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 strncpy(hal->hal_fsname, hsd->fs_name,
232 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
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 timeouted, 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 timeouted 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));
340 /* add the cookie to the list of record to be
341 * canceled by caller */
342 if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
343 __u64 *ptr, *old_ptr;
344 int old_sz, new_sz, new_cnt;
346 /* need to increase vector size */
347 old_sz = sizeof(__u64) * hsd->max_cookie;
348 old_ptr = hsd->cookies;
350 new_cnt = 2 * hsd->max_cookie;
351 new_sz = sizeof(__u64) * new_cnt;
353 OBD_ALLOC(ptr, new_sz);
355 CERROR("%s: Cannot allocate memory "
356 "(%d o) for cookie vector\n",
357 mdt_obd_name(mdt), new_sz);
360 memcpy(ptr, hsd->cookies, old_sz);
362 hsd->max_cookie = new_cnt;
363 OBD_FREE(old_ptr, old_sz);
365 hsd->cookies[hsd->cookie_cnt] =
366 larr->arr_hai.hai_cookie;
374 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
375 cfs_time_current_sec())
376 RETURN(LLOG_DEL_RECORD);
383 * create /proc entries for coordinator
386 * \retval -ve failure
388 int hsm_cdt_procfs_init(struct mdt_device *mdt)
390 struct coordinator *cdt = &mdt->mdt_coordinator;
394 /* init /proc entries, failure is not critical */
395 cdt->cdt_proc_dir = lprocfs_seq_register("hsm",
396 mdt2obd_dev(mdt)->obd_proc_entry,
397 lprocfs_mdt_hsm_vars, mdt);
398 if (IS_ERR(cdt->cdt_proc_dir)) {
399 rc = PTR_ERR(cdt->cdt_proc_dir);
400 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
401 " rc=%d\n", mdt_obd_name(mdt), rc);
402 cdt->cdt_proc_dir = NULL;
410 * remove /proc entries for coordinator
413 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
415 struct coordinator *cdt = &mdt->mdt_coordinator;
417 LASSERT(cdt->cdt_state == CDT_STOPPED);
418 if (cdt->cdt_proc_dir != NULL)
419 lprocfs_remove(&cdt->cdt_proc_dir);
423 * get vector of hsm cdt /proc vars
427 struct lprocfs_seq_vars *hsm_cdt_get_proc_vars(void)
429 return lprocfs_mdt_hsm_vars;
434 * \param data [IN] obd device
436 * \retval -ve failure
438 static int mdt_coordinator(void *data)
440 struct mdt_thread_info *mti = data;
441 struct mdt_device *mdt = mti->mti_mdt;
442 struct coordinator *cdt = &mdt->mdt_coordinator;
443 struct hsm_scan_data hsd = { 0 };
447 cdt->cdt_thread.t_flags = SVC_RUNNING;
448 wake_up(&cdt->cdt_thread.t_ctl_waitq);
450 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
451 mdt_obd_name(mdt), current_pid());
453 /* timeouted cookie vector initialization */
457 /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
458 * increases due to a change from /proc we do not overflow the
459 * hsd.request[] vector
461 hsd.max_requests = cdt->cdt_max_requests;
462 hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
463 OBD_ALLOC(hsd.request, hsd.request_sz);
465 GOTO(out, rc = -ENOMEM);
468 obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
471 struct l_wait_info lwi;
474 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
476 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
477 (cdt->cdt_thread.t_flags &
478 (SVC_STOPPING|SVC_EVENT)),
481 CDEBUG(D_HSM, "coordinator resumes\n");
483 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
484 cdt->cdt_state == CDT_STOPPING) {
485 cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
490 /* wake up before timeout, new work arrives */
491 if (cdt->cdt_thread.t_flags & SVC_EVENT)
492 cdt->cdt_thread.t_flags &= ~SVC_EVENT;
494 /* if coordinator is suspended continue to wait */
495 if (cdt->cdt_state == CDT_DISABLE) {
496 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
500 CDEBUG(D_HSM, "coordinator starts reading llog\n");
502 if (hsd.max_requests != cdt->cdt_max_requests) {
503 /* cdt_max_requests has changed,
504 * we need to allocate a new buffer
506 OBD_FREE(hsd.request, hsd.request_sz);
507 hsd.max_requests = cdt->cdt_max_requests;
509 hsd.max_requests * sizeof(*hsd.request);
510 OBD_ALLOC(hsd.request, hsd.request_sz);
517 /* create canceled cookie vector for an arbitrary size
518 * if needed, vector will grow during llog scan
522 OBD_ALLOC(hsd.cookies, hsd.max_cookie * sizeof(__u64));
529 rc = cdt_llog_process(mti->mti_env, mdt,
530 mdt_coordinator_cb, &hsd);
534 CDEBUG(D_HSM, "Found %d requests to send and %d"
535 " requests to cancel\n",
536 hsd.request_cnt, hsd.cookie_cnt);
537 /* first we cancel llog records of the timeouted requests */
538 if (hsd.cookie_cnt > 0) {
539 rc = mdt_agent_record_update(mti->mti_env, mdt,
544 CERROR("%s: mdt_agent_record_update() failed, "
545 "rc=%d, cannot update status to %s "
547 mdt_obd_name(mdt), rc,
548 agent_req_status2name(ARS_CANCELED),
552 if (list_empty(&cdt->cdt_agents)) {
553 CDEBUG(D_HSM, "no agent available, "
554 "coordinator sleeps\n");
558 /* here hsd contains a list of requests to be started */
559 for (i = 0; i < hsd.max_requests; i++) {
560 struct hsm_action_list *hal;
561 struct hsm_action_item *hai;
564 enum agent_req_status status;
566 /* still room for work ? */
567 if (atomic_read(&cdt->cdt_request_count) ==
568 cdt->cdt_max_requests)
571 if (hsd.request[i].hal == NULL)
574 /* found a request, we start it */
575 /* kuc payload allocation so we avoid an additionnal
576 * allocation in mdt_hsm_agent_send()
578 hal = kuc_alloc(hsd.request[i].hal_used_sz,
579 KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
581 CERROR("%s: Cannot allocate memory (%d o) "
582 "for compound "LPX64"\n",
584 hsd.request[i].hal_used_sz,
585 hsd.request[i].hal->hal_compound_id);
588 memcpy(hal, hsd.request[i].hal,
589 hsd.request[i].hal_used_sz);
591 rc = mdt_hsm_agent_send(mti, hal, 0);
592 /* if failure, we suppose it is temporary
593 * if the copy tool failed to do the request
594 * it has to use hsm_progress
596 status = (rc ? ARS_WAITING : ARS_STARTED);
598 /* set up cookie vector to set records status
599 * after copy tools start or failed
601 sz = hsd.request[i].hal->hal_count * sizeof(__u64);
602 OBD_ALLOC(cookies, sz);
603 if (cookies == NULL) {
604 CERROR("%s: Cannot allocate memory (%d o) "
605 "for cookies vector "LPX64"\n",
606 mdt_obd_name(mdt), sz,
607 hsd.request[i].hal->hal_compound_id);
608 kuc_free(hal, hsd.request[i].hal_used_sz);
611 hai = hai_first(hal);
612 for (j = 0; j < hsd.request[i].hal->hal_count; j++) {
613 cookies[j] = hai->hai_cookie;
617 rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
618 hsd.request[i].hal->hal_count,
621 CERROR("%s: mdt_agent_record_update() failed, "
622 "rc=%d, cannot update status to %s "
624 mdt_obd_name(mdt), rc,
625 agent_req_status2name(status),
626 hsd.request[i].hal->hal_count);
628 OBD_FREE(cookies, sz);
629 kuc_free(hal, hsd.request[i].hal_used_sz);
632 /* free cookie vector allocated for/by callback */
634 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
640 /* free hal allocated by callback */
641 for (i = 0; i < hsd.max_requests; i++) {
642 if (hsd.request[i].hal) {
643 OBD_FREE(hsd.request[i].hal,
644 hsd.request[i].hal_sz);
645 hsd.request[i].hal_sz = 0;
646 hsd.request[i].hal = NULL;
650 LASSERT(hsd.request_cnt == 0);
652 /* reset callback data */
653 memset(hsd.request, 0, hsd.request_sz);
658 OBD_FREE(hsd.request, hsd.request_sz);
661 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
663 if (cdt->cdt_state == CDT_STOPPING) {
664 /* request comes from /proc path, so we need to clean cdt
666 mdt_hsm_cdt_stop(mdt);
667 mdt->mdt_opts.mo_coordinator = 0;
669 /* request comes from a thread event, generated
670 * by mdt_stop_coordinator(), we have to ack
671 * and cdt cleaning will be done by event sender
673 cdt->cdt_thread.t_flags = SVC_STOPPED;
674 wake_up(&cdt->cdt_thread.t_ctl_waitq);
678 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
679 mdt_obd_name(mdt), current_pid(), rc);
681 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
683 mdt_obd_name(mdt), current_pid());
689 * lookup a restore handle by FID
690 * caller needs to hold cdt_restore_lock
691 * \param cdt [IN] coordinator
692 * \param fid [IN] FID
693 * \retval cdt_restore_handle found
694 * \retval NULL not found
696 static struct cdt_restore_handle *hsm_restore_hdl_find(struct coordinator *cdt,
697 const struct lu_fid *fid)
699 struct cdt_restore_handle *crh;
702 list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
703 if (lu_fid_eq(&crh->crh_fid, fid))
710 * data passed to llog_cat_process() callback
711 * to scan requests and take actions
713 struct hsm_restore_data {
714 struct mdt_thread_info *hrd_mti;
718 * llog_cat_process() callback, used to:
719 * - find restore request and allocate the restore handle
720 * \param env [IN] environment
721 * \param llh [IN] llog handle
722 * \param hdr [IN] llog record
723 * \param data [IN/OUT] cb data = struct hsm_restore_data
725 * \retval -ve failure
727 static int hsm_restore_cb(const struct lu_env *env,
728 struct llog_handle *llh,
729 struct llog_rec_hdr *hdr, void *data)
731 struct llog_agent_req_rec *larr;
732 struct hsm_restore_data *hrd;
733 struct cdt_restore_handle *crh;
734 struct hsm_action_item *hai;
735 struct mdt_thread_info *mti;
736 struct coordinator *cdt;
737 struct mdt_object *child;
743 cdt = &mti->mti_mdt->mdt_coordinator;
745 larr = (struct llog_agent_req_rec *)hdr;
746 hai = &larr->arr_hai;
747 if (hai->hai_cookie > cdt->cdt_last_cookie)
748 /* update the cookie to avoid collision */
749 cdt->cdt_last_cookie = hai->hai_cookie + 1;
751 if (hai->hai_action != HSMA_RESTORE ||
752 agent_req_in_final_state(larr->arr_status))
755 /* restore request not in a final state */
757 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
761 crh->crh_fid = hai->hai_fid;
762 /* in V1 all file is restored
763 crh->extent.start = hai->hai_extent.offset;
764 crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
766 crh->crh_extent.start = 0;
767 crh->crh_extent.end = hai->hai_extent.length;
768 /* get the layout lock */
769 mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
770 child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
771 MDS_INODELOCK_LAYOUT);
773 GOTO(out, rc = PTR_ERR(child));
776 /* we choose to not keep a reference
777 * on the object during the restore time which can be very long */
778 mdt_object_put(mti->mti_env, child);
780 mutex_lock(&cdt->cdt_restore_lock);
781 list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
782 mutex_unlock(&cdt->cdt_restore_lock);
789 * restore coordinator state at startup
790 * the goal is to take a layout lock for each registered restore request
791 * \param mti [IN] context
793 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
795 struct hsm_restore_data hrd;
801 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
802 hsm_restore_cb, &hrd);
807 static int hsm_init_ucred(struct lu_ucred *uc)
811 uc->uc_valid = UCRED_OLD;
820 uc->uc_suppgids[0] = -1;
821 uc->uc_suppgids[1] = -1;
822 uc->uc_cap = CFS_CAP_FS_MASK;
825 uc->uc_identity = NULL;
831 * wake up coordinator thread
832 * \param mdt [IN] device
834 * \retval -ve failure
836 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
838 struct coordinator *cdt = &mdt->mdt_coordinator;
841 if (cdt->cdt_state == CDT_STOPPED)
844 /* wake up coordinator */
845 cdt->cdt_thread.t_flags = SVC_EVENT;
846 wake_up(&cdt->cdt_thread.t_ctl_waitq);
852 * initialize coordinator struct
853 * \param mdt [IN] device
855 * \retval -ve failure
857 int mdt_hsm_cdt_init(struct mdt_device *mdt)
859 struct coordinator *cdt = &mdt->mdt_coordinator;
860 struct mdt_thread_info *cdt_mti = NULL;
864 cdt->cdt_state = CDT_STOPPED;
866 init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
867 mutex_init(&cdt->cdt_llog_lock);
868 init_rwsem(&cdt->cdt_agent_lock);
869 init_rwsem(&cdt->cdt_request_lock);
870 mutex_init(&cdt->cdt_restore_lock);
872 CFS_INIT_LIST_HEAD(&cdt->cdt_requests);
873 CFS_INIT_LIST_HEAD(&cdt->cdt_agents);
874 CFS_INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
876 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
880 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
881 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
883 lu_context_enter(&cdt->cdt_session);
884 cdt->cdt_env.le_ses = &cdt->cdt_session;
886 lu_env_fini(&cdt->cdt_env);
890 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
891 LASSERT(cdt_mti != NULL);
893 cdt_mti->mti_env = &cdt->cdt_env;
894 cdt_mti->mti_mdt = mdt;
896 hsm_init_ucred(mdt_ucred(cdt_mti));
898 /* default values for /proc tunnables
899 * can be override by MGS conf */
900 cdt->cdt_default_archive_id = 1;
901 cdt->cdt_grace_delay = 60;
902 cdt->cdt_loop_period = 10;
903 cdt->cdt_max_requests = 3;
904 cdt->cdt_policy = CDT_DEFAULT_POLICY;
905 cdt->cdt_active_req_timeout = 3600;
911 * free a coordinator thread
912 * \param mdt [IN] device
914 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
916 struct coordinator *cdt = &mdt->mdt_coordinator;
919 lu_context_exit(cdt->cdt_env.le_ses);
920 lu_context_fini(cdt->cdt_env.le_ses);
922 lu_env_fini(&cdt->cdt_env);
928 * start a coordinator thread
929 * \param mdt [IN] device
931 * \retval -ve failure
933 int mdt_hsm_cdt_start(struct mdt_device *mdt)
935 struct coordinator *cdt = &mdt->mdt_coordinator;
938 struct mdt_thread_info *cdt_mti;
939 struct task_struct *task;
942 /* functions defined but not yet used
943 * this avoid compilation warning
947 if (cdt->cdt_state != CDT_STOPPED) {
948 CERROR("%s: Coordinator already started\n",
953 CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
954 cdt->cdt_policy = CDT_DEFAULT_POLICY;
956 cdt->cdt_state = CDT_INIT;
958 atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
959 /* just need to be larger than previous one */
960 /* cdt_last_cookie is protected by cdt_llog_lock */
961 cdt->cdt_last_cookie = cfs_time_current_sec();
962 atomic_set(&cdt->cdt_request_count, 0);
963 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
964 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
965 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
967 /* to avoid deadlock when start is made through /proc
968 * /proc entries are created by the coordinator thread */
970 /* set up list of started restore requests */
971 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
972 rc = mdt_hsm_pending_restore(cdt_mti);
974 CERROR("%s: cannot take the layout locks needed"
975 " for registered restore: %d\n",
976 mdt_obd_name(mdt), rc);
978 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
981 cdt->cdt_state = CDT_STOPPED;
982 CERROR("%s: error starting coordinator thread: %d\n",
983 mdt_obd_name(mdt), rc);
986 CDEBUG(D_HSM, "%s: coordinator thread started\n",
991 wait_event(cdt->cdt_thread.t_ctl_waitq,
992 (cdt->cdt_thread.t_flags & SVC_RUNNING));
994 cdt->cdt_state = CDT_RUNNING;
995 mdt->mdt_opts.mo_coordinator = 1;
1000 * stop a coordinator thread
1001 * \param mdt [IN] device
1003 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1005 struct coordinator *cdt = &mdt->mdt_coordinator;
1006 struct cdt_agent_req *car, *tmp1;
1007 struct hsm_agent *ha, *tmp2;
1008 struct cdt_restore_handle *crh, *tmp3;
1009 struct mdt_thread_info *cdt_mti;
1012 if (cdt->cdt_state == CDT_STOPPED) {
1013 CERROR("%s: Coordinator already stopped\n",
1018 if (cdt->cdt_state != CDT_STOPPING) {
1019 /* stop coordinator thread before cleaning */
1020 cdt->cdt_thread.t_flags = SVC_STOPPING;
1021 wake_up(&cdt->cdt_thread.t_ctl_waitq);
1022 wait_event(cdt->cdt_thread.t_ctl_waitq,
1023 cdt->cdt_thread.t_flags & SVC_STOPPED);
1025 cdt->cdt_state = CDT_STOPPED;
1027 /* start cleaning */
1028 down_write(&cdt->cdt_request_lock);
1029 list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
1031 list_del(&car->car_request_list);
1032 mdt_cdt_free_request(car);
1034 up_write(&cdt->cdt_request_lock);
1036 down_write(&cdt->cdt_agent_lock);
1037 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
1038 list_del(&ha->ha_list);
1041 up_write(&cdt->cdt_agent_lock);
1043 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1044 mutex_lock(&cdt->cdt_restore_lock);
1045 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
1046 struct mdt_object *child;
1048 /* give back layout lock */
1049 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
1051 mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
1053 list_del(&crh->crh_list);
1055 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1057 mutex_unlock(&cdt->cdt_restore_lock);
1059 mdt->mdt_opts.mo_coordinator = 0;
1065 * register all requests from an hal in the memory list
1066 * \param mti [IN] context
1067 * \param hal [IN] request
1068 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1069 * which is running the CT
1071 * \retval -ve failure
1073 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1074 struct hsm_action_list *hal, struct obd_uuid *uuid)
1076 struct mdt_device *mdt = mti->mti_mdt;
1077 struct coordinator *cdt = &mdt->mdt_coordinator;
1078 struct hsm_action_item *hai;
1082 /* register request in memory list */
1083 hai = hai_first(hal);
1084 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1085 struct cdt_agent_req *car;
1087 /* in case of a cancel request, we first mark the ondisk
1088 * record of the request we want to stop as canceled
1089 * this does not change the cancel record
1090 * it will be done when updating the request status
1092 if (hai->hai_action == HSMA_CANCEL) {
1093 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1097 CERROR("%s: mdt_agent_record_update() failed, "
1098 "rc=%d, cannot update status to %s "
1099 "for cookie "LPX64"\n",
1100 mdt_obd_name(mdt), rc,
1101 agent_req_status2name(ARS_CANCELED),
1106 /* find the running request to set it canceled */
1107 car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1109 car->car_canceled = 1;
1110 /* uuid has to be changed to the one running the
1111 * request to cancel */
1112 *uuid = car->car_uuid;
1113 mdt_cdt_put_request(car);
1115 /* no need to memorize cancel request
1116 * this also avoid a deadlock when we receive
1117 * a purge all requests command
1122 if (hai->hai_action == HSMA_ARCHIVE) {
1123 struct mdt_object *obj;
1126 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1127 if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1130 GOTO(out, rc = PTR_ERR(obj));
1132 hsm.mh_flags |= HS_EXISTS;
1133 hsm.mh_arch_id = hal->hal_archive_id;
1134 rc = mdt_hsm_attr_set(mti, obj, &hsm);
1135 mdt_object_put(mti->mti_env, obj);
1140 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1141 hal->hal_archive_id, hal->hal_flags,
1144 GOTO(out, rc = PTR_ERR(car));
1146 rc = mdt_cdt_add_request(cdt, car);
1148 mdt_cdt_free_request(car);
1155 * swap layouts between 2 fids
1156 * \param mti [IN] context
1159 * \param mh_common [IN] MD HSM
1161 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1162 const lustre_fid *fid, const lustre_fid *dfid,
1163 struct md_hsm *mh_common)
1165 struct mdt_device *mdt = mti->mti_mdt;
1166 struct mdt_object *child1, *child2;
1167 struct mdt_lock_handle *lh2;
1171 child1 = mdt_object_find(mti->mti_env, mdt, fid);
1173 GOTO(out, rc = PTR_ERR(child1));
1175 /* we already have layout lock on FID so take only
1177 lh2 = &mti->mti_lh[MDT_LH_OLD];
1178 mdt_lock_reg_init(lh2, LCK_EX);
1179 child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1181 GOTO(out_child1, rc = PTR_ERR(child2));
1183 /* if copy tool closes the volatile before sending the final
1184 * progress through llapi_hsm_copy_end(), all the objects
1185 * are removed and mdd_swap_layout LBUG */
1186 if (!mdt_object_exists(child2)) {
1187 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1188 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1189 GOTO(out_child2, rc = -ENOENT);
1191 /* Since we only handle restores here, unconditionally use
1192 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1193 * be preserved in case of failure during swap_layout and not
1194 * leave a file in an intermediate but incoherent state.
1195 * But need to setup HSM xattr of data FID before, reuse
1196 * mti and mh presets for FID in hsm_cdt_request_completed(),
1197 * only need to clear RELEASED and DIRTY.
1199 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1200 rc = mdt_hsm_attr_set(mti, child2, mh_common);
1202 rc = mo_swap_layouts(mti->mti_env,
1203 mdt_object_child(child1),
1204 mdt_object_child(child2),
1205 SWAP_LAYOUTS_MDS_HSM);
1208 mdt_object_unlock_put(mti, child2, lh2, 1);
1210 mdt_object_put(mti->mti_env, child1);
1216 * update status of a completed request
1217 * \param mti [IN] context
1218 * \param pgs [IN] progress of the copy tool
1219 * \param update_record [IN] update llog record
1221 * \retval -ve failure
1223 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1224 struct hsm_progress_kernel *pgs,
1225 const struct cdt_agent_req *car,
1226 enum agent_req_status *status)
1228 const struct lu_env *env = mti->mti_env;
1229 struct mdt_device *mdt = mti->mti_mdt;
1230 struct coordinator *cdt = &mdt->mdt_coordinator;
1231 struct mdt_object *obj = NULL;
1232 int cl_flags = 0, rc = 0;
1237 /* default is to retry */
1238 *status = ARS_WAITING;
1240 /* find object by FID */
1241 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1242 /* we will update MD HSM only if needed */
1243 is_mh_changed = false;
1245 /* object removed */
1246 *status = ARS_SUCCEED;
1250 /* no need to change mh->mh_arch_id
1251 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1253 if (pgs->hpk_errval != 0) {
1254 switch (pgs->hpk_errval) {
1256 /* the copy tool does not support cancel
1257 * so the cancel request is failed
1258 * As we cannot distinguish a cancel progress
1259 * from another action progress (they have the
1260 * same cookie), we suppose here the CT returns
1261 * ENOSYS only if does not support cancel
1263 /* this can also happen when cdt calls it to
1264 * for a timeouted request */
1265 *status = ARS_FAILED;
1266 /* to have a cancel event in changelog */
1267 pgs->hpk_errval = ECANCELED;
1270 /* the request record has already been set to
1271 * ARS_CANCELED, this set the cancel request
1273 *status = ARS_SUCCEED;
1276 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1277 !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1278 ARS_FAILED : ARS_WAITING);
1282 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1283 CERROR("%s: Request "LPX64" on "DFID
1284 " failed, error code %d too large\n",
1286 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1288 hsm_set_cl_error(&cl_flags,
1289 CLF_HSM_ERROVERFLOW);
1292 hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1295 switch (car->car_hai->hai_action) {
1297 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1300 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1303 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1306 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1307 CERROR("%s: Failed request "LPX64" on "DFID
1308 " cannot be a CANCEL\n",
1311 PFID(&pgs->hpk_fid));
1314 CERROR("%s: Failed request "LPX64" on "DFID
1315 " %d is an unknown action\n",
1317 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1318 car->car_hai->hai_action);
1323 *status = ARS_SUCCEED;
1324 switch (car->car_hai->hai_action) {
1326 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1327 /* set ARCHIVE keep EXIST and clear LOST and
1329 mh.mh_arch_ver = pgs->hpk_data_version;
1330 mh.mh_flags |= HS_ARCHIVED;
1331 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1332 is_mh_changed = true;
1335 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1337 /* do not clear RELEASED and DIRTY here
1338 * this will occur in hsm_swap_layouts()
1341 /* Restoring has changed the file version on
1343 mh.mh_arch_ver = pgs->hpk_data_version;
1344 is_mh_changed = true;
1347 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1348 /* clear ARCHIVED EXISTS and LOST */
1349 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1350 is_mh_changed = true;
1353 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1354 CERROR("%s: Successful request "LPX64
1356 " cannot be a CANCEL\n",
1359 PFID(&pgs->hpk_fid));
1362 CERROR("%s: Successful request "LPX64
1364 " %d is an unknown action\n",
1366 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1367 car->car_hai->hai_action);
1373 /* rc != 0 means error when analysing action, it may come from
1374 * a crasy CT no need to manage DIRTY
1377 hsm_set_cl_flags(&cl_flags,
1378 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1380 /* unlock is done later, after layout lock management */
1382 rc = mdt_hsm_attr_set(mti, obj, &mh);
1385 /* we give back layout lock only if restore was successful or
1386 * if restore was canceled or if policy is to not retry
1387 * in other cases we just unlock the object */
1388 if (car->car_hai->hai_action == HSMA_RESTORE &&
1389 (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1390 cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1391 struct cdt_restore_handle *crh;
1393 /* restore in data FID done, we swap the layouts
1394 * only if restore is successfull */
1395 if (pgs->hpk_errval == 0) {
1396 rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1397 &car->car_hai->hai_dfid, &mh);
1399 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1400 *status = ARS_FAILED;
1401 pgs->hpk_errval = -rc;
1404 /* we have to retry, so keep layout lock */
1405 if (*status == ARS_WAITING)
1408 /* give back layout lock */
1409 mutex_lock(&cdt->cdt_restore_lock);
1410 crh = hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1412 list_del(&crh->crh_list);
1413 mutex_unlock(&cdt->cdt_restore_lock);
1414 /* just give back layout lock, we keep
1415 * the reference which is given back
1416 * later with the lock for HSM flags */
1417 if (!IS_ERR(obj) && crh != NULL)
1418 mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1421 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1427 if (obj != NULL && !IS_ERR(obj)) {
1428 mo_changelog(env, CL_HSM, cl_flags,
1429 mdt_object_child(obj));
1430 mdt_object_put(mti->mti_env, obj);
1437 * update status of a request
1438 * \param mti [IN] context
1439 * \param pgs [IN] progress of the copy tool
1440 * \param update_record [IN] update llog record
1442 * \retval -ve failure
1444 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1445 struct hsm_progress_kernel *pgs,
1446 const int update_record)
1448 struct mdt_device *mdt = mti->mti_mdt;
1449 struct coordinator *cdt = &mdt->mdt_coordinator;
1450 struct cdt_agent_req *car;
1454 /* no coordinator started, so we cannot serve requests */
1455 if (cdt->cdt_state == CDT_STOPPED)
1458 /* first do sanity checks */
1459 car = mdt_cdt_update_request(cdt, pgs);
1461 CERROR("%s: Cannot find running request for cookie "LPX64
1464 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1467 RETURN(PTR_ERR(car));
1470 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1471 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1472 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1473 hsm_copytool_action2name(car->car_hai->hai_action),
1474 pgs->hpk_flags, pgs->hpk_errval,
1475 PFID(&car->car_hai->hai_fid),
1476 PFID(&car->car_hai->hai_dfid));
1478 /* progress is done on FID or data FID depending of the action and
1479 * of the copy progress */
1480 /* for restore progress is used to send back the data FID to cdt */
1481 if (car->car_hai->hai_action == HSMA_RESTORE &&
1482 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1483 car->car_hai->hai_dfid = pgs->hpk_fid;
1485 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1486 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1487 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1488 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1489 CERROR("%s: Progress on "DFID" for cookie "LPX64
1490 " does not match request FID "DFID" nor data FID "
1493 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1494 PFID(&car->car_hai->hai_fid),
1495 PFID(&car->car_hai->hai_dfid));
1496 GOTO(out, rc = -EINVAL);
1499 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1500 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1501 " is not coherent (err=%d and not completed"
1504 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1505 hsm_copytool_action2name(car->car_hai->hai_action),
1506 pgs->hpk_errval, pgs->hpk_flags);
1507 GOTO(out, rc = -EINVAL);
1510 /* now progress is valid */
1512 /* we use a root like ucred */
1513 hsm_init_ucred(mdt_ucred(mti));
1515 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1516 enum agent_req_status status;
1518 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1520 /* remove request from memory list */
1521 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1523 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1524 " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1526 hsm_copytool_action2name(car->car_hai->hai_action),
1527 agent_req_status2name(status));
1529 if (update_record) {
1532 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1533 &pgs->hpk_cookie, 1,
1536 CERROR("%s: mdt_agent_record_update() failed,"
1537 " rc=%d, cannot update status to %s"
1538 " for cookie "LPX64"\n",
1539 mdt_obd_name(mdt), rc1,
1540 agent_req_status2name(status),
1542 rc = (rc != 0 ? rc : rc1);
1544 /* ct has completed a request, so a slot is available, wakeup
1545 * cdt to find new work */
1546 mdt_hsm_cdt_wakeup(mdt);
1548 /* if copytool send a progress on a canceled request
1549 * we inform copytool it should stop
1551 if (car->car_canceled == 1)
1557 /* remove ref got from mdt_cdt_update_request() */
1558 mdt_cdt_put_request(car);
1565 * data passed to llog_cat_process() callback
1566 * to cancel requests
1568 struct hsm_cancel_all_data {
1569 struct mdt_device *mdt;
1573 * llog_cat_process() callback, used to:
1574 * - purge all requests
1575 * \param env [IN] environment
1576 * \param llh [IN] llog handle
1577 * \param hdr [IN] llog record
1578 * \param data [IN] cb data = struct hsm_cancel_all_data
1580 * \retval -ve failure
1582 static int mdt_cancel_all_cb(const struct lu_env *env,
1583 struct llog_handle *llh,
1584 struct llog_rec_hdr *hdr, void *data)
1586 struct llog_agent_req_rec *larr;
1587 struct hsm_cancel_all_data *hcad;
1591 larr = (struct llog_agent_req_rec *)hdr;
1593 if (larr->arr_status == ARS_WAITING ||
1594 larr->arr_status == ARS_STARTED) {
1595 larr->arr_status = ARS_CANCELED;
1596 larr->arr_req_change = cfs_time_current_sec();
1597 rc = mdt_agent_llog_update_rec(env, hcad->mdt, llh, larr);
1599 RETURN(LLOG_DEL_RECORD);
1605 * cancel all actions
1606 * \param obd [IN] MDT device
1608 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1610 struct mdt_thread_info *mti;
1611 struct coordinator *cdt = &mdt->mdt_coordinator;
1612 struct cdt_agent_req *car;
1613 struct hsm_action_list *hal = NULL;
1614 struct hsm_action_item *hai;
1615 struct hsm_cancel_all_data hcad;
1616 int hal_sz = 0, hal_len, rc;
1617 enum cdt_states save_state;
1620 /* retrieve coordinator context */
1621 mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1623 /* disable coordinator */
1624 save_state = cdt->cdt_state;
1625 cdt->cdt_state = CDT_DISABLE;
1627 /* send cancel to all running requests */
1628 down_read(&cdt->cdt_request_lock);
1629 list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1630 mdt_cdt_get_request(car);
1631 /* request is not yet removed from list, it will be done
1632 * when copytool will return progress
1635 if (car->car_hai->hai_action == HSMA_CANCEL) {
1636 mdt_cdt_put_request(car);
1641 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1642 cfs_size_round(car->car_hai->hai_len);
1644 if (hal_len > hal_sz && hal_sz > 0) {
1645 /* not enough room, free old buffer */
1646 OBD_FREE(hal, hal_sz);
1650 /* empty buffer, allocate one */
1653 OBD_ALLOC(hal, hal_sz);
1655 mdt_cdt_put_request(car);
1656 up_read(&cdt->cdt_request_lock);
1657 GOTO(out, rc = -ENOMEM);
1661 hal->hal_version = HAL_VERSION;
1662 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1664 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1665 hal->hal_compound_id = car->car_compound_id;
1666 hal->hal_archive_id = car->car_archive_id;
1667 hal->hal_flags = car->car_flags;
1670 hai = hai_first(hal);
1671 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1672 hai->hai_action = HSMA_CANCEL;
1675 /* it is possible to safely call mdt_hsm_agent_send()
1676 * (ie without a deadlock on cdt_request_lock), because the
1677 * write lock is taken only if we are not in purge mode
1678 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1679 * nor mdt_cdt_remove_request())
1681 /* no conflict with cdt thread because cdt is disable and we
1682 * have the request lock */
1683 mdt_hsm_agent_send(mti, hal, 1);
1685 mdt_cdt_put_request(car);
1687 up_read(&cdt->cdt_request_lock);
1690 OBD_FREE(hal, hal_sz);
1692 /* cancel all on-disk records */
1695 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1696 mdt_cancel_all_cb, &hcad);
1698 /* enable coordinator */
1699 cdt->cdt_state = save_state;
1705 * check if a request is comptaible with file status
1706 * \param hai [IN] request description
1707 * \param hal_an [IN] request archive number (not used)
1708 * \param rq_flags [IN] request flags
1709 * \param hsm [IN] file HSM metadata
1712 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1713 const int hal_an, const __u64 rq_flags,
1714 const struct md_hsm *hsm)
1716 int is_compat = false;
1720 hsm_flags = hsm->mh_flags;
1721 switch (hai->hai_action) {
1723 if (!(hsm_flags & HS_NOARCHIVE) &&
1724 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1728 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1729 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1733 if (!(hsm_flags & HS_RELEASED) &&
1734 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1741 CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1742 " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1743 PFID(&hai->hai_fid),
1744 hsm_copytool_action2name(hai->hai_action), rq_flags,
1745 hai->hai_extent.offset, hai->hai_extent.length,
1747 (is_compat ? "compatible" : "uncompatible"));
1753 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1755 static const struct {
1759 } hsm_policy_names[] = {
1760 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1761 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1766 * convert a policy name to a bit
1767 * \param name [IN] policy name
1769 * \retval policy bit
1771 static __u64 hsm_policy_str2bit(const char *name)
1775 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1776 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1777 strcmp(hsm_policy_names[i].name, name) == 0)
1778 return hsm_policy_names[i].bit;
1783 * convert a policy bit field to a string
1784 * \param mask [IN] policy bit field
1785 * \param hexa [IN] print mask before bit names
1786 * \param buffer [OUT] string
1787 * \param count [IN] size of buffer
1789 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1797 seq_printf(m, "("LPX64") ", mask);
1799 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1802 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1803 if (hsm_policy_names[j].bit == bit)
1807 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1809 seq_printf(m, "%s ", hsm_policy_names[j].name);
1811 /* remove last ' ' */
1816 /* methods to read/write HSM policy flags */
1817 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1819 struct mdt_device *mdt = m->private;
1820 struct coordinator *cdt = &mdt->mdt_coordinator;
1823 hsm_policy_bit2str(m, cdt->cdt_policy, false);
1828 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1829 size_t count, loff_t *off)
1831 struct seq_file *m = file->private_data;
1832 struct mdt_device *mdt = m->private;
1833 struct coordinator *cdt = &mdt->mdt_coordinator;
1834 char *start, *token, sign;
1837 __u64 add_mask, remove_mask, set_mask;
1841 if (count + 1 > PAGE_SIZE)
1844 OBD_ALLOC(buf, count + 1);
1848 if (copy_from_user(buf, buffer, count))
1849 GOTO(out, rc = -EFAULT);
1854 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1857 add_mask = remove_mask = set_mask = 0;
1859 token = strsep(&start, "\n ");
1865 if (sign == '-' || sign == '+')
1868 policy = hsm_policy_str2bit(token);
1870 CWARN("%s: '%s' is unknown, "
1871 "supported policies are:\n", mdt_obd_name(mdt),
1873 hsm_policy_bit2str(m, 0, false);
1874 GOTO(out, rc = -EINVAL);
1878 remove_mask |= policy;
1888 } while (start != NULL);
1890 CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1891 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1893 /* if no sign in all string, it is a clear and set
1894 * if some sign found, all unsigned are converted
1896 * P1 P2 = set to P1 and P2
1897 * P1 -P2 = add P1 clear P2 same as +P1 -P2
1899 if (remove_mask == 0 && add_mask == 0) {
1900 cdt->cdt_policy = set_mask;
1902 cdt->cdt_policy |= set_mask | add_mask;
1903 cdt->cdt_policy &= ~remove_mask;
1906 GOTO(out, rc = count);
1909 OBD_FREE(buf, count + 1);
1912 LPROC_SEQ_FOPS(mdt_hsm_policy);
1914 #define GENERATE_PROC_METHOD(VAR) \
1915 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
1917 struct mdt_device *mdt = m->private; \
1918 struct coordinator *cdt = &mdt->mdt_coordinator; \
1921 seq_printf(m, LPU64"\n", (__u64)cdt->VAR); \
1925 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1926 size_t count, loff_t *off) \
1929 struct seq_file *m = file->private_data; \
1930 struct mdt_device *mdt = m->private; \
1931 struct coordinator *cdt = &mdt->mdt_coordinator; \
1936 rc = lprocfs_write_helper(buffer, count, &val); \
1946 GENERATE_PROC_METHOD(cdt_loop_period)
1947 GENERATE_PROC_METHOD(cdt_grace_delay)
1948 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1949 GENERATE_PROC_METHOD(cdt_max_requests)
1950 GENERATE_PROC_METHOD(cdt_default_archive_id)
1953 * procfs write method for MDT/hsm_control
1954 * proc entry is in mdt directory so data is mdt obd_device pointer
1956 #define CDT_ENABLE_CMD "enabled"
1957 #define CDT_STOP_CMD "shutdown"
1958 #define CDT_DISABLE_CMD "disabled"
1959 #define CDT_PURGE_CMD "purge"
1960 #define CDT_HELP_CMD "help"
1961 #define CDT_MAX_CMD_LEN 10
1964 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1965 size_t count, loff_t *off)
1967 struct seq_file *m = file->private_data;
1968 struct obd_device *obd = m->private;
1969 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1970 struct coordinator *cdt = &(mdt->mdt_coordinator);
1972 char kernbuf[CDT_MAX_CMD_LEN];
1975 if (count == 0 || count >= sizeof(kernbuf))
1978 if (copy_from_user(kernbuf, buffer, count))
1982 if (kernbuf[count - 1] == '\n')
1983 kernbuf[count - 1] = 0;
1986 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1987 if (cdt->cdt_state == CDT_DISABLE) {
1988 cdt->cdt_state = CDT_RUNNING;
1989 mdt_hsm_cdt_wakeup(mdt);
1991 rc = mdt_hsm_cdt_start(mdt);
1993 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1994 if ((cdt->cdt_state == CDT_STOPPING) ||
1995 (cdt->cdt_state == CDT_STOPPED)) {
1996 CERROR("%s: Coordinator already stopped\n",
2000 cdt->cdt_state = CDT_STOPPING;
2002 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2003 if ((cdt->cdt_state == CDT_STOPPING) ||
2004 (cdt->cdt_state == CDT_STOPPED)) {
2005 CERROR("%s: Coordinator is stopped\n",
2009 cdt->cdt_state = CDT_DISABLE;
2011 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2012 rc = hsm_cancel_all_actions(mdt);
2013 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2021 CERROR("%s: Valid coordinator control commands are: "
2022 "%s %s %s %s %s\n", mdt_obd_name(mdt),
2023 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2024 CDT_PURGE_CMD, CDT_HELP_CMD);
2032 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2034 struct obd_device *obd = m->private;
2035 struct coordinator *cdt;
2038 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2040 if (cdt->cdt_state == CDT_INIT)
2041 seq_printf(m, "init\n");
2042 else if (cdt->cdt_state == CDT_RUNNING)
2043 seq_printf(m, "enabled\n");
2044 else if (cdt->cdt_state == CDT_STOPPING)
2045 seq_printf(m, "stopping\n");
2046 else if (cdt->cdt_state == CDT_STOPPED)
2047 seq_printf(m, "stopped\n");
2048 else if (cdt->cdt_state == CDT_DISABLE)
2049 seq_printf(m, "disabled\n");
2051 seq_printf(m, "unknown\n");
2057 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2062 for (i = 0; i < 8 * sizeof(mask); i++) {
2063 if (mask & (1UL << i))
2064 rc += seq_printf(m, "%s%s", rc == 0 ? "" : " ",
2065 hsm_copytool_action2name(i));
2067 rc += seq_printf(m, "\n");
2073 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2075 struct mdt_device *mdt = m->private;
2076 struct coordinator *cdt = &mdt->mdt_coordinator;
2078 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2082 mdt_hsm_group_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_group_request_mask);
2091 mdt_hsm_other_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_other_request_mask);
2099 static inline enum hsm_copytool_action
2100 hsm_copytool_name2action(const char *name)
2102 if (strcasecmp(name, "NOOP") == 0)
2104 else if (strcasecmp(name, "ARCHIVE") == 0)
2105 return HSMA_ARCHIVE;
2106 else if (strcasecmp(name, "RESTORE") == 0)
2107 return HSMA_RESTORE;
2108 else if (strcasecmp(name, "REMOVE") == 0)
2110 else if (strcasecmp(name, "CANCEL") == 0)
2117 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2118 size_t user_count, __u64 *mask)
2120 char *buf, *pos, *name;
2126 if (!(user_count < 4096))
2129 buf_size = user_count + 1;
2131 OBD_ALLOC(buf, buf_size);
2135 if (copy_from_user(buf, user_buf, buf_size - 1))
2136 GOTO(out, rc = -EFAULT);
2138 buf[buf_size - 1] = '\0';
2141 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2147 action = hsm_copytool_name2action(name);
2149 GOTO(out, rc = -EINVAL);
2151 new_mask |= (1UL << action);
2157 OBD_FREE(buf, buf_size);
2163 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2164 size_t count, loff_t *off)
2166 struct seq_file *m = file->private_data;
2167 struct mdt_device *mdt = m->private;
2168 struct coordinator *cdt = &mdt->mdt_coordinator;
2170 return mdt_write_hsm_request_mask(file, buf, count,
2171 &cdt->cdt_user_request_mask);
2175 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2176 size_t count, loff_t *off)
2178 struct seq_file *m = file->private_data;
2179 struct mdt_device *mdt = m->private;
2180 struct coordinator *cdt = &mdt->mdt_coordinator;
2182 return mdt_write_hsm_request_mask(file, buf, count,
2183 &cdt->cdt_group_request_mask);
2187 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2188 size_t count, loff_t *off)
2190 struct seq_file *m = file->private_data;
2191 struct mdt_device *mdt = m->private;
2192 struct coordinator *cdt = &mdt->mdt_coordinator;
2194 return mdt_write_hsm_request_mask(file, buf, count,
2195 &cdt->cdt_other_request_mask);
2198 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2199 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2200 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2201 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2202 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2203 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2204 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2205 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2207 static struct lprocfs_seq_vars lprocfs_mdt_hsm_vars[] = {
2209 .fops = &mdt_hsm_agent_fops },
2210 { .name = "actions",
2211 .fops = &mdt_hsm_actions_fops,
2212 .proc_mode = 0444 },
2213 { .name = "default_archive_id",
2214 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2215 { .name = "grace_delay",
2216 .fops = &mdt_hsm_cdt_grace_delay_fops },
2217 { .name = "loop_period",
2218 .fops = &mdt_hsm_cdt_loop_period_fops },
2219 { .name = "max_requests",
2220 .fops = &mdt_hsm_cdt_max_requests_fops },
2222 .fops = &mdt_hsm_policy_fops },
2223 { .name = "active_request_timeout",
2224 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2225 { .name = "active_requests",
2226 .fops = &mdt_hsm_active_requests_fops },
2227 { .name = "user_request_mask",
2228 .fops = &mdt_hsm_user_request_mask_fops, },
2229 { .name = "group_request_mask",
2230 .fops = &mdt_hsm_group_request_mask_fops, },
2231 { .name = "other_request_mask",
2232 .fops = &mdt_hsm_other_request_mask_fops, },