iSharkFly-Open
/
fis-gtm
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
fis-gtm/sr_port/tp.h

807 lines
34 KiB
C
Raw Blame History

/****************************************************************
* *
* Copyright 2001, 2011 Fidelity Information Services, Inc *
* *
* This source code contains the intellectual property *
* of its copyright holder(s), and is made available *
* under a license. If you do not know the terms of *
* the license, please stop and do not read further. *
* *
****************************************************************/
#ifndef __TP_H__
#define __TP_H__
#include <sys/types.h>
error_def(ERR_TPNOTACID);
\
/* HEADER-FILE-DEPENDENCIES : hashtab_int4.h */
#define JNL_LIST_INIT_ALLOC 16 /* initial allocation for si->jnl_list */
#define CW_SET_LIST_INIT_ALLOC 64 /* initial allocation for si->cw_set_list */
#define TLVL_CW_SET_LIST_INIT_ALLOC 64 /* initial allocation for si->tlvl_cw_set_list */
#define NEW_BUFF_LIST_INIT_ALLOC 8 /* initial allocation for si->new_buff_list */
#define RECOMPUTE_LIST_INIT_ALLOC 8 /* initial allocation for si->recompute_list */
#define BLKS_IN_USE_INIT_ELEMS 64 /* initial no. of elements in hash table si->blks_in_use */
#define TLVL_INFO_LIST_INIT_ALLOC 4 /* initial allocation for si->tlvl_info_head */
#define GBL_TLVL_INFO_LIST_INIT_ALLOC 8 /* initial allocation for global_tlvl_info_head (one per region) */
#define INIT_CUR_TP_HIST_SIZE 64 /* initial value of si->cur_tp_hist_size */
#define TP_MAX_MM_TRANSIZE 64*1024
#define JNL_FORMAT_BUFF_INIT_ALLOC 16*1024
#define JFB_ELE_SIZE 8 /* Same as JNL_REC_START_BNDRY */
#define JFB_ELE_SIZE_IN_BITS 3 /* log2 of JFB_ELE_SIZE */
#define TP_BATCH_ID "BATCH"
#define TP_BATCH_LEN (SIZEOF(TP_BATCH_ID) - 1)
#define TP_BATCH_SHRT 2 /* permit abbreviation to two characters */
#define TP_DEADLOCK_FACTOR 5 /* multiplied by dollar_trestart to produce an argument for wcs_backoff */
#define MAX_VISIBLE_TRESTART 4 /* Per Bhaskar on 10/20/98: dollar_trestart is not allowed to visibly exceed 4
* because of errors this causes in legacy Profile versions (any < 6.1)
*/
#define MAX_TP_FINAL_RETRY_TRESTART_CNT 2
#define MAX_TP_FINAL_RETRY_MLOCKRESTART_CNT 16
#define MAX_TRESTARTS (2 * MAX_TP_FINAL_RETRY_MLOCKRESTART_CNT)
#define FAIL_HIST_ARRAY_SIZE MAX_TRESTARTS
/* structure to hold transaction level specific info per segment.
* Aids in incremental rollback, to identify the state of various elements at the BEGINNING of a new transaction level. */
typedef struct tlevel_info_struct
{
que_ent free_que;
struct tlevel_info_struct
*next_tlevel_info;
kill_set *tlvl_kill_set; /* these two fields hold the kill_set state before this t_level started */
int4 tlvl_kill_used;
jnl_format_buffer
*tlvl_jfb_info; /* state of the tp_jnl_format_buff_info list before this t_level started */
srch_blk_status *tlvl_tp_hist_info; /* state of the tp_hist array (tail) before this t_level started */
uint4 t_level;
uint4 update_trans; /* a copy of sgm_info_ptr->update_trans before this t_level started */
uint4 jnl_list_elems; /* # of si->jnl_list elements consumed before this transaction started */
uint4 jfb_list_elems; /* # of si->format_buff_list elements consumed before this transaction started */
} tlevel_info;
/* structure to hold the global (across all segments) dollar_tlevel specific information.
* To identify the state of various elements at the BEGINNING of a given transaction level.
*/
typedef struct global_tlvl_info_struct
{
struct global_tlvl_info_struct
*next_global_tlvl_info;
sgmnt_addrs *global_tlvl_fence_info;
uint4 t_level;
uint4 tlvl_cumul_jrec_len;
# ifdef DEBUG
uint4 tlvl_cumul_index;
# endif
uint4 tlvl_tp_ztp_jnl_upd_num;
# ifdef GTM_TRIGGER
unsigned char *tlvl_prev_ztworm_ptr;
# endif
struct ua_list *curr_ua; /* points to global variable curr_ua at start of this transaction */
char *upd_array_ptr; /* points to global variable update_array_ptr at start of this transaction */
} global_tlvl_info;
/* A note on the buddy lists used in sgm_info structure,
* cw_set_list -> uses get_new_element and free_last_n_elements
* tlvl_cw_set_list -> uses get_new_free_element and free_element
* jnl_list -> uses get_new_element and free_last_n_elements
* tlvl_info_list -> uses get_new_element and free_last_n_elements
* gbl_tlvl_info_list -> uses get_new_element and free_last_n_elements
* new_buff_list -> uses get_new_free_element and free_element
* recompute_list -> uses get_new_element
*/
/* A small comment about the tp_hist_size and cur_tp_hist_size members of the sgm_info structure.
* Currently, we allow a maximum of 64k (TP_MAX_MM_TRANSIZE) blocks in the read-set of a TP transaction and n_bts/2 blocks
* in the write-set. Each block in the read-set has a corresponding srch_blk_status structure used in final validation in tp_tend.
* The current code requires the read-set to be a contiguous array of srch_blk_status structures for performance considerations.
* A process may need anywhere from 1 to 64K elements of this array depending on the TP transaction although the average need
* is most likely to be less than a hundred. But mallocing 64K srch_blk_status structures at one stretch (in order to allow
* the maximum) is too costly in terms of virtual memory (approx. 3 MB memory per region per process).
* Therefore, we malloc only cur_tp_hist_size structures initially. As the need for more structures arises, we malloc an array
* double the size of the original and reset pointers appropriately.
* cur_tp_hist_size always represents the current allocation while tp_hist_size represents the maximum possible allocation.
*/
typedef struct sgm_info_struct
{
struct sgm_info_struct *next_sgm_info,
*next_tp_si_by_ftok; /* List of ALL regions in the TP transaction sorted on ftok order */
srch_blk_status *first_tp_hist,
*last_tp_hist;
hash_table_int4 *blks_in_use;
trans_num start_tn;
gd_region *gv_cur_region; /* Backpointer to the region; Note that it is not necessarily unique since
* multiple regions could point to the same csa (and hence same sgm_info
* structure) with all but one of them having reg->was_open set to TRUE.
*/
uint4 update_trans; /* bitmask indicating among other things whether this region was updated;
* Bit-0 is 1 if cw_set_depth is non-zero or if it is a duplicate set
* (cw_set_depth is zero in that case).
* Bit-1 is 1 if there was a journaled logical update in this region.
* Bit-2 is 1 if transaction commit in this region is beyond point of rollback
*/
cw_set_element *first_cw_set,
*last_cw_set,
*first_cw_bitmap;
buddy_list *cw_set_list; /* list(buddy) of cw_set_elements for this region */
buddy_list *tlvl_cw_set_list; /* list(buddy) of horizontal cw_set_elements for this region */
/* first link in each of these horizontal lists is maintained in
* cw_set_list buddy list */
buddy_list *new_buff_list; /* to hold the new_buff for the cw_set elements */
buddy_list *recompute_list; /* to hold the list of to-be-recomputed keys and values */
buddy_list *tlvl_info_list; /* to hold the list of tlvl_info structures */
cache_rec_ptr_ptr_t cr_array;
sgmnt_data_ptr_t tp_csd;
sgmnt_addrs *tp_csa;
kill_set *kill_set_head,
*kill_set_tail;
tlevel_info *tlvl_info_head;
jnl_format_buffer *jnl_head,
**jnl_tail;
buddy_list *format_buff_list;
buddy_list *jnl_list;
int cw_set_depth,
cr_array_index,
num_of_blks,
tp_hist_size,
cur_tp_hist_size,
total_jnl_rec_size,
cr_array_size;
boolean_t fresh_start;
int4 crash_count;
boolean_t backup_block_saved;
sgmnt_addrs *kip_csa;
int tmp_cw_set_depth; /* used only #ifdef DEBUG. see comments for tmp_cw_set_depth in "tp_tend" */
uint4 tot_jrec_size; /* maximum journal space needs for this transaction */
} sgm_info;
typedef struct
{
#ifdef BIGENDIAN
unsigned flag : 1;
unsigned cw_index : 15;
unsigned next_off : 16;
#else
unsigned next_off : 16;
unsigned cw_index : 15;
unsigned flag : 1;
#endif
} off_chain;
/* The following struct is built into a separate list for each transaction
because it is not thrown away if a transaction restarts. The list keeps
growing so we can lock down all the necessary regions in the correct order
in case one attempt doesn't get very far while later attempts get further.
Items will be put on the list sorted in unique_id order so that they will always
be grab-crit'd in the same order thus avoiding deadlocks. */
/* The structure backup_region_list defined in mupipbckup.h needs to have its first four fields
identical to the first three fields in this structure */
typedef struct tp_region_struct
{
struct tp_region_struct *fPtr; /* Next in list */
gd_region *reg; /* Region pointer. Note that it is not necessarily unique since multiple
* regions could point to the same physical file (with all but one of them
* having reg->was_open set to TRUE.and hence have the same tp_region structure.
*/
union /* we will either use file_id or index */
{
gd_id file_id; /* both for VMS and UNIX */
int4 fid_index; /* copy of csa->fid_index for this region */
} file;
} tp_region;
#ifdef DEBUG
/* Macro to check that the tp_reg_list linked list is sorted properly on the file-id */
#define DBG_CHECK_TP_REG_LIST_SORTING(REGLIST) \
{ \
int4 prev_index; \
tp_region *tr; \
\
prev_index = 0; \
for (tr = REGLIST; NULL != tr; tr = tr->fPtr) \
{ \
if (tr->reg->open) \
{ \
assert(prev_index < tr->file.fid_index); \
DEBUG_ONLY(prev_index = tr->file.fid_index); \
} \
} \
}
#define DBG_CHECK_SI_BUDDY_LIST_IS_REINITIALIZED(si) \
{ \
sgmnt_addrs *csa; \
\
csa = si->tp_csa; \
assert(NULL == si->kill_set_head); \
assert(NULL == si->kill_set_tail); \
assert(NULL == si->jnl_head); \
assert(NULL == csa->next_fenced); \
if (JNL_ALLOWED(csa)) \
{ \
assert(si->total_jnl_rec_size == csa->min_total_tpjnl_rec_size); \
VERIFY_LIST_IS_REINITIALIZED(si->jnl_list); \
VERIFY_LIST_IS_REINITIALIZED(si->format_buff_list); \
assert(si->jnl_tail == &si->jnl_head); \
} else \
{ \
assert(NULL == si->jnl_list); \
assert(NULL == si->format_buff_list); \
assert(NULL == si->jnl_tail); \
} \
VERIFY_LIST_IS_REINITIALIZED(si->recompute_list); \
VERIFY_LIST_IS_REINITIALIZED(si->cw_set_list); \
VERIFY_LIST_IS_REINITIALIZED(si->new_buff_list); \
VERIFY_LIST_IS_REINITIALIZED(si->tlvl_cw_set_list); \
assert(NULL == si->first_cw_set); \
assert(NULL == si->last_cw_set); \
assert(NULL == si->first_cw_bitmap); \
assert(0 == si->cw_set_depth); \
assert(0 == si->update_trans); \
}
#define DBG_CHECK_IN_FIRST_SGM_INFO_LIST(SI) \
{ \
sgm_info *tmpsi; \
\
GBLREF sgm_info *first_sgm_info; \
GBLREF uint4 dollar_tlevel; \
\
assert(dollar_tlevel); \
assert(NULL != first_sgm_info); \
assert(NULL != SI); \
for (tmpsi = first_sgm_info; NULL != tmpsi; tmpsi = tmpsi->next_sgm_info) \
{ \
if (tmpsi == SI) \
break; \
} \
assert(NULL != tmpsi); \
}
#else
#define DBG_CHECK_TP_REG_LIST_SORTING(REGLIST)
#define DBG_CHECK_IN_FIRST_SGM_INFO_LIST(SI)
#endif
/* The below macro is used to check if any block split info (heuristic used by gvcst_put) is no longer relevant
* (due to an incremental rollback or rollback or restart) and if so reset it to a safe value of 0.
*/
#define TP_CLEANUP_GVNH_SPLIT_IF_NEEDED(GVNH, CW_DEPTH) \
{ \
int cw_set_depth; \
int level; \
off_chain chain1; \
\
cw_set_depth = CW_DEPTH; \
if (GVNH->split_cleanup_needed) \
{ /* Created block numbers stored in the blk split array are no longer relevant after the \
* restart. So reset them not to confuse the next call to gvcst_put for this gv_target. \
*/ \
for (level = 0; level < ARRAYSIZE(GVNH->last_split_blk_num); level++) \
{ \
chain1 = *(off_chain *)&GVNH->last_split_blk_num[level]; \
if (chain1.flag && ((unsigned)cw_set_depth <= (unsigned)chain1.cw_index)) \
GVNH->last_split_blk_num[level] = 0; \
} \
} \
}
typedef struct ua_list_struct
{
struct ua_list_struct
*next_ua;
char *update_array;
uint4 update_array_size;
} ua_list;
#define TP_MAX_NEST 127
/* Note gv_orig_key[i] is assigned to tp_pointer->orig_key which then tries to dereference the "begin", "end", "prev", "top"
* fields like it were a gv_currkey pointer. Since these members are 2-byte fields, we need atleast 2 byte alignment.
* We want to be safer and hence give 4-byte alignment by declaring the array as an array of integers.
*/
typedef struct gv_orig_key_struct
{
int4 gv_orig_key[TP_MAX_NEST + 1][DIVIDE_ROUND_UP((SIZEOF(gv_key) + MAX_KEY_SZ + 1), SIZEOF(int4))];
}gv_orig_key_array;
GBLREF int4 tprestart_syslog_delta;
GBLREF block_id t_fail_hist_blk[];
GBLREF gd_region *tp_fail_hist_reg[];
GBLREF gv_namehead *tp_fail_hist[];
GBLREF int4 tp_fail_n;
GBLREF int4 tp_fail_level;
GBLREF trans_num tp_fail_histtn[], tp_fail_bttn[];
#define TP_TRACE_HIST(X, Y) \
{ \
if (tprestart_syslog_delta) \
{ \
tp_fail_hist_reg[t_tries] = gv_cur_region; \
t_fail_hist_blk[t_tries] = ((block_id)X); \
tp_fail_hist[t_tries] = (gv_namehead *)(((int)X & ~(-BLKS_PER_LMAP)) ? Y : NULL); \
} \
}
#define TP_TRACE_HIST_MOD(X, Y, n, csd, histtn, bttn, level) \
{ \
if (tprestart_syslog_delta) \
{ \
tp_fail_hist_reg[t_tries] = gv_cur_region; \
t_fail_hist_blk[t_tries] = ((block_id)X); \
tp_fail_hist[t_tries] = (gv_namehead *)(((int)X & ~(-BLKS_PER_LMAP)) ? Y : NULL); \
(csd)->tp_cdb_sc_blkmod[(n)]++; \
tp_fail_n = (n); \
tp_fail_level = (level); \
tp_fail_histtn[t_tries] = (histtn); \
tp_fail_bttn[t_tries] = (bttn); \
} \
}
#define ASSERT_IS_WITHIN_TP_HIST_ARRAY_BOUNDS(first_tp_srch_status, sgm_info_ptr) \
{ \
assert(NULL == (first_tp_srch_status) \
|| ((first_tp_srch_status) >= (sgm_info_ptr)->first_tp_hist \
&& (first_tp_srch_status) < (sgm_info_ptr)->last_tp_hist)); \
}
#define SET_WC_BLOCKED_FINAL_RETRY_IF_NEEDED(csa, status) \
{ /* set wc_blocked if final retry and cache related failure status */ \
if (CDB_STAGNATE <= t_tries) \
{ \
GBLREF boolean_t is_uchar_wcs_code[]; \
GBLREF boolean_t is_lchar_wcs_code[]; \
boolean_t is_wcs_code = FALSE; \
\
if (ISALPHA_ASCII(status)) \
{ \
if (status > 'Z') \
is_wcs_code = is_lchar_wcs_code[status - 'a']; \
else \
is_wcs_code = is_uchar_wcs_code[status - 'A']; \
} \
if (is_wcs_code) \
{ \
SET_TRACEABLE_VAR(csa->hdr->wc_blocked, TRUE); \
BG_TRACE_PRO_ANY(csa, wc_blocked_wcs_cdb_sc_final_retry); \
} \
} \
}
#define TP_RETRY_ACCOUNTING(csa, cnl, status) \
{ \
GBLREF uint4 dollar_trestart; \
\
switch (dollar_trestart) \
{ \
case 0: \
INCR_GVSTATS_COUNTER(csa, cnl, n_tp_cnflct_retries_0, 1); \
break; \
case 1: \
INCR_GVSTATS_COUNTER(csa, cnl, n_tp_cnflct_retries_1, 1); \
break; \
case 2: \
INCR_GVSTATS_COUNTER(csa, cnl, n_tp_cnflct_retries_2, 1); \
break; \
case 3: \
INCR_GVSTATS_COUNTER(csa, cnl, n_tp_cnflct_retries_3, 1); \
break; \
default: \
INCR_GVSTATS_COUNTER(csa, cnl, n_tp_cnflct_retries_4, 1); \
break; \
} \
}
#define PREV_OFF_INVALID -1
/* JNL_FILE_TAIL_PRESERVE macro indicates maximum number of bytes to ensure allocated at the end of the journal file
* to store the journal records that will be written whenever the journal file gets closed.
* (i) Any process closing the journal file needs to write at most one PINI, one EPOCH, one PFIN and one EOF record
* In case of wcs_recover extra INCTN will be written
* (ii) We may need to give room for twice the above space to accommodate the EOF writing by a process that closes the journal
* and the EOF writing by the first process that reopens it and finds no space left and switches to a new journal.
* (iii) We may need to write one ALIGN record at the most since the total calculated from (i) and (ii) above is
* less than the minimum alignsize that we support (asserted before using JNL_FILE_TAIL_PRESERVE in macros below)
* The variable portion of this ALIGN record can get at the most equal to the maximum of the sizes of the
* PINI/EPOCH/PFIN/EOF record. (We know PINI_RECLEN is maximum of EPOCH_RECLEN, PFIN_RECLEN, EOF_RECLEN)
*/
#define JNL_FILE_TAIL_PRESERVE (MIN_ALIGN_RECLEN + (PINI_RECLEN + EPOCH_RECLEN + INCTN_RECLEN + \
PFIN_RECLEN + EOF_RECLEN) * 2 + PINI_RECLEN)
#define TOTAL_TPJNL_REC_SIZE(total_jnl_rec_size, si, csa) \
{ \
DEBUG_ONLY(si->tmp_cw_set_depth = si->cw_set_depth;) /* save a copy to check later in "tp_tend" */ \
total_jnl_rec_size = si->total_jnl_rec_size; \
if (csa->jnl_before_image) \
total_jnl_rec_size += (si->cw_set_depth * csa->pblk_align_jrecsize); \
/* Since we have already taken into account an align record per journal record and since the size of \
* an align record will be < (size of the journal record written + fixed-size of align record) \
* we can be sure we won't need more than twice the computed space. \
* JNL_FILE_TAIL_PRESERVE is to give allowance for space needed at end of journal file \
* in case journal file close is needed \
*/ \
assert(JNL_FILE_TAIL_PRESERVE < (JNL_MIN_ALIGNSIZE * DISK_BLOCK_SIZE)); \
si->total_jnl_rec_size = total_jnl_rec_size = (total_jnl_rec_size * 2) + (uint4)JNL_FILE_TAIL_PRESERVE; \
}
#define MIN_TOTAL_NONTPJNL_REC_SIZE (PINI_RECLEN + MIN_ALIGN_RECLEN + INCTN_RECLEN + MIN_ALIGN_RECLEN)
/* This macro gives a pessimistic estimate on the total journal record size needed.
* The side effect is that we might end up with a journal file extension when it was actually not needed.
*/
#define TOTAL_NONTPJNL_REC_SIZE(total_jnl_rec_size, non_tp_jfb_ptr, csa, tmp_cw_set_depth) \
{ \
total_jnl_rec_size = (non_tp_jfb_ptr->record_size + (uint4)MIN_TOTAL_NONTPJNL_REC_SIZE); \
if (csa->jnl_before_image) \
/* One PBLK record for each gds block changed by the transaction */ \
total_jnl_rec_size += (tmp_cw_set_depth * csa->pblk_align_jrecsize); \
if (write_after_image) \
total_jnl_rec_size += (uint4)MIN_AIMG_RECLEN + csa->hdr->blk_size + (uint4)MIN_ALIGN_RECLEN; \
/* Since we have already taken into account an align record per journal record and since the size of \
* an align record will be < (size of the journal record written + fixed-size of align record) \
* we can be sure we won't need more than twice the computed space. \
* JNL_FILE_TAIL_PRESERVE is to give allowance for space needed at end of journal file \
* in case journal file close is needed \
*/ \
assert(JNL_FILE_TAIL_PRESERVE < (JNL_MIN_ALIGNSIZE * DISK_BLOCK_SIZE)); \
total_jnl_rec_size = total_jnl_rec_size * 2 + (uint4)JNL_FILE_TAIL_PRESERVE; \
}
#define INVALIDATE_CLUE(cse) \
{ \
off_chain macro_chain; \
\
assert(cse->blk_target); \
cse->blk_target->clue.end = 0; \
macro_chain = *(off_chain *)&cse->blk_target->root; \
if (macro_chain.flag) \
cse->blk_target->root = 0; \
}
/* freeup killset starting from the link 'ks' */
#define FREE_KILL_SET(si, ks) \
{ \
kill_set *macro_next_ks; \
for (; ks; ks = macro_next_ks) \
{ \
macro_next_ks = ks->next_kill_set; \
free(ks); \
} \
}
/* Determine previous jfb, if any */
#define SET_PREV_JFB(SI, PREV_JFB) \
{ \
if (SI->jnl_tail != &SI->jnl_head) \
PREV_JFB = (jnl_format_buffer *)((uchar_ptr_t)SI->jnl_tail - OFFSETOF(jnl_format_buffer, next)); \
else \
PREV_JFB = NULL; \
}
/* freeup si->jnl_list and si->format_buff_list elements which are no more needed
* as the sub-transactions in which they have been created are now being rolled back
*/
#define FREE_JFB_INFO(SI, TLI, FREE_ALL_ELEMS) \
{ \
int jnl_list_cnt, format_buf_list_cnt; \
\
jnl_list_cnt = SI->jnl_list->nElems; \
format_buf_list_cnt = SI->format_buff_list->nElems; \
if (!FREE_ALL_ELEMS) \
{ \
assert(TLI); \
assert(jnl_list_cnt >= TLI->jnl_list_elems); \
assert(format_buf_list_cnt >= TLI->jfb_list_elems); \
jnl_list_cnt -= TLI->jnl_list_elems; \
format_buf_list_cnt -= TLI->jfb_list_elems; \
} \
/* else all elements from si->jnl_head can be free'ed up since the sub-transactions \
* in which journal activity started for this region is been currently rolled \
* back \
*/ \
if (format_buf_list_cnt && !free_last_n_elements(SI->format_buff_list, format_buf_list_cnt)) \
assert(FALSE); \
if (jnl_list_cnt && !free_last_n_elements(SI->jnl_list, jnl_list_cnt)) \
assert(FALSE); \
}
/* Calls FREE_JFB_INFO based on whether journal activity existed in the prior, non-rolled back levels.
* FREE_ALL_ELEMS is TRUE if either (a) No TLI exists for this region OR (b) All such TLI are now
* being actively rolled back. In either case, the entire SI->jnl_list and SI->format_buff_list can be
* freed up.
*/
#define FREE_JFB_INFO_IF_NEEDED(CSA, SI, TLI, FREE_ALL_ELEMS) \
{ \
jnl_format_buffer *jfb; \
\
if (JNL_WRITE_LOGICAL_RECS(CSA)) \
{ \
if (!FREE_ALL_ELEMS && (TLI->tlvl_jfb_info)) \
{ /* Journal activity existed in $TLEVEL <= newlevel which should not be free'ed up */ \
assert(0 < TLI->jnl_list_elems); \
assert(0 < TLI->jfb_list_elems); \
jfb = TLI->tlvl_jfb_info; /* last journal format buffer in the non-rolled back levels */ \
FREE_JFB_INFO(SI, TLI, FALSE); \
jfb->next = NULL; \
SI->jnl_tail = &jfb->next; \
/* Ensure that size of si->jnl_list and si->format_buff_list matches the one noted down at op_tstart \
* of $TLEVEL = (newlevel + 1) \
*/ \
assert(SI->jnl_list->nElems == TLI->jnl_list_elems); \
assert(SI->format_buff_list->nElems == TLI->jfb_list_elems); \
} else \
{ /* Either journal activity did not exist on $TLEVEL <= newlevel OR the sub-transactions in which \
* they existed are now being rolled back. So, free all the jnl_list and format_buff_list buddy lists */\
FREE_JFB_INFO(SI, TLI, TRUE); \
SI->jnl_head = NULL; \
SI->jnl_tail = &SI->jnl_head; \
assert(0 == SI->jnl_list->nElems); \
assert(0 == SI->format_buff_list->nElems); \
} \
} \
} \
/* freeup gbl_tlvl_info_list starting from the link 'gti' */
#define FREE_GBL_TLVL_INFO(gti) \
{ \
int macro_cnt; \
for (macro_cnt = 0; gti; gti = gti->next_global_tlvl_info) \
macro_cnt++; \
if (macro_cnt) \
free_last_n_elements(global_tlvl_info_list, macro_cnt); \
}
#ifdef GTM_TRIGGER
#define INVALIDATE_TRIGGER_CYCLES_IF_NEEDED(INCREMENTAL, COMMIT) \
{ \
GBLREF boolean_t dollar_ztrigger_invoked; \
GBLREF trans_num local_tn; \
GBLREF gv_namehead *gvt_tp_list; \
GBLREF sgm_info *first_sgm_info; \
\
gv_namehead *gvnh, *lcl_hasht_tree; \
cw_set_element *cse; \
sgmnt_addrs *csa; \
sgm_info *si; \
\
if (dollar_ztrigger_invoked) \
{ /* There was at least one region where a $ZTRIGGER() was invoked. */ \
dollar_ztrigger_invoked = FALSE; \
/* Phase 1: Adjust trigger/$ztrigger related fields for all gv_target read/updated in this transaction */ \
if (COMMIT) \
{ /* Reset csa->db_dztrigger_cycle and gvt->db_dztrigger_cycle to zero. This is needed so that globals \
* updated after the TCOMMIT don't re-read triggers when NOT necessary. Such a case is possible if \
* for instance a $ZTRIGGER() happened in a sub-transaction that got rolled back. Though no actual \
* trigger change happened, csa->db_dztrigger_cycle will be incremented and hence gvt(s) whose \
* db_dztrigger_cycle does not match will now re-read triggers. We don't expect $ZTRIGGER() to be \
* frequent. So, it's okay to go through the list of gvt \
*/ \
for (gvnh = gvt_tp_list; NULL != gvnh; gvnh = gvnh->next_tp_gvnh) \
{ \
gvnh->db_dztrigger_cycle = 0; \
gvnh->gd_csa->db_dztrigger_cycle = 0; \
} \
} else \
{ \
/* Now that the transaction is rolled back/restarted, invalidate gvt->gvt_trigger->gv_trigger_cycle \
* (by resetting it to zero) for all gvt read/updated in this transaction. Note that even though we \
* reset db_trigger_cycle (to -1) for non-incremental rollbacks/restarts and increment db_dztrigger_cycle\
* for incremental rollbacks we still need to reset gv_trigger_cycle as otherwise gvtr_init will find \
* that gv_trigger_cycle has NOT changed since it was updated last and will NOT do any trigger reads \
*/ \
for (gvnh = gvt_tp_list; NULL != gvnh; gvnh = gvnh->next_tp_gvnh) \
{ \
assert(gvnh->read_local_tn == local_tn); \
if (NULL != gvnh->gvt_trigger) \
((gvt_trigger_t *)(gvnh->gvt_trigger))->gv_trigger_cycle = 0; \
if (!INCREMENTAL) \
{ /* TROLLBACK(0) or TRESTART. Reset db_dztrigger_cycle to 0 since we are going to start \
* a new transaction. But, we want to ensure that the new transaction re-read triggers \
* since any gvt which updated its gvt_trigger in this transaction will be stale as \
* they never got committed \
*/ \
gvnh->db_dztrigger_cycle = 0; \
gvnh->db_trigger_cycle = (uint4)-1; \
} \
} \
} \
/* Phase 2: Adjust trigger/$ztrigger related fields for all csa accessed in this transaction */ \
if (INCREMENTAL) \
{ /* An incremental rollback. Find out if there are still any cse->blk_target containing ^#t updates. \
* If not, set csa->incr_db_trigger_cycle to FALSE (if already set to TRUE) so that we don't \
* increment csd->db_trigger_cycle during commit time \
*/ \
for (si = first_sgm_info; NULL != si; si = si->next_sgm_info) \
{ \
cse = si->first_cw_set; \
csa = si->tp_csa; \
lcl_hasht_tree = csa->hasht_tree; \
if (NULL != lcl_hasht_tree) \
{ \
/* Walk through the cw_set_elements remaining after the incremental rollback to \
* see if any of them has a ^#t update \
*/ \
while (NULL != cse) \
{ \
if (lcl_hasht_tree == cse->blk_target) \
break; \
cse = cse->next_cw_set; \
} \
if (NULL != cse) \
csa->incr_db_trigger_cycle = FALSE; \
} else \
{ \
assert(!csa->incr_db_trigger_cycle); \
} \
if (csa->db_dztrigger_cycle) \
csa->db_dztrigger_cycle++; /* so that future updates in this TN re-read triggers */ \
} \
/* Keep dollar_ztrigger_invoked as TRUE as the transaction is not yet complete and we need this \
* variable being TRUE to reset csa->db_dztrigger_cycle and gvt->db_dztrigger_cycle to 0 during \
* tp_clean_up of this transaction. \
*/ \
dollar_ztrigger_invoked = TRUE; \
} else if (!COMMIT) \
{ /* This is either a complete rollback or a restart. In either case, set csa->incr_db_trigger_cycle \
* to FALSE for all csa referenced in this transaction as they are anyways not going to be committed. \
*/ \
for (si = first_sgm_info; NULL != si; si = si->next_sgm_info) \
{ \
si->tp_csa->db_dztrigger_cycle = 0; \
si->tp_csa->incr_db_trigger_cycle = FALSE; \
} \
} \
} \
}
# ifdef DEBUG
# define ASSERT_ZTRIGGER_CYCLE_RESET \
{ /* At the end of a transaction (either because of trestart, complete trollback or tcommit) ensure that \
* csa->db_dztrigger_cycle is reset to zero. It's okay not to check if all gvt updated in this transaction \
* also has gvt->db_dztrigger_cycle set back to zero because if they don't there are other asserts that \
* will trip in the subsequent transactions \
*/ \
GBLREF sgm_info *first_sgm_info; \
\
sgm_info *si; \
\
for (si = first_sgm_info; NULL != si; si = si->next_sgm_info) \
assert(0 == si->tp_csa->db_dztrigger_cycle); \
}
# else
# define ASSERT_ZTRIGGER_CYCLE_RESET
# endif
#endif
#ifdef VMS
/* The error below has special handling in a few condition handlers because it not so much signals an error
as it does drive the necessary mechanisms to invoke a restart. Consequently this error can be
overridden by a "real" error. For VMS, the extra parameters are specified to provide "placeholders" on
the stack in the signal array if a real error needs to be overlayed in place of this one (see example
code in mdb_condition_handler). The number of extra parameters need to be 2 more than the largest
number of parameters for an rts_error in tp_restart().
*/
#define INVOKE_RESTART rts_error(VARLSTCNT(6) ERR_TPRETRY, 4, 0, 0, 0, 0, 0, 0, 0, 0);
#else
#define INVOKE_RESTART rts_error(VARLSTCNT(1) ERR_TPRETRY);
#endif
/* the following macros T_BEGIN_READ_NONTP_OR_TP and T_BEGIN_SETORKILL_NONTP_OR_TP are similar except for one difference
* which is that for the SETORKILL case, sgm_info_ptr->update_trans needs to be set. They need to be maintained
* in parallel always. The reason for choosing this duplication is because it saves us an if check which would have
* otherwise been had the two macros been merged and this is used in database code where performance is a concern.
*/
/* the macro below uses "dollar_tlevel", "t_err" and "sgm_info_ptr" */
#define T_BEGIN_SETORKILL_NONTP_OR_TP(err_code) \
{ \
GBLREF sgm_info *sgm_info_ptr; \
GBLREF sgmnt_addrs *cs_addrs; \
GBLREF uint4 t_err; \
error_def(err_code); \
\
if (!dollar_tlevel) \
t_begin(err_code, UPDTRNS_DB_UPDATED_MASK); \
else \
{ \
t_err = err_code; \
assert((NULL != sgm_info_ptr) && (cs_addrs->sgm_info_ptr == sgm_info_ptr)); \
sgm_info_ptr->update_trans |= UPDTRNS_DB_UPDATED_MASK; \
} \
}
/* the macro below uses "dollar_tlevel", "t_err" */
#define T_BEGIN_READ_NONTP_OR_TP(err_code) \
{ \
GBLREF uint4 t_err; \
GBLREF sgm_info *sgm_info_ptr; \
GBLREF sgmnt_addrs *cs_addrs; \
\
error_def(err_code); \
\
if (!dollar_tlevel) \
t_begin(err_code, 0); \
else \
{ \
assert((NULL != sgm_info_ptr) && (cs_addrs->sgm_info_ptr == sgm_info_ptr)); \
t_err = err_code; \
} \
}
/* The following GBLREFs are needed by the IS_TP_AND_FINAL_RETRY macro */
GBLREF uint4 dollar_tlevel;
GBLREF unsigned int t_tries;
#define IS_TP_AND_FINAL_RETRY (dollar_tlevel && (CDB_STAGNATE <= t_tries))
#define TP_REL_CRIT_ALL_REG \
{ \
sgmnt_addrs *csa; \
tp_region *tr; \
\
GBLREF tp_region *tp_reg_list; \
\
for (tr = tp_reg_list; NULL != tr; tr = tr->fPtr) \
{ \
assert(tr->reg->open); \
if (!tr->reg->open) \
continue; \
csa = (sgmnt_addrs *)&FILE_INFO(tr->reg)->s_addrs; \
assert(!csa->hold_onto_crit); \
if (csa->now_crit) \
rel_crit(tr->reg); \
} \
assert(!have_crit(CRIT_HAVE_ANY_REG)); \
}
#define TP_FINAL_RETRY_DECREMENT_T_TRIES_IF_OK \
{ \
GBLREF boolean_t mupip_jnl_recover; \
\
assert(dollar_tlevel); \
assert(CDB_STAGNATE == t_tries); \
/* In case of journal recovery, it operates at t_tries=CDB_STAGNATE but we should not adjust t_tries \
* in that case and it is ok to not print the TPNOTACID message in that case. Since we have standalone \
* access, we dont expect anyone else to interfere with us and cause a restart anyways. \
*/ \
if (!mupip_jnl_recover) \
t_tries = CDB_STAGNATE - 1; \
}
#define TPNOTACID_CHECK(CALLER_STR) \
{ \
mval zpos; \
\
if (IS_TP_AND_FINAL_RETRY) \
{ \
TP_REL_CRIT_ALL_REG; \
TP_FINAL_RETRY_DECREMENT_T_TRIES_IF_OK; \
getzposition(&zpos); \
gtm_putmsg(VARLSTCNT(6) ERR_TPNOTACID, 4, LEN_AND_LIT(CALLER_STR), zpos.str.len, zpos.str.addr); \
send_msg(VARLSTCNT(6) ERR_TPNOTACID, 4, LEN_AND_LIT(CALLER_STR), zpos.str.len, zpos.str.addr); \
} \
}
void tp_get_cw(cw_set_element *cs, int depth, cw_set_element **cs1);
void tp_clean_up(boolean_t rollback_flag);
void tp_cw_list(cw_set_element **cs);
void tp_get_cw(cw_set_element *cs, int depth, cw_set_element **cs1);
void tp_incr_clean_up(uint4 newlevel);
void tp_set_sgm(void);
void tp_start_timer_dummy(int4 timeout_seconds);
void tp_clear_timeout_dummy(void);
void tp_timeout_action_dummy(void);
tp_region *insert_region(gd_region *reg, tp_region **reg_list, tp_region **reg_free_list, int4 size);
boolean_t tp_tend(void);
boolean_t tp_crit_all_regions(void);
#endif
Reference in New Issue View Git Blame Copy Permalink