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