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 spin_lock(&tmt->tmt_sub_lock);
472 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
473 if (st->st_sub_th == sub_th) {
474 st->st_committed = 1;
477 if (!st->st_committed)
478 all_committed = false;
480 spin_unlock(&tmt->tmt_sub_lock);
482 if (tmt->tmt_result == 0)
483 tmt->tmt_result = err;
486 top_trans_committed_cb(tmt);
488 top_multiple_thandle_dump(tmt, D_INFO);
489 top_multiple_thandle_put(tmt);
493 static void sub_thandle_register_commit_cb(struct sub_thandle *st,
494 struct top_multiple_thandle *tmt)
496 LASSERT(st->st_sub_th != NULL);
497 top_multiple_thandle_get(tmt);
498 st->st_commit_dcb.dcb_func = sub_trans_commit_cb;
499 st->st_commit_dcb.dcb_data = tmt;
500 INIT_LIST_HEAD(&st->st_commit_dcb.dcb_linkage);
501 dt_trans_cb_add(st->st_sub_th, &st->st_commit_dcb);
505 * Sub thandle stop call back
507 * After sub thandle is stopped, it will call this callback to notify
510 * \param[in] th sub thandle to be stopped
511 * \param[in] rc result of sub trans
513 static void sub_trans_stop_cb(struct lu_env *env,
514 struct thandle *sub_th,
515 struct dt_txn_commit_cb *cb, int err)
517 struct sub_thandle *st;
518 struct top_multiple_thandle *tmt = cb->dcb_data;
521 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
525 if (st->st_dt == sub_th->th_dev) {
532 wake_up(&tmt->tmt_stop_waitq);
536 static void sub_thandle_register_stop_cb(struct sub_thandle *st,
537 struct top_multiple_thandle *tmt)
539 st->st_stop_dcb.dcb_func = sub_trans_stop_cb;
540 st->st_stop_dcb.dcb_data = tmt;
541 st->st_stop_dcb.dcb_flags = DCB_TRANS_STOP;
542 INIT_LIST_HEAD(&st->st_stop_dcb.dcb_linkage);
543 dt_trans_cb_add(st->st_sub_th, &st->st_stop_dcb);
549 * Create transaction handle for sub_thandle
551 * \param[in] env execution environment
552 * \param[in] th top thandle
553 * \param[in] st sub_thandle
555 * \retval 0 if creation succeeds.
556 * \retval negative errno if creation fails.
558 int sub_thandle_trans_create(const struct lu_env *env,
559 struct top_thandle *top_th,
560 struct sub_thandle *st)
562 struct thandle *sub_th;
564 sub_th = dt_trans_create(env, st->st_dt);
566 return PTR_ERR(sub_th);
568 sub_th->th_top = &top_th->tt_super;
569 st->st_sub_th = sub_th;
571 sub_th->th_wait_submit = 1;
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_IO);
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 top_th->tt_super.th_tags |= child_th->th_tags;
612 return &top_th->tt_super;
614 EXPORT_SYMBOL(top_trans_create);
617 * Declare write update transaction
619 * Check if there are updates being recorded in this transaction,
620 * it will write the record into the disk.
622 * \param[in] env execution environment
623 * \param[in] tmt top multiple transaction handle
625 * \retval 0 if writing succeeds
626 * \retval negative errno if writing fails
628 static int declare_updates_write(const struct lu_env *env,
629 struct top_multiple_thandle *tmt)
631 struct llog_update_record *record;
632 struct sub_thandle *st;
635 record = tmt->tmt_update_records->tur_update_records;
636 /* Declare update write for all other target */
637 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
638 if (st->st_sub_th == NULL)
641 rc = sub_declare_updates_write(env, record, st->st_sub_th,
642 tmt->tmt_record_size);
651 * Assign batchid to the distribute transaction.
653 * Assign batchid to the distribute transaction
655 * \param[in] tmt distribute transaction
657 static void distribute_txn_assign_batchid(struct top_multiple_thandle *new)
659 struct target_distribute_txn_data *tdtd;
660 struct dt_device *dt = new->tmt_master_sub_dt;
663 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
664 spin_lock(&tdtd->tdtd_batchid_lock);
665 new->tmt_batchid = tdtd->tdtd_batchid++;
666 list_add_tail(&new->tmt_commit_list, &tdtd->tdtd_list);
667 spin_unlock(&tdtd->tdtd_batchid_lock);
668 top_multiple_thandle_get(new);
669 top_multiple_thandle_dump(new, D_INFO);
673 * Insert distribute transaction to the distribute txn list.
675 * Insert distribute transaction to the distribute txn list.
677 * \param[in] new the distribute txn to be inserted.
679 void distribute_txn_insert_by_batchid(struct top_multiple_thandle *new)
681 struct dt_device *dt = new->tmt_master_sub_dt;
682 struct top_multiple_thandle *tmt;
683 struct target_distribute_txn_data *tdtd;
684 bool at_head = false;
687 tdtd = dt2lu_dev(dt)->ld_site->ls_tgt->lut_tdtd;
689 spin_lock(&tdtd->tdtd_batchid_lock);
690 list_for_each_entry_reverse(tmt, &tdtd->tdtd_list, tmt_commit_list) {
691 if (new->tmt_batchid > tmt->tmt_batchid) {
692 list_add(&new->tmt_commit_list, &tmt->tmt_commit_list);
696 if (list_empty(&new->tmt_commit_list)) {
698 list_add(&new->tmt_commit_list, &tdtd->tdtd_list);
700 spin_unlock(&tdtd->tdtd_batchid_lock);
701 top_multiple_thandle_get(new);
702 top_multiple_thandle_dump(new, D_INFO);
703 if (new->tmt_committed && at_head)
704 wake_up(&tdtd->tdtd_commit_thread_waitq);
708 * Prepare cross-MDT operation.
710 * Create the update record buffer to record updates for cross-MDT operation,
711 * add master sub transaction to tt_sub_trans_list, and declare the update
714 * During updates packing, all of parameters will be packed in
715 * tur_update_params, and updates will be packed in tur_update_records.
716 * Then in transaction stop, parameters and updates will be merged
717 * into one updates buffer.
719 * And also master thandle will be added to the sub_th list, so it will be
720 * easy to track the commit status.
722 * \param[in] env execution environment
723 * \param[in] th top transaction handle
725 * \retval 0 if preparation succeeds.
726 * \retval negative errno if preparation fails.
728 static int prepare_multiple_node_trans(const struct lu_env *env,
729 struct top_multiple_thandle *tmt)
731 struct thandle_update_records *tur;
735 if (tmt->tmt_update_records == NULL) {
736 tur = &update_env_info(env)->uti_tur;
737 rc = check_and_prepare_update_record(env, tur);
741 tmt->tmt_update_records = tur;
742 distribute_txn_assign_batchid(tmt);
745 rc = declare_updates_write(env, tmt);
751 * start the top transaction.
753 * Start all of its sub transactions, then start master sub transaction.
755 * \param[in] env execution environment
756 * \param[in] master_dev master_dev the top thandle will be start
757 * \param[in] th top thandle
759 * \retval 0 if transaction start succeeds.
760 * \retval negative errno if start fails.
762 int top_trans_start(const struct lu_env *env, struct dt_device *master_dev,
765 struct top_thandle *top_th = container_of(th, struct top_thandle,
767 struct sub_thandle *st;
768 struct top_multiple_thandle *tmt = top_th->tt_multiple_thandle;
773 rc = dt_trans_start(env, top_th->tt_master_sub_thandle->th_dev,
774 top_th->tt_master_sub_thandle);
778 tmt = top_th->tt_multiple_thandle;
779 rc = prepare_multiple_node_trans(env, tmt);
783 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
784 if (st->st_sub_th == NULL)
787 st->st_sub_th->th_sync = th->th_sync;
788 st->st_sub_th->th_local = th->th_local;
789 st->st_sub_th->th_tags = th->th_tags;
790 rc = dt_trans_start(env, st->st_sub_th->th_dev,
795 sub_thandle_register_stop_cb(st, tmt);
796 sub_thandle_register_commit_cb(st, tmt);
802 EXPORT_SYMBOL(top_trans_start);
805 * Check whether we need write updates record
807 * Check if the updates for the top_thandle needs to be writen
808 * to all targets. Only if the transaction succeeds and the updates
809 * number > 2, it will write the updates,
811 * \params [in] top_th top thandle.
813 * \retval true if it needs to write updates
814 * \retval false if it does not need to write updates
816 static bool top_check_write_updates(struct top_thandle *top_th)
818 struct top_multiple_thandle *tmt;
819 struct thandle_update_records *tur;
821 /* Do not write updates to records if the transaction fails */
822 if (top_th->tt_super.th_result != 0)
825 tmt = top_th->tt_multiple_thandle;
829 tur = tmt->tmt_update_records;
833 /* Hmm, false update records, since the cross-MDT operation
834 * should includes both local and remote updates, so the
835 * updates count should >= 2 */
836 if (tur->tur_update_records == NULL ||
837 tur->tur_update_records->lur_update_rec.ur_update_count <= 1)
844 * Check if top transaction is stopped
846 * Check if top transaction is stopped, only if all sub transaction
847 * is stopped, then the top transaction is stopped.
849 * \param [in] top_th top thandle
851 * \retval true if the top transaction is stopped.
852 * \retval false if the top transaction is not stopped.
854 static bool top_trans_is_stopped(struct top_thandle *top_th)
856 struct top_multiple_thandle *tmt;
857 struct sub_thandle *st;
858 bool all_stopped = true;
860 tmt = top_th->tt_multiple_thandle;
861 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
862 if (!st->st_stopped && st->st_sub_th != NULL) {
867 if (st->st_result != 0 &&
868 top_th->tt_super.th_result == 0)
869 top_th->tt_super.th_result = st->st_result;
876 * Wait result of top transaction
878 * Wait until all sub transaction get its result.
880 * \param [in] top_th top thandle.
882 * \retval the result of top thandle.
884 static int top_trans_wait_result(struct top_thandle *top_th)
886 struct l_wait_info lwi = {0};
888 l_wait_event(top_th->tt_multiple_thandle->tmt_stop_waitq,
889 top_trans_is_stopped(top_th), &lwi);
891 RETURN(top_th->tt_super.th_result);
895 * Stop the top transaction.
897 * Stop the transaction on the master device first, then stop transactions
898 * on other sub devices.
900 * \param[in] env execution environment
901 * \param[in] master_dev master_dev the top thandle will be created
902 * \param[in] th top thandle
904 * \retval 0 if stop transaction succeeds.
905 * \retval negative errno if stop transaction fails.
907 int top_trans_stop(const struct lu_env *env, struct dt_device *master_dev,
910 struct top_thandle *top_th = container_of(th, struct top_thandle,
912 struct sub_thandle *st;
913 struct sub_thandle *master_st;
914 struct top_multiple_thandle *tmt;
915 struct thandle_update_records *tur;
916 bool write_updates = false;
920 if (likely(top_th->tt_multiple_thandle == NULL)) {
921 LASSERT(master_dev != NULL);
922 rc = dt_trans_stop(env, master_dev,
923 top_th->tt_master_sub_thandle);
924 OBD_FREE_PTR(top_th);
928 tmt = top_th->tt_multiple_thandle;
929 tur = tmt->tmt_update_records;
931 /* Note: we need stop the master thandle first, then the stop
932 * callback will fill the master transno in the update logs,
933 * then these update logs will be sent to other MDTs */
934 /* get the master sub thandle */
935 master_st = lookup_sub_thandle(tmt, tmt->tmt_master_sub_dt);
936 write_updates = top_check_write_updates(top_th);
938 /* Step 1: write the updates log on Master MDT */
939 if (master_st != NULL && master_st->st_sub_th != NULL &&
941 struct llog_update_record *lur;
943 /* Merge the parameters and updates into one buffer */
944 rc = prepare_writing_updates(env, tmt);
946 CERROR("%s: cannot prepare updates: rc = %d\n",
947 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
949 GOTO(stop_master_trans, rc);
952 lur = tur->tur_update_records;
953 /* Write updates to the master MDT */
954 rc = sub_updates_write(env, lur, master_st);
956 /* Cleanup the common parameters in the update records,
957 * master transno callback might add more parameters.
958 * and we need merge the update records again in the
960 if (tur->tur_update_params != NULL)
961 lur->lur_update_rec.ur_param_count = 0;
964 CERROR("%s: write updates failed: rc = %d\n",
965 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
967 GOTO(stop_master_trans, rc);
972 /* Step 2: Stop the transaction on the master MDT, and fill the
973 * master transno in the update logs to other MDT. */
974 if (master_st != NULL && master_st->st_sub_th != NULL) {
975 master_st->st_sub_th->th_local = th->th_local;
977 master_st->st_sub_th->th_sync = th->th_sync;
978 master_st->st_sub_th->th_tags = th->th_tags;
979 master_st->st_sub_th->th_result = th->th_result;
980 rc = dt_trans_stop(env, master_st->st_dt, master_st->st_sub_th);
983 GOTO(stop_other_trans, rc);
984 } else if (tur != NULL && tur->tur_update_records != NULL) {
985 struct llog_update_record *lur;
987 lur = tur->tur_update_records;
988 if (lur->lur_update_rec.ur_master_transno == 0)
989 /* Update master transno after master stop
991 lur->lur_update_rec.ur_master_transno =
992 tgt_th_info(env)->tti_transno;
996 /* Step 3: write updates to other MDTs */
998 struct llog_update_record *lur;
1000 /* Stop callback of master will add more updates and also update
1001 * master transno, so merge the parameters and updates into one
1003 rc = prepare_writing_updates(env, tmt);
1005 CERROR("%s: prepare updates failed: rc = %d\n",
1006 master_dev->dd_lu_dev.ld_obd->obd_name, rc);
1008 GOTO(stop_other_trans, rc);
1010 lur = tur->tur_update_records;
1011 list_for_each_entry(st, &tmt->tmt_sub_thandle_list,
1013 if (st->st_sub_th == NULL || st == master_st ||
1014 st->st_sub_th->th_result < 0)
1017 rc = sub_updates_write(env, lur, st);
1026 /* Step 4: Stop the transaction on other MDTs */
1027 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1028 if (st == master_st || st->st_sub_th == NULL)
1032 st->st_sub_th->th_sync = th->th_sync;
1033 st->st_sub_th->th_local = th->th_local;
1034 st->st_sub_th->th_tags = th->th_tags;
1035 st->st_sub_th->th_result = th->th_result;
1036 rc = dt_trans_stop(env, st->st_sub_th->th_dev,
1038 if (unlikely(rc < 0 && th->th_result == 0))
1042 rc = top_trans_wait_result(top_th);
1044 tmt->tmt_result = rc;
1046 /* Balance for the refcount in top_trans_create, Note: if it is NOT
1047 * multiple node transaction, the top transaction will be destroyed. */
1048 top_multiple_thandle_put(tmt);
1049 OBD_FREE_PTR(top_th);
1052 EXPORT_SYMBOL(top_trans_stop);
1055 * Create top_multiple_thandle for top_thandle
1057 * Create top_mutilple_thandle to manage the mutiple node transaction
1058 * for top_thandle, and it also needs to add master sub thandle to the
1059 * sub trans list now.
1061 * \param[in] env execution environment
1062 * \param[in] top_th the top thandle
1064 * \retval 0 if creation succeeds
1065 * \retval negative errno if creation fails
1067 int top_trans_create_tmt(const struct lu_env *env,
1068 struct top_thandle *top_th)
1070 struct top_multiple_thandle *tmt;
1076 tmt->tmt_magic = TOP_THANDLE_MAGIC;
1077 INIT_LIST_HEAD(&tmt->tmt_sub_thandle_list);
1078 INIT_LIST_HEAD(&tmt->tmt_commit_list);
1079 atomic_set(&tmt->tmt_refcount, 1);
1080 spin_lock_init(&tmt->tmt_sub_lock);
1081 init_waitqueue_head(&tmt->tmt_stop_waitq);
1083 top_th->tt_multiple_thandle = tmt;
1088 static struct sub_thandle *
1089 create_sub_thandle_with_thandle(struct top_thandle *top_th,
1090 struct thandle *sub_th)
1092 struct sub_thandle *st;
1094 /* create and init sub th to the top trans list */
1095 st = create_sub_thandle(top_th->tt_multiple_thandle,
1100 st->st_sub_th = sub_th;
1102 sub_th->th_top = &top_th->tt_super;
1109 * Get sub thandle from the top thandle according to the sub dt_device.
1111 * \param[in] env execution environment
1112 * \param[in] th thandle on the top layer.
1113 * \param[in] sub_dt sub dt_device used to get sub transaction
1115 * \retval thandle of sub transaction if succeed
1116 * \retval PTR_ERR(errno) if failed
1118 struct thandle *thandle_get_sub_by_dt(const struct lu_env *env,
1120 struct dt_device *sub_dt)
1122 struct sub_thandle *st = NULL;
1123 struct top_thandle *top_th;
1124 struct thandle *sub_th = NULL;
1128 top_th = container_of(th, struct top_thandle, tt_super);
1130 if (likely(sub_dt == top_th->tt_master_sub_thandle->th_dev))
1131 RETURN(top_th->tt_master_sub_thandle);
1133 if (top_th->tt_multiple_thandle != NULL) {
1134 st = lookup_sub_thandle(top_th->tt_multiple_thandle, sub_dt);
1136 RETURN(st->st_sub_th);
1139 sub_th = dt_trans_create(env, sub_dt);
1143 /* Create top_multiple_thandle if necessary */
1144 if (top_th->tt_multiple_thandle == NULL) {
1145 struct top_multiple_thandle *tmt;
1147 rc = top_trans_create_tmt(env, top_th);
1149 GOTO(stop_trans, rc);
1151 tmt = top_th->tt_multiple_thandle;
1153 /* Add master sub th to the top trans list */
1154 tmt->tmt_master_sub_dt =
1155 top_th->tt_master_sub_thandle->th_dev;
1156 st = create_sub_thandle_with_thandle(top_th,
1157 top_th->tt_master_sub_thandle);
1159 GOTO(stop_trans, rc = PTR_ERR(st));
1162 /* create and init sub th to the top trans list */
1163 st = create_sub_thandle_with_thandle(top_th, sub_th);
1164 st->st_sub_th->th_wait_submit = 1;
1169 sub_th->th_result = rc;
1170 dt_trans_stop(env, sub_dt, sub_th);
1171 sub_th = ERR_PTR(rc);
1176 EXPORT_SYMBOL(thandle_get_sub_by_dt);
1179 * Top multiple thandle destroy
1181 * Destroy multiple thandle and all its sub thandle.
1183 * \param[in] tmt top_multiple_thandle to be destroyed.
1185 void top_multiple_thandle_destroy(struct top_multiple_thandle *tmt)
1187 struct sub_thandle *st;
1188 struct sub_thandle *tmp;
1190 LASSERT(tmt->tmt_magic == TOP_THANDLE_MAGIC);
1191 list_for_each_entry_safe(st, tmp, &tmt->tmt_sub_thandle_list,
1193 struct sub_thandle_cookie *stc;
1194 struct sub_thandle_cookie *tmp;
1196 list_del(&st->st_sub_list);
1197 list_for_each_entry_safe(stc, tmp, &st->st_cookie_list,
1199 list_del(&stc->stc_list);
1206 EXPORT_SYMBOL(top_multiple_thandle_destroy);
1209 * Cancel the update log on MDTs
1211 * Cancel the update log on MDTs then destroy the thandle.
1213 * \param[in] env execution environment
1214 * \param[in] tmt the top multiple thandle whose updates records
1215 * will be cancelled.
1217 * \retval 0 if cancellation succeeds.
1218 * \retval negative errno if cancellation fails.
1220 static int distribute_txn_cancel_records(const struct lu_env *env,
1221 struct top_multiple_thandle *tmt)
1223 struct sub_thandle *st;
1226 top_multiple_thandle_dump(tmt, D_INFO);
1227 /* Cancel update logs on other MDTs */
1228 list_for_each_entry(st, &tmt->tmt_sub_thandle_list, st_sub_list) {
1229 struct llog_ctxt *ctxt;
1230 struct obd_device *obd;
1231 struct llog_cookie *cookie;
1232 struct sub_thandle_cookie *stc;
1235 obd = st->st_dt->dd_lu_dev.ld_obd;
1236 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
1238 list_for_each_entry(stc, &st->st_cookie_list, stc_list) {
1239 cookie = &stc->stc_cookie;
1240 if (fid_is_zero(&cookie->lgc_lgl.lgl_oi.oi_fid))
1243 rc = llog_cat_cancel_records(env, ctxt->loc_handle, 1,
1245 CDEBUG(D_HA, "%s: batchid %llu cancel update log "
1246 DOSTID ".%u : rc = %d\n", obd->obd_name,
1248 POSTID(&cookie->lgc_lgl.lgl_oi),
1249 cookie->lgc_index, rc);
1252 llog_ctxt_put(ctxt);
1259 * Check if there are committed transaction
1261 * Check if there are committed transaction in the distribute transaction
1262 * list, then cancel the update records for those committed transaction.
1263 * Because the distribute transaction in the list are sorted by batchid,
1264 * and cancellation will be done by batchid order, so we only check the first
1265 * the transaction(with lowest batchid) in the list.
1267 * \param[in] lod lod device where cancel thread is
1269 * \retval true if it is ready
1270 * \retval false if it is not ready
1272 static bool tdtd_ready_for_cancel_log(struct target_distribute_txn_data *tdtd)
1274 struct top_multiple_thandle *tmt = NULL;
1275 struct obd_device *obd = tdtd->tdtd_lut->lut_obd;
1278 spin_lock(&tdtd->tdtd_batchid_lock);
1279 if (!list_empty(&tdtd->tdtd_list)) {
1280 tmt = list_entry(tdtd->tdtd_list.next,
1281 struct top_multiple_thandle, tmt_commit_list);
1282 if (tmt->tmt_committed &&
1283 (!obd->obd_recovering || (obd->obd_recovering &&
1284 tmt->tmt_batchid <= tdtd->tdtd_committed_batchid)))
1287 spin_unlock(&tdtd->tdtd_batchid_lock);
1292 struct distribute_txn_bid_data {
1293 struct dt_txn_commit_cb dtbd_cb;
1294 struct target_distribute_txn_data *dtbd_tdtd;
1299 * callback of updating commit batchid
1301 * Updating commit batchid then wake up the commit thread to cancel the
1304 * \param[in]env execution environment
1305 * \param[in]th thandle to updating commit batchid
1306 * \param[in]cb commit callback
1307 * \param[in]err result of thandle
1309 static void distribute_txn_batchid_cb(struct lu_env *env,
1311 struct dt_txn_commit_cb *cb,
1314 struct distribute_txn_bid_data *dtbd = NULL;
1315 struct target_distribute_txn_data *tdtd;
1317 dtbd = container_of0(cb, struct distribute_txn_bid_data, dtbd_cb);
1318 tdtd = dtbd->dtbd_tdtd;
1320 CDEBUG(D_HA, "%s: %llu batchid updated\n",
1321 tdtd->tdtd_lut->lut_obd->obd_name, dtbd->dtbd_batchid);
1322 spin_lock(&tdtd->tdtd_batchid_lock);
1323 if (dtbd->dtbd_batchid > tdtd->tdtd_committed_batchid &&
1324 !tdtd->tdtd_lut->lut_obd->obd_no_transno)
1325 tdtd->tdtd_committed_batchid = dtbd->dtbd_batchid;
1326 spin_unlock(&tdtd->tdtd_batchid_lock);
1327 atomic_dec(&tdtd->tdtd_refcount);
1328 wake_up(&tdtd->tdtd_commit_thread_waitq);
1334 * Update the commit batchid in disk
1336 * Update commit batchid in the disk, after this is committed, it can start
1337 * to cancel the update records.
1339 * \param[in] env execution environment
1340 * \param[in] tdtd distribute transaction structure
1341 * \param[in] batchid commit batchid to be updated
1343 * \retval 0 if update succeeds.
1344 * \retval negative errno if update fails.
1347 distribute_txn_commit_batchid_update(const struct lu_env *env,
1348 struct target_distribute_txn_data *tdtd,
1351 struct distribute_txn_bid_data *dtbd = NULL;
1359 OBD_ALLOC_PTR(dtbd);
1362 dtbd->dtbd_batchid = batchid;
1363 dtbd->dtbd_tdtd = tdtd;
1364 dtbd->dtbd_cb.dcb_func = distribute_txn_batchid_cb;
1365 atomic_inc(&tdtd->tdtd_refcount);
1367 th = dt_trans_create(env, tdtd->tdtd_lut->lut_bottom);
1370 RETURN(PTR_ERR(th));
1373 tmp = cpu_to_le64(batchid);
1375 buf.lb_len = sizeof(tmp);
1378 rc = dt_declare_record_write(env, tdtd->tdtd_batchid_obj, &buf, off,
1383 rc = dt_trans_start_local(env, tdtd->tdtd_lut->lut_bottom, th);
1387 rc = dt_trans_cb_add(th, &dtbd->dtbd_cb);
1391 rc = dt_record_write(env, tdtd->tdtd_batchid_obj, &buf,
1394 CDEBUG(D_INFO, "%s: update batchid "LPU64": rc = %d\n",
1395 tdtd->tdtd_lut->lut_obd->obd_name, batchid, rc);
1398 dt_trans_stop(env, tdtd->tdtd_lut->lut_bottom, th);
1405 * Init commit batchid for distribute transaction.
1407 * Initialize the batchid object and get commit batchid from the object.
1409 * \param[in] env execution environment
1410 * \param[in] tdtd distribute transaction whose batchid is initialized.
1412 * \retval 0 if initialization succeeds.
1413 * \retval negative errno if initialization fails.
1416 distribute_txn_commit_batchid_init(const struct lu_env *env,
1417 struct target_distribute_txn_data *tdtd)
1419 struct tgt_thread_info *tti = tgt_th_info(env);
1420 struct lu_target *lut = tdtd->tdtd_lut;
1421 struct lu_attr *attr = &tti->tti_attr;
1422 struct lu_fid *fid = &tti->tti_fid1;
1423 struct dt_object_format *dof = &tti->tti_u.update.tti_update_dof;
1424 struct dt_object *dt_obj = NULL;
1431 memset(attr, 0, sizeof(*attr));
1432 attr->la_valid = LA_MODE;
1433 attr->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1434 dof->dof_type = dt_mode_to_dft(S_IFREG);
1436 lu_local_obj_fid(fid, BATCHID_COMMITTED_OID);
1438 dt_obj = dt_find_or_create(env, lut->lut_bottom, fid, dof,
1440 if (IS_ERR(dt_obj)) {
1441 rc = PTR_ERR(dt_obj);
1446 tdtd->tdtd_batchid_obj = dt_obj;
1449 buf.lb_len = sizeof(tmp);
1451 rc = dt_read(env, dt_obj, &buf, &off);
1452 if (rc < 0 || (rc < buf.lb_len && rc > 0)) {
1453 CERROR("%s can't read last committed batchid: rc = %d\n",
1454 tdtd->tdtd_lut->lut_obd->obd_name, rc);
1458 } else if (rc == buf.lb_len) {
1459 tdtd->tdtd_committed_batchid = le64_to_cpu(tmp);
1460 CDEBUG(D_HA, "%s: committed batchid %llu\n",
1461 tdtd->tdtd_lut->lut_obd->obd_name,
1462 tdtd->tdtd_committed_batchid);
1467 if (rc < 0 && dt_obj != NULL) {
1468 lu_object_put(env, &dt_obj->do_lu);
1469 tdtd->tdtd_batchid_obj = NULL;
1475 * manage the distribute transaction thread
1477 * Distribute transaction are linked to the list, and once the distribute
1478 * transaction is committed, it will update the last committed batchid first,
1479 * after it is committed, it will cancel the records.
1481 * \param[in] _arg argument for commit thread
1483 * \retval 0 if thread is running successfully
1484 * \retval negative errno if the thread can not be run.
1486 static int distribute_txn_commit_thread(void *_arg)
1488 struct target_distribute_txn_data *tdtd = _arg;
1489 struct lu_target *lut = tdtd->tdtd_lut;
1490 struct ptlrpc_thread *thread = &lut->lut_tdtd_commit_thread;
1491 struct l_wait_info lwi = { 0 };
1493 struct list_head list;
1495 struct top_multiple_thandle *tmt;
1496 struct top_multiple_thandle *tmp;
1497 __u64 batchid = 0, committed;
1501 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
1505 spin_lock(&tdtd->tdtd_batchid_lock);
1506 thread->t_flags = SVC_RUNNING;
1507 spin_unlock(&tdtd->tdtd_batchid_lock);
1508 wake_up(&thread->t_ctl_waitq);
1509 INIT_LIST_HEAD(&list);
1511 CDEBUG(D_HA, "%s: start commit thread committed batchid "LPU64"\n",
1512 tdtd->tdtd_lut->lut_obd->obd_name,
1513 tdtd->tdtd_committed_batchid);
1515 while (distribute_txn_commit_thread_running(lut)) {
1516 spin_lock(&tdtd->tdtd_batchid_lock);
1517 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1519 if (tmt->tmt_committed == 0)
1522 /* Note: right now, replay is based on master MDT
1523 * transno, but cancellation is based on batchid.
1524 * so we do not try to cancel the update log until
1525 * the recoverying is done, unless the update records
1526 * batchid < committed_batchid. */
1527 if (tmt->tmt_batchid <= tdtd->tdtd_committed_batchid) {
1528 list_move_tail(&tmt->tmt_commit_list, &list);
1529 } else if (!tdtd->tdtd_lut->lut_obd->obd_recovering) {
1530 LASSERTF(tmt->tmt_batchid >= batchid,
1531 "tmt %p tmt_batchid: "LPU64", batchid "
1532 LPU64"\n", tmt, tmt->tmt_batchid,
1534 /* There are three types of distribution
1535 * transaction result
1537 * 1. If tmt_result < 0, it means the
1538 * distribution transaction fails, which should
1539 * be rare, because once declare phase succeeds,
1540 * the operation should succeeds anyway. Note in
1541 * this case, we will still update batchid so
1542 * cancellation would be stopped.
1544 * 2. If tmt_result == 0, it means the
1545 * distribution transaction succeeds, and we
1546 * will update batchid.
1548 * 3. If tmt_result > 0, it means distribute
1549 * transaction is not yet committed on every
1550 * node, but we need release this tmt before
1551 * that, which usuually happens during umount.
1553 if (tmt->tmt_result <= 0)
1554 batchid = tmt->tmt_batchid;
1555 list_move_tail(&tmt->tmt_commit_list, &list);
1558 spin_unlock(&tdtd->tdtd_batchid_lock);
1560 CDEBUG(D_HA, "%s: batchid: "LPU64" committed batchid "
1561 LPU64"\n", tdtd->tdtd_lut->lut_obd->obd_name, batchid,
1562 tdtd->tdtd_committed_batchid);
1563 /* update globally committed on a storage */
1564 if (batchid > tdtd->tdtd_committed_batchid) {
1565 distribute_txn_commit_batchid_update(&env, tdtd,
1567 spin_lock(&tdtd->tdtd_batchid_lock);
1568 if (batchid > tdtd->tdtd_batchid) {
1569 /* This might happen during recovery,
1570 * batchid is initialized as last transno,
1571 * and the batchid in the update records
1572 * on other MDTs might be bigger than
1573 * the batchid, so we need update it to
1574 * avoid duplicate batchid. */
1575 CDEBUG(D_HA, "%s update batchid from "LPU64
1577 tdtd->tdtd_lut->lut_obd->obd_name,
1578 tdtd->tdtd_batchid, batchid);
1579 tdtd->tdtd_batchid = batchid;
1581 spin_unlock(&tdtd->tdtd_batchid_lock);
1583 /* cancel the records for committed batchid's */
1584 /* XXX: should we postpone cancel's till the end of recovery? */
1585 committed = tdtd->tdtd_committed_batchid;
1586 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1587 if (tmt->tmt_batchid > committed)
1589 list_del_init(&tmt->tmt_commit_list);
1590 if (tmt->tmt_result <= 0)
1591 distribute_txn_cancel_records(&env, tmt);
1592 top_multiple_thandle_put(tmt);
1595 l_wait_event(tdtd->tdtd_commit_thread_waitq,
1596 !distribute_txn_commit_thread_running(lut) ||
1597 committed < tdtd->tdtd_committed_batchid ||
1598 tdtd_ready_for_cancel_log(tdtd), &lwi);
1601 l_wait_event(tdtd->tdtd_commit_thread_waitq,
1602 atomic_read(&tdtd->tdtd_refcount) == 0, &lwi);
1604 spin_lock(&tdtd->tdtd_batchid_lock);
1605 list_for_each_entry_safe(tmt, tmp, &tdtd->tdtd_list,
1607 list_move_tail(&tmt->tmt_commit_list, &list);
1608 spin_unlock(&tdtd->tdtd_batchid_lock);
1610 CDEBUG(D_INFO, "%s stopping distribute txn commit thread.\n",
1611 tdtd->tdtd_lut->lut_obd->obd_name);
1612 list_for_each_entry_safe(tmt, tmp, &list, tmt_commit_list) {
1613 list_del_init(&tmt->tmt_commit_list);
1614 top_multiple_thandle_dump(tmt, D_HA);
1615 top_multiple_thandle_put(tmt);
1618 thread->t_flags = SVC_STOPPED;
1620 wake_up(&thread->t_ctl_waitq);
1626 * Start llog cancel thread
1628 * Start llog cancel(master/slave) thread on LOD
1630 * \param[in]lclt cancel log thread to be started.
1632 * \retval 0 if the thread is started successfully.
1633 * \retval negative errno if the thread is not being
1636 int distribute_txn_init(const struct lu_env *env,
1637 struct lu_target *lut,
1638 struct target_distribute_txn_data *tdtd,
1641 struct task_struct *task;
1642 struct l_wait_info lwi = { 0 };
1646 spin_lock_init(&tdtd->tdtd_batchid_lock);
1647 INIT_LIST_HEAD(&tdtd->tdtd_list);
1649 tdtd->tdtd_batchid = lut->lut_last_transno + 1;
1651 init_waitqueue_head(&lut->lut_tdtd_commit_thread.t_ctl_waitq);
1652 init_waitqueue_head(&tdtd->tdtd_commit_thread_waitq);
1653 atomic_set(&tdtd->tdtd_refcount, 0);
1655 tdtd->tdtd_lut = lut;
1656 rc = distribute_txn_commit_batchid_init(env, tdtd);
1660 task = kthread_run(distribute_txn_commit_thread, tdtd, "tdtd-%u",
1663 RETURN(PTR_ERR(task));
1665 l_wait_event(lut->lut_tdtd_commit_thread.t_ctl_waitq,
1666 distribute_txn_commit_thread_running(lut) ||
1667 distribute_txn_commit_thread_stopped(lut), &lwi);
1670 EXPORT_SYMBOL(distribute_txn_init);
1673 * Stop llog cancel thread
1675 * Stop llog cancel(master/slave) thread on LOD and also destory
1676 * all of transaction in the list.
1678 * \param[in]lclt cancel log thread to be stopped.
1680 void distribute_txn_fini(const struct lu_env *env,
1681 struct target_distribute_txn_data *tdtd)
1683 struct lu_target *lut = tdtd->tdtd_lut;
1685 /* Stop cancel thread */
1686 if (lut == NULL || !distribute_txn_commit_thread_running(lut))
1689 spin_lock(&tdtd->tdtd_batchid_lock);
1690 lut->lut_tdtd_commit_thread.t_flags = SVC_STOPPING;
1691 spin_unlock(&tdtd->tdtd_batchid_lock);
1692 wake_up(&tdtd->tdtd_commit_thread_waitq);
1693 wait_event(lut->lut_tdtd_commit_thread.t_ctl_waitq,
1694 lut->lut_tdtd_commit_thread.t_flags & SVC_STOPPED);
1696 dtrq_list_destroy(tdtd);
1697 if (tdtd->tdtd_batchid_obj != NULL) {
1698 lu_object_put(env, &tdtd->tdtd_batchid_obj->do_lu);
1699 tdtd->tdtd_batchid_obj = NULL;
1702 EXPORT_SYMBOL(distribute_txn_fini);