lustre/mdt/mdt_coordinator.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
  24  *                          alternatives
  25  *
  26  * Copyright (c) 2013, 2017, Intel Corporation.
  27  * Use is subject to license terms.
  28  */
  29 /*
  30  * lustre/mdt/mdt_coordinator.c
  31  *
  32  * Lustre HSM Coordinator
  33  *
  34  * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
  35  * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
  36  * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <linux/kthread.h>
  42 #include <obd_support.h>
  43 #include <lustre_export.h>
  44 #include <obd.h>
  45 #include <lprocfs_status.h>
  46 #include <lustre_log.h>
  47 #include <lustre_kernelcomm.h>
  48 #include "mdt_internal.h"
  49
  50 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
  51
  52 /**
  53  * get obj and HSM attributes on a fid
  54  * \param mti [IN] context
  55  * \param fid [IN] object fid
  56  * \param hsm [OUT] HSM meta data
  57  * \retval obj or error (-ENOENT if not found)
  58  */
  59 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  60                                       const struct lu_fid *fid,
  61                                       struct md_hsm *hsm)
  62 {
  63         struct md_attr          *ma;
  64         struct mdt_object       *obj;
  65         int                      rc;
  66         ENTRY;
  67
  68         ma = &mti->mti_attr;
  69         ma->ma_need = MA_HSM;
  70         ma->ma_valid = 0;
  71
  72         /* find object by FID */
  73         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  74         if (IS_ERR(obj))
  75                 RETURN(obj);
  76
  77         if (!mdt_object_exists(obj)) {
  78                 /* no more object */
  79                 mdt_object_put(mti->mti_env, obj);
  80                 RETURN(ERR_PTR(-ENOENT));
  81         }
  82
  83         rc = mdt_attr_get_complex(mti, obj, ma);
  84         if (rc) {
  85                 mdt_object_put(mti->mti_env, obj);
  86                 RETURN(ERR_PTR(rc));
  87         }
  88
  89         if (ma->ma_valid & MA_HSM)
  90                 *hsm = ma->ma_hsm;
  91         else
  92                 memset(hsm, 0, sizeof(*hsm));
  93         ma->ma_valid = 0;
  94         RETURN(obj);
  95 }
  96
  97 void mdt_hsm_dump_hal(int level, const char *prefix,
  98                       struct hsm_action_list *hal)
  99 {
 100         int                      i, sz;
 101         struct hsm_action_item  *hai;
 102         char                     buf[12];
 103
 104         CDEBUG(level, "%s: HAL header: version %X count %d"
 105                       " archive_id %d flags %#llx\n",
 106                prefix, hal->hal_version, hal->hal_count,
 107                hal->hal_archive_id, hal->hal_flags);
 108
 109         hai = hai_first(hal);
 110         for (i = 0; i < hal->hal_count; i++) {
 111                 sz = hai->hai_len - sizeof(*hai);
 112                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 113                        " cookie=%#llx"
 114                        " action=%s extent=%#llx-%#llx gid=%#llx"
 115                        " datalen=%d data=[%s]\n",
 116                        prefix, i,
 117                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 118                        hai->hai_cookie,
 119                        hsm_copytool_action2name(hai->hai_action),
 120                        hai->hai_extent.offset,
 121                        hai->hai_extent.length,
 122                        hai->hai_gid, sz,
 123                        hai_dump_data_field(hai, buf, sizeof(buf)));
 124                 hai = hai_next(hai);
 125         }
 126 }
 127
 128 /**
 129  * data passed to llog_cat_process() callback
 130  * to scan requests and take actions
 131  */
 132 struct hsm_scan_request {
 133         int                      hal_sz;
 134         int                      hal_used_sz;
 135         struct hsm_action_list  *hal;
 136 };
 137
 138 struct hsm_scan_data {
 139         struct mdt_thread_info  *hsd_mti;
 140         char                     hsd_fsname[MTI_NAME_MAXLEN + 1];
 141         /* are we scanning the logs for housekeeping, or just looking
 142          * for new work?
 143          */
 144         bool                     hsd_housekeeping;
 145         int                      hsd_action_count;
 146         int                      hsd_request_len; /* array alloc len */
 147         int                      hsd_request_count; /* array used count */
 148         struct hsm_scan_request *hsd_request;
 149 };
 150
 151 static int mdt_cdt_waiting_cb(const struct lu_env *env,
 152                               struct mdt_device *mdt,
 153                               struct llog_handle *llh,
 154                               struct llog_agent_req_rec *larr,
 155                               struct hsm_scan_data *hsd)
 156 {
 157         struct coordinator *cdt = &mdt->mdt_coordinator;
 158         struct hsm_scan_request *request;
 159         struct hsm_action_item *hai;
 160         size_t hai_size;
 161         u32 archive_id;
 162         int i;
 163
 164         /* Are agents full? */
 165         if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
 166             cdt->cdt_max_requests) {
 167                 if (hsd->hsd_housekeeping) {
 168                         /* Unknown request and no more room for a new
 169                          * request. Continue to scan to find other
 170                          * entries for already existing requests. */
 171                         RETURN(0);
 172                 } else {
 173                         /* We cannot send and more requests, stop
 174                          * here. There might be more known requests
 175                          * that could be merged, but this avoid
 176                          * analyzing too many llogs for minor
 177                          * gains. */
 178                         RETURN(LLOG_PROC_BREAK);
 179                 }
 180         }
 181
 182         hai_size = cfs_size_round(larr->arr_hai.hai_len);
 183         archive_id = larr->arr_archive_id;
 184
 185         /* Can we add this action to one of the existing HALs in hsd. */
 186         request = NULL;
 187         for (i = 0; i < hsd->hsd_request_count; i++) {
 188                 if (hsd->hsd_request[i].hal->hal_archive_id == archive_id &&
 189                     hsd->hsd_request[i].hal_used_sz + hai_size <=
 190                     LDLM_MAXREQSIZE) {
 191                         request = &hsd->hsd_request[i];
 192                         break;
 193                 }
 194         }
 195
 196         if (!request) {
 197                 struct hsm_action_list *hal;
 198
 199                 if (hsd->hsd_request_count == hsd->hsd_request_len) {
 200                         /* Logic as above. */
 201                         if (hsd->hsd_housekeeping)
 202                                 RETURN(0);
 203                         else
 204                                 RETURN(LLOG_PROC_BREAK);
 205                 }
 206
 207                 request = &hsd->hsd_request[hsd->hsd_request_count];
 208
 209                 /* allocates hai vector size just needs to be large
 210                  * enough */
 211                 request->hal_sz = sizeof(*request->hal) +
 212                         cfs_size_round(MTI_NAME_MAXLEN + 1) + 2 * hai_size;
 213                 OBD_ALLOC_LARGE(hal, request->hal_sz);
 214                 if (!hal)
 215                         RETURN(-ENOMEM);
 216
 217                 hal->hal_version = HAL_VERSION;
 218                 strlcpy(hal->hal_fsname, hsd->hsd_fsname, MTI_NAME_MAXLEN + 1);
 219                 hal->hal_archive_id = larr->arr_archive_id;
 220                 hal->hal_flags = larr->arr_flags;
 221                 hal->hal_count = 0;
 222                 request->hal_used_sz = hal_size(hal);
 223                 request->hal = hal;
 224                 hsd->hsd_request_count++;
 225         } else if (request->hal_sz < request->hal_used_sz + hai_size) {
 226                 /* Not enough room, need an extension */
 227                 void *hal_buffer;
 228                 int sz;
 229
 230                 sz = min_t(int, 2 * request->hal_sz, LDLM_MAXREQSIZE);
 231                 LASSERT(request->hal_used_sz + hai_size < sz);
 232
 233                 OBD_ALLOC_LARGE(hal_buffer, sz);
 234                 if (!hal_buffer)
 235                         RETURN(-ENOMEM);
 236
 237                 memcpy(hal_buffer, request->hal, request->hal_used_sz);
 238                 OBD_FREE_LARGE(request->hal, request->hal_sz);
 239                 request->hal = hal_buffer;
 240                 request->hal_sz = sz;
 241         }
 242
 243         hai = hai_first(request->hal);
 244         for (i = 0; i < request->hal->hal_count; i++)
 245                 hai = hai_next(hai);
 246
 247         memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 248
 249         request->hal_used_sz += cfs_size_round(hai->hai_len);
 250         request->hal->hal_count++;
 251
 252         hsd->hsd_action_count++;
 253
 254         if (hai->hai_action != HSMA_CANCEL)
 255                 cdt_agent_record_hash_add(cdt, hai->hai_cookie,
 256                                           llh->lgh_hdr->llh_cat_idx,
 257                                           larr->arr_hdr.lrh_index);
 258
 259         RETURN(0);
 260 }
 261
 262 static int mdt_cdt_started_cb(const struct lu_env *env,
 263                               struct mdt_device *mdt,
 264                               struct llog_handle *llh,
 265                               struct llog_agent_req_rec *larr,
 266                               struct hsm_scan_data *hsd)
 267 {
 268         struct coordinator *cdt = &mdt->mdt_coordinator;
 269         struct hsm_action_item *hai = &larr->arr_hai;
 270         struct cdt_agent_req *car;
 271         time64_t now = ktime_get_real_seconds();
 272         time64_t last;
 273         int cl_flags;
 274         int rc;
 275
 276         if (!hsd->hsd_housekeeping)
 277                 RETURN(0);
 278
 279         /* we search for a running request
 280          * error may happen if coordinator crashes or stopped
 281          * with running request
 282          */
 283         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
 284         if (car == NULL) {
 285                 last = larr->arr_req_change;
 286         } else {
 287                 last = car->car_req_update;
 288         }
 289
 290         /* test if request too long, if yes cancel it
 291          * the same way the copy tool acknowledge a cancel request */
 292         if (now <= last + cdt->cdt_active_req_timeout)
 293                 GOTO(out_car, rc = 0);
 294
 295         dump_llog_agent_req_rec("request timed out, start cleaning", larr);
 296
 297         if (car != NULL) {
 298                 car->car_req_update = now;
 299                 mdt_hsm_agent_update_statistics(cdt, 0, 1, 0, &car->car_uuid);
 300                 /* Remove car from memory list (LU-9075) */
 301                 mdt_cdt_remove_request(cdt, hai->hai_cookie);
 302         }
 303
 304         /* Emit a changelog record for the failed action.*/
 305         cl_flags = 0;
 306         hsm_set_cl_error(&cl_flags, ECANCELED);
 307
 308         switch (hai->hai_action) {
 309         case HSMA_ARCHIVE:
 310                 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
 311                 break;
 312         case HSMA_RESTORE:
 313                 hsm_set_cl_event(&cl_flags, HE_RESTORE);
 314                 break;
 315         case HSMA_REMOVE:
 316                 hsm_set_cl_event(&cl_flags, HE_REMOVE);
 317                 break;
 318         case HSMA_CANCEL:
 319                 hsm_set_cl_event(&cl_flags, HE_CANCEL);
 320                 break;
 321         default:
 322                 /* Unknown record type, skip changelog. */
 323                 cl_flags = 0;
 324                 break;
 325         }
 326
 327         if (cl_flags != 0)
 328                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
 329                              &hai->hai_fid);
 330
 331         if (hai->hai_action == HSMA_RESTORE)
 332                 cdt_restore_handle_del(hsd->hsd_mti, cdt, &hai->hai_fid);
 333
 334         larr->arr_status = ARS_CANCELED;
 335         larr->arr_req_change = now;
 336         rc = llog_write(hsd->hsd_mti->mti_env, llh, &larr->arr_hdr,
 337                         larr->arr_hdr.lrh_index);
 338         if (rc < 0) {
 339                 CERROR("%s: cannot update agent log: rc = %d\n",
 340                        mdt_obd_name(mdt), rc);
 341                 rc = LLOG_DEL_RECORD;
 342         }
 343
 344         /* ct has completed a request, so a slot is available,
 345          * signal the coordinator to find new work */
 346         mdt_hsm_cdt_event(cdt);
 347 out_car:
 348         if (car != NULL)
 349                 mdt_cdt_put_request(car);
 350
 351         RETURN(rc);
 352 }
 353
 354 /**
 355  *  llog_cat_process() callback, used to:
 356  *  - find waiting request and start action
 357  *  - purge canceled and done requests
 358  * \param env [IN] environment
 359  * \param llh [IN] llog handle
 360  * \param hdr [IN] llog record
 361  * \param data [IN/OUT] cb data = struct hsm_scan_data
 362  * \retval 0 success
 363  * \retval -ve failure
 364  */
 365 static int mdt_coordinator_cb(const struct lu_env *env,
 366                               struct llog_handle *llh,
 367                               struct llog_rec_hdr *hdr,
 368                               void *data)
 369 {
 370         struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
 371         struct hsm_scan_data *hsd = data;
 372         struct mdt_device *mdt = hsd->hsd_mti->mti_mdt;
 373         struct coordinator *cdt = &mdt->mdt_coordinator;
 374         ENTRY;
 375
 376         larr = (struct llog_agent_req_rec *)hdr;
 377         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 378         switch (larr->arr_status) {
 379         case ARS_WAITING:
 380                 RETURN(mdt_cdt_waiting_cb(env, mdt, llh, larr, hsd));
 381         case ARS_STARTED:
 382                 RETURN(mdt_cdt_started_cb(env, mdt, llh, larr, hsd));
 383         default:
 384                 if (!hsd->hsd_housekeeping)
 385                         RETURN(0);
 386
 387                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 388                     ktime_get_real_seconds()) {
 389                         cdt_agent_record_hash_del(cdt,
 390                                                   larr->arr_hai.hai_cookie);
 391                         RETURN(LLOG_DEL_RECORD);
 392                 }
 393
 394                 RETURN(0);
 395         }
 396 }
 397
 398 /**
 399  * create /proc entries for coordinator
 400  * \param mdt [IN]
 401  * \retval 0 success
 402  * \retval -ve failure
 403  */
 404 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 405 {
 406         struct coordinator      *cdt = &mdt->mdt_coordinator;
 407         int                      rc = 0;
 408         ENTRY;
 409
 410         /* init /proc entries, failure is not critical */
 411         cdt->cdt_proc_dir = lprocfs_register("hsm",
 412                                              mdt2obd_dev(mdt)->obd_proc_entry,
 413                                              lprocfs_mdt_hsm_vars, mdt);
 414         if (IS_ERR(cdt->cdt_proc_dir)) {
 415                 rc = PTR_ERR(cdt->cdt_proc_dir);
 416                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 417                        " rc=%d\n", mdt_obd_name(mdt), rc);
 418                 cdt->cdt_proc_dir = NULL;
 419                 RETURN(rc);
 420         }
 421
 422         RETURN(0);
 423 }
 424
 425 /**
 426  * remove /proc entries for coordinator
 427  * \param mdt [IN]
 428  */
 429 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
 430 {
 431         struct coordinator *cdt = &mdt->mdt_coordinator;
 432
 433         if (cdt->cdt_proc_dir != NULL)
 434                 lprocfs_remove(&cdt->cdt_proc_dir);
 435 }
 436
 437 /**
 438  * get vector of hsm cdt /proc vars
 439  * \param none
 440  * \retval var vector
 441  */
 442 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 443 {
 444         return lprocfs_mdt_hsm_vars;
 445 }
 446
 447 /* Release the ressource used by the coordinator. Called when the
 448  * coordinator is stopping. */
 449 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
 450 {
 451         struct coordinator              *cdt = &mdt->mdt_coordinator;
 452         struct cdt_agent_req            *car, *tmp1;
 453         struct hsm_agent                *ha, *tmp2;
 454         struct cdt_restore_handle       *crh, *tmp3;
 455         struct mdt_thread_info          *cdt_mti;
 456
 457         /* start cleaning */
 458         down_write(&cdt->cdt_request_lock);
 459         list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
 460                                  car_request_list) {
 461                 cfs_hash_del(cdt->cdt_request_cookie_hash,
 462                              &car->car_hai->hai_cookie,
 463                              &car->car_cookie_hash);
 464                 list_del(&car->car_request_list);
 465                 mdt_cdt_put_request(car);
 466         }
 467         up_write(&cdt->cdt_request_lock);
 468
 469         down_write(&cdt->cdt_agent_lock);
 470         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 471                 list_del(&ha->ha_list);
 472                 if (ha->ha_archive_cnt != 0)
 473                         OBD_FREE(ha->ha_archive_id, ha->ha_archive_cnt *
 474                                  sizeof(*ha->ha_archive_id));
 475                 OBD_FREE_PTR(ha);
 476         }
 477         up_write(&cdt->cdt_agent_lock);
 478
 479         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 480         mutex_lock(&cdt->cdt_restore_lock);
 481         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_handle_list,
 482                                  crh_list) {
 483                 list_del(&crh->crh_list);
 484                 /* give back layout lock */
 485                 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
 486                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 487         }
 488         mutex_unlock(&cdt->cdt_restore_lock);
 489 }
 490
 491 /*
 492  * Coordinator state transition table, indexed on enum cdt_states, taking
 493  * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
 494  * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
 495  */
 496 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
 497         /* from -> to:    stopped init   running disable stopping */
 498         /* stopped */   { true,   true,  false,  false,  false },
 499         /* init */      { true,   false, true,   false,  false },
 500         /* running */   { false,  false, true,   true,   true },
 501         /* disable */   { false,  false, true,   true,   true },
 502         /* stopping */  { true,   false, false,  false,  false }
 503 };
 504
 505 /**
 506  * Change coordinator thread state
 507  * Some combinations are not valid, so catch them here.
 508  *
 509  * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
 510  * the transition was not possible.
 511  */
 512 static int set_cdt_state_locked(struct coordinator *cdt,
 513                                 enum cdt_states new_state)
 514 {
 515         int rc;
 516         enum cdt_states state;
 517
 518         state = cdt->cdt_state;
 519
 520         if (cdt_transition[state][new_state]) {
 521                 cdt->cdt_state = new_state;
 522                 rc = 0;
 523         } else {
 524                 CDEBUG(D_HSM,
 525                        "unexpected coordinator transition, from=%s, to=%s\n",
 526                        cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
 527                 rc = -EINVAL;
 528         }
 529
 530         return rc;
 531 }
 532
 533 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state)
 534 {
 535         int rc;
 536
 537         mutex_lock(&cdt->cdt_state_lock);
 538         rc = set_cdt_state_locked(cdt, new_state);
 539         mutex_unlock(&cdt->cdt_state_lock);
 540
 541         return rc;
 542 }
 543
 544
 545
 546 /**
 547  * coordinator thread
 548  * \param data [IN] obd device
 549  * \retval 0 success
 550  * \retval -ve failure
 551  */
 552 static int mdt_coordinator(void *data)
 553 {
 554         struct mdt_thread_info  *mti = data;
 555         struct mdt_device       *mdt = mti->mti_mdt;
 556         struct coordinator      *cdt = &mdt->mdt_coordinator;
 557         struct hsm_scan_data     hsd = { NULL };
 558         time64_t                 last_housekeeping = 0;
 559         size_t request_sz = 0;
 560         int rc;
 561         ENTRY;
 562
 563         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 564                mdt_obd_name(mdt), current_pid());
 565
 566         hsd.hsd_mti = mti;
 567         obd_uuid2fsname(hsd.hsd_fsname, mdt_obd_name(mdt),
 568                         sizeof(hsd.hsd_fsname));
 569
 570         set_cdt_state(cdt, CDT_RUNNING);
 571
 572         /* Inform mdt_hsm_cdt_start(). */
 573         wake_up_all(&cdt->cdt_waitq);
 574
 575         while (1) {
 576                 int i;
 577                 int update_idx = 0;
 578                 int updates_sz;
 579                 int updates_cnt;
 580                 struct hsm_record_update *updates;
 581
 582                 /* Limit execution of the expensive requests traversal
 583                  * to at most one second. This prevents repeatedly
 584                  * locking/unlocking the catalog for each request
 585                  * and preventing other HSM operations from happening
 586                  */
 587                 wait_event_interruptible_timeout(cdt->cdt_waitq,
 588                                                  kthread_should_stop() ||
 589                                                  cdt->cdt_wakeup_coordinator,
 590                                                  cfs_time_seconds(1));
 591
 592                 cdt->cdt_wakeup_coordinator = false;
 593                 CDEBUG(D_HSM, "coordinator resumes\n");
 594
 595                 if (kthread_should_stop()) {
 596                         CDEBUG(D_HSM, "Coordinator stops\n");
 597                         rc = 0;
 598                         break;
 599                 }
 600
 601                 /* if coordinator is suspended continue to wait */
 602                 if (cdt->cdt_state == CDT_DISABLE) {
 603                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 604                         continue;
 605                 }
 606
 607                 /* If no event, and no housekeeping to do, continue to
 608                  * wait. */
 609                 if (last_housekeeping + cdt->cdt_loop_period <=
 610                     ktime_get_real_seconds()) {
 611                         last_housekeeping = ktime_get_real_seconds();
 612                         hsd.hsd_housekeeping = true;
 613                 } else if (cdt->cdt_event) {
 614                         hsd.hsd_housekeeping = false;
 615                 } else {
 616                         continue;
 617                 }
 618
 619                 cdt->cdt_event = false;
 620
 621                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 622
 623                 if (hsd.hsd_request_len != cdt->cdt_max_requests) {
 624                         /* cdt_max_requests has changed,
 625                          * we need to allocate a new buffer
 626                          */
 627                         struct hsm_scan_request *tmp = NULL;
 628                         int max_requests = cdt->cdt_max_requests;
 629                         OBD_ALLOC_LARGE(tmp, max_requests *
 630                                         sizeof(struct hsm_scan_request));
 631                         if (!tmp) {
 632                                 CERROR("Failed to resize request buffer, "
 633                                        "keeping it at %d\n",
 634                                        hsd.hsd_request_len);
 635                         } else {
 636                                 if (hsd.hsd_request != NULL)
 637                                         OBD_FREE_LARGE(hsd.hsd_request,
 638                                                        request_sz);
 639
 640                                 hsd.hsd_request_len = max_requests;
 641                                 request_sz = hsd.hsd_request_len *
 642                                         sizeof(struct hsm_scan_request);
 643                                 hsd.hsd_request = tmp;
 644                         }
 645                 }
 646
 647                 hsd.hsd_action_count = 0;
 648                 hsd.hsd_request_count = 0;
 649
 650                 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
 651                                       &hsd, 0, 0, WRITE);
 652                 if (rc < 0)
 653                         goto clean_cb_alloc;
 654
 655                 CDEBUG(D_HSM, "found %d requests to send\n",
 656                        hsd.hsd_request_count);
 657
 658                 if (list_empty(&cdt->cdt_agents)) {
 659                         CDEBUG(D_HSM, "no agent available, "
 660                                       "coordinator sleeps\n");
 661                         goto clean_cb_alloc;
 662                 }
 663
 664                 /* Compute how many HAI we have in all the requests */
 665                 updates_cnt = 0;
 666                 for (i = 0; i < hsd.hsd_request_count; i++) {
 667                         const struct hsm_scan_request *request =
 668                                 &hsd.hsd_request[i];
 669
 670                         updates_cnt += request->hal->hal_count;
 671                 }
 672
 673                 /* Allocate a temporary array to store the cookies to
 674                  * update, and their status. */
 675                 updates_sz = updates_cnt * sizeof(*updates);
 676                 OBD_ALLOC_LARGE(updates, updates_sz);
 677                 if (updates == NULL) {
 678                         CERROR("%s: Cannot allocate memory (%d o) "
 679                                "for %d updates\n",
 680                                mdt_obd_name(mdt), updates_sz, updates_cnt);
 681                         continue;
 682                 }
 683
 684                 /* here hsd contains a list of requests to be started */
 685                 for (i = 0; i < hsd.hsd_request_count; i++) {
 686                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 687                         struct hsm_action_list  *hal = request->hal;
 688                         struct hsm_action_item  *hai;
 689                         int                      j;
 690
 691                         /* still room for work ? */
 692                         if (atomic_read(&cdt->cdt_request_count) >=
 693                             cdt->cdt_max_requests)
 694                                 break;
 695
 696                         rc = mdt_hsm_agent_send(mti, hal, 0);
 697                         /* if failure, we suppose it is temporary
 698                          * if the copy tool failed to do the request
 699                          * it has to use hsm_progress
 700                          */
 701
 702                         /* set up cookie vector to set records status
 703                          * after copy tools start or failed
 704                          */
 705                         hai = hai_first(hal);
 706                         for (j = 0; j < hal->hal_count; j++) {
 707                                 updates[update_idx].cookie = hai->hai_cookie;
 708                                 updates[update_idx].status =
 709                                         (rc ? ARS_WAITING : ARS_STARTED);
 710                                 hai = hai_next(hai);
 711                                 update_idx++;
 712                         }
 713                 }
 714
 715                 if (update_idx) {
 716                         rc = mdt_agent_record_update(mti->mti_env, mdt,
 717                                                      updates, update_idx);
 718                         if (rc)
 719                                 CERROR("%s: mdt_agent_record_update() failed, "
 720                                        "rc=%d, cannot update records "
 721                                        "for %d cookies\n",
 722                                        mdt_obd_name(mdt), rc, update_idx);
 723                 }
 724
 725                 OBD_FREE_LARGE(updates, updates_sz);
 726
 727 clean_cb_alloc:
 728                 /* free hal allocated by callback */
 729                 for (i = 0; i < hsd.hsd_request_count; i++) {
 730                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 731
 732                         OBD_FREE_LARGE(request->hal, request->hal_sz);
 733                 }
 734         }
 735
 736         if (hsd.hsd_request != NULL)
 737                 OBD_FREE_LARGE(hsd.hsd_request, request_sz);
 738
 739         mdt_hsm_cdt_cleanup(mdt);
 740
 741         if (rc != 0)
 742                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 743                        mdt_obd_name(mdt), current_pid(), rc);
 744         else
 745                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 746                               " no error\n",
 747                        mdt_obd_name(mdt), current_pid());
 748
 749         RETURN(rc);
 750 }
 751
 752 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
 753                            const struct lu_fid *fid,
 754                            const struct hsm_extent *he)
 755 {
 756         struct cdt_restore_handle *crh;
 757         struct mdt_object *obj;
 758         int rc;
 759         ENTRY;
 760
 761         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 762         if (crh == NULL)
 763                 RETURN(-ENOMEM);
 764
 765         crh->crh_fid = *fid;
 766         /* in V1 all file is restored
 767          * crh->extent.start = he->offset;
 768          * crh->extent.end = he->offset + he->length;
 769          */
 770         crh->crh_extent.start = 0;
 771         crh->crh_extent.end = he->length;
 772         /* get the layout lock */
 773         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 774         obj = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 775                                    MDS_INODELOCK_LAYOUT);
 776         if (IS_ERR(obj))
 777                 GOTO(out_crh, rc = PTR_ERR(obj));
 778
 779         /* We do not keep a reference on the object during the restore
 780          * which can be very long. */
 781         mdt_object_put(mti->mti_env, obj);
 782
 783         mutex_lock(&cdt->cdt_restore_lock);
 784         if (unlikely(cdt->cdt_state == CDT_STOPPED ||
 785                      cdt->cdt_state == CDT_STOPPING)) {
 786                 mutex_unlock(&cdt->cdt_restore_lock);
 787                 GOTO(out_lh, rc = -EAGAIN);
 788         }
 789
 790         list_add_tail(&crh->crh_list, &cdt->cdt_restore_handle_list);
 791         mutex_unlock(&cdt->cdt_restore_lock);
 792
 793         RETURN(0);
 794 out_lh:
 795         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 796 out_crh:
 797         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 798
 799         return rc;
 800 }
 801
 802 /**
 803  * lookup a restore handle by FID
 804  * caller needs to hold cdt_restore_lock
 805  * \param cdt [IN] coordinator
 806  * \param fid [IN] FID
 807  * \retval cdt_restore_handle found
 808  * \retval NULL not found
 809  */
 810 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
 811                                                    const struct lu_fid *fid)
 812 {
 813         struct cdt_restore_handle *crh;
 814         ENTRY;
 815
 816         list_for_each_entry(crh, &cdt->cdt_restore_handle_list, crh_list) {
 817                 if (lu_fid_eq(&crh->crh_fid, fid))
 818                         RETURN(crh);
 819         }
 820
 821         RETURN(NULL);
 822 }
 823
 824 void cdt_restore_handle_del(struct mdt_thread_info *mti,
 825                             struct coordinator *cdt, const struct lu_fid *fid)
 826 {
 827         struct cdt_restore_handle *crh;
 828
 829         /* give back layout lock */
 830         mutex_lock(&cdt->cdt_restore_lock);
 831         crh = cdt_restore_handle_find(cdt, fid);
 832         if (crh != NULL)
 833                 list_del(&crh->crh_list);
 834         mutex_unlock(&cdt->cdt_restore_lock);
 835
 836         if (crh == NULL)
 837                 return;
 838
 839         /* XXX We pass a NULL object since the restore handle does not
 840          * keep a reference on the object being restored. */
 841         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 842         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 843 }
 844
 845 /**
 846  * data passed to llog_cat_process() callback
 847  * to scan requests and take actions
 848  */
 849 struct hsm_restore_data {
 850         struct mdt_thread_info  *hrd_mti;
 851 };
 852
 853 /**
 854  *  llog_cat_process() callback, used to:
 855  *  - find restore request and allocate the restore handle
 856  * \param env [IN] environment
 857  * \param llh [IN] llog handle
 858  * \param hdr [IN] llog record
 859  * \param data [IN/OUT] cb data = struct hsm_restore_data
 860  * \retval 0 success
 861  * \retval -ve failure
 862  */
 863 static int hsm_restore_cb(const struct lu_env *env,
 864                           struct llog_handle *llh,
 865                           struct llog_rec_hdr *hdr, void *data)
 866 {
 867         struct llog_agent_req_rec       *larr;
 868         struct hsm_restore_data         *hrd;
 869         struct hsm_action_item          *hai;
 870         struct mdt_thread_info          *mti;
 871         struct coordinator              *cdt;
 872         int rc;
 873         ENTRY;
 874
 875         hrd = data;
 876         mti = hrd->hrd_mti;
 877         cdt = &mti->mti_mdt->mdt_coordinator;
 878
 879         larr = (struct llog_agent_req_rec *)hdr;
 880         hai = &larr->arr_hai;
 881         if (hai->hai_cookie > cdt->cdt_last_cookie)
 882                 /* update the cookie to avoid collision */
 883                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 884
 885         if (hai->hai_action != HSMA_RESTORE ||
 886             agent_req_in_final_state(larr->arr_status))
 887                 RETURN(0);
 888
 889         /* restore request not in a final state */
 890
 891         /* force replay of restore requests left in started state from previous
 892          * CDT context, to be canceled later if finally found to be incompatible
 893          * when being re-started */
 894         if (larr->arr_status == ARS_STARTED) {
 895                 larr->arr_status = ARS_WAITING;
 896                 larr->arr_req_change = ktime_get_real_seconds();
 897                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 898                 if (rc != 0)
 899                         GOTO(out, rc);
 900         }
 901
 902         rc = cdt_restore_handle_add(mti, cdt, &hai->hai_fid, &hai->hai_extent);
 903 out:
 904         RETURN(rc);
 905 }
 906
 907 /**
 908  * restore coordinator state at startup
 909  * the goal is to take a layout lock for each registered restore request
 910  * \param mti [IN] context
 911  */
 912 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 913 {
 914         struct hsm_restore_data  hrd;
 915         int                      rc;
 916         ENTRY;
 917
 918         hrd.hrd_mti = mti;
 919
 920         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
 921                               0, 0, WRITE);
 922
 923         RETURN(rc);
 924 }
 925
 926 static int hsm_init_ucred(struct lu_ucred *uc)
 927 {
 928         ENTRY;
 929
 930         uc->uc_valid = UCRED_OLD;
 931         uc->uc_o_uid = 0;
 932         uc->uc_o_gid = 0;
 933         uc->uc_o_fsuid = 0;
 934         uc->uc_o_fsgid = 0;
 935         uc->uc_uid = 0;
 936         uc->uc_gid = 0;
 937         uc->uc_fsuid = 0;
 938         uc->uc_fsgid = 0;
 939         uc->uc_suppgids[0] = -1;
 940         uc->uc_suppgids[1] = -1;
 941         uc->uc_cap = CFS_CAP_FS_MASK;
 942         uc->uc_umask = 0777;
 943         uc->uc_ginfo = NULL;
 944         uc->uc_identity = NULL;
 945         /* always record internal HSM activity if also enabled globally */
 946         uc->uc_enable_audit = 1;
 947
 948         RETURN(0);
 949 }
 950
 951 /**
 952  * initialize coordinator struct
 953  * \param mdt [IN] device
 954  * \retval 0 success
 955  * \retval -ve failure
 956  */
 957 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 958 {
 959         struct coordinator      *cdt = &mdt->mdt_coordinator;
 960         struct mdt_thread_info  *cdt_mti = NULL;
 961         int                      rc;
 962         ENTRY;
 963
 964         init_waitqueue_head(&cdt->cdt_waitq);
 965         init_rwsem(&cdt->cdt_llog_lock);
 966         init_rwsem(&cdt->cdt_agent_lock);
 967         init_rwsem(&cdt->cdt_request_lock);
 968         mutex_init(&cdt->cdt_restore_lock);
 969         mutex_init(&cdt->cdt_state_lock);
 970         set_cdt_state(cdt, CDT_STOPPED);
 971
 972         INIT_LIST_HEAD(&cdt->cdt_request_list);
 973         INIT_LIST_HEAD(&cdt->cdt_agents);
 974         INIT_LIST_HEAD(&cdt->cdt_restore_handle_list);
 975
 976         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
 977                                                        CFS_HASH_BITS_MIN,
 978                                                        CFS_HASH_BITS_MAX,
 979                                                        CFS_HASH_BKT_BITS,
 980                                                        0 /* extra bytes */,
 981                                                        CFS_HASH_MIN_THETA,
 982                                                        CFS_HASH_MAX_THETA,
 983                                                 &cdt_request_cookie_hash_ops,
 984                                                        CFS_HASH_DEFAULT);
 985         if (cdt->cdt_request_cookie_hash == NULL)
 986                 RETURN(-ENOMEM);
 987
 988         cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
 989                                                      CFS_HASH_BITS_MIN,
 990                                                      CFS_HASH_BITS_MAX,
 991                                                      CFS_HASH_BKT_BITS,
 992                                                      0 /* extra bytes */,
 993                                                      CFS_HASH_MIN_THETA,
 994                                                      CFS_HASH_MAX_THETA,
 995                                                      &cdt_agent_record_hash_ops,
 996                                                      CFS_HASH_DEFAULT);
 997         if (cdt->cdt_agent_record_hash == NULL)
 998                 GOTO(out_request_cookie_hash, rc = -ENOMEM);
 999
1000         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
1001         if (rc < 0)
1002                 GOTO(out_agent_record_hash, rc);
1003
1004         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1005         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
1006         if (rc < 0)
1007                 GOTO(out_env, rc);
1008
1009         lu_context_enter(&cdt->cdt_session);
1010         cdt->cdt_env.le_ses = &cdt->cdt_session;
1011
1012         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1013         LASSERT(cdt_mti != NULL);
1014
1015         cdt_mti->mti_env = &cdt->cdt_env;
1016         cdt_mti->mti_mdt = mdt;
1017
1018         hsm_init_ucred(mdt_ucred(cdt_mti));
1019
1020         /* default values for /proc tunnables
1021          * can be override by MGS conf */
1022         cdt->cdt_default_archive_id = 1;
1023         cdt->cdt_grace_delay = 60;
1024         cdt->cdt_loop_period = 10;
1025         cdt->cdt_max_requests = 3;
1026         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1027         cdt->cdt_active_req_timeout = 3600;
1028
1029         /* by default do not remove archives on last unlink */
1030         cdt->cdt_remove_archive_on_last_unlink = false;
1031
1032         RETURN(0);
1033
1034 out_env:
1035         lu_env_fini(&cdt->cdt_env);
1036 out_agent_record_hash:
1037         cfs_hash_putref(cdt->cdt_agent_record_hash);
1038         cdt->cdt_agent_record_hash = NULL;
1039 out_request_cookie_hash:
1040         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1041         cdt->cdt_request_cookie_hash = NULL;
1042
1043         return rc;
1044 }
1045
1046 /**
1047  * free a coordinator thread
1048  * \param mdt [IN] device
1049  */
1050 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
1051 {
1052         struct coordinator *cdt = &mdt->mdt_coordinator;
1053         ENTRY;
1054
1055         lu_context_exit(cdt->cdt_env.le_ses);
1056         lu_context_fini(cdt->cdt_env.le_ses);
1057
1058         lu_env_fini(&cdt->cdt_env);
1059
1060         cfs_hash_putref(cdt->cdt_agent_record_hash);
1061         cdt->cdt_agent_record_hash = NULL;
1062
1063         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1064         cdt->cdt_request_cookie_hash = NULL;
1065
1066         RETURN(0);
1067 }
1068
1069 /**
1070  * start a coordinator thread
1071  * \param mdt [IN] device
1072  * \retval 0 success
1073  * \retval -ve failure
1074  */
1075 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1076 {
1077         struct coordinator      *cdt = &mdt->mdt_coordinator;
1078         struct mdt_thread_info *cdt_mti;
1079         int                      rc;
1080         void                    *ptr;
1081         struct task_struct      *task;
1082         ENTRY;
1083
1084         /* functions defined but not yet used
1085          * this avoid compilation warning
1086          */
1087         ptr = dump_requests;
1088
1089         rc = set_cdt_state(cdt, CDT_INIT);
1090         if (rc) {
1091                 CERROR("%s: Coordinator already started or stopping\n",
1092                        mdt_obd_name(mdt));
1093                 RETURN(-EALREADY);
1094         }
1095
1096         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
1097         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1098
1099         /* just need to be larger than previous one */
1100         /* cdt_last_cookie is protected by cdt_llog_lock */
1101         cdt->cdt_last_cookie = ktime_get_real_seconds();
1102         atomic_set(&cdt->cdt_request_count, 0);
1103         atomic_set(&cdt->cdt_archive_count, 0);
1104         atomic_set(&cdt->cdt_restore_count, 0);
1105         atomic_set(&cdt->cdt_remove_count, 0);
1106         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
1107         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
1108         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
1109
1110         /* to avoid deadlock when start is made through /proc
1111          * /proc entries are created by the coordinator thread */
1112
1113         /* set up list of started restore requests */
1114         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1115         rc = mdt_hsm_pending_restore(cdt_mti);
1116         if (rc)
1117                 CERROR("%s: cannot take the layout locks needed"
1118                        " for registered restore: %d\n",
1119                        mdt_obd_name(mdt), rc);
1120
1121         if (mdt->mdt_bottom->dd_rdonly)
1122                 RETURN(0);
1123
1124         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
1125         if (IS_ERR(task)) {
1126                 rc = PTR_ERR(task);
1127                 set_cdt_state(cdt, CDT_STOPPED);
1128                 CERROR("%s: error starting coordinator thread: %d\n",
1129                        mdt_obd_name(mdt), rc);
1130         } else {
1131                 cdt->cdt_task = task;
1132                 wait_event(cdt->cdt_waitq,
1133                            cdt->cdt_state != CDT_INIT);
1134                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1135                        mdt_obd_name(mdt));
1136                 rc = 0;
1137         }
1138
1139         RETURN(rc);
1140 }
1141
1142 /**
1143  * stop a coordinator thread
1144  * \param mdt [IN] device
1145  */
1146 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1147 {
1148         struct coordinator *cdt = &mdt->mdt_coordinator;
1149         int rc;
1150
1151         ENTRY;
1152         /* stop coordinator thread */
1153         rc = set_cdt_state(cdt, CDT_STOPPING);
1154         if (rc == 0) {
1155                 kthread_stop(cdt->cdt_task);
1156                 cdt->cdt_task = NULL;
1157                 set_cdt_state(cdt, CDT_STOPPED);
1158         }
1159
1160         RETURN(rc);
1161 }
1162
1163 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1164                               const struct lu_fid *fid,
1165                               u32 archive_id)
1166 {
1167         struct mdt_object *obj;
1168         struct md_hsm mh;
1169         int rc;
1170
1171         obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1172         if (IS_ERR(obj))
1173                 GOTO(out, rc = PTR_ERR(obj));
1174
1175         if (mh.mh_flags & HS_EXISTS &&
1176             mh.mh_arch_id == archive_id)
1177                 GOTO(out_obj, rc = 0);
1178
1179         mh.mh_flags |= HS_EXISTS;
1180         mh.mh_arch_id = archive_id;
1181         rc = mdt_hsm_attr_set(mti, obj, &mh);
1182
1183 out_obj:
1184         mdt_object_put(mti->mti_env, obj);
1185 out:
1186         return rc;
1187 }
1188
1189 /**
1190  * register all requests from an hal in the memory list
1191  * \param mti [IN] context
1192  * \param hal [IN] request
1193  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1194  *  which is running the CT
1195  * \retval 0 success
1196  * \retval -ve failure
1197  */
1198 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1199                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1200 {
1201         struct mdt_device       *mdt = mti->mti_mdt;
1202         struct coordinator      *cdt = &mdt->mdt_coordinator;
1203         struct hsm_action_item  *hai;
1204         int                      rc = 0, i;
1205         ENTRY;
1206
1207         /* register request in memory list */
1208         hai = hai_first(hal);
1209         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1210                 struct cdt_agent_req *car;
1211
1212                 /* in case of a cancel request, we first mark the ondisk
1213                  * record of the request we want to stop as canceled
1214                  * this does not change the cancel record
1215                  * it will be done when updating the request status
1216                  */
1217                 if (hai->hai_action == HSMA_CANCEL) {
1218                         struct hsm_record_update update = {
1219                                 .cookie = hai->hai_cookie,
1220                                 .status = ARS_CANCELED,
1221                         };
1222
1223                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1224                                                      &update, 1);
1225                         if (rc) {
1226                                 CERROR("%s: mdt_agent_record_update() failed, "
1227                                        "rc=%d, cannot update status to %s "
1228                                        "for cookie %#llx\n",
1229                                        mdt_obd_name(mdt), rc,
1230                                        agent_req_status2name(ARS_CANCELED),
1231                                        hai->hai_cookie);
1232                                 GOTO(out, rc);
1233                         }
1234
1235                         /* find the running request to set it canceled */
1236                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1237                         if (car != NULL) {
1238                                 car->car_canceled = 1;
1239                                 /* uuid has to be changed to the one running the
1240                                 * request to cancel */
1241                                 *uuid = car->car_uuid;
1242                                 mdt_cdt_put_request(car);
1243                         }
1244                         /* no need to memorize cancel request
1245                          * this also avoid a deadlock when we receive
1246                          * a purge all requests command
1247                          */
1248                         continue;
1249                 }
1250
1251                 if (hai->hai_action == HSMA_ARCHIVE) {
1252                         rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1253                                                 hal->hal_archive_id);
1254                         if (rc == -ENOENT)
1255                                 continue;
1256                         else if (rc < 0)
1257                                 GOTO(out, rc);
1258                 }
1259
1260                 car = mdt_cdt_alloc_request(hal->hal_archive_id, hal->hal_flags,
1261                                             uuid, hai);
1262                 if (IS_ERR(car))
1263                         GOTO(out, rc = PTR_ERR(car));
1264
1265                 rc = mdt_cdt_add_request(cdt, car);
1266                 if (rc != 0)
1267                         mdt_cdt_free_request(car);
1268         }
1269 out:
1270         RETURN(rc);
1271 }
1272
1273 /**
1274  * swap layouts between 2 fids
1275  * \param mti [IN] context
1276  * \param obj [IN]
1277  * \param dfid [IN]
1278  * \param mh_common [IN] MD HSM
1279  */
1280 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1281                             struct mdt_object *obj, const struct lu_fid *dfid,
1282                             struct md_hsm *mh_common)
1283 {
1284         struct mdt_object       *dobj;
1285         struct mdt_lock_handle  *dlh;
1286         int                      rc;
1287         ENTRY;
1288
1289         if (!mdt_object_exists(obj))
1290                 GOTO(out, rc = -ENOENT);
1291
1292         /* we already have layout lock on obj so take only
1293          * on dfid */
1294         dlh = &mti->mti_lh[MDT_LH_OLD];
1295         mdt_lock_reg_init(dlh, LCK_EX);
1296         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1297         if (IS_ERR(dobj))
1298                 GOTO(out, rc = PTR_ERR(dobj));
1299
1300         /* if copy tool closes the volatile before sending the final
1301          * progress through llapi_hsm_copy_end(), all the objects
1302          * are removed and mdd_swap_layout LBUG */
1303         if (!mdt_object_exists(dobj)) {
1304                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1305                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1306                 GOTO(out_dobj, rc = -ENOENT);
1307         }
1308         /* Since we only handle restores here, unconditionally use
1309          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1310          * be preserved in case of failure during swap_layout and not
1311          * leave a file in an intermediate but incoherent state.
1312          * But need to setup HSM xattr of data FID before, reuse
1313          * mti and mh presets for FID in hsm_cdt_request_completed(),
1314          * only need to clear RELEASED and DIRTY.
1315          */
1316         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1317         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1318         if (rc == 0)
1319                 rc = mo_swap_layouts(mti->mti_env,
1320                                      mdt_object_child(obj),
1321                                      mdt_object_child(dobj),
1322                                      SWAP_LAYOUTS_MDS_HSM);
1323         if (rc == 0) {
1324                 rc = mdt_lsom_downgrade(mti, obj);
1325                 if (rc)
1326                         CDEBUG(D_INODE,
1327                                "%s: File fid="DFID" SOM "
1328                                "downgrade failed, rc = %d\n",
1329                                mdt_obd_name(mti->mti_mdt),
1330                                PFID(mdt_object_fid(obj)), rc);
1331         }
1332 out_dobj:
1333         mdt_object_unlock_put(mti, dobj, dlh, 1);
1334 out:
1335         RETURN(rc);
1336 }
1337
1338 /**
1339  * update status of a completed request
1340  * \param mti [IN] context
1341  * \param pgs [IN] progress of the copy tool
1342  * \retval 0 success
1343  * \retval -ve failure
1344  */
1345 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1346                                      struct hsm_progress_kernel *pgs,
1347                                      const struct cdt_agent_req *car,
1348                                      enum agent_req_status *status)
1349 {
1350         const struct lu_env     *env = mti->mti_env;
1351         struct mdt_device       *mdt = mti->mti_mdt;
1352         struct coordinator      *cdt = &mdt->mdt_coordinator;
1353         struct mdt_object       *obj = NULL;
1354         int                      cl_flags = 0, rc = 0;
1355         struct md_hsm            mh;
1356         bool                     is_mh_changed;
1357         bool                     need_changelog = true;
1358         ENTRY;
1359
1360         /* default is to retry */
1361         *status = ARS_WAITING;
1362
1363         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1364          * if error/removed continue anyway to get correct reporting done */
1365         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1366         /* we will update MD HSM only if needed */
1367         is_mh_changed = false;
1368
1369         /* no need to change mh->mh_arch_id
1370          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1371          */
1372         if (pgs->hpk_errval != 0) {
1373                 switch (pgs->hpk_errval) {
1374                 case ENOSYS:
1375                         /* the copy tool does not support cancel
1376                          * so the cancel request is failed
1377                          * As we cannot distinguish a cancel progress
1378                          * from another action progress (they have the
1379                          * same cookie), we suppose here the CT returns
1380                          * ENOSYS only if does not support cancel
1381                          */
1382                         /* this can also happen when cdt calls it to
1383                          * for a timed out request */
1384                         *status = ARS_FAILED;
1385                         /* to have a cancel event in changelog */
1386                         pgs->hpk_errval = ECANCELED;
1387                         break;
1388                 case ECANCELED:
1389                         /* the request record has already been set to
1390                          * ARS_CANCELED, this set the cancel request
1391                          * to ARS_SUCCEED */
1392                         *status = ARS_SUCCEED;
1393                         break;
1394                 default:
1395                         /* retry only if current policy or requested, and
1396                          * object is not on error/removed */
1397                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1398                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1399                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1400                         break;
1401                 }
1402
1403                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1404                         CERROR("%s: Request %#llx on "DFID
1405                                " failed, error code %d too large\n",
1406                                mdt_obd_name(mdt),
1407                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1408                                pgs->hpk_errval);
1409                         hsm_set_cl_error(&cl_flags,
1410                                          CLF_HSM_ERROVERFLOW);
1411                         rc = -EINVAL;
1412                 } else {
1413                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1414                 }
1415
1416                 switch (car->car_hai->hai_action) {
1417                 case HSMA_ARCHIVE:
1418                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1419                         break;
1420                 case HSMA_RESTORE:
1421                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1422                         break;
1423                 case HSMA_REMOVE:
1424                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1425                         break;
1426                 case HSMA_CANCEL:
1427                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1428                         CERROR("%s: Failed request %#llx on "DFID
1429                                " cannot be a CANCEL\n",
1430                                mdt_obd_name(mdt),
1431                                pgs->hpk_cookie,
1432                                PFID(&pgs->hpk_fid));
1433                         break;
1434                 default:
1435                         CERROR("%s: Failed request %#llx on "DFID
1436                                " %d is an unknown action\n",
1437                                mdt_obd_name(mdt),
1438                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1439                                car->car_hai->hai_action);
1440                         rc = -EINVAL;
1441                         break;
1442                 }
1443         } else {
1444                 *status = ARS_SUCCEED;
1445                 switch (car->car_hai->hai_action) {
1446                 case HSMA_ARCHIVE:
1447                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1448                         /* set ARCHIVE keep EXIST and clear LOST and
1449                          * DIRTY */
1450                         mh.mh_arch_ver = pgs->hpk_data_version;
1451                         mh.mh_flags |= HS_ARCHIVED;
1452                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1453                         is_mh_changed = true;
1454                         break;
1455                 case HSMA_RESTORE:
1456                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1457
1458                         /* do not clear RELEASED and DIRTY here
1459                          * this will occur in hsm_swap_layouts()
1460                          */
1461
1462                         /* Restoring has changed the file version on
1463                          * disk. */
1464                         mh.mh_arch_ver = pgs->hpk_data_version;
1465                         is_mh_changed = true;
1466                         break;
1467                 case HSMA_REMOVE:
1468                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1469                         /* clear ARCHIVED EXISTS and LOST */
1470                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1471                         is_mh_changed = true;
1472                         break;
1473                 case HSMA_CANCEL:
1474                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1475                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1476                                mdt_obd_name(mdt),
1477                                pgs->hpk_cookie,
1478                                PFID(&pgs->hpk_fid));
1479                         break;
1480                 default:
1481                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1482                                mdt_obd_name(mdt),
1483                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1484                                car->car_hai->hai_action);
1485                         rc = -EINVAL;
1486                         break;
1487                 }
1488         }
1489
1490         /* rc != 0 means error when analysing action, it may come from
1491          * a crasy CT no need to manage DIRTY
1492          * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1493          * filled
1494          */
1495         if (rc == 0 && !IS_ERR(obj))
1496                 hsm_set_cl_flags(&cl_flags,
1497                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1498
1499         /* unlock is done later, after layout lock management */
1500         if (is_mh_changed && !IS_ERR(obj))
1501                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1502
1503         /* we give back layout lock only if restore was successful or
1504          * if no retry will be attempted and if object is still alive,
1505          * in other cases we just unlock the object */
1506         if (car->car_hai->hai_action == HSMA_RESTORE) {
1507                 /* restore in data FID done, we swap the layouts
1508                  * only if restore is successful */
1509                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1510                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1511                                               &mh);
1512                         if (rc) {
1513                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1514                                         *status = ARS_FAILED;
1515                                 pgs->hpk_errval = -rc;
1516                         }
1517                 }
1518                 /* we have to retry, so keep layout lock */
1519                 if (*status == ARS_WAITING)
1520                         GOTO(out, rc);
1521
1522                 /* restore special case, need to create ChangeLog record
1523                  * before to give back layout lock to avoid concurrent
1524                  * file updater to post out of order ChangeLog */
1525                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1526                              &car->car_hai->hai_fid);
1527                 need_changelog = false;
1528
1529                 cdt_restore_handle_del(mti, cdt, &car->car_hai->hai_fid);
1530         }
1531
1532         GOTO(out, rc);
1533
1534 out:
1535         /* always add a ChangeLog record */
1536         if (need_changelog)
1537                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1538                              &car->car_hai->hai_fid);
1539
1540         if (!IS_ERR(obj))
1541                 mdt_object_put(mti->mti_env, obj);
1542
1543         RETURN(rc);
1544 }
1545
1546 /**
1547  * update status of a request
1548  * \param mti [IN] context
1549  * \param pgs [IN] progress of the copy tool
1550  * \retval 0 success
1551  * \retval -ve failure
1552  */
1553 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1554                                  struct hsm_progress_kernel *pgs)
1555 {
1556         struct mdt_device       *mdt = mti->mti_mdt;
1557         struct coordinator      *cdt = &mdt->mdt_coordinator;
1558         struct cdt_agent_req    *car;
1559         int                      rc = 0;
1560         ENTRY;
1561
1562         /* no coordinator started, so we cannot serve requests */
1563         if (cdt->cdt_state == CDT_STOPPED)
1564                 RETURN(-EAGAIN);
1565
1566         /* first do sanity checks */
1567         car = mdt_cdt_update_request(cdt, pgs);
1568         if (IS_ERR(car)) {
1569                 CERROR("%s: Cannot find running request for cookie %#llx"
1570                        " on fid="DFID"\n",
1571                        mdt_obd_name(mdt),
1572                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1573
1574                 RETURN(PTR_ERR(car));
1575         }
1576
1577         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1578                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1579                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1580                       hsm_copytool_action2name(car->car_hai->hai_action),
1581                       pgs->hpk_flags, pgs->hpk_errval,
1582                       PFID(&car->car_hai->hai_fid),
1583                       PFID(&car->car_hai->hai_dfid));
1584
1585         /* progress is done on FID or data FID depending of the action and
1586          * of the copy progress */
1587         /* for restore progress is used to send back the data FID to cdt */
1588         if (car->car_hai->hai_action == HSMA_RESTORE &&
1589             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1590                 car->car_hai->hai_dfid = pgs->hpk_fid;
1591
1592         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1593              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1594             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1595              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1596                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1597                        " does not match request FID "DFID" nor data FID "
1598                        DFID"\n",
1599                        mdt_obd_name(mdt),
1600                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1601                        PFID(&car->car_hai->hai_fid),
1602                        PFID(&car->car_hai->hai_dfid));
1603                 GOTO(out, rc = -EINVAL);
1604         }
1605
1606         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1607                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1608                        " is not coherent (err=%d and not completed"
1609                        " (flags=%d))\n",
1610                        mdt_obd_name(mdt),
1611                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1612                        hsm_copytool_action2name(car->car_hai->hai_action),
1613                        pgs->hpk_errval, pgs->hpk_flags);
1614                 GOTO(out, rc = -EINVAL);
1615         }
1616
1617         /* now progress is valid */
1618
1619         /* we use a root like ucred */
1620         hsm_init_ucred(mdt_ucred(mti));
1621
1622         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1623                 enum agent_req_status status;
1624                 struct hsm_record_update update;
1625                 int rc1;
1626
1627                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1628
1629                 CDEBUG(D_HSM, "updating record: fid="DFID" cookie=%#llx action=%s "
1630                               "status=%s\n",
1631                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1632                        hsm_copytool_action2name(car->car_hai->hai_action),
1633                        agent_req_status2name(status));
1634
1635                 /* update record first (LU-9075) */
1636                 update.cookie = pgs->hpk_cookie;
1637                 update.status = status;
1638
1639                 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1640                                               &update, 1);
1641                 if (rc1)
1642                         CERROR("%s: mdt_agent_record_update() failed,"
1643                                " rc=%d, cannot update status to %s"
1644                                " for cookie %#llx\n",
1645                                mdt_obd_name(mdt), rc1,
1646                                agent_req_status2name(status),
1647                                pgs->hpk_cookie);
1648                 rc = (rc != 0 ? rc : rc1);
1649
1650                 /* then remove request from memory list (LU-9075) */
1651                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1652
1653                 /* ct has completed a request, so a slot is available,
1654                  * signal the coordinator to find new work */
1655                 mdt_hsm_cdt_event(cdt);
1656         } else {
1657                 /* if copytool send a progress on a canceled request
1658                  * we inform copytool it should stop
1659                  */
1660                 if (car->car_canceled == 1)
1661                         rc = -ECANCELED;
1662         }
1663         GOTO(out, rc);
1664
1665 out:
1666         /* remove ref got from mdt_cdt_update_request() */
1667         mdt_cdt_put_request(car);
1668
1669         return rc;
1670 }
1671
1672
1673 /**
1674  * data passed to llog_cat_process() callback
1675  * to cancel requests
1676  */
1677 struct hsm_cancel_all_data {
1678         struct mdt_device       *mdt;
1679 };
1680
1681 /**
1682  *  llog_cat_process() callback, used to:
1683  *  - purge all requests
1684  * \param env [IN] environment
1685  * \param llh [IN] llog handle
1686  * \param hdr [IN] llog record
1687  * \param data [IN] cb data = struct hsm_cancel_all_data
1688  * \retval 0 success
1689  * \retval -ve failure
1690  */
1691 static int mdt_cancel_all_cb(const struct lu_env *env,
1692                              struct llog_handle *llh,
1693                              struct llog_rec_hdr *hdr, void *data)
1694 {
1695         struct llog_agent_req_rec       *larr;
1696         struct hsm_cancel_all_data      *hcad;
1697         int                              rc = 0;
1698         ENTRY;
1699
1700         larr = (struct llog_agent_req_rec *)hdr;
1701         hcad = data;
1702         if (larr->arr_status == ARS_WAITING ||
1703             larr->arr_status == ARS_STARTED) {
1704                 larr->arr_status = ARS_CANCELED;
1705                 larr->arr_req_change = ktime_get_real_seconds();
1706                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1707         }
1708
1709         RETURN(rc);
1710 }
1711
1712 /**
1713  * cancel all actions
1714  * \param obd [IN] MDT device
1715  */
1716 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1717 {
1718         struct lu_env                    env;
1719         struct lu_context                session;
1720         struct mdt_thread_info          *mti;
1721         struct coordinator              *cdt = &mdt->mdt_coordinator;
1722         struct cdt_agent_req            *car;
1723         struct hsm_action_list          *hal = NULL;
1724         struct hsm_action_item          *hai;
1725         struct hsm_cancel_all_data       hcad;
1726         int                              hal_sz = 0, hal_len, rc;
1727         enum cdt_states                  old_state;
1728         ENTRY;
1729
1730         rc = lu_env_init(&env, LCT_MD_THREAD);
1731         if (rc < 0)
1732                 RETURN(rc);
1733
1734         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1735         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1736         if (rc < 0)
1737                 GOTO(out_env, rc);
1738
1739         lu_context_enter(&session);
1740         env.le_ses = &session;
1741
1742         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1743         LASSERT(mti != NULL);
1744
1745         mti->mti_env = &env;
1746         mti->mti_mdt = mdt;
1747
1748         hsm_init_ucred(mdt_ucred(mti));
1749
1750         mutex_lock(&cdt->cdt_state_lock);
1751         old_state = cdt->cdt_state;
1752
1753         /* disable coordinator */
1754         rc = set_cdt_state_locked(cdt, CDT_DISABLE);
1755         if (rc)
1756                 GOTO(out_cdt_state_unlock, rc);
1757
1758         /* send cancel to all running requests */
1759         down_read(&cdt->cdt_request_lock);
1760         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1761                 mdt_cdt_get_request(car);
1762                 /* request is not yet removed from list, it will be done
1763                  * when copytool will return progress
1764                  */
1765
1766                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1767                         mdt_cdt_put_request(car);
1768                         continue;
1769                 }
1770
1771                 /* needed size */
1772                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1773                           cfs_size_round(car->car_hai->hai_len);
1774
1775                 if (hal_len > hal_sz && hal_sz > 0) {
1776                         /* not enough room, free old buffer */
1777                         OBD_FREE(hal, hal_sz);
1778                         hal = NULL;
1779                 }
1780
1781                 /* empty buffer, allocate one */
1782                 if (hal == NULL) {
1783                         hal_sz = hal_len;
1784                         OBD_ALLOC(hal, hal_sz);
1785                         if (hal == NULL) {
1786                                 mdt_cdt_put_request(car);
1787                                 up_read(&cdt->cdt_request_lock);
1788                                 GOTO(out_cdt_state, rc = -ENOMEM);
1789                         }
1790                 }
1791
1792                 hal->hal_version = HAL_VERSION;
1793                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1794                                 MTI_NAME_MAXLEN);
1795                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1796                 hal->hal_archive_id = car->car_archive_id;
1797                 hal->hal_flags = car->car_flags;
1798                 hal->hal_count = 0;
1799
1800                 hai = hai_first(hal);
1801                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1802                 hai->hai_action = HSMA_CANCEL;
1803                 hal->hal_count = 1;
1804
1805                 /* it is possible to safely call mdt_hsm_agent_send()
1806                  * (ie without a deadlock on cdt_request_lock), because the
1807                  * write lock is taken only if we are not in purge mode
1808                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1809                  *   nor mdt_cdt_remove_request())
1810                  */
1811                 /* no conflict with cdt thread because cdt is disable and we
1812                  * have the request lock */
1813                 mdt_hsm_agent_send(mti, hal, 1);
1814
1815                 mdt_cdt_put_request(car);
1816         }
1817         up_read(&cdt->cdt_request_lock);
1818
1819         if (hal != NULL)
1820                 OBD_FREE(hal, hal_sz);
1821
1822         /* cancel all on-disk records */
1823         hcad.mdt = mdt;
1824
1825         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1826                               &hcad, 0, 0, WRITE);
1827 out_cdt_state:
1828         /* Enable coordinator, unless the coordinator was stopping. */
1829         set_cdt_state_locked(cdt, old_state);
1830 out_cdt_state_unlock:
1831         mutex_unlock(&cdt->cdt_state_lock);
1832
1833         lu_context_exit(&session);
1834         lu_context_fini(&session);
1835 out_env:
1836         lu_env_fini(&env);
1837
1838         RETURN(rc);
1839 }
1840
1841 /**
1842  * check if a request is compatible with file status
1843  * \param hai [IN] request description
1844  * \param archive_id [IN] request archive id
1845  * \param rq_flags [IN] request flags
1846  * \param hsm [IN] file HSM metadata
1847  * \retval boolean
1848  */
1849 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1850                               u32 archive_id, u64 rq_flags,
1851                               const struct md_hsm *hsm)
1852 {
1853         int      is_compat = false;
1854         int      hsm_flags;
1855         ENTRY;
1856
1857         hsm_flags = hsm->mh_flags;
1858         switch (hai->hai_action) {
1859         case HSMA_ARCHIVE:
1860                 if (!(hsm_flags & HS_NOARCHIVE) &&
1861                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1862                         is_compat = true;
1863
1864                 if (hsm_flags & HS_EXISTS &&
1865                     archive_id != 0 &&
1866                     archive_id != hsm->mh_arch_id)
1867                         is_compat = false;
1868
1869                 break;
1870         case HSMA_RESTORE:
1871                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1872                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1873                         is_compat = true;
1874                 break;
1875         case HSMA_REMOVE:
1876                 if (!(hsm_flags & HS_RELEASED) &&
1877                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1878                         is_compat = true;
1879                 break;
1880         case HSMA_CANCEL:
1881                 is_compat = true;
1882                 break;
1883         }
1884         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1885                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1886                       PFID(&hai->hai_fid),
1887                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1888                       hai->hai_extent.offset, hai->hai_extent.length,
1889                       hsm->mh_flags,
1890                       (is_compat ? "compatible" : "uncompatible"));
1891
1892         RETURN(is_compat);
1893 }
1894
1895 /*
1896  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1897  */
1898 static const struct {
1899         __u64            bit;
1900         char            *name;
1901         char            *nickname;
1902 } hsm_policy_names[] = {
1903         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1904         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1905         { 0 },
1906 };
1907
1908 /**
1909  * convert a policy name to a bit
1910  * \param name [IN] policy name
1911  * \retval 0 unknown
1912  * \retval   policy bit
1913  */
1914 static __u64 hsm_policy_str2bit(const char *name)
1915 {
1916         int      i;
1917
1918         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1919                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1920                     strcmp(hsm_policy_names[i].name, name) == 0)
1921                         return hsm_policy_names[i].bit;
1922         return 0;
1923 }
1924
1925 /**
1926  * convert a policy bit field to a string
1927  * \param mask [IN] policy bit field
1928  * \param hexa [IN] print mask before bit names
1929  * \param buffer [OUT] string
1930  * \param count [IN] size of buffer
1931  */
1932 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1933                                 const bool hexa)
1934 {
1935         int      i, j;
1936         __u64    bit;
1937         ENTRY;
1938
1939         if (hexa)
1940                 seq_printf(m, "(%#llx) ", mask);
1941
1942         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1943                 bit = (1ULL << i);
1944
1945                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1946                         if (hsm_policy_names[j].bit == bit)
1947                                 break;
1948                 }
1949                 if (bit & mask)
1950                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1951                 else
1952                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1953         }
1954         /* remove last ' ' */
1955         m->count--;
1956         seq_putc(m, '\n');
1957 }
1958
1959 /* methods to read/write HSM policy flags */
1960 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1961 {
1962         struct mdt_device       *mdt = m->private;
1963         struct coordinator      *cdt = &mdt->mdt_coordinator;
1964         ENTRY;
1965
1966         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1967         RETURN(0);
1968 }
1969
1970 static ssize_t
1971 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1972                          size_t count, loff_t *off)
1973 {
1974         struct seq_file         *m = file->private_data;
1975         struct mdt_device       *mdt = m->private;
1976         struct coordinator      *cdt = &mdt->mdt_coordinator;
1977         char                    *start, *token, sign;
1978         char                    *buf;
1979         __u64                    policy;
1980         __u64                    add_mask, remove_mask, set_mask;
1981         int                      rc;
1982         ENTRY;
1983
1984         if (count + 1 > PAGE_SIZE)
1985                 RETURN(-EINVAL);
1986
1987         OBD_ALLOC(buf, count + 1);
1988         if (buf == NULL)
1989                 RETURN(-ENOMEM);
1990
1991         if (copy_from_user(buf, buffer, count))
1992                 GOTO(out, rc = -EFAULT);
1993
1994         buf[count] = '\0';
1995
1996         start = buf;
1997         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1998                start);
1999
2000         add_mask = remove_mask = set_mask = 0;
2001         do {
2002                 token = strsep(&start, "\n ");
2003                 sign = *token;
2004
2005                 if (sign == '\0')
2006                         continue;
2007
2008                 if (sign == '-' || sign == '+')
2009                         token++;
2010
2011                 policy = hsm_policy_str2bit(token);
2012                 if (policy == 0) {
2013                         CWARN("%s: '%s' is unknown, "
2014                               "supported policies are:\n", mdt_obd_name(mdt),
2015                               token);
2016                         hsm_policy_bit2str(m, 0, false);
2017                         GOTO(out, rc = -EINVAL);
2018                 }
2019                 switch (sign) {
2020                 case '-':
2021                         remove_mask |= policy;
2022                         break;
2023                 case '+':
2024                         add_mask |= policy;
2025                         break;
2026                 default:
2027                         set_mask |= policy;
2028                         break;
2029                 }
2030
2031         } while (start != NULL);
2032
2033         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
2034                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
2035
2036         /* if no sign in all string, it is a clear and set
2037          * if some sign found, all unsigned are converted
2038          * to add
2039          * P1 P2 = set to P1 and P2
2040          * P1 -P2 = add P1 clear P2 same as +P1 -P2
2041          */
2042         if (remove_mask == 0 && add_mask == 0) {
2043                 cdt->cdt_policy = set_mask;
2044         } else {
2045                 cdt->cdt_policy |= set_mask | add_mask;
2046                 cdt->cdt_policy &= ~remove_mask;
2047         }
2048
2049         GOTO(out, rc = count);
2050
2051 out:
2052         OBD_FREE(buf, count + 1);
2053         RETURN(rc);
2054 }
2055 LPROC_SEQ_FOPS(mdt_hsm_policy);
2056
2057 #define GENERATE_PROC_METHOD(VAR)                                       \
2058 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
2059 {                                                                       \
2060         struct mdt_device       *mdt = m->private;                      \
2061         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
2062         ENTRY;                                                          \
2063                                                                         \
2064         seq_printf(m, "%llu\n", (__u64)cdt->VAR);                       \
2065         RETURN(0);                                                      \
2066 }                                                                       \
2067 static ssize_t                                                          \
2068 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
2069                           size_t count, loff_t *off)                    \
2070                                                                         \
2071 {                                                                       \
2072         struct seq_file         *m = file->private_data;                \
2073         struct mdt_device       *mdt = m->private;                      \
2074         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
2075         unsigned int val;                                               \
2076         int rc;                                                         \
2077                                                                         \
2078         ENTRY;                                                          \
2079         rc = kstrtouint_from_user(buffer, count, 0, &val);              \
2080         if (rc)                                                         \
2081                 RETURN(rc);                                             \
2082                                                                         \
2083         if (val !=  0) {                                                \
2084                 cdt->VAR = val;                                         \
2085                 RETURN(count);                                          \
2086         }                                                               \
2087         RETURN(-EINVAL);                                                \
2088 }                                                                       \
2089
2090 GENERATE_PROC_METHOD(cdt_loop_period)
2091 GENERATE_PROC_METHOD(cdt_grace_delay)
2092 GENERATE_PROC_METHOD(cdt_active_req_timeout)
2093 GENERATE_PROC_METHOD(cdt_max_requests)
2094 GENERATE_PROC_METHOD(cdt_default_archive_id)
2095
2096 /*
2097  * procfs write method for MDT/hsm_control
2098  * proc entry is in mdt directory so data is mdt obd_device pointer
2099  */
2100 #define CDT_ENABLE_CMD   "enabled"
2101 #define CDT_STOP_CMD     "shutdown"
2102 #define CDT_DISABLE_CMD  "disabled"
2103 #define CDT_PURGE_CMD    "purge"
2104 #define CDT_HELP_CMD     "help"
2105 #define CDT_MAX_CMD_LEN  10
2106
2107 ssize_t
2108 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
2109                               size_t count, loff_t *off)
2110 {
2111         struct seq_file         *m = file->private_data;
2112         struct obd_device       *obd = m->private;
2113         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
2114         struct coordinator      *cdt = &(mdt->mdt_coordinator);
2115         int                      rc, usage = 0;
2116         char                     kernbuf[CDT_MAX_CMD_LEN];
2117         ENTRY;
2118
2119         if (count == 0 || count >= sizeof(kernbuf))
2120                 RETURN(-EINVAL);
2121
2122         if (copy_from_user(kernbuf, buffer, count))
2123                 RETURN(-EFAULT);
2124         kernbuf[count] = 0;
2125
2126         if (kernbuf[count - 1] == '\n')
2127                 kernbuf[count - 1] = 0;
2128
2129         rc = 0;
2130         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
2131                 if (cdt->cdt_state == CDT_DISABLE) {
2132                         rc = set_cdt_state(cdt, CDT_RUNNING);
2133                         mdt_hsm_cdt_event(cdt);
2134                         wake_up(&cdt->cdt_waitq);
2135                 } else {
2136                         rc = mdt_hsm_cdt_start(mdt);
2137                 }
2138         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2139                 if ((cdt->cdt_state == CDT_STOPPING) ||
2140                     (cdt->cdt_state == CDT_STOPPED)) {
2141                         CERROR("%s: Coordinator already stopped\n",
2142                                mdt_obd_name(mdt));
2143                         rc = -EALREADY;
2144                 } else {
2145                         rc = mdt_hsm_cdt_stop(mdt);
2146                 }
2147         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2148                 if ((cdt->cdt_state == CDT_STOPPING) ||
2149                     (cdt->cdt_state == CDT_STOPPED)) {
2150                         CERROR("%s: Coordinator is stopped\n",
2151                                mdt_obd_name(mdt));
2152                         rc = -EINVAL;
2153                 } else {
2154                         rc = set_cdt_state(cdt, CDT_DISABLE);
2155                 }
2156         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2157                 rc = hsm_cancel_all_actions(mdt);
2158         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2159                 usage = 1;
2160         } else {
2161                 usage = 1;
2162                 rc = -EINVAL;
2163         }
2164
2165         if (usage == 1)
2166                 CERROR("%s: Valid coordinator control commands are: "
2167                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2168                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2169                        CDT_PURGE_CMD, CDT_HELP_CMD);
2170
2171         if (rc)
2172                 RETURN(rc);
2173
2174         RETURN(count);
2175 }
2176
2177 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2178 {
2179         struct obd_device       *obd = m->private;
2180         struct coordinator      *cdt;
2181         ENTRY;
2182
2183         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2184
2185         seq_printf(m, "%s\n", cdt_mdt_state2str(cdt->cdt_state));
2186
2187         RETURN(0);
2188 }
2189
2190 static int
2191 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2192 {
2193         bool first = true;
2194         int i;
2195         ENTRY;
2196
2197         for (i = 0; i < 8 * sizeof(mask); i++) {
2198                 if (mask & (1UL << i)) {
2199                         seq_printf(m, "%s%s", first ? "" : " ",
2200                                    hsm_copytool_action2name(i));
2201                         first = false;
2202                 }
2203         }
2204         seq_putc(m, '\n');
2205
2206         RETURN(0);
2207 }
2208
2209 static int
2210 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2211 {
2212         struct mdt_device *mdt = m->private;
2213         struct coordinator *cdt = &mdt->mdt_coordinator;
2214
2215         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2216 }
2217
2218 static int
2219 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2220 {
2221         struct mdt_device *mdt = m->private;
2222         struct coordinator *cdt = &mdt->mdt_coordinator;
2223
2224         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2225 }
2226
2227 static int
2228 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2229 {
2230         struct mdt_device *mdt = m->private;
2231         struct coordinator *cdt = &mdt->mdt_coordinator;
2232
2233         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2234 }
2235
2236 static inline enum hsm_copytool_action
2237 hsm_copytool_name2action(const char *name)
2238 {
2239         if (strcasecmp(name, "NOOP") == 0)
2240                 return HSMA_NONE;
2241         else if (strcasecmp(name, "ARCHIVE") == 0)
2242                 return HSMA_ARCHIVE;
2243         else if (strcasecmp(name, "RESTORE") == 0)
2244                 return HSMA_RESTORE;
2245         else if (strcasecmp(name, "REMOVE") == 0)
2246                 return HSMA_REMOVE;
2247         else if (strcasecmp(name, "CANCEL") == 0)
2248                 return HSMA_CANCEL;
2249         else
2250                 return -1;
2251 }
2252
2253 static ssize_t
2254 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2255                             size_t user_count, __u64 *mask)
2256 {
2257         char *buf, *pos, *name;
2258         size_t buf_size;
2259         __u64 new_mask = 0;
2260         int rc;
2261         ENTRY;
2262
2263         if (!(user_count < 4096))
2264                 RETURN(-ENOMEM);
2265
2266         buf_size = user_count + 1;
2267
2268         OBD_ALLOC(buf, buf_size);
2269         if (buf == NULL)
2270                 RETURN(-ENOMEM);
2271
2272         if (copy_from_user(buf, user_buf, buf_size - 1))
2273                 GOTO(out, rc = -EFAULT);
2274
2275         buf[buf_size - 1] = '\0';
2276
2277         pos = buf;
2278         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2279                 int action;
2280
2281                 if (*name == '\0')
2282                         continue;
2283
2284                 action = hsm_copytool_name2action(name);
2285                 if (action < 0)
2286                         GOTO(out, rc = -EINVAL);
2287
2288                 new_mask |= (1UL << action);
2289         }
2290
2291         *mask = new_mask;
2292         rc = user_count;
2293 out:
2294         OBD_FREE(buf, buf_size);
2295
2296         RETURN(rc);
2297 }
2298
2299 static ssize_t
2300 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2301                                         size_t count, loff_t *off)
2302 {
2303         struct seq_file         *m = file->private_data;
2304         struct mdt_device       *mdt = m->private;
2305         struct coordinator *cdt = &mdt->mdt_coordinator;
2306
2307         return mdt_write_hsm_request_mask(file, buf, count,
2308                                            &cdt->cdt_user_request_mask);
2309 }
2310
2311 static ssize_t
2312 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2313                                         size_t count, loff_t *off)
2314 {
2315         struct seq_file         *m = file->private_data;
2316         struct mdt_device       *mdt = m->private;
2317         struct coordinator      *cdt = &mdt->mdt_coordinator;
2318
2319         return mdt_write_hsm_request_mask(file, buf, count,
2320                                            &cdt->cdt_group_request_mask);
2321 }
2322
2323 static ssize_t
2324 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2325                                         size_t count, loff_t *off)
2326 {
2327         struct seq_file         *m = file->private_data;
2328         struct mdt_device       *mdt = m->private;
2329         struct coordinator      *cdt = &mdt->mdt_coordinator;
2330
2331         return mdt_write_hsm_request_mask(file, buf, count,
2332                                            &cdt->cdt_other_request_mask);
2333 }
2334
2335 static int mdt_hsm_cdt_raolu_seq_show(struct seq_file *m, void *data)
2336 {
2337         struct mdt_device *mdt = m->private;
2338         struct coordinator *cdt = &mdt->mdt_coordinator;
2339         ENTRY;
2340
2341         seq_printf(m, "%d\n", (int)cdt->cdt_remove_archive_on_last_unlink);
2342         RETURN(0);
2343 }
2344
2345 static ssize_t
2346 mdt_hsm_cdt_raolu_seq_write(struct file *file, const char __user *buffer,
2347                             size_t count, loff_t *off)
2348
2349 {
2350         struct seq_file *m = file->private_data;
2351         struct mdt_device *mdt = m->private;
2352         struct coordinator *cdt = &mdt->mdt_coordinator;
2353         bool val;
2354         int rc;
2355
2356         ENTRY;
2357         rc = kstrtobool_from_user(buffer, count, &val);
2358         if (rc < 0)
2359                 RETURN(rc);
2360
2361         cdt->cdt_remove_archive_on_last_unlink = val;
2362         RETURN(count);
2363 }
2364
2365 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2366 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2367 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2368 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2369 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2370 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2371 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2372 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2373 LPROC_SEQ_FOPS(mdt_hsm_cdt_raolu);
2374
2375 /* Read-only proc files for request counters */
2376 static int mdt_hsm_cdt_archive_count_seq_show(struct seq_file *m, void *data)
2377 {
2378         struct mdt_device *mdt = m->private;
2379         struct coordinator *cdt = &mdt->mdt_coordinator;
2380         ENTRY;
2381
2382         seq_printf(m, "%d\n", atomic_read(&cdt->cdt_archive_count));
2383         RETURN(0);
2384 }
2385
2386 static int mdt_hsm_cdt_restore_count_seq_show(struct seq_file *m, void *data)
2387 {
2388         struct mdt_device *mdt = m->private;
2389         struct coordinator *cdt = &mdt->mdt_coordinator;
2390         ENTRY;
2391
2392         seq_printf(m, "%d\n", atomic_read(&cdt->cdt_restore_count));
2393         RETURN(0);
2394 }
2395
2396 static int mdt_hsm_cdt_remove_count_seq_show(struct seq_file *m, void *data)
2397 {
2398         struct mdt_device *mdt = m->private;
2399         struct coordinator *cdt = &mdt->mdt_coordinator;
2400         ENTRY;
2401
2402         seq_printf(m, "%d\n", atomic_read(&cdt->cdt_remove_count));
2403         RETURN(0);
2404 }
2405
2406 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_archive_count);
2407 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_restore_count);
2408 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_remove_count);
2409
2410 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2411         { .name =       "agents",
2412           .fops =       &mdt_hsm_agent_fops                     },
2413         { .name =       "actions",
2414           .fops =       &mdt_hsm_actions_fops,
2415           .proc_mode =  0444                                    },
2416         { .name =       "default_archive_id",
2417           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2418         { .name =       "grace_delay",
2419           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2420         { .name =       "loop_period",
2421           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2422         { .name =       "max_requests",
2423           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2424         { .name =       "policy",
2425           .fops =       &mdt_hsm_policy_fops                    },
2426         { .name =       "active_request_timeout",
2427           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2428         { .name =       "active_requests",
2429           .fops =       &mdt_hsm_active_requests_fops           },
2430         { .name =       "user_request_mask",
2431           .fops =       &mdt_hsm_user_request_mask_fops,        },
2432         { .name =       "group_request_mask",
2433           .fops =       &mdt_hsm_group_request_mask_fops,       },
2434         { .name =       "other_request_mask",
2435           .fops =       &mdt_hsm_other_request_mask_fops,       },
2436         { .name =       "remove_archive_on_last_unlink",
2437           .fops =       &mdt_hsm_cdt_raolu_fops,                },
2438         { .name =       "archive_count",
2439           .fops =       &mdt_hsm_cdt_archive_count_fops,        },
2440         { .name =       "restore_count",
2441           .fops =       &mdt_hsm_cdt_restore_count_fops,        },
2442         { .name =       "remove_count",
2443           .fops =       &mdt_hsm_cdt_remove_count_fops,         },
2444         { 0 }
2445 };