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(ha->ha_archive_id,
139 ha->ha_archive_cnt * sizeof(*ha->ha_archive_id));
143 /* wake the coordinator to potentially schedule requests */
144 if (rc == -EEXIST || rc == 0)
145 mdt_hsm_cdt_event(cdt);
151 * register a copy tool
152 * \param mti [IN] MDT context
153 * \param uuid [IN] uuid to be registered
154 * \param archive_mask [IN] bitmask of archive number served by the copytool
156 * \retval -ve failure
158 int mdt_hsm_agent_register_mask(struct mdt_thread_info *mti,
159 const struct obd_uuid *uuid, __u32 archive_mask)
161 int rc, i, nr_archives = 0;
162 __u32 *archive_id = NULL;
165 nr_archives = hweight32(archive_mask);
167 if (nr_archives != 0) {
168 OBD_ALLOC(archive_id, nr_archives * sizeof(*archive_id));
173 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
174 if ((1 << i) & archive_mask) {
175 archive_id[nr_archives] = i + 1;
181 rc = mdt_hsm_agent_register(mti, uuid, nr_archives, archive_id);
183 if (archive_id != NULL)
184 OBD_FREE(archive_id, nr_archives * sizeof(*archive_id));
190 * unregister a copy tool
191 * \param mti [IN] MDT context
192 * \param uuid [IN] uuid to be unregistered
194 * \retval -ve failure
196 int mdt_hsm_agent_unregister(struct mdt_thread_info *mti,
197 const struct obd_uuid *uuid)
199 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
200 struct hsm_agent *ha;
204 /* no coordinator started, so we cannot serve requests */
205 if (cdt->cdt_state == CDT_STOPPED)
208 down_write(&cdt->cdt_agent_lock);
210 ha = mdt_hsm_agent_lookup(cdt, uuid);
212 list_del_init(&ha->ha_list);
214 up_write(&cdt->cdt_agent_lock);
217 GOTO(out, rc = -ENOENT);
219 if (ha->ha_archive_cnt != 0)
220 OBD_FREE(ha->ha_archive_id,
221 ha->ha_archive_cnt * sizeof(*ha->ha_archive_id));
226 CDEBUG(D_HSM, "agent %s unregistration: %d\n", obd_uuid2str(uuid), rc);
232 * update agent statistics
233 * \param mdt [IN] MDT device
234 * \param succ_rq [IN] number of success
235 * \param fail_rq [IN] number of failure
236 * \param new_rq [IN] number of new requests
237 * \param uuid [IN] agent uuid
238 * if all counters == 0, clear counters
240 * \retval -ve failure
242 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
243 int succ_rq, int fail_rq, int new_rq,
244 const struct obd_uuid *uuid)
246 struct hsm_agent *ha;
250 down_read(&cdt->cdt_agent_lock);
251 list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
252 if (obd_uuid_equals(&ha->ha_uuid, uuid)) {
253 if (succ_rq == 0 && fail_rq == 0 && new_rq == 0) {
254 atomic_set(&ha->ha_success, 0);
255 atomic_set(&ha->ha_failure, 0);
256 atomic_set(&ha->ha_requests, 0);
258 atomic_add(succ_rq, &ha->ha_success);
259 atomic_add(fail_rq, &ha->ha_failure);
260 atomic_add(new_rq, &ha->ha_requests);
261 atomic_sub(succ_rq, &ha->ha_requests);
262 atomic_sub(fail_rq, &ha->ha_requests);
270 up_read(&cdt->cdt_agent_lock);
275 * find the best agent
276 * \param cdt [IN] coordinator
277 * \param archive [IN] archive number
278 * \param uuid [OUT] agent who can serve archive
280 * \retval -ve failure
282 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
283 struct obd_uuid *uuid)
285 int rc = -EAGAIN, i, load = -1;
286 struct hsm_agent *ha;
289 /* Choose an export to send a copytool req to */
290 down_read(&cdt->cdt_agent_lock);
291 list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
292 for (i = 0; (i < ha->ha_archive_cnt) &&
293 (ha->ha_archive_id[i] != archive); i++) {
294 /* nothing to do, just skip unmatching records */
297 /* archive count == 0 means copy tool serves any backend */
298 if (ha->ha_archive_cnt != 0 && i == ha->ha_archive_cnt)
301 if (load == -1 || load > atomic_read(&ha->ha_requests)) {
302 load = atomic_read(&ha->ha_requests);
306 if (atomic_read(&ha->ha_requests) == 0)
309 up_read(&cdt->cdt_agent_lock);
314 int mdt_hsm_send_action_to_each_archive(struct mdt_thread_info *mti,
315 struct hsm_action_item *hai)
317 struct hsm_agent *ha;
318 __u32 archive_mask = 0;
319 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
321 /* return error by default in case all archive_ids have unregistered */
325 /* send action to all registered archive_ids */
326 down_read(&cdt->cdt_agent_lock);
327 list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
328 for (i = 0; (i < ha->ha_archive_cnt); i++) {
329 /* only send once for each archive_id */
330 if ((1 << ha->ha_archive_id[i]) & archive_mask)
332 archive_mask |= (1 << ha->ha_archive_id[i]);
334 /* XXX: it could make sense to gather all
335 * actions for the same archive_id like in
336 * mdt_hsm_add_actions() ?? */
337 rc = mdt_agent_record_add(mti->mti_env, mti->mti_mdt,
338 ha->ha_archive_id[i], 0,
341 CERROR("%s: unable to add HSM remove request "
342 "for "DFID": rc=%d\n",
343 mdt_obd_name(mti->mti_mdt),
344 PFID(&hai->hai_fid), rc);
347 CDEBUG(D_HSM, "%s: added HSM remove request "
348 "for "DFID", archive_id=%d\n",
349 mdt_obd_name(mti->mti_mdt),
351 ha->ha_archive_id[i]);
354 /* early exit from loop due to error? */
355 if (i != ha->ha_archive_cnt)
358 up_read(&cdt->cdt_agent_lock);
364 * send a HAL to the agent
365 * \param mti [IN] context
366 * \param hal [IN] request (can be a kuc payload)
367 * \param purge [IN] purge mode (no record)
369 * \retval -ve failure
370 * This function supposes:
371 * - all actions are for the same archive number
372 * - in case of cancel, all cancel are for the same agent
373 * This implies that request split has to be done
374 * before when building the hal
376 int mdt_hsm_agent_send(struct mdt_thread_info *mti,
377 struct hsm_action_list *hal, bool purge)
379 struct obd_export *exp;
380 struct mdt_device *mdt = mti->mti_mdt;
381 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
382 struct hsm_action_list *buf = NULL;
383 struct hsm_action_item *hai;
384 struct obd_uuid uuid;
387 bool is_registered = false;
390 rc = mdt_hsm_find_best_agent(cdt, hal->hal_archive_id, &uuid);
391 if (rc && hal->hal_archive_id == 0) {
395 /* special case of remove requests with no archive_id specified,
396 * and no agent registered to serve all archives, then create a
397 * set of new requests, each to be sent to each registered
399 * Todo so, find all HSMA_REMOVE entries, and then :
400 * _ set completed status as SUCCESS (or FAIL?)
401 * _ create a new LLOG record for each archive_id
402 * presently being served by any CT
404 hai = hai_first(hal);
405 for (i = 0; i < hal->hal_count; i++,
406 hai = hai_next(hai)) {
407 struct hsm_record_update update;
409 /* only removes are concerned */
410 if (hai->hai_action != HSMA_REMOVE) {
411 /* count if other actions than HSMA_REMOVE,
412 * to return original error/rc */
417 /* send remove request to all registered archive_ids */
418 rc2 = mdt_hsm_send_action_to_each_archive(mti, hai);
422 /* only update original request as SUCCEED if it has
423 * been successfully broadcasted to all available
425 * XXX: this should only cause duplicates to be sent,
426 * unless a method to record already successfully
427 * reached archive_ids is implemented */
429 update.cookie = hai->hai_cookie;
430 update.status = ARS_SUCCEED;
431 rc2 = mdt_agent_record_update(mti->mti_env, mdt,
434 CERROR("%s: mdt_agent_record_update() "
435 "failed, cannot update "
436 "status to %s for cookie "
439 agent_req_status2name(ARS_SUCCEED),
440 hai->hai_cookie, rc2);
444 /* only remove requests with archive_id=0 */
451 CERROR("%s: Cannot find agent for archive %d: rc = %d\n",
452 mdt_obd_name(mdt), hal->hal_archive_id, rc);
456 CDEBUG(D_HSM, "Agent %s selected for archive %d\n", obd_uuid2str(&uuid),
457 hal->hal_archive_id);
460 buf = kuc_alloc(len, KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
462 RETURN(PTR_ERR(buf));
463 memcpy(buf, hal, len);
465 /* Check if request is still valid (cf file hsm flags) */
466 fail_request = false;
467 hai = hai_first(hal);
468 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
469 struct mdt_object *obj;
472 if (hai->hai_action == HSMA_CANCEL)
475 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
477 mdt_object_put(mti->mti_env, obj);
478 } else if (PTR_ERR(obj) == -ENOENT) {
479 struct hsm_record_update update = {
480 .cookie = hai->hai_cookie,
481 .status = ARS_FAILED,
484 if (hai->hai_action == HSMA_REMOVE)
488 rc = mdt_agent_record_update(mti->mti_env, mdt,
491 CERROR("%s: mdt_agent_record_update() failed, "
492 "cannot update status to %s for cookie "
495 agent_req_status2name(ARS_FAILED),
496 hai->hai_cookie, rc);
502 GOTO(out_buf, rc = PTR_ERR(obj));
505 if (!mdt_hsm_is_action_compat(hai, hal->hal_archive_id,
506 hal->hal_flags, &hsm)) {
507 struct hsm_record_update update = {
508 .cookie = hai->hai_cookie,
509 .status = ARS_FAILED,
512 /* incompatible request, we abort the request */
513 /* next time coordinator will wake up, it will
514 * make the same HAL with valid only
517 rc = mdt_agent_record_update(mti->mti_env, mdt,
520 CERROR("%s: mdt_agent_record_update() failed, "
521 "cannot update status to %s for cookie "
524 agent_req_status2name(ARS_FAILED),
525 hai->hai_cookie, rc);
529 /* if restore and record status updated, give
530 * back granted layout lock */
531 if (hai->hai_action == HSMA_RESTORE)
532 cdt_restore_handle_del(mti, cdt, &hai->hai_fid);
536 /* we found incompatible requests, so the HAL cannot be sent
537 * as is. Bad records have been invalidated in llog.
538 * Valid one will be reschedule next time coordinator will wake up
539 * So no need the rebuild a full valid HAL now
542 GOTO(out_buf, rc = 0);
544 /* Cancel memory registration is useless for purge
545 * non registration avoid a deadlock :
546 * in case of failure we have to take the write lock
547 * to remove entry which conflict with the read loack needed
551 /* set is_registered even if failure because we may have
552 * partial work done */
553 is_registered = true;
554 rc = mdt_hsm_add_hal(mti, hal, &uuid);
559 /* Uses the ldlm reverse import; this rpc will be seen by
560 * the ldlm_callback_handler. Note this sends a request RPC
561 * from a server (MDT) to a client (MDC), backwards of normal comms.
563 exp = cfs_hash_lookup(mdt2obd_dev(mdt)->obd_uuid_hash, &uuid);
564 if (exp == NULL || exp->exp_disconnected) {
566 class_export_put(exp);
567 /* This should clean up agents on evicted exports */
569 CERROR("%s: agent uuid (%s) not found, unregistering:"
571 mdt_obd_name(mdt), obd_uuid2str(&uuid), rc);
572 mdt_hsm_agent_unregister(mti, &uuid);
576 /* send request to agent */
577 rc = do_set_info_async(exp->exp_imp_reverse, LDLM_SET_INFO,
579 sizeof(KEY_HSM_COPYTOOL_SEND),
580 KEY_HSM_COPYTOOL_SEND,
581 kuc_len(len), kuc_ptr(buf), NULL);
584 CERROR("%s: cannot send request to agent '%s': rc = %d\n",
585 mdt_obd_name(mdt), obd_uuid2str(&uuid), rc);
587 class_export_put(exp);
590 CDEBUG(D_HSM, "Lost connection to agent '%s', unregistering\n",
591 obd_uuid2str(&uuid));
592 mdt_hsm_agent_unregister(mti, &uuid);
596 if (rc != 0 && is_registered) {
597 /* in case of error, we have to unregister requests */
598 hai = hai_first(hal);
599 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
600 if (hai->hai_action == HSMA_CANCEL)
602 mdt_cdt_remove_request(cdt, hai->hai_cookie);
613 * seq_file method called to start access to /proc file
615 static void *mdt_hsm_agent_proc_start(struct seq_file *s, loff_t *off)
617 struct mdt_device *mdt = s->private;
618 struct coordinator *cdt = &mdt->mdt_coordinator;
619 struct list_head *pos;
623 down_read(&cdt->cdt_agent_lock);
625 if (list_empty(&cdt->cdt_agents))
629 RETURN(SEQ_START_TOKEN);
632 list_for_each(pos, &cdt->cdt_agents) {
642 * seq_file method called to get next item
643 * just returns NULL at eof
645 static void *mdt_hsm_agent_proc_next(struct seq_file *s, void *v, loff_t *p)
647 struct mdt_device *mdt = s->private;
648 struct coordinator *cdt = &mdt->mdt_coordinator;
649 struct list_head *pos = v;
652 if (pos == SEQ_START_TOKEN)
653 pos = cdt->cdt_agents.next;
658 if (pos != &cdt->cdt_agents)
666 static int mdt_hsm_agent_proc_show(struct seq_file *s, void *v)
668 struct list_head *pos = v;
669 struct hsm_agent *ha;
673 if (pos == SEQ_START_TOKEN)
676 ha = list_entry(pos, struct hsm_agent, ha_list);
677 seq_printf(s, "uuid=%s archive_id=", ha->ha_uuid.uuid);
678 if (ha->ha_archive_cnt == 0) {
679 seq_printf(s, "ANY");
681 seq_printf(s, "%d", ha->ha_archive_id[0]);
682 for (i = 1; i < ha->ha_archive_cnt; i++)
683 seq_printf(s, ",%d", ha->ha_archive_id[i]);
686 seq_printf(s, " requests=[current:%d ok:%d errors:%d]\n",
687 atomic_read(&ha->ha_requests),
688 atomic_read(&ha->ha_success),
689 atomic_read(&ha->ha_failure));
694 * seq_file method called to stop access to /proc file
696 static void mdt_hsm_agent_proc_stop(struct seq_file *s, void *v)
698 struct mdt_device *mdt = s->private;
699 struct coordinator *cdt = &mdt->mdt_coordinator;
701 up_read(&cdt->cdt_agent_lock);
704 /* hsm agent list proc functions */
705 static const struct seq_operations mdt_hsm_agent_proc_ops = {
706 .start = mdt_hsm_agent_proc_start,
707 .next = mdt_hsm_agent_proc_next,
708 .show = mdt_hsm_agent_proc_show,
709 .stop = mdt_hsm_agent_proc_stop,
713 * public function called at open of /proc file to get
716 static int lprocfs_open_hsm_agent(struct inode *inode, struct file *file)
722 rc = seq_open(file, &mdt_hsm_agent_proc_ops);
726 s = file->private_data;
727 s->private = PDE_DATA(inode);
732 /* methods to access hsm agent list */
733 const struct file_operations mdt_hsm_agent_fops = {
734 .owner = THIS_MODULE,
735 .open = lprocfs_open_hsm_agent,
738 .release = lprocfs_seq_release,