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;
145 /* records to be canceled */
146 int max_cookie; /** vector size */
147 int cookie_cnt; /** used count */
152 * llog_cat_process() callback, used to:
153 * - find waiting request and start action
154 * - purge canceled and done requests
155 * \param env [IN] environment
156 * \param llh [IN] llog handle
157 * \param hdr [IN] llog record
158 * \param data [IN/OUT] cb data = struct hsm_scan_data
160 * \retval -ve failure
162 static int mdt_coordinator_cb(const struct lu_env *env,
163 struct llog_handle *llh,
164 struct llog_rec_hdr *hdr,
167 const struct llog_agent_req_rec *larr;
168 struct hsm_scan_data *hsd;
169 struct hsm_action_item *hai;
170 struct mdt_device *mdt;
171 struct coordinator *cdt;
176 mdt = hsd->mti->mti_mdt;
177 cdt = &mdt->mdt_coordinator;
179 larr = (struct llog_agent_req_rec *)hdr;
180 dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
181 switch (larr->arr_status) {
183 int i, empty_slot, found;
185 /* Are agents full? */
186 if (atomic_read(&cdt->cdt_request_count) ==
187 cdt->cdt_max_requests)
190 /* first search whether the request is found in the list we
191 * have built and if there is room in the request vector */
194 for (i = 0; i < hsd->max_requests &&
195 (empty_slot == -1 || found == -1); i++) {
196 if (hsd->request[i].hal == NULL) {
200 if (hsd->request[i].hal->hal_compound_id ==
201 larr->arr_compound_id) {
206 if (found == -1 && empty_slot == -1)
207 /* unknown request and no more room for new request,
208 * continue scan for to find other entries for
209 * already found request
214 struct hsm_action_list *hal;
216 /* request is not already known */
217 /* allocates hai vector size just needs to be large
219 hsd->request[empty_slot].hal_sz =
220 sizeof(*hsd->request[empty_slot].hal) +
221 cfs_size_round(MTI_NAME_MAXLEN+1) +
222 2 * cfs_size_round(larr->arr_hai.hai_len);
223 OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
225 CERROR("%s: Cannot allocate memory (%d o)"
226 "for compound "LPX64"\n",
228 hsd->request[i].hal_sz,
229 larr->arr_compound_id);
232 hal->hal_version = HAL_VERSION;
233 strlcpy(hal->hal_fsname, hsd->fs_name,
234 MTI_NAME_MAXLEN + 1);
235 hal->hal_compound_id = larr->arr_compound_id;
236 hal->hal_archive_id = larr->arr_archive_id;
237 hal->hal_flags = larr->arr_flags;
239 hsd->request[empty_slot].hal_used_sz = hal_size(hal);
240 hsd->request[empty_slot].hal = hal;
243 hai = hai_first(hal);
245 /* request is known */
246 /* we check if record archive num is the same as the
247 * known request, if not we will serve it in multiple
248 * time because we do not know if the agent can serve
250 * a use case is a compound made of multiple restore
251 * where the files are not archived in the same backend
253 if (larr->arr_archive_id !=
254 hsd->request[found].hal->hal_archive_id)
257 if (hsd->request[found].hal_sz <
258 hsd->request[found].hal_used_sz +
259 cfs_size_round(larr->arr_hai.hai_len)) {
260 /* Not enough room, need an extension */
264 sz = 2 * hsd->request[found].hal_sz;
265 OBD_ALLOC(hal_buffer, sz);
267 CERROR("%s: Cannot allocate memory "
268 "(%d o) for compound "LPX64"\n",
269 mdt_obd_name(mdt), sz,
270 larr->arr_compound_id);
273 memcpy(hal_buffer, hsd->request[found].hal,
274 hsd->request[found].hal_used_sz);
275 OBD_FREE(hsd->request[found].hal,
276 hsd->request[found].hal_sz);
277 hsd->request[found].hal = hal_buffer;
278 hsd->request[found].hal_sz = sz;
280 hai = hai_first(hsd->request[found].hal);
281 for (i = 0; i < hsd->request[found].hal->hal_count;
285 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
286 hai->hai_cookie = larr->arr_hai.hai_cookie;
287 hai->hai_gid = larr->arr_hai.hai_gid;
289 hsd->request[found].hal_used_sz +=
290 cfs_size_round(hai->hai_len);
291 hsd->request[found].hal->hal_count++;
295 struct cdt_agent_req *car;
298 /* we search for a running request
299 * error may happen if coordinator crashes or stopped
300 * with running request
302 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
304 last = larr->arr_req_create;
306 last = car->car_req_update;
307 mdt_cdt_put_request(car);
310 /* test if request too long, if yes cancel it
311 * the same way the copy tool acknowledge a cancel request */
312 if ((last + cdt->cdt_active_req_timeout)
313 < cfs_time_current_sec()) {
314 struct hsm_progress_kernel pgs;
316 dump_llog_agent_req_rec("mdt_coordinator_cb(): "
317 "request timed out, start "
319 /* a too old cancel request just needs to be removed
320 * this can happen, if copy tool does not support cancel
321 * for other requests, we have to remove the running
322 * request and notify the copytool
324 pgs.hpk_fid = larr->arr_hai.hai_fid;
325 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
326 pgs.hpk_extent = larr->arr_hai.hai_extent;
327 pgs.hpk_flags = HP_FLAG_COMPLETED;
328 pgs.hpk_errval = ENOSYS;
329 pgs.hpk_data_version = 0;
330 /* update request state, but do not record in llog, to
331 * avoid deadlock on cdt_llog_lock
333 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
335 CERROR("%s: Cannot cleanup timed out request: "
336 DFID" for cookie "LPX64" action=%s\n",
338 PFID(&pgs.hpk_fid), pgs.hpk_cookie,
339 hsm_copytool_action2name(
340 larr->arr_hai.hai_action));
343 /* The request no longer exists, forget
344 * about it, and do not send a cancel request
345 * to the client, for which an error will be
346 * sent back, leading to an endless cycle of
348 RETURN(LLOG_DEL_RECORD);
351 /* add the cookie to the list of record to be
352 * canceled by caller */
353 if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
354 __u64 *ptr, *old_ptr;
355 int old_sz, new_sz, new_cnt;
357 /* need to increase vector size */
358 old_sz = sizeof(__u64) * hsd->max_cookie;
359 old_ptr = hsd->cookies;
361 new_cnt = 2 * hsd->max_cookie;
362 new_sz = sizeof(__u64) * new_cnt;
364 OBD_ALLOC(ptr, new_sz);
366 CERROR("%s: Cannot allocate memory "
367 "(%d o) for cookie vector\n",
368 mdt_obd_name(mdt), new_sz);
371 memcpy(ptr, hsd->cookies, old_sz);
373 hsd->max_cookie = new_cnt;
374 OBD_FREE(old_ptr, old_sz);
376 hsd->cookies[hsd->cookie_cnt] =
377 larr->arr_hai.hai_cookie;
385 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
386 cfs_time_current_sec())
387 RETURN(LLOG_DEL_RECORD);
394 * create /proc entries for coordinator
397 * \retval -ve failure
399 int hsm_cdt_procfs_init(struct mdt_device *mdt)
401 struct coordinator *cdt = &mdt->mdt_coordinator;
405 /* init /proc entries, failure is not critical */
406 cdt->cdt_proc_dir = lprocfs_register("hsm",
407 mdt2obd_dev(mdt)->obd_proc_entry,
408 lprocfs_mdt_hsm_vars, mdt);
409 if (IS_ERR(cdt->cdt_proc_dir)) {
410 rc = PTR_ERR(cdt->cdt_proc_dir);
411 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
412 " rc=%d\n", mdt_obd_name(mdt), rc);
413 cdt->cdt_proc_dir = NULL;
421 * remove /proc entries for coordinator
424 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
426 struct coordinator *cdt = &mdt->mdt_coordinator;
428 LASSERT(cdt->cdt_state == CDT_STOPPED);
429 if (cdt->cdt_proc_dir != NULL)
430 lprocfs_remove(&cdt->cdt_proc_dir);
434 * get vector of hsm cdt /proc vars
438 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
440 return lprocfs_mdt_hsm_vars;
445 * \param data [IN] obd device
447 * \retval -ve failure
449 static int mdt_coordinator(void *data)
451 struct mdt_thread_info *mti = data;
452 struct mdt_device *mdt = mti->mti_mdt;
453 struct coordinator *cdt = &mdt->mdt_coordinator;
454 struct hsm_scan_data hsd = { NULL };
458 cdt->cdt_thread.t_flags = SVC_RUNNING;
459 wake_up(&cdt->cdt_thread.t_ctl_waitq);
461 CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
462 mdt_obd_name(mdt), current_pid());
464 /* timeouted cookie vector initialization */
468 /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
469 * increases due to a change from /proc we do not overflow the
470 * hsd.request[] vector
472 hsd.max_requests = cdt->cdt_max_requests;
473 hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
474 OBD_ALLOC(hsd.request, hsd.request_sz);
476 GOTO(out, rc = -ENOMEM);
479 obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
482 struct l_wait_info lwi;
485 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
487 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
488 (cdt->cdt_thread.t_flags &
489 (SVC_STOPPING|SVC_EVENT)),
492 CDEBUG(D_HSM, "coordinator resumes\n");
494 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
495 cdt->cdt_state == CDT_STOPPING) {
496 cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
501 /* wake up before timeout, new work arrives */
502 if (cdt->cdt_thread.t_flags & SVC_EVENT)
503 cdt->cdt_thread.t_flags &= ~SVC_EVENT;
505 /* if coordinator is suspended continue to wait */
506 if (cdt->cdt_state == CDT_DISABLE) {
507 CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
511 CDEBUG(D_HSM, "coordinator starts reading llog\n");
513 if (hsd.max_requests != cdt->cdt_max_requests) {
514 /* cdt_max_requests has changed,
515 * we need to allocate a new buffer
517 OBD_FREE(hsd.request, hsd.request_sz);
518 hsd.max_requests = cdt->cdt_max_requests;
520 hsd.max_requests * sizeof(*hsd.request);
521 OBD_ALLOC(hsd.request, hsd.request_sz);
528 /* create canceled cookie vector for an arbitrary size
529 * if needed, vector will grow during llog scan
533 OBD_ALLOC(hsd.cookies, hsd.max_cookie * sizeof(__u64));
540 rc = cdt_llog_process(mti->mti_env, mdt,
541 mdt_coordinator_cb, &hsd);
545 CDEBUG(D_HSM, "Found %d requests to send and %d"
546 " requests to cancel\n",
547 hsd.request_cnt, hsd.cookie_cnt);
548 /* first we cancel llog records of the timed out requests */
549 if (hsd.cookie_cnt > 0) {
550 rc = mdt_agent_record_update(mti->mti_env, mdt,
555 CERROR("%s: mdt_agent_record_update() failed, "
556 "rc=%d, cannot update status to %s "
558 mdt_obd_name(mdt), rc,
559 agent_req_status2name(ARS_CANCELED),
563 if (list_empty(&cdt->cdt_agents)) {
564 CDEBUG(D_HSM, "no agent available, "
565 "coordinator sleeps\n");
569 /* here hsd contains a list of requests to be started */
570 for (i = 0; i < hsd.max_requests; i++) {
571 struct hsm_action_list *hal;
572 struct hsm_action_item *hai;
575 enum agent_req_status status;
577 /* still room for work ? */
578 if (atomic_read(&cdt->cdt_request_count) ==
579 cdt->cdt_max_requests)
582 if (hsd.request[i].hal == NULL)
585 /* found a request, we start it */
586 /* kuc payload allocation so we avoid an additionnal
587 * allocation in mdt_hsm_agent_send()
589 hal = kuc_alloc(hsd.request[i].hal_used_sz,
590 KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
592 CERROR("%s: Cannot allocate memory (%d o) "
593 "for compound "LPX64"\n",
595 hsd.request[i].hal_used_sz,
596 hsd.request[i].hal->hal_compound_id);
599 memcpy(hal, hsd.request[i].hal,
600 hsd.request[i].hal_used_sz);
602 rc = mdt_hsm_agent_send(mti, hal, 0);
603 /* if failure, we suppose it is temporary
604 * if the copy tool failed to do the request
605 * it has to use hsm_progress
607 status = (rc ? ARS_WAITING : ARS_STARTED);
609 /* set up cookie vector to set records status
610 * after copy tools start or failed
612 sz = hsd.request[i].hal->hal_count * sizeof(__u64);
613 OBD_ALLOC(cookies, sz);
614 if (cookies == NULL) {
615 CERROR("%s: Cannot allocate memory (%d o) "
616 "for cookies vector "LPX64"\n",
617 mdt_obd_name(mdt), sz,
618 hsd.request[i].hal->hal_compound_id);
619 kuc_free(hal, hsd.request[i].hal_used_sz);
622 hai = hai_first(hal);
623 for (j = 0; j < hsd.request[i].hal->hal_count; j++) {
624 cookies[j] = hai->hai_cookie;
628 rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
629 hsd.request[i].hal->hal_count,
632 CERROR("%s: mdt_agent_record_update() failed, "
633 "rc=%d, cannot update status to %s "
635 mdt_obd_name(mdt), rc,
636 agent_req_status2name(status),
637 hsd.request[i].hal->hal_count);
639 OBD_FREE(cookies, sz);
640 kuc_free(hal, hsd.request[i].hal_used_sz);
643 /* free cookie vector allocated for/by callback */
645 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
651 /* free hal allocated by callback */
652 for (i = 0; i < hsd.max_requests; i++) {
653 if (hsd.request[i].hal) {
654 OBD_FREE(hsd.request[i].hal,
655 hsd.request[i].hal_sz);
656 hsd.request[i].hal_sz = 0;
657 hsd.request[i].hal = NULL;
661 LASSERT(hsd.request_cnt == 0);
663 /* reset callback data */
664 memset(hsd.request, 0, hsd.request_sz);
669 OBD_FREE(hsd.request, hsd.request_sz);
672 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
674 if (cdt->cdt_state == CDT_STOPPING) {
675 /* request comes from /proc path, so we need to clean cdt
677 mdt_hsm_cdt_stop(mdt);
678 mdt->mdt_opts.mo_coordinator = 0;
680 /* request comes from a thread event, generated
681 * by mdt_stop_coordinator(), we have to ack
682 * and cdt cleaning will be done by event sender
684 cdt->cdt_thread.t_flags = SVC_STOPPED;
685 wake_up(&cdt->cdt_thread.t_ctl_waitq);
689 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
690 mdt_obd_name(mdt), current_pid(), rc);
692 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
694 mdt_obd_name(mdt), current_pid());
700 * lookup a restore handle by FID
701 * caller needs to hold cdt_restore_lock
702 * \param cdt [IN] coordinator
703 * \param fid [IN] FID
704 * \retval cdt_restore_handle found
705 * \retval NULL not found
707 static struct cdt_restore_handle *hsm_restore_hdl_find(struct coordinator *cdt,
708 const struct lu_fid *fid)
710 struct cdt_restore_handle *crh;
713 list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
714 if (lu_fid_eq(&crh->crh_fid, fid))
721 * data passed to llog_cat_process() callback
722 * to scan requests and take actions
724 struct hsm_restore_data {
725 struct mdt_thread_info *hrd_mti;
729 * llog_cat_process() callback, used to:
730 * - find restore request and allocate the restore handle
731 * \param env [IN] environment
732 * \param llh [IN] llog handle
733 * \param hdr [IN] llog record
734 * \param data [IN/OUT] cb data = struct hsm_restore_data
736 * \retval -ve failure
738 static int hsm_restore_cb(const struct lu_env *env,
739 struct llog_handle *llh,
740 struct llog_rec_hdr *hdr, void *data)
742 struct llog_agent_req_rec *larr;
743 struct hsm_restore_data *hrd;
744 struct cdt_restore_handle *crh;
745 struct hsm_action_item *hai;
746 struct mdt_thread_info *mti;
747 struct coordinator *cdt;
748 struct mdt_object *child;
754 cdt = &mti->mti_mdt->mdt_coordinator;
756 larr = (struct llog_agent_req_rec *)hdr;
757 hai = &larr->arr_hai;
758 if (hai->hai_cookie > cdt->cdt_last_cookie)
759 /* update the cookie to avoid collision */
760 cdt->cdt_last_cookie = hai->hai_cookie + 1;
762 if (hai->hai_action != HSMA_RESTORE ||
763 agent_req_in_final_state(larr->arr_status))
766 /* restore request not in a final state */
768 OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
772 crh->crh_fid = hai->hai_fid;
773 /* in V1 all file is restored
774 crh->extent.start = hai->hai_extent.offset;
775 crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
777 crh->crh_extent.start = 0;
778 crh->crh_extent.end = hai->hai_extent.length;
779 /* get the layout lock */
780 mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
781 child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
782 MDS_INODELOCK_LAYOUT);
784 GOTO(out, rc = PTR_ERR(child));
787 /* we choose to not keep a reference
788 * on the object during the restore time which can be very long */
789 mdt_object_put(mti->mti_env, child);
791 mutex_lock(&cdt->cdt_restore_lock);
792 list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
793 mutex_unlock(&cdt->cdt_restore_lock);
800 * restore coordinator state at startup
801 * the goal is to take a layout lock for each registered restore request
802 * \param mti [IN] context
804 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
806 struct hsm_restore_data hrd;
812 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
813 hsm_restore_cb, &hrd);
818 static int hsm_init_ucred(struct lu_ucred *uc)
822 uc->uc_valid = UCRED_OLD;
831 uc->uc_suppgids[0] = -1;
832 uc->uc_suppgids[1] = -1;
833 uc->uc_cap = CFS_CAP_FS_MASK;
836 uc->uc_identity = NULL;
842 * wake up coordinator thread
843 * \param mdt [IN] device
845 * \retval -ve failure
847 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
849 struct coordinator *cdt = &mdt->mdt_coordinator;
852 if (cdt->cdt_state == CDT_STOPPED)
855 /* wake up coordinator */
856 cdt->cdt_thread.t_flags = SVC_EVENT;
857 wake_up(&cdt->cdt_thread.t_ctl_waitq);
863 * initialize coordinator struct
864 * \param mdt [IN] device
866 * \retval -ve failure
868 int mdt_hsm_cdt_init(struct mdt_device *mdt)
870 struct coordinator *cdt = &mdt->mdt_coordinator;
871 struct mdt_thread_info *cdt_mti = NULL;
875 cdt->cdt_state = CDT_STOPPED;
877 init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
878 mutex_init(&cdt->cdt_llog_lock);
879 init_rwsem(&cdt->cdt_agent_lock);
880 init_rwsem(&cdt->cdt_request_lock);
881 mutex_init(&cdt->cdt_restore_lock);
883 INIT_LIST_HEAD(&cdt->cdt_requests);
884 INIT_LIST_HEAD(&cdt->cdt_agents);
885 INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
887 rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
891 /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
892 rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
894 lu_context_enter(&cdt->cdt_session);
895 cdt->cdt_env.le_ses = &cdt->cdt_session;
897 lu_env_fini(&cdt->cdt_env);
901 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
902 LASSERT(cdt_mti != NULL);
904 cdt_mti->mti_env = &cdt->cdt_env;
905 cdt_mti->mti_mdt = mdt;
907 hsm_init_ucred(mdt_ucred(cdt_mti));
909 /* default values for /proc tunnables
910 * can be override by MGS conf */
911 cdt->cdt_default_archive_id = 1;
912 cdt->cdt_grace_delay = 60;
913 cdt->cdt_loop_period = 10;
914 cdt->cdt_max_requests = 3;
915 cdt->cdt_policy = CDT_DEFAULT_POLICY;
916 cdt->cdt_active_req_timeout = 3600;
922 * free a coordinator thread
923 * \param mdt [IN] device
925 int mdt_hsm_cdt_fini(struct mdt_device *mdt)
927 struct coordinator *cdt = &mdt->mdt_coordinator;
930 lu_context_exit(cdt->cdt_env.le_ses);
931 lu_context_fini(cdt->cdt_env.le_ses);
933 lu_env_fini(&cdt->cdt_env);
939 * start a coordinator thread
940 * \param mdt [IN] device
942 * \retval -ve failure
944 int mdt_hsm_cdt_start(struct mdt_device *mdt)
946 struct coordinator *cdt = &mdt->mdt_coordinator;
949 struct mdt_thread_info *cdt_mti;
950 struct task_struct *task;
953 /* functions defined but not yet used
954 * this avoid compilation warning
958 if (cdt->cdt_state != CDT_STOPPED) {
959 CERROR("%s: Coordinator already started\n",
964 CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
965 cdt->cdt_policy = CDT_DEFAULT_POLICY;
967 cdt->cdt_state = CDT_INIT;
969 atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
970 /* just need to be larger than previous one */
971 /* cdt_last_cookie is protected by cdt_llog_lock */
972 cdt->cdt_last_cookie = cfs_time_current_sec();
973 atomic_set(&cdt->cdt_request_count, 0);
974 cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
975 cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
976 cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
978 /* to avoid deadlock when start is made through /proc
979 * /proc entries are created by the coordinator thread */
981 /* set up list of started restore requests */
982 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
983 rc = mdt_hsm_pending_restore(cdt_mti);
985 CERROR("%s: cannot take the layout locks needed"
986 " for registered restore: %d\n",
987 mdt_obd_name(mdt), rc);
989 task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
992 cdt->cdt_state = CDT_STOPPED;
993 CERROR("%s: error starting coordinator thread: %d\n",
994 mdt_obd_name(mdt), rc);
997 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1002 wait_event(cdt->cdt_thread.t_ctl_waitq,
1003 (cdt->cdt_thread.t_flags & SVC_RUNNING));
1005 cdt->cdt_state = CDT_RUNNING;
1006 mdt->mdt_opts.mo_coordinator = 1;
1011 * stop a coordinator thread
1012 * \param mdt [IN] device
1014 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1016 struct coordinator *cdt = &mdt->mdt_coordinator;
1017 struct cdt_agent_req *car, *tmp1;
1018 struct hsm_agent *ha, *tmp2;
1019 struct cdt_restore_handle *crh, *tmp3;
1020 struct mdt_thread_info *cdt_mti;
1023 if (cdt->cdt_state == CDT_STOPPED) {
1024 CERROR("%s: Coordinator already stopped\n",
1029 if (cdt->cdt_state != CDT_STOPPING) {
1030 /* stop coordinator thread before cleaning */
1031 cdt->cdt_thread.t_flags = SVC_STOPPING;
1032 wake_up(&cdt->cdt_thread.t_ctl_waitq);
1033 wait_event(cdt->cdt_thread.t_ctl_waitq,
1034 cdt->cdt_thread.t_flags & SVC_STOPPED);
1036 cdt->cdt_state = CDT_STOPPED;
1038 /* start cleaning */
1039 down_write(&cdt->cdt_request_lock);
1040 list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
1042 list_del(&car->car_request_list);
1043 mdt_cdt_free_request(car);
1045 up_write(&cdt->cdt_request_lock);
1047 down_write(&cdt->cdt_agent_lock);
1048 list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
1049 list_del(&ha->ha_list);
1052 up_write(&cdt->cdt_agent_lock);
1054 cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1055 mutex_lock(&cdt->cdt_restore_lock);
1056 list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
1057 struct mdt_object *child;
1059 /* give back layout lock */
1060 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
1062 mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
1064 list_del(&crh->crh_list);
1066 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1068 mutex_unlock(&cdt->cdt_restore_lock);
1070 mdt->mdt_opts.mo_coordinator = 0;
1076 * register all requests from an hal in the memory list
1077 * \param mti [IN] context
1078 * \param hal [IN] request
1079 * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1080 * which is running the CT
1082 * \retval -ve failure
1084 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1085 struct hsm_action_list *hal, struct obd_uuid *uuid)
1087 struct mdt_device *mdt = mti->mti_mdt;
1088 struct coordinator *cdt = &mdt->mdt_coordinator;
1089 struct hsm_action_item *hai;
1093 /* register request in memory list */
1094 hai = hai_first(hal);
1095 for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1096 struct cdt_agent_req *car;
1098 /* in case of a cancel request, we first mark the ondisk
1099 * record of the request we want to stop as canceled
1100 * this does not change the cancel record
1101 * it will be done when updating the request status
1103 if (hai->hai_action == HSMA_CANCEL) {
1104 rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1108 CERROR("%s: mdt_agent_record_update() failed, "
1109 "rc=%d, cannot update status to %s "
1110 "for cookie "LPX64"\n",
1111 mdt_obd_name(mdt), rc,
1112 agent_req_status2name(ARS_CANCELED),
1117 /* find the running request to set it canceled */
1118 car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1120 car->car_canceled = 1;
1121 /* uuid has to be changed to the one running the
1122 * request to cancel */
1123 *uuid = car->car_uuid;
1124 mdt_cdt_put_request(car);
1126 /* no need to memorize cancel request
1127 * this also avoid a deadlock when we receive
1128 * a purge all requests command
1133 if (hai->hai_action == HSMA_ARCHIVE) {
1134 struct mdt_object *obj;
1137 obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1138 if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1141 GOTO(out, rc = PTR_ERR(obj));
1143 hsm.mh_flags |= HS_EXISTS;
1144 hsm.mh_arch_id = hal->hal_archive_id;
1145 rc = mdt_hsm_attr_set(mti, obj, &hsm);
1146 mdt_object_put(mti->mti_env, obj);
1151 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1152 hal->hal_archive_id, hal->hal_flags,
1155 GOTO(out, rc = PTR_ERR(car));
1157 rc = mdt_cdt_add_request(cdt, car);
1159 mdt_cdt_free_request(car);
1166 * swap layouts between 2 fids
1167 * \param mti [IN] context
1170 * \param mh_common [IN] MD HSM
1172 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1173 const lustre_fid *fid, const lustre_fid *dfid,
1174 struct md_hsm *mh_common)
1176 struct mdt_device *mdt = mti->mti_mdt;
1177 struct mdt_object *child1, *child2;
1178 struct mdt_lock_handle *lh2;
1182 child1 = mdt_object_find(mti->mti_env, mdt, fid);
1184 GOTO(out, rc = PTR_ERR(child1));
1186 /* we already have layout lock on FID so take only
1188 lh2 = &mti->mti_lh[MDT_LH_OLD];
1189 mdt_lock_reg_init(lh2, LCK_EX);
1190 child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1192 GOTO(out_child1, rc = PTR_ERR(child2));
1194 /* if copy tool closes the volatile before sending the final
1195 * progress through llapi_hsm_copy_end(), all the objects
1196 * are removed and mdd_swap_layout LBUG */
1197 if (!mdt_object_exists(child2)) {
1198 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1199 mdt_obd_name(mti->mti_mdt), PFID(dfid));
1200 GOTO(out_child2, rc = -ENOENT);
1202 /* Since we only handle restores here, unconditionally use
1203 * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1204 * be preserved in case of failure during swap_layout and not
1205 * leave a file in an intermediate but incoherent state.
1206 * But need to setup HSM xattr of data FID before, reuse
1207 * mti and mh presets for FID in hsm_cdt_request_completed(),
1208 * only need to clear RELEASED and DIRTY.
1210 mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1211 rc = mdt_hsm_attr_set(mti, child2, mh_common);
1213 rc = mo_swap_layouts(mti->mti_env,
1214 mdt_object_child(child1),
1215 mdt_object_child(child2),
1216 SWAP_LAYOUTS_MDS_HSM);
1219 mdt_object_unlock_put(mti, child2, lh2, 1);
1221 mdt_object_put(mti->mti_env, child1);
1227 * update status of a completed request
1228 * \param mti [IN] context
1229 * \param pgs [IN] progress of the copy tool
1230 * \param update_record [IN] update llog record
1232 * \retval -ve failure
1234 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1235 struct hsm_progress_kernel *pgs,
1236 const struct cdt_agent_req *car,
1237 enum agent_req_status *status)
1239 const struct lu_env *env = mti->mti_env;
1240 struct mdt_device *mdt = mti->mti_mdt;
1241 struct coordinator *cdt = &mdt->mdt_coordinator;
1242 struct mdt_object *obj = NULL;
1243 int cl_flags = 0, rc = 0;
1248 /* default is to retry */
1249 *status = ARS_WAITING;
1251 /* find object by FID */
1252 obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1253 /* we will update MD HSM only if needed */
1254 is_mh_changed = false;
1256 /* object removed */
1257 *status = ARS_SUCCEED;
1261 /* no need to change mh->mh_arch_id
1262 * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1264 if (pgs->hpk_errval != 0) {
1265 switch (pgs->hpk_errval) {
1267 /* the copy tool does not support cancel
1268 * so the cancel request is failed
1269 * As we cannot distinguish a cancel progress
1270 * from another action progress (they have the
1271 * same cookie), we suppose here the CT returns
1272 * ENOSYS only if does not support cancel
1274 /* this can also happen when cdt calls it to
1275 * for a timed out request */
1276 *status = ARS_FAILED;
1277 /* to have a cancel event in changelog */
1278 pgs->hpk_errval = ECANCELED;
1281 /* the request record has already been set to
1282 * ARS_CANCELED, this set the cancel request
1284 *status = ARS_SUCCEED;
1287 *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1288 !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1289 ARS_FAILED : ARS_WAITING);
1293 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1294 CERROR("%s: Request "LPX64" on "DFID
1295 " failed, error code %d too large\n",
1297 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1299 hsm_set_cl_error(&cl_flags,
1300 CLF_HSM_ERROVERFLOW);
1303 hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1306 switch (car->car_hai->hai_action) {
1308 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1311 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1314 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1317 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1318 CERROR("%s: Failed request "LPX64" on "DFID
1319 " cannot be a CANCEL\n",
1322 PFID(&pgs->hpk_fid));
1325 CERROR("%s: Failed request "LPX64" on "DFID
1326 " %d is an unknown action\n",
1328 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1329 car->car_hai->hai_action);
1334 *status = ARS_SUCCEED;
1335 switch (car->car_hai->hai_action) {
1337 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1338 /* set ARCHIVE keep EXIST and clear LOST and
1340 mh.mh_arch_ver = pgs->hpk_data_version;
1341 mh.mh_flags |= HS_ARCHIVED;
1342 mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1343 is_mh_changed = true;
1346 hsm_set_cl_event(&cl_flags, HE_RESTORE);
1348 /* do not clear RELEASED and DIRTY here
1349 * this will occur in hsm_swap_layouts()
1352 /* Restoring has changed the file version on
1354 mh.mh_arch_ver = pgs->hpk_data_version;
1355 is_mh_changed = true;
1358 hsm_set_cl_event(&cl_flags, HE_REMOVE);
1359 /* clear ARCHIVED EXISTS and LOST */
1360 mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1361 is_mh_changed = true;
1364 hsm_set_cl_event(&cl_flags, HE_CANCEL);
1365 CERROR("%s: Successful request "LPX64
1367 " cannot be a CANCEL\n",
1370 PFID(&pgs->hpk_fid));
1373 CERROR("%s: Successful request "LPX64
1375 " %d is an unknown action\n",
1377 pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1378 car->car_hai->hai_action);
1384 /* rc != 0 means error when analysing action, it may come from
1385 * a crasy CT no need to manage DIRTY
1388 hsm_set_cl_flags(&cl_flags,
1389 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1391 /* unlock is done later, after layout lock management */
1393 rc = mdt_hsm_attr_set(mti, obj, &mh);
1396 /* we give back layout lock only if restore was successful or
1397 * if restore was canceled or if policy is to not retry
1398 * in other cases we just unlock the object */
1399 if (car->car_hai->hai_action == HSMA_RESTORE &&
1400 (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1401 cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1402 struct cdt_restore_handle *crh;
1404 /* restore in data FID done, we swap the layouts
1405 * only if restore is successful */
1406 if (pgs->hpk_errval == 0) {
1407 rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1408 &car->car_hai->hai_dfid, &mh);
1410 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1411 *status = ARS_FAILED;
1412 pgs->hpk_errval = -rc;
1415 /* we have to retry, so keep layout lock */
1416 if (*status == ARS_WAITING)
1419 /* give back layout lock */
1420 mutex_lock(&cdt->cdt_restore_lock);
1421 crh = hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1423 list_del(&crh->crh_list);
1424 mutex_unlock(&cdt->cdt_restore_lock);
1425 /* just give back layout lock, we keep
1426 * the reference which is given back
1427 * later with the lock for HSM flags */
1428 if (!IS_ERR(obj) && crh != NULL)
1429 mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1432 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1438 if (obj != NULL && !IS_ERR(obj)) {
1439 mo_changelog(env, CL_HSM, cl_flags,
1440 mdt_object_child(obj));
1441 mdt_object_put(mti->mti_env, obj);
1448 * update status of a request
1449 * \param mti [IN] context
1450 * \param pgs [IN] progress of the copy tool
1451 * \param update_record [IN] update llog record
1453 * \retval -ve failure
1455 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1456 struct hsm_progress_kernel *pgs,
1457 const int update_record)
1459 struct mdt_device *mdt = mti->mti_mdt;
1460 struct coordinator *cdt = &mdt->mdt_coordinator;
1461 struct cdt_agent_req *car;
1465 /* no coordinator started, so we cannot serve requests */
1466 if (cdt->cdt_state == CDT_STOPPED)
1469 /* first do sanity checks */
1470 car = mdt_cdt_update_request(cdt, pgs);
1472 CERROR("%s: Cannot find running request for cookie "LPX64
1475 pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1478 RETURN(PTR_ERR(car));
1481 CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1482 " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1483 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1484 hsm_copytool_action2name(car->car_hai->hai_action),
1485 pgs->hpk_flags, pgs->hpk_errval,
1486 PFID(&car->car_hai->hai_fid),
1487 PFID(&car->car_hai->hai_dfid));
1489 /* progress is done on FID or data FID depending of the action and
1490 * of the copy progress */
1491 /* for restore progress is used to send back the data FID to cdt */
1492 if (car->car_hai->hai_action == HSMA_RESTORE &&
1493 lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1494 car->car_hai->hai_dfid = pgs->hpk_fid;
1496 if ((car->car_hai->hai_action == HSMA_RESTORE ||
1497 car->car_hai->hai_action == HSMA_ARCHIVE) &&
1498 (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1499 !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1500 CERROR("%s: Progress on "DFID" for cookie "LPX64
1501 " does not match request FID "DFID" nor data FID "
1504 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1505 PFID(&car->car_hai->hai_fid),
1506 PFID(&car->car_hai->hai_dfid));
1507 GOTO(out, rc = -EINVAL);
1510 if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1511 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1512 " is not coherent (err=%d and not completed"
1515 PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1516 hsm_copytool_action2name(car->car_hai->hai_action),
1517 pgs->hpk_errval, pgs->hpk_flags);
1518 GOTO(out, rc = -EINVAL);
1521 /* now progress is valid */
1523 /* we use a root like ucred */
1524 hsm_init_ucred(mdt_ucred(mti));
1526 if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1527 enum agent_req_status status;
1529 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1531 /* remove request from memory list */
1532 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1534 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1535 " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1537 hsm_copytool_action2name(car->car_hai->hai_action),
1538 agent_req_status2name(status));
1540 if (update_record) {
1543 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1544 &pgs->hpk_cookie, 1,
1547 CERROR("%s: mdt_agent_record_update() failed,"
1548 " rc=%d, cannot update status to %s"
1549 " for cookie "LPX64"\n",
1550 mdt_obd_name(mdt), rc1,
1551 agent_req_status2name(status),
1553 rc = (rc != 0 ? rc : rc1);
1555 /* ct has completed a request, so a slot is available, wakeup
1556 * cdt to find new work */
1557 mdt_hsm_cdt_wakeup(mdt);
1559 /* if copytool send a progress on a canceled request
1560 * we inform copytool it should stop
1562 if (car->car_canceled == 1)
1568 /* remove ref got from mdt_cdt_update_request() */
1569 mdt_cdt_put_request(car);
1576 * data passed to llog_cat_process() callback
1577 * to cancel requests
1579 struct hsm_cancel_all_data {
1580 struct mdt_device *mdt;
1584 * llog_cat_process() callback, used to:
1585 * - purge all requests
1586 * \param env [IN] environment
1587 * \param llh [IN] llog handle
1588 * \param hdr [IN] llog record
1589 * \param data [IN] cb data = struct hsm_cancel_all_data
1591 * \retval -ve failure
1593 static int mdt_cancel_all_cb(const struct lu_env *env,
1594 struct llog_handle *llh,
1595 struct llog_rec_hdr *hdr, void *data)
1597 struct llog_agent_req_rec *larr;
1598 struct hsm_cancel_all_data *hcad;
1602 larr = (struct llog_agent_req_rec *)hdr;
1604 if (larr->arr_status == ARS_WAITING ||
1605 larr->arr_status == ARS_STARTED) {
1606 larr->arr_status = ARS_CANCELED;
1607 larr->arr_req_change = cfs_time_current_sec();
1608 rc = mdt_agent_llog_update_rec(env, hcad->mdt, llh, larr);
1610 RETURN(LLOG_DEL_RECORD);
1616 * cancel all actions
1617 * \param obd [IN] MDT device
1619 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1621 struct mdt_thread_info *mti;
1622 struct coordinator *cdt = &mdt->mdt_coordinator;
1623 struct cdt_agent_req *car;
1624 struct hsm_action_list *hal = NULL;
1625 struct hsm_action_item *hai;
1626 struct hsm_cancel_all_data hcad;
1627 int hal_sz = 0, hal_len, rc;
1628 enum cdt_states save_state;
1631 /* retrieve coordinator context */
1632 mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1634 /* disable coordinator */
1635 save_state = cdt->cdt_state;
1636 cdt->cdt_state = CDT_DISABLE;
1638 /* send cancel to all running requests */
1639 down_read(&cdt->cdt_request_lock);
1640 list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1641 mdt_cdt_get_request(car);
1642 /* request is not yet removed from list, it will be done
1643 * when copytool will return progress
1646 if (car->car_hai->hai_action == HSMA_CANCEL) {
1647 mdt_cdt_put_request(car);
1652 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1653 cfs_size_round(car->car_hai->hai_len);
1655 if (hal_len > hal_sz && hal_sz > 0) {
1656 /* not enough room, free old buffer */
1657 OBD_FREE(hal, hal_sz);
1661 /* empty buffer, allocate one */
1664 OBD_ALLOC(hal, hal_sz);
1666 mdt_cdt_put_request(car);
1667 up_read(&cdt->cdt_request_lock);
1668 GOTO(out, rc = -ENOMEM);
1672 hal->hal_version = HAL_VERSION;
1673 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1675 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1676 hal->hal_compound_id = car->car_compound_id;
1677 hal->hal_archive_id = car->car_archive_id;
1678 hal->hal_flags = car->car_flags;
1681 hai = hai_first(hal);
1682 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1683 hai->hai_action = HSMA_CANCEL;
1686 /* it is possible to safely call mdt_hsm_agent_send()
1687 * (ie without a deadlock on cdt_request_lock), because the
1688 * write lock is taken only if we are not in purge mode
1689 * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1690 * nor mdt_cdt_remove_request())
1692 /* no conflict with cdt thread because cdt is disable and we
1693 * have the request lock */
1694 mdt_hsm_agent_send(mti, hal, 1);
1696 mdt_cdt_put_request(car);
1698 up_read(&cdt->cdt_request_lock);
1701 OBD_FREE(hal, hal_sz);
1703 /* cancel all on-disk records */
1706 rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1707 mdt_cancel_all_cb, &hcad);
1709 /* enable coordinator */
1710 cdt->cdt_state = save_state;
1716 * check if a request is compatible with file status
1717 * \param hai [IN] request description
1718 * \param hal_an [IN] request archive number (not used)
1719 * \param rq_flags [IN] request flags
1720 * \param hsm [IN] file HSM metadata
1723 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1724 const int hal_an, const __u64 rq_flags,
1725 const struct md_hsm *hsm)
1727 int is_compat = false;
1731 hsm_flags = hsm->mh_flags;
1732 switch (hai->hai_action) {
1734 if (!(hsm_flags & HS_NOARCHIVE) &&
1735 (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1739 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1740 hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1744 if (!(hsm_flags & HS_RELEASED) &&
1745 (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1752 CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1753 " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1754 PFID(&hai->hai_fid),
1755 hsm_copytool_action2name(hai->hai_action), rq_flags,
1756 hai->hai_extent.offset, hai->hai_extent.length,
1758 (is_compat ? "compatible" : "uncompatible"));
1764 * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1766 static const struct {
1770 } hsm_policy_names[] = {
1771 { CDT_NONBLOCKING_RESTORE, "NonBlockingRestore", "NBR"},
1772 { CDT_NORETRY_ACTION, "NoRetryAction", "NRA"},
1777 * convert a policy name to a bit
1778 * \param name [IN] policy name
1780 * \retval policy bit
1782 static __u64 hsm_policy_str2bit(const char *name)
1786 for (i = 0; hsm_policy_names[i].bit != 0; i++)
1787 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1788 strcmp(hsm_policy_names[i].name, name) == 0)
1789 return hsm_policy_names[i].bit;
1794 * convert a policy bit field to a string
1795 * \param mask [IN] policy bit field
1796 * \param hexa [IN] print mask before bit names
1797 * \param buffer [OUT] string
1798 * \param count [IN] size of buffer
1800 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1808 seq_printf(m, "("LPX64") ", mask);
1810 for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1813 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1814 if (hsm_policy_names[j].bit == bit)
1818 seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1820 seq_printf(m, "%s ", hsm_policy_names[j].name);
1822 /* remove last ' ' */
1827 /* methods to read/write HSM policy flags */
1828 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1830 struct mdt_device *mdt = m->private;
1831 struct coordinator *cdt = &mdt->mdt_coordinator;
1834 hsm_policy_bit2str(m, cdt->cdt_policy, false);
1839 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1840 size_t count, loff_t *off)
1842 struct seq_file *m = file->private_data;
1843 struct mdt_device *mdt = m->private;
1844 struct coordinator *cdt = &mdt->mdt_coordinator;
1845 char *start, *token, sign;
1848 __u64 add_mask, remove_mask, set_mask;
1852 if (count + 1 > PAGE_SIZE)
1855 OBD_ALLOC(buf, count + 1);
1859 if (copy_from_user(buf, buffer, count))
1860 GOTO(out, rc = -EFAULT);
1865 CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1868 add_mask = remove_mask = set_mask = 0;
1870 token = strsep(&start, "\n ");
1876 if (sign == '-' || sign == '+')
1879 policy = hsm_policy_str2bit(token);
1881 CWARN("%s: '%s' is unknown, "
1882 "supported policies are:\n", mdt_obd_name(mdt),
1884 hsm_policy_bit2str(m, 0, false);
1885 GOTO(out, rc = -EINVAL);
1889 remove_mask |= policy;
1899 } while (start != NULL);
1901 CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1902 mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1904 /* if no sign in all string, it is a clear and set
1905 * if some sign found, all unsigned are converted
1907 * P1 P2 = set to P1 and P2
1908 * P1 -P2 = add P1 clear P2 same as +P1 -P2
1910 if (remove_mask == 0 && add_mask == 0) {
1911 cdt->cdt_policy = set_mask;
1913 cdt->cdt_policy |= set_mask | add_mask;
1914 cdt->cdt_policy &= ~remove_mask;
1917 GOTO(out, rc = count);
1920 OBD_FREE(buf, count + 1);
1923 LPROC_SEQ_FOPS(mdt_hsm_policy);
1925 #define GENERATE_PROC_METHOD(VAR) \
1926 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data) \
1928 struct mdt_device *mdt = m->private; \
1929 struct coordinator *cdt = &mdt->mdt_coordinator; \
1932 seq_printf(m, LPU64"\n", (__u64)cdt->VAR); \
1936 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1937 size_t count, loff_t *off) \
1940 struct seq_file *m = file->private_data; \
1941 struct mdt_device *mdt = m->private; \
1942 struct coordinator *cdt = &mdt->mdt_coordinator; \
1947 rc = lprocfs_write_helper(buffer, count, &val); \
1957 GENERATE_PROC_METHOD(cdt_loop_period)
1958 GENERATE_PROC_METHOD(cdt_grace_delay)
1959 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1960 GENERATE_PROC_METHOD(cdt_max_requests)
1961 GENERATE_PROC_METHOD(cdt_default_archive_id)
1964 * procfs write method for MDT/hsm_control
1965 * proc entry is in mdt directory so data is mdt obd_device pointer
1967 #define CDT_ENABLE_CMD "enabled"
1968 #define CDT_STOP_CMD "shutdown"
1969 #define CDT_DISABLE_CMD "disabled"
1970 #define CDT_PURGE_CMD "purge"
1971 #define CDT_HELP_CMD "help"
1972 #define CDT_MAX_CMD_LEN 10
1975 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1976 size_t count, loff_t *off)
1978 struct seq_file *m = file->private_data;
1979 struct obd_device *obd = m->private;
1980 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1981 struct coordinator *cdt = &(mdt->mdt_coordinator);
1983 char kernbuf[CDT_MAX_CMD_LEN];
1986 if (count == 0 || count >= sizeof(kernbuf))
1989 if (copy_from_user(kernbuf, buffer, count))
1993 if (kernbuf[count - 1] == '\n')
1994 kernbuf[count - 1] = 0;
1997 if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1998 if (cdt->cdt_state == CDT_DISABLE) {
1999 cdt->cdt_state = CDT_RUNNING;
2000 mdt_hsm_cdt_wakeup(mdt);
2002 rc = mdt_hsm_cdt_start(mdt);
2004 } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2005 if ((cdt->cdt_state == CDT_STOPPING) ||
2006 (cdt->cdt_state == CDT_STOPPED)) {
2007 CERROR("%s: Coordinator already stopped\n",
2011 cdt->cdt_state = CDT_STOPPING;
2013 } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2014 if ((cdt->cdt_state == CDT_STOPPING) ||
2015 (cdt->cdt_state == CDT_STOPPED)) {
2016 CERROR("%s: Coordinator is stopped\n",
2020 cdt->cdt_state = CDT_DISABLE;
2022 } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2023 rc = hsm_cancel_all_actions(mdt);
2024 } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2032 CERROR("%s: Valid coordinator control commands are: "
2033 "%s %s %s %s %s\n", mdt_obd_name(mdt),
2034 CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2035 CDT_PURGE_CMD, CDT_HELP_CMD);
2043 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2045 struct obd_device *obd = m->private;
2046 struct coordinator *cdt;
2049 cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2051 if (cdt->cdt_state == CDT_INIT)
2052 seq_printf(m, "init\n");
2053 else if (cdt->cdt_state == CDT_RUNNING)
2054 seq_printf(m, "enabled\n");
2055 else if (cdt->cdt_state == CDT_STOPPING)
2056 seq_printf(m, "stopping\n");
2057 else if (cdt->cdt_state == CDT_STOPPED)
2058 seq_printf(m, "stopped\n");
2059 else if (cdt->cdt_state == CDT_DISABLE)
2060 seq_printf(m, "disabled\n");
2062 seq_printf(m, "unknown\n");
2068 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2074 for (i = 0; i < 8 * sizeof(mask); i++) {
2075 if (mask & (1UL << i)) {
2076 seq_printf(m, "%s%s", first ? "" : " ",
2077 hsm_copytool_action2name(i));
2087 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2089 struct mdt_device *mdt = m->private;
2090 struct coordinator *cdt = &mdt->mdt_coordinator;
2092 return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2096 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2098 struct mdt_device *mdt = m->private;
2099 struct coordinator *cdt = &mdt->mdt_coordinator;
2101 return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2105 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2107 struct mdt_device *mdt = m->private;
2108 struct coordinator *cdt = &mdt->mdt_coordinator;
2110 return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2113 static inline enum hsm_copytool_action
2114 hsm_copytool_name2action(const char *name)
2116 if (strcasecmp(name, "NOOP") == 0)
2118 else if (strcasecmp(name, "ARCHIVE") == 0)
2119 return HSMA_ARCHIVE;
2120 else if (strcasecmp(name, "RESTORE") == 0)
2121 return HSMA_RESTORE;
2122 else if (strcasecmp(name, "REMOVE") == 0)
2124 else if (strcasecmp(name, "CANCEL") == 0)
2131 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2132 size_t user_count, __u64 *mask)
2134 char *buf, *pos, *name;
2140 if (!(user_count < 4096))
2143 buf_size = user_count + 1;
2145 OBD_ALLOC(buf, buf_size);
2149 if (copy_from_user(buf, user_buf, buf_size - 1))
2150 GOTO(out, rc = -EFAULT);
2152 buf[buf_size - 1] = '\0';
2155 while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2161 action = hsm_copytool_name2action(name);
2163 GOTO(out, rc = -EINVAL);
2165 new_mask |= (1UL << action);
2171 OBD_FREE(buf, buf_size);
2177 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2178 size_t count, loff_t *off)
2180 struct seq_file *m = file->private_data;
2181 struct mdt_device *mdt = m->private;
2182 struct coordinator *cdt = &mdt->mdt_coordinator;
2184 return mdt_write_hsm_request_mask(file, buf, count,
2185 &cdt->cdt_user_request_mask);
2189 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2190 size_t count, loff_t *off)
2192 struct seq_file *m = file->private_data;
2193 struct mdt_device *mdt = m->private;
2194 struct coordinator *cdt = &mdt->mdt_coordinator;
2196 return mdt_write_hsm_request_mask(file, buf, count,
2197 &cdt->cdt_group_request_mask);
2201 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2202 size_t count, loff_t *off)
2204 struct seq_file *m = file->private_data;
2205 struct mdt_device *mdt = m->private;
2206 struct coordinator *cdt = &mdt->mdt_coordinator;
2208 return mdt_write_hsm_request_mask(file, buf, count,
2209 &cdt->cdt_other_request_mask);
2212 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2213 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2214 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2215 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2216 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2217 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2218 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2219 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2221 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2223 .fops = &mdt_hsm_agent_fops },
2224 { .name = "actions",
2225 .fops = &mdt_hsm_actions_fops,
2226 .proc_mode = 0444 },
2227 { .name = "default_archive_id",
2228 .fops = &mdt_hsm_cdt_default_archive_id_fops },
2229 { .name = "grace_delay",
2230 .fops = &mdt_hsm_cdt_grace_delay_fops },
2231 { .name = "loop_period",
2232 .fops = &mdt_hsm_cdt_loop_period_fops },
2233 { .name = "max_requests",
2234 .fops = &mdt_hsm_cdt_max_requests_fops },
2236 .fops = &mdt_hsm_policy_fops },
2237 { .name = "active_request_timeout",
2238 .fops = &mdt_hsm_cdt_active_req_timeout_fops },
2239 { .name = "active_requests",
2240 .fops = &mdt_hsm_active_requests_fops },
2241 { .name = "user_request_mask",
2242 .fops = &mdt_hsm_user_request_mask_fops, },
2243 { .name = "group_request_mask",
2244 .fops = &mdt_hsm_group_request_mask_fops, },
2245 { .name = "other_request_mask",
2246 .fops = &mdt_hsm_other_request_mask_fops, },