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, 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 sub_th %p\n",
87 st, st->st_dt->dd_lu_dev.ld_obd->obd_name,
88 st->st_committed, 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);
125 LASSERT(ctxt != NULL);
127 /* Not ready to record updates yet. */
128 if (ctxt->loc_handle == NULL)
129 GOTO(out_put, rc = 0);
131 rc = llog_declare_add(env, ctxt->loc_handle,
132 &record->lur_hdr, sub_th);
136 while (left > ctxt->loc_chunk_size) {
137 rc = llog_declare_add(env, ctxt->loc_handle,
138 &record->lur_hdr, sub_th);
142 left -= ctxt->loc_chunk_size;
152 * write update to sub device
154 * Write llog update record to the sub device during distribute
155 * transaction. If it succeeds, llog cookie of the record will be
156 * returned by @cookie.
158 * \param[in] env execution environment
159 * \param[in] record update records being written
160 * \param[in] sub_th sub transaction handle
161 * \param[out] cookie llog cookie of the update record.
163 * \retval 1 if writing succeeds
164 * \retval negative errno if writing fails
166 static int sub_updates_write(const struct lu_env *env,
167 struct llog_update_record *record,
168 struct sub_thandle *sub_th)
170 struct dt_device *dt = sub_th->st_dt;
171 struct llog_ctxt *ctxt;
173 struct llog_update_record *lur = NULL;
174 struct update_params *params = 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;
185 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
186 LLOG_UPDATELOG_ORIG_CTXT);
187 LASSERT(ctxt != NULL);
189 /* Not ready to record updates yet, usually happens
190 * in error handler path */
191 if (ctxt->loc_handle == NULL)
192 GOTO(llog_put, rc = 0);
194 /* Since the cross-MDT updates will includes both local
195 * and remote updates, the update ops count must > 1 */
196 LASSERT(record->lur_update_rec.ur_update_count > 1);
197 LASSERTF(record->lur_hdr.lrh_len == llog_update_record_size(record),
198 "lrh_len %u record_size %zu\n", record->lur_hdr.lrh_len,
199 llog_update_record_size(record));
201 if (likely(record->lur_hdr.lrh_len <= ctxt->loc_chunk_size)) {
204 GOTO(llog_put, rc = -ENOMEM);
205 INIT_LIST_HEAD(&stc->stc_list);
207 rc = llog_add(env, ctxt->loc_handle, &record->lur_hdr,
208 &stc->stc_cookie, sub_th->st_sub_th);
210 CDEBUG(D_INFO, "%s: Add update log "DOSTID":%u: rc = %d\n",
211 dt->dd_lu_dev.ld_obd->obd_name,
212 POSTID(&stc->stc_cookie.lgc_lgl.lgl_oi),
213 stc->stc_cookie.lgc_index, rc);
216 list_add(&stc->stc_list, &sub_th->st_cookie_list);
225 /* Split the records into chunk_size update record */
226 OBD_ALLOC_LARGE(lur, ctxt->loc_chunk_size);
228 GOTO(llog_put, rc = -ENOMEM);
230 memcpy(lur, &record->lur_hdr, sizeof(record->lur_hdr));
231 lur->lur_update_rec.ur_update_count = 0;
232 lur->lur_update_rec.ur_param_count = 0;
233 src = &record->lur_update_rec.ur_ops;
235 lur->lur_hdr.lrh_len = llog_update_record_size(lur);
236 params = update_records_get_params(&record->lur_update_rec);
240 if (update_count < record->lur_update_rec.ur_update_count) {
241 next = update_op_next_op((struct update_op *)src);
243 if (param_count == 0)
244 next = update_records_get_params(
245 &record->lur_update_rec);
248 object_update_param_size(
249 (struct object_update_param *)src);
252 rec_len = cfs_size_round((unsigned long)(next - src));
253 /* If its size > llog chunk_size, then write current chunk to
254 * the update llog. */
255 if (lur->lur_hdr.lrh_len + rec_len + LLOG_MIN_REC_SIZE >
256 ctxt->loc_chunk_size ||
257 param_count == record->lur_update_rec.ur_param_count) {
258 lur->lur_update_rec.ur_update_count =
259 update_count > last_update_count ?
260 update_count - last_update_count : 0;
261 lur->lur_update_rec.ur_param_count = param_count -
264 memcpy(&lur->lur_update_rec.ur_ops, start,
265 (unsigned long)(src - start));
266 if (last_update_count != 0)
267 lur->lur_update_rec.ur_flags |=
268 UPDATE_RECORD_CONTINUE;
270 update_records_dump(&lur->lur_update_rec, D_INFO, true);
271 lur->lur_hdr.lrh_len = llog_update_record_size(lur);
272 LASSERT(lur->lur_hdr.lrh_len <= ctxt->loc_chunk_size);
276 GOTO(llog_put, rc = -ENOMEM);
277 INIT_LIST_HEAD(&stc->stc_list);
279 rc = llog_add(env, ctxt->loc_handle,
281 &stc->stc_cookie, sub_th->st_sub_th);
283 CDEBUG(D_INFO, "%s: Add update log "DOSTID":%u"
284 " rc = %d\n", dt->dd_lu_dev.ld_obd->obd_name,
285 POSTID(&stc->stc_cookie.lgc_lgl.lgl_oi),
286 stc->stc_cookie.lgc_index, rc);
289 list_add(&stc->stc_list,
290 &sub_th->st_cookie_list);
297 last_update_count = update_count;
298 last_param_count = param_count;
300 lur->lur_update_rec.ur_update_count = 0;
301 lur->lur_update_rec.ur_param_count = 0;
302 lur->lur_hdr.lrh_len = llog_update_record_size(lur);
306 lur->lur_hdr.lrh_len += cfs_size_round(rec_len);
307 if (update_count < record->lur_update_rec.ur_update_count)
309 else if (param_count < record->lur_update_rec.ur_param_count)
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);
451 * sub thandle commit callback
453 * Mark the sub thandle to be committed and if all sub thandle are committed
454 * notify the top thandle.
456 * \param[in] env execution environment
457 * \param[in] sub_th sub thandle being committed
458 * \param[in] cb commit callback
459 * \param[in] err trans result
461 static void sub_trans_commit_cb(struct lu_env *env,
462 struct thandle *sub_th,
463 struct dt_txn_commit_cb *cb, int err)
465 struct sub_thandle *st;
466 struct top_multiple_thandle *tmt = cb->dcb_data;
467 bool all_committed = true;
470 /* Check if all sub thandles are committed */
471 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
472 if (st->st_sub_th == sub_th) {
473 st->st_committed = 1;
476 if (!st->st_committed)
477 all_committed = false;
480 if (tmt->tmt_result == 0)
481 tmt->tmt_result = err;
484 top_trans_committed_cb(tmt);
486 top_multiple_thandle_dump(tmt, D_INFO);
487 top_multiple_thandle_put(tmt);
491 static void sub_thandle_register_commit_cb(struct sub_thandle *st,
492 struct top_multiple_thandle *tmt)
494 LASSERT(st->st_sub_th != NULL);
495 top_multiple_thandle_get(tmt);
496 st->st_commit_dcb.dcb_func = sub_trans_commit_cb;
497 st->st_commit_dcb.dcb_data = tmt;
498 INIT_LIST_HEAD(&st->st_commit_dcb.dcb_linkage);
499 dt_trans_cb_add(st->st_sub_th, &st->st_commit_dcb);
503 * Sub thandle stop call back
505 * After sub thandle is stopped, it will call this callback to notify
508 * \param[in] th sub thandle to be stopped
509 * \param[in] rc result of sub trans
511 static void sub_trans_stop_cb(struct lu_env *env,
512 struct thandle *sub_th,
513 struct dt_txn_commit_cb *cb, int err)
515 struct sub_thandle *st;
516 struct top_multiple_thandle *tmt = cb->dcb_data;
519 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
523 if (st->st_dt == sub_th->th_dev) {
530 wake_up(&tmt->tmt_stop_waitq);
534 static void sub_thandle_register_stop_cb(struct sub_thandle *st,
535 struct top_multiple_thandle *tmt)
537 st->st_stop_dcb.dcb_func = sub_trans_stop_cb;
538 st->st_stop_dcb.dcb_data = tmt;
539 st->st_stop_dcb.dcb_flags = DCB_TRANS_STOP;
540 INIT_LIST_HEAD(&st->st_stop_dcb.dcb_linkage);
541 dt_trans_cb_add(st->st_sub_th, &st->st_stop_dcb);
547 * Create transaction handle for sub_thandle
549 * \param[in] env execution environment
550 * \param[in] th top thandle
551 * \param[in] st sub_thandle
553 * \retval 0 if creation succeeds.
554 * \retval negative errno if creation fails.
556 int sub_thandle_trans_create(const struct lu_env *env,
557 struct top_thandle *top_th,
558 struct sub_thandle *st)
560 struct thandle *sub_th;
562 sub_th = dt_trans_create(env, st->st_dt);
564 return PTR_ERR(sub_th);
566 sub_th->th_top = &top_th->tt_super;
567 st->st_sub_th = sub_th;
569 sub_th->th_wait_submit = 1;
574 * Create the top transaction.
576 * Create the top transaction on the master device. It will create a top
577 * thandle and a sub thandle on the master device.
579 * \param[in] env execution environment
580 * \param[in] master_dev master_dev the top thandle will be created
582 * \retval pointer to the created thandle.
583 * \retval ERR_PTR(errno) if creation failed.
586 top_trans_create(const struct lu_env *env, struct dt_device *master_dev)
588 struct top_thandle *top_th;
589 struct thandle *child_th;
591 OBD_ALLOC_GFP(top_th, sizeof(*top_th), __GFP_IO);
593 return ERR_PTR(-ENOMEM);
595 top_th->tt_super.th_top = &top_th->tt_super;
597 if (master_dev != NULL) {
598 child_th = dt_trans_create(env, master_dev);
599 if (IS_ERR(child_th)) {
600 OBD_FREE_PTR(top_th);
604 child_th->th_top = &top_th->tt_super;
605 child_th->th_wait_submit = 1;
606 top_th->tt_master_sub_thandle = child_th;
608 top_th->tt_super.th_tags |= child_th->th_tags;
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;
661 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
662 spin_lock(&tdtd->tdtd_batchid_lock);
663 new->tmt_batchid = tdtd->tdtd_batchid++;
664 list_add_tail(&new->tmt_commit_list, &tdtd->tdtd_list);
665 spin_unlock(&tdtd->tdtd_batchid_lock);
666 top_multiple_thandle_get(new);
667 top_multiple_thandle_dump(new, D_INFO);
671 * Insert distribute transaction to the distribute txn list.
673 * Insert distribute transaction to the distribute txn list.
675 * \param[in] new the distribute txn to be inserted.
677 void distribute_txn_insert_by_batchid(struct top_multiple_thandle *new)
679 struct dt_device *dt = new->tmt_master_sub_dt;
680 struct top_multiple_thandle *tmt;
681 struct target_distribute_txn_data *tdtd;
682 bool at_head = false;
685 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
687 spin_lock(&tdtd->tdtd_batchid_lock);
688 list_for_each_entry_reverse(tmt, &tdtd->tdtd_list, tmt_commit_list) {
689 if (new->tmt_batchid > tmt->tmt_batchid) {
690 list_add(&new->tmt_commit_list, &tmt->tmt_commit_list);
694 if (list_empty(&new->tmt_commit_list)) {
696 list_add(&new->tmt_commit_list, &tdtd->tdtd_list);
698 spin_unlock(&tdtd->tdtd_batchid_lock);
699 top_multiple_thandle_get(new);
700 top_multiple_thandle_dump(new, D_INFO);
701 if (new->tmt_committed && at_head)
702 wake_up(&tdtd->tdtd_commit_thread_waitq);
706 * Prepare cross-MDT operation.
708 * Create the update record buffer to record updates for cross-MDT operation,
709 * add master sub transaction to tt_sub_trans_list, and declare the update
712 * During updates packing, all of parameters will be packed in
713 * tur_update_params, and updates will be packed in tur_update_records.
714 * Then in transaction stop, parameters and updates will be merged
715 * into one updates buffer.
717 * And also master thandle will be added to the sub_th list, so it will be
718 * easy to track the commit status.
720 * \param[in] env execution environment
721 * \param[in] th top transaction handle
723 * \retval 0 if preparation succeeds.
724 * \retval negative errno if preparation fails.
726 static int prepare_multiple_node_trans(const struct lu_env *env,
727 struct top_multiple_thandle *tmt)
729 struct thandle_update_records *tur;
733 if (tmt->tmt_update_records == NULL) {
734 tur = &update_env_info(env)->uti_tur;
735 rc = check_and_prepare_update_record(env, tur);
739 tmt->tmt_update_records = tur;
740 distribute_txn_assign_batchid(tmt);
743 rc = declare_updates_write(env, tmt);
749 * start the top transaction.
751 * Start all of its sub transactions, then start master sub transaction.
753 * \param[in] env execution environment
754 * \param[in] master_dev master_dev the top thandle will be start
755 * \param[in] th top thandle
757 * \retval 0 if transaction start succeeds.
758 * \retval negative errno if start fails.
760 int top_trans_start(const struct lu_env *env, struct dt_device *master_dev,
763 struct top_thandle *top_th = container_of(th, struct top_thandle,
765 struct sub_thandle *st;
766 struct top_multiple_thandle *tmt = top_th->tt_multiple_thandle;
771 rc = dt_trans_start(env, top_th->tt_master_sub_thandle->th_dev,
772 top_th->tt_master_sub_thandle);
776 tmt = top_th->tt_multiple_thandle;
777 rc = prepare_multiple_node_trans(env, tmt);
781 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
782 if (st->st_sub_th == NULL)
784 st->st_sub_th->th_sync = th->th_sync;
785 st->st_sub_th->th_local = th->th_local;
786 st->st_sub_th->th_tags = th->th_tags;
787 rc = dt_trans_start(env, st->st_sub_th->th_dev,
792 sub_thandle_register_stop_cb(st, tmt);
793 sub_thandle_register_commit_cb(st, tmt);
799 EXPORT_SYMBOL(top_trans_start);
802 * Check whether we need write updates record
804 * Check if the updates for the top_thandle needs to be writen
805 * to all targets. Only if the transaction succeeds and the updates
806 * number > 2, it will write the updates,
808 * \params [in] top_th top thandle.
810 * \retval true if it needs to write updates
811 * \retval false if it does not need to write updates
813 static bool top_check_write_updates(struct top_thandle *top_th)
815 struct top_multiple_thandle *tmt;
816 struct thandle_update_records *tur;
818 /* Do not write updates to records if the transaction fails */
819 if (top_th->tt_super.th_result != 0)
822 tmt = top_th->tt_multiple_thandle;
826 tur = tmt->tmt_update_records;
830 /* Hmm, false update records, since the cross-MDT operation
831 * should includes both local and remote updates, so the
832 * updates count should >= 2 */
833 if (tur->tur_update_records == NULL ||
834 tur->tur_update_records->lur_update_rec.ur_update_count <= 1)
841 * Check if top transaction is stopped
843 * Check if top transaction is stopped, only if all sub transaction
844 * is stopped, then the top transaction is stopped.
846 * \param [in] top_th top thandle
848 * \retval true if the top transaction is stopped.
849 * \retval false if the top transaction is not stopped.
851 static bool top_trans_is_stopped(struct top_thandle *top_th)
853 struct top_multiple_thandle *tmt;
854 struct sub_thandle *st;
855 bool all_stopped = true;
857 tmt = top_th->tt_multiple_thandle;
858 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
859 if (!st->st_stopped && st->st_sub_th != NULL) {
864 if (st->st_result != 0 &&
865 top_th->tt_super.th_result == 0)
866 top_th->tt_super.th_result = st->st_result;
873 * Wait result of top transaction
875 * Wait until all sub transaction get its result.
877 * \param [in] top_th top thandle.
879 * \retval the result of top thandle.
881 static int top_trans_wait_result(struct top_thandle *top_th)
883 struct l_wait_info lwi = {0};
885 l_wait_event(top_th->tt_multiple_thandle->tmt_stop_waitq,
886 top_trans_is_stopped(top_th), &lwi);
888 RETURN(top_th->tt_super.th_result);
892 * Stop the top transaction.
894 * Stop the transaction on the master device first, then stop transactions
895 * on other sub devices.
897 * \param[in] env execution environment
898 * \param[in] master_dev master_dev the top thandle will be created
899 * \param[in] th top thandle
901 * \retval 0 if stop transaction succeeds.
902 * \retval negative errno if stop transaction fails.
904 int top_trans_stop(const struct lu_env *env, struct dt_device *master_dev,
907 struct top_thandle *top_th = container_of(th, struct top_thandle,
909 struct sub_thandle *st;
910 struct sub_thandle *master_st;
911 struct top_multiple_thandle *tmt;
912 struct thandle_update_records *tur;
913 bool write_updates = false;
917 if (likely(top_th->tt_multiple_thandle == NULL)) {
918 LASSERT(master_dev != NULL);
919 rc = dt_trans_stop(env, master_dev,
920 top_th->tt_master_sub_thandle);
921 OBD_FREE_PTR(top_th);
925 tmt = top_th->tt_multiple_thandle;
926 tur = tmt->tmt_update_records;
928 /* Note: we need stop the master thandle first, then the stop
929 * callback will fill the master transno in the update logs,
930 * then these update logs will be sent to other MDTs */
931 /* get the master sub thandle */
932 master_st = lookup_sub_thandle(tmt, tmt->tmt_master_sub_dt);
933 write_updates = top_check_write_updates(top_th);
935 /* Step 1: write the updates log on Master MDT */
936 if (master_st != NULL && master_st->st_sub_th != NULL &&
938 struct llog_update_record *lur;
940 /* Merge the parameters and updates into one buffer */
941 rc = prepare_writing_updates(env, tmt);
943 CERROR("%s: cannot prepare updates: rc = %d\n",
944 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
946 GOTO(stop_master_trans, rc);
949 lur = tur->tur_update_records;
950 /* Write updates to the master MDT */
951 rc = sub_updates_write(env, lur, master_st);
953 /* Cleanup the common parameters in the update records,
954 * master transno callback might add more parameters.
955 * and we need merge the update records again in the
957 if (tur->tur_update_params != NULL)
958 lur->lur_update_rec.ur_param_count = 0;
961 CERROR("%s: write updates failed: rc = %d\n",
962 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
964 GOTO(stop_master_trans, rc);
969 /* Step 2: Stop the transaction on the master MDT, and fill the
970 * master transno in the update logs to other MDT. */
971 if (master_st != NULL && master_st->st_sub_th != NULL) {
972 master_st->st_sub_th->th_local = th->th_local;
973 master_st->st_sub_th->th_sync = th->th_sync;
974 master_st->st_sub_th->th_tags = th->th_tags;
975 master_st->st_sub_th->th_result = th->th_result;
976 rc = dt_trans_stop(env, master_st->st_dt, master_st->st_sub_th);
979 GOTO(stop_other_trans, rc);
980 } else if (tur != NULL && tur->tur_update_records != NULL) {
981 struct llog_update_record *lur;
983 lur = tur->tur_update_records;
984 if (lur->lur_update_rec.ur_master_transno == 0)
985 /* Update master transno after master stop
987 lur->lur_update_rec.ur_master_transno =
988 tgt_th_info(env)->tti_transno;
992 /* Step 3: write updates to other MDTs */
994 struct llog_update_record *lur;
996 /* Stop callback of master will add more updates and also update
997 * master transno, so merge the parameters and updates into one
999 rc = prepare_writing_updates(env, tmt);
1001 CERROR("%s: prepare updates failed: rc = %d\n",
1002 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
1004 GOTO(stop_other_trans, rc);
1006 lur = tur->tur_update_records;
1007 list_for_each_entry(st, &tmt->tmt_sub_thandle_list,
1009 if (st->st_sub_th == NULL || st == master_st ||
1010 st->st_sub_th->th_result < 0)
1013 rc = sub_updates_write(env, lur, st);
1022 /* Step 4: Stop the transaction on other MDTs */
1023 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1024 if (st == master_st || st->st_sub_th == NULL)
1027 st->st_sub_th->th_sync = th->th_sync;
1028 st->st_sub_th->th_local = th->th_local;
1029 st->st_sub_th->th_tags = th->th_tags;
1030 st->st_sub_th->th_result = th->th_result;
1031 rc = dt_trans_stop(env, st->st_sub_th->th_dev,
1033 if (unlikely(rc < 0 && th->th_result == 0))
1037 rc = top_trans_wait_result(top_th);
1039 tmt->tmt_result = rc;
1041 /* Balance for the refcount in top_trans_create, Note: if it is NOT
1042 * multiple node transaction, the top transaction will be destroyed. */
1043 top_multiple_thandle_put(tmt);
1044 OBD_FREE_PTR(top_th);
1047 EXPORT_SYMBOL(top_trans_stop);
1050 * Create top_multiple_thandle for top_thandle
1052 * Create top_mutilple_thandle to manage the mutiple node transaction
1053 * for top_thandle, and it also needs to add master sub thandle to the
1054 * sub trans list now.
1056 * \param[in] env execution environment
1057 * \param[in] top_th the top thandle
1059 * \retval 0 if creation succeeds
1060 * \retval negative errno if creation fails
1062 int top_trans_create_tmt(const struct lu_env *env,
1063 struct top_thandle *top_th)
1065 struct top_multiple_thandle *tmt;
1071 tmt->tmt_magic = TOP_THANDLE_MAGIC;
1072 INIT_LIST_HEAD(&tmt->tmt_sub_thandle_list);
1073 INIT_LIST_HEAD(&tmt->tmt_commit_list);
1074 atomic_set(&tmt->tmt_refcount, 1);
1076 init_waitqueue_head(&tmt->tmt_stop_waitq);
1077 top_th->tt_multiple_thandle = tmt;
1082 static struct sub_thandle *
1083 create_sub_thandle_with_thandle(struct top_thandle *top_th,
1084 struct thandle *sub_th)
1086 struct sub_thandle *st;
1088 /* create and init sub th to the top trans list */
1089 st = create_sub_thandle(top_th->tt_multiple_thandle,
1094 st->st_sub_th = sub_th;
1096 sub_th->th_top = &top_th->tt_super;
1103 * Get sub thandle from the top thandle according to the sub dt_device.
1105 * \param[in] env execution environment
1106 * \param[in] th thandle on the top layer.
1107 * \param[in] sub_dt sub dt_device used to get sub transaction
1109 * \retval thandle of sub transaction if succeed
1110 * \retval PTR_ERR(errno) if failed
1112 struct thandle *thandle_get_sub_by_dt(const struct lu_env *env,
1114 struct dt_device *sub_dt)
1116 struct sub_thandle *st = NULL;
1117 struct top_thandle *top_th;
1118 struct thandle *sub_th = NULL;
1122 top_th = container_of(th, struct top_thandle, tt_super);
1124 if (likely(sub_dt == top_th->tt_master_sub_thandle->th_dev))
1125 RETURN(top_th->tt_master_sub_thandle);
1127 if (top_th->tt_multiple_thandle != NULL) {
1128 st = lookup_sub_thandle(top_th->tt_multiple_thandle, sub_dt);
1130 RETURN(st->st_sub_th);
1133 sub_th = dt_trans_create(env, sub_dt);
1137 /* Create top_multiple_thandle if necessary */
1138 if (top_th->tt_multiple_thandle == NULL) {
1139 struct top_multiple_thandle *tmt;
1141 rc = top_trans_create_tmt(env, top_th);
1143 GOTO(stop_trans, rc);
1145 tmt = top_th->tt_multiple_thandle;
1147 /* Add master sub th to the top trans list */
1148 tmt->tmt_master_sub_dt =
1149 top_th->tt_master_sub_thandle->th_dev;
1150 st = create_sub_thandle_with_thandle(top_th,
1151 top_th->tt_master_sub_thandle);
1153 GOTO(stop_trans, rc = PTR_ERR(st));
1156 /* create and init sub th to the top trans list */
1157 st = create_sub_thandle_with_thandle(top_th, sub_th);
1158 st->st_sub_th->th_wait_submit = 1;
1163 sub_th->th_result = rc;
1164 dt_trans_stop(env, sub_dt, sub_th);
1165 sub_th = ERR_PTR(rc);
1170 EXPORT_SYMBOL(thandle_get_sub_by_dt);
1173 * Top multiple thandle destroy
1175 * Destroy multiple thandle and all its sub thandle.
1177 * \param[in] tmt top_multiple_thandle to be destroyed.
1179 void top_multiple_thandle_destroy(struct top_multiple_thandle *tmt)
1181 struct sub_thandle *st;
1182 struct sub_thandle *tmp;
1184 LASSERT(tmt->tmt_magic == TOP_THANDLE_MAGIC);
1185 list_for_each_entry_safe(st, tmp, &tmt->tmt_sub_thandle_list,
1187 struct sub_thandle_cookie *stc;
1188 struct sub_thandle_cookie *tmp;
1190 list_del(&st->st_sub_list);
1191 list_for_each_entry_safe(stc, tmp, &st->st_cookie_list,
1193 list_del(&stc->stc_list);
1200 EXPORT_SYMBOL(top_multiple_thandle_destroy);
1203 * Cancel the update log on MDTs
1205 * Cancel the update log on MDTs then destroy the thandle.
1207 * \param[in] env execution environment
1208 * \param[in] tmt the top multiple thandle whose updates records
1209 * will be cancelled.
1211 * \retval 0 if cancellation succeeds.
1212 * \retval negative errno if cancellation fails.
1214 static int distribute_txn_cancel_records(const struct lu_env *env,
1215 struct top_multiple_thandle *tmt)
1217 struct sub_thandle *st;
1220 top_multiple_thandle_dump(tmt, D_INFO);
1221 /* Cancel update logs on other MDTs */
1222 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1223 struct llog_ctxt *ctxt;
1224 struct obd_device *obd;
1225 struct llog_cookie *cookie;
1226 struct sub_thandle_cookie *stc;
1229 obd = st->st_dt->dd_lu_dev.ld_obd;
1230 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
1232 list_for_each_entry(stc, &st->st_cookie_list, stc_list) {
1233 cookie = &stc->stc_cookie;
1234 if (fid_is_zero(&cookie->lgc_lgl.lgl_oi.oi_fid))
1237 rc = llog_cat_cancel_records(env, ctxt->loc_handle, 1,
1239 CDEBUG(D_HA, "%s: batchid %llu cancel update log "
1240 DOSTID ".%u : rc = %d\n", obd->obd_name,
1242 POSTID(&cookie->lgc_lgl.lgl_oi),
1243 cookie->lgc_index, rc);
1246 llog_ctxt_put(ctxt);
1253 * Check if there are committed transaction
1255 * Check if there are committed transaction in the distribute transaction
1256 * list, then cancel the update records for those committed transaction.
1257 * Because the distribute transaction in the list are sorted by batchid,
1258 * and cancellation will be done by batchid order, so we only check the first
1259 * the transaction(with lowest batchid) in the list.
1261 * \param[in] lod lod device where cancel thread is
1263 * \retval true if it is ready
1264 * \retval false if it is not ready
1266 static bool tdtd_ready_for_cancel_log(struct target_distribute_txn_data *tdtd)
1268 struct top_multiple_thandle *tmt = NULL;
1269 struct obd_device *obd = tdtd->tdtd_lut->lut_obd;
1272 spin_lock(&tdtd->tdtd_batchid_lock);
1273 if (!list_empty(&tdtd->tdtd_list)) {
1274 tmt = list_entry(tdtd->tdtd_list.next,
1275 struct top_multiple_thandle, tmt_commit_list);
1276 if (tmt->tmt_committed &&
1277 (!obd->obd_recovering || (obd->obd_recovering &&
1278 tmt->tmt_batchid <= tdtd->tdtd_committed_batchid)))
1281 spin_unlock(&tdtd->tdtd_batchid_lock);
1286 struct distribute_txn_bid_data {
1287 struct dt_txn_commit_cb dtbd_cb;
1288 struct target_distribute_txn_data *dtbd_tdtd;
1293 * callback of updating commit batchid
1295 * Updating commit batchid then wake up the commit thread to cancel the
1298 * \param[in]env execution environment
1299 * \param[in]th thandle to updating commit batchid
1300 * \param[in]cb commit callback
1301 * \param[in]err result of thandle
1303 static void distribute_txn_batchid_cb(struct lu_env *env,
1305 struct dt_txn_commit_cb *cb,
1308 struct distribute_txn_bid_data *dtbd = NULL;
1309 struct target_distribute_txn_data *tdtd;
1311 dtbd = container_of0(cb, struct distribute_txn_bid_data, dtbd_cb);
1312 tdtd = dtbd->dtbd_tdtd;
1314 CDEBUG(D_HA, "%s: %llu batchid updated\n",
1315 tdtd->tdtd_lut->lut_obd->obd_name, dtbd->dtbd_batchid);
1316 spin_lock(&tdtd->tdtd_batchid_lock);
1317 if (dtbd->dtbd_batchid > tdtd->tdtd_committed_batchid &&
1318 !tdtd->tdtd_lut->lut_obd->obd_no_transno)
1319 tdtd->tdtd_committed_batchid = dtbd->dtbd_batchid;
1320 spin_unlock(&tdtd->tdtd_batchid_lock);
1321 atomic_dec(&tdtd->tdtd_refcount);
1322 wake_up(&tdtd->tdtd_commit_thread_waitq);
1328 * Update the commit batchid in disk
1330 * Update commit batchid in the disk, after this is committed, it can start
1331 * to cancel the update records.
1333 * \param[in] env execution environment
1334 * \param[in] tdtd distribute transaction structure
1335 * \param[in] batchid commit batchid to be updated
1337 * \retval 0 if update succeeds.
1338 * \retval negative errno if update fails.
1341 distribute_txn_commit_batchid_update(const struct lu_env *env,
1342 struct target_distribute_txn_data *tdtd,
1345 struct distribute_txn_bid_data *dtbd = NULL;
1353 OBD_ALLOC_PTR(dtbd);
1356 dtbd->dtbd_batchid = batchid;
1357 dtbd->dtbd_tdtd = tdtd;
1358 dtbd->dtbd_cb.dcb_func = distribute_txn_batchid_cb;
1359 atomic_inc(&tdtd->tdtd_refcount);
1361 th = dt_trans_create(env, tdtd->tdtd_lut->lut_bottom);
1364 RETURN(PTR_ERR(th));
1367 tmp = cpu_to_le64(batchid);
1369 buf.lb_len = sizeof(tmp);
1372 rc = dt_declare_record_write(env, tdtd->tdtd_batchid_obj, &buf, off,
1377 rc = dt_trans_start_local(env, tdtd->tdtd_lut->lut_bottom, th);
1381 rc = dt_trans_cb_add(th, &dtbd->dtbd_cb);
1385 rc = dt_record_write(env, tdtd->tdtd_batchid_obj, &buf,
1388 CDEBUG(D_INFO, "%s: update batchid "LPU64": rc = %d\n",
1389 tdtd->tdtd_lut->lut_obd->obd_name, batchid, rc);
1392 dt_trans_stop(env, tdtd->tdtd_lut->lut_bottom, th);
1399 * Init commit batchid for distribute transaction.
1401 * Initialize the batchid object and get commit batchid from the object.
1403 * \param[in] env execution environment
1404 * \param[in] tdtd distribute transaction whose batchid is initialized.
1406 * \retval 0 if initialization succeeds.
1407 * \retval negative errno if initialization fails.
1410 distribute_txn_commit_batchid_init(const struct lu_env *env,
1411 struct target_distribute_txn_data *tdtd)
1413 struct tgt_thread_info *tti = tgt_th_info(env);
1414 struct lu_target *lut = tdtd->tdtd_lut;
1415 struct lu_attr *attr = &tti->tti_attr;
1416 struct lu_fid *fid = &tti->tti_fid1;
1417 struct dt_object_format *dof = &tti->tti_u.update.tti_update_dof;
1418 struct dt_object *dt_obj = NULL;
1425 memset(attr, 0, sizeof(*attr));
1426 attr->la_valid = LA_MODE;
1427 attr->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1428 dof->dof_type = dt_mode_to_dft(S_IFREG);
1430 lu_local_obj_fid(fid, BATCHID_COMMITTED_OID);
1432 dt_obj = dt_find_or_create(env, lut->lut_bottom, fid, dof,
1434 if (IS_ERR(dt_obj)) {
1435 rc = PTR_ERR(dt_obj);
1440 tdtd->tdtd_batchid_obj = dt_obj;
1443 buf.lb_len = sizeof(tmp);
1445 rc = dt_read(env, dt_obj, &buf, &off);
1446 if (rc < 0 || (rc < buf.lb_len && rc > 0)) {
1447 CERROR("%s can't read last committed batchid: rc = %d\n",
1448 tdtd->tdtd_lut->lut_obd->obd_name, rc);
1452 } else if (rc == buf.lb_len) {
1453 tdtd->tdtd_committed_batchid = le64_to_cpu(tmp);
1454 CDEBUG(D_HA, "%s: committed batchid %llu\n",
1455 tdtd->tdtd_lut->lut_obd->obd_name,
1456 tdtd->tdtd_committed_batchid);
1461 if (rc < 0 && dt_obj != NULL) {
1462 lu_object_put(env, &dt_obj->do_lu);
1463 tdtd->tdtd_batchid_obj = NULL;
1469 * manage the distribute transaction thread
1471 * Distribute transaction are linked to the list, and once the distribute
1472 * transaction is committed, it will update the last committed batchid first,
1473 * after it is committed, it will cancel the records.
1475 * \param[in] _arg argument for commit thread
1477 * \retval 0 if thread is running successfully
1478 * \retval negative errno if the thread can not be run.
1480 static int distribute_txn_commit_thread(void *_arg)
1482 struct target_distribute_txn_data *tdtd = _arg;
1483 struct lu_target *lut = tdtd->tdtd_lut;
1484 struct ptlrpc_thread *thread = &lut->lut_tdtd_commit_thread;
1485 struct l_wait_info lwi = { 0 };
1487 struct list_head list;
1489 struct top_multiple_thandle *tmt;
1490 struct top_multiple_thandle *tmp;
1491 __u64 batchid = 0, committed;
1495 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
1499 spin_lock(&tdtd->tdtd_batchid_lock);
1500 thread->t_flags = SVC_RUNNING;
1501 spin_unlock(&tdtd->tdtd_batchid_lock);
1502 wake_up(&thread->t_ctl_waitq);
1503 INIT_LIST_HEAD(&list);
1505 CDEBUG(D_HA, "%s: start commit thread committed batchid "LPU64"\n",
1506 tdtd->tdtd_lut->lut_obd->obd_name,
1507 tdtd->tdtd_committed_batchid);
1509 while (distribute_txn_commit_thread_running(lut)) {
1510 spin_lock(&tdtd->tdtd_batchid_lock);
1511 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1513 if (tmt->tmt_committed == 0)
1516 /* Note: right now, replay is based on master MDT
1517 * transno, but cancellation is based on batchid.
1518 * so we do not try to cancel the update log until
1519 * the recoverying is done, unless the update records
1520 * batchid < committed_batchid. */
1521 if (tmt->tmt_batchid <= tdtd->tdtd_committed_batchid) {
1522 list_move_tail(&tmt->tmt_commit_list, &list);
1523 } else if (!tdtd->tdtd_lut->lut_obd->obd_recovering) {
1524 LASSERTF(tmt->tmt_batchid >= batchid,
1525 "tmt %p tmt_batchid: "LPU64", batchid "
1526 LPU64"\n", tmt, tmt->tmt_batchid,
1528 /* There are three types of distribution
1529 * transaction result
1531 * 1. If tmt_result < 0, it means the
1532 * distribution transaction fails, which should
1533 * be rare, because once declare phase succeeds,
1534 * the operation should succeeds anyway. Note in
1535 * this case, we will still update batchid so
1536 * cancellation would be stopped.
1538 * 2. If tmt_result == 0, it means the
1539 * distribution transaction succeeds, and we
1540 * will update batchid.
1542 * 3. If tmt_result > 0, it means distribute
1543 * transaction is not yet committed on every
1544 * node, but we need release this tmt before
1545 * that, which usuually happens during umount.
1547 if (tmt->tmt_result <= 0)
1548 batchid = tmt->tmt_batchid;
1549 list_move_tail(&tmt->tmt_commit_list, &list);
1552 spin_unlock(&tdtd->tdtd_batchid_lock);
1554 CDEBUG(D_HA, "%s: batchid: "LPU64" committed batchid "
1555 LPU64"\n", tdtd->tdtd_lut->lut_obd->obd_name, batchid,
1556 tdtd->tdtd_committed_batchid);
1557 /* update globally committed on a storage */
1558 if (batchid > tdtd->tdtd_committed_batchid) {
1559 distribute_txn_commit_batchid_update(&env, tdtd,
1561 spin_lock(&tdtd->tdtd_batchid_lock);
1562 if (batchid > tdtd->tdtd_batchid) {
1563 /* This might happen during recovery,
1564 * batchid is initialized as last transno,
1565 * and the batchid in the update records
1566 * on other MDTs might be bigger than
1567 * the batchid, so we need update it to
1568 * avoid duplicate batchid. */
1569 CDEBUG(D_HA, "%s update batchid from "LPU64
1571 tdtd->tdtd_lut->lut_obd->obd_name,
1572 tdtd->tdtd_batchid, batchid);
1573 tdtd->tdtd_batchid = batchid;
1575 spin_unlock(&tdtd->tdtd_batchid_lock);
1577 /* cancel the records for committed batchid's */
1578 /* XXX: should we postpone cancel's till the end of recovery? */
1579 committed = tdtd->tdtd_committed_batchid;
1580 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1581 if (tmt->tmt_batchid > committed)
1583 list_del_init(&tmt->tmt_commit_list);
1584 if (tmt->tmt_result <= 0)
1585 distribute_txn_cancel_records(&env, tmt);
1586 top_multiple_thandle_put(tmt);
1589 l_wait_event(tdtd->tdtd_commit_thread_waitq,
1590 !distribute_txn_commit_thread_running(lut) ||
1591 committed < tdtd->tdtd_committed_batchid ||
1592 tdtd_ready_for_cancel_log(tdtd), &lwi);
1595 l_wait_event(tdtd->tdtd_commit_thread_waitq,
1596 atomic_read(&tdtd->tdtd_refcount) == 0, &lwi);
1598 spin_lock(&tdtd->tdtd_batchid_lock);
1599 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1601 list_move_tail(&tmt->tmt_commit_list, &list);
1602 spin_unlock(&tdtd->tdtd_batchid_lock);
1604 CDEBUG(D_INFO, "%s stopping distribute txn commit thread.\n",
1605 tdtd->tdtd_lut->lut_obd->obd_name);
1606 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1607 list_del_init(&tmt->tmt_commit_list);
1608 top_multiple_thandle_dump(tmt, D_HA);
1609 top_multiple_thandle_put(tmt);
1612 thread->t_flags = SVC_STOPPED;
1614 wake_up(&thread->t_ctl_waitq);
1620 * Start llog cancel thread
1622 * Start llog cancel(master/slave) thread on LOD
1624 * \param[in]lclt cancel log thread to be started.
1626 * \retval 0 if the thread is started successfully.
1627 * \retval negative errno if the thread is not being
1630 int distribute_txn_init(const struct lu_env *env,
1631 struct lu_target *lut,
1632 struct target_distribute_txn_data *tdtd,
1635 struct task_struct *task;
1636 struct l_wait_info lwi = { 0 };
1640 spin_lock_init(&tdtd->tdtd_batchid_lock);
1641 INIT_LIST_HEAD(&tdtd->tdtd_list);
1643 tdtd->tdtd_batchid = lut->lut_last_transno + 1;
1645 init_waitqueue_head(&lut->lut_tdtd_commit_thread.t_ctl_waitq);
1646 init_waitqueue_head(&tdtd->tdtd_commit_thread_waitq);
1647 atomic_set(&tdtd->tdtd_refcount, 0);
1649 tdtd->tdtd_lut = lut;
1650 rc = distribute_txn_commit_batchid_init(env, tdtd);
1654 task = kthread_run(distribute_txn_commit_thread, tdtd, "tdtd-%u",
1657 RETURN(PTR_ERR(task));
1659 l_wait_event(lut->lut_tdtd_commit_thread.t_ctl_waitq,
1660 distribute_txn_commit_thread_running(lut) ||
1661 distribute_txn_commit_thread_stopped(lut), &lwi);
1664 EXPORT_SYMBOL(distribute_txn_init);
1667 * Stop llog cancel thread
1669 * Stop llog cancel(master/slave) thread on LOD and also destory
1670 * all of transaction in the list.
1672 * \param[in]lclt cancel log thread to be stopped.
1674 void distribute_txn_fini(const struct lu_env *env,
1675 struct target_distribute_txn_data *tdtd)
1677 struct lu_target *lut = tdtd->tdtd_lut;
1679 /* Stop cancel thread */
1680 if (lut == NULL || !distribute_txn_commit_thread_running(lut))
1683 spin_lock(&tdtd->tdtd_batchid_lock);
1684 lut->lut_tdtd_commit_thread.t_flags = SVC_STOPPING;
1685 spin_unlock(&tdtd->tdtd_batchid_lock);
1686 wake_up(&tdtd->tdtd_commit_thread_waitq);
1687 wait_event(lut->lut_tdtd_commit_thread.t_ctl_waitq,
1688 lut->lut_tdtd_commit_thread.t_flags & SVC_STOPPED);
1690 dtrq_list_destroy(tdtd);
1691 if (tdtd->tdtd_batchid_obj != NULL) {
1692 lu_object_put(env, &tdtd->tdtd_batchid_obj->do_lu);
1693 tdtd->tdtd_batchid_obj = NULL;
1696 EXPORT_SYMBOL(distribute_txn_fini);