4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
26 * Copyright (c) 2013, 2016, Intel Corporation.
27 * Use is subject to license terms.
30 * lustre/mdt/mdt_coordinator.c
32 * Lustre HSM Coordinator
34 * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
35 * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
36 * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <linux/kthread.h>
42 #include <obd_support.h>
43 #include <lustre_net.h>
44 #include <lustre_export.h>
46 #include <lprocfs_status.h>
47 #include <lustre_log.h>
48 #include <lustre_kernelcomm.h>
49 #include "mdt_internal.h"
51 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
54 * get obj and HSM attributes on a fid
55 * \param mti [IN] context
56 * \param fid [IN] object fid
57 * \param hsm [OUT] HSM meta data
58 * \retval obj or error (-ENOENT if not found)
60 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
61 const struct lu_fid *fid,
65 struct mdt_object *obj;
73 /* find object by FID */
74 obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
78 if (!mdt_object_exists(obj)) {
80 mdt_object_put(mti->mti_env, obj);
81 RETURN(ERR_PTR(-ENOENT));
84 rc = mdt_attr_get_complex(mti, obj, ma);
86 mdt_object_put(mti->mti_env, obj);
90 if (ma->ma_valid & MA_HSM)
93 memset(hsm, 0, sizeof(*hsm));
98 void mdt_hsm_dump_hal(int level, const char *prefix,
99 struct hsm_action_list *hal)
102 struct hsm_action_item *hai;
105 CDEBUG(level, "%s: HAL header: version %X count %d compound %#llx"
106 " archive_id %d flags %#llx\n",
107 prefix, hal->hal_version, hal->hal_count,
108 hal->hal_compound_id, hal->hal_archive_id, hal->hal_flags);
110 hai = hai_first(hal);
111 for (i = 0; i < hal->hal_count; i++) {
112 sz = hai->hai_len - sizeof(*hai);
113 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
114 " compound/cookie=%#llx/%#llx"
115 " action=%s extent=%#llx-%#llx gid=%#llx"
116 " datalen=%d data=[%s]\n",
118 PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
119 hal->hal_compound_id, hai->hai_cookie,
120 hsm_copytool_action2name(hai->hai_action),
121 hai->hai_extent.offset,
122 hai->hai_extent.length,
124 hai_dump_data_field(hai, buf, sizeof(buf)));
130 * data passed to llog_cat_process() callback
131 * to scan requests and take actions
133 struct hsm_scan_request {
136 struct hsm_action_list *hal;
139 struct hsm_scan_data {
140 struct mdt_thread_info *mti;
141 char fs_name[MTI_NAME_MAXLEN+1];
142 /* request to be send to agents */
143 int max_requests; /** vector size */
144 int request_cnt; /** used count */
145 struct hsm_scan_request *request;
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 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) {
181 struct hsm_scan_request *request;
183 /* Are agents full? */
184 if (atomic_read(&cdt->cdt_request_count) >=
185 cdt->cdt_max_requests)
188 /* first search whether the request is found in the
189 * list we have built. */
191 for (i = 0; i < hsd->request_cnt; i++) {
192 if (hsd->request[i].hal->hal_compound_id ==
193 larr->arr_compound_id) {
194 request = &hsd->request[i];
200 struct hsm_action_list *hal;
202 if (hsd->request_cnt == hsd->max_requests)
203 /* Unknown request and no more room
204 * for a new request. Continue to scan
205 * to find other entries for already
210 request = &hsd->request[hsd->request_cnt];
212 /* allocates hai vector size just needs to be large
215 sizeof(*request->hal) +
216 cfs_size_round(MTI_NAME_MAXLEN+1) +
217 2 * cfs_size_round(larr->arr_hai.hai_len);
218 OBD_ALLOC(hal, request->hal_sz);
221 hal->hal_version = HAL_VERSION;
222 strlcpy(hal->hal_fsname, hsd->fs_name,
223 MTI_NAME_MAXLEN + 1);
224 hal->hal_compound_id = larr->arr_compound_id;
225 hal->hal_archive_id = larr->arr_archive_id;
226 hal->hal_flags = larr->arr_flags;
228 request->hal_used_sz = hal_size(hal);
231 hai = hai_first(hal);
233 /* request is known */
234 /* we check if record archive num is the same as the
235 * known request, if not we will serve it in multiple
236 * time because we do not know if the agent can serve
238 * a use case is a compound made of multiple restore
239 * where the files are not archived in the same backend
241 if (larr->arr_archive_id !=
242 request->hal->hal_archive_id)
245 if (request->hal_sz <
246 request->hal_used_sz +
247 cfs_size_round(larr->arr_hai.hai_len)) {
248 /* Not enough room, need an extension */
252 sz = 2 * request->hal_sz;
253 OBD_ALLOC(hal_buffer, sz);
256 memcpy(hal_buffer, request->hal,
257 request->hal_used_sz);
258 OBD_FREE(request->hal,
260 request->hal = hal_buffer;
261 request->hal_sz = sz;
263 hai = hai_first(request->hal);
264 for (i = 0; i < request->hal->hal_count; i++)
267 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
268 hai->hai_cookie = larr->arr_hai.hai_cookie;
269 hai->hai_gid = larr->arr_hai.hai_gid;
271 request->hal_used_sz += cfs_size_round(hai->hai_len);
272 request->hal->hal_count++;
276 struct hsm_progress_kernel pgs;
277 struct cdt_agent_req *car;
278 cfs_time_t now = cfs_time_current_sec();
281 /* we search for a running request
282 * error may happen if coordinator crashes or stopped
283 * with running request
285 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
287 last = larr->arr_req_create;
289 last = car->car_req_update;
290 mdt_cdt_put_request(car);
293 /* test if request too long, if yes cancel it
294 * the same way the copy tool acknowledge a cancel request */
295 if (now <= last + cdt->cdt_active_req_timeout)
298 dump_llog_agent_req_rec("request timed out, start cleaning",
300 /* a too old cancel request just needs to be removed
301 * this can happen, if copy tool does not support
302 * cancel for other requests, we have to remove the
303 * running request and notify the copytool */
304 pgs.hpk_fid = larr->arr_hai.hai_fid;
305 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
306 pgs.hpk_extent = larr->arr_hai.hai_extent;
307 pgs.hpk_flags = HP_FLAG_COMPLETED;
308 pgs.hpk_errval = ENOSYS;
309 pgs.hpk_data_version = 0;
311 /* update request state, but do not record in llog, to
312 * avoid deadlock on cdt_llog_lock */
313 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
315 CERROR("%s: cannot cleanup timed out request: "
316 DFID" for cookie %#llx action=%s\n",
318 PFID(&pgs.hpk_fid), pgs.hpk_cookie,
319 hsm_copytool_action2name(
320 larr->arr_hai.hai_action));
323 /* The request no longer exists, forget
324 * about it, and do not send a cancel request
325 * to the client, for which an error will be
326 * sent back, leading to an endless cycle of
328 RETURN(LLOG_DEL_RECORD);
331 /* XXX A cancel request cannot be cancelled. */
332 if (larr->arr_hai.hai_action == HSMA_CANCEL)
335 larr->arr_status = ARS_CANCELED;
336 larr->arr_req_change = now;
337 rc = llog_write(hsd->mti->mti_env, llh, hdr, hdr->lrh_index);
339 CERROR("%s: cannot update agent log: rc = %d\n",
340 mdt_obd_name(mdt), rc);
346 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
347 cfs_time_current_sec())
348 RETURN(LLOG_DEL_RECORD);
355 * create /proc entries for coordinator
358 * \retval -ve failure
360 int hsm_cdt_procfs_init(struct mdt_device *mdt)
362 struct coordinator *cdt = &mdt->mdt_coordinator;
366 /* init /proc entries, failure is not critical */
367 cdt->cdt_proc_dir = lprocfs_register("hsm",
368 mdt2obd_dev(mdt)->obd_proc_entry,
369 lprocfs_mdt_hsm_vars, mdt);
370 if (IS_ERR(cdt->cdt_proc_dir)) {
371 rc = PTR_ERR(cdt->cdt_proc_dir);
372 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
373 " rc=%d\n", mdt_obd_name(mdt), rc);
374 cdt->cdt_proc_dir = NULL;
382 * remove /proc entries for coordinator
385 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
387 struct coordinator *cdt = &mdt->mdt_coordinator;
389 LASSERT(cdt->cdt_state == CDT_STOPPED);
390 if (cdt->cdt_proc_dir != NULL)
391 lprocfs_remove(&cdt->cdt_proc_dir);
395 * get vector of hsm cdt /proc vars
399 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
401 return lprocfs_mdt_hsm_vars;
406 * \param data [IN] obd device
408 * \retval -ve failure
410 static int mdt_coordinator(void *data)
412 struct mdt_thread_info *mti = data;
413 struct mdt_device *mdt = mti->mti_mdt;
414 struct coordinator *cdt = &mdt->mdt_coordinator;
415 struct hsm_scan_data hsd = { NULL };
420 cdt->cdt_flags = SVC_RUNNING;
421 wake_up(&cdt->cdt_waitq);
423 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
424 mdt_obd_name(mdt), current_pid());
426 /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
427 * increases due to a change from /proc we do not overflow the
428 * hsd.request[] vector
430 hsd.max_requests = cdt->cdt_max_requests;
431 request_sz = hsd.max_requests * sizeof(*hsd.request);
432 OBD_ALLOC(hsd.request, request_sz);
434 GOTO(out, rc = -ENOMEM);
437 obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
440 struct l_wait_info lwi;
443 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
445 l_wait_event(cdt->cdt_waitq,
446 cdt->cdt_flags & (SVC_STOPPING|SVC_EVENT),
449 CDEBUG(D_HSM, "coordinator resumes\n");
451 if (cdt->cdt_flags & SVC_STOPPING ||
452 cdt->cdt_state == CDT_STOPPING) {
453 cdt->cdt_flags &= ~SVC_STOPPING;
458 /* wake up before timeout, new work arrives */
459 if (cdt->cdt_flags & SVC_EVENT)
460 cdt->cdt_flags &= ~SVC_EVENT;
462 /* if coordinator is suspended continue to wait */
463 if (cdt->cdt_state == CDT_DISABLE) {
464 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
468 CDEBUG(D_HSM, "coordinator starts reading llog\n");
470 if (hsd.max_requests != cdt->cdt_max_requests) {
471 /* cdt_max_requests has changed,
472 * we need to allocate a new buffer
474 OBD_FREE(hsd.request, request_sz);
475 hsd.max_requests = cdt->cdt_max_requests;
476 request_sz = hsd.max_requests * sizeof(*hsd.request);
477 OBD_ALLOC(hsd.request, request_sz);
486 rc = cdt_llog_process(mti->mti_env, mdt,
487 mdt_coordinator_cb, &hsd);
491 CDEBUG(D_HSM, "found %d requests to send\n", hsd.request_cnt);
493 if (list_empty(&cdt->cdt_agents)) {
494 CDEBUG(D_HSM, "no agent available, "
495 "coordinator sleeps\n");
499 /* here hsd contains a list of requests to be started */
500 for (i = 0; i < hsd.request_cnt; i++) {
501 struct hsm_scan_request *request = &hsd.request[i];
502 struct hsm_action_list *hal = request->hal;
503 struct hsm_action_item *hai;
506 enum agent_req_status status;
508 /* still room for work ? */
509 if (atomic_read(&cdt->cdt_request_count) >=
510 cdt->cdt_max_requests)
513 rc = mdt_hsm_agent_send(mti, hal, 0);
514 /* if failure, we suppose it is temporary
515 * if the copy tool failed to do the request
516 * it has to use hsm_progress
518 status = (rc ? ARS_WAITING : ARS_STARTED);
520 /* set up cookie vector to set records status
521 * after copy tools start or failed
523 sz = hal->hal_count * sizeof(__u64);
524 OBD_ALLOC(cookies, sz);
528 hai = hai_first(hal);
529 for (j = 0; j < hal->hal_count; j++) {
530 cookies[j] = hai->hai_cookie;
534 rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
535 hal->hal_count, status);
537 CERROR("%s: mdt_agent_record_update() failed, "
538 "rc=%d, cannot update status to %s "
540 mdt_obd_name(mdt), rc,
541 agent_req_status2name(status),
544 OBD_FREE(cookies, sz);
547 /* free hal allocated by callback */
548 for (i = 0; i < hsd.request_cnt; i++) {
549 struct hsm_scan_request *request = &hsd.request[i];
551 OBD_FREE(request->hal, request->hal_sz);
557 OBD_FREE(hsd.request, request_sz);
559 if (cdt->cdt_state == CDT_STOPPING) {
560 /* request comes from /proc path, so we need to clean cdt
562 mdt_hsm_cdt_stop(mdt);
563 mdt->mdt_opts.mo_coordinator = 0;
565 /* request comes from a thread event, generated
566 * by mdt_stop_coordinator(), we have to ack
567 * and cdt cleaning will be done by event sender
569 cdt->cdt_flags = SVC_STOPPED;
570 wake_up(&cdt->cdt_waitq);
574 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
575 mdt_obd_name(mdt), current_pid(), rc);
577 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
579 mdt_obd_name(mdt), current_pid());
585 * lookup a restore handle by FID
586 * caller needs to hold cdt_restore_lock
587 * \param cdt [IN] coordinator
588 * \param fid [IN] FID
589 * \retval cdt_restore_handle found
590 * \retval NULL not found
592 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
593 const struct lu_fid *fid)
595 struct cdt_restore_handle *crh;
598 list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
599 if (lu_fid_eq(&crh->crh_fid, fid))
606 * data passed to llog_cat_process() callback
607 * to scan requests and take actions
609 struct hsm_restore_data {
610 struct mdt_thread_info *hrd_mti;
614 * llog_cat_process() callback, used to:
615 * - find restore request and allocate the restore handle
616 * \param env [IN] environment
617 * \param llh [IN] llog handle
618 * \param hdr [IN] llog record
619 * \param data [IN/OUT] cb data = struct hsm_restore_data
621 * \retval -ve failure
623 static int hsm_restore_cb(const struct lu_env *env,
624 struct llog_handle *llh,
625 struct llog_rec_hdr *hdr, void *data)
627 struct llog_agent_req_rec *larr;
628 struct hsm_restore_data *hrd;
629 struct cdt_restore_handle *crh;
630 struct hsm_action_item *hai;
631 struct mdt_thread_info *mti;
632 struct coordinator *cdt;
633 struct mdt_object *child;
639 cdt = &mti->mti_mdt->mdt_coordinator;
641 larr = (struct llog_agent_req_rec *)hdr;
642 hai = &larr->arr_hai;
643 if (hai->hai_cookie > cdt->cdt_last_cookie)
644 /* update the cookie to avoid collision */
645 cdt->cdt_last_cookie = hai->hai_cookie + 1;
647 if (hai->hai_action != HSMA_RESTORE ||
648 agent_req_in_final_state(larr->arr_status))
651 /* restore request not in a final state */
653 /* force replay of restore requests left in started state from previous
654 * CDT context, to be canceled later if finally found to be incompatible
655 * when being re-started */
656 if (larr->arr_status == ARS_STARTED) {
657 larr->arr_status = ARS_WAITING;
658 larr->arr_req_change = cfs_time_current_sec();
659 rc = llog_write(env, llh, hdr, hdr->lrh_index);
664 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
668 crh->crh_fid = hai->hai_fid;
669 /* in V1 all file is restored
670 crh->extent.start = hai->hai_extent.offset;
671 crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
673 crh->crh_extent.start = 0;
674 crh->crh_extent.end = hai->hai_extent.length;
675 /* get the layout lock */
676 mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
677 child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
678 MDS_INODELOCK_LAYOUT);
680 GOTO(out, rc = PTR_ERR(child));
683 /* we choose to not keep a reference
684 * on the object during the restore time which can be very long */
685 mdt_object_put(mti->mti_env, child);
687 mutex_lock(&cdt->cdt_restore_lock);
688 list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
689 mutex_unlock(&cdt->cdt_restore_lock);
696 * restore coordinator state at startup
697 * the goal is to take a layout lock for each registered restore request
698 * \param mti [IN] context
700 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
702 struct hsm_restore_data hrd;
708 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
709 hsm_restore_cb, &hrd);
714 static int hsm_init_ucred(struct lu_ucred *uc)
718 uc->uc_valid = UCRED_OLD;
727 uc->uc_suppgids[0] = -1;
728 uc->uc_suppgids[1] = -1;
729 uc->uc_cap = CFS_CAP_FS_MASK;
732 uc->uc_identity = NULL;
738 * wake up coordinator thread
739 * \param mdt [IN] device
741 * \retval -ve failure
743 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
745 struct coordinator *cdt = &mdt->mdt_coordinator;
748 if (cdt->cdt_state == CDT_STOPPED)
751 /* wake up coordinator */
752 cdt->cdt_flags = SVC_EVENT;
753 wake_up(&cdt->cdt_waitq);
759 * initialize coordinator struct
760 * \param mdt [IN] device
762 * \retval -ve failure
764 int mdt_hsm_cdt_init(struct mdt_device *mdt)
766 struct coordinator *cdt = &mdt->mdt_coordinator;
767 struct mdt_thread_info *cdt_mti = NULL;
771 cdt->cdt_state = CDT_STOPPED;
773 init_waitqueue_head(&cdt->cdt_waitq);
774 mutex_init(&cdt->cdt_llog_lock);
775 init_rwsem(&cdt->cdt_agent_lock);
776 init_rwsem(&cdt->cdt_request_lock);
777 mutex_init(&cdt->cdt_restore_lock);
779 INIT_LIST_HEAD(&cdt->cdt_requests);
780 INIT_LIST_HEAD(&cdt->cdt_agents);
781 INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
783 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
787 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
788 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
790 lu_context_enter(&cdt->cdt_session);
791 cdt->cdt_env.le_ses = &cdt->cdt_session;
793 lu_env_fini(&cdt->cdt_env);
797 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
798 LASSERT(cdt_mti != NULL);
800 cdt_mti->mti_env = &cdt->cdt_env;
801 cdt_mti->mti_mdt = mdt;
803 hsm_init_ucred(mdt_ucred(cdt_mti));
805 /* default values for /proc tunnables
806 * can be override by MGS conf */
807 cdt->cdt_default_archive_id = 1;
808 cdt->cdt_grace_delay = 60;
809 cdt->cdt_loop_period = 10;
810 cdt->cdt_max_requests = 3;
811 cdt->cdt_policy = CDT_DEFAULT_POLICY;
812 cdt->cdt_active_req_timeout = 3600;
818 * free a coordinator thread
819 * \param mdt [IN] device
821 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
823 struct coordinator *cdt = &mdt->mdt_coordinator;
826 lu_context_exit(cdt->cdt_env.le_ses);
827 lu_context_fini(cdt->cdt_env.le_ses);
829 lu_env_fini(&cdt->cdt_env);
835 * start a coordinator thread
836 * \param mdt [IN] device
838 * \retval -ve failure
840 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
842 struct coordinator *cdt = &mdt->mdt_coordinator;
845 struct mdt_thread_info *cdt_mti;
846 struct task_struct *task;
849 /* functions defined but not yet used
850 * this avoid compilation warning
854 if (cdt->cdt_state != CDT_STOPPED) {
855 CERROR("%s: Coordinator already started\n",
860 CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
861 cdt->cdt_policy = CDT_DEFAULT_POLICY;
863 cdt->cdt_state = CDT_INIT;
865 atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
866 /* just need to be larger than previous one */
867 /* cdt_last_cookie is protected by cdt_llog_lock */
868 cdt->cdt_last_cookie = cfs_time_current_sec();
869 atomic_set(&cdt->cdt_request_count, 0);
870 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
871 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
872 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
874 /* to avoid deadlock when start is made through /proc
875 * /proc entries are created by the coordinator thread */
877 /* set up list of started restore requests */
878 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
879 rc = mdt_hsm_pending_restore(cdt_mti);
881 CERROR("%s: cannot take the layout locks needed"
882 " for registered restore: %d\n",
883 mdt_obd_name(mdt), rc);
885 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
888 cdt->cdt_state = CDT_STOPPED;
889 CERROR("%s: error starting coordinator thread: %d\n",
890 mdt_obd_name(mdt), rc);
893 CDEBUG(D_HSM, "%s: coordinator thread started\n",
898 wait_event(cdt->cdt_waitq,
899 (cdt->cdt_flags & SVC_RUNNING));
901 cdt->cdt_state = CDT_RUNNING;
902 mdt->mdt_opts.mo_coordinator = 1;
907 * stop a coordinator thread
908 * \param mdt [IN] device
910 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
912 struct coordinator *cdt = &mdt->mdt_coordinator;
913 struct cdt_agent_req *car, *tmp1;
914 struct hsm_agent *ha, *tmp2;
915 struct cdt_restore_handle *crh, *tmp3;
916 struct mdt_thread_info *cdt_mti;
919 if (cdt->cdt_state == CDT_STOPPED) {
920 CERROR("%s: Coordinator already stopped\n",
925 if (cdt->cdt_state != CDT_STOPPING) {
926 /* stop coordinator thread before cleaning */
927 cdt->cdt_flags = SVC_STOPPING;
928 wake_up(&cdt->cdt_waitq);
929 wait_event(cdt->cdt_waitq,
930 cdt->cdt_flags & SVC_STOPPED);
932 cdt->cdt_state = CDT_STOPPED;
935 down_write(&cdt->cdt_request_lock);
936 list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
938 list_del(&car->car_request_list);
939 mdt_cdt_free_request(car);
941 up_write(&cdt->cdt_request_lock);
943 down_write(&cdt->cdt_agent_lock);
944 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
945 list_del(&ha->ha_list);
948 up_write(&cdt->cdt_agent_lock);
950 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
951 mutex_lock(&cdt->cdt_restore_lock);
952 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
953 struct mdt_object *child;
955 /* give back layout lock */
956 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
958 mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
960 list_del(&crh->crh_list);
962 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
964 mutex_unlock(&cdt->cdt_restore_lock);
966 mdt->mdt_opts.mo_coordinator = 0;
972 * register all requests from an hal in the memory list
973 * \param mti [IN] context
974 * \param hal [IN] request
975 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
976 * which is running the CT
978 * \retval -ve failure
980 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
981 struct hsm_action_list *hal, struct obd_uuid *uuid)
983 struct mdt_device *mdt = mti->mti_mdt;
984 struct coordinator *cdt = &mdt->mdt_coordinator;
985 struct hsm_action_item *hai;
989 /* register request in memory list */
990 hai = hai_first(hal);
991 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
992 struct cdt_agent_req *car;
994 /* in case of a cancel request, we first mark the ondisk
995 * record of the request we want to stop as canceled
996 * this does not change the cancel record
997 * it will be done when updating the request status
999 if (hai->hai_action == HSMA_CANCEL) {
1000 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1004 CERROR("%s: mdt_agent_record_update() failed, "
1005 "rc=%d, cannot update status to %s "
1006 "for cookie %#llx\n",
1007 mdt_obd_name(mdt), rc,
1008 agent_req_status2name(ARS_CANCELED),
1013 /* find the running request to set it canceled */
1014 car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1016 car->car_canceled = 1;
1017 /* uuid has to be changed to the one running the
1018 * request to cancel */
1019 *uuid = car->car_uuid;
1020 mdt_cdt_put_request(car);
1022 /* no need to memorize cancel request
1023 * this also avoid a deadlock when we receive
1024 * a purge all requests command
1029 if (hai->hai_action == HSMA_ARCHIVE) {
1030 struct mdt_object *obj;
1033 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1034 if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1037 GOTO(out, rc = PTR_ERR(obj));
1039 hsm.mh_flags |= HS_EXISTS;
1040 hsm.mh_arch_id = hal->hal_archive_id;
1041 rc = mdt_hsm_attr_set(mti, obj, &hsm);
1042 mdt_object_put(mti->mti_env, obj);
1047 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1048 hal->hal_archive_id, hal->hal_flags,
1051 GOTO(out, rc = PTR_ERR(car));
1053 rc = mdt_cdt_add_request(cdt, car);
1055 mdt_cdt_free_request(car);
1062 * swap layouts between 2 fids
1063 * \param mti [IN] context
1066 * \param mh_common [IN] MD HSM
1068 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1069 struct mdt_object *obj, const struct lu_fid *dfid,
1070 struct md_hsm *mh_common)
1072 struct mdt_object *dobj;
1073 struct mdt_lock_handle *dlh;
1077 if (!mdt_object_exists(obj))
1078 GOTO(out, rc = -ENOENT);
1080 /* we already have layout lock on obj so take only
1082 dlh = &mti->mti_lh[MDT_LH_OLD];
1083 mdt_lock_reg_init(dlh, LCK_EX);
1084 dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1086 GOTO(out, rc = PTR_ERR(dobj));
1088 /* if copy tool closes the volatile before sending the final
1089 * progress through llapi_hsm_copy_end(), all the objects
1090 * are removed and mdd_swap_layout LBUG */
1091 if (!mdt_object_exists(dobj)) {
1092 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1093 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1094 GOTO(out_dobj, rc = -ENOENT);
1096 /* Since we only handle restores here, unconditionally use
1097 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1098 * be preserved in case of failure during swap_layout and not
1099 * leave a file in an intermediate but incoherent state.
1100 * But need to setup HSM xattr of data FID before, reuse
1101 * mti and mh presets for FID in hsm_cdt_request_completed(),
1102 * only need to clear RELEASED and DIRTY.
1104 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1105 rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1107 rc = mo_swap_layouts(mti->mti_env,
1108 mdt_object_child(obj),
1109 mdt_object_child(dobj),
1110 SWAP_LAYOUTS_MDS_HSM);
1113 mdt_object_unlock_put(mti, dobj, dlh, 1);
1119 * update status of a completed request
1120 * \param mti [IN] context
1121 * \param pgs [IN] progress of the copy tool
1122 * \param update_record [IN] update llog record
1124 * \retval -ve failure
1126 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1127 struct hsm_progress_kernel *pgs,
1128 const struct cdt_agent_req *car,
1129 enum agent_req_status *status)
1131 const struct lu_env *env = mti->mti_env;
1132 struct mdt_device *mdt = mti->mti_mdt;
1133 struct coordinator *cdt = &mdt->mdt_coordinator;
1134 struct mdt_object *obj = NULL;
1135 int cl_flags = 0, rc = 0;
1140 /* default is to retry */
1141 *status = ARS_WAITING;
1143 /* find object by FID
1144 * if error/removed continue anyway to get correct reporting done */
1145 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1146 /* we will update MD HSM only if needed */
1147 is_mh_changed = false;
1149 /* no need to change mh->mh_arch_id
1150 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1152 if (pgs->hpk_errval != 0) {
1153 switch (pgs->hpk_errval) {
1155 /* the copy tool does not support cancel
1156 * so the cancel request is failed
1157 * As we cannot distinguish a cancel progress
1158 * from another action progress (they have the
1159 * same cookie), we suppose here the CT returns
1160 * ENOSYS only if does not support cancel
1162 /* this can also happen when cdt calls it to
1163 * for a timed out request */
1164 *status = ARS_FAILED;
1165 /* to have a cancel event in changelog */
1166 pgs->hpk_errval = ECANCELED;
1169 /* the request record has already been set to
1170 * ARS_CANCELED, this set the cancel request
1172 *status = ARS_SUCCEED;
1175 /* retry only if current policy or requested, and
1176 * object is not on error/removed */
1177 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1178 !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1179 IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1183 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1184 CERROR("%s: Request %#llx on "DFID
1185 " failed, error code %d too large\n",
1187 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1189 hsm_set_cl_error(&cl_flags,
1190 CLF_HSM_ERROVERFLOW);
1193 hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1196 switch (car->car_hai->hai_action) {
1198 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1201 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1204 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1207 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1208 CERROR("%s: Failed request %#llx on "DFID
1209 " cannot be a CANCEL\n",
1212 PFID(&pgs->hpk_fid));
1215 CERROR("%s: Failed request %#llx on "DFID
1216 " %d is an unknown action\n",
1218 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1219 car->car_hai->hai_action);
1224 *status = ARS_SUCCEED;
1225 switch (car->car_hai->hai_action) {
1227 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1228 /* set ARCHIVE keep EXIST and clear LOST and
1230 mh.mh_arch_ver = pgs->hpk_data_version;
1231 mh.mh_flags |= HS_ARCHIVED;
1232 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1233 is_mh_changed = true;
1236 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1238 /* do not clear RELEASED and DIRTY here
1239 * this will occur in hsm_swap_layouts()
1242 /* Restoring has changed the file version on
1244 mh.mh_arch_ver = pgs->hpk_data_version;
1245 is_mh_changed = true;
1248 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1249 /* clear ARCHIVED EXISTS and LOST */
1250 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1251 is_mh_changed = true;
1254 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1255 CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1258 PFID(&pgs->hpk_fid));
1261 CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1263 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1264 car->car_hai->hai_action);
1270 /* rc != 0 means error when analysing action, it may come from
1271 * a crasy CT no need to manage DIRTY
1274 hsm_set_cl_flags(&cl_flags,
1275 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1277 /* unlock is done later, after layout lock management */
1278 if (is_mh_changed && !IS_ERR(obj))
1279 rc = mdt_hsm_attr_set(mti, obj, &mh);
1281 /* we give back layout lock only if restore was successful or
1282 * if no retry will be attempted and if object is still alive,
1283 * in other cases we just unlock the object */
1284 if (car->car_hai->hai_action == HSMA_RESTORE) {
1285 struct cdt_restore_handle *crh;
1287 /* restore in data FID done, we swap the layouts
1288 * only if restore is successful */
1289 if (pgs->hpk_errval == 0 && !IS_ERR_OR_NULL(obj)) {
1290 rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1293 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1294 *status = ARS_FAILED;
1295 pgs->hpk_errval = -rc;
1298 /* we have to retry, so keep layout lock */
1299 if (*status == ARS_WAITING)
1302 /* give back layout lock */
1303 mutex_lock(&cdt->cdt_restore_lock);
1304 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1306 list_del(&crh->crh_list);
1307 mutex_unlock(&cdt->cdt_restore_lock);
1308 /* Just give back layout lock, we keep the reference
1309 * which is given back later with the lock for HSM
1311 * XXX obj may be invalid so we do not pass it. */
1313 mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
1316 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1322 /* always add a ChangeLog record */
1323 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1324 &car->car_hai->hai_fid);
1327 mdt_object_put(mti->mti_env, obj);
1333 * update status of a request
1334 * \param mti [IN] context
1335 * \param pgs [IN] progress of the copy tool
1336 * \param update_record [IN] update llog record
1338 * \retval -ve failure
1340 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1341 struct hsm_progress_kernel *pgs,
1342 const int update_record)
1344 struct mdt_device *mdt = mti->mti_mdt;
1345 struct coordinator *cdt = &mdt->mdt_coordinator;
1346 struct cdt_agent_req *car;
1350 /* no coordinator started, so we cannot serve requests */
1351 if (cdt->cdt_state == CDT_STOPPED)
1354 /* first do sanity checks */
1355 car = mdt_cdt_update_request(cdt, pgs);
1357 CERROR("%s: Cannot find running request for cookie %#llx"
1360 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1362 RETURN(PTR_ERR(car));
1365 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1366 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1367 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1368 hsm_copytool_action2name(car->car_hai->hai_action),
1369 pgs->hpk_flags, pgs->hpk_errval,
1370 PFID(&car->car_hai->hai_fid),
1371 PFID(&car->car_hai->hai_dfid));
1373 /* progress is done on FID or data FID depending of the action and
1374 * of the copy progress */
1375 /* for restore progress is used to send back the data FID to cdt */
1376 if (car->car_hai->hai_action == HSMA_RESTORE &&
1377 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1378 car->car_hai->hai_dfid = pgs->hpk_fid;
1380 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1381 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1382 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1383 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1384 CERROR("%s: Progress on "DFID" for cookie %#llx"
1385 " does not match request FID "DFID" nor data FID "
1388 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1389 PFID(&car->car_hai->hai_fid),
1390 PFID(&car->car_hai->hai_dfid));
1391 GOTO(out, rc = -EINVAL);
1394 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1395 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1396 " is not coherent (err=%d and not completed"
1399 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1400 hsm_copytool_action2name(car->car_hai->hai_action),
1401 pgs->hpk_errval, pgs->hpk_flags);
1402 GOTO(out, rc = -EINVAL);
1405 /* now progress is valid */
1407 /* we use a root like ucred */
1408 hsm_init_ucred(mdt_ucred(mti));
1410 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1411 enum agent_req_status status;
1413 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1415 /* remove request from memory list */
1416 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1418 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie=%#llx"
1419 " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1421 hsm_copytool_action2name(car->car_hai->hai_action),
1422 agent_req_status2name(status));
1424 if (update_record) {
1427 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1428 &pgs->hpk_cookie, 1,
1431 CERROR("%s: mdt_agent_record_update() failed,"
1432 " rc=%d, cannot update status to %s"
1433 " for cookie %#llx\n",
1434 mdt_obd_name(mdt), rc1,
1435 agent_req_status2name(status),
1437 rc = (rc != 0 ? rc : rc1);
1439 /* ct has completed a request, so a slot is available, wakeup
1440 * cdt to find new work */
1441 mdt_hsm_cdt_wakeup(mdt);
1443 /* if copytool send a progress on a canceled request
1444 * we inform copytool it should stop
1446 if (car->car_canceled == 1)
1452 /* remove ref got from mdt_cdt_update_request() */
1453 mdt_cdt_put_request(car);
1460 * data passed to llog_cat_process() callback
1461 * to cancel requests
1463 struct hsm_cancel_all_data {
1464 struct mdt_device *mdt;
1468 * llog_cat_process() callback, used to:
1469 * - purge all requests
1470 * \param env [IN] environment
1471 * \param llh [IN] llog handle
1472 * \param hdr [IN] llog record
1473 * \param data [IN] cb data = struct hsm_cancel_all_data
1475 * \retval -ve failure
1477 static int mdt_cancel_all_cb(const struct lu_env *env,
1478 struct llog_handle *llh,
1479 struct llog_rec_hdr *hdr, void *data)
1481 struct llog_agent_req_rec *larr;
1482 struct hsm_cancel_all_data *hcad;
1486 larr = (struct llog_agent_req_rec *)hdr;
1488 if (larr->arr_status == ARS_WAITING ||
1489 larr->arr_status == ARS_STARTED) {
1490 larr->arr_status = ARS_CANCELED;
1491 larr->arr_req_change = cfs_time_current_sec();
1492 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1499 * cancel all actions
1500 * \param obd [IN] MDT device
1502 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1505 struct lu_context session;
1506 struct mdt_thread_info *mti;
1507 struct coordinator *cdt = &mdt->mdt_coordinator;
1508 struct cdt_agent_req *car;
1509 struct hsm_action_list *hal = NULL;
1510 struct hsm_action_item *hai;
1511 struct hsm_cancel_all_data hcad;
1512 int hal_sz = 0, hal_len, rc;
1513 enum cdt_states save_state;
1516 rc = lu_env_init(&env, LCT_MD_THREAD);
1520 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1521 rc = lu_context_init(&session, LCT_SERVER_SESSION);
1525 lu_context_enter(&session);
1526 env.le_ses = &session;
1528 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1529 LASSERT(mti != NULL);
1531 mti->mti_env = &env;
1534 hsm_init_ucred(mdt_ucred(mti));
1536 /* disable coordinator */
1537 save_state = cdt->cdt_state;
1538 cdt->cdt_state = CDT_DISABLE;
1540 /* send cancel to all running requests */
1541 down_read(&cdt->cdt_request_lock);
1542 list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1543 mdt_cdt_get_request(car);
1544 /* request is not yet removed from list, it will be done
1545 * when copytool will return progress
1548 if (car->car_hai->hai_action == HSMA_CANCEL) {
1549 mdt_cdt_put_request(car);
1554 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1555 cfs_size_round(car->car_hai->hai_len);
1557 if (hal_len > hal_sz && hal_sz > 0) {
1558 /* not enough room, free old buffer */
1559 OBD_FREE(hal, hal_sz);
1563 /* empty buffer, allocate one */
1566 OBD_ALLOC(hal, hal_sz);
1568 mdt_cdt_put_request(car);
1569 up_read(&cdt->cdt_request_lock);
1570 GOTO(out_cdt_state, rc = -ENOMEM);
1574 hal->hal_version = HAL_VERSION;
1575 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1577 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1578 hal->hal_compound_id = car->car_compound_id;
1579 hal->hal_archive_id = car->car_archive_id;
1580 hal->hal_flags = car->car_flags;
1583 hai = hai_first(hal);
1584 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1585 hai->hai_action = HSMA_CANCEL;
1588 /* it is possible to safely call mdt_hsm_agent_send()
1589 * (ie without a deadlock on cdt_request_lock), because the
1590 * write lock is taken only if we are not in purge mode
1591 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1592 * nor mdt_cdt_remove_request())
1594 /* no conflict with cdt thread because cdt is disable and we
1595 * have the request lock */
1596 mdt_hsm_agent_send(mti, hal, 1);
1598 mdt_cdt_put_request(car);
1600 up_read(&cdt->cdt_request_lock);
1603 OBD_FREE(hal, hal_sz);
1605 /* cancel all on-disk records */
1608 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1609 mdt_cancel_all_cb, &hcad);
1611 /* enable coordinator */
1612 cdt->cdt_state = save_state;
1613 lu_context_exit(&session);
1614 lu_context_fini(&session);
1622 * check if a request is compatible with file status
1623 * \param hai [IN] request description
1624 * \param hal_an [IN] request archive number (not used)
1625 * \param rq_flags [IN] request flags
1626 * \param hsm [IN] file HSM metadata
1629 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1630 const int hal_an, const __u64 rq_flags,
1631 const struct md_hsm *hsm)
1633 int is_compat = false;
1637 hsm_flags = hsm->mh_flags;
1638 switch (hai->hai_action) {
1640 if (!(hsm_flags & HS_NOARCHIVE) &&
1641 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1645 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1646 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1650 if (!(hsm_flags & HS_RELEASED) &&
1651 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1658 CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1659 " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1660 PFID(&hai->hai_fid),
1661 hsm_copytool_action2name(hai->hai_action), rq_flags,
1662 hai->hai_extent.offset, hai->hai_extent.length,
1664 (is_compat ? "compatible" : "uncompatible"));
1670 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1672 static const struct {
1676 } hsm_policy_names[] = {
1677 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1678 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1683 * convert a policy name to a bit
1684 * \param name [IN] policy name
1686 * \retval policy bit
1688 static __u64 hsm_policy_str2bit(const char *name)
1692 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1693 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1694 strcmp(hsm_policy_names[i].name, name) == 0)
1695 return hsm_policy_names[i].bit;
1700 * convert a policy bit field to a string
1701 * \param mask [IN] policy bit field
1702 * \param hexa [IN] print mask before bit names
1703 * \param buffer [OUT] string
1704 * \param count [IN] size of buffer
1706 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1714 seq_printf(m, "(%#llx) ", mask);
1716 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1719 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1720 if (hsm_policy_names[j].bit == bit)
1724 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1726 seq_printf(m, "%s ", hsm_policy_names[j].name);
1728 /* remove last ' ' */
1733 /* methods to read/write HSM policy flags */
1734 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1736 struct mdt_device *mdt = m->private;
1737 struct coordinator *cdt = &mdt->mdt_coordinator;
1740 hsm_policy_bit2str(m, cdt->cdt_policy, false);
1745 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1746 size_t count, loff_t *off)
1748 struct seq_file *m = file->private_data;
1749 struct mdt_device *mdt = m->private;
1750 struct coordinator *cdt = &mdt->mdt_coordinator;
1751 char *start, *token, sign;
1754 __u64 add_mask, remove_mask, set_mask;
1758 if (count + 1 > PAGE_SIZE)
1761 OBD_ALLOC(buf, count + 1);
1765 if (copy_from_user(buf, buffer, count))
1766 GOTO(out, rc = -EFAULT);
1771 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1774 add_mask = remove_mask = set_mask = 0;
1776 token = strsep(&start, "\n ");
1782 if (sign == '-' || sign == '+')
1785 policy = hsm_policy_str2bit(token);
1787 CWARN("%s: '%s' is unknown, "
1788 "supported policies are:\n", mdt_obd_name(mdt),
1790 hsm_policy_bit2str(m, 0, false);
1791 GOTO(out, rc = -EINVAL);
1795 remove_mask |= policy;
1805 } while (start != NULL);
1807 CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
1808 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1810 /* if no sign in all string, it is a clear and set
1811 * if some sign found, all unsigned are converted
1813 * P1 P2 = set to P1 and P2
1814 * P1 -P2 = add P1 clear P2 same as +P1 -P2
1816 if (remove_mask == 0 && add_mask == 0) {
1817 cdt->cdt_policy = set_mask;
1819 cdt->cdt_policy |= set_mask | add_mask;
1820 cdt->cdt_policy &= ~remove_mask;
1823 GOTO(out, rc = count);
1826 OBD_FREE(buf, count + 1);
1829 LPROC_SEQ_FOPS(mdt_hsm_policy);
1831 #define GENERATE_PROC_METHOD(VAR) \
1832 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
1834 struct mdt_device *mdt = m->private; \
1835 struct coordinator *cdt = &mdt->mdt_coordinator; \
1838 seq_printf(m, "%llu\n", (__u64)cdt->VAR); \
1842 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1843 size_t count, loff_t *off) \
1846 struct seq_file *m = file->private_data; \
1847 struct mdt_device *mdt = m->private; \
1848 struct coordinator *cdt = &mdt->mdt_coordinator; \
1853 rc = lprocfs_str_to_s64(buffer, count, &val); \
1856 if (val > 0 && val < INT_MAX) { \
1863 GENERATE_PROC_METHOD(cdt_loop_period)
1864 GENERATE_PROC_METHOD(cdt_grace_delay)
1865 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1866 GENERATE_PROC_METHOD(cdt_max_requests)
1867 GENERATE_PROC_METHOD(cdt_default_archive_id)
1870 * procfs write method for MDT/hsm_control
1871 * proc entry is in mdt directory so data is mdt obd_device pointer
1873 #define CDT_ENABLE_CMD "enabled"
1874 #define CDT_STOP_CMD "shutdown"
1875 #define CDT_DISABLE_CMD "disabled"
1876 #define CDT_PURGE_CMD "purge"
1877 #define CDT_HELP_CMD "help"
1878 #define CDT_MAX_CMD_LEN 10
1881 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1882 size_t count, loff_t *off)
1884 struct seq_file *m = file->private_data;
1885 struct obd_device *obd = m->private;
1886 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1887 struct coordinator *cdt = &(mdt->mdt_coordinator);
1889 char kernbuf[CDT_MAX_CMD_LEN];
1892 if (count == 0 || count >= sizeof(kernbuf))
1895 if (copy_from_user(kernbuf, buffer, count))
1899 if (kernbuf[count - 1] == '\n')
1900 kernbuf[count - 1] = 0;
1903 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1904 if (cdt->cdt_state == CDT_DISABLE) {
1905 cdt->cdt_state = CDT_RUNNING;
1906 mdt_hsm_cdt_wakeup(mdt);
1908 rc = mdt_hsm_cdt_start(mdt);
1910 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1911 if ((cdt->cdt_state == CDT_STOPPING) ||
1912 (cdt->cdt_state == CDT_STOPPED)) {
1913 CERROR("%s: Coordinator already stopped\n",
1917 cdt->cdt_state = CDT_STOPPING;
1919 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
1920 if ((cdt->cdt_state == CDT_STOPPING) ||
1921 (cdt->cdt_state == CDT_STOPPED)) {
1922 CERROR("%s: Coordinator is stopped\n",
1926 cdt->cdt_state = CDT_DISABLE;
1928 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
1929 rc = hsm_cancel_all_actions(mdt);
1930 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
1938 CERROR("%s: Valid coordinator control commands are: "
1939 "%s %s %s %s %s\n", mdt_obd_name(mdt),
1940 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1941 CDT_PURGE_CMD, CDT_HELP_CMD);
1949 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
1951 struct obd_device *obd = m->private;
1952 struct coordinator *cdt;
1955 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
1957 if (cdt->cdt_state == CDT_INIT)
1958 seq_printf(m, "init\n");
1959 else if (cdt->cdt_state == CDT_RUNNING)
1960 seq_printf(m, "enabled\n");
1961 else if (cdt->cdt_state == CDT_STOPPING)
1962 seq_printf(m, "stopping\n");
1963 else if (cdt->cdt_state == CDT_STOPPED)
1964 seq_printf(m, "stopped\n");
1965 else if (cdt->cdt_state == CDT_DISABLE)
1966 seq_printf(m, "disabled\n");
1968 seq_printf(m, "unknown\n");
1974 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
1980 for (i = 0; i < 8 * sizeof(mask); i++) {
1981 if (mask & (1UL << i)) {
1982 seq_printf(m, "%s%s", first ? "" : " ",
1983 hsm_copytool_action2name(i));
1993 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
1995 struct mdt_device *mdt = m->private;
1996 struct coordinator *cdt = &mdt->mdt_coordinator;
1998 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2002 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2004 struct mdt_device *mdt = m->private;
2005 struct coordinator *cdt = &mdt->mdt_coordinator;
2007 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2011 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2013 struct mdt_device *mdt = m->private;
2014 struct coordinator *cdt = &mdt->mdt_coordinator;
2016 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2019 static inline enum hsm_copytool_action
2020 hsm_copytool_name2action(const char *name)
2022 if (strcasecmp(name, "NOOP") == 0)
2024 else if (strcasecmp(name, "ARCHIVE") == 0)
2025 return HSMA_ARCHIVE;
2026 else if (strcasecmp(name, "RESTORE") == 0)
2027 return HSMA_RESTORE;
2028 else if (strcasecmp(name, "REMOVE") == 0)
2030 else if (strcasecmp(name, "CANCEL") == 0)
2037 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2038 size_t user_count, __u64 *mask)
2040 char *buf, *pos, *name;
2046 if (!(user_count < 4096))
2049 buf_size = user_count + 1;
2051 OBD_ALLOC(buf, buf_size);
2055 if (copy_from_user(buf, user_buf, buf_size - 1))
2056 GOTO(out, rc = -EFAULT);
2058 buf[buf_size - 1] = '\0';
2061 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2067 action = hsm_copytool_name2action(name);
2069 GOTO(out, rc = -EINVAL);
2071 new_mask |= (1UL << action);
2077 OBD_FREE(buf, buf_size);
2083 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2084 size_t count, loff_t *off)
2086 struct seq_file *m = file->private_data;
2087 struct mdt_device *mdt = m->private;
2088 struct coordinator *cdt = &mdt->mdt_coordinator;
2090 return mdt_write_hsm_request_mask(file, buf, count,
2091 &cdt->cdt_user_request_mask);
2095 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2096 size_t count, loff_t *off)
2098 struct seq_file *m = file->private_data;
2099 struct mdt_device *mdt = m->private;
2100 struct coordinator *cdt = &mdt->mdt_coordinator;
2102 return mdt_write_hsm_request_mask(file, buf, count,
2103 &cdt->cdt_group_request_mask);
2107 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2108 size_t count, loff_t *off)
2110 struct seq_file *m = file->private_data;
2111 struct mdt_device *mdt = m->private;
2112 struct coordinator *cdt = &mdt->mdt_coordinator;
2114 return mdt_write_hsm_request_mask(file, buf, count,
2115 &cdt->cdt_other_request_mask);
2118 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2119 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2120 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2121 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2122 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2123 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2124 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2125 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2127 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2129 .fops = &mdt_hsm_agent_fops },
2130 { .name = "actions",
2131 .fops = &mdt_hsm_actions_fops,
2132 .proc_mode = 0444 },
2133 { .name = "default_archive_id",
2134 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2135 { .name = "grace_delay",
2136 .fops = &mdt_hsm_cdt_grace_delay_fops },
2137 { .name = "loop_period",
2138 .fops = &mdt_hsm_cdt_loop_period_fops },
2139 { .name = "max_requests",
2140 .fops = &mdt_hsm_cdt_max_requests_fops },
2142 .fops = &mdt_hsm_policy_fops },
2143 { .name = "active_request_timeout",
2144 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2145 { .name = "active_requests",
2146 .fops = &mdt_hsm_active_requests_fops },
2147 { .name = "user_request_mask",
2148 .fops = &mdt_hsm_user_request_mask_fops, },
2149 { .name = "group_request_mask",
2150 .fops = &mdt_hsm_group_request_mask_fops, },
2151 { .name = "other_request_mask",
2152 .fops = &mdt_hsm_other_request_mask_fops, },