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) 2012, 2013, Intel Corporation.
24 * Use is subject to license terms.
25 * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
29 * lustre/mdt/mdt_coordinator.c
31 * Lustre HSM Coordinator
33 * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
34 * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
35 * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <obd_support.h>
41 #include <lustre_net.h>
42 #include <lustre_export.h>
44 #include <lprocfs_status.h>
45 #include <lustre_log.h>
46 #include "mdt_internal.h"
48 static struct lprocfs_seq_vars lprocfs_mdt_hsm_vars[];
51 * get obj and HSM attributes on a fid
52 * \param mti [IN] context
53 * \param fid [IN] object fid
54 * \param hsm [OUT] HSM meta data
55 * \retval obj or error (-ENOENT if not found)
57 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
58 const struct lu_fid *fid,
62 struct mdt_object *obj;
70 /* find object by FID */
71 obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
75 if (!mdt_object_exists(obj)) {
77 mdt_object_put(mti->mti_env, obj);
78 RETURN(ERR_PTR(-ENOENT));
81 rc = mdt_attr_get_complex(mti, obj, ma);
83 mdt_object_put(mti->mti_env, obj);
87 if (ma->ma_valid & MA_HSM)
90 memset(hsm, 0, sizeof(*hsm));
95 void mdt_hsm_dump_hal(int level, const char *prefix,
96 struct hsm_action_list *hal)
99 struct hsm_action_item *hai;
102 CDEBUG(level, "%s: HAL header: version %X count %d compound "LPX64
103 " archive_id %d flags "LPX64"\n",
104 prefix, hal->hal_version, hal->hal_count,
105 hal->hal_compound_id, hal->hal_archive_id, hal->hal_flags);
107 hai = hai_first(hal);
108 for (i = 0; i < hal->hal_count; i++) {
109 sz = hai->hai_len - sizeof(*hai);
110 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
111 " compound/cookie="LPX64"/"LPX64
112 " action=%s extent="LPX64"-"LPX64" gid="LPX64
113 " datalen=%d data=[%s]\n",
115 PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
116 hal->hal_compound_id, hai->hai_cookie,
117 hsm_copytool_action2name(hai->hai_action),
118 hai->hai_extent.offset,
119 hai->hai_extent.length,
121 hai_dump_data_field(hai, buf, sizeof(buf)));
127 * data passed to llog_cat_process() callback
128 * to scan requests and take actions
130 struct hsm_scan_data {
131 struct mdt_thread_info *mti;
132 char fs_name[MTI_NAME_MAXLEN+1];
133 /* request to be send to agents */
134 int request_sz; /** allocated size */
135 int max_requests; /** vector size */
136 int request_cnt; /** used count */
140 struct hsm_action_list *hal;
142 /* records to be canceled */
143 int max_cookie; /** vector size */
144 int cookie_cnt; /** used count */
149 * llog_cat_process() callback, used to:
150 * - find waiting request and start action
151 * - purge canceled and done requests
152 * \param env [IN] environment
153 * \param llh [IN] llog handle
154 * \param hdr [IN] llog record
155 * \param data [IN/OUT] cb data = struct hsm_scan_data
157 * \retval -ve failure
159 static int mdt_coordinator_cb(const struct lu_env *env,
160 struct llog_handle *llh,
161 struct llog_rec_hdr *hdr,
164 const struct llog_agent_req_rec *larr;
165 struct hsm_scan_data *hsd;
166 struct hsm_action_item *hai;
167 struct mdt_device *mdt;
168 struct coordinator *cdt;
173 mdt = hsd->mti->mti_mdt;
174 cdt = &mdt->mdt_coordinator;
176 larr = (struct llog_agent_req_rec *)hdr;
177 dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
178 switch (larr->arr_status) {
180 int i, empty_slot, found;
182 /* Are agents full? */
183 if (atomic_read(&cdt->cdt_request_count) ==
184 cdt->cdt_max_requests)
187 /* first search if the request if known in the list we have
188 * build and if there is room in the request vector */
191 for (i = 0; i < hsd->max_requests &&
192 (empty_slot == -1 || found == -1); i++) {
193 if (hsd->request[i].hal == NULL) {
197 if (hsd->request[i].hal->hal_compound_id ==
198 larr->arr_compound_id) {
203 if (found == -1 && empty_slot == -1)
204 /* unknown request and no more room for new request,
205 * continue scan for to find other entries for
206 * already found request
211 struct hsm_action_list *hal;
213 /* request is not already known */
214 /* allocates hai vector size just needs to be large
216 hsd->request[empty_slot].hal_sz =
217 sizeof(*hsd->request[empty_slot].hal) +
218 cfs_size_round(MTI_NAME_MAXLEN+1) +
219 2 * cfs_size_round(larr->arr_hai.hai_len);
220 OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
222 CERROR("%s: Cannot allocate memory (%d o)"
223 "for compound "LPX64"\n",
225 hsd->request[i].hal_sz,
226 larr->arr_compound_id);
229 hal->hal_version = HAL_VERSION;
230 strlcpy(hal->hal_fsname, hsd->fs_name,
231 MTI_NAME_MAXLEN + 1);
232 hal->hal_compound_id = larr->arr_compound_id;
233 hal->hal_archive_id = larr->arr_archive_id;
234 hal->hal_flags = larr->arr_flags;
236 hsd->request[empty_slot].hal_used_sz = hal_size(hal);
237 hsd->request[empty_slot].hal = hal;
240 hai = hai_first(hal);
242 /* request is known */
243 /* we check if record archive num is the same as the
244 * known request, if not we will serve it in multiple
245 * time because we do not know if the agent can serve
247 * a use case is a compound made of multiple restore
248 * where the files are not archived in the same backend
250 if (larr->arr_archive_id !=
251 hsd->request[found].hal->hal_archive_id)
254 if (hsd->request[found].hal_sz <
255 hsd->request[found].hal_used_sz +
256 cfs_size_round(larr->arr_hai.hai_len)) {
257 /* Not enough room, need an extension */
261 sz = 2 * hsd->request[found].hal_sz;
262 OBD_ALLOC(hal_buffer, sz);
264 CERROR("%s: Cannot allocate memory "
265 "(%d o) for compound "LPX64"\n",
266 mdt_obd_name(mdt), sz,
267 larr->arr_compound_id);
270 memcpy(hal_buffer, hsd->request[found].hal,
271 hsd->request[found].hal_used_sz);
272 OBD_FREE(hsd->request[found].hal,
273 hsd->request[found].hal_sz);
274 hsd->request[found].hal = hal_buffer;
275 hsd->request[found].hal_sz = sz;
277 hai = hai_first(hsd->request[found].hal);
278 for (i = 0; i < hsd->request[found].hal->hal_count;
282 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
283 hai->hai_cookie = larr->arr_hai.hai_cookie;
284 hai->hai_gid = larr->arr_hai.hai_gid;
286 hsd->request[found].hal_used_sz +=
287 cfs_size_round(hai->hai_len);
288 hsd->request[found].hal->hal_count++;
292 struct cdt_agent_req *car;
295 /* we search for a running request
296 * error may happen if coordinator crashes or stopped
297 * with running request
299 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
301 last = larr->arr_req_create;
303 last = car->car_req_update;
304 mdt_cdt_put_request(car);
307 /* test if request too long, if yes cancel it
308 * the same way the copy tool acknowledge a cancel request */
309 if ((last + cdt->cdt_active_req_timeout)
310 < cfs_time_current_sec()) {
311 struct hsm_progress_kernel pgs;
313 dump_llog_agent_req_rec("mdt_coordinator_cb(): "
314 "request timeouted, start "
316 /* a too old cancel request just needs to be removed
317 * this can happen, if copy tool does not support cancel
318 * for other requests, we have to remove the running
319 * request and notify the copytool
321 pgs.hpk_fid = larr->arr_hai.hai_fid;
322 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
323 pgs.hpk_extent = larr->arr_hai.hai_extent;
324 pgs.hpk_flags = HP_FLAG_COMPLETED;
325 pgs.hpk_errval = ENOSYS;
326 pgs.hpk_data_version = 0;
327 /* update request state, but do not record in llog, to
328 * avoid deadlock on cdt_llog_lock
330 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
332 CERROR("%s: Cannot cleanup timeouted request: "
333 DFID" for cookie "LPX64" action=%s\n",
335 PFID(&pgs.hpk_fid), pgs.hpk_cookie,
336 hsm_copytool_action2name(
337 larr->arr_hai.hai_action));
340 /* The request no longer exists, forget
341 * about it, and do not send a cancel request
342 * to the client, for which an error will be
343 * sent back, leading to an endless cycle of
345 RETURN(LLOG_DEL_RECORD);
348 /* add the cookie to the list of record to be
349 * canceled by caller */
350 if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
351 __u64 *ptr, *old_ptr;
352 int old_sz, new_sz, new_cnt;
354 /* need to increase vector size */
355 old_sz = sizeof(__u64) * hsd->max_cookie;
356 old_ptr = hsd->cookies;
358 new_cnt = 2 * hsd->max_cookie;
359 new_sz = sizeof(__u64) * new_cnt;
361 OBD_ALLOC(ptr, new_sz);
363 CERROR("%s: Cannot allocate memory "
364 "(%d o) for cookie vector\n",
365 mdt_obd_name(mdt), new_sz);
368 memcpy(ptr, hsd->cookies, old_sz);
370 hsd->max_cookie = new_cnt;
371 OBD_FREE(old_ptr, old_sz);
373 hsd->cookies[hsd->cookie_cnt] =
374 larr->arr_hai.hai_cookie;
382 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
383 cfs_time_current_sec())
384 RETURN(LLOG_DEL_RECORD);
391 * create /proc entries for coordinator
394 * \retval -ve failure
396 int hsm_cdt_procfs_init(struct mdt_device *mdt)
398 struct coordinator *cdt = &mdt->mdt_coordinator;
402 /* init /proc entries, failure is not critical */
403 cdt->cdt_proc_dir = lprocfs_seq_register("hsm",
404 mdt2obd_dev(mdt)->obd_proc_entry,
405 lprocfs_mdt_hsm_vars, mdt);
406 if (IS_ERR(cdt->cdt_proc_dir)) {
407 rc = PTR_ERR(cdt->cdt_proc_dir);
408 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
409 " rc=%d\n", mdt_obd_name(mdt), rc);
410 cdt->cdt_proc_dir = NULL;
418 * remove /proc entries for coordinator
421 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
423 struct coordinator *cdt = &mdt->mdt_coordinator;
425 LASSERT(cdt->cdt_state == CDT_STOPPED);
426 if (cdt->cdt_proc_dir != NULL)
427 lprocfs_remove(&cdt->cdt_proc_dir);
431 * get vector of hsm cdt /proc vars
435 struct lprocfs_seq_vars *hsm_cdt_get_proc_vars(void)
437 return lprocfs_mdt_hsm_vars;
442 * \param data [IN] obd device
444 * \retval -ve failure
446 static int mdt_coordinator(void *data)
448 struct mdt_thread_info *mti = data;
449 struct mdt_device *mdt = mti->mti_mdt;
450 struct coordinator *cdt = &mdt->mdt_coordinator;
451 struct hsm_scan_data hsd = { 0 };
455 cdt->cdt_thread.t_flags = SVC_RUNNING;
456 wake_up(&cdt->cdt_thread.t_ctl_waitq);
458 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
459 mdt_obd_name(mdt), current_pid());
461 /* timeouted cookie vector initialization */
465 /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
466 * increases due to a change from /proc we do not overflow the
467 * hsd.request[] vector
469 hsd.max_requests = cdt->cdt_max_requests;
470 hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
471 OBD_ALLOC(hsd.request, hsd.request_sz);
473 GOTO(out, rc = -ENOMEM);
476 obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
479 struct l_wait_info lwi;
482 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
484 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
485 (cdt->cdt_thread.t_flags &
486 (SVC_STOPPING|SVC_EVENT)),
489 CDEBUG(D_HSM, "coordinator resumes\n");
491 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
492 cdt->cdt_state == CDT_STOPPING) {
493 cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
498 /* wake up before timeout, new work arrives */
499 if (cdt->cdt_thread.t_flags & SVC_EVENT)
500 cdt->cdt_thread.t_flags &= ~SVC_EVENT;
502 /* if coordinator is suspended continue to wait */
503 if (cdt->cdt_state == CDT_DISABLE) {
504 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
508 CDEBUG(D_HSM, "coordinator starts reading llog\n");
510 if (hsd.max_requests != cdt->cdt_max_requests) {
511 /* cdt_max_requests has changed,
512 * we need to allocate a new buffer
514 OBD_FREE(hsd.request, hsd.request_sz);
515 hsd.max_requests = cdt->cdt_max_requests;
517 hsd.max_requests * sizeof(*hsd.request);
518 OBD_ALLOC(hsd.request, hsd.request_sz);
525 /* create canceled cookie vector for an arbitrary size
526 * if needed, vector will grow during llog scan
530 OBD_ALLOC(hsd.cookies, hsd.max_cookie * sizeof(__u64));
537 rc = cdt_llog_process(mti->mti_env, mdt,
538 mdt_coordinator_cb, &hsd);
542 CDEBUG(D_HSM, "Found %d requests to send and %d"
543 " requests to cancel\n",
544 hsd.request_cnt, hsd.cookie_cnt);
545 /* first we cancel llog records of the timeouted requests */
546 if (hsd.cookie_cnt > 0) {
547 rc = mdt_agent_record_update(mti->mti_env, mdt,
552 CERROR("%s: mdt_agent_record_update() failed, "
553 "rc=%d, cannot update status to %s "
555 mdt_obd_name(mdt), rc,
556 agent_req_status2name(ARS_CANCELED),
560 if (list_empty(&cdt->cdt_agents)) {
561 CDEBUG(D_HSM, "no agent available, "
562 "coordinator sleeps\n");
566 /* here hsd contains a list of requests to be started */
567 for (i = 0; i < hsd.max_requests; i++) {
568 struct hsm_action_list *hal;
569 struct hsm_action_item *hai;
572 enum agent_req_status status;
574 /* still room for work ? */
575 if (atomic_read(&cdt->cdt_request_count) ==
576 cdt->cdt_max_requests)
579 if (hsd.request[i].hal == NULL)
582 /* found a request, we start it */
583 /* kuc payload allocation so we avoid an additionnal
584 * allocation in mdt_hsm_agent_send()
586 hal = kuc_alloc(hsd.request[i].hal_used_sz,
587 KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
589 CERROR("%s: Cannot allocate memory (%d o) "
590 "for compound "LPX64"\n",
592 hsd.request[i].hal_used_sz,
593 hsd.request[i].hal->hal_compound_id);
596 memcpy(hal, hsd.request[i].hal,
597 hsd.request[i].hal_used_sz);
599 rc = mdt_hsm_agent_send(mti, hal, 0);
600 /* if failure, we suppose it is temporary
601 * if the copy tool failed to do the request
602 * it has to use hsm_progress
604 status = (rc ? ARS_WAITING : ARS_STARTED);
606 /* set up cookie vector to set records status
607 * after copy tools start or failed
609 sz = hsd.request[i].hal->hal_count * sizeof(__u64);
610 OBD_ALLOC(cookies, sz);
611 if (cookies == NULL) {
612 CERROR("%s: Cannot allocate memory (%d o) "
613 "for cookies vector "LPX64"\n",
614 mdt_obd_name(mdt), sz,
615 hsd.request[i].hal->hal_compound_id);
616 kuc_free(hal, hsd.request[i].hal_used_sz);
619 hai = hai_first(hal);
620 for (j = 0; j < hsd.request[i].hal->hal_count; j++) {
621 cookies[j] = hai->hai_cookie;
625 rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
626 hsd.request[i].hal->hal_count,
629 CERROR("%s: mdt_agent_record_update() failed, "
630 "rc=%d, cannot update status to %s "
632 mdt_obd_name(mdt), rc,
633 agent_req_status2name(status),
634 hsd.request[i].hal->hal_count);
636 OBD_FREE(cookies, sz);
637 kuc_free(hal, hsd.request[i].hal_used_sz);
640 /* free cookie vector allocated for/by callback */
642 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
648 /* free hal allocated by callback */
649 for (i = 0; i < hsd.max_requests; i++) {
650 if (hsd.request[i].hal) {
651 OBD_FREE(hsd.request[i].hal,
652 hsd.request[i].hal_sz);
653 hsd.request[i].hal_sz = 0;
654 hsd.request[i].hal = NULL;
658 LASSERT(hsd.request_cnt == 0);
660 /* reset callback data */
661 memset(hsd.request, 0, hsd.request_sz);
666 OBD_FREE(hsd.request, hsd.request_sz);
669 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
671 if (cdt->cdt_state == CDT_STOPPING) {
672 /* request comes from /proc path, so we need to clean cdt
674 mdt_hsm_cdt_stop(mdt);
675 mdt->mdt_opts.mo_coordinator = 0;
677 /* request comes from a thread event, generated
678 * by mdt_stop_coordinator(), we have to ack
679 * and cdt cleaning will be done by event sender
681 cdt->cdt_thread.t_flags = SVC_STOPPED;
682 wake_up(&cdt->cdt_thread.t_ctl_waitq);
686 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
687 mdt_obd_name(mdt), current_pid(), rc);
689 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
691 mdt_obd_name(mdt), current_pid());
697 * lookup a restore handle by FID
698 * caller needs to hold cdt_restore_lock
699 * \param cdt [IN] coordinator
700 * \param fid [IN] FID
701 * \retval cdt_restore_handle found
702 * \retval NULL not found
704 static struct cdt_restore_handle *hsm_restore_hdl_find(struct coordinator *cdt,
705 const struct lu_fid *fid)
707 struct cdt_restore_handle *crh;
710 list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
711 if (lu_fid_eq(&crh->crh_fid, fid))
718 * data passed to llog_cat_process() callback
719 * to scan requests and take actions
721 struct hsm_restore_data {
722 struct mdt_thread_info *hrd_mti;
726 * llog_cat_process() callback, used to:
727 * - find restore request and allocate the restore handle
728 * \param env [IN] environment
729 * \param llh [IN] llog handle
730 * \param hdr [IN] llog record
731 * \param data [IN/OUT] cb data = struct hsm_restore_data
733 * \retval -ve failure
735 static int hsm_restore_cb(const struct lu_env *env,
736 struct llog_handle *llh,
737 struct llog_rec_hdr *hdr, void *data)
739 struct llog_agent_req_rec *larr;
740 struct hsm_restore_data *hrd;
741 struct cdt_restore_handle *crh;
742 struct hsm_action_item *hai;
743 struct mdt_thread_info *mti;
744 struct coordinator *cdt;
745 struct mdt_object *child;
751 cdt = &mti->mti_mdt->mdt_coordinator;
753 larr = (struct llog_agent_req_rec *)hdr;
754 hai = &larr->arr_hai;
755 if (hai->hai_cookie > cdt->cdt_last_cookie)
756 /* update the cookie to avoid collision */
757 cdt->cdt_last_cookie = hai->hai_cookie + 1;
759 if (hai->hai_action != HSMA_RESTORE ||
760 agent_req_in_final_state(larr->arr_status))
763 /* restore request not in a final state */
765 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
769 crh->crh_fid = hai->hai_fid;
770 /* in V1 all file is restored
771 crh->extent.start = hai->hai_extent.offset;
772 crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
774 crh->crh_extent.start = 0;
775 crh->crh_extent.end = hai->hai_extent.length;
776 /* get the layout lock */
777 mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
778 child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
779 MDS_INODELOCK_LAYOUT);
781 GOTO(out, rc = PTR_ERR(child));
784 /* we choose to not keep a reference
785 * on the object during the restore time which can be very long */
786 mdt_object_put(mti->mti_env, child);
788 mutex_lock(&cdt->cdt_restore_lock);
789 list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
790 mutex_unlock(&cdt->cdt_restore_lock);
797 * restore coordinator state at startup
798 * the goal is to take a layout lock for each registered restore request
799 * \param mti [IN] context
801 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
803 struct hsm_restore_data hrd;
809 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
810 hsm_restore_cb, &hrd);
815 static int hsm_init_ucred(struct lu_ucred *uc)
819 uc->uc_valid = UCRED_OLD;
828 uc->uc_suppgids[0] = -1;
829 uc->uc_suppgids[1] = -1;
830 uc->uc_cap = CFS_CAP_FS_MASK;
833 uc->uc_identity = NULL;
839 * wake up coordinator thread
840 * \param mdt [IN] device
842 * \retval -ve failure
844 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
846 struct coordinator *cdt = &mdt->mdt_coordinator;
849 if (cdt->cdt_state == CDT_STOPPED)
852 /* wake up coordinator */
853 cdt->cdt_thread.t_flags = SVC_EVENT;
854 wake_up(&cdt->cdt_thread.t_ctl_waitq);
860 * initialize coordinator struct
861 * \param mdt [IN] device
863 * \retval -ve failure
865 int mdt_hsm_cdt_init(struct mdt_device *mdt)
867 struct coordinator *cdt = &mdt->mdt_coordinator;
868 struct mdt_thread_info *cdt_mti = NULL;
872 cdt->cdt_state = CDT_STOPPED;
874 init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
875 mutex_init(&cdt->cdt_llog_lock);
876 init_rwsem(&cdt->cdt_agent_lock);
877 init_rwsem(&cdt->cdt_request_lock);
878 mutex_init(&cdt->cdt_restore_lock);
880 INIT_LIST_HEAD(&cdt->cdt_requests);
881 INIT_LIST_HEAD(&cdt->cdt_agents);
882 INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
884 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
888 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
889 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
891 lu_context_enter(&cdt->cdt_session);
892 cdt->cdt_env.le_ses = &cdt->cdt_session;
894 lu_env_fini(&cdt->cdt_env);
898 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
899 LASSERT(cdt_mti != NULL);
901 cdt_mti->mti_env = &cdt->cdt_env;
902 cdt_mti->mti_mdt = mdt;
904 hsm_init_ucred(mdt_ucred(cdt_mti));
906 /* default values for /proc tunnables
907 * can be override by MGS conf */
908 cdt->cdt_default_archive_id = 1;
909 cdt->cdt_grace_delay = 60;
910 cdt->cdt_loop_period = 10;
911 cdt->cdt_max_requests = 3;
912 cdt->cdt_policy = CDT_DEFAULT_POLICY;
913 cdt->cdt_active_req_timeout = 3600;
919 * free a coordinator thread
920 * \param mdt [IN] device
922 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
924 struct coordinator *cdt = &mdt->mdt_coordinator;
927 lu_context_exit(cdt->cdt_env.le_ses);
928 lu_context_fini(cdt->cdt_env.le_ses);
930 lu_env_fini(&cdt->cdt_env);
936 * start a coordinator thread
937 * \param mdt [IN] device
939 * \retval -ve failure
941 int mdt_hsm_cdt_start(struct mdt_device *mdt)
943 struct coordinator *cdt = &mdt->mdt_coordinator;
946 struct mdt_thread_info *cdt_mti;
947 struct task_struct *task;
950 /* functions defined but not yet used
951 * this avoid compilation warning
955 if (cdt->cdt_state != CDT_STOPPED) {
956 CERROR("%s: Coordinator already started\n",
961 CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
962 cdt->cdt_policy = CDT_DEFAULT_POLICY;
964 cdt->cdt_state = CDT_INIT;
966 atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
967 /* just need to be larger than previous one */
968 /* cdt_last_cookie is protected by cdt_llog_lock */
969 cdt->cdt_last_cookie = cfs_time_current_sec();
970 atomic_set(&cdt->cdt_request_count, 0);
971 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
972 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
973 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
975 /* to avoid deadlock when start is made through /proc
976 * /proc entries are created by the coordinator thread */
978 /* set up list of started restore requests */
979 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
980 rc = mdt_hsm_pending_restore(cdt_mti);
982 CERROR("%s: cannot take the layout locks needed"
983 " for registered restore: %d\n",
984 mdt_obd_name(mdt), rc);
986 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
989 cdt->cdt_state = CDT_STOPPED;
990 CERROR("%s: error starting coordinator thread: %d\n",
991 mdt_obd_name(mdt), rc);
994 CDEBUG(D_HSM, "%s: coordinator thread started\n",
999 wait_event(cdt->cdt_thread.t_ctl_waitq,
1000 (cdt->cdt_thread.t_flags & SVC_RUNNING));
1002 cdt->cdt_state = CDT_RUNNING;
1003 mdt->mdt_opts.mo_coordinator = 1;
1008 * stop a coordinator thread
1009 * \param mdt [IN] device
1011 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1013 struct coordinator *cdt = &mdt->mdt_coordinator;
1014 struct cdt_agent_req *car, *tmp1;
1015 struct hsm_agent *ha, *tmp2;
1016 struct cdt_restore_handle *crh, *tmp3;
1017 struct mdt_thread_info *cdt_mti;
1020 if (cdt->cdt_state == CDT_STOPPED) {
1021 CERROR("%s: Coordinator already stopped\n",
1026 if (cdt->cdt_state != CDT_STOPPING) {
1027 /* stop coordinator thread before cleaning */
1028 cdt->cdt_thread.t_flags = SVC_STOPPING;
1029 wake_up(&cdt->cdt_thread.t_ctl_waitq);
1030 wait_event(cdt->cdt_thread.t_ctl_waitq,
1031 cdt->cdt_thread.t_flags & SVC_STOPPED);
1033 cdt->cdt_state = CDT_STOPPED;
1035 /* start cleaning */
1036 down_write(&cdt->cdt_request_lock);
1037 list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
1039 list_del(&car->car_request_list);
1040 mdt_cdt_free_request(car);
1042 up_write(&cdt->cdt_request_lock);
1044 down_write(&cdt->cdt_agent_lock);
1045 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
1046 list_del(&ha->ha_list);
1049 up_write(&cdt->cdt_agent_lock);
1051 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1052 mutex_lock(&cdt->cdt_restore_lock);
1053 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
1054 struct mdt_object *child;
1056 /* give back layout lock */
1057 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
1059 mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
1061 list_del(&crh->crh_list);
1063 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1065 mutex_unlock(&cdt->cdt_restore_lock);
1067 mdt->mdt_opts.mo_coordinator = 0;
1073 * register all requests from an hal in the memory list
1074 * \param mti [IN] context
1075 * \param hal [IN] request
1076 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1077 * which is running the CT
1079 * \retval -ve failure
1081 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1082 struct hsm_action_list *hal, struct obd_uuid *uuid)
1084 struct mdt_device *mdt = mti->mti_mdt;
1085 struct coordinator *cdt = &mdt->mdt_coordinator;
1086 struct hsm_action_item *hai;
1090 /* register request in memory list */
1091 hai = hai_first(hal);
1092 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1093 struct cdt_agent_req *car;
1095 /* in case of a cancel request, we first mark the ondisk
1096 * record of the request we want to stop as canceled
1097 * this does not change the cancel record
1098 * it will be done when updating the request status
1100 if (hai->hai_action == HSMA_CANCEL) {
1101 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1105 CERROR("%s: mdt_agent_record_update() failed, "
1106 "rc=%d, cannot update status to %s "
1107 "for cookie "LPX64"\n",
1108 mdt_obd_name(mdt), rc,
1109 agent_req_status2name(ARS_CANCELED),
1114 /* find the running request to set it canceled */
1115 car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1117 car->car_canceled = 1;
1118 /* uuid has to be changed to the one running the
1119 * request to cancel */
1120 *uuid = car->car_uuid;
1121 mdt_cdt_put_request(car);
1123 /* no need to memorize cancel request
1124 * this also avoid a deadlock when we receive
1125 * a purge all requests command
1130 if (hai->hai_action == HSMA_ARCHIVE) {
1131 struct mdt_object *obj;
1134 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1135 if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1138 GOTO(out, rc = PTR_ERR(obj));
1140 hsm.mh_flags |= HS_EXISTS;
1141 hsm.mh_arch_id = hal->hal_archive_id;
1142 rc = mdt_hsm_attr_set(mti, obj, &hsm);
1143 mdt_object_put(mti->mti_env, obj);
1148 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1149 hal->hal_archive_id, hal->hal_flags,
1152 GOTO(out, rc = PTR_ERR(car));
1154 rc = mdt_cdt_add_request(cdt, car);
1156 mdt_cdt_free_request(car);
1163 * swap layouts between 2 fids
1164 * \param mti [IN] context
1167 * \param mh_common [IN] MD HSM
1169 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1170 const lustre_fid *fid, const lustre_fid *dfid,
1171 struct md_hsm *mh_common)
1173 struct mdt_device *mdt = mti->mti_mdt;
1174 struct mdt_object *child1, *child2;
1175 struct mdt_lock_handle *lh2;
1179 child1 = mdt_object_find(mti->mti_env, mdt, fid);
1181 GOTO(out, rc = PTR_ERR(child1));
1183 /* we already have layout lock on FID so take only
1185 lh2 = &mti->mti_lh[MDT_LH_OLD];
1186 mdt_lock_reg_init(lh2, LCK_EX);
1187 child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1189 GOTO(out_child1, rc = PTR_ERR(child2));
1191 /* if copy tool closes the volatile before sending the final
1192 * progress through llapi_hsm_copy_end(), all the objects
1193 * are removed and mdd_swap_layout LBUG */
1194 if (!mdt_object_exists(child2)) {
1195 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1196 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1197 GOTO(out_child2, rc = -ENOENT);
1199 /* Since we only handle restores here, unconditionally use
1200 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1201 * be preserved in case of failure during swap_layout and not
1202 * leave a file in an intermediate but incoherent state.
1203 * But need to setup HSM xattr of data FID before, reuse
1204 * mti and mh presets for FID in hsm_cdt_request_completed(),
1205 * only need to clear RELEASED and DIRTY.
1207 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1208 rc = mdt_hsm_attr_set(mti, child2, mh_common);
1210 rc = mo_swap_layouts(mti->mti_env,
1211 mdt_object_child(child1),
1212 mdt_object_child(child2),
1213 SWAP_LAYOUTS_MDS_HSM);
1216 mdt_object_unlock_put(mti, child2, lh2, 1);
1218 mdt_object_put(mti->mti_env, child1);
1224 * update status of a completed request
1225 * \param mti [IN] context
1226 * \param pgs [IN] progress of the copy tool
1227 * \param update_record [IN] update llog record
1229 * \retval -ve failure
1231 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1232 struct hsm_progress_kernel *pgs,
1233 const struct cdt_agent_req *car,
1234 enum agent_req_status *status)
1236 const struct lu_env *env = mti->mti_env;
1237 struct mdt_device *mdt = mti->mti_mdt;
1238 struct coordinator *cdt = &mdt->mdt_coordinator;
1239 struct mdt_object *obj = NULL;
1240 int cl_flags = 0, rc = 0;
1245 /* default is to retry */
1246 *status = ARS_WAITING;
1248 /* find object by FID */
1249 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1250 /* we will update MD HSM only if needed */
1251 is_mh_changed = false;
1253 /* object removed */
1254 *status = ARS_SUCCEED;
1258 /* no need to change mh->mh_arch_id
1259 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1261 if (pgs->hpk_errval != 0) {
1262 switch (pgs->hpk_errval) {
1264 /* the copy tool does not support cancel
1265 * so the cancel request is failed
1266 * As we cannot distinguish a cancel progress
1267 * from another action progress (they have the
1268 * same cookie), we suppose here the CT returns
1269 * ENOSYS only if does not support cancel
1271 /* this can also happen when cdt calls it to
1272 * for a timeouted request */
1273 *status = ARS_FAILED;
1274 /* to have a cancel event in changelog */
1275 pgs->hpk_errval = ECANCELED;
1278 /* the request record has already been set to
1279 * ARS_CANCELED, this set the cancel request
1281 *status = ARS_SUCCEED;
1284 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1285 !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1286 ARS_FAILED : ARS_WAITING);
1290 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1291 CERROR("%s: Request "LPX64" on "DFID
1292 " failed, error code %d too large\n",
1294 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1296 hsm_set_cl_error(&cl_flags,
1297 CLF_HSM_ERROVERFLOW);
1300 hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1303 switch (car->car_hai->hai_action) {
1305 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1308 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1311 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1314 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1315 CERROR("%s: Failed request "LPX64" on "DFID
1316 " cannot be a CANCEL\n",
1319 PFID(&pgs->hpk_fid));
1322 CERROR("%s: Failed request "LPX64" on "DFID
1323 " %d is an unknown action\n",
1325 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1326 car->car_hai->hai_action);
1331 *status = ARS_SUCCEED;
1332 switch (car->car_hai->hai_action) {
1334 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1335 /* set ARCHIVE keep EXIST and clear LOST and
1337 mh.mh_arch_ver = pgs->hpk_data_version;
1338 mh.mh_flags |= HS_ARCHIVED;
1339 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1340 is_mh_changed = true;
1343 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1345 /* do not clear RELEASED and DIRTY here
1346 * this will occur in hsm_swap_layouts()
1349 /* Restoring has changed the file version on
1351 mh.mh_arch_ver = pgs->hpk_data_version;
1352 is_mh_changed = true;
1355 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1356 /* clear ARCHIVED EXISTS and LOST */
1357 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1358 is_mh_changed = true;
1361 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1362 CERROR("%s: Successful request "LPX64
1364 " cannot be a CANCEL\n",
1367 PFID(&pgs->hpk_fid));
1370 CERROR("%s: Successful request "LPX64
1372 " %d is an unknown action\n",
1374 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1375 car->car_hai->hai_action);
1381 /* rc != 0 means error when analysing action, it may come from
1382 * a crasy CT no need to manage DIRTY
1385 hsm_set_cl_flags(&cl_flags,
1386 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1388 /* unlock is done later, after layout lock management */
1390 rc = mdt_hsm_attr_set(mti, obj, &mh);
1393 /* we give back layout lock only if restore was successful or
1394 * if restore was canceled or if policy is to not retry
1395 * in other cases we just unlock the object */
1396 if (car->car_hai->hai_action == HSMA_RESTORE &&
1397 (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1398 cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1399 struct cdt_restore_handle *crh;
1401 /* restore in data FID done, we swap the layouts
1402 * only if restore is successfull */
1403 if (pgs->hpk_errval == 0) {
1404 rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1405 &car->car_hai->hai_dfid, &mh);
1407 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1408 *status = ARS_FAILED;
1409 pgs->hpk_errval = -rc;
1412 /* we have to retry, so keep layout lock */
1413 if (*status == ARS_WAITING)
1416 /* give back layout lock */
1417 mutex_lock(&cdt->cdt_restore_lock);
1418 crh = hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1420 list_del(&crh->crh_list);
1421 mutex_unlock(&cdt->cdt_restore_lock);
1422 /* just give back layout lock, we keep
1423 * the reference which is given back
1424 * later with the lock for HSM flags */
1425 if (!IS_ERR(obj) && crh != NULL)
1426 mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1429 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1435 if (obj != NULL && !IS_ERR(obj)) {
1436 mo_changelog(env, CL_HSM, cl_flags,
1437 mdt_object_child(obj));
1438 mdt_object_put(mti->mti_env, obj);
1445 * update status of a request
1446 * \param mti [IN] context
1447 * \param pgs [IN] progress of the copy tool
1448 * \param update_record [IN] update llog record
1450 * \retval -ve failure
1452 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1453 struct hsm_progress_kernel *pgs,
1454 const int update_record)
1456 struct mdt_device *mdt = mti->mti_mdt;
1457 struct coordinator *cdt = &mdt->mdt_coordinator;
1458 struct cdt_agent_req *car;
1462 /* no coordinator started, so we cannot serve requests */
1463 if (cdt->cdt_state == CDT_STOPPED)
1466 /* first do sanity checks */
1467 car = mdt_cdt_update_request(cdt, pgs);
1469 CERROR("%s: Cannot find running request for cookie "LPX64
1472 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1475 RETURN(PTR_ERR(car));
1478 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1479 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1480 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1481 hsm_copytool_action2name(car->car_hai->hai_action),
1482 pgs->hpk_flags, pgs->hpk_errval,
1483 PFID(&car->car_hai->hai_fid),
1484 PFID(&car->car_hai->hai_dfid));
1486 /* progress is done on FID or data FID depending of the action and
1487 * of the copy progress */
1488 /* for restore progress is used to send back the data FID to cdt */
1489 if (car->car_hai->hai_action == HSMA_RESTORE &&
1490 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1491 car->car_hai->hai_dfid = pgs->hpk_fid;
1493 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1494 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1495 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1496 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1497 CERROR("%s: Progress on "DFID" for cookie "LPX64
1498 " does not match request FID "DFID" nor data FID "
1501 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1502 PFID(&car->car_hai->hai_fid),
1503 PFID(&car->car_hai->hai_dfid));
1504 GOTO(out, rc = -EINVAL);
1507 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1508 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1509 " is not coherent (err=%d and not completed"
1512 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1513 hsm_copytool_action2name(car->car_hai->hai_action),
1514 pgs->hpk_errval, pgs->hpk_flags);
1515 GOTO(out, rc = -EINVAL);
1518 /* now progress is valid */
1520 /* we use a root like ucred */
1521 hsm_init_ucred(mdt_ucred(mti));
1523 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1524 enum agent_req_status status;
1526 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1528 /* remove request from memory list */
1529 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1531 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1532 " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1534 hsm_copytool_action2name(car->car_hai->hai_action),
1535 agent_req_status2name(status));
1537 if (update_record) {
1540 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1541 &pgs->hpk_cookie, 1,
1544 CERROR("%s: mdt_agent_record_update() failed,"
1545 " rc=%d, cannot update status to %s"
1546 " for cookie "LPX64"\n",
1547 mdt_obd_name(mdt), rc1,
1548 agent_req_status2name(status),
1550 rc = (rc != 0 ? rc : rc1);
1552 /* ct has completed a request, so a slot is available, wakeup
1553 * cdt to find new work */
1554 mdt_hsm_cdt_wakeup(mdt);
1556 /* if copytool send a progress on a canceled request
1557 * we inform copytool it should stop
1559 if (car->car_canceled == 1)
1565 /* remove ref got from mdt_cdt_update_request() */
1566 mdt_cdt_put_request(car);
1573 * data passed to llog_cat_process() callback
1574 * to cancel requests
1576 struct hsm_cancel_all_data {
1577 struct mdt_device *mdt;
1581 * llog_cat_process() callback, used to:
1582 * - purge all requests
1583 * \param env [IN] environment
1584 * \param llh [IN] llog handle
1585 * \param hdr [IN] llog record
1586 * \param data [IN] cb data = struct hsm_cancel_all_data
1588 * \retval -ve failure
1590 static int mdt_cancel_all_cb(const struct lu_env *env,
1591 struct llog_handle *llh,
1592 struct llog_rec_hdr *hdr, void *data)
1594 struct llog_agent_req_rec *larr;
1595 struct hsm_cancel_all_data *hcad;
1599 larr = (struct llog_agent_req_rec *)hdr;
1601 if (larr->arr_status == ARS_WAITING ||
1602 larr->arr_status == ARS_STARTED) {
1603 larr->arr_status = ARS_CANCELED;
1604 larr->arr_req_change = cfs_time_current_sec();
1605 rc = mdt_agent_llog_update_rec(env, hcad->mdt, llh, larr);
1607 RETURN(LLOG_DEL_RECORD);
1613 * cancel all actions
1614 * \param obd [IN] MDT device
1616 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1618 struct mdt_thread_info *mti;
1619 struct coordinator *cdt = &mdt->mdt_coordinator;
1620 struct cdt_agent_req *car;
1621 struct hsm_action_list *hal = NULL;
1622 struct hsm_action_item *hai;
1623 struct hsm_cancel_all_data hcad;
1624 int hal_sz = 0, hal_len, rc;
1625 enum cdt_states save_state;
1628 /* retrieve coordinator context */
1629 mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1631 /* disable coordinator */
1632 save_state = cdt->cdt_state;
1633 cdt->cdt_state = CDT_DISABLE;
1635 /* send cancel to all running requests */
1636 down_read(&cdt->cdt_request_lock);
1637 list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1638 mdt_cdt_get_request(car);
1639 /* request is not yet removed from list, it will be done
1640 * when copytool will return progress
1643 if (car->car_hai->hai_action == HSMA_CANCEL) {
1644 mdt_cdt_put_request(car);
1649 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1650 cfs_size_round(car->car_hai->hai_len);
1652 if (hal_len > hal_sz && hal_sz > 0) {
1653 /* not enough room, free old buffer */
1654 OBD_FREE(hal, hal_sz);
1658 /* empty buffer, allocate one */
1661 OBD_ALLOC(hal, hal_sz);
1663 mdt_cdt_put_request(car);
1664 up_read(&cdt->cdt_request_lock);
1665 GOTO(out, rc = -ENOMEM);
1669 hal->hal_version = HAL_VERSION;
1670 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1672 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1673 hal->hal_compound_id = car->car_compound_id;
1674 hal->hal_archive_id = car->car_archive_id;
1675 hal->hal_flags = car->car_flags;
1678 hai = hai_first(hal);
1679 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1680 hai->hai_action = HSMA_CANCEL;
1683 /* it is possible to safely call mdt_hsm_agent_send()
1684 * (ie without a deadlock on cdt_request_lock), because the
1685 * write lock is taken only if we are not in purge mode
1686 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1687 * nor mdt_cdt_remove_request())
1689 /* no conflict with cdt thread because cdt is disable and we
1690 * have the request lock */
1691 mdt_hsm_agent_send(mti, hal, 1);
1693 mdt_cdt_put_request(car);
1695 up_read(&cdt->cdt_request_lock);
1698 OBD_FREE(hal, hal_sz);
1700 /* cancel all on-disk records */
1703 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1704 mdt_cancel_all_cb, &hcad);
1706 /* enable coordinator */
1707 cdt->cdt_state = save_state;
1713 * check if a request is comptaible with file status
1714 * \param hai [IN] request description
1715 * \param hal_an [IN] request archive number (not used)
1716 * \param rq_flags [IN] request flags
1717 * \param hsm [IN] file HSM metadata
1720 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1721 const int hal_an, const __u64 rq_flags,
1722 const struct md_hsm *hsm)
1724 int is_compat = false;
1728 hsm_flags = hsm->mh_flags;
1729 switch (hai->hai_action) {
1731 if (!(hsm_flags & HS_NOARCHIVE) &&
1732 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1736 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1737 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1741 if (!(hsm_flags & HS_RELEASED) &&
1742 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1749 CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1750 " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1751 PFID(&hai->hai_fid),
1752 hsm_copytool_action2name(hai->hai_action), rq_flags,
1753 hai->hai_extent.offset, hai->hai_extent.length,
1755 (is_compat ? "compatible" : "uncompatible"));
1761 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1763 static const struct {
1767 } hsm_policy_names[] = {
1768 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1769 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1774 * convert a policy name to a bit
1775 * \param name [IN] policy name
1777 * \retval policy bit
1779 static __u64 hsm_policy_str2bit(const char *name)
1783 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1784 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1785 strcmp(hsm_policy_names[i].name, name) == 0)
1786 return hsm_policy_names[i].bit;
1791 * convert a policy bit field to a string
1792 * \param mask [IN] policy bit field
1793 * \param hexa [IN] print mask before bit names
1794 * \param buffer [OUT] string
1795 * \param count [IN] size of buffer
1797 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1805 seq_printf(m, "("LPX64") ", mask);
1807 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1810 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1811 if (hsm_policy_names[j].bit == bit)
1815 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1817 seq_printf(m, "%s ", hsm_policy_names[j].name);
1819 /* remove last ' ' */
1824 /* methods to read/write HSM policy flags */
1825 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1827 struct mdt_device *mdt = m->private;
1828 struct coordinator *cdt = &mdt->mdt_coordinator;
1831 hsm_policy_bit2str(m, cdt->cdt_policy, false);
1836 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1837 size_t count, loff_t *off)
1839 struct seq_file *m = file->private_data;
1840 struct mdt_device *mdt = m->private;
1841 struct coordinator *cdt = &mdt->mdt_coordinator;
1842 char *start, *token, sign;
1845 __u64 add_mask, remove_mask, set_mask;
1849 if (count + 1 > PAGE_SIZE)
1852 OBD_ALLOC(buf, count + 1);
1856 if (copy_from_user(buf, buffer, count))
1857 GOTO(out, rc = -EFAULT);
1862 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1865 add_mask = remove_mask = set_mask = 0;
1867 token = strsep(&start, "\n ");
1873 if (sign == '-' || sign == '+')
1876 policy = hsm_policy_str2bit(token);
1878 CWARN("%s: '%s' is unknown, "
1879 "supported policies are:\n", mdt_obd_name(mdt),
1881 hsm_policy_bit2str(m, 0, false);
1882 GOTO(out, rc = -EINVAL);
1886 remove_mask |= policy;
1896 } while (start != NULL);
1898 CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1899 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1901 /* if no sign in all string, it is a clear and set
1902 * if some sign found, all unsigned are converted
1904 * P1 P2 = set to P1 and P2
1905 * P1 -P2 = add P1 clear P2 same as +P1 -P2
1907 if (remove_mask == 0 && add_mask == 0) {
1908 cdt->cdt_policy = set_mask;
1910 cdt->cdt_policy |= set_mask | add_mask;
1911 cdt->cdt_policy &= ~remove_mask;
1914 GOTO(out, rc = count);
1917 OBD_FREE(buf, count + 1);
1920 LPROC_SEQ_FOPS(mdt_hsm_policy);
1922 #define GENERATE_PROC_METHOD(VAR) \
1923 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
1925 struct mdt_device *mdt = m->private; \
1926 struct coordinator *cdt = &mdt->mdt_coordinator; \
1929 seq_printf(m, LPU64"\n", (__u64)cdt->VAR); \
1933 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1934 size_t count, loff_t *off) \
1937 struct seq_file *m = file->private_data; \
1938 struct mdt_device *mdt = m->private; \
1939 struct coordinator *cdt = &mdt->mdt_coordinator; \
1944 rc = lprocfs_write_helper(buffer, count, &val); \
1954 GENERATE_PROC_METHOD(cdt_loop_period)
1955 GENERATE_PROC_METHOD(cdt_grace_delay)
1956 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1957 GENERATE_PROC_METHOD(cdt_max_requests)
1958 GENERATE_PROC_METHOD(cdt_default_archive_id)
1961 * procfs write method for MDT/hsm_control
1962 * proc entry is in mdt directory so data is mdt obd_device pointer
1964 #define CDT_ENABLE_CMD "enabled"
1965 #define CDT_STOP_CMD "shutdown"
1966 #define CDT_DISABLE_CMD "disabled"
1967 #define CDT_PURGE_CMD "purge"
1968 #define CDT_HELP_CMD "help"
1969 #define CDT_MAX_CMD_LEN 10
1972 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1973 size_t count, loff_t *off)
1975 struct seq_file *m = file->private_data;
1976 struct obd_device *obd = m->private;
1977 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1978 struct coordinator *cdt = &(mdt->mdt_coordinator);
1980 char kernbuf[CDT_MAX_CMD_LEN];
1983 if (count == 0 || count >= sizeof(kernbuf))
1986 if (copy_from_user(kernbuf, buffer, count))
1990 if (kernbuf[count - 1] == '\n')
1991 kernbuf[count - 1] = 0;
1994 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1995 if (cdt->cdt_state == CDT_DISABLE) {
1996 cdt->cdt_state = CDT_RUNNING;
1997 mdt_hsm_cdt_wakeup(mdt);
1999 rc = mdt_hsm_cdt_start(mdt);
2001 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2002 if ((cdt->cdt_state == CDT_STOPPING) ||
2003 (cdt->cdt_state == CDT_STOPPED)) {
2004 CERROR("%s: Coordinator already stopped\n",
2008 cdt->cdt_state = CDT_STOPPING;
2010 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2011 if ((cdt->cdt_state == CDT_STOPPING) ||
2012 (cdt->cdt_state == CDT_STOPPED)) {
2013 CERROR("%s: Coordinator is stopped\n",
2017 cdt->cdt_state = CDT_DISABLE;
2019 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2020 rc = hsm_cancel_all_actions(mdt);
2021 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2029 CERROR("%s: Valid coordinator control commands are: "
2030 "%s %s %s %s %s\n", mdt_obd_name(mdt),
2031 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2032 CDT_PURGE_CMD, CDT_HELP_CMD);
2040 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2042 struct obd_device *obd = m->private;
2043 struct coordinator *cdt;
2046 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2048 if (cdt->cdt_state == CDT_INIT)
2049 seq_printf(m, "init\n");
2050 else if (cdt->cdt_state == CDT_RUNNING)
2051 seq_printf(m, "enabled\n");
2052 else if (cdt->cdt_state == CDT_STOPPING)
2053 seq_printf(m, "stopping\n");
2054 else if (cdt->cdt_state == CDT_STOPPED)
2055 seq_printf(m, "stopped\n");
2056 else if (cdt->cdt_state == CDT_DISABLE)
2057 seq_printf(m, "disabled\n");
2059 seq_printf(m, "unknown\n");
2065 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2070 for (i = 0; i < 8 * sizeof(mask); i++) {
2071 if (mask & (1UL << i))
2072 rc += seq_printf(m, "%s%s", rc == 0 ? "" : " ",
2073 hsm_copytool_action2name(i));
2075 rc += seq_printf(m, "\n");
2081 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2083 struct mdt_device *mdt = m->private;
2084 struct coordinator *cdt = &mdt->mdt_coordinator;
2086 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2090 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2092 struct mdt_device *mdt = m->private;
2093 struct coordinator *cdt = &mdt->mdt_coordinator;
2095 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2099 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2101 struct mdt_device *mdt = m->private;
2102 struct coordinator *cdt = &mdt->mdt_coordinator;
2104 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2107 static inline enum hsm_copytool_action
2108 hsm_copytool_name2action(const char *name)
2110 if (strcasecmp(name, "NOOP") == 0)
2112 else if (strcasecmp(name, "ARCHIVE") == 0)
2113 return HSMA_ARCHIVE;
2114 else if (strcasecmp(name, "RESTORE") == 0)
2115 return HSMA_RESTORE;
2116 else if (strcasecmp(name, "REMOVE") == 0)
2118 else if (strcasecmp(name, "CANCEL") == 0)
2125 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2126 size_t user_count, __u64 *mask)
2128 char *buf, *pos, *name;
2134 if (!(user_count < 4096))
2137 buf_size = user_count + 1;
2139 OBD_ALLOC(buf, buf_size);
2143 if (copy_from_user(buf, user_buf, buf_size - 1))
2144 GOTO(out, rc = -EFAULT);
2146 buf[buf_size - 1] = '\0';
2149 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2155 action = hsm_copytool_name2action(name);
2157 GOTO(out, rc = -EINVAL);
2159 new_mask |= (1UL << action);
2165 OBD_FREE(buf, buf_size);
2171 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2172 size_t count, loff_t *off)
2174 struct seq_file *m = file->private_data;
2175 struct mdt_device *mdt = m->private;
2176 struct coordinator *cdt = &mdt->mdt_coordinator;
2178 return mdt_write_hsm_request_mask(file, buf, count,
2179 &cdt->cdt_user_request_mask);
2183 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2184 size_t count, loff_t *off)
2186 struct seq_file *m = file->private_data;
2187 struct mdt_device *mdt = m->private;
2188 struct coordinator *cdt = &mdt->mdt_coordinator;
2190 return mdt_write_hsm_request_mask(file, buf, count,
2191 &cdt->cdt_group_request_mask);
2195 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2196 size_t count, loff_t *off)
2198 struct seq_file *m = file->private_data;
2199 struct mdt_device *mdt = m->private;
2200 struct coordinator *cdt = &mdt->mdt_coordinator;
2202 return mdt_write_hsm_request_mask(file, buf, count,
2203 &cdt->cdt_other_request_mask);
2206 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2207 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2208 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2209 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2210 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2211 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2212 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2213 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2215 static struct lprocfs_seq_vars lprocfs_mdt_hsm_vars[] = {
2217 .fops = &mdt_hsm_agent_fops },
2218 { .name = "actions",
2219 .fops = &mdt_hsm_actions_fops,
2220 .proc_mode = 0444 },
2221 { .name = "default_archive_id",
2222 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2223 { .name = "grace_delay",
2224 .fops = &mdt_hsm_cdt_grace_delay_fops },
2225 { .name = "loop_period",
2226 .fops = &mdt_hsm_cdt_loop_period_fops },
2227 { .name = "max_requests",
2228 .fops = &mdt_hsm_cdt_max_requests_fops },
2230 .fops = &mdt_hsm_policy_fops },
2231 { .name = "active_request_timeout",
2232 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2233 { .name = "active_requests",
2234 .fops = &mdt_hsm_active_requests_fops },
2235 { .name = "user_request_mask",
2236 .fops = &mdt_hsm_user_request_mask_fops, },
2237 { .name = "group_request_mask",
2238 .fops = &mdt_hsm_group_request_mask_fops, },
2239 { .name = "other_request_mask",
2240 .fops = &mdt_hsm_other_request_mask_fops, },