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,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
25 * Copyright (c) 2016, 2017, Intel Corporation.
30 * lustre/mdt/mdt_hsm_cdt_agent.c
32 * Lustre HSM Coordinator
34 * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
35 * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
38 #define DEBUG_SUBSYSTEM S_MDS
41 #include <obd_support.h>
42 #include <lustre_export.h>
43 #include <lprocfs_status.h>
44 #include <lustre_kernelcomm.h>
45 #include "mdt_internal.h"
52 * find a hsm_agent by uuid
53 * lock cdt_agent_lock needs to be held by caller
54 * \param cdt [IN] coordinator
55 * \param uuid [IN] agent UUID
56 * \retval hsm_agent pointer or NULL if not found
58 static struct hsm_agent *mdt_hsm_agent_lookup(struct coordinator *cdt,
59 const struct obd_uuid *uuid)
63 list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
64 if (obd_uuid_equals(&ha->ha_uuid, uuid))
71 * register a copy tool
72 * \param mti [IN] MDT context
73 * \param uuid [IN] client UUID to be registered
74 * \param count [IN] number of archives agent serves
75 * \param archive_id [IN] vector of archive number served by the copytool
79 int mdt_hsm_agent_register(struct mdt_thread_info *mti,
80 const struct obd_uuid *uuid,
81 int nr_archives, __u32 *archive_id)
83 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
84 struct hsm_agent *ha, *tmp;
88 /* no coordinator started, so we cannot serve requests */
89 if (cdt->cdt_state == CDT_STOPPED) {
90 LCONSOLE_WARN("HSM coordinator thread is not running - "
91 "denying agent registration.\n");
97 GOTO(out, rc = -ENOMEM);
100 ha->ha_archive_cnt = nr_archives;
101 if (ha->ha_archive_cnt != 0) {
104 sz = ha->ha_archive_cnt * sizeof(*ha->ha_archive_id);
105 OBD_ALLOC(ha->ha_archive_id, sz);
106 if (ha->ha_archive_id == NULL)
107 GOTO(out_free, rc = -ENOMEM);
108 memcpy(ha->ha_archive_id, archive_id, sz);
110 atomic_set(&ha->ha_requests, 0);
111 atomic_set(&ha->ha_success, 0);
112 atomic_set(&ha->ha_failure, 0);
114 down_write(&cdt->cdt_agent_lock);
115 tmp = mdt_hsm_agent_lookup(cdt, uuid);
117 LCONSOLE_WARN("HSM agent %s already registered\n",
119 up_write(&cdt->cdt_agent_lock);
120 GOTO(out_free, rc = -EEXIST);
123 list_add_tail(&ha->ha_list, &cdt->cdt_agents);
125 if (ha->ha_archive_cnt == 0)
126 CDEBUG(D_HSM, "agent %s registered for all archives\n",
127 obd_uuid2str(&ha->ha_uuid));
129 CDEBUG(D_HSM, "agent %s registered for %d archives\n",
130 obd_uuid2str(&ha->ha_uuid), ha->ha_archive_cnt);
132 up_write(&cdt->cdt_agent_lock);
137 if (ha != NULL && ha->ha_archive_id != NULL)
138 OBD_FREE_PTR_ARRAY(ha->ha_archive_id, ha->ha_archive_cnt);
142 /* wake the coordinator to potentially schedule requests */
143 if (rc == -EEXIST || rc == 0)
144 mdt_hsm_cdt_event(cdt);
150 * register a copy tool
151 * \param mti [IN] MDT context
152 * \param uuid [IN] uuid to be registered
153 * \param archive_mask [IN] bitmask of archive number served by the copytool
155 * \retval -ve failure
157 int mdt_hsm_agent_register_mask(struct mdt_thread_info *mti,
158 const struct obd_uuid *uuid, __u32 archive_mask)
160 int rc, i, nr_archives = 0;
161 __u32 *archive_id = NULL;
164 nr_archives = hweight32(archive_mask);
166 if (nr_archives != 0) {
167 OBD_ALLOC_PTR_ARRAY(archive_id, nr_archives);
172 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
173 if (BIT(i) & archive_mask) {
174 archive_id[nr_archives] = i + 1;
180 rc = mdt_hsm_agent_register(mti, uuid, nr_archives, archive_id);
182 if (archive_id != NULL)
183 OBD_FREE_PTR_ARRAY(archive_id, nr_archives);
189 * unregister a copy tool
190 * \param mti [IN] MDT context
191 * \param uuid [IN] uuid to be unregistered
193 * \retval -ve failure
195 int mdt_hsm_agent_unregister(struct mdt_thread_info *mti,
196 const struct obd_uuid *uuid)
198 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
199 struct hsm_agent *ha;
203 /* no coordinator started, so we cannot serve requests */
204 if (cdt->cdt_state == CDT_STOPPED)
207 down_write(&cdt->cdt_agent_lock);
209 ha = mdt_hsm_agent_lookup(cdt, uuid);
211 list_del_init(&ha->ha_list);
213 up_write(&cdt->cdt_agent_lock);
216 GOTO(out, rc = -ENOENT);
218 if (ha->ha_archive_cnt != 0)
219 OBD_FREE_PTR_ARRAY(ha->ha_archive_id, ha->ha_archive_cnt);
224 CDEBUG(D_HSM, "agent %s unregistration: %d\n", obd_uuid2str(uuid), rc);
230 * update agent statistics
231 * \param mdt [IN] MDT device
232 * \param succ_rq [IN] number of success
233 * \param fail_rq [IN] number of failure
234 * \param new_rq [IN] number of new requests
235 * \param uuid [IN] agent uuid
236 * if all counters == 0, clear counters
238 * \retval -ve failure
240 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
241 int succ_rq, int fail_rq, int new_rq,
242 const struct obd_uuid *uuid)
244 struct hsm_agent *ha;
248 down_read(&cdt->cdt_agent_lock);
249 list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
250 if (obd_uuid_equals(&ha->ha_uuid, uuid)) {
251 if (succ_rq == 0 && fail_rq == 0 && new_rq == 0) {
252 atomic_set(&ha->ha_success, 0);
253 atomic_set(&ha->ha_failure, 0);
254 atomic_set(&ha->ha_requests, 0);
256 atomic_add(succ_rq, &ha->ha_success);
257 atomic_add(fail_rq, &ha->ha_failure);
258 atomic_add(new_rq, &ha->ha_requests);
259 atomic_sub(succ_rq, &ha->ha_requests);
260 atomic_sub(fail_rq, &ha->ha_requests);
268 up_read(&cdt->cdt_agent_lock);
273 * find the best agent
274 * \param cdt [IN] coordinator
275 * \param archive [IN] archive number
276 * \param uuid [OUT] agent who can serve archive
278 * \retval -ve failure
280 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
281 struct obd_uuid *uuid)
283 int rc = -EAGAIN, i, load = -1;
284 struct hsm_agent *ha;
287 /* Choose an export to send a copytool req to */
288 down_read(&cdt->cdt_agent_lock);
289 list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
290 for (i = 0; (i < ha->ha_archive_cnt) &&
291 (ha->ha_archive_id[i] != archive); i++) {
292 /* nothing to do, just skip unmatching records */
295 /* archive count == 0 means copy tool serves any backend */
296 if (ha->ha_archive_cnt != 0 && i == ha->ha_archive_cnt)
299 if (load == -1 || load > atomic_read(&ha->ha_requests)) {
300 load = atomic_read(&ha->ha_requests);
304 if (atomic_read(&ha->ha_requests) == 0)
307 up_read(&cdt->cdt_agent_lock);
312 int mdt_hsm_send_action_to_each_archive(struct mdt_thread_info *mti,
313 struct hsm_action_item *hai)
315 struct hsm_agent *ha;
316 __u32 archive_mask = 0;
317 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
319 /* return error by default in case all archive_ids have unregistered */
323 /* send action to all registered archive_ids */
324 down_read(&cdt->cdt_agent_lock);
325 list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
326 for (i = 0; (i < ha->ha_archive_cnt); i++) {
327 /* only send once for each archive_id */
328 if (BIT(ha->ha_archive_id[i]) & archive_mask)
330 archive_mask |= BIT(ha->ha_archive_id[i]);
332 /* XXX: it could make sense to gather all
333 * actions for the same archive_id like in
334 * mdt_hsm_add_actions() ?? */
335 rc = mdt_agent_record_add(mti->mti_env, mti->mti_mdt,
336 ha->ha_archive_id[i], 0,
339 CERROR("%s: unable to add HSM remove request "
340 "for "DFID": rc=%d\n",
341 mdt_obd_name(mti->mti_mdt),
342 PFID(&hai->hai_fid), rc);
345 CDEBUG(D_HSM, "%s: added HSM remove request "
346 "for "DFID", archive_id=%d\n",
347 mdt_obd_name(mti->mti_mdt),
349 ha->ha_archive_id[i]);
352 /* early exit from loop due to error? */
353 if (i != ha->ha_archive_cnt)
356 up_read(&cdt->cdt_agent_lock);
362 * send a HAL to the agent
363 * \param mti [IN] context
364 * \param hal [IN] request (can be a kuc payload)
365 * \param purge [IN] purge mode (no record)
367 * \retval -ve failure
368 * This function supposes:
369 * - all actions are for the same archive number
370 * - in case of cancel, all cancel are for the same agent
371 * This implies that request split has to be done
372 * before when building the hal
374 int mdt_hsm_agent_send(struct mdt_thread_info *mti,
375 struct hsm_action_list *hal, bool purge)
377 struct obd_export *exp;
378 struct mdt_device *mdt = mti->mti_mdt;
379 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
380 struct hsm_action_list *buf = NULL;
381 struct hsm_action_item *hai;
382 struct obd_uuid uuid;
385 bool is_registered = false;
388 rc = mdt_hsm_find_best_agent(cdt, hal->hal_archive_id, &uuid);
389 if (rc && hal->hal_archive_id == 0) {
393 /* special case of remove requests with no archive_id specified,
394 * and no agent registered to serve all archives, then create a
395 * set of new requests, each to be sent to each registered
397 * Todo so, find all HSMA_REMOVE entries, and then :
398 * _ set completed status as SUCCESS (or FAIL?)
399 * _ create a new LLOG record for each archive_id
400 * presently being served by any CT
402 hai = hai_first(hal);
403 for (i = 0; i < hal->hal_count; i++,
404 hai = hai_next(hai)) {
405 struct hsm_record_update update;
407 /* only removes are concerned */
408 if (hai->hai_action != HSMA_REMOVE) {
409 /* count if other actions than HSMA_REMOVE,
410 * to return original error/rc */
415 /* send remove request to all registered archive_ids */
416 rc2 = mdt_hsm_send_action_to_each_archive(mti, hai);
420 /* only update original request as SUCCEED if it has
421 * been successfully broadcasted to all available
423 * XXX: this should only cause duplicates to be sent,
424 * unless a method to record already successfully
425 * reached archive_ids is implemented */
427 update.cookie = hai->hai_cookie;
428 update.status = ARS_SUCCEED;
429 rc2 = mdt_agent_record_update(mti, &update, 1);
431 CERROR("%s: mdt_agent_record_update() "
432 "failed, cannot update "
433 "status to %s for cookie "
436 agent_req_status2name(ARS_SUCCEED),
437 hai->hai_cookie, rc2);
441 /* only remove requests with archive_id=0 */
448 CERROR("%s: Cannot find agent for archive %d: rc = %d\n",
449 mdt_obd_name(mdt), hal->hal_archive_id, rc);
453 CDEBUG(D_HSM, "Agent %s selected for archive %d\n", obd_uuid2str(&uuid),
454 hal->hal_archive_id);
457 buf = kuc_alloc(len, KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
459 RETURN(PTR_ERR(buf));
460 memcpy(buf, hal, len);
462 /* Check if request is still valid (cf file hsm flags) */
463 fail_request = false;
464 hai = hai_first(hal);
465 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
466 struct mdt_object *obj;
469 if (hai->hai_action == HSMA_CANCEL)
472 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
474 mdt_object_put(mti->mti_env, obj);
475 } else if (PTR_ERR(obj) == -ENOENT) {
476 struct hsm_record_update update = {
477 .cookie = hai->hai_cookie,
478 .status = ARS_FAILED,
481 if (hai->hai_action == HSMA_REMOVE)
485 rc = mdt_agent_record_update(mti, &update, 1);
487 CERROR("%s: mdt_agent_record_update() failed, "
488 "cannot update status to %s for cookie "
491 agent_req_status2name(ARS_FAILED),
492 hai->hai_cookie, rc);
498 GOTO(out_buf, rc = PTR_ERR(obj));
501 if (!mdt_hsm_is_action_compat(hai, hal->hal_archive_id,
502 hal->hal_flags, &hsm)) {
503 struct hsm_record_update update = {
504 .cookie = hai->hai_cookie,
505 .status = ARS_FAILED,
508 /* incompatible request, we abort the request */
509 /* next time coordinator will wake up, it will
510 * make the same HAL with valid only
513 rc = mdt_agent_record_update(mti, &update, 1);
515 CERROR("%s: mdt_agent_record_update() failed, "
516 "cannot update status to %s for cookie "
519 agent_req_status2name(ARS_FAILED),
520 hai->hai_cookie, rc);
524 /* if restore and record status updated, give
525 * back granted layout lock */
526 if (hai->hai_action == HSMA_RESTORE)
527 cdt_restore_handle_del(mti, cdt, &hai->hai_fid);
531 /* we found incompatible requests, so the HAL cannot be sent
532 * as is. Bad records have been invalidated in llog.
533 * Valid one will be reschedule next time coordinator will wake up
534 * So no need the rebuild a full valid HAL now
537 GOTO(out_buf, rc = -EAGAIN);
539 /* Cancel memory registration is useless for purge
540 * non registration avoid a deadlock :
541 * in case of failure we have to take the write lock
542 * to remove entry which conflict with the read loack needed
546 /* set is_registered even if failure because we may have
547 * partial work done */
548 is_registered = true;
549 rc = mdt_hsm_add_hal(mti, hal, &uuid);
554 /* Uses the ldlm reverse import; this rpc will be seen by
555 * the ldlm_callback_handler. Note this sends a request RPC
556 * from a server (MDT) to a client (MDC), backwards of normal comms.
558 exp = obd_uuid_lookup(mdt2obd_dev(mdt), &uuid);
559 if (exp == NULL || exp->exp_disconnected) {
561 class_export_put(exp);
562 /* This should clean up agents on evicted exports */
564 CERROR("%s: agent uuid (%s) not found, unregistering:"
566 mdt_obd_name(mdt), obd_uuid2str(&uuid), rc);
567 mdt_hsm_agent_unregister(mti, &uuid);
571 /* send request to agent */
572 rc = do_set_info_async(exp->exp_imp_reverse, LDLM_SET_INFO,
574 sizeof(KEY_HSM_COPYTOOL_SEND),
575 KEY_HSM_COPYTOOL_SEND,
576 kuc_len(len), kuc_ptr(buf), NULL);
579 CERROR("%s: cannot send request to agent '%s': rc = %d\n",
580 mdt_obd_name(mdt), obd_uuid2str(&uuid), rc);
582 class_export_put(exp);
585 CDEBUG(D_HSM, "Lost connection to agent '%s', unregistering\n",
586 obd_uuid2str(&uuid));
587 mdt_hsm_agent_unregister(mti, &uuid);
591 if (rc != 0 && is_registered) {
592 /* in case of error, we have to unregister requests */
593 hai = hai_first(hal);
594 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
595 if (hai->hai_action == HSMA_CANCEL)
597 mdt_cdt_remove_request(cdt, hai->hai_cookie);
608 * seq_file method called to start access to debugfs file
610 static void *mdt_hsm_agent_debugfs_start(struct seq_file *s, loff_t *off)
612 struct mdt_device *mdt = s->private;
613 struct coordinator *cdt = &mdt->mdt_coordinator;
614 struct list_head *pos;
618 down_read(&cdt->cdt_agent_lock);
620 if (list_empty(&cdt->cdt_agents))
624 RETURN(SEQ_START_TOKEN);
627 list_for_each(pos, &cdt->cdt_agents) {
637 * seq_file method called to get next item
638 * just returns NULL at eof
640 static void *mdt_hsm_agent_debugfs_next(struct seq_file *s, void *v, loff_t *p)
642 struct mdt_device *mdt = s->private;
643 struct coordinator *cdt = &mdt->mdt_coordinator;
644 struct list_head *pos = v;
647 if (pos == SEQ_START_TOKEN)
648 pos = cdt->cdt_agents.next;
653 if (pos != &cdt->cdt_agents)
661 static int mdt_hsm_agent_debugfs_show(struct seq_file *s, void *v)
663 struct list_head *pos = v;
664 struct hsm_agent *ha;
668 if (pos == SEQ_START_TOKEN)
671 ha = list_entry(pos, struct hsm_agent, ha_list);
672 seq_printf(s, "uuid=%s archive_id=", ha->ha_uuid.uuid);
673 if (ha->ha_archive_cnt == 0) {
674 seq_printf(s, "ANY");
676 seq_printf(s, "%d", ha->ha_archive_id[0]);
677 for (i = 1; i < ha->ha_archive_cnt; i++)
678 seq_printf(s, ",%d", ha->ha_archive_id[i]);
681 seq_printf(s, " requests=[current:%d ok:%d errors:%d]\n",
682 atomic_read(&ha->ha_requests),
683 atomic_read(&ha->ha_success),
684 atomic_read(&ha->ha_failure));
689 * seq_file method called to stop access to debugfs file
691 static void mdt_hsm_agent_debugfs_stop(struct seq_file *s, void *v)
693 struct mdt_device *mdt = s->private;
694 struct coordinator *cdt = &mdt->mdt_coordinator;
696 up_read(&cdt->cdt_agent_lock);
699 /* hsm agent list debugfs functions */
700 static const struct seq_operations mdt_hsm_agent_debugfs_ops = {
701 .start = mdt_hsm_agent_debugfs_start,
702 .next = mdt_hsm_agent_debugfs_next,
703 .show = mdt_hsm_agent_debugfs_show,
704 .stop = mdt_hsm_agent_debugfs_stop,
708 * public function called at open of debugfs file to get
711 static int ldebugfs_open_hsm_agent(struct inode *inode, struct file *file)
717 rc = seq_open(file, &mdt_hsm_agent_debugfs_ops);
721 s = file->private_data;
722 s->private = inode->i_private;
727 /* methods to access hsm agent list */
728 const struct file_operations mdt_hsm_agent_fops = {
729 .owner = THIS_MODULE,
730 .open = ldebugfs_open_hsm_agent,
733 .release = seq_release,