/* 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 mupip_restore(void)
{
static readonly char label[] = GDS_LABEL;
char db_name[MAX_FN_LEN + 1], *inbuf, *p;
inc_list_struct *ptr;
inc_header *inhead;
sgmnt_data *old_data;
short iosb[4];
unsigned short n_len;
int4 status, vbn, rsize, temp, save_errno;
uint4 rest_blks, totblks;
trans_num curr_tn;
uint4 ii;
block_id blk_num;
bool extend;
uint4 cli_status;
BFILE *in;
int i, db_fd;
uint4 old_blk_size, old_tot_blks, bplmap;
short old_start_vbn;
off_t new_eof;
char buff[DISK_BLOCK_SIZE];
char msg_buffer[1024], *newmap, *newmap_bptr;
mstr msg_string;
char addr[SA_MAXLEN+1];
unsigned char tcp[5];
backup_type type;
unsigned short port;
int4 timeout, cut, match;
char debug_info[256];
void (*common_read)();
char *errptr;
pid_t waitpid_res;
error_def(ERR_MUPRESTERR);
error_def(ERR_MUPCLIERR);
error_def(ERR_IOEOF);
extend = TRUE;
if (CLI_NEGATED == (cli_status = cli_present("EXTEND")))
extend = FALSE;
mu_outofband_setup();
mu_gv_cur_reg_init();
n_len = sizeof(db_name);
if (cli_get_str("DATABASE", db_name, &n_len) == FALSE)
mupip_exit(ERR_MUPCLIERR);
strcpy((char *)gv_cur_region->dyn.addr->fname, db_name);
gv_cur_region->dyn.addr->fname_len = n_len;
if (!mu_rndwn_file(gv_cur_region, TRUE))
{
util_out_print("Error securing stand alone access to output file !AD. Aborting restore.", TRUE, n_len, db_name);
mupip_exit(ERR_MUPRESTERR);
}
OPENFILE(db_name, O_RDWR, db_fd);
if (-1 == db_fd)
{
save_errno = errno;
util_out_print("Error accessing output file !AD. Aborting restore.", TRUE, n_len, db_name);
errptr = (char *)STRERROR(save_errno);
util_out_print("open : !AZ", TRUE, errptr);
mupip_exit(save_errno);
}
murgetlst();
inbuf = (char*)malloc(INC_BACKUP_CHUNK_SIZE);
old_data = (sgmnt_data*)malloc(sizeof(sgmnt_data));
LSEEKREAD(db_fd, 0, old_data, sizeof(sgmnt_data), save_errno);
if (0 != save_errno)
{
util_out_print("Error accessing output file !AD. Aborting restore.", TRUE, n_len, db_name);
if (-1 != save_errno)
{
errptr = (char *)STRERROR(save_errno);
util_out_print("read : !AZ", TRUE, errptr);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(save_errno);
} else
{
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_IOEOF);
}
}
if (memcmp(&old_data->label[0], &label[0], GDS_LABEL_SZ))
{
util_out_print("Output file !AD has an unrecognizable format", TRUE, n_len, db_name);
free(old_data);
free(inbuf);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
curr_tn = old_data->trans_hist.curr_tn;
old_blk_size = old_data->blk_size;
old_tot_blks = old_data->trans_hist.total_blks;
old_start_vbn = old_data->start_vbn;
bplmap = old_data->bplmap;
free(old_data);
msg_string.addr = msg_buffer;
msg_string.len = sizeof(msg_buffer);
inhead = (inc_header *)malloc(sizeof(inc_header) + 8);
inhead = (inc_header *)((((int4)inhead) + 7) & -8);
rest_blks = 0;
for (ptr = in_files.next; ptr; ptr = ptr->next)
{ /* --- determine source type --- */
type = backup_to_file;
if (0 == ptr->input_file.len)
continue;
else if ('|' == *(ptr->input_file.addr + ptr->input_file.len - 1))
{
type = backup_to_exec;
ptr->input_file.len--;
*(ptr->input_file.addr + ptr->input_file.len) = '\0';
} else if (ptr->input_file.len > 5)
{
lower_to_upper(tcp, (uchar_ptr_t)ptr->input_file.addr, 5);
if (0 == memcmp(tcp, "TCP:/", 5))
{
type = backup_to_tcp;
cut = 5;
while ('/' == *(ptr->input_file.addr + cut))
cut++;
ptr->input_file.len -= cut;
p = ptr->input_file.addr;
while (p < ptr->input_file.addr + ptr->input_file.len)
{
*p = *(p + cut);
p++;
}
*p = '\0';
}
}
/* --- open the input stream --- */
restore_read_errno = 0;
switch(type)
{
case backup_to_file:
common_read = iob_read;
if ((in = iob_open_rd(ptr->input_file.addr, DISK_BLOCK_SIZE, BLOCKING_FACTOR)) == NULL)
{
save_errno = errno;
util_out_print("Error accessing input file !AD. Aborting restore.", TRUE,
ptr->input_file.len, ptr->input_file.addr);
errptr = (char *)STRERROR(save_errno);
util_out_print("open : !AZ", TRUE, errptr);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(save_errno);
}
ESTABLISH(iob_io_error);
break;
case backup_to_exec:
pipe_child = 0;
common_read = exec_read;
in = (BFILE *)malloc(sizeof(BFILE));
if (0 > (in->fd = gtm_pipe(ptr->input_file.addr, input_from_comm)))
{
util_out_print("Error creating input pipe from !AD.",
TRUE, ptr->input_file.len, ptr->input_file.addr);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
#ifdef DEBUG_ONLINE
PRINTF("file descriptor for the openned pipe is %d.\n", in->fd);
PRINTF("the command passed to gtm_pipe is %s.\n", ptr->input_file.addr);
#endif
break;
case backup_to_tcp:
common_read = tcp_read;
/* parse the input */
switch (match = SSCANF(ptr->input_file.addr, "%[^:]:%hu", addr, &port))
{
case 1 :
port = DEFAULT_BKRS_PORT;
case 2 :
break;
default :
util_out_print("Error : A hostname has to be specified.", TRUE);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
if ((0 == cli_get_int("NETTIMEOUT", &timeout)) || (0 > timeout))
timeout = DEFAULT_BKRS_TIMEOUT;
in = (BFILE *)malloc(sizeof(BFILE));
iotcp_fillroutine();
if (0 > (in->fd = tcp_open(addr, port, timeout, TRUE)))
{
util_out_print("Error establishing TCP connection to !AD.",
TRUE, ptr->input_file.len, ptr->input_file.addr);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
break;
default:
util_out_print("Aborting restore!/", TRUE);
util_out_print("Unrecognized input format !AD", TRUE, ptr->input_file.len, ptr->input_file.addr);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
COMMON_READ(in, inhead, sizeof(inc_header));
if (memcmp(&inhead->label[0], INC_HEADER_LABEL, INC_HDR_LABEL_SZ))
{
util_out_print("Input file !AD has an unrecognizable format", TRUE, ptr->input_file.len,
ptr->input_file.addr);
free(inbuf);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
if (curr_tn != inhead->start_tn)
{
util_out_print("Transaction in input file !AD does not align with database TN.!/DB: !XL!_Input file: !XL",
TRUE, ptr->input_file.len, ptr->input_file.addr, curr_tn, inhead->start_tn);
free(inbuf);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
if (old_blk_size != inhead->blk_size)
{
util_out_print("Incompatable block size. Output file !AD has block size !XL,", TRUE, n_len, db_name);
util_out_print("while input file !AD is from a database with block size !XL,", TRUE, ptr->input_file.len,
ptr->input_file.addr, inhead->blk_size);
free(inbuf);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
if (old_tot_blks != inhead->db_total_blks)
{
if (old_tot_blks > inhead->db_total_blks || !extend)
{
totblks = old_tot_blks - DIVIDE_ROUND_UP(old_tot_blks, DISK_BLOCK_SIZE);
util_out_print("Incompatable database sizes. Output file !AD has!/ !UL (!XL hex) total blocks,",
TRUE, n_len, db_name, totblks, totblks);
totblks = inhead->db_total_blks - DIVIDE_ROUND_UP(inhead->db_total_blks, DISK_BLOCK_SIZE);
util_out_print("while input file !AD is from a database with!/ !UL (!XL hex) total blocks",
TRUE, ptr->input_file.len, ptr->input_file.addr, totblks, totblks);
free(inbuf);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
} else
{ /* this part of the code is similar to gdsfilext except that you don't need to do
* most of the work that gdsfilext does. However, for situations where the database
* extended since the last backup (the beginning of this incremental backup), and
* there are new bitmaps that are never touched later on by GT.M, these bitmaps
* will have tn == 0, which prevents the backup process to pick up these blocks,
* so, we need to initialize these bitmaps here
*/
new_eof = ((off_t)(old_start_vbn - 1) * DISK_BLOCK_SIZE)
+ ((off_t)inhead->db_total_blks * old_blk_size);
memset(buff, 0, DISK_BLOCK_SIZE);
LSEEKWRITE(db_fd, new_eof, buff, DISK_BLOCK_SIZE, status);
if (0 != status)
{
util_out_print("Aborting restore!/", TRUE);
util_out_print("lseek or write error : Unable to extend output file !AD!/",
TRUE, n_len, db_name);
util_out_print(" from !UL (!XL hex) total blocks to !UL (!XL hex) total blocks.!/",
TRUE, old_tot_blks, old_tot_blks, inhead->db_total_blks, inhead->db_total_blks);
util_out_print(" Current input file is !AD with !UL (!XL hex) total blocks!/",
TRUE, ptr->input_file.len, ptr->input_file.addr,
inhead->db_total_blks, inhead->db_total_blks);
gtm_putmsg(VARLSTCNT(1) status);
free(inbuf);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
/* --- initialize all new bitmaps, just in case they are not touched later --- */
if (DIVIDE_ROUND_DOWN(inhead->db_total_blks, bplmap) > DIVIDE_ROUND_DOWN(old_tot_blks, bplmap))
{ /* -- similar logic exist in bml_newmap.c, which need to pick up any new updates here -- */
newmap = (char *)malloc(old_blk_size);
((blk_hdr *)newmap)->bsiz = BM_SIZE(bplmap);
((blk_hdr *)newmap)->levl = LCL_MAP_LEVL;
((blk_hdr *)newmap)->tn = curr_tn;
newmap_bptr = newmap + sizeof(blk_hdr);
*newmap_bptr++ = THREE_BLKS_FREE;
memset(newmap_bptr, FOUR_BLKS_FREE, BM_SIZE(bplmap) - sizeof(blk_hdr) - 1);
for (ii = ROUND_UP(old_tot_blks, bplmap); ii <= inhead->db_total_blks; ii += bplmap)
{
new_eof = (off_t)(old_start_vbn - 1) * DISK_BLOCK_SIZE + (off_t)ii * old_blk_size;
LSEEKWRITE(db_fd, new_eof, newmap, old_blk_size, status);
if (0 != status)
{
util_out_print("Aborting restore!/", TRUE);
util_out_print("Bitmap 0x!XL initialization error!", TRUE, ii);
gtm_putmsg(VARLSTCNT(1) status);
free(inbuf);
free(newmap);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
}
free(newmap);
}
old_tot_blks = inhead->db_total_blks;
}
}
COMMON_READ(in, &rsize, sizeof(int4));
for ( ; ;)
{ /* rsize is the size of the record, including the size, but, since the size has already been
read in, this will read in the current record and the size for the next record */
/* ensure we have a reasonable record size, at least */
if (rsize - sizeof(int4) - sizeof(block_id) > old_blk_size)
{
util_out_print("Invalid information in restore file !AD. Aborting restore.",
TRUE, ptr->input_file.len,
ptr->input_file.addr);
iob_close(in);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(ERR_MUPRESTERR);
}
COMMON_READ(in, inbuf, rsize);
if (!memcmp(inbuf, &end_msg[0], sizeof end_msg - 1))
break;
rest_blks++;
blk_num = *(block_id*)inbuf;
vbn = old_start_vbn - 1 + (old_blk_size / DISK_BLOCK_SIZE * blk_num);
LSEEKWRITE(db_fd,
(off_t)vbn * DISK_BLOCK_SIZE,
inbuf + sizeof(block_id),
rsize - sizeof(block_id) - sizeof(int4),
save_errno);
if (0 != save_errno)
{
util_out_print("Error accessing output file !AD. Aborting restore.",
TRUE, n_len, db_name);
errptr = (char *)STRERROR(save_errno);
util_out_print("write : !AZ", TRUE, errptr);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(save_errno);
}
GET_LONG(temp, (inbuf + rsize - sizeof(int4)));
rsize = temp;
}
GET_LONG(temp, (inbuf + rsize - sizeof(int4)));
rsize = temp;
vbn = 0;
for (i = 0; ; i++) /* Restore file header */
{
COMMON_READ(in, inbuf, rsize);
if (!memcmp(inbuf, &hdr_msg[0], sizeof hdr_msg - 1))
break;
LSEEKWRITE(db_fd,
vbn,
inbuf,
rsize - sizeof(int4),
save_errno);
if (0 != save_errno)
{
util_out_print("Error accessing output file !AD. Aborting restore.",
TRUE, n_len, db_name);
errptr = (char *)STRERROR(save_errno);
util_out_print("write : !AZ", TRUE, errptr);
db_ipcs_reset(gv_cur_region, TRUE);
mu_gv_cur_reg_free();
mupip_exit(save_errno);
}
vbn += rsize - sizeof(int4);
GET_LONG(temp, (inbuf + rsize - sizeof(int4)));
rsize = temp;
}
curr_tn = inhead->end_tn;
switch (type)
{
case backup_to_file:
REVERT;
iob_close(in);
break;
case backup_to_exec:
close(in->fd);
if ((pipe_child > 0) && (FALSE != is_proc_alive(pipe_child, 0)))
WAITPID(pipe_child, (int *)&status, 0, waitpid_res);
break;
case backup_to_tcp:
break;
}
}
util_out_print("!/RESTORE COMPLETED", TRUE);
util_out_print("!UL blocks restored", TRUE, rest_blks);
free(inbuf);
db_ipcs_reset(gv_cur_region, FALSE);
mu_gv_cur_reg_free();
mupip_exit(SS_NORMAL);
}
boolean_t mu_truncate(int4 truncate_percent)
{
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
int num_local_maps;
int lmap_num, lmap_blk_num;
int bml_status, sigkill;
int save_errno;
int ftrunc_status;
uint4 jnl_status;
uint4 old_total, new_total;
uint4 old_free, new_free;
uint4 end_blocks;
int4 blks_in_lmap, blk;
gtm_uint64_t before_trunc_file_size;
off_t trunc_file_size;
off_t padding;
uchar_ptr_t lmap_addr;
boolean_t was_crit;
uint4 found_busy_blk;
srch_blk_status bmphist;
srch_blk_status *blkhist;
srch_hist alt_hist;
trans_num curr_tn;
blk_hdr_ptr_t lmap_blk_hdr;
block_id *blkid_ptr;
unix_db_info *udi;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
char *err_msg;
intrpt_state_t prev_intrpt_state;
off_t offset;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
csa = cs_addrs;
csd = cs_data;
if (dba_mm == csd->acc_meth)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCNOTBG, 2, REG_LEN_STR(gv_cur_region));
return TRUE;
}
if ((GDSVCURR != csd->desired_db_format) || (csd->blks_to_upgrd != 0))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCNOV4, 2, REG_LEN_STR(gv_cur_region));
return TRUE;
}
if (csa->ti->free_blocks < (truncate_percent * csa->ti->total_blks / 100))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_MUTRUNCNOSPACE, 3, REG_LEN_STR(gv_cur_region), truncate_percent);
return TRUE;
}
/* already checked for parallel truncates on this region --- see mupip_reorg.c */
gv_target = NULL;
assert(csa->nl->trunc_pid == process_id);
assert(dba_mm != csd->acc_meth);
old_total = csa->ti->total_blks;
old_free = csa->ti->free_blocks;
sigkill = 0;
found_busy_blk = 0;
memset(&alt_hist, 0, SIZEOF(alt_hist)); /* null-initialize history */
assert(csd->bplmap == BLKS_PER_LMAP);
end_blocks = old_total % BLKS_PER_LMAP; /* blocks in the last lmap (first one we start scanning) */
if (0 == end_blocks)
end_blocks = BLKS_PER_LMAP;
num_local_maps = DIVIDE_ROUND_UP(old_total, BLKS_PER_LMAP);
/* ======================================== PHASE 1 ======================================== */
for (lmap_num = num_local_maps - 1; (lmap_num > 0 && !found_busy_blk); lmap_num--)
{
if (mu_ctrly_occurred || mu_ctrlc_occurred)
return TRUE;
assert(csa->ti->total_blks >= old_total); /* otherwise, a concurrent truncate happened... */
if (csa->ti->total_blks != old_total) /* Extend (likely called by mupip extend) -- don't truncate */
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_MUTRUNCNOSPACE, 3, REG_LEN_STR(gv_cur_region),
truncate_percent);
return TRUE;
}
lmap_blk_num = lmap_num * BLKS_PER_LMAP;
if (csa->nl->highest_lbm_with_busy_blk >= lmap_blk_num)
{
found_busy_blk = lmap_blk_num;
break;
}
blks_in_lmap = (lmap_num == num_local_maps - 1) ? end_blocks : BLKS_PER_LMAP;
/* Loop through non-bitmap blocks of this lmap, do recycled2free */
DBGEHND((stdout, "DBG:: lmap_num = [%lu], lmap_blk_num = [%lu], blks_in_lmap = [%lu]\n",
lmap_num, lmap_blk_num, blks_in_lmap));
for (blk = 1; blk < blks_in_lmap && blk != -1 && !found_busy_blk;)
{
t_begin(ERR_MUTRUNCFAIL, UPDTRNS_DB_UPDATED_MASK);
for (;;) /* retry loop for recycled to free transactions */
{
curr_tn = csd->trans_hist.curr_tn;
/* Read the nth local bitmap into memory */
bmphist.blk_num = lmap_blk_num;
bmphist.buffaddr = t_qread(bmphist.blk_num, &bmphist.cycle, &bmphist.cr);
lmap_blk_hdr = (blk_hdr_ptr_t)bmphist.buffaddr;
if (!(bmphist.buffaddr) || (BM_SIZE(BLKS_PER_LMAP) != lmap_blk_hdr->bsiz))
{ /* Could not read the block successfully. Retry. */
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
lmap_addr = bmphist.buffaddr + SIZEOF(blk_hdr);
/* starting from the hint (blk itself), find the first busy or recycled block */
blk = bml_find_busy_recycled(blk, lmap_addr, blks_in_lmap, &bml_status);
assert(blk < BLKS_PER_LMAP);
if (blk == -1 || blk >= blks_in_lmap)
{ /* done with this lmap, continue to next */
t_abort(gv_cur_region, csa);
break;
}
else if (BLK_BUSY == bml_status || csa->nl->highest_lbm_with_busy_blk >= lmap_blk_num)
{ /* stop processing blocks... skip ahead to phase 2 */
found_busy_blk = lmap_blk_num;
t_abort(gv_cur_region, csa);
break;
}
else if (BLK_RECYCLED == bml_status)
{ /* Write PBLK records for recycled blocks only if before_image journaling is
* enabled. t_end() takes care of checking if journaling is enabled and
* writing PBLK record. We have to at least mark the recycled block as free.
*/
RESET_UPDATE_ARRAY;
update_trans = UPDTRNS_DB_UPDATED_MASK;
*((block_id *)update_array_ptr) = blk;
update_array_ptr += SIZEOF(block_id);
*(int *)update_array_ptr = 0;
alt_hist.h[1].blk_num = 0;
alt_hist.h[0].level = 0;
alt_hist.h[0].cse = NULL;
alt_hist.h[0].tn = curr_tn;
alt_hist.h[0].blk_num = lmap_blk_num + blk;
alt_hist.h[0].buffaddr = t_qread(alt_hist.h[0].blk_num,
&alt_hist.h[0].cycle, &alt_hist.h[0].cr);
if (!alt_hist.h[0].buffaddr)
{
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
if (!t_recycled2free(&alt_hist.h[0]))
{
t_retry(cdb_sc_lostbmlcr);
continue;
}
t_write_map(&bmphist, (unsigned char *)update_array, curr_tn, 0);
/* Set the opcode for INCTN record written by t_end() */
inctn_opcode = inctn_blkmarkfree;
if ((trans_num)0 == t_end(&alt_hist, NULL, TN_NOT_SPECIFIED))
continue;
/* block processed, scan from the next one */
blk++;
break;
} else
{
assert(t_tries < CDB_STAGNATE);
t_retry(cdb_sc_badbitmap);
continue;
}
} /* END recycled2free retry loop */
} /* END scanning blocks of this particular lmap */
/* Write PBLK for the bitmap block, in case it hasn't been written i.e. t_end() was never called above */
/* Do a transaction that just increments the bitmap block's tn so that t_end() can do its thing */
DBGEHND((stdout, "DBG:: bitmap block inctn -- lmap_blk_num = [%lu]\n", lmap_blk_num));
t_begin(ERR_MUTRUNCFAIL, UPDTRNS_DB_UPDATED_MASK);
for (;;)
{
RESET_UPDATE_ARRAY;
BLK_ADDR(blkid_ptr, SIZEOF(block_id), block_id);
*blkid_ptr = 0;
update_trans = UPDTRNS_DB_UPDATED_MASK;
inctn_opcode = inctn_mu_reorg; /* inctn_mu_truncate */
curr_tn = csd->trans_hist.curr_tn;
blkhist = &alt_hist.h[0];
blkhist->blk_num = lmap_blk_num;
blkhist->tn = curr_tn;
blkhist->cse = NULL; /* start afresh (do not use value from previous retry) */
/* Read the nth local bitmap into memory */
blkhist->buffaddr = t_qread(lmap_blk_num, (sm_int_ptr_t)&blkhist->cycle, &blkhist->cr);
lmap_blk_hdr = (blk_hdr_ptr_t)blkhist->buffaddr;
if (!(blkhist->buffaddr) || (BM_SIZE(BLKS_PER_LMAP) != lmap_blk_hdr->bsiz))
{ /* Could not read the block successfully. Retry. */
t_retry((enum cdb_sc)rdfail_detail);
continue;
}
t_write_map(blkhist, (unsigned char *)blkid_ptr, curr_tn, 0);
blkhist->blk_num = 0; /* create empty history for bitmap block */
if ((trans_num)0 == t_end(&alt_hist, NULL, TN_NOT_SPECIFIED))
continue;
break;
}
} /* END scanning lmaps */
/* ======================================== PHASE 2 ======================================== */
assert(!csa->now_crit);
for (;;)
{ /* wait for FREEZE, we don't want to truncate a frozen database */
grab_crit(gv_cur_region);
if (FROZEN_CHILLED(cs_data))
DO_CHILLED_AUTORELEASE(csa, cs_data);
if (!FROZEN(cs_data) && !IS_REPL_INST_FROZEN)
break;
rel_crit(gv_cur_region);
while (FROZEN(cs_data) || IS_REPL_INST_FROZEN)
{
hiber_start(1000);
if (FROZEN_CHILLED(cs_data) && CHILLED_AUTORELEASE(cs_data))
break;
}
}
assert(csa->nl->trunc_pid == process_id);
/* Flush pending updates to disk. If this is not done, old updates can be flushed AFTER ftruncate, extending the file. */
if (!wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_MSYNC_DB))
{
assert(FALSE);
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_BUFFLUFAILED, 4, LEN_AND_LIT("MUPIP REORG TRUNCATE"),
DB_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return FALSE;
}
csa->nl->highest_lbm_with_busy_blk = MAX(found_busy_blk, csa->nl->highest_lbm_with_busy_blk);
assert(IS_BITMAP_BLK(csa->nl->highest_lbm_with_busy_blk));
new_total = MIN(old_total, csa->nl->highest_lbm_with_busy_blk + BLKS_PER_LMAP);
if (mu_ctrly_occurred || mu_ctrlc_occurred)
{
rel_crit(gv_cur_region);
return TRUE;
} else if (csa->ti->total_blks != old_total || new_total == old_total)
{
assert(csa->ti->total_blks >= old_total); /* Better have been an extend, not a truncate... */
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(5) ERR_MUTRUNCNOSPACE, 3, REG_LEN_STR(gv_cur_region), truncate_percent);
rel_crit(gv_cur_region);
return TRUE;
} else if (GDSVCURR != csd->desired_db_format || csd->blks_to_upgrd != 0 || !csd->fully_upgraded)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCNOV4, 2, REG_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return TRUE;
} else if (SNAPSHOTS_IN_PROG(csa->nl))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCSSINPROG, 2, REG_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return TRUE;
} else if (BACKUP_NOT_IN_PROGRESS != cs_addrs->nl->nbb)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_MUTRUNCBACKINPROG, 2, REG_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
return TRUE;
}
DEFER_INTERRUPTS(INTRPT_IN_TRUNC, prev_intrpt_state);
if (JNL_ENABLED(csa))
{ /* Write JRT_TRUNC and INCTN records */
if (!jgbl.dont_reset_gbl_jrec_time)
SET_GBL_JREC_TIME; /* needed before jnl_ensure_open as that can write jnl records */
jpc = csa->jnl;
jbp = jpc->jnl_buff;
/* Before writing to jnlfile, adjust jgbl.gbl_jrec_time if needed to maintain time order
* of jnl records. This needs to be done BEFORE the jnl_ensure_open as that could write
* journal records (if it decides to switch to a new journal file).
*/
ADJUST_GBL_JREC_TIME(jgbl, jbp);
jnl_status = jnl_ensure_open(gv_cur_region, csa);
if (SS_NORMAL != jnl_status)
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
else
{
if (0 == jpc->pini_addr)
jnl_put_jrt_pini(csa);
jnl_write_trunc_rec(csa, old_total, csa->ti->free_blocks, new_total);
inctn_opcode = inctn_mu_reorg;
jnl_write_inctn_rec(csa);
jnl_status = jnl_flush(gv_cur_region);
if (SS_NORMAL != jnl_status)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_JNLFLUSH, 2, JNL_LEN_STR(csd),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error with journal flush during mu_truncate"),
jnl_status);
assert(NOJNL == jpc->channel); /* jnl file lost has been triggered */
}
}
}
/* Good to go ahead and REALLY truncate (reduce total_blks, clear cache_array, FTRUNCATE) */
curr_tn = csa->ti->curr_tn;
CHECK_TN(csa, csd, curr_tn);
udi = FILE_INFO(gv_cur_region);
/* Information used by recover_truncate to check if the file size and csa->ti->total_blks are INCONSISTENT */
trunc_file_size = BLK_ZERO_OFF(csd->start_vbn) + ((off_t)csd->blk_size * (new_total + 1));
csd->after_trunc_total_blks = new_total;
csd->before_trunc_free_blocks = csa->ti->free_blocks;
csd->before_trunc_total_blks = old_total; /* Flags interrupted truncate for recover_truncate */
/* file size and total blocks: INCONSISTENT */
csa->ti->total_blks = new_total;
/* past the point of no return -- shared memory intact */
assert(csa->ti->free_blocks >= DELTA_FREE_BLOCKS(old_total, new_total));
csa->ti->free_blocks -= DELTA_FREE_BLOCKS(old_total, new_total);
new_free = csa->ti->free_blocks;
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_1); /* 55 : Issue a kill -9 before 1st fsync */
fileheader_sync(gv_cur_region);
DB_FSYNC(gv_cur_region, udi, csa, db_fsync_in_prog, save_errno);
CHECK_DBSYNC(gv_cur_region, save_errno);
/* past the point of no return -- shared memory deleted */
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_2); /* 56 : Issue a kill -9 after 1st fsync */
clear_cache_array(csa, csd, gv_cur_region, new_total, old_total);
offset = (off_t)BLK_ZERO_OFF(csd->start_vbn) + (off_t)new_total * csd->blk_size;
save_errno = db_write_eof_block(udi, udi->fd, csd->blk_size, offset, &(TREF(dio_buff)));
if (0 != save_errno)
{
err_msg = (char *)STRERROR(errno);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_MUTRUNCERROR, 4, REG_LEN_STR(gv_cur_region), LEN_AND_STR(err_msg));
return FALSE;
}
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_3); /* 57 : Issue a kill -9 after reducing csa->ti->total_blks, before FTRUNCATE */
/* Execute an ftruncate() and truncate the DB file
* ftruncate() is a SYSTEM CALL on almost all platforms (except SunOS)
* It ignores kill -9 signal till its operation is completed.
* So we can safely assume that the result of ftruncate() will be complete.
*/
FTRUNCATE(FILE_INFO(gv_cur_region)->fd, trunc_file_size, ftrunc_status);
if (0 != ftrunc_status)
{
err_msg = (char *)STRERROR(errno);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_MUTRUNCERROR, 4, REG_LEN_STR(gv_cur_region), LEN_AND_STR(err_msg));
/* should go through recover_truncate now, which will again try to FTRUNCATE */
return FALSE;
}
/* file size and total blocks: CONSISTENT (shrunk) */
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_4); /* 58 : Issue a kill -9 after FTRUNCATE, before 2nd fsync */
csa->nl->root_search_cycle++; /* Force concurrent processes to restart in t_end/tp_tend to make sure no one
* tries to commit updates past the end of the file. Bitmap validations together
* with highest_lbm_with_busy_blk should actually be sufficient, so this is
* just to be safe.
*/
csd->before_trunc_total_blks = 0; /* indicate CONSISTENT */
/* Increment TN */
assert(csa->ti->early_tn == csa->ti->curr_tn);
csd->trans_hist.early_tn = csd->trans_hist.curr_tn + 1;
INCREMENT_CURR_TN(csd);
fileheader_sync(gv_cur_region);
DB_FSYNC(gv_cur_region, udi, csa, db_fsync_in_prog, save_errno);
KILL_TRUNC_TEST(WBTEST_CRASH_TRUNCATE_5); /* 58 : Issue a kill -9 after after 2nd fsync */
CHECK_DBSYNC(gv_cur_region, save_errno);
ENABLE_INTERRUPTS(INTRPT_IN_TRUNC, prev_intrpt_state);
curr_tn = csa->ti->curr_tn;
rel_crit(gv_cur_region);
send_msg_csa(CSA_ARG(csa) VARLSTCNT(7) ERR_MUTRUNCSUCCESS, 5, DB_LEN_STR(gv_cur_region), old_total, new_total, &curr_tn);
util_out_print("Truncated region: !AD. Reduced total blocks from [!UL] to [!UL]. Reduced free blocks from [!UL] to [!UL].",
FLUSH, REG_LEN_STR(gv_cur_region), old_total, new_total, old_free, new_free);
return TRUE;
} /* END of mu_truncate() */
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;
}
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;
}
void mu_int_maps(void)
{
unsigned char *local;
uchar_ptr_t disk;
boolean_t agree, disk_full, local_full, master_full;
int maps, mapsize, mcnt, lcnt, bcnt;
unsigned int level;
uint_ptr_t dskmap_p;
uint4 dskmap, lfree, *lmap, map_blk_size;
block_id blkno, last_bmp;
enum db_ver ondsk_blkver;
mu_int_offset[0] = 0;
maps = (mu_int_data.trans_hist.total_blks + mu_int_data.bplmap - 1) / mu_int_data.bplmap;
local = mu_int_locals;
map_blk_size = BM_SIZE(mu_int_data.bplmap);
last_bmp = ((mu_int_data.trans_hist.total_blks / mu_int_data.bplmap) * mu_int_data.bplmap);
mapsize = mu_int_data.bplmap;
for (mcnt = 0; mcnt < maps; mcnt++, local += BM_MINUS_BLKHDR_SIZE(mapsize))
{
assert(mapsize == mu_int_data.bplmap);
blkno = mcnt * mu_int_data.bplmap;
bml_busy(blkno, mu_int_locals);
disk = mu_int_read(blkno, &ondsk_blkver); /* ondsk_blkver set to GDSV4 or GDSV6 (GDSVCURR) */
if (!disk)
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
mu_int_err(ERR_DBREADBM, 0, 0, 0, 0, 0, 0, LCL_MAP_LEVL);
continue;
}
if (LCL_MAP_LEVL != (level = (unsigned int)((blk_hdr_ptr_t)disk)->levl))
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
mu_int_err(ERR_DBLVLINC, 0, 0, 0, 0, 0, 0, level);
}
if (((blk_hdr_ptr_t)disk)->bsiz < map_blk_size)
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
mu_int_err(ERR_DBMBSIZMN, 0, 0, 0, 0, 0, 0, level);
continue;
}
if (((blk_hdr_ptr_t)disk)->bsiz > map_blk_size)
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
mu_int_err(ERR_DBMBSIZMX, 0, 0, 0, 0, 0, 0, level);
continue;
}
if (tn_reset_this_reg)
{
((blk_hdr_ptr_t)disk)->tn = 0;
mu_int_write(blkno, disk);
if (GDSVCURR != mu_int_data.desired_db_format)
mu_int_blks_to_upgrd++;
} else if (GDSVCURR != ondsk_blkver)
mu_int_blks_to_upgrd++;
if (((blk_hdr_ptr_t)disk)->tn >= mu_int_data.trans_hist.curr_tn)
{
if (trans_errors < disp_trans_errors)
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
mu_int_err(ERR_DBMBTNSIZMX, 0, 0, 0, 0, 0, 0, level);
trans_errors++;
} else
{
mu_int_errknt++;
trans_errors++;
}
}
master_full = !bit_set(mcnt, mu_int_master);
if (last_bmp == blkno)
mapsize = (mu_int_data.trans_hist.total_blks - blkno);
disk_full = (NO_FREE_SPACE == bml_find_free(0, disk + SIZEOF(blk_hdr), mapsize));
agree = TRUE;
for (lcnt = 0, dskmap_p = (uint_ptr_t)(disk + SIZEOF(blk_hdr)), lmap = (uint4 *)local;
lcnt < mapsize;
lcnt += SIZEOF(int4) * BITS_PER_UCHAR / BML_BITS_PER_BLK,
dskmap_p++, lmap++) /* # of bits/ bits per blk */
{
GET_LONG(dskmap, dskmap_p);
/* do a quick check to see if there is anything wrong with the
bitmaps entries that fit into an int4.
We need to check that the free bit matches the in-memory
copy and that there are no illegal combinations.
There are four combinations per block
(00 = busy, 01 = free, 10 = unused (invalid bitmap value),
11 = free block which was previously used).
We use the following calculation to determine if there are
any illegal combinations within the current int4.
00011011 take original value from bitmap on disk.
and 01010101 mask off "reused" bit.
----------
00010001 = free blocks.
xor 01010101 toggle free bit
----------
01000100 = busy blocks
<< 1
----------
10001000 = mask checking the "reused" bit for the busy blocks
and 00011011 original value from bitmap on disk.
----------
00001000 non-zero indicates an illegal combination somewhere
in the int4.
*/
if ((dskmap & SIXTEEN_BLKS_FREE) != *lmap
|| ((((dskmap & SIXTEEN_BLKS_FREE) ^ SIXTEEN_BLKS_FREE) << 1) & dskmap))
{
if (agree)
{
agree = FALSE;
mu_int_path[0] = blkno;
mu_int_plen = 1;
if (mu_map_errs < disp_map_errors)
mu_int_err(ERR_DBLOCMBINC, 0, 0, 0, 0, 0, 0, level);
else
mu_int_errknt++;
}
for (bcnt = 0; bcnt < SIZEOF(int4) * BITS_PER_UCHAR / BML_BITS_PER_BLK; bcnt++)
{
if (!(mu_int_isvalid_mask[bcnt] ^ (dskmap & mu_int_mask[bcnt])))
{
mu_int_path[0] = blkno + lcnt + bcnt;
mu_int_plen = 1;
mu_int_err(ERR_DBBFSTAT,
0, 0, 0, 0, 0, 0, level);
}
else if ((lfree = mu_int_isfree_mask[bcnt] & *(lmap)) ^ mu_int_isfree_mask[bcnt] & dskmap)
{
mu_int_path[0] = blkno + lcnt + bcnt;
mu_int_plen = 1;
/* for the following two mu_int_err(), we should actually be calculating the
* actual level of the mu_int_path[0]. But this would need a read() of the block,
* which might slow down the process. We should consider this however at a
* later time. */
if (!lfree)
mu_int_err(ERR_DBMRKFREE, 0, 0, 0, 0, 0, 0, LCL_MAP_LEVL);
else if (mu_map_errs < disp_map_errors)
{
mu_int_err(ERR_DBMRKBUSY, 0, 0, 0, 0, 0, 0, LCL_MAP_LEVL);
mu_map_errs++;
} else
{
mu_int_errknt++;
mu_map_errs++;
}
}
}
}
}
if (!agree)
{
local_full = (NO_FREE_SPACE == bml_find_free(0, local, mapsize));
if (local_full || disk_full)
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
if (mu_map_errs < disp_map_errors)
{
mu_int_err(master_full ?
(local_full ? ERR_DBMBPFLINT
: ERR_DBMBPFLDLBM)
: (local_full ?
ERR_DBMBPFRDLBM
: ERR_DBMBPFRINT),
0, 0, 0, 0, 0, 0, level);
} else
mu_int_errknt++;
} else if (master_full)
{
if (mu_map_errs < disp_map_errors)
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
mu_int_err( ERR_DBMBPFLDIS,
0, 0, 0, 0, 0, 0, level);
} else
mu_int_errknt++;
}
} else if (disk_full ^ master_full)
{
if (mu_map_errs < disp_map_errors)
{
mu_int_path[0] = blkno;
mu_int_plen = 1;
mu_int_err(master_full ? ERR_DBMBPINCFL
: ERR_DBMBMINCFRE,
0, 0, 0, 0, 0, 0, level);
} else
mu_int_errknt++;
}
free(disk);
}
if (mu_map_errs >= disp_map_errors)
{
util_out_print("Maximum number of incorrectly busy errors to display: !UL, has been exceeded", TRUE,
disp_map_errors);
util_out_print("!UL incorrectly busy errors encountered", TRUE, mu_map_errs);
}
return;
}
bool mubinccpy(backup_reg_list *list)
{
static readonly mval null_str = {MV_STR, 0, 0 , 0 , 0, 0};
int backup_socket;
int4 size, size1, bsize, bm_num, hint, lmsize, save_blks, rsize, match, timeout, outsize;
uint4 status, total_blks, bplmap, gds_ratio, blks_per_buff, counter, i, lcnt, read_size;
uchar_ptr_t bm_blk_buff, ptr1, ptr1_top, ptr, ptr_top;
char_ptr_t outptr, data_ptr;
unsigned short rd_iosb[4], port;
enum db_acc_method access;
blk_hdr *bp, *bptr;
struct FAB *fcb, temp_fab, mubincfab;
struct RAB temp_rab, mubincrab;
inc_header *outbuf;
mval val;
mstr *file;
sgmnt_data_ptr_t header;
char *common, addr[SA_MAXLEN + 1];
void (*common_write)();
void (*common_close)();
muinc_blk_hdr_ptr_t sblkh_p;
trans_num blk_tn;
block_id blk_num_base, blk_num;
boolean_t is_bitmap_blk, backup_this_blk;
enum db_ver dummy_odbv;
int4 blk_bsiz;
error_def(ERR_BCKUPBUFLUSH);
error_def(ERR_COMMITWAITSTUCK);
error_def(ERR_DBCCERR);
error_def(ERR_ERRCALL);
assert(list->reg == gv_cur_region);
assert(incremental);
/* Make sure inc_header can be same size on all platforms. Some platforms pad 8 byte aligned structures
that end on a 4 byte boundary and some do not. It is critical that this structure is the same size on
all platforms as it is sent across TCP connections when doing TCP backup.
*/
assert(0 == (SIZEOF(inc_header) % 8));
/* ================= Initialization and some checks ======================== */
header = list->backup_hdr;
file = &(list->backup_file);
if (!mubtomag)
mubmaxblk = BACKUP_TEMPFILE_BUFF_SIZE;
fcb = ((vms_gds_info *)(gv_cur_region->dyn.addr->file_cntl->file_info))->fab;
if (list->tn >= header->trans_hist.curr_tn)
{
util_out_print("!/TRANSACTION number is greater than or equal to current transaction,", TRUE);
util_out_print("No blocks backed up from database !AD", TRUE, fcb->fab$b_fns, fcb->fab$l_fna);
return TRUE;
}
/* =========== open backup destination and define common_write ================= */
backup_write_errno = 0;
backup_close_errno = 0;
switch(list->backup_to)
{
case backup_to_file:
/* open the file and define the common_write function */
mubincfab = cc$rms_fab;
mubincfab.fab$b_fac = FAB$M_PUT;
mubincfab.fab$l_fop = FAB$M_CBT | FAB$M_MXV | FAB$M_TEF | FAB$M_POS & (~FAB$M_RWC) & (~FAB$M_RWO);
mubincfab.fab$l_fna = file->addr;
mubincfab.fab$b_fns = file->len;
mubincfab.fab$l_alq = cs_addrs->hdr->start_vbn +
STARTING_BLOCKS * cs_addrs->hdr->blk_size / DISK_BLOCK_SIZE;
mubincfab.fab$w_mrs = mubmaxblk;
mubincfab.fab$w_deq = EXTEND_SIZE;
switch (status = sys$create(&mubincfab))
{
case RMS$_NORMAL:
case RMS$_CREATED:
case RMS$_SUPERSEDE:
case RMS$_FILEPURGED:
break;
default:
gtm_putmsg(status, 0, mubincfab.fab$l_stv);
util_out_print("Error: Cannot create backup file !AD.",
TRUE, mubincfab.fab$b_fns, mubincfab.fab$l_fna);
return FALSE;
}
mubincrab = cc$rms_rab;
mubincrab.rab$l_fab = &mubincfab;
mubincrab.rab$l_rop = RAB$M_WBH;
if (RMS$_NORMAL != (status = sys$connect(&mubincrab)))
{
gtm_putmsg(status, 0, mubincrab.rab$l_stv);
util_out_print("Error: Cannot connect to backup file !AD.",
TRUE, mubincfab.fab$b_fns, mubincfab.fab$l_fna);
mubincfab.fab$l_fop |= FAB$M_DLT;
sys$close(&mubincfab);
return FALSE;
}
common = (char *)(&mubincrab);
common_write = file_write;
common_close = file_close;
break;
case backup_to_exec:
util_out_print("Error: Backup to pipe is yet to be implemented.", TRUE);
util_out_print("Error: Your request to backup database !AD to !AD is currently not valid.", TRUE,
fcb->fab$b_fns, fcb->fab$l_fna, file->len, file->addr);
return FALSE;
case backup_to_tcp:
iotcp_fillroutine();
/* parse it first */
switch (match = SSCANF(file->addr, "%[^:]:%hu", addr, &port))
{
case 1 :
port = DEFAULT_BKRS_PORT;
case 2 :
break;
default :
util_out_print("ERROR: A hostname has to be specified to backup through a TCP connection.",
TRUE);
return FALSE;
}
if ((0 == cli_get_int("NETTIMEOUT", &timeout)) || (0 > timeout))
timeout = DEFAULT_BKRS_TIMEOUT;
if (0 > (backup_socket = tcp_open(addr, port, timeout, FALSE)))
{
util_out_print("ERROR: Cannot open tcp connection due to the above error.", TRUE);
return FALSE;
}
common_write = tcp_write;
common_close = tcp_close;
common = (char *)(&backup_socket);
break;
default :
util_out_print("ERROR: Backup format !UL not supported.", TRUE, list->backup_to);
util_out_print("Error: Your request to backup database !AD to !AD is not valid.", TRUE,
fcb->fab$b_fns, fcb->fab$l_fna, file->len, file->addr);
return FALSE;
}
/* ============================= write inc_header =========================================== */
outptr = malloc(SIZEOF(inc_header));
outbuf = (inc_header *)outptr;
MEMCPY_LIT(&outbuf->label[0], INC_HEADER_LABEL);
stringpool.free = stringpool.base;
op_horolog(&val);
stringpool.free = stringpool.base;
op_fnzdate(&val, &mu_bin_datefmt, &null_str, &null_str, &val);
memcpy(&outbuf->date[0], val.str.addr, val.str.len);
memcpy(&outbuf->reg[0], gv_cur_region->rname, MAX_RN_LEN);
outbuf->start_tn = list->tn;
outbuf->end_tn = header->trans_hist.curr_tn;
outbuf->db_total_blks = header->trans_hist.total_blks;
outbuf->blk_size = header->blk_size;
outbuf->blks_to_upgrd = header->blks_to_upgrd;
COMMON_WRITE(common, outptr, SIZEOF(inc_header));
free(outptr);
if (mu_ctrly_occurred || mu_ctrlc_occurred)
{
error_mupip = TRUE;
COMMON_CLOSE(common);
util_out_print("WARNING: DB file !AD backup aborted.", TRUE, fcb->fab$b_fns, fcb->fab$l_fna);
return FALSE;
}
/* ============================ read/write appropriate blocks =============================== */
bsize = header->blk_size;
gds_ratio = bsize / DISK_BLOCK_SIZE;
blks_per_buff = BACKUP_READ_SIZE / bsize;
read_size = blks_per_buff * bsize;
outsize = SIZEOF(muinc_blk_hdr) + bsize;
outptr = (char_ptr_t)malloc(MAX(outsize, mubmaxblk));
sblkh_p = (muinc_blk_hdr_ptr_t)outptr;
data_ptr = (char_ptr_t)(sblkh_p + 1);
bp = (blk_hdr_ptr_t)mubbuf;
bm_blk_buff = (uchar_ptr_t)malloc(SIZEOF(blk_hdr) + (BLKS_PER_LMAP * BML_BITS_PER_BLK / BITS_PER_UCHAR));
mubincrab.rab$l_rbf = outptr;
save_blks = 0;
access = header->acc_meth;
memset(sblkh_p, 0, SIZEOF(*sblkh_p));
if (access == dba_bg)
bp = mubbuf;
else
{
ptr = cs_addrs->db_addrs[0] + (cs_addrs->hdr->start_vbn - 1) * DISK_BLOCK_SIZE;
ptr_top = cs_addrs->db_addrs[1] + 1;
}
sblkh_p->use.bkup.ondsk_blkver = GDSNOVER;
for (blk_num_base = 0; blk_num_base < header->trans_hist.total_blks; blk_num_base += blks_per_buff)
{
if (online && (0 != cs_addrs->shmpool_buffer->failed))
break;
if (header->trans_hist.total_blks - blk_num_base < blks_per_buff)
{
blks_per_buff = header->trans_hist.total_blks - blk_num_base;
read_size = blks_per_buff * bsize;
}
if (access == dba_bg)
{
if ((SS$_NORMAL != (status = sys$qiow(EFN$C_ENF, fcb->fab$l_stv, IO$_READVBLK, &rd_iosb, 0, 0, bp,
read_size, cs_addrs->hdr->start_vbn + (gds_ratio * blk_num_base),
0, 0, 0)))
|| (SS$_NORMAL != (status = rd_iosb[0])))
{
gtm_putmsg(VARLSTCNT(1) status);
util_out_print("Error reading data from database !AD.", TRUE,
fcb->fab$b_fns, fcb->fab$l_fna);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
} else
{
assert(dba_mm == access);
bp = ptr + blk_num_base * bsize;
}
bptr = (blk_hdr *)bp;
/* The blocks we back up will be whatever version they are. There is no implicit conversion in this
part of the backup/restore. Since we aren't even looking at the blocks (and indeed some of these blocks
could potentially contain unintialized garbage data), we set the block version to GDSNOVER to signal
that the block version is unknown. The above applies to "regular" blocks but not to bitmap blocks which
we know are initialized. Because we have to read the bitmap blocks, they will be converted as necessary.
*/
for (i = 0;
i < blks_per_buff && ((blk_num_base + i) < header->trans_hist.total_blks);
i++, bptr = (blk_hdr *)((char *)bptr + bsize))
{
blk_num = blk_num_base + i;
if (mu_ctrly_occurred || mu_ctrlc_occurred)
{
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
util_out_print("WARNING: DB file !AD backup aborted.", TRUE, fcb->fab$b_fns, fcb->fab$l_fna);
return FALSE;
}
/* Before we check if this block needs backing up, check if this is a new bitmap block or not. If it is,
we can fall through and back it up as normal. But if this is NOT a bitmap block, use the
existing bitmap to determine if this block has ever been allocated or not. If not, we don't want to
even look at this block. It could be uninitialized which will just make things run slower if we
go to read it and back it up.
*/
if (0 != ((BLKS_PER_LMAP - 1) & blk_num))
{ /* Not a local bitmap block */
if (!gvcst_blk_ever_allocated(bm_blk_buff + SIZEOF(blk_hdr),
((blk_num * BML_BITS_PER_BLK)
% (BLKS_PER_LMAP * BML_BITS_PER_BLK))))
continue; /* Bypass never-set blocks to avoid conversion problems */
is_bitmap_blk = FALSE;
if (SIZEOF(v15_blk_hdr) <= (blk_bsiz = ((v15_blk_hdr_ptr_t)bptr)->bsiz))
{ /* We have either a V4 block or uninitialized garbage */
if (blk_bsiz > bsize)
/* This is not a valid V4 block so ignore it */
continue;
blk_tn = ((v15_blk_hdr_ptr_t)bptr)->tn;
} else
{ /* Assume V5 block */
if ((blk_bsiz = bptr->bsiz) > bsize)
/* Not a valid V5 block either */
continue;
blk_tn = bptr->tn;
}
} else
{ /* This is a bitmap block so save it into our bitmap block buffer. It is used as the
basis of whether or not we have to process a given block or not. We process allocated and
recycled blocks leaving free (never used) blocks alone as they have no data worth saving.
But after saving it, upgrade it to the current format if necessary.
*/
is_bitmap_blk = TRUE;
memcpy(bm_blk_buff, bptr, BM_SIZE(header->bplmap));
if (SIZEOF(v15_blk_hdr) <= ((v15_blk_hdr_ptr_t)bm_blk_buff)->bsiz)
{ /* This is a V4 format block -- needs upgrading */
status = gds_blk_upgrade(bm_blk_buff, bm_blk_buff, bsize, &dummy_odbv);
if (SS_NORMAL != status)
{
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
util_out_print("Error: Block 0x!XL is too large for automatic upgrade", TRUE,
sblkh_p->blkid);
return FALSE;
}
}
assert(BM_SIZE(header->bplmap) == ((blk_hdr_ptr_t)bm_blk_buff)->bsiz);
assert(LCL_MAP_LEVL == ((blk_hdr_ptr_t)bm_blk_buff)->levl);
assert(gvcst_blk_is_allocated(bm_blk_buff + SIZEOF(blk_hdr),
((blk_num * BML_BITS_PER_BLK)
% (BLKS_PER_LMAP * BML_BITS_PER_BLK))));
blk_bsiz = BM_SIZE(header->bplmap);
blk_tn = ((blk_hdr_ptr_t)bm_blk_buff)->tn;
}
/* The conditions for backing up a block or ignoring it (in order of evaluation):
1) If blk is larger than size of db at time backup was initiated, we ignore the block.
2) Always backup blocks 0, 1, and 2 as these are the only blocks that can contain data
and still have a transaction number of 0.
3) For bitmap blocks, if blks_to_upgrd != 0 and the TN is 0 and the block number >=
last_blk_at_last_bkup, then backup the block. This way we get the correct version of
the bitmap block in the restore (otherwise have no clue what version to create them in
as bitmaps are created with a TN of 0 when before image journaling is enabled).
4) If the block TN is below our TN threshold, ignore the block.
5) Else if none of the above conditions, backup the block.
*/
if (online && (header->trans_hist.curr_tn <= blk_tn))
backup_this_blk = FALSE;
else if (3 > blk_num || (is_bitmap_blk && 0 != header->blks_to_upgrd && (trans_num)0 == blk_tn
&& blk_num >= list->last_blk_at_last_bkup))
backup_this_blk = TRUE;
else if ((blk_tn < list->tn))
backup_this_blk = FALSE;
else
backup_this_blk = TRUE;
if (!backup_this_blk)
{
if (online)
cs_addrs->nl->nbb = blk_num;
continue; /* not applicable */
}
sblkh_p->blkid = blk_num;
memcpy(data_ptr, bptr, blk_bsiz);
sblkh_p->valid_data = TRUE; /* Validation marker */
COMMON_WRITE(common, outptr, outsize);
if (online)
{
if (0 != cs_addrs->shmpool_buffer->failed)
break;
cs_addrs->nl->nbb = blk_num;
}
save_blks++;
}
}
/* ============================= write saved information for online backup ========================== */
if (online && (0 == cs_addrs->shmpool_buffer->failed))
{
/* -------- make sure everyone involved finishes -------- */
cs_addrs->nl->nbb = BACKUP_NOT_IN_PROGRESS;
/* By getting crit here, we ensure that there is no process still in transaction logic that sees
(nbb != BACKUP_NOT_IN_PRORESS). After rel_crit(), any process that enters transaction logic will
see (nbb == BACKUP_NOT_IN_PRORESS) because we just set it to that value. At this point, backup
buffer is complete and there will not be any more new entries in the backup buffer until the next
backup.
*/
grab_crit(gv_cur_region);
assert(cs_data == cs_addrs->hdr);
if (dba_bg == cs_data->acc_meth)
{ /* Now that we have crit, wait for any pending phase2 updates to finish. Since phase2 updates happen
* outside of crit, we dont want them to keep writing to the backup temporary file even after the
* backup is complete and the temporary file has been deleted.
*/
if (cs_addrs->nl->wcs_phase2_commit_pidcnt && !wcs_phase2_commit_wait(cs_addrs, NULL))
{
gtm_putmsg(VARLSTCNT(7) ERR_COMMITWAITSTUCK, 5, process_id, 1,
cs_addrs->nl->wcs_phase2_commit_pidcnt, DB_LEN_STR(gv_cur_region));
rel_crit(gv_cur_region);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
}
if (debug_mupip)
{
util_out_print("MUPIP INFO: Current Transaction # at end of backup is 0x!16@XQ", TRUE,
&cs_data->trans_hist.curr_tn);
}
rel_crit(gv_cur_region);
counter = 0;
while (0 != cs_addrs->shmpool_buffer->backup_cnt)
{
if (0 != cs_addrs->shmpool_buffer->failed)
{
util_out_print("Process !UL encountered the following error.", TRUE,
cs_addrs->shmpool_buffer->failed);
if (0 != cs_addrs->shmpool_buffer->backup_errno)
gtm_putmsg(VARLSTCNT(1) cs_addrs->shmpool_buffer->backup_errno);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
backup_buffer_flush(gv_cur_region);
if (++counter > MAX_BACKUP_FLUSH_TRY)
{
gtm_putmsg(VARLSTCNT(1) ERR_BCKUPBUFLUSH);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
if (counter & 0xF)
wcs_sleep(counter);
else
{ /* Force shmpool recovery to see if it can find the lost blocks */
if (!shmpool_lock_hdr(gv_cur_region))
{
gtm_putmsg(VARLSTCNT(9) ERR_DBCCERR, 2, REG_LEN_STR(gv_cur_region),
ERR_ERRCALL, 3, CALLFROM);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
assert(FALSE);
return FALSE;;
}
shmpool_abandoned_blk_chk(gv_cur_region, TRUE);
shmpool_unlock_hdr(gv_cur_region);
}
}
/* -------- Open the temporary file -------- */
temp_fab = cc$rms_fab;
temp_fab.fab$b_fac = FAB$M_GET;
temp_fab.fab$l_fna = list->backup_tempfile;
temp_fab.fab$b_fns = strlen(list->backup_tempfile);
temp_rab = cc$rms_rab;
temp_rab.rab$l_fab = &temp_fab;
for (lcnt = 1; MAX_OPEN_RETRY >= lcnt; lcnt++)
{
if (RMS$_FLK != (status = sys$open(&temp_fab, NULL, NULL)))
break;
wcs_sleep(lcnt);
}
if (RMS$_NORMAL != status)
{
gtm_putmsg(status, 0, temp_fab.fab$l_stv);
util_out_print("WARNING: DB file !AD backup aborted.", TRUE, fcb->fab$b_fns, fcb->fab$l_fna);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
if (RMS$_NORMAL != (status = sys$connect(&temp_rab)))
{
gtm_putmsg(status, 0, temp_rab.rab$l_stv);
util_out_print("WARNING: DB file !AD backup aborted.", TRUE, fcb->fab$b_fns, fcb->fab$l_fna);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
/* -------- read and write every record in the temporary file -------- */
while (1)
{
temp_rab.rab$w_usz = outsize;
temp_rab.rab$l_ubf = outptr;
status = sys$get(&temp_rab);
if (RMS$_NORMAL != status)
{
if (RMS$_EOF == status)
status = RMS$_NORMAL;
break;
}
assert(outsize == temp_rab.rab$w_rsz);
/* Still validly sized blk? */
assert((outsize - SIZEOF(shmpool_blk_hdr)) >= ((blk_hdr_ptr_t)(outptr + SIZEOF(shmpool_blk_hdr)))->bsiz);
COMMON_WRITE(common, outptr, temp_rab.rab$w_rsz);
}
if (RMS$_NORMAL != status)
{
gtm_putmsg(status, 0, temp_rab.rab$l_stv);
util_out_print("WARNING: DB file !AD backup aborted.", TRUE, fcb->fab$b_fns, fcb->fab$l_fna);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
/* ---------------- Close the temporary file ----------------------- */
if (RMS$_NORMAL != (status = sys$close(&temp_fab)))
{
gtm_putmsg(status, 0, temp_fab.fab$l_stv);
util_out_print("WARNING: DB file !AD backup aborted.", TRUE, fcb->fab$b_fns, fcb->fab$l_fna);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
}
/* ============================= write end_msg and fileheader ======================================= */
if ((!online) || (0 == cs_addrs->shmpool_buffer->failed))
{
MEMCPY_LIT(outptr, END_MSG);
/* Although the write only need be of length SIZEOF(END_MSG) - 1 for file IO, if the write is going
to TCP we have to write all these records with common length so just write the "regular" sized
buffer. The extra garbage left over from the last write will be ignored as we key only on the
this end text.
*/
COMMON_WRITE(common, outptr, outsize);
ptr1 = header;
size1 = ROUND_UP(SIZEOF(sgmnt_data), DISK_BLOCK_SIZE);
ptr1_top = ptr1 + size1;
for (;ptr1 < ptr1_top ; ptr1 += size1)
{
if ((size1 = ptr1_top - ptr1) > mubmaxblk)
size1 = (mubmaxblk / DISK_BLOCK_SIZE) * DISK_BLOCK_SIZE;
COMMON_WRITE(common, ptr1, size1);
}
MEMCPY_LIT(outptr, HDR_MSG);
COMMON_WRITE(common, outptr, SIZEOF(HDR_MSG));
ptr1 = MM_ADDR(header);
size1 = ROUND_UP(MASTER_MAP_SIZE(header), DISK_BLOCK_SIZE);
ptr1_top = ptr1 + size1;
for (;ptr1 < ptr1_top ; ptr1 += size1)
{
if ((size1 = ptr1_top - ptr1) > mubmaxblk)
size1 = (mubmaxblk / DISK_BLOCK_SIZE) * DISK_BLOCK_SIZE;
COMMON_WRITE(common, ptr1, size1);
}
MEMCPY_LIT(outptr, MAP_MSG);
COMMON_WRITE(common, outptr, SIZEOF(MAP_MSG));
}
/* ================== close backup destination, output and return ================================== */
if (online && (0 != cs_addrs->shmpool_buffer->failed))
{
util_out_print("Process !UL encountered the following error.", TRUE,
cs_addrs->shmpool_buffer->failed);
if (0 != cs_addrs->shmpool_buffer->backup_errno)
gtm_putmsg(VARLSTCNT(1) cs_addrs->shmpool_buffer->backup_errno);
free(outptr);
free(bm_blk_buff);
error_mupip = TRUE;
COMMON_CLOSE(common);
return FALSE;
}
COMMON_CLOSE(common);
free(outptr);
free(bm_blk_buff);
util_out_print("DB file !AD incrementally backed up in !AD", TRUE,
fcb->fab$b_fns, fcb->fab$l_fna, file->len, file->addr);
util_out_print("!UL blocks saved.", TRUE, save_blks);
util_out_print("Transactions from 0x!16@XQ to 0x!16@XQ are backed up.", TRUE,
&cs_addrs->shmpool_buffer->inc_backup_tn, &header->trans_hist.curr_tn);
cs_addrs->hdr->last_inc_backup = header->trans_hist.curr_tn;
if (record)
cs_addrs->hdr->last_rec_backup = header->trans_hist.curr_tn;
file_backed_up = TRUE;
return TRUE;
}
block_id bm_getfree(block_id orig_hint, boolean_t *blk_used, unsigned int cw_work, cw_set_element *cs, int *cw_depth_ptr)
{
cw_set_element *cs1;
sm_uc_ptr_t bmp;
block_id bml, hint, hint_cycled, hint_limit;
block_id_ptr_t b_ptr;
int cw_set_top, depth, lcnt;
unsigned int local_maps, map_size, n_decrements = 0, total_blks;
trans_num ctn;
int4 free_bit, offset;
uint4 space_needed;
uint4 status;
srch_blk_status blkhist;
total_blks = (dba_mm == cs_data->acc_meth) ? cs_addrs->total_blks : cs_addrs->ti->total_blks;
if (orig_hint >= total_blks) /* for TP, hint can be > total_blks */
orig_hint = 1;
hint = orig_hint;
hint_cycled = DIVIDE_ROUND_UP(total_blks, BLKS_PER_LMAP);
hint_limit = DIVIDE_ROUND_DOWN(orig_hint, BLKS_PER_LMAP);
local_maps = hint_cycled + 2; /* for (up to) 2 wraps */
for (lcnt = 0; lcnt <= local_maps; lcnt++)
{
bml = bmm_find_free(hint / BLKS_PER_LMAP, (sm_uc_ptr_t)MM_ADDR(cs_data), local_maps);
if ((NO_FREE_SPACE == bml) || (bml >= hint_cycled))
{ /* if no free space or might have looped to original map, extend */
if ((NO_FREE_SPACE != bml) && (hint_limit < hint_cycled))
{
hint_cycled = hint_limit;
hint = 1;
continue;
}
if (SS_NORMAL != (status = gdsfilext(cs_data->extension_size, total_blks)))
return (status);
if (dba_mm == cs_data->acc_meth)
return (FILE_EXTENDED);
hint = total_blks;
total_blks = cs_addrs->ti->total_blks;
hint_cycled = DIVIDE_ROUND_UP(total_blks, BLKS_PER_LMAP);
local_maps = hint_cycled + 2; /* for (up to) 2 wraps */
/*
* note that you can make an optimization of not going back over the whole database and going over
* only the extended section. but since it is very unlikely that a free block won't be found
* in the extended section and the fact that we are starting from the extended section in either
* approach and the fact that we have a GTMASSERT to check that we don't have a lot of
* free blocks while doing an extend and the fact that it is very easy to make the change to do
* a full-pass, the full-pass solution is currently being implemented
*/
lcnt = -1; /* allow it one extra pass to ensure that it can take advantage of the entension */
n_decrements++; /* used only for debugging purposes */
continue;
}
bml *= BLKS_PER_LMAP;
if (ROUND_DOWN2(hint, BLKS_PER_LMAP) != bml)
{ /* not within requested map */
if ((bml < hint) && (hint_cycled)) /* wrap? - second one should force an extend for sure */
hint_cycled = (hint_limit < hint_cycled) ? hint_limit: 0;
hint = bml + 1; /* start at beginning */
}
if (ROUND_DOWN2(total_blks, BLKS_PER_LMAP) == bml)
map_size = (total_blks - bml);
else
map_size = BLKS_PER_LMAP;
if (0 != dollar_tlevel)
{
depth = cw_work;
cw_set_top = *cw_depth_ptr;
if (depth < cw_set_top)
tp_get_cw(cs, cw_work, &cs1);
for (; depth < cw_set_top; depth++, cs1 = cs1->next_cw_set)
{ /* do tp front to back because list is more efficient than tp_get_cw and forward pointers exist */
if (bml == cs1->blk)
{
TRAVERSE_TO_LATEST_CSE(cs1);
break;
}
}
if (depth >= cw_set_top)
{
assert(cw_set_top == depth);
depth = 0;
}
} else
{
for (depth = *cw_depth_ptr - 1; depth >= cw_work; depth--)
{ /* do non-tp back to front, because of adjacency */
if (bml == (cs + depth)->blk)
{
cs1 = cs + depth;
break;
}
}
if (depth < cw_work)
{
assert(cw_work - 1 == depth);
depth = 0;
}
}
if (0 == depth)
{
ctn = cs_addrs->ti->curr_tn;
if (!(bmp = t_qread(bml, (sm_int_ptr_t)&blkhist.cycle, &blkhist.cr)))
return MAP_RD_FAIL;
if ((BM_SIZE(BLKS_PER_LMAP) != ((blk_hdr_ptr_t)bmp)->bsiz) || (LCL_MAP_LEVL != ((blk_hdr_ptr_t)bmp)->levl))
{
assert(CDB_STAGNATE > t_tries);
rdfail_detail = cdb_sc_badbitmap;
return MAP_RD_FAIL;
}
offset = 0;
} else
{
bmp = cs1->old_block;
b_ptr = (block_id_ptr_t)(cs1->upd_addr);
b_ptr += cs1->reference_cnt - 1;
offset = *b_ptr + 1;
}
if (offset < map_size)
{
free_bit = bm_find_blk(offset, (sm_uc_ptr_t)bmp + sizeof(blk_hdr), map_size, blk_used);
if (MAP_RD_FAIL == free_bit)
return MAP_RD_FAIL;
} else
free_bit = NO_FREE_SPACE;
if (NO_FREE_SPACE != free_bit)
break;
if ((hint = bml + BLKS_PER_LMAP) >= total_blks) /* if map is full, start at 1st blk in next map */
{ /* wrap - second one should force an extend for sure */
hint = 1;
if (hint_cycled)
hint_cycled = (hint_limit < hint_cycled) ? hint_limit: 0;
}
if ((0 == depth) && (FALSE != cs_addrs->now_crit)) /* if it's from the cw_set, its state is murky */
bit_clear(bml / BLKS_PER_LMAP, MM_ADDR(cs_data)); /* if crit, repair master map error */
}
/* If not in the final retry, it is possible that free_bit is >= map_size (e.g. if bitmap block gets recycled). */
if (map_size <= (uint4)free_bit && CDB_STAGNATE <= t_tries)
{ /* bad free bit */
assert((NO_FREE_SPACE == free_bit) && (lcnt > local_maps)); /* All maps full, should have extended */
GTMASSERT;
}
if (0 != depth)
{
b_ptr = (block_id_ptr_t)(cs1->upd_addr);
b_ptr += cs1->reference_cnt++;
*b_ptr = free_bit;
} else
{
space_needed = (BLKS_PER_LMAP + 1) * sizeof(block_id);
if (dollar_tlevel)
{
ENSURE_UPDATE_ARRAY_SPACE(space_needed); /* have brackets for "if" for macros */
}
BLK_ADDR(b_ptr, space_needed, block_id);
memset(b_ptr, 0, space_needed);
*b_ptr = free_bit;
blkhist.blk_num = bml;
blkhist.buffaddr = bmp; /* cycle and cr have already been assigned from t_qread */
t_write_map(&blkhist, (uchar_ptr_t)b_ptr, ctn, 1); /* last parameter 1 is what cs->reference_cnt gets set to */
}
return bml + free_bit;
}
void 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;
}
/******************************************************************************************
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_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_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);
}
