asmlinkage int sunos_shmsys(int op, unsigned long arg1, unsigned long arg2,
unsigned long arg3)
{
unsigned long raddr;
int rval;
switch(op) {
case 0:
/* do_shmat(): attach a shared memory area */
rval = do_shmat((int)arg1,(char __user *)arg2,(int)arg3,&raddr);
if (!rval)
rval = (int) raddr;
break;
case 1:
/* sys_shmctl(): modify shared memory area attr. */
rval = sys_shmctl((int)arg1,(int)arg2,(struct shmid_ds __user *)arg3);
break;
case 2:
/* sys_shmdt(): detach a shared memory area */
rval = sys_shmdt((char __user *)arg1);
break;
case 3:
/* sys_shmget(): get a shared memory area */
rval = sys_shmget((key_t)arg1,(int)arg2,(int)arg3);
break;
default:
rval = -EINVAL;
break;
};
return rval;
}
asmlinkage long xtensa_shmat(int shmid, char __user *shmaddr, int shmflg)
{
unsigned long ret;
long err;
err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
if (err)
return err;
return (long)ret;
}
long compat_sys_shmat(int shmid, compat_uptr_t shmaddr, int shmflg)
{
unsigned long ret;
long err;
err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret);
if (err)
return err;
force_successful_syscall_return();
return (long)ret;
}
/*
* Native ABI that is O32 or N64 version
*/
asmlinkage long sys_shmat(int shmid, char __user *shmaddr,
int shmflg, unsigned long *addr)
{
unsigned long raddr;
int err;
err = do_shmat(shmid, shmaddr, shmflg, &raddr);
if (err)
return err;
return put_user(raddr, addr);
}
asmlinkage unsigned long
ia64_shmat (int shmid, void __user *shmaddr, int shmflg)
{
unsigned long raddr;
int retval;
retval = do_shmat(shmid, shmaddr, shmflg, &raddr);
if (retval < 0)
return retval;
force_successful_syscall_return();
return raddr;
}
long compat_sys_shmat(int first, int second, compat_uptr_t third, int version,
void __user *uptr)
{
int err;
unsigned long raddr;
compat_ulong_t __user *uaddr;
if (version == 1)
return -EINVAL;
err = do_shmat(first, uptr, second, &raddr);
if (err < 0)
return err;
uaddr = compat_ptr(third);
return put_user(raddr, uaddr);
}
asmlinkage int sunos_shmsys(int op, u32 arg1, u32 arg2, u32 arg3)
{
struct shmid_ds ksds;
unsigned long raddr;
mm_segment_t old_fs = get_fs();
int rval;
switch(op) {
case 0:
/* do_shmat(): attach a shared memory area */
rval = do_shmat((int)arg1,(char __user *)(unsigned long)arg2,(int)arg3,&raddr);
if (!rval)
rval = (int) raddr;
break;
case 1:
/* sys_shmctl(): modify shared memory area attr. */
if (!sunos_shmid_get((struct shmid_ds32 __user *)(unsigned long)arg3, &ksds)) {
set_fs(KERNEL_DS);
rval = sys_shmctl((int) arg1,(int) arg2,
(struct shmid_ds __user *) &ksds);
set_fs(old_fs);
if (!rval)
rval = sunos_shmid_put((struct shmid_ds32 __user *)(unsigned long)arg3,
&ksds);
} else
rval = -EFAULT;
break;
case 2:
/* sys_shmdt(): detach a shared memory area */
rval = sys_shmdt((char __user *)(unsigned long)arg1);
break;
case 3:
/* sys_shmget(): get a shared memory area */
rval = sys_shmget((key_t)arg1,(int)arg2,(int)arg3);
break;
default:
rval = -EINVAL;
break;
};
return rval;
}
int mu_rndwn_replpool(replpool_identifier *replpool_id, repl_inst_hdr_ptr_t repl_inst_filehdr, int shm_id, boolean_t *ipc_rmvd)
{
int semval, status, save_errno, nattch;
char *instfilename, pool_type;
sm_uc_ptr_t start_addr;
struct shmid_ds shm_buf;
unix_db_info *udi;
sgmnt_addrs *csa;
boolean_t anticipatory_freeze_available, force_attach;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(INVALID_SHMID != shm_id);
instfilename = replpool_id->instfilename;
pool_type = replpool_id->pool_type;
assert((JNLPOOL_SEGMENT == pool_type) || (RECVPOOL_SEGMENT == pool_type));
anticipatory_freeze_available = ANTICIPATORY_FREEZE_AVAILABLE;
force_attach = (jgbl.onlnrlbk || (!jgbl.mur_rollback && !argumentless_rundown && anticipatory_freeze_available));
if (-1 == shmctl(shm_id, IPC_STAT, &shm_buf))
{
save_errno = errno;
ISSUE_REPLPOOLINST_AND_RETURN(save_errno, shm_id, instfilename, "shmctl()");
}
nattch = shm_buf.shm_nattch;
if ((0 != nattch) && !force_attach)
{
util_out_print("Replpool segment (id = !UL) for replication instance !AD is in use by another process.",
TRUE, shm_id, LEN_AND_STR(instfilename));
return -1;
}
if (-1 == (sm_long_t)(start_addr = (sm_uc_ptr_t) do_shmat(shm_id, 0, 0)))
{
save_errno = errno;
ISSUE_REPLPOOLINST_AND_RETURN(save_errno, shm_id, instfilename, "shmat()");
}
ESTABLISH_RET(mu_rndwn_replpool_ch, -1);
/* assert that the identifiers are at the top of replpool control structure */
assert(0 == offsetof(jnlpool_ctl_struct, jnlpool_id));
assert(0 == offsetof(recvpool_ctl_struct, recvpool_id));
memcpy((void *)replpool_id, (void *)start_addr, SIZEOF(replpool_identifier));
if (memcmp(replpool_id->label, GDS_RPL_LABEL, GDS_LABEL_SZ - 1))
{
if (!memcmp(replpool_id->label, GDS_RPL_LABEL, GDS_LABEL_SZ - 3))
util_out_print(
"Incorrect version for the replpool segment (id = !UL) belonging to replication instance !AD",
TRUE, shm_id, LEN_AND_STR(instfilename));
else
util_out_print("Incorrect replpool format for the segment (id = !UL) belonging to replication instance !AD",
TRUE, shm_id, LEN_AND_STR(instfilename));
DETACH_AND_RETURN(start_addr, shm_id, instfilename);
}
if (memcmp(replpool_id->now_running, gtm_release_name, gtm_release_name_len + 1))
{
util_out_print("Attempt to access with version !AD, while already using !AD for replpool segment (id = !UL)"
" belonging to replication instance !AD.", TRUE, gtm_release_name_len, gtm_release_name,
LEN_AND_STR(replpool_id->now_running), shm_id, LEN_AND_STR(instfilename));
DETACH_AND_RETURN(start_addr, shm_id, instfilename);
}
/* Assert that if we haven't yet attached to the journal pool yet, we have the corresponding global vars set to NULL */
assert((JNLPOOL_SEGMENT != pool_type) || ((NULL == jnlpool.jnlpool_ctl) && (NULL == jnlpool_ctl)));
if (JNLPOOL_SEGMENT == pool_type)
{ /* Initialize variables to simulate a "jnlpool_init". This is required by "repl_inst_flush_jnlpool" called below */
jnlpool_ctl = jnlpool.jnlpool_ctl = (jnlpool_ctl_ptr_t)start_addr;
assert(NULL != jnlpool.jnlpool_dummy_reg);
udi = FILE_INFO(jnlpool.jnlpool_dummy_reg);
csa = &udi->s_addrs;
csa->critical = (mutex_struct_ptr_t)((sm_uc_ptr_t)jnlpool.jnlpool_ctl + JNLPOOL_CTL_SIZE);
csa->nl = (node_local_ptr_t)((sm_uc_ptr_t)csa->critical + CRIT_SPACE + SIZEOF(mutex_spin_parms_struct));
/* secshr_db_clnup uses this relationship */
assert(jnlpool.jnlpool_ctl->filehdr_off);
assert(jnlpool.jnlpool_ctl->srclcl_array_off > jnlpool.jnlpool_ctl->filehdr_off);
assert(jnlpool.jnlpool_ctl->sourcelocal_array_off > jnlpool.jnlpool_ctl->srclcl_array_off);
/* Initialize "jnlpool.repl_inst_filehdr" and related fields as "repl_inst_flush_jnlpool" relies on that */
jnlpool.repl_inst_filehdr = (repl_inst_hdr_ptr_t)((sm_uc_ptr_t)jnlpool.jnlpool_ctl
+ jnlpool.jnlpool_ctl->filehdr_off);
jnlpool.gtmsrc_lcl_array = (gtmsrc_lcl_ptr_t)((sm_uc_ptr_t)jnlpool.jnlpool_ctl
+ jnlpool.jnlpool_ctl->srclcl_array_off);
jnlpool.gtmsource_local_array = (gtmsource_local_ptr_t)((sm_uc_ptr_t)jnlpool.jnlpool_ctl
+ jnlpool.jnlpool_ctl->sourcelocal_array_off);
if (0 == nattch)
{ /* No one attached. So, we can safely flush the journal pool so that the gtmsrc_lcl structures in the
* jnlpool and disk are in sync with each other. More importantly we are about to remove the jnlpool
* so we better get things in sync before that. If anticipatory freeze scheme is in effect, then we
* need to keep the journal pool up and running. So, don't reset the crash field in the instance file
* header (dictated by the second parameter to repl_inst_flush_jnlpool below).
* Note:
* If mu_rndwn_repl_instance created new semaphores (in mu_replpool_remove_sem), we need to flush those
* to the instance file as well. So, override the jnlpool_semid and jnlpool_semid_ctime with the new
* values.
*/
assert((INVALID_SEMID != repl_inst_filehdr->jnlpool_semid)
&& (0 != repl_inst_filehdr->jnlpool_semid_ctime));
jnlpool.repl_inst_filehdr->jnlpool_semid = repl_inst_filehdr->jnlpool_semid;
jnlpool.repl_inst_filehdr->jnlpool_semid_ctime = repl_inst_filehdr->jnlpool_semid_ctime;
repl_inst_flush_jnlpool(FALSE, !anticipatory_freeze_available);
assert(!jnlpool.repl_inst_filehdr->crash || anticipatory_freeze_available);
/* Refresh local copy (repl_inst_filehdr) with the copy that was just flushed (jnlpool.repl_inst_filehdr) */
memcpy(repl_inst_filehdr, jnlpool.repl_inst_filehdr, SIZEOF(repl_inst_hdr));
if (!anticipatory_freeze_available)
{ /* Now that jnlpool has been flushed and there is going to be no journal pool, reset
* "jnlpool.repl_inst_filehdr" as otherwise other routines (e.g. "repl_inst_recvpool_reset") are
* affected by whether this is NULL or not.
*/
jnlpool.jnlpool_ctl = NULL;
jnlpool_ctl = NULL;
jnlpool.gtmsrc_lcl_array = NULL;
jnlpool.gtmsource_local_array = NULL;
jnlpool.jnldata_base = NULL;
jnlpool.repl_inst_filehdr = NULL;
}
} /* else we are ONLINE ROLLBACK. repl_inst_flush_jnlpool will be done later after gvcst_init in mur_open_files */
}
if ((0 == nattch) && (!anticipatory_freeze_available || (RECVPOOL_SEGMENT == pool_type)))
{
if (-1 == shmdt((caddr_t)start_addr))
{
save_errno = errno;
ISSUE_REPLPOOLINST_AND_RETURN(save_errno, shm_id, instfilename, "shmdt()");
}
if (0 != shm_rmid(shm_id))
{
save_errno = errno;
ISSUE_REPLPOOLINST_AND_RETURN(save_errno, shm_id, instfilename, "shm_rmid()");
}
if (JNLPOOL_SEGMENT == pool_type)
{
repl_inst_filehdr->jnlpool_shmid = INVALID_SHMID;
repl_inst_filehdr->jnlpool_shmid_ctime = 0;
assert((NULL == jnlpool.jnlpool_ctl) && (NULL == jnlpool_ctl));
*ipc_rmvd = TRUE;
} else
{
repl_inst_filehdr->recvpool_shmid = INVALID_SHMID;
repl_inst_filehdr->recvpool_shmid_ctime = 0;
*ipc_rmvd = TRUE;
}
} else
{ /* Else we are ONLINE ROLLBACK or anticipatory freeze is in effect and so we want to keep the journal pool available
* for the duration of the rollback. Do not remove and/or reset the fields in the file header
*/
assert((JNLPOOL_SEGMENT != pool_type) || ((NULL != jnlpool.jnlpool_ctl) && (NULL != jnlpool_ctl)));
if (JNLPOOL_SEGMENT == pool_type)
*ipc_rmvd = FALSE;
if (RECVPOOL_SEGMENT == pool_type)
*ipc_rmvd = FALSE;
}
REVERT;
return 0;
}
SYSCALL_DEFINE6(sparc_ipc, unsigned int, call, int, first, unsigned long, second,
unsigned long, third, void __user *, ptr, long, fifth)
{
long err;
/* No need for backward compatibility. We can start fresh... */
if (call <= SEMCTL) {
switch (call) {
case SEMOP:
err = sys_semtimedop(first, ptr,
(unsigned)second, NULL);
goto out;
case SEMTIMEDOP:
err = sys_semtimedop(first, ptr, (unsigned)second,
(const struct timespec __user *)
(unsigned long) fifth);
goto out;
case SEMGET:
err = sys_semget(first, (int)second, (int)third);
goto out;
case SEMCTL: {
err = sys_semctl(first, second,
(int)third | IPC_64,
(union semun) ptr);
goto out;
}
default:
err = -ENOSYS;
goto out;
}
}
if (call <= MSGCTL) {
switch (call) {
case MSGSND:
err = sys_msgsnd(first, ptr, (size_t)second,
(int)third);
goto out;
case MSGRCV:
err = sys_msgrcv(first, ptr, (size_t)second, fifth,
(int)third);
goto out;
case MSGGET:
err = sys_msgget((key_t)first, (int)second);
goto out;
case MSGCTL:
err = sys_msgctl(first, (int)second | IPC_64, ptr);
goto out;
default:
err = -ENOSYS;
goto out;
}
}
if (call <= SHMCTL) {
switch (call) {
case SHMAT: {
ulong raddr;
err = do_shmat(first, ptr, (int)second, &raddr);
if (!err) {
if (put_user(raddr,
(ulong __user *) third))
err = -EFAULT;
}
goto out;
}
case SHMDT:
err = sys_shmdt(ptr);
goto out;
case SHMGET:
err = sys_shmget(first, (size_t)second, (int)third);
goto out;
case SHMCTL:
err = sys_shmctl(first, (int)second | IPC_64, ptr);
goto out;
default:
err = -ENOSYS;
goto out;
}
} else {
err = -ENOSYS;
}
out:
return err;
}
/*
* sys_ipc() is the de-multiplexer for the SysV IPC calls..
*
* This is really horribly ugly. This will be remove with new toolchain.
*/
asmlinkage long
sys_ipc(uint call, int first, int second, int third, void *ptr, long fifth)
{
int version, ret;
version = call >> 16; /* hack for backward compatibility */
call &= 0xffff;
ret = -EINVAL;
switch (call) {
case SEMOP:
ret = sys_semop(first, (struct sembuf *)ptr, second);
break;
case SEMGET:
ret = sys_semget(first, second, third);
break;
case SEMCTL:
{
union semun fourth;
if (!ptr)
break;
ret = (access_ok(VERIFY_READ, ptr, sizeof(long)) ? 0 : -EFAULT)
|| (get_user(fourth.__pad, (void **)ptr)) ;
if (ret)
break;
ret = sys_semctl(first, second, third, fourth);
break;
}
case MSGSND:
ret = sys_msgsnd(first, (struct msgbuf *) ptr, second, third);
break;
case MSGRCV:
switch (version) {
case 0: {
struct ipc_kludge tmp;
if (!ptr)
break;
ret = (access_ok(VERIFY_READ, ptr, sizeof(tmp))
? 0 : -EFAULT) || copy_from_user(&tmp,
(struct ipc_kludge *) ptr, sizeof(tmp));
if (ret)
break;
ret = sys_msgrcv(first, tmp.msgp, second, tmp.msgtyp,
third);
break;
}
default:
ret = sys_msgrcv(first, (struct msgbuf *) ptr,
second, fifth, third);
break;
}
break;
case MSGGET:
ret = sys_msgget((key_t) first, second);
break;
case MSGCTL:
ret = sys_msgctl(first, second, (struct msqid_ds *) ptr);
break;
case SHMAT:
switch (version) {
default: {
ulong raddr;
ret = access_ok(VERIFY_WRITE, (ulong *) third,
sizeof(ulong)) ? 0 : -EFAULT;
if (ret)
break;
ret = do_shmat(first, (char *) ptr, second, &raddr);
if (ret)
break;
ret = put_user(raddr, (ulong *) third);
break;
}
case 1: /* iBCS2 emulator entry point */
if (!segment_eq(get_fs(), get_ds()))
break;
ret = do_shmat(first, (char *) ptr, second,
(ulong *) third);
break;
}
break;
case SHMDT:
ret = sys_shmdt((char *)ptr);
break;
case SHMGET:
ret = sys_shmget(first, second, third);
break;
case SHMCTL:
ret = sys_shmctl(first, second, (struct shmid_ds *) ptr);
break;
}
return ret;
}
void db_init(gd_region *reg, sgmnt_data_ptr_t tsd)
{
static boolean_t mutex_init_done = FALSE;
boolean_t is_bg, read_only;
char machine_name[MAX_MCNAMELEN];
file_control *fc;
int gethostname_res, stat_res, mm_prot;
int4 status, semval, dblksize, fbwsize;
sm_long_t status_l;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
struct sembuf sop[3];
struct stat stat_buf;
union semun semarg;
struct semid_ds semstat;
struct shmid_ds shmstat;
struct statvfs dbvfs;
uint4 sopcnt;
unix_db_info *udi;
#ifdef periodic_timer_removed
void periodic_flush_check();
#endif
error_def(ERR_CLSTCONFLICT);
error_def(ERR_CRITSEMFAIL);
error_def(ERR_DBNAMEMISMATCH);
error_def(ERR_DBIDMISMATCH);
error_def(ERR_NLMISMATCHCALC);
error_def(ERR_REQRUNDOWN);
error_def(ERR_SYSCALL);
assert(tsd->acc_meth == dba_bg || tsd->acc_meth == dba_mm);
is_bg = (dba_bg == tsd->acc_meth);
read_only = reg->read_only;
new_dbinit_ipc = FALSE; /* we did not create a new ipc resource */
udi = FILE_INFO(reg);
memset(machine_name, 0, sizeof(machine_name));
if (GETHOSTNAME(machine_name, MAX_MCNAMELEN, gethostname_res))
rts_error(VARLSTCNT(5) ERR_TEXT, 2, LEN_AND_LIT("Unable to get the hostname"), errno);
assert(strlen(machine_name) < MAX_MCNAMELEN);
csa = &udi->s_addrs;
csa->db_addrs[0] = csa->db_addrs[1] = csa->lock_addrs[0] = NULL; /* to help in dbinit_ch and gds_rundown */
reg->opening = TRUE;
/*
* Create ftok semaphore for this region.
* We do not want to make ftok counter semaphore to be 2 for on mupip journal recover process.
*/
if (!ftok_sem_get(reg, !mupip_jnl_recover, GTM_ID, FALSE))
rts_error(VARLSTCNT(4) ERR_DBFILERR, 2, DB_LEN_STR(reg));
/*
* At this point we have ftok_semid sempahore based on ftok key.
* Any ftok conflicted region will block at this point.
* Say, a.dat and b.dat both has same ftok and we have process A to access a.dat and
* process B to access b.dat. In this case only one can continue to do db_init()
*/
fc = reg->dyn.addr->file_cntl;
fc->file_type = reg->dyn.addr->acc_meth;
fc->op = FC_READ;
fc->op_buff = (sm_uc_ptr_t)tsd;
fc->op_len = sizeof(*tsd);
fc->op_pos = 1;
dbfilop(fc); /* Read file header */
udi->shmid = tsd->shmid;
udi->semid = tsd->semid;
udi->sem_ctime = tsd->sem_ctime.ctime;
udi->shm_ctime = tsd->shm_ctime.ctime;
dbsecspc(reg, tsd); /* Find db segment size */
if (!mupip_jnl_recover)
{
if (INVALID_SEMID == udi->semid)
{
if (0 != udi->sem_ctime || INVALID_SHMID != udi->shmid || 0 != udi->shm_ctime)
/* We must have somthing wrong in protocol or, code, if this happens */
GTMASSERT;
/*
* Create new semaphore using IPC_PRIVATE. System guarantees a unique id.
*/
if (-1 == (udi->semid = semget(IPC_PRIVATE, FTOK_SEM_PER_ID, RWDALL | IPC_CREAT)))
{
udi->semid = INVALID_SEMID;
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control semget"), errno);
}
udi->shmid = INVALID_SHMID; /* reset shmid so dbinit_ch does not get confused in case we go there */
new_dbinit_ipc = TRUE;
tsd->semid = udi->semid;
semarg.val = GTM_ID;
/*
* Following will set semaphore number 2 (=FTOK_SEM_PER_ID - 1) value as GTM_ID.
* In case we have orphaned semaphore for some reason, mupip rundown will be
* able to identify GTM semaphores from the value and can remove.
*/
if (-1 == semctl(udi->semid, FTOK_SEM_PER_ID - 1, SETVAL, semarg))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control semctl SETVAL"), errno);
/*
* Warning: We must read the sem_ctime using IPC_STAT after SETVAL, which changes it.
* We must NOT do any more SETVAL after this. Our design is to use
* sem_ctime as creation time of semaphore.
*/
semarg.buf = &semstat;
if (-1 == semctl(udi->semid, FTOK_SEM_PER_ID - 1, IPC_STAT, semarg))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control semctl IPC_STAT"), errno);
tsd->sem_ctime.ctime = udi->sem_ctime = semarg.buf->sem_ctime;
} else
{
if (INVALID_SHMID == udi->shmid)
/* if mu_rndwn_file gets standalone access of this region and
* somehow mupip process crashes, we can have semid != -1 but shmid == -1
*/
rts_error(VARLSTCNT(10) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name),
ERR_TEXT, 2, LEN_AND_LIT("semid is valid but shmid is invalid"));
semarg.buf = &semstat;
if (-1 == semctl(udi->semid, 0, IPC_STAT, semarg))
/* file header has valid semid but semaphore does not exists */
rts_error(VARLSTCNT(6) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name));
else if (semarg.buf->sem_ctime != tsd->sem_ctime.ctime)
rts_error(VARLSTCNT(10) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name),
ERR_TEXT, 2, LEN_AND_LIT("sem_ctime does not match"));
if (-1 == shmctl(udi->shmid, IPC_STAT, &shmstat))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control shmctl"), errno);
else if (shmstat.shm_ctime != tsd->shm_ctime.ctime)
rts_error(VARLSTCNT(10) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(tsd->machine_name),
ERR_TEXT, 2, LEN_AND_LIT("shm_ctime does not match"));
}
/* We already have ftok semaphore of this region, so just plainly do semaphore operation */
/* This is the database access control semaphore for any region */
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;
if (!read_only)
{
sop[2].sem_num = 1; sop[2].sem_op = 1; /* increment r/w access counter */
sopcnt = 3;
}
sop[0].sem_flg = sop[1].sem_flg = sop[2].sem_flg = SEM_UNDO | IPC_NOWAIT;
SEMOP(udi->semid, sop, sopcnt, status);
if (-1 == status)
{
errno_save = errno;
gtm_putmsg(VARLSTCNT(4) ERR_CRITSEMFAIL, 2, DB_LEN_STR(reg));
rts_error(VARLSTCNT(8) ERR_SYSCALL, 5, RTS_ERROR_LITERAL("semop()"), CALLFROM, errno_save);
}
} else /* for mupip_jnl_recover we were already in mu_rndwn_file and got "semid" semaphore */
{
if (INVALID_SEMID == udi->semid || 0 == udi->sem_ctime)
/* make sure mu_rndwn_file() has reset created semaphore for standalone access */
GTMASSERT;
if (INVALID_SHMID != udi->shmid || 0 != udi->shm_ctime)
/* make sure mu_rndwn_file() has reset shared memory */
GTMASSERT;
udi->shmid = INVALID_SHMID; /* reset shmid so dbinit_ch does not get confused in case we go there */
new_dbinit_ipc = TRUE;
}
sem_incremented = TRUE;
if (new_dbinit_ipc)
{
/* Create new shared memory using IPC_PRIVATE. System guarantees a unique id */
#ifdef __MVS__
if (-1 == (status_l = udi->shmid = shmget(IPC_PRIVATE, ROUND_UP(reg->sec_size, MEGA_BOUND),
__IPC_MEGA | IPC_CREAT | RWDALL)))
#else
if (-1 == (status_l = udi->shmid = shmget(IPC_PRIVATE, reg->sec_size, RWDALL | IPC_CREAT)))
#endif
{
udi->shmid = status_l = INVALID_SHMID;
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database shmget"), errno);
}
tsd->shmid = udi->shmid;
if (-1 == shmctl(udi->shmid, IPC_STAT, &shmstat))
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database control shmctl"), errno);
tsd->shm_ctime.ctime = udi->shm_ctime = shmstat.shm_ctime;
}
#ifdef DEBUG_DB64
status_l = (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)do_shmat(udi->shmid, next_smseg, SHM_RND));
next_smseg = (sm_uc_ptr_t)ROUND_UP((sm_long_t)(next_smseg + reg->sec_size), SHMAT_ADDR_INCS);
#else
status_l = (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)do_shmat(udi->shmid, 0, SHM_RND));
#endif
if (-1 == status_l)
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error attaching to database shared memory"), errno);
}
csa->nl = (node_local_ptr_t)csa->db_addrs[0];
csa->critical = (mutex_struct_ptr_t)(csa->db_addrs[0] + NODE_LOCAL_SIZE);
assert(((int)csa->critical & 0xf) == 0); /* critical should be 16-byte aligned */
#ifdef CACHELINE_SIZE
assert(0 == ((int)csa->critical & (CACHELINE_SIZE - 1)));
#endif
/* Note: Here we check jnl_sate from database file and its value cannot change without standalone access.
* The jnl_buff buffer should be initialized irrespective of read/write process */
JNL_INIT(csa, reg, tsd);
csa->backup_buffer = (backup_buff_ptr_t)(csa->db_addrs[0] + NODE_LOCAL_SPACE + JNL_SHARE_SIZE(tsd));
csa->lock_addrs[0] = (sm_uc_ptr_t)csa->backup_buffer + BACKUP_BUFFER_SIZE + 1;
csa->lock_addrs[1] = csa->lock_addrs[0] + LOCK_SPACE_SIZE(tsd) - 1;
csa->total_blks = tsd->trans_hist.total_blks; /* For test to see if file has extended */
if (new_dbinit_ipc)
{
memset(csa->nl, 0, sizeof(*csa->nl)); /* We allocated shared storage -- we have to init it */
if (JNL_ALLOWED(csa))
{ /* initialize jb->cycle to a value different from initial value of jpc->cycle (0). although this is not
* necessary right now, in the future, the plan is to change jnl_ensure_open() to only do a cycle mismatch
* check in order to determine whether to call jnl_file_open() or not. this is in preparation for that.
*/
csa->jnl->jnl_buff->cycle = 1;
}
}
if (is_bg)
csd = csa->hdr = (sgmnt_data_ptr_t)(csa->lock_addrs[1] + 1 + CACHE_CONTROL_SIZE(tsd));
else
{
csa->acc_meth.mm.mmblk_state = (mmblk_que_heads_ptr_t)(csa->lock_addrs[1] + 1);
FSTAT_FILE(udi->fd, &stat_buf, stat_res);
if (-1 == stat_res)
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), errno);
mm_prot = read_only ? PROT_READ : (PROT_READ | PROT_WRITE);
#ifdef DEBUG_DB64
if (-1 == (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)mmap((caddr_t)get_mmseg((size_t)stat_buf.st_size),
(size_t)stat_buf.st_size,
mm_prot,
GTM_MM_FLAGS, udi->fd, (off_t)0)))
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), errno);
put_mmseg((caddr_t)(csa->db_addrs[0]), (size_t)stat_buf.st_size);
#else
if (-1 == (sm_long_t)(csa->db_addrs[0] = (sm_uc_ptr_t)mmap((caddr_t)NULL,
(size_t)stat_buf.st_size,
mm_prot,
GTM_MM_FLAGS, udi->fd, (off_t)0)))
rts_error(VARLSTCNT(5) ERR_DBFILERR, 2, DB_LEN_STR(reg), errno);
#endif
csa->db_addrs[1] = csa->db_addrs[0] + stat_buf.st_size - 1;
csd = csa->hdr = (sgmnt_data_ptr_t)csa->db_addrs[0];
}
if (!csa->nl->glob_sec_init)
{
assert(new_dbinit_ipc);
if (is_bg)
*csd = *tsd;
if (csd->machine_name[0]) /* crash occured */
{
if (0 != memcmp(csd->machine_name, machine_name, MAX_MCNAMELEN)) /* crashed on some other node */
rts_error(VARLSTCNT(6) ERR_CLSTCONFLICT, 4, DB_LEN_STR(reg), LEN_AND_STR(csd->machine_name));
else
rts_error(VARLSTCNT(6) ERR_REQRUNDOWN, 4, DB_LEN_STR(reg), LEN_AND_STR(csd->machine_name));
}
if (is_bg)
{
bt_malloc(csa);
csa->nl->cache_off = -CACHE_CONTROL_SIZE(tsd);
db_csh_ini(csa);
}
db_csh_ref(csa);
strcpy(csa->nl->machine_name, machine_name); /* machine name */
assert(MAX_REL_NAME > gtm_release_name_len);
memcpy(csa->nl->now_running, gtm_release_name, gtm_release_name_len + 1); /* GT.M release name */
memcpy(csa->nl->label, GDS_LABEL, GDS_LABEL_SZ - 1); /* GDS label */
memcpy(csa->nl->fname, reg->dyn.addr->fname, reg->dyn.addr->fname_len); /* database filename */
csa->nl->creation_date_time = csd->creation.date_time;
csa->nl->highest_lbm_blk_changed = -1;
csa->nl->wcs_timers = -1;
csa->nl->nbb = BACKUP_NOT_IN_PROGRESS;
csa->nl->unique_id.uid = FILE_INFO(reg)->fileid; /* save what file we initialized this storage for */
/* save pointers in csa to access shared memory */
csa->nl->critical = (sm_off_t)((sm_uc_ptr_t)csa->critical - (sm_uc_ptr_t)csa->nl);
if (JNL_ALLOWED(csa))
csa->nl->jnl_buff = (sm_off_t)((sm_uc_ptr_t)csa->jnl->jnl_buff - (sm_uc_ptr_t)csa->nl);
csa->nl->backup_buffer = (sm_off_t)((sm_uc_ptr_t)csa->backup_buffer - (sm_uc_ptr_t)csa->nl);
csa->nl->hdr = (sm_off_t)((sm_uc_ptr_t)csd - (sm_uc_ptr_t)csa->nl);
csa->nl->lock_addrs = (sm_off_t)((sm_uc_ptr_t)csa->lock_addrs[0] - (sm_uc_ptr_t)csa->nl);
if (!read_only || is_bg)
{
csd->trans_hist.early_tn = csd->trans_hist.curr_tn;
csd->max_update_array_size = csd->max_non_bm_update_array_size
= ROUND_UP2(MAX_NON_BITMAP_UPDATE_ARRAY_SIZE(csd), UPDATE_ARRAY_ALIGN_SIZE);
csd->max_update_array_size += ROUND_UP2(MAX_BITMAP_UPDATE_ARRAY_SIZE, UPDATE_ARRAY_ALIGN_SIZE);
/* add current db_csh counters into the cumulative counters and reset the current counters */
#define TAB_DB_CSH_ACCT_REC(COUNTER, DUMMY1, DUMMY2) \
csd->COUNTER.cumul_count += csd->COUNTER.curr_count; \
csd->COUNTER.curr_count = 0;
#include "tab_db_csh_acct_rec.h"
#undef TAB_DB_CSH_ACCT_REC
}
if (!read_only)
{
if (is_bg)
{
assert(memcmp(csd, GDS_LABEL, GDS_LABEL_SZ - 1) == 0);
LSEEKWRITE(udi->fd, (off_t)0, (sm_uc_ptr_t)csd, sizeof(sgmnt_data), errno_save);
if (0 != errno_save)
{
rts_error(VARLSTCNT(9) ERR_DBFILERR, 2, DB_LEN_STR(reg),
ERR_TEXT, 2, LEN_AND_LIT("Error with database write"), errno_save);
}
}
}
reg->dyn.addr->ext_blk_count = csd->extension_size;
mlk_shr_init(csa->lock_addrs[0], csd->lock_space_size, csa, (FALSE == read_only));
DEBUG_ONLY(locknl = csa->nl;) /* for DEBUG_ONLY LOCK_HIST macro */
int mur_forward_multi_proc(reg_ctl_list *rctl)
{
boolean_t multi_proc, this_reg_stuck, release_latch, ok_to_play;
boolean_t cancelled_dbsync_timer, cancelled_timer;
reg_ctl_list *rctl_top, *prev_rctl;
jnl_ctl_list *jctl;
gd_region *reg;
sgmnt_addrs *csa;
seq_num rec_token_seq;
jnl_tm_t rec_time;
enum broken_type recstat;
jnl_record *rec;
enum jnl_record_type rectype;
char errstr[256];
int i, rctl_index, save_errno, num_procs_stuck, num_reg_stuck;
uint4 status, regcnt_stuck, num_partners, start_hrtbt_cntr;
forw_multi_struct *forw_multi;
shm_forw_multi_t *sfm;
multi_struct *multi;
jnl_tm_t adjusted_resolve_time;
shm_reg_ctl_t *shm_rctl_start, *shm_rctl, *first_shm_rctl;
size_t shm_size, reccnt, copy_size;
int4 *size_ptr;
char *shmPtr; /* not using "shm_ptr" since it is already used in an AIX include file */
int shmid;
multi_proc_shm_hdr_t *mp_hdr; /* Pointer to "multi_proc_shm_hdr_t" structure in shared memory */
status = 0;
/* Although we made sure the # of tasks is the same as the # of processes forked off (in the "gtm_multi_proc"
* invocation in "mur_forward"), it is possible one of the forked process finishes one invocation of
* "mur_forward_multi_proc" before even another forked process gets assigned one task in "gtm_multi_proc_helper".
* In this case, we would be invoked more than once. But the first invocation would have done all the needed stuff
* so return for later invocations.
*/
if (mur_forward_multi_proc_done)
return 0;
mur_forward_multi_proc_done = TRUE;
/* Note: "rctl" is unused. But cannot avoid passing it since "gtm_multi_proc" expects something */
prev_rctl = NULL;
rctl_start = NULL;
adjusted_resolve_time = murgbl.adjusted_resolve_time;
assert(0 == murgbl.regcnt_remaining);
multi_proc = multi_proc_in_use; /* cache value in "local" to speed up access inside loops below */
if (multi_proc)
{
mp_hdr = multi_proc_shm_hdr;
shm_rctl_start = mur_shm_hdr->shm_rctl_start;
if (jgbl.onlnrlbk)
{
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
assert(rctl->csa->hold_onto_crit); /* would have been set in parent process */
rctl->csa->hold_onto_crit = FALSE; /* reset since we dont own this region */
assert(rctl->csa->now_crit); /* would have been set in parent process */
rctl->csa->now_crit = FALSE; /* reset since we dont own this region */
}
}
START_HEARTBEAT_IF_NEEDED; /* heartbeat timer needed later (in case not already started by "gtm_multi_proc") */
}
first_shm_rctl = NULL;
/* Phase1 of forward recovery starts */
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
/* Check if "rctl" is available for us or if some other concurrent process has taken it */
if (multi_proc)
{
rctl_index = rctl - &mur_ctl[0];
shm_rctl = &shm_rctl_start[rctl_index];
if (shm_rctl->owning_pid)
{
assert(process_id != shm_rctl->owning_pid);
continue;
}
GRAB_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
assert(release_latch);
for ( ; rctl < rctl_top; rctl++, shm_rctl++)
{
if (shm_rctl->owning_pid)
{
assert(process_id != shm_rctl->owning_pid);
continue;
}
shm_rctl->owning_pid = process_id; /* Declare ownership */
rctl->this_pid_is_owner = TRUE;
if (jgbl.onlnrlbk)
{ /* This is an online rollback and crit was grabbed on all regions by the parent rollback
* process. But this child process now owns this region and does the actual rollback on
* this region so borrow crit for the duration of this child process.
*/
csa = rctl->csa;
csa->hold_onto_crit = TRUE;
csa->now_crit = TRUE;
assert(csa->nl->in_crit == mp_hdr->parent_pid);
csa->nl->in_crit = process_id;
assert(csa->nl->onln_rlbk_pid == mp_hdr->parent_pid);
csa->nl->onln_rlbk_pid = process_id;
}
if (NULL == first_shm_rctl)
first_shm_rctl = shm_rctl;
break;
}
REL_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
if (rctl >= rctl_top)
{
assert(rctl == rctl_top);
break;
}
/* Set key to print this rctl'ss region-name as prefix in case this forked off process prints any output */
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key);
# ifdef MUR_DEBUG
fprintf(stderr, "pid = %d : Owns region %s\n", process_id, multi_proc_key);
# endif
} else
rctl->this_pid_is_owner = TRUE;
if (mur_options.forward)
{
assert(NULL == rctl->jctl_turn_around);
jctl = rctl->jctl = rctl->jctl_head;
assert(jctl->reg_ctl == rctl);
jctl->rec_offset = JNL_HDR_LEN;
jnl_fence_ctl.fence_list = JNL_FENCE_LIST_END; /* initialized to reflect journaling is not enabled */
if (mur_options.rollback)
jgbl.mur_jrec_seqno = jctl->jfh->start_seqno;
} else
{
jctl = rctl->jctl = (NULL == rctl->jctl_turn_around) ? rctl->jctl_head : rctl->jctl_turn_around;
assert(jctl->reg_ctl == rctl);
jctl->rec_offset = jctl->turn_around_offset;
jgbl.mur_jrec_seqno = jctl->turn_around_seqno;
assert((NULL != rctl->jctl_turn_around) || (0 == jctl->rec_offset));
}
if (mur_options.rollback)
{
if (murgbl.consist_jnl_seqno < jgbl.mur_jrec_seqno)
{
/* Assert that murgbl.losttn_seqno is never lesser than jgbl.mur_jrec_seqno (the turnaround
* point seqno) as this is what murgbl.consist_jnl_seqno is going to be set to and will
* eventually be the post-rollback seqno. If this condition is violated, the result of the
* recovery is a compromised database (the file header will indicate a Region Seqno which
* is not necessarily correct since seqnos prior to it might be absent in the database).
* Therefore, this is an out-of-design situation with respect to rollback and so stop it.
*/
assert(murgbl.losttn_seqno >= jgbl.mur_jrec_seqno);
murgbl.consist_jnl_seqno = jgbl.mur_jrec_seqno;
}
assert(murgbl.consist_jnl_seqno <= murgbl.losttn_seqno);
}
if (mur_options.update || mur_options.extr[GOOD_TN])
{
reg = rctl->gd;
gv_cur_region = reg;
tp_change_reg(); /* note : sets cs_addrs to non-NULL value even if gv_cur_region->open is FALSE
* (cs_data could still be NULL). */
rctl->csa = cs_addrs;
cs_addrs->miscptr = (void *)rctl;
rctl->csd = cs_data;
rctl->sgm_info_ptr = cs_addrs->sgm_info_ptr;
assert(!reg->open || (NULL != cs_addrs->dir_tree));
gv_target = cs_addrs->dir_tree;
}
jctl->after_end_of_data = FALSE;
status = mur_next(jctl, jctl->rec_offset);
assert(ERR_JNLREADEOF != status); /* cannot get EOF at start of forward processing */
if (SS_NORMAL != status)
goto finish;
PRINT_VERBOSE_STAT(jctl, "mur_forward:at the start");
rctl->process_losttn = FALSE;
/* Any multi-region TP transaction will be processed as multiple single-region TP transactions up
* until the tp-resolve-time is reached. From then on, they will be treated as one multi-region TP
* transaction. This is needed for proper lost-tn determination (any multi-region transaction that
* gets played in a region AFTER it has already encountered a broken tn should treat this as a lost tn).
*/
do
{
if (multi_proc && IS_FORCED_MULTI_PROC_EXIT(mp_hdr))
{ /* We are at a logical point. So exit if signaled by parent */
status = ERR_FORCEDHALT;
goto finish;
}
assert(jctl == rctl->jctl);
rec = rctl->mur_desc->jnlrec;
rec_time = rec->prefix.time;
if (rec_time > mur_options.before_time)
break; /* Records after -BEFORE_TIME do not go to extract or losttrans or brkntrans files */
if (rec_time < mur_options.after_time)
{
status = mur_next_rec(&jctl);
continue; /* Records before -AFTER_TIME do not go to extract or losttrans or brkntrans files */
}
if (rec_time >= adjusted_resolve_time)
break; /* Records after this adjusted resolve_time will be processed below in phase2 */
/* Note: Since we do hashtable token processing only for records from tp_resolve_time onwards,
* it is possible that if we encounter any broken transactions here we wont know they are broken
* but will play them as is. That is unavoidable. Specify -SINCE_TIME (for -BACKWARD rollback/recover)
* and -VERIFY (for -FORWARD rollback/recover) to control tp_resolve_time (and in turn more
* effective broken tn determination).
*/
status = mur_forward_play_cur_jrec(rctl);
if (SS_NORMAL != status)
break;
status = mur_next_rec(&jctl);
} while (SS_NORMAL == status);
CHECK_IF_EOF_REACHED(rctl, status); /* sets rctl->forw_eof_seen if needed; resets "status" to SS_NORMAL */
if (SS_NORMAL != status)
{ /* ERR_FILENOTCREATE is possible from "mur_cre_file_extfmt" OR ERR_FORCEDHALT is possible
* from "mur_forward_play_cur_jrec". No other errors are known to occur here. Assert accordingly.
*/
assert((ERR_FILENOTCREATE == status) || (ERR_FORCEDHALT == status));
goto finish;
}
if (rctl->forw_eof_seen)
{
PRINT_VERBOSE_STAT(jctl, "mur_forward:Reached EOF before tp_resolve_time");
continue; /* Reached EOF before even getting to tp_resolve_time.
* Do not even consider region for next processing loop */
}
rctl->last_tn = 0;
murgbl.regcnt_remaining++; /* # of regions participating in recovery at this point */
if (NULL == rctl_start)
rctl_start = rctl;
if (NULL != prev_rctl)
{
prev_rctl->next_rctl = rctl;
rctl->prev_rctl = prev_rctl;
}
prev_rctl = rctl;
assert(murgbl.ok_to_update_db || !rctl->db_updated);
PRINT_VERBOSE_STAT(jctl, "mur_forward:at tp_resolve_time");
}
if (multi_proc)
multi_proc_key = NULL; /* reset key until it can be set to rctl's region-name again */
/* Note that it is possible for rctl_start to be NULL at this point. That is there is no journal record in any region
* AFTER the calculated tp-resolve-time. This is possible if for example -AFTER_TIME was used and has a time later
* than any journal record in all journal files. If rctl_start is NULL, prev_rctl should also be NULL and vice versa.
*/
if (NULL != rctl_start)
{
assert(NULL != prev_rctl);
prev_rctl->next_rctl = rctl_start;
rctl_start->prev_rctl = prev_rctl;
}
rctl = rctl_start;
regcnt_stuck = 0; /* # of regions we are stuck in waiting for other regions to resolve a multi-region TP transaction */
assert((NULL == rctl) || (NULL == rctl->forw_multi));
gv_cur_region = NULL; /* clear out any previous value to ensure gv_cur_region/cs_addrs/cs_data
* all get set in sync by the MUR_CHANGE_REG macro below.
*/
/* Phase2 of forward recovery starts */
while (NULL != rctl)
{ /* while there is at least one region remaining with unprocessed journal records */
assert(NULL != rctl_start);
assert(0 < murgbl.regcnt_remaining);
if (NULL != rctl->forw_multi)
{ /* This region's current journal record is part of a TP transaction waiting for other regions */
regcnt_stuck++;
assert(regcnt_stuck <= murgbl.regcnt_remaining);
if (regcnt_stuck == murgbl.regcnt_remaining)
{
assertpro(multi_proc_in_use); /* Else : Out-of-design situation. Stuck in ALL regions. */
/* Check one last time if all regions are stuck waiting for another process to resolve the
* multi-region TP transaction. If so, wait in a sleep loop. If not, we can proceed.
*/
rctl = rctl_start;
start_hrtbt_cntr = heartbeat_counter;
do
{
if (IS_FORCED_MULTI_PROC_EXIT(mp_hdr))
{ /* We are at a logical point. So exit if signaled by parent */
status = ERR_FORCEDHALT;
goto finish;
}
forw_multi = rctl->forw_multi;
assert(NULL != forw_multi);
sfm = forw_multi->shm_forw_multi;
assert(NULL != sfm);
assert(sfm->num_reg_seen_forward <= sfm->num_reg_seen_backward);
# ifdef MUR_DEBUG
fprintf(stderr, "Pid = %d : Line %d : token = %llu : forward = %d : backward = %d\n",
process_id, __LINE__, (long long int)sfm->token,
sfm->num_reg_seen_forward, sfm->num_reg_seen_backward);
# endif
if (sfm->num_reg_seen_forward == sfm->num_reg_seen_backward)
{ /* We are no longer stuck in this region */
assert(!forw_multi->no_longer_stuck);
forw_multi->no_longer_stuck = TRUE;
break;
}
rctl = rctl->next_rctl; /* Move on to the next available region */
assert(NULL != rctl);
if (rctl == rctl_start)
{ /* We went through all regions once and are still stuck.
* Sleep until at leat TWO heartbeats have elapsed after which check for deadlock.
* Do this only in the child process that owns the FIRST region in the region list.
* This way we dont have contention for the GRAB_MULTI_PROC_LATCH from
* all children at more or less the same time.
*/
if ((rctl == mur_ctl) && (heartbeat_counter > (start_hrtbt_cntr + 2)))
{ /* Check if all processes are stuck for a while. If so assertpro */
GRAB_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
assert(release_latch);
shm_rctl_start = mur_shm_hdr->shm_rctl_start;
num_reg_stuck = 0;
for (i = 0; i < murgbl.reg_total; i++)
{
shm_rctl = &shm_rctl_start[i];
sfm = shm_rctl->shm_forw_multi;
if (NULL != sfm)
{
if (sfm->num_reg_seen_forward != sfm->num_reg_seen_backward)
num_reg_stuck++;
}
}
REL_MULTI_PROC_LATCH_IF_NEEDED(release_latch);
/* If everyone is stuck at this point, it is an out-of-design situation */
assertpro(num_reg_stuck < murgbl.reg_total);
start_hrtbt_cntr = heartbeat_counter;
} else
{ /* Sleep and recheck if any region we are stuck in got resolved.
* To minimize time spent sleeping, we just yield our timeslice.
*/
rel_quant();
continue;
}
}
} while (TRUE);
} else
{
rctl = rctl->next_rctl; /* Move on to the next available region */
assert(NULL != rctl);
continue;
}
}
regcnt_stuck = 0; /* restart the counter now that we found at least one non-stuck region */
MUR_CHANGE_REG(rctl);
jctl = rctl->jctl;
this_reg_stuck = FALSE;
for ( status = SS_NORMAL; SS_NORMAL == status; )
{
if (multi_proc && IS_FORCED_MULTI_PROC_EXIT(mp_hdr))
{ /* We are at a logical point. So exit if signaled by parent */
status = ERR_FORCEDHALT;
goto finish;
}
assert(jctl == rctl->jctl);
rec = rctl->mur_desc->jnlrec;
rec_time = rec->prefix.time;
if (rec_time > mur_options.before_time)
break; /* Records after -BEFORE_TIME do not go to extract or losttrans or brkntrans files */
assert((rec_time >= adjusted_resolve_time) || (mur_options.notncheck && !mur_options.verify));
assert((0 == mur_options.after_time) || (mur_options.forward && !rctl->db_updated));
if (rec_time < mur_options.after_time)
{
status = mur_next_rec(&jctl);
continue; /* Records before -AFTER_TIME do not go to extract or losttrans or brkntrans files */
}
/* Check if current journal record can be played right away or need to wait for corresponding journal
* records from other participating TP regions to be reached. A non-TP or ZTP transaction can be played
* without issues (i.e. has no dependencies with any other regions). A single-region TP transaction too
* falls in the same category. A multi-region TP transaction needs to wait until all participating regions
* have played all journal records BEFORE this TP in order to ensure recover plays records in the exact
* same order that GT.M performed them in.
*/
/* If FENCE_NONE is specified, we would not have maintained any multi hashtable in mur_back_process for
* broken transaction processing. So we process multi-region TP transactions as multiple single-region
* TP transactions in forward phase.
*/
if (FENCE_NONE != mur_options.fences)
{
rectype = (enum jnl_record_type)rec->prefix.jrec_type;
if (IS_TP(rectype) && IS_TUPD(rectype))
{
assert(IS_SET_KILL_ZKILL_ZTWORM_LGTRIG_ZTRIG(rectype));
assert(&rec->jrec_set_kill.num_participants == &rec->jrec_ztworm.num_participants);
assert(&rec->jrec_set_kill.num_participants == &rec->jrec_lgtrig.num_participants);
num_partners = rec->jrec_set_kill.num_participants;
assert(0 < num_partners);
if (1 < num_partners)
{
this_reg_stuck = TRUE;
assert(&rec->jrec_set_kill.update_num == &rec->jrec_ztworm.update_num);
assert(&rec->jrec_set_kill.update_num == &rec->jrec_lgtrig.update_num);
}
}
}
if (this_reg_stuck)
{
rec_token_seq = GET_JNL_SEQNO(rec);
MUR_FORW_TOKEN_LOOKUP(forw_multi, rec_token_seq, rec_time);
if (NULL != forw_multi)
{ /* This token has already been seen in another region in forward processing.
* Add current region as well. If all regions have been resolved, then play
* the entire transaction maintaining the exact same order of updates within.
*/
if (!forw_multi->no_longer_stuck)
MUR_FORW_TOKEN_ONE_MORE_REG(forw_multi, rctl);
} else
{ /* First time we are seeing this token in forward processing. Check if this
* has already been determined to be a broken transaction.
*/
recstat = GOOD_TN;
multi = NULL;
if (IS_REC_POSSIBLY_BROKEN(rec_time, rec_token_seq))
{
multi = MUR_TOKEN_LOOKUP(rec_token_seq, rec_time, TPFENCE);
if ((NULL != multi) && (0 < multi->partner))
recstat = BROKEN_TN;
}
MUR_FORW_TOKEN_ADD(forw_multi, rec_token_seq, rec_time, rctl, num_partners,
recstat, multi);
}
/* Check that "tabent" field has been initialized above (by either the MUR_FORW_TOKEN_LOOKUP
* or MUR_FORW_TOKEN_ADD macros). This is relied upon by "mur_forward_play_multireg_tp" below.
*/
assert(NULL != forw_multi->u.tabent);
assert(forw_multi->num_reg_seen_forward <= forw_multi->num_reg_seen_backward);
if (multi_proc)
{
sfm = forw_multi->shm_forw_multi;
ok_to_play = (NULL == sfm) || (sfm->num_reg_seen_forward == sfm->num_reg_seen_backward);
} else
ok_to_play = (forw_multi->num_reg_seen_forward == forw_multi->num_reg_seen_backward);
assert(ok_to_play || !forw_multi->no_longer_stuck);
if (ok_to_play )
{ /* We have enough information to proceed with playing this multi-region TP in
* forward processing (even if we might not have seen all needed regions). Now play it.
* Note that the TP could be BROKEN_TN or GOOD_TN. The callee handles it.
*/
assert(forw_multi == rctl->forw_multi);
status = mur_forward_play_multireg_tp(forw_multi, rctl);
this_reg_stuck = FALSE;
/* Note that as part of playing the TP transaction, we could have reached
* the EOF of rctl. In this case, we need to break out of the loop.
*/
if ((SS_NORMAL != status) || rctl->forw_eof_seen)
break;
assert(NULL == rctl->forw_multi);
assert(!dollar_tlevel);
jctl = rctl->jctl; /* In case the first record after the most recently processed
* TP transaction is in the next generation journal file */
continue;
}
break;
} else
{
status = mur_forward_play_cur_jrec(rctl);
if (SS_NORMAL != status)
break;
}
assert(!this_reg_stuck);
status = mur_next_rec(&jctl);
}
assert((NULL == rctl->forw_multi) || this_reg_stuck);
assert((NULL != rctl->forw_multi) || !this_reg_stuck);
if (!this_reg_stuck)
{ /* We are not stuck in this region (to resolve a multi-region TP).
* This means we are done processing all the records of this region.
*/
assert(NULL == rctl->forw_multi);
if (!rctl->forw_eof_seen)
{
CHECK_IF_EOF_REACHED(rctl, status);
/* sets rctl->forw_eof_seen if needed; resets "status" to SS_NORMAL */
if (SS_NORMAL != status)
{
assert(ERR_FILENOTCREATE == status);
goto finish;
}
assert(!dollar_tlevel);
DELETE_RCTL_FROM_UNPROCESSED_LIST(rctl); /* since all of its records should have been processed */
} else
{ /* EOF was seen in rctl inside "mur_forward_play_multireg_tp" and it was removed
* from the unprocessed list of rctls. At the time rctl was removed, its "next_rctl"
* field could have been pointing to another <rctl> that has since then also been
* removed inside the same function. Therefore the "next_rctl" field is not reliable
* in this case but instead we should rely on the global variable "rctl_start" which
* points to the list of unprocessed rctls. Set "next_rctl" accordingly.
*/
rctl->next_rctl = rctl_start;
if (ERR_JNLREADEOF == status)
status = SS_NORMAL;
}
assert(rctl->deleted_from_unprocessed_list);
}
assert(SS_NORMAL == status);
assert(!this_reg_stuck || !rctl->forw_eof_seen);
assert((NULL == rctl->next_rctl) || (NULL != rctl_start));
assert((NULL == rctl->next_rctl) || (0 < murgbl.regcnt_remaining));
rctl = rctl->next_rctl; /* Note : even though "rctl" could have been deleted from the doubly linked list above,
* rctl->next_rctl is not touched so we can still use it to get to the next element. */
}
assert(0 == murgbl.regcnt_remaining);
jgbl.mur_pini_addr_reset_fnptr = NULL; /* No more simulation of GT.M activity for any region */
prc_vec = murgbl.prc_vec; /* Use process-vector of MUPIP RECOVER (not any simulating GT.M process) now onwards */
assert(0 == dollar_tlevel);
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++)
{
if (!rctl->this_pid_is_owner)
{
assert(multi_proc_in_use);
continue; /* in a parallel processing environment, process only regions we own */
}
if (multi_proc)
{ /* Set key to print this rctl's region-name as prefix in case this forked off process prints any output */
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key);
}
PRINT_VERBOSE_STAT(rctl->jctl, "mur_forward:at the end");
assert(!mur_options.rollback || (0 != murgbl.consist_jnl_seqno));
assert(mur_options.rollback || (0 == murgbl.consist_jnl_seqno));
assert(!dollar_tlevel); /* In case it applied a broken TUPD */
assert(murgbl.ok_to_update_db || !rctl->db_updated);
rctl->mur_plst = NULL; /* reset now that simulation of GT.M updates is done */
/* Ensure mur_block_count_correct is called if updates allowed */
if (murgbl.ok_to_update_db && (SS_NORMAL != mur_block_count_correct(rctl)))
{
gtm_putmsg_csa(CSA_ARG(rctl->csa) VARLSTCNT(4) ERR_BLKCNTEDITFAIL, 2, DB_LEN_STR(rctl->gd));
murgbl.wrn_count++;
}
}
finish:
if (multi_proc)
multi_proc_key = NULL; /* reset key until it can be set to rctl's region-name again */
if ((SS_NORMAL == status) && mur_options.show)
mur_output_show();
if (NULL != first_shm_rctl)
{ /* Transfer needed process-private information to shared memory so parent process can later inherit this. */
first_shm_rctl->err_cnt = murgbl.err_cnt;
first_shm_rctl->wrn_count = murgbl.wrn_count;
first_shm_rctl->consist_jnl_seqno = murgbl.consist_jnl_seqno;
/* If extract files were created by this process for one or more regions, then copy that information to
* shared memory so parent process can use this information to do a merge sort.
*/
shm_rctl = mur_shm_hdr->shm_rctl_start;
for (rctl = mur_ctl, rctl_top = mur_ctl + murgbl.reg_total; rctl < rctl_top; rctl++, shm_rctl++)
{
assert(multi_proc_in_use);
if (!rctl->this_pid_is_owner)
continue; /* in a parallel processing environment, process only regions we own */
/* Cancel any flush/dbsync timers by this child process for this region. This is because the
* child is not going to go through exit handling code (no gds_rundown etc.). And we need to
* clear up csa->nl->wcs_timers. (normally done by gds_rundown).
*/
if (NULL != rctl->csa) /* rctl->csa can be NULL in case of "mupip journal -extract" etc. */
CANCEL_DB_TIMERS(rctl->gd, rctl->csa, cancelled_timer, cancelled_dbsync_timer);
reccnt = 0;
for (size_ptr = &rctl->jnlext_multi_list_size[0], recstat = 0;
recstat < TOT_EXTR_TYPES;
recstat++, size_ptr++)
{ /* Assert "extr_file_created" information is in sync between rctl and shm_rctl.
* This was done at the end of "mur_cre_file_extfmt".
*/
assert(shm_rctl->extr_file_created[recstat] == rctl->extr_file_created[recstat]);
/* Assert that if *size_ptr is non-zero, then we better have created an extract file.
* Note that the converse is not true. It is possible we created a file for example to
* write an INCTN record but decided to not write anything because it was not a -detail
* type of extract. So *sizeptr could be 0 even though we created the extract file.
*/
assert(!*size_ptr || rctl->extr_file_created[recstat]);
shm_rctl->jnlext_list_size[recstat] = *size_ptr;
reccnt += *size_ptr;
}
assert(INVALID_SHMID == shm_rctl->jnlext_shmid);
shm_size = reccnt * SIZEOF(jnlext_multi_t);
/* If we are quitting because of an abnormal status OR a forced signal to terminate
* OR if the parent is dead (kill -9) dont bother creating shmid to communicate back with parent.
*/
if (mp_hdr->parent_pid != getppid())
{
SET_FORCED_MULTI_PROC_EXIT; /* Also signal sibling children to stop processing */
if (SS_NORMAL != status)
status = ERR_FORCEDHALT;
}
if ((SS_NORMAL == status) && shm_size)
{
shmid = shmget(IPC_PRIVATE, shm_size, 0600 | IPC_CREAT);
if (-1 == shmid)
{
save_errno = errno;
SNPRINTF(errstr, SIZEOF(errstr),
"shmget() : shmsize=0x%llx", shm_size);
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key); /* to print region name prefix */
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8)
ERR_SYSCALL, 5, LEN_AND_STR(errstr), CALLFROM, save_errno);
}
shmPtr = (char *)do_shmat(shmid, 0, 0);
if (-1 == (sm_long_t)shmPtr)
{
save_errno = errno;
SNPRINTF(errstr, SIZEOF(errstr),
"shmat() : shmid=%d shmsize=0x%llx", shmid, shm_size);
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key); /* to print region name prefix */
rts_error_csa(CSA_ARG(NULL) VARLSTCNT(8)
ERR_SYSCALL, 5, LEN_AND_STR(errstr), CALLFROM, save_errno);
}
shm_rctl->jnlext_shmid = shmid;
shm_rctl->jnlext_shm_size = shm_size;
for (size_ptr = &rctl->jnlext_multi_list_size[0], recstat = 0;
recstat < TOT_EXTR_TYPES;
recstat++, size_ptr++)
{
shm_size = *size_ptr;
if (shm_size)
{
copy_size = copy_list_to_buf(rctl->jnlext_multi_list[recstat],
(int4)shm_size, shmPtr);
assert(copy_size == (shm_size * SIZEOF(jnlext_multi_t)));
shmPtr += copy_size;
}
}
}
}
}
mur_close_file_extfmt(IN_MUR_CLOSE_FILES_FALSE); /* Need to flush buffered extract/losttrans/brokentrans files */
return (int)status;
}
/* Takes an entry from 'ipcs -am' and checks for its validity to be a GT.M replication segment.
* Returns TRUE if the shared memory segment is a valid GT.M replication segment
* (based on a check on some fields in the shared memory) else FALSE.
* If the segment belongs to GT.M, it returns the replication id of the segment
* by the second argument.
* Sets exit_stat to ERR_MUNOTALLSEC if appropriate.
*/
boolean_t validate_replpool_shm_entry(shm_parms *parm_buff, replpool_id_ptr_t replpool_id, int *exit_stat)
{
boolean_t remove_shmid, jnlpool_segment;
int fd;
repl_inst_hdr repl_instance;
sm_uc_ptr_t start_addr;
int save_errno, status, shmid;
struct shmid_ds shmstat;
char msgbuff[OUT_BUFF_SIZE], *instfilename;
if (NULL == parm_buff)
return FALSE;
/* Check for the bare minimum size of the replic shared segment that we expect */
/* if (parm_buff->sgmnt_siz < (SIZEOF(replpool_identifier) + MIN(MIN_JNLPOOL_SIZE, MIN_RECVPOOL_SIZE))) */
if (parm_buff->sgmnt_siz < MIN(MIN_JNLPOOL_SIZE, MIN_RECVPOOL_SIZE))
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;
memcpy((void *)replpool_id, (void *)start_addr, SIZEOF(replpool_identifier));
instfilename = replpool_id->instfilename;
/* Even though we could be looking at a replication pool structure that has been created by an older version
* or newer version of GT.M, the format of the "replpool_identifier" structure is expected to be the same
* across all versions so we can safely dereference the "label" and "instfilename" fields in order to generate
* user-friendly error messages. Asserts for the layout are in "mu_rndwn_repl_instance" (not here) with a
* comment there as to why that location was chosen.
*/
if (memcmp(replpool_id->label, GDS_RPL_LABEL, GDS_LABEL_SZ - 1))
{
if (!memcmp(replpool_id->label, GDS_RPL_LABEL, GDS_LABEL_SZ - 3))
{
util_out_print("Cannot rundown replpool shmid = !UL as it has format !AD "
"created by !AD but this mupip is version and uses format !AD",
TRUE, shmid, GDS_LABEL_SZ - 1, replpool_id->label,
LEN_AND_STR(replpool_id->now_running), gtm_release_name_len, gtm_release_name,
GDS_LABEL_SZ - 1, GDS_RPL_LABEL);
*exit_stat = ERR_MUNOTALLSEC;
}
shmdt((void *)start_addr);
return FALSE;
}
assert(JNLPOOL_SEGMENT == replpool_id->pool_type || RECVPOOL_SEGMENT == replpool_id->pool_type);
if(JNLPOOL_SEGMENT != replpool_id->pool_type && RECVPOOL_SEGMENT != replpool_id->pool_type)
{
shmdt((void *)start_addr);
return FALSE;
}
jnlpool_segment = (JNLPOOL_SEGMENT == replpool_id->pool_type);
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, LEN_AND_STR(instfilename), shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
/* Check if instance filename reported in shared memory still exists. If not, clean this
* shared memory section without even invoking "mu_rndwn_repl_instance" as that expects
* the instance file to exist. Same case if shared memory points back to an instance file
* whose file header does not have this shmid.
*/
OPENFILE(instfilename, O_RDONLY, fd); /* check if we can open it */
msgbuff[0] = '\0';
remove_shmid = FALSE;
if (FD_INVALID == fd)
{
if (ENOENT == errno)
{
SNPRINTF(msgbuff, OUT_BUFF_SIZE, "File %s does not exist", instfilename);
if (1 < shmstat.shm_nattch)
{
PRINT_AND_SEND_REPLPOOL_FAILURE_MSG(msgbuff, replpool_id, shmid);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = TRUE;
} else
{ /* open() errored out e.g. due to file permissions. Log that */
save_errno = errno;
util_out_print("Cannot rundown replpool shmid !UL for instance file"
" !AD as open() on the file returned the following error",
TRUE, shmid, LEN_AND_STR(instfilename));
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
} else
{
LSEEKREAD(fd, 0, &repl_instance, SIZEOF(repl_inst_hdr), status);
if (0 != status)
{
save_errno = errno;
util_out_print("!AD -> Error with LSEEKREAD for shmid = !UL", TRUE,
LEN_AND_STR(instfilename), shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
if ((jnlpool_segment && (repl_instance.jnlpool_shmid != shmid))
|| (!jnlpool_segment && (repl_instance.recvpool_shmid != shmid)))
{
SNPRINTF(msgbuff, OUT_BUFF_SIZE, "%s SHMID (%d) in the instance file header does not match with the"
" one reported by \"ipcs\" command (%d)", jnlpool_segment ? "Journal Pool" : "Receive Pool",
jnlpool_segment ? repl_instance.jnlpool_shmid : repl_instance.recvpool_shmid, shmid);
if (1 < shmstat.shm_nattch)
{
PRINT_AND_SEND_REPLPOOL_FAILURE_MSG(msgbuff, replpool_id, shmid);
*exit_stat = ERR_MUNOTALLSEC;
shmdt((void *)start_addr);
return FALSE;
}
remove_shmid = TRUE;
}
CLOSEFILE_RESET(fd, status); /* resets "fd" to FD_INVALID */
}
shmdt((void *)start_addr);
if (remove_shmid)
{
assert('\0' != msgbuff[0]);
if (0 != shm_rmid(shmid))
{
save_errno = errno;
util_out_print("!AD -> Error removing shared memory for shmid = !UL",
TRUE, LEN_AND_STR(instfilename), shmid);
gtm_putmsg_csa(CSA_ARG(NULL) VARLSTCNT(1) save_errno);
*exit_stat = ERR_MUNOTALLSEC;
return FALSE;
}
PRINT_AND_SEND_SHMREMOVED_MSG(msgbuff, STRLEN(instfilename), instfilename, shmid);
*exit_stat = ERR_SHMREMOVED;
} else
*exit_stat = SS_NORMAL;
return TRUE;
}
/* 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;
}
SYSCALL_DEFINE6(ipc, unsigned int, call, int, first, unsigned long, second,
unsigned long, third, void __user *, ptr, long, fifth)
{
int version, ret;
version = call >> 16; /* hack for backward compatibility */
call &= 0xffff;
switch (call) {
case SEMOP:
return sys_semtimedop(first, (struct sembuf __user *)ptr,
second, NULL);
case SEMTIMEDOP:
return sys_semtimedop(first, (struct sembuf __user *)ptr,
second,
(const struct timespec __user *)fifth);
case SEMGET:
return sys_semget(first, second, third);
case SEMCTL: {
unsigned long arg;
if (!ptr)
return -EINVAL;
if (get_user(arg, (unsigned long __user *) ptr))
return -EFAULT;
return sys_semctl(first, second, third, arg);
}
case MSGSND:
return sys_msgsnd(first, (struct msgbuf __user *) ptr,
second, third);
case MSGRCV:
switch (version) {
case 0: {
struct ipc_kludge tmp;
if (!ptr)
return -EINVAL;
if (copy_from_user(&tmp,
(struct ipc_kludge __user *) ptr,
sizeof(tmp)))
return -EFAULT;
return sys_msgrcv(first, tmp.msgp, second,
tmp.msgtyp, third);
}
default:
return sys_msgrcv(first,
(struct msgbuf __user *) ptr,
second, fifth, third);
}
case MSGGET:
return sys_msgget((key_t) first, second);
case MSGCTL:
return sys_msgctl(first, second, (struct msqid_ds __user *)ptr);
case SHMAT:
switch (version) {
default: {
unsigned long raddr;
ret = do_shmat(first, (char __user *)ptr,
second, &raddr, SHMLBA);
if (ret)
return ret;
return put_user(raddr, (unsigned long __user *) third);
}
case 1:
/*
* This was the entry point for kernel-originating calls
* from iBCS2 in 2.2 days.
*/
return -EINVAL;
}
case SHMDT:
return sys_shmdt((char __user *)ptr);
case SHMGET:
return sys_shmget(first, second, third);
case SHMCTL:
return sys_shmctl(first, second,
(struct shmid_ds __user *) ptr);
default:
return -ENOSYS;
}
}
/*
* sys_ipc() is the de-multiplexer for the SysV IPC calls..
*
* This is really horribly ugly.
*/
asmlinkage int sys_ipc (uint call, int first, int second,
unsigned long third, void *ptr, long fifth)
{
int version, ret;
version = call >> 16; /* hack for backward compatibility */
call &= 0xffff;
switch (call) {
case SEMOP:
return sys_semtimedop (first, (struct sembuf *)ptr, second,
NULL);
case SEMTIMEDOP:
return sys_semtimedop (first, (struct sembuf *)ptr, second,
(const struct timespec __user *)fifth);
case SEMGET:
return sys_semget (first, second, third);
case SEMCTL: {
union semun fourth;
if (!ptr)
return -EINVAL;
if (get_user(fourth.__pad, (void **) ptr))
return -EFAULT;
return sys_semctl (first, second, third, fourth);
}
case MSGSND:
return sys_msgsnd (first, (struct msgbuf *) ptr,
second, third);
case MSGRCV:
switch (version) {
case 0: {
struct ipc_kludge tmp;
if (!ptr)
return -EINVAL;
if (copy_from_user(&tmp,
(struct ipc_kludge *) ptr,
sizeof (tmp)))
return -EFAULT;
return sys_msgrcv (first, tmp.msgp, second,
tmp.msgtyp, third);
}
default:
return sys_msgrcv (first,
(struct msgbuf *) ptr,
second, fifth, third);
}
case MSGGET:
return sys_msgget ((key_t) first, second);
case MSGCTL:
return sys_msgctl (first, second, (struct msqid_ds *) ptr);
case SHMAT:
switch (version) {
default: {
ulong raddr;
ret = do_shmat (first, (char *) ptr, second, &raddr);
if (ret)
return ret;
return put_user (raddr, (ulong *) third);
}
case 1: /* iBCS2 emulator entry point */
if (!segment_eq(get_fs(), get_ds()))
return -EINVAL;
return do_shmat (first, (char *) ptr, second, (ulong *) third);
}
case SHMDT:
return sys_shmdt ((char *)ptr);
case SHMGET:
return sys_shmget (first, second, third);
case SHMCTL:
return sys_shmctl (first, second,
(struct shmid_ds *) ptr);
default:
return -ENOSYS;
}
}
