4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2014, 2015, Intel Corporation.
26 * lustre/target/update_trans.c
28 * This file implements the update distribute transaction API.
30 * To manage the cross-MDT operation (distribute operation) transaction,
31 * the transaction will also be separated two layers on MD stack, top
32 * transaction and sub transaction.
34 * During the distribute operation, top transaction is created in the LOD
35 * layer, and represent the operation. Sub transaction is created by
36 * each OSD or OSP. Top transaction start/stop will trigger all of its sub
37 * transaction start/stop. Top transaction (the whole operation) is committed
38 * only all of its sub transaction are committed.
40 * there are three kinds of transactions
41 * 1. local transaction: All updates are in a single local OSD.
42 * 2. Remote transaction: All Updates are only in the remote OSD,
43 * i.e. locally all updates are in OSP.
44 * 3. Mixed transaction: Updates are both in local OSD and remote
47 * Author: Di Wang <di.wang@intel.com>
50 #define DEBUG_SUBSYSTEM S_CLASS
52 #include <linux/kthread.h>
53 #include <lu_target.h>
54 #include <lustre_log.h>
55 #include <lustre_update.h>
57 #include <obd_class.h>
58 #include <tgt_internal.h>
60 #include <tgt_internal.h>
62 * Dump top mulitple thandle
64 * Dump top multiple thandle and all of its sub thandle to the debug log.
66 * \param[in]mask debug mask
67 * \param[in]top_th top_thandle to be dumped
69 static void top_multiple_thandle_dump(struct top_multiple_thandle *tmt,
72 struct sub_thandle *st;
74 LASSERT(tmt->tmt_magic == TOP_THANDLE_MAGIC);
75 CDEBUG(mask, "%s tmt %p refcount %d committed %d result %d "
77 tmt->tmt_master_sub_dt ?
78 tmt->tmt_master_sub_dt->dd_lu_dev.ld_obd->obd_name :
80 tmt, atomic_read(&tmt->tmt_refcount), tmt->tmt_committed,
81 tmt->tmt_result, tmt->tmt_batchid);
83 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
84 struct sub_thandle_cookie *stc;
86 CDEBUG(mask, "st %p obd %s committed %d stopped %d sub_th %p\n",
87 st, st->st_dt->dd_lu_dev.ld_obd->obd_name,
88 st->st_committed, st->st_stopped, st->st_sub_th);
90 list_for_each_entry(stc, &st->st_cookie_list, stc_list) {
91 CDEBUG(mask, " cookie "DOSTID": %u\n",
92 POSTID(&stc->stc_cookie.lgc_lgl.lgl_oi),
93 stc->stc_cookie.lgc_index);
99 * Declare write update to sub device
101 * Declare Write updates llog records to the sub device during distribute
104 * \param[in] env execution environment
105 * \param[in] record update records being written
106 * \param[in] sub_th sub transaction handle
107 * \param[in] record_size total update record size
109 * \retval 0 if writing succeeds
110 * \retval negative errno if writing fails
112 static int sub_declare_updates_write(const struct lu_env *env,
113 struct llog_update_record *record,
114 struct thandle *sub_th, size_t record_size)
116 struct llog_ctxt *ctxt;
117 struct dt_device *dt = sub_th->th_dev;
118 int left = record_size;
121 /* If ctxt is NULL, it means not need to write update,
122 * for example if the the OSP is used to connect to OST */
123 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
124 LLOG_UPDATELOG_ORIG_CTXT);
126 /* Not ready to record updates yet. */
127 if (ctxt == NULL || ctxt->loc_handle == NULL) {
132 rc = llog_declare_add(env, ctxt->loc_handle,
133 &record->lur_hdr, sub_th);
137 while (left > ctxt->loc_chunk_size) {
138 rc = llog_declare_add(env, ctxt->loc_handle,
139 &record->lur_hdr, sub_th);
143 left -= ctxt->loc_chunk_size;
153 * write update to sub device
155 * Write llog update record to the sub device during distribute
156 * transaction. If it succeeds, llog cookie of the record will be
157 * returned by @cookie.
159 * \param[in] env execution environment
160 * \param[in] record update records being written
161 * \param[in] sub_th sub transaction handle
162 * \param[out] cookie llog cookie of the update record.
164 * \retval 1 if writing succeeds
165 * \retval negative errno if writing fails
167 static int sub_updates_write(const struct lu_env *env,
168 struct llog_update_record *record,
169 struct sub_thandle *sub_th)
171 struct dt_device *dt = sub_th->st_dt;
172 struct llog_ctxt *ctxt;
173 struct llog_update_record *lur = NULL;
174 __u32 update_count = 0;
175 __u32 param_count = 0;
176 __u32 last_update_count = 0;
177 __u32 last_param_count = 0;
181 struct sub_thandle_cookie *stc;
187 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
188 LLOG_UPDATELOG_ORIG_CTXT);
189 /* If ctxt == NULL, then it means updates on OST (only happens
190 * during migration), and we do not track those updates for now */
191 /* If ctxt->loc_handle == NULL, then it does not need to record
192 * update, usually happens in error handler path */
193 if (ctxt == NULL || ctxt->loc_handle == NULL) {
198 /* Since the cross-MDT updates will includes both local
199 * and remote updates, the update ops count must > 1 */
200 LASSERT(record->lur_update_rec.ur_update_count > 1);
201 LASSERTF(record->lur_hdr.lrh_len == llog_update_record_size(record),
202 "lrh_len %u record_size %zu\n", record->lur_hdr.lrh_len,
203 llog_update_record_size(record));
205 if (likely(record->lur_hdr.lrh_len <= ctxt->loc_chunk_size)) {
208 GOTO(llog_put, rc = -ENOMEM);
209 INIT_LIST_HEAD(&stc->stc_list);
211 rc = llog_add(env, ctxt->loc_handle, &record->lur_hdr,
212 &stc->stc_cookie, sub_th->st_sub_th);
214 CDEBUG(D_INFO, "%s: Add update log "DOSTID":%u: rc = %d\n",
215 dt->dd_lu_dev.ld_obd->obd_name,
216 POSTID(&stc->stc_cookie.lgc_lgl.lgl_oi),
217 stc->stc_cookie.lgc_index, rc);
220 list_add(&stc->stc_list, &sub_th->st_cookie_list);
229 /* Split the records into chunk_size update record */
230 OBD_ALLOC_LARGE(lur, ctxt->loc_chunk_size);
232 GOTO(llog_put, rc = -ENOMEM);
234 memcpy(lur, &record->lur_hdr, sizeof(record->lur_hdr));
235 lur->lur_update_rec.ur_update_count = 0;
236 lur->lur_update_rec.ur_param_count = 0;
237 start = (char *)&record->lur_update_rec.ur_ops;
240 if (update_count < record->lur_update_rec.ur_update_count)
241 next = (char *)update_op_next_op(
242 (struct update_op *)cur);
243 else if (param_count < record->lur_update_rec.ur_param_count)
244 next = (char *)update_param_next_param(
245 (struct update_param *)cur);
250 * If its size > llog chunk_size, then write current chunk to
251 * the update llog, NB the padding should >= LLOG_MIN_REC_SIZE.
253 * So check padding length is either >= LLOG_MIN_REC_SIZE or is
254 * 0 (record length just matches the chunk size).
256 reclen = __llog_update_record_size(
257 __update_records_size(next - start));
258 if ((reclen + LLOG_MIN_REC_SIZE <= ctxt->loc_chunk_size ||
259 reclen == ctxt->loc_chunk_size) &&
264 record->lur_update_rec.ur_update_count)
266 else if (param_count <
267 record->lur_update_rec.ur_param_count)
272 lur->lur_update_rec.ur_update_count = update_count -
274 lur->lur_update_rec.ur_param_count = param_count -
276 memcpy(&lur->lur_update_rec.ur_ops, start, cur - start);
277 lur->lur_hdr.lrh_len = llog_update_record_size(lur);
279 LASSERT(lur->lur_hdr.lrh_len ==
280 __llog_update_record_size(
281 __update_records_size(cur - start)));
282 LASSERT(lur->lur_hdr.lrh_len <= ctxt->loc_chunk_size);
284 update_records_dump(&lur->lur_update_rec, D_INFO, true);
288 GOTO(llog_put, rc = -ENOMEM);
289 INIT_LIST_HEAD(&stc->stc_list);
291 rc = llog_add(env, ctxt->loc_handle, &lur->lur_hdr,
292 &stc->stc_cookie, sub_th->st_sub_th);
294 CDEBUG(D_INFO, "%s: Add update log "DOSTID":%u rc = %d\n",
295 dt->dd_lu_dev.ld_obd->obd_name,
296 POSTID(&stc->stc_cookie.lgc_lgl.lgl_oi),
297 stc->stc_cookie.lgc_index, rc);
300 list_add(&stc->stc_list, &sub_th->st_cookie_list);
307 last_update_count = update_count;
308 last_param_count = param_count;
310 lur->lur_update_rec.ur_update_count = 0;
311 lur->lur_update_rec.ur_param_count = 0;
312 lur->lur_update_rec.ur_flags |= UPDATE_RECORD_CONTINUE;
317 OBD_FREE_LARGE(lur, ctxt->loc_chunk_size);
324 * Prepare the update records.
326 * Merge params and ops into the update records, then initializing
329 * During transaction execution phase, parameters and update ops
330 * are collected in two different buffers (see lod_updates_pack()),
331 * during transaction stop, it needs to be merged in one buffer,
332 * so it will be written in the update log.
334 * \param[in] env execution environment
335 * \param[in] tmt top_multiple_thandle for distribute txn
337 * \retval 0 if merging succeeds.
338 * \retval negaitive errno if merging fails.
340 static int prepare_writing_updates(const struct lu_env *env,
341 struct top_multiple_thandle *tmt)
343 struct thandle_update_records *tur = tmt->tmt_update_records;
344 struct llog_update_record *lur;
345 struct update_params *params;
349 if (tur == NULL || tur->tur_update_records == NULL ||
350 tur->tur_update_params == NULL)
353 lur = tur->tur_update_records;
354 /* Extends the update records buffer if needed */
355 params_size = update_params_size(tur->tur_update_params,
356 tur->tur_update_param_count);
357 LASSERT(lur->lur_update_rec.ur_param_count == 0);
358 update_size = llog_update_record_size(lur);
359 if (cfs_size_round(update_size + params_size) >
360 tur->tur_update_records_buf_size) {
363 rc = tur_update_records_extend(tur,
364 cfs_size_round(update_size + params_size));
368 lur = tur->tur_update_records;
371 params = update_records_get_params(&lur->lur_update_rec);
372 memcpy(params, tur->tur_update_params, params_size);
374 lur->lur_update_rec.ur_param_count = tur->tur_update_param_count;
375 lur->lur_update_rec.ur_batchid = tmt->tmt_batchid;
376 /* Init update record header */
377 lur->lur_hdr.lrh_len = llog_update_record_size(lur);
378 lur->lur_hdr.lrh_type = UPDATE_REC;
380 /* Dump updates for debugging purpose */
381 update_records_dump(&lur->lur_update_rec, D_INFO, true);
387 distribute_txn_commit_thread_running(struct lu_target *lut)
389 return lut->lut_tdtd_commit_thread.t_flags & SVC_RUNNING;
393 distribute_txn_commit_thread_stopped(struct lu_target *lut)
395 return lut->lut_tdtd_commit_thread.t_flags & SVC_STOPPED;
399 * Top thandle commit callback
401 * This callback will be called when all of sub transactions are committed.
403 * \param[in] th top thandle to be committed.
405 static void top_trans_committed_cb(struct top_multiple_thandle *tmt)
407 struct lu_target *lut;
410 LASSERT(atomic_read(&tmt->tmt_refcount) > 0);
412 top_multiple_thandle_dump(tmt, D_HA);
413 tmt->tmt_committed = 1;
414 lut = dt2lu_dev(tmt->tmt_master_sub_dt)->ld_site->ls_tgt;
415 if (distribute_txn_commit_thread_running(lut))
416 wake_up(&lut->lut_tdtd->tdtd_commit_thread_waitq);
420 struct sub_thandle *lookup_sub_thandle(struct top_multiple_thandle *tmt,
421 struct dt_device *dt_dev)
423 struct sub_thandle *st;
425 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
426 if (st->st_dt == dt_dev)
431 EXPORT_SYMBOL(lookup_sub_thandle);
433 struct sub_thandle *create_sub_thandle(struct top_multiple_thandle *tmt,
434 struct dt_device *dt_dev)
436 struct sub_thandle *st;
440 RETURN(ERR_PTR(-ENOMEM));
442 INIT_LIST_HEAD(&st->st_sub_list);
443 INIT_LIST_HEAD(&st->st_cookie_list);
446 list_add(&st->st_sub_list, &tmt->tmt_sub_thandle_list);
450 static void sub_trans_commit_cb_internal(struct top_multiple_thandle *tmt,
451 struct thandle *sub_th, int err)
453 struct sub_thandle *st;
454 bool all_committed = true;
456 /* Check if all sub thandles are committed */
457 spin_lock(&tmt->tmt_sub_lock);
458 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
459 if (st->st_sub_th == sub_th) {
460 st->st_committed = 1;
463 if (!st->st_committed)
464 all_committed = false;
466 spin_unlock(&tmt->tmt_sub_lock);
468 if (tmt->tmt_result == 0)
469 tmt->tmt_result = err;
472 top_trans_committed_cb(tmt);
474 top_multiple_thandle_dump(tmt, D_INFO);
475 top_multiple_thandle_put(tmt);
480 * sub thandle commit callback
482 * Mark the sub thandle to be committed and if all sub thandle are committed
483 * notify the top thandle.
485 * \param[in] env execution environment
486 * \param[in] sub_th sub thandle being committed
487 * \param[in] cb commit callback
488 * \param[in] err trans result
490 static void sub_trans_commit_cb(struct lu_env *env,
491 struct thandle *sub_th,
492 struct dt_txn_commit_cb *cb, int err)
494 struct top_multiple_thandle *tmt = cb->dcb_data;
496 sub_trans_commit_cb_internal(tmt, sub_th, err);
499 static void sub_thandle_register_commit_cb(struct sub_thandle *st,
500 struct top_multiple_thandle *tmt)
502 LASSERT(st->st_sub_th != NULL);
503 top_multiple_thandle_get(tmt);
504 st->st_commit_dcb.dcb_func = sub_trans_commit_cb;
505 st->st_commit_dcb.dcb_data = tmt;
506 INIT_LIST_HEAD(&st->st_commit_dcb.dcb_linkage);
507 dt_trans_cb_add(st->st_sub_th, &st->st_commit_dcb);
511 * Sub thandle stop call back
513 * After sub thandle is stopped, it will call this callback to notify
516 * \param[in] th sub thandle to be stopped
517 * \param[in] rc result of sub trans
519 static void sub_trans_stop_cb(struct lu_env *env,
520 struct thandle *sub_th,
521 struct dt_txn_commit_cb *cb, int err)
523 struct sub_thandle *st;
524 struct top_multiple_thandle *tmt = cb->dcb_data;
527 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
531 if (st->st_dt == sub_th->th_dev) {
538 wake_up(&tmt->tmt_stop_waitq);
542 static void sub_thandle_register_stop_cb(struct sub_thandle *st,
543 struct top_multiple_thandle *tmt)
545 st->st_stop_dcb.dcb_func = sub_trans_stop_cb;
546 st->st_stop_dcb.dcb_data = tmt;
547 st->st_stop_dcb.dcb_flags = DCB_TRANS_STOP;
548 INIT_LIST_HEAD(&st->st_stop_dcb.dcb_linkage);
549 dt_trans_cb_add(st->st_sub_th, &st->st_stop_dcb);
555 * Create transaction handle for sub_thandle
557 * \param[in] env execution environment
558 * \param[in] th top thandle
559 * \param[in] st sub_thandle
561 * \retval 0 if creation succeeds.
562 * \retval negative errno if creation fails.
564 int sub_thandle_trans_create(const struct lu_env *env,
565 struct top_thandle *top_th,
566 struct sub_thandle *st)
568 struct thandle *sub_th;
570 sub_th = dt_trans_create(env, st->st_dt);
572 return PTR_ERR(sub_th);
574 sub_th->th_top = &top_th->tt_super;
575 st->st_sub_th = sub_th;
577 sub_th->th_wait_submit = 1;
578 sub_thandle_register_stop_cb(st, top_th->tt_multiple_thandle);
583 * Create the top transaction.
585 * Create the top transaction on the master device. It will create a top
586 * thandle and a sub thandle on the master device.
588 * \param[in] env execution environment
589 * \param[in] master_dev master_dev the top thandle will be created
591 * \retval pointer to the created thandle.
592 * \retval ERR_PTR(errno) if creation failed.
595 top_trans_create(const struct lu_env *env, struct dt_device *master_dev)
597 struct top_thandle *top_th;
598 struct thandle *child_th;
600 OBD_ALLOC_GFP(top_th, sizeof(*top_th), __GFP_IO);
602 return ERR_PTR(-ENOMEM);
604 top_th->tt_super.th_top = &top_th->tt_super;
606 if (master_dev != NULL) {
607 child_th = dt_trans_create(env, master_dev);
608 if (IS_ERR(child_th)) {
609 OBD_FREE_PTR(top_th);
613 child_th->th_top = &top_th->tt_super;
614 child_th->th_wait_submit = 1;
615 top_th->tt_master_sub_thandle = child_th;
617 top_th->tt_super.th_tags |= child_th->th_tags;
619 return &top_th->tt_super;
621 EXPORT_SYMBOL(top_trans_create);
624 * Declare write update transaction
626 * Check if there are updates being recorded in this transaction,
627 * it will write the record into the disk.
629 * \param[in] env execution environment
630 * \param[in] tmt top multiple transaction handle
632 * \retval 0 if writing succeeds
633 * \retval negative errno if writing fails
635 static int declare_updates_write(const struct lu_env *env,
636 struct top_multiple_thandle *tmt)
638 struct llog_update_record *record;
639 struct sub_thandle *st;
642 record = tmt->tmt_update_records->tur_update_records;
643 /* Declare update write for all other target */
644 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
645 if (st->st_sub_th == NULL)
648 rc = sub_declare_updates_write(env, record, st->st_sub_th,
649 tmt->tmt_record_size);
658 * Assign batchid to the distribute transaction.
660 * Assign batchid to the distribute transaction
662 * \param[in] tmt distribute transaction
664 static void distribute_txn_assign_batchid(struct top_multiple_thandle *new)
666 struct target_distribute_txn_data *tdtd;
667 struct dt_device *dt = new->tmt_master_sub_dt;
668 struct sub_thandle *st;
671 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
672 spin_lock(&tdtd->tdtd_batchid_lock);
673 new->tmt_batchid = tdtd->tdtd_batchid++;
674 list_add_tail(&new->tmt_commit_list, &tdtd->tdtd_list);
675 spin_unlock(&tdtd->tdtd_batchid_lock);
676 list_for_each_entry(st, &new->tmt_sub_thandle_list, st_sub_list) {
677 if (st->st_sub_th != NULL)
678 sub_thandle_register_commit_cb(st, new);
680 top_multiple_thandle_get(new);
681 top_multiple_thandle_dump(new, D_INFO);
685 * Insert distribute transaction to the distribute txn list.
687 * Insert distribute transaction to the distribute txn list.
689 * \param[in] new the distribute txn to be inserted.
691 void distribute_txn_insert_by_batchid(struct top_multiple_thandle *new)
693 struct dt_device *dt = new->tmt_master_sub_dt;
694 struct top_multiple_thandle *tmt;
695 struct target_distribute_txn_data *tdtd;
696 struct sub_thandle *st;
697 bool at_head = false;
700 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
702 spin_lock(&tdtd->tdtd_batchid_lock);
703 list_for_each_entry_reverse(tmt, &tdtd->tdtd_list, tmt_commit_list) {
704 if (new->tmt_batchid > tmt->tmt_batchid) {
705 list_add(&new->tmt_commit_list, &tmt->tmt_commit_list);
709 if (list_empty(&new->tmt_commit_list)) {
711 list_add(&new->tmt_commit_list, &tdtd->tdtd_list);
713 spin_unlock(&tdtd->tdtd_batchid_lock);
715 list_for_each_entry(st, &new->tmt_sub_thandle_list, st_sub_list) {
716 if (st->st_sub_th != NULL)
717 sub_thandle_register_commit_cb(st, new);
720 top_multiple_thandle_get(new);
721 top_multiple_thandle_dump(new, D_INFO);
722 if (new->tmt_committed && at_head)
723 wake_up(&tdtd->tdtd_commit_thread_waitq);
727 * Prepare cross-MDT operation.
729 * Create the update record buffer to record updates for cross-MDT operation,
730 * add master sub transaction to tt_sub_trans_list, and declare the update
733 * During updates packing, all of parameters will be packed in
734 * tur_update_params, and updates will be packed in tur_update_records.
735 * Then in transaction stop, parameters and updates will be merged
736 * into one updates buffer.
738 * And also master thandle will be added to the sub_th list, so it will be
739 * easy to track the commit status.
741 * \param[in] env execution environment
742 * \param[in] th top transaction handle
744 * \retval 0 if preparation succeeds.
745 * \retval negative errno if preparation fails.
747 static int prepare_multiple_node_trans(const struct lu_env *env,
748 struct top_multiple_thandle *tmt)
750 struct thandle_update_records *tur;
754 if (tmt->tmt_update_records == NULL) {
755 tur = &update_env_info(env)->uti_tur;
756 rc = check_and_prepare_update_record(env, tur);
760 tmt->tmt_update_records = tur;
761 distribute_txn_assign_batchid(tmt);
764 rc = declare_updates_write(env, tmt);
770 * start the top transaction.
772 * Start all of its sub transactions, then start master sub transaction.
774 * \param[in] env execution environment
775 * \param[in] master_dev master_dev the top thandle will be start
776 * \param[in] th top thandle
778 * \retval 0 if transaction start succeeds.
779 * \retval negative errno if start fails.
781 int top_trans_start(const struct lu_env *env, struct dt_device *master_dev,
784 struct top_thandle *top_th = container_of(th, struct top_thandle,
786 struct sub_thandle *st;
787 struct top_multiple_thandle *tmt = top_th->tt_multiple_thandle;
793 top_th->tt_master_sub_thandle->th_sync = th->th_sync;
795 top_th->tt_master_sub_thandle->th_local = th->th_local;
796 top_th->tt_master_sub_thandle->th_tags = th->th_tags;
797 rc = dt_trans_start(env, top_th->tt_master_sub_thandle->th_dev,
798 top_th->tt_master_sub_thandle);
802 tmt = top_th->tt_multiple_thandle;
803 rc = prepare_multiple_node_trans(env, tmt);
807 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
808 if (st->st_sub_th == NULL)
811 st->st_sub_th->th_sync = th->th_sync;
813 st->st_sub_th->th_local = th->th_local;
814 st->st_sub_th->th_tags = th->th_tags;
815 rc = dt_trans_start(env, st->st_sub_th->th_dev,
820 LASSERT(st->st_started == 0);
827 EXPORT_SYMBOL(top_trans_start);
830 * Check whether we need write updates record
832 * Check if the updates for the top_thandle needs to be writen
833 * to all targets. Only if the transaction succeeds and the updates
834 * number > 2, it will write the updates,
836 * \params [in] top_th top thandle.
838 * \retval true if it needs to write updates
839 * \retval false if it does not need to write updates
841 static bool top_check_write_updates(struct top_thandle *top_th)
843 struct top_multiple_thandle *tmt;
844 struct thandle_update_records *tur;
846 /* Do not write updates to records if the transaction fails */
847 if (top_th->tt_super.th_result != 0)
850 tmt = top_th->tt_multiple_thandle;
854 tur = tmt->tmt_update_records;
858 /* Hmm, false update records, since the cross-MDT operation
859 * should includes both local and remote updates, so the
860 * updates count should >= 2 */
861 if (tur->tur_update_records == NULL ||
862 tur->tur_update_records->lur_update_rec.ur_update_count <= 1)
869 * Check if top transaction is stopped
871 * Check if top transaction is stopped, only if all sub transaction
872 * is stopped, then the top transaction is stopped.
874 * \param [in] top_th top thandle
876 * \retval true if the top transaction is stopped.
877 * \retval false if the top transaction is not stopped.
879 static bool top_trans_is_stopped(struct top_thandle *top_th)
881 struct top_multiple_thandle *tmt;
882 struct sub_thandle *st;
883 bool all_stopped = true;
885 tmt = top_th->tt_multiple_thandle;
886 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
887 if (!st->st_stopped && st->st_sub_th != NULL) {
892 if (st->st_result != 0 &&
893 top_th->tt_super.th_result == 0)
894 top_th->tt_super.th_result = st->st_result;
901 * Wait result of top transaction
903 * Wait until all sub transaction get its result.
905 * \param [in] top_th top thandle.
907 * \retval the result of top thandle.
909 static int top_trans_wait_result(struct top_thandle *top_th)
911 struct l_wait_info lwi = {0};
913 l_wait_event(top_th->tt_multiple_thandle->tmt_stop_waitq,
914 top_trans_is_stopped(top_th), &lwi);
916 RETURN(top_th->tt_super.th_result);
920 * Stop the top transaction.
922 * Stop the transaction on the master device first, then stop transactions
923 * on other sub devices.
925 * \param[in] env execution environment
926 * \param[in] master_dev master_dev the top thandle will be created
927 * \param[in] th top thandle
929 * \retval 0 if stop transaction succeeds.
930 * \retval negative errno if stop transaction fails.
932 int top_trans_stop(const struct lu_env *env, struct dt_device *master_dev,
935 struct top_thandle *top_th = container_of(th, struct top_thandle,
937 struct sub_thandle *st;
938 struct sub_thandle *master_st;
939 struct top_multiple_thandle *tmt;
940 struct thandle_update_records *tur;
941 bool write_updates = false;
945 if (likely(top_th->tt_multiple_thandle == NULL)) {
946 LASSERT(master_dev != NULL);
949 top_th->tt_master_sub_thandle->th_sync = th->th_sync;
951 top_th->tt_master_sub_thandle->th_local = th->th_local;
952 top_th->tt_master_sub_thandle->th_tags = th->th_tags;
953 rc = dt_trans_stop(env, master_dev,
954 top_th->tt_master_sub_thandle);
955 OBD_FREE_PTR(top_th);
959 tmt = top_th->tt_multiple_thandle;
960 tur = tmt->tmt_update_records;
962 /* Note: we need stop the master thandle first, then the stop
963 * callback will fill the master transno in the update logs,
964 * then these update logs will be sent to other MDTs */
965 /* get the master sub thandle */
966 master_st = lookup_sub_thandle(tmt, tmt->tmt_master_sub_dt);
967 write_updates = top_check_write_updates(top_th);
969 /* Step 1: write the updates log on Master MDT */
970 if (master_st != NULL && master_st->st_sub_th != NULL &&
972 struct llog_update_record *lur;
974 /* Merge the parameters and updates into one buffer */
975 rc = prepare_writing_updates(env, tmt);
977 CERROR("%s: cannot prepare updates: rc = %d\n",
978 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
980 write_updates = false;
981 GOTO(stop_master_trans, rc);
984 lur = tur->tur_update_records;
985 /* Write updates to the master MDT */
986 rc = sub_updates_write(env, lur, master_st);
988 /* Cleanup the common parameters in the update records,
989 * master transno callback might add more parameters.
990 * and we need merge the update records again in the
992 if (tur->tur_update_params != NULL)
993 lur->lur_update_rec.ur_param_count = 0;
996 CERROR("%s: write updates failed: rc = %d\n",
997 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
999 write_updates = false;
1000 GOTO(stop_master_trans, rc);
1005 /* Step 2: Stop the transaction on the master MDT, and fill the
1006 * master transno in the update logs to other MDT. */
1007 if (master_st != NULL && master_st->st_sub_th != NULL) {
1009 master_st->st_sub_th->th_local = th->th_local;
1011 master_st->st_sub_th->th_sync = th->th_sync;
1012 master_st->st_sub_th->th_tags = th->th_tags;
1013 master_st->st_sub_th->th_result = th->th_result;
1014 rc = dt_trans_stop(env, master_st->st_dt, master_st->st_sub_th);
1015 /* If it does not write_updates, then we call submit callback
1016 * here, otherwise callback is done through
1017 * osd(osp)_trans_commit_cb() */
1018 if (!master_st->st_started &&
1019 !list_empty(&tmt->tmt_commit_list))
1020 sub_trans_commit_cb_internal(tmt,
1021 master_st->st_sub_th, rc);
1024 GOTO(stop_other_trans, rc);
1025 } else if (tur != NULL && tur->tur_update_records != NULL) {
1026 struct llog_update_record *lur;
1028 lur = tur->tur_update_records;
1029 if (lur->lur_update_rec.ur_master_transno == 0)
1030 /* Update master transno after master stop
1032 lur->lur_update_rec.ur_master_transno =
1033 tgt_th_info(env)->tti_transno;
1037 /* Step 3: write updates to other MDTs */
1038 if (write_updates) {
1039 struct llog_update_record *lur;
1041 /* Stop callback of master will add more updates and also update
1042 * master transno, so merge the parameters and updates into one
1044 rc = prepare_writing_updates(env, tmt);
1046 CERROR("%s: prepare updates failed: rc = %d\n",
1047 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
1049 GOTO(stop_other_trans, rc);
1051 lur = tur->tur_update_records;
1052 list_for_each_entry(st, &tmt->tmt_sub_thandle_list,
1054 if (st->st_sub_th == NULL || st == master_st ||
1055 st->st_sub_th->th_result < 0)
1058 rc = sub_updates_write(env, lur, st);
1067 /* Step 4: Stop the transaction on other MDTs */
1068 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1069 if (st == master_st || st->st_sub_th == NULL)
1073 st->st_sub_th->th_sync = th->th_sync;
1075 st->st_sub_th->th_local = th->th_local;
1076 st->st_sub_th->th_tags = th->th_tags;
1077 st->st_sub_th->th_result = th->th_result;
1078 rc = dt_trans_stop(env, st->st_sub_th->th_dev,
1080 if (unlikely(rc < 0 && th->th_result == 0))
1084 rc = top_trans_wait_result(top_th);
1086 tmt->tmt_result = rc;
1088 /* Balance for the refcount in top_trans_create, Note: if it is NOT
1089 * multiple node transaction, the top transaction will be destroyed. */
1090 top_multiple_thandle_put(tmt);
1091 OBD_FREE_PTR(top_th);
1094 EXPORT_SYMBOL(top_trans_stop);
1097 * Create top_multiple_thandle for top_thandle
1099 * Create top_mutilple_thandle to manage the mutiple node transaction
1100 * for top_thandle, and it also needs to add master sub thandle to the
1101 * sub trans list now.
1103 * \param[in] env execution environment
1104 * \param[in] top_th the top thandle
1106 * \retval 0 if creation succeeds
1107 * \retval negative errno if creation fails
1109 int top_trans_create_tmt(const struct lu_env *env,
1110 struct top_thandle *top_th)
1112 struct top_multiple_thandle *tmt;
1118 tmt->tmt_magic = TOP_THANDLE_MAGIC;
1119 INIT_LIST_HEAD(&tmt->tmt_sub_thandle_list);
1120 INIT_LIST_HEAD(&tmt->tmt_commit_list);
1121 atomic_set(&tmt->tmt_refcount, 1);
1122 spin_lock_init(&tmt->tmt_sub_lock);
1123 init_waitqueue_head(&tmt->tmt_stop_waitq);
1125 top_th->tt_multiple_thandle = tmt;
1130 static struct sub_thandle *
1131 create_sub_thandle_with_thandle(struct top_thandle *top_th,
1132 struct thandle *sub_th)
1134 struct sub_thandle *st;
1136 /* create and init sub th to the top trans list */
1137 st = create_sub_thandle(top_th->tt_multiple_thandle,
1142 st->st_sub_th = sub_th;
1144 sub_th->th_top = &top_th->tt_super;
1145 sub_thandle_register_stop_cb(st, top_th->tt_multiple_thandle);
1152 * Get sub thandle from the top thandle according to the sub dt_device.
1154 * \param[in] env execution environment
1155 * \param[in] th thandle on the top layer.
1156 * \param[in] sub_dt sub dt_device used to get sub transaction
1158 * \retval thandle of sub transaction if succeed
1159 * \retval PTR_ERR(errno) if failed
1161 struct thandle *thandle_get_sub_by_dt(const struct lu_env *env,
1163 struct dt_device *sub_dt)
1165 struct sub_thandle *st = NULL;
1166 struct sub_thandle *master_st = NULL;
1167 struct top_thandle *top_th;
1168 struct thandle *sub_th = NULL;
1172 top_th = container_of(th, struct top_thandle, tt_super);
1174 if (likely(sub_dt == top_th->tt_master_sub_thandle->th_dev))
1175 RETURN(top_th->tt_master_sub_thandle);
1177 if (top_th->tt_multiple_thandle != NULL) {
1178 st = lookup_sub_thandle(top_th->tt_multiple_thandle, sub_dt);
1180 RETURN(st->st_sub_th);
1183 sub_th = dt_trans_create(env, sub_dt);
1187 /* Create top_multiple_thandle if necessary */
1188 if (top_th->tt_multiple_thandle == NULL) {
1189 struct top_multiple_thandle *tmt;
1191 rc = top_trans_create_tmt(env, top_th);
1193 GOTO(stop_trans, rc);
1195 tmt = top_th->tt_multiple_thandle;
1197 /* Add master sub th to the top trans list */
1198 tmt->tmt_master_sub_dt =
1199 top_th->tt_master_sub_thandle->th_dev;
1200 master_st = create_sub_thandle_with_thandle(top_th,
1201 top_th->tt_master_sub_thandle);
1202 if (IS_ERR(master_st)) {
1203 rc = PTR_ERR(master_st);
1205 GOTO(stop_trans, rc);
1209 /* create and init sub th to the top trans list */
1210 st = create_sub_thandle_with_thandle(top_th, sub_th);
1214 GOTO(stop_trans, rc);
1216 st->st_sub_th->th_wait_submit = 1;
1219 if (master_st != NULL) {
1220 list_del(&master_st->st_sub_list);
1221 OBD_FREE_PTR(master_st);
1223 sub_th->th_result = rc;
1224 dt_trans_stop(env, sub_dt, sub_th);
1225 sub_th = ERR_PTR(rc);
1230 EXPORT_SYMBOL(thandle_get_sub_by_dt);
1233 * Top multiple thandle destroy
1235 * Destroy multiple thandle and all its sub thandle.
1237 * \param[in] tmt top_multiple_thandle to be destroyed.
1239 void top_multiple_thandle_destroy(struct top_multiple_thandle *tmt)
1241 struct sub_thandle *st;
1242 struct sub_thandle *tmp;
1244 LASSERT(tmt->tmt_magic == TOP_THANDLE_MAGIC);
1245 list_for_each_entry_safe(st, tmp, &tmt->tmt_sub_thandle_list,
1247 struct sub_thandle_cookie *stc;
1248 struct sub_thandle_cookie *tmp;
1250 list_del(&st->st_sub_list);
1251 list_for_each_entry_safe(stc, tmp, &st->st_cookie_list,
1253 list_del(&stc->stc_list);
1260 EXPORT_SYMBOL(top_multiple_thandle_destroy);
1263 * Cancel the update log on MDTs
1265 * Cancel the update log on MDTs then destroy the thandle.
1267 * \param[in] env execution environment
1268 * \param[in] tmt the top multiple thandle whose updates records
1269 * will be cancelled.
1271 * \retval 0 if cancellation succeeds.
1272 * \retval negative errno if cancellation fails.
1274 static int distribute_txn_cancel_records(const struct lu_env *env,
1275 struct top_multiple_thandle *tmt)
1277 struct sub_thandle *st;
1280 top_multiple_thandle_dump(tmt, D_INFO);
1281 /* Cancel update logs on other MDTs */
1282 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1283 struct llog_ctxt *ctxt;
1284 struct obd_device *obd;
1285 struct llog_cookie *cookie;
1286 struct sub_thandle_cookie *stc;
1289 obd = st->st_dt->dd_lu_dev.ld_obd;
1290 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
1293 list_for_each_entry(stc, &st->st_cookie_list, stc_list) {
1294 cookie = &stc->stc_cookie;
1295 if (fid_is_zero(&cookie->lgc_lgl.lgl_oi.oi_fid))
1298 rc = llog_cat_cancel_records(env, ctxt->loc_handle, 1,
1300 CDEBUG(D_HA, "%s: batchid %llu cancel update log "
1301 DOSTID ".%u : rc = %d\n", obd->obd_name,
1303 POSTID(&cookie->lgc_lgl.lgl_oi),
1304 cookie->lgc_index, rc);
1307 llog_ctxt_put(ctxt);
1314 * Check if there are committed transaction
1316 * Check if there are committed transaction in the distribute transaction
1317 * list, then cancel the update records for those committed transaction.
1318 * Because the distribute transaction in the list are sorted by batchid,
1319 * and cancellation will be done by batchid order, so we only check the first
1320 * the transaction(with lowest batchid) in the list.
1322 * \param[in] lod lod device where cancel thread is
1324 * \retval true if it is ready
1325 * \retval false if it is not ready
1327 static bool tdtd_ready_for_cancel_log(struct target_distribute_txn_data *tdtd)
1329 struct top_multiple_thandle *tmt = NULL;
1330 struct obd_device *obd = tdtd->tdtd_lut->lut_obd;
1333 spin_lock(&tdtd->tdtd_batchid_lock);
1334 if (!list_empty(&tdtd->tdtd_list)) {
1335 tmt = list_entry(tdtd->tdtd_list.next,
1336 struct top_multiple_thandle, tmt_commit_list);
1337 if (tmt->tmt_committed &&
1338 (!obd->obd_recovering || (obd->obd_recovering &&
1339 tmt->tmt_batchid <= tdtd->tdtd_committed_batchid)))
1342 spin_unlock(&tdtd->tdtd_batchid_lock);
1347 struct distribute_txn_bid_data {
1348 struct dt_txn_commit_cb dtbd_cb;
1349 struct target_distribute_txn_data *dtbd_tdtd;
1354 * callback of updating commit batchid
1356 * Updating commit batchid then wake up the commit thread to cancel the
1359 * \param[in]env execution environment
1360 * \param[in]th thandle to updating commit batchid
1361 * \param[in]cb commit callback
1362 * \param[in]err result of thandle
1364 static void distribute_txn_batchid_cb(struct lu_env *env,
1366 struct dt_txn_commit_cb *cb,
1369 struct distribute_txn_bid_data *dtbd = NULL;
1370 struct target_distribute_txn_data *tdtd;
1372 dtbd = container_of0(cb, struct distribute_txn_bid_data, dtbd_cb);
1373 tdtd = dtbd->dtbd_tdtd;
1375 CDEBUG(D_HA, "%s: %llu batchid updated\n",
1376 tdtd->tdtd_lut->lut_obd->obd_name, dtbd->dtbd_batchid);
1377 spin_lock(&tdtd->tdtd_batchid_lock);
1378 if (dtbd->dtbd_batchid > tdtd->tdtd_committed_batchid &&
1379 !tdtd->tdtd_lut->lut_obd->obd_no_transno)
1380 tdtd->tdtd_committed_batchid = dtbd->dtbd_batchid;
1381 spin_unlock(&tdtd->tdtd_batchid_lock);
1382 atomic_dec(&tdtd->tdtd_refcount);
1383 wake_up(&tdtd->tdtd_commit_thread_waitq);
1389 * Update the commit batchid in disk
1391 * Update commit batchid in the disk, after this is committed, it can start
1392 * to cancel the update records.
1394 * \param[in] env execution environment
1395 * \param[in] tdtd distribute transaction structure
1396 * \param[in] batchid commit batchid to be updated
1398 * \retval 0 if update succeeds.
1399 * \retval negative errno if update fails.
1402 distribute_txn_commit_batchid_update(const struct lu_env *env,
1403 struct target_distribute_txn_data *tdtd,
1406 struct distribute_txn_bid_data *dtbd = NULL;
1414 OBD_ALLOC_PTR(dtbd);
1417 dtbd->dtbd_batchid = batchid;
1418 dtbd->dtbd_tdtd = tdtd;
1419 dtbd->dtbd_cb.dcb_func = distribute_txn_batchid_cb;
1420 atomic_inc(&tdtd->tdtd_refcount);
1422 th = dt_trans_create(env, tdtd->tdtd_lut->lut_bottom);
1425 RETURN(PTR_ERR(th));
1428 tmp = cpu_to_le64(batchid);
1430 buf.lb_len = sizeof(tmp);
1433 rc = dt_declare_record_write(env, tdtd->tdtd_batchid_obj, &buf, off,
1438 rc = dt_trans_start_local(env, tdtd->tdtd_lut->lut_bottom, th);
1442 rc = dt_trans_cb_add(th, &dtbd->dtbd_cb);
1446 rc = dt_record_write(env, tdtd->tdtd_batchid_obj, &buf,
1449 CDEBUG(D_INFO, "%s: update batchid "LPU64": rc = %d\n",
1450 tdtd->tdtd_lut->lut_obd->obd_name, batchid, rc);
1453 dt_trans_stop(env, tdtd->tdtd_lut->lut_bottom, th);
1460 * Init commit batchid for distribute transaction.
1462 * Initialize the batchid object and get commit batchid from the object.
1464 * \param[in] env execution environment
1465 * \param[in] tdtd distribute transaction whose batchid is initialized.
1467 * \retval 0 if initialization succeeds.
1468 * \retval negative errno if initialization fails.
1471 distribute_txn_commit_batchid_init(const struct lu_env *env,
1472 struct target_distribute_txn_data *tdtd)
1474 struct tgt_thread_info *tti = tgt_th_info(env);
1475 struct lu_target *lut = tdtd->tdtd_lut;
1476 struct lu_attr *attr = &tti->tti_attr;
1477 struct lu_fid *fid = &tti->tti_fid1;
1478 struct dt_object_format *dof = &tti->tti_u.update.tti_update_dof;
1479 struct dt_object *dt_obj = NULL;
1486 memset(attr, 0, sizeof(*attr));
1487 attr->la_valid = LA_MODE;
1488 attr->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1489 dof->dof_type = dt_mode_to_dft(S_IFREG);
1491 lu_local_obj_fid(fid, BATCHID_COMMITTED_OID);
1493 dt_obj = dt_find_or_create(env, lut->lut_bottom, fid, dof,
1495 if (IS_ERR(dt_obj)) {
1496 rc = PTR_ERR(dt_obj);
1501 tdtd->tdtd_batchid_obj = dt_obj;
1504 buf.lb_len = sizeof(tmp);
1506 rc = dt_read(env, dt_obj, &buf, &off);
1507 if (rc < 0 || (rc < buf.lb_len && rc > 0)) {
1508 CERROR("%s can't read last committed batchid: rc = %d\n",
1509 tdtd->tdtd_lut->lut_obd->obd_name, rc);
1513 } else if (rc == buf.lb_len) {
1514 tdtd->tdtd_committed_batchid = le64_to_cpu(tmp);
1515 CDEBUG(D_HA, "%s: committed batchid %llu\n",
1516 tdtd->tdtd_lut->lut_obd->obd_name,
1517 tdtd->tdtd_committed_batchid);
1522 if (rc < 0 && dt_obj != NULL) {
1523 lu_object_put(env, &dt_obj->do_lu);
1524 tdtd->tdtd_batchid_obj = NULL;
1530 * manage the distribute transaction thread
1532 * Distribute transaction are linked to the list, and once the distribute
1533 * transaction is committed, it will update the last committed batchid first,
1534 * after it is committed, it will cancel the records.
1536 * \param[in] _arg argument for commit thread
1538 * \retval 0 if thread is running successfully
1539 * \retval negative errno if the thread can not be run.
1541 static int distribute_txn_commit_thread(void *_arg)
1543 struct target_distribute_txn_data *tdtd = _arg;
1544 struct lu_target *lut = tdtd->tdtd_lut;
1545 struct ptlrpc_thread *thread = &lut->lut_tdtd_commit_thread;
1546 struct l_wait_info lwi = { 0 };
1548 struct list_head list;
1550 struct top_multiple_thandle *tmt;
1551 struct top_multiple_thandle *tmp;
1552 __u64 batchid = 0, committed;
1556 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
1560 spin_lock(&tdtd->tdtd_batchid_lock);
1561 thread->t_flags = SVC_RUNNING;
1562 spin_unlock(&tdtd->tdtd_batchid_lock);
1563 wake_up(&thread->t_ctl_waitq);
1564 INIT_LIST_HEAD(&list);
1566 CDEBUG(D_HA, "%s: start commit thread committed batchid "LPU64"\n",
1567 tdtd->tdtd_lut->lut_obd->obd_name,
1568 tdtd->tdtd_committed_batchid);
1570 while (distribute_txn_commit_thread_running(lut)) {
1571 spin_lock(&tdtd->tdtd_batchid_lock);
1572 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1574 if (tmt->tmt_committed == 0)
1577 /* Note: right now, replay is based on master MDT
1578 * transno, but cancellation is based on batchid.
1579 * so we do not try to cancel the update log until
1580 * the recoverying is done, unless the update records
1581 * batchid < committed_batchid. */
1582 if (tmt->tmt_batchid <= tdtd->tdtd_committed_batchid) {
1583 list_move_tail(&tmt->tmt_commit_list, &list);
1584 } else if (!tdtd->tdtd_lut->lut_obd->obd_recovering) {
1585 LASSERTF(tmt->tmt_batchid >= batchid,
1586 "tmt %p tmt_batchid: "LPU64", batchid "
1587 LPU64"\n", tmt, tmt->tmt_batchid,
1589 /* There are three types of distribution
1590 * transaction result
1592 * 1. If tmt_result < 0, it means the
1593 * distribution transaction fails, which should
1594 * be rare, because once declare phase succeeds,
1595 * the operation should succeeds anyway. Note in
1596 * this case, we will still update batchid so
1597 * cancellation would be stopped.
1599 * 2. If tmt_result == 0, it means the
1600 * distribution transaction succeeds, and we
1601 * will update batchid.
1603 * 3. If tmt_result > 0, it means distribute
1604 * transaction is not yet committed on every
1605 * node, but we need release this tmt before
1606 * that, which usuually happens during umount.
1608 if (tmt->tmt_result <= 0)
1609 batchid = tmt->tmt_batchid;
1610 list_move_tail(&tmt->tmt_commit_list, &list);
1613 spin_unlock(&tdtd->tdtd_batchid_lock);
1615 CDEBUG(D_HA, "%s: batchid: "LPU64" committed batchid "
1616 LPU64"\n", tdtd->tdtd_lut->lut_obd->obd_name, batchid,
1617 tdtd->tdtd_committed_batchid);
1618 /* update globally committed on a storage */
1619 if (batchid > tdtd->tdtd_committed_batchid) {
1620 rc = distribute_txn_commit_batchid_update(&env, tdtd,
1625 /* cancel the records for committed batchid's */
1626 /* XXX: should we postpone cancel's till the end of recovery? */
1627 committed = tdtd->tdtd_committed_batchid;
1628 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1629 if (tmt->tmt_batchid > committed)
1631 list_del_init(&tmt->tmt_commit_list);
1632 if (tmt->tmt_result <= 0)
1633 distribute_txn_cancel_records(&env, tmt);
1634 top_multiple_thandle_put(tmt);
1637 l_wait_event(tdtd->tdtd_commit_thread_waitq,
1638 !distribute_txn_commit_thread_running(lut) ||
1639 committed < tdtd->tdtd_committed_batchid ||
1640 tdtd_ready_for_cancel_log(tdtd), &lwi);
1643 l_wait_event(tdtd->tdtd_commit_thread_waitq,
1644 atomic_read(&tdtd->tdtd_refcount) == 0, &lwi);
1646 spin_lock(&tdtd->tdtd_batchid_lock);
1647 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1649 list_move_tail(&tmt->tmt_commit_list, &list);
1650 spin_unlock(&tdtd->tdtd_batchid_lock);
1652 CDEBUG(D_INFO, "%s stopping distribute txn commit thread.\n",
1653 tdtd->tdtd_lut->lut_obd->obd_name);
1654 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1655 list_del_init(&tmt->tmt_commit_list);
1656 top_multiple_thandle_dump(tmt, D_HA);
1657 top_multiple_thandle_put(tmt);
1660 thread->t_flags = SVC_STOPPED;
1662 wake_up(&thread->t_ctl_waitq);
1668 * Start llog cancel thread
1670 * Start llog cancel(master/slave) thread on LOD
1672 * \param[in]lclt cancel log thread to be started.
1674 * \retval 0 if the thread is started successfully.
1675 * \retval negative errno if the thread is not being
1678 int distribute_txn_init(const struct lu_env *env,
1679 struct lu_target *lut,
1680 struct target_distribute_txn_data *tdtd,
1683 struct task_struct *task;
1684 struct l_wait_info lwi = { 0 };
1688 INIT_LIST_HEAD(&tdtd->tdtd_list);
1689 INIT_LIST_HEAD(&tdtd->tdtd_replay_finish_list);
1690 INIT_LIST_HEAD(&tdtd->tdtd_replay_list);
1691 spin_lock_init(&tdtd->tdtd_batchid_lock);
1692 spin_lock_init(&tdtd->tdtd_replay_list_lock);
1693 tdtd->tdtd_replay_handler = distribute_txn_replay_handle;
1694 tdtd->tdtd_replay_ready = 0;
1696 tdtd->tdtd_batchid = lut->lut_last_transno + 1;
1698 init_waitqueue_head(&lut->lut_tdtd_commit_thread.t_ctl_waitq);
1699 init_waitqueue_head(&tdtd->tdtd_commit_thread_waitq);
1700 init_waitqueue_head(&tdtd->tdtd_recovery_threads_waitq);
1701 atomic_set(&tdtd->tdtd_refcount, 0);
1702 atomic_set(&tdtd->tdtd_recovery_threads_count, 0);
1704 tdtd->tdtd_lut = lut;
1705 rc = distribute_txn_commit_batchid_init(env, tdtd);
1709 task = kthread_run(distribute_txn_commit_thread, tdtd, "tdtd-%u",
1712 RETURN(PTR_ERR(task));
1714 l_wait_event(lut->lut_tdtd_commit_thread.t_ctl_waitq,
1715 distribute_txn_commit_thread_running(lut) ||
1716 distribute_txn_commit_thread_stopped(lut), &lwi);
1719 EXPORT_SYMBOL(distribute_txn_init);
1722 * Stop llog cancel thread
1724 * Stop llog cancel(master/slave) thread on LOD and also destory
1725 * all of transaction in the list.
1727 * \param[in]lclt cancel log thread to be stopped.
1729 void distribute_txn_fini(const struct lu_env *env,
1730 struct target_distribute_txn_data *tdtd)
1732 struct lu_target *lut = tdtd->tdtd_lut;
1734 /* Stop cancel thread */
1735 if (lut == NULL || !distribute_txn_commit_thread_running(lut))
1738 spin_lock(&tdtd->tdtd_batchid_lock);
1739 lut->lut_tdtd_commit_thread.t_flags = SVC_STOPPING;
1740 spin_unlock(&tdtd->tdtd_batchid_lock);
1741 wake_up(&tdtd->tdtd_commit_thread_waitq);
1742 wait_event(lut->lut_tdtd_commit_thread.t_ctl_waitq,
1743 lut->lut_tdtd_commit_thread.t_flags & SVC_STOPPED);
1745 dtrq_list_destroy(tdtd);
1746 if (tdtd->tdtd_batchid_obj != NULL) {
1747 lu_object_put(env, &tdtd->tdtd_batchid_obj->do_lu);
1748 tdtd->tdtd_batchid_obj = NULL;
1751 EXPORT_SYMBOL(distribute_txn_fini);