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.
27 * lustre/target/update_recovery.c
29 * This file implement the methods to handle the update recovery.
31 * During DNE recovery, the recovery thread will redo the operation according
32 * to the transaction no, and these replay are either from client replay req
33 * or update replay records(for distribute transaction) in the update log.
34 * For distribute transaction replay, the replay thread will call
35 * distribute_txn_replay_handle() to handle the updates.
37 * After the Master MDT restarts, it will retrieve the update records from all
38 * of MDTs, for each distributed operation, it will check updates on all MDTs,
39 * if some updates records are missing on some MDTs, the replay thread will redo
40 * updates on these MDTs.
42 * Author: Di Wang <di.wang@intel.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
46 #include <lu_target.h>
47 #include <md_object.h>
48 #include <lustre_update.h>
50 #include <obd_class.h>
51 #include "tgt_internal.h"
54 * Lookup distribute_txn_replay req
56 * Lookup distribute_txn_replay in the replay list by batchid.
57 * It is assumed the list has been locked before calling this function.
59 * \param[in] tdtd distribute_txn_data, which holds the replay
61 * \param[in] batchid batchid used by lookup.
63 * \retval pointer of the replay if succeeds.
64 * \retval NULL if can not find it.
66 static struct distribute_txn_replay_req *
67 dtrq_lookup(struct target_distribute_txn_data *tdtd, __u64 batchid)
69 struct distribute_txn_replay_req *tmp;
70 struct distribute_txn_replay_req *dtrq = NULL;
72 list_for_each_entry(tmp, &tdtd->tdtd_replay_list, dtrq_list) {
73 if (tmp->dtrq_batchid == batchid) {
82 * insert distribute txn replay req
84 * Insert distribute txn replay to the replay list, and it assumes the
85 * list has been looked. Note: the replay list is a sorted list, which
86 * is sorted by master transno. It is assumed the replay list has been
87 * locked before calling this function.
89 * \param[in] tdtd target distribute txn data where replay list is
90 * \param[in] new distribute txn replay to be inserted
92 * \retval 0 if insertion succeeds
93 * \retval EEXIST if the dtrq already exists
95 static int dtrq_insert(struct target_distribute_txn_data *tdtd,
96 struct distribute_txn_replay_req *new)
98 struct distribute_txn_replay_req *iter;
100 /* Check if the dtrq has been added to the list */
101 iter = dtrq_lookup(tdtd, new->dtrq_batchid);
105 list_for_each_entry_reverse(iter, &tdtd->tdtd_replay_list, dtrq_list) {
106 if (iter->dtrq_master_transno > new->dtrq_master_transno)
109 /* If there are mulitple replay req with same transno, then
110 * sort them with batchid */
111 if (iter->dtrq_master_transno == new->dtrq_master_transno &&
112 iter->dtrq_batchid > new->dtrq_batchid)
115 list_add(&new->dtrq_list, &iter->dtrq_list);
119 if (list_empty(&new->dtrq_list))
120 list_add(&new->dtrq_list, &tdtd->tdtd_replay_list);
126 * create distribute txn replay req
128 * Allocate distribute txn replay req according to the update records.
130 * \param[in] tdtd target distribute txn data where replay list is.
131 * \param[in] record update records from the update log.
133 * \retval the pointer of distribute txn replay req if
134 * the creation succeeds.
135 * \retval NULL if the creation fails.
137 static struct distribute_txn_replay_req *
138 dtrq_create(struct target_distribute_txn_data *tdtd,
139 struct llog_update_record *lur)
141 struct distribute_txn_replay_req *new;
145 RETURN(ERR_PTR(-ENOMEM));
147 new->dtrq_lur_size = llog_update_record_size(lur);
148 OBD_ALLOC_LARGE(new->dtrq_lur, new->dtrq_lur_size);
149 if (new->dtrq_lur == NULL) {
151 RETURN(ERR_PTR(-ENOMEM));
154 memcpy(new->dtrq_lur, lur, new->dtrq_lur_size);
156 /* If the transno in the update record is 0, it means the
157 * update are from master MDT, and it will use the master
158 * last committed transno as its master transno. Later, if
159 * the update records are gotten from slave MDTs, then these
160 * transno will be replaced.
161 * See insert_update_records_to_replay_list(). */
162 if (lur->lur_update_rec.ur_master_transno == 0) {
163 new->dtrq_lur->lur_update_rec.ur_master_transno =
164 tdtd->tdtd_lut->lut_obd->obd_last_committed;
165 new->dtrq_master_transno =
166 tdtd->tdtd_lut->lut_obd->obd_last_committed;
168 new->dtrq_master_transno =
169 lur->lur_update_rec.ur_master_transno;
172 new->dtrq_batchid = lur->lur_update_rec.ur_batchid;
174 spin_lock_init(&new->dtrq_sub_list_lock);
175 INIT_LIST_HEAD(&new->dtrq_sub_list);
176 INIT_LIST_HEAD(&new->dtrq_list);
182 * Lookup distribute sub replay
184 * Lookup distribute sub replay in the sub list of distribute_txn_replay by
187 * \param[in] distribute_txn_replay_req the distribute txn replay req to lookup
188 * \param[in] mdt_index the mdt_index as the key of lookup
190 * \retval the pointer of sub replay if it can be found.
191 * \retval NULL if it can not find.
193 struct distribute_txn_replay_req_sub *
194 dtrq_sub_lookup(struct distribute_txn_replay_req *dtrq, __u32 mdt_index)
196 struct distribute_txn_replay_req_sub *dtrqs = NULL;
197 struct distribute_txn_replay_req_sub *tmp;
199 list_for_each_entry(tmp, &dtrq->dtrq_sub_list, dtrqs_list) {
200 if (tmp->dtrqs_mdt_index == mdt_index) {
209 * Try to add cookie to sub distribute txn request
211 * Check if the update log cookie has been added to the request, if not,
212 * add it to the dtrqs_cookie_list.
214 * \param[in] dtrqs sub replay req where cookies to be added.
215 * \param[in] cookie cookie to be added.
217 * \retval 0 if the cookie is adding succeeds.
218 * \retval negative errno if adding fails.
220 static int dtrq_sub_add_cookie(struct distribute_txn_replay_req_sub *dtrqs,
221 struct llog_cookie *cookie)
223 struct sub_thandle_cookie *new;
229 INIT_LIST_HEAD(&new->stc_list);
230 new->stc_cookie = *cookie;
231 /* Note: only single thread will access one sub_request each time,
232 * so no need lock here */
233 list_add(&new->stc_list, &dtrqs->dtrqs_cookie_list);
239 * Insert distribute txn sub req replay
241 * Allocate sub replay req and insert distribute txn replay list.
243 * \param[in] dtrq d to be added
244 * \param[in] cookie the cookie of the update record
245 * \param[in] mdt_index the mdt_index of the update record
247 * \retval 0 if the adding succeeds.
248 * \retval negative errno if the adding fails.
251 dtrq_sub_create_and_insert(struct distribute_txn_replay_req *dtrq,
252 struct llog_cookie *cookie,
255 struct distribute_txn_replay_req_sub *dtrqs = NULL;
256 struct distribute_txn_replay_req_sub *new;
260 spin_lock(&dtrq->dtrq_sub_list_lock);
261 dtrqs = dtrq_sub_lookup(dtrq, mdt_index);
262 spin_unlock(&dtrq->dtrq_sub_list_lock);
264 rc = dtrq_sub_add_cookie(dtrqs, cookie);
272 INIT_LIST_HEAD(&new->dtrqs_list);
273 INIT_LIST_HEAD(&new->dtrqs_cookie_list);
274 new->dtrqs_mdt_index = mdt_index;
275 spin_lock(&dtrq->dtrq_sub_list_lock);
276 dtrqs = dtrq_sub_lookup(dtrq, mdt_index);
278 list_add(&new->dtrqs_list, &dtrq->dtrq_sub_list);
283 spin_unlock(&dtrq->dtrq_sub_list_lock);
285 rc = dtrq_sub_add_cookie(dtrqs, cookie);
291 * append updates to the current replay updates
293 * Append more updates to the existent replay update. And this is only
294 * used when combining mulitple updates into one large updates during
297 * \param[in] dtrq the update replay request where the new update
298 * records will be added.
299 * \param[in] lur the new update record.
301 * \retval 0 if appending succeeds.
302 * \retval negative errno if appending fails.
304 static int dtrq_append_updates(struct distribute_txn_replay_req *dtrq,
305 struct update_records *record)
307 struct llog_update_record *new_lur;
308 size_t lur_size = dtrq->dtrq_lur_size;
312 /* Because several threads might retrieve the same records from
313 * different targets, and we only need one copy of records. So
314 * we will check if the records is in the next one, if not, just
316 spin_lock(&dtrq->dtrq_sub_list_lock);
317 if (dtrq->dtrq_lur->lur_update_rec.ur_index + 1 != record->ur_index) {
318 spin_unlock(&dtrq->dtrq_sub_list_lock);
321 dtrq->dtrq_lur->lur_update_rec.ur_index++;
322 spin_unlock(&dtrq->dtrq_sub_list_lock);
324 lur_size += update_records_size(record);
325 OBD_ALLOC_LARGE(new_lur, lur_size);
326 if (new_lur == NULL) {
327 spin_lock(&dtrq->dtrq_sub_list_lock);
328 dtrq->dtrq_lur->lur_update_rec.ur_index--;
329 spin_unlock(&dtrq->dtrq_sub_list_lock);
333 /* Copy the old and new records to the new allocated buffer */
334 memcpy(new_lur, dtrq->dtrq_lur, dtrq->dtrq_lur_size);
335 ptr = (char *)&new_lur->lur_update_rec +
336 update_records_size(&new_lur->lur_update_rec);
337 memcpy(ptr, &record->ur_ops,
338 update_records_size(record) -
339 offsetof(struct update_records, ur_ops));
341 new_lur->lur_update_rec.ur_update_count += record->ur_update_count;
342 new_lur->lur_update_rec.ur_param_count += record->ur_param_count;
343 new_lur->lur_hdr.lrh_len = llog_update_record_size(new_lur);
345 /* Replace the records */
346 OBD_FREE_LARGE(dtrq->dtrq_lur, dtrq->dtrq_lur_size);
347 dtrq->dtrq_lur = new_lur;
348 dtrq->dtrq_lur_size = lur_size;
349 dtrq->dtrq_lur->lur_update_rec.ur_flags = record->ur_flags;
350 update_records_dump(&new_lur->lur_update_rec, D_INFO, true);
355 * Insert update records to the replay list.
357 * Allocate distribute txn replay req and insert it into the replay
358 * list, then insert the update records into the replay req.
360 * \param[in] tdtd distribute txn replay data where the replay list
362 * \param[in] record the update record
363 * \param[in] cookie cookie of the record
364 * \param[in] index mdt index of the record
366 * \retval 0 if the adding succeeds.
367 * \retval negative errno if the adding fails.
370 insert_update_records_to_replay_list(struct target_distribute_txn_data *tdtd,
371 struct llog_update_record *lur,
372 struct llog_cookie *cookie,
375 struct distribute_txn_replay_req *dtrq;
376 struct update_records *record = &lur->lur_update_rec;
377 bool replace_record = false;
381 CDEBUG(D_HA, "%s: insert record batchid = "LPU64" transno = "LPU64
382 " mdt_index %u\n", tdtd->tdtd_lut->lut_obd->obd_name,
383 record->ur_batchid, record->ur_master_transno, mdt_index);
385 /* Update batchid if necessary */
386 spin_lock(&tdtd->tdtd_batchid_lock);
387 if (record->ur_batchid >= tdtd->tdtd_batchid) {
388 CDEBUG(D_HA, "%s update batchid from "LPU64 " to "LPU64"\n",
389 tdtd->tdtd_lut->lut_obd->obd_name,
390 tdtd->tdtd_batchid, record->ur_batchid);
391 tdtd->tdtd_batchid = record->ur_batchid + 1;
393 spin_unlock(&tdtd->tdtd_batchid_lock);
396 spin_lock(&tdtd->tdtd_replay_list_lock);
397 /* First try to build the replay update request with the records */
398 dtrq = dtrq_lookup(tdtd, record->ur_batchid);
400 spin_unlock(&tdtd->tdtd_replay_list_lock);
401 dtrq = dtrq_create(tdtd, lur);
403 RETURN(PTR_ERR(dtrq));
405 spin_lock(&tdtd->tdtd_replay_list_lock);
406 rc = dtrq_insert(tdtd, dtrq);
408 spin_unlock(&tdtd->tdtd_replay_list_lock);
415 /* If the master transno in update header is not
416 * matched with the one in the record, then it means
417 * the dtrq is originally created by master record,
418 * so we need update master transno and reposition
419 * the dtrq(by master transno) in the list and also
420 * replace update record */
421 if (record->ur_master_transno != 0 &&
422 dtrq->dtrq_master_transno != record->ur_master_transno &&
423 dtrq->dtrq_lur != NULL) {
424 list_del_init(&dtrq->dtrq_list);
425 dtrq->dtrq_lur->lur_update_rec.ur_master_transno =
426 record->ur_master_transno;
428 dtrq->dtrq_master_transno = record->ur_master_transno;
429 replace_record = true;
430 /* try to insert again */
431 rc = dtrq_insert(tdtd, dtrq);
433 spin_unlock(&tdtd->tdtd_replay_list_lock);
439 spin_unlock(&tdtd->tdtd_replay_list_lock);
441 /* Because there should be only thread access the update record, so
442 * we do not need lock here */
443 if (replace_record) {
444 /* Replace the update record and master transno */
445 OBD_FREE(dtrq->dtrq_lur, dtrq->dtrq_lur_size);
446 dtrq->dtrq_lur = NULL;
447 dtrq->dtrq_lur_size = llog_update_record_size(lur);
448 OBD_ALLOC_LARGE(dtrq->dtrq_lur, dtrq->dtrq_lur_size);
449 if (dtrq->dtrq_lur == NULL)
452 memcpy(dtrq->dtrq_lur, lur, dtrq->dtrq_lur_size);
455 /* This is a partial update records, let's try to append
456 * the record to the current replay request */
457 if (record->ur_flags & UPDATE_RECORD_CONTINUE)
458 rc = dtrq_append_updates(dtrq, record);
460 /* Then create and add sub update request */
461 rc = dtrq_sub_create_and_insert(dtrq, cookie, mdt_index);
465 EXPORT_SYMBOL(insert_update_records_to_replay_list);
468 * Dump updates of distribute txns.
470 * Output all of recovery updates in the distribute txn list to the
473 * \param[in] tdtd distribute txn data where all of distribute txn
475 * \param[in] mask debug mask
477 void dtrq_list_dump(struct target_distribute_txn_data *tdtd, unsigned int mask)
479 struct distribute_txn_replay_req *dtrq;
481 spin_lock(&tdtd->tdtd_replay_list_lock);
482 list_for_each_entry(dtrq, &tdtd->tdtd_replay_list, dtrq_list)
483 update_records_dump(&dtrq->dtrq_lur->lur_update_rec, mask,
485 spin_unlock(&tdtd->tdtd_replay_list_lock);
487 EXPORT_SYMBOL(dtrq_list_dump);
490 * Destroy distribute txn replay req
492 * Destroy distribute txn replay req and all of subs.
494 * \param[in] dtrq distribute txn replqy req to be destroyed.
496 void dtrq_destroy(struct distribute_txn_replay_req *dtrq)
498 struct distribute_txn_replay_req_sub *dtrqs;
499 struct distribute_txn_replay_req_sub *tmp;
501 LASSERT(list_empty(&dtrq->dtrq_list));
502 spin_lock(&dtrq->dtrq_sub_list_lock);
503 list_for_each_entry_safe(dtrqs, tmp, &dtrq->dtrq_sub_list, dtrqs_list) {
504 struct sub_thandle_cookie *stc;
505 struct sub_thandle_cookie *tmp;
507 list_del(&dtrqs->dtrqs_list);
508 list_for_each_entry_safe(stc, tmp, &dtrqs->dtrqs_cookie_list,
510 list_del(&stc->stc_list);
515 spin_unlock(&dtrq->dtrq_sub_list_lock);
517 if (dtrq->dtrq_lur != NULL)
518 OBD_FREE_LARGE(dtrq->dtrq_lur, dtrq->dtrq_lur_size);
522 EXPORT_SYMBOL(dtrq_destroy);
525 * Destroy all of replay req.
527 * Destroy all of replay req in the replay list.
529 * \param[in] tdtd target distribute txn data where the replay list is.
531 void dtrq_list_destroy(struct target_distribute_txn_data *tdtd)
533 struct distribute_txn_replay_req *dtrq;
534 struct distribute_txn_replay_req *tmp;
536 spin_lock(&tdtd->tdtd_replay_list_lock);
537 list_for_each_entry_safe(dtrq, tmp, &tdtd->tdtd_replay_list,
539 list_del_init(&dtrq->dtrq_list);
542 spin_unlock(&tdtd->tdtd_replay_list_lock);
544 EXPORT_SYMBOL(dtrq_list_destroy);
547 * Get next req in the replay list
549 * Get next req needs to be replayed, since it is a sorted list
550 * (by master MDT transno)
552 * \param[in] tdtd distribute txn data where the replay list is
554 * \retval the pointer of update recovery header
556 struct distribute_txn_replay_req *
557 distribute_txn_get_next_req(struct target_distribute_txn_data *tdtd)
559 struct distribute_txn_replay_req *dtrq = NULL;
561 spin_lock(&tdtd->tdtd_replay_list_lock);
562 if (!list_empty(&tdtd->tdtd_replay_list)) {
563 dtrq = list_entry(tdtd->tdtd_replay_list.next,
564 struct distribute_txn_replay_req, dtrq_list);
565 list_del_init(&dtrq->dtrq_list);
567 spin_unlock(&tdtd->tdtd_replay_list_lock);
571 EXPORT_SYMBOL(distribute_txn_get_next_req);
574 * Get next transno in the replay list, because this is the sorted
575 * list, so it will return the transno of next req in the list.
577 * \param[in] tdtd distribute txn data where the replay list is
579 * \retval the transno of next update in the list
581 __u64 distribute_txn_get_next_transno(struct target_distribute_txn_data *tdtd)
583 struct distribute_txn_replay_req *dtrq = NULL;
586 spin_lock(&tdtd->tdtd_replay_list_lock);
587 if (!list_empty(&tdtd->tdtd_replay_list)) {
588 dtrq = list_entry(tdtd->tdtd_replay_list.next,
589 struct distribute_txn_replay_req, dtrq_list);
590 transno = dtrq->dtrq_master_transno;
592 spin_unlock(&tdtd->tdtd_replay_list_lock);
594 CDEBUG(D_HA, "%s: Next update transno "LPU64"\n",
595 tdtd->tdtd_lut->lut_obd->obd_name, transno);
598 EXPORT_SYMBOL(distribute_txn_get_next_transno);
601 * Check if the update of one object is committed
603 * Check whether the update for the object is committed by checking whether
604 * the correspondent sub exists in the replay req. If it is committed, mark
605 * the committed flag in correspondent the sub thandle.
607 * \param[in] env execution environment
608 * \param[in] dtrq replay request
609 * \param[in] dt_obj object for the update
610 * \param[in] top_th top thandle
611 * \param[in] sub_th sub thandle which the update belongs to
613 * \retval 1 if the update is not committed.
614 * \retval 0 if the update is committed.
615 * \retval negative errno if some other failures happen.
617 static int update_is_committed(const struct lu_env *env,
618 struct distribute_txn_replay_req *dtrq,
619 struct dt_object *dt_obj,
620 struct top_thandle *top_th,
621 struct sub_thandle *st)
623 struct seq_server_site *seq_site;
624 const struct lu_fid *fid = lu_object_fid(&dt_obj->do_lu);
625 struct distribute_txn_replay_req_sub *dtrqs;
629 if (st->st_sub_th != NULL)
632 if (st->st_committed)
635 seq_site = lu_site2seq(dt_obj->do_lu.lo_dev->ld_site);
636 if (fid_is_update_log(fid) || fid_is_update_log_dir(fid)) {
637 mdt_index = fid_oid(fid);
638 } else if (!fid_seq_in_fldb(fid_seq(fid))) {
639 mdt_index = seq_site->ss_node_id;
641 struct lu_server_fld *fld;
642 struct lu_seq_range range = {0};
645 fld = seq_site->ss_server_fld;
646 fld_range_set_type(&range, LU_SEQ_RANGE_MDT);
647 LASSERT(fld->lsf_seq_lookup != NULL);
648 rc = fld->lsf_seq_lookup(env, fld, fid_seq(fid),
652 mdt_index = range.lsr_index;
655 dtrqs = dtrq_sub_lookup(dtrq, mdt_index);
656 if (dtrqs != NULL || top_th->tt_multiple_thandle->tmt_committed) {
657 st->st_committed = 1;
659 struct sub_thandle_cookie *stc;
660 struct sub_thandle_cookie *tmp;
662 list_for_each_entry_safe(stc, tmp,
663 &dtrqs->dtrqs_cookie_list,
665 list_move(&stc->stc_list, &st->st_cookie_list);
670 CDEBUG(D_HA, "Update of "DFID "on MDT%u is not committed\n", PFID(fid),
677 * Implementation of different update methods for update recovery.
679 * These following functions update_recovery_$(update_name) implement
680 * different updates recovery methods. They will extract the parameters
681 * from the common parameters area and call correspondent dt API to redo
684 * \param[in] env execution environment
685 * \param[in] op update operation to be replayed
686 * \param[in] params common update parameters which holds all parameters
688 * \param[in] th transaction handle
689 * \param[in] declare indicate it will do declare or real execution, true
690 * means declare, false means real execution
692 * \retval 0 if it succeeds.
693 * \retval negative errno if it fails.
695 static int update_recovery_create(const struct lu_env *env,
696 struct dt_object *dt_obj,
697 const struct update_op *op,
698 const struct update_params *params,
699 struct thandle_exec_args *ta,
702 struct update_thread_info *uti = update_env_info(env);
703 struct llog_update_record *lur = uti->uti_dtrq->dtrq_lur;
704 struct lu_attr *attr = &uti->uti_attr;
706 struct obdo *lobdo = &uti->uti_obdo;
707 struct dt_object_format dof;
709 unsigned int param_count;
713 if (dt_object_exists(dt_obj))
716 param_count = lur->lur_update_rec.ur_param_count;
717 wobdo = update_params_get_param_buf(params, op->uop_params_off[0],
721 if (size != sizeof(*wobdo))
724 if (LLOG_REC_HDR_NEEDS_SWABBING(&lur->lur_hdr))
725 lustre_swab_obdo(wobdo);
727 lustre_get_wire_obdo(NULL, lobdo, wobdo);
728 la_from_obdo(attr, lobdo, lobdo->o_valid);
730 dof.dof_type = dt_mode_to_dft(attr->la_mode);
732 rc = out_tx_create(env, dt_obj, attr, NULL, &dof,
738 static int update_recovery_destroy(const struct lu_env *env,
739 struct dt_object *dt_obj,
740 const struct update_op *op,
741 const struct update_params *params,
742 struct thandle_exec_args *ta,
748 rc = out_tx_destroy(env, dt_obj, ta, th, NULL, 0);
753 static int update_recovery_ref_add(const struct lu_env *env,
754 struct dt_object *dt_obj,
755 const struct update_op *op,
756 const struct update_params *params,
757 struct thandle_exec_args *ta,
763 rc = out_tx_ref_add(env, dt_obj, ta, th, NULL, 0);
768 static int update_recovery_ref_del(const struct lu_env *env,
769 struct dt_object *dt_obj,
770 const struct update_op *op,
771 const struct update_params *params,
772 struct thandle_exec_args *ta,
778 rc = out_tx_ref_del(env, dt_obj, ta, th, NULL, 0);
783 static int update_recovery_attr_set(const struct lu_env *env,
784 struct dt_object *dt_obj,
785 const struct update_op *op,
786 const struct update_params *params,
787 struct thandle_exec_args *ta,
790 struct update_thread_info *uti = update_env_info(env);
791 struct llog_update_record *lur = uti->uti_dtrq->dtrq_lur;
793 struct obdo *lobdo = &uti->uti_obdo;
794 struct lu_attr *attr = &uti->uti_attr;
796 unsigned int param_count;
800 param_count = lur->lur_update_rec.ur_param_count;
801 wobdo = update_params_get_param_buf(params, op->uop_params_off[0],
805 if (size != sizeof(*wobdo))
808 if (LLOG_REC_HDR_NEEDS_SWABBING(&lur->lur_hdr))
809 lustre_swab_obdo(wobdo);
811 lustre_get_wire_obdo(NULL, lobdo, wobdo);
812 la_from_obdo(attr, lobdo, lobdo->o_valid);
814 rc = out_tx_attr_set(env, dt_obj, attr, ta, th, NULL, 0);
819 static int update_recovery_xattr_set(const struct lu_env *env,
820 struct dt_object *dt_obj,
821 const struct update_op *op,
822 const struct update_params *params,
823 struct thandle_exec_args *ta,
826 struct update_thread_info *uti = update_env_info(env);
835 param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
836 name = update_params_get_param_buf(params,
837 op->uop_params_off[0],
842 buf = update_params_get_param_buf(params,
843 op->uop_params_off[1],
848 uti->uti_buf.lb_buf = buf;
849 uti->uti_buf.lb_len = (size_t)size;
851 buf = update_params_get_param_buf(params, op->uop_params_off[2],
855 if (size != sizeof(fl))
858 fl = le32_to_cpu(*(int *)buf);
860 rc = out_tx_xattr_set(env, dt_obj, &uti->uti_buf, name, fl, ta, th,
866 static int update_recovery_index_insert(const struct lu_env *env,
867 struct dt_object *dt_obj,
868 const struct update_op *op,
869 const struct update_params *params,
870 struct thandle_exec_args *ta,
873 struct update_thread_info *uti = update_env_info(env);
883 param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
884 name = update_params_get_param_buf(params, op->uop_params_off[0],
889 fid = update_params_get_param_buf(params, op->uop_params_off[1],
893 if (size != sizeof(*fid))
896 fid_le_to_cpu(fid, fid);
898 ptype = update_params_get_param_buf(params, op->uop_params_off[2],
902 if (size != sizeof(*ptype))
904 type = le32_to_cpu(*ptype);
906 if (dt_try_as_dir(env, dt_obj) == 0)
909 uti->uti_rec.rec_fid = fid;
910 uti->uti_rec.rec_type = type;
912 rc = out_tx_index_insert(env, dt_obj,
913 (const struct dt_rec *)&uti->uti_rec,
914 (const struct dt_key *)name, ta, th,
920 static int update_recovery_index_delete(const struct lu_env *env,
921 struct dt_object *dt_obj,
922 const struct update_op *op,
923 const struct update_params *params,
924 struct thandle_exec_args *ta,
927 struct update_thread_info *uti = update_env_info(env);
934 param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
935 name = update_params_get_param_buf(params, op->uop_params_off[0],
940 if (dt_try_as_dir(env, dt_obj) == 0)
943 rc = out_tx_index_delete(env, dt_obj,
944 (const struct dt_key *)name, ta, th, NULL, 0);
949 static int update_recovery_write(const struct lu_env *env,
950 struct dt_object *dt_obj,
951 const struct update_op *op,
952 const struct update_params *params,
953 struct thandle_exec_args *ta,
956 struct update_thread_info *uti = update_env_info(env);
964 param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
965 buf = update_params_get_param_buf(params, op->uop_params_off[0],
970 uti->uti_buf.lb_buf = buf;
971 uti->uti_buf.lb_len = size;
973 buf = update_params_get_param_buf(params, op->uop_params_off[1],
978 pos = le64_to_cpu(*(__u64 *)buf);
980 rc = out_tx_write(env, dt_obj, &uti->uti_buf, pos,
986 static int update_recovery_xattr_del(const struct lu_env *env,
987 struct dt_object *dt_obj,
988 const struct update_op *op,
989 const struct update_params *params,
990 struct thandle_exec_args *ta,
993 struct update_thread_info *uti = update_env_info(env);
1000 param_count = uti->uti_dtrq->dtrq_lur->lur_update_rec.ur_param_count;
1001 name = update_params_get_param_buf(params, op->uop_params_off[0],
1002 param_count, &size);
1006 rc = out_tx_xattr_del(env, dt_obj, name, ta, th, NULL, 0);
1012 * Execute updates in the update replay records
1014 * Declare distribute txn replay by update records and add the updates
1015 * to the execution list. Note: it will check if the update has been
1016 * committed, and only execute the updates if it is not committed to
1019 * \param[in] env execution environment
1020 * \param[in] tdtd distribute txn replay data which hold all of replay
1021 * reqs and all replay parameters.
1022 * \param[in] dtrq distribute transaction replay req.
1023 * \param[in] ta thandle execute args.
1025 * \retval 0 if declare succeeds.
1026 * \retval negative errno if declare fails.
1028 static int update_recovery_exec(const struct lu_env *env,
1029 struct target_distribute_txn_data *tdtd,
1030 struct distribute_txn_replay_req *dtrq,
1031 struct thandle_exec_args *ta)
1033 struct llog_update_record *lur = dtrq->dtrq_lur;
1034 struct update_records *records = &lur->lur_update_rec;
1035 struct update_ops *ops = &records->ur_ops;
1036 struct update_params *params = update_records_get_params(records);
1037 struct top_thandle *top_th = container_of(ta->ta_handle,
1040 struct top_multiple_thandle *tmt = top_th->tt_multiple_thandle;
1041 struct update_op *op;
1046 /* These records have been swabbed in llog_cat_process() */
1047 for (i = 0, op = &ops->uops_op[0]; i < records->ur_update_count;
1048 i++, op = update_op_next_op(op)) {
1049 struct lu_fid *fid = &op->uop_fid;
1050 struct dt_object *dt_obj;
1051 struct dt_object *sub_dt_obj;
1052 struct dt_device *sub_dt;
1053 struct sub_thandle *st;
1055 if (op->uop_type == OUT_NOOP)
1058 dt_obj = dt_locate(env, tdtd->tdtd_dt, fid);
1059 if (IS_ERR(dt_obj)) {
1060 rc = PTR_ERR(dt_obj);
1063 sub_dt_obj = dt_object_child(dt_obj);
1065 /* Create sub thandle if not */
1066 sub_dt = lu2dt_dev(sub_dt_obj->do_lu.lo_dev);
1067 st = lookup_sub_thandle(tmt, sub_dt);
1069 st = create_sub_thandle(tmt, sub_dt);
1071 GOTO(next, rc = PTR_ERR(st));
1074 /* check if updates on the OSD/OSP are committed */
1075 rc = update_is_committed(env, dtrq, dt_obj, top_th, st);
1077 /* If this is committed, goto next */
1083 /* Create thandle for sub thandle if needed */
1084 if (st->st_sub_th == NULL) {
1085 rc = sub_thandle_trans_create(env, top_th, st);
1090 CDEBUG(D_HA, "replay %uth update\n", i);
1091 switch (op->uop_type) {
1093 rc = update_recovery_create(env, sub_dt_obj,
1098 rc = update_recovery_destroy(env, sub_dt_obj,
1103 rc = update_recovery_ref_add(env, sub_dt_obj,
1108 rc = update_recovery_ref_del(env, sub_dt_obj,
1113 rc = update_recovery_attr_set(env, sub_dt_obj,
1118 rc = update_recovery_xattr_set(env, sub_dt_obj,
1122 case OUT_INDEX_INSERT:
1123 rc = update_recovery_index_insert(env, sub_dt_obj,
1127 case OUT_INDEX_DELETE:
1128 rc = update_recovery_index_delete(env, sub_dt_obj,
1133 rc = update_recovery_write(env, sub_dt_obj,
1138 rc = update_recovery_xattr_del(env, sub_dt_obj,
1143 CERROR("Unknown update type %u\n", (__u32)op->uop_type);
1148 lu_object_put(env, &dt_obj->do_lu);
1153 ta->ta_handle->th_result = rc;
1158 * redo updates on MDT if needed.
1160 * During DNE recovery, the recovery thread (target_recovery_thread) will call
1161 * this function to replay distribute txn updates on all MDTs. It only replay
1162 * updates on the MDT where the update record is missing.
1164 * If the update already exists on the MDT, then it does not need replay the
1165 * updates on that MDT, and only mark the sub transaction has been committed
1168 * \param[in] env execution environment
1169 * \param[in] tdtd target distribute txn data, which holds the replay list
1170 * and all parameters needed by replay process.
1171 * \param[in] dtrq distribute txn replay req.
1173 * \retval 0 if replay succeeds.
1174 * \retval negative errno if replay failes.
1176 int distribute_txn_replay_handle(struct lu_env *env,
1177 struct target_distribute_txn_data *tdtd,
1178 struct distribute_txn_replay_req *dtrq)
1180 struct update_records *records = &dtrq->dtrq_lur->lur_update_rec;
1181 struct thandle_exec_args *ta;
1182 struct lu_context session_env;
1183 struct thandle *th = NULL;
1184 struct top_thandle *top_th;
1185 struct top_multiple_thandle *tmt;
1186 struct thandle_update_records *tur = NULL;
1191 /* initialize session, it is needed for the handler of target */
1192 rc = lu_context_init(&session_env, LCT_SERVER_SESSION | LCT_NOREF);
1194 CERROR("%s: failure to initialize session: rc = %d\n",
1195 tdtd->tdtd_lut->lut_obd->obd_name, rc);
1198 lu_context_enter(&session_env);
1199 env->le_ses = &session_env;
1201 update_records_dump(records, D_HA, true);
1202 th = top_trans_create(env, NULL);
1204 GOTO(exit_session, rc = PTR_ERR(th));
1206 ta = &update_env_info(env)->uti_tea;
1209 update_env_info(env)->uti_dtrq = dtrq;
1210 /* Create distribute transaction structure for this top thandle */
1211 top_th = container_of(th, struct top_thandle, tt_super);
1212 rc = top_trans_create_tmt(env, top_th);
1214 GOTO(stop_trans, rc);
1218 /* check if the distribute transaction has been committed */
1219 tmt = top_th->tt_multiple_thandle;
1220 tmt->tmt_master_sub_dt = tdtd->tdtd_lut->lut_bottom;
1221 tmt->tmt_batchid = dtrq->dtrq_batchid;
1222 tgt_th_info(env)->tti_transno = dtrq->dtrq_master_transno;
1224 if (tmt->tmt_batchid <= tdtd->tdtd_committed_batchid)
1225 tmt->tmt_committed = 1;
1227 rc = update_recovery_exec(env, tdtd, dtrq, ta);
1229 GOTO(stop_trans, rc);
1231 /* If no updates are needed to be replayed, then
1232 * mark this records as committed, so commit thread
1233 * distribute_txn_commit_thread() will delete the
1235 if (ta->ta_argno == 0)
1236 tmt->tmt_committed = 1;
1238 tur = &update_env_info(env)->uti_tur;
1239 tur->tur_update_records = dtrq->dtrq_lur;
1240 tur->tur_update_records_buf_size = dtrq->dtrq_lur_size;
1241 tur->tur_update_params = NULL;
1242 tur->tur_update_param_count = 0;
1243 tmt->tmt_update_records = tur;
1245 distribute_txn_insert_by_batchid(tmt);
1246 rc = top_trans_start(env, NULL, th);
1248 GOTO(stop_trans, rc);
1250 for (i = 0; i < ta->ta_argno; i++) {
1251 struct tx_arg *ta_arg;
1252 struct dt_object *dt_obj;
1253 struct dt_device *sub_dt;
1254 struct sub_thandle *st;
1256 ta_arg = ta->ta_args[i];
1257 dt_obj = ta_arg->object;
1259 LASSERT(tmt->tmt_committed == 0);
1260 sub_dt = lu2dt_dev(dt_obj->do_lu.lo_dev);
1261 st = lookup_sub_thandle(tmt, sub_dt);
1262 LASSERT(st != NULL);
1263 LASSERT(st->st_sub_th != NULL);
1264 rc = ta->ta_args[i]->exec_fn(env, st->st_sub_th,
1266 if (unlikely(rc < 0)) {
1267 CDEBUG(D_HA, "error during execution of #%u from"
1268 " %s:%d: rc = %d\n", i, ta->ta_args[i]->file,
1269 ta->ta_args[i]->line, rc);
1271 if (ta->ta_args[i]->undo_fn != NULL) {
1272 dt_obj = ta->ta_args[i]->object;
1274 lu2dt_dev(dt_obj->do_lu.lo_dev);
1275 st = lookup_sub_thandle(tmt, sub_dt);
1276 LASSERT(st != NULL);
1277 LASSERT(st->st_sub_th != NULL);
1279 ta->ta_args[i]->undo_fn(env,
1283 CERROR("%s: undo for %s:%d: rc = %d\n",
1284 dt_obd_name(ta->ta_handle->th_dev),
1285 ta->ta_args[i]->file,
1286 ta->ta_args[i]->line, -ENOTSUPP);
1291 CDEBUG(D_HA, "%s: executed %u/%u: rc = %d\n",
1292 dt_obd_name(sub_dt), i, ta->ta_argno, rc);
1298 rc = top_trans_stop(env, tdtd->tdtd_dt, th);
1299 for (i = 0; i < ta->ta_argno; i++) {
1300 if (ta->ta_args[i]->object != NULL) {
1301 lu_object_put(env, &ta->ta_args[i]->object->do_lu);
1302 ta->ta_args[i]->object = NULL;
1307 tur->tur_update_records = NULL;
1309 lu_context_exit(&session_env);
1310 lu_context_fini(&session_env);
1313 EXPORT_SYMBOL(distribute_txn_replay_handle);