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 <obd_support.h>
43 #include <lustre_export.h>
45 #include <lprocfs_status.h>
46 #include <lustre_log.h>
47 #include <lustre_kernelcomm.h>
48 #include "mdt_internal.h"
50 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
53 * get obj and HSM attributes on a fid
54 * \param mti [IN] context
55 * \param fid [IN] object fid
56 * \param hsm [OUT] HSM meta data
57 * \retval obj or error (-ENOENT if not found)
59 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
60 const struct lu_fid *fid,
64 struct mdt_object *obj;
72 /* find object by FID */
73 obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
77 if (!mdt_object_exists(obj)) {
79 mdt_object_put(mti->mti_env, obj);
80 RETURN(ERR_PTR(-ENOENT));
83 rc = mdt_attr_get_complex(mti, obj, ma);
85 mdt_object_put(mti->mti_env, obj);
89 if (ma->ma_valid & MA_HSM)
92 memset(hsm, 0, sizeof(*hsm));
97 void mdt_hsm_dump_hal(int level, const char *prefix,
98 struct hsm_action_list *hal)
101 struct hsm_action_item *hai;
104 CDEBUG(level, "%s: HAL header: version %X count %d"
105 " archive_id %d flags %#llx\n",
106 prefix, hal->hal_version, hal->hal_count,
107 hal->hal_archive_id, hal->hal_flags);
109 hai = hai_first(hal);
110 for (i = 0; i < hal->hal_count; i++) {
111 sz = hai->hai_len - sizeof(*hai);
112 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
114 " action=%s extent=%#llx-%#llx gid=%#llx"
115 " datalen=%d data=[%s]\n",
117 PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
119 hsm_copytool_action2name(hai->hai_action),
120 hai->hai_extent.offset,
121 hai->hai_extent.length,
123 hai_dump_data_field(hai, buf, sizeof(buf)));
129 * data passed to llog_cat_process() callback
130 * to scan requests and take actions
132 struct hsm_scan_request {
135 struct hsm_action_list *hal;
138 struct hsm_scan_data {
139 struct mdt_thread_info *hsd_mti;
140 char hsd_fsname[MTI_NAME_MAXLEN + 1];
141 /* are we scanning the logs for housekeeping, or just looking
144 bool hsd_housekeeping;
145 bool hsd_one_restore;
146 int hsd_action_count;
147 int hsd_request_len; /* array alloc len */
148 int hsd_request_count; /* array used count */
149 struct hsm_scan_request *hsd_request;
152 static int mdt_cdt_waiting_cb(const struct lu_env *env,
153 struct mdt_device *mdt,
154 struct llog_handle *llh,
155 struct llog_agent_req_rec *larr,
156 struct hsm_scan_data *hsd)
158 struct coordinator *cdt = &mdt->mdt_coordinator;
159 struct hsm_scan_request *request;
160 struct hsm_action_item *hai;
165 /* Are agents full? */
166 if (atomic_read(&cdt->cdt_request_count) >= cdt->cdt_max_requests)
167 RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
169 if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
170 cdt->cdt_max_requests) {
171 /* We cannot send any more request
173 * *** SPECIAL CASE ***
175 * Restore requests are too important not to schedule at least
176 * one, everytime we can.
178 if (larr->arr_hai.hai_action != HSMA_RESTORE ||
179 hsd->hsd_one_restore)
180 RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
183 hai_size = cfs_size_round(larr->arr_hai.hai_len);
184 archive_id = larr->arr_archive_id;
186 /* Can we add this action to one of the existing HALs in hsd. */
188 for (i = 0; i < hsd->hsd_request_count; i++) {
189 if (hsd->hsd_request[i].hal->hal_archive_id == archive_id &&
190 hsd->hsd_request[i].hal_used_sz + hai_size <=
192 request = &hsd->hsd_request[i];
197 /* Are we trying to force-schedule a request? */
198 if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
199 cdt->cdt_max_requests) {
200 /* Is there really no compatible hsm_scan_request? */
202 for (i -= 1; i >= 0; i--) {
203 if (hsd->hsd_request[i].hal->hal_archive_id ==
205 request = &hsd->hsd_request[i];
211 /* Make room for the hai */
213 /* Discard the last hai until there is enough space */
215 request->hal->hal_count--;
217 hai = hai_first(request->hal);
218 for (i = 0; i < request->hal->hal_count; i++)
220 request->hal_used_sz -=
221 cfs_size_round(hai->hai_len);
222 hsd->hsd_action_count--;
223 } while (request->hal_used_sz + hai_size >
225 } else if (hsd->hsd_housekeeping) {
226 struct hsm_scan_request *tmp;
228 /* Discard the (whole) last hal */
229 hsd->hsd_request_count--;
230 LASSERT(hsd->hsd_request_count >= 0);
231 tmp = &hsd->hsd_request[hsd->hsd_request_count];
232 hsd->hsd_action_count -= tmp->hal->hal_count;
233 LASSERT(hsd->hsd_action_count >= 0);
234 OBD_FREE(tmp->hal, tmp->hal_sz);
236 /* Bailing out, this code path is too hot */
237 RETURN(LLOG_PROC_BREAK);
243 struct hsm_action_list *hal;
245 LASSERT(hsd->hsd_request_count < hsd->hsd_request_len);
246 request = &hsd->hsd_request[hsd->hsd_request_count];
248 /* allocates hai vector size just needs to be large
250 request->hal_sz = sizeof(*request->hal) +
251 cfs_size_round(MTI_NAME_MAXLEN + 1) + 2 * hai_size;
252 OBD_ALLOC_LARGE(hal, request->hal_sz);
256 hal->hal_version = HAL_VERSION;
257 strlcpy(hal->hal_fsname, hsd->hsd_fsname, MTI_NAME_MAXLEN + 1);
258 hal->hal_archive_id = larr->arr_archive_id;
259 hal->hal_flags = larr->arr_flags;
261 request->hal_used_sz = hal_size(hal);
263 hsd->hsd_request_count++;
264 } else if (request->hal_sz < request->hal_used_sz + hai_size) {
265 /* Not enough room, need an extension */
269 sz = min_t(int, 2 * request->hal_sz, LDLM_MAXREQSIZE);
270 LASSERT(request->hal_used_sz + hai_size < sz);
272 OBD_ALLOC_LARGE(hal_buffer, sz);
276 memcpy(hal_buffer, request->hal, request->hal_used_sz);
277 OBD_FREE_LARGE(request->hal, request->hal_sz);
278 request->hal = hal_buffer;
279 request->hal_sz = sz;
282 hai = hai_first(request->hal);
283 for (i = 0; i < request->hal->hal_count; i++)
286 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
288 request->hal_used_sz += hai_size;
289 request->hal->hal_count++;
291 hsd->hsd_action_count++;
293 switch (hai->hai_action) {
297 hsd->hsd_one_restore = true;
298 /* Intentional fallthrough */
300 cdt_agent_record_hash_add(cdt, hai->hai_cookie,
301 llh->lgh_hdr->llh_cat_idx,
302 larr->arr_hdr.lrh_index);
308 static int mdt_cdt_started_cb(const struct lu_env *env,
309 struct mdt_device *mdt,
310 struct llog_handle *llh,
311 struct llog_agent_req_rec *larr,
312 struct hsm_scan_data *hsd)
314 struct coordinator *cdt = &mdt->mdt_coordinator;
315 struct hsm_action_item *hai = &larr->arr_hai;
316 struct cdt_agent_req *car;
317 time64_t now = ktime_get_real_seconds();
319 enum changelog_rec_flags clf_flags;
322 if (!hsd->hsd_housekeeping)
325 /* we search for a running request
326 * error may happen if coordinator crashes or stopped
327 * with running request
329 car = mdt_cdt_find_request(cdt, hai->hai_cookie);
331 last = larr->arr_req_change;
333 last = car->car_req_update;
336 /* test if request too long, if yes cancel it
337 * the same way the copy tool acknowledge a cancel request */
338 if (now <= last + cdt->cdt_active_req_timeout)
339 GOTO(out_car, rc = 0);
341 dump_llog_agent_req_rec("request timed out, start cleaning", larr);
344 car->car_req_update = now;
345 mdt_hsm_agent_update_statistics(cdt, 0, 1, 0, &car->car_uuid);
346 /* Remove car from memory list (LU-9075) */
347 mdt_cdt_remove_request(cdt, hai->hai_cookie);
350 /* Emit a changelog record for the failed action.*/
352 hsm_set_cl_error(&clf_flags, ECANCELED);
354 switch (hai->hai_action) {
356 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
359 hsm_set_cl_event(&clf_flags, HE_RESTORE);
362 hsm_set_cl_event(&clf_flags, HE_REMOVE);
365 hsm_set_cl_event(&clf_flags, HE_CANCEL);
368 /* Unknown record type, skip changelog. */
374 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
377 if (hai->hai_action == HSMA_RESTORE)
378 cdt_restore_handle_del(hsd->hsd_mti, cdt, &hai->hai_fid);
380 larr->arr_status = ARS_CANCELED;
381 larr->arr_req_change = now;
382 rc = llog_write(hsd->hsd_mti->mti_env, llh, &larr->arr_hdr,
383 larr->arr_hdr.lrh_index);
385 CERROR("%s: cannot update agent log: rc = %d\n",
386 mdt_obd_name(mdt), rc);
387 rc = LLOG_DEL_RECORD;
390 /* ct has completed a request, so a slot is available,
391 * signal the coordinator to find new work */
392 mdt_hsm_cdt_event(cdt);
395 mdt_cdt_put_request(car);
401 * llog_cat_process() callback, used to:
402 * - find waiting request and start action
403 * - purge canceled and done requests
404 * \param env [IN] environment
405 * \param llh [IN] llog handle
406 * \param hdr [IN] llog record
407 * \param data [IN/OUT] cb data = struct hsm_scan_data
409 * \retval -ve failure
411 static int mdt_coordinator_cb(const struct lu_env *env,
412 struct llog_handle *llh,
413 struct llog_rec_hdr *hdr,
416 struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
417 struct hsm_scan_data *hsd = data;
418 struct mdt_device *mdt = hsd->hsd_mti->mti_mdt;
419 struct coordinator *cdt = &mdt->mdt_coordinator;
422 larr = (struct llog_agent_req_rec *)hdr;
423 dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
424 switch (larr->arr_status) {
426 RETURN(mdt_cdt_waiting_cb(env, mdt, llh, larr, hsd));
428 RETURN(mdt_cdt_started_cb(env, mdt, llh, larr, hsd));
430 if (!hsd->hsd_housekeeping)
433 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
434 ktime_get_real_seconds()) {
435 cdt_agent_record_hash_del(cdt,
436 larr->arr_hai.hai_cookie);
437 RETURN(LLOG_DEL_RECORD);
445 * create /proc entries for coordinator
448 * \retval -ve failure
450 int hsm_cdt_procfs_init(struct mdt_device *mdt)
452 struct coordinator *cdt = &mdt->mdt_coordinator;
456 /* init /proc entries, failure is not critical */
457 cdt->cdt_proc_dir = lprocfs_register("hsm",
458 mdt2obd_dev(mdt)->obd_proc_entry,
459 lprocfs_mdt_hsm_vars, mdt);
460 if (IS_ERR(cdt->cdt_proc_dir)) {
461 rc = PTR_ERR(cdt->cdt_proc_dir);
462 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
463 " rc=%d\n", mdt_obd_name(mdt), rc);
464 cdt->cdt_proc_dir = NULL;
472 * remove /proc entries for coordinator
475 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
477 struct coordinator *cdt = &mdt->mdt_coordinator;
479 if (cdt->cdt_proc_dir != NULL)
480 lprocfs_remove(&cdt->cdt_proc_dir);
484 * get vector of hsm cdt /proc vars
488 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
490 return lprocfs_mdt_hsm_vars;
493 /* Release the ressource used by the coordinator. Called when the
494 * coordinator is stopping. */
495 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
497 struct coordinator *cdt = &mdt->mdt_coordinator;
498 struct cdt_agent_req *car, *tmp1;
499 struct hsm_agent *ha, *tmp2;
500 struct cdt_restore_handle *crh, *tmp3;
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(ha->ha_archive_id, ha->ha_archive_cnt *
520 sizeof(*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 mutex_lock(&cdt->cdt_restore_lock);
527 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_handle_list,
529 list_del(&crh->crh_list);
530 /* give back layout lock */
531 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
532 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
534 mutex_unlock(&cdt->cdt_restore_lock);
538 * Coordinator state transition table, indexed on enum cdt_states, taking
539 * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
540 * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
542 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
543 /* from -> to: stopped init running disable stopping */
544 /* stopped */ { true, true, false, false, false },
545 /* init */ { true, false, true, false, false },
546 /* running */ { false, false, true, true, true },
547 /* disable */ { false, false, true, true, true },
548 /* stopping */ { true, false, false, false, false }
552 * Change coordinator thread state
553 * Some combinations are not valid, so catch them here.
555 * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
556 * the transition was not possible.
558 static int set_cdt_state_locked(struct coordinator *cdt,
559 enum cdt_states new_state)
562 enum cdt_states state;
564 state = cdt->cdt_state;
566 if (cdt_transition[state][new_state]) {
567 cdt->cdt_state = new_state;
571 "unexpected coordinator transition, from=%s, to=%s\n",
572 cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
579 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state)
583 mutex_lock(&cdt->cdt_state_lock);
584 rc = set_cdt_state_locked(cdt, new_state);
585 mutex_unlock(&cdt->cdt_state_lock);
594 * \param data [IN] obd device
596 * \retval -ve failure
598 static int mdt_coordinator(void *data)
600 struct mdt_thread_info *mti = data;
601 struct mdt_device *mdt = mti->mti_mdt;
602 struct coordinator *cdt = &mdt->mdt_coordinator;
603 struct hsm_scan_data hsd = { NULL };
604 time64_t last_housekeeping = 0;
605 size_t request_sz = 0;
609 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
610 mdt_obd_name(mdt), current_pid());
613 obd_uuid2fsname(hsd.hsd_fsname, mdt_obd_name(mdt),
614 sizeof(hsd.hsd_fsname));
616 set_cdt_state(cdt, CDT_RUNNING);
618 /* Inform mdt_hsm_cdt_start(). */
619 wake_up_all(&cdt->cdt_waitq);
626 struct hsm_record_update *updates;
628 /* Limit execution of the expensive requests traversal
629 * to at most one second. This prevents repeatedly
630 * locking/unlocking the catalog for each request
631 * and preventing other HSM operations from happening
633 wait_event_interruptible_timeout(cdt->cdt_waitq,
634 kthread_should_stop() ||
635 cdt->cdt_wakeup_coordinator,
636 cfs_time_seconds(1));
638 cdt->cdt_wakeup_coordinator = false;
639 CDEBUG(D_HSM, "coordinator resumes\n");
641 if (kthread_should_stop()) {
642 CDEBUG(D_HSM, "Coordinator stops\n");
647 /* if coordinator is suspended continue to wait */
648 if (cdt->cdt_state == CDT_DISABLE) {
649 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
653 /* If no event, and no housekeeping to do, continue to
655 if (last_housekeeping + cdt->cdt_loop_period <=
656 ktime_get_real_seconds()) {
657 last_housekeeping = ktime_get_real_seconds();
658 hsd.hsd_housekeeping = true;
659 } else if (cdt->cdt_event) {
660 hsd.hsd_housekeeping = false;
665 cdt->cdt_event = false;
667 CDEBUG(D_HSM, "coordinator starts reading llog\n");
669 if (hsd.hsd_request_len != cdt->cdt_max_requests) {
670 /* cdt_max_requests has changed,
671 * we need to allocate a new buffer
673 struct hsm_scan_request *tmp = NULL;
674 int max_requests = cdt->cdt_max_requests;
675 OBD_ALLOC_LARGE(tmp, max_requests *
676 sizeof(struct hsm_scan_request));
678 CERROR("Failed to resize request buffer, "
679 "keeping it at %d\n",
680 hsd.hsd_request_len);
682 if (hsd.hsd_request != NULL)
683 OBD_FREE_LARGE(hsd.hsd_request,
686 hsd.hsd_request_len = max_requests;
687 request_sz = hsd.hsd_request_len *
688 sizeof(struct hsm_scan_request);
689 hsd.hsd_request = tmp;
693 hsd.hsd_action_count = 0;
694 hsd.hsd_request_count = 0;
695 hsd.hsd_one_restore = false;
697 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
702 CDEBUG(D_HSM, "found %d requests to send\n",
703 hsd.hsd_request_count);
705 if (list_empty(&cdt->cdt_agents)) {
706 CDEBUG(D_HSM, "no agent available, "
707 "coordinator sleeps\n");
711 /* Compute how many HAI we have in all the requests */
713 for (i = 0; i < hsd.hsd_request_count; i++) {
714 const struct hsm_scan_request *request =
717 updates_cnt += request->hal->hal_count;
720 /* Allocate a temporary array to store the cookies to
721 * update, and their status. */
722 updates_sz = updates_cnt * sizeof(*updates);
723 OBD_ALLOC_LARGE(updates, updates_sz);
724 if (updates == NULL) {
725 CERROR("%s: Cannot allocate memory (%d o) "
727 mdt_obd_name(mdt), updates_sz, updates_cnt);
731 /* here hsd contains a list of requests to be started */
732 for (i = 0; i < hsd.hsd_request_count; i++) {
733 struct hsm_scan_request *request = &hsd.hsd_request[i];
734 struct hsm_action_list *hal = request->hal;
735 struct hsm_action_item *hai;
738 /* still room for work ? */
739 if (atomic_read(&cdt->cdt_request_count) >=
740 cdt->cdt_max_requests)
743 rc = mdt_hsm_agent_send(mti, hal, 0);
744 /* if failure, we suppose it is temporary
745 * if the copy tool failed to do the request
746 * it has to use hsm_progress
749 /* set up cookie vector to set records status
750 * after copy tools start or failed
752 hai = hai_first(hal);
753 for (j = 0; j < hal->hal_count; j++) {
754 updates[update_idx].cookie = hai->hai_cookie;
755 updates[update_idx].status =
756 (rc ? ARS_WAITING : ARS_STARTED);
763 rc = mdt_agent_record_update(mti->mti_env, mdt,
764 updates, update_idx);
766 CERROR("%s: mdt_agent_record_update() failed, "
767 "rc=%d, cannot update records "
769 mdt_obd_name(mdt), rc, update_idx);
772 OBD_FREE_LARGE(updates, updates_sz);
775 /* free hal allocated by callback */
776 for (i = 0; i < hsd.hsd_request_count; i++) {
777 struct hsm_scan_request *request = &hsd.hsd_request[i];
779 OBD_FREE_LARGE(request->hal, request->hal_sz);
783 if (hsd.hsd_request != NULL)
784 OBD_FREE_LARGE(hsd.hsd_request, request_sz);
786 mdt_hsm_cdt_cleanup(mdt);
789 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
790 mdt_obd_name(mdt), current_pid(), rc);
792 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
794 mdt_obd_name(mdt), current_pid());
799 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
800 const struct lu_fid *fid,
801 const struct hsm_extent *he)
803 struct cdt_restore_handle *crh;
804 struct mdt_object *obj;
808 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
813 /* in V1 all file is restored
814 * crh->extent.start = he->offset;
815 * crh->extent.end = he->offset + he->length;
817 crh->crh_extent.start = 0;
818 crh->crh_extent.end = he->length;
819 /* get the layout lock */
820 mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
821 obj = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
822 MDS_INODELOCK_LAYOUT);
824 GOTO(out_crh, rc = PTR_ERR(obj));
826 /* We do not keep a reference on the object during the restore
827 * which can be very long. */
828 mdt_object_put(mti->mti_env, obj);
830 mutex_lock(&cdt->cdt_restore_lock);
831 if (unlikely(cdt->cdt_state == CDT_STOPPED ||
832 cdt->cdt_state == CDT_STOPPING)) {
833 mutex_unlock(&cdt->cdt_restore_lock);
834 GOTO(out_lh, rc = -EAGAIN);
837 list_add_tail(&crh->crh_list, &cdt->cdt_restore_handle_list);
838 mutex_unlock(&cdt->cdt_restore_lock);
842 mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
844 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
850 * lookup a restore handle by FID
851 * caller needs to hold cdt_restore_lock
852 * \param cdt [IN] coordinator
853 * \param fid [IN] FID
854 * \retval cdt_restore_handle found
855 * \retval NULL not found
857 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
858 const struct lu_fid *fid)
860 struct cdt_restore_handle *crh;
863 list_for_each_entry(crh, &cdt->cdt_restore_handle_list, crh_list) {
864 if (lu_fid_eq(&crh->crh_fid, fid))
871 void cdt_restore_handle_del(struct mdt_thread_info *mti,
872 struct coordinator *cdt, const struct lu_fid *fid)
874 struct cdt_restore_handle *crh;
876 /* give back layout lock */
877 mutex_lock(&cdt->cdt_restore_lock);
878 crh = cdt_restore_handle_find(cdt, fid);
880 list_del(&crh->crh_list);
881 mutex_unlock(&cdt->cdt_restore_lock);
886 /* XXX We pass a NULL object since the restore handle does not
887 * keep a reference on the object being restored. */
888 mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
889 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
893 * data passed to llog_cat_process() callback
894 * to scan requests and take actions
896 struct hsm_restore_data {
897 struct mdt_thread_info *hrd_mti;
901 * llog_cat_process() callback, used to:
902 * - find restore request and allocate the restore handle
903 * \param env [IN] environment
904 * \param llh [IN] llog handle
905 * \param hdr [IN] llog record
906 * \param data [IN/OUT] cb data = struct hsm_restore_data
908 * \retval -ve failure
910 static int hsm_restore_cb(const struct lu_env *env,
911 struct llog_handle *llh,
912 struct llog_rec_hdr *hdr, void *data)
914 struct llog_agent_req_rec *larr;
915 struct hsm_restore_data *hrd;
916 struct hsm_action_item *hai;
917 struct mdt_thread_info *mti;
918 struct coordinator *cdt;
924 cdt = &mti->mti_mdt->mdt_coordinator;
926 larr = (struct llog_agent_req_rec *)hdr;
927 hai = &larr->arr_hai;
928 if (hai->hai_cookie > cdt->cdt_last_cookie)
929 /* update the cookie to avoid collision */
930 cdt->cdt_last_cookie = hai->hai_cookie + 1;
932 if (hai->hai_action != HSMA_RESTORE ||
933 agent_req_in_final_state(larr->arr_status))
936 /* restore request not in a final state */
938 /* force replay of restore requests left in started state from previous
939 * CDT context, to be canceled later if finally found to be incompatible
940 * when being re-started */
941 if (larr->arr_status == ARS_STARTED) {
942 larr->arr_status = ARS_WAITING;
943 larr->arr_req_change = ktime_get_real_seconds();
944 rc = llog_write(env, llh, hdr, hdr->lrh_index);
949 rc = cdt_restore_handle_add(mti, cdt, &hai->hai_fid, &hai->hai_extent);
955 * restore coordinator state at startup
956 * the goal is to take a layout lock for each registered restore request
957 * \param mti [IN] context
959 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
961 struct hsm_restore_data hrd;
967 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
973 static int hsm_init_ucred(struct lu_ucred *uc)
977 uc->uc_valid = UCRED_OLD;
986 uc->uc_suppgids[0] = -1;
987 uc->uc_suppgids[1] = -1;
988 uc->uc_cap = CFS_CAP_FS_MASK;
991 uc->uc_identity = NULL;
992 /* always record internal HSM activity if also enabled globally */
993 uc->uc_enable_audit = 1;
999 * initialize coordinator struct
1000 * \param mdt [IN] device
1002 * \retval -ve failure
1004 int mdt_hsm_cdt_init(struct mdt_device *mdt)
1006 struct coordinator *cdt = &mdt->mdt_coordinator;
1007 struct mdt_thread_info *cdt_mti = NULL;
1011 init_waitqueue_head(&cdt->cdt_waitq);
1012 init_rwsem(&cdt->cdt_llog_lock);
1013 init_rwsem(&cdt->cdt_agent_lock);
1014 init_rwsem(&cdt->cdt_request_lock);
1015 mutex_init(&cdt->cdt_restore_lock);
1016 mutex_init(&cdt->cdt_state_lock);
1017 set_cdt_state(cdt, CDT_STOPPED);
1019 INIT_LIST_HEAD(&cdt->cdt_request_list);
1020 INIT_LIST_HEAD(&cdt->cdt_agents);
1021 INIT_LIST_HEAD(&cdt->cdt_restore_handle_list);
1023 cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
1027 0 /* extra bytes */,
1030 &cdt_request_cookie_hash_ops,
1032 if (cdt->cdt_request_cookie_hash == NULL)
1035 cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
1039 0 /* extra bytes */,
1042 &cdt_agent_record_hash_ops,
1044 if (cdt->cdt_agent_record_hash == NULL)
1045 GOTO(out_request_cookie_hash, rc = -ENOMEM);
1047 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
1049 GOTO(out_agent_record_hash, rc);
1051 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1052 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
1056 lu_context_enter(&cdt->cdt_session);
1057 cdt->cdt_env.le_ses = &cdt->cdt_session;
1059 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1060 LASSERT(cdt_mti != NULL);
1062 cdt_mti->mti_env = &cdt->cdt_env;
1063 cdt_mti->mti_mdt = mdt;
1065 hsm_init_ucred(mdt_ucred(cdt_mti));
1067 /* default values for /proc tunnables
1068 * can be override by MGS conf */
1069 cdt->cdt_default_archive_id = 1;
1070 cdt->cdt_grace_delay = 60;
1071 cdt->cdt_loop_period = 10;
1072 cdt->cdt_max_requests = 3;
1073 cdt->cdt_policy = CDT_DEFAULT_POLICY;
1074 cdt->cdt_active_req_timeout = 3600;
1076 /* by default do not remove archives on last unlink */
1077 cdt->cdt_remove_archive_on_last_unlink = false;
1082 lu_env_fini(&cdt->cdt_env);
1083 out_agent_record_hash:
1084 cfs_hash_putref(cdt->cdt_agent_record_hash);
1085 cdt->cdt_agent_record_hash = NULL;
1086 out_request_cookie_hash:
1087 cfs_hash_putref(cdt->cdt_request_cookie_hash);
1088 cdt->cdt_request_cookie_hash = NULL;
1094 * free a coordinator thread
1095 * \param mdt [IN] device
1097 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
1099 struct coordinator *cdt = &mdt->mdt_coordinator;
1102 lu_context_exit(cdt->cdt_env.le_ses);
1103 lu_context_fini(cdt->cdt_env.le_ses);
1105 lu_env_fini(&cdt->cdt_env);
1107 cfs_hash_putref(cdt->cdt_agent_record_hash);
1108 cdt->cdt_agent_record_hash = NULL;
1110 cfs_hash_putref(cdt->cdt_request_cookie_hash);
1111 cdt->cdt_request_cookie_hash = NULL;
1117 * start a coordinator thread
1118 * \param mdt [IN] device
1120 * \retval -ve failure
1122 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1124 struct coordinator *cdt = &mdt->mdt_coordinator;
1125 struct mdt_thread_info *cdt_mti;
1128 struct task_struct *task;
1131 /* functions defined but not yet used
1132 * this avoid compilation warning
1134 ptr = dump_requests;
1136 rc = set_cdt_state(cdt, CDT_INIT);
1138 CERROR("%s: Coordinator already started or stopping\n",
1143 CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
1144 cdt->cdt_policy = CDT_DEFAULT_POLICY;
1146 /* just need to be larger than previous one */
1147 /* cdt_last_cookie is protected by cdt_llog_lock */
1148 cdt->cdt_last_cookie = ktime_get_real_seconds();
1149 atomic_set(&cdt->cdt_request_count, 0);
1150 atomic_set(&cdt->cdt_archive_count, 0);
1151 atomic_set(&cdt->cdt_restore_count, 0);
1152 atomic_set(&cdt->cdt_remove_count, 0);
1153 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
1154 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
1155 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
1157 /* to avoid deadlock when start is made through /proc
1158 * /proc entries are created by the coordinator thread */
1160 /* set up list of started restore requests */
1161 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1162 rc = mdt_hsm_pending_restore(cdt_mti);
1164 CERROR("%s: cannot take the layout locks needed"
1165 " for registered restore: %d\n",
1166 mdt_obd_name(mdt), rc);
1168 if (mdt->mdt_bottom->dd_rdonly)
1171 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
1174 set_cdt_state(cdt, CDT_STOPPED);
1175 CERROR("%s: error starting coordinator thread: %d\n",
1176 mdt_obd_name(mdt), rc);
1178 cdt->cdt_task = task;
1179 wait_event(cdt->cdt_waitq,
1180 cdt->cdt_state != CDT_INIT);
1181 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1190 * stop a coordinator thread
1191 * \param mdt [IN] device
1193 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1195 struct coordinator *cdt = &mdt->mdt_coordinator;
1199 /* stop coordinator thread */
1200 rc = set_cdt_state(cdt, CDT_STOPPING);
1202 kthread_stop(cdt->cdt_task);
1203 cdt->cdt_task = NULL;
1204 set_cdt_state(cdt, CDT_STOPPED);
1210 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1211 const struct lu_fid *fid,
1214 struct mdt_object *obj;
1218 obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1220 GOTO(out, rc = PTR_ERR(obj));
1222 if (mh.mh_flags & HS_EXISTS &&
1223 mh.mh_arch_id == archive_id)
1224 GOTO(out_obj, rc = 0);
1226 mh.mh_flags |= HS_EXISTS;
1227 mh.mh_arch_id = archive_id;
1228 rc = mdt_hsm_attr_set(mti, obj, &mh);
1231 mdt_object_put(mti->mti_env, obj);
1237 * register all requests from an hal in the memory list
1238 * \param mti [IN] context
1239 * \param hal [IN] request
1240 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1241 * which is running the CT
1243 * \retval -ve failure
1245 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1246 struct hsm_action_list *hal, struct obd_uuid *uuid)
1248 struct mdt_device *mdt = mti->mti_mdt;
1249 struct coordinator *cdt = &mdt->mdt_coordinator;
1250 struct hsm_action_item *hai;
1254 /* register request in memory list */
1255 hai = hai_first(hal);
1256 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1257 struct cdt_agent_req *car;
1259 /* in case of a cancel request, we first mark the ondisk
1260 * record of the request we want to stop as canceled
1261 * this does not change the cancel record
1262 * it will be done when updating the request status
1264 if (hai->hai_action == HSMA_CANCEL) {
1265 struct hsm_record_update update = {
1266 .cookie = hai->hai_cookie,
1267 .status = ARS_CANCELED,
1270 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1273 CERROR("%s: mdt_agent_record_update() failed, "
1274 "rc=%d, cannot update status to %s "
1275 "for cookie %#llx\n",
1276 mdt_obd_name(mdt), rc,
1277 agent_req_status2name(ARS_CANCELED),
1282 /* find the running request to set it canceled */
1283 car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1285 car->car_canceled = 1;
1286 /* uuid has to be changed to the one running the
1287 * request to cancel */
1288 *uuid = car->car_uuid;
1289 mdt_cdt_put_request(car);
1291 /* no need to memorize cancel request
1292 * this also avoid a deadlock when we receive
1293 * a purge all requests command
1298 if (hai->hai_action == HSMA_ARCHIVE) {
1299 rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1300 hal->hal_archive_id);
1307 car = mdt_cdt_alloc_request(hal->hal_archive_id, hal->hal_flags,
1310 GOTO(out, rc = PTR_ERR(car));
1312 rc = mdt_cdt_add_request(cdt, car);
1314 mdt_cdt_free_request(car);
1321 * swap layouts between 2 fids
1322 * \param mti [IN] context
1325 * \param mh_common [IN] MD HSM
1327 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1328 struct mdt_object *obj, const struct lu_fid *dfid,
1329 struct md_hsm *mh_common)
1331 struct mdt_object *dobj;
1332 struct mdt_lock_handle *dlh;
1336 if (!mdt_object_exists(obj))
1337 GOTO(out, rc = -ENOENT);
1339 /* we already have layout lock on obj so take only
1341 dlh = &mti->mti_lh[MDT_LH_OLD];
1342 mdt_lock_reg_init(dlh, LCK_EX);
1343 dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1345 GOTO(out, rc = PTR_ERR(dobj));
1347 /* if copy tool closes the volatile before sending the final
1348 * progress through llapi_hsm_copy_end(), all the objects
1349 * are removed and mdd_swap_layout LBUG */
1350 if (!mdt_object_exists(dobj)) {
1351 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1352 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1353 GOTO(out_dobj, rc = -ENOENT);
1355 /* Since we only handle restores here, unconditionally use
1356 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1357 * be preserved in case of failure during swap_layout and not
1358 * leave a file in an intermediate but incoherent state.
1359 * But need to setup HSM xattr of data FID before, reuse
1360 * mti and mh presets for FID in hsm_cdt_request_completed(),
1361 * only need to clear RELEASED and DIRTY.
1363 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1364 rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1366 rc = mo_swap_layouts(mti->mti_env,
1367 mdt_object_child(obj),
1368 mdt_object_child(dobj),
1369 SWAP_LAYOUTS_MDS_HSM);
1371 rc = mdt_lsom_downgrade(mti, obj);
1374 "%s: File fid="DFID" SOM "
1375 "downgrade failed, rc = %d\n",
1376 mdt_obd_name(mti->mti_mdt),
1377 PFID(mdt_object_fid(obj)), rc);
1380 mdt_object_unlock_put(mti, dobj, dlh, 1);
1386 * update status of a completed request
1387 * \param mti [IN] context
1388 * \param pgs [IN] progress of the copy tool
1390 * \retval -ve failure
1392 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1393 struct hsm_progress_kernel *pgs,
1394 const struct cdt_agent_req *car,
1395 enum agent_req_status *status)
1397 const struct lu_env *env = mti->mti_env;
1398 struct mdt_device *mdt = mti->mti_mdt;
1399 struct coordinator *cdt = &mdt->mdt_coordinator;
1400 struct mdt_object *obj = NULL;
1401 enum changelog_rec_flags clf_flags = 0;
1404 bool need_changelog = true;
1408 /* default is to retry */
1409 *status = ARS_WAITING;
1411 /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1412 * if error/removed continue anyway to get correct reporting done */
1413 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1414 /* we will update MD HSM only if needed */
1415 is_mh_changed = false;
1417 /* no need to change mh->mh_arch_id
1418 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1420 if (pgs->hpk_errval != 0) {
1421 switch (pgs->hpk_errval) {
1423 /* the copy tool does not support cancel
1424 * so the cancel request is failed
1425 * As we cannot distinguish a cancel progress
1426 * from another action progress (they have the
1427 * same cookie), we suppose here the CT returns
1428 * ENOSYS only if does not support cancel
1430 /* this can also happen when cdt calls it to
1431 * for a timed out request */
1432 *status = ARS_FAILED;
1433 /* to have a cancel event in changelog */
1434 pgs->hpk_errval = ECANCELED;
1437 /* the request record has already been set to
1438 * ARS_CANCELED, this set the cancel request
1440 *status = ARS_SUCCEED;
1443 /* retry only if current policy or requested, and
1444 * object is not on error/removed */
1445 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1446 !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1447 IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1451 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1452 CERROR("%s: Request %#llx on "DFID
1453 " failed, error code %d too large\n",
1455 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1457 hsm_set_cl_error(&clf_flags, CLF_HSM_ERROVERFLOW);
1460 hsm_set_cl_error(&clf_flags, pgs->hpk_errval);
1463 switch (car->car_hai->hai_action) {
1465 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1468 hsm_set_cl_event(&clf_flags, HE_RESTORE);
1471 hsm_set_cl_event(&clf_flags, HE_REMOVE);
1474 hsm_set_cl_event(&clf_flags, HE_CANCEL);
1475 CERROR("%s: Failed request %#llx on "DFID
1476 " cannot be a CANCEL\n",
1479 PFID(&pgs->hpk_fid));
1482 CERROR("%s: Failed request %#llx on "DFID
1483 " %d is an unknown action\n",
1485 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1486 car->car_hai->hai_action);
1491 *status = ARS_SUCCEED;
1492 switch (car->car_hai->hai_action) {
1494 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1495 /* set ARCHIVE keep EXIST and clear LOST and
1497 mh.mh_arch_ver = pgs->hpk_data_version;
1498 mh.mh_flags |= HS_ARCHIVED;
1499 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1500 is_mh_changed = true;
1503 hsm_set_cl_event(&clf_flags, HE_RESTORE);
1505 /* do not clear RELEASED and DIRTY here
1506 * this will occur in hsm_swap_layouts()
1509 /* Restoring has changed the file version on
1511 mh.mh_arch_ver = pgs->hpk_data_version;
1512 is_mh_changed = true;
1515 hsm_set_cl_event(&clf_flags, HE_REMOVE);
1516 /* clear ARCHIVED EXISTS and LOST */
1517 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1518 is_mh_changed = true;
1521 hsm_set_cl_event(&clf_flags, HE_CANCEL);
1522 CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1525 PFID(&pgs->hpk_fid));
1528 CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1530 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1531 car->car_hai->hai_action);
1537 /* rc != 0 means error when analysing action, it may come from
1538 * a crasy CT no need to manage DIRTY
1539 * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1542 if (rc == 0 && !IS_ERR(obj))
1543 hsm_set_cl_flags(&clf_flags,
1544 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1546 /* unlock is done later, after layout lock management */
1547 if (is_mh_changed && !IS_ERR(obj))
1548 rc = mdt_hsm_attr_set(mti, obj, &mh);
1550 /* we give back layout lock only if restore was successful or
1551 * if no retry will be attempted and if object is still alive,
1552 * in other cases we just unlock the object */
1553 if (car->car_hai->hai_action == HSMA_RESTORE) {
1554 /* restore in data FID done, we swap the layouts
1555 * only if restore is successful */
1556 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1557 rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1560 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1561 *status = ARS_FAILED;
1562 pgs->hpk_errval = -rc;
1565 /* we have to retry, so keep layout lock */
1566 if (*status == ARS_WAITING)
1569 /* restore special case, need to create ChangeLog record
1570 * before to give back layout lock to avoid concurrent
1571 * file updater to post out of order ChangeLog */
1572 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1573 &car->car_hai->hai_fid);
1574 need_changelog = false;
1576 cdt_restore_handle_del(mti, cdt, &car->car_hai->hai_fid);
1582 /* always add a ChangeLog record */
1584 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1585 &car->car_hai->hai_fid);
1588 mdt_object_put(mti->mti_env, obj);
1594 * update status of a request
1595 * \param mti [IN] context
1596 * \param pgs [IN] progress of the copy tool
1598 * \retval -ve failure
1600 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1601 struct hsm_progress_kernel *pgs)
1603 struct mdt_device *mdt = mti->mti_mdt;
1604 struct coordinator *cdt = &mdt->mdt_coordinator;
1605 struct cdt_agent_req *car;
1609 /* no coordinator started, so we cannot serve requests */
1610 if (cdt->cdt_state == CDT_STOPPED)
1613 /* first do sanity checks */
1614 car = mdt_cdt_update_request(cdt, pgs);
1616 CERROR("%s: Cannot find running request for cookie %#llx"
1619 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1621 RETURN(PTR_ERR(car));
1624 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1625 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1626 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1627 hsm_copytool_action2name(car->car_hai->hai_action),
1628 pgs->hpk_flags, pgs->hpk_errval,
1629 PFID(&car->car_hai->hai_fid),
1630 PFID(&car->car_hai->hai_dfid));
1632 /* progress is done on FID or data FID depending of the action and
1633 * of the copy progress */
1634 /* for restore progress is used to send back the data FID to cdt */
1635 if (car->car_hai->hai_action == HSMA_RESTORE &&
1636 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1637 car->car_hai->hai_dfid = pgs->hpk_fid;
1639 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1640 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1641 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1642 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1643 CERROR("%s: Progress on "DFID" for cookie %#llx"
1644 " does not match request FID "DFID" nor data FID "
1647 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1648 PFID(&car->car_hai->hai_fid),
1649 PFID(&car->car_hai->hai_dfid));
1650 GOTO(out, rc = -EINVAL);
1653 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1654 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1655 " is not coherent (err=%d and not completed"
1658 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1659 hsm_copytool_action2name(car->car_hai->hai_action),
1660 pgs->hpk_errval, pgs->hpk_flags);
1661 GOTO(out, rc = -EINVAL);
1664 /* now progress is valid */
1666 /* we use a root like ucred */
1667 hsm_init_ucred(mdt_ucred(mti));
1669 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1670 enum agent_req_status status;
1671 struct hsm_record_update update;
1674 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1676 CDEBUG(D_HSM, "updating record: fid="DFID" cookie=%#llx action=%s "
1678 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1679 hsm_copytool_action2name(car->car_hai->hai_action),
1680 agent_req_status2name(status));
1682 /* update record first (LU-9075) */
1683 update.cookie = pgs->hpk_cookie;
1684 update.status = status;
1686 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1689 CERROR("%s: mdt_agent_record_update() failed,"
1690 " rc=%d, cannot update status to %s"
1691 " for cookie %#llx\n",
1692 mdt_obd_name(mdt), rc1,
1693 agent_req_status2name(status),
1695 rc = (rc != 0 ? rc : rc1);
1697 /* then remove request from memory list (LU-9075) */
1698 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1700 /* ct has completed a request, so a slot is available,
1701 * signal the coordinator to find new work */
1702 mdt_hsm_cdt_event(cdt);
1704 /* if copytool send a progress on a canceled request
1705 * we inform copytool it should stop
1707 if (car->car_canceled == 1)
1713 /* remove ref got from mdt_cdt_update_request() */
1714 mdt_cdt_put_request(car);
1721 * data passed to llog_cat_process() callback
1722 * to cancel requests
1724 struct hsm_cancel_all_data {
1725 struct mdt_device *mdt;
1729 * llog_cat_process() callback, used to:
1730 * - purge all requests
1731 * \param env [IN] environment
1732 * \param llh [IN] llog handle
1733 * \param hdr [IN] llog record
1734 * \param data [IN] cb data = struct hsm_cancel_all_data
1736 * \retval -ve failure
1738 static int mdt_cancel_all_cb(const struct lu_env *env,
1739 struct llog_handle *llh,
1740 struct llog_rec_hdr *hdr, void *data)
1742 struct llog_agent_req_rec *larr;
1743 struct hsm_cancel_all_data *hcad;
1747 larr = (struct llog_agent_req_rec *)hdr;
1749 if (larr->arr_status == ARS_WAITING ||
1750 larr->arr_status == ARS_STARTED) {
1751 larr->arr_status = ARS_CANCELED;
1752 larr->arr_req_change = ktime_get_real_seconds();
1753 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1760 * cancel all actions
1761 * \param obd [IN] MDT device
1763 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1766 struct lu_context session;
1767 struct mdt_thread_info *mti;
1768 struct coordinator *cdt = &mdt->mdt_coordinator;
1769 struct cdt_agent_req *car;
1770 struct hsm_action_list *hal = NULL;
1771 struct hsm_action_item *hai;
1772 struct hsm_cancel_all_data hcad;
1773 int hal_sz = 0, hal_len, rc;
1774 enum cdt_states old_state;
1777 rc = lu_env_init(&env, LCT_MD_THREAD);
1781 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1782 rc = lu_context_init(&session, LCT_SERVER_SESSION);
1786 lu_context_enter(&session);
1787 env.le_ses = &session;
1789 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1790 LASSERT(mti != NULL);
1792 mti->mti_env = &env;
1795 hsm_init_ucred(mdt_ucred(mti));
1797 mutex_lock(&cdt->cdt_state_lock);
1798 old_state = cdt->cdt_state;
1800 /* disable coordinator */
1801 rc = set_cdt_state_locked(cdt, CDT_DISABLE);
1803 GOTO(out_cdt_state_unlock, rc);
1805 /* send cancel to all running requests */
1806 down_read(&cdt->cdt_request_lock);
1807 list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1808 mdt_cdt_get_request(car);
1809 /* request is not yet removed from list, it will be done
1810 * when copytool will return progress
1813 if (car->car_hai->hai_action == HSMA_CANCEL) {
1814 mdt_cdt_put_request(car);
1819 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1820 cfs_size_round(car->car_hai->hai_len);
1822 if (hal_len > hal_sz && hal_sz > 0) {
1823 /* not enough room, free old buffer */
1824 OBD_FREE(hal, hal_sz);
1828 /* empty buffer, allocate one */
1831 OBD_ALLOC(hal, hal_sz);
1833 mdt_cdt_put_request(car);
1834 up_read(&cdt->cdt_request_lock);
1835 GOTO(out_cdt_state, rc = -ENOMEM);
1839 hal->hal_version = HAL_VERSION;
1840 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1842 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1843 hal->hal_archive_id = car->car_archive_id;
1844 hal->hal_flags = car->car_flags;
1847 hai = hai_first(hal);
1848 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1849 hai->hai_action = HSMA_CANCEL;
1852 /* it is possible to safely call mdt_hsm_agent_send()
1853 * (ie without a deadlock on cdt_request_lock), because the
1854 * write lock is taken only if we are not in purge mode
1855 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1856 * nor mdt_cdt_remove_request())
1858 /* no conflict with cdt thread because cdt is disable and we
1859 * have the request lock */
1860 mdt_hsm_agent_send(mti, hal, 1);
1862 mdt_cdt_put_request(car);
1864 up_read(&cdt->cdt_request_lock);
1867 OBD_FREE(hal, hal_sz);
1869 /* cancel all on-disk records */
1872 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1873 &hcad, 0, 0, WRITE);
1875 /* Enable coordinator, unless the coordinator was stopping. */
1876 set_cdt_state_locked(cdt, old_state);
1877 out_cdt_state_unlock:
1878 mutex_unlock(&cdt->cdt_state_lock);
1880 lu_context_exit(&session);
1881 lu_context_fini(&session);
1889 * check if a request is compatible with file status
1890 * \param hai [IN] request description
1891 * \param archive_id [IN] request archive id
1892 * \param rq_flags [IN] request flags
1893 * \param hsm [IN] file HSM metadata
1896 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1897 u32 archive_id, u64 rq_flags,
1898 const struct md_hsm *hsm)
1900 int is_compat = false;
1904 hsm_flags = hsm->mh_flags;
1905 switch (hai->hai_action) {
1907 if (!(hsm_flags & HS_NOARCHIVE) &&
1908 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1911 if (hsm_flags & HS_EXISTS &&
1913 archive_id != hsm->mh_arch_id)
1918 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1919 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1923 if (!(hsm_flags & HS_RELEASED) &&
1924 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1931 CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1932 " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1933 PFID(&hai->hai_fid),
1934 hsm_copytool_action2name(hai->hai_action), rq_flags,
1935 hai->hai_extent.offset, hai->hai_extent.length,
1937 (is_compat ? "compatible" : "uncompatible"));
1943 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1945 static const struct {
1949 } hsm_policy_names[] = {
1950 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1951 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1956 * convert a policy name to a bit
1957 * \param name [IN] policy name
1959 * \retval policy bit
1961 static __u64 hsm_policy_str2bit(const char *name)
1965 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1966 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1967 strcmp(hsm_policy_names[i].name, name) == 0)
1968 return hsm_policy_names[i].bit;
1973 * convert a policy bit field to a string
1974 * \param mask [IN] policy bit field
1975 * \param hexa [IN] print mask before bit names
1976 * \param buffer [OUT] string
1977 * \param count [IN] size of buffer
1979 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1987 seq_printf(m, "(%#llx) ", mask);
1989 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1992 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1993 if (hsm_policy_names[j].bit == bit)
1997 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1999 seq_printf(m, "%s ", hsm_policy_names[j].name);
2001 /* remove last ' ' */
2006 /* methods to read/write HSM policy flags */
2007 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
2009 struct mdt_device *mdt = m->private;
2010 struct coordinator *cdt = &mdt->mdt_coordinator;
2013 hsm_policy_bit2str(m, cdt->cdt_policy, false);
2018 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
2019 size_t count, loff_t *off)
2021 struct seq_file *m = file->private_data;
2022 struct mdt_device *mdt = m->private;
2023 struct coordinator *cdt = &mdt->mdt_coordinator;
2024 char *start, *token, sign;
2027 __u64 add_mask, remove_mask, set_mask;
2031 if (count + 1 > PAGE_SIZE)
2034 OBD_ALLOC(buf, count + 1);
2038 if (copy_from_user(buf, buffer, count))
2039 GOTO(out, rc = -EFAULT);
2044 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
2047 add_mask = remove_mask = set_mask = 0;
2049 token = strsep(&start, "\n ");
2055 if (sign == '-' || sign == '+')
2058 policy = hsm_policy_str2bit(token);
2060 CWARN("%s: '%s' is unknown, "
2061 "supported policies are:\n", mdt_obd_name(mdt),
2063 hsm_policy_bit2str(m, 0, false);
2064 GOTO(out, rc = -EINVAL);
2068 remove_mask |= policy;
2078 } while (start != NULL);
2080 CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
2081 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
2083 /* if no sign in all string, it is a clear and set
2084 * if some sign found, all unsigned are converted
2086 * P1 P2 = set to P1 and P2
2087 * P1 -P2 = add P1 clear P2 same as +P1 -P2
2089 if (remove_mask == 0 && add_mask == 0) {
2090 cdt->cdt_policy = set_mask;
2092 cdt->cdt_policy |= set_mask | add_mask;
2093 cdt->cdt_policy &= ~remove_mask;
2096 GOTO(out, rc = count);
2099 OBD_FREE(buf, count + 1);
2102 LPROC_SEQ_FOPS(mdt_hsm_policy);
2104 #define GENERATE_PROC_METHOD(VAR) \
2105 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
2107 struct mdt_device *mdt = m->private; \
2108 struct coordinator *cdt = &mdt->mdt_coordinator; \
2111 seq_printf(m, "%llu\n", (__u64)cdt->VAR); \
2115 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
2116 size_t count, loff_t *off) \
2119 struct seq_file *m = file->private_data; \
2120 struct mdt_device *mdt = m->private; \
2121 struct coordinator *cdt = &mdt->mdt_coordinator; \
2126 rc = kstrtouint_from_user(buffer, count, 0, &val); \
2137 GENERATE_PROC_METHOD(cdt_loop_period)
2138 GENERATE_PROC_METHOD(cdt_grace_delay)
2139 GENERATE_PROC_METHOD(cdt_active_req_timeout)
2140 GENERATE_PROC_METHOD(cdt_max_requests)
2141 GENERATE_PROC_METHOD(cdt_default_archive_id)
2144 * procfs write method for MDT/hsm_control
2145 * proc entry is in mdt directory so data is mdt obd_device pointer
2147 #define CDT_ENABLE_CMD "enabled"
2148 #define CDT_STOP_CMD "shutdown"
2149 #define CDT_DISABLE_CMD "disabled"
2150 #define CDT_PURGE_CMD "purge"
2151 #define CDT_HELP_CMD "help"
2152 #define CDT_MAX_CMD_LEN 10
2155 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
2156 size_t count, loff_t *off)
2158 struct seq_file *m = file->private_data;
2159 struct obd_device *obd = m->private;
2160 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2161 struct coordinator *cdt = &(mdt->mdt_coordinator);
2163 char kernbuf[CDT_MAX_CMD_LEN];
2166 if (count == 0 || count >= sizeof(kernbuf))
2169 if (copy_from_user(kernbuf, buffer, count))
2173 if (kernbuf[count - 1] == '\n')
2174 kernbuf[count - 1] = 0;
2177 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
2178 if (cdt->cdt_state == CDT_DISABLE) {
2179 rc = set_cdt_state(cdt, CDT_RUNNING);
2180 mdt_hsm_cdt_event(cdt);
2181 wake_up(&cdt->cdt_waitq);
2183 rc = mdt_hsm_cdt_start(mdt);
2185 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2186 if ((cdt->cdt_state == CDT_STOPPING) ||
2187 (cdt->cdt_state == CDT_STOPPED)) {
2188 CERROR("%s: Coordinator already stopped\n",
2192 rc = mdt_hsm_cdt_stop(mdt);
2194 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2195 if ((cdt->cdt_state == CDT_STOPPING) ||
2196 (cdt->cdt_state == CDT_STOPPED)) {
2197 CERROR("%s: Coordinator is stopped\n",
2201 rc = set_cdt_state(cdt, CDT_DISABLE);
2203 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2204 rc = hsm_cancel_all_actions(mdt);
2205 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2213 CERROR("%s: Valid coordinator control commands are: "
2214 "%s %s %s %s %s\n", mdt_obd_name(mdt),
2215 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2216 CDT_PURGE_CMD, CDT_HELP_CMD);
2224 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2226 struct obd_device *obd = m->private;
2227 struct coordinator *cdt;
2230 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2232 seq_printf(m, "%s\n", cdt_mdt_state2str(cdt->cdt_state));
2238 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2244 for (i = 0; i < 8 * sizeof(mask); i++) {
2245 if (mask & (1UL << i)) {
2246 seq_printf(m, "%s%s", first ? "" : " ",
2247 hsm_copytool_action2name(i));
2257 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2259 struct mdt_device *mdt = m->private;
2260 struct coordinator *cdt = &mdt->mdt_coordinator;
2262 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2266 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2268 struct mdt_device *mdt = m->private;
2269 struct coordinator *cdt = &mdt->mdt_coordinator;
2271 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2275 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2277 struct mdt_device *mdt = m->private;
2278 struct coordinator *cdt = &mdt->mdt_coordinator;
2280 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2283 static inline enum hsm_copytool_action
2284 hsm_copytool_name2action(const char *name)
2286 if (strcasecmp(name, "NOOP") == 0)
2288 else if (strcasecmp(name, "ARCHIVE") == 0)
2289 return HSMA_ARCHIVE;
2290 else if (strcasecmp(name, "RESTORE") == 0)
2291 return HSMA_RESTORE;
2292 else if (strcasecmp(name, "REMOVE") == 0)
2294 else if (strcasecmp(name, "CANCEL") == 0)
2301 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2302 size_t user_count, __u64 *mask)
2304 char *buf, *pos, *name;
2310 if (!(user_count < 4096))
2313 buf_size = user_count + 1;
2315 OBD_ALLOC(buf, buf_size);
2319 if (copy_from_user(buf, user_buf, buf_size - 1))
2320 GOTO(out, rc = -EFAULT);
2322 buf[buf_size - 1] = '\0';
2325 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2331 action = hsm_copytool_name2action(name);
2333 GOTO(out, rc = -EINVAL);
2335 new_mask |= (1UL << action);
2341 OBD_FREE(buf, buf_size);
2347 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2348 size_t count, loff_t *off)
2350 struct seq_file *m = file->private_data;
2351 struct mdt_device *mdt = m->private;
2352 struct coordinator *cdt = &mdt->mdt_coordinator;
2354 return mdt_write_hsm_request_mask(file, buf, count,
2355 &cdt->cdt_user_request_mask);
2359 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2360 size_t count, loff_t *off)
2362 struct seq_file *m = file->private_data;
2363 struct mdt_device *mdt = m->private;
2364 struct coordinator *cdt = &mdt->mdt_coordinator;
2366 return mdt_write_hsm_request_mask(file, buf, count,
2367 &cdt->cdt_group_request_mask);
2371 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2372 size_t count, loff_t *off)
2374 struct seq_file *m = file->private_data;
2375 struct mdt_device *mdt = m->private;
2376 struct coordinator *cdt = &mdt->mdt_coordinator;
2378 return mdt_write_hsm_request_mask(file, buf, count,
2379 &cdt->cdt_other_request_mask);
2382 static int mdt_hsm_cdt_raolu_seq_show(struct seq_file *m, void *data)
2384 struct mdt_device *mdt = m->private;
2385 struct coordinator *cdt = &mdt->mdt_coordinator;
2388 seq_printf(m, "%d\n", (int)cdt->cdt_remove_archive_on_last_unlink);
2393 mdt_hsm_cdt_raolu_seq_write(struct file *file, const char __user *buffer,
2394 size_t count, loff_t *off)
2397 struct seq_file *m = file->private_data;
2398 struct mdt_device *mdt = m->private;
2399 struct coordinator *cdt = &mdt->mdt_coordinator;
2404 rc = kstrtobool_from_user(buffer, count, &val);
2408 cdt->cdt_remove_archive_on_last_unlink = val;
2412 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2413 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2414 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2415 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2416 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2417 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2418 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2419 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2420 LPROC_SEQ_FOPS(mdt_hsm_cdt_raolu);
2422 /* Read-only proc files for request counters */
2423 static int mdt_hsm_cdt_archive_count_seq_show(struct seq_file *m, void *data)
2425 struct mdt_device *mdt = m->private;
2426 struct coordinator *cdt = &mdt->mdt_coordinator;
2429 seq_printf(m, "%d\n", atomic_read(&cdt->cdt_archive_count));
2433 static int mdt_hsm_cdt_restore_count_seq_show(struct seq_file *m, void *data)
2435 struct mdt_device *mdt = m->private;
2436 struct coordinator *cdt = &mdt->mdt_coordinator;
2439 seq_printf(m, "%d\n", atomic_read(&cdt->cdt_restore_count));
2443 static int mdt_hsm_cdt_remove_count_seq_show(struct seq_file *m, void *data)
2445 struct mdt_device *mdt = m->private;
2446 struct coordinator *cdt = &mdt->mdt_coordinator;
2449 seq_printf(m, "%d\n", atomic_read(&cdt->cdt_remove_count));
2453 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_archive_count);
2454 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_restore_count);
2455 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_remove_count);
2457 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2459 .fops = &mdt_hsm_agent_fops },
2460 { .name = "actions",
2461 .fops = &mdt_hsm_actions_fops,
2462 .proc_mode = 0444 },
2463 { .name = "default_archive_id",
2464 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2465 { .name = "grace_delay",
2466 .fops = &mdt_hsm_cdt_grace_delay_fops },
2467 { .name = "loop_period",
2468 .fops = &mdt_hsm_cdt_loop_period_fops },
2469 { .name = "max_requests",
2470 .fops = &mdt_hsm_cdt_max_requests_fops },
2472 .fops = &mdt_hsm_policy_fops },
2473 { .name = "active_request_timeout",
2474 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2475 { .name = "active_requests",
2476 .fops = &mdt_hsm_active_requests_fops },
2477 { .name = "user_request_mask",
2478 .fops = &mdt_hsm_user_request_mask_fops, },
2479 { .name = "group_request_mask",
2480 .fops = &mdt_hsm_group_request_mask_fops, },
2481 { .name = "other_request_mask",
2482 .fops = &mdt_hsm_other_request_mask_fops, },
2483 { .name = "remove_archive_on_last_unlink",
2484 .fops = &mdt_hsm_cdt_raolu_fops, },
2485 { .name = "archive_count",
2486 .fops = &mdt_hsm_cdt_archive_count_fops, },
2487 { .name = "restore_count",
2488 .fops = &mdt_hsm_cdt_restore_count_fops, },
2489 { .name = "remove_count",
2490 .fops = &mdt_hsm_cdt_remove_count_fops, },