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) 2011, 2012 Commissariat a l'energie atomique et aux energies
  24  *                          alternatives
  25  *
  26  * Copyright (c) 2013, 2017, Intel Corporation.
  27  * Use is subject to license terms.
  28  */
  29 /*
  30  * lustre/mdt/mdt_coordinator.c
  31  *
  32  * Lustre HSM Coordinator
  33  *
  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>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <linux/kthread.h>
  42 #include <obd_support.h>
  43 #include <lustre_export.h>
  44 #include <obd.h>
  45 #include <lprocfs_status.h>
  46 #include <lustre_log.h>
  47 #include <lustre_kernelcomm.h>
  48 #include "mdt_internal.h"
  49
  50 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
  51
  52 /**
  53  * get obj and HSM attributes on a fid
  54  * \param mti [IN] context
  55  * \param fid [IN] object fid
  56  * \param hsm [OUT] HSM meta data
  57  * \retval obj or error (-ENOENT if not found)
  58  */
  59 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  60                                       const struct lu_fid *fid,
  61                                       struct md_hsm *hsm)
  62 {
  63         struct md_attr          *ma;
  64         struct mdt_object       *obj;
  65         int                      rc;
  66         ENTRY;
  67
  68         ma = &mti->mti_attr;
  69         ma->ma_need = MA_HSM;
  70         ma->ma_valid = 0;
  71
  72         /* find object by FID */
  73         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  74         if (IS_ERR(obj))
  75                 RETURN(obj);
  76
  77         if (!mdt_object_exists(obj)) {
  78                 /* no more object */
  79                 mdt_object_put(mti->mti_env, obj);
  80                 RETURN(ERR_PTR(-ENOENT));
  81         }
  82
  83         rc = mdt_attr_get_complex(mti, obj, ma);
  84         if (rc) {
  85                 mdt_object_put(mti->mti_env, obj);
  86                 RETURN(ERR_PTR(rc));
  87         }
  88
  89         if (ma->ma_valid & MA_HSM)
  90                 *hsm = ma->ma_hsm;
  91         else
  92                 memset(hsm, 0, sizeof(*hsm));
  93         ma->ma_valid = 0;
  94         RETURN(obj);
  95 }
  96
  97 void mdt_hsm_dump_hal(int level, const char *prefix,
  98                       struct hsm_action_list *hal)
  99 {
 100         int                      i, sz;
 101         struct hsm_action_item  *hai;
 102         char                     buf[12];
 103
 104         CDEBUG(level, "%s: HAL header: version %X count %d"
 105                       " archive_id %d flags %#llx\n",
 106                prefix, hal->hal_version, hal->hal_count,
 107                hal->hal_archive_id, hal->hal_flags);
 108
 109         hai = hai_first(hal);
 110         for (i = 0; i < hal->hal_count; i++) {
 111                 sz = hai->hai_len - sizeof(*hai);
 112                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 113                        " cookie=%#llx"
 114                        " action=%s extent=%#llx-%#llx gid=%#llx"
 115                        " datalen=%d data=[%s]\n",
 116                        prefix, i,
 117                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 118                        hai->hai_cookie,
 119                        hsm_copytool_action2name(hai->hai_action),
 120                        hai->hai_extent.offset,
 121                        hai->hai_extent.length,
 122                        hai->hai_gid, sz,
 123                        hai_dump_data_field(hai, buf, sizeof(buf)));
 124                 hai = hai_next(hai);
 125         }
 126 }
 127
 128 /**
 129  * data passed to llog_cat_process() callback
 130  * to scan requests and take actions
 131  */
 132 struct hsm_scan_request {
 133         int                      hal_sz;
 134         int                      hal_used_sz;
 135         struct hsm_action_list  *hal;
 136 };
 137
 138 struct hsm_scan_data {
 139         struct mdt_thread_info  *hsd_mti;
 140         char                     hsd_fsname[MTI_NAME_MAXLEN + 1];
 141         /* are we scanning the logs for housekeeping, or just looking
 142          * for new work?
 143          */
 144         bool                     hsd_housekeeping;
 145         bool                     hsd_one_restore;
 146         int                      hsd_action_count;
 147         int                      hsd_request_len; /* array alloc len */
 148         int                      hsd_request_count; /* array used count */
 149         struct hsm_scan_request *hsd_request;
 150 };
 151
 152 static int mdt_cdt_waiting_cb(const struct lu_env *env,
 153                               struct mdt_device *mdt,
 154                               struct llog_handle *llh,
 155                               struct llog_agent_req_rec *larr,
 156                               struct hsm_scan_data *hsd)
 157 {
 158         struct coordinator *cdt = &mdt->mdt_coordinator;
 159         struct hsm_scan_request *request;
 160         struct hsm_action_item *hai;
 161         size_t hai_size;
 162         u32 archive_id;
 163         int i;
 164
 165         /* Are agents full? */
 166         if (atomic_read(&cdt->cdt_request_count) >= cdt->cdt_max_requests)
 167                 RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
 168
 169         if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
 170             cdt->cdt_max_requests) {
 171                 /* We cannot send any more request
 172                  *
 173                  *                     *** SPECIAL CASE ***
 174                  *
 175                  * Restore requests are too important not to schedule at least
 176                  * one, everytime we can.
 177                  */
 178                 if (larr->arr_hai.hai_action != HSMA_RESTORE ||
 179                     hsd->hsd_one_restore)
 180                         RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
 181         }
 182
 183         hai_size = cfs_size_round(larr->arr_hai.hai_len);
 184         archive_id = larr->arr_archive_id;
 185
 186         /* Can we add this action to one of the existing HALs in hsd. */
 187         request = NULL;
 188         for (i = 0; i < hsd->hsd_request_count; i++) {
 189                 if (hsd->hsd_request[i].hal->hal_archive_id == archive_id &&
 190                     hsd->hsd_request[i].hal_used_sz + hai_size <=
 191                     LDLM_MAXREQSIZE) {
 192                         request = &hsd->hsd_request[i];
 193                         break;
 194                 }
 195         }
 196
 197         /* Are we trying to force-schedule a request? */
 198         if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
 199             cdt->cdt_max_requests) {
 200                 /* Is there really no compatible hsm_scan_request? */
 201                 if (!request) {
 202                         for (i -= 1; i >= 0; i--) {
 203                                 if (hsd->hsd_request[i].hal->hal_archive_id ==
 204                                     archive_id) {
 205                                         request = &hsd->hsd_request[i];
 206                                         break;
 207                                 }
 208                         }
 209                 }
 210
 211                 /* Make room for the hai */
 212                 if (request) {
 213                         /* Discard the last hai until there is enough space */
 214                         do {
 215                                 request->hal->hal_count--;
 216
 217                                 hai = hai_first(request->hal);
 218                                 for (i = 0; i < request->hal->hal_count; i++)
 219                                         hai = hai_next(hai);
 220                                 request->hal_used_sz -=
 221                                         cfs_size_round(hai->hai_len);
 222                                 hsd->hsd_action_count--;
 223                         } while (request->hal_used_sz + hai_size >
 224                                  LDLM_MAXREQSIZE);
 225                 } else if (hsd->hsd_housekeeping) {
 226                         struct hsm_scan_request *tmp;
 227
 228                         /* Discard the (whole) last hal */
 229                         hsd->hsd_request_count--;
 230                         LASSERT(hsd->hsd_request_count >= 0);
 231                         tmp = &hsd->hsd_request[hsd->hsd_request_count];
 232                         hsd->hsd_action_count -= tmp->hal->hal_count;
 233                         LASSERT(hsd->hsd_action_count >= 0);
 234                         OBD_FREE(tmp->hal, tmp->hal_sz);
 235                 } else {
 236                         /* Bailing out, this code path is too hot */
 237                         RETURN(LLOG_PROC_BREAK);
 238
 239                 }
 240         }
 241
 242         if (!request) {
 243                 struct hsm_action_list *hal;
 244
 245                 LASSERT(hsd->hsd_request_count < hsd->hsd_request_len);
 246                 request = &hsd->hsd_request[hsd->hsd_request_count];
 247
 248                 /* allocates hai vector size just needs to be large
 249                  * enough */
 250                 request->hal_sz = sizeof(*request->hal) +
 251                         cfs_size_round(MTI_NAME_MAXLEN + 1) + 2 * hai_size;
 252                 OBD_ALLOC_LARGE(hal, request->hal_sz);
 253                 if (!hal)
 254                         RETURN(-ENOMEM);
 255
 256                 hal->hal_version = HAL_VERSION;
 257                 strlcpy(hal->hal_fsname, hsd->hsd_fsname, MTI_NAME_MAXLEN + 1);
 258                 hal->hal_archive_id = larr->arr_archive_id;
 259                 hal->hal_flags = larr->arr_flags;
 260                 hal->hal_count = 0;
 261                 request->hal_used_sz = hal_size(hal);
 262                 request->hal = hal;
 263                 hsd->hsd_request_count++;
 264         } else if (request->hal_sz < request->hal_used_sz + hai_size) {
 265                 /* Not enough room, need an extension */
 266                 void *hal_buffer;
 267                 int sz;
 268
 269                 sz = min_t(int, 2 * request->hal_sz, LDLM_MAXREQSIZE);
 270                 LASSERT(request->hal_used_sz + hai_size < sz);
 271
 272                 OBD_ALLOC_LARGE(hal_buffer, sz);
 273                 if (!hal_buffer)
 274                         RETURN(-ENOMEM);
 275
 276                 memcpy(hal_buffer, request->hal, request->hal_used_sz);
 277                 OBD_FREE_LARGE(request->hal, request->hal_sz);
 278                 request->hal = hal_buffer;
 279                 request->hal_sz = sz;
 280         }
 281
 282         hai = hai_first(request->hal);
 283         for (i = 0; i < request->hal->hal_count; i++)
 284                 hai = hai_next(hai);
 285
 286         memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 287
 288         request->hal_used_sz += hai_size;
 289         request->hal->hal_count++;
 290
 291         hsd->hsd_action_count++;
 292
 293         switch (hai->hai_action) {
 294         case HSMA_CANCEL:
 295                 break;
 296         case HSMA_RESTORE:
 297                 hsd->hsd_one_restore = true;
 298                 /* Intentional fallthrough */
 299         default:
 300                 cdt_agent_record_hash_add(cdt, hai->hai_cookie,
 301                                           llh->lgh_hdr->llh_cat_idx,
 302                                           larr->arr_hdr.lrh_index);
 303         }
 304
 305         RETURN(0);
 306 }
 307
 308 static int mdt_cdt_started_cb(const struct lu_env *env,
 309                               struct mdt_device *mdt,
 310                               struct llog_handle *llh,
 311                               struct llog_agent_req_rec *larr,
 312                               struct hsm_scan_data *hsd)
 313 {
 314         struct coordinator *cdt = &mdt->mdt_coordinator;
 315         struct hsm_action_item *hai = &larr->arr_hai;
 316         struct cdt_agent_req *car;
 317         time64_t now = ktime_get_real_seconds();
 318         time64_t last;
 319         enum changelog_rec_flags clf_flags;
 320         int rc;
 321
 322         if (!hsd->hsd_housekeeping)
 323                 RETURN(0);
 324
 325         /* we search for a running request
 326          * error may happen if coordinator crashes or stopped
 327          * with running request
 328          */
 329         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
 330         if (car == NULL) {
 331                 last = larr->arr_req_change;
 332         } else {
 333                 last = car->car_req_update;
 334         }
 335
 336         /* test if request too long, if yes cancel it
 337          * the same way the copy tool acknowledge a cancel request */
 338         if (now <= last + cdt->cdt_active_req_timeout)
 339                 GOTO(out_car, rc = 0);
 340
 341         dump_llog_agent_req_rec("request timed out, start cleaning", larr);
 342
 343         if (car != NULL) {
 344                 car->car_req_update = now;
 345                 mdt_hsm_agent_update_statistics(cdt, 0, 1, 0, &car->car_uuid);
 346                 /* Remove car from memory list (LU-9075) */
 347                 mdt_cdt_remove_request(cdt, hai->hai_cookie);
 348         }
 349
 350         /* Emit a changelog record for the failed action.*/
 351         clf_flags = 0;
 352         hsm_set_cl_error(&clf_flags, ECANCELED);
 353
 354         switch (hai->hai_action) {
 355         case HSMA_ARCHIVE:
 356                 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
 357                 break;
 358         case HSMA_RESTORE:
 359                 hsm_set_cl_event(&clf_flags, HE_RESTORE);
 360                 break;
 361         case HSMA_REMOVE:
 362                 hsm_set_cl_event(&clf_flags, HE_REMOVE);
 363                 break;
 364         case HSMA_CANCEL:
 365                 hsm_set_cl_event(&clf_flags, HE_CANCEL);
 366                 break;
 367         default:
 368                 /* Unknown record type, skip changelog. */
 369                 clf_flags = 0;
 370                 break;
 371         }
 372
 373         if (clf_flags != 0)
 374                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
 375                              &hai->hai_fid);
 376
 377         if (hai->hai_action == HSMA_RESTORE)
 378                 cdt_restore_handle_del(hsd->hsd_mti, cdt, &hai->hai_fid);
 379
 380         larr->arr_status = ARS_CANCELED;
 381         larr->arr_req_change = now;
 382         rc = llog_write(hsd->hsd_mti->mti_env, llh, &larr->arr_hdr,
 383                         larr->arr_hdr.lrh_index);
 384         if (rc < 0) {
 385                 CERROR("%s: cannot update agent log: rc = %d\n",
 386                        mdt_obd_name(mdt), rc);
 387                 rc = LLOG_DEL_RECORD;
 388         }
 389
 390         /* ct has completed a request, so a slot is available,
 391          * signal the coordinator to find new work */
 392         mdt_hsm_cdt_event(cdt);
 393 out_car:
 394         if (car != NULL)
 395                 mdt_cdt_put_request(car);
 396
 397         RETURN(rc);
 398 }
 399
 400 /**
 401  *  llog_cat_process() callback, used to:
 402  *  - find waiting request and start action
 403  *  - purge canceled and done requests
 404  * \param env [IN] environment
 405  * \param llh [IN] llog handle
 406  * \param hdr [IN] llog record
 407  * \param data [IN/OUT] cb data = struct hsm_scan_data
 408  * \retval 0 success
 409  * \retval -ve failure
 410  */
 411 static int mdt_coordinator_cb(const struct lu_env *env,
 412                               struct llog_handle *llh,
 413                               struct llog_rec_hdr *hdr,
 414                               void *data)
 415 {
 416         struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
 417         struct hsm_scan_data *hsd = data;
 418         struct mdt_device *mdt = hsd->hsd_mti->mti_mdt;
 419         struct coordinator *cdt = &mdt->mdt_coordinator;
 420         ENTRY;
 421
 422         larr = (struct llog_agent_req_rec *)hdr;
 423         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 424         switch (larr->arr_status) {
 425         case ARS_WAITING:
 426                 RETURN(mdt_cdt_waiting_cb(env, mdt, llh, larr, hsd));
 427         case ARS_STARTED:
 428                 RETURN(mdt_cdt_started_cb(env, mdt, llh, larr, hsd));
 429         default:
 430                 if (!hsd->hsd_housekeeping)
 431                         RETURN(0);
 432
 433                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 434                     ktime_get_real_seconds()) {
 435                         cdt_agent_record_hash_del(cdt,
 436                                                   larr->arr_hai.hai_cookie);
 437                         RETURN(LLOG_DEL_RECORD);
 438                 }
 439
 440                 RETURN(0);
 441         }
 442 }
 443
 444 /* Release the ressource used by the coordinator. Called when the
 445  * coordinator is stopping. */
 446 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
 447 {
 448         struct coordinator              *cdt = &mdt->mdt_coordinator;
 449         struct cdt_agent_req            *car, *tmp1;
 450         struct hsm_agent                *ha, *tmp2;
 451         struct cdt_restore_handle       *crh, *tmp3;
 452         struct mdt_thread_info          *cdt_mti;
 453
 454         /* start cleaning */
 455         down_write(&cdt->cdt_request_lock);
 456         list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
 457                                  car_request_list) {
 458                 cfs_hash_del(cdt->cdt_request_cookie_hash,
 459                              &car->car_hai->hai_cookie,
 460                              &car->car_cookie_hash);
 461                 list_del(&car->car_request_list);
 462                 mdt_cdt_put_request(car);
 463         }
 464         up_write(&cdt->cdt_request_lock);
 465
 466         down_write(&cdt->cdt_agent_lock);
 467         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 468                 list_del(&ha->ha_list);
 469                 if (ha->ha_archive_cnt != 0)
 470                         OBD_FREE(ha->ha_archive_id, ha->ha_archive_cnt *
 471                                  sizeof(*ha->ha_archive_id));
 472                 OBD_FREE_PTR(ha);
 473         }
 474         up_write(&cdt->cdt_agent_lock);
 475
 476         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 477         mutex_lock(&cdt->cdt_restore_lock);
 478         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_handle_list,
 479                                  crh_list) {
 480                 list_del(&crh->crh_list);
 481                 /* give back layout lock */
 482                 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
 483                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 484         }
 485         mutex_unlock(&cdt->cdt_restore_lock);
 486 }
 487
 488 /*
 489  * Coordinator state transition table, indexed on enum cdt_states, taking
 490  * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
 491  * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
 492  */
 493 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
 494         /* from -> to:    stopped init   running disable stopping */
 495         /* stopped */   { true,   true,  false,  false,  false },
 496         /* init */      { true,   false, true,   false,  false },
 497         /* running */   { false,  false, true,   true,   true },
 498         /* disable */   { false,  false, true,   true,   true },
 499         /* stopping */  { true,   false, false,  false,  false }
 500 };
 501
 502 /**
 503  * Change coordinator thread state
 504  * Some combinations are not valid, so catch them here.
 505  *
 506  * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
 507  * the transition was not possible.
 508  */
 509 static int set_cdt_state_locked(struct coordinator *cdt,
 510                                 enum cdt_states new_state)
 511 {
 512         int rc;
 513         enum cdt_states state;
 514
 515         state = cdt->cdt_state;
 516
 517         if (cdt_transition[state][new_state]) {
 518                 cdt->cdt_state = new_state;
 519                 rc = 0;
 520         } else {
 521                 CDEBUG(D_HSM,
 522                        "unexpected coordinator transition, from=%s, to=%s\n",
 523                        cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
 524                 rc = -EINVAL;
 525         }
 526
 527         return rc;
 528 }
 529
 530 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state)
 531 {
 532         int rc;
 533
 534         mutex_lock(&cdt->cdt_state_lock);
 535         rc = set_cdt_state_locked(cdt, new_state);
 536         mutex_unlock(&cdt->cdt_state_lock);
 537
 538         return rc;
 539 }
 540
 541
 542
 543 /**
 544  * coordinator thread
 545  * \param data [IN] obd device
 546  * \retval 0 success
 547  * \retval -ve failure
 548  */
 549 static int mdt_coordinator(void *data)
 550 {
 551         struct mdt_thread_info  *mti = data;
 552         struct mdt_device       *mdt = mti->mti_mdt;
 553         struct coordinator      *cdt = &mdt->mdt_coordinator;
 554         struct hsm_scan_data     hsd = { NULL };
 555         time64_t                 last_housekeeping = 0;
 556         size_t request_sz = 0;
 557         int rc;
 558         ENTRY;
 559
 560         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 561                mdt_obd_name(mdt), current_pid());
 562
 563         hsd.hsd_mti = mti;
 564         obd_uuid2fsname(hsd.hsd_fsname, mdt_obd_name(mdt),
 565                         sizeof(hsd.hsd_fsname));
 566
 567         set_cdt_state(cdt, CDT_RUNNING);
 568
 569         /* Inform mdt_hsm_cdt_start(). */
 570         wake_up_all(&cdt->cdt_waitq);
 571
 572         while (1) {
 573                 int i;
 574                 int update_idx = 0;
 575                 int updates_sz;
 576                 int updates_cnt;
 577                 struct hsm_record_update *updates;
 578
 579                 /* Limit execution of the expensive requests traversal
 580                  * to at most one second. This prevents repeatedly
 581                  * locking/unlocking the catalog for each request
 582                  * and preventing other HSM operations from happening
 583                  */
 584                 wait_event_interruptible_timeout(cdt->cdt_waitq,
 585                                                  kthread_should_stop() ||
 586                                                  cdt->cdt_wakeup_coordinator,
 587                                                  cfs_time_seconds(1));
 588
 589                 cdt->cdt_wakeup_coordinator = false;
 590                 CDEBUG(D_HSM, "coordinator resumes\n");
 591
 592                 if (kthread_should_stop()) {
 593                         CDEBUG(D_HSM, "Coordinator stops\n");
 594                         rc = 0;
 595                         break;
 596                 }
 597
 598                 /* if coordinator is suspended continue to wait */
 599                 if (cdt->cdt_state == CDT_DISABLE) {
 600                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 601                         continue;
 602                 }
 603
 604                 /* If no event, and no housekeeping to do, continue to
 605                  * wait. */
 606                 if (last_housekeeping + cdt->cdt_loop_period <=
 607                     ktime_get_real_seconds()) {
 608                         last_housekeeping = ktime_get_real_seconds();
 609                         hsd.hsd_housekeeping = true;
 610                 } else if (cdt->cdt_event) {
 611                         hsd.hsd_housekeeping = false;
 612                 } else {
 613                         continue;
 614                 }
 615
 616                 cdt->cdt_event = false;
 617
 618                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 619
 620                 if (hsd.hsd_request_len != cdt->cdt_max_requests) {
 621                         /* cdt_max_requests has changed,
 622                          * we need to allocate a new buffer
 623                          */
 624                         struct hsm_scan_request *tmp = NULL;
 625                         int max_requests = cdt->cdt_max_requests;
 626                         OBD_ALLOC_LARGE(tmp, max_requests *
 627                                         sizeof(struct hsm_scan_request));
 628                         if (!tmp) {
 629                                 CERROR("Failed to resize request buffer, "
 630                                        "keeping it at %d\n",
 631                                        hsd.hsd_request_len);
 632                         } else {
 633                                 if (hsd.hsd_request != NULL)
 634                                         OBD_FREE_LARGE(hsd.hsd_request,
 635                                                        request_sz);
 636
 637                                 hsd.hsd_request_len = max_requests;
 638                                 request_sz = hsd.hsd_request_len *
 639                                         sizeof(struct hsm_scan_request);
 640                                 hsd.hsd_request = tmp;
 641                         }
 642                 }
 643
 644                 hsd.hsd_action_count = 0;
 645                 hsd.hsd_request_count = 0;
 646                 hsd.hsd_one_restore = false;
 647
 648                 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
 649                                       &hsd, 0, 0, WRITE);
 650                 if (rc < 0)
 651                         goto clean_cb_alloc;
 652
 653                 CDEBUG(D_HSM, "found %d requests to send\n",
 654                        hsd.hsd_request_count);
 655
 656                 if (list_empty(&cdt->cdt_agents)) {
 657                         CDEBUG(D_HSM, "no agent available, "
 658                                       "coordinator sleeps\n");
 659                         goto clean_cb_alloc;
 660                 }
 661
 662                 /* Compute how many HAI we have in all the requests */
 663                 updates_cnt = 0;
 664                 for (i = 0; i < hsd.hsd_request_count; i++) {
 665                         const struct hsm_scan_request *request =
 666                                 &hsd.hsd_request[i];
 667
 668                         updates_cnt += request->hal->hal_count;
 669                 }
 670
 671                 /* Allocate a temporary array to store the cookies to
 672                  * update, and their status. */
 673                 updates_sz = updates_cnt * sizeof(*updates);
 674                 OBD_ALLOC_LARGE(updates, updates_sz);
 675                 if (updates == NULL) {
 676                         CERROR("%s: Cannot allocate memory (%d bytes) "
 677                                "for %d updates. Too many HSM requests?\n",
 678                                mdt_obd_name(mdt), updates_sz, updates_cnt);
 679                         goto clean_cb_alloc;
 680                 }
 681
 682                 /* here hsd contains a list of requests to be started */
 683                 for (i = 0; i < hsd.hsd_request_count; i++) {
 684                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 685                         struct hsm_action_list  *hal = request->hal;
 686                         struct hsm_action_item  *hai;
 687                         int                      j;
 688
 689                         /* still room for work ? */
 690                         if (atomic_read(&cdt->cdt_request_count) >=
 691                             cdt->cdt_max_requests)
 692                                 break;
 693
 694                         rc = mdt_hsm_agent_send(mti, hal, 0);
 695                         /* if failure, we suppose it is temporary
 696                          * if the copy tool failed to do the request
 697                          * it has to use hsm_progress
 698                          */
 699
 700                         /* set up cookie vector to set records status
 701                          * after copy tools start or failed
 702                          */
 703                         hai = hai_first(hal);
 704                         for (j = 0; j < hal->hal_count; j++) {
 705                                 updates[update_idx].cookie = hai->hai_cookie;
 706                                 updates[update_idx].status =
 707                                         (rc ? ARS_WAITING : ARS_STARTED);
 708                                 hai = hai_next(hai);
 709                                 update_idx++;
 710                         }
 711                 }
 712
 713                 if (update_idx) {
 714                         rc = mdt_agent_record_update(mti->mti_env, mdt,
 715                                                      updates, update_idx);
 716                         if (rc)
 717                                 CERROR("%s: mdt_agent_record_update() failed, "
 718                                        "rc=%d, cannot update records "
 719                                        "for %d cookies\n",
 720                                        mdt_obd_name(mdt), rc, update_idx);
 721                 }
 722
 723                 OBD_FREE_LARGE(updates, updates_sz);
 724
 725 clean_cb_alloc:
 726                 /* free hal allocated by callback */
 727                 for (i = 0; i < hsd.hsd_request_count; i++) {
 728                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 729
 730                         OBD_FREE_LARGE(request->hal, request->hal_sz);
 731                 }
 732         }
 733
 734         if (hsd.hsd_request != NULL)
 735                 OBD_FREE_LARGE(hsd.hsd_request, request_sz);
 736
 737         mdt_hsm_cdt_cleanup(mdt);
 738
 739         if (rc != 0)
 740                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 741                        mdt_obd_name(mdt), current_pid(), rc);
 742         else
 743                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 744                               " no error\n",
 745                        mdt_obd_name(mdt), current_pid());
 746
 747         RETURN(rc);
 748 }
 749
 750 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
 751                            const struct lu_fid *fid,
 752                            const struct hsm_extent *he)
 753 {
 754         struct cdt_restore_handle *crh;
 755         struct mdt_object *obj;
 756         int rc;
 757         ENTRY;
 758
 759         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 760         if (crh == NULL)
 761                 RETURN(-ENOMEM);
 762
 763         crh->crh_fid = *fid;
 764         /* in V1 all file is restored
 765          * crh->extent.start = he->offset;
 766          * crh->extent.end = he->offset + he->length;
 767          */
 768         crh->crh_extent.start = 0;
 769         crh->crh_extent.end = he->length;
 770         /* get the layout lock */
 771         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 772         obj = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 773                                    MDS_INODELOCK_LAYOUT);
 774         if (IS_ERR(obj))
 775                 GOTO(out_crh, rc = PTR_ERR(obj));
 776
 777         /* We do not keep a reference on the object during the restore
 778          * which can be very long. */
 779         mdt_object_put(mti->mti_env, obj);
 780
 781         mutex_lock(&cdt->cdt_restore_lock);
 782         if (unlikely(cdt->cdt_state == CDT_STOPPED ||
 783                      cdt->cdt_state == CDT_STOPPING)) {
 784                 mutex_unlock(&cdt->cdt_restore_lock);
 785                 GOTO(out_lh, rc = -EAGAIN);
 786         }
 787
 788         list_add_tail(&crh->crh_list, &cdt->cdt_restore_handle_list);
 789         mutex_unlock(&cdt->cdt_restore_lock);
 790
 791         RETURN(0);
 792 out_lh:
 793         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 794 out_crh:
 795         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 796
 797         return rc;
 798 }
 799
 800 /**
 801  * lookup a restore handle by FID
 802  * caller needs to hold cdt_restore_lock
 803  * \param cdt [IN] coordinator
 804  * \param fid [IN] FID
 805  * \retval cdt_restore_handle found
 806  * \retval NULL not found
 807  */
 808 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
 809                                                    const struct lu_fid *fid)
 810 {
 811         struct cdt_restore_handle *crh;
 812         ENTRY;
 813
 814         list_for_each_entry(crh, &cdt->cdt_restore_handle_list, crh_list) {
 815                 if (lu_fid_eq(&crh->crh_fid, fid))
 816                         RETURN(crh);
 817         }
 818
 819         RETURN(NULL);
 820 }
 821
 822 void cdt_restore_handle_del(struct mdt_thread_info *mti,
 823                             struct coordinator *cdt, const struct lu_fid *fid)
 824 {
 825         struct cdt_restore_handle *crh;
 826
 827         /* give back layout lock */
 828         mutex_lock(&cdt->cdt_restore_lock);
 829         crh = cdt_restore_handle_find(cdt, fid);
 830         if (crh != NULL)
 831                 list_del(&crh->crh_list);
 832         mutex_unlock(&cdt->cdt_restore_lock);
 833
 834         if (crh == NULL)
 835                 return;
 836
 837         /* XXX We pass a NULL object since the restore handle does not
 838          * keep a reference on the object being restored. */
 839         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 840         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 841 }
 842
 843 /**
 844  * data passed to llog_cat_process() callback
 845  * to scan requests and take actions
 846  */
 847 struct hsm_restore_data {
 848         struct mdt_thread_info  *hrd_mti;
 849 };
 850
 851 /**
 852  *  llog_cat_process() callback, used to:
 853  *  - find restore request and allocate the restore handle
 854  * \param env [IN] environment
 855  * \param llh [IN] llog handle
 856  * \param hdr [IN] llog record
 857  * \param data [IN/OUT] cb data = struct hsm_restore_data
 858  * \retval 0 success
 859  * \retval -ve failure
 860  */
 861 static int hsm_restore_cb(const struct lu_env *env,
 862                           struct llog_handle *llh,
 863                           struct llog_rec_hdr *hdr, void *data)
 864 {
 865         struct llog_agent_req_rec       *larr;
 866         struct hsm_restore_data         *hrd;
 867         struct hsm_action_item          *hai;
 868         struct mdt_thread_info          *mti;
 869         struct coordinator              *cdt;
 870         int rc;
 871         ENTRY;
 872
 873         hrd = data;
 874         mti = hrd->hrd_mti;
 875         cdt = &mti->mti_mdt->mdt_coordinator;
 876
 877         larr = (struct llog_agent_req_rec *)hdr;
 878         hai = &larr->arr_hai;
 879         if (hai->hai_cookie > cdt->cdt_last_cookie)
 880                 /* update the cookie to avoid collision */
 881                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 882
 883         if (hai->hai_action != HSMA_RESTORE ||
 884             agent_req_in_final_state(larr->arr_status))
 885                 RETURN(0);
 886
 887         /* restore request not in a final state */
 888
 889         /* force replay of restore requests left in started state from previous
 890          * CDT context, to be canceled later if finally found to be incompatible
 891          * when being re-started */
 892         if (larr->arr_status == ARS_STARTED) {
 893                 larr->arr_status = ARS_WAITING;
 894                 larr->arr_req_change = ktime_get_real_seconds();
 895                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 896                 if (rc != 0)
 897                         GOTO(out, rc);
 898         }
 899
 900         rc = cdt_restore_handle_add(mti, cdt, &hai->hai_fid, &hai->hai_extent);
 901 out:
 902         RETURN(rc);
 903 }
 904
 905 /**
 906  * restore coordinator state at startup
 907  * the goal is to take a layout lock for each registered restore request
 908  * \param mti [IN] context
 909  */
 910 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 911 {
 912         struct hsm_restore_data  hrd;
 913         int                      rc;
 914         ENTRY;
 915
 916         hrd.hrd_mti = mti;
 917
 918         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
 919                               0, 0, WRITE);
 920
 921         RETURN(rc);
 922 }
 923
 924 int hsm_init_ucred(struct lu_ucred *uc)
 925 {
 926         ENTRY;
 927
 928         uc->uc_valid = UCRED_OLD;
 929         uc->uc_o_uid = 0;
 930         uc->uc_o_gid = 0;
 931         uc->uc_o_fsuid = 0;
 932         uc->uc_o_fsgid = 0;
 933         uc->uc_uid = 0;
 934         uc->uc_gid = 0;
 935         uc->uc_fsuid = 0;
 936         uc->uc_fsgid = 0;
 937         uc->uc_suppgids[0] = -1;
 938         uc->uc_suppgids[1] = -1;
 939         uc->uc_cap = CFS_CAP_FS_MASK;
 940         uc->uc_umask = 0777;
 941         uc->uc_ginfo = NULL;
 942         uc->uc_identity = NULL;
 943         /* always record internal HSM activity if also enabled globally */
 944         uc->uc_enable_audit = 1;
 945
 946         RETURN(0);
 947 }
 948
 949 /**
 950  * initialize coordinator struct
 951  * \param mdt [IN] device
 952  * \retval 0 success
 953  * \retval -ve failure
 954  */
 955 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 956 {
 957         struct coordinator      *cdt = &mdt->mdt_coordinator;
 958         struct mdt_thread_info  *cdt_mti = NULL;
 959         int                      rc;
 960         ENTRY;
 961
 962         init_waitqueue_head(&cdt->cdt_waitq);
 963         init_rwsem(&cdt->cdt_llog_lock);
 964         init_rwsem(&cdt->cdt_agent_lock);
 965         init_rwsem(&cdt->cdt_request_lock);
 966         mutex_init(&cdt->cdt_restore_lock);
 967         mutex_init(&cdt->cdt_state_lock);
 968         set_cdt_state(cdt, CDT_STOPPED);
 969
 970         INIT_LIST_HEAD(&cdt->cdt_request_list);
 971         INIT_LIST_HEAD(&cdt->cdt_agents);
 972         INIT_LIST_HEAD(&cdt->cdt_restore_handle_list);
 973
 974         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
 975                                                        CFS_HASH_BITS_MIN,
 976                                                        CFS_HASH_BITS_MAX,
 977                                                        CFS_HASH_BKT_BITS,
 978                                                        0 /* extra bytes */,
 979                                                        CFS_HASH_MIN_THETA,
 980                                                        CFS_HASH_MAX_THETA,
 981                                                 &cdt_request_cookie_hash_ops,
 982                                                        CFS_HASH_DEFAULT);
 983         if (cdt->cdt_request_cookie_hash == NULL)
 984                 RETURN(-ENOMEM);
 985
 986         cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
 987                                                      CFS_HASH_BITS_MIN,
 988                                                      CFS_HASH_BITS_MAX,
 989                                                      CFS_HASH_BKT_BITS,
 990                                                      0 /* extra bytes */,
 991                                                      CFS_HASH_MIN_THETA,
 992                                                      CFS_HASH_MAX_THETA,
 993                                                      &cdt_agent_record_hash_ops,
 994                                                      CFS_HASH_DEFAULT);
 995         if (cdt->cdt_agent_record_hash == NULL)
 996                 GOTO(out_request_cookie_hash, rc = -ENOMEM);
 997
 998         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 999         if (rc < 0)
1000                 GOTO(out_agent_record_hash, rc);
1001
1002         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1003         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
1004         if (rc < 0)
1005                 GOTO(out_env, rc);
1006
1007         lu_context_enter(&cdt->cdt_session);
1008         cdt->cdt_env.le_ses = &cdt->cdt_session;
1009
1010         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1011         LASSERT(cdt_mti != NULL);
1012
1013         cdt_mti->mti_env = &cdt->cdt_env;
1014         cdt_mti->mti_mdt = mdt;
1015
1016         hsm_init_ucred(mdt_ucred(cdt_mti));
1017
1018         /* default values for sysfs tunnables
1019          * can be override by MGS conf */
1020         cdt->cdt_default_archive_id = 1;
1021         cdt->cdt_grace_delay = 60;
1022         cdt->cdt_loop_period = 10;
1023         cdt->cdt_max_requests = 3;
1024         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1025         cdt->cdt_active_req_timeout = 3600;
1026
1027         /* by default do not remove archives on last unlink */
1028         cdt->cdt_remove_archive_on_last_unlink = false;
1029
1030         RETURN(0);
1031
1032 out_env:
1033         lu_env_fini(&cdt->cdt_env);
1034 out_agent_record_hash:
1035         cfs_hash_putref(cdt->cdt_agent_record_hash);
1036         cdt->cdt_agent_record_hash = NULL;
1037 out_request_cookie_hash:
1038         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1039         cdt->cdt_request_cookie_hash = NULL;
1040
1041         return rc;
1042 }
1043
1044 /**
1045  * free a coordinator thread
1046  * \param mdt [IN] device
1047  */
1048 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
1049 {
1050         struct coordinator *cdt = &mdt->mdt_coordinator;
1051         ENTRY;
1052
1053         lu_context_exit(cdt->cdt_env.le_ses);
1054         lu_context_fini(cdt->cdt_env.le_ses);
1055
1056         lu_env_fini(&cdt->cdt_env);
1057
1058         cfs_hash_putref(cdt->cdt_agent_record_hash);
1059         cdt->cdt_agent_record_hash = NULL;
1060
1061         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1062         cdt->cdt_request_cookie_hash = NULL;
1063
1064         RETURN(0);
1065 }
1066
1067 /**
1068  * start a coordinator thread
1069  * \param mdt [IN] device
1070  * \retval 0 success
1071  * \retval -ve failure
1072  */
1073 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1074 {
1075         struct coordinator      *cdt = &mdt->mdt_coordinator;
1076         struct mdt_thread_info *cdt_mti;
1077         int                      rc;
1078         void                    *ptr;
1079         struct task_struct      *task;
1080         ENTRY;
1081
1082         /* functions defined but not yet used
1083          * this avoid compilation warning
1084          */
1085         ptr = dump_requests;
1086
1087         rc = set_cdt_state(cdt, CDT_INIT);
1088         if (rc) {
1089                 CERROR("%s: Coordinator already started or stopping\n",
1090                        mdt_obd_name(mdt));
1091                 RETURN(-EALREADY);
1092         }
1093
1094         BUILD_BUG_ON(BIT(CDT_POLICY_SHIFT_COUNT - 1) != CDT_POLICY_LAST);
1095         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1096
1097         /* just need to be larger than previous one */
1098         /* cdt_last_cookie is protected by cdt_llog_lock */
1099         cdt->cdt_last_cookie = ktime_get_real_seconds();
1100         atomic_set(&cdt->cdt_request_count, 0);
1101         atomic_set(&cdt->cdt_archive_count, 0);
1102         atomic_set(&cdt->cdt_restore_count, 0);
1103         atomic_set(&cdt->cdt_remove_count, 0);
1104         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
1105         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
1106         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
1107
1108         /* to avoid deadlock when start is made through sysfs
1109          * sysfs entries are created by the coordinator thread
1110          */
1111         /* set up list of started restore requests */
1112         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1113         rc = mdt_hsm_pending_restore(cdt_mti);
1114         if (rc)
1115                 CERROR("%s: cannot take the layout locks needed"
1116                        " for registered restore: %d\n",
1117                        mdt_obd_name(mdt), rc);
1118
1119         if (mdt->mdt_bottom->dd_rdonly)
1120                 RETURN(0);
1121
1122         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
1123         if (IS_ERR(task)) {
1124                 rc = PTR_ERR(task);
1125                 set_cdt_state(cdt, CDT_STOPPED);
1126                 CERROR("%s: error starting coordinator thread: %d\n",
1127                        mdt_obd_name(mdt), rc);
1128         } else {
1129                 cdt->cdt_task = task;
1130                 wait_event(cdt->cdt_waitq,
1131                            cdt->cdt_state != CDT_INIT);
1132                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1133                        mdt_obd_name(mdt));
1134                 rc = 0;
1135         }
1136
1137         RETURN(rc);
1138 }
1139
1140 /**
1141  * stop a coordinator thread
1142  * \param mdt [IN] device
1143  */
1144 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1145 {
1146         struct coordinator *cdt = &mdt->mdt_coordinator;
1147         int rc;
1148
1149         ENTRY;
1150         /* stop coordinator thread */
1151         rc = set_cdt_state(cdt, CDT_STOPPING);
1152         if (rc == 0) {
1153                 kthread_stop(cdt->cdt_task);
1154                 cdt->cdt_task = NULL;
1155                 set_cdt_state(cdt, CDT_STOPPED);
1156         }
1157
1158         RETURN(rc);
1159 }
1160
1161 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1162                               const struct lu_fid *fid,
1163                               u32 archive_id)
1164 {
1165         struct mdt_object *obj;
1166         struct md_hsm mh;
1167         int rc;
1168
1169         obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1170         if (IS_ERR(obj))
1171                 GOTO(out, rc = PTR_ERR(obj));
1172
1173         if (mh.mh_flags & HS_EXISTS &&
1174             mh.mh_arch_id == archive_id)
1175                 GOTO(out_obj, rc = 0);
1176
1177         mh.mh_flags |= HS_EXISTS;
1178         mh.mh_arch_id = archive_id;
1179         rc = mdt_hsm_attr_set(mti, obj, &mh);
1180
1181 out_obj:
1182         mdt_object_put(mti->mti_env, obj);
1183 out:
1184         return rc;
1185 }
1186
1187 /**
1188  * register all requests from an hal in the memory list
1189  * \param mti [IN] context
1190  * \param hal [IN] request
1191  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1192  *  which is running the CT
1193  * \retval 0 success
1194  * \retval -ve failure
1195  */
1196 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1197                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1198 {
1199         struct mdt_device       *mdt = mti->mti_mdt;
1200         struct coordinator      *cdt = &mdt->mdt_coordinator;
1201         struct hsm_action_item  *hai;
1202         int                      rc = 0, i;
1203         ENTRY;
1204
1205         /* register request in memory list */
1206         hai = hai_first(hal);
1207         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1208                 struct cdt_agent_req *car;
1209
1210                 /* in case of a cancel request, we first mark the ondisk
1211                  * record of the request we want to stop as canceled
1212                  * this does not change the cancel record
1213                  * it will be done when updating the request status
1214                  */
1215                 if (hai->hai_action == HSMA_CANCEL) {
1216                         struct hsm_record_update update = {
1217                                 .cookie = hai->hai_cookie,
1218                                 .status = ARS_CANCELED,
1219                         };
1220
1221                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1222                                                      &update, 1);
1223                         if (rc) {
1224                                 CERROR("%s: mdt_agent_record_update() failed, "
1225                                        "rc=%d, cannot update status to %s "
1226                                        "for cookie %#llx\n",
1227                                        mdt_obd_name(mdt), rc,
1228                                        agent_req_status2name(ARS_CANCELED),
1229                                        hai->hai_cookie);
1230                                 GOTO(out, rc);
1231                         }
1232
1233                         /* find the running request to set it canceled */
1234                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1235                         if (car != NULL) {
1236                                 car->car_canceled = 1;
1237                                 /* uuid has to be changed to the one running the
1238                                 * request to cancel */
1239                                 *uuid = car->car_uuid;
1240                                 mdt_cdt_put_request(car);
1241                         }
1242                         /* no need to memorize cancel request
1243                          * this also avoid a deadlock when we receive
1244                          * a purge all requests command
1245                          */
1246                         continue;
1247                 }
1248
1249                 if (hai->hai_action == HSMA_ARCHIVE) {
1250                         rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1251                                                 hal->hal_archive_id);
1252                         if (rc == -ENOENT)
1253                                 continue;
1254                         else if (rc < 0)
1255                                 GOTO(out, rc);
1256                 }
1257
1258                 car = mdt_cdt_alloc_request(hal->hal_archive_id, hal->hal_flags,
1259                                             uuid, hai);
1260                 if (IS_ERR(car))
1261                         GOTO(out, rc = PTR_ERR(car));
1262
1263                 rc = mdt_cdt_add_request(cdt, car);
1264                 if (rc != 0)
1265                         mdt_cdt_free_request(car);
1266         }
1267 out:
1268         RETURN(rc);
1269 }
1270
1271 /**
1272  * swap layouts between 2 fids
1273  * \param mti [IN] context
1274  * \param obj [IN]
1275  * \param dfid [IN]
1276  * \param mh_common [IN] MD HSM
1277  */
1278 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1279                             struct mdt_object *obj, const struct lu_fid *dfid,
1280                             struct md_hsm *mh_common)
1281 {
1282         struct mdt_object       *dobj;
1283         struct mdt_lock_handle  *dlh;
1284         int                      rc;
1285         ENTRY;
1286
1287         if (!mdt_object_exists(obj))
1288                 GOTO(out, rc = -ENOENT);
1289
1290         /* we already have layout lock on obj so take only
1291          * on dfid */
1292         dlh = &mti->mti_lh[MDT_LH_OLD];
1293         mdt_lock_reg_init(dlh, LCK_EX);
1294         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1295         if (IS_ERR(dobj))
1296                 GOTO(out, rc = PTR_ERR(dobj));
1297
1298         /* if copy tool closes the volatile before sending the final
1299          * progress through llapi_hsm_copy_end(), all the objects
1300          * are removed and mdd_swap_layout LBUG */
1301         if (!mdt_object_exists(dobj)) {
1302                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1303                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1304                 GOTO(out_dobj, rc = -ENOENT);
1305         }
1306         /* Since we only handle restores here, unconditionally use
1307          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1308          * be preserved in case of failure during swap_layout and not
1309          * leave a file in an intermediate but incoherent state.
1310          * But need to setup HSM xattr of data FID before, reuse
1311          * mti and mh presets for FID in hsm_cdt_request_completed(),
1312          * only need to clear RELEASED and DIRTY.
1313          */
1314         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1315         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1316         if (rc == 0)
1317                 rc = mo_swap_layouts(mti->mti_env,
1318                                      mdt_object_child(obj),
1319                                      mdt_object_child(dobj),
1320                                      SWAP_LAYOUTS_MDS_HSM);
1321         if (rc == 0) {
1322                 rc = mdt_lsom_downgrade(mti, obj);
1323                 if (rc)
1324                         CDEBUG(D_INODE,
1325                                "%s: File fid="DFID" SOM "
1326                                "downgrade failed, rc = %d\n",
1327                                mdt_obd_name(mti->mti_mdt),
1328                                PFID(mdt_object_fid(obj)), rc);
1329         }
1330 out_dobj:
1331         mdt_object_unlock_put(mti, dobj, dlh, 1);
1332 out:
1333         RETURN(rc);
1334 }
1335
1336 /**
1337  * update status of a completed request
1338  * \param mti [IN] context
1339  * \param pgs [IN] progress of the copy tool
1340  * \retval 0 success
1341  * \retval -ve failure
1342  */
1343 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1344                                      struct hsm_progress_kernel *pgs,
1345                                      const struct cdt_agent_req *car,
1346                                      enum agent_req_status *status)
1347 {
1348         const struct lu_env *env = mti->mti_env;
1349         struct mdt_device *mdt = mti->mti_mdt;
1350         struct coordinator *cdt = &mdt->mdt_coordinator;
1351         struct mdt_object *obj = NULL;
1352         enum changelog_rec_flags clf_flags = 0;
1353         struct md_hsm mh;
1354         bool is_mh_changed;
1355         bool need_changelog = true;
1356         int rc = 0;
1357
1358         ENTRY;
1359         /* default is to retry */
1360         *status = ARS_WAITING;
1361
1362         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1363          * if error/removed continue anyway to get correct reporting done */
1364         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1365         /* we will update MD HSM only if needed */
1366         is_mh_changed = false;
1367
1368         /* no need to change mh->mh_arch_id
1369          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1370          */
1371         if (pgs->hpk_errval != 0) {
1372                 switch (pgs->hpk_errval) {
1373                 case ENOSYS:
1374                         /* the copy tool does not support cancel
1375                          * so the cancel request is failed
1376                          * As we cannot distinguish a cancel progress
1377                          * from another action progress (they have the
1378                          * same cookie), we suppose here the CT returns
1379                          * ENOSYS only if does not support cancel
1380                          */
1381                         /* this can also happen when cdt calls it to
1382                          * for a timed out request */
1383                         *status = ARS_FAILED;
1384                         /* to have a cancel event in changelog */
1385                         pgs->hpk_errval = ECANCELED;
1386                         break;
1387                 case ECANCELED:
1388                         /* the request record has already been set to
1389                          * ARS_CANCELED, this set the cancel request
1390                          * to ARS_SUCCEED */
1391                         *status = ARS_SUCCEED;
1392                         break;
1393                 default:
1394                         /* retry only if current policy or requested, and
1395                          * object is not on error/removed */
1396                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1397                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1398                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1399                         break;
1400                 }
1401
1402                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1403                         CERROR("%s: Request %#llx on "DFID
1404                                " failed, error code %d too large\n",
1405                                mdt_obd_name(mdt),
1406                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1407                                pgs->hpk_errval);
1408                         hsm_set_cl_error(&clf_flags, CLF_HSM_ERROVERFLOW);
1409                         rc = -EINVAL;
1410                 } else {
1411                         hsm_set_cl_error(&clf_flags, pgs->hpk_errval);
1412                 }
1413
1414                 switch (car->car_hai->hai_action) {
1415                 case HSMA_ARCHIVE:
1416                         hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1417                         break;
1418                 case HSMA_RESTORE:
1419                         hsm_set_cl_event(&clf_flags, HE_RESTORE);
1420                         break;
1421                 case HSMA_REMOVE:
1422                         hsm_set_cl_event(&clf_flags, HE_REMOVE);
1423                         break;
1424                 case HSMA_CANCEL:
1425                         hsm_set_cl_event(&clf_flags, HE_CANCEL);
1426                         CERROR("%s: Failed request %#llx on "DFID
1427                                " cannot be a CANCEL\n",
1428                                mdt_obd_name(mdt),
1429                                pgs->hpk_cookie,
1430                                PFID(&pgs->hpk_fid));
1431                         break;
1432                 default:
1433                         CERROR("%s: Failed request %#llx on "DFID
1434                                " %d is an unknown action\n",
1435                                mdt_obd_name(mdt),
1436                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1437                                car->car_hai->hai_action);
1438                         rc = -EINVAL;
1439                         break;
1440                 }
1441         } else {
1442                 *status = ARS_SUCCEED;
1443                 switch (car->car_hai->hai_action) {
1444                 case HSMA_ARCHIVE:
1445                         hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1446                         /* set ARCHIVE keep EXIST and clear LOST and
1447                          * DIRTY */
1448                         mh.mh_arch_ver = pgs->hpk_data_version;
1449                         mh.mh_flags |= HS_ARCHIVED;
1450                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1451                         is_mh_changed = true;
1452                         break;
1453                 case HSMA_RESTORE:
1454                         hsm_set_cl_event(&clf_flags, HE_RESTORE);
1455
1456                         /* do not clear RELEASED and DIRTY here
1457                          * this will occur in hsm_swap_layouts()
1458                          */
1459
1460                         /* Restoring has changed the file version on
1461                          * disk. */
1462                         mh.mh_arch_ver = pgs->hpk_data_version;
1463                         is_mh_changed = true;
1464                         break;
1465                 case HSMA_REMOVE:
1466                         hsm_set_cl_event(&clf_flags, HE_REMOVE);
1467                         /* clear ARCHIVED EXISTS and LOST */
1468                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1469                         is_mh_changed = true;
1470                         break;
1471                 case HSMA_CANCEL:
1472                         hsm_set_cl_event(&clf_flags, HE_CANCEL);
1473                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1474                                mdt_obd_name(mdt),
1475                                pgs->hpk_cookie,
1476                                PFID(&pgs->hpk_fid));
1477                         break;
1478                 default:
1479                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1480                                mdt_obd_name(mdt),
1481                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1482                                car->car_hai->hai_action);
1483                         rc = -EINVAL;
1484                         break;
1485                 }
1486         }
1487
1488         /* rc != 0 means error when analysing action, it may come from
1489          * a crasy CT no need to manage DIRTY
1490          * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1491          * filled
1492          */
1493         if (rc == 0 && !IS_ERR(obj))
1494                 hsm_set_cl_flags(&clf_flags,
1495                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1496
1497         /* unlock is done later, after layout lock management */
1498         if (is_mh_changed && !IS_ERR(obj))
1499                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1500
1501         /* we give back layout lock only if restore was successful or
1502          * if no retry will be attempted and if object is still alive,
1503          * in other cases we just unlock the object */
1504         if (car->car_hai->hai_action == HSMA_RESTORE) {
1505                 struct mdt_lock_handle *lh;
1506
1507                 /* restore in data FID done, we swap the layouts
1508                  * only if restore is successful */
1509                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1510                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1511                                               &mh);
1512                         if (rc) {
1513                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1514                                         *status = ARS_FAILED;
1515                                 pgs->hpk_errval = -rc;
1516                         }
1517                 }
1518                 /* we have to retry, so keep layout lock */
1519                 if (*status == ARS_WAITING)
1520                         GOTO(out, rc);
1521
1522                 /* restore special case, need to create ChangeLog record
1523                  * before to give back layout lock to avoid concurrent
1524                  * file updater to post out of order ChangeLog */
1525                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1526                              &car->car_hai->hai_fid);
1527                 need_changelog = false;
1528
1529                 cdt_restore_handle_del(mti, cdt, &car->car_hai->hai_fid);
1530                 if (!IS_ERR_OR_NULL(obj)) {
1531                         /* flush UPDATE lock so attributes are upadated */
1532                         lh = &mti->mti_lh[MDT_LH_OLD];
1533                         mdt_lock_reg_init(lh, LCK_EX);
1534                         mdt_object_lock(mti, obj, lh, MDS_INODELOCK_UPDATE);
1535                         mdt_object_unlock(mti, obj, lh, 1);
1536                 }
1537         }
1538
1539         GOTO(out, rc);
1540
1541 out:
1542         /* always add a ChangeLog record */
1543         if (need_changelog)
1544                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1545                              &car->car_hai->hai_fid);
1546
1547         if (!IS_ERR(obj))
1548                 mdt_object_put(mti->mti_env, obj);
1549
1550         RETURN(rc);
1551 }
1552
1553 /**
1554  * update status of a request
1555  * \param mti [IN] context
1556  * \param pgs [IN] progress of the copy tool
1557  * \retval 0 success
1558  * \retval -ve failure
1559  */
1560 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1561                                  struct hsm_progress_kernel *pgs)
1562 {
1563         struct mdt_device       *mdt = mti->mti_mdt;
1564         struct coordinator      *cdt = &mdt->mdt_coordinator;
1565         struct cdt_agent_req    *car;
1566         int                      rc = 0;
1567         ENTRY;
1568
1569         /* no coordinator started, so we cannot serve requests */
1570         if (cdt->cdt_state == CDT_STOPPED)
1571                 RETURN(-EAGAIN);
1572
1573         /* first do sanity checks */
1574         car = mdt_cdt_update_request(cdt, pgs);
1575         if (IS_ERR(car)) {
1576                 CERROR("%s: Cannot find running request for cookie %#llx"
1577                        " on fid="DFID"\n",
1578                        mdt_obd_name(mdt),
1579                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1580
1581                 RETURN(PTR_ERR(car));
1582         }
1583
1584         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1585                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1586                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1587                       hsm_copytool_action2name(car->car_hai->hai_action),
1588                       pgs->hpk_flags, pgs->hpk_errval,
1589                       PFID(&car->car_hai->hai_fid),
1590                       PFID(&car->car_hai->hai_dfid));
1591
1592         /* progress is done on FID or data FID depending of the action and
1593          * of the copy progress */
1594         /* for restore progress is used to send back the data FID to cdt */
1595         if (car->car_hai->hai_action == HSMA_RESTORE &&
1596             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1597                 car->car_hai->hai_dfid = pgs->hpk_fid;
1598
1599         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1600              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1601             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1602              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1603                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1604                        " does not match request FID "DFID" nor data FID "
1605                        DFID"\n",
1606                        mdt_obd_name(mdt),
1607                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1608                        PFID(&car->car_hai->hai_fid),
1609                        PFID(&car->car_hai->hai_dfid));
1610                 GOTO(out, rc = -EINVAL);
1611         }
1612
1613         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1614                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1615                        " is not coherent (err=%d and not completed"
1616                        " (flags=%d))\n",
1617                        mdt_obd_name(mdt),
1618                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1619                        hsm_copytool_action2name(car->car_hai->hai_action),
1620                        pgs->hpk_errval, pgs->hpk_flags);
1621                 GOTO(out, rc = -EINVAL);
1622         }
1623
1624         /* now progress is valid */
1625
1626         /* we use a root like ucred */
1627         hsm_init_ucred(mdt_ucred(mti));
1628
1629         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1630                 enum agent_req_status status;
1631                 struct hsm_record_update update;
1632                 int rc1;
1633
1634                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1635
1636                 CDEBUG(D_HSM, "updating record: fid="DFID" cookie=%#llx action=%s "
1637                               "status=%s\n",
1638                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1639                        hsm_copytool_action2name(car->car_hai->hai_action),
1640                        agent_req_status2name(status));
1641
1642                 /* update record first (LU-9075) */
1643                 update.cookie = pgs->hpk_cookie;
1644                 update.status = status;
1645
1646                 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1647                                               &update, 1);
1648                 if (rc1)
1649                         CERROR("%s: mdt_agent_record_update() failed,"
1650                                " rc=%d, cannot update status to %s"
1651                                " for cookie %#llx\n",
1652                                mdt_obd_name(mdt), rc1,
1653                                agent_req_status2name(status),
1654                                pgs->hpk_cookie);
1655                 rc = (rc != 0 ? rc : rc1);
1656
1657                 /* then remove request from memory list (LU-9075) */
1658                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1659
1660                 /* ct has completed a request, so a slot is available,
1661                  * signal the coordinator to find new work */
1662                 mdt_hsm_cdt_event(cdt);
1663         } else {
1664                 /* if copytool send a progress on a canceled request
1665                  * we inform copytool it should stop
1666                  */
1667                 if (car->car_canceled == 1)
1668                         rc = -ECANCELED;
1669         }
1670         GOTO(out, rc);
1671
1672 out:
1673         /* remove ref got from mdt_cdt_update_request() */
1674         mdt_cdt_put_request(car);
1675
1676         return rc;
1677 }
1678
1679
1680 /**
1681  * data passed to llog_cat_process() callback
1682  * to cancel requests
1683  */
1684 struct hsm_cancel_all_data {
1685         struct mdt_device       *mdt;
1686 };
1687
1688 /**
1689  *  llog_cat_process() callback, used to:
1690  *  - purge all requests
1691  * \param env [IN] environment
1692  * \param llh [IN] llog handle
1693  * \param hdr [IN] llog record
1694  * \param data [IN] cb data = struct hsm_cancel_all_data
1695  * \retval 0 success
1696  * \retval -ve failure
1697  */
1698 static int mdt_cancel_all_cb(const struct lu_env *env,
1699                              struct llog_handle *llh,
1700                              struct llog_rec_hdr *hdr, void *data)
1701 {
1702         struct llog_agent_req_rec       *larr;
1703         struct hsm_cancel_all_data      *hcad;
1704         int                              rc = 0;
1705         ENTRY;
1706
1707         larr = (struct llog_agent_req_rec *)hdr;
1708         hcad = data;
1709         if (larr->arr_status == ARS_WAITING ||
1710             larr->arr_status == ARS_STARTED) {
1711                 larr->arr_status = ARS_CANCELED;
1712                 larr->arr_req_change = ktime_get_real_seconds();
1713                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1714         }
1715
1716         RETURN(rc);
1717 }
1718
1719 /**
1720  * cancel all actions
1721  * \param obd [IN] MDT device
1722  */
1723 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1724 {
1725         struct lu_env                    env;
1726         struct lu_context                session;
1727         struct mdt_thread_info          *mti;
1728         struct coordinator              *cdt = &mdt->mdt_coordinator;
1729         struct cdt_agent_req            *car;
1730         struct hsm_action_list          *hal = NULL;
1731         struct hsm_action_item          *hai;
1732         struct hsm_cancel_all_data       hcad;
1733         int                              hal_sz = 0, hal_len, rc;
1734         enum cdt_states                  old_state;
1735         ENTRY;
1736
1737         rc = lu_env_init(&env, LCT_MD_THREAD);
1738         if (rc < 0)
1739                 RETURN(rc);
1740
1741         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1742         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1743         if (rc < 0)
1744                 GOTO(out_env, rc);
1745
1746         lu_context_enter(&session);
1747         env.le_ses = &session;
1748
1749         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1750         LASSERT(mti != NULL);
1751
1752         mti->mti_env = &env;
1753         mti->mti_mdt = mdt;
1754
1755         hsm_init_ucred(mdt_ucred(mti));
1756
1757         mutex_lock(&cdt->cdt_state_lock);
1758         old_state = cdt->cdt_state;
1759
1760         /* disable coordinator */
1761         rc = set_cdt_state_locked(cdt, CDT_DISABLE);
1762         if (rc)
1763                 GOTO(out_cdt_state_unlock, rc);
1764
1765         /* send cancel to all running requests */
1766         down_read(&cdt->cdt_request_lock);
1767         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1768                 mdt_cdt_get_request(car);
1769                 /* request is not yet removed from list, it will be done
1770                  * when copytool will return progress
1771                  */
1772
1773                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1774                         mdt_cdt_put_request(car);
1775                         continue;
1776                 }
1777
1778                 /* needed size */
1779                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1780                           cfs_size_round(car->car_hai->hai_len);
1781
1782                 if (hal_len > hal_sz && hal_sz > 0) {
1783                         /* not enough room, free old buffer */
1784                         OBD_FREE(hal, hal_sz);
1785                         hal = NULL;
1786                 }
1787
1788                 /* empty buffer, allocate one */
1789                 if (hal == NULL) {
1790                         hal_sz = hal_len;
1791                         OBD_ALLOC(hal, hal_sz);
1792                         if (hal == NULL) {
1793                                 mdt_cdt_put_request(car);
1794                                 up_read(&cdt->cdt_request_lock);
1795                                 GOTO(out_cdt_state, rc = -ENOMEM);
1796                         }
1797                 }
1798
1799                 hal->hal_version = HAL_VERSION;
1800                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1801                                 MTI_NAME_MAXLEN);
1802                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1803                 hal->hal_archive_id = car->car_archive_id;
1804                 hal->hal_flags = car->car_flags;
1805                 hal->hal_count = 0;
1806
1807                 hai = hai_first(hal);
1808                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1809                 hai->hai_action = HSMA_CANCEL;
1810                 hal->hal_count = 1;
1811
1812                 /* it is possible to safely call mdt_hsm_agent_send()
1813                  * (ie without a deadlock on cdt_request_lock), because the
1814                  * write lock is taken only if we are not in purge mode
1815                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1816                  *   nor mdt_cdt_remove_request())
1817                  */
1818                 /* no conflict with cdt thread because cdt is disable and we
1819                  * have the request lock */
1820                 mdt_hsm_agent_send(mti, hal, 1);
1821
1822                 mdt_cdt_put_request(car);
1823         }
1824         up_read(&cdt->cdt_request_lock);
1825
1826         if (hal != NULL)
1827                 OBD_FREE(hal, hal_sz);
1828
1829         /* cancel all on-disk records */
1830         hcad.mdt = mdt;
1831
1832         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1833                               &hcad, 0, 0, WRITE);
1834 out_cdt_state:
1835         /* Enable coordinator, unless the coordinator was stopping. */
1836         set_cdt_state_locked(cdt, old_state);
1837 out_cdt_state_unlock:
1838         mutex_unlock(&cdt->cdt_state_lock);
1839
1840         lu_context_exit(&session);
1841         lu_context_fini(&session);
1842 out_env:
1843         lu_env_fini(&env);
1844
1845         RETURN(rc);
1846 }
1847
1848 /**
1849  * check if a request is compatible with file status
1850  * \param hai [IN] request description
1851  * \param archive_id [IN] request archive id
1852  * \param rq_flags [IN] request flags
1853  * \param hsm [IN] file HSM metadata
1854  * \retval boolean
1855  */
1856 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1857                               u32 archive_id, u64 rq_flags,
1858                               const struct md_hsm *hsm)
1859 {
1860         int      is_compat = false;
1861         int      hsm_flags;
1862         ENTRY;
1863
1864         hsm_flags = hsm->mh_flags;
1865         switch (hai->hai_action) {
1866         case HSMA_ARCHIVE:
1867                 if (!(hsm_flags & HS_NOARCHIVE) &&
1868                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1869                         is_compat = true;
1870
1871                 if (hsm_flags & HS_EXISTS &&
1872                     archive_id != 0 &&
1873                     archive_id != hsm->mh_arch_id)
1874                         is_compat = false;
1875
1876                 break;
1877         case HSMA_RESTORE:
1878                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1879                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1880                         is_compat = true;
1881                 break;
1882         case HSMA_REMOVE:
1883                 if (!(hsm_flags & HS_RELEASED) &&
1884                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1885                         is_compat = true;
1886                 break;
1887         case HSMA_CANCEL:
1888                 is_compat = true;
1889                 break;
1890         }
1891         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1892                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1893                       PFID(&hai->hai_fid),
1894                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1895                       hai->hai_extent.offset, hai->hai_extent.length,
1896                       hsm->mh_flags,
1897                       (is_compat ? "compatible" : "uncompatible"));
1898
1899         RETURN(is_compat);
1900 }
1901
1902 /*
1903  * sysfs interface used to get/set HSM behaviour (cdt->cdt_policy)
1904  */
1905 static const struct {
1906         __u64            bit;
1907         char            *name;
1908         char            *nickname;
1909 } hsm_policy_names[] = {
1910         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1911         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1912         { 0 },
1913 };
1914
1915 /**
1916  * convert a policy name to a bit
1917  * \param name [IN] policy name
1918  * \retval 0 unknown
1919  * \retval   policy bit
1920  */
1921 static __u64 hsm_policy_str2bit(const char *name)
1922 {
1923         int      i;
1924
1925         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1926                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1927                     strcmp(hsm_policy_names[i].name, name) == 0)
1928                         return hsm_policy_names[i].bit;
1929         return 0;
1930 }
1931
1932 /**
1933  * convert a policy bit field to a string
1934  * \param mask [IN] policy bit field
1935  * \param hexa [IN] print mask before bit names
1936  * \param buffer [OUT] string
1937  * \param count [IN] size of buffer
1938  */
1939 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1940                                 const bool hexa)
1941 {
1942         int      i, j;
1943         __u64    bit;
1944         ENTRY;
1945
1946         if (hexa)
1947                 seq_printf(m, "(%#llx) ", mask);
1948
1949         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1950                 bit = (1ULL << i);
1951
1952                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1953                         if (hsm_policy_names[j].bit == bit)
1954                                 break;
1955                 }
1956                 if (bit & mask)
1957                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1958                 else
1959                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1960         }
1961         /* remove last ' ' */
1962         m->count--;
1963         seq_putc(m, '\n');
1964 }
1965
1966 /* methods to read/write HSM policy flags */
1967 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1968 {
1969         struct mdt_device       *mdt = m->private;
1970         struct coordinator      *cdt = &mdt->mdt_coordinator;
1971         ENTRY;
1972
1973         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1974         RETURN(0);
1975 }
1976
1977 static ssize_t
1978 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1979                          size_t count, loff_t *off)
1980 {
1981         struct seq_file         *m = file->private_data;
1982         struct mdt_device       *mdt = m->private;
1983         struct coordinator      *cdt = &mdt->mdt_coordinator;
1984         char                    *start, *token, sign;
1985         char                    *buf;
1986         __u64                    policy;
1987         __u64                    add_mask, remove_mask, set_mask;
1988         int                      rc;
1989         ENTRY;
1990
1991         if (count + 1 > PAGE_SIZE)
1992                 RETURN(-EINVAL);
1993
1994         OBD_ALLOC(buf, count + 1);
1995         if (buf == NULL)
1996                 RETURN(-ENOMEM);
1997
1998         if (copy_from_user(buf, buffer, count))
1999                 GOTO(out, rc = -EFAULT);
2000
2001         buf[count] = '\0';
2002
2003         start = buf;
2004         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
2005                start);
2006
2007         add_mask = remove_mask = set_mask = 0;
2008         do {
2009                 token = strsep(&start, "\n ");
2010                 sign = *token;
2011
2012                 if (sign == '\0')
2013                         continue;
2014
2015                 if (sign == '-' || sign == '+')
2016                         token++;
2017
2018                 policy = hsm_policy_str2bit(token);
2019                 if (policy == 0) {
2020                         CWARN("%s: '%s' is unknown, "
2021                               "supported policies are:\n", mdt_obd_name(mdt),
2022                               token);
2023                         hsm_policy_bit2str(m, 0, false);
2024                         GOTO(out, rc = -EINVAL);
2025                 }
2026                 switch (sign) {
2027                 case '-':
2028                         remove_mask |= policy;
2029                         break;
2030                 case '+':
2031                         add_mask |= policy;
2032                         break;
2033                 default:
2034                         set_mask |= policy;
2035                         break;
2036                 }
2037
2038         } while (start != NULL);
2039
2040         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
2041                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
2042
2043         /* if no sign in all string, it is a clear and set
2044          * if some sign found, all unsigned are converted
2045          * to add
2046          * P1 P2 = set to P1 and P2
2047          * P1 -P2 = add P1 clear P2 same as +P1 -P2
2048          */
2049         if (remove_mask == 0 && add_mask == 0) {
2050                 cdt->cdt_policy = set_mask;
2051         } else {
2052                 cdt->cdt_policy |= set_mask | add_mask;
2053                 cdt->cdt_policy &= ~remove_mask;
2054         }
2055
2056         GOTO(out, rc = count);
2057
2058 out:
2059         OBD_FREE(buf, count + 1);
2060         RETURN(rc);
2061 }
2062 LDEBUGFS_SEQ_FOPS(mdt_hsm_policy);
2063
2064 ssize_t loop_period_show(struct kobject *kobj, struct attribute *attr,
2065                          char *buf)
2066 {
2067         struct coordinator *cdt = container_of(kobj, struct coordinator,
2068                                                cdt_hsm_kobj);
2069
2070         return scnprintf(buf, PAGE_SIZE, "%lu\n", cdt->cdt_loop_period);
2071 }
2072
2073 ssize_t loop_period_store(struct kobject *kobj, struct attribute *attr,
2074                           const char *buffer, size_t count)
2075 {
2076         struct coordinator *cdt = container_of(kobj, struct coordinator,
2077                                                cdt_hsm_kobj);
2078         unsigned int val;
2079         int rc;
2080
2081         rc = kstrtouint(buffer, 0, &val);
2082         if (rc)
2083                 return rc;
2084
2085         if (val != 0)
2086                 cdt->cdt_loop_period = val;
2087
2088         return val ? count : -EINVAL;
2089 }
2090 LUSTRE_RW_ATTR(loop_period);
2091
2092 ssize_t grace_delay_show(struct kobject *kobj, struct attribute *attr,
2093                          char *buf)
2094 {
2095         struct coordinator *cdt = container_of(kobj, struct coordinator,
2096                                                cdt_hsm_kobj);
2097
2098         return scnprintf(buf, PAGE_SIZE, "%lu\n", cdt->cdt_grace_delay);
2099 }
2100
2101 ssize_t grace_delay_store(struct kobject *kobj, struct attribute *attr,
2102                           const char *buffer, size_t count)
2103 {
2104         struct coordinator *cdt = container_of(kobj, struct coordinator,
2105                                                cdt_hsm_kobj);
2106         unsigned int val;
2107         int rc;
2108
2109         rc = kstrtouint(buffer, 0, &val);
2110         if (rc)
2111                 return rc;
2112
2113         if (val != 0)
2114                 cdt->cdt_grace_delay = val;
2115
2116         return val ? count : -EINVAL;
2117 }
2118 LUSTRE_RW_ATTR(grace_delay);
2119
2120 ssize_t active_request_timeout_show(struct kobject *kobj,
2121                                     struct attribute *attr,
2122                                     char *buf)
2123 {
2124         struct coordinator *cdt = container_of(kobj, struct coordinator,
2125                                                cdt_hsm_kobj);
2126
2127         return scnprintf(buf, PAGE_SIZE, "%lu\n", cdt->cdt_active_req_timeout);
2128 }
2129
2130 ssize_t active_request_timeout_store(struct kobject *kobj,
2131                                      struct attribute *attr,
2132                                      const char *buffer, size_t count)
2133 {
2134         struct coordinator *cdt = container_of(kobj, struct coordinator,
2135                                                cdt_hsm_kobj);
2136         unsigned int val;
2137         int rc;
2138
2139         rc = kstrtouint(buffer, 0, &val);
2140         if (rc)
2141                 return rc;
2142
2143         if (val != 0)
2144                 cdt->cdt_active_req_timeout = val;
2145
2146         return val ? count : -EINVAL;
2147 }
2148 LUSTRE_RW_ATTR(active_request_timeout);
2149
2150 ssize_t max_requests_show(struct kobject *kobj, struct attribute *attr,
2151                           char *buf)
2152 {
2153         struct coordinator *cdt = container_of(kobj, struct coordinator,
2154                                                cdt_hsm_kobj);
2155
2156         return scnprintf(buf, PAGE_SIZE, "%llu\n", cdt->cdt_max_requests);
2157 }
2158
2159 ssize_t max_requests_store(struct kobject *kobj, struct attribute *attr,
2160                            const char *buffer, size_t count)
2161 {
2162         struct coordinator *cdt = container_of(kobj, struct coordinator,
2163                                                cdt_hsm_kobj);
2164         unsigned long long val;
2165         int rc;
2166
2167         rc = kstrtoull(buffer, 0, &val);
2168         if (rc)
2169                 return rc;
2170
2171         if (val != 0)
2172                 cdt->cdt_max_requests = val;
2173
2174         return val ? count : -EINVAL;
2175 }
2176 LUSTRE_RW_ATTR(max_requests);
2177
2178 ssize_t default_archive_id_show(struct kobject *kobj, struct attribute *attr,
2179                                 char *buf)
2180 {
2181         struct coordinator *cdt = container_of(kobj, struct coordinator,
2182                                                cdt_hsm_kobj);
2183
2184         return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_default_archive_id);
2185 }
2186
2187 ssize_t default_archive_id_store(struct kobject *kobj, struct attribute *attr,
2188                                  const char *buffer, size_t count)
2189 {
2190         struct coordinator *cdt = container_of(kobj, struct coordinator,
2191                                                cdt_hsm_kobj);
2192         unsigned int val;
2193         int rc;
2194
2195         rc = kstrtouint(buffer, 0, &val);
2196         if (rc)
2197                 return rc;
2198
2199         if (val != 0)
2200                 cdt->cdt_default_archive_id = val;
2201
2202         return val ? count : -EINVAL;
2203 }
2204 LUSTRE_RW_ATTR(default_archive_id);
2205
2206 /*
2207  * procfs write method for MDT/hsm_control
2208  * proc entry is in mdt directory so data is mdt obd_device pointer
2209  */
2210 #define CDT_ENABLE_CMD   "enabled"
2211 #define CDT_STOP_CMD     "shutdown"
2212 #define CDT_DISABLE_CMD  "disabled"
2213 #define CDT_PURGE_CMD    "purge"
2214 #define CDT_HELP_CMD     "help"
2215 #define CDT_MAX_CMD_LEN  10
2216
2217 ssize_t hsm_control_store(struct kobject *kobj, struct attribute *attr,
2218                           const char *buffer, size_t count)
2219 {
2220         struct obd_device *obd = container_of(kobj, struct obd_device,
2221                                               obd_kset.kobj);
2222         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2223         struct coordinator *cdt = &(mdt->mdt_coordinator);
2224         int usage = 0;
2225         int rc = 0;
2226
2227         if (count == 0 || count >= CDT_MAX_CMD_LEN)
2228                 return -EINVAL;
2229
2230         if (strncmp(buffer, CDT_ENABLE_CMD, strlen(CDT_ENABLE_CMD)) == 0) {
2231                 if (cdt->cdt_state == CDT_DISABLE) {
2232                         rc = set_cdt_state(cdt, CDT_RUNNING);
2233                         mdt_hsm_cdt_event(cdt);
2234                         wake_up(&cdt->cdt_waitq);
2235                 } else {
2236                         rc = mdt_hsm_cdt_start(mdt);
2237                 }
2238         } else if (strncmp(buffer, CDT_STOP_CMD, strlen(CDT_STOP_CMD)) == 0) {
2239                 if ((cdt->cdt_state == CDT_STOPPING) ||
2240                     (cdt->cdt_state == CDT_STOPPED)) {
2241                         CERROR("%s: Coordinator already stopped\n",
2242                                mdt_obd_name(mdt));
2243                         rc = -EALREADY;
2244                 } else {
2245                         rc = mdt_hsm_cdt_stop(mdt);
2246                 }
2247         } else if (strncmp(buffer, CDT_DISABLE_CMD,
2248                            strlen(CDT_DISABLE_CMD)) == 0) {
2249                 if ((cdt->cdt_state == CDT_STOPPING) ||
2250                     (cdt->cdt_state == CDT_STOPPED)) {
2251                         CERROR("%s: Coordinator is stopped\n",
2252                                mdt_obd_name(mdt));
2253                         rc = -EINVAL;
2254                 } else {
2255                         rc = set_cdt_state(cdt, CDT_DISABLE);
2256                 }
2257         } else if (strncmp(buffer, CDT_PURGE_CMD,
2258                            strlen(CDT_PURGE_CMD)) == 0) {
2259                 rc = hsm_cancel_all_actions(mdt);
2260         } else if (strncmp(buffer, CDT_HELP_CMD,
2261                            strlen(CDT_HELP_CMD)) == 0) {
2262                 usage = 1;
2263         } else {
2264                 usage = 1;
2265                 rc = -EINVAL;
2266         }
2267
2268         if (usage == 1)
2269                 CERROR("%s: Valid coordinator control commands are: "
2270                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2271                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2272                        CDT_PURGE_CMD, CDT_HELP_CMD);
2273
2274         if (rc)
2275                 RETURN(rc);
2276
2277         RETURN(count);
2278 }
2279
2280 ssize_t hsm_control_show(struct kobject *kobj, struct attribute *attr,
2281                          char *buf)
2282 {
2283         struct obd_device *obd = container_of(kobj, struct obd_device,
2284                                               obd_kset.kobj);
2285         struct coordinator *cdt;
2286
2287         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2288
2289         return scnprintf(buf, PAGE_SIZE, "%s\n",
2290                          cdt_mdt_state2str(cdt->cdt_state));
2291 }
2292
2293 static int
2294 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2295 {
2296         bool first = true;
2297         int i;
2298         ENTRY;
2299
2300         for (i = 0; i < 8 * sizeof(mask); i++) {
2301                 if (mask & (1UL << i)) {
2302                         seq_printf(m, "%s%s", first ? "" : " ",
2303                                    hsm_copytool_action2name(i));
2304                         first = false;
2305                 }
2306         }
2307         seq_putc(m, '\n');
2308
2309         RETURN(0);
2310 }
2311
2312 static int
2313 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2314 {
2315         struct mdt_device *mdt = m->private;
2316         struct coordinator *cdt = &mdt->mdt_coordinator;
2317
2318         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2319 }
2320
2321 static int
2322 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2323 {
2324         struct mdt_device *mdt = m->private;
2325         struct coordinator *cdt = &mdt->mdt_coordinator;
2326
2327         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2328 }
2329
2330 static int
2331 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2332 {
2333         struct mdt_device *mdt = m->private;
2334         struct coordinator *cdt = &mdt->mdt_coordinator;
2335
2336         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2337 }
2338
2339 static inline enum hsm_copytool_action
2340 hsm_copytool_name2action(const char *name)
2341 {
2342         if (strcasecmp(name, "NOOP") == 0)
2343                 return HSMA_NONE;
2344         else if (strcasecmp(name, "ARCHIVE") == 0)
2345                 return HSMA_ARCHIVE;
2346         else if (strcasecmp(name, "RESTORE") == 0)
2347                 return HSMA_RESTORE;
2348         else if (strcasecmp(name, "REMOVE") == 0)
2349                 return HSMA_REMOVE;
2350         else if (strcasecmp(name, "CANCEL") == 0)
2351                 return HSMA_CANCEL;
2352         else
2353                 return -1;
2354 }
2355
2356 static ssize_t
2357 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2358                             size_t user_count, __u64 *mask)
2359 {
2360         char *buf, *pos, *name;
2361         size_t buf_size;
2362         __u64 new_mask = 0;
2363         int rc;
2364         ENTRY;
2365
2366         if (!(user_count < 4096))
2367                 RETURN(-ENOMEM);
2368
2369         buf_size = user_count + 1;
2370
2371         OBD_ALLOC(buf, buf_size);
2372         if (buf == NULL)
2373                 RETURN(-ENOMEM);
2374
2375         if (copy_from_user(buf, user_buf, buf_size - 1))
2376                 GOTO(out, rc = -EFAULT);
2377
2378         buf[buf_size - 1] = '\0';
2379
2380         pos = buf;
2381         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2382                 int action;
2383
2384                 if (*name == '\0')
2385                         continue;
2386
2387                 action = hsm_copytool_name2action(name);
2388                 if (action < 0)
2389                         GOTO(out, rc = -EINVAL);
2390
2391                 new_mask |= (1UL << action);
2392         }
2393
2394         *mask = new_mask;
2395         rc = user_count;
2396 out:
2397         OBD_FREE(buf, buf_size);
2398
2399         RETURN(rc);
2400 }
2401
2402 static ssize_t
2403 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2404                                         size_t count, loff_t *off)
2405 {
2406         struct seq_file         *m = file->private_data;
2407         struct mdt_device       *mdt = m->private;
2408         struct coordinator *cdt = &mdt->mdt_coordinator;
2409
2410         return mdt_write_hsm_request_mask(file, buf, count,
2411                                            &cdt->cdt_user_request_mask);
2412 }
2413
2414 static ssize_t
2415 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2416                                         size_t count, loff_t *off)
2417 {
2418         struct seq_file         *m = file->private_data;
2419         struct mdt_device       *mdt = m->private;
2420         struct coordinator      *cdt = &mdt->mdt_coordinator;
2421
2422         return mdt_write_hsm_request_mask(file, buf, count,
2423                                            &cdt->cdt_group_request_mask);
2424 }
2425
2426 static ssize_t
2427 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2428                                         size_t count, loff_t *off)
2429 {
2430         struct seq_file         *m = file->private_data;
2431         struct mdt_device       *mdt = m->private;
2432         struct coordinator      *cdt = &mdt->mdt_coordinator;
2433
2434         return mdt_write_hsm_request_mask(file, buf, count,
2435                                            &cdt->cdt_other_request_mask);
2436 }
2437
2438 static ssize_t remove_archive_on_last_unlink_show(struct kobject *kobj,
2439                                                   struct attribute *attr,
2440                                                   char *buf)
2441 {
2442         struct coordinator *cdt = container_of(kobj, struct coordinator,
2443                                                cdt_hsm_kobj);
2444
2445         return scnprintf(buf, PAGE_SIZE, "%u\n",
2446                          cdt->cdt_remove_archive_on_last_unlink);
2447 }
2448
2449 static ssize_t remove_archive_on_last_unlink_store(struct kobject *kobj,
2450                                                    struct attribute *attr,
2451                                                    const char *buffer,
2452                                                    size_t count)
2453 {
2454         struct coordinator *cdt = container_of(kobj, struct coordinator,
2455                                                cdt_hsm_kobj);
2456         bool val;
2457         int rc;
2458
2459         rc = kstrtobool(buffer, &val);
2460         if (rc < 0)
2461                 return rc;
2462
2463         cdt->cdt_remove_archive_on_last_unlink = val;
2464         return count;
2465 }
2466 LUSTRE_RW_ATTR(remove_archive_on_last_unlink);
2467
2468 LDEBUGFS_SEQ_FOPS(mdt_hsm_user_request_mask);
2469 LDEBUGFS_SEQ_FOPS(mdt_hsm_group_request_mask);
2470 LDEBUGFS_SEQ_FOPS(mdt_hsm_other_request_mask);
2471
2472 /* Read-only sysfs files for request counters */
2473 static ssize_t archive_count_show(struct kobject *kobj, struct attribute *attr,
2474                                   char *buf)
2475 {
2476         struct coordinator *cdt = container_of(kobj, struct coordinator,
2477                                                cdt_hsm_kobj);
2478
2479         return scnprintf(buf, PAGE_SIZE, "%d\n",
2480                          atomic_read(&cdt->cdt_archive_count));
2481 }
2482 LUSTRE_RO_ATTR(archive_count);
2483
2484 static ssize_t restore_count_show(struct kobject *kobj, struct attribute *attr,
2485                                   char *buf)
2486 {
2487         struct coordinator *cdt = container_of(kobj, struct coordinator,
2488                                                cdt_hsm_kobj);
2489
2490         return scnprintf(buf, PAGE_SIZE, "%d\n",
2491                          atomic_read(&cdt->cdt_restore_count));
2492 }
2493 LUSTRE_RO_ATTR(restore_count);
2494
2495 static ssize_t remove_count_show(struct kobject *kobj, struct attribute *attr,
2496                                  char *buf)
2497 {
2498         struct coordinator *cdt = container_of(kobj, struct coordinator,
2499                                                cdt_hsm_kobj);
2500
2501         return scnprintf(buf, PAGE_SIZE, "%d\n",
2502                          atomic_read(&cdt->cdt_remove_count));
2503 }
2504 LUSTRE_RO_ATTR(remove_count);
2505
2506 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2507         { .name =       "agents",
2508           .fops =       &mdt_hsm_agent_fops                     },
2509         { .name =       "actions",
2510           .fops =       &mdt_hsm_actions_fops,
2511           .proc_mode =  0444                                    },
2512         { .name =       "policy",
2513           .fops =       &mdt_hsm_policy_fops                    },
2514         { .name =       "active_requests",
2515           .fops =       &mdt_hsm_active_requests_fops           },
2516         { .name =       "user_request_mask",
2517           .fops =       &mdt_hsm_user_request_mask_fops,        },
2518         { .name =       "group_request_mask",
2519           .fops =       &mdt_hsm_group_request_mask_fops,       },
2520         { .name =       "other_request_mask",
2521           .fops =       &mdt_hsm_other_request_mask_fops,       },
2522         { 0 }
2523 };
2524
2525 static struct attribute *hsm_attrs[] = {
2526         &lustre_attr_loop_period.attr,
2527         &lustre_attr_grace_delay.attr,
2528         &lustre_attr_active_request_timeout.attr,
2529         &lustre_attr_max_requests.attr,
2530         &lustre_attr_default_archive_id.attr,
2531         &lustre_attr_remove_archive_on_last_unlink.attr,
2532         &lustre_attr_archive_count.attr,
2533         &lustre_attr_restore_count.attr,
2534         &lustre_attr_remove_count.attr,
2535         NULL,
2536 };
2537
2538 static void hsm_kobj_release(struct kobject *kobj)
2539 {
2540         struct coordinator *cdt = container_of(kobj, struct coordinator,
2541                                                cdt_hsm_kobj);
2542
2543         debugfs_remove_recursive(cdt->cdt_debugfs_dir);
2544         cdt->cdt_debugfs_dir = NULL;
2545
2546         complete(&cdt->cdt_kobj_unregister);
2547 }
2548
2549 static struct kobj_type hsm_ktype = {
2550         .default_attrs  = hsm_attrs,
2551         .sysfs_ops      = &lustre_sysfs_ops,
2552         .release        = hsm_kobj_release,
2553 };
2554
2555 /**
2556  * create sysfs entries for coordinator
2557  * \param mdt [IN]
2558  * \retval 0 success
2559  * \retval -ve failure
2560  */
2561 int hsm_cdt_tunables_init(struct mdt_device *mdt)
2562 {
2563         struct coordinator *cdt = &mdt->mdt_coordinator;
2564         struct obd_device *obd = mdt2obd_dev(mdt);
2565         int rc;
2566
2567         init_completion(&cdt->cdt_kobj_unregister);
2568         rc = kobject_init_and_add(&cdt->cdt_hsm_kobj, &hsm_ktype,
2569                                   &obd->obd_kset.kobj, "%s", "hsm");
2570         if (rc) {
2571                 kobject_put(&cdt->cdt_hsm_kobj);
2572                 return rc;
2573         }
2574
2575         /* init debugfs entries, failure is not critical */
2576         cdt->cdt_debugfs_dir = ldebugfs_register("hsm",
2577                                                  obd->obd_debugfs_entry,
2578                                                  lprocfs_mdt_hsm_vars, mdt);
2579         if (IS_ERR_OR_NULL(cdt->cdt_debugfs_dir)) {
2580                 rc = cdt->cdt_debugfs_dir ? PTR_ERR(cdt->cdt_debugfs_dir) :
2581                                             -ENOMEM;
2582                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir, rc = %d\n",
2583                        mdt_obd_name(mdt), rc);
2584                 cdt->cdt_debugfs_dir = NULL;
2585                 kobject_put(&cdt->cdt_hsm_kobj);
2586                 return rc;
2587         }
2588
2589         return 0;
2590 }
2591
2592 /**
2593  * remove sysfs entries for coordinator
2594  *
2595  * @mdt
2596  */
2597 void hsm_cdt_tunables_fini(struct mdt_device *mdt)
2598 {
2599         struct coordinator *cdt = &mdt->mdt_coordinator;
2600
2601         kobject_put(&cdt->cdt_hsm_kobj);
2602         wait_for_completion(&cdt->cdt_kobj_unregister);
2603 }