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) 2015, 2017, 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 batchid %llu\n",
76 tmt->tmt_master_sub_dt ?
77 tmt->tmt_master_sub_dt->dd_lu_dev.ld_obd->obd_name :
79 tmt, kref_read(&tmt->tmt_refcount), tmt->tmt_committed,
80 tmt->tmt_result, tmt->tmt_batchid);
82 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
83 struct sub_thandle_cookie *stc;
85 CDEBUG(mask, "st %p obd %s committed %d started %d stopped %d "
86 "result %d sub_th %p\n",
87 st, st->st_dt->dd_lu_dev.ld_obd->obd_name,
88 st->st_committed, st->st_started, st->st_stopped,
89 st->st_result, st->st_sub_th);
91 list_for_each_entry(stc, &st->st_cookie_list, stc_list) {
92 CDEBUG(mask, " cookie "DFID".%u\n",
93 PLOGID(&stc->stc_cookie.lgc_lgl),
94 stc->stc_cookie.lgc_index);
100 * Declare write update to sub device
102 * Declare Write updates llog records to the sub device during distribute
105 * \param[in] env execution environment
106 * \param[in] record update records being written
107 * \param[in] sub_th sub transaction handle
108 * \param[in] record_size total update record size
110 * \retval 0 if writing succeeds
111 * \retval negative errno if writing fails
113 static int sub_declare_updates_write(const struct lu_env *env,
114 struct llog_update_record *record,
115 struct thandle *sub_th, size_t record_size)
117 struct llog_ctxt *ctxt;
118 struct dt_device *dt = sub_th->th_dev;
119 int left = record_size;
122 /* If ctxt is NULL, it means not need to write update,
123 * for example if the the OSP is used to connect to OST */
124 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
125 LLOG_UPDATELOG_ORIG_CTXT);
127 /* Not ready to record updates yet. */
128 if (ctxt == NULL || ctxt->loc_handle == NULL) {
133 rc = llog_declare_add(env, ctxt->loc_handle,
134 &record->lur_hdr, sub_th);
138 while (left > ctxt->loc_chunk_size) {
139 rc = llog_declare_add(env, ctxt->loc_handle,
140 &record->lur_hdr, sub_th);
144 left -= ctxt->loc_chunk_size;
154 * write update to sub device
156 * Write llog update record to the sub device during distribute
157 * transaction. If it succeeds, llog cookie of the record will be
158 * returned by @cookie.
160 * \param[in] env execution environment
161 * \param[in] record update records being written
162 * \param[in] sub_th sub transaction handle
163 * \param[out] cookie llog cookie of the update record.
165 * \retval 1 if writing succeeds
166 * \retval negative errno if writing fails
168 static int sub_updates_write(const struct lu_env *env,
169 struct llog_update_record *record,
170 struct sub_thandle *sub_th)
172 struct dt_device *dt = sub_th->st_dt;
173 struct llog_ctxt *ctxt;
174 struct llog_update_record *lur = NULL;
175 __u32 update_count = 0;
176 __u32 param_count = 0;
177 __u32 last_update_count = 0;
178 __u32 last_param_count = 0;
182 struct sub_thandle_cookie *stc;
188 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
189 LLOG_UPDATELOG_ORIG_CTXT);
190 /* If ctxt == NULL, then it means updates on OST (only happens
191 * during migration), and we do not track those updates for now */
192 /* If ctxt->loc_handle == NULL, then it does not need to record
193 * update, usually happens in error handler path */
194 if (ctxt == NULL || ctxt->loc_handle == NULL) {
199 /* Since the cross-MDT updates will includes both local
200 * and remote updates, the update ops count must > 1 */
201 LASSERT(record->lur_update_rec.ur_update_count > 1);
202 LASSERTF(record->lur_hdr.lrh_len == llog_update_record_size(record),
203 "lrh_len %u record_size %zu\n", record->lur_hdr.lrh_len,
204 llog_update_record_size(record));
207 * If its size > llog chunk_size, then write current chunk to the update
208 * llog, NB the padding should >= LLOG_MIN_REC_SIZE.
210 * So check padding length is either >= LLOG_MIN_REC_SIZE or is 0
211 * (record length just matches the chunk size).
214 reclen = record->lur_hdr.lrh_len;
215 if (reclen + LLOG_MIN_REC_SIZE <= ctxt->loc_chunk_size ||
216 reclen == ctxt->loc_chunk_size) {
219 GOTO(llog_put, rc = -ENOMEM);
220 INIT_LIST_HEAD(&stc->stc_list);
222 rc = llog_add(env, ctxt->loc_handle, &record->lur_hdr,
223 &stc->stc_cookie, sub_th->st_sub_th);
225 CDEBUG(D_INFO, "%s: Add update log "DFID".%u: rc = %d\n",
226 dt->dd_lu_dev.ld_obd->obd_name,
227 PLOGID(&stc->stc_cookie.lgc_lgl),
228 stc->stc_cookie.lgc_index, rc);
231 list_add(&stc->stc_list, &sub_th->st_cookie_list);
240 /* Split the records into chunk_size update record */
241 OBD_ALLOC_LARGE(lur, ctxt->loc_chunk_size);
243 GOTO(llog_put, rc = -ENOMEM);
245 memcpy(lur, &record->lur_hdr, sizeof(record->lur_hdr));
246 lur->lur_update_rec.ur_update_count = 0;
247 lur->lur_update_rec.ur_param_count = 0;
248 start = (char *)&record->lur_update_rec.ur_ops;
251 if (update_count < record->lur_update_rec.ur_update_count)
252 next = (char *)update_op_next_op(
253 (struct update_op *)cur);
254 else if (param_count < record->lur_update_rec.ur_param_count)
255 next = (char *)update_param_next_param(
256 (struct update_param *)cur);
260 reclen = __llog_update_record_size(
261 __update_records_size(next - start));
262 if ((reclen + LLOG_MIN_REC_SIZE <= ctxt->loc_chunk_size ||
263 reclen == ctxt->loc_chunk_size) &&
268 record->lur_update_rec.ur_update_count)
270 else if (param_count <
271 record->lur_update_rec.ur_param_count)
276 lur->lur_update_rec.ur_update_count = update_count -
278 lur->lur_update_rec.ur_param_count = param_count -
280 memcpy(&lur->lur_update_rec.ur_ops, start, cur - start);
281 lur->lur_hdr.lrh_len = llog_update_record_size(lur);
283 LASSERT(lur->lur_hdr.lrh_len ==
284 __llog_update_record_size(
285 __update_records_size(cur - start)));
286 LASSERT(lur->lur_hdr.lrh_len <= ctxt->loc_chunk_size);
288 update_records_dump(&lur->lur_update_rec, D_INFO, true);
292 GOTO(llog_put, rc = -ENOMEM);
293 INIT_LIST_HEAD(&stc->stc_list);
295 rc = llog_add(env, ctxt->loc_handle, &lur->lur_hdr,
296 &stc->stc_cookie, sub_th->st_sub_th);
298 CDEBUG(D_INFO, "%s: Add update log "DFID".%u: rc = %d\n",
299 dt->dd_lu_dev.ld_obd->obd_name,
300 PLOGID(&stc->stc_cookie.lgc_lgl),
301 stc->stc_cookie.lgc_index, rc);
304 list_add(&stc->stc_list, &sub_th->st_cookie_list);
311 last_update_count = update_count;
312 last_param_count = param_count;
314 lur->lur_update_rec.ur_update_count = 0;
315 lur->lur_update_rec.ur_param_count = 0;
316 lur->lur_update_rec.ur_flags |= UPDATE_RECORD_CONTINUE;
321 OBD_FREE_LARGE(lur, ctxt->loc_chunk_size);
328 * Prepare the update records.
330 * Merge params and ops into the update records, then initializing
333 * During transaction execution phase, parameters and update ops
334 * are collected in two different buffers (see lod_updates_pack()),
335 * during transaction stop, it needs to be merged in one buffer,
336 * so it will be written in the update log.
338 * \param[in] env execution environment
339 * \param[in] tmt top_multiple_thandle for distribute txn
341 * \retval 0 if merging succeeds.
342 * \retval negaitive errno if merging fails.
344 static int prepare_writing_updates(const struct lu_env *env,
345 struct top_multiple_thandle *tmt)
347 struct thandle_update_records *tur = tmt->tmt_update_records;
348 struct llog_update_record *lur;
349 struct update_params *params;
353 if (tur == NULL || tur->tur_update_records == NULL ||
354 tur->tur_update_params == NULL)
357 lur = tur->tur_update_records;
358 /* Extends the update records buffer if needed */
359 params_size = update_params_size(tur->tur_update_params,
360 tur->tur_update_param_count);
361 LASSERT(lur->lur_update_rec.ur_param_count == 0);
362 update_size = round_up(llog_update_record_size(lur) + params_size, 8);
363 if (update_size > tur->tur_update_records_buf_size) {
366 rc = tur_update_records_extend(tur, update_size);
370 lur = tur->tur_update_records;
373 params = update_records_get_params(&lur->lur_update_rec);
374 memcpy(params, tur->tur_update_params, params_size);
376 lur->lur_update_rec.ur_param_count = tur->tur_update_param_count;
377 lur->lur_update_rec.ur_batchid = tmt->tmt_batchid;
378 /* Init update record header */
379 lur->lur_hdr.lrh_len = llog_update_record_size(lur);
380 lur->lur_hdr.lrh_type = UPDATE_REC;
382 /* Dump updates for debugging purpose */
383 update_records_dump(&lur->lur_update_rec, D_INFO, true);
389 * Top thandle commit callback
391 * This callback will be called when all of sub transactions are committed.
393 * \param[in] th top thandle to be committed.
395 static void top_trans_committed_cb(struct top_multiple_thandle *tmt)
397 struct lu_target *lut;
400 LASSERT(kref_read(&tmt->tmt_refcount) > 0);
402 top_multiple_thandle_dump(tmt, D_HA);
403 tmt->tmt_committed = 1;
404 lut = dt2lu_dev(tmt->tmt_master_sub_dt)->ld_site->ls_tgt;
405 if (lut->lut_tdtd && lut->lut_tdtd->tdtd_commit_task)
406 wake_up_process(lut->lut_tdtd->tdtd_commit_task);
411 struct sub_thandle *lookup_sub_thandle(struct top_multiple_thandle *tmt,
412 struct dt_device *dt_dev)
414 struct sub_thandle *st;
416 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
417 if (st->st_dt == dt_dev)
422 EXPORT_SYMBOL(lookup_sub_thandle);
424 struct sub_thandle *create_sub_thandle(struct top_multiple_thandle *tmt,
425 struct dt_device *dt_dev)
427 struct sub_thandle *st;
431 RETURN(ERR_PTR(-ENOMEM));
433 INIT_LIST_HEAD(&st->st_sub_list);
434 INIT_LIST_HEAD(&st->st_cookie_list);
437 list_add(&st->st_sub_list, &tmt->tmt_sub_thandle_list);
441 static void sub_trans_commit_cb_internal(struct top_multiple_thandle *tmt,
442 struct thandle *sub_th, int err)
444 struct sub_thandle *st;
445 bool all_committed = true;
447 /* Check if all sub thandles are committed */
448 spin_lock(&tmt->tmt_sub_lock);
449 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
450 if (st->st_sub_th == sub_th) {
451 st->st_committed = 1;
454 if (!st->st_committed)
455 all_committed = false;
457 spin_unlock(&tmt->tmt_sub_lock);
459 if (tmt->tmt_result == 0)
460 tmt->tmt_result = err;
463 top_trans_committed_cb(tmt);
465 top_multiple_thandle_dump(tmt, D_INFO);
466 top_multiple_thandle_put(tmt);
471 * sub thandle commit callback
473 * Mark the sub thandle to be committed and if all sub thandle are committed
474 * notify the top thandle.
476 * \param[in] env execution environment
477 * \param[in] sub_th sub thandle being committed
478 * \param[in] cb commit callback
479 * \param[in] err trans result
481 static void sub_trans_commit_cb(struct lu_env *env,
482 struct thandle *sub_th,
483 struct dt_txn_commit_cb *cb, int err)
485 struct top_multiple_thandle *tmt = cb->dcb_data;
487 sub_trans_commit_cb_internal(tmt, sub_th, err);
490 static void sub_thandle_register_commit_cb(struct sub_thandle *st,
491 struct top_multiple_thandle *tmt)
493 LASSERT(st->st_sub_th != NULL);
494 top_multiple_thandle_get(tmt);
495 st->st_commit_dcb.dcb_func = sub_trans_commit_cb;
496 st->st_commit_dcb.dcb_data = tmt;
497 INIT_LIST_HEAD(&st->st_commit_dcb.dcb_linkage);
498 dt_trans_cb_add(st->st_sub_th, &st->st_commit_dcb);
502 * Sub thandle stop call back
504 * After sub thandle is stopped, it will call this callback to notify
507 * \param[in] th sub thandle to be stopped
508 * \param[in] rc result of sub trans
510 static void sub_trans_stop_cb(struct lu_env *env,
511 struct thandle *sub_th,
512 struct dt_txn_commit_cb *cb, int err)
514 struct sub_thandle *st;
515 struct top_multiple_thandle *tmt = cb->dcb_data;
518 spin_lock(&tmt->tmt_sub_lock);
519 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
523 if (st->st_dt == sub_th->th_dev) {
529 spin_unlock(&tmt->tmt_sub_lock);
531 wake_up(&tmt->tmt_stop_waitq);
535 static void sub_thandle_register_stop_cb(struct sub_thandle *st,
536 struct top_multiple_thandle *tmt)
538 st->st_stop_dcb.dcb_func = sub_trans_stop_cb;
539 st->st_stop_dcb.dcb_data = tmt;
540 st->st_stop_dcb.dcb_flags = DCB_TRANS_STOP;
541 INIT_LIST_HEAD(&st->st_stop_dcb.dcb_linkage);
542 dt_trans_cb_add(st->st_sub_th, &st->st_stop_dcb);
548 * Create transaction handle for sub_thandle
550 * \param[in] env execution environment
551 * \param[in] th top thandle
552 * \param[in] st sub_thandle
554 * \retval 0 if creation succeeds.
555 * \retval negative errno if creation fails.
557 int sub_thandle_trans_create(const struct lu_env *env,
558 struct top_thandle *top_th,
559 struct sub_thandle *st)
561 struct thandle *sub_th;
563 sub_th = dt_trans_create(env, st->st_dt);
565 return PTR_ERR(sub_th);
567 sub_th->th_top = &top_th->tt_super;
568 st->st_sub_th = sub_th;
570 sub_th->th_wait_submit = 1;
571 sub_thandle_register_stop_cb(st, top_th->tt_multiple_thandle);
576 * Create the top transaction.
578 * Create the top transaction on the master device. It will create a top
579 * thandle and a sub thandle on the master device.
581 * \param[in] env execution environment
582 * \param[in] master_dev master_dev the top thandle will be created
584 * \retval pointer to the created thandle.
585 * \retval ERR_PTR(errno) if creation failed.
588 top_trans_create(const struct lu_env *env, struct dt_device *master_dev)
590 struct top_thandle *top_th;
591 struct thandle *child_th;
593 OBD_ALLOC_GFP(top_th, sizeof(*top_th), GFP_NOFS);
595 return ERR_PTR(-ENOMEM);
597 top_th->tt_super.th_top = &top_th->tt_super;
599 if (master_dev != NULL) {
600 child_th = dt_trans_create(env, master_dev);
601 if (IS_ERR(child_th)) {
602 OBD_FREE_PTR(top_th);
606 child_th->th_top = &top_th->tt_super;
607 child_th->th_wait_submit = 1;
608 top_th->tt_master_sub_thandle = child_th;
610 return &top_th->tt_super;
612 EXPORT_SYMBOL(top_trans_create);
615 * Declare write update transaction
617 * Check if there are updates being recorded in this transaction,
618 * it will write the record into the disk.
620 * \param[in] env execution environment
621 * \param[in] tmt top multiple transaction handle
623 * \retval 0 if writing succeeds
624 * \retval negative errno if writing fails
626 static int declare_updates_write(const struct lu_env *env,
627 struct top_multiple_thandle *tmt)
629 struct llog_update_record *record;
630 struct sub_thandle *st;
633 record = tmt->tmt_update_records->tur_update_records;
634 /* Declare update write for all other target */
635 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
636 if (st->st_sub_th == NULL)
639 rc = sub_declare_updates_write(env, record, st->st_sub_th,
640 tmt->tmt_record_size);
649 * Assign batchid to the distribute transaction.
651 * Assign batchid to the distribute transaction
653 * \param[in] tmt distribute transaction
655 static void distribute_txn_assign_batchid(struct top_multiple_thandle *new)
657 struct target_distribute_txn_data *tdtd;
658 struct dt_device *dt = new->tmt_master_sub_dt;
659 struct sub_thandle *st;
662 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
663 spin_lock(&tdtd->tdtd_batchid_lock);
664 new->tmt_batchid = tdtd->tdtd_batchid++;
665 list_add_tail(&new->tmt_commit_list, &tdtd->tdtd_list);
666 spin_unlock(&tdtd->tdtd_batchid_lock);
667 list_for_each_entry(st, &new->tmt_sub_thandle_list, st_sub_list) {
668 if (st->st_sub_th != NULL)
669 sub_thandle_register_commit_cb(st, new);
671 top_multiple_thandle_get(new);
672 top_multiple_thandle_dump(new, D_INFO);
676 * Insert distribute transaction to the distribute txn list.
678 * Insert distribute transaction to the distribute txn list.
680 * \param[in] new the distribute txn to be inserted.
682 void distribute_txn_insert_by_batchid(struct top_multiple_thandle *new)
684 struct dt_device *dt = new->tmt_master_sub_dt;
685 struct top_multiple_thandle *tmt;
686 struct target_distribute_txn_data *tdtd;
687 struct sub_thandle *st;
688 bool at_head = false;
691 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
693 spin_lock(&tdtd->tdtd_batchid_lock);
694 list_for_each_entry_reverse(tmt, &tdtd->tdtd_list, tmt_commit_list) {
695 if (new->tmt_batchid > tmt->tmt_batchid) {
696 list_add(&new->tmt_commit_list, &tmt->tmt_commit_list);
700 if (list_empty(&new->tmt_commit_list)) {
702 list_add(&new->tmt_commit_list, &tdtd->tdtd_list);
704 spin_unlock(&tdtd->tdtd_batchid_lock);
706 list_for_each_entry(st, &new->tmt_sub_thandle_list, st_sub_list) {
707 if (st->st_sub_th != NULL)
708 sub_thandle_register_commit_cb(st, new);
711 top_multiple_thandle_get(new);
712 top_multiple_thandle_dump(new, D_INFO);
713 if (new->tmt_committed && at_head && tdtd->tdtd_commit_task)
714 wake_up_process(tdtd->tdtd_commit_task);
718 * Prepare cross-MDT operation.
720 * Create the update record buffer to record updates for cross-MDT operation,
721 * add master sub transaction to tt_sub_trans_list, and declare the update
724 * During updates packing, all of parameters will be packed in
725 * tur_update_params, and updates will be packed in tur_update_records.
726 * Then in transaction stop, parameters and updates will be merged
727 * into one updates buffer.
729 * And also master thandle will be added to the sub_th list, so it will be
730 * easy to track the commit status.
732 * \param[in] env execution environment
733 * \param[in] th top transaction handle
735 * \retval 0 if preparation succeeds.
736 * \retval negative errno if preparation fails.
738 static int prepare_multiple_node_trans(const struct lu_env *env,
739 struct top_multiple_thandle *tmt)
741 struct thandle_update_records *tur;
745 if (tmt->tmt_update_records == NULL) {
746 tur = &update_env_info(env)->uti_tur;
747 rc = check_and_prepare_update_record(env, tur);
751 tmt->tmt_update_records = tur;
752 distribute_txn_assign_batchid(tmt);
755 rc = declare_updates_write(env, tmt);
761 * start the top transaction.
763 * Start all of its sub transactions, then start master sub transaction.
765 * \param[in] env execution environment
766 * \param[in] master_dev master_dev the top thandle will be start
767 * \param[in] th top thandle
769 * \retval 0 if transaction start succeeds.
770 * \retval negative errno if start fails.
772 int top_trans_start(const struct lu_env *env, struct dt_device *master_dev,
775 struct top_thandle *top_th = container_of(th, struct top_thandle,
777 struct sub_thandle *st;
778 struct top_multiple_thandle *tmt = top_th->tt_multiple_thandle;
784 top_th->tt_master_sub_thandle->th_sync = th->th_sync;
786 top_th->tt_master_sub_thandle->th_local = th->th_local;
787 rc = dt_trans_start(env, top_th->tt_master_sub_thandle->th_dev,
788 top_th->tt_master_sub_thandle);
792 tmt = top_th->tt_multiple_thandle;
793 rc = prepare_multiple_node_trans(env, tmt);
797 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
798 if (st->st_sub_th == NULL)
801 st->st_sub_th->th_sync = th->th_sync;
803 st->st_sub_th->th_local = th->th_local;
804 rc = dt_trans_start(env, st->st_sub_th->th_dev,
809 LASSERT(st->st_started == 0);
816 EXPORT_SYMBOL(top_trans_start);
819 * Check whether we need write updates record
821 * Check if the updates for the top_thandle needs to be writen
822 * to all targets. Only if the transaction succeeds and the updates
823 * number > 2, it will write the updates,
825 * \params [in] top_th top thandle.
827 * \retval true if it needs to write updates
828 * \retval false if it does not need to write updates
830 static bool top_check_write_updates(struct top_thandle *top_th)
832 struct top_multiple_thandle *tmt;
833 struct thandle_update_records *tur;
835 /* Do not write updates to records if the transaction fails */
836 if (top_th->tt_super.th_result != 0)
839 tmt = top_th->tt_multiple_thandle;
843 tur = tmt->tmt_update_records;
847 /* Hmm, false update records, since the cross-MDT operation
848 * should includes both local and remote updates, so the
849 * updates count should >= 2 */
850 if (tur->tur_update_records == NULL ||
851 tur->tur_update_records->lur_update_rec.ur_update_count <= 1)
858 * Check if top transaction is stopped
860 * Check if top transaction is stopped, only if all sub transaction
861 * is stopped, then the top transaction is stopped.
863 * \param [in] top_th top thandle
865 * \retval true if the top transaction is stopped.
866 * \retval false if the top transaction is not stopped.
868 static bool top_trans_is_stopped(struct top_thandle *top_th)
870 struct top_multiple_thandle *tmt;
871 struct sub_thandle *st;
872 bool all_stopped = true;
874 tmt = top_th->tt_multiple_thandle;
875 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
876 if (!st->st_stopped && st->st_sub_th != NULL) {
881 if (st->st_result != 0 &&
882 top_th->tt_super.th_result == 0)
883 top_th->tt_super.th_result = st->st_result;
890 * Wait result of top transaction
892 * Wait until all sub transaction get its result.
894 * \param [in] top_th top thandle.
896 * \retval the result of top thandle.
898 static int top_trans_wait_result(struct top_thandle *top_th)
900 wait_event_idle(top_th->tt_multiple_thandle->tmt_stop_waitq,
901 top_trans_is_stopped(top_th));
903 RETURN(top_th->tt_super.th_result);
907 * Stop the top transaction.
909 * Stop the transaction on the master device first, then stop transactions
910 * on other sub devices.
912 * \param[in] env execution environment
913 * \param[in] master_dev master_dev the top thandle will be created
914 * \param[in] th top thandle
916 * \retval 0 if stop transaction succeeds.
917 * \retval negative errno if stop transaction fails.
919 int top_trans_stop(const struct lu_env *env, struct dt_device *master_dev,
922 struct top_thandle *top_th = container_of(th, struct top_thandle,
924 struct sub_thandle *st;
925 struct sub_thandle *master_st;
926 struct top_multiple_thandle *tmt;
927 struct thandle_update_records *tur;
928 bool write_updates = false;
932 if (likely(top_th->tt_multiple_thandle == NULL)) {
933 LASSERT(master_dev != NULL);
936 top_th->tt_master_sub_thandle->th_sync = th->th_sync;
938 top_th->tt_master_sub_thandle->th_local = th->th_local;
939 rc = dt_trans_stop(env, master_dev,
940 top_th->tt_master_sub_thandle);
941 OBD_FREE_PTR(top_th);
945 tmt = top_th->tt_multiple_thandle;
946 tur = tmt->tmt_update_records;
948 /* Note: we need stop the master thandle first, then the stop
949 * callback will fill the master transno in the update logs,
950 * then these update logs will be sent to other MDTs */
951 /* get the master sub thandle */
952 master_st = lookup_sub_thandle(tmt, tmt->tmt_master_sub_dt);
953 write_updates = top_check_write_updates(top_th);
955 /* Step 1: write the updates log on Master MDT */
956 if (master_st != NULL && master_st->st_sub_th != NULL &&
958 struct llog_update_record *lur;
960 /* Merge the parameters and updates into one buffer */
961 rc = prepare_writing_updates(env, tmt);
963 CERROR("%s: cannot prepare updates: rc = %d\n",
964 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
966 write_updates = false;
967 GOTO(stop_master_trans, rc);
970 lur = tur->tur_update_records;
971 /* Write updates to the master MDT */
972 rc = sub_updates_write(env, lur, master_st);
974 /* Cleanup the common parameters in the update records,
975 * master transno callback might add more parameters.
976 * and we need merge the update records again in the
978 if (tur->tur_update_params != NULL)
979 lur->lur_update_rec.ur_param_count = 0;
982 CERROR("%s: write updates failed: rc = %d\n",
983 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
985 write_updates = false;
986 GOTO(stop_master_trans, rc);
991 /* Step 2: Stop the transaction on the master MDT, and fill the
992 * master transno in the update logs to other MDT. */
993 if (master_st != NULL && master_st->st_sub_th != NULL) {
995 master_st->st_sub_th->th_local = th->th_local;
997 master_st->st_sub_th->th_sync = th->th_sync;
998 master_st->st_sub_th->th_result = th->th_result;
999 rc = dt_trans_stop(env, master_st->st_dt, master_st->st_sub_th);
1000 /* If it does not write_updates, then we call submit callback
1001 * here, otherwise callback is done through
1002 * osd(osp)_trans_commit_cb() */
1003 if (!master_st->st_started &&
1004 !list_empty(&tmt->tmt_commit_list))
1005 sub_trans_commit_cb_internal(tmt,
1006 master_st->st_sub_th, rc);
1008 CERROR("%s: stop trans failed: rc = %d\n",
1009 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
1011 GOTO(stop_other_trans, rc);
1012 } else if (tur != NULL && tur->tur_update_records != NULL) {
1013 struct llog_update_record *lur;
1015 lur = tur->tur_update_records;
1016 if (lur->lur_update_rec.ur_master_transno == 0)
1017 /* Update master transno after master stop
1019 lur->lur_update_rec.ur_master_transno =
1020 tgt_th_info(env)->tti_transno;
1024 /* Step 3: write updates to other MDTs */
1025 if (write_updates) {
1026 struct llog_update_record *lur;
1027 if (CFS_FAIL_PRECHECK(OBD_FAIL_OUT_OBJECT_MISS)) {
1028 if (cfs_fail_val == 1) {
1029 long timeout = cfs_time_seconds(1) / 10;
1031 CFS_RACE(OBD_FAIL_OUT_OBJECT_MISS);
1032 set_current_state(TASK_UNINTERRUPTIBLE);
1033 schedule_timeout(schedule_timeout(timeout));
1039 /* Stop callback of master will add more updates and also update
1040 * master transno, so merge the parameters and updates into one
1042 rc = prepare_writing_updates(env, tmt);
1044 CERROR("%s: prepare updates failed: rc = %d\n",
1045 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
1047 GOTO(stop_other_trans, rc);
1049 lur = tur->tur_update_records;
1050 list_for_each_entry(st, &tmt->tmt_sub_thandle_list,
1052 if (st->st_sub_th == NULL || st == master_st ||
1053 st->st_sub_th->th_result < 0)
1056 rc = sub_updates_write(env, lur, st);
1058 CERROR("%s: write updates failed: rc = %d\n",
1059 st->st_dt->dd_lu_dev.ld_obd->obd_name,
1068 /* Step 4: Stop the transaction on other MDTs */
1069 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1070 if (st == master_st || st->st_sub_th == NULL)
1074 st->st_sub_th->th_sync = th->th_sync;
1076 st->st_sub_th->th_local = th->th_local;
1077 st->st_sub_th->th_result = th->th_result;
1078 rc = dt_trans_stop(env, st->st_sub_th->th_dev,
1081 CERROR("%s: stop trans failed: rc = %d\n",
1082 st->st_dt->dd_lu_dev.ld_obd->obd_name, rc);
1083 if (th->th_result == 0)
1088 rc = top_trans_wait_result(top_th);
1090 tmt->tmt_result = rc;
1092 /* Balance for the refcount in top_trans_create, Note: if it is NOT
1093 * multiple node transaction, the top transaction will be destroyed. */
1094 top_multiple_thandle_put(tmt);
1095 OBD_FREE_PTR(top_th);
1098 EXPORT_SYMBOL(top_trans_stop);
1101 * Create top_multiple_thandle for top_thandle
1103 * Create top_mutilple_thandle to manage the mutiple node transaction
1104 * for top_thandle, and it also needs to add master sub thandle to the
1105 * sub trans list now.
1107 * \param[in] env execution environment
1108 * \param[in] top_th the top thandle
1110 * \retval 0 if creation succeeds
1111 * \retval negative errno if creation fails
1113 int top_trans_create_tmt(const struct lu_env *env,
1114 struct top_thandle *top_th)
1116 struct top_multiple_thandle *tmt;
1122 tmt->tmt_magic = TOP_THANDLE_MAGIC;
1123 INIT_LIST_HEAD(&tmt->tmt_sub_thandle_list);
1124 INIT_LIST_HEAD(&tmt->tmt_commit_list);
1125 kref_init(&tmt->tmt_refcount);
1126 spin_lock_init(&tmt->tmt_sub_lock);
1127 init_waitqueue_head(&tmt->tmt_stop_waitq);
1129 top_th->tt_multiple_thandle = tmt;
1134 static struct sub_thandle *
1135 create_sub_thandle_with_thandle(struct top_thandle *top_th,
1136 struct thandle *sub_th)
1138 struct sub_thandle *st;
1140 /* create and init sub th to the top trans list */
1141 st = create_sub_thandle(top_th->tt_multiple_thandle,
1146 st->st_sub_th = sub_th;
1148 sub_th->th_top = &top_th->tt_super;
1149 sub_thandle_register_stop_cb(st, top_th->tt_multiple_thandle);
1156 * Get sub thandle from the top thandle according to the sub dt_device.
1158 * \param[in] env execution environment
1159 * \param[in] th thandle on the top layer.
1160 * \param[in] sub_dt sub dt_device used to get sub transaction
1162 * \retval thandle of sub transaction if succeed
1163 * \retval PTR_ERR(errno) if failed
1165 struct thandle *thandle_get_sub_by_dt(const struct lu_env *env,
1167 struct dt_device *sub_dt)
1169 struct sub_thandle *st = NULL;
1170 struct sub_thandle *master_st = NULL;
1171 struct top_thandle *top_th;
1172 struct thandle *sub_th = NULL;
1176 top_th = container_of(th, struct top_thandle, tt_super);
1178 if (likely(sub_dt == top_th->tt_master_sub_thandle->th_dev))
1179 RETURN(top_th->tt_master_sub_thandle);
1181 if (top_th->tt_multiple_thandle != NULL) {
1182 st = lookup_sub_thandle(top_th->tt_multiple_thandle, sub_dt);
1184 RETURN(st->st_sub_th);
1187 sub_th = dt_trans_create(env, sub_dt);
1191 /* Create top_multiple_thandle if necessary */
1192 if (top_th->tt_multiple_thandle == NULL) {
1193 struct top_multiple_thandle *tmt;
1195 rc = top_trans_create_tmt(env, top_th);
1197 GOTO(stop_trans, rc);
1199 tmt = top_th->tt_multiple_thandle;
1201 /* Add master sub th to the top trans list */
1202 tmt->tmt_master_sub_dt =
1203 top_th->tt_master_sub_thandle->th_dev;
1204 master_st = create_sub_thandle_with_thandle(top_th,
1205 top_th->tt_master_sub_thandle);
1206 if (IS_ERR(master_st)) {
1207 rc = PTR_ERR(master_st);
1209 GOTO(stop_trans, rc);
1213 /* create and init sub th to the top trans list */
1214 st = create_sub_thandle_with_thandle(top_th, sub_th);
1218 GOTO(stop_trans, rc);
1220 st->st_sub_th->th_wait_submit = 1;
1223 if (master_st != NULL) {
1224 list_del(&master_st->st_sub_list);
1225 OBD_FREE_PTR(master_st);
1227 sub_th->th_result = rc;
1228 dt_trans_stop(env, sub_dt, sub_th);
1229 sub_th = ERR_PTR(rc);
1234 EXPORT_SYMBOL(thandle_get_sub_by_dt);
1237 * Top multiple thandle destroy
1239 * Destroy multiple thandle and all its sub thandle.
1241 * \param[in] kref Pointer to struct kref
1243 void top_multiple_thandle_destroy(struct kref *kref)
1245 struct top_multiple_thandle *tmt;
1246 struct sub_thandle *st;
1247 struct sub_thandle *tmp;
1249 tmt = container_of(kref, struct top_multiple_thandle, tmt_refcount);
1251 LASSERT(tmt->tmt_magic == TOP_THANDLE_MAGIC);
1252 list_for_each_entry_safe(st, tmp, &tmt->tmt_sub_thandle_list,
1254 struct sub_thandle_cookie *stc;
1255 struct sub_thandle_cookie *tmp;
1257 list_del(&st->st_sub_list);
1258 list_for_each_entry_safe(stc, tmp, &st->st_cookie_list,
1260 list_del(&stc->stc_list);
1267 EXPORT_SYMBOL(top_multiple_thandle_destroy);
1270 * Cancel the update log on MDTs
1272 * Cancel the update log on MDTs then destroy the thandle.
1274 * \param[in] env execution environment
1275 * \param[in] tmt the top multiple thandle whose updates records
1276 * will be cancelled.
1278 * \retval 0 if cancellation succeeds.
1279 * \retval negative errno if cancellation fails.
1281 static int distribute_txn_cancel_records(const struct lu_env *env,
1282 struct top_multiple_thandle *tmt)
1284 struct sub_thandle *st;
1287 if (CFS_FAIL_CHECK(OBD_FAIL_TGT_TXN_NO_CANCEL))
1290 top_multiple_thandle_dump(tmt, D_INFO);
1291 /* Cancel update logs on other MDTs */
1292 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1293 struct llog_ctxt *ctxt;
1294 struct obd_device *obd;
1295 struct llog_cookie *cookie;
1296 struct sub_thandle_cookie *stc;
1299 obd = st->st_dt->dd_lu_dev.ld_obd;
1300 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
1303 list_for_each_entry(stc, &st->st_cookie_list, stc_list) {
1304 cookie = &stc->stc_cookie;
1305 if (fid_is_zero(&cookie->lgc_lgl.lgl_oi.oi_fid))
1308 rc = llog_cat_cancel_records(env, ctxt->loc_handle, 1,
1310 CDEBUG(D_HA, "%s: batchid %llu cancel update log "
1311 DFID".%u: rc = %d\n", obd->obd_name,
1312 tmt->tmt_batchid, PLOGID(&cookie->lgc_lgl),
1313 cookie->lgc_index, rc);
1316 llog_ctxt_put(ctxt);
1322 struct distribute_txn_bid_data {
1323 struct dt_txn_commit_cb dtbd_cb;
1324 struct target_distribute_txn_data *dtbd_tdtd;
1329 * callback of updating commit batchid
1331 * Updating commit batchid then wake up the commit thread to cancel the
1334 * \param[in]env execution environment
1335 * \param[in]th thandle to updating commit batchid
1336 * \param[in]cb commit callback
1337 * \param[in]err result of thandle
1339 static void distribute_txn_batchid_cb(struct lu_env *env,
1341 struct dt_txn_commit_cb *cb,
1344 struct distribute_txn_bid_data *dtbd = NULL;
1345 struct target_distribute_txn_data *tdtd;
1347 dtbd = container_of(cb, struct distribute_txn_bid_data, dtbd_cb);
1348 tdtd = dtbd->dtbd_tdtd;
1350 CDEBUG(D_HA, "%s: %llu batchid updated\n",
1351 tdtd->tdtd_lut->lut_obd->obd_name, dtbd->dtbd_batchid);
1352 spin_lock(&tdtd->tdtd_batchid_lock);
1353 if (dtbd->dtbd_batchid > tdtd->tdtd_committed_batchid &&
1354 !tdtd->tdtd_lut->lut_obd->obd_no_transno)
1355 tdtd->tdtd_committed_batchid = dtbd->dtbd_batchid;
1356 spin_unlock(&tdtd->tdtd_batchid_lock);
1357 if (atomic_dec_and_test(&tdtd->tdtd_refcount))
1358 wake_up_process(tdtd->tdtd_commit_task);
1364 * Update the commit batchid in disk
1366 * Update commit batchid in the disk, after this is committed, it can start
1367 * to cancel the update records.
1369 * \param[in] env execution environment
1370 * \param[in] tdtd distribute transaction structure
1371 * \param[in] batchid commit batchid to be updated
1373 * \retval 0 if update succeeds.
1374 * \retval negative errno if update fails.
1377 distribute_txn_commit_batchid_update(const struct lu_env *env,
1378 struct target_distribute_txn_data *tdtd,
1381 struct distribute_txn_bid_data *dtbd = NULL;
1389 OBD_ALLOC_PTR(dtbd);
1392 dtbd->dtbd_batchid = batchid;
1393 dtbd->dtbd_tdtd = tdtd;
1394 dtbd->dtbd_cb.dcb_func = distribute_txn_batchid_cb;
1395 atomic_inc(&tdtd->tdtd_refcount);
1397 th = dt_trans_create(env, tdtd->tdtd_lut->lut_bottom);
1399 atomic_dec(&tdtd->tdtd_refcount);
1401 RETURN(PTR_ERR(th));
1404 tmp = cpu_to_le64(batchid);
1406 buf.lb_len = sizeof(tmp);
1409 rc = dt_declare_record_write(env, tdtd->tdtd_batchid_obj, &buf, off,
1414 rc = dt_trans_start_local(env, tdtd->tdtd_lut->lut_bottom, th);
1418 rc = dt_trans_cb_add(th, &dtbd->dtbd_cb);
1422 rc = dt_record_write(env, tdtd->tdtd_batchid_obj, &buf,
1425 CDEBUG(D_INFO, "%s: update batchid %llu: rc = %d\n",
1426 tdtd->tdtd_lut->lut_obd->obd_name, batchid, rc);
1429 dt_trans_stop(env, tdtd->tdtd_lut->lut_bottom, th);
1431 atomic_dec(&tdtd->tdtd_refcount);
1438 * Init commit batchid for distribute transaction.
1440 * Initialize the batchid object and get commit batchid from the object.
1442 * \param[in] env execution environment
1443 * \param[in] tdtd distribute transaction whose batchid is initialized.
1445 * \retval 0 if initialization succeeds.
1446 * \retval negative errno if initialization fails.
1449 distribute_txn_commit_batchid_init(const struct lu_env *env,
1450 struct target_distribute_txn_data *tdtd)
1452 struct tgt_thread_info *tti = tgt_th_info(env);
1453 struct lu_target *lut = tdtd->tdtd_lut;
1454 struct lu_attr *attr = &tti->tti_attr;
1455 struct lu_fid *fid = &tti->tti_fid1;
1456 struct dt_object_format *dof = &tti->tti_u.update.tti_update_dof;
1457 struct dt_object *dt_obj = NULL;
1464 memset(attr, 0, sizeof(*attr));
1465 attr->la_valid = LA_MODE;
1466 attr->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1467 dof->dof_type = dt_mode_to_dft(S_IFREG);
1469 lu_local_obj_fid(fid, BATCHID_COMMITTED_OID);
1471 dt_obj = dt_find_or_create(env, lut->lut_bottom, fid, dof,
1473 if (IS_ERR(dt_obj)) {
1474 rc = PTR_ERR(dt_obj);
1479 tdtd->tdtd_batchid_obj = dt_obj;
1482 buf.lb_len = sizeof(tmp);
1484 rc = dt_read(env, dt_obj, &buf, &off);
1485 if (rc < 0 || (rc < buf.lb_len && rc > 0)) {
1486 CERROR("%s can't read last committed batchid: rc = %d\n",
1487 tdtd->tdtd_lut->lut_obd->obd_name, rc);
1491 } else if (rc == buf.lb_len) {
1492 tdtd->tdtd_committed_batchid = le64_to_cpu(tmp);
1493 CDEBUG(D_HA, "%s: committed batchid %llu\n",
1494 tdtd->tdtd_lut->lut_obd->obd_name,
1495 tdtd->tdtd_committed_batchid);
1500 if (rc < 0 && dt_obj != NULL) {
1501 dt_object_put(env, dt_obj);
1502 tdtd->tdtd_batchid_obj = NULL;
1508 #define TASK_IDLE TASK_INTERRUPTIBLE
1512 * manage the distribute transaction thread
1514 * Distribute transaction are linked to the list, and once the distribute
1515 * transaction is committed, it will update the last committed batchid first,
1516 * after it is committed, it will cancel the records.
1518 * \param[in] _arg argument for commit thread
1520 * \retval 0 if thread is running successfully
1521 * \retval negative errno if the thread can not be run.
1523 static int distribute_txn_commit_thread(void *_arg)
1525 struct target_distribute_txn_data *tdtd = _arg;
1526 struct lu_env *env = &tdtd->tdtd_env;
1529 struct top_multiple_thandle *tmt;
1530 struct top_multiple_thandle *tmp;
1531 __u64 batchid = 0, committed;
1536 CDEBUG(D_HA, "%s: start commit thread committed batchid %llu\n",
1537 tdtd->tdtd_lut->lut_obd->obd_name,
1538 tdtd->tdtd_committed_batchid);
1540 while (({set_current_state(TASK_IDLE);
1541 !kthread_should_stop(); })) {
1542 spin_lock(&tdtd->tdtd_batchid_lock);
1543 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1545 if (tmt->tmt_committed == 0)
1548 /* Note: right now, replay is based on master MDT
1549 * transno, but cancellation is based on batchid.
1550 * so we do not try to cancel the update log until
1551 * the recoverying is done, unless the update records
1552 * batchid < committed_batchid. */
1553 if (tmt->tmt_batchid <= tdtd->tdtd_committed_batchid) {
1554 __set_current_state(TASK_RUNNING);
1555 list_move_tail(&tmt->tmt_commit_list, &list);
1556 } else if (!tdtd->tdtd_lut->lut_obd->obd_recovering) {
1557 __set_current_state(TASK_RUNNING);
1558 LASSERTF(tmt->tmt_batchid >= batchid,
1559 "tmt %px tmt_batchid: %llu, batchid %llu\n",
1560 tmt, tmt->tmt_batchid, batchid);
1561 /* There are three types of distribution
1562 * transaction result
1564 * 1. If tmt_result < 0, it means the
1565 * distribution transaction fails, which should
1566 * be rare, because once declare phase succeeds,
1567 * the operation should succeeds anyway. Note in
1568 * this case, we will still update batchid so
1569 * cancellation would be stopped.
1571 * 2. If tmt_result == 0, it means the
1572 * distribution transaction succeeds, and we
1573 * will update batchid.
1575 * 3. If tmt_result > 0, it means distribute
1576 * transaction is not yet committed on every
1577 * node, but we need release this tmt before
1578 * that, which usuually happens during umount.
1580 if (tmt->tmt_result <= 0)
1581 batchid = tmt->tmt_batchid;
1582 list_move_tail(&tmt->tmt_commit_list, &list);
1585 spin_unlock(&tdtd->tdtd_batchid_lock);
1587 CDEBUG(D_HA, "%s: batchid: %llu committed batchid "
1588 "%llu\n", tdtd->tdtd_lut->lut_obd->obd_name, batchid,
1589 tdtd->tdtd_committed_batchid);
1590 /* update globally committed on a storage */
1591 if (batchid > tdtd->tdtd_committed_batchid) {
1592 rc = distribute_txn_commit_batchid_update(env, tdtd,
1597 /* cancel the records for committed batchid's */
1598 /* XXX: should we postpone cancel's till the end of recovery? */
1599 committed = tdtd->tdtd_committed_batchid;
1600 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1601 if (tmt->tmt_batchid > committed)
1603 __set_current_state(TASK_RUNNING);
1604 list_del_init(&tmt->tmt_commit_list);
1605 if (tmt->tmt_result <= 0)
1606 distribute_txn_cancel_records(env, tmt);
1607 top_multiple_thandle_put(tmt);
1610 if (!task_is_running(current))
1613 if (CFS_FAIL_PRECHECK(OBD_FAIL_OUT_OBJECT_MISS)) {
1614 set_current_state(TASK_UNINTERRUPTIBLE);
1615 schedule_timeout(cfs_time_seconds(5));
1619 while (({set_current_state(TASK_IDLE);
1620 atomic_read(&tdtd->tdtd_refcount) != 0; }))
1622 __set_current_state(TASK_RUNNING);
1624 spin_lock(&tdtd->tdtd_batchid_lock);
1625 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1627 list_move_tail(&tmt->tmt_commit_list, &list);
1628 spin_unlock(&tdtd->tdtd_batchid_lock);
1630 CDEBUG(D_INFO, "%s stopping distribute txn commit thread.\n",
1631 tdtd->tdtd_lut->lut_obd->obd_name);
1632 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1633 list_del_init(&tmt->tmt_commit_list);
1634 top_multiple_thandle_dump(tmt, D_HA);
1635 top_multiple_thandle_put(tmt);
1641 * Start llog cancel thread
1643 * Start llog cancel(master/slave) thread on LOD
1645 * \param[in]lclt cancel log thread to be started.
1647 * \retval 0 if the thread is started successfully.
1648 * \retval negative errno if the thread is not being
1651 int distribute_txn_init(const struct lu_env *env,
1652 struct lu_target *lut,
1653 struct target_distribute_txn_data *tdtd,
1656 struct task_struct *task;
1660 INIT_LIST_HEAD(&tdtd->tdtd_list);
1661 INIT_LIST_HEAD(&tdtd->tdtd_replay_finish_list);
1662 INIT_LIST_HEAD(&tdtd->tdtd_replay_list);
1663 spin_lock_init(&tdtd->tdtd_batchid_lock);
1664 spin_lock_init(&tdtd->tdtd_replay_list_lock);
1665 tdtd->tdtd_replay_handler = distribute_txn_replay_handle;
1666 tdtd->tdtd_replay_ready = 0;
1668 tdtd->tdtd_batchid = lut->lut_last_transno + 1;
1670 init_waitqueue_head(&tdtd->tdtd_recovery_threads_waitq);
1671 atomic_set(&tdtd->tdtd_refcount, 0);
1672 atomic_set(&tdtd->tdtd_recovery_threads_count, 0);
1674 tdtd->tdtd_lut = lut;
1675 if (lut->lut_bottom->dd_rdonly)
1678 rc = distribute_txn_commit_batchid_init(env, tdtd);
1682 rc = lu_env_init(&tdtd->tdtd_env, LCT_LOCAL | LCT_MD_THREAD);
1686 task = kthread_create(distribute_txn_commit_thread, tdtd, "dist_txn-%u",
1689 lu_env_fini(&tdtd->tdtd_env);
1690 RETURN(PTR_ERR(task));
1692 tdtd->tdtd_commit_task = task;
1693 wake_up_process(task);
1697 EXPORT_SYMBOL(distribute_txn_init);
1700 * Stop llog cancel thread
1702 * Stop llog cancel(master/slave) thread on LOD and also destory
1703 * all of transaction in the list.
1705 * \param[in]lclt cancel log thread to be stopped.
1707 void distribute_txn_fini(const struct lu_env *env,
1708 struct target_distribute_txn_data *tdtd)
1710 struct top_multiple_thandle *tmt;
1713 /* Stop cancel thread */
1714 if (!tdtd->tdtd_commit_task)
1717 kthread_stop(tdtd->tdtd_commit_task);
1718 tdtd->tdtd_commit_task = NULL;
1720 spin_lock(&tdtd->tdtd_batchid_lock);
1721 list_splice_init(&tdtd->tdtd_list, &list);
1722 spin_unlock(&tdtd->tdtd_batchid_lock);
1724 CDEBUG(D_INFO, "%s stopping distribute txn commit thread.\n",
1725 tdtd->tdtd_lut->lut_obd->obd_name);
1726 while ((tmt = list_first_entry_or_null(&list,
1727 struct top_multiple_thandle,
1728 tmt_commit_list)) != NULL) {
1729 list_del_init(&tmt->tmt_commit_list);
1730 top_multiple_thandle_dump(tmt, D_HA);
1731 top_multiple_thandle_put(tmt);
1734 lu_env_fini(&tdtd->tdtd_env);
1736 dtrq_list_destroy(tdtd);
1737 if (tdtd->tdtd_batchid_obj != NULL) {
1738 dt_object_put(env, tdtd->tdtd_batchid_obj);
1739 tdtd->tdtd_batchid_obj = NULL;
1742 EXPORT_SYMBOL(distribute_txn_fini);