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(&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         uc->uc_valid = UCRED_OLD;
 927         uc->uc_o_uid = 0;
 928         uc->uc_o_gid = 0;
 929         uc->uc_o_fsuid = 0;
 930         uc->uc_o_fsgid = 0;
 931         uc->uc_uid = 0;
 932         uc->uc_gid = 0;
 933         uc->uc_fsuid = 0;
 934         uc->uc_fsgid = 0;
 935         uc->uc_suppgids[0] = -1;
 936         uc->uc_suppgids[1] = -1;
 937         uc->uc_cap = cap_combine(CAP_FS_SET, CAP_NFSD_SET);
 938         uc->uc_umask = 0777;
 939         uc->uc_ginfo = NULL;
 940         uc->uc_identity = NULL;
 941         /* always record internal HSM activity if also enabled globally */
 942         uc->uc_enable_audit = 1;
 943
 944         RETURN(0);
 945 }
 946
 947 /**
 948  * initialize coordinator struct
 949  * \param mdt [IN] device
 950  * \retval 0 success
 951  * \retval -ve failure
 952  */
 953 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 954 {
 955         struct coordinator      *cdt = &mdt->mdt_coordinator;
 956         struct mdt_thread_info  *cdt_mti = NULL;
 957         int                      rc;
 958         ENTRY;
 959
 960         init_waitqueue_head(&cdt->cdt_waitq);
 961         init_rwsem(&cdt->cdt_llog_lock);
 962         init_rwsem(&cdt->cdt_agent_lock);
 963         init_rwsem(&cdt->cdt_request_lock);
 964         mutex_init(&cdt->cdt_restore_lock);
 965         mutex_init(&cdt->cdt_state_lock);
 966         set_cdt_state(cdt, CDT_STOPPED);
 967
 968         INIT_LIST_HEAD(&cdt->cdt_request_list);
 969         INIT_LIST_HEAD(&cdt->cdt_agents);
 970         INIT_LIST_HEAD(&cdt->cdt_restore_handle_list);
 971
 972         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
 973                                                        CFS_HASH_BITS_MIN,
 974                                                        CFS_HASH_BITS_MAX,
 975                                                        CFS_HASH_BKT_BITS,
 976                                                        0 /* extra bytes */,
 977                                                        CFS_HASH_MIN_THETA,
 978                                                        CFS_HASH_MAX_THETA,
 979                                                 &cdt_request_cookie_hash_ops,
 980                                                        CFS_HASH_DEFAULT);
 981         if (cdt->cdt_request_cookie_hash == NULL)
 982                 RETURN(-ENOMEM);
 983
 984         cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
 985                                                      CFS_HASH_BITS_MIN,
 986                                                      CFS_HASH_BITS_MAX,
 987                                                      CFS_HASH_BKT_BITS,
 988                                                      0 /* extra bytes */,
 989                                                      CFS_HASH_MIN_THETA,
 990                                                      CFS_HASH_MAX_THETA,
 991                                                      &cdt_agent_record_hash_ops,
 992                                                      CFS_HASH_DEFAULT);
 993         if (cdt->cdt_agent_record_hash == NULL)
 994                 GOTO(out_request_cookie_hash, rc = -ENOMEM);
 995
 996         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 997         if (rc < 0)
 998                 GOTO(out_agent_record_hash, rc);
 999
1000         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1001         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
1002         if (rc < 0)
1003                 GOTO(out_env, rc);
1004
1005         lu_context_enter(&cdt->cdt_session);
1006         cdt->cdt_env.le_ses = &cdt->cdt_session;
1007
1008         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1009         LASSERT(cdt_mti != NULL);
1010
1011         cdt_mti->mti_env = &cdt->cdt_env;
1012         cdt_mti->mti_mdt = mdt;
1013
1014         hsm_init_ucred(mdt_ucred(cdt_mti));
1015
1016         /* default values for sysfs tunnables
1017          * can be override by MGS conf */
1018         cdt->cdt_default_archive_id = 1;
1019         cdt->cdt_grace_delay = 60;
1020         cdt->cdt_loop_period = 10;
1021         cdt->cdt_max_requests = 3;
1022         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1023         cdt->cdt_active_req_timeout = 3600;
1024
1025         /* by default do not remove archives on last unlink */
1026         cdt->cdt_remove_archive_on_last_unlink = false;
1027
1028         RETURN(0);
1029
1030 out_env:
1031         lu_env_fini(&cdt->cdt_env);
1032 out_agent_record_hash:
1033         cfs_hash_putref(cdt->cdt_agent_record_hash);
1034         cdt->cdt_agent_record_hash = NULL;
1035 out_request_cookie_hash:
1036         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1037         cdt->cdt_request_cookie_hash = NULL;
1038
1039         return rc;
1040 }
1041
1042 /**
1043  * free a coordinator thread
1044  * \param mdt [IN] device
1045  */
1046 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
1047 {
1048         struct coordinator *cdt = &mdt->mdt_coordinator;
1049         ENTRY;
1050
1051         lu_context_exit(cdt->cdt_env.le_ses);
1052         lu_context_fini(cdt->cdt_env.le_ses);
1053
1054         lu_env_fini(&cdt->cdt_env);
1055
1056         cfs_hash_putref(cdt->cdt_agent_record_hash);
1057         cdt->cdt_agent_record_hash = NULL;
1058
1059         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1060         cdt->cdt_request_cookie_hash = NULL;
1061
1062         RETURN(0);
1063 }
1064
1065 /**
1066  * start a coordinator thread
1067  * \param mdt [IN] device
1068  * \retval 0 success
1069  * \retval -ve failure
1070  */
1071 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1072 {
1073         struct coordinator *cdt = &mdt->mdt_coordinator;
1074         struct mdt_thread_info *cdt_mti;
1075         unsigned int i = 0;
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         /* wait until MDD initialize hsm actions llog */
1108         while (!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state) && i < obd_timeout) {
1109                 schedule_timeout_interruptible(cfs_time_seconds(1));
1110                 i++;
1111         }
1112         if (!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state))
1113                 CWARN("%s: trying to init HSM before MDD\n", mdt_obd_name(mdt));
1114
1115         /* to avoid deadlock when start is made through sysfs
1116          * sysfs entries are created by the coordinator thread
1117          */
1118         /* set up list of started restore requests */
1119         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1120         rc = mdt_hsm_pending_restore(cdt_mti);
1121         if (rc)
1122                 CERROR("%s: cannot take the layout locks needed"
1123                        " for registered restore: %d\n",
1124                        mdt_obd_name(mdt), rc);
1125
1126         if (mdt->mdt_bottom->dd_rdonly)
1127                 RETURN(0);
1128
1129         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
1130         if (IS_ERR(task)) {
1131                 rc = PTR_ERR(task);
1132                 set_cdt_state(cdt, CDT_STOPPED);
1133                 CERROR("%s: error starting coordinator thread: %d\n",
1134                        mdt_obd_name(mdt), rc);
1135         } else {
1136                 cdt->cdt_task = task;
1137                 wait_event(cdt->cdt_waitq,
1138                            cdt->cdt_state != CDT_INIT);
1139                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1140                        mdt_obd_name(mdt));
1141                 rc = 0;
1142         }
1143
1144         RETURN(rc);
1145 }
1146
1147 /**
1148  * stop a coordinator thread
1149  * \param mdt [IN] device
1150  */
1151 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1152 {
1153         struct coordinator *cdt = &mdt->mdt_coordinator;
1154         int rc;
1155
1156         ENTRY;
1157         /* stop coordinator thread */
1158         rc = set_cdt_state(cdt, CDT_STOPPING);
1159         if (rc == 0) {
1160                 kthread_stop(cdt->cdt_task);
1161                 cdt->cdt_task = NULL;
1162                 set_cdt_state(cdt, CDT_STOPPED);
1163         }
1164
1165         RETURN(rc);
1166 }
1167
1168 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1169                               const struct lu_fid *fid,
1170                               u32 archive_id)
1171 {
1172         struct mdt_object *obj;
1173         struct md_hsm mh;
1174         int rc;
1175
1176         obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1177         if (IS_ERR(obj))
1178                 GOTO(out, rc = PTR_ERR(obj));
1179
1180         if (mh.mh_flags & HS_EXISTS &&
1181             mh.mh_arch_id == archive_id)
1182                 GOTO(out_obj, rc = 0);
1183
1184         mh.mh_flags |= HS_EXISTS;
1185         mh.mh_arch_id = archive_id;
1186         rc = mdt_hsm_attr_set(mti, obj, &mh);
1187
1188 out_obj:
1189         mdt_object_put(mti->mti_env, obj);
1190 out:
1191         return rc;
1192 }
1193
1194 /**
1195  * register all requests from an hal in the memory list
1196  * \param mti [IN] context
1197  * \param hal [IN] request
1198  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1199  *  which is running the CT
1200  * \retval 0 success
1201  * \retval -ve failure
1202  */
1203 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1204                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1205 {
1206         struct mdt_device       *mdt = mti->mti_mdt;
1207         struct coordinator      *cdt = &mdt->mdt_coordinator;
1208         struct hsm_action_item  *hai;
1209         int                      rc = 0, i;
1210         ENTRY;
1211
1212         /* register request in memory list */
1213         hai = hai_first(hal);
1214         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1215                 struct cdt_agent_req *car;
1216
1217                 /* in case of a cancel request, we first mark the ondisk
1218                  * record of the request we want to stop as canceled
1219                  * this does not change the cancel record
1220                  * it will be done when updating the request status
1221                  */
1222                 if (hai->hai_action == HSMA_CANCEL) {
1223                         struct hsm_record_update update = {
1224                                 .cookie = hai->hai_cookie,
1225                                 .status = ARS_CANCELED,
1226                         };
1227
1228                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1229                                                      &update, 1);
1230                         if (rc) {
1231                                 CERROR("%s: mdt_agent_record_update() failed, "
1232                                        "rc=%d, cannot update status to %s "
1233                                        "for cookie %#llx\n",
1234                                        mdt_obd_name(mdt), rc,
1235                                        agent_req_status2name(ARS_CANCELED),
1236                                        hai->hai_cookie);
1237                                 GOTO(out, rc);
1238                         }
1239
1240                         /* find the running request to set it canceled */
1241                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1242                         if (car != NULL) {
1243                                 car->car_canceled = 1;
1244                                 /* uuid has to be changed to the one running the
1245                                 * request to cancel */
1246                                 *uuid = car->car_uuid;
1247                                 mdt_cdt_put_request(car);
1248                         }
1249                         /* no need to memorize cancel request
1250                          * this also avoid a deadlock when we receive
1251                          * a purge all requests command
1252                          */
1253                         continue;
1254                 }
1255
1256                 if (hai->hai_action == HSMA_ARCHIVE) {
1257                         rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1258                                                 hal->hal_archive_id);
1259                         if (rc == -ENOENT)
1260                                 continue;
1261                         else if (rc < 0)
1262                                 GOTO(out, rc);
1263                 }
1264
1265                 car = mdt_cdt_alloc_request(hal->hal_archive_id, hal->hal_flags,
1266                                             uuid, hai);
1267                 if (IS_ERR(car))
1268                         GOTO(out, rc = PTR_ERR(car));
1269
1270                 rc = mdt_cdt_add_request(cdt, car);
1271                 if (rc != 0)
1272                         mdt_cdt_free_request(car);
1273         }
1274 out:
1275         RETURN(rc);
1276 }
1277
1278 /**
1279  * swap layouts between 2 fids
1280  * \param mti [IN] context
1281  * \param obj [IN]
1282  * \param dfid [IN]
1283  * \param mh_common [IN] MD HSM
1284  */
1285 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1286                             struct mdt_object *obj, const struct lu_fid *dfid,
1287                             struct md_hsm *mh_common)
1288 {
1289         struct mdt_object       *dobj;
1290         struct mdt_lock_handle  *dlh;
1291         int                      rc;
1292         ENTRY;
1293
1294         if (!mdt_object_exists(obj))
1295                 GOTO(out, rc = -ENOENT);
1296
1297         /* we already have layout lock on obj so take only
1298          * on dfid */
1299         dlh = &mti->mti_lh[MDT_LH_OLD];
1300         mdt_lock_reg_init(dlh, LCK_EX);
1301         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1302         if (IS_ERR(dobj))
1303                 GOTO(out, rc = PTR_ERR(dobj));
1304
1305         /* if copy tool closes the volatile before sending the final
1306          * progress through llapi_hsm_copy_end(), all the objects
1307          * are removed and mdd_swap_layout LBUG */
1308         if (!mdt_object_exists(dobj)) {
1309                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1310                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1311                 GOTO(out_dobj, rc = -ENOENT);
1312         }
1313         /* Since we only handle restores here, unconditionally use
1314          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1315          * be preserved in case of failure during swap_layout and not
1316          * leave a file in an intermediate but incoherent state.
1317          * But need to setup HSM xattr of data FID before, reuse
1318          * mti and mh presets for FID in hsm_cdt_request_completed(),
1319          * only need to clear RELEASED and DIRTY.
1320          */
1321         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1322         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1323         if (rc == 0)
1324                 rc = mo_swap_layouts(mti->mti_env,
1325                                      mdt_object_child(obj),
1326                                      mdt_object_child(dobj),
1327                                      SWAP_LAYOUTS_MDS_HSM);
1328         if (rc == 0) {
1329                 rc = mdt_lsom_downgrade(mti, obj);
1330                 if (rc)
1331                         CDEBUG(D_INODE,
1332                                "%s: File fid="DFID" SOM "
1333                                "downgrade failed, rc = %d\n",
1334                                mdt_obd_name(mti->mti_mdt),
1335                                PFID(mdt_object_fid(obj)), rc);
1336         }
1337 out_dobj:
1338         mdt_object_unlock_put(mti, dobj, dlh, 1);
1339 out:
1340         RETURN(rc);
1341 }
1342
1343 /**
1344  * update status of a completed request
1345  * \param mti [IN] context
1346  * \param pgs [IN] progress of the copy tool
1347  * \retval 0 success
1348  * \retval -ve failure
1349  */
1350 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1351                                      struct hsm_progress_kernel *pgs,
1352                                      const struct cdt_agent_req *car,
1353                                      enum agent_req_status *status)
1354 {
1355         const struct lu_env *env = mti->mti_env;
1356         struct mdt_device *mdt = mti->mti_mdt;
1357         struct coordinator *cdt = &mdt->mdt_coordinator;
1358         struct mdt_object *obj = NULL;
1359         enum changelog_rec_flags clf_flags = 0;
1360         struct md_hsm mh;
1361         bool is_mh_changed;
1362         bool need_changelog = true;
1363         int rc = 0;
1364
1365         ENTRY;
1366         /* default is to retry */
1367         *status = ARS_WAITING;
1368
1369         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1370          * if error/removed continue anyway to get correct reporting done */
1371         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1372         /* we will update MD HSM only if needed */
1373         is_mh_changed = false;
1374
1375         /* no need to change mh->mh_arch_id
1376          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1377          */
1378         if (pgs->hpk_errval != 0) {
1379                 switch (pgs->hpk_errval) {
1380                 case ENOSYS:
1381                         /* the copy tool does not support cancel
1382                          * so the cancel request is failed
1383                          * As we cannot distinguish a cancel progress
1384                          * from another action progress (they have the
1385                          * same cookie), we suppose here the CT returns
1386                          * ENOSYS only if does not support cancel
1387                          */
1388                         /* this can also happen when cdt calls it to
1389                          * for a timed out request */
1390                         *status = ARS_FAILED;
1391                         /* to have a cancel event in changelog */
1392                         pgs->hpk_errval = ECANCELED;
1393                         break;
1394                 case ECANCELED:
1395                         /* the request record has already been set to
1396                          * ARS_CANCELED, this set the cancel request
1397                          * to ARS_SUCCEED */
1398                         *status = ARS_SUCCEED;
1399                         break;
1400                 default:
1401                         /* retry only if current policy or requested, and
1402                          * object is not on error/removed */
1403                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1404                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1405                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1406                         break;
1407                 }
1408
1409                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1410                         CERROR("%s: Request %#llx on "DFID
1411                                " failed, error code %d too large\n",
1412                                mdt_obd_name(mdt),
1413                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1414                                pgs->hpk_errval);
1415                         hsm_set_cl_error(&clf_flags, CLF_HSM_ERROVERFLOW);
1416                         rc = -EINVAL;
1417                 } else {
1418                         hsm_set_cl_error(&clf_flags, pgs->hpk_errval);
1419                 }
1420
1421                 switch (car->car_hai->hai_action) {
1422                 case HSMA_ARCHIVE:
1423                         hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1424                         break;
1425                 case HSMA_RESTORE:
1426                         hsm_set_cl_event(&clf_flags, HE_RESTORE);
1427                         break;
1428                 case HSMA_REMOVE:
1429                         hsm_set_cl_event(&clf_flags, HE_REMOVE);
1430                         break;
1431                 case HSMA_CANCEL:
1432                         hsm_set_cl_event(&clf_flags, HE_CANCEL);
1433                         CERROR("%s: Failed request %#llx on "DFID
1434                                " cannot be a CANCEL\n",
1435                                mdt_obd_name(mdt),
1436                                pgs->hpk_cookie,
1437                                PFID(&pgs->hpk_fid));
1438                         break;
1439                 default:
1440                         CERROR("%s: Failed request %#llx on "DFID
1441                                " %d is an unknown action\n",
1442                                mdt_obd_name(mdt),
1443                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1444                                car->car_hai->hai_action);
1445                         rc = -EINVAL;
1446                         break;
1447                 }
1448         } else {
1449                 *status = ARS_SUCCEED;
1450                 switch (car->car_hai->hai_action) {
1451                 case HSMA_ARCHIVE:
1452                         hsm_set_cl_event(&clf_flags, HE_ARCHIVE);
1453                         /* set ARCHIVE keep EXIST and clear LOST and
1454                          * DIRTY */
1455                         mh.mh_arch_ver = pgs->hpk_data_version;
1456                         mh.mh_flags |= HS_ARCHIVED;
1457                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1458                         is_mh_changed = true;
1459                         break;
1460                 case HSMA_RESTORE:
1461                         hsm_set_cl_event(&clf_flags, HE_RESTORE);
1462
1463                         /* do not clear RELEASED and DIRTY here
1464                          * this will occur in hsm_swap_layouts()
1465                          */
1466
1467                         /* Restoring has changed the file version on
1468                          * disk. */
1469                         mh.mh_arch_ver = pgs->hpk_data_version;
1470                         is_mh_changed = true;
1471                         break;
1472                 case HSMA_REMOVE:
1473                         hsm_set_cl_event(&clf_flags, HE_REMOVE);
1474                         /* clear ARCHIVED EXISTS and LOST */
1475                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1476                         is_mh_changed = true;
1477                         break;
1478                 case HSMA_CANCEL:
1479                         hsm_set_cl_event(&clf_flags, HE_CANCEL);
1480                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1481                                mdt_obd_name(mdt),
1482                                pgs->hpk_cookie,
1483                                PFID(&pgs->hpk_fid));
1484                         break;
1485                 default:
1486                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1487                                mdt_obd_name(mdt),
1488                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1489                                car->car_hai->hai_action);
1490                         rc = -EINVAL;
1491                         break;
1492                 }
1493         }
1494
1495         /* rc != 0 means error when analysing action, it may come from
1496          * a crasy CT no need to manage DIRTY
1497          * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1498          * filled
1499          */
1500         if (rc == 0 && !IS_ERR(obj))
1501                 hsm_set_cl_flags(&clf_flags,
1502                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1503
1504         /* unlock is done later, after layout lock management */
1505         if (is_mh_changed && !IS_ERR(obj))
1506                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1507
1508         /* we give back layout lock only if restore was successful or
1509          * if no retry will be attempted and if object is still alive,
1510          * in other cases we just unlock the object */
1511         if (car->car_hai->hai_action == HSMA_RESTORE) {
1512                 struct mdt_lock_handle *lh;
1513
1514                 /* restore in data FID done, we swap the layouts
1515                  * only if restore is successful */
1516                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1517                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1518                                               &mh);
1519                         if (rc) {
1520                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1521                                         *status = ARS_FAILED;
1522                                 pgs->hpk_errval = -rc;
1523                         }
1524                 }
1525                 /* we have to retry, so keep layout lock */
1526                 if (*status == ARS_WAITING)
1527                         GOTO(out, rc);
1528
1529                 /* restore special case, need to create ChangeLog record
1530                  * before to give back layout lock to avoid concurrent
1531                  * file updater to post out of order ChangeLog */
1532                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1533                              &car->car_hai->hai_fid);
1534                 need_changelog = false;
1535
1536                 cdt_restore_handle_del(mti, cdt, &car->car_hai->hai_fid);
1537                 if (!IS_ERR_OR_NULL(obj)) {
1538                         /* flush UPDATE lock so attributes are upadated */
1539                         lh = &mti->mti_lh[MDT_LH_OLD];
1540                         mdt_lock_reg_init(lh, LCK_EX);
1541                         mdt_object_lock(mti, obj, lh, MDS_INODELOCK_UPDATE);
1542                         mdt_object_unlock(mti, obj, lh, 1);
1543                 }
1544         }
1545
1546         GOTO(out, rc);
1547
1548 out:
1549         /* always add a ChangeLog record */
1550         if (need_changelog)
1551                 mo_changelog(env, CL_HSM, clf_flags, mdt->mdt_child,
1552                              &car->car_hai->hai_fid);
1553
1554         if (!IS_ERR(obj))
1555                 mdt_object_put(mti->mti_env, obj);
1556
1557         RETURN(rc);
1558 }
1559
1560 /**
1561  * update status of a request
1562  * \param mti [IN] context
1563  * \param pgs [IN] progress of the copy tool
1564  * \retval 0 success
1565  * \retval -ve failure
1566  */
1567 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1568                                  struct hsm_progress_kernel *pgs)
1569 {
1570         struct mdt_device       *mdt = mti->mti_mdt;
1571         struct coordinator      *cdt = &mdt->mdt_coordinator;
1572         struct cdt_agent_req    *car;
1573         int                      rc = 0;
1574         ENTRY;
1575
1576         /* no coordinator started, so we cannot serve requests */
1577         if (cdt->cdt_state == CDT_STOPPED)
1578                 RETURN(-EAGAIN);
1579
1580         /* first do sanity checks */
1581         car = mdt_cdt_update_request(cdt, pgs);
1582         if (IS_ERR(car)) {
1583                 CERROR("%s: Cannot find running request for cookie %#llx"
1584                        " on fid="DFID"\n",
1585                        mdt_obd_name(mdt),
1586                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1587
1588                 RETURN(PTR_ERR(car));
1589         }
1590
1591         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1592                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1593                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1594                       hsm_copytool_action2name(car->car_hai->hai_action),
1595                       pgs->hpk_flags, pgs->hpk_errval,
1596                       PFID(&car->car_hai->hai_fid),
1597                       PFID(&car->car_hai->hai_dfid));
1598
1599         /* progress is done on FID or data FID depending of the action and
1600          * of the copy progress */
1601         /* for restore progress is used to send back the data FID to cdt */
1602         if (car->car_hai->hai_action == HSMA_RESTORE &&
1603             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1604                 car->car_hai->hai_dfid = pgs->hpk_fid;
1605
1606         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1607              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1608             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1609              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1610                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1611                        " does not match request FID "DFID" nor data FID "
1612                        DFID"\n",
1613                        mdt_obd_name(mdt),
1614                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1615                        PFID(&car->car_hai->hai_fid),
1616                        PFID(&car->car_hai->hai_dfid));
1617                 GOTO(out, rc = -EINVAL);
1618         }
1619
1620         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1621                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1622                        " is not coherent (err=%d and not completed"
1623                        " (flags=%d))\n",
1624                        mdt_obd_name(mdt),
1625                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1626                        hsm_copytool_action2name(car->car_hai->hai_action),
1627                        pgs->hpk_errval, pgs->hpk_flags);
1628                 GOTO(out, rc = -EINVAL);
1629         }
1630
1631         /* now progress is valid */
1632
1633         /* we use a root like ucred */
1634         hsm_init_ucred(mdt_ucred(mti));
1635
1636         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1637                 enum agent_req_status status;
1638                 struct hsm_record_update update;
1639                 int rc1;
1640
1641                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1642
1643                 CDEBUG(D_HSM, "updating record: fid="DFID" cookie=%#llx action=%s "
1644                               "status=%s\n",
1645                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1646                        hsm_copytool_action2name(car->car_hai->hai_action),
1647                        agent_req_status2name(status));
1648
1649                 /* update record first (LU-9075) */
1650                 update.cookie = pgs->hpk_cookie;
1651                 update.status = status;
1652
1653                 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1654                                               &update, 1);
1655                 if (rc1)
1656                         CERROR("%s: mdt_agent_record_update() failed,"
1657                                " rc=%d, cannot update status to %s"
1658                                " for cookie %#llx\n",
1659                                mdt_obd_name(mdt), rc1,
1660                                agent_req_status2name(status),
1661                                pgs->hpk_cookie);
1662                 rc = (rc != 0 ? rc : rc1);
1663
1664                 /* then remove request from memory list (LU-9075) */
1665                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1666
1667                 /* ct has completed a request, so a slot is available,
1668                  * signal the coordinator to find new work */
1669                 mdt_hsm_cdt_event(cdt);
1670         } else {
1671                 /* if copytool send a progress on a canceled request
1672                  * we inform copytool it should stop
1673                  */
1674                 if (car->car_canceled == 1)
1675                         rc = -ECANCELED;
1676         }
1677         GOTO(out, rc);
1678
1679 out:
1680         /* remove ref got from mdt_cdt_update_request() */
1681         mdt_cdt_put_request(car);
1682
1683         return rc;
1684 }
1685
1686
1687 /**
1688  * data passed to llog_cat_process() callback
1689  * to cancel requests
1690  */
1691 struct hsm_cancel_all_data {
1692         struct mdt_device       *mdt;
1693 };
1694
1695 /**
1696  *  llog_cat_process() callback, used to:
1697  *  - purge all requests
1698  * \param env [IN] environment
1699  * \param llh [IN] llog handle
1700  * \param hdr [IN] llog record
1701  * \param data [IN] cb data = struct hsm_cancel_all_data
1702  * \retval 0 success
1703  * \retval -ve failure
1704  */
1705 static int mdt_cancel_all_cb(const struct lu_env *env,
1706                              struct llog_handle *llh,
1707                              struct llog_rec_hdr *hdr, void *data)
1708 {
1709         struct llog_agent_req_rec       *larr;
1710         struct hsm_cancel_all_data      *hcad;
1711         int                              rc = 0;
1712         ENTRY;
1713
1714         larr = (struct llog_agent_req_rec *)hdr;
1715         hcad = data;
1716         if (larr->arr_status == ARS_WAITING ||
1717             larr->arr_status == ARS_STARTED) {
1718                 larr->arr_status = ARS_CANCELED;
1719                 larr->arr_req_change = ktime_get_real_seconds();
1720                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1721         }
1722
1723         RETURN(rc);
1724 }
1725
1726 /**
1727  * cancel all actions
1728  * \param obd [IN] MDT device
1729  */
1730 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1731 {
1732         struct lu_env                    env;
1733         struct lu_context                session;
1734         struct mdt_thread_info          *mti;
1735         struct coordinator              *cdt = &mdt->mdt_coordinator;
1736         struct cdt_agent_req            *car;
1737         struct hsm_action_list          *hal = NULL;
1738         struct hsm_action_item          *hai;
1739         struct hsm_cancel_all_data       hcad;
1740         int                              hal_sz = 0, hal_len, rc;
1741         enum cdt_states                  old_state;
1742         ENTRY;
1743
1744         rc = lu_env_init(&env, LCT_MD_THREAD);
1745         if (rc < 0)
1746                 RETURN(rc);
1747
1748         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1749         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1750         if (rc < 0)
1751                 GOTO(out_env, rc);
1752
1753         lu_context_enter(&session);
1754         env.le_ses = &session;
1755
1756         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1757         LASSERT(mti != NULL);
1758
1759         mti->mti_env = &env;
1760         mti->mti_mdt = mdt;
1761
1762         hsm_init_ucred(mdt_ucred(mti));
1763
1764         mutex_lock(&cdt->cdt_state_lock);
1765         old_state = cdt->cdt_state;
1766
1767         /* disable coordinator */
1768         rc = set_cdt_state_locked(cdt, CDT_DISABLE);
1769         if (rc)
1770                 GOTO(out_cdt_state_unlock, rc);
1771
1772         /* send cancel to all running requests */
1773         down_read(&cdt->cdt_request_lock);
1774         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1775                 mdt_cdt_get_request(car);
1776                 /* request is not yet removed from list, it will be done
1777                  * when copytool will return progress
1778                  */
1779
1780                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1781                         mdt_cdt_put_request(car);
1782                         continue;
1783                 }
1784
1785                 /* needed size */
1786                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1787                           cfs_size_round(car->car_hai->hai_len);
1788
1789                 if (hal_len > hal_sz && hal_sz > 0) {
1790                         /* not enough room, free old buffer */
1791                         OBD_FREE(hal, hal_sz);
1792                         hal = NULL;
1793                 }
1794
1795                 /* empty buffer, allocate one */
1796                 if (hal == NULL) {
1797                         hal_sz = hal_len;
1798                         OBD_ALLOC(hal, hal_sz);
1799                         if (hal == NULL) {
1800                                 mdt_cdt_put_request(car);
1801                                 up_read(&cdt->cdt_request_lock);
1802                                 GOTO(out_cdt_state, rc = -ENOMEM);
1803                         }
1804                 }
1805
1806                 hal->hal_version = HAL_VERSION;
1807                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1808                                 MTI_NAME_MAXLEN);
1809                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1810                 hal->hal_archive_id = car->car_archive_id;
1811                 hal->hal_flags = car->car_flags;
1812                 hal->hal_count = 0;
1813
1814                 hai = hai_first(hal);
1815                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1816                 hai->hai_action = HSMA_CANCEL;
1817                 hal->hal_count = 1;
1818
1819                 /* it is possible to safely call mdt_hsm_agent_send()
1820                  * (ie without a deadlock on cdt_request_lock), because the
1821                  * write lock is taken only if we are not in purge mode
1822                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1823                  *   nor mdt_cdt_remove_request())
1824                  */
1825                 /* no conflict with cdt thread because cdt is disable and we
1826                  * have the request lock */
1827                 mdt_hsm_agent_send(mti, hal, 1);
1828
1829                 mdt_cdt_put_request(car);
1830         }
1831         up_read(&cdt->cdt_request_lock);
1832
1833         if (hal != NULL)
1834                 OBD_FREE(hal, hal_sz);
1835
1836         /* cancel all on-disk records */
1837         hcad.mdt = mdt;
1838
1839         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1840                               &hcad, 0, 0, WRITE);
1841 out_cdt_state:
1842         /* Enable coordinator, unless the coordinator was stopping. */
1843         set_cdt_state_locked(cdt, old_state);
1844 out_cdt_state_unlock:
1845         mutex_unlock(&cdt->cdt_state_lock);
1846
1847         lu_context_exit(&session);
1848         lu_context_fini(&session);
1849 out_env:
1850         lu_env_fini(&env);
1851
1852         RETURN(rc);
1853 }
1854
1855 /**
1856  * check if a request is compatible with file status
1857  * \param hai [IN] request description
1858  * \param archive_id [IN] request archive id
1859  * \param rq_flags [IN] request flags
1860  * \param hsm [IN] file HSM metadata
1861  * \retval boolean
1862  */
1863 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1864                               u32 archive_id, u64 rq_flags,
1865                               const struct md_hsm *hsm)
1866 {
1867         int      is_compat = false;
1868         int      hsm_flags;
1869         ENTRY;
1870
1871         hsm_flags = hsm->mh_flags;
1872         switch (hai->hai_action) {
1873         case HSMA_ARCHIVE:
1874                 if (!(hsm_flags & HS_NOARCHIVE) &&
1875                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1876                         is_compat = true;
1877
1878                 if (hsm_flags & HS_EXISTS &&
1879                     archive_id != 0 &&
1880                     archive_id != hsm->mh_arch_id)
1881                         is_compat = false;
1882
1883                 break;
1884         case HSMA_RESTORE:
1885                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1886                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1887                         is_compat = true;
1888                 break;
1889         case HSMA_REMOVE:
1890                 if (!(hsm_flags & HS_RELEASED) &&
1891                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1892                         is_compat = true;
1893                 break;
1894         case HSMA_CANCEL:
1895                 is_compat = true;
1896                 break;
1897         }
1898         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1899                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1900                       PFID(&hai->hai_fid),
1901                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1902                       hai->hai_extent.offset, hai->hai_extent.length,
1903                       hsm->mh_flags,
1904                       (is_compat ? "compatible" : "uncompatible"));
1905
1906         RETURN(is_compat);
1907 }
1908
1909 /*
1910  * sysfs interface used to get/set HSM behaviour (cdt->cdt_policy)
1911  */
1912 static const struct {
1913         __u64            bit;
1914         char            *name;
1915         char            *nickname;
1916 } hsm_policy_names[] = {
1917         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1918         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1919         { 0 },
1920 };
1921
1922 /**
1923  * convert a policy name to a bit
1924  * \param name [IN] policy name
1925  * \retval 0 unknown
1926  * \retval   policy bit
1927  */
1928 static __u64 hsm_policy_str2bit(const char *name)
1929 {
1930         int      i;
1931
1932         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1933                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1934                     strcmp(hsm_policy_names[i].name, name) == 0)
1935                         return hsm_policy_names[i].bit;
1936         return 0;
1937 }
1938
1939 /**
1940  * convert a policy bit field to a string
1941  * \param mask [IN] policy bit field
1942  * \param hexa [IN] print mask before bit names
1943  * \param buffer [OUT] string
1944  * \param count [IN] size of buffer
1945  */
1946 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1947                                 const bool hexa)
1948 {
1949         int      i, j;
1950         __u64    bit;
1951         ENTRY;
1952
1953         if (hexa)
1954                 seq_printf(m, "(%#llx) ", mask);
1955
1956         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1957                 bit = (1ULL << i);
1958
1959                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1960                         if (hsm_policy_names[j].bit == bit)
1961                                 break;
1962                 }
1963                 if (bit & mask)
1964                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1965                 else
1966                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1967         }
1968         /* remove last ' ' */
1969         m->count--;
1970         seq_putc(m, '\n');
1971 }
1972
1973 /* methods to read/write HSM policy flags */
1974 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1975 {
1976         struct mdt_device       *mdt = m->private;
1977         struct coordinator      *cdt = &mdt->mdt_coordinator;
1978         ENTRY;
1979
1980         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1981         RETURN(0);
1982 }
1983
1984 static ssize_t
1985 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1986                          size_t count, loff_t *off)
1987 {
1988         struct seq_file         *m = file->private_data;
1989         struct mdt_device       *mdt = m->private;
1990         struct coordinator      *cdt = &mdt->mdt_coordinator;
1991         char                    *start, *token, sign;
1992         char                    *buf;
1993         __u64                    policy;
1994         __u64                    add_mask, remove_mask, set_mask;
1995         int                      rc;
1996         ENTRY;
1997
1998         if (count + 1 > PAGE_SIZE)
1999                 RETURN(-EINVAL);
2000
2001         OBD_ALLOC(buf, count + 1);
2002         if (buf == NULL)
2003                 RETURN(-ENOMEM);
2004
2005         if (copy_from_user(buf, buffer, count))
2006                 GOTO(out, rc = -EFAULT);
2007
2008         buf[count] = '\0';
2009
2010         start = buf;
2011         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
2012                start);
2013
2014         add_mask = remove_mask = set_mask = 0;
2015         do {
2016                 token = strsep(&start, "\n ");
2017                 sign = *token;
2018
2019                 if (sign == '\0')
2020                         continue;
2021
2022                 if (sign == '-' || sign == '+')
2023                         token++;
2024
2025                 policy = hsm_policy_str2bit(token);
2026                 if (policy == 0) {
2027                         CWARN("%s: '%s' is unknown, "
2028                               "supported policies are:\n", mdt_obd_name(mdt),
2029                               token);
2030                         hsm_policy_bit2str(m, 0, false);
2031                         GOTO(out, rc = -EINVAL);
2032                 }
2033                 switch (sign) {
2034                 case '-':
2035                         remove_mask |= policy;
2036                         break;
2037                 case '+':
2038                         add_mask |= policy;
2039                         break;
2040                 default:
2041                         set_mask |= policy;
2042                         break;
2043                 }
2044
2045         } while (start != NULL);
2046
2047         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
2048                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
2049
2050         /* if no sign in all string, it is a clear and set
2051          * if some sign found, all unsigned are converted
2052          * to add
2053          * P1 P2 = set to P1 and P2
2054          * P1 -P2 = add P1 clear P2 same as +P1 -P2
2055          */
2056         if (remove_mask == 0 && add_mask == 0) {
2057                 cdt->cdt_policy = set_mask;
2058         } else {
2059                 cdt->cdt_policy |= set_mask | add_mask;
2060                 cdt->cdt_policy &= ~remove_mask;
2061         }
2062
2063         GOTO(out, rc = count);
2064
2065 out:
2066         OBD_FREE(buf, count + 1);
2067         RETURN(rc);
2068 }
2069 LDEBUGFS_SEQ_FOPS(mdt_hsm_policy);
2070
2071 ssize_t loop_period_show(struct kobject *kobj, struct attribute *attr,
2072                          char *buf)
2073 {
2074         struct coordinator *cdt = container_of(kobj, struct coordinator,
2075                                                cdt_hsm_kobj);
2076
2077         return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_loop_period);
2078 }
2079
2080 ssize_t loop_period_store(struct kobject *kobj, struct attribute *attr,
2081                           const char *buffer, size_t count)
2082 {
2083         struct coordinator *cdt = container_of(kobj, struct coordinator,
2084                                                cdt_hsm_kobj);
2085         unsigned int val;
2086         int rc;
2087
2088         rc = kstrtouint(buffer, 0, &val);
2089         if (rc)
2090                 return rc;
2091
2092         if (val != 0)
2093                 cdt->cdt_loop_period = val;
2094
2095         return val ? count : -EINVAL;
2096 }
2097 LUSTRE_RW_ATTR(loop_period);
2098
2099 ssize_t grace_delay_show(struct kobject *kobj, struct attribute *attr,
2100                          char *buf)
2101 {
2102         struct coordinator *cdt = container_of(kobj, struct coordinator,
2103                                                cdt_hsm_kobj);
2104
2105         return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_grace_delay);
2106 }
2107
2108 ssize_t grace_delay_store(struct kobject *kobj, struct attribute *attr,
2109                           const char *buffer, size_t count)
2110 {
2111         struct coordinator *cdt = container_of(kobj, struct coordinator,
2112                                                cdt_hsm_kobj);
2113         unsigned int val;
2114         int rc;
2115
2116         rc = kstrtouint(buffer, 0, &val);
2117         if (rc)
2118                 return rc;
2119
2120         if (val != 0)
2121                 cdt->cdt_grace_delay = val;
2122
2123         return val ? count : -EINVAL;
2124 }
2125 LUSTRE_RW_ATTR(grace_delay);
2126
2127 ssize_t active_request_timeout_show(struct kobject *kobj,
2128                                     struct attribute *attr,
2129                                     char *buf)
2130 {
2131         struct coordinator *cdt = container_of(kobj, struct coordinator,
2132                                                cdt_hsm_kobj);
2133
2134         return scnprintf(buf, PAGE_SIZE, "%d\n", cdt->cdt_active_req_timeout);
2135 }
2136
2137 ssize_t active_request_timeout_store(struct kobject *kobj,
2138                                      struct attribute *attr,
2139                                      const char *buffer, size_t count)
2140 {
2141         struct coordinator *cdt = container_of(kobj, struct coordinator,
2142                                                cdt_hsm_kobj);
2143         unsigned int val;
2144         int rc;
2145
2146         rc = kstrtouint(buffer, 0, &val);
2147         if (rc)
2148                 return rc;
2149
2150         if (val != 0)
2151                 cdt->cdt_active_req_timeout = val;
2152
2153         return val ? count : -EINVAL;
2154 }
2155 LUSTRE_RW_ATTR(active_request_timeout);
2156
2157 ssize_t max_requests_show(struct kobject *kobj, struct attribute *attr,
2158                           char *buf)
2159 {
2160         struct coordinator *cdt = container_of(kobj, struct coordinator,
2161                                                cdt_hsm_kobj);
2162
2163         return scnprintf(buf, PAGE_SIZE, "%llu\n", cdt->cdt_max_requests);
2164 }
2165
2166 ssize_t max_requests_store(struct kobject *kobj, struct attribute *attr,
2167                            const char *buffer, size_t count)
2168 {
2169         struct coordinator *cdt = container_of(kobj, struct coordinator,
2170                                                cdt_hsm_kobj);
2171         unsigned long long val;
2172         int rc;
2173
2174         rc = kstrtoull(buffer, 0, &val);
2175         if (rc)
2176                 return rc;
2177
2178         if (val != 0)
2179                 cdt->cdt_max_requests = val;
2180
2181         return val ? count : -EINVAL;
2182 }
2183 LUSTRE_RW_ATTR(max_requests);
2184
2185 ssize_t default_archive_id_show(struct kobject *kobj, struct attribute *attr,
2186                                 char *buf)
2187 {
2188         struct coordinator *cdt = container_of(kobj, struct coordinator,
2189                                                cdt_hsm_kobj);
2190
2191         return scnprintf(buf, PAGE_SIZE, "%u\n", cdt->cdt_default_archive_id);
2192 }
2193
2194 ssize_t default_archive_id_store(struct kobject *kobj, struct attribute *attr,
2195                                  const char *buffer, size_t count)
2196 {
2197         struct coordinator *cdt = container_of(kobj, struct coordinator,
2198                                                cdt_hsm_kobj);
2199         unsigned int val;
2200         int rc;
2201
2202         rc = kstrtouint(buffer, 0, &val);
2203         if (rc)
2204                 return rc;
2205
2206         if (val != 0)
2207                 cdt->cdt_default_archive_id = val;
2208
2209         return val ? count : -EINVAL;
2210 }
2211 LUSTRE_RW_ATTR(default_archive_id);
2212
2213 /*
2214  * procfs write method for MDT/hsm_control
2215  * proc entry is in mdt directory so data is mdt obd_device pointer
2216  */
2217 #define CDT_ENABLE_CMD   "enabled"
2218 #define CDT_STOP_CMD     "shutdown"
2219 #define CDT_DISABLE_CMD  "disabled"
2220 #define CDT_PURGE_CMD    "purge"
2221 #define CDT_HELP_CMD     "help"
2222 #define CDT_MAX_CMD_LEN  10
2223
2224 ssize_t hsm_control_store(struct kobject *kobj, struct attribute *attr,
2225                           const char *buffer, size_t count)
2226 {
2227         struct obd_device *obd = container_of(kobj, struct obd_device,
2228                                               obd_kset.kobj);
2229         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2230         struct coordinator *cdt = &(mdt->mdt_coordinator);
2231         int usage = 0;
2232         int rc = 0;
2233
2234         if (count == 0 || count >= CDT_MAX_CMD_LEN)
2235                 return -EINVAL;
2236
2237         if (strncmp(buffer, CDT_ENABLE_CMD, strlen(CDT_ENABLE_CMD)) == 0) {
2238                 if (cdt->cdt_state == CDT_DISABLE) {
2239                         rc = set_cdt_state(cdt, CDT_RUNNING);
2240                         mdt_hsm_cdt_event(cdt);
2241                         wake_up(&cdt->cdt_waitq);
2242                 } else if (cdt->cdt_state == CDT_RUNNING) {
2243                         rc = 0;
2244                 } else {
2245                         rc = mdt_hsm_cdt_start(mdt);
2246                 }
2247         } else if (strncmp(buffer, CDT_STOP_CMD, strlen(CDT_STOP_CMD)) == 0) {
2248                 if (cdt->cdt_state == CDT_STOPPING) {
2249                         CERROR("%s: Coordinator is already stopping\n",
2250                                mdt_obd_name(mdt));
2251                         rc = -EALREADY;
2252                 } else if (cdt->cdt_state == CDT_STOPPED) {
2253                         rc = 0;
2254                 } else {
2255                         rc = mdt_hsm_cdt_stop(mdt);
2256                 }
2257         } else if (strncmp(buffer, CDT_DISABLE_CMD,
2258                            strlen(CDT_DISABLE_CMD)) == 0) {
2259                 if ((cdt->cdt_state == CDT_STOPPING) ||
2260                     (cdt->cdt_state == CDT_STOPPED)) {
2261                         CERROR("%s: Coordinator is stopped\n",
2262                                mdt_obd_name(mdt));
2263                         rc = -EINVAL;
2264                 } else {
2265                         rc = set_cdt_state(cdt, CDT_DISABLE);
2266                 }
2267         } else if (strncmp(buffer, CDT_PURGE_CMD,
2268                            strlen(CDT_PURGE_CMD)) == 0) {
2269                 rc = hsm_cancel_all_actions(mdt);
2270         } else if (strncmp(buffer, CDT_HELP_CMD,
2271                            strlen(CDT_HELP_CMD)) == 0) {
2272                 usage = 1;
2273         } else {
2274                 usage = 1;
2275                 rc = -EINVAL;
2276         }
2277
2278         if (usage == 1)
2279                 CERROR("%s: Valid coordinator control commands are: "
2280                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2281                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2282                        CDT_PURGE_CMD, CDT_HELP_CMD);
2283
2284         if (rc)
2285                 RETURN(rc);
2286
2287         RETURN(count);
2288 }
2289
2290 ssize_t hsm_control_show(struct kobject *kobj, struct attribute *attr,
2291                          char *buf)
2292 {
2293         struct obd_device *obd = container_of(kobj, struct obd_device,
2294                                               obd_kset.kobj);
2295         struct coordinator *cdt;
2296
2297         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2298
2299         return scnprintf(buf, PAGE_SIZE, "%s\n",
2300                          cdt_mdt_state2str(cdt->cdt_state));
2301 }
2302
2303 static int
2304 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2305 {
2306         bool first = true;
2307         int i;
2308         ENTRY;
2309
2310         for (i = 0; i < 8 * sizeof(mask); i++) {
2311                 if (mask & (1UL << i)) {
2312                         seq_printf(m, "%s%s", first ? "" : " ",
2313                                    hsm_copytool_action2name(i));
2314                         first = false;
2315                 }
2316         }
2317         seq_putc(m, '\n');
2318
2319         RETURN(0);
2320 }
2321
2322 static int
2323 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2324 {
2325         struct mdt_device *mdt = m->private;
2326         struct coordinator *cdt = &mdt->mdt_coordinator;
2327
2328         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2329 }
2330
2331 static int
2332 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2333 {
2334         struct mdt_device *mdt = m->private;
2335         struct coordinator *cdt = &mdt->mdt_coordinator;
2336
2337         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2338 }
2339
2340 static int
2341 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2342 {
2343         struct mdt_device *mdt = m->private;
2344         struct coordinator *cdt = &mdt->mdt_coordinator;
2345
2346         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2347 }
2348
2349 static inline enum hsm_copytool_action
2350 hsm_copytool_name2action(const char *name)
2351 {
2352         if (strcasecmp(name, "NOOP") == 0)
2353                 return HSMA_NONE;
2354         else if (strcasecmp(name, "ARCHIVE") == 0)
2355                 return HSMA_ARCHIVE;
2356         else if (strcasecmp(name, "RESTORE") == 0)
2357                 return HSMA_RESTORE;
2358         else if (strcasecmp(name, "REMOVE") == 0)
2359                 return HSMA_REMOVE;
2360         else if (strcasecmp(name, "CANCEL") == 0)
2361                 return HSMA_CANCEL;
2362         else
2363                 return -1;
2364 }
2365
2366 static ssize_t
2367 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2368                             size_t user_count, __u64 *mask)
2369 {
2370         char *buf, *pos, *name;
2371         size_t buf_size;
2372         __u64 new_mask = 0;
2373         int rc;
2374         ENTRY;
2375
2376         if (!(user_count < 4096))
2377                 RETURN(-ENOMEM);
2378
2379         buf_size = user_count + 1;
2380
2381         OBD_ALLOC(buf, buf_size);
2382         if (buf == NULL)
2383                 RETURN(-ENOMEM);
2384
2385         if (copy_from_user(buf, user_buf, buf_size - 1))
2386                 GOTO(out, rc = -EFAULT);
2387
2388         buf[buf_size - 1] = '\0';
2389
2390         pos = buf;
2391         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2392                 int action;
2393
2394                 if (*name == '\0')
2395                         continue;
2396
2397                 action = hsm_copytool_name2action(name);
2398                 if (action < 0)
2399                         GOTO(out, rc = -EINVAL);
2400
2401                 new_mask |= (1UL << action);
2402         }
2403
2404         *mask = new_mask;
2405         rc = user_count;
2406 out:
2407         OBD_FREE(buf, buf_size);
2408
2409         RETURN(rc);
2410 }
2411
2412 static ssize_t
2413 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2414                                         size_t count, loff_t *off)
2415 {
2416         struct seq_file         *m = file->private_data;
2417         struct mdt_device       *mdt = m->private;
2418         struct coordinator *cdt = &mdt->mdt_coordinator;
2419
2420         return mdt_write_hsm_request_mask(file, buf, count,
2421                                            &cdt->cdt_user_request_mask);
2422 }
2423
2424 static ssize_t
2425 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2426                                         size_t count, loff_t *off)
2427 {
2428         struct seq_file         *m = file->private_data;
2429         struct mdt_device       *mdt = m->private;
2430         struct coordinator      *cdt = &mdt->mdt_coordinator;
2431
2432         return mdt_write_hsm_request_mask(file, buf, count,
2433                                            &cdt->cdt_group_request_mask);
2434 }
2435
2436 static ssize_t
2437 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2438                                         size_t count, loff_t *off)
2439 {
2440         struct seq_file         *m = file->private_data;
2441         struct mdt_device       *mdt = m->private;
2442         struct coordinator      *cdt = &mdt->mdt_coordinator;
2443
2444         return mdt_write_hsm_request_mask(file, buf, count,
2445                                            &cdt->cdt_other_request_mask);
2446 }
2447
2448 static ssize_t remove_archive_on_last_unlink_show(struct kobject *kobj,
2449                                                   struct attribute *attr,
2450                                                   char *buf)
2451 {
2452         struct coordinator *cdt = container_of(kobj, struct coordinator,
2453                                                cdt_hsm_kobj);
2454
2455         return scnprintf(buf, PAGE_SIZE, "%u\n",
2456                          cdt->cdt_remove_archive_on_last_unlink);
2457 }
2458
2459 static ssize_t remove_archive_on_last_unlink_store(struct kobject *kobj,
2460                                                    struct attribute *attr,
2461                                                    const char *buffer,
2462                                                    size_t count)
2463 {
2464         struct coordinator *cdt = container_of(kobj, struct coordinator,
2465                                                cdt_hsm_kobj);
2466         bool val;
2467         int rc;
2468
2469         rc = kstrtobool(buffer, &val);
2470         if (rc < 0)
2471                 return rc;
2472
2473         cdt->cdt_remove_archive_on_last_unlink = val;
2474         return count;
2475 }
2476 LUSTRE_RW_ATTR(remove_archive_on_last_unlink);
2477
2478 LDEBUGFS_SEQ_FOPS(mdt_hsm_user_request_mask);
2479 LDEBUGFS_SEQ_FOPS(mdt_hsm_group_request_mask);
2480 LDEBUGFS_SEQ_FOPS(mdt_hsm_other_request_mask);
2481
2482 /* Read-only sysfs files for request counters */
2483 static ssize_t archive_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_archive_count));
2491 }
2492 LUSTRE_RO_ATTR(archive_count);
2493
2494 static ssize_t restore_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_restore_count));
2502 }
2503 LUSTRE_RO_ATTR(restore_count);
2504
2505 static ssize_t remove_count_show(struct kobject *kobj, struct attribute *attr,
2506                                  char *buf)
2507 {
2508         struct coordinator *cdt = container_of(kobj, struct coordinator,
2509                                                cdt_hsm_kobj);
2510
2511         return scnprintf(buf, PAGE_SIZE, "%d\n",
2512                          atomic_read(&cdt->cdt_remove_count));
2513 }
2514 LUSTRE_RO_ATTR(remove_count);
2515
2516 static struct ldebugfs_vars ldebugfs_mdt_hsm_vars[] = {
2517         { .name =       "agents",
2518           .fops =       &mdt_hsm_agent_fops                     },
2519         { .name =       "actions",
2520           .fops =       &mdt_hsm_actions_fops,
2521           .proc_mode =  0444                                    },
2522         { .name =       "policy",
2523           .fops =       &mdt_hsm_policy_fops                    },
2524         { .name =       "active_requests",
2525           .fops =       &mdt_hsm_active_requests_fops           },
2526         { .name =       "user_request_mask",
2527           .fops =       &mdt_hsm_user_request_mask_fops,        },
2528         { .name =       "group_request_mask",
2529           .fops =       &mdt_hsm_group_request_mask_fops,       },
2530         { .name =       "other_request_mask",
2531           .fops =       &mdt_hsm_other_request_mask_fops,       },
2532         { 0 }
2533 };
2534
2535 static struct attribute *hsm_attrs[] = {
2536         &lustre_attr_loop_period.attr,
2537         &lustre_attr_grace_delay.attr,
2538         &lustre_attr_active_request_timeout.attr,
2539         &lustre_attr_max_requests.attr,
2540         &lustre_attr_default_archive_id.attr,
2541         &lustre_attr_remove_archive_on_last_unlink.attr,
2542         &lustre_attr_archive_count.attr,
2543         &lustre_attr_restore_count.attr,
2544         &lustre_attr_remove_count.attr,
2545         NULL,
2546 };
2547
2548 static void hsm_kobj_release(struct kobject *kobj)
2549 {
2550         struct coordinator *cdt = container_of(kobj, struct coordinator,
2551                                                cdt_hsm_kobj);
2552
2553         debugfs_remove_recursive(cdt->cdt_debugfs_dir);
2554         cdt->cdt_debugfs_dir = NULL;
2555
2556         complete(&cdt->cdt_kobj_unregister);
2557 }
2558
2559 static struct kobj_type hsm_ktype = {
2560         .default_attrs  = hsm_attrs,
2561         .sysfs_ops      = &lustre_sysfs_ops,
2562         .release        = hsm_kobj_release,
2563 };
2564
2565 /**
2566  * create sysfs entries for coordinator
2567  * \param mdt [IN]
2568  * \retval 0 success
2569  * \retval -ve failure
2570  */
2571 int hsm_cdt_tunables_init(struct mdt_device *mdt)
2572 {
2573         struct coordinator *cdt = &mdt->mdt_coordinator;
2574         struct obd_device *obd = mdt2obd_dev(mdt);
2575         int rc;
2576
2577         init_completion(&cdt->cdt_kobj_unregister);
2578         rc = kobject_init_and_add(&cdt->cdt_hsm_kobj, &hsm_ktype,
2579                                   &obd->obd_kset.kobj, "%s", "hsm");
2580         if (rc) {
2581                 kobject_put(&cdt->cdt_hsm_kobj);
2582                 return rc;
2583         }
2584
2585         /* init debugfs entries, failure is not critical */
2586         cdt->cdt_debugfs_dir = debugfs_create_dir("hsm",
2587                                                   obd->obd_debugfs_entry);
2588         ldebugfs_add_vars(cdt->cdt_debugfs_dir, ldebugfs_mdt_hsm_vars, mdt);
2589
2590         return 0;
2591 }
2592
2593 /**
2594  * remove sysfs entries for coordinator
2595  *
2596  * @mdt
2597  */
2598 void hsm_cdt_tunables_fini(struct mdt_device *mdt)
2599 {
2600         struct coordinator *cdt = &mdt->mdt_coordinator;
2601
2602         kobject_put(&cdt->cdt_hsm_kobj);
2603         wait_for_completion(&cdt->cdt_kobj_unregister);
2604 }