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