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 if (mdt->mdt_bottom->dd_rdonly)
888 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
891 cdt->cdt_state = CDT_STOPPED;
892 CERROR("%s: error starting coordinator thread: %d\n",
893 mdt_obd_name(mdt), rc);
896 CDEBUG(D_HSM, "%s: coordinator thread started\n",
901 wait_event(cdt->cdt_waitq,
902 (cdt->cdt_flags & SVC_RUNNING));
904 cdt->cdt_state = CDT_RUNNING;
905 mdt->mdt_opts.mo_coordinator = 1;
910 * stop a coordinator thread
911 * \param mdt [IN] device
913 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
915 struct coordinator *cdt = &mdt->mdt_coordinator;
916 struct cdt_agent_req *car, *tmp1;
917 struct hsm_agent *ha, *tmp2;
918 struct cdt_restore_handle *crh, *tmp3;
919 struct mdt_thread_info *cdt_mti;
922 if (mdt->mdt_opts.mo_coordinator == 0)
925 if (cdt->cdt_state == CDT_STOPPED) {
926 CERROR("%s: Coordinator already stopped\n",
931 if (cdt->cdt_state != CDT_STOPPING) {
932 /* stop coordinator thread before cleaning */
933 cdt->cdt_flags = SVC_STOPPING;
934 wake_up(&cdt->cdt_waitq);
935 wait_event(cdt->cdt_waitq,
936 cdt->cdt_flags & SVC_STOPPED);
938 cdt->cdt_state = CDT_STOPPED;
941 down_write(&cdt->cdt_request_lock);
942 list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
944 list_del(&car->car_request_list);
945 mdt_cdt_free_request(car);
947 up_write(&cdt->cdt_request_lock);
949 down_write(&cdt->cdt_agent_lock);
950 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
951 list_del(&ha->ha_list);
954 up_write(&cdt->cdt_agent_lock);
956 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
957 mutex_lock(&cdt->cdt_restore_lock);
958 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
959 struct mdt_object *child;
961 /* give back layout lock */
962 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
964 mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
966 list_del(&crh->crh_list);
968 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
970 mutex_unlock(&cdt->cdt_restore_lock);
972 mdt->mdt_opts.mo_coordinator = 0;
978 * register all requests from an hal in the memory list
979 * \param mti [IN] context
980 * \param hal [IN] request
981 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
982 * which is running the CT
984 * \retval -ve failure
986 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
987 struct hsm_action_list *hal, struct obd_uuid *uuid)
989 struct mdt_device *mdt = mti->mti_mdt;
990 struct coordinator *cdt = &mdt->mdt_coordinator;
991 struct hsm_action_item *hai;
995 /* register request in memory list */
996 hai = hai_first(hal);
997 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
998 struct cdt_agent_req *car;
1000 /* in case of a cancel request, we first mark the ondisk
1001 * record of the request we want to stop as canceled
1002 * this does not change the cancel record
1003 * it will be done when updating the request status
1005 if (hai->hai_action == HSMA_CANCEL) {
1006 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1010 CERROR("%s: mdt_agent_record_update() failed, "
1011 "rc=%d, cannot update status to %s "
1012 "for cookie %#llx\n",
1013 mdt_obd_name(mdt), rc,
1014 agent_req_status2name(ARS_CANCELED),
1019 /* find the running request to set it canceled */
1020 car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1022 car->car_canceled = 1;
1023 /* uuid has to be changed to the one running the
1024 * request to cancel */
1025 *uuid = car->car_uuid;
1026 mdt_cdt_put_request(car);
1028 /* no need to memorize cancel request
1029 * this also avoid a deadlock when we receive
1030 * a purge all requests command
1035 if (hai->hai_action == HSMA_ARCHIVE) {
1036 struct mdt_object *obj;
1039 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1040 if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1043 GOTO(out, rc = PTR_ERR(obj));
1045 hsm.mh_flags |= HS_EXISTS;
1046 hsm.mh_arch_id = hal->hal_archive_id;
1047 rc = mdt_hsm_attr_set(mti, obj, &hsm);
1048 mdt_object_put(mti->mti_env, obj);
1053 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1054 hal->hal_archive_id, hal->hal_flags,
1057 GOTO(out, rc = PTR_ERR(car));
1059 rc = mdt_cdt_add_request(cdt, car);
1061 mdt_cdt_free_request(car);
1068 * swap layouts between 2 fids
1069 * \param mti [IN] context
1072 * \param mh_common [IN] MD HSM
1074 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1075 struct mdt_object *obj, const struct lu_fid *dfid,
1076 struct md_hsm *mh_common)
1078 struct mdt_object *dobj;
1079 struct mdt_lock_handle *dlh;
1083 if (!mdt_object_exists(obj))
1084 GOTO(out, rc = -ENOENT);
1086 /* we already have layout lock on obj so take only
1088 dlh = &mti->mti_lh[MDT_LH_OLD];
1089 mdt_lock_reg_init(dlh, LCK_EX);
1090 dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1092 GOTO(out, rc = PTR_ERR(dobj));
1094 /* if copy tool closes the volatile before sending the final
1095 * progress through llapi_hsm_copy_end(), all the objects
1096 * are removed and mdd_swap_layout LBUG */
1097 if (!mdt_object_exists(dobj)) {
1098 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1099 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1100 GOTO(out_dobj, rc = -ENOENT);
1102 /* Since we only handle restores here, unconditionally use
1103 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1104 * be preserved in case of failure during swap_layout and not
1105 * leave a file in an intermediate but incoherent state.
1106 * But need to setup HSM xattr of data FID before, reuse
1107 * mti and mh presets for FID in hsm_cdt_request_completed(),
1108 * only need to clear RELEASED and DIRTY.
1110 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1111 rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1113 rc = mo_swap_layouts(mti->mti_env,
1114 mdt_object_child(obj),
1115 mdt_object_child(dobj),
1116 SWAP_LAYOUTS_MDS_HSM);
1119 mdt_object_unlock_put(mti, dobj, dlh, 1);
1125 * update status of a completed request
1126 * \param mti [IN] context
1127 * \param pgs [IN] progress of the copy tool
1128 * \param update_record [IN] update llog record
1130 * \retval -ve failure
1132 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1133 struct hsm_progress_kernel *pgs,
1134 const struct cdt_agent_req *car,
1135 enum agent_req_status *status)
1137 const struct lu_env *env = mti->mti_env;
1138 struct mdt_device *mdt = mti->mti_mdt;
1139 struct coordinator *cdt = &mdt->mdt_coordinator;
1140 struct mdt_object *obj = NULL;
1141 int cl_flags = 0, rc = 0;
1146 /* default is to retry */
1147 *status = ARS_WAITING;
1149 /* find object by FID
1150 * if error/removed continue anyway to get correct reporting done */
1151 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1152 /* we will update MD HSM only if needed */
1153 is_mh_changed = false;
1155 /* no need to change mh->mh_arch_id
1156 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1158 if (pgs->hpk_errval != 0) {
1159 switch (pgs->hpk_errval) {
1161 /* the copy tool does not support cancel
1162 * so the cancel request is failed
1163 * As we cannot distinguish a cancel progress
1164 * from another action progress (they have the
1165 * same cookie), we suppose here the CT returns
1166 * ENOSYS only if does not support cancel
1168 /* this can also happen when cdt calls it to
1169 * for a timed out request */
1170 *status = ARS_FAILED;
1171 /* to have a cancel event in changelog */
1172 pgs->hpk_errval = ECANCELED;
1175 /* the request record has already been set to
1176 * ARS_CANCELED, this set the cancel request
1178 *status = ARS_SUCCEED;
1181 /* retry only if current policy or requested, and
1182 * object is not on error/removed */
1183 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1184 !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1185 IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1189 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1190 CERROR("%s: Request %#llx on "DFID
1191 " failed, error code %d too large\n",
1193 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1195 hsm_set_cl_error(&cl_flags,
1196 CLF_HSM_ERROVERFLOW);
1199 hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1202 switch (car->car_hai->hai_action) {
1204 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1207 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1210 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1213 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1214 CERROR("%s: Failed request %#llx on "DFID
1215 " cannot be a CANCEL\n",
1218 PFID(&pgs->hpk_fid));
1221 CERROR("%s: Failed request %#llx on "DFID
1222 " %d is an unknown action\n",
1224 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1225 car->car_hai->hai_action);
1230 *status = ARS_SUCCEED;
1231 switch (car->car_hai->hai_action) {
1233 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1234 /* set ARCHIVE keep EXIST and clear LOST and
1236 mh.mh_arch_ver = pgs->hpk_data_version;
1237 mh.mh_flags |= HS_ARCHIVED;
1238 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1239 is_mh_changed = true;
1242 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1244 /* do not clear RELEASED and DIRTY here
1245 * this will occur in hsm_swap_layouts()
1248 /* Restoring has changed the file version on
1250 mh.mh_arch_ver = pgs->hpk_data_version;
1251 is_mh_changed = true;
1254 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1255 /* clear ARCHIVED EXISTS and LOST */
1256 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1257 is_mh_changed = true;
1260 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1261 CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1264 PFID(&pgs->hpk_fid));
1267 CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1269 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1270 car->car_hai->hai_action);
1276 /* rc != 0 means error when analysing action, it may come from
1277 * a crasy CT no need to manage DIRTY
1280 hsm_set_cl_flags(&cl_flags,
1281 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1283 /* unlock is done later, after layout lock management */
1284 if (is_mh_changed && !IS_ERR(obj))
1285 rc = mdt_hsm_attr_set(mti, obj, &mh);
1287 /* we give back layout lock only if restore was successful or
1288 * if no retry will be attempted and if object is still alive,
1289 * in other cases we just unlock the object */
1290 if (car->car_hai->hai_action == HSMA_RESTORE) {
1291 struct cdt_restore_handle *crh;
1293 /* restore in data FID done, we swap the layouts
1294 * only if restore is successful */
1295 if (pgs->hpk_errval == 0 && !IS_ERR_OR_NULL(obj)) {
1296 rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1299 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1300 *status = ARS_FAILED;
1301 pgs->hpk_errval = -rc;
1304 /* we have to retry, so keep layout lock */
1305 if (*status == ARS_WAITING)
1308 /* give back layout lock */
1309 mutex_lock(&cdt->cdt_restore_lock);
1310 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1312 list_del(&crh->crh_list);
1313 mutex_unlock(&cdt->cdt_restore_lock);
1314 /* Just give back layout lock, we keep the reference
1315 * which is given back later with the lock for HSM
1317 * XXX obj may be invalid so we do not pass it. */
1319 mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
1322 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1328 /* always add a ChangeLog record */
1329 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1330 &car->car_hai->hai_fid);
1333 mdt_object_put(mti->mti_env, obj);
1339 * update status of a request
1340 * \param mti [IN] context
1341 * \param pgs [IN] progress of the copy tool
1342 * \param update_record [IN] update llog record
1344 * \retval -ve failure
1346 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1347 struct hsm_progress_kernel *pgs,
1348 const int update_record)
1350 struct mdt_device *mdt = mti->mti_mdt;
1351 struct coordinator *cdt = &mdt->mdt_coordinator;
1352 struct cdt_agent_req *car;
1356 /* no coordinator started, so we cannot serve requests */
1357 if (cdt->cdt_state == CDT_STOPPED)
1360 /* first do sanity checks */
1361 car = mdt_cdt_update_request(cdt, pgs);
1363 CERROR("%s: Cannot find running request for cookie %#llx"
1366 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1368 RETURN(PTR_ERR(car));
1371 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1372 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1373 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1374 hsm_copytool_action2name(car->car_hai->hai_action),
1375 pgs->hpk_flags, pgs->hpk_errval,
1376 PFID(&car->car_hai->hai_fid),
1377 PFID(&car->car_hai->hai_dfid));
1379 /* progress is done on FID or data FID depending of the action and
1380 * of the copy progress */
1381 /* for restore progress is used to send back the data FID to cdt */
1382 if (car->car_hai->hai_action == HSMA_RESTORE &&
1383 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1384 car->car_hai->hai_dfid = pgs->hpk_fid;
1386 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1387 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1388 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1389 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1390 CERROR("%s: Progress on "DFID" for cookie %#llx"
1391 " does not match request FID "DFID" nor data FID "
1394 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1395 PFID(&car->car_hai->hai_fid),
1396 PFID(&car->car_hai->hai_dfid));
1397 GOTO(out, rc = -EINVAL);
1400 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1401 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1402 " is not coherent (err=%d and not completed"
1405 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1406 hsm_copytool_action2name(car->car_hai->hai_action),
1407 pgs->hpk_errval, pgs->hpk_flags);
1408 GOTO(out, rc = -EINVAL);
1411 /* now progress is valid */
1413 /* we use a root like ucred */
1414 hsm_init_ucred(mdt_ucred(mti));
1416 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1417 enum agent_req_status status;
1419 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1421 /* remove request from memory list */
1422 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1424 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie=%#llx"
1425 " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1427 hsm_copytool_action2name(car->car_hai->hai_action),
1428 agent_req_status2name(status));
1430 if (update_record) {
1433 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1434 &pgs->hpk_cookie, 1,
1437 CERROR("%s: mdt_agent_record_update() failed,"
1438 " rc=%d, cannot update status to %s"
1439 " for cookie %#llx\n",
1440 mdt_obd_name(mdt), rc1,
1441 agent_req_status2name(status),
1443 rc = (rc != 0 ? rc : rc1);
1445 /* ct has completed a request, so a slot is available, wakeup
1446 * cdt to find new work */
1447 mdt_hsm_cdt_wakeup(mdt);
1449 /* if copytool send a progress on a canceled request
1450 * we inform copytool it should stop
1452 if (car->car_canceled == 1)
1458 /* remove ref got from mdt_cdt_update_request() */
1459 mdt_cdt_put_request(car);
1466 * data passed to llog_cat_process() callback
1467 * to cancel requests
1469 struct hsm_cancel_all_data {
1470 struct mdt_device *mdt;
1474 * llog_cat_process() callback, used to:
1475 * - purge all requests
1476 * \param env [IN] environment
1477 * \param llh [IN] llog handle
1478 * \param hdr [IN] llog record
1479 * \param data [IN] cb data = struct hsm_cancel_all_data
1481 * \retval -ve failure
1483 static int mdt_cancel_all_cb(const struct lu_env *env,
1484 struct llog_handle *llh,
1485 struct llog_rec_hdr *hdr, void *data)
1487 struct llog_agent_req_rec *larr;
1488 struct hsm_cancel_all_data *hcad;
1492 larr = (struct llog_agent_req_rec *)hdr;
1494 if (larr->arr_status == ARS_WAITING ||
1495 larr->arr_status == ARS_STARTED) {
1496 larr->arr_status = ARS_CANCELED;
1497 larr->arr_req_change = cfs_time_current_sec();
1498 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1505 * cancel all actions
1506 * \param obd [IN] MDT device
1508 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1511 struct lu_context session;
1512 struct mdt_thread_info *mti;
1513 struct coordinator *cdt = &mdt->mdt_coordinator;
1514 struct cdt_agent_req *car;
1515 struct hsm_action_list *hal = NULL;
1516 struct hsm_action_item *hai;
1517 struct hsm_cancel_all_data hcad;
1518 int hal_sz = 0, hal_len, rc;
1519 enum cdt_states save_state;
1522 rc = lu_env_init(&env, LCT_MD_THREAD);
1526 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1527 rc = lu_context_init(&session, LCT_SERVER_SESSION);
1531 lu_context_enter(&session);
1532 env.le_ses = &session;
1534 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1535 LASSERT(mti != NULL);
1537 mti->mti_env = &env;
1540 hsm_init_ucred(mdt_ucred(mti));
1542 /* disable coordinator */
1543 save_state = cdt->cdt_state;
1544 cdt->cdt_state = CDT_DISABLE;
1546 /* send cancel to all running requests */
1547 down_read(&cdt->cdt_request_lock);
1548 list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1549 mdt_cdt_get_request(car);
1550 /* request is not yet removed from list, it will be done
1551 * when copytool will return progress
1554 if (car->car_hai->hai_action == HSMA_CANCEL) {
1555 mdt_cdt_put_request(car);
1560 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1561 cfs_size_round(car->car_hai->hai_len);
1563 if (hal_len > hal_sz && hal_sz > 0) {
1564 /* not enough room, free old buffer */
1565 OBD_FREE(hal, hal_sz);
1569 /* empty buffer, allocate one */
1572 OBD_ALLOC(hal, hal_sz);
1574 mdt_cdt_put_request(car);
1575 up_read(&cdt->cdt_request_lock);
1576 GOTO(out_cdt_state, rc = -ENOMEM);
1580 hal->hal_version = HAL_VERSION;
1581 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1583 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1584 hal->hal_compound_id = car->car_compound_id;
1585 hal->hal_archive_id = car->car_archive_id;
1586 hal->hal_flags = car->car_flags;
1589 hai = hai_first(hal);
1590 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1591 hai->hai_action = HSMA_CANCEL;
1594 /* it is possible to safely call mdt_hsm_agent_send()
1595 * (ie without a deadlock on cdt_request_lock), because the
1596 * write lock is taken only if we are not in purge mode
1597 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1598 * nor mdt_cdt_remove_request())
1600 /* no conflict with cdt thread because cdt is disable and we
1601 * have the request lock */
1602 mdt_hsm_agent_send(mti, hal, 1);
1604 mdt_cdt_put_request(car);
1606 up_read(&cdt->cdt_request_lock);
1609 OBD_FREE(hal, hal_sz);
1611 /* cancel all on-disk records */
1614 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1615 mdt_cancel_all_cb, &hcad);
1617 /* enable coordinator */
1618 cdt->cdt_state = save_state;
1619 lu_context_exit(&session);
1620 lu_context_fini(&session);
1628 * check if a request is compatible with file status
1629 * \param hai [IN] request description
1630 * \param hal_an [IN] request archive number (not used)
1631 * \param rq_flags [IN] request flags
1632 * \param hsm [IN] file HSM metadata
1635 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1636 const int hal_an, const __u64 rq_flags,
1637 const struct md_hsm *hsm)
1639 int is_compat = false;
1643 hsm_flags = hsm->mh_flags;
1644 switch (hai->hai_action) {
1646 if (!(hsm_flags & HS_NOARCHIVE) &&
1647 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1651 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1652 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1656 if (!(hsm_flags & HS_RELEASED) &&
1657 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1664 CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1665 " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1666 PFID(&hai->hai_fid),
1667 hsm_copytool_action2name(hai->hai_action), rq_flags,
1668 hai->hai_extent.offset, hai->hai_extent.length,
1670 (is_compat ? "compatible" : "uncompatible"));
1676 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1678 static const struct {
1682 } hsm_policy_names[] = {
1683 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1684 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1689 * convert a policy name to a bit
1690 * \param name [IN] policy name
1692 * \retval policy bit
1694 static __u64 hsm_policy_str2bit(const char *name)
1698 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1699 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1700 strcmp(hsm_policy_names[i].name, name) == 0)
1701 return hsm_policy_names[i].bit;
1706 * convert a policy bit field to a string
1707 * \param mask [IN] policy bit field
1708 * \param hexa [IN] print mask before bit names
1709 * \param buffer [OUT] string
1710 * \param count [IN] size of buffer
1712 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1720 seq_printf(m, "(%#llx) ", mask);
1722 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1725 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1726 if (hsm_policy_names[j].bit == bit)
1730 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1732 seq_printf(m, "%s ", hsm_policy_names[j].name);
1734 /* remove last ' ' */
1739 /* methods to read/write HSM policy flags */
1740 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1742 struct mdt_device *mdt = m->private;
1743 struct coordinator *cdt = &mdt->mdt_coordinator;
1746 hsm_policy_bit2str(m, cdt->cdt_policy, false);
1751 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1752 size_t count, loff_t *off)
1754 struct seq_file *m = file->private_data;
1755 struct mdt_device *mdt = m->private;
1756 struct coordinator *cdt = &mdt->mdt_coordinator;
1757 char *start, *token, sign;
1760 __u64 add_mask, remove_mask, set_mask;
1764 if (count + 1 > PAGE_SIZE)
1767 OBD_ALLOC(buf, count + 1);
1771 if (copy_from_user(buf, buffer, count))
1772 GOTO(out, rc = -EFAULT);
1777 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1780 add_mask = remove_mask = set_mask = 0;
1782 token = strsep(&start, "\n ");
1788 if (sign == '-' || sign == '+')
1791 policy = hsm_policy_str2bit(token);
1793 CWARN("%s: '%s' is unknown, "
1794 "supported policies are:\n", mdt_obd_name(mdt),
1796 hsm_policy_bit2str(m, 0, false);
1797 GOTO(out, rc = -EINVAL);
1801 remove_mask |= policy;
1811 } while (start != NULL);
1813 CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
1814 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1816 /* if no sign in all string, it is a clear and set
1817 * if some sign found, all unsigned are converted
1819 * P1 P2 = set to P1 and P2
1820 * P1 -P2 = add P1 clear P2 same as +P1 -P2
1822 if (remove_mask == 0 && add_mask == 0) {
1823 cdt->cdt_policy = set_mask;
1825 cdt->cdt_policy |= set_mask | add_mask;
1826 cdt->cdt_policy &= ~remove_mask;
1829 GOTO(out, rc = count);
1832 OBD_FREE(buf, count + 1);
1835 LPROC_SEQ_FOPS(mdt_hsm_policy);
1837 #define GENERATE_PROC_METHOD(VAR) \
1838 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
1840 struct mdt_device *mdt = m->private; \
1841 struct coordinator *cdt = &mdt->mdt_coordinator; \
1844 seq_printf(m, "%llu\n", (__u64)cdt->VAR); \
1848 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1849 size_t count, loff_t *off) \
1852 struct seq_file *m = file->private_data; \
1853 struct mdt_device *mdt = m->private; \
1854 struct coordinator *cdt = &mdt->mdt_coordinator; \
1859 rc = lprocfs_str_to_s64(buffer, count, &val); \
1862 if (val > 0 && val < INT_MAX) { \
1869 GENERATE_PROC_METHOD(cdt_loop_period)
1870 GENERATE_PROC_METHOD(cdt_grace_delay)
1871 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1872 GENERATE_PROC_METHOD(cdt_max_requests)
1873 GENERATE_PROC_METHOD(cdt_default_archive_id)
1876 * procfs write method for MDT/hsm_control
1877 * proc entry is in mdt directory so data is mdt obd_device pointer
1879 #define CDT_ENABLE_CMD "enabled"
1880 #define CDT_STOP_CMD "shutdown"
1881 #define CDT_DISABLE_CMD "disabled"
1882 #define CDT_PURGE_CMD "purge"
1883 #define CDT_HELP_CMD "help"
1884 #define CDT_MAX_CMD_LEN 10
1887 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1888 size_t count, loff_t *off)
1890 struct seq_file *m = file->private_data;
1891 struct obd_device *obd = m->private;
1892 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1893 struct coordinator *cdt = &(mdt->mdt_coordinator);
1895 char kernbuf[CDT_MAX_CMD_LEN];
1898 if (count == 0 || count >= sizeof(kernbuf))
1901 if (copy_from_user(kernbuf, buffer, count))
1905 if (kernbuf[count - 1] == '\n')
1906 kernbuf[count - 1] = 0;
1909 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1910 if (cdt->cdt_state == CDT_DISABLE) {
1911 cdt->cdt_state = CDT_RUNNING;
1912 mdt_hsm_cdt_wakeup(mdt);
1914 rc = mdt_hsm_cdt_start(mdt);
1916 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1917 if ((cdt->cdt_state == CDT_STOPPING) ||
1918 (cdt->cdt_state == CDT_STOPPED)) {
1919 CERROR("%s: Coordinator already stopped\n",
1923 cdt->cdt_state = CDT_STOPPING;
1925 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
1926 if ((cdt->cdt_state == CDT_STOPPING) ||
1927 (cdt->cdt_state == CDT_STOPPED)) {
1928 CERROR("%s: Coordinator is stopped\n",
1932 cdt->cdt_state = CDT_DISABLE;
1934 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
1935 rc = hsm_cancel_all_actions(mdt);
1936 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
1944 CERROR("%s: Valid coordinator control commands are: "
1945 "%s %s %s %s %s\n", mdt_obd_name(mdt),
1946 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1947 CDT_PURGE_CMD, CDT_HELP_CMD);
1955 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
1957 struct obd_device *obd = m->private;
1958 struct coordinator *cdt;
1961 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
1963 if (cdt->cdt_state == CDT_INIT)
1964 seq_printf(m, "init\n");
1965 else if (cdt->cdt_state == CDT_RUNNING)
1966 seq_printf(m, "enabled\n");
1967 else if (cdt->cdt_state == CDT_STOPPING)
1968 seq_printf(m, "stopping\n");
1969 else if (cdt->cdt_state == CDT_STOPPED)
1970 seq_printf(m, "stopped\n");
1971 else if (cdt->cdt_state == CDT_DISABLE)
1972 seq_printf(m, "disabled\n");
1974 seq_printf(m, "unknown\n");
1980 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
1986 for (i = 0; i < 8 * sizeof(mask); i++) {
1987 if (mask & (1UL << i)) {
1988 seq_printf(m, "%s%s", first ? "" : " ",
1989 hsm_copytool_action2name(i));
1999 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2001 struct mdt_device *mdt = m->private;
2002 struct coordinator *cdt = &mdt->mdt_coordinator;
2004 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2008 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2010 struct mdt_device *mdt = m->private;
2011 struct coordinator *cdt = &mdt->mdt_coordinator;
2013 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2017 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2019 struct mdt_device *mdt = m->private;
2020 struct coordinator *cdt = &mdt->mdt_coordinator;
2022 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2025 static inline enum hsm_copytool_action
2026 hsm_copytool_name2action(const char *name)
2028 if (strcasecmp(name, "NOOP") == 0)
2030 else if (strcasecmp(name, "ARCHIVE") == 0)
2031 return HSMA_ARCHIVE;
2032 else if (strcasecmp(name, "RESTORE") == 0)
2033 return HSMA_RESTORE;
2034 else if (strcasecmp(name, "REMOVE") == 0)
2036 else if (strcasecmp(name, "CANCEL") == 0)
2043 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2044 size_t user_count, __u64 *mask)
2046 char *buf, *pos, *name;
2052 if (!(user_count < 4096))
2055 buf_size = user_count + 1;
2057 OBD_ALLOC(buf, buf_size);
2061 if (copy_from_user(buf, user_buf, buf_size - 1))
2062 GOTO(out, rc = -EFAULT);
2064 buf[buf_size - 1] = '\0';
2067 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2073 action = hsm_copytool_name2action(name);
2075 GOTO(out, rc = -EINVAL);
2077 new_mask |= (1UL << action);
2083 OBD_FREE(buf, buf_size);
2089 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2090 size_t count, loff_t *off)
2092 struct seq_file *m = file->private_data;
2093 struct mdt_device *mdt = m->private;
2094 struct coordinator *cdt = &mdt->mdt_coordinator;
2096 return mdt_write_hsm_request_mask(file, buf, count,
2097 &cdt->cdt_user_request_mask);
2101 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2102 size_t count, loff_t *off)
2104 struct seq_file *m = file->private_data;
2105 struct mdt_device *mdt = m->private;
2106 struct coordinator *cdt = &mdt->mdt_coordinator;
2108 return mdt_write_hsm_request_mask(file, buf, count,
2109 &cdt->cdt_group_request_mask);
2113 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2114 size_t count, loff_t *off)
2116 struct seq_file *m = file->private_data;
2117 struct mdt_device *mdt = m->private;
2118 struct coordinator *cdt = &mdt->mdt_coordinator;
2120 return mdt_write_hsm_request_mask(file, buf, count,
2121 &cdt->cdt_other_request_mask);
2124 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2125 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2126 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2127 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2128 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2129 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2130 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2131 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2133 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2135 .fops = &mdt_hsm_agent_fops },
2136 { .name = "actions",
2137 .fops = &mdt_hsm_actions_fops,
2138 .proc_mode = 0444 },
2139 { .name = "default_archive_id",
2140 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2141 { .name = "grace_delay",
2142 .fops = &mdt_hsm_cdt_grace_delay_fops },
2143 { .name = "loop_period",
2144 .fops = &mdt_hsm_cdt_loop_period_fops },
2145 { .name = "max_requests",
2146 .fops = &mdt_hsm_cdt_max_requests_fops },
2148 .fops = &mdt_hsm_policy_fops },
2149 { .name = "active_request_timeout",
2150 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2151 { .name = "active_requests",
2152 .fops = &mdt_hsm_active_requests_fops },
2153 { .name = "user_request_mask",
2154 .fops = &mdt_hsm_user_request_mask_fops, },
2155 { .name = "group_request_mask",
2156 .fops = &mdt_hsm_group_request_mask_fops, },
2157 { .name = "other_request_mask",
2158 .fops = &mdt_hsm_other_request_mask_fops, },