rec_size = *keycmpc = *keylen = 0;
while (curptr < rec_top)
{
GET_RSIZ(*rec_size, curptr);
if (0 == cur_level || BSTAR_REC_SIZE != *rec_size)
{
READ_RECORD(cur_level, curptr, *keycmpc, *rec_size, expanded_key, *keylen, status);
if (cdb_sc_normal != status)
{
assert(t_tries < CDB_STAGNATE);
return status;
}
else
{
curptr += *rec_size;
if (curptr >= rec_top)
break;
}
}
else /* a star record in index block */
{
if (curptr + *rec_size != rec_top || NULL == hist_ptr)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_rmisalign;
}
while (0 != cur_level)
{
tblk_size = ((blk_hdr_ptr_t)blk_base)->bsiz;
GET_LONG(tblk_num, blk_base + tblk_size - sizeof(block_id));
if (0 == tblk_num || cs_data->trans_hist.total_blks - 1 < tblk_num)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_badlvl;
}
cur_level--;
hist_ptr->h[cur_level].tn = cs_addrs->ti->curr_tn;
if (!(blk_base = t_qread(tblk_num, (sm_int_ptr_t)(&(hist_ptr->h[cur_level].cycle)),
&(hist_ptr->h[cur_level].cr) )))
{
assert(t_tries < CDB_STAGNATE);
return rdfail_detail;
}
if (((blk_hdr_ptr_t)blk_base)->levl != cur_level)
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_badlvl;
}
hist_ptr->h[cur_level].buffaddr = blk_base;
hist_ptr->h[cur_level].blk_num = tblk_num;
hist_ptr->h[cur_level].prev_rec.match = 0;
hist_ptr->h[cur_level].prev_rec.offset = 0;
hist_ptr->h[cur_level].curr_rec.match = 0;
hist_ptr->h[cur_level].curr_rec.offset = 0;
}
tblk_size = ((blk_hdr_ptr_t)blk_base)->bsiz;
/* expand *-key from right most leaf level block of the
sub-tree, of which, the original block is root */
if (cdb_sc_normal != (status = (gvcst_expand_any_key(blk_base, blk_base + tblk_size,
expanded_star_key, &star_rec_size, &star_keylen, &star_keycmpc, hist_ptr))))
return status;
if (*keylen + *keycmpc) /* Previous key exists */
{
GET_CMPC(*keycmpc, expanded_key, &expanded_star_key[0]);
}
memcpy(expanded_key, expanded_star_key, star_keylen + star_keycmpc);
*keylen = star_keylen + star_keycmpc - *keycmpc;
*rec_size = *keylen + *keycmpc + BSTAR_REC_SIZE;
return cdb_sc_normal;
} /* end else if *-record */
}/* end of "while" loop */
if (curptr == rec_top)
{
return cdb_sc_normal;
}
else
{
assert(t_tries < CDB_STAGNATE);
return cdb_sc_rmisalign;
}
}
void dse_m_rest (
block_id blk, /* block number */
unsigned char *bml_list, /* start of local list of local bit maps */
int4 bml_size, /* size of each entry in *bml_list */
sm_vuint_ptr_t blks_ptr, /* total free blocks */
bool in_dir_tree)
{
sm_uc_ptr_t bp, b_top, rp, r_top, bml_ptr, np, ptr;
unsigned char util_buff[MAX_UTIL_LEN];
block_id next;
int util_len;
int4 dummy_int;
cache_rec_ptr_t dummy_cr;
int4 bml_index;
short level, rsize;
int4 bplmap;
error_def(ERR_DSEBLKRDFAIL);
if(!(bp = t_qread (blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
if (((blk_hdr_ptr_t) bp)->bsiz > cs_addrs->hdr->blk_size)
b_top = bp + cs_addrs->hdr->blk_size;
else if (((blk_hdr_ptr_t) bp)->bsiz < sizeof(blk_hdr))
b_top = bp + sizeof(blk_hdr);
else
b_top = bp + ((blk_hdr_ptr_t) bp)->bsiz;
level = ((blk_hdr_ptr_t)bp)->levl;
bplmap = cs_addrs->hdr->bplmap;
for (rp = bp + sizeof (blk_hdr); rp < b_top ;rp = r_top)
{ if (in_dir_tree || level > 1) /* reread block because it may have been flushed from read */
{ if (!(np = t_qread(blk,&dummy_int,&dummy_cr))) /* cache due to LRU buffer scheme and reads in recursive */
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL); /* calls to dse_m_rest. */
if (np != bp)
{ b_top = np + (b_top - bp);
rp = np + (rp - bp);
r_top = np + (r_top - bp);
bp = np;
}
}
GET_SHORT(rsize,&((rec_hdr_ptr_t)rp)->rsiz);
r_top = rp + rsize;
if (r_top > b_top)
r_top = b_top;
if (r_top - rp < (sizeof (rec_hdr) + sizeof (block_id)))
break;
if (in_dir_tree && level == 0)
{
for (ptr = rp + sizeof(rec_hdr); ; )
{
if (*ptr++ == 0 && *ptr++ == 0)
break;
}
GET_LONG(next,ptr);
}
else
GET_LONG(next,r_top - sizeof (block_id));
if (next < 0 || next >= cs_addrs->ti->total_blks ||
(next / bplmap * bplmap == next))
{ memcpy(util_buff,"Invalid pointer in block ",25);
util_len = 25;
util_len += i2hex_nofill(blk, &util_buff[util_len], 8);
memcpy(&util_buff[util_len], " record offset ",15);
util_len += 15;
util_len += i2hex_nofill((int)(rp - bp), &util_buff[util_len], 4);
util_buff[util_len] = 0;
util_out_print((char*)util_buff,TRUE);
continue;
}
bml_index = next / bplmap;
bml_ptr = bml_list + bml_index * bml_size;
if (bml_busy(next - next / bplmap * bplmap, bml_ptr + sizeof(blk_hdr)))
{ *blks_ptr = *blks_ptr - 1;
if (((blk_hdr_ptr_t) bp)->levl > 1)
{ dse_m_rest (next, bml_list, bml_size, blks_ptr, in_dir_tree);
}
else if (in_dir_tree)
{ assert(((blk_hdr_ptr_t) bp)->levl == 0 || ((blk_hdr_ptr_t) bp)->levl == 1);
dse_m_rest (next, bml_list, bml_size, blks_ptr, ((blk_hdr_ptr_t)bp)->levl);
}
}
}
return;
}
void dse_chng_rhead(void)
{
block_id blk;
sm_uc_ptr_t bp, b_top, cp, rp;
boolean_t chng_rec;
rec_hdr new_rec;
uint4 x;
blk_segment *bs1, *bs_ptr;
int4 blk_seg_cnt, blk_size;
srch_blk_status blkhist;
error_def(ERR_DBRDONLY);
error_def(ERR_DSEBLKRDFAIL);
error_def(ERR_DSEFAIL);
if (gv_cur_region->read_only)
rts_error(VARLSTCNT(4) ERR_DBRDONLY, 2, DB_LEN_STR(gv_cur_region));
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
if (cli_present("BLOCK") == CLI_PRESENT)
{
if(!cli_get_hex("BLOCK", (uint4 *)&blk))
return;
patch_curr_blk = blk;
}
if (patch_curr_blk < 0 || patch_curr_blk >= cs_addrs->ti->total_blks || !(patch_curr_blk % cs_addrs->hdr->bplmap))
{
util_out_print("Error: invalid block number.", TRUE);
return;
}
t_begin_crit(ERR_DSEFAIL);
blkhist.blk_num = patch_curr_blk;
if (!(blkhist.buffaddr = t_qread(blkhist.blk_num, &blkhist.cycle, &blkhist.cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
bp = blkhist.buffaddr;
blk_size = cs_addrs->hdr->blk_size;
chng_rec = FALSE;
b_top = bp + ((blk_hdr_ptr_t)bp)->bsiz;
if (((blk_hdr_ptr_t)bp)->bsiz > blk_size || ((blk_hdr_ptr_t)bp)->bsiz < SIZEOF(blk_hdr))
chng_rec = TRUE; /* force rewrite to correct size */
if (cli_present("RECORD") == CLI_PRESENT)
{
if (!(rp = skan_rnum(bp, FALSE)))
{
t_abort(gv_cur_region, cs_addrs);
return;
}
} else if (!(rp = skan_offset(bp, FALSE)))
{
t_abort(gv_cur_region, cs_addrs);
return;
}
GET_SHORT(new_rec.rsiz, &((rec_hdr_ptr_t)rp)->rsiz);
new_rec.cmpc = ((rec_hdr_ptr_t)rp)->cmpc;
if (cli_present("CMPC") == CLI_PRESENT)
{
if (!cli_get_hex("CMPC", &x))
{
t_abort(gv_cur_region, cs_addrs);
return;
}
if (x > 0x7f)
{
util_out_print("Error: invalid cmpc.",TRUE);
t_abort(gv_cur_region, cs_addrs);
return;
}
if (x > patch_comp_count)
util_out_print("Warning: specified compression count is larger than the current expanded key size.", TRUE);
new_rec.cmpc = x;
chng_rec = TRUE;
}
if (cli_present("RSIZ") == CLI_PRESENT)
{
if (!cli_get_hex("RSIZ", &x))
{
t_abort(gv_cur_region, cs_addrs);
return;
}
if (x < SIZEOF(rec_hdr) || x > blk_size)
{
util_out_print("Error: invalid rsiz.", TRUE);
t_abort(gv_cur_region, cs_addrs);
return;
}
new_rec.rsiz = x;
chng_rec = TRUE;
}
if (chng_rec)
{
BLK_INIT(bs_ptr, bs1);
cp = bp;
cp += SIZEOF(blk_hdr);
if (chng_rec)
{
BLK_SEG(bs_ptr, cp, rp - cp);
BLK_SEG(bs_ptr, (uchar_ptr_t)&new_rec, SIZEOF(rec_hdr));
cp = rp + SIZEOF(rec_hdr);
}
if (b_top - cp)
BLK_SEG(bs_ptr, cp, b_top - cp);
if (!BLK_FINI(bs_ptr, bs1))
{
util_out_print("Error: bad blk build.", TRUE);
t_abort(gv_cur_region, cs_addrs);
return;
}
t_write(&blkhist, (unsigned char *)bs1, 0, 0, ((blk_hdr_ptr_t)bp)->levl, TRUE, FALSE, GDS_WRITE_KILLTN);
BUILD_AIMG_IF_JNL_ENABLED(cs_data, non_tp_jfb_buff_ptr, cs_addrs->ti->curr_tn);
t_end(&dummy_hist, NULL, TN_NOT_SPECIFIED);
}
return;
}
trans_num gvcst_bmp_mark_free(kill_set *ks)
{
block_id bit_map, next_bm, *updptr;
blk_ident *blk, *blk_top, *nextblk;
trans_num ctn, start_db_fmt_tn;
unsigned int len;
# if defined(UNIX) && defined(DEBUG)
unsigned int lcl_t_tries;
# endif
int4 blk_prev_version;
srch_hist alt_hist;
trans_num ret_tn = 0;
boolean_t visit_blks;
srch_blk_status bmphist;
cache_rec_ptr_t cr;
enum db_ver ondsk_blkver;
enum cdb_sc status;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
TREF(in_gvcst_bmp_mark_free) = TRUE;
assert(inctn_bmp_mark_free_gtm == inctn_opcode || inctn_bmp_mark_free_mu_reorg == inctn_opcode);
/* Note down the desired_db_format_tn before you start relying on cs_data->fully_upgraded.
* If the db is fully_upgraded, take the optimal path that does not need to read each block being freed.
* But in order to detect concurrent desired_db_format changes, note down the tn (when the last format change occurred)
* before the fully_upgraded check and after having noted down the database current_tn.
* If they are the same, then we are guaranteed no concurrent desired_db_format change occurred.
* If they are not, then fall through to the non-optimal path where each to-be-killed block has to be visited.
* The reason we need to visit every block in case desired_db_format changes is to take care of the case where
* MUPIP REORG DOWNGRADE concurrently changes a block that we are about to free.
*/
start_db_fmt_tn = cs_data->desired_db_format_tn;
visit_blks = (!cs_data->fully_upgraded); /* Local evaluation */
assert(!visit_blks || (visit_blks && dba_bg == cs_addrs->hdr->acc_meth)); /* must have blks_to_upgrd == 0 for non-BG */
assert(!dollar_tlevel); /* Should NOT be in TP now */
blk = &ks->blk[0];
blk_top = &ks->blk[ks->used];
if (!visit_blks)
{ /* Database has been completely upgraded. Free all blocks in one bitmap as part of one transaction. */
assert(cs_data->db_got_to_v5_once); /* assert all V4 fmt blocks (including RECYCLED) have space for V5 upgrade */
inctn_detail.blknum_struct.blknum = 0; /* to indicate no adjustment to "blks_to_upgrd" necessary */
/* If any of the mini transaction below restarts because of an online rollback, we don't want the application
* refresh to happen (like $ZONLNRLBK++ or rts_error(DBROLLEDBACK). This is because, although we are currently in
* non-tp (dollar_tleve = 0), we could actually be in a TP transaction and have actually faked dollar_tlevel. In
* such a case, we should NOT * be issuing a DBROLLEDBACK error as TP transactions are supposed to just restart in
* case of an online rollback. So, set the global variable that gtm_onln_rlbk_clnup can check and skip doing the
* application refresh, but will reset the clues. The next update will see the cycle mismatch and will accordingly
* take the right action.
*/
for ( ; blk < blk_top; blk = nextblk)
{
if (0 != blk->flag)
{
nextblk = blk + 1;
continue;
}
assert(0 < blk->block);
assert((int4)blk->block < cs_addrs->ti->total_blks);
bit_map = ROUND_DOWN2((int)blk->block, BLKS_PER_LMAP);
next_bm = bit_map + BLKS_PER_LMAP;
CHECK_AND_RESET_UPDATE_ARRAY; /* reset update_array_ptr to update_array */
/* Scan for the next local bitmap */
updptr = (block_id *)update_array_ptr;
for (nextblk = blk;
(0 == nextblk->flag) && (nextblk < blk_top) && ((block_id)nextblk->block < next_bm);
++nextblk)
{
assert((block_id)nextblk->block - bit_map);
*updptr++ = (block_id)nextblk->block - bit_map;
}
len = (unsigned int)((char *)nextblk - (char *)blk);
update_array_ptr = (char *)updptr;
alt_hist.h[0].blk_num = 0; /* need for calls to T_END for bitmaps */
alt_hist.h[0].blk_target = NULL; /* need to initialize for calls to T_END */
/* the following assumes SIZEOF(blk_ident) == SIZEOF(int) */
assert(SIZEOF(blk_ident) == SIZEOF(int));
*(int *)update_array_ptr = 0;
t_begin(ERR_GVKILLFAIL, UPDTRNS_DB_UPDATED_MASK);
for (;;)
{
ctn = cs_addrs->ti->curr_tn;
/* Need a read fence before reading fields from cs_data as we are reading outside
* of crit and relying on this value to detect desired db format state change.
*/
SHM_READ_MEMORY_BARRIER;
if (start_db_fmt_tn != cs_data->desired_db_format_tn)
{ /* Concurrent db format change has occurred. Need to visit every block to be killed
* to determine its block format. Fall through to the non-optimal path below
*/
ret_tn = 0;
break;
}
bmphist.blk_num = bit_map;
if (NULL == (bmphist.buffaddr = t_qread(bmphist.blk_num, (sm_int_ptr_t)&bmphist.cycle,
&bmphist.cr)))
{
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
t_write_map(&bmphist, (uchar_ptr_t)update_array, ctn, -(int4)(nextblk - blk));
UNIX_ONLY(DEBUG_ONLY(lcl_t_tries = t_tries));
if ((trans_num)0 == (ret_tn = t_end(&alt_hist, NULL, TN_NOT_SPECIFIED)))
{
# ifdef UNIX
assert((CDB_STAGNATE == t_tries) || (lcl_t_tries == t_tries - 1));
status = LAST_RESTART_CODE;
if ((cdb_sc_onln_rlbk1 == status) || (cdb_sc_onln_rlbk2 == status)
|| TREF(rlbk_during_redo_root))
{ /* t_end restarted due to online rollback. Discard bitmap free-up and return control
* to the application. But, before that reset only_reset_clues_if_onln_rlbk to FALSE
*/
TREF(in_gvcst_bmp_mark_free) = FALSE;
send_msg(VARLSTCNT(6) ERR_IGNBMPMRKFREE, 4, REG_LEN_STR(gv_cur_region),
DB_LEN_STR(gv_cur_region));
t_abort(gv_cur_region, cs_addrs);
return ret_tn; /* actually 0 */
}
# endif
continue;
}
break;
}
if (0 == ret_tn) /* db format change occurred. Fall through to below for loop to visit each block */
{
/* Abort any active transaction to get rid of lingering Non-TP artifacts */
t_abort(gv_cur_region, cs_addrs);
break;
}
}
} /* for all blocks in the kill_set */
void dse_save(void)
{
block_id blk;
unsigned i, j, util_len;
unsigned short buff_len;
bool was_block, was_crit;
char buff[100], *ptr, util_buff[MAX_UTIL_LEN];
sm_uc_ptr_t bp;
int4 dummy_int, nocrit_present;
cache_rec_ptr_t dummy_cr;
error_def(ERR_DSEBLKRDFAIL);
memset(util_buff, 0, MAX_UTIL_LEN);
if (was_block = (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;
} else
blk = patch_curr_blk;
if (cli_present("LIST") == CLI_PRESENT)
{
if (was_block)
{
util_len = sizeof("!/Saved versions of block ");
memcpy(util_buff, "!/Saved versions of block ", util_len);
util_len += i2hex_nofill(blk, (uchar_ptr_t)&util_buff[util_len-1], 8);
util_buff[util_len-1] = 0;
util_out_print(util_buff, TRUE);
for (i = j = 0; i < patch_save_count; i++)
if (patch_save_set[i].blk == blk)
{
j++;
if (*patch_save_set[i].comment)
util_out_print("Version !UL Region !AD Comment: !AD!/", TRUE,
patch_save_set[i].ver, REG_LEN_STR(patch_save_set[i].region),
LEN_AND_STR(patch_save_set[i].comment));
else
util_out_print("Version !UL Region !AD!/", TRUE, patch_save_set[i].ver,
REG_LEN_STR(patch_save_set[i].region));
}
if (!j)
util_out_print("None.!/", TRUE);
return;
}
util_out_print("!/Save history:!/", TRUE);
for (i = j = 0; i < patch_save_count; i++)
{
util_len = sizeof("Block ");
memcpy(util_buff, "Block ", util_len);
util_len += i2hex_nofill(patch_save_set[i].blk, (uchar_ptr_t)&util_buff[util_len-1], 8);
util_buff[util_len-1] = 0;
util_out_print(util_buff, TRUE);
j++;
if (*patch_save_set[i].comment)
{
util_out_print("Version !UL Region !AD Comment: !AD!/", TRUE,
patch_save_set[i].ver, REG_LEN_STR(patch_save_set[i].region),
LEN_AND_STR(patch_save_set[i].comment));
} else
{
util_out_print("Version !UL Region !AD!/", TRUE, patch_save_set[i].ver,
REG_LEN_STR(patch_save_set[i].region));
}
}
if (!j)
util_out_print(" None.!/", TRUE);
return;
}
j = 1;
for (i = 0; i < patch_save_count; i++)
if (patch_save_set[i].blk == blk && patch_save_set[i].region == gv_cur_region
&& patch_save_set[i].ver >= j)
j = patch_save_set[i].ver + 1;
util_len = sizeof("!/Saving version !UL of block ");
memcpy(util_buff, "!/Saving version !UL of block ", util_len);
util_len += i2hex_nofill(blk, (uchar_ptr_t)&util_buff[util_len-1], 8);
util_buff[util_len-1] = 0;
util_out_print(util_buff, TRUE, j);
patch_save_set[patch_save_count].ver = j;
patch_save_set[patch_save_count].blk = blk;
patch_save_set[patch_save_count].region = gv_cur_region;
patch_save_set[patch_save_count].bp = (char *)malloc(cs_addrs->hdr->blk_size);
if (blk >= cs_addrs->ti->total_blks)
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
was_crit = cs_addrs->now_crit;
nocrit_present = (CLI_NEGATED == cli_present("CRIT"));
if (!was_crit)
{
if (nocrit_present)
cs_addrs->now_crit = TRUE;
else
grab_crit(gv_cur_region);
}
if (!(bp = t_qread(blk, &dummy_int, &dummy_cr)))
rts_error(VARLSTCNT(1) ERR_DSEBLKRDFAIL);
memcpy(patch_save_set[patch_save_count].bp, bp, cs_addrs->hdr->blk_size);
if (!was_crit)
{
if (nocrit_present)
cs_addrs->now_crit = FALSE;
else
rel_crit(gv_cur_region);
}
buff_len = sizeof(buff);
if ((cli_present("COMMENT") == CLI_PRESENT) && cli_get_str("COMMENT", buff, &buff_len))
{
ptr = &buff[buff_len];
*ptr = 0;
j = ptr - &buff[0] + 1;
patch_save_set[patch_save_count].comment = (char *)malloc(j);
memcpy(patch_save_set[patch_save_count].comment, &buff[0], j);
} else
patch_save_set[patch_save_count].comment = "";
patch_save_count++;
return;
}