int create_server(void)
{
int child_pid, done_pid, status = 0;
# ifdef _BSD
union wait chld_status;
# define CSTAT chld_status
# else
# define CSTAT status
# endif
int save_errno;
FORK(child_pid);
if (0 == child_pid)
{
process_id = getpid();
/* Do exec using gtmsecshr_path, which was initialize in file check code - send_mesg2gtmsecshr */
if (WBTEST_ENABLED(WBTEST_BADEXEC_SECSHR_PROCESS))
STRCPY(gtmsecshr_path, "");
status = EXECL(gtmsecshr_path, gtmsecshr_path, 0);
if (-1 == status)
{
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(9) ERR_GTMSECSHRSTART, 3, RTS_ERROR_TEXT("Client"), process_id,
ERR_TEXT, 2, RTS_ERROR_STRING(secshrstart_error_code[UNABLETOEXECGTMSECSHR]));
UNDERSCORE_EXIT(UNABLETOEXECGTMSECSHR);
}
} else
{
if (-1 == child_pid)
{
status = GNDCHLDFORKFLD;
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(10) ERR_GTMSECSHRSTART, 3, RTS_ERROR_TEXT("Client"), process_id,
ERR_TEXT, 2, RTS_ERROR_TEXT("Failed to fork off gtmsecshr"), errno);
/* Sleep for a while and hope a subsequent fork will succeed */
hiber_start(1000);
}
for (; !status ;)
{
/* To prevent a warning message that the compiler issues */
done_pid = wait(&CSTAT);
if (done_pid == child_pid)
{
status = WEXITSTATUS(CSTAT);
break;
} else if (-1 == done_pid)
{
if (ECHILD == errno) /* Assume normal exit status */
break;
else if (EINTR != errno)
{
status = GNDCHLDFORKFLD;
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(10) ERR_GTMSECSHRSTART, 3,
RTS_ERROR_TEXT("Client"), process_id,
ERR_TEXT, 2, RTS_ERROR_TEXT("Error spawning gtmsecshr"), errno);
}
}
}
}
return status;
}
void gtmsecshr_sock_cleanup(int caller)
{
int save_errno;
struct sockaddr_un *sock_ptr;
/* Close the secshr client socket */
if (-1 != gtmsecshr_sockfd)
{
close(gtmsecshr_sockfd);
gtmsecshr_sockfd = -1;
}
/* do the unlink */
sock_ptr = (CLIENT == caller) ? >msecshr_cli_sock_name : >msecshr_sock_name;
if (('\0' != sock_ptr->sun_path[0]) && (-1 == UNLINK(sock_ptr->sun_path))
&& (ENOENT != errno))
{
save_errno = errno;
send_msg(VARLSTCNT(12) ERR_GTMSECSHR, 1, process_id, ERR_TEXT, 2, RTS_ERROR_TEXT("unlinking socket"),
ERR_TEXT, 2, RTS_ERROR_STRING(sock_ptr->sun_path), save_errno);
}
sock_ptr->sun_path[0] = '\0'; /* Even if error unlinking since it is useless now */
gtmsecshr_sock_init_done = FALSE;
}
void util_help(void)
{
int rc;
char *help_option;
char help_cmd_string[HELP_CMD_STRING_SIZE];
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(1 >= TREF(parms_cnt));
assert(GTM_IMAGE < image_type && UTIL_HELP_IMAGES > image_type);
if (0 == TREF(parms_cnt))
help_option = utilImageGLDs[INVALID_IMAGE];
else
{
assert(TAREF1(parm_ary, TREF(parms_cnt) - 1));
assert((char *)-1L != (TAREF1(parm_ary, TREF(parms_cnt) - 1)));
help_option = (TAREF1(parm_ary, TREF(parms_cnt) - 1));
}
SNPRINTF(help_cmd_string, SIZEOF(help_cmd_string),
"$gtm_dist/mumps -run %%XCMD 'do ^GTMHELP(\"%s\",\"$gtm_dist/%shelp.gld\")'",
help_option, utilImageGLDs[image_type]);
rc = SYSTEM(help_cmd_string);
if (0 != rc)
rts_error_csa(NULL, VARLSTCNT(5) ERR_TEXT, 2, RTS_ERROR_TEXT("HELP command error"), rc);
}
static void clean_initialize(mutex_struct_ptr_t addr, int n, bool crash)
{
mutex_que_entry_ptr_t q_free_entry;
# if defined(MUTEX_MSEM_WAKE) && !defined(POSIX_MSEM)
msemaphore *status;
# endif
assert(n > 0);
addr->queslots = n;
/* Initialize the waiting process queue to be empty */
addr->prochead.que.fl = addr->prochead.que.bl = 0;
SET_LATCH_GLOBAL(&addr->prochead.latch, LOCK_AVAILABLE);
/* Initialize the free queue to be empty */
addr->freehead.que.fl = addr->freehead.que.bl = 0;
SET_LATCH_GLOBAL(&addr->freehead.latch, LOCK_AVAILABLE);
/* Clear the first free entry */
q_free_entry = (mutex_que_entry_ptr_t)((sm_uc_ptr_t)&addr->freehead + SIZEOF(mutex_que_head));
q_free_entry->que.fl = q_free_entry->que.bl = 0;
q_free_entry->pid = 0;
q_free_entry->super_crit = (void *)NULL;
q_free_entry->mutex_wake_instance = 0;
while (n--)
{
# ifdef MUTEX_MSEM_WAKE
# ifdef POSIX_MSEM
if (-1 == sem_init(&q_free_entry->mutex_wake_msem, TRUE, 0)) /* Shared lock with no initial resources (locked) */
# else
if ((NULL == (status = msem_init(&q_free_entry->mutex_wake_msem, MSEM_LOCKED))) || ((msemaphore *)-1 == status))
# endif
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_MUTEXERR, 0, ERR_TEXT, 2,
RTS_ERROR_TEXT("Error with mutex wait memory semaphore initialization"), errno);
# endif
/* Initialize fl,bl links to 0 before INSQTI as it (gtm_insqti in relqueopi.c) asserts this */
DEBUG_ONLY(((que_ent_ptr_t)q_free_entry)->fl = 0;)
DEBUG_ONLY(((que_ent_ptr_t)q_free_entry)->bl = 0;)
if (INTERLOCK_FAIL == INSQTI((que_ent_ptr_t)q_free_entry++, (que_head_ptr_t)&addr->freehead))
{
# ifdef MUTEX_MSEM_WAKE
# ifdef POSIX_MSEM
if (-1 == sem_init(&q_free_entry->mutex_wake_msem, TRUE, 0)) /* Shared lock with no initial resources (locked) */
# else
if ((NULL == (status = msem_init(&q_free_entry->mutex_wake_msem, MSEM_LOCKED))) || ((msemaphore *)-1 == status))
# endif
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(7) ERR_MUTEXERR, 0, ERR_TEXT, 2,
RTS_ERROR_TEXT("Error with mutex wait memory semaphore initialization"), errno);
# endif
/* Initialize fl,bl links to 0 before INSQTI as it (gtm_insqti in relqueopi.c) asserts this */
DEBUG_ONLY(((que_ent_ptr_t)q_free_entry)->fl = 0;)
DEBUG_ONLY(((que_ent_ptr_t)q_free_entry)->bl = 0;)
if (INTERLOCK_FAIL == INSQTI((que_ent_ptr_t)q_free_entry++, (que_head_ptr_t)&addr->freehead))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_MUTEXERR, 0, ERR_TEXT, 2,
RTS_ERROR_TEXT("Interlock instruction failure in mutex initialize"));
}
SET_LATCH_GLOBAL(&addr->semaphore, LOCK_AVAILABLE);
SET_LATCH_GLOBAL((global_latch_t *)&addr->stuckexec, LOCK_AVAILABLE);
if (!crash)
{
SET_LATCH(&addr->crashcnt, 0);
SET_LATCH_GLOBAL(&addr->crashcnt_latch, LOCK_AVAILABLE);
}
return;
}
static void crash_initialize(mutex_struct_ptr_t addr, int n, bool crash)
{
/*
* mutex_wake_proc() is not declared here because its return value
void mutex_sock_init(void)
{
mstr mutex_sock_dir_lognam, mutex_sock_dir_transnam;
int mutex_sock_path_len;
uint4 mutex_sock_trans_status;
char mutex_sock_path[MAX_TRANS_NAME_LEN];
int mutex_sock_len, save_errno;
struct stat mutex_sock_stat_buf;
int status;
unsigned char pid_str[2 * sizeof(pid_t) + 1];
error_def(ERR_MUTEXERR);
error_def(ERR_TEXT);
error_def(ERR_MUTEXRSRCCLNUP);
if (mutex_sock_fd != -1) /* Initialization done already */
return;
/* Create the socket used for sending and receiving mutex wake mesgs */
if ((mutex_sock_fd = socket(AF_UNIX, SOCK_DGRAM, 0)) == -1)
rts_error(VARLSTCNT(7) ERR_MUTEXERR, 0, ERR_TEXT, 2, RTS_ERROR_TEXT("Error with mutex socket create"), errno);
memset((char *)&mutex_sock_address, 0, sizeof(mutex_sock_address));
/* Get the socket path */
mutex_sock_dir_lognam.len = sizeof(MUTEX_SOCK_DIR) - 1;
mutex_sock_dir_lognam.addr = MUTEX_SOCK_DIR;
mutex_sock_trans_status = trans_log_name(&mutex_sock_dir_lognam,
&mutex_sock_dir_transnam,
mutex_sock_path);
if (mutex_sock_trans_status != SS_NORMAL)
{
strcpy(mutex_sock_path, DEFAULT_MUTEX_SOCK_DIR);
mutex_sock_path_len = sizeof(DEFAULT_MUTEX_SOCK_DIR) - 1;
} else
mutex_sock_path_len = mutex_sock_dir_transnam.len;
/* If the path doesn't already end with a '/' pad a '/' */
if (mutex_sock_path[mutex_sock_path_len - 1] != '/')
{
mutex_sock_path[mutex_sock_path_len++] = '/';
mutex_sock_path[mutex_sock_path_len] = '\0';
}
strcpy(mutex_sock_path + mutex_sock_path_len, MUTEX_SOCK_FILE_PREFIX);
mutex_sock_path_len += (sizeof(MUTEX_SOCK_FILE_PREFIX) - 1);
mutex_wake_this_proc_prefix_len = mutex_sock_path_len;
/* Extend mutex_sock_path with pid */
strcpy(mutex_sock_path + mutex_sock_path_len, (char *)pid2ascx(pid_str, process_id));
mutex_sock_path_len += strlen(pid_str);
if (mutex_sock_path_len > sizeof(mutex_sock_address.sun_path))
rts_error(VARLSTCNT(6) ERR_MUTEXERR, 0, ERR_TEXT, 2, RTS_ERROR_TEXT("Mutex socket path too long"));
mutex_sock_address.sun_family = AF_UNIX;
strcpy(mutex_sock_address.sun_path, mutex_sock_path);
mutex_sock_len = sizeof(mutex_sock_address.sun_family) + mutex_sock_path_len + 1; /* Include NULL byte in length */
if (UNLINK(mutex_sock_address.sun_path) == -1) /* in case it was left from last time */
{
if (errno != ENOENT)
{
if ((status = send_mesg2gtmsecshr(REMOVE_FILE, (unsigned int)-1, mutex_sock_address.sun_path,
mutex_sock_path_len + 1)) == 0)
send_msg(VARLSTCNT(8) ERR_MUTEXRSRCCLNUP, 2, mutex_sock_path_len, mutex_sock_path,
ERR_TEXT, 2, LEN_AND_LIT("Resource removed by gtmsecshr"));
else if (status != ENOENT) /* don't bother if somebody removed the file before gtmsecshr got to it */
rts_error(VARLSTCNT(11) ERR_MUTEXERR, 0, ERR_TEXT, 2,
LEN_AND_LIT("gtmsecshr failed to remove leftover mutex resource"),
ERR_TEXT, 2, mutex_sock_path_len, mutex_sock_path, status);
}
} else /* unlink succeeded */
send_msg(VARLSTCNT(4) ERR_MUTEXRSRCCLNUP, 2, mutex_sock_path_len, mutex_sock_path);
if (BIND(mutex_sock_fd, (struct sockaddr *)&mutex_sock_address, mutex_sock_len) < 0)
rts_error(VARLSTCNT(7) ERR_MUTEXERR, 0, ERR_TEXT, 2,
RTS_ERROR_TEXT("Error with mutex socket bind"),
errno);
/*
* Set the socket permissions to override any umask settings.
* Allow owner and group read and write access.
*/
STAT_FILE(mutex_sock_address.sun_path, &mutex_sock_stat_buf, status);
if (-1 == status)
rts_error(VARLSTCNT(7) ERR_MUTEXERR, 0, ERR_TEXT, 2,
RTS_ERROR_TEXT("Error with mutex socket stat"),
errno);
mutex_sock_stat_buf.st_mode |= (S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
if (-1 == CHMOD(mutex_sock_address.sun_path, mutex_sock_stat_buf.st_mode))
rts_error(VARLSTCNT(7) ERR_MUTEXERR, 0, ERR_TEXT, 2,
RTS_ERROR_TEXT("Error with mutex socket chmod"),
errno);
/* Clear the descriptor set used to sense wake up message */
FD_ZERO(&mutex_wait_on_descs);
/*
* To make mutex_wake_proc faster, pre-initialize portions of
* mutex_wake_this_proc which are invariant of the pid to be woken up.
*/
memset((char *)&mutex_wake_this_proc, 0, sizeof(mutex_wake_this_proc));
mutex_wake_this_proc.sun_family = AF_UNIX;
strcpy(mutex_wake_this_proc.sun_path, mutex_sock_path);
mutex_wake_this_proc_len = mutex_sock_len;
}
void bin_load(uint4 begin, uint4 end)
{
unsigned char *ptr, *cp1, *cp2, *btop, *gvkey_char_ptr, *tmp_ptr, *tmp_key_ptr;
unsigned char hdr_lvl, src_buff[MAX_KEY_SZ + 1], dest_buff[MAX_ZWR_KEY_SZ],
cmpc_str[MAX_KEY_SZ + 1], dup_key_str[MAX_KEY_SZ + 1];
unsigned char *end_buff;
unsigned short len, rec_len, next_cmpc;
int current, last, length, max_blk_siz, max_key, status;
uint4 iter, max_data_len, max_subsc_len, key_count;
ssize_t rec_count, global_key_count, subsc_len,extr_std_null_coll;
boolean_t need_xlation, new_gvn, utf8_extract;
rec_hdr *rp, *next_rp;
mval v, tmp_mval;
mstr mstr_src, mstr_dest;
collseq *extr_collseq, *db_collseq, *save_gv_target_collseq;
coll_hdr extr_collhdr, db_collhdr;
gv_key *tmp_gvkey;
char std_null_coll[BIN_HEADER_NUMSZ + 1];
error_def(ERR_GVIS);
error_def(ERR_TEXT);
error_def(ERR_LDBINFMT);
error_def(ERR_LOADCTRLY);
error_def(ERR_LOADEOF);
error_def(ERR_MUNOFINISH);
error_def(ERR_COLLTYPVERSION);
error_def(ERR_COLLATIONUNDEF);
error_def(ERR_OLDBINEXTRACT);
error_def(ERR_LOADINVCHSET);
tmp_gvkey = (gv_key *)malloc(sizeof(gv_key) + MAX_KEY_SZ - 1);
assert(4 == sizeof(coll_hdr));
gvinit();
v.mvtype = MV_STR;
len = mu_bin_get((char **)&ptr);
hdr_lvl = EXTR_HEADER_LEVEL(ptr);
if (!((hdr_lvl == '4' && len == BIN_HEADER_SZ) || (hdr_lvl < '4' && len == V3_BIN_HEADER_SZ)))
{
rts_error(VARLSTCNT(1) ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
/* assert the assumption that the level can be represented in a single character */
assert(' ' == *(ptr + sizeof(BIN_HEADER_LABEL) - 3));
if (0 != memcmp(ptr, BIN_HEADER_LABEL, sizeof(BIN_HEADER_LABEL) - 2) || hdr_lvl < '2' || *(BIN_HEADER_VERSION) < hdr_lvl)
{ /* ignore the level check */
rts_error(VARLSTCNT(1) ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
/* check if extract was generated in UTF-8 mode */
utf8_extract = (0 == MEMCMP_LIT(&ptr[len - BIN_HEADER_LABELSZ], UTF8_NAME)) ? TRUE : FALSE;
if ((utf8_extract && !gtm_utf8_mode) || (!utf8_extract && gtm_utf8_mode))
{ /* extract CHSET doesn't match $ZCHSET */
if (utf8_extract)
rts_error(VARLSTCNT(4) ERR_LOADINVCHSET, 2, LEN_AND_LIT("UTF-8"));
else
rts_error(VARLSTCNT(4) ERR_LOADINVCHSET, 2, LEN_AND_LIT("M"));
mupip_exit(ERR_LDBINFMT);
}
util_out_print("Label = !AD\n", TRUE, len, ptr);
new_gvn = FALSE;
if (hdr_lvl > '3')
{
memcpy(std_null_coll, ptr + BIN_HEADER_NULLCOLLOFFSET, BIN_HEADER_NUMSZ);
std_null_coll[BIN_HEADER_NUMSZ] = '\0';
extr_std_null_coll = STRTOUL(std_null_coll, NULL, 10);
if (0 != extr_std_null_coll && 1!= extr_std_null_coll)
{
rts_error(VARLSTCNT(5) ERR_TEXT, 2, RTS_ERROR_TEXT("Corrupted null collation field in header"),
ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
} else
extr_std_null_coll = 0;
if (hdr_lvl > '2')
{
len = mu_bin_get((char **)&ptr);
if (sizeof(coll_hdr) != len)
{
rts_error(VARLSTCNT(5) ERR_TEXT, 2, RTS_ERROR_TEXT("Corrupt collation header"), ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
extr_collhdr = *((coll_hdr *)(ptr));
new_gvn = TRUE;
} else
gtm_putmsg(VARLSTCNT(3) ERR_OLDBINEXTRACT, 1, hdr_lvl - '0');
if (begin < 2)
begin = 2;
for (iter = 2; iter < begin; iter++)
{
if (!(len = mu_bin_get((char **)&ptr)))
{
gtm_putmsg(VARLSTCNT(3) ERR_LOADEOF, 1, begin);
util_out_print("Error reading record number: !UL\n", TRUE, iter);
mupip_error_occurred = TRUE;
return;
} else if (len == sizeof(coll_hdr))
{
extr_collhdr = *((coll_hdr *)(ptr));
assert(hdr_lvl > '2');
iter--;
}
}
assert(iter == begin);
util_out_print("Beginning LOAD at record number: !UL\n", TRUE, begin);
max_data_len = 0;
max_subsc_len = 0;
key_count = 0;
rec_count = begin - 1;
extr_collseq = db_collseq = NULL;
need_xlation = FALSE;
for (; !mupip_DB_full ;)
{
if (++rec_count > end)
break;
next_cmpc = 0;
mupip_error_occurred = FALSE;
if (mu_ctrly_occurred)
break;
if (mu_ctrlc_occurred)
{
util_out_print("!AD:!_ Key cnt: !UL max subsc len: !UL max data len: !UL", TRUE,
LEN_AND_LIT(gt_lit), key_count, max_subsc_len, max_data_len);
util_out_print("Last LOAD record number: !UL", TRUE, key_count ? (rec_count - 1) : 0);
mu_gvis();
util_out_print(0, TRUE);
mu_ctrlc_occurred = FALSE;
}
/* reset the stringpool for every record in order to avoid garbage collection */
stringpool.free = stringpool.base;
if (!(len = mu_bin_get((char **)&ptr)) || mupip_error_occurred)
break;
else if (len == sizeof(coll_hdr))
{
extr_collhdr = *((coll_hdr *)(ptr));
assert(hdr_lvl > '2');
new_gvn = TRUE; /* next record will contain a new gvn */
rec_count--; /* Decrement as this record does not count as a record for loading purposes */
continue;
}
global_key_count = 1;
rp = (rec_hdr*)ptr;
btop = ptr + len;
cp1 = (unsigned char*)(rp + 1);
v.str.addr = (char*)cp1;
while (*cp1++) ;
v.str.len =INTCAST((char*)cp1 - v.str.addr - 1);
if (hdr_lvl <= '2' || new_gvn)
{
bin_call_db(BIN_BIND, (INTPTR_T)gd_header, (INTPTR_T)&v.str);
max_key = gv_cur_region->max_key_size;
db_collhdr.act = gv_target->act;
db_collhdr.ver = gv_target->ver;
db_collhdr.nct = gv_target->nct;
}
GET_SHORT(rec_len, &rp->rsiz);
if (rp->cmpc != 0 || v.str.len > rec_len || mupip_error_occurred)
{
bin_call_db(ERR_COR, rec_count, global_key_count);
mu_gvis();
util_out_print(0, TRUE);
continue;
}
if (new_gvn)
{
if ((db_collhdr.act != extr_collhdr.act || db_collhdr.ver != extr_collhdr.ver
|| db_collhdr.nct != extr_collhdr.nct
|| gv_cur_region->std_null_coll != extr_std_null_coll))
{
if (extr_collhdr.act)
{
if (extr_collseq = ready_collseq((int)extr_collhdr.act))
{
if (!do_verify(extr_collseq, extr_collhdr.act, extr_collhdr.ver))
{
gtm_putmsg(VARLSTCNT(8) ERR_COLLTYPVERSION, 2, extr_collhdr.act,
extr_collhdr.ver, ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLTYPVERSION);
}
} else
{
gtm_putmsg(VARLSTCNT(7) ERR_COLLATIONUNDEF, 1, extr_collhdr.act,
ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLATIONUNDEF);
}
}
if (db_collhdr.act)
{
if (db_collseq = ready_collseq((int)db_collhdr.act))
{
if (!do_verify(db_collseq, db_collhdr.act, db_collhdr.ver))
{
gtm_putmsg(VARLSTCNT(8) ERR_COLLTYPVERSION, 2, db_collhdr.act,
db_collhdr.ver, ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLTYPVERSION);
}
} else
{
gtm_putmsg(VARLSTCNT(7) ERR_COLLATIONUNDEF, 1, db_collhdr.act,
ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLATIONUNDEF);
}
}
need_xlation = TRUE;
} else
need_xlation = FALSE;
}
new_gvn = FALSE;
for (; rp < (rec_hdr*)btop; rp = (rec_hdr*)((unsigned char *)rp + rec_len))
{
GET_SHORT(rec_len, &rp->rsiz);
if (rec_len + (unsigned char *)rp > btop)
{
bin_call_db(ERR_COR, rec_count, global_key_count);
mu_gvis();
util_out_print(0, TRUE);
break;
}
cp1 = (unsigned char*)(rp + 1);
cp2 = gv_currkey->base + rp->cmpc;
current = 1;
for (;;)
{
last = current;
current = *cp2++ = *cp1++;
if (0 == last && 0 == current)
break;
if (cp1 > (unsigned char *) rp + rec_len ||
cp2 > (unsigned char *) gv_currkey + gv_currkey->top)
{
bin_call_db(ERR_COR, rec_count, global_key_count);
mu_gvis();
util_out_print(0, TRUE);
break;
}
}
if (mupip_error_occurred)
break;
gv_currkey->end = cp2 - gv_currkey->base - 1;
if (need_xlation)
{
assert(hdr_lvl >= '3');
assert(extr_collhdr.act || db_collhdr.act || extr_collhdr.nct || db_collhdr.nct ||
extr_std_null_coll != gv_cur_region->std_null_coll);
/* gv_currkey would have been modified/translated in the earlier put */
memcpy(gv_currkey->base, cmpc_str, next_cmpc);
next_rp = (rec_hdr *)((unsigned char*)rp + rec_len);
if ((unsigned char*)next_rp < btop)
{
next_cmpc = next_rp->cmpc;
assert(next_cmpc <= gv_currkey->end);
memcpy(cmpc_str, gv_currkey->base, next_cmpc);
} else
next_cmpc = 0;
/* length of the key might change (due to nct variation),
* so get a copy of the original key from the extract */
memcpy(dup_key_str, gv_currkey->base, gv_currkey->end + 1);
gvkey_char_ptr = dup_key_str;
while (*gvkey_char_ptr++) ;
gv_currkey->prev = 0;
gv_currkey->end = gvkey_char_ptr - dup_key_str;
tmp_gvkey->top = gv_keysize;
while (*gvkey_char_ptr)
{
/* get next subscript (in GT.M internal subsc format) */
subsc_len = 0;
tmp_ptr = src_buff;
while (*gvkey_char_ptr)
*tmp_ptr++ = *gvkey_char_ptr++;
subsc_len = tmp_ptr - src_buff;
src_buff[subsc_len] = '\0';
if (extr_collseq)
{
/* undo the extract time collation */
transform = TRUE;
save_gv_target_collseq = gv_target->collseq;
gv_target->collseq = extr_collseq;
} else
transform = FALSE;
/* convert the subscript to string format */
end_buff = gvsub2str(src_buff, dest_buff, FALSE);
/* transform the string to the current subsc format */
transform = TRUE;
tmp_mval.mvtype = MV_STR;
tmp_mval.str.addr = (char *)dest_buff;
tmp_mval.str.len = INTCAST(end_buff - dest_buff);
tmp_gvkey->prev = 0;
tmp_gvkey->end = 0;
if (extr_collseq)
gv_target->collseq = save_gv_target_collseq;
mval2subsc(&tmp_mval, tmp_gvkey);
/* we now have the correctly transformed subscript */
tmp_key_ptr = gv_currkey->base + gv_currkey->end;
memcpy(tmp_key_ptr, tmp_gvkey->base, tmp_gvkey->end + 1);
gv_currkey->prev = gv_currkey->end;
gv_currkey->end += tmp_gvkey->end;
gvkey_char_ptr++;
}
if ( gv_cur_region->std_null_coll != extr_std_null_coll && gv_currkey->prev)
{
if (extr_std_null_coll == 0)
{
GTM2STDNULLCOLL(gv_currkey->base, gv_currkey->end);
} else
{
STD2GTMNULLCOLL(gv_currkey->base, gv_currkey->end);
}
}
}
if (gv_currkey->end >= max_key)
{
bin_call_db(ERR_COR, rec_count, global_key_count);
mu_gvis();
util_out_print(0, TRUE);
continue;
}
if (max_subsc_len < (gv_currkey->end + 1))
max_subsc_len = gv_currkey->end + 1;
v.str.addr = (char*)cp1;
v.str.len =INTCAST(rec_len - (cp1 - (unsigned char *)rp) );
if (max_data_len < v.str.len)
max_data_len = v.str.len;
bin_call_db(BIN_PUT, (INTPTR_T)&v, 0);
if (mupip_error_occurred)
{
if (!mupip_DB_full)
{
bin_call_db(ERR_COR, rec_count, global_key_count);
util_out_print(0, TRUE);
}
break;
}
key_count++;
global_key_count++;
}
}
free(tmp_gvkey);
mu_load_close();
util_out_print("LOAD TOTAL!_!_Key Cnt: !UL Max Subsc Len: !UL Max Data Len: !UL", TRUE, key_count, max_subsc_len,
max_data_len);
util_out_print("Last LOAD record number: !UL\n", TRUE, key_count ? (rec_count - 1) : 0);
if (mu_ctrly_occurred)
{
gtm_putmsg(VARLSTCNT(1) ERR_LOADCTRLY);
mupip_exit(ERR_MUNOFINISH);
}
}
void op_zrupdate(int argcnt, ...)
{
mval *objfilespec;
va_list var;
mval objpath;
char tranbuf[MAX_FBUFF + 1], *chptr, chr;
open_relinkctl_sgm *linkctl;
relinkrec_t *rec;
plength plen;
int status, fextlen, fnamlen, fcnt;
parse_blk pblk;
struct stat outbuf;
int stat_res;
boolean_t seenfext, fileexists;
mstr objdir, rtnname;
uint4 hash, prev_hash_index;
/* Currently only expecting one value per invocation right now. That will change in phase 2 hence the stdarg setup */
va_start(var, argcnt);
assert(1 == argcnt);
objfilespec = va_arg(var, mval *);
va_end(var);
MV_FORCE_STR(objfilespec);
/* First some pre-processing to determine if an explicit file name or type was specified. If so, it must be ".o" for
* this phase of implementation. Later phases may allow ".m" to be specified but not initially. Use parse_file() to
* parse everything out and isolate any extention.
*/
memset(&pblk, 0, SIZEOF(pblk));
pblk.buffer = tranbuf;
pblk.buff_size = (unsigned char)(MAX_FBUFF); /* Pass size of buffer - 1 (standard protocol for parse_file) */
pblk.def1_buf = DOTOBJEXT; /* Default .o file type if not specified */
pblk.def1_size = SIZEOF(DOTOBJEXT) - 1;
pblk.fop = F_SYNTAXO; /* Syntax check only - bypass directory existence check */
status = parse_file(&objfilespec->str, &pblk);
if (ERR_PARNORMAL != status)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(5) ERR_FILEPARSE, 2, objfilespec->str.len, objfilespec->str.addr, status);
tranbuf[pblk.b_esl] = '\0'; /* Needed for subsequent STAT_FILE */
seenfext = FALSE;
if (0 != pblk.b_ext)
{ /* If a file extension was specified - get the extension sans any potential wildcard character */
for (chptr = pblk.l_ext + 1, fextlen = pblk.b_ext - 1; 0 < fextlen; chptr++, fextlen--)
{ /* Check each character in the extension except first which is the dot if ext exists at all */
if (WILDCARD != *chptr)
{ /* We see a char that isn't a wildcard character. If we've already seen our "o" file extension,
* this char makes our requirement filetype impossible so raise an error.
*/
if (seenfext || (OBJEXT != *chptr))
/* No return from this error */
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8) ERR_FILEPARSE,
2, objfilespec->str.len, objfilespec->str.addr,
ERR_TEXT, 2, RTS_ERROR_TEXT("Unsupported filetype specified"));
seenfext = TRUE;
}
}
}
/* Do a simlar check for the file type */
if (0 != pblk.b_name)
{ /* A file name was specified (if not, tiz probably hard to find the file but that can be dealt with later).
* Like in the above, the name must be comprised of valid chars for routine names.
*/
for (chptr = pblk.l_name, fnamlen = pblk.b_name; 0 < fnamlen; chptr++, fnamlen--)
{
if (WILDCARD != *chptr)
{ /* Substitute '%' for '_'. While this substitution is really only valid on the first char, only the
* first char check allows "%" so a 2nd or later char check would fail the '%' substitution anyway.
*/
chr = ('_' == *chptr) ? '%' : *chptr;
/* We see a char that isn't a wildcard character. If this is the first character, it can be
* alpha or percent. If the second or later character, it can be alphanumeric.
*/
if (((fnamlen != pblk.b_name) && !VALID_MNAME_NFCHAR(chr)) /* 2nd char or later check */
|| ((fnamlen == pblk.b_name) && !VALID_MNAME_FCHAR(chr))) /* 1st char check */
{
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8) ERR_FILEPARSE,
2, objfilespec->str.len, objfilespec->str.addr,
ERR_TEXT, 2, RTS_ERROR_TEXT("Filename is not a valid routine name"));
}
}
}
}
/* When specifying a non-wildcarded object file, it is possible for the file to have been removed, in which case we still
* need to update its relinkctl entry (if exists) to notify other processes about the object's deletion.
*/
if (!(pblk.fnb & F_WILD) & seenfext)
{ /* If no wildcards in path and saw the extension we want - no need to wash through zsearch() */
objdir.addr = pblk.l_dir;
objdir.len = pblk.b_dir;
linkctl = relinkctl_attach(&objdir); /* Create/attach/open relinkctl file */
if (NULL == linkctl) /* Non-existant path and no associated relinkctl file */
/* Note this reference to errno depends on nothing in relinkctl_attach() doing anything to modify
* errno after the realpath() call. No return from this error.
*/
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(5) ERR_FILEPARSE, 2, objfilespec->str.len, objfilespec->str.addr,
errno);
/* What we do at this point depends on the following conditions:
*
* 1) If the specified file exists, we can add it to relinkctl file and/or update its cycle.
* 2) If the file doesn't exist on disk but the routine is found in the relinkctl file, update cycle.
* 3) If no file and no entry, just ignore it and do nothing (info error removed by request).
*/
STAT_FILE(tranbuf, &outbuf, stat_res);
if (-1 == stat_res)
{
if (ENOENT != errno)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(5) ERR_FILEPARSE, 2, objfilespec->str.len,
objfilespec->str.addr, errno);
fileexists = FALSE;
} else
fileexists = TRUE;
rtnname.len = pblk.b_name;
rtnname.addr = pblk.l_name;
CONVERT_FILENAME_TO_RTNNAME(rtnname); /* Set rtnname right before searching in relinkctl file */
assert(valid_mname(&rtnname));
COMPUTE_RELINKCTL_HASH(&rtnname, hash);
rec = relinkctl_find_record(linkctl, &rtnname, hash, &prev_hash_index);
if ((NULL == rec) && !fileexists)
return; /* No file and no entry - ignore */
/* Either the file exists or the entry exists so add or update it */
rec = relinkctl_insert_record(linkctl, &rtnname);
RELINKCTL_CYCLE_INCR(rec, linkctl); /* Increment cycle indicating change to world */
return;
}
/* If we have a wildcarded request or one without the object filetype, reprocess the string with $ZSEARCH using our
* defined stream to resolve wildcards. Then loop through processing each file returned. In this loop, we just ignore
* any file that doesn't have a ".o" extension.
*/
op_fnzsearch((mval *)&literal_null, STRM_ZRUPDATE, 0, &objpath); /* Clear any existing cache */
for (fcnt = 0; ; fcnt++)
{
plen.p.pint = op_fnzsearch(objfilespec, STRM_ZRUPDATE, 0, &objpath);
if (0 == objpath.str.len)
{ /* End of file list */
if (0 == fcnt)
/* Still looking to process our first file - give no objects found message as a "soft" message
* (INFO level message - supressed in other than direct mode)
*/
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8) MAKE_MSG_INFO(ERR_FILEPARSE), 2, objfilespec->str.len,
objfilespec->str.addr, ERR_TEXT, 2, RTS_ERROR_TEXT("No object files found"));
break;
}
/* The extension contains the extension-start character ('.') so we are looking for the extension '.o' hence
* the length must be 2 and the 2nd char must be OBJEXT.
*/
if ((2 == plen.p.pblk.b_ext) && (OBJEXT == *(objpath.str.addr + plen.p.pblk.b_dir + plen.p.pblk.b_name + 1)))
{ /* This is (probably) an object file. Double check file is a file and not a directory */
memcpy(tranbuf, objpath.str.addr, objpath.str.len); /* Need null terminated version for STAT */
tranbuf[objpath.str.len] = '\0'; /* Not guaranteed null termination from op_fnzsearch */
STAT_FILE(tranbuf, &outbuf, stat_res);
/* If either something happened to the file since op_fnzsearch() saw it or the file is not a file, then
* ignore it.
*/
if ((-1 == stat_res) || !S_ISREG(outbuf.st_mode))
{
fcnt--; /* Don't count files not found for whatever reason */
continue;
}
/* Before opening the relinkctl file, verify we actually do have a valid routine name */
rtnname.len = plen.p.pblk.b_name;
rtnname.addr = objpath.str.addr + plen.p.pblk.b_dir;
CONVERT_FILENAME_TO_RTNNAME(rtnname); /* Set rtnname right before searching in relinkctl file */
if (!valid_mname(&rtnname))
{
fcnt--;
continue; /* Ignore files that are invalid wildcard matches */
}
objdir.addr = objpath.str.addr;
objdir.len = plen.p.pblk.b_dir;
linkctl = relinkctl_attach(&objdir); /* Create/attach/open relinkctl file */
if (NULL == linkctl)
{
fcnt--; /* Path disappeared - don't count it */
continue;
}
rec = relinkctl_insert_record(linkctl, &rtnname);
RELINKCTL_CYCLE_INCR(rec, linkctl); /* Increment cycle indicating change to world */
} else
fcnt--; /* Don't count ignored files */
}
}
void op_svput(int varnum, mval *v)
{
int i, ok, state;
char *vptr;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
switch (varnum)
{
case SV_X:
MV_FORCE_NUM(v);
io_curr_device.out->dollar.x = (short)MV_FORCE_INT(v);
if ((short)(io_curr_device.out->dollar.x) < 0)
io_curr_device.out->dollar.x = 0;
break;
case SV_Y:
MV_FORCE_NUM(v);
io_curr_device.out->dollar.y = (short)MV_FORCE_INT(v);
if ((short)(io_curr_device.out->dollar.y) < 0)
io_curr_device.out->dollar.y = 0;
break;
case SV_ZCOMPILE:
MV_FORCE_STR(v);
if ((TREF(dollar_zcompile)).addr)
free ((TREF(dollar_zcompile)).addr);
(TREF(dollar_zcompile)).addr = (char *)malloc(v->str.len);
memcpy((TREF(dollar_zcompile)).addr, v->str.addr, v->str.len);
(TREF(dollar_zcompile)).len = v->str.len;
break;
case SV_ZSTEP:
MV_FORCE_STR(v);
op_commarg(v,indir_linetail);
op_unwind();
dollar_zstep = *v;
break;
case SV_ZGBLDIR:
MV_FORCE_STR(v);
if ((dollar_zgbldir.str.len != v->str.len)
|| memcmp(dollar_zgbldir.str.addr, v->str.addr, dollar_zgbldir.str.len))
{
if (0 == v->str.len)
{
/* set $zgbldir="" */
dpzgbini();
gd_header = NULL;
} else
{
gd_header = zgbldir(v);
/* update the gd_map */
SET_GD_MAP;
dollar_zgbldir.str.len = v->str.len;
dollar_zgbldir.str.addr = v->str.addr;
s2pool(&dollar_zgbldir.str);
}
if (NULL != gv_currkey)
{
gv_currkey->base[0] = '\0';
gv_currkey->prev = gv_currkey->end = 0;
} else if (NULL != gd_header)
gvinit();
if (NULL != gv_target)
gv_target->clue.end = 0;
}
break;
case SV_ZMAXTPTIME:
dollar_zmaxtptime = mval2i(v);
break;
case SV_ZROUTINES:
MV_FORCE_STR(v);
/* The string(v) should be parsed and loaded before setting $zroutines
* to retain the old value in case errors occur while loading */
zro_load(&v->str);
if ((TREF(dollar_zroutines)).addr)
free ((TREF(dollar_zroutines)).addr);
(TREF(dollar_zroutines)).addr = (char *)malloc(v->str.len);
memcpy((TREF(dollar_zroutines)).addr, v->str.addr, v->str.len);
(TREF(dollar_zroutines)).len = v->str.len;
break;
case SV_ZSOURCE:
MV_FORCE_STR(v);
dollar_zsource.mvtype = MV_STR;
dollar_zsource.str = v->str;
break;
case SV_ZTRAP:
# ifdef GTM_TRIGGER
if (0 < gtm_trigger_depth)
rts_error(VARLSTCNT(1) ERR_NOZTRAPINTRIG);
# endif
MV_FORCE_STR(v);
if (ztrap_new)
op_newintrinsic(SV_ZTRAP);
dollar_ztrap.mvtype = MV_STR;
dollar_ztrap.str = v->str;
/* Setting either $ZTRAP or $ETRAP to empty causes any current error trapping to be canceled */
if (!v->str.len)
{
dollar_etrap.mvtype = MV_STR;
dollar_etrap.str = v->str;
ztrap_explicit_null = TRUE;
} else /* Ensure that $ETRAP and $ZTRAP are not both active at the same time */
{
ztrap_explicit_null = FALSE;
if (dollar_etrap.str.len > 0)
gtm_newintrinsic(&dollar_etrap);
}
if (ztrap_form & ZTRAP_POP)
ztrap_save_ctxt();
if (tp_timeout_deferred && !dollar_zininterrupt)
/* A tp timeout was deferred. Now that $ETRAP is no longer in effect and no job interrupt is in
* effect, the timeout need no longer be deferred and can be recognized.
*/
tptimeout_set(0);
break;
case SV_ZSTATUS:
MV_FORCE_STR(v);
dollar_zstatus.mvtype = MV_STR;
dollar_zstatus.str = v->str;
break;
case SV_PROMPT:
MV_FORCE_STR(v);
MV_FORCE_LEN_STRICT(v); /* Ensure that direct mode prompt will not have BADCHARs,
* otherwise the BADCHAR error may fill up the filesystem
*/
if (v->str.len <= SIZEOF_prombuf)
(TREF(gtmprompt)).len = v->str.len;
else if (!gtm_utf8_mode)
(TREF(gtmprompt)).len = SIZEOF_prombuf;
# ifdef UNICODE_SUPPORTED
else
{
UTF8_LEADING_BYTE(v->str.addr + SIZEOF_prombuf, v->str.addr, vptr);
(TREF(gtmprompt)).len = INTCAST(vptr - v->str.addr);
}
# endif
memcpy((TREF(gtmprompt)).addr, v->str.addr, (TREF(gtmprompt)).len);
break;
case SV_ECODE:
MV_FORCE_STR(v);
if (v->str.len)
{
/* Format must be like ,Mnnn,Mnnn,Zxxx,Uxxx,
* Mnnn are ANSI standard error codes
* Zxxx are implementation-specific codes
* Uxxx are end-user defined codes
* Note that there must be commas at the start and at the end
*/
for (state = 2, i = 0; (i < v->str.len) && (state <= 2); i++)
{
switch(state)
{
case 2: state = (v->str.addr[i] == ',') ? 1 : 101;
break;
case 1: state = ((v->str.addr[i] == 'M') ||
(v->str.addr[i] == 'U') ||
(v->str.addr[i] == 'Z')) ? 0 : 101;
break;
case 0: state = (v->str.addr[i] == ',') ? 1 : 0;
break;
}
}
/* The above check would pass strings like ","
* so double-check that there are at least three characters
* (starting comma, ending comma, and something in between)
*/
if ((state != 1) || (v->str.len < 3))
{
/* error, ecode = M101 */
rts_error(VARLSTCNT(4) ERR_INVECODEVAL, 2, v->str.len, v->str.addr);
}
}
if (v->str.len > 0)
{
ecode_add(&v->str);
rts_error(VARLSTCNT(2) ERR_SETECODE, 0);
} else
{
NULLIFY_DOLLAR_ECODE; /* reset $ECODE related variables to correspond to $ECODE = NULL state */
NULLIFY_ERROR_FRAME; /* we are no more in error-handling mode */
if (tp_timeout_deferred && !dollar_zininterrupt)
/* A tp timeout was deferred. Now that we are clear of error handling and no job interrupt
* is in process, allow the timeout to be recognized.
*/
tptimeout_set(0);
}
break;
case SV_ETRAP:
MV_FORCE_STR(v);
dollar_etrap.mvtype = MV_STR;
dollar_etrap.str = v->str;
/* Setting either $ZTRAP or $ETRAP to empty causes any current error trapping to be canceled */
if (!v->str.len)
{
dollar_ztrap.mvtype = MV_STR;
dollar_ztrap.str = v->str;
} else if (dollar_ztrap.str.len > 0)
{ /* Ensure that $ETRAP and $ZTRAP are not both active at the same time */
assert(FALSE == ztrap_explicit_null);
gtm_newintrinsic(&dollar_ztrap);
}
ztrap_explicit_null = FALSE;
break;
case SV_ZERROR:
MV_FORCE_STR(v);
dollar_zerror.mvtype = MV_STR;
dollar_zerror.str = v->str;
break;
case SV_ZYERROR:
MV_FORCE_STR(v);
dollar_zyerror.mvtype = MV_STR;
dollar_zyerror.str = v->str;
break;
case SV_SYSTEM:
assert(FALSE);
rts_error(VARLSTCNT(4) ERR_SYSTEMVALUE, 2, v->str.len, v->str.addr);
break;
case SV_ZDIR:
setzdir(v, NULL); /* change directory to v */
getzdir(); /* update dollar_zdir with current working directory */
break;
case SV_ZINTERRUPT:
MV_FORCE_STR(v);
dollar_zinterrupt.mvtype = MV_STR;
dollar_zinterrupt.str = v->str;
break;
case SV_ZDATE_FORM:
MV_FORCE_NUM(v);
TREF(zdate_form) = (short)MV_FORCE_INT(v);
break;
case SV_ZTEXIT:
MV_FORCE_STR(v);
dollar_ztexit.mvtype = MV_STR;
dollar_ztexit.str = v->str;
/* Coercing $ZTEXIT to boolean at SET command is more efficient than coercing before each
* rethrow at TR/TRO. Since we want to maintain dollar_ztexit as a string, coercion should
* not be performed on dollar_ztext, but on a temporary (i.e. parameter v)
*/
dollar_ztexit_bool = MV_FORCE_BOOL(v);
break;
case SV_ZQUIT:
dollar_zquit_anyway = MV_FORCE_BOOL(v);
break;
case SV_ZTVALUE:
# ifdef GTM_TRIGGER
assert(!dollar_tlevel || (tstart_trigger_depth <= gtm_trigger_depth));
if (!dollar_tlevel || (tstart_trigger_depth == gtm_trigger_depth))
rts_error(VARLSTCNT(4) ERR_SETINTRIGONLY, 2, RTS_ERROR_TEXT("$ZTVALUE"));
if (dollar_ztriggerop != &gvtr_cmd_mval[GVTR_CMDTYPE_SET])
rts_error(VARLSTCNT(4) ERR_SETINSETTRIGONLY, 2, RTS_ERROR_TEXT("$ZTVALUE"));
assert(0 < gtm_trigger_depth);
memcpy(dollar_ztvalue, v, SIZEOF(mval));
dollar_ztvalue->mvtype &= ~MV_ALIASCONT; /* Make sure to shut off alias container flag on copy */
assert(NULL != ztvalue_changed_ptr);
*ztvalue_changed_ptr = TRUE;
break;
# else
rts_error(VARLSTCNT(1) ERR_UNIMPLOP);
# endif
case SV_ZTWORMHOLE:
# ifdef GTM_TRIGGER
MV_FORCE_STR(v);
/* See jnl.h for why MAX_ZTWORMHOLE_SIZE should be less than minimum alignsize */
assert(MAX_ZTWORMHOLE_SIZE < (JNL_MIN_ALIGNSIZE * DISK_BLOCK_SIZE));
if (MAX_ZTWORMHOLE_SIZE < v->str.len)
rts_error(VARLSTCNT(4) ERR_ZTWORMHOLE2BIG, 2, v->str.len, MAX_ZTWORMHOLE_SIZE);
dollar_ztwormhole.mvtype = MV_STR;
dollar_ztwormhole.str = v->str;
break;
# else
rts_error(VARLSTCNT(1) ERR_UNIMPLOP);
# endif
case SV_ZTSLATE:
# ifdef GTM_TRIGGER
assert(!dollar_tlevel || (tstart_trigger_depth <= gtm_trigger_depth));
if (!dollar_tlevel || (tstart_trigger_depth == gtm_trigger_depth))
rts_error(VARLSTCNT(4) ERR_SETINTRIGONLY, 2, RTS_ERROR_TEXT("$ZTSLATE"));
assert(0 < gtm_trigger_depth);
MV_FORCE_DEFINED(v);
memcpy((char *)&dollar_ztslate, v, SIZEOF(mval));
dollar_ztslate.mvtype &= ~MV_ALIASCONT; /* Make sure to shut off alias container flag on copy */
break;
# else
rts_error(VARLSTCNT(1) ERR_UNIMPLOP);
# endif
default:
GTMASSERT;
}
return;
}
uint4 mur_process_intrpt_recov()
{
jnl_ctl_list *jctl, *last_jctl;
reg_ctl_list *rctl, *rctl_top;
int rename_fn_len, save_name_len;
char prev_jnl_fn[MAX_FN_LEN + 1], rename_fn[MAX_FN_LEN + 1], save_name[MAX_FN_LEN + 1];
jnl_create_info jnl_info;
uint4 status, status2;
uint4 max_autoswitchlimit, max_jnl_alq, max_jnl_deq, freeblks;
sgmnt_data_ptr_t csd;
#if defined(VMS)
io_status_block_disk iosb;
#endif
boolean_t jfh_changed;
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
gv_cur_region = rctl->gd; /* wcs_flu requires this to be set */
cs_addrs = rctl->csa;
csd = cs_data = rctl->csd; /* MM logic after wcs_flu call requires this to be set */
assert(csd == rctl->csa->hdr);
jctl = rctl->jctl_turn_around;
max_jnl_alq = max_jnl_deq = max_autoswitchlimit = 0;
for (last_jctl = NULL ; (NULL != jctl); last_jctl = jctl, jctl = jctl->next_gen)
{
if (max_autoswitchlimit < jctl->jfh->autoswitchlimit)
{ /* Note that max_jnl_alq, max_jnl_deq are not the maximum journal allocation/extensions across
* generations, but rather the allocation/extension corresponding to the maximum autoswitchlimit.
*/
max_autoswitchlimit = jctl->jfh->autoswitchlimit;
max_jnl_alq = jctl->jfh->jnl_alq;
max_jnl_deq = jctl->jfh->jnl_deq;
}
/* Until now, "rctl->blks_to_upgrd_adjust" holds the number of V4 format newly created bitmap blocks
* seen in INCTN records in backward processing. It is possible that backward processing might have
* missed out on seeing those INCTN records which are part of virtually-truncated or completely-rolled-bak
* journal files. The journal file-header has a separate field "prev_recov_blks_to_upgrd_adjust" which
* maintains exactly this count. Therefore adjust the rctl counter accordingly.
*/
assert(!jctl->jfh->prev_recov_blks_to_upgrd_adjust || !jctl->jfh->recover_interrupted);
assert(!jctl->jfh->prev_recov_blks_to_upgrd_adjust || jctl->jfh->prev_recov_end_of_data);
rctl->blks_to_upgrd_adjust += jctl->jfh->prev_recov_blks_to_upgrd_adjust;
}
if (max_autoswitchlimit > last_jctl->jfh->autoswitchlimit)
{
csd->jnl_alq = max_jnl_alq;
csd->jnl_deq = max_jnl_deq;
csd->autoswitchlimit = max_autoswitchlimit;
} else
{
assert(csd->jnl_alq == last_jctl->jfh->jnl_alq);
assert(csd->jnl_deq == last_jctl->jfh->jnl_deq);
assert(csd->autoswitchlimit == last_jctl->jfh->autoswitchlimit);
}
/* now that rctl->blks_to_upgrd_adjust is completely computed, use that to increment filehdr blks_to_upgrd. */
csd->blks_to_upgrd += rctl->blks_to_upgrd_adjust;
if (csd->blks_to_upgrd)
csd->fully_upgraded = FALSE;
jctl = rctl->jctl_turn_around;
csd->trans_hist.early_tn = jctl->turn_around_tn;
csd->trans_hist.curr_tn = csd->trans_hist.early_tn; /* INCREMENT_CURR_TN macro not used but noted in comment
* to identify all places that set curr_tn */
csd->jnl_eovtn = csd->trans_hist.curr_tn;
csd->turn_around_point = TRUE;
/* MUPIP REORG UPGRADE/DOWNGRADE stores its partially processed state in the database file header.
* It is difficult for recovery to restore those fields to a correct partial value.
* Hence reset the related fields as if the desired_db_format got set just ONE tn BEFORE the EPOCH record
* and that there was no more processing that happened.
* This might potentially mean some duplicate processing for MUPIP REORG UPGRADE/DOWNGRADE after the recovery.
* But that will only be the case as long as the database is in compatibility (mixed) mode (hopefully not long).
*/
if (csd->desired_db_format_tn >= jctl->turn_around_tn)
csd->desired_db_format_tn = jctl->turn_around_tn - 1;
if (csd->reorg_db_fmt_start_tn >= jctl->turn_around_tn)
csd->reorg_db_fmt_start_tn = jctl->turn_around_tn - 1;
if (csd->tn_upgrd_blks_0 > jctl->turn_around_tn)
csd->tn_upgrd_blks_0 = (trans_num)-1;
csd->reorg_upgrd_dwngrd_restart_block = 0;
/* Compute current value of "free_blocks" based on the value of "free_blocks" at the turnaround point epoch
* record and the change in "total_blks" since that epoch to the present form of the database. Any difference
* in "total_blks" implies database file extensions happened since the turnaround point. A backward rollback
* undoes everything (including all updates) except file extensions (it does not truncate the file size).
* Therefore every block that was newly allocated as part of those file extensions should be considered FREE
* for the current calculations except for the local bitmap blocks which are BUSY the moment they are created.
*/
assert(rctl->trnarnd_total_blks <= csd->trans_hist.total_blks);
csd->trans_hist.free_blocks = rctl->trnarnd_free_blocks + (csd->trans_hist.total_blks - rctl->trnarnd_total_blks)
- DIVIDE_ROUND_UP(csd->trans_hist.total_blks, BLKS_PER_LMAP)
+ DIVIDE_ROUND_UP(rctl->trnarnd_total_blks, BLKS_PER_LMAP);
assert((freeblks = mur_blocks_free(rctl)) == csd->trans_hist.free_blocks);
if (dba_bg == csd->acc_meth)
/* This is taken from bt_refresh() */
((th_rec *)((uchar_ptr_t)cs_addrs->th_base + cs_addrs->th_base->tnque.fl))->tn = jctl->turn_around_tn - 1;
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_FSYNC_DB);
csd->turn_around_point = FALSE;
/* In case this is MM and wcs_flu() remapped an extended database, reset rctl->csd */
assert((dba_mm == cs_data->acc_meth) || (rctl->csd == cs_data));
rctl->csd = cs_data;
}
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
if (!rctl->jfh_recov_interrupted)
jctl = rctl->jctl_turn_around;
else
{
DEBUG_ONLY(
for (jctl = rctl->jctl_turn_around; NULL != jctl->next_gen; jctl = jctl->next_gen)
;
/* check that latest gener file name does not match db header */
assert((rctl->csd->jnl_file_len != jctl->jnl_fn_len)
|| (0 != memcmp(rctl->csd->jnl_file_name, jctl->jnl_fn, jctl->jnl_fn_len)));
)
jctl = rctl->jctl_alt_head;
}
assert(NULL != jctl);
for ( ; NULL != jctl->next_gen; jctl = jctl->next_gen)
;
assert(rctl->csd->jnl_file_len == jctl->jnl_fn_len); /* latest gener file name */
assert(0 == memcmp(rctl->csd->jnl_file_name, jctl->jnl_fn, jctl->jnl_fn_len)); /* should match db header */
if (SS_NORMAL != (status = prepare_unique_name((char *)jctl->jnl_fn, jctl->jnl_fn_len, "", "",
rename_fn, &rename_fn_len, &status2)))
return status;
jctl->jnl_fn_len = rename_fn_len; /* change the name in memory to the proposed name */
memcpy(jctl->jnl_fn, rename_fn, rename_fn_len + 1);
/* Rename hasn't happened yet at the filesystem level. In case current recover command is interrupted,
* we need to update jfh->next_jnl_file_name before mur_forward(). Update jfh->next_jnl_file_name for
* all journal files from which PBLK records were applied. Create new journal files for forward play.
*/
assert(NULL != rctl->jctl_turn_around);
jctl = rctl->jctl_turn_around; /* points to journal file which has current recover's turn around point */
assert(0 != jctl->turn_around_offset);
jctl->jfh->turn_around_offset = jctl->turn_around_offset; /* save progress in file header for */
jctl->jfh->turn_around_time = jctl->turn_around_time; /* possible re-issue of recover */
jfh_changed = TRUE;
for ( ; NULL != jctl; jctl = jctl->next_gen)
{ /* setup the next_jnl links. note that in the case of interrupted recovery, next_jnl links
* would have been already set starting from the turn-around point journal file of the
* interrupted recovery but the new recovery MIGHT have taken us to a still previous
* generation journal file that needs its next_jnl link set. this is why we do the next_jnl
* link setup even in the case of interrupted recovery although in most cases it is unnecessary.
*/
if (NULL != jctl->next_gen)
{
jctl->jfh->next_jnl_file_name_length = jctl->next_gen->jnl_fn_len;
memcpy(jctl->jfh->next_jnl_file_name, jctl->next_gen->jnl_fn, jctl->next_gen->jnl_fn_len);
jfh_changed = TRUE;
} else
assert(0 == jctl->jfh->next_jnl_file_name_length); /* null link from latest generation */
if (jctl->jfh->turn_around_offset && (jctl != rctl->jctl_turn_around))
{ /* It is possible that the current recovery has a turn-around-point much before the
* previously interrupted recovery. If it happens to be a previous generation journal
* file then we have to reset the original turn-around-point to be zero in the journal
* file header in order to ensure if this recovery gets interrupted we do interrupted
* recovery processing until the new turn-around-point instead of stopping incorrectly
* at the original turn-around-point itself. Note that there could be more than one
* journal file with a non-zero turn_around_offset (depending on how many previous
* recoveries got interrupted in this loop) that need to be reset.
*/
assert(!jctl->turn_around_offset);
assert(rctl->recov_interrupted); /* rctl->jfh_recov_interrupted can fail */
jctl->jfh->turn_around_offset = 0;
jctl->jfh->turn_around_time = 0;
jfh_changed = TRUE;
}
if (jfh_changed)
{
DO_FILE_WRITE(jctl->channel, 0, jctl->jfh, REAL_JNL_HDR_LEN, jctl->status, jctl->status2);
if (SS_NORMAL != jctl->status)
{
assert(FALSE);
if (SS_NORMAL == jctl->status2)
gtm_putmsg(VARLSTCNT(5) ERR_JNLWRERR, 2, jctl->jnl_fn_len,
jctl->jnl_fn, jctl->status);
else
gtm_putmsg(VARLSTCNT1(6) ERR_JNLWRERR, 2, jctl->jnl_fn_len,
jctl->jnl_fn, jctl->status, PUT_SYS_ERRNO(jctl->status2));
return jctl->status;
}
UNIX_ONLY(
GTM_FSYNC(jctl->channel, jctl->status);
if (-1 == jctl->status)
{
jctl->status2 = errno;
assert(FALSE);
gtm_putmsg(VARLSTCNT(9) ERR_JNLFSYNCERR, 2,
jctl->jnl_fn_len, jctl->jnl_fn,
ERR_TEXT, 2, RTS_ERROR_TEXT("Error with fsync"), jctl->status2);
return ERR_JNLFSYNCERR;
}
)
}
jfh_changed = FALSE;
}
memset(&jnl_info, 0, SIZEOF(jnl_info));
jnl_info.status = jnl_info.status2 = SS_NORMAL;
jnl_info.prev_jnl = &prev_jnl_fn[0];
set_jnl_info(rctl->gd, &jnl_info);
jnl_info.prev_jnl_len = rctl->jctl_turn_around->jnl_fn_len;
memcpy(jnl_info.prev_jnl, rctl->jctl_turn_around->jnl_fn, rctl->jctl_turn_around->jnl_fn_len);
jnl_info.prev_jnl[jnl_info.prev_jnl_len] = 0;
jnl_info.jnl_len = rctl->csd->jnl_file_len;
memcpy(jnl_info.jnl, rctl->csd->jnl_file_name, jnl_info.jnl_len);
jnl_info.jnl[jnl_info.jnl_len] = 0;
assert(!mur_options.rollback || jgbl.mur_rollback);
jnl_info.reg_seqno = rctl->jctl_turn_around->turn_around_seqno;
jgbl.gbl_jrec_time = rctl->jctl_turn_around->turn_around_time; /* time needed for cre_jnl_file_common() */
if (EXIT_NRM != cre_jnl_file_common(&jnl_info, rename_fn, rename_fn_len))
{
gtm_putmsg(VARLSTCNT(4) ERR_JNLNOCREATE, 2, jnl_info.jnl_len, jnl_info.jnl);
return jnl_info.status;
}
if (NULL != rctl->jctl_alt_head) /* remove the journal files created by last interrupted recover process */
{
mur_rem_jctls(rctl);
rctl->jctl_alt_head = NULL;
}
/* From this point on, journal records are written into the newly created journal file. However, we still read
* from old journal files.
*/
}
int send_mesg2gtmsecshr(unsigned int code, unsigned int id, char *path, int path_len)
{
int client_sockfd, create_server_status, fcntl_res;
int req_code, wait_count = 0;
int recv_len, send_len;
ssize_t num_chars_recvd, num_chars_sent;
int save_errno, ret_code = 0, init_ret_code = 0;
int loop_count = 0;
int recv_complete, send_complete;
boolean_t retry = FALSE;
size_t server_proc_len;
int semop_res;
int selstat, status;
char *recv_ptr, *send_ptr;
struct sockaddr_un server_proc;
struct sembuf sop[4];
fd_set wait_on_fd;
gtmsecshr_mesg mesg;
TID timer_id;
int4 msec_timeout;
char *gtm_tmp_ptr;
struct stat stat_buf;
struct shmid_ds shm_info;
int len;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
DBGGSSHR((LOGFLAGS, "secshr_client: New send request\n"));
if (!gtm_dist_ok_to_use)
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(6) ERR_GTMDISTUNVERIF, 4, STRLEN(gtm_dist), gtm_dist,
gtmImageNames[image_type].imageNameLen, gtmImageNames[image_type].imageName);
/* Create communication key (hash of release name) if it has not already been done */
if (0 == TREF(gtmsecshr_comkey))
{
STR_HASH((char *)gtm_release_name, gtm_release_name_len, TREF(gtmsecshr_comkey), 0);
}
timer_id = (TID)send_mesg2gtmsecshr;
if (!gtmsecshr_file_check_done)
{
len = STRLEN(gtm_dist);
memcpy(gtmsecshr_path, gtm_dist, len);
gtmsecshr_path[len] = '/';
memcpy(gtmsecshr_path + len + 1, GTMSECSHR_EXECUTABLE, STRLEN(GTMSECSHR_EXECUTABLE));
gtmsecshr_pathname.addr = gtmsecshr_path;
gtmsecshr_pathname.len = len + 1 + STRLEN(GTMSECSHR_EXECUTABLE);
assertpro(GTM_PATH_MAX > gtmsecshr_pathname.len);
gtmsecshr_pathname.addr[gtmsecshr_pathname.len] = '\0';
if (-1 == Stat(gtmsecshr_pathname.addr, &stat_buf))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8) ERR_SYSCALL, 5,
LEN_AND_LIT("stat"), CALLFROM, errno);
if ((ROOTUID != stat_buf.st_uid)
|| !(stat_buf.st_mode & S_ISUID)
|| (0 != ACCESS(gtmsecshr_pathname.addr, (X_OK))))
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(1) ERR_GTMSECSHRPERM);
gtmsecshr_file_check_done = TRUE;
}
if (!gtmsecshr_sock_init_done && (0 < (init_ret_code = gtmsecshr_sock_init(CLIENT)))) /* Note assignment */
return init_ret_code;
DEBUG_ONLY(mesg.usesecshr = TREF(gtm_usesecshr)); /* Flag ignored in PRO build */
while (MAX_COMM_ATTEMPTS >= loop_count)
{ /* first, try the sendto */
req_code = mesg.code = code;
send_len = (int4)(GTM_MESG_HDR_SIZE);
if (REMOVE_FILE == code)
{
assert(GTM_PATH_MAX > path_len); /* Name is not user supplied so simple check */
memcpy(mesg.mesg.path, path, path_len);
send_len += path_len;
} else if (FLUSH_DB_IPCS_INFO == code)
{
assert(GTM_PATH_MAX > db_ipcs.fn_len);
memcpy(&mesg.mesg.db_ipcs, &db_ipcs, (offsetof(ipcs_mesg, fn[0]) + db_ipcs.fn_len + 1));
/* Most of the time file length is much smaller than GTM_PATH_MAX */
send_len += offsetof(ipcs_mesg, fn[0]);
send_len += mesg.mesg.db_ipcs.fn_len + 1;
} else
{
mesg.mesg.id = id;
send_len += SIZEOF(mesg.mesg.id);
}
DBGGSSHR((LOGFLAGS, "secshr_client: loop %d frm-pid: %d to-pid: %d send_len: %d code: %d\n", loop_count,
process_id, id, send_len, code));
mesg.comkey = TREF(gtmsecshr_comkey); /* Version communication key */
mesg.pid = process_id; /* Process id of client */
mesg.seqno = ++cur_seqno;
send_ptr = (char *)&mesg;
send_complete = FALSE;
SENDTO_SOCK(gtmsecshr_sockfd, send_ptr, send_len, 0, (struct sockaddr *)>msecshr_sock_name,
(GTM_SOCKLEN_TYPE)gtmsecshr_sockpath_len, num_chars_sent); /* This form handles EINTR internally */
save_errno = errno;
DBGGSSHR((LOGFLAGS, "secshr_client: sendto rc: %d errno: %d (only important if rc=-1)\n", (int)num_chars_sent,
save_errno));
if (0 >= num_chars_sent)
{ /* SENDTO_SOCK failed - start server and attempt to resend */
if ((EISCONN == save_errno) || (EBADF == save_errno))
{
gtmsecshr_sock_cleanup(CLIENT);
gtmsecshr_sock_init(CLIENT);
wcs_backoff(loop_count + 1);
DBGGSSHR((LOGFLAGS, "secshr_client: Connection error, reset socket\n"));
} else
{
if (0 < loop_count)
/* No message unless attempted server start at least once */
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(11) ERR_GTMSECSHRSRVF, 4,
RTS_ERROR_TEXT("Client"), process_id,
loop_count - 1, ERR_TEXT, 2,
RTS_ERROR_TEXT("sendto to gtmsecshr failed"), save_errno);
START_SERVER;
DBGGSSHR((LOGFLAGS, "secshr_client: sendto() failed - restarting server\n"));
}
loop_count++;
continue;
}
SETUP_FOR_RECV; /* Sets timer, recvcomplete = FALSE */
do
{ /* Note RECVFROM does not loop on EINTR return codes so must be handled. Note also we only expect
* to receive the message header back as an acknowledgement.
*/
num_chars_recvd = RECVFROM(gtmsecshr_sockfd, recv_ptr, GTM_MESG_HDR_SIZE, 0, (struct sockaddr *)0,
(GTM_SOCKLEN_TYPE *)0);
save_errno = errno;
DBGGSSHR((LOGFLAGS, "secshr_client: recvfrom rc: %d errno: %d (only important if rc=-1)\n",
(int)num_chars_recvd, save_errno));
if (0 <= num_chars_recvd)
{ /* Message received - make sure it is large enough to have set seqno before we do anything
* to rely on it.
*/
if ((GTM_MESG_HDR_SIZE <= num_chars_recvd) && (mesg.seqno == cur_seqno)
&& (TREF(gtmsecshr_comkey) == mesg.comkey))
recv_complete = TRUE;
else
{ /* Message too short or not correct sequence */
cancel_timer(timer_id);
/* Print True/False for the possibilities we failed */
DBGGSSHR((LOGFLAGS, "secshr_client: Message incorrect - chars: %d, seq: %d\n",
(GTM_MESG_HDR_SIZE <= num_chars_recvd), (mesg.seqno == cur_seqno)));
SETUP_FOR_RECV;
continue;
}
} else
{ /* Something untoward happened */
if (client_timer_popped)
break;
if (EINTR == save_errno) /* Had an irrelevant interrupt - ignore */
continue;
if (EBADF == save_errno)
break;
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(11) ERR_GTMSECSHRSRVF, 4,
RTS_ERROR_TEXT("Client"), process_id, loop_count - 1, ERR_TEXT, 2,
RTS_ERROR_TEXT("recvfrom from gtmsecshr failed"), save_errno);
if ((ECONNRESET == save_errno) || (ENOTCONN == save_errno))
{
num_chars_recvd = 0;
break;
}
gtmsecshr_sock_cleanup(CLIENT);
return save_errno;
}
} while (!recv_complete);
cancel_timer(timer_id);
if (client_timer_popped || (EBADF == save_errno) || (0 == num_chars_recvd))
{ /* Timeout, connection issues, bad descriptor block - retry */
gtmsecshr_sock_cleanup(CLIENT);
gtmsecshr_sock_init(CLIENT);
retry = TRUE;
if (client_timer_popped)
{
START_SERVER;
DBGGSSHR((LOGFLAGS, "secshr_client: Read timer popped - restarting server\n"));
} else
DBGGSSHR((LOGFLAGS, "secshr_client: Read error - socket reset, retrying\n"));
loop_count++;
continue;
}
/* Response to *our* latest message available */
assert(recv_complete);
if (ret_code = mesg.code) /* Warning - assignment */
{
DBGGSSHR((LOGFLAGS, "secshr_client: non-zero response from gtmsecshr - request: %d retcode: %d\n",
req_code, ret_code));
if (INVALID_COMKEY == ret_code)
{ /* Comkey mismatch means for a different version of GT.M - we will not handle it */
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(13) ERR_GTMSECSHRSRVFIL, 7, RTS_ERROR_TEXT("Client"),
process_id, mesg.pid, req_code, RTS_ERROR_TEXT(mesg.mesg.path),
ERR_TEXT, 2, RTS_ERROR_STRING("Communicating with wrong GT.M version"));
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(13) ERR_GTMSECSHRSRVFIL, 7, RTS_ERROR_TEXT("Client"),
process_id, mesg.pid, req_code, RTS_ERROR_TEXT(mesg.mesg.path),
ERR_TEXT, 2, RTS_ERROR_STRING("Communicating with wrong GT.M version"));
break; /* rts_error should not return */
}
switch(req_code)
{
case REMOVE_FILE:
/* Called from mutex_sock_init(). Path (and length) contain null terminator byte.
* See if file still exists (may have been deleted by earlier attempt). Caller
* handles actual error.
*/
if ((-1 != Stat(path, &stat_buf)) || (ENOENT != ret_code))
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(14) ERR_GTMSECSHRSRVFIL, 7,
RTS_ERROR_TEXT("Client"),
process_id, mesg.pid, req_code, RTS_ERROR_TEXT(mesg.mesg.path),
ERR_TEXT, 2, RTS_ERROR_STRING(secshr_fail_mesg_code[req_code]),
mesg.code);
else
ret_code = 0; /* File is gone so this or a previous try actually worked */
break;
case REMOVE_SEM:
/* See if semaphore still eixsts (may have been removed by earlier attempt that
* got a reply confused or lost). If not there, no error. Else error to op-log.
*/
if ((-1 != semctl(id, 0, GETVAL)) && !SEM_REMOVED(errno))
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(13) ERR_GTMSECSHRSRVFID, 6,
RTS_ERROR_TEXT("Client"),
process_id, mesg.pid, req_code, mesg.mesg.id, ERR_TEXT, 2,
RTS_ERROR_STRING(secshr_fail_mesg_code[req_code]),
mesg.code);
else
ret_code = 0; /* File is gone so this or a previous try actually worked */
case REMOVE_SHM:
/* See if shmem still eixsts (may have been removed by earlier attempt that
* got a reply confused or lost). If not there, no error. Else error to op-log.
* Note -
*/
if ((-1 != shmctl(id, IPC_STAT, &shm_info)) && !SEM_REMOVED(errno))
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(13) ERR_GTMSECSHRSRVFID, 6,
RTS_ERROR_TEXT("Client"),
process_id, mesg.pid, req_code, mesg.mesg.id, ERR_TEXT, 2,
RTS_ERROR_STRING(secshr_fail_mesg_code[req_code]),
mesg.code);
else
ret_code = 0; /* File is gone so this or a previous try actually worked */
break;
case FLUSH_DB_IPCS_INFO: /* Errors handled by caller */
break;
default:
if (EPERM != mesg.code && EACCES != mesg.code)
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(13) ERR_GTMSECSHRSRVFID, 6,
RTS_ERROR_TEXT("Client"),
process_id, mesg.pid, req_code, mesg.mesg.id, ERR_TEXT, 2,
RTS_ERROR_STRING(secshr_fail_mesg_code[req_code]),
mesg.code);
break;
}
}
break;
}
if (MAX_COMM_ATTEMPTS < loop_count)
{
ret_code = -1;
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(10) ERR_GTMSECSHRSRVF, 4,
RTS_ERROR_TEXT("Client"), process_id, loop_count - 1,
ERR_TEXT, 2, RTS_ERROR_TEXT("Unable to communicate with gtmsecshr"));
/* If gtm_tmp is not defined, show default path */
if (gtm_tmp_ptr = GETENV("gtm_tmp"))
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(8) ERR_GTMSECSHRTMPPATH, 2,
RTS_ERROR_TEXT(gtm_tmp_ptr),
ERR_TEXT, 2, RTS_ERROR_TEXT("(from $gtm_tmp)"));
else
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(4)
ERR_GTMSECSHRTMPPATH, 2, RTS_ERROR_TEXT("/tmp"));
}
if (ONETIMESOCKET == init_ret_code)
gtmsecshr_sock_cleanup(CLIENT);
return ret_code;
}
void trigger_delete_all(void)
{
int count;
char count_str[MAX_DIGITS_IN_INT + 1];
sgmnt_addrs *csa;
mval curr_gbl_name;
int cycle;
mstr gbl_name;
mname_entry gvent;
gv_namehead *hasht_tree, *gvt;
mval *mv_count_ptr;
mval *mv_cycle_ptr;
mval mv_indx;
gd_region *reg;
int reg_indx;
int4 result;
char save_currkey[SIZEOF(gv_key) + DBKEYSIZE(MAX_KEY_SZ)];
gv_key *save_gv_currkey;
gd_region *save_gv_cur_region;
gv_namehead *save_gv_target;
sgm_info *save_sgm_info_ptr;
int trig_indx;
mval trigger_cycle;
mval trigger_count;
mval val;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(0 < dollar_tlevel);
/* Before we delete any triggers, verify that none of the triggers have been fired in this transaction. If they have,
* this creates an un-commitable transaction that will end in a TPFAIL error. Since that error indicates database
* damage, we'd rather detect this avoidable condition and give a descriptive error instead (TRIGMODINTP).
*/
for (gvt = gv_target_list; NULL != gvt; gvt = gvt->next_gvnh)
{
if (gvt->trig_local_tn == local_tn)
rts_error_csa(CSA_ARG(gvt->gd_csa) VARLSTCNT(1) ERR_TRIGMODINTP);
}
SWITCH_TO_DEFAULT_REGION;
if (gv_cur_region->read_only)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_TRIGMODREGNOTRW, 2, REG_LEN_STR(gv_cur_region));
INITIAL_HASHT_ROOT_SEARCH_IF_NEEDED;
if (0 != gv_target->root)
{
/* kill ^#t("#TRHASH") */
BUILD_HASHT_SUB_CURRKEY(LITERAL_HASHTRHASH, STRLEN(LITERAL_HASHTRHASH));
gvcst_kill(TRUE);
/* kill ^#t("#TNAME") */
BUILD_HASHT_SUB_CURRKEY(LITERAL_HASHTNAME, STRLEN(LITERAL_HASHTNAME));
gvcst_kill(TRUE);
}
for (reg_indx = 0, reg = gd_header->regions; reg_indx < gd_header->n_regions; reg_indx++, reg++)
{
if (!reg->open)
gv_init_reg(reg);
gv_cur_region = reg;
change_reg();
csa = cs_addrs;
SETUP_TRIGGER_GLOBAL;
INITIAL_HASHT_ROOT_SEARCH_IF_NEEDED;
/* There might not be any ^#t in this region, so check */
if (0 != gv_target->root)
{ /* Give error on region only if it has #t global indicating the presence
* of triggers.
*/
if (reg->read_only)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_TRIGMODREGNOTRW, REG_LEN_STR(reg));
/* Kill all descendents of ^#t(trigvn, indx) where trigvn is any global with a trigger,
* but skip the "#XYZ" entries. setup ^#t(trigvn,"$") as the PREV key for op_gvorder
*/
BUILD_HASHT_SUB_CURRKEY(LITERAL_MAXHASHVAL, STRLEN(LITERAL_MAXHASHVAL));
TREF(gv_last_subsc_null) = FALSE; /* We know its not null, but prior state is unreliable */
while (TRUE)
{
op_gvorder(&curr_gbl_name);
/* quit:$length(curr_gbl_name)=0 */
if (0 == curr_gbl_name.str.len)
break;
/* $get(^#t(curr_gbl_name,#COUNT)) */
BUILD_HASHT_SUB_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len,
LITERAL_HASHCOUNT, STRLEN(LITERAL_HASHCOUNT));
if (gvcst_get(&trigger_count))
{
mv_count_ptr = &trigger_count;
count = MV_FORCE_INT(mv_count_ptr);
/* $get(^#t(curr_gbl_name,#CYCLE)) */
BUILD_HASHT_SUB_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len,
LITERAL_HASHCYCLE, STRLEN(LITERAL_HASHCYCLE));
if (!gvcst_get(&trigger_cycle))
{ /* Found #COUNT, there must be #CYCLE */
if (CDB_STAGNATE > t_tries)
t_retry(cdb_sc_triggermod);
else
{
assert(WBTEST_HELPOUT_TRIGDEFBAD == gtm_white_box_test_case_number);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_TRIGDEFBAD, 6,
curr_gbl_name.str.len, curr_gbl_name.str.addr,
curr_gbl_name.str.len, curr_gbl_name.str.addr,
LEN_AND_LIT("\"#CYCLE\""),
ERR_TEXT, 2,
RTS_ERROR_TEXT("#CYCLE field is missing"));
}
}
mv_cycle_ptr = &trigger_cycle;
cycle = MV_FORCE_INT(mv_cycle_ptr);
/* kill ^#t(curr_gbl_name) */
BUILD_HASHT_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len);
gvcst_kill(TRUE);
cycle++;
MV_FORCE_MVAL(&trigger_cycle, cycle);
/* set ^#t(curr_gbl_name,#CYCLE)=trigger_cycle */
SET_TRIGGER_GLOBAL_SUB_SUB_MVAL(curr_gbl_name.str.addr, curr_gbl_name.str.len,
LITERAL_HASHCYCLE, STRLEN(LITERAL_HASHCYCLE), trigger_cycle, result);
assert(PUT_SUCCESS == result);
} /* else there is no #COUNT, then no triggers, leave #CYCLE alone */
/* get ready for op_gvorder() call for next trigger under ^#t */
BUILD_HASHT_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len);
}
csa->incr_db_trigger_cycle = TRUE;
if (dollar_ztrigger_invoked)
{ /* increment db_dztrigger_cycle so that next gvcst_put/gvcst_kill in this transaction,
* on this region, will re-read. See trigger_update.c for a comment on why it is okay
* for db_dztrigger_cycle to be incremented more than once in the same transaction
*/
csa->db_dztrigger_cycle++;
}
}
}
util_out_print_gtmio("All existing triggers deleted", FLUSH);
}
int gtm_trigger_complink(gv_trigger_t *trigdsc, boolean_t dolink)
{
char rtnname[GTM_PATH_MAX + 1], rtnname_template[GTM_PATH_MAX + 1];
char objname[GTM_PATH_MAX + 1];
char zcomp_parms[(GTM_PATH_MAX * 2) + SIZEOF(mident_fixed) + SIZEOF(OBJECT_PARM) + SIZEOF(NAMEOFRTN_PARM)];
mstr save_zsource;
int rtnfd, rc, lenrtnname, lenobjname, len, alphnum_len, retry, save_errno;
char *mident_suffix_p1, *mident_suffix_p2, *mident_suffix_top, *namesub1, *namesub2, *zcomp_parms_ptr;
mval zlfile, zcompprm;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
DBGTRIGR_ONLY(memcpy(rtnname, trigdsc->rtn_desc.rt_name.addr, trigdsc->rtn_desc.rt_name.len));
DBGTRIGR_ONLY(rtnname[trigdsc->rtn_desc.rt_name.len] = 0);
DBGTRIGR((stderr, "gtm_trigger_complink: (Re)compiling trigger %s\n", rtnname));
ESTABLISH_RET(gtm_trigger_complink_ch, ((0 == error_condition) ? TREF(dollar_zcstatus) : error_condition ));
/* Verify there are 2 available chars for uniqueness */
assert((MAX_MIDENT_LEN - TRIGGER_NAME_RESERVED_SPACE) >= (trigdsc->rtn_desc.rt_name.len));
assert(NULL == trigdsc->rtn_desc.rt_adr);
gtm_trigger_comp_prev_run_time = run_time;
run_time = TRUE; /* Required by compiler */
/* Verify the routine name set by MUPIP TRIGGER and read by gvtr_db_read_hasht() is not in use */
if (NULL != find_rtn_hdr(&trigdsc->rtn_desc.rt_name))
{ /* Ooops .. need name to be more unique.. */
/* Though variable definitions are conventionally done at the function entry, the reason alphanumeric_table
* definition is done here is to minimize the time taken to initialize the below table in the most common case
* (i.e. no trigger name collisions).
*/
char alphanumeric_table[] = {'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O',
'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd',
'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's',
't', 'u', 'v', 'w', 'x', 'y', 'z', '0', '1', '2', '3', '4', '5', '6', '7',
'8', '9', '\0'};
alphnum_len = STR_LIT_LEN(alphanumeric_table);
namesub1 = trigdsc->rtn_desc.rt_name.addr + trigdsc->rtn_desc.rt_name.len++;
/* If WBTEST_HELPOUT_TRIGNAMEUNIQ is defined, set alphnum_len to 1. This way, we make the maximum
* possible combinations for the uniqe trigger names to be 3 which is significantly lesser than
* the actual number of combinations (62x62 = 3844). For eg., if ^a is a global having triggers defined
* in 4 global directories, then the possible unique trigger names are a#1# ; a#1#A ; a#1#AA.
*/
GTM_WHITE_BOX_TEST(WBTEST_HELPOUT_TRIGNAMEUNIQ, alphnum_len, 1);
mident_suffix_top = (char *)alphanumeric_table + alphnum_len;
/* Phase 1. See if any single character can add uniqueness */
for (mident_suffix_p1 = (char *)alphanumeric_table; mident_suffix_p1 < mident_suffix_top; mident_suffix_p1++)
{
*namesub1 = *mident_suffix_p1;
if (NULL == find_rtn_hdr(&trigdsc->rtn_desc.rt_name))
break;
}
if (mident_suffix_p1 == mident_suffix_top)
{ /* Phase 2. Phase 1 could not find uniqueness .. Find it with 2 char variations */
namesub2 = trigdsc->rtn_desc.rt_name.addr + trigdsc->rtn_desc.rt_name.len++;
for (mident_suffix_p1 = (char *)alphanumeric_table; mident_suffix_p1 < mident_suffix_top;
mident_suffix_p1++)
{ /* First char loop */
for (mident_suffix_p2 = (char *)alphanumeric_table; mident_suffix_p2 < mident_suffix_top;
mident_suffix_p2++)
{ /* 2nd char loop */
*namesub1 = *mident_suffix_p1;
*namesub2 = *mident_suffix_p2;
if (NULL == find_rtn_hdr(&trigdsc->rtn_desc.rt_name))
{
mident_suffix_p1 = mident_suffix_top + 1; /* Break out of both loops */
break;
}
}
}
if (mident_suffix_p1 == mident_suffix_top)
{ /* Phase 3: Punt */
assert(WBTEST_HELPOUT_TRIGNAMEUNIQ == gtm_white_box_test_case_number);
rts_error(VARLSTCNT(5) ERR_TRIGNAMEUNIQ, 3, trigdsc->rtn_desc.rt_name.len - 2,
trigdsc->rtn_desc.rt_name.addr, alphnum_len * alphnum_len);
}
}
}
/* Write trigger execute string out to temporary file and compile it */
assert(MAX_XECUTE_LEN >= trigdsc->xecute_str.str.len);
rc = SNPRINTF(rtnname_template, GTM_PATH_MAX, "%s/trgtmpXXXXXX", DEFAULT_GTM_TMP);
assert(0 < rc); /* Note rc is return code aka length - we expect a non-zero length */
assert(GTM_PATH_MAX >= rc);
/* The mkstemp() routine is known to bogus-fail for no apparent reason at all especially on AIX 6.1. In the event
* this shortcoming plagues other platforms as well, we add a low-cost retry wrapper.
*/
retry = MAX_MKSTEMP_RETRIES;
do
{
strcpy(rtnname, rtnname_template);
rtnfd = mkstemp(rtnname);
} while ((-1 == rtnfd) && (EEXIST == errno) && (0 < --retry));
if (-1 == rtnfd)
{
save_errno = errno;
assert(FALSE);
rts_error(VARLSTCNT(12) ERR_SYSCALL, 5, RTS_ERROR_LITERAL("mkstemp()"), CALLFROM,
ERR_TEXT, 2, RTS_ERROR_TEXT(rtnname), save_errno);
}
assert(0 < rtnfd); /* Verify file descriptor */
rc = 0;
# ifdef GEN_TRIGCOMPFAIL_ERROR
{ /* Used ONLY to generate an error in a trigger compile by adding some junk in a previous line */
DOWRITERC(rtnfd, ERROR_CAUSING_JUNK, strlen(ERROR_CAUSING_JUNK), rc); /* BYPASSOK */
if (0 != rc)
{
UNLINK(rtnname);
rts_error(VARLSTCNT(8) ERR_SYSCALL, 5, RTS_ERROR_LITERAL("write()"), CALLFROM, rc);
}
}
# endif
DOWRITERC(rtnfd, trigdsc->xecute_str.str.addr, trigdsc->xecute_str.str.len, rc);
if (0 != rc)
{
UNLINK(rtnname);
rts_error(VARLSTCNT(8) ERR_SYSCALL, 5, RTS_ERROR_LITERAL("write()"), CALLFROM, rc);
}
if (NULL == memchr(trigdsc->xecute_str.str.addr, '\n', trigdsc->xecute_str.str.len))
{
DOWRITERC(rtnfd, NEWLINE, strlen(NEWLINE), rc); /* BYPASSOK */
if (0 != rc)
{
UNLINK(rtnname);
rts_error(VARLSTCNT(8) ERR_SYSCALL, 5, RTS_ERROR_LITERAL("write()"), CALLFROM, rc);
}
}
CLOSEFILE(rtnfd, rc);
if (0 != rc)
{
UNLINK(rtnname);
rts_error(VARLSTCNT(8) ERR_SYSCALL, 5, RTS_ERROR_LITERAL("close()"), CALLFROM, rc);
}
assert(MAX_MIDENT_LEN > trigdsc->rtn_desc.rt_name.len);
zcomp_parms_ptr = zcomp_parms;
lenrtnname = STRLEN(rtnname);
MEMCPY_LIT(zcomp_parms_ptr, NAMEOFRTN_PARM);
zcomp_parms_ptr += STRLEN(NAMEOFRTN_PARM);
memcpy(zcomp_parms_ptr, trigdsc->rtn_desc.rt_name.addr, trigdsc->rtn_desc.rt_name.len);
zcomp_parms_ptr += trigdsc->rtn_desc.rt_name.len;
MEMCPY_LIT(zcomp_parms_ptr, OBJECT_PARM);
zcomp_parms_ptr += STRLEN(OBJECT_PARM);
strcpy(objname, rtnname); /* Make copy of rtnname to become object name */
strcat(objname, OBJECT_FTYPE); /* Turn into object file reference */
lenobjname = lenrtnname + STRLEN(OBJECT_FTYPE);
memcpy(zcomp_parms_ptr, objname, lenobjname);
zcomp_parms_ptr += lenobjname;
*zcomp_parms_ptr++ = ' ';
memcpy(zcomp_parms_ptr, rtnname, lenrtnname);
zcomp_parms_ptr += lenrtnname;
*zcomp_parms_ptr = '\0'; /* Null tail */
len = INTCAST(zcomp_parms_ptr - zcomp_parms);
assert((SIZEOF(zcomp_parms) - 1) > len); /* Verify no overflow */
zcompprm.mvtype = MV_STR;
zcompprm.str.addr = zcomp_parms;
zcompprm.str.len = len;
/* Backup dollar_zsource so trigger doesn't show */
PUSH_MV_STENT(MVST_MSAV);
mv_chain->mv_st_cont.mvs_msav.v = dollar_zsource;
mv_chain->mv_st_cont.mvs_msav.addr = &dollar_zsource;
TREF(trigger_compile) = TRUE; /* Set flag so compiler knows this is a special trigger compile */
op_zcompile(&zcompprm, FALSE); /* Compile but don't require a .m file extension */
TREF(trigger_compile) = FALSE; /* compile_source_file() establishes handler so always returns */
if (0 != TREF(dollar_zcstatus))
{ /* Someone err'd.. */
run_time = gtm_trigger_comp_prev_run_time;
REVERT;
UNLINK(objname); /* Remove files before return error */
UNLINK(rtnname);
return ERR_TRIGCOMPFAIL;
}
if (dolink)
{ /* Link is optional as MUPIP TRIGGER doesn't need link */
zlfile.mvtype = MV_STR;
zlfile.str.addr = objname;
zlfile.str.len = lenobjname;
/* Specifying literal_null for a second arg (as opposed to NULL or 0) allows us to specify
* linking the object file (no compilation or looking for source). The 2nd arg is parms for
* recompilation and is non-null in an explicit zlink which we need to emulate.
*/
# ifdef GEN_TRIGLINKFAIL_ERROR
UNLINK(objname); /* delete object before it can be used */
# endif
op_zlink(&zlfile, (mval *)&literal_null); /* need cast due to "extern const" attributes */
/* No return here if link fails for some reason */
trigdsc->rtn_desc.rt_adr = find_rtn_hdr(&trigdsc->rtn_desc.rt_name);
if (NULL == trigdsc->rtn_desc.rt_adr)
GTMASSERT; /* Can't find routine we just put there? Catastrophic if happens */
/* Replace the randomly generated source name with the constant "GTM Trigger" */
trigdsc->rtn_desc.rt_adr->src_full_name.addr = GTM_TRIGGER_SOURCE_NAME;
trigdsc->rtn_desc.rt_adr->src_full_name.len = STRLEN(GTM_TRIGGER_SOURCE_NAME);
trigdsc->rtn_desc.rt_adr->trigr_handle = trigdsc; /* Back pointer to trig def */
}
if (MVST_MSAV == mv_chain->mv_st_type && &dollar_zsource == mv_chain->mv_st_cont.mvs_msav.addr)
{ /* Top mv_stent is one we pushed on there - restore dollar_zsource and get rid of it */
dollar_zsource = mv_chain->mv_st_cont.mvs_msav.v;
POP_MV_STENT();
} else
assert(FALSE); /* This mv_stent should be the one we just pushed */
/* Remove temporary files created */
UNLINK(objname); /* Delete the object file first since rtnname is the unique key */
UNLINK(rtnname); /* Delete the source file */
run_time = gtm_trigger_comp_prev_run_time;
REVERT;
return 0;
}
void bin_load(uint4 begin, uint4 end)
{
unsigned char *ptr, *cp1, *cp2, *btop, *gvkey_char_ptr, *tmp_ptr, *tmp_key_ptr, *c, *ctop, *ptr_base;
unsigned char hdr_lvl, src_buff[MAX_KEY_SZ + 1], dest_buff[MAX_ZWR_KEY_SZ],
cmpc_str[MAX_KEY_SZ + 1], dup_key_str[MAX_KEY_SZ + 1], sn_key_str[MAX_KEY_SZ + 1], *sn_key_str_end;
unsigned char *end_buff;
unsigned short rec_len, next_cmpc, numsubs;
int len;
int current, last, length, max_blk_siz, max_key, status;
int tmp_cmpc, sn_chunk_number, expected_sn_chunk_number = 0, sn_hold_buff_pos, sn_hold_buff_size;
uint4 iter, max_data_len, max_subsc_len, key_count, gblsize;
ssize_t rec_count, global_key_count, subsc_len,extr_std_null_coll, last_sn_error_offset=0,
file_offset_base=0, file_offset=0;
boolean_t need_xlation, new_gvn, utf8_extract;
boolean_t is_hidden_subscript, ok_to_put = TRUE, putting_a_sn = FALSE, sn_incmp_gbl_already_killed = FALSE;
rec_hdr *rp, *next_rp;
mval v, tmp_mval;
mstr mstr_src, mstr_dest;
collseq *extr_collseq, *db_collseq, *save_gv_target_collseq;
coll_hdr extr_collhdr, db_collhdr;
gv_key *tmp_gvkey = NULL; /* null-initialize at start, will be malloced later */
gv_key *sn_gvkey = NULL; /* null-initialize at start, will be malloced later */
gv_key *sn_savekey = NULL; /* null-initialize at start, will be malloced later */
char std_null_coll[BIN_HEADER_NUMSZ + 1], *sn_hold_buff = NULL, *sn_hold_buff_temp = NULL;
# ifdef GTM_CRYPT
gtmcrypt_key_t *encr_key_handles;
char *inbuf;
int4 index;
int req_dec_blk_size, init_status, crypt_status;
muext_hash_hdr_ptr_t hash_array = NULL;
# endif
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(4 == SIZEOF(coll_hdr));
gvinit();
v.mvtype = MV_STR;
len = file_input_bin_get((char **)&ptr, &file_offset_base, (char **)&ptr_base);
hdr_lvl = EXTR_HEADER_LEVEL(ptr);
if (!(((('4' == hdr_lvl) || ('5' == hdr_lvl)) && (V5_BIN_HEADER_SZ == len)) ||
(('6' == hdr_lvl) && (BIN_HEADER_SZ == len)) ||
(('7' == hdr_lvl) && (BIN_HEADER_SZ == len)) ||
(('4' > hdr_lvl) && (V3_BIN_HEADER_SZ == len))))
{
rts_error(VARLSTCNT(1) ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
/* expecting the level in a single character */
assert(' ' == *(ptr + SIZEOF(BIN_HEADER_LABEL) - 3));
if (0 != memcmp(ptr, BIN_HEADER_LABEL, SIZEOF(BIN_HEADER_LABEL) - 2) || ('2' > hdr_lvl) ||
*(BIN_HEADER_VERSION_ENCR) < hdr_lvl)
{ /* ignore the level check */
rts_error(VARLSTCNT(1) ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
/* check if extract was generated in UTF-8 mode */
utf8_extract = (0 == MEMCMP_LIT(&ptr[len - BIN_HEADER_LABELSZ], UTF8_NAME)) ? TRUE : FALSE;
if ((utf8_extract && !gtm_utf8_mode) || (!utf8_extract && gtm_utf8_mode))
{ /* extract CHSET doesn't match $ZCHSET */
if (utf8_extract)
rts_error(VARLSTCNT(4) ERR_LOADINVCHSET, 2, LEN_AND_LIT("UTF-8"));
else
rts_error(VARLSTCNT(4) ERR_LOADINVCHSET, 2, LEN_AND_LIT("M"));
mupip_exit(ERR_LDBINFMT);
}
if ('4' >= hdr_lvl)
{ /* Binary extracts in V50000-to-V52000 (label=4) and pre-V50000 (label=3) could have a '\0' byte (NULL byte)
* in the middle of the string. Replace it with ' ' (space) like it would be in V52000 binary extracts and above.
*/
for (c = ptr, ctop = c + len; c < ctop; c++)
{
if ('\0' == *c)
*c = ' ';
}
}
util_out_print("Label = !AD\n", TRUE, len, ptr);
new_gvn = FALSE;
if (hdr_lvl > '3')
{
if (hdr_lvl > '5')
{
memcpy(std_null_coll, ptr + BIN_HEADER_NULLCOLLOFFSET, BIN_HEADER_NUMSZ);
std_null_coll[BIN_HEADER_NUMSZ] = '\0';
}
else
{
memcpy(std_null_coll, ptr + V5_BIN_HEADER_NULLCOLLOFFSET, V5_BIN_HEADER_NUMSZ);
std_null_coll[V5_BIN_HEADER_NUMSZ] = '\0';
}
extr_std_null_coll = STRTOUL(std_null_coll, NULL, 10);
if (0 != extr_std_null_coll && 1!= extr_std_null_coll)
{
rts_error(VARLSTCNT(5) ERR_TEXT, 2, RTS_ERROR_TEXT("Corrupted null collation field in header"),
ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
} else
extr_std_null_coll = 0;
# ifdef GTM_CRYPT
if ('7' <= hdr_lvl)
{
int i, num_indexes;
len = file_input_bin_get((char **)&ptr, &file_offset_base, (char **)&ptr_base);
hash_array = (muext_hash_hdr *)malloc(len);
/* store hashes of all the files used during extract into muext_hash_hdr structure */
memcpy((char *)hash_array, ptr, len);
num_indexes = len / GTMCRYPT_HASH_LEN;
encr_key_handles = (gtmcrypt_key_t *)malloc(SIZEOF(gtmcrypt_key_t) * num_indexes);
INIT_PROC_ENCRYPTION(crypt_status);
GC_BIN_LOAD_ERR(crypt_status);
for (index = 0; index < num_indexes; index++)
{
if (0 == memcmp(hash_array[index].gtmcrypt_hash, EMPTY_GTMCRYPT_HASH, GTMCRYPT_HASH_LEN))
continue;
GTMCRYPT_GETKEY(hash_array[index].gtmcrypt_hash, encr_key_handles[index], crypt_status);
GC_BIN_LOAD_ERR(crypt_status);
}
}
# endif
if ('2' < hdr_lvl)
{
len = file_input_bin_get((char **)&ptr, &file_offset_base, (char **)&ptr_base);
if (SIZEOF(coll_hdr) != len)
{
rts_error(VARLSTCNT(5) ERR_TEXT, 2, RTS_ERROR_TEXT("Corrupt collation header"), ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
extr_collhdr = *((coll_hdr *)(ptr));
new_gvn = TRUE;
} else
gtm_putmsg(VARLSTCNT(3) ERR_OLDBINEXTRACT, 1, hdr_lvl - '0');
if (begin < 2)
begin = 2;
for (iter = 2; iter < begin; iter++)
{
if (!(len = file_input_bin_get((char **)&ptr, &file_offset_base, (char **)&ptr_base)))
{
gtm_putmsg(VARLSTCNT(3) ERR_LOADEOF, 1, begin);
util_out_print("Error reading record number: !UL\n", TRUE, iter);
mupip_error_occurred = TRUE;
return;
} else if (len == SIZEOF(coll_hdr))
{
extr_collhdr = *((coll_hdr *)(ptr));
assert(hdr_lvl > '2');
iter--;
}
}
assert(iter == begin);
util_out_print("Beginning LOAD at record number: !UL\n", TRUE, begin);
max_data_len = 0;
max_subsc_len = 0;
global_key_count = key_count = 0;
rec_count = begin - 1;
extr_collseq = db_collseq = NULL;
need_xlation = FALSE;
assert(NULL == tmp_gvkey); /* GVKEY_INIT macro relies on this */
GVKEY_INIT(tmp_gvkey, DBKEYSIZE(MAX_KEY_SZ)); /* tmp_gvkey will point to malloced memory after this */
assert(NULL == sn_gvkey); /* GVKEY_INIT macro relies on this */
GVKEY_INIT(sn_gvkey, DBKEYSIZE(MAX_KEY_SZ)); /* sn_gvkey will point to malloced memory after this */
assert(NULL == sn_savekey); /* GVKEY_INIT macro relies on this */
GVKEY_INIT(sn_savekey, DBKEYSIZE(MAX_KEY_SZ)); /* sn_gvkey will point to malloced memory after this */
for (; !mupip_DB_full ;)
{
if (++rec_count > end)
break;
next_cmpc = 0;
mupip_error_occurred = FALSE;
if (mu_ctrly_occurred)
break;
if (mu_ctrlc_occurred)
{
util_out_print("!AD:!_ Key cnt: !UL max subsc len: !UL max data len: !UL", TRUE,
LEN_AND_LIT(gt_lit), key_count, max_subsc_len, max_data_len);
util_out_print("Last LOAD record number: !UL", TRUE, key_count ? (rec_count - 1) : 0);
mu_gvis();
util_out_print(0, TRUE);
mu_ctrlc_occurred = FALSE;
}
if (!(len = file_input_bin_get((char **)&ptr, &file_offset_base, (char **)&ptr_base)) || mupip_error_occurred)
break;
else if (len == SIZEOF(coll_hdr))
{
extr_collhdr = *((coll_hdr *)(ptr));
assert(hdr_lvl > '2');
new_gvn = TRUE; /* next record will contain a new gvn */
rec_count--; /* Decrement as this record does not count as a record for loading purposes */
continue;
}
rp = (rec_hdr*)(ptr);
# ifdef GTM_CRYPT
if ('7' <= hdr_lvl)
{ /* Getting index value from the extracted file. It indicates which database file this record belongs to */
GET_LONG(index, ptr);
if (-1 != index) /* Indicates that the record is encrypted. */
{
req_dec_blk_size = len - SIZEOF(int4);
inbuf = (char *)(ptr + SIZEOF(int4));
GTMCRYPT_DECODE_FAST(encr_key_handles[index], inbuf, req_dec_blk_size, NULL, crypt_status);
GC_BIN_LOAD_ERR(crypt_status);
}
rp = (rec_hdr*)(ptr + SIZEOF(int4));
}
# endif
btop = ptr + len;
cp1 = (unsigned char*)(rp + 1);
v.str.addr = (char*)cp1;
while (*cp1++)
;
v.str.len =INTCAST((char*)cp1 - v.str.addr - 1);
if (('2' >= hdr_lvl) || new_gvn)
{
if ((HASHT_GBLNAME_LEN == v.str.len) && (0 == memcmp(v.str.addr, HASHT_GBLNAME, HASHT_GBLNAME_LEN)))
continue;
bin_call_db(BIN_BIND, (INTPTR_T)gd_header, (INTPTR_T)&v.str);
max_key = gv_cur_region->max_key_size;
db_collhdr.act = gv_target->act;
db_collhdr.ver = gv_target->ver;
db_collhdr.nct = gv_target->nct;
}
GET_USHORT(rec_len, &rp->rsiz);
if (EVAL_CMPC(rp) != 0 || v.str.len > rec_len || mupip_error_occurred)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
DISPLAY_FILE_OFFSET_OF_RECORD_AND_REST_OF_BLOCK;
continue;
}
if (new_gvn)
{
global_key_count = 1;
if ((db_collhdr.act != extr_collhdr.act || db_collhdr.ver != extr_collhdr.ver
|| db_collhdr.nct != extr_collhdr.nct
|| gv_cur_region->std_null_coll != extr_std_null_coll))
{
if (extr_collhdr.act)
{
if (extr_collseq = ready_collseq((int)extr_collhdr.act))
{
if (!do_verify(extr_collseq, extr_collhdr.act, extr_collhdr.ver))
{
gtm_putmsg(VARLSTCNT(8) ERR_COLLTYPVERSION, 2, extr_collhdr.act,
extr_collhdr.ver, ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLTYPVERSION);
}
} else
{
gtm_putmsg(VARLSTCNT(7) ERR_COLLATIONUNDEF, 1, extr_collhdr.act,
ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLATIONUNDEF);
}
}
if (db_collhdr.act)
{
if (db_collseq = ready_collseq((int)db_collhdr.act))
{
if (!do_verify(db_collseq, db_collhdr.act, db_collhdr.ver))
{
gtm_putmsg(VARLSTCNT(8) ERR_COLLTYPVERSION, 2, db_collhdr.act,
db_collhdr.ver, ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLTYPVERSION);
}
} else
{
gtm_putmsg(VARLSTCNT(7) ERR_COLLATIONUNDEF, 1, db_collhdr.act,
ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLATIONUNDEF);
}
}
need_xlation = TRUE;
} else
need_xlation = FALSE;
}
new_gvn = FALSE;
for (; rp < (rec_hdr*)btop; rp = (rec_hdr*)((unsigned char *)rp + rec_len))
{
GET_USHORT(rec_len, &rp->rsiz);
if (rec_len + (unsigned char *)rp > btop)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
DISPLAY_FILE_OFFSET_OF_RECORD_AND_REST_OF_BLOCK;
break;
}
cp1 = (unsigned char*)(rp + 1);
cp2 = gv_currkey->base + EVAL_CMPC(rp);
current = 1;
for (;;)
{
last = current;
current = *cp2++ = *cp1++;
if (0 == last && 0 == current)
break;
if (cp1 > (unsigned char *)rp + rec_len ||
cp2 > (unsigned char *)gv_currkey + gv_currkey->top)
{
gv_currkey->end = cp2 - gv_currkey->base - 1;
gv_currkey->base[gv_currkey->end] = 0;
gv_currkey->base[gv_currkey->end - 1] = 0;
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
DISPLAY_FILE_OFFSET_OF_RECORD_AND_REST_OF_BLOCK;
break;
}
}
if (mupip_error_occurred)
break;
gv_currkey->end = cp2 - gv_currkey->base - 1;
if (need_xlation)
{
assert(hdr_lvl >= '3');
assert(extr_collhdr.act || db_collhdr.act || extr_collhdr.nct || db_collhdr.nct ||
extr_std_null_coll != gv_cur_region->std_null_coll);
/* gv_currkey would have been modified/translated in the earlier put */
memcpy(gv_currkey->base, cmpc_str, next_cmpc);
next_rp = (rec_hdr *)((unsigned char*)rp + rec_len);
if ((unsigned char*)next_rp < btop)
{
next_cmpc = EVAL_CMPC(next_rp);
assert(next_cmpc <= gv_currkey->end);
memcpy(cmpc_str, gv_currkey->base, next_cmpc);
} else
next_cmpc = 0;
/* length of the key might change (due to nct variation),
* so get a copy of the original key from the extract */
memcpy(dup_key_str, gv_currkey->base, gv_currkey->end + 1);
gvkey_char_ptr = dup_key_str;
while (*gvkey_char_ptr++)
;
gv_currkey->prev = 0;
gv_currkey->end = gvkey_char_ptr - dup_key_str;
assert(gv_keysize <= tmp_gvkey->top);
while (*gvkey_char_ptr)
{
/* get next subscript (in GT.M internal subsc format) */
subsc_len = 0;
tmp_ptr = src_buff;
while (*gvkey_char_ptr)
*tmp_ptr++ = *gvkey_char_ptr++;
subsc_len = tmp_ptr - src_buff;
src_buff[subsc_len] = '\0';
if (extr_collseq)
{
/* undo the extract time collation */
TREF(transform) = TRUE;
save_gv_target_collseq = gv_target->collseq;
gv_target->collseq = extr_collseq;
} else
TREF(transform) = FALSE;
/* convert the subscript to string format */
end_buff = gvsub2str(src_buff, dest_buff, FALSE);
/* transform the string to the current subsc format */
TREF(transform) = TRUE;
tmp_mval.mvtype = MV_STR;
tmp_mval.str.addr = (char *)dest_buff;
tmp_mval.str.len = INTCAST(end_buff - dest_buff);
tmp_gvkey->prev = 0;
tmp_gvkey->end = 0;
if (extr_collseq)
gv_target->collseq = save_gv_target_collseq;
mval2subsc(&tmp_mval, tmp_gvkey);
/* we now have the correctly transformed subscript */
tmp_key_ptr = gv_currkey->base + gv_currkey->end;
memcpy(tmp_key_ptr, tmp_gvkey->base, tmp_gvkey->end + 1);
gv_currkey->prev = gv_currkey->end;
gv_currkey->end += tmp_gvkey->end;
gvkey_char_ptr++;
}
if ( gv_cur_region->std_null_coll != extr_std_null_coll && gv_currkey->prev)
{
if (extr_std_null_coll == 0)
{
GTM2STDNULLCOLL(gv_currkey->base, gv_currkey->end);
} else
{
STD2GTMNULLCOLL(gv_currkey->base, gv_currkey->end);
}
}
}
if (gv_currkey->end >= max_key)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
DISPLAY_FILE_OFFSET_OF_RECORD_AND_REST_OF_BLOCK;
continue;
}
/*
* Spanning node-related variables and their usage:
*
* expected_sn_chunk_number: 0 - looking for spanning nodes (regular nodes are OK, too)
* !0 - number of the next chunk needed (implies we are building
* a spanning node's value)
*
* While building a spanning node's value:
* numsubs: the number of chunks needed to build the spanning node's value
* gblsize: the expected size of the completed value
* sn_chunk_number: The chunk number of the chunk from the current record from the extract
*
* Managing the value
* sn_hold_buff: buffer used to accumulate the spanning node's value
* sn_hold_buff_size: Allocated size of buffer
* sn_hold_buff_pos: amount of the buffer used; where to place the next chunk
* sn_hold_buff_temp: used when we have to increase the size of the buffer
*
* Controlling the placing of the key,value in the database:
* ok_to_put: means we are ready to place the key,value in the database, i.e., we have the full value
* (either of the spanning node or a regular node).
* putting_a_sn: we are placing a spanning node in the database, i.e, use the key from sn_gvkey and
* the value from sn_hold_buff.
*/
CHECK_HIDDEN_SUBSCRIPT(gv_currkey,is_hidden_subscript);
if (!is_hidden_subscript && (max_subsc_len < (gv_currkey->end + 1)))
max_subsc_len = gv_currkey->end + 1;
v.str.addr = (char*)cp1;
v.str.len =INTCAST(rec_len - (cp1 - (unsigned char *)rp));
if (expected_sn_chunk_number && !is_hidden_subscript)
{ /* we were expecting a chunk of an spanning node and we did not get one */
DISPLAY_INCMP_SN_MSG;
util_out_print("!_!_Expected chunk number : !UL but found a non-spanning node", TRUE,
expected_sn_chunk_number + 1);
if (sn_hold_buff_pos)
DISPLAY_PARTIAL_SN_HOLD_BUFF;
KILL_INCMP_SN_IF_NEEDED;
sn_hold_buff_pos = 0;
expected_sn_chunk_number = 0;
ok_to_put = TRUE;
putting_a_sn = FALSE;
numsubs = 0;
}
if (is_hidden_subscript)
{ /* it's a chunk and we were expecting one */
sn_chunk_number = SPAN_GVSUBS2INT((span_subs *) &(gv_currkey->base[gv_currkey->end - 4]));
if (!expected_sn_chunk_number && is_hidden_subscript && sn_chunk_number)
{ /* we not expecting a payload chunk (as opposed to a control record) but we got one */
DISPLAY_INCMP_SN_MSG;
util_out_print("!_!_Not expecting a spanning node chunk but found chunk : !UL", TRUE,
sn_chunk_number + 1);
if (v.str.len)
DISPLAY_VALUE("!_!_Errant Chunk :");
continue;
}
if (0 == sn_chunk_number)
{ /* first spanning node chunk, get ctrl info */
if (0 != expected_sn_chunk_number)
{
DISPLAY_INCMP_SN_MSG;
util_out_print("!_!_Expected chunk number : !UL but found chunk number : !UL", TRUE,
expected_sn_chunk_number + 1, sn_chunk_number + 1);
if (sn_hold_buff_pos)
DISPLAY_PARTIAL_SN_HOLD_BUFF;
KILL_INCMP_SN_IF_NEEDED;
}
/* start building a new spanning node */
sn_gvkey->end = gv_currkey->end - (SPAN_SUBS_LEN + 1);
memcpy(sn_gvkey->base, gv_currkey->base, sn_gvkey->end);
sn_gvkey->base[sn_gvkey->end] = 0;
sn_gvkey->prev = gv_currkey->prev;
sn_gvkey->top = gv_currkey->top;
GET_NSBCTRL(v.str.addr, numsubs, gblsize);
/* look for first payload chunk */
expected_sn_chunk_number = 1;
sn_hold_buff_pos = 0;
ok_to_put = FALSE;
sn_incmp_gbl_already_killed = FALSE;
} else
{ /* we only need to compare the key before the hidden subscripts */
if ((expected_sn_chunk_number == sn_chunk_number)
&& (sn_gvkey->end == gv_currkey->end - (SPAN_SUBS_LEN + 1))
&& !memcmp(sn_gvkey->base,gv_currkey->base, sn_gvkey->end)
&& ((sn_hold_buff_pos + v.str.len) <= gblsize))
{
if (NULL == sn_hold_buff)
{
sn_hold_buff_size = DEFAULT_SN_HOLD_BUFF_SIZE;
sn_hold_buff = (char *)malloc(DEFAULT_SN_HOLD_BUFF_SIZE);
}
if ((sn_hold_buff_pos + v.str.len) > sn_hold_buff_size)
{
sn_hold_buff_size = sn_hold_buff_size * 2;
sn_hold_buff_temp = (char *)malloc(sn_hold_buff_size);
memcpy(sn_hold_buff_temp, sn_hold_buff, sn_hold_buff_pos);
free (sn_hold_buff);
sn_hold_buff = sn_hold_buff_temp;
}
memcpy(sn_hold_buff + sn_hold_buff_pos, v.str.addr, v.str.len);
sn_hold_buff_pos += v.str.len;
if (expected_sn_chunk_number == numsubs)
{
if (sn_hold_buff_pos != gblsize)
{ /* we don't have the expected size even though */
/* we have all the expected chunks. */
DISPLAY_INCMP_SN_MSG;
util_out_print("!_!_Expected size : !UL actual size : !UL", TRUE,
gblsize, sn_hold_buff_pos);
if (sn_hold_buff_pos)
DISPLAY_PARTIAL_SN_HOLD_BUFF;
KILL_INCMP_SN_IF_NEEDED;
expected_sn_chunk_number = 0;
ok_to_put = FALSE;
sn_hold_buff_pos = 0;
}
else
{
expected_sn_chunk_number = 0;
ok_to_put = TRUE;
putting_a_sn = TRUE;
}
}else
expected_sn_chunk_number++;
}else
{
DISPLAY_INCMP_SN_MSG;
if ((sn_hold_buff_pos + v.str.len) <= gblsize)
util_out_print("!_!_Expected chunk number : !UL but found chunk number : !UL", /*BYPASSOK*/
TRUE, expected_sn_chunk_number + 1, sn_chunk_number + 1);
else
util_out_print("!_!_Global value too large: expected size : !UL actual size : !UL chunk number : !UL", TRUE, /*BYPASSOK*/
gblsize, sn_hold_buff_pos + v.str.len, sn_chunk_number + 1);
if (sn_hold_buff_pos)
DISPLAY_PARTIAL_SN_HOLD_BUFF;
if (v.str.len)
DISPLAY_VALUE("!_!_Errant Chunk :");
KILL_INCMP_SN_IF_NEEDED;
sn_hold_buff_pos = 0;
expected_sn_chunk_number = 0;
}
}
} else
ok_to_put = TRUE;
if (ok_to_put)
{
if (putting_a_sn)
{
gv_currkey->base[gv_currkey->end - (SPAN_SUBS_LEN + 1)] = 0;
gv_currkey->end -= (SPAN_SUBS_LEN + 1);
v.str.addr = sn_hold_buff;
v.str.len = sn_hold_buff_pos;
}
if (max_data_len < v.str.len)
max_data_len = v.str.len;
bin_call_db(BIN_PUT, (INTPTR_T)&v, 0);
if (mupip_error_occurred)
{
if (!mupip_DB_full)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
file_offset = file_offset_base + ((unsigned char *)rp - ptr_base);
util_out_print("!_!_at File offset : [0x!XL]", TRUE, file_offset);
DISPLAY_CURRKEY;
DISPLAY_VALUE("!_!_Value :");
}
break;
}
if (putting_a_sn)
putting_a_sn = FALSE;
else
{
key_count++;
global_key_count++;
}
}
}
}
GTMCRYPT_ONLY(
if (NULL != hash_array)
free(hash_array);
)
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;)
void trigger_delete_all(char *trigger_rec, uint4 len, uint4 *trig_stats)
{
int count;
sgmnt_addrs *csa;
mval curr_gbl_name;
int cycle;
mval *mv_count_ptr;
mval *mv_cycle_ptr;
gd_region *reg, *reg_top;
int4 result;
gd_region *lgtrig_reg;
mval trigger_cycle;
mval trigger_count;
boolean_t this_db_updated;
uint4 triggers_deleted;
mval trigjrec;
boolean_t jnl_format_done;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(0 < dollar_tlevel);
jnl_format_done = FALSE;
lgtrig_reg = NULL;
trigjrec.mvtype = MV_STR;
trigjrec.str.len = len;
trigjrec.str.addr = trigger_rec;
triggers_deleted = 0;
for (reg = gd_header->regions, reg_top = reg + gd_header->n_regions; reg < reg_top; reg++)
{
GVTR_SWITCH_REG_AND_HASHT_BIND_NAME(reg);
csa = cs_addrs;
if (NULL == csa) /* not BG or MM access method */
continue;
/* gv_target now points to ^#t in region "reg" */
/* To write the LGTRIG logical jnl record, choose some region that has journaling enabled */
if (!reg->read_only && !jnl_format_done && JNL_WRITE_LOGICAL_RECS(csa))
lgtrig_reg = reg;
if (!gv_target->root)
continue;
/* kill ^#t("#TNAME") */
BUILD_HASHT_SUB_CURRKEY(LITERAL_HASHTNAME, STRLEN(LITERAL_HASHTNAME));
if (0 != gvcst_data())
{ /* Issue error if we dont have permissions to touch ^#t global */
if (reg->read_only)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_TRIGMODREGNOTRW, 2, REG_LEN_STR(reg));
gvcst_kill(TRUE);
}
/* Kill all descendents of ^#t(trigvn, ...) where trigvn is any global with a trigger,
* but skip the ^#t("#...",...) entries. Setup ^#t("$") as the key for op_gvorder
*/
BUILD_HASHT_SUB_CURRKEY(LITERAL_MAXHASHVAL, STRLEN(LITERAL_MAXHASHVAL));
TREF(gv_last_subsc_null) = FALSE; /* We know its not null, but prior state is unreliable */
this_db_updated = FALSE;
while (TRUE)
{
op_gvorder(&curr_gbl_name);
/* quit:$length(curr_gbl_name)=0 */
if (0 == curr_gbl_name.str.len)
break;
/* $get(^#t(curr_gbl_name,#COUNT)) */
BUILD_HASHT_SUB_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len,
LITERAL_HASHCOUNT, STRLEN(LITERAL_HASHCOUNT));
if (gvcst_get(&trigger_count))
{
/* Now that we know there is something to kill, check if we have permissions to touch ^#t global */
if (reg->read_only)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_TRIGMODREGNOTRW, 2, REG_LEN_STR(reg));
mv_count_ptr = &trigger_count;
count = MV_FORCE_UINT(mv_count_ptr);
/* $get(^#t(curr_gbl_name,#CYCLE)) */
BUILD_HASHT_SUB_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len,
LITERAL_HASHCYCLE, STRLEN(LITERAL_HASHCYCLE));
if (!gvcst_get(&trigger_cycle))
{ /* Found #COUNT, there must be #CYCLE */
if (CDB_STAGNATE > t_tries)
t_retry(cdb_sc_triggermod);
assert(WBTEST_HELPOUT_TRIGDEFBAD == gtm_white_box_test_case_number);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_TRIGDEFBAD, 6,
curr_gbl_name.str.len, curr_gbl_name.str.addr,
curr_gbl_name.str.len, curr_gbl_name.str.addr, LEN_AND_LIT("\"#CYCLE\""),
ERR_TEXT, 2, RTS_ERROR_TEXT("#CYCLE field is missing"));
}
mv_cycle_ptr = &trigger_cycle;
cycle = MV_FORCE_UINT(mv_cycle_ptr);
if (!jnl_format_done && JNL_WRITE_LOGICAL_RECS(csa))
{
jnl_format(JNL_LGTRIG, NULL, &trigjrec, 0);
jnl_format_done = TRUE;
}
/* kill ^#t(curr_gbl_name) */
BUILD_HASHT_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len);
gvcst_kill(TRUE);
/* Note : ^#t(curr_gbl_name,"#TRHASH") is also killed as part of the above */
cycle++;
MV_FORCE_MVAL(&trigger_cycle, cycle);
/* set ^#t(curr_gbl_name,#CYCLE)=trigger_cycle */
SET_TRIGGER_GLOBAL_SUB_SUB_MVAL(curr_gbl_name.str.addr, curr_gbl_name.str.len,
LITERAL_HASHCYCLE, STRLEN(LITERAL_HASHCYCLE), trigger_cycle, result);
assert(PUT_SUCCESS == result);
this_db_updated = TRUE;
triggers_deleted += count;
} /* else there is no #COUNT, then no triggers, leave #CYCLE alone */
/* get ready for op_gvorder() call for next trigger under ^#t */
BUILD_HASHT_SUB_CURRKEY(curr_gbl_name.str.addr, curr_gbl_name.str.len);
}
if (this_db_updated)
{
csa->incr_db_trigger_cycle = TRUE;
if (dollar_ztrigger_invoked)
{ /* increment db_dztrigger_cycle so that next gvcst_put/gvcst_kill in this transaction,
* on this region, will re-read. See trigger_update.c for a comment on why it is okay
* for db_dztrigger_cycle to be incremented more than once in the same transaction
*/
csa->db_dztrigger_cycle++;
}
}
}
if (!jnl_format_done && (NULL != lgtrig_reg))
{ /* There was no journaled region that had a ^#t update, but found at least one journaled region
* so write a LGTRIG logical jnl record there.
*/
GVTR_SWITCH_REG_AND_HASHT_BIND_NAME(lgtrig_reg);
csa = cs_addrs;
JNLPOOL_INIT_IF_NEEDED(csa, csa->hdr, csa->nl); /* see previous usage for comment on why it is needed */
assert(dollar_tlevel);
T_BEGIN_SETORKILL_NONTP_OR_TP(ERR_TRIGLOADFAIL); /* needed to set update_trans TRUE on this region
* even if NO db updates happen to ^#t nodes. */
jnl_format(JNL_LGTRIG, NULL, &trigjrec, 0);
jnl_format_done = TRUE;
}
if (triggers_deleted)
{
util_out_print_gtmio("All existing triggers (count = !UL) deleted", FLUSH, triggers_deleted);
trig_stats[STATS_DELETED] += triggers_deleted;
trig_stats[STATS_NOERROR_TRIGFILE]++;
} else
{
util_out_print_gtmio("No matching triggers found for deletion", FLUSH);
trig_stats[STATS_UNCHANGED_TRIGFILE]++;
}
}
void gds_rundown(void)
{
bool is_mm, we_are_last_user, we_are_last_writer;
boolean_t ipc_deleted, remove_shm, cancelled_timer, cancelled_dbsync_timer, vermismatch;
now_t now; /* for GET_CUR_TIME macro */
char *time_ptr, time_str[CTIME_BEFORE_NL + 2]; /* for GET_CUR_TIME macro */
gd_region *reg;
int save_errno, status;
int4 semval, ftok_semval, sopcnt, ftok_sopcnt;
short crash_count;
sm_long_t munmap_len;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
struct shmid_ds shm_buf;
struct sembuf sop[2], ftok_sop[2];
uint4 jnl_status;
unix_db_info *udi;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
error_def(ERR_CRITSEMFAIL);
error_def(ERR_DBCCERR);
error_def(ERR_DBFILERR);
error_def(ERR_DBRNDWNWRN);
error_def(ERR_ERRCALL);
error_def(ERR_GBLOFLOW);
error_def(ERR_GTMASSERT);
error_def(ERR_IPCNOTDEL);
error_def(ERR_JNLFLUSH);
error_def(ERR_RNDWNSEMFAIL);
error_def(ERR_TEXT);
error_def(ERR_WCBLOCKED);
forced_exit = FALSE; /* Okay, we're dying already -- let rel_crit live in peace now.
* If coming through a DAL, not necessarily dying. what to do then? -- nars -- 8/15/2001
*/
grabbed_access_sem = FALSE;
jnl_status = 0;
reg = gv_cur_region; /* Local copy */
/*
* early out for cluster regions
* to avoid tripping the assert below.
* Note:
* This early out is consistent with VMS. It has been
* noted that all of the gtcm assignments
* to gv_cur_region should use the TP_CHANGE_REG
* macro. This would also avoid the assert problem
* and should be done eventually.
*/
if (dba_cm == reg->dyn.addr->acc_meth)
return;
udi = FILE_INFO(reg);
csa = &udi->s_addrs;
csd = csa->hdr;
assert(csa == cs_addrs && csd == cs_data);
if ((reg->open) && (dba_usr == csd->acc_meth))
{
change_reg();
gvusr_rundown();
return;
}
ESTABLISH(gds_rundown_ch);
if (!reg->open) /* Not open, no point to rundown */
{
if (reg->opening) /* Died partway open, kill rest of way */
{
rel_crit(reg);
mutex_cleanup(reg);
/* revist this to handle MM properly SMW 98/12/16
if (NULL != csa->nl)
{
status = shmdt((caddr_t)csa->nl);
if (-1 == status)
send_msg(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error during shmdt"), errno);
}
*/
shmdt((caddr_t)csa->nl);
csa->nl = NULL;
}
REVERT;
return;
}
switch(csd->acc_meth)
{ /* Pass mm and bg through */
case dba_bg:
is_mm = FALSE;
break;
case dba_mm:
is_mm = TRUE;
break;
case dba_usr:
assert(FALSE);
default:
REVERT;
return;
}
/* Cancel any pending flush timer for this region by this task */
CANCEL_DB_TIMERS(reg, cancelled_timer, cancelled_dbsync_timer);
we_are_last_user = FALSE;
if (!csa->persistent_freeze)
region_freeze(reg, FALSE, FALSE, FALSE);
assert(!csa->read_lock);
rel_crit(reg); /* get locks to known state */
mutex_cleanup(reg);
/*
* We need to guarantee that none else access database file header when semid/shmid fields are reset.
* We already have created ftok semaphore in db_init or, mu_rndwn_file and did not remove it.
* So just lock it. We do it in blocking mode.
*/
if (!ftok_sem_lock(reg, FALSE, FALSE))
rts_error(VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
/*
* For mupip_jnl_recover we already have database access control semaphore.
* We do not release it. We release it from mur_close_files.
*/
if (!mupip_jnl_recover)
{
sop[0].sem_num = 0; sop[0].sem_op = 0; /* Wait for 0 */
sop[1].sem_num = 0; sop[1].sem_op = 1; /* Lock */
sopcnt = 2;
sop[0].sem_flg = sop[1].sem_flg = SEM_UNDO | IPC_NOWAIT; /* Don't wait the first time thru */
SEMOP(udi->semid, sop, sopcnt, status);
if (-1 == status) /* We couldn't get it in one shot -- see if we already have it */
{
save_errno = errno;
/* see comment about Linux specific difference in behaviour of semctl() with GETPID in gds_rundown_ch() */
if (semctl(udi->semid, 0, GETPID) == process_id)
{
send_msg(VARLSTCNT(5) MAKE_MSG_INFO(ERR_CRITSEMFAIL), 2,
DB_LEN_STR(reg),
ERR_RNDWNSEMFAIL);
REVERT;
return; /* Already in rundown for this region */
}
if (EAGAIN != save_errno)
{
assert(FALSE);
rts_error(VARLSTCNT(9) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("gds_rundown first semop/semctl"), save_errno);
}
sop[0].sem_flg = sop[1].sem_flg = SEM_UNDO; /* Try again - blocking this time */
SEMOP(udi->semid, sop, 2, status);
if (-1 == status) /* We couldn't get it at all.. */
rts_error(VARLSTCNT(5) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), errno);
}
}
grabbed_access_sem = TRUE;
/*
* We now have the dbinit/rundown lock, so we are alone in this code for this region
* and nobody else can attach.
* See if we are all alone in accessing this database shared memory.
*/
assert(csa->ref_cnt); /* decrement private ref_cnt before shared ref_cnt decrement. */
csa->ref_cnt--; /* Currently journaling logic in gds_rundown() in VMS relies on this order to detect last writer */
assert(!csa->ref_cnt);
--csa->nl->ref_cnt;
if (memcmp(csa->nl->now_running, gtm_release_name, gtm_release_name_len + 1))
{ /* VERMISMATCH condition. Possible only if DSE */
assert(dse_running);
vermismatch = TRUE;
} else
vermismatch = FALSE;
if (-1 == shmctl(udi->shmid, IPC_STAT, &shm_buf))
{
save_errno = errno;
rts_error(VARLSTCNT(9) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("gds_rundown shmctl"), save_errno);
} else
we_are_last_user = (1 == shm_buf.shm_nattch) && !vermismatch;
assert(!mupip_jnl_recover || we_are_last_user); /* recover => one user */
if (-1 == (semval = semctl(udi->semid, 1, GETVAL)))
rts_error(VARLSTCNT(5) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), errno);
we_are_last_writer = (1 == semval) && (FALSE == reg->read_only) && !vermismatch;/* There's one writer left and I am it */
assert(!(mupip_jnl_recover && !reg->read_only) || we_are_last_writer); /* recover + R/W region => one writer */
if (-1 == (ftok_semval = semctl(udi->ftok_semid, 1, GETVAL)))
rts_error(VARLSTCNT(5) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), errno);
/* If csa->nl->donotflush_dbjnl is set, it means mupip recover/rollback was interrupted and therefore we should
* not flush shared memory contents to disk as they might be in an inconsistent state.
* In this case, we will go ahead and remove shared memory (without flushing the contents) in this routine.
* A reissue of the recover/rollback command will restore the database to a consistent state.
* Otherwise, if we have write access to this region, let us perform a few writing tasks.
*/
if (csa->nl->donotflush_dbjnl)
csa->wbuf_dqd = 0; /* ignore csa->wbuf_dqd status as we do not care about the cache contents */
else if (!reg->read_only && !vermismatch)
{ /* If we had an orphaned block and were interrupted, set wc_blocked so we can invoke wcs_recover */
if (csa->wbuf_dqd)
{
grab_crit(reg);
SET_TRACEABLE_VAR(csd->wc_blocked, TRUE);
BG_TRACE_PRO_ANY(csa, wcb_gds_rundown);
send_msg(VARLSTCNT(8) ERR_WCBLOCKED, 6, LEN_AND_LIT("wcb_gds_rundown"),
process_id, &csa->ti->curr_tn, DB_LEN_STR(reg));
csa->wbuf_dqd = 0;
wcs_recover(reg);
if (is_mm)
{
assert(FALSE);
csd = csa->hdr;
}
BG_TRACE_PRO_ANY(csa, lost_block_recovery);
rel_crit(reg);
}
if (JNL_ENABLED(csd) && (GTCM_GNP_SERVER_IMAGE == image_type))
originator_prc_vec = NULL;
/* If we are the last writing user, then everything must be flushed */
if (we_are_last_writer)
{ /* Time to flush out all of our buffers */
if (is_mm)
{
if (csa->total_blks != csa->ti->total_blks) /* do remap if file had been extended */
{
grab_crit(reg);
wcs_mm_recover(reg);
csd = csa->hdr;
rel_crit(reg);
}
csa->nl->remove_shm = TRUE;
}
/* Note WCSFLU_SYNC_EPOCH ensures the epoch is synced to the journal and indirectly
* also ensures that the db is fsynced. We don't want to use it in the calls to
* wcs_flu() from t_end() and tp_tend() since we can defer it to out-of-crit there.
* In this case, since we are running down, we don't have any such option.
*/
csa->nl->remove_shm = wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SYNC_EPOCH);
/* Since we_are_last_writer, we should be guaranteed that wcs_flu() did not change csd, (in
* case of MM for potential file extension), even if it did a grab_crit(). Therefore, make
* sure that's true.
*/
assert(csd == csa->hdr);
assert(0 == memcmp(csd->label, GDS_LABEL, GDS_LABEL_SZ - 1));
csd->trans_hist.header_open_tn = csd->trans_hist.curr_tn;
} else if ((cancelled_timer && (0 > csa->nl->wcs_timers)) || cancelled_dbsync_timer)
{ /* cancelled pending db or jnl flush timers - flush database and journal buffers to disk */
grab_crit(reg);
/* we need to sync the epoch as the fact that there is no active pending flush timer implies
* there will be noone else who will flush the dirty buffers and EPOCH to disk in a timely fashion
*/
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SYNC_EPOCH);
rel_crit(reg);
assert((dba_mm == cs_data->acc_meth) || (csd == cs_data));
csd = cs_data; /* In case this is MM and wcs_flu() remapped an extended database, reset csd */
}
/* Do rundown journal processing after buffer flushes since they require jnl to be open */
if (JNL_ENABLED(csd))
{ /* the following tp_change_reg() is not needed due to the assert csa == cs_addrs at the beginning
* of gds_rundown(), but just to be safe. To be removed by 2002!! --- nars -- 2001/04/25.
*/
tp_change_reg(); /* call this because jnl_ensure_open checks cs_addrs rather than gv_cur_region */
jpc = csa->jnl;
jbp = jpc->jnl_buff;
if (jbp->fsync_in_prog_latch.u.parts.latch_pid == process_id)
{
assert(FALSE);
COMPSWAP_UNLOCK(&jbp->fsync_in_prog_latch, process_id, 0, LOCK_AVAILABLE, 0);
}
if (jbp->io_in_prog_latch.u.parts.latch_pid == process_id)
{
assert(FALSE);
COMPSWAP_UNLOCK(&jbp->io_in_prog_latch, process_id, 0, LOCK_AVAILABLE, 0);
}
if (((NOJNL != jpc->channel) && !JNL_FILE_SWITCHED(jpc))
|| we_are_last_writer && (0 != csa->nl->jnl_file.u.inode))
{ /* We need to close the journal file cleanly if we have the latest generation journal file open
* or if we are the last writer and the journal file is open in shared memory (not necessarily
* by ourselves e.g. the only process that opened the journal got shot abnormally)
* Note: we should not infer anything from the shared memory value of csa->nl->jnl_file.u.inode
* if we are not the last writer as it can be concurrently updated.
*/
grab_crit(reg);
if (JNL_ENABLED(csd))
{
SET_GBL_JREC_TIME; /* jnl_ensure_open/jnl_put_jrt_pini/pfin/jnl_file_close all need it */
/* 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 (0 == jnl_status)
{ /* If we_are_last_writer, we would have already done a wcs_flu() which would
* have written an epoch record and we are guaranteed no further updates
* since we are the last writer. So, just close the journal.
* Although we assert pini_addr should be non-zero for last_writer, we
* play it safe in PRO and write a PINI record if not written already.
*/
assert(!jbp->before_images || is_mm
|| !we_are_last_writer || 0 != jpc->pini_addr);
if (we_are_last_writer && 0 == jpc->pini_addr)
jnl_put_jrt_pini(csa);
if (0 != jpc->pini_addr)
jnl_put_jrt_pfin(csa);
/* If not the last writer and no pending flush timer left, do jnl flush now */
if (!we_are_last_writer && (0 > csa->nl->wcs_timers))
{
if (SS_NORMAL == (jnl_status = jnl_flush(reg)))
{
assert(jbp->freeaddr == jbp->dskaddr);
jnl_fsync(reg, jbp->dskaddr);
assert(jbp->fsync_dskaddr == jbp->dskaddr);
} else
{
send_msg(VARLSTCNT(9) ERR_JNLFLUSH, 2, JNL_LEN_STR(csd),
ERR_TEXT, 2,
RTS_ERROR_TEXT("Error with journal flush in gds_rundown"),
jnl_status);
assert(NOJNL == jpc->channel);/* jnl file lost has been triggered */
/* In this routine, all code that follows from here on does not
* assume anything about the journaling characteristics of this
* database so it is safe to continue execution even though
* journaling got closed in the middle.
*/
}
}
jnl_file_close(reg, we_are_last_writer, FALSE);
} else
send_msg(VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd), DB_LEN_STR(reg));
}
rel_crit(reg);
}
}
if (we_are_last_writer) /* Flush the fileheader last and harden the file to disk */
{
grab_crit(reg); /* To satisfy crit requirement in fileheader_sync() */
memset(csd->machine_name, 0, MAX_MCNAMELEN); /* clear the machine_name field */
if (!mupip_jnl_recover && we_are_last_user)
{ /* mupip_jnl_recover will do this after mur_close_file */
csd->semid = INVALID_SEMID;
csd->shmid = INVALID_SHMID;
csd->gt_sem_ctime.ctime = 0;
csd->gt_shm_ctime.ctime = 0;
}
fileheader_sync(reg);
rel_crit(reg);
if (FALSE == is_mm)
{
if (-1 == fsync(udi->fd)) /* Sync it all */
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error during file sync at close"), errno);
}
} else
{ /* Now do final MM file sync before exit */
#if !defined(TARGETED_MSYNC) && !defined(NO_MSYNC)
if (-1 == fsync(udi->fd)) /* Sync it all */
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error during file sync at close"), errno);
}
#else
if (-1 == msync((caddr_t)csa->db_addrs[0], (size_t)(csa->db_addrs[1] - csa->db_addrs[0]), MS_SYNC))
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error during file msync at close"), errno);
}
#endif
}
}
} /* end if (!reg->read_only && !csa->nl->donotflush_dbjnl) */
if (reg->read_only && we_are_last_user && !mupip_jnl_recover)
{ /* mupip_jnl_recover will do this after mur_close_file */
db_ipcs.semid = INVALID_SEMID;
db_ipcs.shmid = INVALID_SHMID;
db_ipcs.gt_sem_ctime = 0;
db_ipcs.gt_shm_ctime = 0;
db_ipcs.fn_len = reg->dyn.addr->fname_len;
memcpy(db_ipcs.fn, reg->dyn.addr->fname, reg->dyn.addr->fname_len);
db_ipcs.fn[reg->dyn.addr->fname_len] = 0;
/* request gtmsecshr to flush. read_only cannot flush itself */
if (0 != send_mesg2gtmsecshr(FLUSH_DB_IPCS_INFO, 0, (char *)NULL, 0))
rts_error(VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("gtmsecshr failed to update database file header"));
}
/* Done with file now, close it */
if (-1 == close(udi->fd))
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error during file close"), errno);
}
/* Unmap storage if mm mode but only the part that is not the fileheader (so shows up in dumps) */
if (is_mm)
{
munmap_len = (sm_long_t)((csa->db_addrs[1] - csa->db_addrs[0]) - ROUND_UP(SIZEOF_FILE_HDR(csa->hdr),
MSYNC_ADDR_INCS));
if (munmap_len > 0)
{
munmap((caddr_t)(csa->db_addrs[0] + ROUND_UP(SIZEOF_FILE_HDR(csa->hdr), MSYNC_ADDR_INCS)),
(size_t)(munmap_len));
#ifdef DEBUG_DB64
rel_mmseg((caddr_t)csa->db_addrs[0]);
#endif
}
}
/* Detach our shared memory while still under lock so reference counts will be
* correct for the next process to run down this region.
* In the process also get the remove_shm status from node_local before detaching.
* If csa->nl->donotflush_dbjnl is TRUE, it means we can safely remove shared memory without compromising data
* integrity as a reissue of recover will restore the database to a consistent state.
*/
remove_shm = !vermismatch && (csa->nl->remove_shm || csa->nl->donotflush_dbjnl);
status = shmdt((caddr_t)csa->nl);
csa->nl = NULL; /* dereferencing nl after detach is not right, so we set it to NULL so that we can test before dereference*/
if (-1 == status)
send_msg(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg), ERR_TEXT, 2, LEN_AND_LIT("Error during shmdt"), errno);
reg->open = FALSE;
/* If file is still not in good shape, die here and now before we get rid of our storage */
if (csa->wbuf_dqd)
GTMASSERT;
ipc_deleted = FALSE;
/* If we are the very last user, remove shared storage id and the semaphores */
if (we_are_last_user)
{ /* remove shared storage, only if last writer to rundown did a successful wcs_flu() */
assert(!vermismatch);
if (remove_shm)
{
ipc_deleted = TRUE;
if (0 != shm_rmid(udi->shmid))
rts_error(VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Unable to remove shared memory"));
} else if (is_src_server || is_updproc)
{
gtm_putmsg(VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
send_msg(VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
} else
send_msg(VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
/*
* Don't release semaphore in case of mupip recover/rollback; since it has standalone access.
* It will release the semaphore in mur_close_files.
*/
if (!mupip_jnl_recover)
{
if (0 != sem_rmid(udi->semid))
rts_error(VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Unable to remove semaphore"));
grabbed_access_sem = FALSE;
}
} else
{
assert(!mupip_jnl_recover);
/* If we were writing, get rid of our writer access count semaphore */
if (!reg->read_only)
if (0 != (save_errno = do_semop(udi->semid, 1, -1, SEM_UNDO)))
rts_error(VARLSTCNT(9) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("gds_rundown write semaphore release"), save_errno);
/* Now remove the rundown lock */
if (0 != (save_errno = do_semop(udi->semid, 0, -1, SEM_UNDO)))
rts_error(VARLSTCNT(9) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("gds_rundown rundown semaphore release"), save_errno);
grabbed_access_sem = FALSE;
}
if (!ftok_sem_release(reg, !mupip_jnl_recover, FALSE))
rts_error(VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
if (!ipc_deleted)
{
GET_CUR_TIME;
if (is_src_server)
gtm_putmsg(VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_ptr,
LEN_AND_LIT("Source server"), REG_LEN_STR(reg));
if (is_updproc)
gtm_putmsg(VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_ptr,
LEN_AND_LIT("Update process"), REG_LEN_STR(reg));
if (mupip_jnl_recover)
{
gtm_putmsg(VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_ptr,
LEN_AND_LIT("Mupip journal process"), REG_LEN_STR(reg));
send_msg(VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_ptr,
LEN_AND_LIT("Mupip journal process"), REG_LEN_STR(reg));
}
}
REVERT;
}
/* Takes an entry from 'ipcs -m' and checks for its validity to be a GT.M db segment.
* Returns TRUE if the shared memory segment is a valid GT.M db segment
* (based on a check on some fields in the shared memory) else FALSE.
* If the segment belongs to GT.M it returns the database file name by the second argument.
* Sets exit_stat to ERR_MUNOTALLSEC if appropriate.
*/
boolean_t validate_db_shm_entry(shm_parms *parm_buff, char *fname, int *exit_stat)
{
boolean_t remove_shmid;
file_control *fc;
int fname_len, save_errno, status, shmid;
node_local_ptr_t nl_addr;
sm_uc_ptr_t start_addr;
struct stat st_buff;
struct shmid_ds shmstat;
sgmnt_data tsd;
unix_db_info *udi;
char msgbuff[OUT_BUFF_SIZE];
if (NULL == parm_buff)
return FALSE;
/* check for the bare minimum size of the shared memory segment that we expect
* (with no fileheader related information at hand) */
if (MIN_NODE_LOCAL_SPACE + SHMPOOL_SECTION_SIZE > parm_buff->sgmnt_siz)
return FALSE;
if (IPC_PRIVATE != parm_buff->key)
return FALSE;
shmid = parm_buff->shmid;
/* we do not need to lock the shm for reading the rundown information as
* the other rundowns (if any) can also be allowed to share reading the
* same info concurrently.
*/
if (-1 == (sm_long_t)(start_addr = (sm_uc_ptr_t) do_shmat(shmid, 0, SHM_RND)))
return FALSE;
nl_addr = (node_local_ptr_t)start_addr;
memcpy(fname, nl_addr->fname, MAX_FN_LEN + 1);
fname[MAX_FN_LEN] = '\0'; /* make sure the fname is null terminated */
fname_len = STRLEN(fname);
msgbuff[0] = '\0';
if (memcmp(nl_addr->label, GDS_LABEL, GDS_LABEL_SZ - 1))
{
if (!memcmp(nl_addr->label, GDS_LABEL, GDS_LABEL_SZ - 3))
{
util_out_print("Cannot rundown shmid = !UL for database !AD as it has format !AD "
"but this mupip uses format !AD", TRUE, shmid,
fname_len, fname, GDS_LABEL_SZ - 1, nl_addr->label, GDS_LABEL_SZ - 1, GDS_LABEL);
*exit_stat = ERR_MUNOTALLSEC;
}
shmdt((void *)start_addr);
return FALSE;
}
if (-1 == shmctl(shmid, IPC_STAT, &shmstat))
{
save_errno = errno;
assert(FALSE);/* we were able to attach to this shmid before so should be able to get stats on it */
util_out_print("!AD -> Error with shmctl for shmid = !UL",
TRUE, fname_len, fname, shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = FALSE;
/* Check if db filename reported in shared memory still exists. If not, clean this shared memory section
* without even invoking "mu_rndwn_file" as that expects the db file to exist. Same case if shared memory
* points back to a database whose file header does not have this shmid.
*/
if (-1 == Stat(fname, &st_buff))
{
if (ENOENT == errno)
{
SNPRINTF(msgbuff, OUT_BUFF_SIZE, "File %s does not exist", fname);
if (1 < shmstat.shm_nattch)
{
PRINT_AND_SEND_DBRNDWN_FAILURE_MSG(msgbuff, fname, shmid);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = TRUE;
} else
{ /* Stat errored out e.g. due to file permissions. Log that */
save_errno = errno;
util_out_print("Cannot rundown shmid !UL for database file !AD as stat() on the file"
" returned the following error", TRUE, shmid, fname_len, fname);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
} else
{
mu_gv_cur_reg_init();
gv_cur_region->dyn.addr->fname_len = strlen(fname);
STRNCPY_STR(gv_cur_region->dyn.addr->fname, fname, gv_cur_region->dyn.addr->fname_len);
fc = gv_cur_region->dyn.addr->file_cntl;
fc->op = FC_OPEN;
status = dbfilop(fc);
if (SS_NORMAL != status)
{
util_out_print("!AD -> Error with dbfilop for shmid = !UL", TRUE, fname_len, fname, shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(5) status, 2, DB_LEN_STR(gv_cur_region), errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
udi = FILE_INFO(gv_cur_region);
LSEEKREAD(udi->fd, 0, &tsd, SIZEOF(sgmnt_data), status);
if (0 != status)
{
save_errno = errno;
util_out_print("!AD -> Error with LSEEKREAD for shmid = !UL", TRUE, fname_len, fname, shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
mu_gv_cur_reg_free();
if (tsd.shmid != shmid)
{
SNPRINTF(msgbuff, OUT_BUFF_SIZE, "Shared memory ID (%d) in the DB file header does not match with the one"
" reported by \"ipcs\" command (%d)", tsd.shmid, shmid);
if (1 < shmstat.shm_nattch)
{
PRINT_AND_SEND_DBRNDWN_FAILURE_MSG(msgbuff, fname, shmid);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = TRUE;
} else if (tsd.gt_shm_ctime.ctime != shmstat.shm_ctime)
{
SNPRINTF(msgbuff, OUT_BUFF_SIZE, "Shared memory creation time in the DB file header does not match with"
" the one reported by shmctl");
if (1 < shmstat.shm_nattch)
{
PRINT_AND_SEND_DBRNDWN_FAILURE_MSG(msgbuff, fname, shmid);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = TRUE;
}
}
shmdt((void *)start_addr);
if (remove_shmid)
{
assert('\0' != msgbuff[0]);
if (0 != shm_rmid(shmid))
{
save_errno = errno;
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(8) ERR_DBFILERR, 2, fname_len, fname,
ERR_TEXT, 2, RTS_ERROR_TEXT("Error removing shared memory"));
util_out_print("!AD -> Error removing shared memory for shmid = !UL", TRUE, fname_len, fname, shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
return FALSE;
}
PRINT_AND_SEND_SHMREMOVED_MSG(msgbuff, fname_len, fname, shmid);
*exit_stat = ERR_SHMREMOVED;
} else
*exit_stat = SS_NORMAL;
return TRUE;
}
void jnl_fsync(gd_region *reg, uint4 fsync_addr)
{
jnl_private_control *jpc;
jnl_buffer_ptr_t jb;
uint4 lcnt, saved_dsk_addr, saved_status;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
int4 lck_state;
int fsync_ret, save_errno;
error_def(ERR_JNLFSYNCERR);
error_def(ERR_FSYNCTIMOUT);
error_def(ERR_TEXT);
error_def(ERR_JNLFRCDTERM);
error_def(ERR_JNLFSYNCLSTCK);
csa = &FILE_INFO(reg)->s_addrs;
jpc = csa->jnl;
jb = jpc->jnl_buff;
if ((NOJNL != jpc->channel) && !JNL_FILE_SWITCHED(jpc))
{
csd = csa->hdr;
for (lcnt = 1; fsync_addr > jb->fsync_dskaddr && !JNL_FILE_SWITCHED(jpc); lcnt++)
{
if (MAX_FSYNC_WAIT_CNT / 2 == lcnt) /* half way into max.patience*/
{
saved_status = jpc->status;
jpc->status = SS_NORMAL;
jnl_send_oper(jpc, ERR_JNLFSYNCLSTCK);
jpc->status = saved_status ;
}
if (MAX_FSYNC_WAIT_CNT == lcnt) /* tried a long */
{
saved_status = jpc->status;
jpc->status = SS_NORMAL;
jnl_send_oper(jpc, ERR_JNLFSYNCLSTCK);
jpc->status = saved_status ;
send_msg(VARLSTCNT(4) ERR_FSYNCTIMOUT, 2, JNL_LEN_STR(csd));
GTMASSERT;
}
BG_TRACE_PRO_ANY(csa, n_jnl_fsync_tries);
if (GET_SWAPLOCK(&jb->fsync_in_prog_latch))
break;
wcs_sleep(lcnt);
performCASLatchCheck(&jb->fsync_in_prog_latch, lcnt);
}
if (fsync_addr > jb->fsync_dskaddr && !JNL_FILE_SWITCHED(jpc))
{
assert(process_id == jb->fsync_in_prog_latch.u.parts.latch_pid); /* assert we have the lock */
saved_dsk_addr = jb->dskaddr;
if (jpc->sync_io)
{
/* We need to maintain the fsync control fields irrespective of the type of IO, because we might
* switch between these at any time.
*/
jb->fsync_dskaddr = saved_dsk_addr;
} else
{
GTM_FSYNC(jpc->channel, fsync_ret);
if (-1 == fsync_ret)
{
save_errno = errno;
assert(FALSE);
send_msg(VARLSTCNT(9) ERR_JNLFSYNCERR, 2, JNL_LEN_STR(csd),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error with fsync"), save_errno);
rts_error(VARLSTCNT(9) ERR_JNLFSYNCERR, 2, JNL_LEN_STR(csd),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error with fsync"), save_errno);
} else
{
jb->fsync_dskaddr = saved_dsk_addr;
BG_TRACE_PRO_ANY(csa, n_jnl_fsyncs);
}
}
}
if (process_id == jb->fsync_in_prog_latch.u.parts.latch_pid)
RELEASE_SWAPLOCK(&jb->fsync_in_prog_latch);
}
return;
}
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() */
int4 gds_rundown(void)
{
boolean_t canceled_dbsync_timer, canceled_flush_timer, ok_to_write_pfin;
boolean_t have_standalone_access, ipc_deleted, err_caught;
boolean_t is_cur_process_ss_initiator, remove_shm, vermismatch, we_are_last_user, we_are_last_writer, is_mm;
boolean_t unsafe_last_writer;
char time_str[CTIME_BEFORE_NL + 2]; /* for GET_CUR_TIME macro */
gd_region *reg;
int save_errno, status, rc;
int4 semval, ftok_semval, sopcnt, ftok_sopcnt;
short crash_count;
sm_long_t munmap_len;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
node_local_ptr_t cnl;
struct shmid_ds shm_buf;
struct sembuf sop[2], ftok_sop[2];
uint4 jnl_status;
unix_db_info *udi;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
shm_snapshot_t *ss_shm_ptr;
uint4 ss_pid, onln_rlbk_pid, holder_pid;
boolean_t was_crit;
boolean_t safe_mode; /* Do not flush or take down shared memory. */
boolean_t bypassed_ftok = FALSE, bypassed_access = FALSE, may_bypass_ftok, inst_is_frozen,
ftok_counter_halted,
access_counter_halted;
int secshrstat;
intrpt_state_t prev_intrpt_state;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
jnl_status = 0;
reg = gv_cur_region; /* Local copy */
/* early out for cluster regions
* to avoid tripping the assert below.
* Note:
* This early out is consistent with VMS. It has been
* noted that all of the gtcm assignments
* to gv_cur_region should use the TP_CHANGE_REG
* macro. This would also avoid the assert problem
* and should be done eventually.
*/
if (dba_cm == reg->dyn.addr->acc_meth)
return EXIT_NRM;
udi = FILE_INFO(reg);
csa = &udi->s_addrs;
csd = csa->hdr;
assert(csa == cs_addrs && csd == cs_data);
if ((reg->open) && (dba_usr == csd->acc_meth))
{
change_reg();
gvusr_rundown();
return EXIT_NRM;
}
/* If the process has standalone access, it has udi->grabbed_access_sem set to TRUE at this point. Note that down in a local
* variable as the udi->grabbed_access_sem is set to TRUE even for non-standalone access below and hence we can't rely on
* that later to determine if the process had standalone access or not when it entered this function. We need to guarantee
* that none else access database file header when semid/shmid fields are reset. We already have created ftok semaphore in
* db_init or, mu_rndwn_file and did not remove it. So just lock it. We do it in blocking mode.
*/
have_standalone_access = udi->grabbed_access_sem; /* process holds standalone access */
DEFER_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
ESTABLISH_NORET(gds_rundown_ch, err_caught);
if (err_caught)
{
REVERT;
WITH_CH(gds_rundown_ch, gds_rundown_err_cleanup(have_standalone_access), 0);
ENABLE_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
DEBUG_ONLY(ok_to_UNWIND_in_exit_handling = FALSE);
return EXIT_ERR;
}
assert(reg->open); /* if we failed to open, dbinit_ch should have taken care of proper clean up */
assert(!reg->opening); /* see comment above */
assert((dba_bg == csd->acc_meth) || (dba_mm == csd->acc_meth));
is_mm = (dba_bg != csd->acc_meth);
assert(!csa->hold_onto_crit || (csa->now_crit && jgbl.onlnrlbk));
/* If we are online rollback, we should already be holding crit and should release it only at the end of this module. This
* is usually done by noting down csa->now_crit in a local variable (was_crit) and using it whenever we are about to
* grab_crit. But, there are instances (like mupip_set_journal.c) where we grab_crit but invoke gds_rundown without any
* preceeding rel_crit. Such code relies on the fact that gds_rundown does rel_crit unconditionally (to get locks to a known
* state). So, augment csa->now_crit with jgbl.onlnrlbk to track if we can rel_crit unconditionally or not in gds_rundown.
*/
was_crit = (csa->now_crit && jgbl.onlnrlbk);
/* Cancel any pending flush timer for this region by this task */
canceled_flush_timer = FALSE;
canceled_dbsync_timer = FALSE;
CANCEL_DB_TIMERS(reg, csa, canceled_flush_timer, canceled_dbsync_timer);
we_are_last_user = FALSE;
inst_is_frozen = IS_REPL_INST_FROZEN && REPL_ALLOWED(csa->hdr);
if (!csa->persistent_freeze)
region_freeze(reg, FALSE, FALSE, FALSE);
if (!was_crit)
{
rel_crit(reg); /* get locks to known state */
mutex_cleanup(reg);
}
/* The only process that can invoke gds_rundown while holding access control semaphore is RECOVER/ROLLBACK. All the others
* (like MUPIP SET -FILE/MUPIP EXTEND would have invoked db_ipcs_reset() before invoking gds_rundown (from
* mupip_exit_handler). The only exception is when these processes encounter a terminate signal and they reach
* mupip_exit_handler while holding access control semaphore. Assert accordingly.
*/
assert(!have_standalone_access || mupip_jnl_recover || process_exiting);
/* If we have standalone access, then ensure that a concurrent online rollback cannot be running at the same time as it
* needs the access control lock as well. The only expection is we are online rollback and currently running down.
*/
cnl = csa->nl;
onln_rlbk_pid = cnl->onln_rlbk_pid;
assert(!have_standalone_access || mupip_jnl_recover || !onln_rlbk_pid || !is_proc_alive(onln_rlbk_pid, 0));
if (!have_standalone_access)
{
if (-1 == (ftok_semval = semctl(udi->ftok_semid, DB_COUNTER_SEM, GETVAL))) /* Check # of procs counted on FTOK */
{
save_errno = errno;
assert(FALSE);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get ftok_semval"), CALLFROM, errno);
}
may_bypass_ftok = CAN_BYPASS(ftok_semval, csd, inst_is_frozen); /* Do we need a blocking wait? */
/* We need to guarantee that no one else access database file header when semid/shmid fields are reset.
* We already have created ftok semaphore in db_init or mu_rndwn_file and did not remove it. So just lock it.
*/
if (!ftok_sem_lock(reg, may_bypass_ftok))
{
if (may_bypass_ftok)
{ /* We did a non-blocking wait. It's ok to proceed without locking */
bypassed_ftok = TRUE;
holder_pid = semctl(udi->ftok_semid, DB_CONTROL_SEM, GETPID);
if ((uint4)-1 == holder_pid)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get holder_pid"),
CALLFROM, errno);
if (!IS_GTM_IMAGE) /* MUMPS processes should not flood syslog with bypass messages. */
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_RESRCINTRLCKBYPAS, 10,
LEN_AND_STR(gtmImageNames[image_type].imageName), process_id, LEN_AND_LIT("FTOK"),
REG_LEN_STR(reg), DB_LEN_STR(reg), holder_pid);
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_TEXT, 2,
LEN_AND_LIT("FTOK bypassed at rundown"));
}
} else
{ /* We did a blocking wait but something bad happened. */
FTOK_TRACE(csa, csa->ti->curr_tn, ftok_ops_lock, process_id);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
}
}
sop[0].sem_num = DB_CONTROL_SEM; sop[0].sem_op = 0; /* Wait for 0 */
sop[1].sem_num = DB_CONTROL_SEM; sop[1].sem_op = 1; /* Lock */
sopcnt = 2;
sop[0].sem_flg = sop[1].sem_flg = SEM_UNDO | IPC_NOWAIT; /* Don't wait the first time thru */
SEMOP(udi->semid, sop, sopcnt, status, NO_WAIT);
if (0 != status)
{
save_errno = errno;
/* Check # of processes counted on access sem. */
if (-1 == (semval = semctl(udi->semid, DB_COUNTER_SEM, GETVAL)))
{
assert(FALSE);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get semval"), CALLFROM, errno);
}
bypassed_access = CAN_BYPASS(semval, csd, inst_is_frozen) || onln_rlbk_pid || csd->file_corrupt;
/* Before attempting again in the blocking mode, see if the holding process is an online rollback.
* If so, it is likely we won't get the access control semaphore anytime soon. In that case, we
* are better off skipping rundown and continuing with sanity cleanup and exit.
*/
holder_pid = semctl(udi->semid, DB_CONTROL_SEM, GETPID);
if ((uint4)-1 == holder_pid)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get holder_pid"), CALLFROM, errno);
if (!bypassed_access)
{ /* We couldn't get it in one shot-- see if we already have it */
if (holder_pid == process_id)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(5) MAKE_MSG_INFO(ERR_CRITSEMFAIL), 2, DB_LEN_STR(reg),
ERR_RNDWNSEMFAIL);
REVERT;
ENABLE_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
assert(FALSE);
return EXIT_ERR;
}
if (EAGAIN != save_errno)
{
assert(FALSE);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMOP on access control semaphore"),
CALLFROM, save_errno);
}
sop[0].sem_flg = sop[1].sem_flg = SEM_UNDO; /* Try again - blocking this time */
SEMOP(udi->semid, sop, 2, status, FORCED_WAIT);
if (-1 == status) /* We couldn't get it at all.. */
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMOP on access control semaphore"),
CALLFROM, errno);
} else if (!IS_GTM_IMAGE)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_RESRCINTRLCKBYPAS, 10,
LEN_AND_STR(gtmImageNames[image_type].imageName), process_id,
LEN_AND_LIT("access control"), REG_LEN_STR(reg), DB_LEN_STR(reg), holder_pid);
send_msg_csa(CSA_ARG(NULL) VARLSTCNT(4) ERR_TEXT, 2,
LEN_AND_LIT("Access control bypassed at rundown"));
}
udi->grabbed_access_sem = !bypassed_access;
}
} /* else we we hold the access control semaphore and therefore have standalone access. We do not release it now - we
* release it later in mupip_exit_handler.c. Since we already hold the access control semaphore, we don't need the
* ftok semaphore and trying it could cause deadlock
*/
/* Note that in the case of online rollback, "udi->grabbed_access_sem" (and in turn "have_standalone_access") is TRUE.
* But there could be other processes still having the database open so we cannot safely reset the halted fields.
*/
if (have_standalone_access && !jgbl.onlnrlbk)
csd->ftok_counter_halted = csd->access_counter_halted = FALSE;
ftok_counter_halted = csd->ftok_counter_halted;
access_counter_halted = csd->access_counter_halted;
/* If we bypassed any of the semaphores, activate safe mode.
* Also, if the replication instance is frozen and this db has replication turned on (which means
* no flushes of dirty buffers to this db can happen while the instance is frozen) activate safe mode.
*/
ok_to_write_pfin = !(bypassed_access || bypassed_ftok || inst_is_frozen);
safe_mode = !ok_to_write_pfin || ftok_counter_halted || access_counter_halted;
/* At this point we are guaranteed no one else is doing a db_init/rundown as we hold the access control semaphore */
assert(csa->ref_cnt); /* decrement private ref_cnt before shared ref_cnt decrement. */
csa->ref_cnt--; /* Currently journaling logic in gds_rundown() in VMS relies on this order to detect last writer */
assert(!csa->ref_cnt);
--cnl->ref_cnt;
if (memcmp(cnl->now_running, gtm_release_name, gtm_release_name_len + 1))
{ /* VERMISMATCH condition. Possible only if DSE */
assert(dse_running);
vermismatch = TRUE;
} else
vermismatch = FALSE;
if (-1 == shmctl(udi->shmid, IPC_STAT, &shm_buf))
{
save_errno = errno;
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown shmctl"), CALLFROM, save_errno);
} else
we_are_last_user = (1 == shm_buf.shm_nattch) && !vermismatch && !safe_mode;
/* recover => one user except ONLINE ROLLBACK, or standalone with frozen instance */
assert(!have_standalone_access || we_are_last_user || jgbl.onlnrlbk || inst_is_frozen);
if (-1 == (semval = semctl(udi->semid, DB_COUNTER_SEM, GETVAL)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get semval"), CALLFROM, errno);
/* There's one writer left and I am it */
assert(reg->read_only || semval >= 0);
unsafe_last_writer = (DB_COUNTER_SEM_INCR == semval) && (FALSE == reg->read_only) && !vermismatch;
we_are_last_writer = unsafe_last_writer && !safe_mode;
assert(!we_are_last_writer || !safe_mode);
assert(!we_are_last_user || !safe_mode);
/* recover + R/W region => one writer except ONLINE ROLLBACK, or standalone with frozen instance, leading to safe_mode */
assert(!(have_standalone_access && !reg->read_only) || we_are_last_writer || jgbl.onlnrlbk || inst_is_frozen);
GTM_WHITE_BOX_TEST(WBTEST_ANTIFREEZE_JNLCLOSE, we_are_last_writer, 1); /* Assume we are the last writer to invoke wcs_flu */
if (!have_standalone_access && (-1 == (ftok_semval = semctl(udi->ftok_semid, DB_COUNTER_SEM, GETVAL))))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg), ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown SEMCTL failed to get ftok_semval"), CALLFROM, errno);
if (NULL != csa->ss_ctx)
ss_destroy_context(csa->ss_ctx);
/* SS_MULTI: If multiple snapshots are supported, then we have to run through each of the snapshots */
assert(1 == MAX_SNAPSHOTS);
ss_shm_ptr = (shm_snapshot_ptr_t)SS_GETSTARTPTR(csa);
ss_pid = ss_shm_ptr->ss_info.ss_pid;
is_cur_process_ss_initiator = (process_id == ss_pid);
if (ss_pid && (is_cur_process_ss_initiator || we_are_last_user))
{
/* Try getting snapshot crit latch. If we don't get latch, we won't hang for eternity and will skip
* doing the orphaned snapshot cleanup. It will be cleaned up eventually either by subsequent MUPIP
* INTEG or by a MUPIP RUNDOWN.
*/
if (ss_get_lock_nowait(reg) && (ss_pid == ss_shm_ptr->ss_info.ss_pid)
&& (is_cur_process_ss_initiator || !is_proc_alive(ss_pid, 0)))
{
ss_release(NULL);
ss_release_lock(reg);
}
}
/* If cnl->donotflush_dbjnl is set, it means mupip recover/rollback was interrupted and therefore we need not flush
* shared memory contents to disk as they might be in an inconsistent state. Moreover, any more flushing will only cause
* future rollback to undo more journal records (PBLKs). In this case, we will go ahead and remove shared memory (without
* flushing the contents) in this routine. A reissue of the recover/rollback command will restore the database to a
* consistent state.
*/
if (!cnl->donotflush_dbjnl && !reg->read_only && !vermismatch)
{ /* If we had an orphaned block and were interrupted, set wc_blocked so we can invoke wcs_recover. Do it ONLY
* if there is NO concurrent online rollback running (as we need crit to set wc_blocked)
*/
if (csa->wbuf_dqd && !is_mm)
{ /* If we had an orphaned block and were interrupted, mupip_exit_handler will invoke secshr_db_clnup which
* will clear this field and so we should never come to gds_rundown with a non-zero wbuf_dqd. The only
* exception is if we are recover/rollback in which case gds_rundown (from mur_close_files) is invoked
* BEFORE secshr_db_clnup in mur_close_files.
* Note: It is NOT possible for online rollback to reach here with wbuf_dqd being non-zero. This is because
* the moment we apply the first PBLK, we stop all interrupts and hence can never be interrupted in
* wcs_wtstart or wcs_get_space. Assert accordingly.
*/
assert(mupip_jnl_recover && !jgbl.onlnrlbk && !safe_mode);
if (!was_crit)
grab_crit(reg);
SET_TRACEABLE_VAR(cnl->wc_blocked, TRUE);
BG_TRACE_PRO_ANY(csa, wcb_gds_rundown);
send_msg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_WCBLOCKED, 6, LEN_AND_LIT("wcb_gds_rundown"),
process_id, &csa->ti->curr_tn, DB_LEN_STR(reg));
csa->wbuf_dqd = 0;
wcs_recover(reg);
BG_TRACE_PRO_ANY(csa, lost_block_recovery);
if (!was_crit)
rel_crit(reg);
}
if (JNL_ENABLED(csd) && IS_GTCM_GNP_SERVER_IMAGE)
originator_prc_vec = NULL;
/* If we are the last writing user, then everything must be flushed */
if (we_are_last_writer)
{ /* Time to flush out all of our buffers */
assert(!safe_mode);
if (is_mm)
{
MM_DBFILEXT_REMAP_IF_NEEDED(csa, reg);
cnl->remove_shm = TRUE;
}
if (cnl->wc_blocked && jgbl.onlnrlbk)
{ /* if the last update done by online rollback was not committed in the normal code-path but was
* completed by secshr_db_clnup, wc_blocked will be set to TRUE. But, since online rollback never
* invokes grab_crit (since csa->hold_onto_crit is set to TRUE), wcs_recover is never invoked. This
* could result in the last update never getting flushed to the disk and if online rollback happened
* to be the last writer then the shared memory will be flushed and removed and the last update will
* be lost. So, force wcs_recover if we find ourselves in such a situation. But, wc_blocked is
* possible only if phase1 or phase2 errors are induced using white box test cases
*/
assert(WB_COMMIT_ERR_ENABLED);
wcs_recover(reg);
}
/* Note WCSFLU_SYNC_EPOCH ensures the epoch is synced to the journal and indirectly
* also ensures that the db is fsynced. We don't want to use it in the calls to
* wcs_flu() from t_end() and tp_tend() since we can defer it to out-of-crit there.
* In this case, since we are running down, we don't have any such option.
*/
cnl->remove_shm = wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SYNC_EPOCH);
/* Since we_are_last_writer, we should be guaranteed that wcs_flu() did not change csd, (in
* case of MM for potential file extension), even if it did a grab_crit(). Therefore, make
* sure that's true.
*/
assert(csd == csa->hdr);
assert(0 == memcmp(csd->label, GDS_LABEL, GDS_LABEL_SZ - 1));
} else if (((canceled_flush_timer && (0 > cnl->wcs_timers)) || canceled_dbsync_timer) && !inst_is_frozen)
{ /* canceled pending db or jnl flush timers - flush database and journal buffers to disk */
if (!was_crit)
grab_crit(reg);
/* we need to sync the epoch as the fact that there is no active pending flush timer implies
* there will be noone else who will flush the dirty buffers and EPOCH to disk in a timely fashion
*/
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_WRITE_EPOCH | WCSFLU_SYNC_EPOCH);
if (!was_crit)
rel_crit(reg);
assert((dba_mm == cs_data->acc_meth) || (csd == cs_data));
csd = cs_data; /* In case this is MM and wcs_flu() remapped an extended database, reset csd */
}
/* Do rundown journal processing after buffer flushes since they require jnl to be open */
if (JNL_ENABLED(csd))
{ /* the following tp_change_reg() is not needed due to the assert csa == cs_addrs at the beginning
* of gds_rundown(), but just to be safe. To be removed by 2002!! --- nars -- 2001/04/25.
*/
tp_change_reg(); /* call this because jnl_ensure_open checks cs_addrs rather than gv_cur_region */
jpc = csa->jnl;
jbp = jpc->jnl_buff;
if (jbp->fsync_in_prog_latch.u.parts.latch_pid == process_id)
{
assert(FALSE);
COMPSWAP_UNLOCK(&jbp->fsync_in_prog_latch, process_id, 0, LOCK_AVAILABLE, 0);
}
if (jbp->io_in_prog_latch.u.parts.latch_pid == process_id)
{
assert(FALSE);
COMPSWAP_UNLOCK(&jbp->io_in_prog_latch, process_id, 0, LOCK_AVAILABLE, 0);
}
if ((((NOJNL != jpc->channel) && !JNL_FILE_SWITCHED(jpc))
|| we_are_last_writer && (0 != cnl->jnl_file.u.inode)) && ok_to_write_pfin)
{ /* We need to close the journal file cleanly if we have the latest generation journal file open
* or if we are the last writer and the journal file is open in shared memory (not necessarily
* by ourselves e.g. the only process that opened the journal got shot abnormally)
* Note: we should not infer anything from the shared memory value of cnl->jnl_file.u.inode
* if we are not the last writer as it can be concurrently updated.
*/
if (!was_crit)
grab_crit(reg);
if (JNL_ENABLED(csd))
{
SET_GBL_JREC_TIME; /* jnl_ensure_open/jnl_put_jrt_pini/pfin/jnl_file_close all need it */
/* 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 (0 == jnl_status)
{ /* If we_are_last_writer, we would have already done a wcs_flu() which would
* have written an epoch record and we are guaranteed no further updates
* since we are the last writer. So, just close the journal.
* If the freeaddr == post_epoch_freeaddr, wcs_flu may have skipped writing
* a pini, so allow for that.
*/
assert(!jbp->before_images || is_mm
|| !we_are_last_writer || (0 != jpc->pini_addr) || jgbl.mur_extract
|| (jpc->jnl_buff->freeaddr == jpc->jnl_buff->post_epoch_freeaddr));
/* If we haven't written a pini, let jnl_file_close write the pini/pfin. */
if (!jgbl.mur_extract && (0 != jpc->pini_addr))
jnl_put_jrt_pfin(csa);
/* If not the last writer and no pending flush timer left, do jnl flush now */
if (!we_are_last_writer && (0 > cnl->wcs_timers))
{
if (SS_NORMAL == (jnl_status = jnl_flush(reg)))
{
assert(jbp->freeaddr == jbp->dskaddr);
jnl_fsync(reg, jbp->dskaddr);
assert(jbp->fsync_dskaddr == jbp->dskaddr);
} else
{
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 in gds_rundown"),
jnl_status);
assert(NOJNL == jpc->channel);/* jnl file lost has been triggered */
/* In this routine, all code that follows from here on does not
* assume anything about the journaling characteristics of this
* database so it is safe to continue execution even though
* journaling got closed in the middle.
*/
}
}
jnl_file_close(reg, we_are_last_writer, FALSE);
} else
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) jnl_status, 4, JNL_LEN_STR(csd),
DB_LEN_STR(reg));
}
if (!was_crit)
rel_crit(reg);
}
}
if (we_are_last_writer) /* Flush the fileheader last and harden the file to disk */
{
if (!was_crit)
grab_crit(reg); /* To satisfy crit requirement in fileheader_sync() */
memset(csd->machine_name, 0, MAX_MCNAMELEN); /* clear the machine_name field */
if (!have_standalone_access && we_are_last_user)
{ /* mupip_exit_handler will do this after mur_close_file */
csd->semid = INVALID_SEMID;
csd->shmid = INVALID_SHMID;
csd->gt_sem_ctime.ctime = 0;
csd->gt_shm_ctime.ctime = 0;
}
fileheader_sync(reg);
if (!was_crit)
rel_crit(reg);
if (!is_mm)
{
GTM_DB_FSYNC(csa, udi->fd, rc); /* Sync it all */
if (-1 == rc)
{
rts_error_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error during file sync at close"), errno);
}
} else
{ /* Now do final MM file sync before exit */
assert(csa->ti->total_blks == csa->total_blks);
#ifdef _AIX
GTM_DB_FSYNC(csa, udi->fd, rc);
if (-1 == rc)
#else
if (-1 == MSYNC((caddr_t)csa->db_addrs[0], (caddr_t)csa->db_addrs[1]))
#endif
{
rts_error_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Error during file sync at close"), errno);
}
}
} else if (unsafe_last_writer && !cnl->lastwriterbypas_msg_issued)
{
send_msg_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_LASTWRITERBYPAS, 2, DB_LEN_STR(reg));
cnl->lastwriterbypas_msg_issued = TRUE;
}
} /* end if (!reg->read_only && !cnl->donotflush_dbjnl) */
/* We had canceled all db timers at start of rundown. In case as part of rundown (wcs_flu above), we had started
* any timers, cancel them BEFORE setting reg->open to FALSE (assert in wcs_clean_dbsync relies on this).
*/
CANCEL_DB_TIMERS(reg, csa, canceled_flush_timer, canceled_dbsync_timer);
if (reg->read_only && we_are_last_user && !have_standalone_access && cnl->remove_shm)
{ /* mupip_exit_handler will do this after mur_close_file */
db_ipcs.semid = INVALID_SEMID;
db_ipcs.shmid = INVALID_SHMID;
db_ipcs.gt_sem_ctime = 0;
db_ipcs.gt_shm_ctime = 0;
db_ipcs.fn_len = reg->dyn.addr->fname_len;
memcpy(db_ipcs.fn, reg->dyn.addr->fname, reg->dyn.addr->fname_len);
db_ipcs.fn[reg->dyn.addr->fname_len] = 0;
/* request gtmsecshr to flush. read_only cannot flush itself */
WAIT_FOR_REPL_INST_UNFREEZE_SAFE(csa);
if (!csa->read_only_fs)
{
secshrstat = send_mesg2gtmsecshr(FLUSH_DB_IPCS_INFO, 0, (char *)NULL, 0);
if (0 != secshrstat)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("gtmsecshr failed to update database file header"));
}
}
/* Done with file now, close it */
CLOSEFILE_RESET(udi->fd, rc); /* resets "udi->fd" to FD_INVALID */
if (-1 == rc)
{
rts_error_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error during file close"), errno);
}
/* Unmap storage if mm mode but only the part that is not the fileheader (so shows up in dumps) */
# if !defined(_AIX)
if (is_mm && (NULL != csa->db_addrs[0]))
{
assert(csa->db_addrs[1] > csa->db_addrs[0]);
munmap_len = (sm_long_t)(csa->db_addrs[1] - csa->db_addrs[0]);
if (0 < munmap_len)
munmap((caddr_t)(csa->db_addrs[0]), (size_t)(munmap_len));
}
# endif
/* Detach our shared memory while still under lock so reference counts will be correct for the next process to run down
* this region. In the process also get the remove_shm status from node_local before detaching.
* If cnl->donotflush_dbjnl is TRUE, it means we can safely remove shared memory without compromising data
* integrity as a reissue of recover will restore the database to a consistent state.
*/
remove_shm = !vermismatch && (cnl->remove_shm || cnl->donotflush_dbjnl);
/* We are done with online rollback on this region. Indicate to other processes by setting the onln_rlbk_pid to 0.
* Do it before releasing crit (t_end relies on this ordering when accessing cnl->onln_rlbk_pid).
*/
if (jgbl.onlnrlbk)
cnl->onln_rlbk_pid = 0;
rel_crit(reg); /* Since we are about to detach from the shared memory, release crit and reset onln_rlbk_pid */
/* If we had skipped flushing journal and database buffers due to a concurrent online rollback, increment the counter
* indicating that in the shared memory so that online rollback can report the # of such processes when it shuts down.
* The same thing is done for both FTOK and access control semaphores when there are too many MUMPS processes.
*/
if (safe_mode) /* indicates flushing was skipped */
{
if (bypassed_access)
cnl->dbrndwn_access_skip++; /* Access semaphore can be bypassed during online rollback */
if (bypassed_ftok)
cnl->dbrndwn_ftok_skip++;
}
if (jgbl.onlnrlbk)
csa->hold_onto_crit = FALSE;
GTM_WHITE_BOX_TEST(WBTEST_HOLD_SEM_BYPASS, cnl->wbox_test_seq_num, 0);
status = shmdt((caddr_t)cnl);
csa->nl = NULL; /* dereferencing nl after detach is not right, so we set it to NULL so that we can test before dereference*/
/* Note that although csa->nl is NULL, we use CSA_ARG(csa) below (not CSA_ARG(NULL)) to be consistent with similar
* usages before csa->nl became NULL. The "is_anticipatory_freeze_needed" function (which is in turn called by the
* CHECK_IF_FREEZE_ON_ERROR_NEEDED macro) does a check of csa->nl before dereferencing shared memory contents so
* we are safe passing "csa".
*/
if (-1 == status)
send_msg_csa(CSA_ARG(csa) VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg), ERR_TEXT, 2,
LEN_AND_LIT("Error during shmdt"), errno);
REMOVE_CSA_FROM_CSADDRSLIST(csa); /* remove "csa" from list of open regions (cs_addrs_list) */
reg->open = FALSE;
/* If file is still not in good shape, die here and now before we get rid of our storage */
assertpro(0 == csa->wbuf_dqd);
ipc_deleted = FALSE;
/* If we are the very last user, remove shared storage id and the semaphores */
if (we_are_last_user)
{ /* remove shared storage, only if last writer to rundown did a successful wcs_flu() */
assert(!vermismatch);
if (remove_shm)
{
ipc_deleted = TRUE;
if (0 != shm_rmid(udi->shmid))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Unable to remove shared memory"));
/* Note that we no longer have a new shared memory. Currently only used/usable for standalone rollback. */
udi->new_shm = FALSE;
/* mupip recover/rollback don't release the semaphore here, but do it later in db_ipcs_reset (invoked from
* mur_close_files())
*/
if (!have_standalone_access)
{
if (0 != sem_rmid(udi->semid))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, RTS_ERROR_TEXT("Unable to remove semaphore"));
udi->new_sem = FALSE; /* Note that we no longer have a new semaphore */
udi->grabbed_access_sem = FALSE;
udi->counter_acc_incremented = FALSE;
}
} else if (is_src_server || is_updproc)
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
} else
send_msg_csa(CSA_ARG(csa) VARLSTCNT(6) ERR_DBRNDWNWRN, 4, DB_LEN_STR(reg), process_id, process_id);
} else
{
assert(!have_standalone_access || jgbl.onlnrlbk || safe_mode);
if (!jgbl.onlnrlbk && !have_standalone_access)
{ /* If we were writing, get rid of our writer access count semaphore */
if (!reg->read_only)
{
if (!access_counter_halted)
{
save_errno = do_semop(udi->semid, DB_COUNTER_SEM, -DB_COUNTER_SEM_INCR, SEM_UNDO);
if (0 != save_errno)
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown access control semaphore decrement"),
CALLFROM, save_errno);
}
udi->counter_acc_incremented = FALSE;
}
assert(safe_mode || !bypassed_access);
/* Now remove the rundown lock */
if (!bypassed_access)
{
if (0 != (save_errno = do_semop(udi->semid, DB_CONTROL_SEM, -1, SEM_UNDO)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(12) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg),
ERR_SYSCALL, 5,
RTS_ERROR_TEXT("gds_rundown access control semaphore release"),
CALLFROM, save_errno);
udi->grabbed_access_sem = FALSE;
}
} /* else access control semaphore will be released in db_ipcs_reset */
}
if (!have_standalone_access)
{
if (bypassed_ftok)
{
if (!ftok_counter_halted)
if (0 != (save_errno = do_semop(udi->ftok_semid, DB_COUNTER_SEM, -DB_COUNTER_SEM_INCR, SEM_UNDO)))
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
} else if (!ftok_sem_release(reg, !ftok_counter_halted, FALSE))
{
FTOK_TRACE(csa, csa->ti->curr_tn, ftok_ops_release, process_id);
rts_error_csa(CSA_ARG(csa) VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
}
udi->grabbed_ftok_sem = FALSE;
udi->counter_ftok_incremented = FALSE;
}
ENABLE_INTERRUPTS(INTRPT_IN_GDS_RUNDOWN, prev_intrpt_state);
if (!ipc_deleted)
{
GET_CUR_TIME(time_str);
if (is_src_server)
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Source server"), REG_LEN_STR(reg));
if (is_updproc)
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Update process"), REG_LEN_STR(reg));
if (mupip_jnl_recover && (!jgbl.onlnrlbk || !we_are_last_user))
{
gtm_putmsg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Mupip journal process"), REG_LEN_STR(reg));
send_msg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_IPCNOTDEL, 6, CTIME_BEFORE_NL, time_str,
LEN_AND_LIT("Mupip journal process"), REG_LEN_STR(reg));
}
}
REVERT;
return EXIT_NRM;
}
void bin_load(uint4 begin, uint4 end)
{
unsigned char *ptr, *cp1, *cp2, *btop, *gvkey_char_ptr, *tmp_ptr, *tmp_key_ptr, *c, *ctop;
unsigned char hdr_lvl, src_buff[MAX_KEY_SZ + 1], dest_buff[MAX_ZWR_KEY_SZ],
cmpc_str[MAX_KEY_SZ + 1], dup_key_str[MAX_KEY_SZ + 1];
unsigned char *end_buff;
unsigned short rec_len, next_cmpc;
int len;
int current, last, length, max_blk_siz, max_key, status;
uint4 iter, max_data_len, max_subsc_len, key_count;
ssize_t rec_count, global_key_count, subsc_len,extr_std_null_coll;
boolean_t need_xlation, new_gvn, utf8_extract;
rec_hdr *rp, *next_rp;
mval v, tmp_mval;
mstr mstr_src, mstr_dest;
collseq *extr_collseq, *db_collseq, *save_gv_target_collseq;
coll_hdr extr_collhdr, db_collhdr;
gv_key *tmp_gvkey = NULL; /* null-initialize at start, will be malloced later */
char std_null_coll[BIN_HEADER_NUMSZ + 1];
# ifdef GTM_CRYPT
gtmcrypt_key_t *encr_key_handles;
char *inbuf;
int4 index;
int req_dec_blk_size, init_status, crypt_status;
muext_hash_hdr_ptr_t hash_array = NULL;
# endif
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(4 == SIZEOF(coll_hdr));
gvinit();
v.mvtype = MV_STR;
len = file_input_bin_get((char **)&ptr);
hdr_lvl = EXTR_HEADER_LEVEL(ptr);
if (!(((('4' == hdr_lvl) || ('5' == hdr_lvl)) && (BIN_HEADER_SZ == len)) || (('4' > hdr_lvl) && (V3_BIN_HEADER_SZ == len))))
{
rts_error(VARLSTCNT(1) ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
/* expecting the level in a single character */
assert(' ' == *(ptr + SIZEOF(BIN_HEADER_LABEL) - 3));
if (0 != memcmp(ptr, BIN_HEADER_LABEL, SIZEOF(BIN_HEADER_LABEL) - 2) || ('2' > hdr_lvl) || *(BIN_HEADER_VERSION) < hdr_lvl)
{ /* ignore the level check */
rts_error(VARLSTCNT(1) ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
/* check if extract was generated in UTF-8 mode */
utf8_extract = (0 == MEMCMP_LIT(&ptr[len - BIN_HEADER_LABELSZ], UTF8_NAME)) ? TRUE : FALSE;
if ((utf8_extract && !gtm_utf8_mode) || (!utf8_extract && gtm_utf8_mode))
{ /* extract CHSET doesn't match $ZCHSET */
if (utf8_extract)
rts_error(VARLSTCNT(4) ERR_LOADINVCHSET, 2, LEN_AND_LIT("UTF-8"));
else
rts_error(VARLSTCNT(4) ERR_LOADINVCHSET, 2, LEN_AND_LIT("M"));
mupip_exit(ERR_LDBINFMT);
}
if ('4' >= hdr_lvl)
{ /* Binary extracts in V50000-to-V52000 (label=4) and pre-V50000 (label=3) could have a '\0' byte (NULL byte)
* in the middle of the string. Replace it with ' ' (space) like it would be in V52000 binary extracts and above.
*/
for (c = ptr, ctop = c + len; c < ctop; c++)
{
if ('\0' == *c)
*c = ' ';
}
}
util_out_print("Label = !AD\n", TRUE, len, ptr);
new_gvn = FALSE;
if (hdr_lvl > '3')
{
memcpy(std_null_coll, ptr + BIN_HEADER_NULLCOLLOFFSET, BIN_HEADER_NUMSZ);
std_null_coll[BIN_HEADER_NUMSZ] = '\0';
extr_std_null_coll = STRTOUL(std_null_coll, NULL, 10);
if (0 != extr_std_null_coll && 1!= extr_std_null_coll)
{
rts_error(VARLSTCNT(5) ERR_TEXT, 2, RTS_ERROR_TEXT("Corrupted null collation field in header"),
ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
} else
extr_std_null_coll = 0;
# ifdef GTM_CRYPT
if ('5' <= hdr_lvl)
{
int i, num_indexes;
len = file_input_bin_get((char **)&ptr);
hash_array = (muext_hash_hdr *)malloc(len);
/* store hashes of all the files used during extract into muext_hash_hdr structure */
memcpy((char *)hash_array, ptr, len);
num_indexes = len / GTMCRYPT_HASH_LEN;
encr_key_handles = (gtmcrypt_key_t *)malloc(SIZEOF(gtmcrypt_key_t) * num_indexes);
INIT_PROC_ENCRYPTION(crypt_status);
GC_BIN_LOAD_ERR(crypt_status);
for (index = 0; index < num_indexes; index++)
{
if (0 == memcmp(hash_array[index].gtmcrypt_hash, EMPTY_GTMCRYPT_HASH, GTMCRYPT_HASH_LEN))
continue;
GTMCRYPT_GETKEY(hash_array[index].gtmcrypt_hash, encr_key_handles[index], crypt_status);
GC_BIN_LOAD_ERR(crypt_status);
}
}
# endif
if ('2' < hdr_lvl)
{
len = file_input_bin_get((char **)&ptr);
if (SIZEOF(coll_hdr) != len)
{
rts_error(VARLSTCNT(5) ERR_TEXT, 2, RTS_ERROR_TEXT("Corrupt collation header"), ERR_LDBINFMT);
mupip_exit(ERR_LDBINFMT);
}
extr_collhdr = *((coll_hdr *)(ptr));
new_gvn = TRUE;
} else
gtm_putmsg(VARLSTCNT(3) ERR_OLDBINEXTRACT, 1, hdr_lvl - '0');
if (begin < 2)
begin = 2;
for (iter = 2; iter < begin; iter++)
{
if (!(len = file_input_bin_get((char **)&ptr)))
{
gtm_putmsg(VARLSTCNT(3) ERR_LOADEOF, 1, begin);
util_out_print("Error reading record number: !UL\n", TRUE, iter);
mupip_error_occurred = TRUE;
return;
} else if (len == SIZEOF(coll_hdr))
{
extr_collhdr = *((coll_hdr *)(ptr));
assert(hdr_lvl > '2');
iter--;
}
}
assert(iter == begin);
util_out_print("Beginning LOAD at record number: !UL\n", TRUE, begin);
max_data_len = 0;
max_subsc_len = 0;
global_key_count = key_count = 0;
rec_count = begin - 1;
extr_collseq = db_collseq = NULL;
need_xlation = FALSE;
assert(NULL == tmp_gvkey); /* GVKEY_INIT macro relies on this */
GVKEY_INIT(tmp_gvkey, DBKEYSIZE(MAX_KEY_SZ)); /* tmp_gvkey will point to malloced memory after this */
for (; !mupip_DB_full ;)
{
if (++rec_count > end)
break;
next_cmpc = 0;
mupip_error_occurred = FALSE;
if (mu_ctrly_occurred)
break;
if (mu_ctrlc_occurred)
{
util_out_print("!AD:!_ Key cnt: !UL max subsc len: !UL max data len: !UL", TRUE,
LEN_AND_LIT(gt_lit), key_count, max_subsc_len, max_data_len);
util_out_print("Last LOAD record number: !UL", TRUE, key_count ? (rec_count - 1) : 0);
mu_gvis();
util_out_print(0, TRUE);
mu_ctrlc_occurred = FALSE;
}
/* reset the stringpool for every record in order to avoid garbage collection */
stringpool.free = stringpool.base;
if (!(len = file_input_bin_get((char **)&ptr)) || mupip_error_occurred)
break;
else if (len == SIZEOF(coll_hdr))
{
extr_collhdr = *((coll_hdr *)(ptr));
assert(hdr_lvl > '2');
new_gvn = TRUE; /* next record will contain a new gvn */
rec_count--; /* Decrement as this record does not count as a record for loading purposes */
continue;
}
rp = (rec_hdr*)(ptr);
# ifdef GTM_CRYPT
if ('5' <= hdr_lvl)
{ /* Getting index value from the extracted file. It indicates which database file this record belongs to */
GET_LONG(index, ptr);
if (-1 != index) /* Indicates that the record is encrypted. */
{
req_dec_blk_size = len - SIZEOF(int4);
inbuf = (char *)(ptr + SIZEOF(int4));
GTMCRYPT_DECODE_FAST(encr_key_handles[index], inbuf, req_dec_blk_size, NULL, crypt_status);
GC_BIN_LOAD_ERR(crypt_status);
}
rp = (rec_hdr*)(ptr + SIZEOF(int4));
}
# endif
btop = ptr + len;
cp1 = (unsigned char*)(rp + 1);
v.str.addr = (char*)cp1;
while (*cp1++)
;
v.str.len =INTCAST((char*)cp1 - v.str.addr - 1);
if (('2' >= hdr_lvl) || new_gvn)
{
if ((HASHT_GBLNAME_LEN == v.str.len) && (0 == memcmp(v.str.addr, HASHT_GBLNAME, HASHT_GBLNAME_LEN)))
continue;
bin_call_db(BIN_BIND, (INTPTR_T)gd_header, (INTPTR_T)&v.str);
max_key = gv_cur_region->max_key_size;
db_collhdr.act = gv_target->act;
db_collhdr.ver = gv_target->ver;
db_collhdr.nct = gv_target->nct;
}
GET_USHORT(rec_len, &rp->rsiz);
if (rp->cmpc != 0 || v.str.len > rec_len || mupip_error_occurred)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
util_out_print(0, TRUE);
continue;
}
if (new_gvn)
{
global_key_count = 1;
if ((db_collhdr.act != extr_collhdr.act || db_collhdr.ver != extr_collhdr.ver
|| db_collhdr.nct != extr_collhdr.nct
|| gv_cur_region->std_null_coll != extr_std_null_coll))
{
if (extr_collhdr.act)
{
if (extr_collseq = ready_collseq((int)extr_collhdr.act))
{
if (!do_verify(extr_collseq, extr_collhdr.act, extr_collhdr.ver))
{
gtm_putmsg(VARLSTCNT(8) ERR_COLLTYPVERSION, 2, extr_collhdr.act,
extr_collhdr.ver, ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLTYPVERSION);
}
} else
{
gtm_putmsg(VARLSTCNT(7) ERR_COLLATIONUNDEF, 1, extr_collhdr.act,
ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLATIONUNDEF);
}
}
if (db_collhdr.act)
{
if (db_collseq = ready_collseq((int)db_collhdr.act))
{
if (!do_verify(db_collseq, db_collhdr.act, db_collhdr.ver))
{
gtm_putmsg(VARLSTCNT(8) ERR_COLLTYPVERSION, 2, db_collhdr.act,
db_collhdr.ver, ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLTYPVERSION);
}
} else
{
gtm_putmsg(VARLSTCNT(7) ERR_COLLATIONUNDEF, 1, db_collhdr.act,
ERR_GVIS, 2, gv_altkey->end - 1, gv_altkey->base);
mupip_exit(ERR_COLLATIONUNDEF);
}
}
need_xlation = TRUE;
} else
need_xlation = FALSE;
}
new_gvn = FALSE;
for (; rp < (rec_hdr*)btop; rp = (rec_hdr*)((unsigned char *)rp + rec_len))
{
GET_USHORT(rec_len, &rp->rsiz);
if (rec_len + (unsigned char *)rp > btop)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
util_out_print(0, TRUE);
break;
}
cp1 = (unsigned char*)(rp + 1);
cp2 = gv_currkey->base + rp->cmpc;
current = 1;
for (;;)
{
last = current;
current = *cp2++ = *cp1++;
if (0 == last && 0 == current)
break;
if (cp1 > (unsigned char *)rp + rec_len ||
cp2 > (unsigned char *)gv_currkey + gv_currkey->top)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
util_out_print(0, TRUE);
break;
}
}
if (mupip_error_occurred)
break;
gv_currkey->end = cp2 - gv_currkey->base - 1;
if (need_xlation)
{
assert(hdr_lvl >= '3');
assert(extr_collhdr.act || db_collhdr.act || extr_collhdr.nct || db_collhdr.nct ||
extr_std_null_coll != gv_cur_region->std_null_coll);
/* gv_currkey would have been modified/translated in the earlier put */
memcpy(gv_currkey->base, cmpc_str, next_cmpc);
next_rp = (rec_hdr *)((unsigned char*)rp + rec_len);
if ((unsigned char*)next_rp < btop)
{
next_cmpc = next_rp->cmpc;
assert(next_cmpc <= gv_currkey->end);
memcpy(cmpc_str, gv_currkey->base, next_cmpc);
} else
next_cmpc = 0;
/* length of the key might change (due to nct variation),
* so get a copy of the original key from the extract */
memcpy(dup_key_str, gv_currkey->base, gv_currkey->end + 1);
gvkey_char_ptr = dup_key_str;
while (*gvkey_char_ptr++)
;
gv_currkey->prev = 0;
gv_currkey->end = gvkey_char_ptr - dup_key_str;
assert(gv_keysize <= tmp_gvkey->top);
while (*gvkey_char_ptr)
{
/* get next subscript (in GT.M internal subsc format) */
subsc_len = 0;
tmp_ptr = src_buff;
while (*gvkey_char_ptr)
*tmp_ptr++ = *gvkey_char_ptr++;
subsc_len = tmp_ptr - src_buff;
src_buff[subsc_len] = '\0';
if (extr_collseq)
{
/* undo the extract time collation */
TREF(transform) = TRUE;
save_gv_target_collseq = gv_target->collseq;
gv_target->collseq = extr_collseq;
} else
TREF(transform) = FALSE;
/* convert the subscript to string format */
end_buff = gvsub2str(src_buff, dest_buff, FALSE);
/* transform the string to the current subsc format */
TREF(transform) = TRUE;
tmp_mval.mvtype = MV_STR;
tmp_mval.str.addr = (char *)dest_buff;
tmp_mval.str.len = INTCAST(end_buff - dest_buff);
tmp_gvkey->prev = 0;
tmp_gvkey->end = 0;
if (extr_collseq)
gv_target->collseq = save_gv_target_collseq;
mval2subsc(&tmp_mval, tmp_gvkey);
/* we now have the correctly transformed subscript */
tmp_key_ptr = gv_currkey->base + gv_currkey->end;
memcpy(tmp_key_ptr, tmp_gvkey->base, tmp_gvkey->end + 1);
gv_currkey->prev = gv_currkey->end;
gv_currkey->end += tmp_gvkey->end;
gvkey_char_ptr++;
}
if ( gv_cur_region->std_null_coll != extr_std_null_coll && gv_currkey->prev)
{
if (extr_std_null_coll == 0)
{
GTM2STDNULLCOLL(gv_currkey->base, gv_currkey->end);
} else
{
STD2GTMNULLCOLL(gv_currkey->base, gv_currkey->end);
}
}
}
if (gv_currkey->end >= max_key)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
mu_gvis();
util_out_print(0, TRUE);
continue;
}
if (max_subsc_len < (gv_currkey->end + 1))
max_subsc_len = gv_currkey->end + 1;
v.str.addr = (char*)cp1;
v.str.len =INTCAST(rec_len - (cp1 - (unsigned char *)rp));
if (max_data_len < v.str.len)
max_data_len = v.str.len;
bin_call_db(BIN_PUT, (INTPTR_T)&v, 0);
if (mupip_error_occurred)
{
if (!mupip_DB_full)
{
bin_call_db(ERR_COR, (INTPTR_T)rec_count, (INTPTR_T)global_key_count);
util_out_print(0, TRUE);
}
break;
}
key_count++;
global_key_count++;
}
}
GTMCRYPT_ONLY(
if (NULL != hash_array)
free(hash_array);
)
uint4 mur_process_intrpt_recov()
{
jnl_ctl_list *jctl, *last_jctl;
reg_ctl_list *rctl, *rctl_top;
int rename_fn_len, save_name_len, idx;
char prev_jnl_fn[MAX_FN_LEN + 1], rename_fn[MAX_FN_LEN + 1], save_name[MAX_FN_LEN + 1];
jnl_create_info jnl_info;
uint4 status, status2;
uint4 max_autoswitchlimit, max_jnl_alq, max_jnl_deq, freeblks;
sgmnt_data_ptr_t csd;
jnl_private_control *jpc;
jnl_buffer_ptr_t jbp;
boolean_t jfh_changed;
jnl_record *jnlrec;
jnl_file_header *jfh;
jnl_tm_t now;
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
TP_CHANGE_REG(rctl->gd);
csd = cs_data; /* MM logic after wcs_flu call requires this to be set */
assert(csd == rctl->csa->hdr);
jctl = rctl->jctl_turn_around;
max_jnl_alq = max_jnl_deq = max_autoswitchlimit = 0;
for (last_jctl = NULL ; (NULL != jctl); last_jctl = jctl, jctl = jctl->next_gen)
{
jfh = jctl->jfh;
if (max_autoswitchlimit < jfh->autoswitchlimit)
{ /* Note that max_jnl_alq, max_jnl_deq are not the maximum journal allocation/extensions across
* generations, but rather the allocation/extension corresponding to the maximum autoswitchlimit.
*/
max_autoswitchlimit = jfh->autoswitchlimit;
max_jnl_alq = jfh->jnl_alq;
max_jnl_deq = jfh->jnl_deq;
}
/* Until now, "rctl->blks_to_upgrd_adjust" holds the number of V4 format newly created bitmap blocks
* seen in INCTN records in backward processing. It is possible that backward processing might have
* missed out on seeing those INCTN records which are part of virtually-truncated or completely-rolled-bak
* journal files. The journal file-header has a separate field "prev_recov_blks_to_upgrd_adjust" which
* maintains exactly this count. Therefore adjust the rctl counter accordingly.
*/
assert(!jfh->prev_recov_blks_to_upgrd_adjust || !jfh->recover_interrupted);
assert(!jfh->prev_recov_blks_to_upgrd_adjust || jfh->prev_recov_end_of_data);
rctl->blks_to_upgrd_adjust += jfh->prev_recov_blks_to_upgrd_adjust;
}
if (max_autoswitchlimit > last_jctl->jfh->autoswitchlimit)
{
csd->jnl_alq = max_jnl_alq;
csd->jnl_deq = max_jnl_deq;
csd->autoswitchlimit = max_autoswitchlimit;
} else
{
assert(csd->jnl_alq == last_jctl->jfh->jnl_alq);
assert(csd->jnl_deq == last_jctl->jfh->jnl_deq);
assert(csd->autoswitchlimit == last_jctl->jfh->autoswitchlimit);
}
jctl = rctl->jctl_turn_around;
/* Get a pointer to the turn around point EPOCH record */
jnlrec = rctl->mur_desc->jnlrec;
assert(JRT_EPOCH == jnlrec->prefix.jrec_type);
assert(jctl->turn_around_time == jnlrec->prefix.time);
assert(jctl->turn_around_seqno == jnlrec->jrec_epoch.jnl_seqno);
assert(jctl->turn_around_tn == jnlrec->prefix.tn);
assert(jctl->rec_offset == jctl->turn_around_offset);
/* Reset file-header "blks_to_upgrd" counter to the turn around point epoch value. Adjust this to include
* the number of new V4 format bitmaps created by post-turnaround-point db file extensions.
* The adjustment value is maintained in rctl->blks_to_upgrd_adjust.
*/
csd->blks_to_upgrd = jnlrec->jrec_epoch.blks_to_upgrd;
csd->blks_to_upgrd += rctl->blks_to_upgrd_adjust;
# ifdef GTM_TRIGGER
/* online rollback can potentially take the database to a point in the past where the triggers that were
* previously installed are no longer a part of the current database state and so any process that restarts
* AFTER online rollback completes SHOULD reload triggers and the only way to do that is by incrementing the
* db_trigger_cycle in the file header.
*/
if (jgbl.onlnrlbk && (0 < csd->db_trigger_cycle))
{ /* check for non-zero db_trigger_cycle is to prevent other processes (continuing after online rollback)
* to establish implicit TP (on seeing the trigger cycle mismatch) when there are actually no triggers
* installed in the database (because there were none at the start of online rollback).
*/
csd->db_trigger_cycle++;
if (0 == csd->db_trigger_cycle)
csd->db_trigger_cycle = 1; /* Don't allow cycle set to 0 which means uninitialized */
}
# endif
assert((WBTEST_ALLOW_ARBITRARY_FULLY_UPGRADED == gtm_white_box_test_case_number) ||
(FALSE == jctl->turn_around_fullyupgraded) || (TRUE == jctl->turn_around_fullyupgraded));
/* Set csd->fully_upgraded to FALSE if:
* a) The turn around EPOCH had the fully_upgraded field set to FALSE
* OR
* b) If csd->blks_to_upgrd counter is non-zero. This field can be non-zero even if the turnaround EPOCH's
* fully_upgraded field is TRUE. This is possible if the database was downgraded to V4 (post turnaround EPOCH)
* format and database extensions happened causing new V4 format bitmap blocks to be written. The count of V4
* format bitmap blocks is maintained ONLY as part of INCTN records (with INCTN opcode SET_JNL_FILE_CLOSE_EXTEND)
* noted down in rctl->blks_to_upgrd_adjust counter as part of BACKWARD processing which are finally added to
* csd->blks_to_upgrd.
*/
if (!jctl->turn_around_fullyupgraded || csd->blks_to_upgrd)
csd->fully_upgraded = FALSE;
csd->trans_hist.early_tn = jctl->turn_around_tn;
csd->trans_hist.curr_tn = csd->trans_hist.early_tn; /* INCREMENT_CURR_TN macro not used but noted in comment
* to identify all places that set curr_tn */
csd->jnl_eovtn = csd->trans_hist.curr_tn;
csd->turn_around_point = TRUE;
/* MUPIP REORG UPGRADE/DOWNGRADE stores its partially processed state in the database file header.
* It is difficult for recovery to restore those fields to a correct partial value.
* Hence reset the related fields as if the desired_db_format got set just ONE tn BEFORE the EPOCH record
* and that there was no more processing that happened.
* This might potentially mean some duplicate processing for MUPIP REORG UPGRADE/DOWNGRADE after the recovery.
* But that will only be the case as long as the database is in compatibility (mixed) mode (hopefully not long).
*/
if (csd->desired_db_format_tn >= jctl->turn_around_tn)
csd->desired_db_format_tn = jctl->turn_around_tn - 1;
if (csd->reorg_db_fmt_start_tn >= jctl->turn_around_tn)
csd->reorg_db_fmt_start_tn = jctl->turn_around_tn - 1;
if (csd->tn_upgrd_blks_0 > jctl->turn_around_tn)
csd->tn_upgrd_blks_0 = (trans_num)-1;
csd->reorg_upgrd_dwngrd_restart_block = 0;
/* Compute current value of "free_blocks" based on the value of "free_blocks" at the turnaround point epoch
* record and the change in "total_blks" since that epoch to the present form of the database. Any difference
* in "total_blks" implies database file extensions happened since the turnaround point. A backward rollback
* undoes everything (including all updates) except file extensions (it does not truncate the file size).
* Therefore every block that was newly allocated as part of those file extensions should be considered FREE
* for the current calculations except for the local bitmap blocks which are BUSY the moment they are created.
*/
assert(jnlrec->jrec_epoch.total_blks <= csd->trans_hist.total_blks);
csd->trans_hist.free_blocks = jnlrec->jrec_epoch.free_blocks
+ (csd->trans_hist.total_blks - jnlrec->jrec_epoch.total_blks)
- DIVIDE_ROUND_UP(csd->trans_hist.total_blks, BLKS_PER_LMAP)
+ DIVIDE_ROUND_UP(jnlrec->jrec_epoch.total_blks, BLKS_PER_LMAP);
assert(!csd->blks_to_upgrd || !csd->fully_upgraded);
assert((freeblks = mur_blocks_free(rctl)) == csd->trans_hist.free_blocks);
/* Update strm_reg_seqno[] in db file header to reflect the turn around point.
* Before updating "strm_reg_seqno", make sure value is saved into "save_strm_reg_seqno".
* This is relied upon by the function "mur_get_max_strm_reg_seqno" in case of interrupted rollback.
*/
for (idx = 0; idx < MAX_SUPPL_STRMS; idx++)
{
if (!csd->save_strm_reg_seqno[idx])
csd->save_strm_reg_seqno[idx] = csd->strm_reg_seqno[idx];
csd->strm_reg_seqno[idx] = jnlrec->jrec_epoch.strm_seqno[idx];
}
wcs_flu(WCSFLU_FLUSH_HDR | WCSFLU_FSYNC_DB);
assert(cs_addrs->ti->curr_tn == jctl->turn_around_tn);
# ifdef UNIX
if (jgbl.onlnrlbk)
{
if (dba_bg == cs_addrs->hdr->acc_meth)
{ /* dryclean the cache (basically reset the cycle fields in all teh cache records) so as to make
* GT.M processes that only does 'reads' to require crit and hence realize that online rollback
* is in progress
*/
bt_refresh(cs_addrs, FALSE); /* sets earliest bt TN to be the turn around TN */
}
db_csh_ref(cs_addrs, FALSE);
assert(NULL != cs_addrs->jnl);
jpc = cs_addrs->jnl;
assert(NULL != jpc->jnl_buff);
jbp = jpc->jnl_buff;
/* Since Rollback simulates the journal record along with the timestamp at which the update was made, it
* sets jgbl.dont_reset_gbl_jrec_time to TRUE so that during forward processing t_end or tp_tend does not
* reset the gbl_jrec_time to reflect the current time. But, with Online Rollback, one can have the shared
* memory up and running and hence can have jbp->prev_jrec_time to be the time of the most recent journal
* update made. Later in t_end/tp_tend, ADJUST_GBL_JREC_TIME is invoked which ensures that if ever
* gbl_jrec_time (the time of the current update) is less than jbp->prev_jrec_time (time of the latest
* journal update), dont_reset_gbl_jrec_time better be FALSE. But, this assert will trip since Rollback
* sets the latter to TRUE. To fix this, set jbp->prev_jrec_time to the turn around time stamp. This way
* we are guaranteed that all the updates done in the forward processing will have a timestamp that is
* greater than the turn around timestamp
*/
SET_JNLBUFF_PREV_JREC_TIME(jbp, jctl->turn_around_time, DO_GBL_JREC_TIME_CHECK_FALSE);
} else if (dba_bg == csd->acc_meth)
{ /* set earliest bt TN to be the turn-around TN (taken from bt_refresh()) */
SET_OLDEST_HIST_TN(cs_addrs, cs_addrs->ti->curr_tn - 1);
}
# else
if (dba_bg == csd->acc_meth)
{ /* set earliest bt TN to be the turn-around TN (taken from bt_refresh()) */
SET_OLDEST_HIST_TN(cs_addrs, cs_addrs->ti->curr_tn - 1);
}
# endif
csd->turn_around_point = FALSE;
assert(OLDEST_HIST_TN(cs_addrs) == (cs_addrs->ti->curr_tn - 1));
/* In case this is MM and wcs_flu() remapped an extended database, reset rctl->csd */
assert((dba_mm == cs_data->acc_meth) || (rctl->csd == cs_data));
rctl->csd = cs_data;
}
JNL_SHORT_TIME(now);
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
TP_CHANGE_REG_IF_NEEDED(rctl->gd);
if (!rctl->jfh_recov_interrupted)
jctl = rctl->jctl_turn_around;
else
{
DEBUG_ONLY(
for (jctl = rctl->jctl_turn_around; NULL != jctl->next_gen; jctl = jctl->next_gen)
;
/* check that latest gener file name does not match db header */
assert((rctl->csd->jnl_file_len != jctl->jnl_fn_len)
|| (0 != memcmp(rctl->csd->jnl_file_name, jctl->jnl_fn, jctl->jnl_fn_len)));
)
jctl = rctl->jctl_alt_head;
}
assert(NULL != jctl);
for ( ; NULL != jctl->next_gen; jctl = jctl->next_gen)
;
assert(rctl->csd->jnl_file_len == jctl->jnl_fn_len); /* latest gener file name */
assert(0 == memcmp(rctl->csd->jnl_file_name, jctl->jnl_fn, jctl->jnl_fn_len)); /* should match db header */
if (SS_NORMAL != (status = prepare_unique_name((char *)jctl->jnl_fn, jctl->jnl_fn_len, "", "",
rename_fn, &rename_fn_len, now, &status2)))
return status;
jctl->jnl_fn_len = rename_fn_len; /* change the name in memory to the proposed name */
memcpy(jctl->jnl_fn, rename_fn, rename_fn_len + 1);
/* Rename hasn't happened yet at the filesystem level. In case current recover command is interrupted,
* we need to update jfh->next_jnl_file_name before mur_forward(). Update jfh->next_jnl_file_name for
* all journal files from which PBLK records were applied. Create new journal files for forward play.
*/
assert(NULL != rctl->jctl_turn_around);
jctl = rctl->jctl_turn_around; /* points to journal file which has current recover's turn around point */
assert(0 != jctl->turn_around_offset);
jfh = jctl->jfh;
jfh->turn_around_offset = jctl->turn_around_offset; /* save progress in file header for */
jfh->turn_around_time = jctl->turn_around_time; /* possible re-issue of recover */
for (idx = 0; idx < MAX_SUPPL_STRMS; idx++)
jfh->strm_end_seqno[idx] = csd->strm_reg_seqno[idx];
jfh_changed = TRUE;
/* We are about to update the journal file header of the turnaround-point journal file to store the
* non-zero jfh->turn_around_offset. Ensure corresponding database is considered updated.
* This is needed in case journal recovery/rollback terminates abnormally and we go to mur_close_files.
* We need to ensure csd->recov_interrupted does not get reset to FALSE even if this region did not have
* have any updates to the corresponding database file otherwise. (GTM-8394)
*/
rctl->db_updated = TRUE;
for ( ; NULL != jctl; jctl = jctl->next_gen)
{ /* setup the next_jnl links. note that in the case of interrupted recovery, next_jnl links
* would have been already set starting from the turn-around point journal file of the
* interrupted recovery but the new recovery MIGHT have taken us to a still previous
* generation journal file that needs its next_jnl link set. this is why we do the next_jnl
* link setup even in the case of interrupted recovery although in most cases it is unnecessary.
*/
jfh = jctl->jfh;
if (NULL != jctl->next_gen)
{
jfh->next_jnl_file_name_length = jctl->next_gen->jnl_fn_len;
memcpy(jfh->next_jnl_file_name, jctl->next_gen->jnl_fn, jctl->next_gen->jnl_fn_len);
jfh_changed = TRUE;
} else
assert(0 == jfh->next_jnl_file_name_length); /* null link from latest generation */
if (jfh->turn_around_offset && (jctl != rctl->jctl_turn_around))
{ /* It is possible that the current recovery has a turn-around-point much before the
* previously interrupted recovery. If it happens to be a previous generation journal
* file then we have to reset the original turn-around-point to be zero in the journal
* file header in order to ensure if this recovery gets interrupted we do interrupted
* recovery processing until the new turn-around-point instead of stopping incorrectly
* at the original turn-around-point itself. Note that there could be more than one
* journal file with a non-zero turn_around_offset (depending on how many previous
* recoveries got interrupted in this loop) that need to be reset.
*/
assert(!jctl->turn_around_offset);
assert(rctl->recov_interrupted || rctl->jctl_apply_pblk); /* rctl->jfh_recov_interrupted can fail */
jfh->turn_around_offset = 0;
jfh->turn_around_time = 0;
jfh_changed = TRUE;
}
if (jfh_changed)
{
/* Since overwriting the journal file header (an already allocated block
* in the file) should not cause ENOSPC, we dont take the trouble of
* passing csa or jnl_fn (first two parameters). Instead we pass NULL.
*/
JNL_DO_FILE_WRITE(NULL, NULL, jctl->channel, 0, jfh,
REAL_JNL_HDR_LEN, jctl->status, jctl->status2);
if (SS_NORMAL != jctl->status)
{
assert(FALSE);
if (SS_NORMAL == jctl->status2)
gtm_putmsg_csa(CSA_ARG(rctl->csa) VARLSTCNT(5) ERR_JNLWRERR, 2, jctl->jnl_fn_len,
jctl->jnl_fn, jctl->status);
else
gtm_putmsg_csa(CSA_ARG(rctl->csa) VARLSTCNT1(6) ERR_JNLWRERR, 2, jctl->jnl_fn_len,
jctl->jnl_fn, jctl->status, PUT_SYS_ERRNO(jctl->status2));
return jctl->status;
}
GTM_JNL_FSYNC(rctl->csa, jctl->channel, jctl->status);
if (-1 == jctl->status)
{
jctl->status2 = errno;
assert(FALSE);
gtm_putmsg_csa(CSA_ARG(rctl->csa) VARLSTCNT(9) ERR_JNLFSYNCERR, 2,
jctl->jnl_fn_len, jctl->jnl_fn,
ERR_TEXT, 2, RTS_ERROR_TEXT("Error with fsync"), jctl->status2);
return ERR_JNLFSYNCERR;
}
}
jfh_changed = FALSE;
}
memset(&jnl_info, 0, SIZEOF(jnl_info));
jnl_info.status = jnl_info.status2 = SS_NORMAL;
jnl_info.prev_jnl = &prev_jnl_fn[0];
set_jnl_info(rctl->gd, &jnl_info);
jnl_info.prev_jnl_len = rctl->jctl_turn_around->jnl_fn_len;
memcpy(jnl_info.prev_jnl, rctl->jctl_turn_around->jnl_fn, rctl->jctl_turn_around->jnl_fn_len);
jnl_info.prev_jnl[jnl_info.prev_jnl_len] = 0;
jnl_info.jnl_len = rctl->csd->jnl_file_len;
memcpy(jnl_info.jnl, rctl->csd->jnl_file_name, jnl_info.jnl_len);
jnl_info.jnl[jnl_info.jnl_len] = 0;
assert(!mur_options.rollback || jgbl.mur_rollback);
jnl_info.reg_seqno = rctl->jctl_turn_around->turn_around_seqno;
jgbl.gbl_jrec_time = rctl->jctl_turn_around->turn_around_time; /* time needed for cre_jnl_file_common() */
if (EXIT_NRM != cre_jnl_file_common(&jnl_info, rename_fn, rename_fn_len))
{
gtm_putmsg_csa(CSA_ARG(rctl->csa) VARLSTCNT(4) ERR_JNLNOCREATE, 2, jnl_info.jnl_len, jnl_info.jnl);
return jnl_info.status;
}
# ifdef UNIX
if (jgbl.onlnrlbk)
{
cs_addrs = rctl->csa;
/* Mimic what jnl_file_close in case of cleanly a closed journal file */
jpc = cs_addrs->jnl; /* the previous loop makes sure cs_addrs->jnl->jnl_buff is valid*/
NULLIFY_JNL_FILE_ID(cs_addrs);
jpc->jnl_buff->cycle++; /* so that, all other processes knows to switch to newer journal file */
jpc->cycle--; /* decrement cycle so jnl_ensure_open() knows to reopen the journal */
}
# endif
if (NULL != rctl->jctl_alt_head) /* remove the journal files created by last interrupted recover process */
{
mur_rem_jctls(rctl);
rctl->jctl_alt_head = NULL;
}
/* From this point on, journal records are written into the newly created journal file. However, we still read
* from old journal files.
*/
}
