4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
26 * Copyright (c) 2013, 2014, Intel Corporation.
27 * Use is subject to license terms.
30 * lustre/mdt/mdt_coordinator.c
32 * Lustre HSM Coordinator
34 * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
35 * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
36 * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <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 "LPX64
106 " archive_id %d flags "LPX64"\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="LPX64"/"LPX64
115 " action=%s extent="LPX64"-"LPX64" gid="LPX64
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_data {
134 struct mdt_thread_info *mti;
135 char fs_name[MTI_NAME_MAXLEN+1];
136 /* request to be send to agents */
137 int request_sz; /** allocated size */
138 int max_requests; /** vector size */
139 int request_cnt; /** used count */
143 struct hsm_action_list *hal;
148 * llog_cat_process() callback, used to:
149 * - find waiting request and start action
150 * - purge canceled and done requests
151 * \param env [IN] environment
152 * \param llh [IN] llog handle
153 * \param hdr [IN] llog record
154 * \param data [IN/OUT] cb data = struct hsm_scan_data
156 * \retval -ve failure
158 static int mdt_coordinator_cb(const struct lu_env *env,
159 struct llog_handle *llh,
160 struct llog_rec_hdr *hdr,
163 struct llog_agent_req_rec *larr;
164 struct hsm_scan_data *hsd;
165 struct hsm_action_item *hai;
166 struct mdt_device *mdt;
167 struct coordinator *cdt;
172 mdt = hsd->mti->mti_mdt;
173 cdt = &mdt->mdt_coordinator;
175 larr = (struct llog_agent_req_rec *)hdr;
176 dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
177 switch (larr->arr_status) {
179 int i, empty_slot, found;
181 /* Are agents full? */
182 if (atomic_read(&cdt->cdt_request_count) ==
183 cdt->cdt_max_requests)
186 /* first search whether the request is found in the list we
187 * have built and if there is room in the request vector */
190 for (i = 0; i < hsd->max_requests &&
191 (empty_slot == -1 || found == -1); i++) {
192 if (hsd->request[i].hal == NULL) {
196 if (hsd->request[i].hal->hal_compound_id ==
197 larr->arr_compound_id) {
202 if (found == -1 && empty_slot == -1)
203 /* unknown request and no more room for new request,
204 * continue scan for to find other entries for
205 * already found request
210 struct hsm_action_list *hal;
212 /* request is not already known */
213 /* allocates hai vector size just needs to be large
215 hsd->request[empty_slot].hal_sz =
216 sizeof(*hsd->request[empty_slot].hal) +
217 cfs_size_round(MTI_NAME_MAXLEN+1) +
218 2 * cfs_size_round(larr->arr_hai.hai_len);
219 OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
221 CERROR("%s: Cannot allocate memory (%d o)"
222 "for compound "LPX64"\n",
224 hsd->request[i].hal_sz,
225 larr->arr_compound_id);
228 hal->hal_version = HAL_VERSION;
229 strlcpy(hal->hal_fsname, hsd->fs_name,
230 MTI_NAME_MAXLEN + 1);
231 hal->hal_compound_id = larr->arr_compound_id;
232 hal->hal_archive_id = larr->arr_archive_id;
233 hal->hal_flags = larr->arr_flags;
235 hsd->request[empty_slot].hal_used_sz = hal_size(hal);
236 hsd->request[empty_slot].hal = hal;
239 hai = hai_first(hal);
241 /* request is known */
242 /* we check if record archive num is the same as the
243 * known request, if not we will serve it in multiple
244 * time because we do not know if the agent can serve
246 * a use case is a compound made of multiple restore
247 * where the files are not archived in the same backend
249 if (larr->arr_archive_id !=
250 hsd->request[found].hal->hal_archive_id)
253 if (hsd->request[found].hal_sz <
254 hsd->request[found].hal_used_sz +
255 cfs_size_round(larr->arr_hai.hai_len)) {
256 /* Not enough room, need an extension */
260 sz = 2 * hsd->request[found].hal_sz;
261 OBD_ALLOC(hal_buffer, sz);
263 CERROR("%s: Cannot allocate memory "
264 "(%d o) for compound "LPX64"\n",
265 mdt_obd_name(mdt), sz,
266 larr->arr_compound_id);
269 memcpy(hal_buffer, hsd->request[found].hal,
270 hsd->request[found].hal_used_sz);
271 OBD_FREE(hsd->request[found].hal,
272 hsd->request[found].hal_sz);
273 hsd->request[found].hal = hal_buffer;
274 hsd->request[found].hal_sz = sz;
276 hai = hai_first(hsd->request[found].hal);
277 for (i = 0; i < hsd->request[found].hal->hal_count;
281 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
282 hai->hai_cookie = larr->arr_hai.hai_cookie;
283 hai->hai_gid = larr->arr_hai.hai_gid;
285 hsd->request[found].hal_used_sz +=
286 cfs_size_round(hai->hai_len);
287 hsd->request[found].hal->hal_count++;
291 struct hsm_progress_kernel pgs;
292 struct cdt_agent_req *car;
293 cfs_time_t now = cfs_time_current_sec();
296 /* we search for a running request
297 * error may happen if coordinator crashes or stopped
298 * with running request
300 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
302 last = larr->arr_req_create;
304 last = car->car_req_update;
305 mdt_cdt_put_request(car);
308 /* test if request too long, if yes cancel it
309 * the same way the copy tool acknowledge a cancel request */
310 if (now <= last + cdt->cdt_active_req_timeout)
313 dump_llog_agent_req_rec("request timed out, start cleaning",
315 /* a too old cancel request just needs to be removed
316 * this can happen, if copy tool does not support
317 * cancel for other requests, we have to remove the
318 * running request and notify the copytool */
319 pgs.hpk_fid = larr->arr_hai.hai_fid;
320 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
321 pgs.hpk_extent = larr->arr_hai.hai_extent;
322 pgs.hpk_flags = HP_FLAG_COMPLETED;
323 pgs.hpk_errval = ENOSYS;
324 pgs.hpk_data_version = 0;
326 /* update request state, but do not record in llog, to
327 * avoid deadlock on cdt_llog_lock */
328 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
330 CERROR("%s: cannot cleanup timed out request: "
331 DFID" for cookie "LPX64" action=%s\n",
333 PFID(&pgs.hpk_fid), pgs.hpk_cookie,
334 hsm_copytool_action2name(
335 larr->arr_hai.hai_action));
338 /* The request no longer exists, forget
339 * about it, and do not send a cancel request
340 * to the client, for which an error will be
341 * sent back, leading to an endless cycle of
343 RETURN(LLOG_DEL_RECORD);
346 /* XXX A cancel request cannot be cancelled. */
347 if (larr->arr_hai.hai_action == HSMA_CANCEL)
350 larr->arr_status = ARS_CANCELED;
351 larr->arr_req_change = now;
352 rc = llog_write(hsd->mti->mti_env, llh, hdr, hdr->lrh_index);
354 CERROR("%s: cannot update agent log: rc = %d\n",
355 mdt_obd_name(mdt), rc);
361 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
362 cfs_time_current_sec())
363 RETURN(LLOG_DEL_RECORD);
370 * create /proc entries for coordinator
373 * \retval -ve failure
375 int hsm_cdt_procfs_init(struct mdt_device *mdt)
377 struct coordinator *cdt = &mdt->mdt_coordinator;
381 /* init /proc entries, failure is not critical */
382 cdt->cdt_proc_dir = lprocfs_register("hsm",
383 mdt2obd_dev(mdt)->obd_proc_entry,
384 lprocfs_mdt_hsm_vars, mdt);
385 if (IS_ERR(cdt->cdt_proc_dir)) {
386 rc = PTR_ERR(cdt->cdt_proc_dir);
387 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
388 " rc=%d\n", mdt_obd_name(mdt), rc);
389 cdt->cdt_proc_dir = NULL;
397 * remove /proc entries for coordinator
400 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
402 struct coordinator *cdt = &mdt->mdt_coordinator;
404 LASSERT(cdt->cdt_state == CDT_STOPPED);
405 if (cdt->cdt_proc_dir != NULL)
406 lprocfs_remove(&cdt->cdt_proc_dir);
410 * get vector of hsm cdt /proc vars
414 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
416 return lprocfs_mdt_hsm_vars;
421 * \param data [IN] obd device
423 * \retval -ve failure
425 static int mdt_coordinator(void *data)
427 struct mdt_thread_info *mti = data;
428 struct mdt_device *mdt = mti->mti_mdt;
429 struct coordinator *cdt = &mdt->mdt_coordinator;
430 struct hsm_scan_data hsd = { NULL };
434 cdt->cdt_thread.t_flags = SVC_RUNNING;
435 wake_up(&cdt->cdt_thread.t_ctl_waitq);
437 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
438 mdt_obd_name(mdt), current_pid());
440 /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
441 * increases due to a change from /proc we do not overflow the
442 * hsd.request[] vector
444 hsd.max_requests = cdt->cdt_max_requests;
445 hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
446 OBD_ALLOC(hsd.request, hsd.request_sz);
448 GOTO(out, rc = -ENOMEM);
451 obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
454 struct l_wait_info lwi;
457 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
459 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
460 (cdt->cdt_thread.t_flags &
461 (SVC_STOPPING|SVC_EVENT)),
464 CDEBUG(D_HSM, "coordinator resumes\n");
466 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
467 cdt->cdt_state == CDT_STOPPING) {
468 cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
473 /* wake up before timeout, new work arrives */
474 if (cdt->cdt_thread.t_flags & SVC_EVENT)
475 cdt->cdt_thread.t_flags &= ~SVC_EVENT;
477 /* if coordinator is suspended continue to wait */
478 if (cdt->cdt_state == CDT_DISABLE) {
479 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
483 CDEBUG(D_HSM, "coordinator starts reading llog\n");
485 if (hsd.max_requests != cdt->cdt_max_requests) {
486 /* cdt_max_requests has changed,
487 * we need to allocate a new buffer
489 OBD_FREE(hsd.request, hsd.request_sz);
490 hsd.max_requests = cdt->cdt_max_requests;
492 hsd.max_requests * sizeof(*hsd.request);
493 OBD_ALLOC(hsd.request, hsd.request_sz);
502 rc = cdt_llog_process(mti->mti_env, mdt,
503 mdt_coordinator_cb, &hsd);
507 CDEBUG(D_HSM, "found %d requests to send\n", hsd.request_cnt);
509 if (list_empty(&cdt->cdt_agents)) {
510 CDEBUG(D_HSM, "no agent available, "
511 "coordinator sleeps\n");
515 /* here hsd contains a list of requests to be started */
516 for (i = 0; i < hsd.max_requests; i++) {
517 struct hsm_action_list *hal = hsd.request[i].hal;
518 struct hsm_action_item *hai;
521 enum agent_req_status status;
523 /* still room for work ? */
524 if (atomic_read(&cdt->cdt_request_count) ==
525 cdt->cdt_max_requests)
531 /* found a request, we start it */
532 rc = mdt_hsm_agent_send(mti, hal, 0);
533 /* if failure, we suppose it is temporary
534 * if the copy tool failed to do the request
535 * it has to use hsm_progress
537 status = (rc ? ARS_WAITING : ARS_STARTED);
539 /* set up cookie vector to set records status
540 * after copy tools start or failed
542 sz = hal->hal_count * sizeof(__u64);
543 OBD_ALLOC(cookies, sz);
544 if (cookies == NULL) {
545 CERROR("%s: Cannot allocate memory (%d o) "
546 "for cookies vector "LPX64"\n",
547 mdt_obd_name(mdt), sz,
548 hal->hal_compound_id);
551 hai = hai_first(hal);
552 for (j = 0; j < hal->hal_count; j++) {
553 cookies[j] = hai->hai_cookie;
557 rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
558 hal->hal_count, status);
560 CERROR("%s: mdt_agent_record_update() failed, "
561 "rc=%d, cannot update status to %s "
563 mdt_obd_name(mdt), rc,
564 agent_req_status2name(status),
567 OBD_FREE(cookies, sz);
570 /* free hal allocated by callback */
571 for (i = 0; i < hsd.max_requests; i++) {
572 if (hsd.request[i].hal) {
573 OBD_FREE(hsd.request[i].hal,
574 hsd.request[i].hal_sz);
575 hsd.request[i].hal_sz = 0;
576 hsd.request[i].hal = NULL;
580 LASSERT(hsd.request_cnt == 0);
582 /* reset callback data */
583 memset(hsd.request, 0, hsd.request_sz);
588 OBD_FREE(hsd.request, hsd.request_sz);
590 if (cdt->cdt_state == CDT_STOPPING) {
591 /* request comes from /proc path, so we need to clean cdt
593 mdt_hsm_cdt_stop(mdt);
594 mdt->mdt_opts.mo_coordinator = 0;
596 /* request comes from a thread event, generated
597 * by mdt_stop_coordinator(), we have to ack
598 * and cdt cleaning will be done by event sender
600 cdt->cdt_thread.t_flags = SVC_STOPPED;
601 wake_up(&cdt->cdt_thread.t_ctl_waitq);
605 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
606 mdt_obd_name(mdt), current_pid(), rc);
608 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
610 mdt_obd_name(mdt), current_pid());
616 * lookup a restore handle by FID
617 * caller needs to hold cdt_restore_lock
618 * \param cdt [IN] coordinator
619 * \param fid [IN] FID
620 * \retval cdt_restore_handle found
621 * \retval NULL not found
623 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
624 const struct lu_fid *fid)
626 struct cdt_restore_handle *crh;
629 list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
630 if (lu_fid_eq(&crh->crh_fid, fid))
637 * data passed to llog_cat_process() callback
638 * to scan requests and take actions
640 struct hsm_restore_data {
641 struct mdt_thread_info *hrd_mti;
645 * llog_cat_process() callback, used to:
646 * - find restore request and allocate the restore handle
647 * \param env [IN] environment
648 * \param llh [IN] llog handle
649 * \param hdr [IN] llog record
650 * \param data [IN/OUT] cb data = struct hsm_restore_data
652 * \retval -ve failure
654 static int hsm_restore_cb(const struct lu_env *env,
655 struct llog_handle *llh,
656 struct llog_rec_hdr *hdr, void *data)
658 struct llog_agent_req_rec *larr;
659 struct hsm_restore_data *hrd;
660 struct cdt_restore_handle *crh;
661 struct hsm_action_item *hai;
662 struct mdt_thread_info *mti;
663 struct coordinator *cdt;
664 struct mdt_object *child;
670 cdt = &mti->mti_mdt->mdt_coordinator;
672 larr = (struct llog_agent_req_rec *)hdr;
673 hai = &larr->arr_hai;
674 if (hai->hai_cookie > cdt->cdt_last_cookie)
675 /* update the cookie to avoid collision */
676 cdt->cdt_last_cookie = hai->hai_cookie + 1;
678 if (hai->hai_action != HSMA_RESTORE ||
679 agent_req_in_final_state(larr->arr_status))
682 /* restore request not in a final state */
684 /* force replay of restore requests left in started state from previous
685 * CDT context, to be canceled later if finally found to be incompatible
686 * when being re-started */
687 if (larr->arr_status == ARS_STARTED) {
688 larr->arr_status = ARS_WAITING;
689 larr->arr_req_change = cfs_time_current_sec();
690 rc = llog_write(env, llh, hdr, hdr->lrh_index);
695 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
699 crh->crh_fid = hai->hai_fid;
700 /* in V1 all file is restored
701 crh->extent.start = hai->hai_extent.offset;
702 crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
704 crh->crh_extent.start = 0;
705 crh->crh_extent.end = hai->hai_extent.length;
706 /* get the layout lock */
707 mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
708 child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
709 MDS_INODELOCK_LAYOUT);
711 GOTO(out, rc = PTR_ERR(child));
714 /* we choose to not keep a reference
715 * on the object during the restore time which can be very long */
716 mdt_object_put(mti->mti_env, child);
718 mutex_lock(&cdt->cdt_restore_lock);
719 list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
720 mutex_unlock(&cdt->cdt_restore_lock);
727 * restore coordinator state at startup
728 * the goal is to take a layout lock for each registered restore request
729 * \param mti [IN] context
731 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
733 struct hsm_restore_data hrd;
739 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
740 hsm_restore_cb, &hrd);
745 static int hsm_init_ucred(struct lu_ucred *uc)
749 uc->uc_valid = UCRED_OLD;
758 uc->uc_suppgids[0] = -1;
759 uc->uc_suppgids[1] = -1;
760 uc->uc_cap = CFS_CAP_FS_MASK;
763 uc->uc_identity = NULL;
769 * wake up coordinator thread
770 * \param mdt [IN] device
772 * \retval -ve failure
774 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
776 struct coordinator *cdt = &mdt->mdt_coordinator;
779 if (cdt->cdt_state == CDT_STOPPED)
782 /* wake up coordinator */
783 cdt->cdt_thread.t_flags = SVC_EVENT;
784 wake_up(&cdt->cdt_thread.t_ctl_waitq);
790 * initialize coordinator struct
791 * \param mdt [IN] device
793 * \retval -ve failure
795 int mdt_hsm_cdt_init(struct mdt_device *mdt)
797 struct coordinator *cdt = &mdt->mdt_coordinator;
798 struct mdt_thread_info *cdt_mti = NULL;
802 cdt->cdt_state = CDT_STOPPED;
804 init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
805 mutex_init(&cdt->cdt_llog_lock);
806 init_rwsem(&cdt->cdt_agent_lock);
807 init_rwsem(&cdt->cdt_request_lock);
808 mutex_init(&cdt->cdt_restore_lock);
810 INIT_LIST_HEAD(&cdt->cdt_requests);
811 INIT_LIST_HEAD(&cdt->cdt_agents);
812 INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
814 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
818 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
819 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
821 lu_context_enter(&cdt->cdt_session);
822 cdt->cdt_env.le_ses = &cdt->cdt_session;
824 lu_env_fini(&cdt->cdt_env);
828 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
829 LASSERT(cdt_mti != NULL);
831 cdt_mti->mti_env = &cdt->cdt_env;
832 cdt_mti->mti_mdt = mdt;
834 hsm_init_ucred(mdt_ucred(cdt_mti));
836 /* default values for /proc tunnables
837 * can be override by MGS conf */
838 cdt->cdt_default_archive_id = 1;
839 cdt->cdt_grace_delay = 60;
840 cdt->cdt_loop_period = 10;
841 cdt->cdt_max_requests = 3;
842 cdt->cdt_policy = CDT_DEFAULT_POLICY;
843 cdt->cdt_active_req_timeout = 3600;
849 * free a coordinator thread
850 * \param mdt [IN] device
852 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
854 struct coordinator *cdt = &mdt->mdt_coordinator;
857 lu_context_exit(cdt->cdt_env.le_ses);
858 lu_context_fini(cdt->cdt_env.le_ses);
860 lu_env_fini(&cdt->cdt_env);
866 * start a coordinator thread
867 * \param mdt [IN] device
869 * \retval -ve failure
871 int mdt_hsm_cdt_start(struct mdt_device *mdt)
873 struct coordinator *cdt = &mdt->mdt_coordinator;
876 struct mdt_thread_info *cdt_mti;
877 struct task_struct *task;
880 /* functions defined but not yet used
881 * this avoid compilation warning
885 if (cdt->cdt_state != CDT_STOPPED) {
886 CERROR("%s: Coordinator already started\n",
891 CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
892 cdt->cdt_policy = CDT_DEFAULT_POLICY;
894 cdt->cdt_state = CDT_INIT;
896 atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
897 /* just need to be larger than previous one */
898 /* cdt_last_cookie is protected by cdt_llog_lock */
899 cdt->cdt_last_cookie = cfs_time_current_sec();
900 atomic_set(&cdt->cdt_request_count, 0);
901 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
902 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
903 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
905 /* to avoid deadlock when start is made through /proc
906 * /proc entries are created by the coordinator thread */
908 /* set up list of started restore requests */
909 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
910 rc = mdt_hsm_pending_restore(cdt_mti);
912 CERROR("%s: cannot take the layout locks needed"
913 " for registered restore: %d\n",
914 mdt_obd_name(mdt), rc);
916 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
919 cdt->cdt_state = CDT_STOPPED;
920 CERROR("%s: error starting coordinator thread: %d\n",
921 mdt_obd_name(mdt), rc);
924 CDEBUG(D_HSM, "%s: coordinator thread started\n",
929 wait_event(cdt->cdt_thread.t_ctl_waitq,
930 (cdt->cdt_thread.t_flags & SVC_RUNNING));
932 cdt->cdt_state = CDT_RUNNING;
933 mdt->mdt_opts.mo_coordinator = 1;
938 * stop a coordinator thread
939 * \param mdt [IN] device
941 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
943 struct coordinator *cdt = &mdt->mdt_coordinator;
944 struct cdt_agent_req *car, *tmp1;
945 struct hsm_agent *ha, *tmp2;
946 struct cdt_restore_handle *crh, *tmp3;
947 struct mdt_thread_info *cdt_mti;
950 if (cdt->cdt_state == CDT_STOPPED) {
951 CERROR("%s: Coordinator already stopped\n",
956 if (cdt->cdt_state != CDT_STOPPING) {
957 /* stop coordinator thread before cleaning */
958 cdt->cdt_thread.t_flags = SVC_STOPPING;
959 wake_up(&cdt->cdt_thread.t_ctl_waitq);
960 wait_event(cdt->cdt_thread.t_ctl_waitq,
961 cdt->cdt_thread.t_flags & SVC_STOPPED);
963 cdt->cdt_state = CDT_STOPPED;
966 down_write(&cdt->cdt_request_lock);
967 list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
969 list_del(&car->car_request_list);
970 mdt_cdt_free_request(car);
972 up_write(&cdt->cdt_request_lock);
974 down_write(&cdt->cdt_agent_lock);
975 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
976 list_del(&ha->ha_list);
979 up_write(&cdt->cdt_agent_lock);
981 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
982 mutex_lock(&cdt->cdt_restore_lock);
983 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
984 struct mdt_object *child;
986 /* give back layout lock */
987 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
989 mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
991 list_del(&crh->crh_list);
993 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
995 mutex_unlock(&cdt->cdt_restore_lock);
997 mdt->mdt_opts.mo_coordinator = 0;
1003 * register all requests from an hal in the memory list
1004 * \param mti [IN] context
1005 * \param hal [IN] request
1006 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1007 * which is running the CT
1009 * \retval -ve failure
1011 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1012 struct hsm_action_list *hal, struct obd_uuid *uuid)
1014 struct mdt_device *mdt = mti->mti_mdt;
1015 struct coordinator *cdt = &mdt->mdt_coordinator;
1016 struct hsm_action_item *hai;
1020 /* register request in memory list */
1021 hai = hai_first(hal);
1022 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1023 struct cdt_agent_req *car;
1025 /* in case of a cancel request, we first mark the ondisk
1026 * record of the request we want to stop as canceled
1027 * this does not change the cancel record
1028 * it will be done when updating the request status
1030 if (hai->hai_action == HSMA_CANCEL) {
1031 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1035 CERROR("%s: mdt_agent_record_update() failed, "
1036 "rc=%d, cannot update status to %s "
1037 "for cookie "LPX64"\n",
1038 mdt_obd_name(mdt), rc,
1039 agent_req_status2name(ARS_CANCELED),
1044 /* find the running request to set it canceled */
1045 car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1047 car->car_canceled = 1;
1048 /* uuid has to be changed to the one running the
1049 * request to cancel */
1050 *uuid = car->car_uuid;
1051 mdt_cdt_put_request(car);
1053 /* no need to memorize cancel request
1054 * this also avoid a deadlock when we receive
1055 * a purge all requests command
1060 if (hai->hai_action == HSMA_ARCHIVE) {
1061 struct mdt_object *obj;
1064 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1065 if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1068 GOTO(out, rc = PTR_ERR(obj));
1070 hsm.mh_flags |= HS_EXISTS;
1071 hsm.mh_arch_id = hal->hal_archive_id;
1072 rc = mdt_hsm_attr_set(mti, obj, &hsm);
1073 mdt_object_put(mti->mti_env, obj);
1078 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1079 hal->hal_archive_id, hal->hal_flags,
1082 GOTO(out, rc = PTR_ERR(car));
1084 rc = mdt_cdt_add_request(cdt, car);
1086 mdt_cdt_free_request(car);
1093 * swap layouts between 2 fids
1094 * \param mti [IN] context
1097 * \param mh_common [IN] MD HSM
1099 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1100 const lustre_fid *fid, const lustre_fid *dfid,
1101 struct md_hsm *mh_common)
1103 struct mdt_device *mdt = mti->mti_mdt;
1104 struct mdt_object *child1, *child2;
1105 struct mdt_lock_handle *lh2;
1109 child1 = mdt_object_find(mti->mti_env, mdt, fid);
1111 GOTO(out, rc = PTR_ERR(child1));
1113 /* we already have layout lock on FID so take only
1115 lh2 = &mti->mti_lh[MDT_LH_OLD];
1116 mdt_lock_reg_init(lh2, LCK_EX);
1117 child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1119 GOTO(out_child1, rc = PTR_ERR(child2));
1121 /* if copy tool closes the volatile before sending the final
1122 * progress through llapi_hsm_copy_end(), all the objects
1123 * are removed and mdd_swap_layout LBUG */
1124 if (!mdt_object_exists(child2)) {
1125 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1126 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1127 GOTO(out_child2, rc = -ENOENT);
1129 /* Since we only handle restores here, unconditionally use
1130 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1131 * be preserved in case of failure during swap_layout and not
1132 * leave a file in an intermediate but incoherent state.
1133 * But need to setup HSM xattr of data FID before, reuse
1134 * mti and mh presets for FID in hsm_cdt_request_completed(),
1135 * only need to clear RELEASED and DIRTY.
1137 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1138 rc = mdt_hsm_attr_set(mti, child2, mh_common);
1140 rc = mo_swap_layouts(mti->mti_env,
1141 mdt_object_child(child1),
1142 mdt_object_child(child2),
1143 SWAP_LAYOUTS_MDS_HSM);
1146 mdt_object_unlock_put(mti, child2, lh2, 1);
1148 mdt_object_put(mti->mti_env, child1);
1154 * update status of a completed request
1155 * \param mti [IN] context
1156 * \param pgs [IN] progress of the copy tool
1157 * \param update_record [IN] update llog record
1159 * \retval -ve failure
1161 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1162 struct hsm_progress_kernel *pgs,
1163 const struct cdt_agent_req *car,
1164 enum agent_req_status *status)
1166 const struct lu_env *env = mti->mti_env;
1167 struct mdt_device *mdt = mti->mti_mdt;
1168 struct coordinator *cdt = &mdt->mdt_coordinator;
1169 struct mdt_object *obj = NULL;
1170 int cl_flags = 0, rc = 0;
1175 /* default is to retry */
1176 *status = ARS_WAITING;
1178 /* find object by FID */
1179 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1180 /* we will update MD HSM only if needed */
1181 is_mh_changed = false;
1183 /* object removed */
1184 *status = ARS_SUCCEED;
1188 /* no need to change mh->mh_arch_id
1189 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1191 if (pgs->hpk_errval != 0) {
1192 switch (pgs->hpk_errval) {
1194 /* the copy tool does not support cancel
1195 * so the cancel request is failed
1196 * As we cannot distinguish a cancel progress
1197 * from another action progress (they have the
1198 * same cookie), we suppose here the CT returns
1199 * ENOSYS only if does not support cancel
1201 /* this can also happen when cdt calls it to
1202 * for a timed out request */
1203 *status = ARS_FAILED;
1204 /* to have a cancel event in changelog */
1205 pgs->hpk_errval = ECANCELED;
1208 /* the request record has already been set to
1209 * ARS_CANCELED, this set the cancel request
1211 *status = ARS_SUCCEED;
1214 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1215 !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1216 ARS_FAILED : ARS_WAITING);
1220 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1221 CERROR("%s: Request "LPX64" on "DFID
1222 " failed, error code %d too large\n",
1224 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1226 hsm_set_cl_error(&cl_flags,
1227 CLF_HSM_ERROVERFLOW);
1230 hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1233 switch (car->car_hai->hai_action) {
1235 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1238 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1241 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1244 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1245 CERROR("%s: Failed request "LPX64" on "DFID
1246 " cannot be a CANCEL\n",
1249 PFID(&pgs->hpk_fid));
1252 CERROR("%s: Failed request "LPX64" on "DFID
1253 " %d is an unknown action\n",
1255 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1256 car->car_hai->hai_action);
1261 *status = ARS_SUCCEED;
1262 switch (car->car_hai->hai_action) {
1264 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1265 /* set ARCHIVE keep EXIST and clear LOST and
1267 mh.mh_arch_ver = pgs->hpk_data_version;
1268 mh.mh_flags |= HS_ARCHIVED;
1269 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1270 is_mh_changed = true;
1273 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1275 /* do not clear RELEASED and DIRTY here
1276 * this will occur in hsm_swap_layouts()
1279 /* Restoring has changed the file version on
1281 mh.mh_arch_ver = pgs->hpk_data_version;
1282 is_mh_changed = true;
1285 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1286 /* clear ARCHIVED EXISTS and LOST */
1287 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1288 is_mh_changed = true;
1291 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1292 CERROR("%s: Successful request "LPX64
1294 " cannot be a CANCEL\n",
1297 PFID(&pgs->hpk_fid));
1300 CERROR("%s: Successful request "LPX64
1302 " %d is an unknown action\n",
1304 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1305 car->car_hai->hai_action);
1311 /* rc != 0 means error when analysing action, it may come from
1312 * a crasy CT no need to manage DIRTY
1315 hsm_set_cl_flags(&cl_flags,
1316 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1318 /* unlock is done later, after layout lock management */
1320 rc = mdt_hsm_attr_set(mti, obj, &mh);
1323 /* we give back layout lock only if restore was successful or
1324 * if restore was canceled or if policy is to not retry
1325 * in other cases we just unlock the object */
1326 if (car->car_hai->hai_action == HSMA_RESTORE &&
1327 (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1328 cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1329 struct cdt_restore_handle *crh;
1331 /* restore in data FID done, we swap the layouts
1332 * only if restore is successful */
1333 if (pgs->hpk_errval == 0) {
1334 rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1335 &car->car_hai->hai_dfid, &mh);
1337 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1338 *status = ARS_FAILED;
1339 pgs->hpk_errval = -rc;
1342 /* we have to retry, so keep layout lock */
1343 if (*status == ARS_WAITING)
1346 /* give back layout lock */
1347 mutex_lock(&cdt->cdt_restore_lock);
1348 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1350 list_del(&crh->crh_list);
1351 mutex_unlock(&cdt->cdt_restore_lock);
1352 /* just give back layout lock, we keep
1353 * the reference which is given back
1354 * later with the lock for HSM flags */
1355 if (!IS_ERR(obj) && crh != NULL)
1356 mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1359 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1365 if (obj != NULL && !IS_ERR(obj)) {
1366 mo_changelog(env, CL_HSM, cl_flags,
1367 mdt_object_child(obj));
1368 mdt_object_put(mti->mti_env, obj);
1375 * update status of a request
1376 * \param mti [IN] context
1377 * \param pgs [IN] progress of the copy tool
1378 * \param update_record [IN] update llog record
1380 * \retval -ve failure
1382 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1383 struct hsm_progress_kernel *pgs,
1384 const int update_record)
1386 struct mdt_device *mdt = mti->mti_mdt;
1387 struct coordinator *cdt = &mdt->mdt_coordinator;
1388 struct cdt_agent_req *car;
1392 /* no coordinator started, so we cannot serve requests */
1393 if (cdt->cdt_state == CDT_STOPPED)
1396 /* first do sanity checks */
1397 car = mdt_cdt_update_request(cdt, pgs);
1399 CERROR("%s: Cannot find running request for cookie "LPX64
1402 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1404 RETURN(PTR_ERR(car));
1407 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1408 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1409 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1410 hsm_copytool_action2name(car->car_hai->hai_action),
1411 pgs->hpk_flags, pgs->hpk_errval,
1412 PFID(&car->car_hai->hai_fid),
1413 PFID(&car->car_hai->hai_dfid));
1415 /* progress is done on FID or data FID depending of the action and
1416 * of the copy progress */
1417 /* for restore progress is used to send back the data FID to cdt */
1418 if (car->car_hai->hai_action == HSMA_RESTORE &&
1419 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1420 car->car_hai->hai_dfid = pgs->hpk_fid;
1422 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1423 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1424 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1425 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1426 CERROR("%s: Progress on "DFID" for cookie "LPX64
1427 " does not match request FID "DFID" nor data FID "
1430 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1431 PFID(&car->car_hai->hai_fid),
1432 PFID(&car->car_hai->hai_dfid));
1433 GOTO(out, rc = -EINVAL);
1436 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1437 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1438 " is not coherent (err=%d and not completed"
1441 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1442 hsm_copytool_action2name(car->car_hai->hai_action),
1443 pgs->hpk_errval, pgs->hpk_flags);
1444 GOTO(out, rc = -EINVAL);
1447 /* now progress is valid */
1449 /* we use a root like ucred */
1450 hsm_init_ucred(mdt_ucred(mti));
1452 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1453 enum agent_req_status status;
1455 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1457 /* remove request from memory list */
1458 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1460 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1461 " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1463 hsm_copytool_action2name(car->car_hai->hai_action),
1464 agent_req_status2name(status));
1466 if (update_record) {
1469 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1470 &pgs->hpk_cookie, 1,
1473 CERROR("%s: mdt_agent_record_update() failed,"
1474 " rc=%d, cannot update status to %s"
1475 " for cookie "LPX64"\n",
1476 mdt_obd_name(mdt), rc1,
1477 agent_req_status2name(status),
1479 rc = (rc != 0 ? rc : rc1);
1481 /* ct has completed a request, so a slot is available, wakeup
1482 * cdt to find new work */
1483 mdt_hsm_cdt_wakeup(mdt);
1485 /* if copytool send a progress on a canceled request
1486 * we inform copytool it should stop
1488 if (car->car_canceled == 1)
1494 /* remove ref got from mdt_cdt_update_request() */
1495 mdt_cdt_put_request(car);
1502 * data passed to llog_cat_process() callback
1503 * to cancel requests
1505 struct hsm_cancel_all_data {
1506 struct mdt_device *mdt;
1510 * llog_cat_process() callback, used to:
1511 * - purge all requests
1512 * \param env [IN] environment
1513 * \param llh [IN] llog handle
1514 * \param hdr [IN] llog record
1515 * \param data [IN] cb data = struct hsm_cancel_all_data
1517 * \retval -ve failure
1519 static int mdt_cancel_all_cb(const struct lu_env *env,
1520 struct llog_handle *llh,
1521 struct llog_rec_hdr *hdr, void *data)
1523 struct llog_agent_req_rec *larr;
1524 struct hsm_cancel_all_data *hcad;
1528 larr = (struct llog_agent_req_rec *)hdr;
1530 if (larr->arr_status == ARS_WAITING ||
1531 larr->arr_status == ARS_STARTED) {
1532 larr->arr_status = ARS_CANCELED;
1533 larr->arr_req_change = cfs_time_current_sec();
1534 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1541 * cancel all actions
1542 * \param obd [IN] MDT device
1544 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1546 struct mdt_thread_info *mti;
1547 struct coordinator *cdt = &mdt->mdt_coordinator;
1548 struct cdt_agent_req *car;
1549 struct hsm_action_list *hal = NULL;
1550 struct hsm_action_item *hai;
1551 struct hsm_cancel_all_data hcad;
1552 int hal_sz = 0, hal_len, rc;
1553 enum cdt_states save_state;
1556 /* retrieve coordinator context */
1557 mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1559 /* disable coordinator */
1560 save_state = cdt->cdt_state;
1561 cdt->cdt_state = CDT_DISABLE;
1563 /* send cancel to all running requests */
1564 down_read(&cdt->cdt_request_lock);
1565 list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1566 mdt_cdt_get_request(car);
1567 /* request is not yet removed from list, it will be done
1568 * when copytool will return progress
1571 if (car->car_hai->hai_action == HSMA_CANCEL) {
1572 mdt_cdt_put_request(car);
1577 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1578 cfs_size_round(car->car_hai->hai_len);
1580 if (hal_len > hal_sz && hal_sz > 0) {
1581 /* not enough room, free old buffer */
1582 OBD_FREE(hal, hal_sz);
1586 /* empty buffer, allocate one */
1589 OBD_ALLOC(hal, hal_sz);
1591 mdt_cdt_put_request(car);
1592 up_read(&cdt->cdt_request_lock);
1593 GOTO(out, rc = -ENOMEM);
1597 hal->hal_version = HAL_VERSION;
1598 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1600 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1601 hal->hal_compound_id = car->car_compound_id;
1602 hal->hal_archive_id = car->car_archive_id;
1603 hal->hal_flags = car->car_flags;
1606 hai = hai_first(hal);
1607 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1608 hai->hai_action = HSMA_CANCEL;
1611 /* it is possible to safely call mdt_hsm_agent_send()
1612 * (ie without a deadlock on cdt_request_lock), because the
1613 * write lock is taken only if we are not in purge mode
1614 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1615 * nor mdt_cdt_remove_request())
1617 /* no conflict with cdt thread because cdt is disable and we
1618 * have the request lock */
1619 mdt_hsm_agent_send(mti, hal, 1);
1621 mdt_cdt_put_request(car);
1623 up_read(&cdt->cdt_request_lock);
1626 OBD_FREE(hal, hal_sz);
1628 /* cancel all on-disk records */
1631 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1632 mdt_cancel_all_cb, &hcad);
1634 /* enable coordinator */
1635 cdt->cdt_state = save_state;
1641 * check if a request is compatible with file status
1642 * \param hai [IN] request description
1643 * \param hal_an [IN] request archive number (not used)
1644 * \param rq_flags [IN] request flags
1645 * \param hsm [IN] file HSM metadata
1648 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1649 const int hal_an, const __u64 rq_flags,
1650 const struct md_hsm *hsm)
1652 int is_compat = false;
1656 hsm_flags = hsm->mh_flags;
1657 switch (hai->hai_action) {
1659 if (!(hsm_flags & HS_NOARCHIVE) &&
1660 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1664 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1665 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1669 if (!(hsm_flags & HS_RELEASED) &&
1670 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1677 CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1678 " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1679 PFID(&hai->hai_fid),
1680 hsm_copytool_action2name(hai->hai_action), rq_flags,
1681 hai->hai_extent.offset, hai->hai_extent.length,
1683 (is_compat ? "compatible" : "uncompatible"));
1689 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1691 static const struct {
1695 } hsm_policy_names[] = {
1696 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1697 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1702 * convert a policy name to a bit
1703 * \param name [IN] policy name
1705 * \retval policy bit
1707 static __u64 hsm_policy_str2bit(const char *name)
1711 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1712 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1713 strcmp(hsm_policy_names[i].name, name) == 0)
1714 return hsm_policy_names[i].bit;
1719 * convert a policy bit field to a string
1720 * \param mask [IN] policy bit field
1721 * \param hexa [IN] print mask before bit names
1722 * \param buffer [OUT] string
1723 * \param count [IN] size of buffer
1725 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1733 seq_printf(m, "("LPX64") ", mask);
1735 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1738 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1739 if (hsm_policy_names[j].bit == bit)
1743 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1745 seq_printf(m, "%s ", hsm_policy_names[j].name);
1747 /* remove last ' ' */
1752 /* methods to read/write HSM policy flags */
1753 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1755 struct mdt_device *mdt = m->private;
1756 struct coordinator *cdt = &mdt->mdt_coordinator;
1759 hsm_policy_bit2str(m, cdt->cdt_policy, false);
1764 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1765 size_t count, loff_t *off)
1767 struct seq_file *m = file->private_data;
1768 struct mdt_device *mdt = m->private;
1769 struct coordinator *cdt = &mdt->mdt_coordinator;
1770 char *start, *token, sign;
1773 __u64 add_mask, remove_mask, set_mask;
1777 if (count + 1 > PAGE_SIZE)
1780 OBD_ALLOC(buf, count + 1);
1784 if (copy_from_user(buf, buffer, count))
1785 GOTO(out, rc = -EFAULT);
1790 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1793 add_mask = remove_mask = set_mask = 0;
1795 token = strsep(&start, "\n ");
1801 if (sign == '-' || sign == '+')
1804 policy = hsm_policy_str2bit(token);
1806 CWARN("%s: '%s' is unknown, "
1807 "supported policies are:\n", mdt_obd_name(mdt),
1809 hsm_policy_bit2str(m, 0, false);
1810 GOTO(out, rc = -EINVAL);
1814 remove_mask |= policy;
1824 } while (start != NULL);
1826 CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1827 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1829 /* if no sign in all string, it is a clear and set
1830 * if some sign found, all unsigned are converted
1832 * P1 P2 = set to P1 and P2
1833 * P1 -P2 = add P1 clear P2 same as +P1 -P2
1835 if (remove_mask == 0 && add_mask == 0) {
1836 cdt->cdt_policy = set_mask;
1838 cdt->cdt_policy |= set_mask | add_mask;
1839 cdt->cdt_policy &= ~remove_mask;
1842 GOTO(out, rc = count);
1845 OBD_FREE(buf, count + 1);
1848 LPROC_SEQ_FOPS(mdt_hsm_policy);
1850 #define GENERATE_PROC_METHOD(VAR) \
1851 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
1853 struct mdt_device *mdt = m->private; \
1854 struct coordinator *cdt = &mdt->mdt_coordinator; \
1857 seq_printf(m, LPU64"\n", (__u64)cdt->VAR); \
1861 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1862 size_t count, loff_t *off) \
1865 struct seq_file *m = file->private_data; \
1866 struct mdt_device *mdt = m->private; \
1867 struct coordinator *cdt = &mdt->mdt_coordinator; \
1872 rc = lprocfs_str_to_s64(buffer, count, &val); \
1875 if (val > 0 && val < INT_MAX) { \
1882 GENERATE_PROC_METHOD(cdt_loop_period)
1883 GENERATE_PROC_METHOD(cdt_grace_delay)
1884 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1885 GENERATE_PROC_METHOD(cdt_max_requests)
1886 GENERATE_PROC_METHOD(cdt_default_archive_id)
1889 * procfs write method for MDT/hsm_control
1890 * proc entry is in mdt directory so data is mdt obd_device pointer
1892 #define CDT_ENABLE_CMD "enabled"
1893 #define CDT_STOP_CMD "shutdown"
1894 #define CDT_DISABLE_CMD "disabled"
1895 #define CDT_PURGE_CMD "purge"
1896 #define CDT_HELP_CMD "help"
1897 #define CDT_MAX_CMD_LEN 10
1900 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1901 size_t count, loff_t *off)
1903 struct seq_file *m = file->private_data;
1904 struct obd_device *obd = m->private;
1905 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1906 struct coordinator *cdt = &(mdt->mdt_coordinator);
1908 char kernbuf[CDT_MAX_CMD_LEN];
1911 if (count == 0 || count >= sizeof(kernbuf))
1914 if (copy_from_user(kernbuf, buffer, count))
1918 if (kernbuf[count - 1] == '\n')
1919 kernbuf[count - 1] = 0;
1922 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1923 if (cdt->cdt_state == CDT_DISABLE) {
1924 cdt->cdt_state = CDT_RUNNING;
1925 mdt_hsm_cdt_wakeup(mdt);
1927 rc = mdt_hsm_cdt_start(mdt);
1929 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1930 if ((cdt->cdt_state == CDT_STOPPING) ||
1931 (cdt->cdt_state == CDT_STOPPED)) {
1932 CERROR("%s: Coordinator already stopped\n",
1936 cdt->cdt_state = CDT_STOPPING;
1938 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
1939 if ((cdt->cdt_state == CDT_STOPPING) ||
1940 (cdt->cdt_state == CDT_STOPPED)) {
1941 CERROR("%s: Coordinator is stopped\n",
1945 cdt->cdt_state = CDT_DISABLE;
1947 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
1948 rc = hsm_cancel_all_actions(mdt);
1949 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
1957 CERROR("%s: Valid coordinator control commands are: "
1958 "%s %s %s %s %s\n", mdt_obd_name(mdt),
1959 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1960 CDT_PURGE_CMD, CDT_HELP_CMD);
1968 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
1970 struct obd_device *obd = m->private;
1971 struct coordinator *cdt;
1974 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
1976 if (cdt->cdt_state == CDT_INIT)
1977 seq_printf(m, "init\n");
1978 else if (cdt->cdt_state == CDT_RUNNING)
1979 seq_printf(m, "enabled\n");
1980 else if (cdt->cdt_state == CDT_STOPPING)
1981 seq_printf(m, "stopping\n");
1982 else if (cdt->cdt_state == CDT_STOPPED)
1983 seq_printf(m, "stopped\n");
1984 else if (cdt->cdt_state == CDT_DISABLE)
1985 seq_printf(m, "disabled\n");
1987 seq_printf(m, "unknown\n");
1993 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
1999 for (i = 0; i < 8 * sizeof(mask); i++) {
2000 if (mask & (1UL << i)) {
2001 seq_printf(m, "%s%s", first ? "" : " ",
2002 hsm_copytool_action2name(i));
2012 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2014 struct mdt_device *mdt = m->private;
2015 struct coordinator *cdt = &mdt->mdt_coordinator;
2017 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2021 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2023 struct mdt_device *mdt = m->private;
2024 struct coordinator *cdt = &mdt->mdt_coordinator;
2026 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2030 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2032 struct mdt_device *mdt = m->private;
2033 struct coordinator *cdt = &mdt->mdt_coordinator;
2035 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2038 static inline enum hsm_copytool_action
2039 hsm_copytool_name2action(const char *name)
2041 if (strcasecmp(name, "NOOP") == 0)
2043 else if (strcasecmp(name, "ARCHIVE") == 0)
2044 return HSMA_ARCHIVE;
2045 else if (strcasecmp(name, "RESTORE") == 0)
2046 return HSMA_RESTORE;
2047 else if (strcasecmp(name, "REMOVE") == 0)
2049 else if (strcasecmp(name, "CANCEL") == 0)
2056 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2057 size_t user_count, __u64 *mask)
2059 char *buf, *pos, *name;
2065 if (!(user_count < 4096))
2068 buf_size = user_count + 1;
2070 OBD_ALLOC(buf, buf_size);
2074 if (copy_from_user(buf, user_buf, buf_size - 1))
2075 GOTO(out, rc = -EFAULT);
2077 buf[buf_size - 1] = '\0';
2080 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2086 action = hsm_copytool_name2action(name);
2088 GOTO(out, rc = -EINVAL);
2090 new_mask |= (1UL << action);
2096 OBD_FREE(buf, buf_size);
2102 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2103 size_t count, loff_t *off)
2105 struct seq_file *m = file->private_data;
2106 struct mdt_device *mdt = m->private;
2107 struct coordinator *cdt = &mdt->mdt_coordinator;
2109 return mdt_write_hsm_request_mask(file, buf, count,
2110 &cdt->cdt_user_request_mask);
2114 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2115 size_t count, loff_t *off)
2117 struct seq_file *m = file->private_data;
2118 struct mdt_device *mdt = m->private;
2119 struct coordinator *cdt = &mdt->mdt_coordinator;
2121 return mdt_write_hsm_request_mask(file, buf, count,
2122 &cdt->cdt_group_request_mask);
2126 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2127 size_t count, loff_t *off)
2129 struct seq_file *m = file->private_data;
2130 struct mdt_device *mdt = m->private;
2131 struct coordinator *cdt = &mdt->mdt_coordinator;
2133 return mdt_write_hsm_request_mask(file, buf, count,
2134 &cdt->cdt_other_request_mask);
2137 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2138 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2139 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2140 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2141 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2142 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2143 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2144 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2146 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2148 .fops = &mdt_hsm_agent_fops },
2149 { .name = "actions",
2150 .fops = &mdt_hsm_actions_fops,
2151 .proc_mode = 0444 },
2152 { .name = "default_archive_id",
2153 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2154 { .name = "grace_delay",
2155 .fops = &mdt_hsm_cdt_grace_delay_fops },
2156 { .name = "loop_period",
2157 .fops = &mdt_hsm_cdt_loop_period_fops },
2158 { .name = "max_requests",
2159 .fops = &mdt_hsm_cdt_max_requests_fops },
2161 .fops = &mdt_hsm_policy_fops },
2162 { .name = "active_request_timeout",
2163 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2164 { .name = "active_requests",
2165 .fops = &mdt_hsm_active_requests_fops },
2166 { .name = "user_request_mask",
2167 .fops = &mdt_hsm_user_request_mask_fops, },
2168 { .name = "group_request_mask",
2169 .fops = &mdt_hsm_group_request_mask_fops, },
2170 { .name = "other_request_mask",
2171 .fops = &mdt_hsm_other_request_mask_fops, },