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 * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
26 * Copyright (c) 2013, 2017, Intel Corporation.
27 * Use is subject to license terms.
30 * lustre/mdt/mdt_coordinator.c
32 * Lustre HSM Coordinator
34 * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
35 * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
36 * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <linux/kthread.h>
42 #include <linux/kernel.h>
43 #include <obd_support.h>
44 #include <lustre_export.h>
46 #include <lprocfs_status.h>
47 #include <lustre_log.h>
48 #include <lustre_kernelcomm.h>
49 #include "mdt_internal.h"
52 * get obj and HSM attributes on a fid
53 * \param mti [IN] context
54 * \param fid [IN] object fid
55 * \param hsm [OUT] HSM meta data
56 * \retval obj or error (-ENOENT if not found)
58 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
59 const struct lu_fid *fid,
63 struct mdt_object *obj;
71 /* find object by FID */
72 obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
76 if (!mdt_object_exists(obj)) {
78 mdt_object_put(mti->mti_env, obj);
79 RETURN(ERR_PTR(-ENOENT));
82 rc = mdt_attr_get_complex(mti, obj, ma);
84 mdt_object_put(mti->mti_env, obj);
88 if (ma->ma_valid & MA_HSM)
91 memset(hsm, 0, sizeof(*hsm));
96 void mdt_hsm_dump_hal(int level, const char *prefix,
97 struct hsm_action_list *hal)
100 struct hsm_action_item *hai;
103 CDEBUG(level, "%s: HAL header: version %X count %d"
104 " archive_id %d flags %#llx\n",
105 prefix, hal->hal_version, hal->hal_count,
106 hal->hal_archive_id, hal->hal_flags);
108 hai = hai_first(hal);
109 for (i = 0; i < hal->hal_count; i++) {
110 sz = hai->hai_len - sizeof(*hai);
111 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
113 " action=%s extent=%#llx-%#llx gid=%#llx"
114 " datalen=%d data=[%s]\n",
116 PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
118 hsm_copytool_action2name(hai->hai_action),
119 hai->hai_extent.offset,
120 hai->hai_extent.length,
122 hai_dump_data_field(hai, buf, sizeof(buf)));
128 * data passed to llog_cat_process() callback
129 * to scan requests and take actions
131 struct hsm_scan_request {
134 struct hsm_action_list *hal;
137 struct hsm_scan_data {
138 struct mdt_thread_info *hsd_mti;
139 char hsd_fsname[MTI_NAME_MAXLEN + 1];
140 /* are we scanning the logs for housekeeping, or just looking
143 bool hsd_housekeeping;
144 bool hsd_one_restore;
145 u32 hsd_start_cat_idx;
146 u32 hsd_start_rec_idx;
147 int hsd_action_count;
148 int hsd_request_len; /* array alloc len */
149 int hsd_request_count; /* array used count */
150 struct hsm_scan_request *hsd_request;
153 static int mdt_cdt_waiting_cb(const struct lu_env *env,
154 struct mdt_device *mdt,
155 struct llog_handle *llh,
156 struct llog_agent_req_rec *larr,
157 struct hsm_scan_data *hsd)
159 struct coordinator *cdt = &mdt->mdt_coordinator;
160 struct hsm_scan_request *request;
161 struct hsm_action_item *hai;
167 /* Are agents full? */
168 if (atomic_read(&cdt->cdt_request_count) >= cdt->cdt_max_requests)
169 RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
171 if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
172 cdt->cdt_max_requests) {
173 /* We cannot send any more request
175 * *** SPECIAL CASE ***
177 * Restore requests are too important not to schedule at least
178 * one, everytime we can.
180 if (larr->arr_hai.hai_action != HSMA_RESTORE ||
181 hsd->hsd_one_restore)
182 RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
185 hai_size = round_up(larr->arr_hai.hai_len, 8);
186 archive_id = larr->arr_archive_id;
188 /* Can we add this action to one of the existing HALs in hsd. */
190 for (i = 0; i < hsd->hsd_request_count; i++) {
191 if (hsd->hsd_request[i].hal->hal_archive_id == archive_id &&
192 hsd->hsd_request[i].hal_used_sz + hai_size <=
194 request = &hsd->hsd_request[i];
199 /* Are we trying to force-schedule a request? */
200 if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
201 cdt->cdt_max_requests) {
202 /* Is there really no compatible hsm_scan_request? */
204 for (i -= 1; i >= 0; i--) {
205 if (hsd->hsd_request[i].hal->hal_archive_id ==
207 request = &hsd->hsd_request[i];
213 /* Make room for the hai */
215 /* Discard the last hai until there is enough space */
217 request->hal->hal_count--;
219 hai = hai_first(request->hal);
220 for (i = 0; i < request->hal->hal_count; i++)
222 request->hal_used_sz -=
223 round_up(hai->hai_len, 8);
224 hsd->hsd_action_count--;
225 } while (request->hal_used_sz + hai_size >
227 } else if (hsd->hsd_housekeeping) {
228 struct hsm_scan_request *tmp;
230 /* Discard the (whole) last hal */
231 hsd->hsd_request_count--;
232 LASSERT(hsd->hsd_request_count >= 0);
233 tmp = &hsd->hsd_request[hsd->hsd_request_count];
234 hsd->hsd_action_count -= tmp->hal->hal_count;
235 LASSERT(hsd->hsd_action_count >= 0);
236 OBD_FREE(tmp->hal, tmp->hal_sz);
238 /* Bailing out, this code path is too hot */
239 RETURN(LLOG_PROC_BREAK);
245 size_t count = round_up(MTI_NAME_MAXLEN + 1, 8) + 2 * hai_size;
246 struct hsm_action_list *hal;
248 LASSERT(hsd->hsd_request_count < hsd->hsd_request_len);
249 request = &hsd->hsd_request[hsd->hsd_request_count];
251 /* allocates hai vector size just needs to be large
253 request->hal_sz = sizeof(*request->hal) + count;
254 OBD_ALLOC_LARGE(hal, request->hal_sz);
258 hal->hal_version = HAL_VERSION;
259 strscpy(hal->hal_fsname, hsd->hsd_fsname, MTI_NAME_MAXLEN + 1);
260 hal->hal_archive_id = larr->arr_archive_id;
261 hal->hal_flags = larr->arr_flags;
263 request->hal_used_sz = hal_size(hal);
265 hsd->hsd_request_count++;
266 } else if (request->hal_sz < request->hal_used_sz + hai_size) {
267 /* Not enough room, need an extension */
271 sz = min_t(int, 2 * request->hal_sz, LDLM_MAXREQSIZE);
272 LASSERT(request->hal_used_sz + hai_size < sz);
274 OBD_ALLOC_LARGE(hal_buffer, sz);
278 memcpy(hal_buffer, request->hal, request->hal_used_sz);
279 OBD_FREE_LARGE(request->hal, request->hal_sz);
280 request->hal = hal_buffer;
281 request->hal_sz = sz;
284 hai = hai_first(request->hal);
285 for (i = 0; i < request->hal->hal_count; i++)
288 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
290 request->hal_used_sz += hai_size;
291 request->hal->hal_count++;
293 hsd->hsd_action_count++;
295 switch (hai->hai_action) {
299 hsd->hsd_one_restore = true;
302 cdt_agent_record_hash_add(cdt, hai->hai_cookie,
303 llh->lgh_hdr->llh_cat_idx,
304 larr->arr_hdr.lrh_index);
307 wrapped = llh->lgh_hdr->llh_cat_idx >= llh->lgh_last_idx &&
308 llh->lgh_hdr->llh_count > 1;
309 if ((!wrapped && llh->lgh_hdr->llh_cat_idx > hsd->hsd_start_cat_idx) ||
310 (wrapped && llh->lgh_hdr->llh_cat_idx < hsd->hsd_start_cat_idx) ||
311 (llh->lgh_hdr->llh_cat_idx == hsd->hsd_start_cat_idx &&
312 larr->arr_hdr.lrh_index > hsd->hsd_start_rec_idx)) {
313 hsd->hsd_start_cat_idx = llh->lgh_hdr->llh_cat_idx;
314 hsd->hsd_start_rec_idx = larr->arr_hdr.lrh_index;
320 static int mdt_cdt_started_cb(const struct lu_env *env,
321 struct mdt_device *mdt,
322 struct llog_handle *llh,
323 struct llog_agent_req_rec *larr,
324 struct hsm_scan_data *hsd)
326 struct coordinator *cdt = &mdt->mdt_coordinator;
327 struct hsm_action_item *hai = &larr->arr_hai;
328 struct cdt_agent_req *car;
329 time64_t now = ktime_get_real_seconds();
331 enum changelog_rec_flags clf_flags;
334 if (!hsd->hsd_housekeeping)
337 /* we search for a running request
338 * error may happen if coordinator crashes or stopped
339 * with running request
341 car = mdt_cdt_find_request(cdt, hai->hai_cookie);
343 last = larr->arr_req_change;
345 last = car->car_req_update;
348 /* test if request too long, if yes cancel it
349 * the same way the copy tool acknowledge a cancel request */
350 if (now <= last + cdt->cdt_active_req_timeout)
351 GOTO(out_car, rc = 0);
353 dump_llog_agent_req_rec("request timed out, start cleaning", larr);
356 car->car_req_update = now;
357 mdt_hsm_agent_update_statistics(cdt, 0, 1, 0, &car->car_uuid);
358 /* Remove car from memory list (LU-9075) */
359 mdt_cdt_remove_request(cdt, hai->hai_cookie);
362 /* Emit a changelog record for the failed action.*/
364 hsm_set_cl_error(&clf_flags, ECANCELED);
366 switch (hai->hai_action) {
368 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
371 hsm_set_cl_event(&clf_flags, HE_RESTORE);
374 hsm_set_cl_event(&clf_flags, HE_REMOVE);
377 hsm_set_cl_event(&clf_flags, HE_CANCEL);
380 /* Unknown record type, skip changelog. */
386 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
389 if (hai->hai_action == HSMA_RESTORE)
390 cdt_restore_handle_del(hsd->hsd_mti, cdt, &hai->hai_fid);
392 larr->arr_status = ARS_CANCELED;
393 larr->arr_req_change = now;
394 rc = llog_write(hsd->hsd_mti->mti_env, llh, &larr->arr_hdr,
395 larr->arr_hdr.lrh_index);
397 CERROR("%s: cannot update agent log: rc = %d\n",
398 mdt_obd_name(mdt), rc);
399 rc = LLOG_DEL_RECORD;
402 /* ct has completed a request, so a slot is available,
403 * signal the coordinator to find new work */
404 mdt_hsm_cdt_event(cdt);
407 mdt_cdt_put_request(car);
413 * llog_cat_process() callback, used to:
414 * - find waiting request and start action
415 * - purge canceled and done requests
416 * \param env [IN] environment
417 * \param llh [IN] llog handle
418 * \param hdr [IN] llog record
419 * \param data [IN/OUT] cb data = struct hsm_scan_data
421 * \retval -ve failure
423 static int mdt_coordinator_cb(const struct lu_env *env,
424 struct llog_handle *llh,
425 struct llog_rec_hdr *hdr,
428 struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
429 struct hsm_scan_data *hsd = data;
430 struct mdt_device *mdt = hsd->hsd_mti->mti_mdt;
431 struct coordinator *cdt = &mdt->mdt_coordinator;
434 larr = (struct llog_agent_req_rec *)hdr;
435 dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
436 switch (larr->arr_status) {
438 RETURN(mdt_cdt_waiting_cb(env, mdt, llh, larr, hsd));
440 RETURN(mdt_cdt_started_cb(env, mdt, llh, larr, hsd));
442 if (!hsd->hsd_housekeeping)
445 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
446 ktime_get_real_seconds()) {
447 cdt_agent_record_hash_del(cdt,
448 larr->arr_hai.hai_cookie);
449 RETURN(LLOG_DEL_RECORD);
456 static void cdt_crh_free(struct rcu_head *head)
458 struct cdt_restore_handle *crh;
460 crh = container_of(head, struct cdt_restore_handle, crh_rcu);
461 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
465 cdt_crh_put(struct cdt_restore_handle *crh, struct mdt_thread_info *cdt_mti)
467 if (atomic_dec_and_test(&crh->crh_refc)) {
468 /* XXX We pass a NULL object since the restore handle does not
469 * keep a reference on the object being restored.
471 if (lustre_handle_is_used(&crh->crh_lh.mlh_reg_lh))
472 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
473 call_rcu(&crh->crh_rcu, cdt_crh_free);
477 static void crh_free_hash(void *vcrh, void *vcdt_mti)
479 struct cdt_restore_handle *crh = vcrh;
480 struct mdt_thread_info *cdt_mti = vcdt_mti;
482 /* put last reference */
483 cdt_crh_put(crh, cdt_mti);
486 static const struct rhashtable_params crh_hash_params = {
487 .key_len = sizeof(struct lu_fid),
488 .key_offset = offsetof(struct cdt_restore_handle, crh_fid),
489 .head_offset = offsetof(struct cdt_restore_handle, crh_hash),
490 .hashfn = lu_fid_hash,
491 .automatic_shrinking = true,
494 /* Release the ressource used by the coordinator. Called when the
495 * coordinator is stopping. */
496 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
498 struct coordinator *cdt = &mdt->mdt_coordinator;
499 struct cdt_agent_req *car, *tmp1;
500 struct hsm_agent *ha, *tmp2;
501 struct mdt_thread_info *cdt_mti;
504 down_write(&cdt->cdt_request_lock);
505 list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
507 cfs_hash_del(cdt->cdt_request_cookie_hash,
508 &car->car_hai->hai_cookie,
509 &car->car_cookie_hash);
510 list_del(&car->car_request_list);
511 mdt_cdt_put_request(car);
513 up_write(&cdt->cdt_request_lock);
515 down_write(&cdt->cdt_agent_lock);
516 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
517 list_del(&ha->ha_list);
518 if (ha->ha_archive_cnt != 0)
519 OBD_FREE_PTR_ARRAY(ha->ha_archive_id,
523 up_write(&cdt->cdt_agent_lock);
525 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
526 rhashtable_free_and_destroy(&cdt->cdt_restore_hash, crh_free_hash,
532 * Coordinator state transition table, indexed on enum cdt_states, taking
533 * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
534 * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
536 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
537 /* from -> to: stopped init running disable stopping */
538 /* stopped */ { true, true, false, false, false },
539 /* init */ { true, false, true, false, false },
540 /* running */ { false, false, true, true, true },
541 /* disable */ { false, false, true, true, true },
542 /* stopping */ { true, false, false, false, false }
546 * Change coordinator thread state
547 * Some combinations are not valid, so catch them here.
549 * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
550 * the transition was not possible.
552 static int set_cdt_state_locked(struct coordinator *cdt,
553 enum cdt_states new_state)
556 enum cdt_states state;
558 state = cdt->cdt_state;
560 if (cdt_transition[state][new_state]) {
561 cdt->cdt_state = new_state;
565 "unexpected coordinator transition, from=%s, to=%s\n",
566 cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
573 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state)
577 mutex_lock(&cdt->cdt_state_lock);
578 rc = set_cdt_state_locked(cdt, new_state);
579 mutex_unlock(&cdt->cdt_state_lock);
584 int cdt_getref_try(struct coordinator *cdt)
586 return refcount_inc_not_zero(&cdt->cdt_ref);
589 void cdt_putref(struct coordinator *cdt)
591 if (refcount_dec_and_test(&cdt->cdt_ref))
592 wake_up(&cdt->cdt_waitq);
595 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti);
597 static int cdt_start_pending_restore(struct mdt_device *mdt,
598 struct coordinator *cdt)
600 struct mdt_thread_info *cdt_mti;
604 /* wait until MDD initialize hsm actions llog */
605 while (!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state) && i < obd_timeout) {
606 schedule_timeout_interruptible(cfs_time_seconds(1));
607 if (kthread_should_stop())
611 if (!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state))
612 CWARN("%s: trying to init HSM before MDD\n", mdt_obd_name(mdt));
614 /* set up list of started restore requests */
615 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
616 rc = mdt_hsm_pending_restore(cdt_mti);
618 CERROR("%s: cannot take the layout locks needed for registered restore: %d\n",
619 mdt_obd_name(mdt), rc);
626 * \param data [IN] obd device
628 * \retval -ve failure
630 static int mdt_coordinator(void *data)
632 struct mdt_thread_info *mti = data;
633 struct mdt_device *mdt = mti->mti_mdt;
634 struct coordinator *cdt = &mdt->mdt_coordinator;
635 struct hsm_scan_data hsd = { NULL };
636 time64_t last_housekeeping = 0;
637 size_t request_sz = 0;
641 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
642 mdt_obd_name(mdt), current->pid);
645 obd_uuid2fsname(hsd.hsd_fsname, mdt_obd_name(mdt),
646 sizeof(hsd.hsd_fsname));
648 set_cdt_state(cdt, CDT_RUNNING);
650 /* Inform mdt_hsm_cdt_start(). */
651 wake_up(&cdt->cdt_waitq);
653 /* this initilazes cdt_last_cookie too */
654 rc = cdt_start_pending_restore(mdt, cdt);
655 if (rc < 0 || kthread_should_stop())
656 GOTO(fail_to_start, rc);
658 refcount_set(&cdt->cdt_ref, 1);
667 struct hsm_record_update *updates;
669 /* Limit execution of the expensive requests traversal
670 * to at most one second. This prevents repeatedly
671 * locking/unlocking the catalog for each request
672 * and preventing other HSM operations from happening
674 wait_event_interruptible_timeout(cdt->cdt_waitq,
675 kthread_should_stop() ||
676 cdt->cdt_wakeup_coordinator,
677 cfs_time_seconds(1));
679 cdt->cdt_wakeup_coordinator = false;
680 CDEBUG(D_HSM, "coordinator resumes\n");
682 if (kthread_should_stop()) {
683 CDEBUG(D_HSM, "Coordinator stops\n");
685 /* Drop the running ref */
687 /* Wait threads to finish */
688 wait_event(cdt->cdt_waitq,
689 refcount_read(&cdt->cdt_ref) == 0);
694 /* if coordinator is suspended continue to wait */
695 if (cdt->cdt_state == CDT_DISABLE) {
696 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
700 /* If no event, and no housekeeping to do, continue to
702 if (last_housekeeping + cdt->cdt_loop_period <=
703 ktime_get_real_seconds()) {
704 last_housekeeping = ktime_get_real_seconds();
705 hsd.hsd_housekeeping = true;
708 } else if (cdt->cdt_event) {
709 hsd.hsd_housekeeping = false;
710 start_cat_idx = hsd.hsd_start_cat_idx;
711 start_rec_idx = hsd.hsd_start_rec_idx;
716 cdt->cdt_event = false;
718 CDEBUG(D_HSM, "coordinator starts reading llog\n");
720 if (hsd.hsd_request_len != cdt->cdt_max_requests) {
721 /* cdt_max_requests has changed,
722 * we need to allocate a new buffer
724 struct hsm_scan_request *tmp = NULL;
725 int max_requests = cdt->cdt_max_requests;
726 OBD_ALLOC_LARGE(tmp, max_requests *
727 sizeof(struct hsm_scan_request));
729 CERROR("Failed to resize request buffer, "
730 "keeping it at %d\n",
731 hsd.hsd_request_len);
733 if (hsd.hsd_request != NULL)
734 OBD_FREE_LARGE(hsd.hsd_request,
737 hsd.hsd_request_len = max_requests;
738 request_sz = hsd.hsd_request_len *
739 sizeof(struct hsm_scan_request);
740 hsd.hsd_request = tmp;
744 hsd.hsd_action_count = 0;
745 hsd.hsd_request_count = 0;
746 hsd.hsd_one_restore = false;
748 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
749 &hsd, start_cat_idx, start_rec_idx,
754 CDEBUG(D_HSM, "found %d requests to send\n",
755 hsd.hsd_request_count);
757 if (list_empty(&cdt->cdt_agents)) {
758 CDEBUG(D_HSM, "no agent available, "
759 "coordinator sleeps\n");
760 /* reset HSM scanning index range. */
761 hsd.hsd_start_cat_idx = start_cat_idx;
762 hsd.hsd_start_rec_idx = start_rec_idx;
766 /* Compute how many HAI we have in all the requests */
768 for (i = 0; i < hsd.hsd_request_count; i++) {
769 const struct hsm_scan_request *request =
772 updates_cnt += request->hal->hal_count;
775 /* Allocate a temporary array to store the cookies to
776 * update, and their status. */
777 updates_sz = updates_cnt * sizeof(*updates);
778 OBD_ALLOC_LARGE(updates, updates_sz);
779 if (updates == NULL) {
780 CERROR("%s: Cannot allocate memory (%d bytes) "
781 "for %d updates. Too many HSM requests?\n",
782 mdt_obd_name(mdt), updates_sz, updates_cnt);
786 /* here hsd contains a list of requests to be started */
787 for (i = 0; i < hsd.hsd_request_count; i++) {
788 struct hsm_scan_request *request = &hsd.hsd_request[i];
789 struct hsm_action_list *hal = request->hal;
790 struct hsm_action_item *hai;
793 /* still room for work ? */
794 if (atomic_read(&cdt->cdt_request_count) >=
795 cdt->cdt_max_requests)
798 rc = mdt_hsm_agent_send(mti, hal, 0);
799 /* if failure, we suppose it is temporary
800 * if the copy tool failed to do the request
801 * it has to use hsm_progress
804 /* set up cookie vector to set records status
805 * after copy tools start or failed
807 hai = hai_first(hal);
808 for (j = 0; j < hal->hal_count; j++) {
809 updates[update_idx].cookie = hai->hai_cookie;
810 updates[update_idx].status =
811 (rc ? ARS_WAITING : ARS_STARTED);
816 /* TODO: narrow down the HSM action range that already
817 * scanned accroding to the cookies when a failure
821 hsd.hsd_start_cat_idx = start_cat_idx;
822 hsd.hsd_start_rec_idx = start_rec_idx;
827 rc = mdt_agent_record_update(mti, updates, update_idx);
829 CERROR("%s: mdt_agent_record_update() failed, "
830 "rc=%d, cannot update records "
832 mdt_obd_name(mdt), rc, update_idx);
835 OBD_FREE_LARGE(updates, updates_sz);
838 /* free hal allocated by callback */
839 for (i = 0; i < hsd.hsd_request_count; i++) {
840 struct hsm_scan_request *request = &hsd.hsd_request[i];
842 OBD_FREE_LARGE(request->hal, request->hal_sz);
846 if (hsd.hsd_request != NULL)
847 OBD_FREE_LARGE(hsd.hsd_request, request_sz);
850 mdt_hsm_cdt_cleanup(mdt);
853 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
854 mdt_obd_name(mdt), current->pid, rc);
856 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
858 mdt_obd_name(mdt), current->pid);
860 set_cdt_state(cdt, CDT_STOPPED);
862 /* Inform mdt_hsm_cdt_stop(). */
863 wake_up(&cdt->cdt_waitq);
869 * register a new HSM restore handle for a file and take EX lock on the layout
870 * \param mti [IN] thread info
871 * \param cdt [IN] coordinator
872 * \param fid [IN] fid of the file to restore
873 * \param he [IN] HSM extent
875 * \retval 1 restore handle already exists for the fid
876 * \retval -ve failure
878 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
879 const struct lu_fid *fid,
880 const struct hsm_extent *he)
882 struct cdt_restore_handle *crh;
883 struct mdt_object *obj;
887 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
892 /* in V1 all file is restored
893 * crh->extent.start = he->offset;
894 * crh->extent.end = he->offset + he->length;
896 crh->crh_extent.start = 0;
897 crh->crh_extent.end = he->length;
898 atomic_set(&crh->crh_refc, 2);
900 rc = rhashtable_lookup_insert_fast(&cdt->cdt_restore_hash,
901 &crh->crh_hash, crh_hash_params);
903 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
907 /* get the layout lock */
908 obj = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
909 MDS_INODELOCK_LAYOUT, LCK_EX);
911 rc = rhashtable_remove_fast(&cdt->cdt_restore_hash,
912 &crh->crh_hash, crh_hash_params);
913 /* rc < 0 means it has been removed in a parallel thread.
914 * This shouldn't happen by design as at current stage record
915 * hasn't been added in llog yet.
918 cdt_crh_put(crh, mti);
919 cdt_crh_put(crh, mti);
921 RETURN(PTR_ERR(obj));
924 /* We do not keep a reference on the object during the restore
925 * which can be very long.
927 mdt_object_put(mti->mti_env, obj);
928 cdt_crh_put(crh, mti);
933 * lookup a restore handle by FID
934 * \param cdt [IN] coordinator
935 * \param fid [IN] FID
936 * \retval true cdt_restore_handle found
937 * \retval false not found
939 bool cdt_restore_handle_exists(struct coordinator *cdt,
940 const struct lu_fid *fid)
942 return rhashtable_lookup_fast(&cdt->cdt_restore_hash, fid,
946 void cdt_restore_handle_del(struct mdt_thread_info *mti,
947 struct coordinator *cdt, const struct lu_fid *fid)
949 struct cdt_restore_handle *crh;
951 /* give back layout lock */
953 crh = rhashtable_lookup(&cdt->cdt_restore_hash, fid, crh_hash_params);
955 rhashtable_remove_fast(&cdt->cdt_restore_hash, &crh->crh_hash,
960 /* crh has been removed in a parallel thread */
964 cdt_crh_put(crh, mti);
968 * data passed to llog_cat_process() callback
969 * to scan requests and take actions
971 struct hsm_restore_data {
972 struct mdt_thread_info *hrd_mti;
976 * llog_cat_process() callback, used to:
977 * - find restore request and allocate the restore handle
978 * \param env [IN] environment
979 * \param llh [IN] llog handle
980 * \param hdr [IN] llog record
981 * \param data [IN/OUT] cb data = struct hsm_restore_data
983 * \retval -ve failure
985 static int hsm_restore_cb(const struct lu_env *env,
986 struct llog_handle *llh,
987 struct llog_rec_hdr *hdr, void *data)
989 struct llog_agent_req_rec *larr;
990 struct hsm_restore_data *hrd;
991 struct hsm_action_item *hai;
992 struct mdt_thread_info *mti;
993 struct coordinator *cdt;
999 cdt = &mti->mti_mdt->mdt_coordinator;
1001 larr = (struct llog_agent_req_rec *)hdr;
1002 hai = &larr->arr_hai;
1003 if (hai->hai_cookie >= cdt->cdt_last_cookie) {
1004 /* update the cookie to avoid collision */
1005 cdt->cdt_last_cookie = hai->hai_cookie + 1;
1008 if (hai->hai_action != HSMA_RESTORE ||
1009 agent_req_in_final_state(larr->arr_status))
1012 /* restore request not in a final state */
1014 /* force replay of restore requests left in started state from previous
1015 * CDT context, to be canceled later if finally found to be incompatible
1016 * when being re-started */
1017 if (larr->arr_status == ARS_STARTED) {
1018 larr->arr_status = ARS_WAITING;
1019 larr->arr_req_change = ktime_get_real_seconds();
1020 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1025 rc = cdt_restore_handle_add(mti, cdt, &hai->hai_fid, &hai->hai_extent);
1026 if (rc == -EEXIST) {
1027 CWARN("%s: duplicate restore record for fid="DFID" found in the llog: rc = %d\n",
1028 mdt_obd_name(mti->mti_mdt), PFID(&hai->hai_fid), rc);
1036 * restore coordinator state at startup
1037 * the goal is to take a layout lock for each registered restore request
1038 * \param mti [IN] context
1040 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
1042 struct hsm_restore_data hrd;
1048 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
1054 int hsm_init_ucred(struct lu_ucred *uc)
1057 uc->uc_valid = UCRED_OLD;
1066 uc->uc_suppgids[0] = -1;
1067 uc->uc_suppgids[1] = -1;
1068 uc->uc_cap = cap_combine(CAP_FS_SET, CAP_NFSD_SET);
1069 uc->uc_umask = 0777;
1070 uc->uc_ginfo = NULL;
1071 uc->uc_identity = NULL;
1072 /* always record internal HSM activity if also enabled globally */
1073 uc->uc_enable_audit = 1;
1074 /* do not let rbac interfere with HSM internal processing */
1075 uc->uc_rbac_file_perms = 1;
1076 uc->uc_rbac_dne_ops = 1;
1077 uc->uc_rbac_quota_ops = 1;
1078 uc->uc_rbac_byfid_ops = 1;
1079 uc->uc_rbac_chlg_ops = 1;
1080 uc->uc_rbac_fscrypt_admin = 1;
1086 * initialize coordinator struct
1087 * \param mdt [IN] device
1089 * \retval -ve failure
1091 int mdt_hsm_cdt_init(struct mdt_device *mdt)
1093 struct coordinator *cdt = &mdt->mdt_coordinator;
1094 struct mdt_thread_info *cdt_mti = NULL;
1098 init_waitqueue_head(&cdt->cdt_waitq);
1099 init_rwsem(&cdt->cdt_llog_lock);
1100 init_rwsem(&cdt->cdt_agent_lock);
1101 init_rwsem(&cdt->cdt_request_lock);
1102 mutex_init(&cdt->cdt_state_lock);
1103 set_cdt_state(cdt, CDT_STOPPED);
1105 INIT_LIST_HEAD(&cdt->cdt_request_list);
1106 INIT_LIST_HEAD(&cdt->cdt_agents);
1108 cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
1112 0 /* extra bytes */,
1115 &cdt_request_cookie_hash_ops,
1117 if (cdt->cdt_request_cookie_hash == NULL)
1120 cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
1124 0 /* extra bytes */,
1127 &cdt_agent_record_hash_ops,
1129 if (cdt->cdt_agent_record_hash == NULL)
1130 GOTO(out_request_cookie_hash, rc = -ENOMEM);
1132 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
1134 GOTO(out_agent_record_hash, rc);
1136 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1137 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
1141 lu_context_enter(&cdt->cdt_session);
1142 cdt->cdt_env.le_ses = &cdt->cdt_session;
1144 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1145 LASSERT(cdt_mti != NULL);
1147 cdt_mti->mti_env = &cdt->cdt_env;
1148 cdt_mti->mti_mdt = mdt;
1150 hsm_init_ucred(mdt_ucred(cdt_mti));
1152 /* default values for sysfs tunnables
1153 * can be override by MGS conf */
1154 cdt->cdt_default_archive_id = 1;
1155 cdt->cdt_grace_delay = 60;
1156 cdt->cdt_loop_period = 10;
1157 cdt->cdt_max_requests = 3;
1158 cdt->cdt_policy = CDT_DEFAULT_POLICY;
1159 cdt->cdt_active_req_timeout = 3600;
1161 /* by default do not remove archives on last unlink */
1162 cdt->cdt_remove_archive_on_last_unlink = false;
1167 lu_env_fini(&cdt->cdt_env);
1168 out_agent_record_hash:
1169 cfs_hash_putref(cdt->cdt_agent_record_hash);
1170 cdt->cdt_agent_record_hash = NULL;
1171 out_request_cookie_hash:
1172 cfs_hash_putref(cdt->cdt_request_cookie_hash);
1173 cdt->cdt_request_cookie_hash = NULL;
1179 * free a coordinator thread
1180 * \param mdt [IN] device
1182 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
1184 struct coordinator *cdt = &mdt->mdt_coordinator;
1187 lu_context_exit(cdt->cdt_env.le_ses);
1188 lu_context_fini(cdt->cdt_env.le_ses);
1190 lu_env_fini(&cdt->cdt_env);
1192 cfs_hash_putref(cdt->cdt_agent_record_hash);
1193 cdt->cdt_agent_record_hash = NULL;
1195 cfs_hash_putref(cdt->cdt_request_cookie_hash);
1196 cdt->cdt_request_cookie_hash = NULL;
1202 * start a coordinator thread
1203 * \param mdt [IN] device
1205 * \retval -ve failure
1207 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1209 struct coordinator *cdt = &mdt->mdt_coordinator;
1210 struct mdt_thread_info *cdt_mti;
1213 struct task_struct *task;
1216 /* functions defined but not yet used
1217 * this avoid compilation warning
1219 ptr = dump_requests;
1221 rc = set_cdt_state(cdt, CDT_INIT);
1223 CERROR("%s: Coordinator already started or stopping\n",
1228 BUILD_BUG_ON(BIT(CDT_POLICY_SHIFT_COUNT - 1) != CDT_POLICY_LAST);
1229 cdt->cdt_policy = CDT_DEFAULT_POLICY;
1231 /* just need to be larger than previous one */
1232 /* cdt_last_cookie is protected by cdt_llog_lock */
1233 cdt->cdt_last_cookie = ktime_get_real_seconds();
1234 atomic_set(&cdt->cdt_request_count, 0);
1235 atomic_set(&cdt->cdt_archive_count, 0);
1236 atomic_set(&cdt->cdt_restore_count, 0);
1237 atomic_set(&cdt->cdt_remove_count, 0);
1238 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
1239 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
1240 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
1241 rc = rhashtable_init(&cdt->cdt_restore_hash, &crh_hash_params);
1243 CERROR("%s: failed to create cdt_restore hash: rc = %d\n",
1244 mdt_obd_name(mdt), rc);
1245 set_cdt_state(cdt, CDT_STOPPED);
1249 /* to avoid deadlock when start is made through sysfs
1250 * sysfs entries are created by the coordinator thread
1252 if (mdt->mdt_bottom->dd_rdonly)
1255 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1256 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
1259 set_cdt_state(cdt, CDT_STOPPED);
1260 CERROR("%s: error starting coordinator thread: %d\n",
1261 mdt_obd_name(mdt), rc);
1263 cdt->cdt_task = task;
1264 wait_event(cdt->cdt_waitq,
1265 cdt->cdt_state != CDT_INIT);
1266 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1275 * stop a coordinator thread
1276 * \param mdt [IN] device
1278 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1280 struct coordinator *cdt = &mdt->mdt_coordinator;
1285 /* stop coordinator thread */
1286 rc = set_cdt_state(cdt, CDT_STOPPING);
1290 kthread_stop(cdt->cdt_task);
1291 rc = wait_event_interruptible(cdt->cdt_waitq,
1292 cdt->cdt_state == CDT_STOPPED);
1296 cdt->cdt_task = NULL;
1300 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1301 const struct lu_fid *fid,
1304 struct mdt_object *obj;
1308 obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1310 GOTO(out, rc = PTR_ERR(obj));
1312 if (mh.mh_flags & HS_EXISTS &&
1313 mh.mh_arch_id == archive_id)
1314 GOTO(out_obj, rc = 0);
1316 mh.mh_flags |= HS_EXISTS;
1317 mh.mh_arch_id = archive_id;
1318 rc = mdt_hsm_attr_set(mti, obj, &mh);
1321 mdt_object_put(mti->mti_env, obj);
1327 * register all requests from an hal in the memory list
1328 * \param mti [IN] context
1329 * \param hal [IN] request
1330 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1331 * which is running the CT
1333 * \retval -ve failure
1335 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1336 struct hsm_action_list *hal, struct obd_uuid *uuid)
1338 struct mdt_device *mdt = mti->mti_mdt;
1339 struct coordinator *cdt = &mdt->mdt_coordinator;
1340 struct hsm_action_item *hai;
1344 /* register request in memory list */
1345 hai = hai_first(hal);
1346 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1347 struct cdt_agent_req *car;
1349 /* in case of a cancel request, we first mark the ondisk
1350 * record of the request we want to stop as canceled
1351 * this does not change the cancel record
1352 * it will be done when updating the request status
1354 if (hai->hai_action == HSMA_CANCEL) {
1355 struct hsm_record_update update = {
1356 .cookie = hai->hai_cookie,
1357 .status = ARS_CANCELED,
1360 rc = mdt_agent_record_update(mti, &update, 1);
1362 CERROR("%s: mdt_agent_record_update() failed, "
1363 "rc=%d, cannot update status to %s "
1364 "for cookie %#llx\n",
1365 mdt_obd_name(mdt), rc,
1366 agent_req_status2name(ARS_CANCELED),
1371 /* find the running request to set it canceled */
1372 car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1374 car->car_canceled = 1;
1375 /* uuid has to be changed to the one running the
1376 * request to cancel */
1377 *uuid = car->car_uuid;
1378 mdt_cdt_put_request(car);
1380 /* no need to memorize cancel request
1381 * this also avoid a deadlock when we receive
1382 * a purge all requests command
1387 if (hai->hai_action == HSMA_ARCHIVE) {
1388 rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1389 hal->hal_archive_id);
1396 car = mdt_cdt_alloc_request(hal->hal_archive_id, hal->hal_flags,
1399 GOTO(out, rc = PTR_ERR(car));
1401 rc = mdt_cdt_add_request(cdt, car);
1403 mdt_cdt_free_request(car);
1410 * swap layouts between 2 fids
1411 * \param mti [IN] context
1414 * \param mh_common [IN] MD HSM
1416 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1417 struct mdt_object *obj, const struct lu_fid *dfid,
1418 struct md_hsm *mh_common)
1420 struct mdt_object *dobj;
1421 struct mdt_lock_handle *dlh;
1425 if (!mdt_object_exists(obj))
1426 GOTO(out, rc = -ENOENT);
1428 /* we already have layout lock on obj so take only
1430 dlh = &mti->mti_lh[MDT_LH_OLD];
1431 dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT,
1434 GOTO(out, rc = PTR_ERR(dobj));
1436 /* if copy tool closes the volatile before sending the final
1437 * progress through llapi_hsm_copy_end(), all the objects
1438 * are removed and mdd_swap_layout LBUG */
1439 if (!mdt_object_exists(dobj)) {
1440 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1441 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1442 GOTO(out_dobj, rc = -ENOENT);
1444 /* Since we only handle restores here, unconditionally use
1445 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1446 * be preserved in case of failure during swap_layout and not
1447 * leave a file in an intermediate but incoherent state.
1448 * But need to setup HSM xattr of data FID before, reuse
1449 * mti and mh presets for FID in hsm_cdt_request_completed(),
1450 * only need to clear RELEASED and DIRTY.
1452 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1453 rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1457 rc = mo_swap_layouts(mti->mti_env, mdt_object_child(obj),
1458 mdt_object_child(dobj), 0, 0, 0);
1462 rc = mdt_lsom_downgrade(mti, obj);
1465 "%s: File fid="DFID" SOM downgrade failed, rc = %d\n",
1466 mdt_obd_name(mti->mti_mdt),
1467 PFID(mdt_object_fid(obj)), rc);
1469 mdt_object_unlock_put(mti, dobj, dlh, 1);
1475 * update status of a completed request
1476 * \param mti [IN] context
1477 * \param pgs [IN] progress of the copy tool
1479 * \retval -ve failure
1481 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1482 struct hsm_progress_kernel *pgs,
1483 const struct cdt_agent_req *car,
1484 enum agent_req_status *status)
1486 const struct lu_env *env = mti->mti_env;
1487 struct mdt_device *mdt = mti->mti_mdt;
1488 struct coordinator *cdt = &mdt->mdt_coordinator;
1489 struct mdt_object *obj = NULL;
1490 enum changelog_rec_flags clf_flags = 0;
1493 bool need_changelog = true;
1497 /* default is to retry */
1498 *status = ARS_WAITING;
1500 /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1501 * if error/removed continue anyway to get correct reporting done */
1502 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1503 /* we will update MD HSM only if needed */
1504 is_mh_changed = false;
1506 /* no need to change mh->mh_arch_id
1507 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1509 if (pgs->hpk_errval != 0) {
1510 switch (pgs->hpk_errval) {
1512 /* the copy tool does not support cancel
1513 * so the cancel request is failed
1514 * As we cannot distinguish a cancel progress
1515 * from another action progress (they have the
1516 * same cookie), we suppose here the CT returns
1517 * ENOSYS only if does not support cancel
1519 /* this can also happen when cdt calls it to
1520 * for a timed out request */
1521 *status = ARS_FAILED;
1522 /* to have a cancel event in changelog */
1523 pgs->hpk_errval = ECANCELED;
1526 /* the request record has already been set to
1527 * ARS_CANCELED, this set the cancel request
1529 *status = ARS_SUCCEED;
1532 /* retry only if current policy or requested, and
1533 * object is not on error/removed */
1534 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1535 !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1536 IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1540 rc = hsm_set_cl_error(&clf_flags, pgs->hpk_errval);
1541 if (rc == -EOVERFLOW) {
1542 CERROR("%s: Request %#llx on "DFID" failed, error code %d too large\n",
1543 mdt_obd_name(mdt), pgs->hpk_cookie,
1544 PFID(&pgs->hpk_fid), (int)abs(pgs->hpk_errval));
1548 switch (car->car_hai->hai_action) {
1550 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1553 hsm_set_cl_event(&clf_flags, HE_RESTORE);
1556 hsm_set_cl_event(&clf_flags, HE_REMOVE);
1559 hsm_set_cl_event(&clf_flags, HE_CANCEL);
1560 CERROR("%s: Failed request %#llx on "DFID
1561 " cannot be a CANCEL\n",
1564 PFID(&pgs->hpk_fid));
1567 CERROR("%s: Failed request %#llx on "DFID
1568 " %d is an unknown action\n",
1570 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1571 car->car_hai->hai_action);
1576 *status = ARS_SUCCEED;
1577 switch (car->car_hai->hai_action) {
1579 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1580 /* set ARCHIVE keep EXIST and clear LOST and
1582 mh.mh_arch_ver = pgs->hpk_data_version;
1583 mh.mh_flags |= HS_ARCHIVED;
1584 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1585 is_mh_changed = true;
1588 hsm_set_cl_event(&clf_flags, HE_RESTORE);
1590 /* do not clear RELEASED and DIRTY here
1591 * this will occur in hsm_swap_layouts()
1594 /* Restoring has changed the file version on
1596 mh.mh_arch_ver = pgs->hpk_data_version;
1597 is_mh_changed = true;
1600 hsm_set_cl_event(&clf_flags, HE_REMOVE);
1601 /* clear ARCHIVED EXISTS and LOST */
1602 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1603 is_mh_changed = true;
1606 hsm_set_cl_event(&clf_flags, HE_CANCEL);
1607 CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1610 PFID(&pgs->hpk_fid));
1613 CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1615 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1616 car->car_hai->hai_action);
1622 /* rc != 0 means error when analysing action, it may come from
1623 * a crasy CT no need to manage DIRTY
1624 * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1627 if (rc == 0 && !IS_ERR(obj))
1628 hsm_set_cl_flags(&clf_flags,
1629 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1631 /* unlock is done later, after layout lock management */
1632 if (is_mh_changed && !IS_ERR(obj))
1633 rc = mdt_hsm_attr_set(mti, obj, &mh);
1635 /* we give back layout lock only if restore was successful or
1636 * if no retry will be attempted and if object is still alive,
1637 * in other cases we just unlock the object */
1638 if (car->car_hai->hai_action == HSMA_RESTORE) {
1639 struct mdt_lock_handle *lh;
1641 /* restore in data FID done, we swap the layouts
1642 * only if restore is successful */
1643 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1644 rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1647 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1648 *status = ARS_FAILED;
1649 pgs->hpk_errval = -rc;
1650 hsm_set_cl_error(&clf_flags, pgs->hpk_errval);
1653 /* we have to retry, so keep layout lock */
1654 if (*status == ARS_WAITING)
1657 /* restore special case, need to create ChangeLog record
1658 * before to give back layout lock to avoid concurrent
1659 * file updater to post out of order ChangeLog */
1660 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1661 &car->car_hai->hai_fid);
1662 need_changelog = false;
1664 cdt_restore_handle_del(mti, cdt, &car->car_hai->hai_fid);
1665 if (!IS_ERR_OR_NULL(obj)) {
1666 /* flush UPDATE lock so attributes are upadated */
1667 lh = &mti->mti_lh[MDT_LH_OLD];
1668 mdt_object_lock(mti, obj, lh, MDS_INODELOCK_UPDATE,
1670 mdt_object_unlock(mti, obj, lh, 1);
1677 /* always add a ChangeLog record */
1679 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1680 &car->car_hai->hai_fid);
1683 mdt_object_put(mti->mti_env, obj);
1689 * update status of a request
1690 * \param mti [IN] context
1691 * \param pgs [IN] progress of the copy tool
1693 * \retval -ve failure
1695 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1696 struct hsm_progress_kernel *pgs)
1698 struct mdt_device *mdt = mti->mti_mdt;
1699 struct coordinator *cdt = &mdt->mdt_coordinator;
1700 struct cdt_agent_req *car;
1704 /* no coordinator started, so we cannot serve requests */
1705 if (!cdt_getref_try(cdt))
1708 /* first do sanity checks */
1709 car = mdt_cdt_update_request(cdt, pgs);
1711 CERROR("%s: Cannot find running request for cookie %#llx"
1714 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1716 GOTO(putref, rc = PTR_ERR(car));
1719 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1720 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1721 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1722 hsm_copytool_action2name(car->car_hai->hai_action),
1723 pgs->hpk_flags, pgs->hpk_errval,
1724 PFID(&car->car_hai->hai_fid),
1725 PFID(&car->car_hai->hai_dfid));
1727 /* progress is done on FID or data FID depending of the action and
1728 * of the copy progress */
1729 /* for restore progress is used to send back the data FID to cdt */
1730 if (car->car_hai->hai_action == HSMA_RESTORE &&
1731 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1732 car->car_hai->hai_dfid = pgs->hpk_fid;
1734 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1735 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1736 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1737 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1738 CERROR("%s: Progress on "DFID" for cookie %#llx"
1739 " does not match request FID "DFID" nor data FID "
1742 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1743 PFID(&car->car_hai->hai_fid),
1744 PFID(&car->car_hai->hai_dfid));
1745 GOTO(out, rc = -EINVAL);
1748 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1749 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1750 " is not coherent (err=%d and not completed"
1753 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1754 hsm_copytool_action2name(car->car_hai->hai_action),
1755 pgs->hpk_errval, pgs->hpk_flags);
1756 GOTO(out, rc = -EINVAL);
1759 /* now progress is valid */
1761 /* we use a root like ucred */
1762 hsm_init_ucred(mdt_ucred(mti));
1764 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1765 enum agent_req_status status;
1766 struct hsm_record_update update;
1769 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1771 CDEBUG(D_HSM, "updating record: fid="DFID" cookie=%#llx action=%s "
1773 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1774 hsm_copytool_action2name(car->car_hai->hai_action),
1775 agent_req_status2name(status));
1777 /* update record first (LU-9075) */
1778 update.cookie = pgs->hpk_cookie;
1779 update.status = status;
1781 rc1 = mdt_agent_record_update(mti, &update, 1);
1783 CERROR("%s: mdt_agent_record_update() failed,"
1784 " rc=%d, cannot update status to %s"
1785 " for cookie %#llx\n",
1786 mdt_obd_name(mdt), rc1,
1787 agent_req_status2name(status),
1789 rc = (rc != 0 ? rc : rc1);
1791 /* then remove request from memory list (LU-9075) */
1792 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1794 /* ct has completed a request, so a slot is available,
1795 * signal the coordinator to find new work */
1796 mdt_hsm_cdt_event(cdt);
1798 /* if copytool send a progress on a canceled request
1799 * we inform copytool it should stop
1801 if (car->car_canceled == 1)
1807 /* remove ref got from mdt_cdt_update_request() */
1808 mdt_cdt_put_request(car);
1817 * llog_cat_process() callback, used to:
1818 * - purge all requests
1819 * \param env [IN] environment
1820 * \param llh [IN] llog handle
1821 * \param hdr [IN] llog record
1822 * \param data [IN] cb data = struct mdt_thread_info
1824 * \retval -ve failure
1826 static int mdt_cancel_all_cb(const struct lu_env *env,
1827 struct llog_handle *llh,
1828 struct llog_rec_hdr *hdr, void *data)
1830 struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
1831 struct hsm_action_item *hai = &larr->arr_hai;
1832 struct mdt_thread_info *mti = data;
1833 struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
1837 if (larr->arr_status != ARS_WAITING &&
1838 larr->arr_status != ARS_STARTED)
1841 /* Unlock the EX layout lock */
1842 if (hai->hai_action == HSMA_RESTORE)
1843 cdt_restore_handle_del(mti, cdt, &hai->hai_fid);
1845 larr->arr_status = ARS_CANCELED;
1846 larr->arr_req_change = ktime_get_real_seconds();
1847 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1849 CERROR("%s: cannot update agent log: rc = %d\n",
1850 mdt_obd_name(mti->mti_mdt), rc);
1851 rc = LLOG_DEL_RECORD;
1858 * cancel all actions
1859 * \param obd [IN] MDT device
1861 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1864 struct lu_context session;
1865 struct mdt_thread_info *mti;
1866 struct coordinator *cdt = &mdt->mdt_coordinator;
1867 struct cdt_agent_req *car;
1868 struct hsm_action_list *hal = NULL;
1869 struct hsm_action_item *hai;
1870 int hal_sz = 0, hal_len, rc;
1871 enum cdt_states old_state;
1874 rc = lu_env_init(&env, LCT_MD_THREAD);
1878 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1879 rc = lu_context_init(&session, LCT_SERVER_SESSION);
1883 lu_context_enter(&session);
1884 env.le_ses = &session;
1886 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1887 LASSERT(mti != NULL);
1889 mti->mti_env = &env;
1892 hsm_init_ucred(mdt_ucred(mti));
1894 mutex_lock(&cdt->cdt_state_lock);
1895 old_state = cdt->cdt_state;
1897 /* disable coordinator */
1898 rc = set_cdt_state_locked(cdt, CDT_DISABLE);
1900 GOTO(out_cdt_state_unlock, rc);
1902 /* send cancel to all running requests */
1903 down_read(&cdt->cdt_request_lock);
1904 list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1905 mdt_cdt_get_request(car);
1906 /* request is not yet removed from list, it will be done
1907 * when copytool will return progress
1910 if (car->car_hai->hai_action == HSMA_CANCEL) {
1911 mdt_cdt_put_request(car);
1916 hal_len = sizeof(*hal) + round_up(MTI_NAME_MAXLEN + 1, 8) +
1917 round_up(car->car_hai->hai_len, 8);
1919 if (hal_len > hal_sz && hal_sz > 0) {
1920 /* not enough room, free old buffer */
1921 OBD_FREE(hal, hal_sz);
1925 /* empty buffer, allocate one */
1928 OBD_ALLOC(hal, hal_sz);
1930 mdt_cdt_put_request(car);
1931 up_read(&cdt->cdt_request_lock);
1932 GOTO(out_cdt_state, rc = -ENOMEM);
1936 hal->hal_version = HAL_VERSION;
1937 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1939 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1940 hal->hal_archive_id = car->car_archive_id;
1941 hal->hal_flags = car->car_flags;
1944 hai = hai_first(hal);
1945 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1946 hai->hai_action = HSMA_CANCEL;
1949 /* it is possible to safely call mdt_hsm_agent_send()
1950 * (ie without a deadlock on cdt_request_lock), because the
1951 * write lock is taken only if we are not in purge mode
1952 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1953 * nor mdt_cdt_remove_request())
1955 /* no conflict with cdt thread because cdt is disable and we
1956 * have the request lock */
1957 mdt_hsm_agent_send(mti, hal, 1);
1959 mdt_cdt_put_request(car);
1961 up_read(&cdt->cdt_request_lock);
1964 OBD_FREE(hal, hal_sz);
1966 /* cancel all on-disk records */
1967 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1968 (void *)mti, 0, 0, WRITE);
1970 /* Enable coordinator, unless the coordinator was stopping. */
1971 set_cdt_state_locked(cdt, old_state);
1972 out_cdt_state_unlock:
1973 mutex_unlock(&cdt->cdt_state_lock);
1975 lu_context_exit(&session);
1976 lu_context_fini(&session);
1984 * check if a request is compatible with file status
1985 * \param hai [IN] request description
1986 * \param archive_id [IN] request archive id
1987 * \param rq_flags [IN] request flags
1988 * \param hsm [IN] file HSM metadata
1991 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1992 u32 archive_id, u64 rq_flags,
1993 const struct md_hsm *hsm)
1995 int is_compat = false;
1999 hsm_flags = hsm->mh_flags;
2000 switch (hai->hai_action) {
2002 if (!(hsm_flags & HS_NOARCHIVE) &&
2003 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
2006 if (hsm_flags & HS_EXISTS &&
2008 archive_id != hsm->mh_arch_id)
2013 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
2014 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
2018 if (!(hsm_flags & HS_RELEASED) &&
2019 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
2026 CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
2027 " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
2028 PFID(&hai->hai_fid),
2029 hsm_copytool_action2name(hai->hai_action), rq_flags,
2030 hai->hai_extent.offset, hai->hai_extent.length,
2032 (is_compat ? "compatible" : "uncompatible"));
2038 * sysfs interface used to get/set HSM behaviour (cdt->cdt_policy)
2040 static const struct {
2044 } hsm_policy_names[] = {
2045 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
2046 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
2051 * convert a policy name to a bit
2052 * \param name [IN] policy name
2054 * \retval policy bit
2056 static __u64 hsm_policy_str2bit(const char *name)
2060 for (i = 0; hsm_policy_names[i].bit != 0; i++)
2061 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
2062 strcmp(hsm_policy_names[i].name, name) == 0)
2063 return hsm_policy_names[i].bit;
2068 * convert a policy bit field to a string
2069 * \param mask [IN] policy bit field
2070 * \param hexa [IN] print mask before bit names
2071 * \param buffer [OUT] string
2072 * \param count [IN] size of buffer
2074 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
2082 seq_printf(m, "(%#llx) ", mask);
2084 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
2087 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
2088 if (hsm_policy_names[j].bit == bit)
2092 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
2094 seq_printf(m, "%s ", hsm_policy_names[j].name);
2096 /* remove last ' ' */
2101 /* methods to read/write HSM policy flags */
2102 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
2104 struct mdt_device *mdt = m->private;
2105 struct coordinator *cdt = &mdt->mdt_coordinator;
2108 hsm_policy_bit2str(m, cdt->cdt_policy, false);
2113 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
2114 size_t count, loff_t *off)
2116 struct seq_file *m = file->private_data;
2117 struct mdt_device *mdt = m->private;
2118 struct coordinator *cdt = &mdt->mdt_coordinator;
2119 char *start, *token, sign;
2122 __u64 add_mask, remove_mask, set_mask;
2126 if (count + 1 > PAGE_SIZE)
2129 OBD_ALLOC(buf, count + 1);
2133 if (copy_from_user(buf, buffer, count))
2134 GOTO(out, rc = -EFAULT);
2139 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
2142 add_mask = remove_mask = set_mask = 0;
2144 token = strsep(&start, "\n ");
2150 if (sign == '-' || sign == '+')
2153 policy = hsm_policy_str2bit(token);
2155 CWARN("%s: '%s' is unknown, "
2156 "supported policies are:\n", mdt_obd_name(mdt),
2158 hsm_policy_bit2str(m, 0, false);
2159 GOTO(out, rc = -EINVAL);
2163 remove_mask |= policy;
2173 } while (start != NULL);
2175 CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
2176 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
2178 /* if no sign in all string, it is a clear and set
2179 * if some sign found, all unsigned are converted
2181 * P1 P2 = set to P1 and P2
2182 * P1 -P2 = add P1 clear P2 same as +P1 -P2
2184 if (remove_mask == 0 && add_mask == 0) {
2185 cdt->cdt_policy = set_mask;
2187 cdt->cdt_policy |= set_mask | add_mask;
2188 cdt->cdt_policy &= ~remove_mask;
2191 GOTO(out, rc = count);
2194 OBD_FREE(buf, count + 1);
2197 LDEBUGFS_SEQ_FOPS(mdt_hsm_policy);
2199 static ssize_t loop_period_show(struct kobject *kobj, struct attribute *attr,
2202 struct coordinator *cdt = container_of(kobj, struct coordinator,
2205 return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_loop_period);
2208 static ssize_t loop_period_store(struct kobject *kobj, struct attribute *attr,
2209 const char *buffer, size_t count)
2211 struct coordinator *cdt = container_of(kobj, struct coordinator,
2216 rc = kstrtouint(buffer, 0, &val);
2221 cdt->cdt_loop_period = val;
2223 return val ? count : -EINVAL;
2225 LUSTRE_RW_ATTR(loop_period);
2227 static ssize_t grace_delay_show(struct kobject *kobj, struct attribute *attr,
2230 struct coordinator *cdt = container_of(kobj, struct coordinator,
2233 return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_grace_delay);
2236 static ssize_t grace_delay_store(struct kobject *kobj, struct attribute *attr,
2237 const char *buffer, size_t count)
2239 struct coordinator *cdt = container_of(kobj, struct coordinator,
2244 rc = kstrtouint(buffer, 0, &val);
2249 cdt->cdt_grace_delay = val;
2251 return val ? count : -EINVAL;
2253 LUSTRE_RW_ATTR(grace_delay);
2255 static ssize_t active_request_timeout_show(struct kobject *kobj,
2256 struct attribute *attr,
2259 struct coordinator *cdt = container_of(kobj, struct coordinator,
2262 return scnprintf(buf, PAGE_SIZE, "%d\n", cdt->cdt_active_req_timeout);
2265 static ssize_t active_request_timeout_store(struct kobject *kobj,
2266 struct attribute *attr,
2267 const char *buffer, size_t count)
2269 struct coordinator *cdt = container_of(kobj, struct coordinator,
2274 rc = kstrtouint(buffer, 0, &val);
2279 cdt->cdt_active_req_timeout = val;
2281 return val ? count : -EINVAL;
2283 LUSTRE_RW_ATTR(active_request_timeout);
2285 static ssize_t max_requests_show(struct kobject *kobj, struct attribute *attr,
2288 struct coordinator *cdt = container_of(kobj, struct coordinator,
2291 return scnprintf(buf, PAGE_SIZE, "%llu\n", cdt->cdt_max_requests);
2294 static ssize_t max_requests_store(struct kobject *kobj, struct attribute *attr,
2295 const char *buffer, size_t count)
2297 struct coordinator *cdt = container_of(kobj, struct coordinator,
2299 unsigned long long val;
2302 rc = kstrtoull(buffer, 0, &val);
2307 cdt->cdt_max_requests = val;
2309 return val ? count : -EINVAL;
2311 LUSTRE_RW_ATTR(max_requests);
2313 static ssize_t default_archive_id_show(struct kobject *kobj,
2314 struct attribute *attr, char *buf)
2316 struct coordinator *cdt = container_of(kobj, struct coordinator,
2319 return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_default_archive_id);
2322 static ssize_t default_archive_id_store(struct kobject *kobj,
2323 struct attribute *attr,
2324 const char *buffer, size_t count)
2326 struct coordinator *cdt = container_of(kobj, struct coordinator,
2331 rc = kstrtouint(buffer, 0, &val);
2336 cdt->cdt_default_archive_id = val;
2338 return val ? count : -EINVAL;
2340 LUSTRE_RW_ATTR(default_archive_id);
2343 * procfs write method for MDT/hsm_control
2344 * proc entry is in mdt directory so data is mdt obd_device pointer
2346 #define CDT_ENABLE_CMD "enabled"
2347 #define CDT_STOP_CMD "shutdown"
2348 #define CDT_DISABLE_CMD "disabled"
2349 #define CDT_PURGE_CMD "purge"
2350 #define CDT_HELP_CMD "help"
2351 #define CDT_MAX_CMD_LEN 10
2353 ssize_t hsm_control_store(struct kobject *kobj, struct attribute *attr,
2354 const char *buffer, size_t count)
2356 struct obd_device *obd = container_of(kobj, struct obd_device,
2358 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2359 struct coordinator *cdt = &(mdt->mdt_coordinator);
2363 if (count == 0 || count >= CDT_MAX_CMD_LEN)
2366 if (strncmp(buffer, CDT_ENABLE_CMD, strlen(CDT_ENABLE_CMD)) == 0) {
2367 if (cdt->cdt_state == CDT_DISABLE) {
2368 rc = set_cdt_state(cdt, CDT_RUNNING);
2369 mdt_hsm_cdt_event(cdt);
2370 wake_up(&cdt->cdt_waitq);
2371 } else if (cdt->cdt_state == CDT_RUNNING) {
2374 rc = mdt_hsm_cdt_start(mdt);
2376 } else if (strncmp(buffer, CDT_STOP_CMD, strlen(CDT_STOP_CMD)) == 0) {
2377 if (cdt->cdt_state == CDT_STOPPING) {
2378 CERROR("%s: Coordinator is already stopping\n",
2381 } else if (cdt->cdt_state == CDT_STOPPED) {
2384 rc = mdt_hsm_cdt_stop(mdt);
2386 } else if (strncmp(buffer, CDT_DISABLE_CMD,
2387 strlen(CDT_DISABLE_CMD)) == 0) {
2388 if ((cdt->cdt_state == CDT_STOPPING) ||
2389 (cdt->cdt_state == CDT_STOPPED)) {
2390 /* exit gracefully if coordinator is being stopped
2391 * or stopped already.
2395 rc = set_cdt_state(cdt, CDT_DISABLE);
2397 } else if (strncmp(buffer, CDT_PURGE_CMD,
2398 strlen(CDT_PURGE_CMD)) == 0) {
2399 rc = hsm_cancel_all_actions(mdt);
2400 } else if (strncmp(buffer, CDT_HELP_CMD,
2401 strlen(CDT_HELP_CMD)) == 0) {
2409 CERROR("%s: Valid coordinator control commands are: "
2410 "%s %s %s %s %s\n", mdt_obd_name(mdt),
2411 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2412 CDT_PURGE_CMD, CDT_HELP_CMD);
2420 ssize_t hsm_control_show(struct kobject *kobj, struct attribute *attr,
2423 struct obd_device *obd = container_of(kobj, struct obd_device,
2425 struct coordinator *cdt;
2427 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2429 return scnprintf(buf, PAGE_SIZE, "%s\n",
2430 cdt_mdt_state2str(cdt->cdt_state));
2434 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2440 for (i = 0; i < 8 * sizeof(mask); i++) {
2441 if (mask & (1UL << i)) {
2442 seq_printf(m, "%s%s", first ? "" : " ",
2443 hsm_copytool_action2name(i));
2453 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2455 struct mdt_device *mdt = m->private;
2456 struct coordinator *cdt = &mdt->mdt_coordinator;
2458 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2462 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2464 struct mdt_device *mdt = m->private;
2465 struct coordinator *cdt = &mdt->mdt_coordinator;
2467 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2471 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2473 struct mdt_device *mdt = m->private;
2474 struct coordinator *cdt = &mdt->mdt_coordinator;
2476 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2479 static inline enum hsm_copytool_action
2480 hsm_copytool_name2action(const char *name)
2482 if (strcasecmp(name, "NOOP") == 0)
2484 else if (strcasecmp(name, "ARCHIVE") == 0)
2485 return HSMA_ARCHIVE;
2486 else if (strcasecmp(name, "RESTORE") == 0)
2487 return HSMA_RESTORE;
2488 else if (strcasecmp(name, "REMOVE") == 0)
2490 else if (strcasecmp(name, "CANCEL") == 0)
2497 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2498 size_t user_count, __u64 *mask)
2500 char *buf, *pos, *name;
2506 if (!(user_count < 4096))
2509 buf_size = user_count + 1;
2511 OBD_ALLOC(buf, buf_size);
2515 if (copy_from_user(buf, user_buf, buf_size - 1))
2516 GOTO(out, rc = -EFAULT);
2518 buf[buf_size - 1] = '\0';
2521 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2527 action = hsm_copytool_name2action(name);
2529 GOTO(out, rc = -EINVAL);
2531 new_mask |= (1UL << action);
2537 OBD_FREE(buf, buf_size);
2543 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2544 size_t count, loff_t *off)
2546 struct seq_file *m = file->private_data;
2547 struct mdt_device *mdt = m->private;
2548 struct coordinator *cdt = &mdt->mdt_coordinator;
2550 return mdt_write_hsm_request_mask(file, buf, count,
2551 &cdt->cdt_user_request_mask);
2555 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2556 size_t count, loff_t *off)
2558 struct seq_file *m = file->private_data;
2559 struct mdt_device *mdt = m->private;
2560 struct coordinator *cdt = &mdt->mdt_coordinator;
2562 return mdt_write_hsm_request_mask(file, buf, count,
2563 &cdt->cdt_group_request_mask);
2567 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2568 size_t count, loff_t *off)
2570 struct seq_file *m = file->private_data;
2571 struct mdt_device *mdt = m->private;
2572 struct coordinator *cdt = &mdt->mdt_coordinator;
2574 return mdt_write_hsm_request_mask(file, buf, count,
2575 &cdt->cdt_other_request_mask);
2578 static ssize_t remove_archive_on_last_unlink_show(struct kobject *kobj,
2579 struct attribute *attr,
2582 struct coordinator *cdt = container_of(kobj, struct coordinator,
2585 return scnprintf(buf, PAGE_SIZE, "%u\n",
2586 cdt->cdt_remove_archive_on_last_unlink);
2589 static ssize_t remove_archive_on_last_unlink_store(struct kobject *kobj,
2590 struct attribute *attr,
2594 struct coordinator *cdt = container_of(kobj, struct coordinator,
2599 rc = kstrtobool(buffer, &val);
2603 cdt->cdt_remove_archive_on_last_unlink = val;
2606 LUSTRE_RW_ATTR(remove_archive_on_last_unlink);
2608 LDEBUGFS_SEQ_FOPS(mdt_hsm_user_request_mask);
2609 LDEBUGFS_SEQ_FOPS(mdt_hsm_group_request_mask);
2610 LDEBUGFS_SEQ_FOPS(mdt_hsm_other_request_mask);
2612 /* Read-only sysfs files for request counters */
2613 static ssize_t archive_count_show(struct kobject *kobj, struct attribute *attr,
2616 struct coordinator *cdt = container_of(kobj, struct coordinator,
2619 return scnprintf(buf, PAGE_SIZE, "%d\n",
2620 atomic_read(&cdt->cdt_archive_count));
2622 LUSTRE_RO_ATTR(archive_count);
2624 static ssize_t restore_count_show(struct kobject *kobj, struct attribute *attr,
2627 struct coordinator *cdt = container_of(kobj, struct coordinator,
2630 return scnprintf(buf, PAGE_SIZE, "%d\n",
2631 atomic_read(&cdt->cdt_restore_count));
2633 LUSTRE_RO_ATTR(restore_count);
2635 static ssize_t remove_count_show(struct kobject *kobj, struct attribute *attr,
2638 struct coordinator *cdt = container_of(kobj, struct coordinator,
2641 return scnprintf(buf, PAGE_SIZE, "%d\n",
2642 atomic_read(&cdt->cdt_remove_count));
2644 LUSTRE_RO_ATTR(remove_count);
2646 static struct ldebugfs_vars ldebugfs_mdt_hsm_vars[] = {
2648 .fops = &mdt_hsm_agent_fops },
2649 { .name = "actions",
2650 .fops = &mdt_hsm_actions_fops,
2651 .proc_mode = 0444 },
2653 .fops = &mdt_hsm_policy_fops },
2654 { .name = "active_requests",
2655 .fops = &mdt_hsm_active_requests_fops },
2656 { .name = "user_request_mask",
2657 .fops = &mdt_hsm_user_request_mask_fops, },
2658 { .name = "group_request_mask",
2659 .fops = &mdt_hsm_group_request_mask_fops, },
2660 { .name = "other_request_mask",
2661 .fops = &mdt_hsm_other_request_mask_fops, },
2665 static struct attribute *hsm_attrs[] = {
2666 &lustre_attr_loop_period.attr,
2667 &lustre_attr_grace_delay.attr,
2668 &lustre_attr_active_request_timeout.attr,
2669 &lustre_attr_max_requests.attr,
2670 &lustre_attr_default_archive_id.attr,
2671 &lustre_attr_remove_archive_on_last_unlink.attr,
2672 &lustre_attr_archive_count.attr,
2673 &lustre_attr_restore_count.attr,
2674 &lustre_attr_remove_count.attr,
2678 KOBJ_ATTRIBUTE_GROUPS(hsm); /* creates hsm_groups from hsm_attrs */
2680 static void hsm_kobj_release(struct kobject *kobj)
2682 struct coordinator *cdt = container_of(kobj, struct coordinator,
2685 debugfs_remove_recursive(cdt->cdt_debugfs_dir);
2686 cdt->cdt_debugfs_dir = NULL;
2688 complete(&cdt->cdt_kobj_unregister);
2691 static struct kobj_type hsm_ktype = {
2692 .default_groups = KOBJ_ATTR_GROUPS(hsm),
2693 .sysfs_ops = &lustre_sysfs_ops,
2694 .release = hsm_kobj_release,
2698 * create sysfs entries for coordinator
2701 * \retval -ve failure
2703 int hsm_cdt_tunables_init(struct mdt_device *mdt)
2705 struct coordinator *cdt = &mdt->mdt_coordinator;
2706 struct obd_device *obd = mdt2obd_dev(mdt);
2709 init_completion(&cdt->cdt_kobj_unregister);
2710 rc = kobject_init_and_add(&cdt->cdt_hsm_kobj, &hsm_ktype,
2711 &obd->obd_kset.kobj, "%s", "hsm");
2713 kobject_put(&cdt->cdt_hsm_kobj);
2717 /* init debugfs entries, failure is not critical */
2718 cdt->cdt_debugfs_dir = debugfs_create_dir("hsm",
2719 obd->obd_debugfs_entry);
2720 ldebugfs_add_vars(cdt->cdt_debugfs_dir, ldebugfs_mdt_hsm_vars, mdt);
2726 * remove sysfs entries for coordinator
2730 void hsm_cdt_tunables_fini(struct mdt_device *mdt)
2732 struct coordinator *cdt = &mdt->mdt_coordinator;
2734 kobject_put(&cdt->cdt_hsm_kobj);
2735 wait_for_completion(&cdt->cdt_kobj_unregister);