int jnl_v11tov15(uchar_ptr_t jnl_buff, uint4 *jnl_len, uchar_ptr_t conv_buff, uint4 *conv_len, uint4 conv_bufsiz)
{
/* Convert a transaction from jnl version 11 (V4.2-002) to 15 (V.4.4-002) */
unsigned char *jb, *cb, *cstart, *jstart, rectype;
int status, reclen;
unsigned short key_len;
unsigned int long_data_len, jlen, total_data, nzeros, conv_reclen, clen_without_sfx, total_key;
jrec_prefix prefix;
jrec_suffix suffix;
seq_num jsno;
jb = jnl_buff;
cb = conv_buff;
status = SS_NORMAL;
jlen = *jnl_len;
while (0 < jlen)
{
if (0 < (reclen = v11_jnl_record_length((jnl_record *)jb, jlen)))
{
if (reclen <= jlen)
{
rectype = REF_CHAR(jb + V11_JREC_TYPE_OFFSET);
total_key = total_data = 0;
assert(IS_REPLICATED(rectype));
if (IS_ZTP(rectype))
GTMASSERT; /* ZTP not supported */
if (IS_SET_KILL_ZKILL(rectype))
{
GET_USHORT(key_len, jb + V11_JREC_PREFIX_SIZE + v11_jnl_fixed_size[rectype]);
total_key = key_len + sizeof(unsigned short);
if (IS_SET(rectype))
{
GET_MSTR_LEN(long_data_len, jb + V11_JREC_PREFIX_SIZE +
v11_jnl_fixed_size[rectype] + total_key);
total_data = long_data_len + sizeof(mstr_len_t);
}
conv_reclen = JREC_PREFIX_SIZE + FIXED_UPD_RECLEN +
total_key + total_data + JREC_SUFFIX_SIZE;
conv_reclen = ROUND_UP2(conv_reclen, JNL_REC_START_BNDRY);
} else if (IS_COM(rectype))
conv_reclen = JREC_PREFIX_SIZE + TCOM_RECLEN + JREC_SUFFIX_SIZE;
clen_without_sfx = conv_reclen - JREC_SUFFIX_SIZE;
if (cb - conv_buff + conv_reclen > conv_bufsiz)
{
repl_errno = EREPL_INTLFILTER_NOSPC;
status = -1;
break;
}
cstart = cb;
jstart = jb;
prefix.jrec_type = rectype;
suffix.backptr = prefix.forwptr = conv_reclen;
prefix.pini_addr = 0;
prefix.time = 0;
prefix.tn = 0;
suffix.suffix_code = JNL_REC_SUFFIX_CODE;
memcpy(cb, (unsigned char*)&prefix, JREC_PREFIX_SIZE);
cb += JREC_PREFIX_SIZE;
memcpy(cb, jb + V11_JREC_PREFIX_SIZE + V11_JNL_SEQNO_OFFSET, sizeof(seq_num));
cb += sizeof(seq_num);
if (IS_SET_KILL_ZKILL(rectype))
{
PUT_JNL_STR_LEN(cb, key_len);
jb += (V11_JREC_PREFIX_SIZE + V11_MUMPS_NODE_OFFSET + sizeof(unsigned short));
if (IS_FENCED(rectype))
jb += TP_TOKEN_TID_SIZE;
cb += sizeof(jnl_str_len_t);
memcpy(cb, jb, key_len);
cb += key_len;
jb += key_len;
if (IS_SET(rectype))
{
PUT_MSTR_LEN(cb, long_data_len);
cb += sizeof(mstr_len_t);
jb += sizeof(mstr_len_t);
memcpy(cb, jb, long_data_len);
cb += long_data_len;
}
} else if (IS_COM(rectype))
{
assert(JRT_TCOM == rectype);
memset(cb, 0, TID_STR_SIZE);
cb += TID_STR_SIZE;
memcpy(cb, jb + V11_JREC_PREFIX_SIZE + V11_TCOM_PARTICIPANTS_OFFSET, sizeof(uint4));
cb += sizeof(uint4);
} else
assert(FALSE);
nzeros = (cstart + clen_without_sfx - cb);
if (nzeros > 0)
{
memset(cb, 0, nzeros);
cb += nzeros;
}
jb = jstart + reclen;
memcpy(cb, (unsigned char*)&suffix, JREC_SUFFIX_SIZE);
cb += JREC_SUFFIX_SIZE;
assert(cb == cstart + conv_reclen);
jlen -= reclen;
continue;
}
repl_errno = EREPL_INTLFILTER_INCMPLREC;
status = -1;
break;
}
repl_errno = EREPL_INTLFILTER_BADREC;
status = -1;
break;
}
assert(0 == jlen || -1 == status);
*jnl_len = jb - jnl_buff;
*conv_len = cb - conv_buff;
return(status);
}
uint4 mur_forward_play_cur_jrec(reg_ctl_list *rctl)
{
boolean_t process_losttn;
boolean_t is_set_kill_zkill_ztworm_lgtrig_ztrig, is_set_kill_zkill_ztrig;
trans_num curr_tn;
enum jnl_record_type rectype;
enum rec_fence_type rec_fence;
enum broken_type recstat;
jnl_tm_t rec_time;
uint4 status;
mval mv;
seq_num rec_token_seq, rec_strm_seqno, resync_strm_seqno;
jnl_record *rec;
jnl_string *keystr;
multi_struct *multi;
jnl_ctl_list *jctl;
ht_ent_mname *tabent;
mname_entry gvent;
gvnh_reg_t *gvnh_reg;
pini_list_struct *plst;
int4 gtmcrypt_errno;
boolean_t use_new_key;
forw_multi_struct *forw_multi;
# if (defined(DEBUG) && defined(UNIX))
int4 strm_idx;
# endif
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(!rctl->forw_eof_seen);
if (multi_proc_in_use)
{ /* Set key to print this rctl's region-name as prefix in case this forked off process prints any output.
* e.g. If this function ends up calling t_end/op_tcommit which in turn needs to do a jnl autoswitch
* inside jnl_file_extend and prints a GTM-I-FILERENAME message.
*/
MUR_SET_MULTI_PROC_KEY(rctl, multi_proc_key);
}
jctl = rctl->jctl;
/* Ensure we never DOUBLE process the same journal record in the forward phase */
assert((jctl != rctl->last_processed_jctl) || (jctl->rec_offset != rctl->last_processed_rec_offset));
# ifdef DEBUG
rctl->last_processed_jctl = jctl;
rctl->last_processed_rec_offset = jctl->rec_offset;
# endif
rec = rctl->mur_desc->jnlrec;
rectype = (enum jnl_record_type)rec->prefix.jrec_type;
rec_time = rec->prefix.time;
assert(rec_time <= mur_options.before_time);
assert(rec_time >= mur_options.after_time);
assert((0 == mur_options.after_time) || (mur_options.forward && !rctl->db_updated));
is_set_kill_zkill_ztworm_lgtrig_ztrig = (boolean_t)(IS_SET_KILL_ZKILL_ZTWORM_LGTRIG_ZTRIG(rectype));
if (is_set_kill_zkill_ztworm_lgtrig_ztrig)
{
keystr = (jnl_string *)&rec->jrec_set_kill.mumps_node;
if (USES_ANY_KEY(jctl->jfh))
{
use_new_key = USES_NEW_KEY(jctl->jfh);
assert(NEEDS_NEW_KEY(jctl->jfh, rec->prefix.tn) == use_new_key);
MUR_DECRYPT_LOGICAL_RECS(
keystr,
(use_new_key ? TRUE : jctl->jfh->non_null_iv),
rec->prefix.forwptr,
(use_new_key ? jctl->encr_key_handle2 : jctl->encr_key_handle),
gtmcrypt_errno);
if (0 != gtmcrypt_errno)
{
GTMCRYPT_REPORT_ERROR(gtmcrypt_errno, gtm_putmsg, jctl->jnl_fn_len, jctl->jnl_fn);
return gtmcrypt_errno;
}
}
}
if (mur_options.selection && !mur_select_rec(jctl))
return SS_NORMAL;
rec_token_seq = (REC_HAS_TOKEN_SEQ(rectype)) ? GET_JNL_SEQNO(rec) : 0;
process_losttn = rctl->process_losttn;
if (!process_losttn && mur_options.rollback)
{
if (IS_REPLICATED(rectype) && (rec_token_seq >= murgbl.losttn_seqno))
process_losttn = rctl->process_losttn = TRUE;
# if (defined(UNIX) && defined(DEBUG))
if ((rec_token_seq < murgbl.losttn_seqno) && murgbl.resync_strm_seqno_nonzero && IS_REPLICATED(rectype))
{
assert(IS_SET_KILL_ZKILL_ZTWORM_LGTRIG_ZTRIG(rectype) || IS_COM(rectype) || (JRT_NULL == (rectype)));
assert(&rec->jrec_set_kill.strm_seqno == &rec->jrec_null.strm_seqno);
assert(&rec->jrec_set_kill.strm_seqno == &rec->jrec_tcom.strm_seqno);
rec_strm_seqno = GET_STRM_SEQNO(rec);
if (rec_strm_seqno)
{
strm_idx = GET_STRM_INDEX(rec_strm_seqno);
rec_strm_seqno = GET_STRM_SEQ60(rec_strm_seqno);
resync_strm_seqno = murgbl.resync_strm_seqno[strm_idx];
assert(!resync_strm_seqno || (rec_strm_seqno < resync_strm_seqno));
}
}
# endif
}
/* Note: Broken transaction determination is done below only based on the records that got selected as
* part of the mur_options.selection criteria. Therefore depending on whether a broken transaction gets
* selected or not, future complete transactions might either go to the lost transaction or extract file.
*/
recstat = process_losttn ? LOST_TN : GOOD_TN;
status = SS_NORMAL;
if (FENCE_NONE != mur_options.fences)
{
if (IS_FENCED(rectype))
{
assert(rec_token_seq);
# ifdef DEBUG
/* assert that all TP records before min_broken_time are not broken */
if (IS_TP(rectype) && ((!mur_options.rollback && rec_time < murgbl.min_broken_time)
|| (mur_options.rollback && rec_token_seq < murgbl.min_broken_seqno)))
{
rec_fence = GET_REC_FENCE_TYPE(rectype);
if (NULL != (multi = MUR_TOKEN_LOOKUP(rec_token_seq, rec_time, rec_fence)))
{
assert(0 == multi->partner);
assert(FALSE == multi->this_is_broken);
}
}
# endif
/* In most cases, the fact whether a TP tn is broken or not would have been determined already in
* mur_forward. In this case, rctl->forw_multi would be set appropriately. So use that to get to
* "multi" and avoid a hashtable lookup. If forw_multi is NULL (e.g. for ZTP or single-region TP),
* the hash-table lookup cannot be avoided.
*/
multi = NULL;
forw_multi = rctl->forw_multi;
if (NULL != forw_multi)
{
multi = forw_multi->multi;
/* Always honor the "recstat" from the forw_multi since that has been determined taking into
* consideration the BROKEN_TN status of ALL participating regions.
*/
assert((GOOD_TN != forw_multi->recstat) || (GOOD_TN == recstat));
recstat = forw_multi->recstat;
} else if (IS_REC_POSSIBLY_BROKEN(rec_time, rec_token_seq))
{
assert(!mur_options.rollback || process_losttn);
rec_fence = GET_REC_FENCE_TYPE(rectype);
assert(rec_token_seq == ((struct_jrec_upd *)rec)->token_seq.token);
multi = MUR_TOKEN_LOOKUP(rec_token_seq, rec_time, rec_fence);
if ((NULL != multi) && (0 < multi->partner))
{
process_losttn = rctl->process_losttn = TRUE;
recstat = BROKEN_TN;
}
}
/* Check that if the hashtable reports a tn as GOOD, it better have had the same
* # of participants in the TCOM records across all the participating regions.
*/
assert((NULL == multi) || (BROKEN_TN == recstat) || (FALSE == multi->this_is_broken));
} else if ((FENCE_ALWAYS == mur_options.fences) && is_set_kill_zkill_ztworm_lgtrig_ztrig)
{
process_losttn = rctl->process_losttn = TRUE;
recstat = BROKEN_TN;
}
} else
forw_multi = NULL;
if (mur_options.show)
{
assert(SS_NORMAL == status);
if (BROKEN_TN != recstat)
{
if (JRT_PFIN == rectype)
status = mur_pini_state(jctl, rec->prefix.pini_addr, FINISHED_PROC);
else if ((JRT_EOF != rectype)
&& ((JRT_ALIGN != rectype) || (JNL_HDR_LEN != rec->prefix.pini_addr)))
{ /* Note that it is possible that we have a PINI record followed by a PFIN record
* and later an ALIGN record with the pini_addr pointing to the original PINI
* record (see comment in jnl_write.c where pini_addr gets assigned to JNL_HDR_LEN)
* In this case we do not want the ALIGN record to cause the process to become
* ACTIVE although it has written a PFIN record. Hence the check above.
*/
status = mur_pini_state(jctl, rec->prefix.pini_addr, ACTIVE_PROC);
}
} else
status = mur_pini_state(jctl, rec->prefix.pini_addr, BROKEN_PROC);
if (SS_NORMAL != status)
return status; /* "mur_pini_state" failed due to bad pini_addr */
++jctl->jnlrec_cnt[rectype]; /* for -show=STATISTICS */
}
if (!mur_options.update && !jgbl.mur_extract)
return SS_NORMAL;
if (murgbl.ok_to_update_db && IS_TUPD(rectype) && (GOOD_TN == recstat))
{ /* Even for FENCE_NONE we apply fences. Otherwise a TUPD becomes UPD etc.
* If forw_multi is non-NULL, a multi-region TP transaction is being played as a SINGLE
* TP transaction across all the involved regions. Therefore only ONE op_tstart is done
* even though more than one TSET might be encountered. In this case, do not issue JNLTPNEST error.
*/
if (dollar_tlevel && (NULL == forw_multi))
{
assert(FALSE);
murgbl.wrn_count++;
gtm_putmsg_csa(CSA_ARG(rctl->csa) VARLSTCNT(6) ERR_JNLTPNEST, 4, jctl->jnl_fn_len,
jctl->jnl_fn, jctl->rec_offset, &rec->prefix.tn);
OP_TROLLBACK(0);
}
if (!dollar_tlevel)
{ /* Note: op_tstart resets gv_currkey. So set gv_currkey later. */
/* mv is used to determine transaction id. But it is ignored by recover/rollback */
mv.mvtype = MV_STR;
mv.str.len = 0;
mv.str.addr = NULL;
op_tstart(IMPLICIT_TSTART, TRUE, &mv, -1);
DEBUG_ONLY(jgbl.max_tp_ztp_jnl_upd_num = 0;)
}
tp_set_sgm(); /* needed to set "sgm_info_ptr" to correspond to "rctl" */
}
/* This routine is called only for recover and rollback (that is, mur_options.update).
* It applies the set/kill/zkill, tcom, inctn, and aimg records during forward processing.
* Some fields like jnl_seqno, rec_seqno and prefix.time are saved here from original journal files.
* Later jnl_write routines copies them to journal records instead of generating them like the runtime system */
uint4 mur_output_record(reg_ctl_list *rctl)
{
mval mv;
jnl_record *rec;
char *val_ptr;
int strm_num;
uint4 dummy;
off_jnl_t pini_addr;
jnl_string *keystr;
enum jnl_record_type rectype;
uint4 jnl_status, status;
pini_list_struct *plst;
boolean_t jnl_enabled, was_crit;
struct_jrec_null null_record;
gd_region *reg;
seq_num strm_seqno;
sgmnt_addrs *csa;
sgmnt_data_ptr_t csd;
jnl_ctl_list *jctl;
jnl_format_buffer *ztworm_jfb;
blk_hdr_ptr_t aimg_blk_ptr;
int in_len, gtmcrypt_errno;
boolean_t use_new_key;
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
assert(mur_options.update);
rec = rctl->mur_desc->jnlrec;
rectype = (enum jnl_record_type)rec->prefix.jrec_type;
switch (rectype)
{
case JRT_ALIGN:
case JRT_EOF:
case JRT_EPOCH:
case JRT_PBLK:
case JRT_PINI:
case JRT_TRUNC:
return SS_NORMAL;
break;
default:
break;
}
jgbl.gbl_jrec_time = rec->prefix.time;
pini_addr = rec->prefix.pini_addr;
reg = rctl->gd;
jctl = rctl->jctl;
assert(jctl->reg_ctl == rctl);
assert(gv_cur_region == reg);
csa = rctl->csa;
assert(cs_addrs == csa);
csd = csa->hdr;
assert(cs_data == csd);
jnl_enabled = JNL_ENABLED(csa);
if (jnl_enabled)
{
status = mur_get_pini(jctl, pini_addr, &plst);
if (SS_NORMAL != status)
return status;
prc_vec = &plst->jpv;
csa->jnl->pini_addr = plst->new_pini_addr;
rctl->mur_plst = plst;
}
if (mur_options.rollback && IS_REPLICATED(rectype))
{
jgbl.mur_jrec_seqno = GET_JNL_SEQNO(rec);
if (jgbl.mur_jrec_seqno >= murgbl.consist_jnl_seqno)
{
assert(murgbl.losttn_seqno >= (jgbl.mur_jrec_seqno + 1));
murgbl.consist_jnl_seqno = jgbl.mur_jrec_seqno + 1;
}
jgbl.mur_jrec_strm_seqno = GET_STRM_SEQNO(rec);
strm_seqno = jgbl.mur_jrec_strm_seqno;
if (strm_seqno)
{ /* maintain csd->strm_reg_seqno */
strm_num = GET_STRM_INDEX(strm_seqno);
strm_seqno = GET_STRM_SEQ60(strm_seqno);
assert(csd->strm_reg_seqno[strm_num] <= (strm_seqno + 1));
csd->strm_reg_seqno[strm_num] = strm_seqno + 1;
}
}
/* Assert that TREF(gd_targ_gvnh_reg) is NULL for every update that journal recovery/rollback plays forward;
* This is necessary to ensure every update is played in only the database file where the journal record is seen
* instead of across all regions that span the particular global reference. For example if ^a(1) spans db files
* a.dat and b.dat, and a KILL ^a(1) is done at the user level, we would see KILL ^a(1) journal records in a.mjl
* and b.mjl. When journal recovery processes the journal record in a.mjl, it should do the kill only in a.dat
* When it gets to the same journal record in b.mjl, it would do the same kill in b.dat and effectively complete
* the user level KILL ^a(1). If instead recovery does the KILL across all spanned regions, we would be basically
* doing duplicate work let alone do it out-of-order since recovery goes region by region for the most part.
*/
assert(NULL == TREF(gd_targ_gvnh_reg));
if (IS_SET_KILL_ZKILL_ZTRIG(rectype))
{ /* TP and non-TP has same format */
keystr = (jnl_string *)&rec->jrec_set_kill.mumps_node;
if (jnl_enabled)
{
MUR_SET_JNL_FENCE_CTL_TOKEN(rec->jrec_set_kill.token_seq.token, rctl);
jnl_fence_ctl.strm_seqno = rec->jrec_set_kill.strm_seqno;
jgbl.tp_ztp_jnl_upd_num = rec->jrec_set_kill.update_num;
DEBUG_ONLY(jgbl.max_tp_ztp_jnl_upd_num = MAX(jgbl.max_tp_ztp_jnl_upd_num, jgbl.tp_ztp_jnl_upd_num);)
jgbl.mur_jrec_nodeflags = keystr->nodeflags;
}
if (IS_FENCED(rectype))
{ /* Even for FENCE_NONE we apply fences. Otherwise an [F/G/T/U]UPD becomes UPD etc. */
/* op_tstart is called in "mur_forward_play_cur_jrec" already */
if (IS_FUPD(rectype))
{
jnl_fence_ctl.level = 1;
if (jnl_enabled)
{
jnl_fence_ctl.fence_list = JNL_FENCE_LIST_END;
csa->next_fenced = NULL;
}
} else if (IS_GUPD(rectype))
{
jnl_fence_ctl.level = 1;
if (jnl_enabled)
{
jnl_fence_ctl.fence_list = csa;
csa->next_fenced = JNL_FENCE_LIST_END;
}
} else if (IS_TP(rectype))
tp_set_sgm();
}
# ifdef GTM_TRIGGER
/* Check if ^#t and if so need to increment trigger cycle in file header. Note that the below 'if' check could cause
* csd->db_trigger_cycle to be incremented even for the region that actually did NOT get any trigger updates. This
* is because some of the ^#t subscripts (like ^#t(#TNAME)) go to the DEFAULT region. So, even though a trigger was
* loaded only for ^a (corresponding to AREG), csd->db_trigger_cycle will be incremented for DEFAULT region as well.
* To avoid this, the below check should be modified to set csa->incr_db_trigger_cycle only if the ^#t subscript
* does not begin with '#' (similar to what is done in UPD_GV_BIND_NAME_APPROPRIATE). However, since journal
* recovery operates in standalone mode, the db_trigger_cycle increment to DEFAULT region should be okay since it
* will NOT cause any restarts
*/
if (IS_GVKEY_HASHT_GBLNAME(keystr->length, keystr->text))
{
assert(cs_addrs == csa);
csa->incr_db_trigger_cycle = TRUE;
}
# endif
if (IS_SET(rectype))
{
val_ptr = &keystr->text[keystr->length];
GET_MSTR_LEN(mv.str.len, val_ptr);
mv.str.addr = val_ptr + SIZEOF(mstr_len_t);
mv.mvtype = MV_STR;
op_gvput(&mv);
} else if (IS_KILL(rectype))
{
if (IS_TP(rectype))
tp_set_sgm();
op_gvkill();
# ifdef GTM_TRIGGER
} else if (IS_ZTRIG(rectype))
{
if (IS_TP(rectype))
tp_set_sgm();
op_ztrigger();
# endif
} else
{
assert(IS_ZKILL(rectype));
if (IS_TP(rectype))
tp_set_sgm();
op_gvzwithdraw();
}
if (IS_ZTP(rectype))
{ /* Even for FENCE_NONE we apply fences. Otherwise an FUPD/GUPD becomes UPD etc. */
assert(jnl_enabled || (JNL_FENCE_LIST_END == jnl_fence_ctl.fence_list && NULL == csa->next_fenced));
jnl_fence_ctl.level = 0;
if (jnl_enabled)
{
jnl_fence_ctl.fence_list = JNL_FENCE_LIST_END;
csa->next_fenced = NULL;
}
}
return SS_NORMAL;
}