Whamcloud - gitweb
LU-3756 mdt: Change HSM policy display
[fs/lustre-release.git] / lustre / mdt / mdt_coordinator.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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.
15  *
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2012, 2013, Intel Corporation.
24  * Use is subject to license terms.
25  * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
26  *                          alternatives
27  */
28 /*
29  * lustre/mdt/mdt_coordinator.c
30  *
31  * Lustre HSM Coordinator
32  *
33  * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
34  * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
35  * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
36  */
37
38 #define DEBUG_SUBSYSTEM S_MDS
39
40 #include <obd_support.h>
41 #include <lustre_net.h>
42 #include <lustre_export.h>
43 #include <obd.h>
44 #include <obd_lov.h>
45 #include <lprocfs_status.h>
46 #include <lustre_log.h>
47 #include "mdt_internal.h"
48
49 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
50
51 /**
52  * get obj and HSM attributes on a fid
53  * \param mti [IN] context
54  * \param fid [IN] object fid
55  * \param hsm [OUT] HSM meta data
56  * \retval obj or error (-ENOENT if not found)
57  */
58 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
59                                       const struct lu_fid *fid,
60                                       struct md_hsm *hsm)
61 {
62         struct md_attr          *ma;
63         struct mdt_object       *obj;
64         int                      rc;
65         ENTRY;
66
67         ma = &mti->mti_attr;
68         ma->ma_need = MA_HSM;
69         ma->ma_valid = 0;
70
71         /* find object by FID */
72         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
73         if (IS_ERR(obj))
74                 RETURN(obj);
75
76         if (!mdt_object_exists(obj)) {
77                 /* no more object */
78                 mdt_object_put(mti->mti_env, obj);
79                 RETURN(ERR_PTR(-ENOENT));
80         }
81
82         rc = mdt_attr_get_complex(mti, obj, ma);
83         if (rc) {
84                 mdt_object_put(mti->mti_env, obj);
85                 RETURN(ERR_PTR(rc));
86         }
87
88         if (ma->ma_valid & MA_HSM)
89                 *hsm = ma->ma_hsm;
90         else
91                 memset(hsm, 0, sizeof(*hsm));
92         ma->ma_valid = 0;
93         RETURN(obj);
94 }
95
96 void mdt_hsm_dump_hal(int level, const char *prefix,
97                       struct hsm_action_list *hal)
98 {
99         int                      i, sz;
100         struct hsm_action_item  *hai;
101         char                     buf[12];
102
103         CDEBUG(level, "%s: HAL header: version %X count %d compound "LPX64
104                       " archive_id %d flags "LPX64"\n",
105                prefix, hal->hal_version, hal->hal_count,
106                hal->hal_compound_id, hal->hal_archive_id, hal->hal_flags);
107
108         hai = hai_first(hal);
109         for (i = 0; i < hal->hal_count; i++) {
110                 sz = hai->hai_len - sizeof(*hai);
111                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
112                        " compound/cookie="LPX64"/"LPX64
113                        " action=%s extent="LPX64"-"LPX64" gid="LPX64
114                        " datalen=%d data=[%s]\n",
115                        prefix, i,
116                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
117                        hal->hal_compound_id, hai->hai_cookie,
118                        hsm_copytool_action2name(hai->hai_action),
119                        hai->hai_extent.offset,
120                        hai->hai_extent.length,
121                        hai->hai_gid, sz,
122                        hai_dump_data_field(hai, buf, sizeof(buf)));
123                 hai = hai_next(hai);
124         }
125 }
126
127 /**
128  * data passed to llog_cat_process() callback
129  * to scan requests and take actions
130  */
131 struct hsm_scan_data {
132         struct mdt_thread_info          *mti;
133         char                             fs_name[MTI_NAME_MAXLEN+1];
134         /* request to be send to agents */
135         int                              request_sz;    /** allocated size */
136         int                              max_request;   /** vector size */
137         int                              request_cnt;   /** used count */
138         struct {
139                 int                      hal_sz;
140                 int                      hal_used_sz;
141                 struct hsm_action_list  *hal;
142         } *request;
143         /* records to be canceled */
144         int                              max_cookie;    /** vector size */
145         int                              cookie_cnt;    /** used count */
146         __u64                           *cookies;
147 };
148
149 /**
150  *  llog_cat_process() callback, used to:
151  *  - find waiting request and start action
152  *  - purge canceled and done requests
153  * \param env [IN] environment
154  * \param llh [IN] llog handle
155  * \param hdr [IN] llog record
156  * \param data [IN/OUT] cb data = struct hsm_scan_data
157  * \retval 0 success
158  * \retval -ve failure
159  */
160 static int mdt_coordinator_cb(const struct lu_env *env,
161                               struct llog_handle *llh,
162                               struct llog_rec_hdr *hdr,
163                               void *data)
164 {
165         const struct llog_agent_req_rec *larr;
166         struct hsm_scan_data            *hsd;
167         struct hsm_action_item          *hai;
168         struct mdt_device               *mdt;
169         struct coordinator              *cdt;
170         int                              rc;
171         ENTRY;
172
173         hsd = data;
174         mdt = hsd->mti->mti_mdt;
175         cdt = &mdt->mdt_coordinator;
176
177         larr = (struct llog_agent_req_rec *)hdr;
178         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
179         switch (larr->arr_status) {
180         case ARS_WAITING: {
181                 int i, empty_slot, found;
182
183                 /* Are agents full? */
184                 if (atomic_read(&cdt->cdt_request_count) ==
185                     cdt->cdt_max_request)
186                         break;
187
188                 /* first search if the request if known in the list we have
189                  * build and if there is room in the request vector */
190                 empty_slot = -1;
191                 found = -1;
192                 for (i = 0; i < hsd->max_request &&
193                             (empty_slot == -1 || found == -1); i++) {
194                         if (hsd->request[i].hal == NULL) {
195                                 empty_slot = i;
196                                 continue;
197                         }
198                         if (hsd->request[i].hal->hal_compound_id ==
199                                 larr->arr_compound_id) {
200                                 found = i;
201                                 continue;
202                         }
203                 }
204                 if (found == -1 && empty_slot == -1)
205                         /* unknown request and no more room for new request,
206                          * continue scan for to find other entries for
207                          * already found request
208                          */
209                         RETURN(0);
210
211                 if (found == -1) {
212                         struct hsm_action_list *hal;
213
214                         /* request is not already known */
215                         /* allocates hai vector size just needs to be large
216                          * enough */
217                         hsd->request[empty_slot].hal_sz =
218                                      sizeof(*hsd->request[empty_slot].hal) +
219                                      cfs_size_round(MTI_NAME_MAXLEN+1) +
220                                      2 * cfs_size_round(larr->arr_hai.hai_len);
221                         OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
222                         if (!hal) {
223                                 CERROR("%s: Cannot allocate memory (%d o)"
224                                        "for compound "LPX64"\n",
225                                        mdt_obd_name(mdt),
226                                        hsd->request[i].hal_sz,
227                                        larr->arr_compound_id);
228                                 RETURN(-ENOMEM);
229                         }
230                         hal->hal_version = HAL_VERSION;
231                         strncpy(hal->hal_fsname, hsd->fs_name,
232                                 MTI_NAME_MAXLEN);
233                         hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
234                         hal->hal_compound_id = larr->arr_compound_id;
235                         hal->hal_archive_id = larr->arr_archive_id;
236                         hal->hal_flags = larr->arr_flags;
237                         hal->hal_count = 0;
238                         hsd->request[empty_slot].hal_used_sz = hal_size(hal);
239                         hsd->request[empty_slot].hal = hal;
240                         hsd->request_cnt++;
241                         found = empty_slot;
242                         hai = hai_first(hal);
243                 } else {
244                         /* request is known */
245                         /* we check if record archive num is the same as the
246                          * known request, if not we will serve it in multiple
247                          * time because we do not know if the agent can serve
248                          * multiple backend
249                          * a use case is a compound made of multiple restore
250                          * where the files are not archived in the same backend
251                          */
252                         if (larr->arr_archive_id !=
253                             hsd->request[found].hal->hal_archive_id)
254                                 RETURN(0);
255
256                         if (hsd->request[found].hal_sz <
257                             hsd->request[found].hal_used_sz +
258                              cfs_size_round(larr->arr_hai.hai_len)) {
259                                 /* Not enough room, need an extension */
260                                 void *hal_buffer;
261                                 int sz;
262
263                                 sz = 2 * hsd->request[found].hal_sz;
264                                 OBD_ALLOC(hal_buffer, sz);
265                                 if (!hal_buffer) {
266                                         CERROR("%s: Cannot allocate memory "
267                                                "(%d o) for compound "LPX64"\n",
268                                                mdt_obd_name(mdt), sz,
269                                                larr->arr_compound_id);
270                                         RETURN(-ENOMEM);
271                                 }
272                                 memcpy(hal_buffer, hsd->request[found].hal,
273                                        hsd->request[found].hal_used_sz);
274                                 OBD_FREE(hsd->request[found].hal,
275                                          hsd->request[found].hal_sz);
276                                 hsd->request[found].hal = hal_buffer;
277                                 hsd->request[found].hal_sz = sz;
278                         }
279                         hai = hai_first(hsd->request[found].hal);
280                         for (i = 0; i < hsd->request[found].hal->hal_count;
281                              i++)
282                                 hai = hai_next(hai);
283                 }
284                 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
285                 hai->hai_cookie = larr->arr_hai.hai_cookie;
286                 hai->hai_gid = larr->arr_hai.hai_gid;
287
288                 hsd->request[found].hal_used_sz +=
289                                                    cfs_size_round(hai->hai_len);
290                 hsd->request[found].hal->hal_count++;
291                 break;
292         }
293         case ARS_STARTED: {
294                 struct cdt_agent_req *car;
295                 cfs_time_t last;
296
297                 /* we search for a running request
298                  * error may happen if coordinator crashes or stopped
299                  * with running request
300                  */
301                 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
302                 if (car == NULL) {
303                         last = larr->arr_req_create;
304                 } else {
305                         last = car->car_req_update;
306                         mdt_cdt_put_request(car);
307                 }
308
309                 /* test if request too long, if yes cancel it
310                  * the same way the copy tool acknowledge a cancel request */
311                 if ((last + cdt->cdt_timeout) < cfs_time_current_sec()) {
312                         struct hsm_progress_kernel pgs;
313
314                         dump_llog_agent_req_rec("mdt_coordinator_cb(): "
315                                                 "request timeouted, start "
316                                                 "cleaning", larr);
317                         /* a too old cancel request just needs to be removed
318                          * this can happen, if copy tool does not support cancel
319                          * for other requests, we have to remove the running
320                          * request and notify the copytool
321                          */
322                         pgs.hpk_fid = larr->arr_hai.hai_fid;
323                         pgs.hpk_cookie = larr->arr_hai.hai_cookie;
324                         pgs.hpk_extent = larr->arr_hai.hai_extent;
325                         pgs.hpk_flags = HP_FLAG_COMPLETED;
326                         pgs.hpk_errval = ENOSYS;
327                         pgs.hpk_data_version = 0;
328                         /* update request state, but do not record in llog, to
329                          * avoid deadlock on cdt_llog_lock
330                          */
331                         rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
332                         if (rc)
333                                 CERROR("%s: Cannot cleanup timeouted request: "
334                                        DFID" for cookie "LPX64" action=%s\n",
335                                        mdt_obd_name(mdt),
336                                        PFID(&pgs.hpk_fid), pgs.hpk_cookie,
337                                        hsm_copytool_action2name(
338                                                      larr->arr_hai.hai_action));
339
340                         /* add the cookie to the list of record to be
341                          * canceled by caller */
342                         if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
343                                 __u64 *ptr, *old_ptr;
344                                 int old_sz, new_sz, new_cnt;
345
346                                 /* need to increase vector size */
347                                 old_sz = sizeof(__u64) * hsd->max_cookie;
348                                 old_ptr = hsd->cookies;
349
350                                 new_cnt = 2 * hsd->max_cookie;
351                                 new_sz = sizeof(__u64) * new_cnt;
352
353                                 OBD_ALLOC(ptr, new_sz);
354                                 if (!ptr) {
355                                         CERROR("%s: Cannot allocate memory "
356                                                "(%d o) for cookie vector\n",
357                                                mdt_obd_name(mdt), new_sz);
358                                         RETURN(-ENOMEM);
359                                 }
360                                 memcpy(ptr, hsd->cookies, old_sz);
361                                 hsd->cookies = ptr;
362                                 hsd->max_cookie = new_cnt;
363                                 OBD_FREE(old_ptr, old_sz);
364                         }
365                         hsd->cookies[hsd->cookie_cnt] =
366                                                        larr->arr_hai.hai_cookie;
367                         hsd->cookie_cnt++;
368                 }
369                 break;
370         }
371         case ARS_FAILED:
372         case ARS_CANCELED:
373         case ARS_SUCCEED:
374                 if ((larr->arr_req_change + cdt->cdt_delay) <
375                     cfs_time_current_sec())
376                         RETURN(LLOG_DEL_RECORD);
377                 break;
378         }
379         RETURN(0);
380 }
381
382 /**
383  * create /proc entries for coordinator
384  * \param mdt [IN]
385  * \retval 0 success
386  * \retval -ve failure
387  */
388 static int hsm_cdt_procfs_init(struct mdt_device *mdt)
389 {
390         struct coordinator      *cdt = &mdt->mdt_coordinator;
391         int                      rc = 0;
392         ENTRY;
393
394         /* init /proc entries, failure is not critical */
395         cdt->cdt_proc_dir = lprocfs_register("hsm",
396                                              mdt2obd_dev(mdt)->obd_proc_entry,
397                                              lprocfs_mdt_hsm_vars, mdt);
398         if (IS_ERR(cdt->cdt_proc_dir)) {
399                 rc = PTR_ERR(cdt->cdt_proc_dir);
400                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
401                        " rc=%d\n", mdt_obd_name(mdt), rc);
402                 cdt->cdt_proc_dir = NULL;
403                 RETURN(rc);
404         }
405
406         RETURN(0);
407 }
408
409 /**
410  * coordinator thread
411  * \param data [IN] obd device
412  * \retval 0 success
413  * \retval -ve failure
414  */
415 static int mdt_coordinator(void *data)
416 {
417         struct mdt_thread_info  *mti = data;
418         struct mdt_device       *mdt = mti->mti_mdt;
419         struct coordinator      *cdt = &mdt->mdt_coordinator;
420         struct hsm_scan_data     hsd = { 0 };
421         int                      rc = 0;
422         ENTRY;
423
424         cdt->cdt_thread.t_flags = SVC_RUNNING;
425         wake_up(&cdt->cdt_thread.t_ctl_waitq);
426
427         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
428                mdt_obd_name(mdt), current_pid());
429
430         /*
431          * create /proc entries for coordinator
432          */
433         hsm_cdt_procfs_init(mdt);
434         /* timeouted cookie vector initialization */
435         hsd.max_cookie = 0;
436         hsd.cookie_cnt = 0;
437         hsd.cookies = NULL;
438         /* we use a copy of cdt_max_request in the cb, so if cdt_max_request
439          * increases due to a change from /proc we do not overflow the
440          * hsd.request[] vector
441          */
442         hsd.max_request = cdt->cdt_max_request;
443         hsd.request_sz = hsd.max_request * sizeof(*hsd.request);
444         OBD_ALLOC(hsd.request, hsd.request_sz);
445         if (!hsd.request)
446                 GOTO(out, rc = -ENOMEM);
447
448         hsd.mti = mti;
449         obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
450
451         while (1) {
452                 struct l_wait_info lwi;
453                 int i;
454
455                 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
456                                   NULL, NULL);
457                 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
458                              (cdt->cdt_thread.t_flags &
459                               (SVC_STOPPING|SVC_EVENT)),
460                              &lwi);
461
462                 CDEBUG(D_HSM, "coordinator resumes\n");
463
464                 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
465                     cdt->cdt_state == CDT_STOPPING) {
466                         cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
467                         rc = 0;
468                         break;
469                 }
470
471                 /* wake up before timeout, new work arrives */
472                 if (cdt->cdt_thread.t_flags & SVC_EVENT)
473                         cdt->cdt_thread.t_flags &= ~SVC_EVENT;
474
475                 /* if coordinator is suspended continue to wait */
476                 if (cdt->cdt_state == CDT_DISABLE) {
477                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
478                         continue;
479                 }
480
481                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
482
483                 if (hsd.max_request != cdt->cdt_max_request) {
484                         /* cdt_max_request has changed,
485                          * we need to allocate a new buffer
486                          */
487                         OBD_FREE(hsd.request, hsd.request_sz);
488                         hsd.max_request = cdt->cdt_max_request;
489                         hsd.request_sz =
490                                    hsd.max_request * sizeof(*hsd.request);
491                         OBD_ALLOC(hsd.request, hsd.request_sz);
492                         if (!hsd.request) {
493                                 rc = -ENOMEM;
494                                 break;
495                         }
496                 }
497
498                 /* create canceled cookie vector for an arbitrary size
499                  * if needed, vector will grow during llog scan
500                  */
501                 hsd.max_cookie = 10;
502                 hsd.cookie_cnt = 0;
503                 OBD_ALLOC(hsd.cookies, hsd.max_cookie * sizeof(__u64));
504                 if (!hsd.cookies) {
505                         rc = -ENOMEM;
506                         goto clean_cb_alloc;
507                 }
508                 hsd.request_cnt = 0;
509
510                 rc = cdt_llog_process(mti->mti_env, mdt,
511                                       mdt_coordinator_cb, &hsd);
512                 if (rc < 0)
513                         goto clean_cb_alloc;
514
515                 CDEBUG(D_HSM, "Found %d requests to send and %d"
516                               " requests to cancel\n",
517                        hsd.request_cnt, hsd.cookie_cnt);
518                 /* first we cancel llog records of the timeouted requests */
519                 if (hsd.cookie_cnt > 0) {
520                         rc = mdt_agent_record_update(mti->mti_env, mdt,
521                                                      hsd.cookies,
522                                                      hsd.cookie_cnt,
523                                                      ARS_CANCELED);
524                         if (rc)
525                                 CERROR("%s: mdt_agent_record_update() failed, "
526                                        "rc=%d, cannot update status to %s "
527                                        "for %d cookies\n",
528                                        mdt_obd_name(mdt), rc,
529                                        agent_req_status2name(ARS_CANCELED),
530                                        hsd.cookie_cnt);
531                 }
532
533                 if (list_empty(&cdt->cdt_agents)) {
534                         CDEBUG(D_HSM, "no agent available, "
535                                       "coordinator sleeps\n");
536                         goto clean_cb_alloc;
537                 }
538
539                 /* here hsd contains a list of requests to be started */
540                 for (i = 0; i < hsd.max_request; i++) {
541                         struct hsm_action_list  *hal;
542                         struct hsm_action_item  *hai;
543                         __u64                   *cookies;
544                         int                      sz, j;
545                         enum agent_req_status    status;
546
547                         /* still room for work ? */
548                         if (atomic_read(&cdt->cdt_request_count) ==
549                             cdt->cdt_max_request)
550                                 break;
551
552                         if (hsd.request[i].hal == NULL)
553                                 continue;
554
555                         /* found a request, we start it */
556                         /* kuc payload allocation so we avoid an additionnal
557                          * allocation in mdt_hsm_agent_send()
558                          */
559                         hal = kuc_alloc(hsd.request[i].hal_used_sz,
560                                         KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
561                         if (IS_ERR(hal)) {
562                                 CERROR("%s: Cannot allocate memory (%d o) "
563                                        "for compound "LPX64"\n",
564                                        mdt_obd_name(mdt),
565                                        hsd.request[i].hal_used_sz,
566                                        hsd.request[i].hal->hal_compound_id);
567                                 continue;
568                         }
569                         memcpy(hal, hsd.request[i].hal,
570                                hsd.request[i].hal_used_sz);
571
572                         rc = mdt_hsm_agent_send(mti, hal, 0);
573                         /* if failure, we suppose it is temporary
574                          * if the copy tool failed to do the request
575                          * it has to use hsm_progress
576                          */
577                         status = (rc ? ARS_WAITING : ARS_STARTED);
578
579                         /* set up cookie vector to set records status
580                          * after copy tools start or failed
581                          */
582                         sz = hsd.request[i].hal->hal_count * sizeof(__u64);
583                         OBD_ALLOC(cookies, sz);
584                         if (cookies == NULL) {
585                                 CERROR("%s: Cannot allocate memory (%d o) "
586                                        "for cookies vector "LPX64"\n",
587                                        mdt_obd_name(mdt), sz,
588                                        hsd.request[i].hal->hal_compound_id);
589                                 kuc_free(hal, hsd.request[i].hal_used_sz);
590                                 continue;
591                         }
592                         hai = hai_first(hal);
593                         for (j = 0; j < hsd.request[i].hal->hal_count; j++) {
594                                 cookies[j] = hai->hai_cookie;
595                                 hai = hai_next(hai);
596                         }
597
598                         rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
599                                                 hsd.request[i].hal->hal_count,
600                                                 status);
601                         if (rc)
602                                 CERROR("%s: mdt_agent_record_update() failed, "
603                                        "rc=%d, cannot update status to %s "
604                                        "for %d cookies\n",
605                                        mdt_obd_name(mdt), rc,
606                                        agent_req_status2name(status),
607                                        hsd.request[i].hal->hal_count);
608
609                         OBD_FREE(cookies, sz);
610                         kuc_free(hal, hsd.request[i].hal_used_sz);
611                 }
612 clean_cb_alloc:
613                 /* free cookie vector allocated for/by callback */
614                 if (hsd.cookies) {
615                         OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
616                         hsd.max_cookie = 0;
617                         hsd.cookie_cnt = 0;
618                         hsd.cookies = NULL;
619                 }
620
621                 /* free hal allocated by callback */
622                 for (i = 0; i < hsd.max_request; i++) {
623                         if (hsd.request[i].hal) {
624                                 OBD_FREE(hsd.request[i].hal,
625                                          hsd.request[i].hal_sz);
626                                 hsd.request[i].hal_sz = 0;
627                                 hsd.request[i].hal = NULL;
628                                 hsd.request_cnt--;
629                         }
630                 }
631                 LASSERT(hsd.request_cnt == 0);
632
633                 /* reset callback data */
634                 memset(hsd.request, 0, hsd.request_sz);
635         }
636         EXIT;
637 out:
638         if (hsd.request)
639                 OBD_FREE(hsd.request, hsd.request_sz);
640
641         if (hsd.cookies)
642                 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
643
644         if (cdt->cdt_state == CDT_STOPPING) {
645                 /* request comes from /proc path, so we need to clean cdt
646                  * struct */
647                  mdt_hsm_cdt_stop(mdt);
648                  mdt->mdt_opts.mo_coordinator = 0;
649         } else {
650                 /* request comes from a thread event, generated
651                  * by mdt_stop_coordinator(), we have to ack
652                  * and cdt cleaning will be done by event sender
653                  */
654                 cdt->cdt_thread.t_flags = SVC_STOPPED;
655                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
656         }
657
658         if (rc != 0)
659                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
660                        mdt_obd_name(mdt), current_pid(), rc);
661         else
662                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
663                               " no error\n",
664                        mdt_obd_name(mdt), current_pid());
665
666         return rc;
667 }
668
669 /**
670  * lookup a restore handle by FID
671  * caller needs to hold cdt_restore_lock
672  * \param cdt [IN] coordinator
673  * \param fid [IN] FID
674  * \retval cdt_restore_handle found
675  * \retval NULL not found
676  */
677 static struct cdt_restore_handle *hsm_restore_hdl_find(struct coordinator *cdt,
678                                                        const struct lu_fid *fid)
679 {
680         struct cdt_restore_handle       *crh;
681         ENTRY;
682
683         list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
684                 if (lu_fid_eq(&crh->crh_fid, fid))
685                         RETURN(crh);
686         }
687         RETURN(NULL);
688 }
689
690 /**
691  * data passed to llog_cat_process() callback
692  * to scan requests and take actions
693  */
694 struct hsm_restore_data {
695         struct mdt_thread_info  *hrd_mti;
696 };
697
698 /**
699  *  llog_cat_process() callback, used to:
700  *  - find restore request and allocate the restore handle
701  * \param env [IN] environment
702  * \param llh [IN] llog handle
703  * \param hdr [IN] llog record
704  * \param data [IN/OUT] cb data = struct hsm_restore_data
705  * \retval 0 success
706  * \retval -ve failure
707  */
708 static int hsm_restore_cb(const struct lu_env *env,
709                           struct llog_handle *llh,
710                           struct llog_rec_hdr *hdr, void *data)
711 {
712         struct llog_agent_req_rec       *larr;
713         struct hsm_restore_data         *hrd;
714         struct cdt_restore_handle       *crh;
715         struct hsm_action_item          *hai;
716         struct mdt_thread_info          *mti;
717         struct coordinator              *cdt;
718         struct mdt_object               *child;
719         int rc;
720         ENTRY;
721
722         hrd = data;
723         mti = hrd->hrd_mti;
724         cdt = &mti->mti_mdt->mdt_coordinator;
725
726         larr = (struct llog_agent_req_rec *)hdr;
727         hai = &larr->arr_hai;
728         if (hai->hai_cookie > cdt->cdt_last_cookie)
729                 /* update the cookie to avoid collision */
730                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
731
732         if (hai->hai_action != HSMA_RESTORE ||
733             agent_req_in_final_state(larr->arr_status))
734                 RETURN(0);
735
736         /* restore request not in a final state */
737
738         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
739         if (crh == NULL)
740                 RETURN(-ENOMEM);
741
742         crh->crh_fid = hai->hai_fid;
743         /* in V1 all file is restored
744         crh->extent.start = hai->hai_extent.offset;
745         crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
746         */
747         crh->crh_extent.start = 0;
748         crh->crh_extent.end = hai->hai_extent.length;
749         /* get the layout lock */
750         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
751         child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
752                                      MDS_INODELOCK_LAYOUT);
753         if (IS_ERR(child))
754                 GOTO(out, rc = PTR_ERR(child));
755
756         rc = 0;
757         /* we choose to not keep a reference
758          * on the object during the restore time which can be very long */
759         mdt_object_put(mti->mti_env, child);
760
761         mutex_lock(&cdt->cdt_restore_lock);
762         list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
763         mutex_unlock(&cdt->cdt_restore_lock);
764
765 out:
766         RETURN(rc);
767 }
768
769 /**
770  * restore coordinator state at startup
771  * the goal is to take a layout lock for each registered restore request
772  * \param mti [IN] context
773  */
774 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
775 {
776         struct hsm_restore_data  hrd;
777         int                      rc;
778         ENTRY;
779
780         hrd.hrd_mti = mti;
781
782         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
783                               hsm_restore_cb, &hrd);
784
785         RETURN(rc);
786 }
787
788 static int hsm_init_ucred(struct lu_ucred *uc)
789 {
790         ENTRY;
791
792         uc->uc_valid = UCRED_OLD;
793         uc->uc_o_uid = 0;
794         uc->uc_o_gid = 0;
795         uc->uc_o_fsuid = 0;
796         uc->uc_o_fsgid = 0;
797         uc->uc_uid = 0;
798         uc->uc_gid = 0;
799         uc->uc_fsuid = 0;
800         uc->uc_fsgid = 0;
801         uc->uc_suppgids[0] = -1;
802         uc->uc_suppgids[1] = -1;
803         uc->uc_cap = CFS_CAP_FS_MASK;
804         uc->uc_umask = 0777;
805         uc->uc_ginfo = NULL;
806         uc->uc_identity = NULL;
807
808         RETURN(0);
809 }
810
811 /**
812  * wake up coordinator thread
813  * \param mdt [IN] device
814  * \retval 0 success
815  * \retval -ve failure
816  */
817 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
818 {
819         struct coordinator      *cdt = &mdt->mdt_coordinator;
820         ENTRY;
821
822         if (cdt->cdt_state == CDT_STOPPED)
823                 RETURN(-ESRCH);
824
825         /* wake up coordinator */
826         cdt->cdt_thread.t_flags = SVC_EVENT;
827         wake_up(&cdt->cdt_thread.t_ctl_waitq);
828
829         RETURN(0);
830 }
831
832 /**
833  * initialize coordinator struct
834  * \param mdt [IN] device
835  * \retval 0 success
836  * \retval -ve failure
837  */
838 int mdt_hsm_cdt_init(struct mdt_device *mdt)
839 {
840         struct coordinator      *cdt = &mdt->mdt_coordinator;
841         struct mdt_thread_info  *cdt_mti = NULL;
842         int                      rc;
843         ENTRY;
844
845         cdt->cdt_state = CDT_STOPPED;
846
847         init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
848         mutex_init(&cdt->cdt_llog_lock);
849         init_rwsem(&cdt->cdt_agent_lock);
850         init_rwsem(&cdt->cdt_request_lock);
851         mutex_init(&cdt->cdt_restore_lock);
852
853         CFS_INIT_LIST_HEAD(&cdt->cdt_requests);
854         CFS_INIT_LIST_HEAD(&cdt->cdt_agents);
855         CFS_INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
856
857         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
858         if (rc < 0)
859                 RETURN(rc);
860
861         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
862         rc = lu_context_init(&cdt->cdt_session, LCT_SESSION);
863         if (rc == 0) {
864                 lu_context_enter(&cdt->cdt_session);
865                 cdt->cdt_env.le_ses = &cdt->cdt_session;
866         } else {
867                 lu_env_fini(&cdt->cdt_env);
868                 RETURN(rc);
869         }
870
871         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
872         LASSERT(cdt_mti != NULL);
873
874         cdt_mti->mti_env = &cdt->cdt_env;
875         cdt_mti->mti_mdt = mdt;
876
877         hsm_init_ucred(mdt_ucred(cdt_mti));
878
879         RETURN(0);
880 }
881
882 /**
883  * free a coordinator thread
884  * \param mdt [IN] device
885  */
886 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
887 {
888         struct coordinator *cdt = &mdt->mdt_coordinator;
889         ENTRY;
890
891         lu_context_exit(cdt->cdt_env.le_ses);
892         lu_context_fini(cdt->cdt_env.le_ses);
893
894         lu_env_fini(&cdt->cdt_env);
895
896         RETURN(0);
897 }
898
899 /**
900  * start a coordinator thread
901  * \param mdt [IN] device
902  * \retval 0 success
903  * \retval -ve failure
904  */
905 int mdt_hsm_cdt_start(struct mdt_device *mdt)
906 {
907         struct coordinator      *cdt = &mdt->mdt_coordinator;
908         int                      rc;
909         void                    *ptr;
910         struct mdt_thread_info  *cdt_mti;
911         struct task_struct      *task;
912         ENTRY;
913
914         /* functions defined but not yet used
915          * this avoid compilation warning
916          */
917         ptr = dump_requests;
918
919         if (cdt->cdt_state != CDT_STOPPED) {
920                 CERROR("%s: Coordinator already started\n",
921                        mdt_obd_name(mdt));
922                 RETURN(-EALREADY);
923         }
924
925         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
926         cdt->cdt_policy = CDT_DEFAULT_POLICY;
927         cdt->cdt_state = CDT_INIT;
928
929         atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
930         /* just need to be larger than previous one */
931         /* cdt_last_cookie is protected by cdt_llog_lock */
932         cdt->cdt_last_cookie = cfs_time_current_sec();
933         cdt->cdt_loop_period = 10;
934         cdt->cdt_delay = 60;
935         cdt->cdt_timeout = 3600;
936         cdt->cdt_max_request = 3;
937         cdt->cdt_archive_id = 1;
938         atomic_set(&cdt->cdt_request_count, 0);
939
940         /* to avoid deadlock when start is made through /proc
941          * /proc entries are created by the coordinator thread */
942
943         /* set up list of started restore requests */
944         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
945         rc = mdt_hsm_pending_restore(cdt_mti);
946         if (rc)
947                 CERROR("%s: cannot take the layout locks needed"
948                        " for registered restore: %d",
949                        mdt_obd_name(mdt), rc);
950
951         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
952         if (IS_ERR(task)) {
953                 rc = PTR_ERR(task);
954                 cdt->cdt_state = CDT_STOPPED;
955                 CERROR("%s: error starting coordinator thread: %d\n",
956                        mdt_obd_name(mdt), rc);
957                 RETURN(rc);
958         } else {
959                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
960                        mdt_obd_name(mdt));
961                 rc = 0;
962         }
963
964         wait_event(cdt->cdt_thread.t_ctl_waitq,
965                        (cdt->cdt_thread.t_flags & SVC_RUNNING));
966
967         cdt->cdt_state = CDT_RUNNING;
968         mdt->mdt_opts.mo_coordinator = 1;
969         RETURN(0);
970 }
971
972 /**
973  * stop a coordinator thread
974  * \param mdt [IN] device
975  */
976 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
977 {
978         struct coordinator              *cdt = &mdt->mdt_coordinator;
979         struct cdt_agent_req            *car, *tmp1;
980         struct hsm_agent                *ha, *tmp2;
981         struct cdt_restore_handle       *crh, *tmp3;
982         struct mdt_thread_info          *cdt_mti;
983         ENTRY;
984
985         if (cdt->cdt_state == CDT_STOPPED) {
986                 CERROR("%s: Coordinator already stopped\n",
987                        mdt_obd_name(mdt));
988                 RETURN(-EALREADY);
989         }
990
991         /* remove proc entries */
992         if (cdt->cdt_proc_dir != NULL)
993                 lprocfs_remove(&cdt->cdt_proc_dir);
994
995         if (cdt->cdt_state != CDT_STOPPING) {
996                 /* stop coordinator thread before cleaning */
997                 cdt->cdt_thread.t_flags = SVC_STOPPING;
998                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
999                 wait_event(cdt->cdt_thread.t_ctl_waitq,
1000                            cdt->cdt_thread.t_flags & SVC_STOPPED);
1001         }
1002         cdt->cdt_state = CDT_STOPPED;
1003
1004         /* start cleaning */
1005         down_write(&cdt->cdt_request_lock);
1006         list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
1007                                  car_request_list) {
1008                 list_del(&car->car_request_list);
1009                 mdt_cdt_free_request(car);
1010         }
1011         up_write(&cdt->cdt_request_lock);
1012
1013         down_write(&cdt->cdt_agent_lock);
1014         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
1015                 list_del(&ha->ha_list);
1016                 OBD_FREE_PTR(ha);
1017         }
1018         up_write(&cdt->cdt_agent_lock);
1019
1020         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1021         mutex_lock(&cdt->cdt_restore_lock);
1022         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
1023                 struct mdt_object       *child;
1024
1025                 /* give back layout lock */
1026                 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
1027                 if (!IS_ERR(child))
1028                         mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
1029
1030                 list_del(&crh->crh_list);
1031
1032                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1033         }
1034         mutex_unlock(&cdt->cdt_restore_lock);
1035
1036         mdt->mdt_opts.mo_coordinator = 0;
1037
1038         RETURN(0);
1039 }
1040
1041 /**
1042  * register all requests from an hal in the memory list
1043  * \param mti [IN] context
1044  * \param hal [IN] request
1045  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1046  *  which is running the CT
1047  * \retval 0 success
1048  * \retval -ve failure
1049  */
1050 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1051                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1052 {
1053         struct mdt_device       *mdt = mti->mti_mdt;
1054         struct coordinator      *cdt = &mdt->mdt_coordinator;
1055         struct hsm_action_item  *hai;
1056         int                      rc = 0, i;
1057         ENTRY;
1058
1059         /* register request in memory list */
1060         hai = hai_first(hal);
1061         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1062                 struct cdt_agent_req *car;
1063
1064                 /* in case of a cancel request, we first mark the ondisk
1065                  * record of the request we want to stop as canceled
1066                  * this does not change the cancel record
1067                  * it will be done when updating the request status
1068                  */
1069                 if (hai->hai_action == HSMA_CANCEL) {
1070                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1071                                                      &hai->hai_cookie,
1072                                                      1, ARS_CANCELED);
1073                         if (rc) {
1074                                 CERROR("%s: mdt_agent_record_update() failed, "
1075                                        "rc=%d, cannot update status to %s "
1076                                        "for cookie "LPX64"\n",
1077                                        mdt_obd_name(mdt), rc,
1078                                        agent_req_status2name(ARS_CANCELED),
1079                                        hai->hai_cookie);
1080                                 GOTO(out, rc);
1081                         }
1082
1083                         /* find the running request to set it canceled */
1084                         car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1085                         if (car != NULL) {
1086                                 car->car_canceled = 1;
1087                                 /* uuid has to be changed to the one running the
1088                                 * request to cancel */
1089                                 *uuid = car->car_uuid;
1090                                 mdt_cdt_put_request(car);
1091                         }
1092                         /* no need to memorize cancel request
1093                          * this also avoid a deadlock when we receive
1094                          * a purge all requests command
1095                          */
1096                         continue;
1097                 }
1098
1099                 if (hai->hai_action == HSMA_ARCHIVE) {
1100                         struct mdt_object *obj;
1101                         struct md_hsm hsm;
1102
1103                         obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1104                         if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1105                                 continue;
1106                         if (IS_ERR(obj))
1107                                 GOTO(out, rc = PTR_ERR(obj));
1108
1109                         hsm.mh_flags |= HS_EXISTS;
1110                         hsm.mh_arch_id = hal->hal_archive_id;
1111                         rc = mdt_hsm_attr_set(mti, obj, &hsm);
1112                         mdt_object_put(mti->mti_env, obj);
1113                         if (rc)
1114                                 GOTO(out, rc);
1115                 }
1116
1117                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1118                                             hal->hal_archive_id, hal->hal_flags,
1119                                             uuid, hai);
1120                 if (IS_ERR(car))
1121                         GOTO(out, rc = PTR_ERR(car));
1122
1123                 rc = mdt_cdt_add_request(cdt, car);
1124                 if (rc != 0)
1125                         mdt_cdt_free_request(car);
1126         }
1127 out:
1128         RETURN(rc);
1129 }
1130
1131 /**
1132  * swap layouts between 2 fids
1133  * \param mti [IN] context
1134  * \param fid1 [IN]
1135  * \param fid2 [IN]
1136  */
1137 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1138                             const lustre_fid *fid, const lustre_fid *dfid)
1139 {
1140         struct mdt_device       *mdt = mti->mti_mdt;
1141         struct mdt_object       *child1, *child2;
1142         struct mdt_lock_handle  *lh2;
1143         int                      rc;
1144         ENTRY;
1145
1146         child1 = mdt_object_find(mti->mti_env, mdt, fid);
1147         if (IS_ERR(child1))
1148                 GOTO(out, rc = PTR_ERR(child1));
1149
1150         /* we already have layout lock on FID so take only
1151          * on dfid */
1152         lh2 = &mti->mti_lh[MDT_LH_OLD];
1153         mdt_lock_reg_init(lh2, LCK_EX);
1154         child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1155         if (IS_ERR(child2))
1156                 GOTO(out_child1, rc = PTR_ERR(child2));
1157
1158         /* if copy tool closes the volatile before sending the final
1159          * progress through llapi_hsm_copy_end(), all the objects
1160          * are removed and mdd_swap_layout LBUG */
1161         if (mdt_object_exists(child2)) {
1162                 rc = mo_swap_layouts(mti->mti_env, mdt_object_child(child1),
1163                                      mdt_object_child(child2), 0);
1164         } else {
1165                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1166                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1167                 rc = -ENOENT;
1168         }
1169
1170         mdt_object_unlock_put(mti, child2, lh2, 1);
1171 out_child1:
1172         mdt_object_put(mti->mti_env, child1);
1173 out:
1174         RETURN(rc);
1175 }
1176
1177 /**
1178  * update status of a completed request
1179  * \param mti [IN] context
1180  * \param pgs [IN] progress of the copy tool
1181  * \param update_record [IN] update llog record
1182  * \retval 0 success
1183  * \retval -ve failure
1184  */
1185 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1186                                      struct hsm_progress_kernel *pgs,
1187                                      const struct cdt_agent_req *car,
1188                                      enum agent_req_status *status)
1189 {
1190         const struct lu_env     *env = mti->mti_env;
1191         struct mdt_device       *mdt = mti->mti_mdt;
1192         struct coordinator      *cdt = &mdt->mdt_coordinator;
1193         struct mdt_object       *obj = NULL;
1194         int                      cl_flags = 0, rc = 0;
1195         struct md_hsm            mh;
1196         bool                     is_mh_changed;
1197         ENTRY;
1198
1199         /* default is to retry */
1200         *status = ARS_WAITING;
1201
1202         /* find object by FID */
1203         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1204         /* we will update MD HSM only if needed */
1205         is_mh_changed = false;
1206         if (IS_ERR(obj)) {
1207                 /* object removed */
1208                 *status = ARS_SUCCEED;
1209                 goto unlock;
1210         }
1211
1212         /* no need to change mh->mh_arch_id
1213          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1214          */
1215         if (pgs->hpk_errval != 0) {
1216                 switch (pgs->hpk_errval) {
1217                 case ENOSYS:
1218                         /* the copy tool does not support cancel
1219                          * so the cancel request is failed
1220                          * As we cannot distinguish a cancel progress
1221                          * from another action progress (they have the
1222                          * same cookie), we suppose here the CT returns
1223                          * ENOSYS only if does not support cancel
1224                          */
1225                         /* this can also happen when cdt calls it to
1226                          * for a timeouted request */
1227                         *status = ARS_FAILED;
1228                         /* to have a cancel event in changelog */
1229                         pgs->hpk_errval = ECANCELED;
1230                         break;
1231                 case ECANCELED:
1232                         /* the request record has already been set to
1233                          * ARS_CANCELED, this set the cancel request
1234                          * to ARS_SUCCEED */
1235                         *status = ARS_SUCCEED;
1236                         break;
1237                 default:
1238                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1239                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1240                                    ARS_FAILED : ARS_WAITING);
1241                         break;
1242                 }
1243
1244                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1245                         CERROR("%s: Request "LPX64" on "DFID
1246                                " failed, error code %d too large\n",
1247                                mdt_obd_name(mdt),
1248                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1249                                pgs->hpk_errval);
1250                         hsm_set_cl_error(&cl_flags,
1251                                          CLF_HSM_ERROVERFLOW);
1252                         rc = -EINVAL;
1253                 } else {
1254                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1255                 }
1256
1257                 switch (car->car_hai->hai_action) {
1258                 case HSMA_ARCHIVE:
1259                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1260                         break;
1261                 case HSMA_RESTORE:
1262                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1263                         break;
1264                 case HSMA_REMOVE:
1265                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1266                         break;
1267                 case HSMA_CANCEL:
1268                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1269                         CERROR("%s: Failed request "LPX64" on "DFID
1270                                " cannot be a CANCEL\n",
1271                                mdt_obd_name(mdt),
1272                                pgs->hpk_cookie,
1273                                PFID(&pgs->hpk_fid));
1274                         break;
1275                 default:
1276                         CERROR("%s: Failed request "LPX64" on "DFID
1277                                " %d is an unknown action\n",
1278                                mdt_obd_name(mdt),
1279                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1280                                car->car_hai->hai_action);
1281                         rc = -EINVAL;
1282                         break;
1283                 }
1284         } else {
1285                 *status = ARS_SUCCEED;
1286                 switch (car->car_hai->hai_action) {
1287                 case HSMA_ARCHIVE:
1288                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1289                         /* set ARCHIVE keep EXIST and clear LOST and
1290                          * DIRTY */
1291                         mh.mh_arch_ver = pgs->hpk_data_version;
1292                         mh.mh_flags |= HS_ARCHIVED;
1293                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1294                         is_mh_changed = true;
1295                         break;
1296                 case HSMA_RESTORE:
1297                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1298
1299                         /* clear RELEASED and DIRTY */
1300                         mh.mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1301                         /* Restoring has changed the file version on
1302                          * disk. */
1303                         mh.mh_arch_ver = pgs->hpk_data_version;
1304                         is_mh_changed = true;
1305                         break;
1306                 case HSMA_REMOVE:
1307                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1308                         /* clear ARCHIVED EXISTS and LOST */
1309                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1310                         is_mh_changed = true;
1311                         break;
1312                 case HSMA_CANCEL:
1313                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1314                         CERROR("%s: Successful request "LPX64
1315                                " on "DFID
1316                                " cannot be a CANCEL\n",
1317                                mdt_obd_name(mdt),
1318                                pgs->hpk_cookie,
1319                                PFID(&pgs->hpk_fid));
1320                         break;
1321                 default:
1322                         CERROR("%s: Successful request "LPX64
1323                                " on "DFID
1324                                " %d is an unknown action\n",
1325                                mdt_obd_name(mdt),
1326                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1327                                car->car_hai->hai_action);
1328                         rc = -EINVAL;
1329                         break;
1330                 }
1331         }
1332
1333         /* rc != 0 means error when analysing action, it may come from
1334          * a crasy CT no need to manage DIRTY
1335          */
1336         if (rc == 0)
1337                 hsm_set_cl_flags(&cl_flags,
1338                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1339
1340         /* unlock is done later, after layout lock management */
1341         if (is_mh_changed)
1342                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1343
1344 unlock:
1345         /* we give back layout lock only if restore was successful or
1346          * if restore was canceled or if policy is to not retry
1347          * in other cases we just unlock the object */
1348         if (car->car_hai->hai_action == HSMA_RESTORE &&
1349             (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1350              cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1351                 struct cdt_restore_handle       *crh;
1352
1353                 /* restore in data FID done, we swap the layouts
1354                  * only if restore is successfull */
1355                 if (pgs->hpk_errval == 0) {
1356                         rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1357                                               &car->car_hai->hai_dfid);
1358                         if (rc) {
1359                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1360                                         *status = ARS_FAILED;
1361                                 pgs->hpk_errval = -rc;
1362                         }
1363                 }
1364                 /* we have to retry, so keep layout lock */
1365                 if (*status == ARS_WAITING)
1366                         GOTO(out, rc);
1367
1368                 /* give back layout lock */
1369                 mutex_lock(&cdt->cdt_restore_lock);
1370                 crh = hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1371                 if (crh != NULL)
1372                         list_del(&crh->crh_list);
1373                 mutex_unlock(&cdt->cdt_restore_lock);
1374                 /* just give back layout lock, we keep
1375                  * the reference which is given back
1376                  * later with the lock for HSM flags */
1377                 if (!IS_ERR(obj) && crh != NULL)
1378                         mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1379
1380                 if (crh != NULL)
1381                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1382         }
1383
1384         GOTO(out, rc);
1385
1386 out:
1387         if (obj != NULL && !IS_ERR(obj)) {
1388                 mo_changelog(env, CL_HSM, cl_flags,
1389                              mdt_object_child(obj));
1390                 mdt_object_put(mti->mti_env, obj);
1391         }
1392
1393         RETURN(rc);
1394 }
1395
1396 /**
1397  * update status of a request
1398  * \param mti [IN] context
1399  * \param pgs [IN] progress of the copy tool
1400  * \param update_record [IN] update llog record
1401  * \retval 0 success
1402  * \retval -ve failure
1403  */
1404 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1405                                  struct hsm_progress_kernel *pgs,
1406                                  const int update_record)
1407 {
1408         struct mdt_device       *mdt = mti->mti_mdt;
1409         struct coordinator      *cdt = &mdt->mdt_coordinator;
1410         struct cdt_agent_req    *car;
1411         int                      rc = 0;
1412         ENTRY;
1413
1414         /* no coordinator started, so we cannot serve requests */
1415         if (cdt->cdt_state == CDT_STOPPED)
1416                 RETURN(-EAGAIN);
1417
1418         /* first do sanity checks */
1419         car = mdt_cdt_update_request(cdt, pgs);
1420         if (IS_ERR(car)) {
1421                 CERROR("%s: Cannot find running request for cookie "LPX64
1422                        " on fid="DFID"\n",
1423                        mdt_obd_name(mdt),
1424                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1425                 if (car == NULL)
1426                         RETURN(-ENOENT);
1427                 RETURN(PTR_ERR(car));
1428         }
1429
1430         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1431                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1432                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1433                       hsm_copytool_action2name(car->car_hai->hai_action),
1434                       pgs->hpk_flags, pgs->hpk_errval,
1435                       PFID(&car->car_hai->hai_fid),
1436                       PFID(&car->car_hai->hai_dfid));
1437
1438         /* progress is done on FID or data FID depending of the action and
1439          * of the copy progress */
1440         /* for restore progress is used to send back the data FID to cdt */
1441         if (car->car_hai->hai_action == HSMA_RESTORE &&
1442             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1443                 car->car_hai->hai_dfid = pgs->hpk_fid;
1444
1445         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1446              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1447             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1448              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1449                 CERROR("%s: Progress on "DFID" for cookie "LPX64
1450                        " does not match request FID "DFID" nor data FID "
1451                        DFID"\n",
1452                        mdt_obd_name(mdt),
1453                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1454                        PFID(&car->car_hai->hai_fid),
1455                        PFID(&car->car_hai->hai_dfid));
1456                 GOTO(out, rc = -EINVAL);
1457         }
1458
1459         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1460                 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1461                        " is not coherent (err=%d and not completed"
1462                        " (flags=%d))\n",
1463                        mdt_obd_name(mdt),
1464                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1465                        hsm_copytool_action2name(car->car_hai->hai_action),
1466                        pgs->hpk_errval, pgs->hpk_flags);
1467                 GOTO(out, rc = -EINVAL);
1468         }
1469
1470         /* now progress is valid */
1471
1472         /* we use a root like ucred */
1473         hsm_init_ucred(mdt_ucred(mti));
1474
1475         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1476                 enum agent_req_status    status;
1477
1478                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1479
1480                 /* remove request from memory list */
1481                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1482
1483                 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1484                               " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1485                        pgs->hpk_cookie,
1486                        hsm_copytool_action2name(car->car_hai->hai_action),
1487                        agent_req_status2name(status));
1488
1489                 if (update_record) {
1490                         int rc1;
1491
1492                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1493                                                      &pgs->hpk_cookie, 1,
1494                                                      status);
1495                         if (rc1)
1496                                 CERROR("%s: mdt_agent_record_update() failed,"
1497                                        " rc=%d, cannot update status to %s"
1498                                        " for cookie "LPX64"\n",
1499                                        mdt_obd_name(mdt), rc1,
1500                                        agent_req_status2name(status),
1501                                        pgs->hpk_cookie);
1502                         rc = (rc != 0 ? rc : rc1);
1503                 }
1504                 /* ct has completed a request, so a slot is available, wakeup
1505                  * cdt to find new work */
1506                 mdt_hsm_cdt_wakeup(mdt);
1507         } else {
1508                 /* if copytool send a progress on a canceled request
1509                  * we inform copytool it should stop
1510                  */
1511                 if (car->car_canceled == 1)
1512                         rc = -ECANCELED;
1513         }
1514         GOTO(out, rc);
1515
1516 out:
1517         /* remove ref got from mdt_cdt_update_request() */
1518         mdt_cdt_put_request(car);
1519
1520         return rc;
1521 }
1522
1523
1524 /**
1525  * data passed to llog_cat_process() callback
1526  * to cancel requests
1527  */
1528 struct hsm_cancel_all_data {
1529         struct mdt_device       *mdt;
1530 };
1531
1532 /**
1533  *  llog_cat_process() callback, used to:
1534  *  - purge all requests
1535  * \param env [IN] environment
1536  * \param llh [IN] llog handle
1537  * \param hdr [IN] llog record
1538  * \param data [IN] cb data = struct hsm_cancel_all_data
1539  * \retval 0 success
1540  * \retval -ve failure
1541  */
1542 static int mdt_cancel_all_cb(const struct lu_env *env,
1543                              struct llog_handle *llh,
1544                              struct llog_rec_hdr *hdr, void *data)
1545 {
1546         struct llog_agent_req_rec       *larr;
1547         struct hsm_cancel_all_data      *hcad;
1548         int                              rc = 0;
1549         ENTRY;
1550
1551         larr = (struct llog_agent_req_rec *)hdr;
1552         hcad = data;
1553         if (larr->arr_status == ARS_WAITING ||
1554             larr->arr_status == ARS_STARTED) {
1555                 larr->arr_status = ARS_CANCELED;
1556                 larr->arr_req_change = cfs_time_current_sec();
1557                 rc = mdt_agent_llog_update_rec(env, hcad->mdt, llh, larr);
1558                 if (rc == 0)
1559                         RETURN(LLOG_DEL_RECORD);
1560         }
1561         RETURN(rc);
1562 }
1563
1564 /**
1565  * cancel all actions
1566  * \param obd [IN] MDT device
1567  */
1568 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1569 {
1570         struct mdt_thread_info          *mti;
1571         struct coordinator              *cdt = &mdt->mdt_coordinator;
1572         struct cdt_agent_req            *car;
1573         struct hsm_action_list          *hal = NULL;
1574         struct hsm_action_item          *hai;
1575         struct hsm_cancel_all_data       hcad;
1576         int                              hal_sz = 0, hal_len, rc;
1577         enum cdt_states                  save_state;
1578         ENTRY;
1579
1580         /* retrieve coordinator context */
1581         mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1582
1583         /* disable coordinator */
1584         save_state = cdt->cdt_state;
1585         cdt->cdt_state = CDT_DISABLE;
1586
1587         /* send cancel to all running requests */
1588         down_read(&cdt->cdt_request_lock);
1589         list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1590                 mdt_cdt_get_request(car);
1591                 /* request is not yet removed from list, it will be done
1592                  * when copytool will return progress
1593                  */
1594
1595                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1596                         mdt_cdt_put_request(car);
1597                         continue;
1598                 }
1599
1600                 /* needed size */
1601                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1602                           cfs_size_round(car->car_hai->hai_len);
1603
1604                 if (hal_len > hal_sz && hal_sz > 0) {
1605                         /* not enough room, free old buffer */
1606                         OBD_FREE(hal, hal_sz);
1607                         hal = NULL;
1608                 }
1609
1610                 /* empty buffer, allocate one */
1611                 if (hal == NULL) {
1612                         hal_sz = hal_len;
1613                         OBD_ALLOC(hal, hal_sz);
1614                         if (hal == NULL) {
1615                                 mdt_cdt_put_request(car);
1616                                 up_read(&cdt->cdt_request_lock);
1617                                 GOTO(out, rc = -ENOMEM);
1618                         }
1619                 }
1620
1621                 hal->hal_version = HAL_VERSION;
1622                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1623                                 MTI_NAME_MAXLEN);
1624                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1625                 hal->hal_compound_id = car->car_compound_id;
1626                 hal->hal_archive_id = car->car_archive_id;
1627                 hal->hal_flags = car->car_flags;
1628                 hal->hal_count = 0;
1629
1630                 hai = hai_first(hal);
1631                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1632                 hai->hai_action = HSMA_CANCEL;
1633                 hal->hal_count = 1;
1634
1635                 /* it is possible to safely call mdt_hsm_agent_send()
1636                  * (ie without a deadlock on cdt_request_lock), because the
1637                  * write lock is taken only if we are not in purge mode
1638                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1639                  *   nor mdt_cdt_remove_request())
1640                  */
1641                 /* no conflict with cdt thread because cdt is disable and we
1642                  * have the request lock */
1643                 mdt_hsm_agent_send(mti, hal, 1);
1644
1645                 mdt_cdt_put_request(car);
1646         }
1647         up_read(&cdt->cdt_request_lock);
1648
1649         if (hal != NULL)
1650                 OBD_FREE(hal, hal_sz);
1651
1652         /* cancel all on-disk records */
1653         hcad.mdt = mdt;
1654
1655         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1656                               mdt_cancel_all_cb, &hcad);
1657 out:
1658         /* enable coordinator */
1659         cdt->cdt_state = save_state;
1660
1661         RETURN(rc);
1662 }
1663
1664 /**
1665  * check if a request is comptaible with file status
1666  * \param hai [IN] request description
1667  * \param hal_an [IN] request archive number (not used)
1668  * \param rq_flags [IN] request flags
1669  * \param hsm [IN] file HSM metadata
1670  * \retval boolean
1671  */
1672 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1673                               const int hal_an, const __u64 rq_flags,
1674                               const struct md_hsm *hsm)
1675 {
1676         int      is_compat = false;
1677         int      hsm_flags;
1678         ENTRY;
1679
1680         hsm_flags = hsm->mh_flags;
1681         switch (hai->hai_action) {
1682         case HSMA_ARCHIVE:
1683                 if (!(hsm_flags & HS_NOARCHIVE) &&
1684                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1685                         is_compat = true;
1686                 break;
1687         case HSMA_RESTORE:
1688                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1689                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1690                         is_compat = true;
1691                 break;
1692         case HSMA_REMOVE:
1693                 if (!(hsm_flags & HS_RELEASED) &&
1694                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1695                         is_compat = true;
1696                 break;
1697         case HSMA_CANCEL:
1698                 is_compat = true;
1699                 break;
1700         }
1701         CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1702                       " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1703                       PFID(&hai->hai_fid),
1704                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1705                       hai->hai_extent.offset, hai->hai_extent.length,
1706                       hsm->mh_flags,
1707                       (is_compat ? "compatible" : "uncompatible"));
1708
1709         RETURN(is_compat);
1710 }
1711
1712 /*
1713  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1714  */
1715 static const struct {
1716         __u64            bit;
1717         char            *name;
1718         char            *nickname;
1719 } hsm_policy_names[] = {
1720         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1721         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1722         { 0 },
1723 };
1724
1725 /**
1726  * convert a policy name to a bit
1727  * \param name [IN] policy name
1728  * \retval 0 unknown
1729  * \retval   policy bit
1730  */
1731 static __u64 hsm_policy_str2bit(const char *name)
1732 {
1733         int      i;
1734
1735         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1736                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1737                     strcmp(hsm_policy_names[i].name, name) == 0)
1738                         return hsm_policy_names[i].bit;
1739         return 0;
1740 }
1741
1742 /**
1743  * convert a policy bit field to a string
1744  * \param mask [IN] policy bit field
1745  * \param hexa [IN] print mask before bit names
1746  * \param buffer [OUT] string
1747  * \param count [IN] size of buffer
1748  * \retval size filled in buffer
1749  */
1750 static int hsm_policy_bit2str(const __u64 mask, const bool hexa, char *buffer,
1751                               int count)
1752 {
1753         int      i, j, sz;
1754         char    *ptr;
1755         __u64    bit;
1756         ENTRY;
1757
1758         ptr = buffer;
1759         if (hexa) {
1760                 sz = snprintf(buffer, count, "("LPX64") ", mask);
1761                 ptr += sz;
1762                 count -= sz;
1763         }
1764         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1765                 bit = (1ULL << i);
1766
1767                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1768                         if (hsm_policy_names[j].bit == bit)
1769                                 break;
1770                 }
1771                 if (bit & mask)
1772                         sz = snprintf(ptr, count, "[%s] ",
1773                                       hsm_policy_names[j].name);
1774                 else
1775                         sz = snprintf(ptr, count, "%s ",
1776                                       hsm_policy_names[j].name);
1777
1778                 ptr += sz;
1779                 count -= sz;
1780         }
1781         /* remove last ' ' */
1782         *ptr = '\0';
1783         ptr--;
1784         RETURN(ptr - buffer);
1785 }
1786
1787 /* methods to read/write HSM policy flags */
1788 static int lprocfs_rd_hsm_policy(char *page, char **start, off_t off,
1789                                  int count, int *eof, void *data)
1790 {
1791         struct mdt_device       *mdt = data;
1792         struct coordinator      *cdt = &mdt->mdt_coordinator;
1793         int                      sz;
1794         ENTRY;
1795
1796         sz = hsm_policy_bit2str(cdt->cdt_policy, false, page, count);
1797         page[sz] = '\n';
1798         sz++;
1799         page[sz] = '\0';
1800         *eof = 1;
1801         RETURN(sz);
1802 }
1803
1804 static int lprocfs_wr_hsm_policy(struct file *file, const char *buffer,
1805                                  unsigned long count, void *data)
1806 {
1807         struct mdt_device       *mdt = data;
1808         struct coordinator      *cdt = &mdt->mdt_coordinator;
1809         char                    *start, *token, sign;
1810         char                    *buf;
1811         __u64                    policy;
1812         __u64                    add_mask, remove_mask, set_mask;
1813         int                      sz;
1814         int                      rc;
1815         ENTRY;
1816
1817         if (count + 1 > PAGE_SIZE)
1818                 RETURN(-EINVAL);
1819
1820         OBD_ALLOC(buf, count + 1);
1821         if (buf == NULL)
1822                 RETURN(-ENOMEM);
1823
1824         if (copy_from_user(buf, buffer, count))
1825                 RETURN(-EFAULT);
1826         buf[count] = '\0';
1827
1828         start = buf;
1829         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1830                start);
1831
1832         add_mask = remove_mask = set_mask = 0;
1833         do {
1834                 token = strsep(&start, "\n ");
1835                 sign = *token;
1836
1837                 if (sign == '\0')
1838                         continue;
1839
1840                 if (sign == '-' || sign == '+')
1841                         token++;
1842
1843                 policy = hsm_policy_str2bit(token);
1844                 if (policy == 0) {
1845                         char *msg;
1846
1847                         sz = PAGE_SIZE;
1848                         OBD_ALLOC(msg, sz);
1849                         if (!msg)
1850                                 RETURN(-ENOMEM);
1851
1852                         hsm_policy_bit2str(0, false, msg, sz);
1853                         CWARN("%s: '%s' is unknown, "
1854                               "supported policies are: %s\n", mdt_obd_name(mdt),
1855                               token, msg);
1856                         OBD_FREE(msg, sz);
1857                         GOTO(out, rc = -EINVAL);
1858                 }
1859                 switch (sign) {
1860                 case '-':
1861                         remove_mask |= policy;
1862                         break;
1863                 case '+':
1864                         add_mask |= policy;
1865                         break;
1866                 default:
1867                         set_mask |= policy;
1868                         break;
1869                 }
1870
1871         } while (start != NULL);
1872
1873         CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1874                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1875
1876         /* if no sign in all string, it is a clear and set
1877          * if some sign found, all unsigned are converted
1878          * to add
1879          * P1 P2 = set to P1 and P2
1880          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1881          */
1882         if (remove_mask == 0 && add_mask == 0) {
1883                 cdt->cdt_policy = set_mask;
1884         } else {
1885                 cdt->cdt_policy |= set_mask | add_mask;
1886                 cdt->cdt_policy &= ~remove_mask;
1887         }
1888
1889         GOTO(out, rc = count);
1890
1891 out:
1892         OBD_FREE(buf, count + 1);
1893         RETURN(rc);
1894 }
1895
1896 #define GENERATE_PROC_METHOD(VAR)                                       \
1897 static int lprocfs_rd_hsm_##VAR(char *page, char **start, off_t off,    \
1898                                 int count, int *eof, void *data)        \
1899 {                                                                       \
1900         struct mdt_device       *mdt = data;                            \
1901         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1902         int                      sz;                                    \
1903         ENTRY;                                                          \
1904                                                                         \
1905         sz = snprintf(page, count, LPU64"\n", (__u64)cdt->VAR);         \
1906         *eof = 1;                                                       \
1907         RETURN(sz);                                                     \
1908 }                                                                       \
1909 static int lprocfs_wr_hsm_##VAR(struct file *file, const char *buffer,  \
1910                                 unsigned long count, void *data)        \
1911                                                                         \
1912 {                                                                       \
1913         struct mdt_device       *mdt = data;                            \
1914         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1915         int                      val;                                   \
1916         int                      rc;                                    \
1917         ENTRY;                                                          \
1918                                                                         \
1919         rc = lprocfs_write_helper(buffer, count, &val);                 \
1920         if (rc)                                                         \
1921                 RETURN(rc);                                             \
1922         if (val > 0) {                                                  \
1923                 cdt->VAR = val;                                         \
1924                 RETURN(count);                                          \
1925         }                                                               \
1926         RETURN(-EINVAL);                                                \
1927 }
1928
1929 GENERATE_PROC_METHOD(cdt_loop_period)
1930 GENERATE_PROC_METHOD(cdt_delay)
1931 GENERATE_PROC_METHOD(cdt_timeout)
1932 GENERATE_PROC_METHOD(cdt_max_request)
1933 GENERATE_PROC_METHOD(cdt_archive_id)
1934
1935 /*
1936  * procfs write method for MDT/hsm_control
1937  * proc entry is in mdt directory so data is mdt obd_device pointer
1938  */
1939 #define CDT_ENABLE_CMD   "enabled"
1940 #define CDT_STOP_CMD     "shutdown"
1941 #define CDT_DISABLE_CMD  "disabled"
1942 #define CDT_PURGE_CMD    "purge"
1943 #define CDT_HELP_CMD     "help"
1944
1945 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
1946                                unsigned long count, void *data)
1947 {
1948         struct obd_device       *obd = data;
1949         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
1950         struct coordinator      *cdt = &(mdt->mdt_coordinator);
1951         int                      rc, usage = 0;
1952         ENTRY;
1953
1954         rc = 0;
1955         if (strncmp(buffer, CDT_ENABLE_CMD, strlen(CDT_ENABLE_CMD)) == 0) {
1956                 if (cdt->cdt_state == CDT_DISABLE) {
1957                         cdt->cdt_state = CDT_RUNNING;
1958                         mdt_hsm_cdt_wakeup(mdt);
1959                 } else {
1960                         rc = mdt_hsm_cdt_start(mdt);
1961                 }
1962         } else if (strncmp(buffer, CDT_STOP_CMD, strlen(CDT_STOP_CMD)) == 0) {
1963                 cdt->cdt_state = CDT_STOPPING;
1964         } else if (strncmp(buffer, CDT_DISABLE_CMD,
1965                            strlen(CDT_DISABLE_CMD)) == 0) {
1966                 cdt->cdt_state = CDT_DISABLE;
1967         } else if (strncmp(buffer, CDT_PURGE_CMD, strlen(CDT_PURGE_CMD)) == 0) {
1968                 rc = hsm_cancel_all_actions(mdt);
1969         } else if (strncmp(buffer, CDT_HELP_CMD, strlen(CDT_HELP_CMD)) == 0) {
1970                 usage = 1;
1971         } else {
1972                 usage = 1;
1973                 rc = -EINVAL;
1974         }
1975
1976         if (usage == 1)
1977                 CERROR("%s: Valid coordinator control commands are: "
1978                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
1979                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1980                        CDT_PURGE_CMD, CDT_HELP_CMD);
1981
1982         if (rc)
1983                 RETURN(rc);
1984
1985         RETURN(count);
1986 }
1987
1988 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
1989                                int count, int *eof, void *data)
1990 {
1991         struct obd_device       *obd = data;
1992         struct coordinator      *cdt;
1993         int                      sz;
1994         ENTRY;
1995
1996         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
1997         *eof = 1;
1998
1999         if (cdt->cdt_state == CDT_INIT)
2000                 sz = snprintf(page, count, "init\n");
2001         else if (cdt->cdt_state == CDT_RUNNING)
2002                 sz = snprintf(page, count, "enabled\n");
2003         else if (cdt->cdt_state == CDT_STOPPING)
2004                 sz = snprintf(page, count, "stopping\n");
2005         else if (cdt->cdt_state == CDT_STOPPED)
2006                 sz = snprintf(page, count, "stopped\n");
2007         else if (cdt->cdt_state == CDT_DISABLE)
2008                 sz = snprintf(page, count, "disabled\n");
2009         else
2010                 sz = snprintf(page, count, "unknown\n");
2011
2012         RETURN(sz);
2013 }
2014
2015 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2016         { "agents",             NULL, NULL, NULL, &mdt_hsm_agent_fops, 0 },
2017         { "agent_actions",      NULL, NULL, NULL,
2018                                 &mdt_agent_actions_fops, 0444 },
2019         { "archive_id",         lprocfs_rd_hsm_cdt_archive_id,
2020                                 lprocfs_wr_hsm_cdt_archive_id,
2021                                 NULL, NULL, 0 },
2022         { "grace_delay",        lprocfs_rd_hsm_cdt_delay,
2023                                 lprocfs_wr_hsm_cdt_delay,
2024                                 NULL, NULL, 0 },
2025         { "loop_period",        lprocfs_rd_hsm_cdt_loop_period,
2026                                 lprocfs_wr_hsm_cdt_loop_period,
2027                                 NULL, NULL, 0 },
2028         { "max_requests",       lprocfs_rd_hsm_cdt_max_request,
2029                                 lprocfs_wr_hsm_cdt_max_request,
2030                                 NULL, NULL, 0 },
2031         { "policy",             lprocfs_rd_hsm_policy, lprocfs_wr_hsm_policy,
2032                                 NULL, NULL, 0 },
2033         { "request_timeout",    lprocfs_rd_hsm_cdt_timeout,
2034                                 lprocfs_wr_hsm_cdt_timeout,
2035                                 NULL, NULL, 0 },
2036         { "requests",           NULL, NULL, NULL, &mdt_hsm_request_fops, 0 },
2037         { 0 }
2038 };