int jnl_file_extend(jnl_private_control *jpc, uint4 total_jnl_rec_size)
{
file_control *fc;
boolean_t need_extend;
jnl_buffer_ptr_t jb;
jnl_create_info jnl_info;
jnl_file_header header;
uint4 new_alq;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
char prev_jnl_fn[JNL_NAME_SIZE];
uint4 jnl_status = 0, status;
int new_blocks, result;
GTM_BAVAIL_TYPE avail_blocks;
uint4 aligned_tot_jrec_size, count;
switch(jpc->region->dyn.addr->acc_meth)
{
case dba_mm:
case dba_bg:
csa = &FILE_INFO(jpc->region)->s_addrs;
break;
default:
GTMASSERT;
}
csd = csa->hdr;
assert(csa == cs_addrs && csd == cs_data);
assert(csa->now_crit || (csd->clustered && (CCST_CLOSED == csa->nl->ccp_state)));
assert(jpc->region == gv_cur_region);
assert(csa->jnl_state == csd->jnl_state);
if (!JNL_ENABLED(csa) || (NOJNL == jpc->channel) || (JNL_FILE_SWITCHED(jpc)))
GTMASSERT; /* crit and messing with the journal file - how could it have vanished? */
if (!csd->jnl_deq)
{
assert(DIVIDE_ROUND_UP(total_jnl_rec_size, DISK_BLOCK_SIZE) <= csd->jnl_alq);
assert(csd->jnl_alq == csd->autoswitchlimit);
new_blocks = csd->jnl_alq;
} else
/* May cause extension of csd->jnl_deq * n blocks where n > 0 */
new_blocks = ROUND_UP(DIVIDE_ROUND_UP(total_jnl_rec_size, DISK_BLOCK_SIZE), csd->jnl_deq);
jpc->status = SS_NORMAL;
jb = jpc->jnl_buff;
assert(0 <= new_blocks);
DEBUG_ONLY(count = 0);
for (need_extend = (0 != new_blocks); need_extend; )
{
DEBUG_ONLY(count++);
/* usually we will do the loop just once where we do the file extension.
* rarely we might need to do an autoswitch instead after which again rarely
* we might need to do an extension on the new journal to fit in the transaction's journal requirements.
* therefore we should do this loop a maximum of twice. hence the assert below.
*/
assert(count <= 2);
need_extend = FALSE;
if (SS_NORMAL == (status = disk_block_available(jpc->channel, &avail_blocks, TRUE)))
{
if ((new_blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
if (new_blocks > avail_blocks)
{ /* if we cannot satisfy the request, it is an error */
send_msg(VARLSTCNT(6) ERR_NOSPACEEXT, 4, JNL_LEN_STR(csd),
new_blocks, avail_blocks);
new_blocks = 0;
jpc->status = SS_NORMAL;
break;
} else
send_msg(VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, JNL_LEN_STR(csd),
(avail_blocks - new_blocks));
}
} else
send_msg(VARLSTCNT(5) ERR_JNLFILEXTERR, 2, JNL_LEN_STR(csd), status);
new_alq = jb->filesize + new_blocks;
/* ensure current journal file size is well within autoswitchlimit --> design constraint */
assert(csd->autoswitchlimit >= jb->filesize);
if (csd->autoswitchlimit < (jb->filesize + (EXTEND_WARNING_FACTOR * new_blocks))) /* close to max */
send_msg(VARLSTCNT(5) ERR_JNLSPACELOW, 3, JNL_LEN_STR(csd), csd->autoswitchlimit - jb->filesize);
if (csd->autoswitchlimit < new_alq)
{ /* Reached max, need to autoswitch */
/* Ensure new journal file can hold the entire current transaction's journal record requirements */
assert(csd->autoswitchlimit >= MAX_REQD_JNL_FILE_SIZE(total_jnl_rec_size));
memset(&jnl_info, 0, sizeof(jnl_info));
jnl_info.prev_jnl = &prev_jnl_fn[0];
set_jnl_info(gv_cur_region, &jnl_info);
assert(JNL_ENABLED(csa) && (NOJNL != jpc->channel) && !(JNL_FILE_SWITCHED(jpc)));
jnl_status = jnl_ensure_open();
if (0 == jnl_status)
{ /* flush the cache and jnl-buffer-contents to current journal file before
* switching to a new journal. */
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH);
jnl_file_close(gv_cur_region, TRUE, TRUE);
} else
rts_error(VARLSTCNT(7) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region), jpc->status);
assert(!jgbl.forw_phase_recovery || (NULL != jgbl.mur_pini_addr_reset_fnptr));
if (jgbl.forw_phase_recovery && (NULL != jgbl.mur_pini_addr_reset_fnptr))
(*jgbl.mur_pini_addr_reset_fnptr)();
assert(!jnl_info.no_rename);
assert(!jnl_info.no_prev_link);
if (EXIT_NRM == cre_jnl_file(&jnl_info))
{
assert(0 == memcmp(csd->jnl_file_name, jnl_info.jnl, jnl_info.jnl_len));
assert(csd->jnl_file_name[jnl_info.jnl_len] == '\0');
assert(csd->jnl_file_len == jnl_info.jnl_len);
assert(csd->jnl_buffer_size == jnl_info.buffer);
assert(csd->jnl_alq == jnl_info.alloc);
assert(csd->jnl_deq == jnl_info.extend);
assert(csd->jnl_before_image == jnl_info.before_images);
csd->trans_hist.header_open_tn = jnl_info.tn; /* needed for successful jnl_file_open() */
send_msg(VARLSTCNT(4) ERR_NEWJNLFILECREATE, 2, JNL_LEN_STR(csd));
fc = gv_cur_region->dyn.addr->file_cntl;
fc->op = FC_WRITE;
fc->op_buff = (sm_uc_ptr_t)csd;
status = dbfilop(fc);
if (SS_NORMAL != status)
send_msg(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), status);
assert(JNL_ENABLED(csa));
/* call jnl_ensure_open instead of jnl_file_open to make sure jpc->pini_addr is set to 0 */
jnl_status = jnl_ensure_open(); /* sets jpc->status */
if (0 != jnl_status)
rts_error(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
assert(jb->filesize == csd->jnl_alq);
aligned_tot_jrec_size = ALIGNED_ROUND_UP(MAX_REQD_JNL_FILE_SIZE(total_jnl_rec_size),
csd->jnl_alq, csd->jnl_deq);
if (aligned_tot_jrec_size > csd->jnl_alq)
{ /* need to extend more than initial allocation in the new journal file
* to accommodate the current transaction.
*/
new_blocks = aligned_tot_jrec_size - csd->jnl_alq;
assert(new_blocks);
assert(0 == new_blocks % csd->jnl_deq);
need_extend = TRUE;
}
} else
{
send_msg(VARLSTCNT(4) ERR_JNLCREATERR, 2, JNL_LEN_STR(csd));
jpc->status = ERR_JNLNOCREATE;
new_blocks = -1;
}
} else
{
assert(!need_extend); /* ensure we won't go through the for loop again */
/* Virtually extend currently used journal file */
jb->filesize = new_alq; /* Actually this is virtual file size blocks */
DO_FILE_READ(jpc->channel, 0, &header, JNL_HDR_LEN, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
rts_error(VARLSTCNT(5) ERR_JNLRDERR, 2, JNL_LEN_STR(csd), jpc->status);
assert((header.virtual_size + new_blocks) == new_alq);
header.virtual_size = new_alq;
DO_FILE_WRITE(jpc->channel, 0, &header, JNL_HDR_LEN, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
rts_error(VARLSTCNT(5) ERR_JNLWRERR, 2, JNL_LEN_STR(csd), jpc->status);
}
if (0 >= new_blocks)
break;
}
if (0 >= new_blocks)
{
jpc->status = ERR_JNLREADEOF;
jnl_file_lost(jpc, ERR_JNLEXTEND);
new_blocks = -1;
}
return new_blocks;
}
uint4 gdsfilext(uint4 blocks, uint4 filesize, boolean_t trans_in_prog)
{
sm_uc_ptr_t old_base[2], mmap_retaddr;
boolean_t was_crit, is_mm;
int result, save_errno, status;
DEBUG_ONLY(int first_save_errno);
uint4 new_bit_maps, bplmap, map, new_blocks, new_total, max_tot_blks, old_total;
uint4 jnl_status;
gtm_uint64_t avail_blocks, mmap_sz;
off_t new_eof, new_size;
trans_num curr_tn;
unix_db_info *udi;
inctn_opcode_t save_inctn_opcode;
int4 prev_extend_blks_to_upgrd;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
cache_rec_ptr_t cr;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(!IS_DSE_IMAGE);
assert((cs_addrs->nl == NULL) || (process_id != cs_addrs->nl->trunc_pid)); /* mu_truncate shouldn't extend file... */
assert(!process_exiting);
DEBUG_ONLY(old_base[0] = old_base[1] = NULL);
assert(!gv_cur_region->read_only);
udi = FILE_INFO(gv_cur_region);
is_mm = (dba_mm == cs_addrs->hdr->acc_meth);
# if !defined(MM_FILE_EXT_OK)
if (!udi->grabbed_access_sem && is_mm)
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not allowed ? */
# endif
/* Both blocks and total blocks are unsigned ints so make sure we aren't asking for huge numbers that will
overflow and end up doing silly things.
*/
assert((blocks <= (MAXTOTALBLKS(cs_data) - cs_data->trans_hist.total_blks)) || WBTEST_ENABLED(WBTEST_FILE_EXTEND_ERROR));
# if defined(__sun) || defined(__hpux)
cs_data->defer_allocate = TRUE;
# endif
if (!blocks && (cs_data->defer_allocate || (TRANS_IN_PROG_TRUE == trans_in_prog)))
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not enabled ? */
bplmap = cs_data->bplmap;
/* New total of non-bitmap blocks will be number of current, non-bitmap blocks, plus new blocks desired
* There are (bplmap - 1) non-bitmap blocks per bitmap, so add (bplmap - 2) to number of non-bitmap blocks
* and divide by (bplmap - 1) to get total number of bitmaps for expanded database. (must round up in this
* manner as every non-bitmap block must have an associated bitmap)
* Current number of bitmaps is (total number of current blocks + bplmap - 1) / bplmap.
* Subtract current number of bitmaps from number needed for expanded database to get number of new bitmaps needed.
*/
new_bit_maps = DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap) + blocks, bplmap - 1)
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap);
new_blocks = blocks + new_bit_maps;
assert((0 < (int)new_blocks) || (!cs_data->defer_allocate && (0 == new_blocks)));
if (new_blocks + cs_data->trans_hist.total_blks > MAXTOTALBLKS(cs_data))
{
assert(WBTEST_ENABLED(WBTEST_FILE_EXTEND_ERROR));
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(1) ERR_TOTALBLKMAX);
return (uint4)(NO_FREE_SPACE);
}
if (0 != (save_errno = disk_block_available(udi->fd, &avail_blocks, FALSE)))
{
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
} else
{
if (!(gtmDebugLevel & GDL_IgnoreAvailSpace))
{ /* Bypass this space check if debug flag above is on. Allows us to create a large sparce DB
* in space it could never fit it if wasn't sparse. Needed for some tests.
*/
avail_blocks = avail_blocks / (cs_data->blk_size / DISK_BLOCK_SIZE);
if ((blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
if (blocks > (uint4)avail_blocks)
{
if (!INST_FREEZE_ON_NOSPC_ENABLED(cs_addrs))
return (uint4)(NO_FREE_SPACE);
else
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(6) MAKE_MSG_WARNING(ERR_NOSPACEEXT), 4,
DB_LEN_STR(gv_cur_region), new_blocks, (uint4)avail_blocks);
} else
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, DB_LEN_STR(gv_cur_region),
(uint4)(avail_blocks - ((new_blocks <= avail_blocks) ? new_blocks : 0)));
}
}
}
# ifdef DEBUG
if (WBTEST_ENABLED(WBTEST_MM_CONCURRENT_FILE_EXTEND) && dollar_tlevel && !MEMCMP_LIT(gv_cur_region->rname, "DEFAULT"))
{
SYSTEM("$gtm_dist/mumps -run $gtm_wbox_mrtn");
assert(1 == cs_addrs->nl->wbox_test_seq_num); /* should have been set by mubfilcpy */
cs_addrs->nl->wbox_test_seq_num = 2; /* signal mupip backup to stop sleeping in mubfilcpy */
}
# endif
/* From here on, we need to use GDSFILEXT_CLNUP before returning to the caller */
was_crit = cs_addrs->now_crit;
assert(!cs_addrs->hold_onto_crit || was_crit);
/* If we are coming from mupip_extend (which gets crit itself) we better have waited for any unfreezes to occur.
* If we are coming from online rollback (when that feature is available), we will come in holding crit and in
* the final retry. In that case too, we expect to have waited for unfreezes to occur in the caller itself.
* Therefore if we are coming in holding crit from MUPIP, we expect the db to be unfrozen so no need to wait for
* freeze.
* If we are coming from GT.M and final retry (in which case we come in holding crit) we expect to have waited
* for any unfreezes (by invoking tp_crit_all_regions) to occur (TP or non-TP) before coming into this
* function. However, there is one exception. In the final retry, if tp_crit_all_regions notices that
* at least one of the participating regions did ONLY READs, it will not wait for any freeze on THAT region
* to complete before grabbing crit. Later, in the final retry, if THAT region did an update which caused
* op_tcommit to invoke bm_getfree->gdsfilext, then we would have come here with a frozen region on which
* we hold crit.
*/
assert(!was_crit || !FROZEN_HARD(cs_data) || (dollar_tlevel && (CDB_STAGNATE <= t_tries)));
/*
* If we are in the final retry and already hold crit, it is possible that csa->nl->wc_blocked is also set to TRUE
* (by a concurrent process in phase2 which encountered an error in the midst of commit and secshr_db_clnup
* finished the job for it). In this case we do NOT want to invoke wcs_recover as that will update the "bt"
* transaction numbers without correspondingly updating the history transaction numbers (effectively causing
* a cdb_sc_blkmod type of restart). Therefore do NOT call grab_crit (which unconditionally invokes wcs_recover)
* if we already hold crit.
*/
if (!was_crit)
{
for ( ; ; )
{
grab_crit(gv_cur_region);
if (FROZEN_CHILLED(cs_data))
DO_CHILLED_AUTORELEASE(cs_addrs, cs_data);
if (!FROZEN(cs_data) && !IS_REPL_INST_FROZEN)
break;
rel_crit(gv_cur_region);
while (FROZEN(cs_data) || IS_REPL_INST_FROZEN)
{
hiber_start(1000);
if (FROZEN_CHILLED(cs_data) && CHILLED_AUTORELEASE(cs_data))
break;
}
}
} else if (FROZEN_HARD(cs_data) && dollar_tlevel)
{ /* We don't want to continue with file extension as explained above. Hence return with an error code which
* op_tcommit will recognize (as a cdb_sc_needcrit/cdb_sc_instancefreeze type of restart) and restart accordingly.
*/
assert(CDB_STAGNATE <= t_tries);
GDSFILEXT_CLNUP;
return (uint4)FINAL_RETRY_FREEZE_PROG;
} else
WAIT_FOR_REGION_TO_UNCHILL(cs_addrs, cs_data);
if (IS_REPL_INST_FROZEN && trans_in_prog)
{
assert(CDB_STAGNATE <= t_tries);
GDSFILEXT_CLNUP;
return (uint4)FINAL_RETRY_INST_FREEZE;
}
assert(cs_addrs->ti->total_blks == cs_data->trans_hist.total_blks);
old_total = cs_data->trans_hist.total_blks;
if (old_total != filesize)
{ /* Somebody else has already extended it, since we are in crit, this is trust-worthy. However, in case of MM,
* we still need to remap the database
*/
assert((old_total > filesize) || !is_mm);
/* For BG, someone else could have truncated or extended - we have no idea */
GDSFILEXT_CLNUP;
return (SS_NORMAL);
}
if (trans_in_prog && SUSPICIOUS_EXTEND)
{
if (!was_crit)
{
GDSFILEXT_CLNUP;
return (uint4)(EXTEND_SUSPECT);
}
/* If free_blocks counter is not ok, then correct it. Do the check again. If still fails, then it means we held
* crit through bm_getfree into gdsfilext and still didn't get it right.
*/
assertpro(!is_free_blks_ctr_ok() && !SUSPICIOUS_EXTEND);
}
if (JNL_ENABLED(cs_data))
{
if (!jgbl.dont_reset_gbl_jrec_time)
SET_GBL_JREC_TIME; /* needed before jnl_ensure_open as that can write jnl records */
jpc = cs_addrs->jnl;
jbp = jpc->jnl_buff;
/* Before writing to jnlfile, adjust jgbl.gbl_jrec_time if needed to maintain time order
* of jnl records. This needs to be done BEFORE the jnl_ensure_open as that could write
* journal records (if it decides to switch to a new journal file).
*/
ADJUST_GBL_JREC_TIME(jgbl, jbp);
jnl_status = jnl_ensure_open(gv_cur_region, cs_addrs);
if (jnl_status)
{
GDSFILEXT_CLNUP;
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(cs_data), DB_LEN_STR(gv_cur_region));
return (uint4)(NO_FREE_SPACE); /* should have better return status */
}
}
if (is_mm)
{
cs_addrs->nl->mm_extender_pid = process_id;
status = wcs_wtstart(gv_cur_region, 0, NULL, NULL);
cs_addrs->nl->mm_extender_pid = 0;
assertpro(SS_NORMAL == status);
old_base[0] = cs_addrs->db_addrs[0];
old_base[1] = cs_addrs->db_addrs[1];
cs_addrs->db_addrs[0] = NULL; /* don't rely on it until the mmap below */
# ifdef _AIX
status = shmdt(old_base[0] - BLK_ZERO_OFF(cs_data->start_vbn));
# else
status = munmap((caddr_t)old_base[0], (size_t)(old_base[1] - old_base[0]));
# endif
if (0 != status)
{
save_errno = errno;
GDSFILEXT_CLNUP;
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(12) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region),
ERR_SYSCALL, 5, LEN_AND_STR(MEM_UNMAP_SYSCALL), CALLFROM, save_errno);
return (uint4)(NO_FREE_SPACE);
}
} else
{ /* Due to concurrency issues, it is possible some process had issued a disk read of the GDS block# corresponding
* to "old_total" right after a truncate wrote a GDS-block of zeros on disk (to signal end of the db file).
* If so, the global buffer containing this block needs to be invalidated now as part of the extend. If not, it is
* possible the EOF block on disk is now going to be overwritten by a properly initialized bitmap block (as part
* of the gdsfilext below) while the global buffer continues to have an incorrect copy of that bitmap block and
* this in turn would cause XXXX failures due to a bad bitmap block in shared memory. (GTM-7519)
*/
cr = db_csh_get((block_id)old_total);
if ((NULL != cr) && ((cache_rec_ptr_t)CR_NOTVALID != cr))
{
assert((0 == cr->dirty) && (0 == cr->bt_index) && !cr->stopped);
cr->cycle++;
cr->blk = CR_BLKEMPTY;
}
}
CHECK_TN(cs_addrs, cs_data, cs_data->trans_hist.curr_tn); /* can issue rts_error TNTOOLARGE */
new_total = old_total + new_blocks;
new_eof = BLK_ZERO_OFF(cs_data->start_vbn) + ((off_t)new_total * cs_data->blk_size);
# if !defined(__sun) && !defined(__hpux)
if (!cs_data->defer_allocate)
{
new_size = new_eof + cs_data->blk_size;
save_errno = posix_fallocate(udi->fd, 0, new_size);
DEBUG_ONLY(first_save_errno = save_errno);
if ((ENOSPC == save_errno) && IS_GTM_IMAGE)
save_errno = extend_wait_for_fallocate(udi, new_size);
if (0 != save_errno)
{
GDSFILEXT_CLNUP;
assert(ENOSPC == save_errno);
if (ENOSPC != save_errno)
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_PREALLOCATEFAIL, 2, DB_LEN_STR(gv_cur_region),
save_errno);
return (uint4)(NO_FREE_SPACE);
}
}
# endif
save_errno = db_write_eof_block(udi, udi->fd, cs_data->blk_size, new_eof, &(TREF(dio_buff)));
if ((ENOSPC == save_errno) && IS_GTM_IMAGE)
save_errno = extend_wait_for_write(udi, cs_data->blk_size, new_eof);
if (0 != save_errno)
{
GDSFILEXT_CLNUP;
if (ENOSPC != save_errno)
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
return (uint4)(NO_FREE_SPACE);
}
if (WBTEST_ENABLED(WBTEST_FILE_EXTEND_INTERRUPT_1))
{
LONG_SLEEP(600);
assert(FALSE);
}
/* Ensure the EOF and metadata get to disk BEFORE any bitmap writes. Otherwise, the file size could no longer reflect
* a proper extent and subsequent invocations of gdsfilext could corrupt the database.
*/
if (!IS_STATSDB_CSA(cs_addrs))
{
GTM_DB_FSYNC(cs_addrs, udi->fd, status);
assert(0 == status);
if (0 != status)
{
GDSFILEXT_CLNUP;
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(8) ERR_DBFILERR, 5,
RTS_ERROR_LITERAL("fsync1()"), CALLFROM, status);
return (uint4)(NO_FREE_SPACE);
}
}
if (WBTEST_ENABLED(WBTEST_FILE_EXTEND_INTERRUPT_2))
{
LONG_SLEEP(600);
assert(FALSE); /* Should be killed before that */
}
DEBUG_ONLY(prev_extend_blks_to_upgrd = cs_data->blks_to_upgrd;)
unsigned char mu_cre_file(void)
{
unsigned char *inadr[2], *c, exit_stat;
enum db_acc_method temp_acc_meth;
uint4 lcnt, retadr[2];
int4 blk_init_size, initial_alq, free_blocks;
gtm_uint64_t free_blocks_ll, blocks_for_extension;
char buff[GLO_NAME_MAXLEN], fn_buff[MAX_FN_LEN];
unsigned int status;
int free_space;
struct FAB *fcb;
struct NAM nam;
gds_file_id new_id;
io_status_block_disk iosb;
char node[16];
short len;
struct {
short blen;
short code;
char *buf;
short *len;
int4 terminator;
} item = {15, SYI$_NODENAME, &node, &len, 0};
$DESCRIPTOR(desc, buff);
exit_stat = EXIT_NRM;
/* The following calculations should duplicate the BT_SIZE macro from GDSBT and the LOCK_BLOCK macro from GDSFHEAD.H,
* but without using a sgmnt_data which is not yet set up at this point
*/
#ifdef GT_CX_DEF
/* This section needs serious chnages for the fileheader changes in V5 if it is ever resurrected */
over_head = DIVIDE_ROUND_UP(SIZEOF_FILE_HDR_DFLT
+ (WC_MAX_BUFFS + getprime(WC_MAX_BUFFS) + 1) * SIZEOF(bt_rec), DISK_BLOCK_SIZE);
if (gv_cur_region->dyn.addr->acc_meth == dba_bg)
{
free_space = over_head - DIVIDE_ROUND_UP(SIZEOF_FILE_HDR_DFLT
+ (gv_cur_region->dyn.addr->global_buffers + getprime(gv_cur_region->dyn.addr->global_buffers) + 1)
* SIZEOF(bt_rec), DISK_BLOCK_SIZE);
over_head += gv_cur_region->dyn.addr->lock_space ? gv_cur_region->dyn.addr->lock_space
: DEF_LOCK_SIZE / OS_PAGELET_SIZE;
} else if (gv_cur_region->dyn.addr->acc_meth == dba_mm)
{
free_space = over_head - DIVIDE_ROUND_UP(SIZEOF_FILE_HDR_DFLT, DISK_BLOCK_SIZE);
if (gv_cur_region->dyn.addr->lock_space)
{
over_head += gv_cur_region->dyn.addr->lock_space;
free_space += gv_cur_region->dyn.addr->lock_space;
} else
{
over_head += DEF_LOCK_SIZE / OS_PAGELET_SIZE;
free_space += DEF_LOCK_SIZE / OS_PAGELET_SIZE;
}
}
free_space *= DISK_BLOCK_SIZE;
#else
assert(START_VBN_CURRENT > DIVIDE_ROUND_UP(SIZEOF_FILE_HDR_DFLT, DISK_BLOCK_SIZE));
free_space = ((START_VBN_CURRENT - 1) * DISK_BLOCK_SIZE) - SIZEOF_FILE_HDR_DFLT;
#endif
switch (gv_cur_region->dyn.addr->acc_meth)
{
case dba_bg:
case dba_mm:
mu_cre_vms_structs(gv_cur_region);
fcb = ((vms_gds_info *)(gv_cur_region->dyn.addr->file_cntl->file_info))->fab;
cs_addrs = &((vms_gds_info *)(gv_cur_region->dyn.addr->file_cntl->file_info))->s_addrs;
fcb->fab$b_shr &= FAB$M_NIL; /* No access to this file while it is created */
fcb->fab$l_nam = &nam;
nam = cc$rms_nam;
/* There are (bplmap - 1) non-bitmap blocks per bitmap, so add (bplmap - 2) to number of non-bitmap blocks
* and divide by (bplmap - 1) to get total number of bitmaps for expanded database. (must round up in this
* manner as every non-bitmap block must have an associated bitmap)
*/
fcb->fab$l_alq += DIVIDE_ROUND_UP(fcb->fab$l_alq, BLKS_PER_LMAP - 1); /* Bitmaps */
blk_init_size = fcb->fab$l_alq;
fcb->fab$l_alq *= BLK_SIZE / DISK_BLOCK_SIZE;
fcb->fab$l_alq += START_VBN_CURRENT - 1;
initial_alq = fcb->fab$l_alq;
fcb->fab$w_mrs = 512; /* no longer a relevent field to us */
break;
case dba_usr:
util_out_print("Database file for region !AD not created; access method is not GDS.", TRUE,
REG_LEN_STR(gv_cur_region));
return EXIT_WRN;
default:
gtm_putmsg(VARLSTCNT(1) ERR_BADACCMTHD);
return EXIT_ERR;
}
nam.nam$b_ess = SIZEOF(fn_buff);
nam.nam$l_esa = fn_buff;
nam.nam$b_nop |= NAM$M_SYNCHK;
status = sys$parse(fcb, 0, 0);
if (RMS$_NORMAL != status)
{
gtm_putmsg(VARLSTCNT(8) ERR_DBFILERR, 2, fcb->fab$b_fns, fcb->fab$l_fna, status, 0, fcb->fab$l_stv, 0);
return EXIT_ERR;
}
if (nam.nam$b_node != 0)
{
status = sys$getsyiw(EFN$C_ENF, 0, 0, &item, &iosb, 0, 0);
if (SS$_NORMAL == status)
status = iosb.cond;
if (SS$_NORMAL == status)
{
if (len == nam.nam$b_node-2 && !memcmp(nam.nam$l_esa, node, len))
{
fcb->fab$l_fna = nam.nam$l_esa + nam.nam$b_node;
fcb->fab$b_fns = nam.nam$b_esl - nam.nam$b_node;
}
} else
{
util_out_print("Could not get node for !AD.", TRUE, REG_LEN_STR(gv_cur_region));
exit_stat = EXIT_WRN;
}
}
assert(gv_cur_region->dyn.addr->acc_meth == dba_bg || gv_cur_region->dyn.addr->acc_meth == dba_mm);
nam.nam$l_esa = NULL;
nam.nam$b_esl = 0;
status = sys$create(fcb);
if (status != RMS$_CREATED && status != RMS$_FILEPURGED)
{
switch(status)
{
case RMS$_FLK:
util_out_print("Database file for region !AD not created; currently locked by another user.", TRUE,
REG_LEN_STR(gv_cur_region));
exit_stat = EXIT_INF;
break;
case RMS$_NORMAL:
util_out_print("Database file for region !AD not created; already exists.", TRUE,
REG_LEN_STR(gv_cur_region));
exit_stat = EXIT_INF;
break;
case RMS$_SUPPORT:
util_out_print("Database file for region !AD not created; cannot create across network.", TRUE,
REG_LEN_STR(gv_cur_region));
exit_stat = EXIT_WRN;
break;
case RMS$_FUL:
send_msg(VARLSTCNT(8) ERR_DBFILERR, 2, fcb->fab$b_fns, fcb->fab$l_fna,
status, 0, fcb->fab$l_stv, 0);
/* intentionally falling through */
default:
gtm_putmsg(VARLSTCNT(8) ERR_DBFILERR, 2, fcb->fab$b_fns, fcb->fab$l_fna,
status, 0, fcb->fab$l_stv, 0);
exit_stat = EXIT_ERR;
}
sys$dassgn(fcb->fab$l_stv);
return exit_stat;
}
memcpy(new_id.dvi, nam.nam$t_dvi, SIZEOF(nam.nam$t_dvi));
memcpy(new_id.did, nam.nam$w_did, SIZEOF(nam.nam$w_did));
memcpy(new_id.fid, nam.nam$w_fid, SIZEOF(nam.nam$w_fid));
global_name("GT$S", &new_id, buff); /* 2nd parm is actually a gds_file_id * in global_name */
desc.dsc$w_length = buff[0]; /* By definition, a gds_file_id is dvi,fid,did from nam */
desc.dsc$a_pointer = &buff[1];
cs_addrs->db_addrs[0] = cs_addrs->db_addrs[1] = inadr[0] = inadr[1] = inadr; /* used to determine p0 or p1 allocation */
status = init_sec(cs_addrs->db_addrs, &desc, fcb->fab$l_stv, (START_VBN_CURRENT - 1),
SEC$M_DZRO|SEC$M_GBL|SEC$M_WRT|SEC$M_EXPREG);
if ((SS$_CREATED != status) && (SS$_NORMAL != status))
{
gtm_putmsg(VARLSTCNT(8) ERR_DBFILERR, 2, fcb->fab$b_fns, fcb->fab$l_fna, status, 0, fcb->fab$l_stv, 0);
sys$dassgn(fcb->fab$l_stv);
return EXIT_ERR;
}
cs_data = (sgmnt_data *)cs_addrs->db_addrs[0];
memset(cs_data, 0, SIZEOF_FILE_HDR_DFLT);
cs_data->createinprogress = TRUE;
cs_data->trans_hist.total_blks = (initial_alq - (START_VBN_CURRENT - 1)) / (BLK_SIZE / DISK_BLOCK_SIZE);
/* assert that total_blks stored in file-header = non-bitmap blocks (initial allocation) + bitmap blocks */
assert(cs_data->trans_hist.total_blks == gv_cur_region->dyn.addr->allocation +
DIVIDE_ROUND_UP(gv_cur_region->dyn.addr->allocation, BLKS_PER_LMAP - 1));
cs_data->start_vbn = START_VBN_CURRENT;
temp_acc_meth = gv_cur_region->dyn.addr->acc_meth;
cs_data->acc_meth = gv_cur_region->dyn.addr->acc_meth = dba_bg;
cs_data->extension_size = gv_cur_region->dyn.addr->ext_blk_count;
mucregini(blk_init_size);
cs_addrs->hdr->free_space = free_space;
#ifndef GT_CX_DEF
cs_addrs->hdr->unbacked_cache = TRUE;
#endif
cs_data->acc_meth = gv_cur_region->dyn.addr->acc_meth = temp_acc_meth;
cs_data->createinprogress = FALSE;
if (SS$_NORMAL == (status = disk_block_available(fcb->fab$l_stv, &free_blocks)))
{
blocks_for_extension = (cs_data->blk_size / DISK_BLOCK_SIZE *
(DIVIDE_ROUND_UP(EXTEND_WARNING_FACTOR * (gtm_uint64_t)cs_data->extension_size, BLKS_PER_LMAP - 1)
+ EXTEND_WARNING_FACTOR * (gtm_uint64_t)cs_data->extension_size));
if ((gtm_uint64_t)free_blocks < blocks_for_extension)
{
free_blocks_ll = (gtm_uint64_t)free_blocks;
gtm_putmsg(VARLSTCNT(8) ERR_LOWSPACECRE, 6, fcb->fab$b_fns, fcb->fab$l_fna, EXTEND_WARNING_FACTOR,
&blocks_for_extension, DISK_BLOCK_SIZE, &free_blocks_ll);
send_msg(VARLSTCNT(8) ERR_LOWSPACECRE, 6, fcb->fab$b_fns, fcb->fab$l_fna, EXTEND_WARNING_FACTOR,
&blocks_for_extension, DISK_BLOCK_SIZE, &free_blocks_ll);
}
}
if (SS$_NORMAL == (status = sys$updsec(((vms_gds_info *)(gv_cur_region->dyn.addr->file_cntl->file_info))->s_addrs.db_addrs,
NULL, PSL$C_USER, 0, efn_immed_wait, &iosb, NULL, 0)))
{
status = sys$synch(efn_immed_wait, &iosb);
if (SS$_NORMAL == status)
status = iosb.cond;
} else if (SS$_NOTMODIFIED == status)
status = SS$_NORMAL;
if (SS$_NORMAL == status)
status = del_sec(SEC$M_GBL, &desc, 0);
if (SS$_NORMAL == status)
status = sys$deltva(cs_addrs->db_addrs, retadr, PSL$C_USER);
if (SS$_NORMAL == status)
status = sys$dassgn(fcb->fab$l_stv);
if (SS$_NORMAL == status)
{
util_out_print("Database file for region !AD created.", TRUE, REG_LEN_STR(gv_cur_region));
/* the open and close are an attempt to ensure that the file is available, not under the control of an ACP,
* before MUPIP exits */
fcb->fab$b_shr = FAB$M_SHRPUT | FAB$M_SHRGET | FAB$M_UPI;
fcb->fab$l_fop = 0;
for (lcnt = 1; (60 * MAX_OPEN_RETRY) >= lcnt; lcnt++)
{ /* per VMS engineering a delay is expected. We will wait up to an hour as a
* Delete Global Section operation is essentially and inherently asynchronous in nature
* and could take an arbitrary amount of time.
*/
if (RMS$_FLK != (status = sys$open(fcb, NULL, NULL)))
break;
wcs_sleep(lcnt);
}
assert(RMS$_NORMAL == status);
if (RMS$_NORMAL == status)
{
status = sys$close(fcb);
assert(RMS$_NORMAL == status);
}
if (RMS$_NORMAL != status)
exit_stat = EXIT_WRN;
} else
exit_stat = EXIT_ERR;
if (RMS$_NORMAL != status)
gtm_putmsg(VARLSTCNT(8) ERR_DBFILERR, 2, fcb->fab$b_fns, fcb->fab$l_fna, status, 0, fcb->fab$l_stv, 0);
if ((MAX_RMS_RECORDSIZE - SIZEOF(shmpool_blk_hdr)) < cs_data->blk_size)
gtm_putmsg(VARLSTCNT(5) ERR_MUNOSTRMBKUP, 3, fcb->fab$b_fns, fcb->fab$l_fna, 32 * 1024 - DISK_BLOCK_SIZE);
return exit_stat;
}
OS_PAGE_SIZE_DECLARE
uint4 gdsfilext(uint4 blocks, uint4 filesize, boolean_t trans_in_prog)
{
sm_uc_ptr_t old_base[2], mmap_retaddr;
boolean_t was_crit, is_mm;
char buff[DISK_BLOCK_SIZE];
int result, save_errno, status;
uint4 new_bit_maps, bplmap, map, new_blocks, new_total, max_tot_blks, old_total;
uint4 jnl_status, to_wait, to_msg, wait_period;
gtm_uint64_t avail_blocks, mmap_sz;
off_t new_eof;
trans_num curr_tn;
unix_db_info *udi;
inctn_opcode_t save_inctn_opcode;
int4 prev_extend_blks_to_upgrd;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
cache_rec_ptr_t cr;
DCL_THREADGBL_ACCESS;
assert(!IS_DSE_IMAGE);
assert((cs_addrs->nl == NULL) || (process_id != cs_addrs->nl->trunc_pid)); /* mu_truncate shouldn't extend file... */
assert(!process_exiting);
DEBUG_ONLY(old_base[0] = old_base[1] = NULL);
assert(!gv_cur_region->read_only);
udi = FILE_INFO(gv_cur_region);
is_mm = (dba_mm == cs_addrs->hdr->acc_meth);
# if !defined(MM_FILE_EXT_OK)
if (!udi->grabbed_access_sem && is_mm)
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not allowed ? */
# endif
/* Both blocks and total blocks are unsigned ints so make sure we aren't asking for huge numbers that will
overflow and end up doing silly things.
*/
assert((blocks <= (MAXTOTALBLKS(cs_data) - cs_data->trans_hist.total_blks)) || WBTEST_ENABLED(WBTEST_FILE_EXTEND_ERROR));
if (!blocks)
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not enabled ? */
bplmap = cs_data->bplmap;
/* New total of non-bitmap blocks will be number of current, non-bitmap blocks, plus new blocks desired
* There are (bplmap - 1) non-bitmap blocks per bitmap, so add (bplmap - 2) to number of non-bitmap blocks
* and divide by (bplmap - 1) to get total number of bitmaps for expanded database. (must round up in this
* manner as every non-bitmap block must have an associated bitmap)
* Current number of bitmaps is (total number of current blocks + bplmap - 1) / bplmap.
* Subtract current number of bitmaps from number needed for expanded database to get number of new bitmaps needed.
*/
new_bit_maps = DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap) + blocks, bplmap - 1)
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap);
new_blocks = blocks + new_bit_maps;
assert(0 < (int)new_blocks);
if (new_blocks + cs_data->trans_hist.total_blks > MAXTOTALBLKS(cs_data))
{
assert(FALSE);
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(1) ERR_TOTALBLKMAX);
return (uint4)(NO_FREE_SPACE);
}
if (0 != (save_errno = disk_block_available(udi->fd, &avail_blocks, FALSE)))
{
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
rts_error_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
} else
{
if (!(gtmDebugLevel & GDL_IgnoreAvailSpace))
{ /* Bypass this space check if debug flag above is on. Allows us to create a large sparce DB
* in space it could never fit it if wasn't sparse. Needed for some tests.
*/
avail_blocks = avail_blocks / (cs_data->blk_size / DISK_BLOCK_SIZE);
if ((blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
if (blocks > (uint4)avail_blocks)
{
SETUP_THREADGBL_ACCESS;
if (!INST_FREEZE_ON_NOSPC_ENABLED(cs_addrs))
return (uint4)(NO_FREE_SPACE);
else
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(6) MAKE_MSG_WARNING(ERR_NOSPACEEXT), 4,
DB_LEN_STR(gv_cur_region), new_blocks, (uint4)avail_blocks);
} else
send_msg_csa(CSA_ARG(cs_addrs) VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, DB_LEN_STR(gv_cur_region),
(uint4)(avail_blocks - ((new_blocks <= avail_blocks) ? new_blocks : 0)));
}
}
}
/* From here on, we need to use GDSFILEXT_CLNUP before returning to the caller */
was_crit = cs_addrs->now_crit;
assert(!cs_addrs->hold_onto_crit || was_crit);
/* If we are coming from mupip_extend (which gets crit itself) we better have waited for any unfreezes to occur.
* If we are coming from online rollback (when that feature is available), we will come in holding crit and in
* the final retry. In that case too, we expect to have waited for unfreezes to occur in the caller itself.
* Therefore if we are coming in holding crit from MUPIP, we expect the db to be unfrozen so no need to wait for
* freeze.
* If we are coming from GT.M and final retry (in which case we come in holding crit) we expect to have waited
* for any unfreezes (by invoking tp_crit_all_regions) to occur (TP or non-TP) before coming into this
* function. However, there is one exception. In the final retry, if tp_crit_all_regions notices that
* at least one of the participating regions did ONLY READs, it will not wait for any freeze on THAT region
* to complete before grabbing crit. Later, in the final retry, if THAT region did an update which caused
* op_tcommit to invoke bm_getfree->gdsfilext, then we would have come here with a frozen region on which
* we hold crit.
*/
assert(!was_crit || !cs_data->freeze || (dollar_tlevel && (CDB_STAGNATE <= t_tries)));
/*
* If we are in the final retry and already hold crit, it is possible that csa->nl->wc_blocked is also set to TRUE
* (by a concurrent process in phase2 which encountered an error in the midst of commit and secshr_db_clnup
* finished the job for it). In this case we do NOT want to invoke wcs_recover as that will update the "bt"
* transaction numbers without correspondingly updating the history transaction numbers (effectively causing
* a cdb_sc_blkmod type of restart). Therefore do NOT call grab_crit (which unconditionally invokes wcs_recover)
* if we already hold crit.
*/
if (!was_crit)
{
for ( ; ; )
{
grab_crit(gv_cur_region);
if (!cs_data->freeze && !IS_REPL_INST_FROZEN)
break;
rel_crit(gv_cur_region);
while (cs_data->freeze || IS_REPL_INST_FROZEN)
hiber_start(1000);
}
} else if (cs_data->freeze && dollar_tlevel)
{ /* We don't want to continue with file extension as explained above. Hence return with an error code which
* op_tcommit will recognize (as a cdb_sc_needcrit/cdb_sc_instancefreeze type of restart) and restart accordingly.
*/
assert(CDB_STAGNATE <= t_tries);
GDSFILEXT_CLNUP;
return (uint4)FINAL_RETRY_FREEZE_PROG;
}
if (IS_REPL_INST_FROZEN && trans_in_prog)
{
assert(CDB_STAGNATE <= t_tries);
GDSFILEXT_CLNUP;
return (uint4)FINAL_RETRY_INST_FREEZE;
}
assert(cs_addrs->ti->total_blks == cs_data->trans_hist.total_blks);
old_total = cs_data->trans_hist.total_blks;
if (old_total != filesize)
{ /* Somebody else has already extended it, since we are in crit, this is trust-worthy. However, in case of MM,
* we still need to remap the database
*/
assert((old_total > filesize) GTM_TRUNCATE_ONLY( || !is_mm));
/* For BG, someone else could have truncated or extended - we have no idea */
GDSFILEXT_CLNUP;
return (SS_NORMAL);
}
uint4 jnl_file_extend(jnl_private_control *jpc, uint4 total_jnl_rec_size)
{
file_control *fc;
boolean_t need_extend;
jnl_buffer_ptr_t jb;
jnl_create_info jnl_info;
jnl_file_header *header;
unsigned char hdr_buff[REAL_JNL_HDR_LEN + MAX_IO_BLOCK_SIZE];
uint4 new_alq;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
char prev_jnl_fn[JNL_NAME_SIZE];
uint4 jnl_status = 0, status;
int new_blocks, warn_blocks, result;
gtm_uint64_t avail_blocks;
uint4 aligned_tot_jrec_size, count;
uint4 jnl_fs_block_size, read_write_size;
DCL_THREADGBL_ACCESS;
switch(jpc->region->dyn.addr->acc_meth)
{
case dba_mm:
case dba_bg:
csa = &FILE_INFO(jpc->region)->s_addrs;
break;
default:
GTMASSERT;
}
csd = csa->hdr;
assert(csa == cs_addrs && csd == cs_data);
assert(csa->now_crit || (csd->clustered && (CCST_CLOSED == csa->nl->ccp_state)));
assert(&FILE_INFO(jpc->region)->s_addrs == csa);
assert(csa->jnl_state == csd->jnl_state);
assertpro(JNL_ENABLED(csa) && (NOJNL != jpc->channel) && (!JNL_FILE_SWITCHED(jpc)));
/* crit and messing with the journal file - how could it have vanished? */
if (!csd->jnl_deq || (csd->jnl_alq + csd->jnl_deq > csd->autoswitchlimit))
{
assert(DIVIDE_ROUND_UP(total_jnl_rec_size, DISK_BLOCK_SIZE) <= csd->jnl_alq);
assert(csd->jnl_alq == csd->autoswitchlimit);
new_blocks = csd->jnl_alq;
} else
/* May cause extension of csd->jnl_deq * n blocks where n > 0 */
new_blocks = ROUND_UP(DIVIDE_ROUND_UP(total_jnl_rec_size, DISK_BLOCK_SIZE), csd->jnl_deq);
jpc->status = SS_NORMAL;
jb = jpc->jnl_buff;
assert(0 <= new_blocks);
DEBUG_ONLY(count = 0);
for (need_extend = (0 != new_blocks); need_extend; )
{
DEBUG_ONLY(count++);
/* usually we will do the loop just once where we do the file extension.
* rarely we might need to do an autoswitch instead after which again rarely
* we might need to do an extension on the new journal to fit in the transaction's journal requirements.
* therefore we should do this loop a maximum of twice. hence the assert below.
*/
assert(count <= 2);
need_extend = FALSE;
if (SS_NORMAL == (status = disk_block_available(jpc->channel, &avail_blocks, TRUE)))
{
warn_blocks = (csd->jnl_alq + csd->jnl_deq > csd->autoswitchlimit)
? ((csd->jnl_deq > csd->autoswitchlimit) ? csd->jnl_deq : csd->autoswitchlimit)
: new_blocks;
if ((warn_blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
if (new_blocks > avail_blocks)
{ /* If we cannot satisfy the request, it is an error, unless the anticipatory freeze
* scheme is in effect in which case, we will assume space is available even if
* it is not and go ahead with writes to the disk. If the writes fail with ENOSPC
* we will freeze the instance and wait for space to become available and keep
* retrying the writes. Therefore, we make the NOSPACEEXT a warning in this case.
*/
SETUP_THREADGBL_ACCESS;
if (!ANTICIPATORY_FREEZE_ENABLED(csa))
{
send_msg(VARLSTCNT(6) ERR_NOSPACEEXT, 4,
JNL_LEN_STR(csd), new_blocks, avail_blocks);
new_blocks = 0;
jpc->status = SS_NORMAL;
break;
} else
send_msg(VARLSTCNT(6) MAKE_MSG_WARNING(ERR_NOSPACEEXT), 4,
JNL_LEN_STR(csd), new_blocks, avail_blocks);
} else
send_msg(VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, JNL_LEN_STR(csd), (avail_blocks - warn_blocks));
}
} else
send_msg(VARLSTCNT(5) ERR_JNLFILEXTERR, 2, JNL_LEN_STR(csd), status);
new_alq = jb->filesize + new_blocks;
/* ensure current journal file size is well within autoswitchlimit --> design constraint */
assert(csd->autoswitchlimit >= jb->filesize);
if (csd->autoswitchlimit < (jb->filesize + (EXTEND_WARNING_FACTOR * new_blocks))) /* close to max */
send_msg(VARLSTCNT(5) ERR_JNLSPACELOW, 3, JNL_LEN_STR(csd), csd->autoswitchlimit - jb->filesize);
if (csd->autoswitchlimit < new_alq)
{ /* Reached max, need to autoswitch */
/* Ensure new journal file can hold the entire current transaction's journal record requirements */
assert(csd->autoswitchlimit >= MAX_REQD_JNL_FILE_SIZE(total_jnl_rec_size));
memset(&jnl_info, 0, SIZEOF(jnl_info));
jnl_info.prev_jnl = &prev_jnl_fn[0];
set_jnl_info(gv_cur_region, &jnl_info);
assert(JNL_ENABLED(csa) && (NOJNL != jpc->channel) && !(JNL_FILE_SWITCHED(jpc)));
jnl_status = jnl_ensure_open();
if (0 == jnl_status)
{ /* flush the cache and jnl-buffer-contents to current journal file before
* switching to a new journal. Set a global variable in_jnl_file_autoswitch
* so jnl_write can know not to do the padding check. But because this is a global
* variable, we also need to make sure it is reset in case of errors during the
* autoswitch (or else calls to jnl_write after we are out of the autoswitch logic
* will continue to incorrectly not do the padding check. Hence a condition handler.
*/
assert(!in_jnl_file_autoswitch);
in_jnl_file_autoswitch = TRUE;
/* Also make sure time is not changed. This way if "jnl_write" as part of writing a
* journal record invokes jnl_file_extend, when the autoswitch is done and writing
* of the parent jnl_write resumes, we want it to continue with the same timestamp
* and not have to reset its time (non-trivial task) to reflect any changes since then.
*/
assert(!jgbl.save_dont_reset_gbl_jrec_time);
jgbl.save_dont_reset_gbl_jrec_time = jgbl.dont_reset_gbl_jrec_time;
jgbl.dont_reset_gbl_jrec_time = TRUE;
/* Establish a condition handler so we reset a few global variables that have
* temporarily been modified in case of errors inside wcs_flu/jnl_file_close.
*/
ESTABLISH_RET(jnl_file_autoswitch_ch, EXIT_ERR);
/* It is possible we still have not written a PINI record in this journal file
* (e.g. mupip extend saw the need to do jnl_file_extend inside jnl_write while
* trying to write a PINI record). Write a PINI record in that case before closing
* the journal file that way the EOF record will have a non-zero pini_addr.
*/
if (0 == jpc->pini_addr)
jnl_put_jrt_pini(csa);
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SPEEDUP_NOBEFORE);
jnl_file_close(gv_cur_region, TRUE, TRUE);
REVERT;
in_jnl_file_autoswitch = FALSE;
jgbl.dont_reset_gbl_jrec_time = jgbl.save_dont_reset_gbl_jrec_time;
DEBUG_ONLY(jgbl.save_dont_reset_gbl_jrec_time = FALSE);
assert((dba_mm == cs_data->acc_meth) || (csd == cs_data));
csd = cs_data; /* In MM, wcs_flu() can remap an extended DB, so reset csd to be sure */
} else
{
if (SS_NORMAL != jpc->status)
rts_error(VARLSTCNT(7) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region),
jpc->status);
else
rts_error(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
}
assert(!jgbl.forw_phase_recovery || (NULL != jgbl.mur_pini_addr_reset_fnptr));
assert(jgbl.forw_phase_recovery || (NULL == jgbl.mur_pini_addr_reset_fnptr));
if (NULL != jgbl.mur_pini_addr_reset_fnptr)
(*jgbl.mur_pini_addr_reset_fnptr)(csa);
assert(!jnl_info.no_rename);
assert(!jnl_info.no_prev_link);
if (EXIT_NRM == cre_jnl_file(&jnl_info))
{
assert(0 == memcmp(csd->jnl_file_name, jnl_info.jnl, jnl_info.jnl_len));
assert(csd->jnl_file_name[jnl_info.jnl_len] == '\0');
assert(csd->jnl_file_len == jnl_info.jnl_len);
assert(csd->jnl_buffer_size == jnl_info.buffer);
assert(csd->jnl_alq == jnl_info.alloc);
assert(csd->jnl_deq == jnl_info.extend);
assert(csd->jnl_before_image == jnl_info.before_images);
csd->jnl_checksum = jnl_info.checksum;
csd->jnl_eovtn = csd->trans_hist.curr_tn;
send_msg(VARLSTCNT(4) ERR_NEWJNLFILECREAT, 2, JNL_LEN_STR(csd));
fc = gv_cur_region->dyn.addr->file_cntl;
fc->op = FC_WRITE;
fc->op_buff = (sm_uc_ptr_t)csd;
fc->op_len = SGMNT_HDR_LEN;
fc->op_pos = 1;
status = dbfilop(fc);
if (SS_NORMAL != status)
send_msg(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), status);
assert(JNL_ENABLED(csa));
/* call jnl_ensure_open instead of jnl_file_open to make sure jpc->pini_addr is set to 0 */
jnl_status = jnl_ensure_open(); /* sets jpc->status */
if (0 != jnl_status)
{
if (jpc->status)
rts_error(VARLSTCNT(7) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region),
jpc->status);
else
rts_error(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(gv_cur_region));
}
assert(jb->filesize == csd->jnl_alq);
if (csd->jnl_alq + csd->jnl_deq <= csd->autoswitchlimit)
{
aligned_tot_jrec_size = ALIGNED_ROUND_UP(MAX_REQD_JNL_FILE_SIZE(total_jnl_rec_size),
csd->jnl_alq, csd->jnl_deq);
if (aligned_tot_jrec_size > csd->jnl_alq)
{ /* need to extend more than initial allocation in the new journal file
* to accommodate the current transaction.
*/
new_blocks = aligned_tot_jrec_size - csd->jnl_alq;
assert(new_blocks);
assert(0 == new_blocks % csd->jnl_deq);
need_extend = TRUE;
}
}
} else
{
send_msg(VARLSTCNT(4) ERR_JNLNOCREATE, 2, JNL_LEN_STR(csd));
jpc->status = ERR_JNLNOCREATE;
new_blocks = -1;
}
} else
{
assert(!need_extend); /* ensure we won't go through the for loop again */
/* Virtually extend currently used journal file */
jnl_fs_block_size = jb->fs_block_size;
header = (jnl_file_header *)(ROUND_UP2((uintszofptr_t)hdr_buff, jnl_fs_block_size));
read_write_size = ROUND_UP2(REAL_JNL_HDR_LEN, jnl_fs_block_size);
assert((unsigned char *)header + read_write_size <= ARRAYTOP(hdr_buff));
DO_FILE_READ(jpc->channel, 0, header, read_write_size, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
{
assert(FALSE);
rts_error(VARLSTCNT(5) ERR_JNLRDERR, 2, JNL_LEN_STR(csd), jpc->status);
}
assert((header->virtual_size + new_blocks) == new_alq);
jb->filesize = new_alq; /* Actually this is virtual file size blocks */
header->virtual_size = new_alq;
JNL_DO_FILE_WRITE(csa, csd->jnl_file_name, jpc->channel, 0,
header, read_write_size, jpc->status, jpc->status2);
if (SS_NORMAL != jpc->status)
{
assert(FALSE);
rts_error(VARLSTCNT(5) ERR_JNLWRERR, 2, JNL_LEN_STR(csd), jpc->status);
}
}
if (0 >= new_blocks)
break;
}
if (0 < new_blocks)
{
INCR_GVSTATS_COUNTER(csa, csa->nl, n_jnl_extends, 1);
return EXIT_NRM;
}
jpc->status = ERR_JNLREADEOF;
jnl_file_lost(jpc, ERR_JNLEXTEND);
return EXIT_ERR;
}
uint4 gdsfilext (uint4 blocks, uint4 filesize)
{
sm_uc_ptr_t old_base[2];
boolean_t was_crit, need_to_restore_mask = FALSE;
char *buff;
int mm_prot, result, save_errno, status;
uint4 new_bit_maps, bplmap, map, new_blocks, new_total, max_tot_blks;
uint4 jnl_status, to_wait, to_msg, wait_period;
GTM_BAVAIL_TYPE avail_blocks;
sgmnt_data_ptr_t tmp_csd;
off_t new_eof;
trans_num curr_tn;
unix_db_info *udi;
sigset_t savemask;
inctn_opcode_t save_inctn_opcode;
int4 prev_extend_blks_to_upgrd;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
error_def(ERR_DBFILERR);
error_def(ERR_DBFILEXT);
error_def(ERR_DSKSPACEFLOW);
error_def(ERR_JNLFLUSH);
error_def(ERR_TEXT);
error_def(ERR_TOTALBLKMAX);
error_def(ERR_WAITDSKSPACE);
#ifdef __hppa
if (dba_mm == cs_addrs->hdr->acc_meth)
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not allowed ? */
#endif
/* Both blocks and total blocks are unsigned ints so make sure we aren't asking for huge numbers that will
overflow and end up doing silly things.
*/
assert(blocks <= (MAXTOTALBLKS(cs_data) - cs_data->trans_hist.total_blks));
if (!blocks)
return (uint4)(NO_FREE_SPACE); /* should this be changed to show extension not enabled ? */
bplmap = cs_data->bplmap;
/* new total of non-bitmap blocks will be number of current, non-bitmap blocks, plus new blocks desired
There are (bplmap - 1) non-bitmap blocks per bitmap, so add (bplmap - 2) to number of non-bitmap blocks
and divide by (bplmap - 1) to get total number of bitmaps for expanded database. (must round up in this
manner as every non-bitmap block must have an associated bitmap)
Current number of bitmaps is (total number of current blocks + bplmap - 1) / bplmap.
Subtract current number of bitmaps from number needed for expanded database to get number of new bitmaps needed.
*/
new_bit_maps = DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap) + blocks, bplmap - 1)
- DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, bplmap);
new_blocks = blocks + new_bit_maps;
assert(0 < (int)new_blocks);
udi = FILE_INFO(gv_cur_region);
if (0 != (save_errno = disk_block_available(udi->fd, &avail_blocks, FALSE)))
{
send_msg(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
} else
{
avail_blocks = avail_blocks / (cs_data->blk_size / DISK_BLOCK_SIZE);
if ((blocks * EXTEND_WARNING_FACTOR) > avail_blocks)
{
send_msg(VARLSTCNT(5) ERR_DSKSPACEFLOW, 3, DB_LEN_STR(gv_cur_region),
(uint4)(avail_blocks - ((new_blocks <= avail_blocks) ? new_blocks : 0)));
#ifndef __MVS__
if (blocks > (uint4)avail_blocks)
return (uint4)(NO_FREE_SPACE);
#endif
}
}
cs_addrs->extending = TRUE;
was_crit = cs_addrs->now_crit;
/* If we are coming from mupip_extend (which gets crit itself) we better have waited for any unfreezes to occur */
assert(!was_crit || CDB_STAGNATE == t_tries || FALSE == cs_data->freeze);
for ( ; ; )
{ /* If we are in the final retry and already hold crit, it is possible that csd->wc_blocked is also set to TRUE
* (by a concurrent process in phase2 which encountered an error in the midst of commit and secshr_db_clnup
* finished the job for it). In this case we do NOT want to invoke wcs_recover as that will update the "bt"
* transaction numbers without correspondingly updating the history transaction numbers (effectively causing
* a cdb_sc_blkmod type of restart). Therefore do NOT call grab_crit (which unconditionally invokes wcs_recover)
* if we already hold crit.
*/
if (!was_crit)
grab_crit(gv_cur_region);
if (FALSE == cs_data->freeze)
break;
rel_crit(gv_cur_region);
if (was_crit)
{ /* Two cases.
* (i) Final retry and in TP. We might be holding crit in other regions too.
* We can't do a grab_crit() on this region again unless it is deadlock-safe.
* To be on the safer side, we do a restart. The tp_restart() logic will wait
* for this region's freeze to be removed before grabbing crit.
* (ii) Final retry and not in TP. In that case too, it is better to restart in case there is
* some validation code that shortcuts the checking for the final retry assuming we were
* in crit from t_begin() to t_end(). t_retry() has logic that will wait for unfreeze.
* In either case, we need to restart. Returning EXTEND_UNFREEZECRIT will cause one in t_end/tp_tend.
*/
return (uint4)(EXTEND_UNFREEZECRIT);
}
while (cs_data->freeze)
hiber_start(1000);
}
assert(cs_addrs->ti->total_blks == cs_data->trans_hist.total_blks);
if (cs_data->trans_hist.total_blks != filesize)
{
/* somebody else has already extended it, since we are in crit, this is trust-worthy
* however, in case of MM, we still need to remap the database */
assert(cs_data->trans_hist.total_blks > filesize);
GDSFILEXT_CLNUP;
return (SS_NORMAL);
}
if (run_time && (2 * ((0 < dollar_tlevel) ? sgm_info_ptr->cw_set_depth : cw_set_depth) < cs_addrs->ti->free_blocks))
{
if (FALSE == was_crit)
{
rel_crit(gv_cur_region);
return (uint4)(EXTEND_SUSPECT);
}
/* If free_blocks counter is not ok, then correct it. Do the check again. If still fails, then GTMASSERT. */
if (is_free_blks_ctr_ok() ||
(2 * ((0 < dollar_tlevel) ? sgm_info_ptr->cw_set_depth : cw_set_depth) < cs_addrs->ti->free_blocks))
GTMASSERT; /* held crit through bm_getfree into gdsfilext and still didn't get it right */
}
if (JNL_ENABLED(cs_data))
{
if (!jgbl.dont_reset_gbl_jrec_time)
SET_GBL_JREC_TIME; /* needed before jnl_ensure_open as that can write jnl records */
jpc = cs_addrs->jnl;
jbp = jpc->jnl_buff;
/* Before writing to jnlfile, adjust jgbl.gbl_jrec_time if needed to maintain time order
* of jnl records. This needs to be done BEFORE the jnl_ensure_open as that could write
* journal records (if it decides to switch to a new journal file).
*/
ADJUST_GBL_JREC_TIME(jgbl, jbp);
jnl_status = jnl_ensure_open();
if (jnl_status)
{
GDSFILEXT_CLNUP;
send_msg(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(cs_data), DB_LEN_STR(gv_cur_region));
return (uint4)(NO_FREE_SPACE); /* should have better return status */
}
}
if (dba_mm == cs_addrs->hdr->acc_meth)
{
#if defined(UNTARGETED_MSYNC)
status = msync((caddr_t)cs_addrs->db_addrs[0], (size_t)(cs_addrs->db_addrs[1] - cs_addrs->db_addrs[0]), MS_SYNC);
#else
cs_addrs->nl->mm_extender_pid = process_id;
status = wcs_wtstart(gv_cur_region, 0);
cs_addrs->nl->mm_extender_pid = 0;
if (0 != cs_addrs->acc_meth.mm.mmblk_state->mmblkq_active.fl)
GTMASSERT;
status = 0;
#endif
if (0 == status)
{
/* Block SIGALRM for the duration when cs_data and cs_addrs are out of sync */
sigprocmask(SIG_BLOCK, &blockalrm, &savemask);
need_to_restore_mask = TRUE;
tmp_csd = cs_data;
cs_data = (sgmnt_data_ptr_t)malloc(sizeof(*cs_data));
memcpy((sm_uc_ptr_t)cs_data, (uchar_ptr_t)tmp_csd, sizeof(*cs_data));
status = munmap((caddr_t)cs_addrs->db_addrs[0],
(size_t)(cs_addrs->db_addrs[1] - cs_addrs->db_addrs[0]));
#ifdef DEBUG_DB64
if (-1 != status)
rel_mmseg((caddr_t)cs_addrs->db_addrs[0]);
#endif
} else
tmp_csd = NULL;
if (0 != status)
{
if (tmp_csd)
{
free(cs_data);
cs_data = tmp_csd;
}
GDSFILEXT_CLNUP;
send_msg(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), status);
return (uint4)(NO_FREE_SPACE);
}
cs_addrs->hdr = cs_data;
cs_addrs->ti = &cs_data->trans_hist;
}
if (new_blocks + cs_data->trans_hist.total_blks > MAXTOTALBLKS(cs_data))
{
GDSFILEXT_CLNUP;
send_msg(VARLSTCNT(1) ERR_TOTALBLKMAX);
return (uint4)(NO_FREE_SPACE);
}
CHECK_TN(cs_addrs, cs_data, cs_data->trans_hist.curr_tn); /* can issue rts_error TNTOOLARGE */
assert(0 < (int)new_blocks);
new_total = cs_data->trans_hist.total_blks + new_blocks;
new_eof = ((off_t)(cs_data->start_vbn - 1) * DISK_BLOCK_SIZE) + ((off_t)new_total * cs_data->blk_size);
buff = (char *)malloc(DISK_BLOCK_SIZE);
memset(buff, 0, DISK_BLOCK_SIZE);
LSEEKWRITE(udi->fd, new_eof, buff, DISK_BLOCK_SIZE, save_errno);
if ((ENOSPC == save_errno) && run_time)
{
/* try to write it every second, and send message to operator
* log every 1/20 of cs_data->wait_disk_space
*/
wait_period = to_wait = DIVIDE_ROUND_UP(cs_data->wait_disk_space, CDB_STAGNATE + 1);
to_msg = (to_wait / 8) ? (to_wait / 8) : 1; /* send around 8 messages during 1 wait_period */
while ((to_wait > 0) && (ENOSPC == save_errno))
{
if ((to_wait == cs_data->wait_disk_space) || (to_wait % to_msg == 0))
{
send_msg(VARLSTCNT(11) ERR_WAITDSKSPACE, 4, process_id,
to_wait + (CDB_STAGNATE - t_tries) * wait_period, DB_LEN_STR(gv_cur_region),
ERR_TEXT, 2,
RTS_ERROR_TEXT("Please make more disk space available or shutdown GT.M to avoid data loss"),
save_errno);
gtm_putmsg(VARLSTCNT(11) ERR_WAITDSKSPACE, 4, process_id,
to_wait + (CDB_STAGNATE - t_tries) * wait_period, DB_LEN_STR(gv_cur_region),
ERR_TEXT, 2,
RTS_ERROR_TEXT("Please make more disk space available or shutdown GT.M to avoid data loss"),
save_errno);
}
if (!was_crit)
rel_crit(gv_cur_region);
hiber_start(1000);
to_wait--;
if (!was_crit)
grab_crit(gv_cur_region);
LSEEKWRITE(udi->fd, new_eof, buff, DISK_BLOCK_SIZE, save_errno);
}
}
free(buff);
if (0 != save_errno)
{
GDSFILEXT_CLNUP;
if (ENOSPC == save_errno)
return (uint4)(NO_FREE_SPACE);
send_msg(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(gv_cur_region), save_errno);
return (uint4)(NO_FREE_SPACE);
}
DEBUG_ONLY(prev_extend_blks_to_upgrd = cs_data->blks_to_upgrd;)
unsigned char mu_cre_file(void)
{
char *cc = NULL, path[MAX_FBUFF + 1], errbuff[512];
unsigned char buff[DISK_BLOCK_SIZE];
int fd = -1, i, lower, upper, status, padded_len, padded_vbn, norm_vbn;
uint4 raw_dev_size; /* size of a raw device, in bytes */
int4 blocks_for_create, blocks_for_extension, save_errno;
GTM_BAVAIL_TYPE avail_blocks;
file_control fc;
mstr file;
parse_blk pblk;
unix_db_info udi_struct, *udi;
char *fgets_res;
gd_segment *seg;
error_def(ERR_NOSPACECRE);
error_def(ERR_LOWSPACECRE);
assert((-(sizeof(uint4) * 2) & sizeof(sgmnt_data)) == sizeof(sgmnt_data));
cs_addrs = &udi_struct.s_addrs;
cs_data = (sgmnt_data_ptr_t)NULL; /* for CLEANUP */
memset(&pblk, 0, sizeof(pblk));
pblk.fop = (F_SYNTAXO | F_PARNODE);
pblk.buffer = path;
pblk.buff_size = MAX_FBUFF;
file.addr = (char*)gv_cur_region->dyn.addr->fname;
file.len = gv_cur_region->dyn.addr->fname_len;
strncpy(path,file.addr,file.len);
*(path+file.len) = '\0';
if (is_raw_dev(path))
{ /* do not use a default extension for raw device files */
pblk.def1_buf = DEF_NODBEXT;
pblk.def1_size = sizeof(DEF_NODBEXT) - 1;
} else
{
pblk.def1_buf = DEF_DBEXT;
pblk.def1_size = sizeof(DEF_DBEXT) - 1;
}
if (1 != (parse_file(&file, &pblk) & 1))
{
PRINTF("Error translating filename %s.\n", gv_cur_region->dyn.addr->fname);
return EXIT_ERR;
}
path[pblk.b_esl] = 0;
if (pblk.fnb & F_HAS_NODE)
{ /* Remote node specification given */
assert(pblk.b_node);
PRINTF("Database file for region %s not created; cannot create across network.\n", path);
return EXIT_WRN;
}
udi = &udi_struct;
udi->raw = is_raw_dev(pblk.l_dir);
if (udi->raw)
{
fd = OPEN(pblk.l_dir,O_EXCL | O_RDWR);
if (-1 == fd)
{
SPRINTF_AND_PERROR("Error opening file %s\n");
return EXIT_ERR;
}
if (-1 != (status = (ssize_t)lseek(fd, 0, SEEK_SET)))
{
DOREADRC(fd, buff, sizeof(buff), status);
} else
status = errno;
if (0 != status)
{
SPRINTF_AND_PERROR("Error reading header for file %s\n");
return EXIT_ERR;
}
if (!memcmp(buff, GDS_LABEL, STR_LIT_LEN(GDS_LABEL)))
{
char rsp[80];
PRINTF("Database already exists on device %s\n", path);
PRINTF("Do you wish to re-initialize (all current data will be lost) [y/n] ? ");
FGETS(rsp, 79, stdin, fgets_res);
if ('y' != *rsp)
return EXIT_NRM;
}
PRINTF("Determining size of raw device...\n");
for(i = 1; read(fd, buff, sizeof(buff)) == sizeof(buff);)
{
i *= 2;
lseek(fd, (off_t)i * BUFSIZ, SEEK_SET);
}
lower = i / 2;
upper = i;
while ((lower + upper) / 2 != lower)
{
i = (lower + upper) / 2;
lseek(fd, (off_t)i * BUFSIZ, SEEK_SET);
if (read(fd, buff, sizeof(buff)) == sizeof(buff))
lower = i;
else
upper = i;
}
raw_dev_size = i * BUFSIZ;
} else
{
fd = OPEN3(pblk.l_dir, O_CREAT | O_EXCL | O_RDWR, 0600);
if (-1 == fd)
{
SPRINTF_AND_PERROR("Error opening file %s\n");
return EXIT_ERR;
}
if (0 != (save_errno = disk_block_available(fd, &avail_blocks, FALSE)))
{
errno = save_errno;
SPRINTF_AND_PERROR("Error checking available disk space for %s\n");
CLEANUP(EXIT_ERR);
return EXIT_ERR;
}
seg = gv_cur_region->dyn.addr;
/* blocks_for_create is in the unit of DISK_BLOCK_SIZE */
blocks_for_create = DIVIDE_ROUND_UP(sizeof(sgmnt_data), DISK_BLOCK_SIZE) + 1 +
(seg->blk_size / DISK_BLOCK_SIZE *
((DIVIDE_ROUND_UP(seg->allocation, BLKS_PER_LMAP - 1)) + seg->allocation));
if ((uint4)avail_blocks < blocks_for_create)
{
gtm_putmsg(VARLSTCNT(6) ERR_NOSPACECRE, 4, LEN_AND_STR(path), blocks_for_create, (uint4)avail_blocks);
send_msg(VARLSTCNT(6) ERR_NOSPACECRE, 4, LEN_AND_STR(path), blocks_for_create, (uint4)avail_blocks);
CLEANUP(EXIT_ERR);
return EXIT_ERR;
}
blocks_for_extension = (seg->blk_size / DISK_BLOCK_SIZE *
((DIVIDE_ROUND_UP(EXTEND_WARNING_FACTOR * seg->ext_blk_count, BLKS_PER_LMAP - 1))
+ EXTEND_WARNING_FACTOR * seg->ext_blk_count));
if ((uint4)(avail_blocks - blocks_for_create) < blocks_for_extension)
{
gtm_putmsg(VARLSTCNT(8) ERR_LOWSPACECRE, 6, LEN_AND_STR(path), EXTEND_WARNING_FACTOR, blocks_for_extension,
DISK_BLOCK_SIZE, (uint4)(avail_blocks - blocks_for_create));
send_msg(VARLSTCNT(8) ERR_LOWSPACECRE, 6, LEN_AND_STR(path), EXTEND_WARNING_FACTOR, blocks_for_extension,
DISK_BLOCK_SIZE, (uint4)(avail_blocks - blocks_for_create));
}
}
gv_cur_region->dyn.addr->file_cntl = &fc;
fc.file_info = (void*)&udi_struct;
udi->fd = fd;
cs_data = (sgmnt_data_ptr_t)malloc(sizeof(sgmnt_data));
memset(cs_data, 0, sizeof(*cs_data));
cs_data->createinprogress = TRUE;
cs_data->semid = INVALID_SEMID;
cs_data->shmid = INVALID_SHMID;
/* We want our datablocks to start on what would be a block boundary within the file so pad the fileheader
if necessary to make this happen.
*/
padded_len = ROUND_UP(sizeof(sgmnt_data), BLK_SIZE);
padded_vbn = DIVIDE_ROUND_UP(padded_len, DISK_BLOCK_SIZE) + 1;
norm_vbn = DIVIDE_ROUND_UP(sizeof(sgmnt_data), DISK_BLOCK_SIZE) + 1;
cs_data->start_vbn = padded_vbn;
cs_data->free_space += (padded_vbn - norm_vbn) * DISK_BLOCK_SIZE;
cs_data->acc_meth = gv_cur_region->dyn.addr->acc_meth;
if (udi->raw)
{
/* calculate total blocks, reduce to make room for the
* database header (size rounded up to a block), then
* make into a multiple of BLKS_PER_LMAP to have a complete bitmap
* for each set of blocks.
*/
cs_data->trans_hist.total_blks = raw_dev_size - ROUND_UP(sizeof(sgmnt_data), DISK_BLOCK_SIZE);
cs_data->trans_hist.total_blks /= (uint4)(((gd_segment *)gv_cur_region->dyn.addr)->blk_size);
if (0 == (cs_data->trans_hist.total_blks - DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, BLKS_PER_LMAP - 1)
% (BLKS_PER_LMAP - 1)))
cs_data->trans_hist.total_blks -= 1; /* don't create a bitmap with no data blocks */
cs_data->extension_size = 0;
PRINTF("Raw device size is %dK, %d GDS blocks\n",
raw_dev_size / 1000,
cs_data->trans_hist.total_blks);
} else
{
cs_data->trans_hist.total_blks = gv_cur_region->dyn.addr->allocation;
/* There are (bplmap - 1) non-bitmap blocks per bitmap, so add (bplmap - 2) to number of non-bitmap blocks
* and divide by (bplmap - 1) to get total number of bitmaps for expanded database. (must round up in this
* manner as every non-bitmap block must have an associated bitmap)
*/
cs_data->trans_hist.total_blks += DIVIDE_ROUND_UP(cs_data->trans_hist.total_blks, BLKS_PER_LMAP - 1);
cs_data->extension_size = gv_cur_region->dyn.addr->ext_blk_count;
}
mucregini(cs_data->trans_hist.total_blks);
cs_data->createinprogress = FALSE;
LSEEKWRITE(udi->fd, 0, cs_data, sizeof(sgmnt_data), status);
if (0 != status)
{
SPRINTF_AND_PERROR("Error writing out header for file %s\n");
CLEANUP(EXIT_ERR);
return EXIT_ERR;
}
cc = (char*)malloc(DISK_BLOCK_SIZE);
memset(cc, 0, DISK_BLOCK_SIZE);
LSEEKWRITE(udi->fd,
(cs_data->start_vbn - 1) * DISK_BLOCK_SIZE + ((off_t)(cs_data->trans_hist.total_blks) * cs_data->blk_size),
cc,
DISK_BLOCK_SIZE,
status);
if (0 != status)
{
SPRINTF_AND_PERROR("Error writing out end of file %s\n");
CLEANUP(EXIT_ERR);
return EXIT_ERR;
}
if ((!udi->raw) && (-1 == CHMOD(pblk.l_dir, 0666)))
{
SPRINTF_AND_PERROR("Error changing file mode on file %s\n");
CLEANUP(EXIT_WRN);
return EXIT_WRN;
}
CLEANUP(EXIT_NRM);
PRINTF("Created file %s\n", path);
return EXIT_NRM;
}
