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, 2014, 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 <obd_support.h>
42 #include <lustre_net.h>
43 #include <lustre_export.h>
45 #include <lprocfs_status.h>
46 #include <lustre_log.h>
47 #include "mdt_internal.h"
49 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
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 compound "LPX64
104 " archive_id %d flags "LPX64"\n",
105 prefix, hal->hal_version, hal->hal_count,
106 hal->hal_compound_id, 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
112 " compound/cookie="LPX64"/"LPX64
113 " action=%s extent="LPX64"-"LPX64" gid="LPX64
114 " datalen=%d data=[%s]\n",
116 PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
117 hal->hal_compound_id, hai->hai_cookie,
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_data {
132 struct mdt_thread_info *mti;
133 char fs_name[MTI_NAME_MAXLEN+1];
134 /* request to be send to agents */
135 int request_sz; /** allocated size */
136 int max_requests; /** vector size */
137 int request_cnt; /** used count */
141 struct hsm_action_list *hal;
143 /* records to be canceled */
144 int max_cookie; /** vector size */
145 int cookie_cnt; /** used count */
150 * llog_cat_process() callback, used to:
151 * - find waiting request and start action
152 * - purge canceled and done requests
153 * \param env [IN] environment
154 * \param llh [IN] llog handle
155 * \param hdr [IN] llog record
156 * \param data [IN/OUT] cb data = struct hsm_scan_data
158 * \retval -ve failure
160 static int mdt_coordinator_cb(const struct lu_env *env,
161 struct llog_handle *llh,
162 struct llog_rec_hdr *hdr,
165 const struct llog_agent_req_rec *larr;
166 struct hsm_scan_data *hsd;
167 struct hsm_action_item *hai;
168 struct mdt_device *mdt;
169 struct coordinator *cdt;
174 mdt = hsd->mti->mti_mdt;
175 cdt = &mdt->mdt_coordinator;
177 larr = (struct llog_agent_req_rec *)hdr;
178 dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
179 switch (larr->arr_status) {
181 int i, empty_slot, found;
183 /* Are agents full? */
184 if (atomic_read(&cdt->cdt_request_count) ==
185 cdt->cdt_max_requests)
188 /* first search if the request if known in the list we have
189 * build and if there is room in the request vector */
192 for (i = 0; i < hsd->max_requests &&
193 (empty_slot == -1 || found == -1); i++) {
194 if (hsd->request[i].hal == NULL) {
198 if (hsd->request[i].hal->hal_compound_id ==
199 larr->arr_compound_id) {
204 if (found == -1 && empty_slot == -1)
205 /* unknown request and no more room for new request,
206 * continue scan for to find other entries for
207 * already found request
212 struct hsm_action_list *hal;
214 /* request is not already known */
215 /* allocates hai vector size just needs to be large
217 hsd->request[empty_slot].hal_sz =
218 sizeof(*hsd->request[empty_slot].hal) +
219 cfs_size_round(MTI_NAME_MAXLEN+1) +
220 2 * cfs_size_round(larr->arr_hai.hai_len);
221 OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
223 CERROR("%s: Cannot allocate memory (%d o)"
224 "for compound "LPX64"\n",
226 hsd->request[i].hal_sz,
227 larr->arr_compound_id);
230 hal->hal_version = HAL_VERSION;
231 strlcpy(hal->hal_fsname, hsd->fs_name,
232 MTI_NAME_MAXLEN + 1);
233 hal->hal_compound_id = larr->arr_compound_id;
234 hal->hal_archive_id = larr->arr_archive_id;
235 hal->hal_flags = larr->arr_flags;
237 hsd->request[empty_slot].hal_used_sz = hal_size(hal);
238 hsd->request[empty_slot].hal = hal;
241 hai = hai_first(hal);
243 /* request is known */
244 /* we check if record archive num is the same as the
245 * known request, if not we will serve it in multiple
246 * time because we do not know if the agent can serve
248 * a use case is a compound made of multiple restore
249 * where the files are not archived in the same backend
251 if (larr->arr_archive_id !=
252 hsd->request[found].hal->hal_archive_id)
255 if (hsd->request[found].hal_sz <
256 hsd->request[found].hal_used_sz +
257 cfs_size_round(larr->arr_hai.hai_len)) {
258 /* Not enough room, need an extension */
262 sz = 2 * hsd->request[found].hal_sz;
263 OBD_ALLOC(hal_buffer, sz);
265 CERROR("%s: Cannot allocate memory "
266 "(%d o) for compound "LPX64"\n",
267 mdt_obd_name(mdt), sz,
268 larr->arr_compound_id);
271 memcpy(hal_buffer, hsd->request[found].hal,
272 hsd->request[found].hal_used_sz);
273 OBD_FREE(hsd->request[found].hal,
274 hsd->request[found].hal_sz);
275 hsd->request[found].hal = hal_buffer;
276 hsd->request[found].hal_sz = sz;
278 hai = hai_first(hsd->request[found].hal);
279 for (i = 0; i < hsd->request[found].hal->hal_count;
283 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
284 hai->hai_cookie = larr->arr_hai.hai_cookie;
285 hai->hai_gid = larr->arr_hai.hai_gid;
287 hsd->request[found].hal_used_sz +=
288 cfs_size_round(hai->hai_len);
289 hsd->request[found].hal->hal_count++;
293 struct cdt_agent_req *car;
296 /* we search for a running request
297 * error may happen if coordinator crashes or stopped
298 * with running request
300 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
302 last = larr->arr_req_create;
304 last = car->car_req_update;
305 mdt_cdt_put_request(car);
308 /* test if request too long, if yes cancel it
309 * the same way the copy tool acknowledge a cancel request */
310 if ((last + cdt->cdt_active_req_timeout)
311 < cfs_time_current_sec()) {
312 struct hsm_progress_kernel pgs;
314 dump_llog_agent_req_rec("mdt_coordinator_cb(): "
315 "request timed out, start "
317 /* a too old cancel request just needs to be removed
318 * this can happen, if copy tool does not support cancel
319 * for other requests, we have to remove the running
320 * request and notify the copytool
322 pgs.hpk_fid = larr->arr_hai.hai_fid;
323 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
324 pgs.hpk_extent = larr->arr_hai.hai_extent;
325 pgs.hpk_flags = HP_FLAG_COMPLETED;
326 pgs.hpk_errval = ENOSYS;
327 pgs.hpk_data_version = 0;
328 /* update request state, but do not record in llog, to
329 * avoid deadlock on cdt_llog_lock
331 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
333 CERROR("%s: Cannot cleanup timed out request: "
334 DFID" for cookie "LPX64" action=%s\n",
336 PFID(&pgs.hpk_fid), pgs.hpk_cookie,
337 hsm_copytool_action2name(
338 larr->arr_hai.hai_action));
341 /* The request no longer exists, forget
342 * about it, and do not send a cancel request
343 * to the client, for which an error will be
344 * sent back, leading to an endless cycle of
346 RETURN(LLOG_DEL_RECORD);
349 /* add the cookie to the list of record to be
350 * canceled by caller */
351 if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
352 __u64 *ptr, *old_ptr;
353 int old_sz, new_sz, new_cnt;
355 /* need to increase vector size */
356 old_sz = sizeof(__u64) * hsd->max_cookie;
357 old_ptr = hsd->cookies;
359 new_cnt = 2 * hsd->max_cookie;
360 new_sz = sizeof(__u64) * new_cnt;
362 OBD_ALLOC(ptr, new_sz);
364 CERROR("%s: Cannot allocate memory "
365 "(%d o) for cookie vector\n",
366 mdt_obd_name(mdt), new_sz);
369 memcpy(ptr, hsd->cookies, old_sz);
371 hsd->max_cookie = new_cnt;
372 OBD_FREE(old_ptr, old_sz);
374 hsd->cookies[hsd->cookie_cnt] =
375 larr->arr_hai.hai_cookie;
383 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
384 cfs_time_current_sec())
385 RETURN(LLOG_DEL_RECORD);
392 * create /proc entries for coordinator
395 * \retval -ve failure
397 int hsm_cdt_procfs_init(struct mdt_device *mdt)
399 struct coordinator *cdt = &mdt->mdt_coordinator;
403 /* init /proc entries, failure is not critical */
404 cdt->cdt_proc_dir = lprocfs_register("hsm",
405 mdt2obd_dev(mdt)->obd_proc_entry,
406 lprocfs_mdt_hsm_vars, mdt);
407 if (IS_ERR(cdt->cdt_proc_dir)) {
408 rc = PTR_ERR(cdt->cdt_proc_dir);
409 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
410 " rc=%d\n", mdt_obd_name(mdt), rc);
411 cdt->cdt_proc_dir = NULL;
419 * remove /proc entries for coordinator
422 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
424 struct coordinator *cdt = &mdt->mdt_coordinator;
426 LASSERT(cdt->cdt_state == CDT_STOPPED);
427 if (cdt->cdt_proc_dir != NULL)
428 lprocfs_remove(&cdt->cdt_proc_dir);
432 * get vector of hsm cdt /proc vars
436 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
438 return lprocfs_mdt_hsm_vars;
443 * \param data [IN] obd device
445 * \retval -ve failure
447 static int mdt_coordinator(void *data)
449 struct mdt_thread_info *mti = data;
450 struct mdt_device *mdt = mti->mti_mdt;
451 struct coordinator *cdt = &mdt->mdt_coordinator;
452 struct hsm_scan_data hsd = { NULL };
456 cdt->cdt_thread.t_flags = SVC_RUNNING;
457 wake_up(&cdt->cdt_thread.t_ctl_waitq);
459 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
460 mdt_obd_name(mdt), current_pid());
462 /* timeouted cookie vector initialization */
466 /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
467 * increases due to a change from /proc we do not overflow the
468 * hsd.request[] vector
470 hsd.max_requests = cdt->cdt_max_requests;
471 hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
472 OBD_ALLOC(hsd.request, hsd.request_sz);
474 GOTO(out, rc = -ENOMEM);
477 obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
480 struct l_wait_info lwi;
483 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
485 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
486 (cdt->cdt_thread.t_flags &
487 (SVC_STOPPING|SVC_EVENT)),
490 CDEBUG(D_HSM, "coordinator resumes\n");
492 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
493 cdt->cdt_state == CDT_STOPPING) {
494 cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
499 /* wake up before timeout, new work arrives */
500 if (cdt->cdt_thread.t_flags & SVC_EVENT)
501 cdt->cdt_thread.t_flags &= ~SVC_EVENT;
503 /* if coordinator is suspended continue to wait */
504 if (cdt->cdt_state == CDT_DISABLE) {
505 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
509 CDEBUG(D_HSM, "coordinator starts reading llog\n");
511 if (hsd.max_requests != cdt->cdt_max_requests) {
512 /* cdt_max_requests has changed,
513 * we need to allocate a new buffer
515 OBD_FREE(hsd.request, hsd.request_sz);
516 hsd.max_requests = cdt->cdt_max_requests;
518 hsd.max_requests * sizeof(*hsd.request);
519 OBD_ALLOC(hsd.request, hsd.request_sz);
526 /* create canceled cookie vector for an arbitrary size
527 * if needed, vector will grow during llog scan
531 OBD_ALLOC(hsd.cookies, hsd.max_cookie * sizeof(__u64));
538 rc = cdt_llog_process(mti->mti_env, mdt,
539 mdt_coordinator_cb, &hsd);
543 CDEBUG(D_HSM, "Found %d requests to send and %d"
544 " requests to cancel\n",
545 hsd.request_cnt, hsd.cookie_cnt);
546 /* first we cancel llog records of the timed out requests */
547 if (hsd.cookie_cnt > 0) {
548 rc = mdt_agent_record_update(mti->mti_env, mdt,
553 CERROR("%s: mdt_agent_record_update() failed, "
554 "rc=%d, cannot update status to %s "
556 mdt_obd_name(mdt), rc,
557 agent_req_status2name(ARS_CANCELED),
561 if (list_empty(&cdt->cdt_agents)) {
562 CDEBUG(D_HSM, "no agent available, "
563 "coordinator sleeps\n");
567 /* here hsd contains a list of requests to be started */
568 for (i = 0; i < hsd.max_requests; i++) {
569 struct hsm_action_list *hal;
570 struct hsm_action_item *hai;
573 enum agent_req_status status;
575 /* still room for work ? */
576 if (atomic_read(&cdt->cdt_request_count) ==
577 cdt->cdt_max_requests)
580 if (hsd.request[i].hal == NULL)
583 /* found a request, we start it */
584 /* kuc payload allocation so we avoid an additionnal
585 * allocation in mdt_hsm_agent_send()
587 hal = kuc_alloc(hsd.request[i].hal_used_sz,
588 KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
590 CERROR("%s: Cannot allocate memory (%d o) "
591 "for compound "LPX64"\n",
593 hsd.request[i].hal_used_sz,
594 hsd.request[i].hal->hal_compound_id);
597 memcpy(hal, hsd.request[i].hal,
598 hsd.request[i].hal_used_sz);
600 rc = mdt_hsm_agent_send(mti, hal, 0);
601 /* if failure, we suppose it is temporary
602 * if the copy tool failed to do the request
603 * it has to use hsm_progress
605 status = (rc ? ARS_WAITING : ARS_STARTED);
607 /* set up cookie vector to set records status
608 * after copy tools start or failed
610 sz = hsd.request[i].hal->hal_count * sizeof(__u64);
611 OBD_ALLOC(cookies, sz);
612 if (cookies == NULL) {
613 CERROR("%s: Cannot allocate memory (%d o) "
614 "for cookies vector "LPX64"\n",
615 mdt_obd_name(mdt), sz,
616 hsd.request[i].hal->hal_compound_id);
617 kuc_free(hal, hsd.request[i].hal_used_sz);
620 hai = hai_first(hal);
621 for (j = 0; j < hsd.request[i].hal->hal_count; j++) {
622 cookies[j] = hai->hai_cookie;
626 rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
627 hsd.request[i].hal->hal_count,
630 CERROR("%s: mdt_agent_record_update() failed, "
631 "rc=%d, cannot update status to %s "
633 mdt_obd_name(mdt), rc,
634 agent_req_status2name(status),
635 hsd.request[i].hal->hal_count);
637 OBD_FREE(cookies, sz);
638 kuc_free(hal, hsd.request[i].hal_used_sz);
641 /* free cookie vector allocated for/by callback */
643 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
649 /* free hal allocated by callback */
650 for (i = 0; i < hsd.max_requests; i++) {
651 if (hsd.request[i].hal) {
652 OBD_FREE(hsd.request[i].hal,
653 hsd.request[i].hal_sz);
654 hsd.request[i].hal_sz = 0;
655 hsd.request[i].hal = NULL;
659 LASSERT(hsd.request_cnt == 0);
661 /* reset callback data */
662 memset(hsd.request, 0, hsd.request_sz);
667 OBD_FREE(hsd.request, hsd.request_sz);
670 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
672 if (cdt->cdt_state == CDT_STOPPING) {
673 /* request comes from /proc path, so we need to clean cdt
675 mdt_hsm_cdt_stop(mdt);
676 mdt->mdt_opts.mo_coordinator = 0;
678 /* request comes from a thread event, generated
679 * by mdt_stop_coordinator(), we have to ack
680 * and cdt cleaning will be done by event sender
682 cdt->cdt_thread.t_flags = SVC_STOPPED;
683 wake_up(&cdt->cdt_thread.t_ctl_waitq);
687 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
688 mdt_obd_name(mdt), current_pid(), rc);
690 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
692 mdt_obd_name(mdt), current_pid());
698 * lookup a restore handle by FID
699 * caller needs to hold cdt_restore_lock
700 * \param cdt [IN] coordinator
701 * \param fid [IN] FID
702 * \retval cdt_restore_handle found
703 * \retval NULL not found
705 static struct cdt_restore_handle *hsm_restore_hdl_find(struct coordinator *cdt,
706 const struct lu_fid *fid)
708 struct cdt_restore_handle *crh;
711 list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
712 if (lu_fid_eq(&crh->crh_fid, fid))
719 * data passed to llog_cat_process() callback
720 * to scan requests and take actions
722 struct hsm_restore_data {
723 struct mdt_thread_info *hrd_mti;
727 * llog_cat_process() callback, used to:
728 * - find restore request and allocate the restore handle
729 * \param env [IN] environment
730 * \param llh [IN] llog handle
731 * \param hdr [IN] llog record
732 * \param data [IN/OUT] cb data = struct hsm_restore_data
734 * \retval -ve failure
736 static int hsm_restore_cb(const struct lu_env *env,
737 struct llog_handle *llh,
738 struct llog_rec_hdr *hdr, void *data)
740 struct llog_agent_req_rec *larr;
741 struct hsm_restore_data *hrd;
742 struct cdt_restore_handle *crh;
743 struct hsm_action_item *hai;
744 struct mdt_thread_info *mti;
745 struct coordinator *cdt;
746 struct mdt_object *child;
752 cdt = &mti->mti_mdt->mdt_coordinator;
754 larr = (struct llog_agent_req_rec *)hdr;
755 hai = &larr->arr_hai;
756 if (hai->hai_cookie > cdt->cdt_last_cookie)
757 /* update the cookie to avoid collision */
758 cdt->cdt_last_cookie = hai->hai_cookie + 1;
760 if (hai->hai_action != HSMA_RESTORE ||
761 agent_req_in_final_state(larr->arr_status))
764 /* restore request not in a final state */
766 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
770 crh->crh_fid = hai->hai_fid;
771 /* in V1 all file is restored
772 crh->extent.start = hai->hai_extent.offset;
773 crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
775 crh->crh_extent.start = 0;
776 crh->crh_extent.end = hai->hai_extent.length;
777 /* get the layout lock */
778 mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
779 child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
780 MDS_INODELOCK_LAYOUT);
782 GOTO(out, rc = PTR_ERR(child));
785 /* we choose to not keep a reference
786 * on the object during the restore time which can be very long */
787 mdt_object_put(mti->mti_env, child);
789 mutex_lock(&cdt->cdt_restore_lock);
790 list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
791 mutex_unlock(&cdt->cdt_restore_lock);
798 * restore coordinator state at startup
799 * the goal is to take a layout lock for each registered restore request
800 * \param mti [IN] context
802 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
804 struct hsm_restore_data hrd;
810 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
811 hsm_restore_cb, &hrd);
816 static int hsm_init_ucred(struct lu_ucred *uc)
820 uc->uc_valid = UCRED_OLD;
829 uc->uc_suppgids[0] = -1;
830 uc->uc_suppgids[1] = -1;
831 uc->uc_cap = CFS_CAP_FS_MASK;
834 uc->uc_identity = NULL;
840 * wake up coordinator thread
841 * \param mdt [IN] device
843 * \retval -ve failure
845 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
847 struct coordinator *cdt = &mdt->mdt_coordinator;
850 if (cdt->cdt_state == CDT_STOPPED)
853 /* wake up coordinator */
854 cdt->cdt_thread.t_flags = SVC_EVENT;
855 wake_up(&cdt->cdt_thread.t_ctl_waitq);
861 * initialize coordinator struct
862 * \param mdt [IN] device
864 * \retval -ve failure
866 int mdt_hsm_cdt_init(struct mdt_device *mdt)
868 struct coordinator *cdt = &mdt->mdt_coordinator;
869 struct mdt_thread_info *cdt_mti = NULL;
873 cdt->cdt_state = CDT_STOPPED;
875 init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
876 mutex_init(&cdt->cdt_llog_lock);
877 init_rwsem(&cdt->cdt_agent_lock);
878 init_rwsem(&cdt->cdt_request_lock);
879 mutex_init(&cdt->cdt_restore_lock);
881 INIT_LIST_HEAD(&cdt->cdt_requests);
882 INIT_LIST_HEAD(&cdt->cdt_agents);
883 INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
885 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
889 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
890 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
892 lu_context_enter(&cdt->cdt_session);
893 cdt->cdt_env.le_ses = &cdt->cdt_session;
895 lu_env_fini(&cdt->cdt_env);
899 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
900 LASSERT(cdt_mti != NULL);
902 cdt_mti->mti_env = &cdt->cdt_env;
903 cdt_mti->mti_mdt = mdt;
905 hsm_init_ucred(mdt_ucred(cdt_mti));
907 /* default values for /proc tunnables
908 * can be override by MGS conf */
909 cdt->cdt_default_archive_id = 1;
910 cdt->cdt_grace_delay = 60;
911 cdt->cdt_loop_period = 10;
912 cdt->cdt_max_requests = 3;
913 cdt->cdt_policy = CDT_DEFAULT_POLICY;
914 cdt->cdt_active_req_timeout = 3600;
920 * free a coordinator thread
921 * \param mdt [IN] device
923 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
925 struct coordinator *cdt = &mdt->mdt_coordinator;
928 lu_context_exit(cdt->cdt_env.le_ses);
929 lu_context_fini(cdt->cdt_env.le_ses);
931 lu_env_fini(&cdt->cdt_env);
937 * start a coordinator thread
938 * \param mdt [IN] device
940 * \retval -ve failure
942 int mdt_hsm_cdt_start(struct mdt_device *mdt)
944 struct coordinator *cdt = &mdt->mdt_coordinator;
947 struct mdt_thread_info *cdt_mti;
948 struct task_struct *task;
951 /* functions defined but not yet used
952 * this avoid compilation warning
956 if (cdt->cdt_state != CDT_STOPPED) {
957 CERROR("%s: Coordinator already started\n",
962 CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
963 cdt->cdt_policy = CDT_DEFAULT_POLICY;
965 cdt->cdt_state = CDT_INIT;
967 atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
968 /* just need to be larger than previous one */
969 /* cdt_last_cookie is protected by cdt_llog_lock */
970 cdt->cdt_last_cookie = cfs_time_current_sec();
971 atomic_set(&cdt->cdt_request_count, 0);
972 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
973 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
974 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
976 /* to avoid deadlock when start is made through /proc
977 * /proc entries are created by the coordinator thread */
979 /* set up list of started restore requests */
980 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
981 rc = mdt_hsm_pending_restore(cdt_mti);
983 CERROR("%s: cannot take the layout locks needed"
984 " for registered restore: %d\n",
985 mdt_obd_name(mdt), rc);
987 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
990 cdt->cdt_state = CDT_STOPPED;
991 CERROR("%s: error starting coordinator thread: %d\n",
992 mdt_obd_name(mdt), rc);
995 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1000 wait_event(cdt->cdt_thread.t_ctl_waitq,
1001 (cdt->cdt_thread.t_flags & SVC_RUNNING));
1003 cdt->cdt_state = CDT_RUNNING;
1004 mdt->mdt_opts.mo_coordinator = 1;
1009 * stop a coordinator thread
1010 * \param mdt [IN] device
1012 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1014 struct coordinator *cdt = &mdt->mdt_coordinator;
1015 struct cdt_agent_req *car, *tmp1;
1016 struct hsm_agent *ha, *tmp2;
1017 struct cdt_restore_handle *crh, *tmp3;
1018 struct mdt_thread_info *cdt_mti;
1021 if (cdt->cdt_state == CDT_STOPPED) {
1022 CERROR("%s: Coordinator already stopped\n",
1027 if (cdt->cdt_state != CDT_STOPPING) {
1028 /* stop coordinator thread before cleaning */
1029 cdt->cdt_thread.t_flags = SVC_STOPPING;
1030 wake_up(&cdt->cdt_thread.t_ctl_waitq);
1031 wait_event(cdt->cdt_thread.t_ctl_waitq,
1032 cdt->cdt_thread.t_flags & SVC_STOPPED);
1034 cdt->cdt_state = CDT_STOPPED;
1036 /* start cleaning */
1037 down_write(&cdt->cdt_request_lock);
1038 list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
1040 list_del(&car->car_request_list);
1041 mdt_cdt_free_request(car);
1043 up_write(&cdt->cdt_request_lock);
1045 down_write(&cdt->cdt_agent_lock);
1046 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
1047 list_del(&ha->ha_list);
1050 up_write(&cdt->cdt_agent_lock);
1052 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1053 mutex_lock(&cdt->cdt_restore_lock);
1054 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
1055 struct mdt_object *child;
1057 /* give back layout lock */
1058 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
1060 mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
1062 list_del(&crh->crh_list);
1064 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1066 mutex_unlock(&cdt->cdt_restore_lock);
1068 mdt->mdt_opts.mo_coordinator = 0;
1074 * register all requests from an hal in the memory list
1075 * \param mti [IN] context
1076 * \param hal [IN] request
1077 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1078 * which is running the CT
1080 * \retval -ve failure
1082 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1083 struct hsm_action_list *hal, struct obd_uuid *uuid)
1085 struct mdt_device *mdt = mti->mti_mdt;
1086 struct coordinator *cdt = &mdt->mdt_coordinator;
1087 struct hsm_action_item *hai;
1091 /* register request in memory list */
1092 hai = hai_first(hal);
1093 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1094 struct cdt_agent_req *car;
1096 /* in case of a cancel request, we first mark the ondisk
1097 * record of the request we want to stop as canceled
1098 * this does not change the cancel record
1099 * it will be done when updating the request status
1101 if (hai->hai_action == HSMA_CANCEL) {
1102 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1106 CERROR("%s: mdt_agent_record_update() failed, "
1107 "rc=%d, cannot update status to %s "
1108 "for cookie "LPX64"\n",
1109 mdt_obd_name(mdt), rc,
1110 agent_req_status2name(ARS_CANCELED),
1115 /* find the running request to set it canceled */
1116 car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1118 car->car_canceled = 1;
1119 /* uuid has to be changed to the one running the
1120 * request to cancel */
1121 *uuid = car->car_uuid;
1122 mdt_cdt_put_request(car);
1124 /* no need to memorize cancel request
1125 * this also avoid a deadlock when we receive
1126 * a purge all requests command
1131 if (hai->hai_action == HSMA_ARCHIVE) {
1132 struct mdt_object *obj;
1135 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1136 if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1139 GOTO(out, rc = PTR_ERR(obj));
1141 hsm.mh_flags |= HS_EXISTS;
1142 hsm.mh_arch_id = hal->hal_archive_id;
1143 rc = mdt_hsm_attr_set(mti, obj, &hsm);
1144 mdt_object_put(mti->mti_env, obj);
1149 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1150 hal->hal_archive_id, hal->hal_flags,
1153 GOTO(out, rc = PTR_ERR(car));
1155 rc = mdt_cdt_add_request(cdt, car);
1157 mdt_cdt_free_request(car);
1164 * swap layouts between 2 fids
1165 * \param mti [IN] context
1168 * \param mh_common [IN] MD HSM
1170 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1171 const lustre_fid *fid, const lustre_fid *dfid,
1172 struct md_hsm *mh_common)
1174 struct mdt_device *mdt = mti->mti_mdt;
1175 struct mdt_object *child1, *child2;
1176 struct mdt_lock_handle *lh2;
1180 child1 = mdt_object_find(mti->mti_env, mdt, fid);
1182 GOTO(out, rc = PTR_ERR(child1));
1184 /* we already have layout lock on FID so take only
1186 lh2 = &mti->mti_lh[MDT_LH_OLD];
1187 mdt_lock_reg_init(lh2, LCK_EX);
1188 child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1190 GOTO(out_child1, rc = PTR_ERR(child2));
1192 /* if copy tool closes the volatile before sending the final
1193 * progress through llapi_hsm_copy_end(), all the objects
1194 * are removed and mdd_swap_layout LBUG */
1195 if (!mdt_object_exists(child2)) {
1196 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1197 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1198 GOTO(out_child2, rc = -ENOENT);
1200 /* Since we only handle restores here, unconditionally use
1201 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1202 * be preserved in case of failure during swap_layout and not
1203 * leave a file in an intermediate but incoherent state.
1204 * But need to setup HSM xattr of data FID before, reuse
1205 * mti and mh presets for FID in hsm_cdt_request_completed(),
1206 * only need to clear RELEASED and DIRTY.
1208 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1209 rc = mdt_hsm_attr_set(mti, child2, mh_common);
1211 rc = mo_swap_layouts(mti->mti_env,
1212 mdt_object_child(child1),
1213 mdt_object_child(child2),
1214 SWAP_LAYOUTS_MDS_HSM);
1217 mdt_object_unlock_put(mti, child2, lh2, 1);
1219 mdt_object_put(mti->mti_env, child1);
1225 * update status of a completed request
1226 * \param mti [IN] context
1227 * \param pgs [IN] progress of the copy tool
1228 * \param update_record [IN] update llog record
1230 * \retval -ve failure
1232 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1233 struct hsm_progress_kernel *pgs,
1234 const struct cdt_agent_req *car,
1235 enum agent_req_status *status)
1237 const struct lu_env *env = mti->mti_env;
1238 struct mdt_device *mdt = mti->mti_mdt;
1239 struct coordinator *cdt = &mdt->mdt_coordinator;
1240 struct mdt_object *obj = NULL;
1241 int cl_flags = 0, rc = 0;
1246 /* default is to retry */
1247 *status = ARS_WAITING;
1249 /* find object by FID */
1250 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1251 /* we will update MD HSM only if needed */
1252 is_mh_changed = false;
1254 /* object removed */
1255 *status = ARS_SUCCEED;
1259 /* no need to change mh->mh_arch_id
1260 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1262 if (pgs->hpk_errval != 0) {
1263 switch (pgs->hpk_errval) {
1265 /* the copy tool does not support cancel
1266 * so the cancel request is failed
1267 * As we cannot distinguish a cancel progress
1268 * from another action progress (they have the
1269 * same cookie), we suppose here the CT returns
1270 * ENOSYS only if does not support cancel
1272 /* this can also happen when cdt calls it to
1273 * for a timed out request */
1274 *status = ARS_FAILED;
1275 /* to have a cancel event in changelog */
1276 pgs->hpk_errval = ECANCELED;
1279 /* the request record has already been set to
1280 * ARS_CANCELED, this set the cancel request
1282 *status = ARS_SUCCEED;
1285 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1286 !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1287 ARS_FAILED : ARS_WAITING);
1291 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1292 CERROR("%s: Request "LPX64" on "DFID
1293 " failed, error code %d too large\n",
1295 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1297 hsm_set_cl_error(&cl_flags,
1298 CLF_HSM_ERROVERFLOW);
1301 hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1304 switch (car->car_hai->hai_action) {
1306 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1309 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1312 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1315 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1316 CERROR("%s: Failed request "LPX64" on "DFID
1317 " cannot be a CANCEL\n",
1320 PFID(&pgs->hpk_fid));
1323 CERROR("%s: Failed request "LPX64" on "DFID
1324 " %d is an unknown action\n",
1326 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1327 car->car_hai->hai_action);
1332 *status = ARS_SUCCEED;
1333 switch (car->car_hai->hai_action) {
1335 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1336 /* set ARCHIVE keep EXIST and clear LOST and
1338 mh.mh_arch_ver = pgs->hpk_data_version;
1339 mh.mh_flags |= HS_ARCHIVED;
1340 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1341 is_mh_changed = true;
1344 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1346 /* do not clear RELEASED and DIRTY here
1347 * this will occur in hsm_swap_layouts()
1350 /* Restoring has changed the file version on
1352 mh.mh_arch_ver = pgs->hpk_data_version;
1353 is_mh_changed = true;
1356 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1357 /* clear ARCHIVED EXISTS and LOST */
1358 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1359 is_mh_changed = true;
1362 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1363 CERROR("%s: Successful request "LPX64
1365 " cannot be a CANCEL\n",
1368 PFID(&pgs->hpk_fid));
1371 CERROR("%s: Successful request "LPX64
1373 " %d is an unknown action\n",
1375 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1376 car->car_hai->hai_action);
1382 /* rc != 0 means error when analysing action, it may come from
1383 * a crasy CT no need to manage DIRTY
1386 hsm_set_cl_flags(&cl_flags,
1387 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1389 /* unlock is done later, after layout lock management */
1391 rc = mdt_hsm_attr_set(mti, obj, &mh);
1394 /* we give back layout lock only if restore was successful or
1395 * if restore was canceled or if policy is to not retry
1396 * in other cases we just unlock the object */
1397 if (car->car_hai->hai_action == HSMA_RESTORE &&
1398 (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1399 cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1400 struct cdt_restore_handle *crh;
1402 /* restore in data FID done, we swap the layouts
1403 * only if restore is successfull */
1404 if (pgs->hpk_errval == 0) {
1405 rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1406 &car->car_hai->hai_dfid, &mh);
1408 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1409 *status = ARS_FAILED;
1410 pgs->hpk_errval = -rc;
1413 /* we have to retry, so keep layout lock */
1414 if (*status == ARS_WAITING)
1417 /* give back layout lock */
1418 mutex_lock(&cdt->cdt_restore_lock);
1419 crh = hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1421 list_del(&crh->crh_list);
1422 mutex_unlock(&cdt->cdt_restore_lock);
1423 /* just give back layout lock, we keep
1424 * the reference which is given back
1425 * later with the lock for HSM flags */
1426 if (!IS_ERR(obj) && crh != NULL)
1427 mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1430 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1436 if (obj != NULL && !IS_ERR(obj)) {
1437 mo_changelog(env, CL_HSM, cl_flags,
1438 mdt_object_child(obj));
1439 mdt_object_put(mti->mti_env, obj);
1446 * update status of a request
1447 * \param mti [IN] context
1448 * \param pgs [IN] progress of the copy tool
1449 * \param update_record [IN] update llog record
1451 * \retval -ve failure
1453 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1454 struct hsm_progress_kernel *pgs,
1455 const int update_record)
1457 struct mdt_device *mdt = mti->mti_mdt;
1458 struct coordinator *cdt = &mdt->mdt_coordinator;
1459 struct cdt_agent_req *car;
1463 /* no coordinator started, so we cannot serve requests */
1464 if (cdt->cdt_state == CDT_STOPPED)
1467 /* first do sanity checks */
1468 car = mdt_cdt_update_request(cdt, pgs);
1470 CERROR("%s: Cannot find running request for cookie "LPX64
1473 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1476 RETURN(PTR_ERR(car));
1479 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1480 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1481 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1482 hsm_copytool_action2name(car->car_hai->hai_action),
1483 pgs->hpk_flags, pgs->hpk_errval,
1484 PFID(&car->car_hai->hai_fid),
1485 PFID(&car->car_hai->hai_dfid));
1487 /* progress is done on FID or data FID depending of the action and
1488 * of the copy progress */
1489 /* for restore progress is used to send back the data FID to cdt */
1490 if (car->car_hai->hai_action == HSMA_RESTORE &&
1491 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1492 car->car_hai->hai_dfid = pgs->hpk_fid;
1494 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1495 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1496 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1497 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1498 CERROR("%s: Progress on "DFID" for cookie "LPX64
1499 " does not match request FID "DFID" nor data FID "
1502 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1503 PFID(&car->car_hai->hai_fid),
1504 PFID(&car->car_hai->hai_dfid));
1505 GOTO(out, rc = -EINVAL);
1508 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1509 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1510 " is not coherent (err=%d and not completed"
1513 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1514 hsm_copytool_action2name(car->car_hai->hai_action),
1515 pgs->hpk_errval, pgs->hpk_flags);
1516 GOTO(out, rc = -EINVAL);
1519 /* now progress is valid */
1521 /* we use a root like ucred */
1522 hsm_init_ucred(mdt_ucred(mti));
1524 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1525 enum agent_req_status status;
1527 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1529 /* remove request from memory list */
1530 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1532 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1533 " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1535 hsm_copytool_action2name(car->car_hai->hai_action),
1536 agent_req_status2name(status));
1538 if (update_record) {
1541 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1542 &pgs->hpk_cookie, 1,
1545 CERROR("%s: mdt_agent_record_update() failed,"
1546 " rc=%d, cannot update status to %s"
1547 " for cookie "LPX64"\n",
1548 mdt_obd_name(mdt), rc1,
1549 agent_req_status2name(status),
1551 rc = (rc != 0 ? rc : rc1);
1553 /* ct has completed a request, so a slot is available, wakeup
1554 * cdt to find new work */
1555 mdt_hsm_cdt_wakeup(mdt);
1557 /* if copytool send a progress on a canceled request
1558 * we inform copytool it should stop
1560 if (car->car_canceled == 1)
1566 /* remove ref got from mdt_cdt_update_request() */
1567 mdt_cdt_put_request(car);
1574 * data passed to llog_cat_process() callback
1575 * to cancel requests
1577 struct hsm_cancel_all_data {
1578 struct mdt_device *mdt;
1582 * llog_cat_process() callback, used to:
1583 * - purge all requests
1584 * \param env [IN] environment
1585 * \param llh [IN] llog handle
1586 * \param hdr [IN] llog record
1587 * \param data [IN] cb data = struct hsm_cancel_all_data
1589 * \retval -ve failure
1591 static int mdt_cancel_all_cb(const struct lu_env *env,
1592 struct llog_handle *llh,
1593 struct llog_rec_hdr *hdr, void *data)
1595 struct llog_agent_req_rec *larr;
1596 struct hsm_cancel_all_data *hcad;
1600 larr = (struct llog_agent_req_rec *)hdr;
1602 if (larr->arr_status == ARS_WAITING ||
1603 larr->arr_status == ARS_STARTED) {
1604 larr->arr_status = ARS_CANCELED;
1605 larr->arr_req_change = cfs_time_current_sec();
1606 rc = mdt_agent_llog_update_rec(env, hcad->mdt, llh, larr);
1608 RETURN(LLOG_DEL_RECORD);
1614 * cancel all actions
1615 * \param obd [IN] MDT device
1617 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1619 struct mdt_thread_info *mti;
1620 struct coordinator *cdt = &mdt->mdt_coordinator;
1621 struct cdt_agent_req *car;
1622 struct hsm_action_list *hal = NULL;
1623 struct hsm_action_item *hai;
1624 struct hsm_cancel_all_data hcad;
1625 int hal_sz = 0, hal_len, rc;
1626 enum cdt_states save_state;
1629 /* retrieve coordinator context */
1630 mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1632 /* disable coordinator */
1633 save_state = cdt->cdt_state;
1634 cdt->cdt_state = CDT_DISABLE;
1636 /* send cancel to all running requests */
1637 down_read(&cdt->cdt_request_lock);
1638 list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1639 mdt_cdt_get_request(car);
1640 /* request is not yet removed from list, it will be done
1641 * when copytool will return progress
1644 if (car->car_hai->hai_action == HSMA_CANCEL) {
1645 mdt_cdt_put_request(car);
1650 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1651 cfs_size_round(car->car_hai->hai_len);
1653 if (hal_len > hal_sz && hal_sz > 0) {
1654 /* not enough room, free old buffer */
1655 OBD_FREE(hal, hal_sz);
1659 /* empty buffer, allocate one */
1662 OBD_ALLOC(hal, hal_sz);
1664 mdt_cdt_put_request(car);
1665 up_read(&cdt->cdt_request_lock);
1666 GOTO(out, rc = -ENOMEM);
1670 hal->hal_version = HAL_VERSION;
1671 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1673 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1674 hal->hal_compound_id = car->car_compound_id;
1675 hal->hal_archive_id = car->car_archive_id;
1676 hal->hal_flags = car->car_flags;
1679 hai = hai_first(hal);
1680 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1681 hai->hai_action = HSMA_CANCEL;
1684 /* it is possible to safely call mdt_hsm_agent_send()
1685 * (ie without a deadlock on cdt_request_lock), because the
1686 * write lock is taken only if we are not in purge mode
1687 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1688 * nor mdt_cdt_remove_request())
1690 /* no conflict with cdt thread because cdt is disable and we
1691 * have the request lock */
1692 mdt_hsm_agent_send(mti, hal, 1);
1694 mdt_cdt_put_request(car);
1696 up_read(&cdt->cdt_request_lock);
1699 OBD_FREE(hal, hal_sz);
1701 /* cancel all on-disk records */
1704 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1705 mdt_cancel_all_cb, &hcad);
1707 /* enable coordinator */
1708 cdt->cdt_state = save_state;
1714 * check if a request is comptaible with file status
1715 * \param hai [IN] request description
1716 * \param hal_an [IN] request archive number (not used)
1717 * \param rq_flags [IN] request flags
1718 * \param hsm [IN] file HSM metadata
1721 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1722 const int hal_an, const __u64 rq_flags,
1723 const struct md_hsm *hsm)
1725 int is_compat = false;
1729 hsm_flags = hsm->mh_flags;
1730 switch (hai->hai_action) {
1732 if (!(hsm_flags & HS_NOARCHIVE) &&
1733 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1737 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1738 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1742 if (!(hsm_flags & HS_RELEASED) &&
1743 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1750 CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1751 " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1752 PFID(&hai->hai_fid),
1753 hsm_copytool_action2name(hai->hai_action), rq_flags,
1754 hai->hai_extent.offset, hai->hai_extent.length,
1756 (is_compat ? "compatible" : "uncompatible"));
1762 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1764 static const struct {
1768 } hsm_policy_names[] = {
1769 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1770 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1775 * convert a policy name to a bit
1776 * \param name [IN] policy name
1778 * \retval policy bit
1780 static __u64 hsm_policy_str2bit(const char *name)
1784 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1785 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1786 strcmp(hsm_policy_names[i].name, name) == 0)
1787 return hsm_policy_names[i].bit;
1792 * convert a policy bit field to a string
1793 * \param mask [IN] policy bit field
1794 * \param hexa [IN] print mask before bit names
1795 * \param buffer [OUT] string
1796 * \param count [IN] size of buffer
1798 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1806 seq_printf(m, "("LPX64") ", mask);
1808 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1811 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1812 if (hsm_policy_names[j].bit == bit)
1816 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1818 seq_printf(m, "%s ", hsm_policy_names[j].name);
1820 /* remove last ' ' */
1825 /* methods to read/write HSM policy flags */
1826 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1828 struct mdt_device *mdt = m->private;
1829 struct coordinator *cdt = &mdt->mdt_coordinator;
1832 hsm_policy_bit2str(m, cdt->cdt_policy, false);
1837 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1838 size_t count, loff_t *off)
1840 struct seq_file *m = file->private_data;
1841 struct mdt_device *mdt = m->private;
1842 struct coordinator *cdt = &mdt->mdt_coordinator;
1843 char *start, *token, sign;
1846 __u64 add_mask, remove_mask, set_mask;
1850 if (count + 1 > PAGE_SIZE)
1853 OBD_ALLOC(buf, count + 1);
1857 if (copy_from_user(buf, buffer, count))
1858 GOTO(out, rc = -EFAULT);
1863 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1866 add_mask = remove_mask = set_mask = 0;
1868 token = strsep(&start, "\n ");
1874 if (sign == '-' || sign == '+')
1877 policy = hsm_policy_str2bit(token);
1879 CWARN("%s: '%s' is unknown, "
1880 "supported policies are:\n", mdt_obd_name(mdt),
1882 hsm_policy_bit2str(m, 0, false);
1883 GOTO(out, rc = -EINVAL);
1887 remove_mask |= policy;
1897 } while (start != NULL);
1899 CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1900 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1902 /* if no sign in all string, it is a clear and set
1903 * if some sign found, all unsigned are converted
1905 * P1 P2 = set to P1 and P2
1906 * P1 -P2 = add P1 clear P2 same as +P1 -P2
1908 if (remove_mask == 0 && add_mask == 0) {
1909 cdt->cdt_policy = set_mask;
1911 cdt->cdt_policy |= set_mask | add_mask;
1912 cdt->cdt_policy &= ~remove_mask;
1915 GOTO(out, rc = count);
1918 OBD_FREE(buf, count + 1);
1921 LPROC_SEQ_FOPS(mdt_hsm_policy);
1923 #define GENERATE_PROC_METHOD(VAR) \
1924 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
1926 struct mdt_device *mdt = m->private; \
1927 struct coordinator *cdt = &mdt->mdt_coordinator; \
1930 seq_printf(m, LPU64"\n", (__u64)cdt->VAR); \
1934 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1935 size_t count, loff_t *off) \
1938 struct seq_file *m = file->private_data; \
1939 struct mdt_device *mdt = m->private; \
1940 struct coordinator *cdt = &mdt->mdt_coordinator; \
1945 rc = lprocfs_write_helper(buffer, count, &val); \
1955 GENERATE_PROC_METHOD(cdt_loop_period)
1956 GENERATE_PROC_METHOD(cdt_grace_delay)
1957 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1958 GENERATE_PROC_METHOD(cdt_max_requests)
1959 GENERATE_PROC_METHOD(cdt_default_archive_id)
1962 * procfs write method for MDT/hsm_control
1963 * proc entry is in mdt directory so data is mdt obd_device pointer
1965 #define CDT_ENABLE_CMD "enabled"
1966 #define CDT_STOP_CMD "shutdown"
1967 #define CDT_DISABLE_CMD "disabled"
1968 #define CDT_PURGE_CMD "purge"
1969 #define CDT_HELP_CMD "help"
1970 #define CDT_MAX_CMD_LEN 10
1973 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1974 size_t count, loff_t *off)
1976 struct seq_file *m = file->private_data;
1977 struct obd_device *obd = m->private;
1978 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1979 struct coordinator *cdt = &(mdt->mdt_coordinator);
1981 char kernbuf[CDT_MAX_CMD_LEN];
1984 if (count == 0 || count >= sizeof(kernbuf))
1987 if (copy_from_user(kernbuf, buffer, count))
1991 if (kernbuf[count - 1] == '\n')
1992 kernbuf[count - 1] = 0;
1995 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1996 if (cdt->cdt_state == CDT_DISABLE) {
1997 cdt->cdt_state = CDT_RUNNING;
1998 mdt_hsm_cdt_wakeup(mdt);
2000 rc = mdt_hsm_cdt_start(mdt);
2002 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2003 if ((cdt->cdt_state == CDT_STOPPING) ||
2004 (cdt->cdt_state == CDT_STOPPED)) {
2005 CERROR("%s: Coordinator already stopped\n",
2009 cdt->cdt_state = CDT_STOPPING;
2011 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2012 if ((cdt->cdt_state == CDT_STOPPING) ||
2013 (cdt->cdt_state == CDT_STOPPED)) {
2014 CERROR("%s: Coordinator is stopped\n",
2018 cdt->cdt_state = CDT_DISABLE;
2020 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2021 rc = hsm_cancel_all_actions(mdt);
2022 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2030 CERROR("%s: Valid coordinator control commands are: "
2031 "%s %s %s %s %s\n", mdt_obd_name(mdt),
2032 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2033 CDT_PURGE_CMD, CDT_HELP_CMD);
2041 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2043 struct obd_device *obd = m->private;
2044 struct coordinator *cdt;
2047 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2049 if (cdt->cdt_state == CDT_INIT)
2050 seq_printf(m, "init\n");
2051 else if (cdt->cdt_state == CDT_RUNNING)
2052 seq_printf(m, "enabled\n");
2053 else if (cdt->cdt_state == CDT_STOPPING)
2054 seq_printf(m, "stopping\n");
2055 else if (cdt->cdt_state == CDT_STOPPED)
2056 seq_printf(m, "stopped\n");
2057 else if (cdt->cdt_state == CDT_DISABLE)
2058 seq_printf(m, "disabled\n");
2060 seq_printf(m, "unknown\n");
2066 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2071 for (i = 0; i < 8 * sizeof(mask); i++) {
2072 if (mask & (1UL << i))
2073 rc += seq_printf(m, "%s%s", rc == 0 ? "" : " ",
2074 hsm_copytool_action2name(i));
2076 rc += seq_printf(m, "\n");
2082 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2084 struct mdt_device *mdt = m->private;
2085 struct coordinator *cdt = &mdt->mdt_coordinator;
2087 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2091 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2093 struct mdt_device *mdt = m->private;
2094 struct coordinator *cdt = &mdt->mdt_coordinator;
2096 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2100 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2102 struct mdt_device *mdt = m->private;
2103 struct coordinator *cdt = &mdt->mdt_coordinator;
2105 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2108 static inline enum hsm_copytool_action
2109 hsm_copytool_name2action(const char *name)
2111 if (strcasecmp(name, "NOOP") == 0)
2113 else if (strcasecmp(name, "ARCHIVE") == 0)
2114 return HSMA_ARCHIVE;
2115 else if (strcasecmp(name, "RESTORE") == 0)
2116 return HSMA_RESTORE;
2117 else if (strcasecmp(name, "REMOVE") == 0)
2119 else if (strcasecmp(name, "CANCEL") == 0)
2126 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2127 size_t user_count, __u64 *mask)
2129 char *buf, *pos, *name;
2135 if (!(user_count < 4096))
2138 buf_size = user_count + 1;
2140 OBD_ALLOC(buf, buf_size);
2144 if (copy_from_user(buf, user_buf, buf_size - 1))
2145 GOTO(out, rc = -EFAULT);
2147 buf[buf_size - 1] = '\0';
2150 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2156 action = hsm_copytool_name2action(name);
2158 GOTO(out, rc = -EINVAL);
2160 new_mask |= (1UL << action);
2166 OBD_FREE(buf, buf_size);
2172 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2173 size_t count, loff_t *off)
2175 struct seq_file *m = file->private_data;
2176 struct mdt_device *mdt = m->private;
2177 struct coordinator *cdt = &mdt->mdt_coordinator;
2179 return mdt_write_hsm_request_mask(file, buf, count,
2180 &cdt->cdt_user_request_mask);
2184 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2185 size_t count, loff_t *off)
2187 struct seq_file *m = file->private_data;
2188 struct mdt_device *mdt = m->private;
2189 struct coordinator *cdt = &mdt->mdt_coordinator;
2191 return mdt_write_hsm_request_mask(file, buf, count,
2192 &cdt->cdt_group_request_mask);
2196 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2197 size_t count, loff_t *off)
2199 struct seq_file *m = file->private_data;
2200 struct mdt_device *mdt = m->private;
2201 struct coordinator *cdt = &mdt->mdt_coordinator;
2203 return mdt_write_hsm_request_mask(file, buf, count,
2204 &cdt->cdt_other_request_mask);
2207 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2208 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2209 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2210 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2211 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2212 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2213 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2214 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2216 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2218 .fops = &mdt_hsm_agent_fops },
2219 { .name = "actions",
2220 .fops = &mdt_hsm_actions_fops,
2221 .proc_mode = 0444 },
2222 { .name = "default_archive_id",
2223 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2224 { .name = "grace_delay",
2225 .fops = &mdt_hsm_cdt_grace_delay_fops },
2226 { .name = "loop_period",
2227 .fops = &mdt_hsm_cdt_loop_period_fops },
2228 { .name = "max_requests",
2229 .fops = &mdt_hsm_cdt_max_requests_fops },
2231 .fops = &mdt_hsm_policy_fops },
2232 { .name = "active_request_timeout",
2233 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2234 { .name = "active_requests",
2235 .fops = &mdt_hsm_active_requests_fops },
2236 { .name = "user_request_mask",
2237 .fops = &mdt_hsm_user_request_mask_fops, },
2238 { .name = "group_request_mask",
2239 .fops = &mdt_hsm_group_request_mask_fops, },
2240 { .name = "other_request_mask",
2241 .fops = &mdt_hsm_other_request_mask_fops, },