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 <linux/kernel.h>
  43 #include <obd_support.h>
  44 #include <lustre_export.h>
  45 #include <obd.h>
  46 #include <lprocfs_status.h>
  47 #include <lustre_log.h>
  48 #include <lustre_kernelcomm.h>
  49 #include "mdt_internal.h"
  50
  51 /**
  52  * get obj and HSM attributes on a fid
  53  * \param mti [IN] context
  54  * \param fid [IN] object fid
  55  * \param hsm [OUT] HSM meta data
  56  * \retval obj or error (-ENOENT if not found)
  57  */
  58 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  59                                       const struct lu_fid *fid,
  60                                       struct md_hsm *hsm)
  61 {
  62         struct md_attr          *ma;
  63         struct mdt_object       *obj;
  64         int                      rc;
  65         ENTRY;
  66
  67         ma = &mti->mti_attr;
  68         ma->ma_need = MA_HSM;
  69         ma->ma_valid = 0;
  70
  71         /* find object by FID */
  72         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  73         if (IS_ERR(obj))
  74                 RETURN(obj);
  75
  76         if (!mdt_object_exists(obj)) {
  77                 /* no more object */
  78                 mdt_object_put(mti->mti_env, obj);
  79                 RETURN(ERR_PTR(-ENOENT));
  80         }
  81
  82         rc = mdt_attr_get_complex(mti, obj, ma);
  83         if (rc) {
  84                 mdt_object_put(mti->mti_env, obj);
  85                 RETURN(ERR_PTR(rc));
  86         }
  87
  88         if (ma->ma_valid & MA_HSM)
  89                 *hsm = ma->ma_hsm;
  90         else
  91                 memset(hsm, 0, sizeof(*hsm));
  92         ma->ma_valid = 0;
  93         RETURN(obj);
  94 }
  95
  96 void mdt_hsm_dump_hal(int level, const char *prefix,
  97                       struct hsm_action_list *hal)
  98 {
  99         int                      i, sz;
 100         struct hsm_action_item  *hai;
 101         char                     buf[12];
 102
 103         CDEBUG(level, "%s: HAL header: version %X count %d"
 104                       " archive_id %d flags %#llx\n",
 105                prefix, hal->hal_version, hal->hal_count,
 106                hal->hal_archive_id, hal->hal_flags);
 107
 108         hai = hai_first(hal);
 109         for (i = 0; i < hal->hal_count; i++) {
 110                 sz = hai->hai_len - sizeof(*hai);
 111                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 112                        " cookie=%#llx"
 113                        " action=%s extent=%#llx-%#llx gid=%#llx"
 114                        " datalen=%d data=[%s]\n",
 115                        prefix, i,
 116                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 117                        hai->hai_cookie,
 118                        hsm_copytool_action2name(hai->hai_action),
 119                        hai->hai_extent.offset,
 120                        hai->hai_extent.length,
 121                        hai->hai_gid, sz,
 122                        hai_dump_data_field(hai, buf, sizeof(buf)));
 123                 hai = hai_next(hai);
 124         }
 125 }
 126
 127 /**
 128  * data passed to llog_cat_process() callback
 129  * to scan requests and take actions
 130  */
 131 struct hsm_scan_request {
 132         int                      hal_sz;
 133         int                      hal_used_sz;
 134         struct hsm_action_list  *hal;
 135 };
 136
 137 struct hsm_scan_data {
 138         struct mdt_thread_info  *hsd_mti;
 139         char                     hsd_fsname[MTI_NAME_MAXLEN + 1];
 140         /* are we scanning the logs for housekeeping, or just looking
 141          * for new work?
 142          */
 143         bool                     hsd_housekeeping;
 144         bool                     hsd_one_restore;
 145         u32                      hsd_start_cat_idx;
 146         u32                      hsd_start_rec_idx;
 147         int                      hsd_action_count;
 148         int                      hsd_request_len; /* array alloc len */
 149         int                      hsd_request_count; /* array used count */
 150         struct hsm_scan_request *hsd_request;
 151 };
 152
 153 static int mdt_cdt_waiting_cb(const struct lu_env *env,
 154                               struct mdt_device *mdt,
 155                               struct llog_handle *llh,
 156                               struct llog_agent_req_rec *larr,
 157                               struct hsm_scan_data *hsd)
 158 {
 159         struct coordinator *cdt = &mdt->mdt_coordinator;
 160         struct hsm_scan_request *request;
 161         struct hsm_action_item *hai;
 162         size_t hai_size;
 163         u32 archive_id;
 164         bool wrapped;
 165         int i;
 166
 167         /* Are agents full? */
 168         if (atomic_read(&cdt->cdt_request_count) >= cdt->cdt_max_requests)
 169                 RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
 170
 171         if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
 172             cdt->cdt_max_requests) {
 173                 /* We cannot send any more request
 174                  *
 175                  *                     *** SPECIAL CASE ***
 176                  *
 177                  * Restore requests are too important not to schedule at least
 178                  * one, everytime we can.
 179                  */
 180                 if (larr->arr_hai.hai_action != HSMA_RESTORE ||
 181                     hsd->hsd_one_restore)
 182                         RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
 183         }
 184
 185         hai_size = cfs_size_round(larr->arr_hai.hai_len);
 186         archive_id = larr->arr_archive_id;
 187
 188         /* Can we add this action to one of the existing HALs in hsd. */
 189         request = NULL;
 190         for (i = 0; i < hsd->hsd_request_count; i++) {
 191                 if (hsd->hsd_request[i].hal->hal_archive_id == archive_id &&
 192                     hsd->hsd_request[i].hal_used_sz + hai_size <=
 193                     LDLM_MAXREQSIZE) {
 194                         request = &hsd->hsd_request[i];
 195                         break;
 196                 }
 197         }
 198
 199         /* Are we trying to force-schedule a request? */
 200         if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
 201             cdt->cdt_max_requests) {
 202                 /* Is there really no compatible hsm_scan_request? */
 203                 if (!request) {
 204                         for (i -= 1; i >= 0; i--) {
 205                                 if (hsd->hsd_request[i].hal->hal_archive_id ==
 206                                     archive_id) {
 207                                         request = &hsd->hsd_request[i];
 208                                         break;
 209                                 }
 210                         }
 211                 }
 212
 213                 /* Make room for the hai */
 214                 if (request) {
 215                         /* Discard the last hai until there is enough space */
 216                         do {
 217                                 request->hal->hal_count--;
 218
 219                                 hai = hai_first(request->hal);
 220                                 for (i = 0; i < request->hal->hal_count; i++)
 221                                         hai = hai_next(hai);
 222                                 request->hal_used_sz -=
 223                                         cfs_size_round(hai->hai_len);
 224                                 hsd->hsd_action_count--;
 225                         } while (request->hal_used_sz + hai_size >
 226                                  LDLM_MAXREQSIZE);
 227                 } else if (hsd->hsd_housekeeping) {
 228                         struct hsm_scan_request *tmp;
 229
 230                         /* Discard the (whole) last hal */
 231                         hsd->hsd_request_count--;
 232                         LASSERT(hsd->hsd_request_count >= 0);
 233                         tmp = &hsd->hsd_request[hsd->hsd_request_count];
 234                         hsd->hsd_action_count -= tmp->hal->hal_count;
 235                         LASSERT(hsd->hsd_action_count >= 0);
 236                         OBD_FREE(tmp->hal, tmp->hal_sz);
 237                 } else {
 238                         /* Bailing out, this code path is too hot */
 239                         RETURN(LLOG_PROC_BREAK);
 240
 241                 }
 242         }
 243
 244         if (!request) {
 245                 struct hsm_action_list *hal;
 246
 247                 LASSERT(hsd->hsd_request_count < hsd->hsd_request_len);
 248                 request = &hsd->hsd_request[hsd->hsd_request_count];
 249
 250                 /* allocates hai vector size just needs to be large
 251                  * enough */
 252                 request->hal_sz = sizeof(*request->hal) +
 253                         cfs_size_round(MTI_NAME_MAXLEN + 1) + 2 * hai_size;
 254                 OBD_ALLOC_LARGE(hal, request->hal_sz);
 255                 if (!hal)
 256                         RETURN(-ENOMEM);
 257
 258                 hal->hal_version = HAL_VERSION;
 259                 strlcpy(hal->hal_fsname, hsd->hsd_fsname, MTI_NAME_MAXLEN + 1);
 260                 hal->hal_archive_id = larr->arr_archive_id;
 261                 hal->hal_flags = larr->arr_flags;
 262                 hal->hal_count = 0;
 263                 request->hal_used_sz = hal_size(hal);
 264                 request->hal = hal;
 265                 hsd->hsd_request_count++;
 266         } else if (request->hal_sz < request->hal_used_sz + hai_size) {
 267                 /* Not enough room, need an extension */
 268                 void *hal_buffer;
 269                 int sz;
 270
 271                 sz = min_t(int, 2 * request->hal_sz, LDLM_MAXREQSIZE);
 272                 LASSERT(request->hal_used_sz + hai_size < sz);
 273
 274                 OBD_ALLOC_LARGE(hal_buffer, sz);
 275                 if (!hal_buffer)
 276                         RETURN(-ENOMEM);
 277
 278                 memcpy(hal_buffer, request->hal, request->hal_used_sz);
 279                 OBD_FREE_LARGE(request->hal, request->hal_sz);
 280                 request->hal = hal_buffer;
 281                 request->hal_sz = sz;
 282         }
 283
 284         hai = hai_first(request->hal);
 285         for (i = 0; i < request->hal->hal_count; i++)
 286                 hai = hai_next(hai);
 287
 288         memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 289
 290         request->hal_used_sz += hai_size;
 291         request->hal->hal_count++;
 292
 293         hsd->hsd_action_count++;
 294
 295         switch (hai->hai_action) {
 296         case HSMA_CANCEL:
 297                 break;
 298         case HSMA_RESTORE:
 299                 hsd->hsd_one_restore = true;
 300                 fallthrough;
 301         default:
 302                 cdt_agent_record_hash_add(cdt, hai->hai_cookie,
 303                                           llh->lgh_hdr->llh_cat_idx,
 304                                           larr->arr_hdr.lrh_index);
 305         }
 306
 307         wrapped = llh->lgh_hdr->llh_cat_idx >= llh->lgh_last_idx &&
 308                   llh->lgh_hdr->llh_count > 1;
 309         if ((!wrapped && llh->lgh_hdr->llh_cat_idx > hsd->hsd_start_cat_idx) ||
 310             (wrapped && llh->lgh_hdr->llh_cat_idx < hsd->hsd_start_cat_idx) ||
 311             (llh->lgh_hdr->llh_cat_idx == hsd->hsd_start_cat_idx &&
 312              larr->arr_hdr.lrh_index > hsd->hsd_start_rec_idx)) {
 313                 hsd->hsd_start_cat_idx = llh->lgh_hdr->llh_cat_idx;
 314                 hsd->hsd_start_rec_idx = larr->arr_hdr.lrh_index;
 315         }
 316
 317         RETURN(0);
 318 }
 319
 320 static int mdt_cdt_started_cb(const struct lu_env *env,
 321                               struct mdt_device *mdt,
 322                               struct llog_handle *llh,
 323                               struct llog_agent_req_rec *larr,
 324                               struct hsm_scan_data *hsd)
 325 {
 326         struct coordinator *cdt = &mdt->mdt_coordinator;
 327         struct hsm_action_item *hai = &larr->arr_hai;
 328         struct cdt_agent_req *car;
 329         time64_t now = ktime_get_real_seconds();
 330         time64_t last;
 331         enum changelog_rec_flags clf_flags;
 332         int rc;
 333
 334         if (!hsd->hsd_housekeeping)
 335                 RETURN(0);
 336
 337         /* we search for a running request
 338          * error may happen if coordinator crashes or stopped
 339          * with running request
 340          */
 341         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
 342         if (car == NULL) {
 343                 last = larr->arr_req_change;
 344         } else {
 345                 last = car->car_req_update;
 346         }
 347
 348         /* test if request too long, if yes cancel it
 349          * the same way the copy tool acknowledge a cancel request */
 350         if (now <= last + cdt->cdt_active_req_timeout)
 351                 GOTO(out_car, rc = 0);
 352
 353         dump_llog_agent_req_rec("request timed out, start cleaning", larr);
 354
 355         if (car != NULL) {
 356                 car->car_req_update = now;
 357                 mdt_hsm_agent_update_statistics(cdt, 0, 1, 0, &car->car_uuid);
 358                 /* Remove car from memory list (LU-9075) */
 359                 mdt_cdt_remove_request(cdt, hai->hai_cookie);
 360         }
 361
 362         /* Emit a changelog record for the failed action.*/
 363         clf_flags = 0;
 364         hsm_set_cl_error(&clf_flags, ECANCELED);
 365
 366         switch (hai->hai_action) {
 367         case HSMA_ARCHIVE:
 368                 hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
 369                 break;
 370         case HSMA_RESTORE:
 371                 hsm_set_cl_event(&clf_flags, HE_RESTORE);
 372                 break;
 373         case HSMA_REMOVE:
 374                 hsm_set_cl_event(&clf_flags, HE_REMOVE);
 375                 break;
 376         case HSMA_CANCEL:
 377                 hsm_set_cl_event(&clf_flags, HE_CANCEL);
 378                 break;
 379         default:
 380                 /* Unknown record type, skip changelog. */
 381                 clf_flags = 0;
 382                 break;
 383         }
 384
 385         if (clf_flags != 0)
 386                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
 387                              &hai->hai_fid);
 388
 389         if (hai->hai_action == HSMA_RESTORE)
 390                 cdt_restore_handle_del(hsd->hsd_mti, cdt, &hai->hai_fid);
 391
 392         larr->arr_status = ARS_CANCELED;
 393         larr->arr_req_change = now;
 394         rc = llog_write(hsd->hsd_mti->mti_env, llh, &larr->arr_hdr,
 395                         larr->arr_hdr.lrh_index);
 396         if (rc < 0) {
 397                 CERROR("%s: cannot update agent log: rc = %d\n",
 398                        mdt_obd_name(mdt), rc);
 399                 rc = LLOG_DEL_RECORD;
 400         }
 401
 402         /* ct has completed a request, so a slot is available,
 403          * signal the coordinator to find new work */
 404         mdt_hsm_cdt_event(cdt);
 405 out_car:
 406         if (car != NULL)
 407                 mdt_cdt_put_request(car);
 408
 409         RETURN(rc);
 410 }
 411
 412 /**
 413  *  llog_cat_process() callback, used to:
 414  *  - find waiting request and start action
 415  *  - purge canceled and done requests
 416  * \param env [IN] environment
 417  * \param llh [IN] llog handle
 418  * \param hdr [IN] llog record
 419  * \param data [IN/OUT] cb data = struct hsm_scan_data
 420  * \retval 0 success
 421  * \retval -ve failure
 422  */
 423 static int mdt_coordinator_cb(const struct lu_env *env,
 424                               struct llog_handle *llh,
 425                               struct llog_rec_hdr *hdr,
 426                               void *data)
 427 {
 428         struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
 429         struct hsm_scan_data *hsd = data;
 430         struct mdt_device *mdt = hsd->hsd_mti->mti_mdt;
 431         struct coordinator *cdt = &mdt->mdt_coordinator;
 432         ENTRY;
 433
 434         larr = (struct llog_agent_req_rec *)hdr;
 435         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 436         switch (larr->arr_status) {
 437         case ARS_WAITING:
 438                 RETURN(mdt_cdt_waiting_cb(env, mdt, llh, larr, hsd));
 439         case ARS_STARTED:
 440                 RETURN(mdt_cdt_started_cb(env, mdt, llh, larr, hsd));
 441         default:
 442                 if (!hsd->hsd_housekeeping)
 443                         RETURN(0);
 444
 445                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 446                     ktime_get_real_seconds()) {
 447                         cdt_agent_record_hash_del(cdt,
 448                                                   larr->arr_hai.hai_cookie);
 449                         RETURN(LLOG_DEL_RECORD);
 450                 }
 451
 452                 RETURN(0);
 453         }
 454 }
 455
 456 /* Release the ressource used by the coordinator. Called when the
 457  * coordinator is stopping. */
 458 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
 459 {
 460         struct coordinator              *cdt = &mdt->mdt_coordinator;
 461         struct cdt_agent_req            *car, *tmp1;
 462         struct hsm_agent                *ha, *tmp2;
 463         struct cdt_restore_handle       *crh, *tmp3;
 464         struct mdt_thread_info          *cdt_mti;
 465
 466         /* start cleaning */
 467         down_write(&cdt->cdt_request_lock);
 468         list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
 469                                  car_request_list) {
 470                 cfs_hash_del(cdt->cdt_request_cookie_hash,
 471                              &car->car_hai->hai_cookie,
 472                              &car->car_cookie_hash);
 473                 list_del(&car->car_request_list);
 474                 mdt_cdt_put_request(car);
 475         }
 476         up_write(&cdt->cdt_request_lock);
 477
 478         down_write(&cdt->cdt_agent_lock);
 479         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 480                 list_del(&ha->ha_list);
 481                 if (ha->ha_archive_cnt != 0)
 482                         OBD_FREE_PTR_ARRAY(ha->ha_archive_id,
 483                                            ha->ha_archive_cnt);
 484                 OBD_FREE_PTR(ha);
 485         }
 486         up_write(&cdt->cdt_agent_lock);
 487
 488         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 489         mutex_lock(&cdt->cdt_restore_lock);
 490         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_handle_list,
 491                                  crh_list) {
 492                 /* not locked yet, cleanup by cdt_restore_handle_add() */
 493                 if (crh->crh_lh.mlh_type == MDT_NUL_LOCK)
 494                         continue;
 495                 list_del(&crh->crh_list);
 496                 /* give back layout lock */
 497                 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
 498                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 499         }
 500         mutex_unlock(&cdt->cdt_restore_lock);
 501 }
 502
 503 /*
 504  * Coordinator state transition table, indexed on enum cdt_states, taking
 505  * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
 506  * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
 507  */
 508 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
 509         /* from -> to:    stopped init   running disable stopping */
 510         /* stopped */   { true,   true,  false,  false,  false },
 511         /* init */      { true,   false, true,   false,  false },
 512         /* running */   { false,  false, true,   true,   true },
 513         /* disable */   { false,  false, true,   true,   true },
 514         /* stopping */  { true,   false, false,  false,  false }
 515 };
 516
 517 /**
 518  * Change coordinator thread state
 519  * Some combinations are not valid, so catch them here.
 520  *
 521  * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
 522  * the transition was not possible.
 523  */
 524 static int set_cdt_state_locked(struct coordinator *cdt,
 525                                 enum cdt_states new_state)
 526 {
 527         int rc;
 528         enum cdt_states state;
 529
 530         state = cdt->cdt_state;
 531
 532         if (cdt_transition[state][new_state]) {
 533                 cdt->cdt_state = new_state;
 534                 rc = 0;
 535         } else {
 536                 CDEBUG(D_HSM,
 537                        "unexpected coordinator transition, from=%s, to=%s\n",
 538                        cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
 539                 rc = -EINVAL;
 540         }
 541
 542         return rc;
 543 }
 544
 545 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state)
 546 {
 547         int rc;
 548
 549         mutex_lock(&cdt->cdt_state_lock);
 550         rc = set_cdt_state_locked(cdt, new_state);
 551         mutex_unlock(&cdt->cdt_state_lock);
 552
 553         return rc;
 554 }
 555
 556 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti);
 557
 558 static void cdt_start_pending_restore(struct mdt_device *mdt,
 559                                       struct coordinator *cdt)
 560 {
 561         struct mdt_thread_info *cdt_mti;
 562         unsigned int i = 0;
 563         int rc;
 564
 565         /* wait until MDD initialize hsm actions llog */
 566         while (!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state) && i < obd_timeout) {
 567                 schedule_timeout_interruptible(cfs_time_seconds(1));
 568                 i++;
 569         }
 570         if (!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state))
 571                 CWARN("%s: trying to init HSM before MDD\n", mdt_obd_name(mdt));
 572
 573         /* set up list of started restore requests */
 574         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 575         rc = mdt_hsm_pending_restore(cdt_mti);
 576         if (rc)
 577                 CERROR("%s: cannot take the layout locks needed for registered restore: %d\n",
 578                        mdt_obd_name(mdt), rc);
 579
 580 }
 581
 582 /**
 583  * coordinator thread
 584  * \param data [IN] obd device
 585  * \retval 0 success
 586  * \retval -ve failure
 587  */
 588 static int mdt_coordinator(void *data)
 589 {
 590         struct mdt_thread_info  *mti = data;
 591         struct mdt_device       *mdt = mti->mti_mdt;
 592         struct coordinator      *cdt = &mdt->mdt_coordinator;
 593         struct hsm_scan_data     hsd = { NULL };
 594         time64_t                 last_housekeeping = 0;
 595         size_t request_sz = 0;
 596         int rc;
 597         ENTRY;
 598
 599         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 600                mdt_obd_name(mdt), current->pid);
 601
 602         hsd.hsd_mti = mti;
 603         obd_uuid2fsname(hsd.hsd_fsname, mdt_obd_name(mdt),
 604                         sizeof(hsd.hsd_fsname));
 605
 606         set_cdt_state(cdt, CDT_RUNNING);
 607
 608         /* Inform mdt_hsm_cdt_start(). */
 609         wake_up(&cdt->cdt_waitq);
 610         cdt_start_pending_restore(mdt, cdt);
 611
 612         while (1) {
 613                 int i;
 614                 int update_idx = 0;
 615                 int updates_sz;
 616                 int updates_cnt;
 617                 u32 start_cat_idx;
 618                 u32 start_rec_idx;
 619                 struct hsm_record_update *updates;
 620
 621                 /* Limit execution of the expensive requests traversal
 622                  * to at most one second. This prevents repeatedly
 623                  * locking/unlocking the catalog for each request
 624                  * and preventing other HSM operations from happening
 625                  */
 626                 wait_event_interruptible_timeout(cdt->cdt_waitq,
 627                                                  kthread_should_stop() ||
 628                                                  cdt->cdt_wakeup_coordinator,
 629                                                  cfs_time_seconds(1));
 630
 631                 cdt->cdt_wakeup_coordinator = false;
 632                 CDEBUG(D_HSM, "coordinator resumes\n");
 633
 634                 if (kthread_should_stop()) {
 635                         CDEBUG(D_HSM, "Coordinator stops\n");
 636                         rc = 0;
 637                         break;
 638                 }
 639
 640                 /* if coordinator is suspended continue to wait */
 641                 if (cdt->cdt_state == CDT_DISABLE) {
 642                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 643                         continue;
 644                 }
 645
 646                 /* If no event, and no housekeeping to do, continue to
 647                  * wait. */
 648                 if (last_housekeeping + cdt->cdt_loop_period <=
 649                     ktime_get_real_seconds()) {
 650                         last_housekeeping = ktime_get_real_seconds();
 651                         hsd.hsd_housekeeping = true;
 652                         start_cat_idx = 0;
 653                         start_rec_idx = 0;
 654                 } else if (cdt->cdt_event) {
 655                         hsd.hsd_housekeeping = false;
 656                         start_cat_idx = hsd.hsd_start_cat_idx;
 657                         start_rec_idx = hsd.hsd_start_rec_idx;
 658                 } else {
 659                         continue;
 660                 }
 661
 662                 cdt->cdt_event = false;
 663
 664                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 665
 666                 if (hsd.hsd_request_len != cdt->cdt_max_requests) {
 667                         /* cdt_max_requests has changed,
 668                          * we need to allocate a new buffer
 669                          */
 670                         struct hsm_scan_request *tmp = NULL;
 671                         int max_requests = cdt->cdt_max_requests;
 672                         OBD_ALLOC_LARGE(tmp, max_requests *
 673                                         sizeof(struct hsm_scan_request));
 674                         if (!tmp) {
 675                                 CERROR("Failed to resize request buffer, "
 676                                        "keeping it at %d\n",
 677                                        hsd.hsd_request_len);
 678                         } else {
 679                                 if (hsd.hsd_request != NULL)
 680                                         OBD_FREE_LARGE(hsd.hsd_request,
 681                                                        request_sz);
 682
 683                                 hsd.hsd_request_len = max_requests;
 684                                 request_sz = hsd.hsd_request_len *
 685                                         sizeof(struct hsm_scan_request);
 686                                 hsd.hsd_request = tmp;
 687                         }
 688                 }
 689
 690                 hsd.hsd_action_count = 0;
 691                 hsd.hsd_request_count = 0;
 692                 hsd.hsd_one_restore = false;
 693
 694                 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
 695                                       &hsd, start_cat_idx, start_rec_idx,
 696                                       WRITE);
 697                 if (rc < 0)
 698                         goto clean_cb_alloc;
 699
 700                 CDEBUG(D_HSM, "found %d requests to send\n",
 701                        hsd.hsd_request_count);
 702
 703                 if (list_empty(&cdt->cdt_agents)) {
 704                         CDEBUG(D_HSM, "no agent available, "
 705                                       "coordinator sleeps\n");
 706                         /* reset HSM scanning index range. */
 707                         hsd.hsd_start_cat_idx = start_cat_idx;
 708                         hsd.hsd_start_rec_idx = start_rec_idx;
 709                         goto clean_cb_alloc;
 710                 }
 711
 712                 /* Compute how many HAI we have in all the requests */
 713                 updates_cnt = 0;
 714                 for (i = 0; i < hsd.hsd_request_count; i++) {
 715                         const struct hsm_scan_request *request =
 716                                 &hsd.hsd_request[i];
 717
 718                         updates_cnt += request->hal->hal_count;
 719                 }
 720
 721                 /* Allocate a temporary array to store the cookies to
 722                  * update, and their status. */
 723                 updates_sz = updates_cnt * sizeof(*updates);
 724                 OBD_ALLOC_LARGE(updates, updates_sz);
 725                 if (updates == NULL) {
 726                         CERROR("%s: Cannot allocate memory (%d bytes) "
 727                                "for %d updates. Too many HSM requests?\n",
 728                                mdt_obd_name(mdt), updates_sz, updates_cnt);
 729                         goto clean_cb_alloc;
 730                 }
 731
 732                 /* here hsd contains a list of requests to be started */
 733                 for (i = 0; i < hsd.hsd_request_count; i++) {
 734                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 735                         struct hsm_action_list  *hal = request->hal;
 736                         struct hsm_action_item  *hai;
 737                         int                      j;
 738
 739                         /* still room for work ? */
 740                         if (atomic_read(&cdt->cdt_request_count) >=
 741                             cdt->cdt_max_requests)
 742                                 break;
 743
 744                         rc = mdt_hsm_agent_send(mti, hal, 0);
 745                         /* if failure, we suppose it is temporary
 746                          * if the copy tool failed to do the request
 747                          * it has to use hsm_progress
 748                          */
 749
 750                         /* set up cookie vector to set records status
 751                          * after copy tools start or failed
 752                          */
 753                         hai = hai_first(hal);
 754                         for (j = 0; j < hal->hal_count; j++) {
 755                                 updates[update_idx].cookie = hai->hai_cookie;
 756                                 updates[update_idx].status =
 757                                         (rc ? ARS_WAITING : ARS_STARTED);
 758                                 hai = hai_next(hai);
 759                                 update_idx++;
 760                         }
 761
 762                         /* TODO: narrow down the HSM action range that already
 763                          * scanned accroding to the cookies when a failure
 764                          * occurs.
 765                          */
 766                         if (rc) {
 767                                 hsd.hsd_start_cat_idx = start_cat_idx;
 768                                 hsd.hsd_start_rec_idx = start_rec_idx;
 769                         }
 770                 }
 771
 772                 if (update_idx) {
 773                         rc = mdt_agent_record_update(mti, updates, update_idx);
 774                         if (rc)
 775                                 CERROR("%s: mdt_agent_record_update() failed, "
 776                                        "rc=%d, cannot update records "
 777                                        "for %d cookies\n",
 778                                        mdt_obd_name(mdt), rc, update_idx);
 779                 }
 780
 781                 OBD_FREE_LARGE(updates, updates_sz);
 782
 783 clean_cb_alloc:
 784                 /* free hal allocated by callback */
 785                 for (i = 0; i < hsd.hsd_request_count; i++) {
 786                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 787
 788                         OBD_FREE_LARGE(request->hal, request->hal_sz);
 789                 }
 790         }
 791
 792         if (hsd.hsd_request != NULL)
 793                 OBD_FREE_LARGE(hsd.hsd_request, request_sz);
 794
 795         mdt_hsm_cdt_cleanup(mdt);
 796
 797         if (rc != 0)
 798                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 799                        mdt_obd_name(mdt), current->pid, rc);
 800         else
 801                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 802                               " no error\n",
 803                        mdt_obd_name(mdt), current->pid);
 804
 805         RETURN(rc);
 806 }
 807
 808 /**
 809  * register a new HSM restore handle for a file and take EX lock on the layout
 810  * \param mti [IN] thread info
 811  * \param cdt [IN] coordinator
 812  * \param fid [IN] fid of the file to restore
 813  * \param he  [IN] HSM extent
 814  * \retval 0 success
 815  * \retval 1 restore handle already exists for the fid
 816  * \retval -ve failure
 817  */
 818 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
 819                            const struct lu_fid *fid,
 820                            const struct hsm_extent *he)
 821 {
 822         struct mdt_lock_handle lh = { 0 };
 823         struct cdt_restore_handle *crh;
 824         struct mdt_object *obj;
 825         int rc;
 826         ENTRY;
 827
 828         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 829         if (crh == NULL)
 830                 RETURN(-ENOMEM);
 831
 832         crh->crh_fid = *fid;
 833         /* in V1 all file is restored
 834          * crh->extent.start = he->offset;
 835          * crh->extent.end = he->offset + he->length;
 836          */
 837         crh->crh_extent.start = 0;
 838         crh->crh_extent.end = he->length;
 839         crh->crh_lh.mlh_type = MDT_NUL_LOCK;
 840
 841         mutex_lock(&cdt->cdt_restore_lock);
 842         if (cdt_restore_handle_find(cdt, fid) != NULL)
 843                 GOTO(out_crl, rc = 1);
 844
 845         if (unlikely(cdt->cdt_state == CDT_STOPPED ||
 846                      cdt->cdt_state == CDT_STOPPING))
 847                 GOTO(out_crl, rc = -EAGAIN);
 848
 849         list_add_tail(&crh->crh_list, &cdt->cdt_restore_handle_list);
 850         mutex_unlock(&cdt->cdt_restore_lock);
 851
 852         /* get the layout lock */
 853         mdt_lock_reg_init(&lh, LCK_EX);
 854         obj = mdt_object_find_lock(mti, &crh->crh_fid, &lh,
 855                                    MDS_INODELOCK_LAYOUT);
 856         if (IS_ERR(obj)) {
 857                 mutex_lock(&cdt->cdt_restore_lock);
 858                 GOTO(out_ldel, rc = PTR_ERR(obj));
 859         }
 860
 861         /* We do not keep a reference on the object during the restore
 862          * which can be very long.
 863          */
 864         mdt_object_put(mti->mti_env, obj);
 865
 866         mutex_lock(&cdt->cdt_restore_lock);
 867         if (unlikely(cdt->cdt_state == CDT_STOPPED ||
 868                      cdt->cdt_state == CDT_STOPPING))
 869                 GOTO(out_lh, rc = -EAGAIN);
 870
 871         crh->crh_lh = lh;
 872         mutex_unlock(&cdt->cdt_restore_lock);
 873
 874         RETURN(0);
 875 out_lh:
 876         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 877 out_ldel:
 878         list_del(&crh->crh_list);
 879 out_crl:
 880         mutex_unlock(&cdt->cdt_restore_lock);
 881         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 882
 883         return rc;
 884 }
 885
 886 /**
 887  * lookup a restore handle by FID
 888  * caller needs to hold cdt_restore_lock
 889  * \param cdt [IN] coordinator
 890  * \param fid [IN] FID
 891  * \retval cdt_restore_handle found
 892  * \retval NULL not found
 893  */
 894 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
 895                                                    const struct lu_fid *fid)
 896 {
 897         struct cdt_restore_handle *crh;
 898         ENTRY;
 899
 900         list_for_each_entry(crh, &cdt->cdt_restore_handle_list, crh_list) {
 901                 if (lu_fid_eq(&crh->crh_fid, fid))
 902                         RETURN(crh);
 903         }
 904
 905         RETURN(NULL);
 906 }
 907
 908 void cdt_restore_handle_del(struct mdt_thread_info *mti,
 909                             struct coordinator *cdt, const struct lu_fid *fid)
 910 {
 911         struct cdt_restore_handle *crh;
 912
 913         /* give back layout lock */
 914         mutex_lock(&cdt->cdt_restore_lock);
 915         crh = cdt_restore_handle_find(cdt, fid);
 916         if (crh != NULL)
 917                 list_del(&crh->crh_list);
 918         mutex_unlock(&cdt->cdt_restore_lock);
 919
 920         if (crh == NULL)
 921                 return;
 922
 923         /* XXX We pass a NULL object since the restore handle does not
 924          * keep a reference on the object being restored. */
 925         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 926         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 927 }
 928
 929 /**
 930  * data passed to llog_cat_process() callback
 931  * to scan requests and take actions
 932  */
 933 struct hsm_restore_data {
 934         struct mdt_thread_info  *hrd_mti;
 935 };
 936
 937 /**
 938  *  llog_cat_process() callback, used to:
 939  *  - find restore request and allocate the restore handle
 940  * \param env [IN] environment
 941  * \param llh [IN] llog handle
 942  * \param hdr [IN] llog record
 943  * \param data [IN/OUT] cb data = struct hsm_restore_data
 944  * \retval 0 success
 945  * \retval -ve failure
 946  */
 947 static int hsm_restore_cb(const struct lu_env *env,
 948                           struct llog_handle *llh,
 949                           struct llog_rec_hdr *hdr, void *data)
 950 {
 951         struct llog_agent_req_rec       *larr;
 952         struct hsm_restore_data         *hrd;
 953         struct hsm_action_item          *hai;
 954         struct mdt_thread_info          *mti;
 955         struct coordinator              *cdt;
 956         int rc;
 957         ENTRY;
 958
 959         hrd = data;
 960         mti = hrd->hrd_mti;
 961         cdt = &mti->mti_mdt->mdt_coordinator;
 962
 963         larr = (struct llog_agent_req_rec *)hdr;
 964         hai = &larr->arr_hai;
 965         if (hai->hai_cookie >= cdt->cdt_last_cookie) {
 966                 /* update the cookie to avoid collision */
 967                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 968         }
 969
 970         if (hai->hai_action != HSMA_RESTORE ||
 971             agent_req_in_final_state(larr->arr_status))
 972                 RETURN(0);
 973
 974         /* restore request not in a final state */
 975
 976         /* force replay of restore requests left in started state from previous
 977          * CDT context, to be canceled later if finally found to be incompatible
 978          * when being re-started */
 979         if (larr->arr_status == ARS_STARTED) {
 980                 larr->arr_status = ARS_WAITING;
 981                 larr->arr_req_change = ktime_get_real_seconds();
 982                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 983                 if (rc != 0)
 984                         GOTO(out, rc);
 985         }
 986
 987         rc = cdt_restore_handle_add(mti, cdt, &hai->hai_fid, &hai->hai_extent);
 988         if (rc == 1)
 989                 rc = 0;
 990 out:
 991         RETURN(rc);
 992 }
 993
 994 /**
 995  * restore coordinator state at startup
 996  * the goal is to take a layout lock for each registered restore request
 997  * \param mti [IN] context
 998  */
 999 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
1000 {
1001         struct hsm_restore_data  hrd;
1002         int                      rc;
1003         ENTRY;
1004
1005         hrd.hrd_mti = mti;
1006
1007         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
1008                               0, 0, WRITE);
1009
1010         RETURN(rc);
1011 }
1012
1013 int hsm_init_ucred(struct lu_ucred *uc)
1014 {
1015         ENTRY;
1016         uc->uc_valid = UCRED_OLD;
1017         uc->uc_o_uid = 0;
1018         uc->uc_o_gid = 0;
1019         uc->uc_o_fsuid = 0;
1020         uc->uc_o_fsgid = 0;
1021         uc->uc_uid = 0;
1022         uc->uc_gid = 0;
1023         uc->uc_fsuid = 0;
1024         uc->uc_fsgid = 0;
1025         uc->uc_suppgids[0] = -1;
1026         uc->uc_suppgids[1] = -1;
1027         uc->uc_cap = cap_combine(CAP_FS_SET, CAP_NFSD_SET);
1028         uc->uc_umask = 0777;
1029         uc->uc_ginfo = NULL;
1030         uc->uc_identity = NULL;
1031         /* always record internal HSM activity if also enabled globally */
1032         uc->uc_enable_audit = 1;
1033
1034         RETURN(0);
1035 }
1036
1037 /**
1038  * initialize coordinator struct
1039  * \param mdt [IN] device
1040  * \retval 0 success
1041  * \retval -ve failure
1042  */
1043 int mdt_hsm_cdt_init(struct mdt_device *mdt)
1044 {
1045         struct coordinator      *cdt = &mdt->mdt_coordinator;
1046         struct mdt_thread_info  *cdt_mti = NULL;
1047         int                      rc;
1048         ENTRY;
1049
1050         init_waitqueue_head(&cdt->cdt_waitq);
1051         init_rwsem(&cdt->cdt_llog_lock);
1052         init_rwsem(&cdt->cdt_agent_lock);
1053         init_rwsem(&cdt->cdt_request_lock);
1054         mutex_init(&cdt->cdt_restore_lock);
1055         mutex_init(&cdt->cdt_state_lock);
1056         set_cdt_state(cdt, CDT_STOPPED);
1057
1058         INIT_LIST_HEAD(&cdt->cdt_request_list);
1059         INIT_LIST_HEAD(&cdt->cdt_agents);
1060         INIT_LIST_HEAD(&cdt->cdt_restore_handle_list);
1061
1062         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
1063                                                        CFS_HASH_BITS_MIN,
1064                                                        CFS_HASH_BITS_MAX,
1065                                                        CFS_HASH_BKT_BITS,
1066                                                        0 /* extra bytes */,
1067                                                        CFS_HASH_MIN_THETA,
1068                                                        CFS_HASH_MAX_THETA,
1069                                                 &cdt_request_cookie_hash_ops,
1070                                                        CFS_HASH_DEFAULT);
1071         if (cdt->cdt_request_cookie_hash == NULL)
1072                 RETURN(-ENOMEM);
1073
1074         cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
1075                                                      CFS_HASH_BITS_MIN,
1076                                                      CFS_HASH_BITS_MAX,
1077                                                      CFS_HASH_BKT_BITS,
1078                                                      0 /* extra bytes */,
1079                                                      CFS_HASH_MIN_THETA,
1080                                                      CFS_HASH_MAX_THETA,
1081                                                      &cdt_agent_record_hash_ops,
1082                                                      CFS_HASH_DEFAULT);
1083         if (cdt->cdt_agent_record_hash == NULL)
1084                 GOTO(out_request_cookie_hash, rc = -ENOMEM);
1085
1086         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
1087         if (rc < 0)
1088                 GOTO(out_agent_record_hash, rc);
1089
1090         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1091         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
1092         if (rc < 0)
1093                 GOTO(out_env, rc);
1094
1095         lu_context_enter(&cdt->cdt_session);
1096         cdt->cdt_env.le_ses = &cdt->cdt_session;
1097
1098         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1099         LASSERT(cdt_mti != NULL);
1100
1101         cdt_mti->mti_env = &cdt->cdt_env;
1102         cdt_mti->mti_mdt = mdt;
1103
1104         hsm_init_ucred(mdt_ucred(cdt_mti));
1105
1106         /* default values for sysfs tunnables
1107          * can be override by MGS conf */
1108         cdt->cdt_default_archive_id = 1;
1109         cdt->cdt_grace_delay = 60;
1110         cdt->cdt_loop_period = 10;
1111         cdt->cdt_max_requests = 3;
1112         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1113         cdt->cdt_active_req_timeout = 3600;
1114
1115         /* by default do not remove archives on last unlink */
1116         cdt->cdt_remove_archive_on_last_unlink = false;
1117
1118         RETURN(0);
1119
1120 out_env:
1121         lu_env_fini(&cdt->cdt_env);
1122 out_agent_record_hash:
1123         cfs_hash_putref(cdt->cdt_agent_record_hash);
1124         cdt->cdt_agent_record_hash = NULL;
1125 out_request_cookie_hash:
1126         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1127         cdt->cdt_request_cookie_hash = NULL;
1128
1129         return rc;
1130 }
1131
1132 /**
1133  * free a coordinator thread
1134  * \param mdt [IN] device
1135  */
1136 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
1137 {
1138         struct coordinator *cdt = &mdt->mdt_coordinator;
1139         ENTRY;
1140
1141         lu_context_exit(cdt->cdt_env.le_ses);
1142         lu_context_fini(cdt->cdt_env.le_ses);
1143
1144         lu_env_fini(&cdt->cdt_env);
1145
1146         cfs_hash_putref(cdt->cdt_agent_record_hash);
1147         cdt->cdt_agent_record_hash = NULL;
1148
1149         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1150         cdt->cdt_request_cookie_hash = NULL;
1151
1152         RETURN(0);
1153 }
1154
1155 /**
1156  * start a coordinator thread
1157  * \param mdt [IN] device
1158  * \retval 0 success
1159  * \retval -ve failure
1160  */
1161 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1162 {
1163         struct coordinator *cdt = &mdt->mdt_coordinator;
1164         struct mdt_thread_info *cdt_mti;
1165         int rc;
1166         void *ptr;
1167         struct task_struct *task;
1168         ENTRY;
1169
1170         /* functions defined but not yet used
1171          * this avoid compilation warning
1172          */
1173         ptr = dump_requests;
1174
1175         rc = set_cdt_state(cdt, CDT_INIT);
1176         if (rc) {
1177                 CERROR("%s: Coordinator already started or stopping\n",
1178                        mdt_obd_name(mdt));
1179                 RETURN(-EALREADY);
1180         }
1181
1182         BUILD_BUG_ON(BIT(CDT_POLICY_SHIFT_COUNT - 1) != CDT_POLICY_LAST);
1183         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1184
1185         /* just need to be larger than previous one */
1186         /* cdt_last_cookie is protected by cdt_llog_lock */
1187         cdt->cdt_last_cookie = ktime_get_real_seconds();
1188         atomic_set(&cdt->cdt_request_count, 0);
1189         atomic_set(&cdt->cdt_archive_count, 0);
1190         atomic_set(&cdt->cdt_restore_count, 0);
1191         atomic_set(&cdt->cdt_remove_count, 0);
1192         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
1193         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
1194         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
1195
1196         /* to avoid deadlock when start is made through sysfs
1197          * sysfs entries are created by the coordinator thread
1198          */
1199         if (mdt->mdt_bottom->dd_rdonly)
1200                 RETURN(0);
1201
1202         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1203         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
1204         if (IS_ERR(task)) {
1205                 rc = PTR_ERR(task);
1206                 set_cdt_state(cdt, CDT_STOPPED);
1207                 CERROR("%s: error starting coordinator thread: %d\n",
1208                        mdt_obd_name(mdt), rc);
1209         } else {
1210                 cdt->cdt_task = task;
1211                 wait_event(cdt->cdt_waitq,
1212                            cdt->cdt_state != CDT_INIT);
1213                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1214                        mdt_obd_name(mdt));
1215                 rc = 0;
1216         }
1217
1218         RETURN(rc);
1219 }
1220
1221 /**
1222  * stop a coordinator thread
1223  * \param mdt [IN] device
1224  */
1225 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1226 {
1227         struct coordinator *cdt = &mdt->mdt_coordinator;
1228         int rc;
1229
1230         ENTRY;
1231         /* stop coordinator thread */
1232         rc = set_cdt_state(cdt, CDT_STOPPING);
1233         if (rc == 0) {
1234                 kthread_stop(cdt->cdt_task);
1235                 cdt->cdt_task = NULL;
1236                 set_cdt_state(cdt, CDT_STOPPED);
1237         }
1238
1239         RETURN(rc);
1240 }
1241
1242 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1243                               const struct lu_fid *fid,
1244                               u32 archive_id)
1245 {
1246         struct mdt_object *obj;
1247         struct md_hsm mh;
1248         int rc;
1249
1250         obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1251         if (IS_ERR(obj))
1252                 GOTO(out, rc = PTR_ERR(obj));
1253
1254         if (mh.mh_flags & HS_EXISTS &&
1255             mh.mh_arch_id == archive_id)
1256                 GOTO(out_obj, rc = 0);
1257
1258         mh.mh_flags |= HS_EXISTS;
1259         mh.mh_arch_id = archive_id;
1260         rc = mdt_hsm_attr_set(mti, obj, &mh);
1261
1262 out_obj:
1263         mdt_object_put(mti->mti_env, obj);
1264 out:
1265         return rc;
1266 }
1267
1268 /**
1269  * register all requests from an hal in the memory list
1270  * \param mti [IN] context
1271  * \param hal [IN] request
1272  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1273  *  which is running the CT
1274  * \retval 0 success
1275  * \retval -ve failure
1276  */
1277 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1278                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1279 {
1280         struct mdt_device       *mdt = mti->mti_mdt;
1281         struct coordinator      *cdt = &mdt->mdt_coordinator;
1282         struct hsm_action_item  *hai;
1283         int                      rc = 0, i;
1284         ENTRY;
1285
1286         /* register request in memory list */
1287         hai = hai_first(hal);
1288         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1289                 struct cdt_agent_req *car;
1290
1291                 /* in case of a cancel request, we first mark the ondisk
1292                  * record of the request we want to stop as canceled
1293                  * this does not change the cancel record
1294                  * it will be done when updating the request status
1295                  */
1296                 if (hai->hai_action == HSMA_CANCEL) {
1297                         struct hsm_record_update update = {
1298                                 .cookie = hai->hai_cookie,
1299                                 .status = ARS_CANCELED,
1300                         };
1301
1302                         rc = mdt_agent_record_update(mti, &update, 1);
1303                         if (rc) {
1304                                 CERROR("%s: mdt_agent_record_update() failed, "
1305                                        "rc=%d, cannot update status to %s "
1306                                        "for cookie %#llx\n",
1307                                        mdt_obd_name(mdt), rc,
1308                                        agent_req_status2name(ARS_CANCELED),
1309                                        hai->hai_cookie);
1310                                 GOTO(out, rc);
1311                         }
1312
1313                         /* find the running request to set it canceled */
1314                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1315                         if (car != NULL) {
1316                                 car->car_canceled = 1;
1317                                 /* uuid has to be changed to the one running the
1318                                 * request to cancel */
1319                                 *uuid = car->car_uuid;
1320                                 mdt_cdt_put_request(car);
1321                         }
1322                         /* no need to memorize cancel request
1323                          * this also avoid a deadlock when we receive
1324                          * a purge all requests command
1325                          */
1326                         continue;
1327                 }
1328
1329                 if (hai->hai_action == HSMA_ARCHIVE) {
1330                         rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1331                                                 hal->hal_archive_id);
1332                         if (rc == -ENOENT)
1333                                 continue;
1334                         else if (rc < 0)
1335                                 GOTO(out, rc);
1336                 }
1337
1338                 car = mdt_cdt_alloc_request(hal->hal_archive_id, hal->hal_flags,
1339                                             uuid, hai);
1340                 if (IS_ERR(car))
1341                         GOTO(out, rc = PTR_ERR(car));
1342
1343                 rc = mdt_cdt_add_request(cdt, car);
1344                 if (rc != 0)
1345                         mdt_cdt_free_request(car);
1346         }
1347 out:
1348         RETURN(rc);
1349 }
1350
1351 /**
1352  * swap layouts between 2 fids
1353  * \param mti [IN] context
1354  * \param obj [IN]
1355  * \param dfid [IN]
1356  * \param mh_common [IN] MD HSM
1357  */
1358 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1359                             struct mdt_object *obj, const struct lu_fid *dfid,
1360                             struct md_hsm *mh_common)
1361 {
1362         struct mdt_object       *dobj;
1363         struct mdt_lock_handle  *dlh;
1364         int                      rc;
1365         ENTRY;
1366
1367         if (!mdt_object_exists(obj))
1368                 GOTO(out, rc = -ENOENT);
1369
1370         /* we already have layout lock on obj so take only
1371          * on dfid */
1372         dlh = &mti->mti_lh[MDT_LH_OLD];
1373         mdt_lock_reg_init(dlh, LCK_EX);
1374         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1375         if (IS_ERR(dobj))
1376                 GOTO(out, rc = PTR_ERR(dobj));
1377
1378         /* if copy tool closes the volatile before sending the final
1379          * progress through llapi_hsm_copy_end(), all the objects
1380          * are removed and mdd_swap_layout LBUG */
1381         if (!mdt_object_exists(dobj)) {
1382                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1383                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1384                 GOTO(out_dobj, rc = -ENOENT);
1385         }
1386         /* Since we only handle restores here, unconditionally use
1387          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1388          * be preserved in case of failure during swap_layout and not
1389          * leave a file in an intermediate but incoherent state.
1390          * But need to setup HSM xattr of data FID before, reuse
1391          * mti and mh presets for FID in hsm_cdt_request_completed(),
1392          * only need to clear RELEASED and DIRTY.
1393          */
1394         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1395         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1396         if (rc == 0)
1397                 rc = mo_swap_layouts(mti->mti_env,
1398                                      mdt_object_child(obj),
1399                                      mdt_object_child(dobj),
1400                                      SWAP_LAYOUTS_MDS_HSM);
1401         if (rc == 0) {
1402                 rc = mdt_lsom_downgrade(mti, obj);
1403                 if (rc)
1404                         CDEBUG(D_INODE,
1405                                "%s: File fid="DFID" SOM "
1406                                "downgrade failed, rc = %d\n",
1407                                mdt_obd_name(mti->mti_mdt),
1408                                PFID(mdt_object_fid(obj)), rc);
1409         }
1410 out_dobj:
1411         mdt_object_unlock_put(mti, dobj, dlh, 1);
1412 out:
1413         RETURN(rc);
1414 }
1415
1416 /**
1417  * update status of a completed request
1418  * \param mti [IN] context
1419  * \param pgs [IN] progress of the copy tool
1420  * \retval 0 success
1421  * \retval -ve failure
1422  */
1423 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1424                                      struct hsm_progress_kernel *pgs,
1425                                      const struct cdt_agent_req *car,
1426                                      enum agent_req_status *status)
1427 {
1428         const struct lu_env *env = mti->mti_env;
1429         struct mdt_device *mdt = mti->mti_mdt;
1430         struct coordinator *cdt = &mdt->mdt_coordinator;
1431         struct mdt_object *obj = NULL;
1432         enum changelog_rec_flags clf_flags = 0;
1433         struct md_hsm mh;
1434         bool is_mh_changed;
1435         bool need_changelog = true;
1436         int rc = 0;
1437
1438         ENTRY;
1439         /* default is to retry */
1440         *status = ARS_WAITING;
1441
1442         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1443          * if error/removed continue anyway to get correct reporting done */
1444         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1445         /* we will update MD HSM only if needed */
1446         is_mh_changed = false;
1447
1448         /* no need to change mh->mh_arch_id
1449          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1450          */
1451         if (pgs->hpk_errval != 0) {
1452                 switch (pgs->hpk_errval) {
1453                 case ENOSYS:
1454                         /* the copy tool does not support cancel
1455                          * so the cancel request is failed
1456                          * As we cannot distinguish a cancel progress
1457                          * from another action progress (they have the
1458                          * same cookie), we suppose here the CT returns
1459                          * ENOSYS only if does not support cancel
1460                          */
1461                         /* this can also happen when cdt calls it to
1462                          * for a timed out request */
1463                         *status = ARS_FAILED;
1464                         /* to have a cancel event in changelog */
1465                         pgs->hpk_errval = ECANCELED;
1466                         break;
1467                 case ECANCELED:
1468                         /* the request record has already been set to
1469                          * ARS_CANCELED, this set the cancel request
1470                          * to ARS_SUCCEED */
1471                         *status = ARS_SUCCEED;
1472                         break;
1473                 default:
1474                         /* retry only if current policy or requested, and
1475                          * object is not on error/removed */
1476                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1477                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1478                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1479                         break;
1480                 }
1481
1482                 rc = hsm_set_cl_error(&clf_flags, pgs->hpk_errval);
1483                 if (rc == -EOVERFLOW) {
1484                         CERROR("%s: Request %#llx on "DFID" failed, error code %d too large\n",
1485                                mdt_obd_name(mdt), pgs->hpk_cookie,
1486                                PFID(&pgs->hpk_fid), (int)abs(pgs->hpk_errval));
1487                         rc = 0;
1488                 }
1489
1490                 switch (car->car_hai->hai_action) {
1491                 case HSMA_ARCHIVE:
1492                         hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1493                         break;
1494                 case HSMA_RESTORE:
1495                         hsm_set_cl_event(&clf_flags, HE_RESTORE);
1496                         break;
1497                 case HSMA_REMOVE:
1498                         hsm_set_cl_event(&clf_flags, HE_REMOVE);
1499                         break;
1500                 case HSMA_CANCEL:
1501                         hsm_set_cl_event(&clf_flags, HE_CANCEL);
1502                         CERROR("%s: Failed request %#llx on "DFID
1503                                " cannot be a CANCEL\n",
1504                                mdt_obd_name(mdt),
1505                                pgs->hpk_cookie,
1506                                PFID(&pgs->hpk_fid));
1507                         break;
1508                 default:
1509                         CERROR("%s: Failed request %#llx on "DFID
1510                                " %d is an unknown action\n",
1511                                mdt_obd_name(mdt),
1512                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1513                                car->car_hai->hai_action);
1514                         rc = -EINVAL;
1515                         break;
1516                 }
1517         } else {
1518                 *status = ARS_SUCCEED;
1519                 switch (car->car_hai->hai_action) {
1520                 case HSMA_ARCHIVE:
1521                         hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1522                         /* set ARCHIVE keep EXIST and clear LOST and
1523                          * DIRTY */
1524                         mh.mh_arch_ver = pgs->hpk_data_version;
1525                         mh.mh_flags |= HS_ARCHIVED;
1526                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1527                         is_mh_changed = true;
1528                         break;
1529                 case HSMA_RESTORE:
1530                         hsm_set_cl_event(&clf_flags, HE_RESTORE);
1531
1532                         /* do not clear RELEASED and DIRTY here
1533                          * this will occur in hsm_swap_layouts()
1534                          */
1535
1536                         /* Restoring has changed the file version on
1537                          * disk. */
1538                         mh.mh_arch_ver = pgs->hpk_data_version;
1539                         is_mh_changed = true;
1540                         break;
1541                 case HSMA_REMOVE:
1542                         hsm_set_cl_event(&clf_flags, HE_REMOVE);
1543                         /* clear ARCHIVED EXISTS and LOST */
1544                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1545                         is_mh_changed = true;
1546                         break;
1547                 case HSMA_CANCEL:
1548                         hsm_set_cl_event(&clf_flags, HE_CANCEL);
1549                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1550                                mdt_obd_name(mdt),
1551                                pgs->hpk_cookie,
1552                                PFID(&pgs->hpk_fid));
1553                         break;
1554                 default:
1555                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1556                                mdt_obd_name(mdt),
1557                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1558                                car->car_hai->hai_action);
1559                         rc = -EINVAL;
1560                         break;
1561                 }
1562         }
1563
1564         /* rc != 0 means error when analysing action, it may come from
1565          * a crasy CT no need to manage DIRTY
1566          * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1567          * filled
1568          */
1569         if (rc == 0 && !IS_ERR(obj))
1570                 hsm_set_cl_flags(&clf_flags,
1571                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1572
1573         /* unlock is done later, after layout lock management */
1574         if (is_mh_changed && !IS_ERR(obj))
1575                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1576
1577         /* we give back layout lock only if restore was successful or
1578          * if no retry will be attempted and if object is still alive,
1579          * in other cases we just unlock the object */
1580         if (car->car_hai->hai_action == HSMA_RESTORE) {
1581                 struct mdt_lock_handle *lh;
1582
1583                 /* restore in data FID done, we swap the layouts
1584                  * only if restore is successful */
1585                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1586                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1587                                               &mh);
1588                         if (rc) {
1589                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1590                                         *status = ARS_FAILED;
1591                                 pgs->hpk_errval = -rc;
1592                                 hsm_set_cl_error(&clf_flags, pgs->hpk_errval);
1593                         }
1594                 }
1595                 /* we have to retry, so keep layout lock */
1596                 if (*status == ARS_WAITING)
1597                         GOTO(out, rc);
1598
1599                 /* restore special case, need to create ChangeLog record
1600                  * before to give back layout lock to avoid concurrent
1601                  * file updater to post out of order ChangeLog */
1602                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1603                              &car->car_hai->hai_fid);
1604                 need_changelog = false;
1605
1606                 cdt_restore_handle_del(mti, cdt, &car->car_hai->hai_fid);
1607                 if (!IS_ERR_OR_NULL(obj)) {
1608                         /* flush UPDATE lock so attributes are upadated */
1609                         lh = &mti->mti_lh[MDT_LH_OLD];
1610                         mdt_lock_reg_init(lh, LCK_EX);
1611                         mdt_object_lock(mti, obj, lh, MDS_INODELOCK_UPDATE);
1612                         mdt_object_unlock(mti, obj, lh, 1);
1613                 }
1614         }
1615
1616         GOTO(out, rc);
1617
1618 out:
1619         /* always add a ChangeLog record */
1620         if (need_changelog)
1621                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1622                              &car->car_hai->hai_fid);
1623
1624         if (!IS_ERR(obj))
1625                 mdt_object_put(mti->mti_env, obj);
1626
1627         RETURN(rc);
1628 }
1629
1630 /**
1631  * update status of a request
1632  * \param mti [IN] context
1633  * \param pgs [IN] progress of the copy tool
1634  * \retval 0 success
1635  * \retval -ve failure
1636  */
1637 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1638                                  struct hsm_progress_kernel *pgs)
1639 {
1640         struct mdt_device       *mdt = mti->mti_mdt;
1641         struct coordinator      *cdt = &mdt->mdt_coordinator;
1642         struct cdt_agent_req    *car;
1643         int                      rc = 0;
1644         ENTRY;
1645
1646         /* no coordinator started, so we cannot serve requests */
1647         if (cdt->cdt_state == CDT_STOPPED)
1648                 RETURN(-EAGAIN);
1649
1650         /* first do sanity checks */
1651         car = mdt_cdt_update_request(cdt, pgs);
1652         if (IS_ERR(car)) {
1653                 CERROR("%s: Cannot find running request for cookie %#llx"
1654                        " on fid="DFID"\n",
1655                        mdt_obd_name(mdt),
1656                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1657
1658                 RETURN(PTR_ERR(car));
1659         }
1660
1661         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1662                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1663                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1664                       hsm_copytool_action2name(car->car_hai->hai_action),
1665                       pgs->hpk_flags, pgs->hpk_errval,
1666                       PFID(&car->car_hai->hai_fid),
1667                       PFID(&car->car_hai->hai_dfid));
1668
1669         /* progress is done on FID or data FID depending of the action and
1670          * of the copy progress */
1671         /* for restore progress is used to send back the data FID to cdt */
1672         if (car->car_hai->hai_action == HSMA_RESTORE &&
1673             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1674                 car->car_hai->hai_dfid = pgs->hpk_fid;
1675
1676         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1677              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1678             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1679              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1680                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1681                        " does not match request FID "DFID" nor data FID "
1682                        DFID"\n",
1683                        mdt_obd_name(mdt),
1684                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1685                        PFID(&car->car_hai->hai_fid),
1686                        PFID(&car->car_hai->hai_dfid));
1687                 GOTO(out, rc = -EINVAL);
1688         }
1689
1690         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1691                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1692                        " is not coherent (err=%d and not completed"
1693                        " (flags=%d))\n",
1694                        mdt_obd_name(mdt),
1695                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1696                        hsm_copytool_action2name(car->car_hai->hai_action),
1697                        pgs->hpk_errval, pgs->hpk_flags);
1698                 GOTO(out, rc = -EINVAL);
1699         }
1700
1701         /* now progress is valid */
1702
1703         /* we use a root like ucred */
1704         hsm_init_ucred(mdt_ucred(mti));
1705
1706         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1707                 enum agent_req_status status;
1708                 struct hsm_record_update update;
1709                 int rc1;
1710
1711                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1712
1713                 CDEBUG(D_HSM, "updating record: fid="DFID" cookie=%#llx action=%s "
1714                               "status=%s\n",
1715                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1716                        hsm_copytool_action2name(car->car_hai->hai_action),
1717                        agent_req_status2name(status));
1718
1719                 /* update record first (LU-9075) */
1720                 update.cookie = pgs->hpk_cookie;
1721                 update.status = status;
1722
1723                 rc1 = mdt_agent_record_update(mti, &update, 1);
1724                 if (rc1)
1725                         CERROR("%s: mdt_agent_record_update() failed,"
1726                                " rc=%d, cannot update status to %s"
1727                                " for cookie %#llx\n",
1728                                mdt_obd_name(mdt), rc1,
1729                                agent_req_status2name(status),
1730                                pgs->hpk_cookie);
1731                 rc = (rc != 0 ? rc : rc1);
1732
1733                 /* then remove request from memory list (LU-9075) */
1734                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1735
1736                 /* ct has completed a request, so a slot is available,
1737                  * signal the coordinator to find new work */
1738                 mdt_hsm_cdt_event(cdt);
1739         } else {
1740                 /* if copytool send a progress on a canceled request
1741                  * we inform copytool it should stop
1742                  */
1743                 if (car->car_canceled == 1)
1744                         rc = -ECANCELED;
1745         }
1746         GOTO(out, rc);
1747
1748 out:
1749         /* remove ref got from mdt_cdt_update_request() */
1750         mdt_cdt_put_request(car);
1751
1752         return rc;
1753 }
1754
1755
1756 /**
1757  *  llog_cat_process() callback, used to:
1758  *  - purge all requests
1759  * \param env [IN] environment
1760  * \param llh [IN] llog handle
1761  * \param hdr [IN] llog record
1762  * \param data [IN] cb data = struct mdt_thread_info
1763  * \retval 0 success
1764  * \retval -ve failure
1765  */
1766 static int mdt_cancel_all_cb(const struct lu_env *env,
1767                              struct llog_handle *llh,
1768                              struct llog_rec_hdr *hdr, void *data)
1769 {
1770         struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
1771         struct hsm_action_item *hai = &larr->arr_hai;
1772         struct mdt_thread_info  *mti = data;
1773         struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
1774         int rc;
1775         ENTRY;
1776
1777         if (larr->arr_status != ARS_WAITING &&
1778             larr->arr_status != ARS_STARTED)
1779                 RETURN(0);
1780
1781         /* Unlock the EX layout lock */
1782         if (hai->hai_action == HSMA_RESTORE)
1783                 cdt_restore_handle_del(mti, cdt, &hai->hai_fid);
1784
1785         larr->arr_status = ARS_CANCELED;
1786         larr->arr_req_change = ktime_get_real_seconds();
1787         rc = llog_write(env, llh, hdr, hdr->lrh_index);
1788         if (rc < 0) {
1789                 CERROR("%s: cannot update agent log: rc = %d\n",
1790                        mdt_obd_name(mti->mti_mdt), rc);
1791                 rc = LLOG_DEL_RECORD;
1792         }
1793
1794         RETURN(rc);
1795 }
1796
1797 /**
1798  * cancel all actions
1799  * \param obd [IN] MDT device
1800  */
1801 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1802 {
1803         struct lu_env                    env;
1804         struct lu_context                session;
1805         struct mdt_thread_info          *mti;
1806         struct coordinator              *cdt = &mdt->mdt_coordinator;
1807         struct cdt_agent_req            *car;
1808         struct hsm_action_list          *hal = NULL;
1809         struct hsm_action_item          *hai;
1810         int                              hal_sz = 0, hal_len, rc;
1811         enum cdt_states                  old_state;
1812         ENTRY;
1813
1814         rc = lu_env_init(&env, LCT_MD_THREAD);
1815         if (rc < 0)
1816                 RETURN(rc);
1817
1818         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1819         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1820         if (rc < 0)
1821                 GOTO(out_env, rc);
1822
1823         lu_context_enter(&session);
1824         env.le_ses = &session;
1825
1826         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1827         LASSERT(mti != NULL);
1828
1829         mti->mti_env = &env;
1830         mti->mti_mdt = mdt;
1831
1832         hsm_init_ucred(mdt_ucred(mti));
1833
1834         mutex_lock(&cdt->cdt_state_lock);
1835         old_state = cdt->cdt_state;
1836
1837         /* disable coordinator */
1838         rc = set_cdt_state_locked(cdt, CDT_DISABLE);
1839         if (rc)
1840                 GOTO(out_cdt_state_unlock, rc);
1841
1842         /* send cancel to all running requests */
1843         down_read(&cdt->cdt_request_lock);
1844         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1845                 mdt_cdt_get_request(car);
1846                 /* request is not yet removed from list, it will be done
1847                  * when copytool will return progress
1848                  */
1849
1850                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1851                         mdt_cdt_put_request(car);
1852                         continue;
1853                 }
1854
1855                 /* needed size */
1856                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1857                           cfs_size_round(car->car_hai->hai_len);
1858
1859                 if (hal_len > hal_sz && hal_sz > 0) {
1860                         /* not enough room, free old buffer */
1861                         OBD_FREE(hal, hal_sz);
1862                         hal = NULL;
1863                 }
1864
1865                 /* empty buffer, allocate one */
1866                 if (hal == NULL) {
1867                         hal_sz = hal_len;
1868                         OBD_ALLOC(hal, hal_sz);
1869                         if (hal == NULL) {
1870                                 mdt_cdt_put_request(car);
1871                                 up_read(&cdt->cdt_request_lock);
1872                                 GOTO(out_cdt_state, rc = -ENOMEM);
1873                         }
1874                 }
1875
1876                 hal->hal_version = HAL_VERSION;
1877                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1878                                 MTI_NAME_MAXLEN);
1879                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1880                 hal->hal_archive_id = car->car_archive_id;
1881                 hal->hal_flags = car->car_flags;
1882                 hal->hal_count = 0;
1883
1884                 hai = hai_first(hal);
1885                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1886                 hai->hai_action = HSMA_CANCEL;
1887                 hal->hal_count = 1;
1888
1889                 /* it is possible to safely call mdt_hsm_agent_send()
1890                  * (ie without a deadlock on cdt_request_lock), because the
1891                  * write lock is taken only if we are not in purge mode
1892                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1893                  *   nor mdt_cdt_remove_request())
1894                  */
1895                 /* no conflict with cdt thread because cdt is disable and we
1896                  * have the request lock */
1897                 mdt_hsm_agent_send(mti, hal, 1);
1898
1899                 mdt_cdt_put_request(car);
1900         }
1901         up_read(&cdt->cdt_request_lock);
1902
1903         if (hal != NULL)
1904                 OBD_FREE(hal, hal_sz);
1905
1906         /* cancel all on-disk records */
1907         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1908                               (void *)mti, 0, 0, WRITE);
1909 out_cdt_state:
1910         /* Enable coordinator, unless the coordinator was stopping. */
1911         set_cdt_state_locked(cdt, old_state);
1912 out_cdt_state_unlock:
1913         mutex_unlock(&cdt->cdt_state_lock);
1914
1915         lu_context_exit(&session);
1916         lu_context_fini(&session);
1917 out_env:
1918         lu_env_fini(&env);
1919
1920         RETURN(rc);
1921 }
1922
1923 /**
1924  * check if a request is compatible with file status
1925  * \param hai [IN] request description
1926  * \param archive_id [IN] request archive id
1927  * \param rq_flags [IN] request flags
1928  * \param hsm [IN] file HSM metadata
1929  * \retval boolean
1930  */
1931 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1932                               u32 archive_id, u64 rq_flags,
1933                               const struct md_hsm *hsm)
1934 {
1935         int      is_compat = false;
1936         int      hsm_flags;
1937         ENTRY;
1938
1939         hsm_flags = hsm->mh_flags;
1940         switch (hai->hai_action) {
1941         case HSMA_ARCHIVE:
1942                 if (!(hsm_flags & HS_NOARCHIVE) &&
1943                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1944                         is_compat = true;
1945
1946                 if (hsm_flags & HS_EXISTS &&
1947                     archive_id != 0 &&
1948                     archive_id != hsm->mh_arch_id)
1949                         is_compat = false;
1950
1951                 break;
1952         case HSMA_RESTORE:
1953                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1954                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1955                         is_compat = true;
1956                 break;
1957         case HSMA_REMOVE:
1958                 if (!(hsm_flags & HS_RELEASED) &&
1959                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1960                         is_compat = true;
1961                 break;
1962         case HSMA_CANCEL:
1963                 is_compat = true;
1964                 break;
1965         }
1966         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1967                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1968                       PFID(&hai->hai_fid),
1969                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1970                       hai->hai_extent.offset, hai->hai_extent.length,
1971                       hsm->mh_flags,
1972                       (is_compat ? "compatible" : "uncompatible"));
1973
1974         RETURN(is_compat);
1975 }
1976
1977 /*
1978  * sysfs interface used to get/set HSM behaviour (cdt->cdt_policy)
1979  */
1980 static const struct {
1981         __u64            bit;
1982         char            *name;
1983         char            *nickname;
1984 } hsm_policy_names[] = {
1985         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1986         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1987         { 0 },
1988 };
1989
1990 /**
1991  * convert a policy name to a bit
1992  * \param name [IN] policy name
1993  * \retval 0 unknown
1994  * \retval   policy bit
1995  */
1996 static __u64 hsm_policy_str2bit(const char *name)
1997 {
1998         int      i;
1999
2000         for (i = 0; hsm_policy_names[i].bit != 0; i++)
2001                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
2002                     strcmp(hsm_policy_names[i].name, name) == 0)
2003                         return hsm_policy_names[i].bit;
2004         return 0;
2005 }
2006
2007 /**
2008  * convert a policy bit field to a string
2009  * \param mask [IN] policy bit field
2010  * \param hexa [IN] print mask before bit names
2011  * \param buffer [OUT] string
2012  * \param count [IN] size of buffer
2013  */
2014 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
2015                                 const bool hexa)
2016 {
2017         int      i, j;
2018         __u64    bit;
2019         ENTRY;
2020
2021         if (hexa)
2022                 seq_printf(m, "(%#llx) ", mask);
2023
2024         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
2025                 bit = (1ULL << i);
2026
2027                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
2028                         if (hsm_policy_names[j].bit == bit)
2029                                 break;
2030                 }
2031                 if (bit & mask)
2032                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
2033                 else
2034                         seq_printf(m, "%s ", hsm_policy_names[j].name);
2035         }
2036         /* remove last ' ' */
2037         m->count--;
2038         seq_putc(m, '\n');
2039 }
2040
2041 /* methods to read/write HSM policy flags */
2042 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
2043 {
2044         struct mdt_device       *mdt = m->private;
2045         struct coordinator      *cdt = &mdt->mdt_coordinator;
2046         ENTRY;
2047
2048         hsm_policy_bit2str(m, cdt->cdt_policy, false);
2049         RETURN(0);
2050 }
2051
2052 static ssize_t
2053 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
2054                          size_t count, loff_t *off)
2055 {
2056         struct seq_file         *m = file->private_data;
2057         struct mdt_device       *mdt = m->private;
2058         struct coordinator      *cdt = &mdt->mdt_coordinator;
2059         char                    *start, *token, sign;
2060         char                    *buf;
2061         __u64                    policy;
2062         __u64                    add_mask, remove_mask, set_mask;
2063         int                      rc;
2064         ENTRY;
2065
2066         if (count + 1 > PAGE_SIZE)
2067                 RETURN(-EINVAL);
2068
2069         OBD_ALLOC(buf, count + 1);
2070         if (buf == NULL)
2071                 RETURN(-ENOMEM);
2072
2073         if (copy_from_user(buf, buffer, count))
2074                 GOTO(out, rc = -EFAULT);
2075
2076         buf[count] = '\0';
2077
2078         start = buf;
2079         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
2080                start);
2081
2082         add_mask = remove_mask = set_mask = 0;
2083         do {
2084                 token = strsep(&start, "\n ");
2085                 sign = *token;
2086
2087                 if (sign == '\0')
2088                         continue;
2089
2090                 if (sign == '-' || sign == '+')
2091                         token++;
2092
2093                 policy = hsm_policy_str2bit(token);
2094                 if (policy == 0) {
2095                         CWARN("%s: '%s' is unknown, "
2096                               "supported policies are:\n", mdt_obd_name(mdt),
2097                               token);
2098                         hsm_policy_bit2str(m, 0, false);
2099                         GOTO(out, rc = -EINVAL);
2100                 }
2101                 switch (sign) {
2102                 case '-':
2103                         remove_mask |= policy;
2104                         break;
2105                 case '+':
2106                         add_mask |= policy;
2107                         break;
2108                 default:
2109                         set_mask |= policy;
2110                         break;
2111                 }
2112
2113         } while (start != NULL);
2114
2115         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
2116                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
2117
2118         /* if no sign in all string, it is a clear and set
2119          * if some sign found, all unsigned are converted
2120          * to add
2121          * P1 P2 = set to P1 and P2
2122          * P1 -P2 = add P1 clear P2 same as +P1 -P2
2123          */
2124         if (remove_mask == 0 && add_mask == 0) {
2125                 cdt->cdt_policy = set_mask;
2126         } else {
2127                 cdt->cdt_policy |= set_mask | add_mask;
2128                 cdt->cdt_policy &= ~remove_mask;
2129         }
2130
2131         GOTO(out, rc = count);
2132
2133 out:
2134         OBD_FREE(buf, count + 1);
2135         RETURN(rc);
2136 }
2137 LDEBUGFS_SEQ_FOPS(mdt_hsm_policy);
2138
2139 ssize_t loop_period_show(struct kobject *kobj, struct attribute *attr,
2140                          char *buf)
2141 {
2142         struct coordinator *cdt = container_of(kobj, struct coordinator,
2143                                                cdt_hsm_kobj);
2144
2145         return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_loop_period);
2146 }
2147
2148 ssize_t loop_period_store(struct kobject *kobj, struct attribute *attr,
2149                           const char *buffer, size_t count)
2150 {
2151         struct coordinator *cdt = container_of(kobj, struct coordinator,
2152                                                cdt_hsm_kobj);
2153         unsigned int val;
2154         int rc;
2155
2156         rc = kstrtouint(buffer, 0, &val);
2157         if (rc)
2158                 return rc;
2159
2160         if (val != 0)
2161                 cdt->cdt_loop_period = val;
2162
2163         return val ? count : -EINVAL;
2164 }
2165 LUSTRE_RW_ATTR(loop_period);
2166
2167 ssize_t grace_delay_show(struct kobject *kobj, struct attribute *attr,
2168                          char *buf)
2169 {
2170         struct coordinator *cdt = container_of(kobj, struct coordinator,
2171                                                cdt_hsm_kobj);
2172
2173         return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_grace_delay);
2174 }
2175
2176 ssize_t grace_delay_store(struct kobject *kobj, struct attribute *attr,
2177                           const char *buffer, size_t count)
2178 {
2179         struct coordinator *cdt = container_of(kobj, struct coordinator,
2180                                                cdt_hsm_kobj);
2181         unsigned int val;
2182         int rc;
2183
2184         rc = kstrtouint(buffer, 0, &val);
2185         if (rc)
2186                 return rc;
2187
2188         if (val != 0)
2189                 cdt->cdt_grace_delay = val;
2190
2191         return val ? count : -EINVAL;
2192 }
2193 LUSTRE_RW_ATTR(grace_delay);
2194
2195 ssize_t active_request_timeout_show(struct kobject *kobj,
2196                                     struct attribute *attr,
2197                                     char *buf)
2198 {
2199         struct coordinator *cdt = container_of(kobj, struct coordinator,
2200                                                cdt_hsm_kobj);
2201
2202         return scnprintf(buf, PAGE_SIZE, "%d\n", cdt->cdt_active_req_timeout);
2203 }
2204
2205 ssize_t active_request_timeout_store(struct kobject *kobj,
2206                                      struct attribute *attr,
2207                                      const char *buffer, size_t count)
2208 {
2209         struct coordinator *cdt = container_of(kobj, struct coordinator,
2210                                                cdt_hsm_kobj);
2211         unsigned int val;
2212         int rc;
2213
2214         rc = kstrtouint(buffer, 0, &val);
2215         if (rc)
2216                 return rc;
2217
2218         if (val != 0)
2219                 cdt->cdt_active_req_timeout = val;
2220
2221         return val ? count : -EINVAL;
2222 }
2223 LUSTRE_RW_ATTR(active_request_timeout);
2224
2225 ssize_t max_requests_show(struct kobject *kobj, struct attribute *attr,
2226                           char *buf)
2227 {
2228         struct coordinator *cdt = container_of(kobj, struct coordinator,
2229                                                cdt_hsm_kobj);
2230
2231         return scnprintf(buf, PAGE_SIZE, "%llu\n", cdt->cdt_max_requests);
2232 }
2233
2234 ssize_t max_requests_store(struct kobject *kobj, struct attribute *attr,
2235                            const char *buffer, size_t count)
2236 {
2237         struct coordinator *cdt = container_of(kobj, struct coordinator,
2238                                                cdt_hsm_kobj);
2239         unsigned long long val;
2240         int rc;
2241
2242         rc = kstrtoull(buffer, 0, &val);
2243         if (rc)
2244                 return rc;
2245
2246         if (val != 0)
2247                 cdt->cdt_max_requests = val;
2248
2249         return val ? count : -EINVAL;
2250 }
2251 LUSTRE_RW_ATTR(max_requests);
2252
2253 ssize_t default_archive_id_show(struct kobject *kobj, struct attribute *attr,
2254                                 char *buf)
2255 {
2256         struct coordinator *cdt = container_of(kobj, struct coordinator,
2257                                                cdt_hsm_kobj);
2258
2259         return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_default_archive_id);
2260 }
2261
2262 ssize_t default_archive_id_store(struct kobject *kobj, struct attribute *attr,
2263                                  const char *buffer, size_t count)
2264 {
2265         struct coordinator *cdt = container_of(kobj, struct coordinator,
2266                                                cdt_hsm_kobj);
2267         unsigned int val;
2268         int rc;
2269
2270         rc = kstrtouint(buffer, 0, &val);
2271         if (rc)
2272                 return rc;
2273
2274         if (val != 0)
2275                 cdt->cdt_default_archive_id = val;
2276
2277         return val ? count : -EINVAL;
2278 }
2279 LUSTRE_RW_ATTR(default_archive_id);
2280
2281 /*
2282  * procfs write method for MDT/hsm_control
2283  * proc entry is in mdt directory so data is mdt obd_device pointer
2284  */
2285 #define CDT_ENABLE_CMD   "enabled"
2286 #define CDT_STOP_CMD     "shutdown"
2287 #define CDT_DISABLE_CMD  "disabled"
2288 #define CDT_PURGE_CMD    "purge"
2289 #define CDT_HELP_CMD     "help"
2290 #define CDT_MAX_CMD_LEN  10
2291
2292 ssize_t hsm_control_store(struct kobject *kobj, struct attribute *attr,
2293                           const char *buffer, size_t count)
2294 {
2295         struct obd_device *obd = container_of(kobj, struct obd_device,
2296                                               obd_kset.kobj);
2297         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2298         struct coordinator *cdt = &(mdt->mdt_coordinator);
2299         int usage = 0;
2300         int rc = 0;
2301
2302         if (count == 0 || count >= CDT_MAX_CMD_LEN)
2303                 return -EINVAL;
2304
2305         if (strncmp(buffer, CDT_ENABLE_CMD, strlen(CDT_ENABLE_CMD)) == 0) {
2306                 if (cdt->cdt_state == CDT_DISABLE) {
2307                         rc = set_cdt_state(cdt, CDT_RUNNING);
2308                         mdt_hsm_cdt_event(cdt);
2309                         wake_up(&cdt->cdt_waitq);
2310                 } else if (cdt->cdt_state == CDT_RUNNING) {
2311                         rc = 0;
2312                 } else {
2313                         rc = mdt_hsm_cdt_start(mdt);
2314                 }
2315         } else if (strncmp(buffer, CDT_STOP_CMD, strlen(CDT_STOP_CMD)) == 0) {
2316                 if (cdt->cdt_state == CDT_STOPPING) {
2317                         CERROR("%s: Coordinator is already stopping\n",
2318                                mdt_obd_name(mdt));
2319                         rc = -EALREADY;
2320                 } else if (cdt->cdt_state == CDT_STOPPED) {
2321                         rc = 0;
2322                 } else {
2323                         rc = mdt_hsm_cdt_stop(mdt);
2324                 }
2325         } else if (strncmp(buffer, CDT_DISABLE_CMD,
2326                            strlen(CDT_DISABLE_CMD)) == 0) {
2327                 if ((cdt->cdt_state == CDT_STOPPING) ||
2328                     (cdt->cdt_state == CDT_STOPPED)) {
2329                         /* exit gracefully if coordinator is being stopped
2330                          * or stopped already.
2331                          */
2332                         rc = 0;
2333                 } else {
2334                         rc = set_cdt_state(cdt, CDT_DISABLE);
2335                 }
2336         } else if (strncmp(buffer, CDT_PURGE_CMD,
2337                            strlen(CDT_PURGE_CMD)) == 0) {
2338                 rc = hsm_cancel_all_actions(mdt);
2339         } else if (strncmp(buffer, CDT_HELP_CMD,
2340                            strlen(CDT_HELP_CMD)) == 0) {
2341                 usage = 1;
2342         } else {
2343                 usage = 1;
2344                 rc = -EINVAL;
2345         }
2346
2347         if (usage == 1)
2348                 CERROR("%s: Valid coordinator control commands are: "
2349                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2350                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2351                        CDT_PURGE_CMD, CDT_HELP_CMD);
2352
2353         if (rc)
2354                 RETURN(rc);
2355
2356         RETURN(count);
2357 }
2358
2359 ssize_t hsm_control_show(struct kobject *kobj, struct attribute *attr,
2360                          char *buf)
2361 {
2362         struct obd_device *obd = container_of(kobj, struct obd_device,
2363                                               obd_kset.kobj);
2364         struct coordinator *cdt;
2365
2366         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2367
2368         return scnprintf(buf, PAGE_SIZE, "%s\n",
2369                          cdt_mdt_state2str(cdt->cdt_state));
2370 }
2371
2372 static int
2373 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2374 {
2375         bool first = true;
2376         int i;
2377         ENTRY;
2378
2379         for (i = 0; i < 8 * sizeof(mask); i++) {
2380                 if (mask & (1UL << i)) {
2381                         seq_printf(m, "%s%s", first ? "" : " ",
2382                                    hsm_copytool_action2name(i));
2383                         first = false;
2384                 }
2385         }
2386         seq_putc(m, '\n');
2387
2388         RETURN(0);
2389 }
2390
2391 static int
2392 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2393 {
2394         struct mdt_device *mdt = m->private;
2395         struct coordinator *cdt = &mdt->mdt_coordinator;
2396
2397         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2398 }
2399
2400 static int
2401 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2402 {
2403         struct mdt_device *mdt = m->private;
2404         struct coordinator *cdt = &mdt->mdt_coordinator;
2405
2406         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2407 }
2408
2409 static int
2410 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2411 {
2412         struct mdt_device *mdt = m->private;
2413         struct coordinator *cdt = &mdt->mdt_coordinator;
2414
2415         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2416 }
2417
2418 static inline enum hsm_copytool_action
2419 hsm_copytool_name2action(const char *name)
2420 {
2421         if (strcasecmp(name, "NOOP") == 0)
2422                 return HSMA_NONE;
2423         else if (strcasecmp(name, "ARCHIVE") == 0)
2424                 return HSMA_ARCHIVE;
2425         else if (strcasecmp(name, "RESTORE") == 0)
2426                 return HSMA_RESTORE;
2427         else if (strcasecmp(name, "REMOVE") == 0)
2428                 return HSMA_REMOVE;
2429         else if (strcasecmp(name, "CANCEL") == 0)
2430                 return HSMA_CANCEL;
2431         else
2432                 return -1;
2433 }
2434
2435 static ssize_t
2436 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2437                             size_t user_count, __u64 *mask)
2438 {
2439         char *buf, *pos, *name;
2440         size_t buf_size;
2441         __u64 new_mask = 0;
2442         int rc;
2443         ENTRY;
2444
2445         if (!(user_count < 4096))
2446                 RETURN(-ENOMEM);
2447
2448         buf_size = user_count + 1;
2449
2450         OBD_ALLOC(buf, buf_size);
2451         if (buf == NULL)
2452                 RETURN(-ENOMEM);
2453
2454         if (copy_from_user(buf, user_buf, buf_size - 1))
2455                 GOTO(out, rc = -EFAULT);
2456
2457         buf[buf_size - 1] = '\0';
2458
2459         pos = buf;
2460         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2461                 int action;
2462
2463                 if (*name == '\0')
2464                         continue;
2465
2466                 action = hsm_copytool_name2action(name);
2467                 if (action < 0)
2468                         GOTO(out, rc = -EINVAL);
2469
2470                 new_mask |= (1UL << action);
2471         }
2472
2473         *mask = new_mask;
2474         rc = user_count;
2475 out:
2476         OBD_FREE(buf, buf_size);
2477
2478         RETURN(rc);
2479 }
2480
2481 static ssize_t
2482 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2483                                         size_t count, loff_t *off)
2484 {
2485         struct seq_file         *m = file->private_data;
2486         struct mdt_device       *mdt = m->private;
2487         struct coordinator *cdt = &mdt->mdt_coordinator;
2488
2489         return mdt_write_hsm_request_mask(file, buf, count,
2490                                            &cdt->cdt_user_request_mask);
2491 }
2492
2493 static ssize_t
2494 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2495                                         size_t count, loff_t *off)
2496 {
2497         struct seq_file         *m = file->private_data;
2498         struct mdt_device       *mdt = m->private;
2499         struct coordinator      *cdt = &mdt->mdt_coordinator;
2500
2501         return mdt_write_hsm_request_mask(file, buf, count,
2502                                            &cdt->cdt_group_request_mask);
2503 }
2504
2505 static ssize_t
2506 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2507                                         size_t count, loff_t *off)
2508 {
2509         struct seq_file         *m = file->private_data;
2510         struct mdt_device       *mdt = m->private;
2511         struct coordinator      *cdt = &mdt->mdt_coordinator;
2512
2513         return mdt_write_hsm_request_mask(file, buf, count,
2514                                            &cdt->cdt_other_request_mask);
2515 }
2516
2517 static ssize_t remove_archive_on_last_unlink_show(struct kobject *kobj,
2518                                                   struct attribute *attr,
2519                                                   char *buf)
2520 {
2521         struct coordinator *cdt = container_of(kobj, struct coordinator,
2522                                                cdt_hsm_kobj);
2523
2524         return scnprintf(buf, PAGE_SIZE, "%u\n",
2525                          cdt->cdt_remove_archive_on_last_unlink);
2526 }
2527
2528 static ssize_t remove_archive_on_last_unlink_store(struct kobject *kobj,
2529                                                    struct attribute *attr,
2530                                                    const char *buffer,
2531                                                    size_t count)
2532 {
2533         struct coordinator *cdt = container_of(kobj, struct coordinator,
2534                                                cdt_hsm_kobj);
2535         bool val;
2536         int rc;
2537
2538         rc = kstrtobool(buffer, &val);
2539         if (rc < 0)
2540                 return rc;
2541
2542         cdt->cdt_remove_archive_on_last_unlink = val;
2543         return count;
2544 }
2545 LUSTRE_RW_ATTR(remove_archive_on_last_unlink);
2546
2547 LDEBUGFS_SEQ_FOPS(mdt_hsm_user_request_mask);
2548 LDEBUGFS_SEQ_FOPS(mdt_hsm_group_request_mask);
2549 LDEBUGFS_SEQ_FOPS(mdt_hsm_other_request_mask);
2550
2551 /* Read-only sysfs files for request counters */
2552 static ssize_t archive_count_show(struct kobject *kobj, struct attribute *attr,
2553                                   char *buf)
2554 {
2555         struct coordinator *cdt = container_of(kobj, struct coordinator,
2556                                                cdt_hsm_kobj);
2557
2558         return scnprintf(buf, PAGE_SIZE, "%d\n",
2559                          atomic_read(&cdt->cdt_archive_count));
2560 }
2561 LUSTRE_RO_ATTR(archive_count);
2562
2563 static ssize_t restore_count_show(struct kobject *kobj, struct attribute *attr,
2564                                   char *buf)
2565 {
2566         struct coordinator *cdt = container_of(kobj, struct coordinator,
2567                                                cdt_hsm_kobj);
2568
2569         return scnprintf(buf, PAGE_SIZE, "%d\n",
2570                          atomic_read(&cdt->cdt_restore_count));
2571 }
2572 LUSTRE_RO_ATTR(restore_count);
2573
2574 static ssize_t remove_count_show(struct kobject *kobj, struct attribute *attr,
2575                                  char *buf)
2576 {
2577         struct coordinator *cdt = container_of(kobj, struct coordinator,
2578                                                cdt_hsm_kobj);
2579
2580         return scnprintf(buf, PAGE_SIZE, "%d\n",
2581                          atomic_read(&cdt->cdt_remove_count));
2582 }
2583 LUSTRE_RO_ATTR(remove_count);
2584
2585 static struct ldebugfs_vars ldebugfs_mdt_hsm_vars[] = {
2586         { .name =       "agents",
2587           .fops =       &mdt_hsm_agent_fops                     },
2588         { .name =       "actions",
2589           .fops =       &mdt_hsm_actions_fops,
2590           .proc_mode =  0444                                    },
2591         { .name =       "policy",
2592           .fops =       &mdt_hsm_policy_fops                    },
2593         { .name =       "active_requests",
2594           .fops =       &mdt_hsm_active_requests_fops           },
2595         { .name =       "user_request_mask",
2596           .fops =       &mdt_hsm_user_request_mask_fops,        },
2597         { .name =       "group_request_mask",
2598           .fops =       &mdt_hsm_group_request_mask_fops,       },
2599         { .name =       "other_request_mask",
2600           .fops =       &mdt_hsm_other_request_mask_fops,       },
2601         { 0 }
2602 };
2603
2604 static struct attribute *hsm_attrs[] = {
2605         &lustre_attr_loop_period.attr,
2606         &lustre_attr_grace_delay.attr,
2607         &lustre_attr_active_request_timeout.attr,
2608         &lustre_attr_max_requests.attr,
2609         &lustre_attr_default_archive_id.attr,
2610         &lustre_attr_remove_archive_on_last_unlink.attr,
2611         &lustre_attr_archive_count.attr,
2612         &lustre_attr_restore_count.attr,
2613         &lustre_attr_remove_count.attr,
2614         NULL,
2615 };
2616
2617 KOBJ_ATTRIBUTE_GROUPS(hsm); /* creates hsm_groups from hsm_attrs */
2618
2619 static void hsm_kobj_release(struct kobject *kobj)
2620 {
2621         struct coordinator *cdt = container_of(kobj, struct coordinator,
2622                                                cdt_hsm_kobj);
2623
2624         debugfs_remove_recursive(cdt->cdt_debugfs_dir);
2625         cdt->cdt_debugfs_dir = NULL;
2626
2627         complete(&cdt->cdt_kobj_unregister);
2628 }
2629
2630 static struct kobj_type hsm_ktype = {
2631         .default_groups = KOBJ_ATTR_GROUPS(hsm),
2632         .sysfs_ops      = &lustre_sysfs_ops,
2633         .release        = hsm_kobj_release,
2634 };
2635
2636 /**
2637  * create sysfs entries for coordinator
2638  * \param mdt [IN]
2639  * \retval 0 success
2640  * \retval -ve failure
2641  */
2642 int hsm_cdt_tunables_init(struct mdt_device *mdt)
2643 {
2644         struct coordinator *cdt = &mdt->mdt_coordinator;
2645         struct obd_device *obd = mdt2obd_dev(mdt);
2646         int rc;
2647
2648         init_completion(&cdt->cdt_kobj_unregister);
2649         rc = kobject_init_and_add(&cdt->cdt_hsm_kobj, &hsm_ktype,
2650                                   &obd->obd_kset.kobj, "%s", "hsm");
2651         if (rc) {
2652                 kobject_put(&cdt->cdt_hsm_kobj);
2653                 return rc;
2654         }
2655
2656         /* init debugfs entries, failure is not critical */
2657         cdt->cdt_debugfs_dir = debugfs_create_dir("hsm",
2658                                                   obd->obd_debugfs_entry);
2659         ldebugfs_add_vars(cdt->cdt_debugfs_dir, ldebugfs_mdt_hsm_vars, mdt);
2660
2661         return 0;
2662 }
2663
2664 /**
2665  * remove sysfs entries for coordinator
2666  *
2667  * @mdt
2668  */
2669 void hsm_cdt_tunables_fini(struct mdt_device *mdt)
2670 {
2671         struct coordinator *cdt = &mdt->mdt_coordinator;
2672
2673         kobject_put(&cdt->cdt_hsm_kobj);
2674         wait_for_completion(&cdt->cdt_kobj_unregister);
2675 }