lustre/mdt/mdt_coordinator.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
  24  *                          alternatives
  25  *
  26  * Copyright (c) 2013, 2017, Intel Corporation.
  27  * Use is subject to license terms.
  28  */
  29 /*
  30  * lustre/mdt/mdt_coordinator.c
  31  *
  32  * Lustre HSM Coordinator
  33  *
  34  * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
  35  * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
  36  * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <linux/kthread.h>
  42 #include <obd_support.h>
  43 #include <lustre_export.h>
  44 #include <obd.h>
  45 #include <lprocfs_status.h>
  46 #include <lustre_log.h>
  47 #include <lustre_kernelcomm.h>
  48 #include "mdt_internal.h"
  49
  50 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
  51
  52 /**
  53  * get obj and HSM attributes on a fid
  54  * \param mti [IN] context
  55  * \param fid [IN] object fid
  56  * \param hsm [OUT] HSM meta data
  57  * \retval obj or error (-ENOENT if not found)
  58  */
  59 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  60                                       const struct lu_fid *fid,
  61                                       struct md_hsm *hsm)
  62 {
  63         struct md_attr          *ma;
  64         struct mdt_object       *obj;
  65         int                      rc;
  66         ENTRY;
  67
  68         ma = &mti->mti_attr;
  69         ma->ma_need = MA_HSM;
  70         ma->ma_valid = 0;
  71
  72         /* find object by FID */
  73         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  74         if (IS_ERR(obj))
  75                 RETURN(obj);
  76
  77         if (!mdt_object_exists(obj)) {
  78                 /* no more object */
  79                 mdt_object_put(mti->mti_env, obj);
  80                 RETURN(ERR_PTR(-ENOENT));
  81         }
  82
  83         rc = mdt_attr_get_complex(mti, obj, ma);
  84         if (rc) {
  85                 mdt_object_put(mti->mti_env, obj);
  86                 RETURN(ERR_PTR(rc));
  87         }
  88
  89         if (ma->ma_valid & MA_HSM)
  90                 *hsm = ma->ma_hsm;
  91         else
  92                 memset(hsm, 0, sizeof(*hsm));
  93         ma->ma_valid = 0;
  94         RETURN(obj);
  95 }
  96
  97 void mdt_hsm_dump_hal(int level, const char *prefix,
  98                       struct hsm_action_list *hal)
  99 {
 100         int                      i, sz;
 101         struct hsm_action_item  *hai;
 102         char                     buf[12];
 103
 104         CDEBUG(level, "%s: HAL header: version %X count %d"
 105                       " archive_id %d flags %#llx\n",
 106                prefix, hal->hal_version, hal->hal_count,
 107                hal->hal_archive_id, hal->hal_flags);
 108
 109         hai = hai_first(hal);
 110         for (i = 0; i < hal->hal_count; i++) {
 111                 sz = hai->hai_len - sizeof(*hai);
 112                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 113                        " cookie=%#llx"
 114                        " action=%s extent=%#llx-%#llx gid=%#llx"
 115                        " datalen=%d data=[%s]\n",
 116                        prefix, i,
 117                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 118                        hai->hai_cookie,
 119                        hsm_copytool_action2name(hai->hai_action),
 120                        hai->hai_extent.offset,
 121                        hai->hai_extent.length,
 122                        hai->hai_gid, sz,
 123                        hai_dump_data_field(hai, buf, sizeof(buf)));
 124                 hai = hai_next(hai);
 125         }
 126 }
 127
 128 /**
 129  * data passed to llog_cat_process() callback
 130  * to scan requests and take actions
 131  */
 132 struct hsm_scan_request {
 133         int                      hal_sz;
 134         int                      hal_used_sz;
 135         struct hsm_action_list  *hal;
 136 };
 137
 138 struct hsm_scan_data {
 139         struct mdt_thread_info  *hsd_mti;
 140         char                     hsd_fsname[MTI_NAME_MAXLEN + 1];
 141         /* are we scanning the logs for housekeeping, or just looking
 142          * for new work?
 143          */
 144         bool                     hsd_housekeeping;
 145         bool                     hsd_one_restore;
 146         int                      hsd_action_count;
 147         int                      hsd_request_len; /* array alloc len */
 148         int                      hsd_request_count; /* array used count */
 149         struct hsm_scan_request *hsd_request;
 150 };
 151
 152 static int mdt_cdt_waiting_cb(const struct lu_env *env,
 153                               struct mdt_device *mdt,
 154                               struct llog_handle *llh,
 155                               struct llog_agent_req_rec *larr,
 156                               struct hsm_scan_data *hsd)
 157 {
 158         struct coordinator *cdt = &mdt->mdt_coordinator;
 159         struct hsm_scan_request *request;
 160         struct hsm_action_item *hai;
 161         size_t hai_size;
 162         u32 archive_id;
 163         int i;
 164
 165         /* Are agents full? */
 166         if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
 167             cdt->cdt_max_requests) {
 168                 /* We cannot send any more request
 169                  *
 170                  *                     *** SPECIAL CASE ***
 171                  *
 172                  * Restore requests are too important not to schedule at least
 173                  * one, everytime we can.
 174                  */
 175                 if (larr->arr_hai.hai_action != HSMA_RESTORE ||
 176                     hsd->hsd_one_restore)
 177                         RETURN(hsd->hsd_housekeeping ? 0 : LLOG_PROC_BREAK);
 178         }
 179
 180         hai_size = cfs_size_round(larr->arr_hai.hai_len);
 181         archive_id = larr->arr_archive_id;
 182
 183         /* Can we add this action to one of the existing HALs in hsd. */
 184         request = NULL;
 185         for (i = 0; i < hsd->hsd_request_count; i++) {
 186                 if (hsd->hsd_request[i].hal->hal_archive_id == archive_id &&
 187                     hsd->hsd_request[i].hal_used_sz + hai_size <=
 188                     LDLM_MAXREQSIZE) {
 189                         request = &hsd->hsd_request[i];
 190                         break;
 191                 }
 192         }
 193
 194         /* Are we trying to force-schedule a request? */
 195         if (hsd->hsd_action_count + atomic_read(&cdt->cdt_request_count) >=
 196             cdt->cdt_max_requests) {
 197                 /* Is there really no compatible hsm_scan_request? */
 198                 if (!request) {
 199                         for (i -= 1; i >= 0; i--) {
 200                                 if (hsd->hsd_request[i].hal->hal_archive_id ==
 201                                     archive_id) {
 202                                         request = &hsd->hsd_request[i];
 203                                         break;
 204                                 }
 205                         }
 206                 }
 207
 208                 /* Make room for the hai */
 209                 if (request) {
 210                         /* Discard the last hai until there is enough space */
 211                         do {
 212                                 request->hal->hal_count--;
 213
 214                                 hai = hai_first(request->hal);
 215                                 for (i = 0; i < request->hal->hal_count; i++)
 216                                         hai = hai_next(hai);
 217                                 request->hal_used_sz -=
 218                                         cfs_size_round(hai->hai_len);
 219                                 hsd->hsd_action_count--;
 220                         } while (request->hal_used_sz + hai_size >
 221                                  LDLM_MAXREQSIZE);
 222                 } else if (hsd->hsd_housekeeping) {
 223                         struct hsm_scan_request *tmp;
 224
 225                         /* Discard the (whole) last hal */
 226                         hsd->hsd_request_count--;
 227                         tmp = &hsd->hsd_request[hsd->hsd_request_count];
 228                         hsd->hsd_action_count -= tmp->hal->hal_count;
 229                         OBD_FREE(tmp->hal, tmp->hal_sz);
 230                 } else {
 231                         /* Bailing out, this code path is too hot */
 232                         RETURN(LLOG_PROC_BREAK);
 233
 234                 }
 235         }
 236
 237         if (!request) {
 238                 struct hsm_action_list *hal;
 239
 240                 LASSERT(hsd->hsd_request_count < hsd->hsd_request_len);
 241                 request = &hsd->hsd_request[hsd->hsd_request_count];
 242
 243                 /* allocates hai vector size just needs to be large
 244                  * enough */
 245                 request->hal_sz = sizeof(*request->hal) +
 246                         cfs_size_round(MTI_NAME_MAXLEN + 1) + 2 * hai_size;
 247                 OBD_ALLOC_LARGE(hal, request->hal_sz);
 248                 if (!hal)
 249                         RETURN(-ENOMEM);
 250
 251                 hal->hal_version = HAL_VERSION;
 252                 strlcpy(hal->hal_fsname, hsd->hsd_fsname, MTI_NAME_MAXLEN + 1);
 253                 hal->hal_archive_id = larr->arr_archive_id;
 254                 hal->hal_flags = larr->arr_flags;
 255                 hal->hal_count = 0;
 256                 request->hal_used_sz = hal_size(hal);
 257                 request->hal = hal;
 258                 hsd->hsd_request_count++;
 259         } else if (request->hal_sz < request->hal_used_sz + hai_size) {
 260                 /* Not enough room, need an extension */
 261                 void *hal_buffer;
 262                 int sz;
 263
 264                 sz = min_t(int, 2 * request->hal_sz, LDLM_MAXREQSIZE);
 265                 LASSERT(request->hal_used_sz + hai_size < sz);
 266
 267                 OBD_ALLOC_LARGE(hal_buffer, sz);
 268                 if (!hal_buffer)
 269                         RETURN(-ENOMEM);
 270
 271                 memcpy(hal_buffer, request->hal, request->hal_used_sz);
 272                 OBD_FREE_LARGE(request->hal, request->hal_sz);
 273                 request->hal = hal_buffer;
 274                 request->hal_sz = sz;
 275         }
 276
 277         hai = hai_first(request->hal);
 278         for (i = 0; i < request->hal->hal_count; i++)
 279                 hai = hai_next(hai);
 280
 281         memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 282
 283         request->hal_used_sz += hai_size;
 284         request->hal->hal_count++;
 285
 286         hsd->hsd_action_count++;
 287
 288         switch (hai->hai_action) {
 289         case HSMA_CANCEL:
 290                 break;
 291         case HSMA_RESTORE:
 292                 hsd->hsd_one_restore = true;
 293                 /* Intentional fallthrough */
 294         default:
 295                 cdt_agent_record_hash_add(cdt, hai->hai_cookie,
 296                                           llh->lgh_hdr->llh_cat_idx,
 297                                           larr->arr_hdr.lrh_index);
 298         }
 299
 300         RETURN(0);
 301 }
 302
 303 static int mdt_cdt_started_cb(const struct lu_env *env,
 304                               struct mdt_device *mdt,
 305                               struct llog_handle *llh,
 306                               struct llog_agent_req_rec *larr,
 307                               struct hsm_scan_data *hsd)
 308 {
 309         struct coordinator *cdt = &mdt->mdt_coordinator;
 310         struct hsm_action_item *hai = &larr->arr_hai;
 311         struct cdt_agent_req *car;
 312         time64_t now = ktime_get_real_seconds();
 313         time64_t last;
 314         int cl_flags;
 315         int rc;
 316
 317         if (!hsd->hsd_housekeeping)
 318                 RETURN(0);
 319
 320         /* we search for a running request
 321          * error may happen if coordinator crashes or stopped
 322          * with running request
 323          */
 324         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
 325         if (car == NULL) {
 326                 last = larr->arr_req_change;
 327         } else {
 328                 last = car->car_req_update;
 329         }
 330
 331         /* test if request too long, if yes cancel it
 332          * the same way the copy tool acknowledge a cancel request */
 333         if (now <= last + cdt->cdt_active_req_timeout)
 334                 GOTO(out_car, rc = 0);
 335
 336         dump_llog_agent_req_rec("request timed out, start cleaning", larr);
 337
 338         if (car != NULL) {
 339                 car->car_req_update = now;
 340                 mdt_hsm_agent_update_statistics(cdt, 0, 1, 0, &car->car_uuid);
 341                 /* Remove car from memory list (LU-9075) */
 342                 mdt_cdt_remove_request(cdt, hai->hai_cookie);
 343         }
 344
 345         /* Emit a changelog record for the failed action.*/
 346         cl_flags = 0;
 347         hsm_set_cl_error(&cl_flags, ECANCELED);
 348
 349         switch (hai->hai_action) {
 350         case HSMA_ARCHIVE:
 351                 hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
 352                 break;
 353         case HSMA_RESTORE:
 354                 hsm_set_cl_event(&cl_flags, HE_RESTORE);
 355                 break;
 356         case HSMA_REMOVE:
 357                 hsm_set_cl_event(&cl_flags, HE_REMOVE);
 358                 break;
 359         case HSMA_CANCEL:
 360                 hsm_set_cl_event(&cl_flags, HE_CANCEL);
 361                 break;
 362         default:
 363                 /* Unknown record type, skip changelog. */
 364                 cl_flags = 0;
 365                 break;
 366         }
 367
 368         if (cl_flags != 0)
 369                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
 370                              &hai->hai_fid);
 371
 372         if (hai->hai_action == HSMA_RESTORE)
 373                 cdt_restore_handle_del(hsd->hsd_mti, cdt, &hai->hai_fid);
 374
 375         larr->arr_status = ARS_CANCELED;
 376         larr->arr_req_change = now;
 377         rc = llog_write(hsd->hsd_mti->mti_env, llh, &larr->arr_hdr,
 378                         larr->arr_hdr.lrh_index);
 379         if (rc < 0) {
 380                 CERROR("%s: cannot update agent log: rc = %d\n",
 381                        mdt_obd_name(mdt), rc);
 382                 rc = LLOG_DEL_RECORD;
 383         }
 384
 385         /* ct has completed a request, so a slot is available,
 386          * signal the coordinator to find new work */
 387         mdt_hsm_cdt_event(cdt);
 388 out_car:
 389         if (car != NULL)
 390                 mdt_cdt_put_request(car);
 391
 392         RETURN(rc);
 393 }
 394
 395 /**
 396  *  llog_cat_process() callback, used to:
 397  *  - find waiting request and start action
 398  *  - purge canceled and done requests
 399  * \param env [IN] environment
 400  * \param llh [IN] llog handle
 401  * \param hdr [IN] llog record
 402  * \param data [IN/OUT] cb data = struct hsm_scan_data
 403  * \retval 0 success
 404  * \retval -ve failure
 405  */
 406 static int mdt_coordinator_cb(const struct lu_env *env,
 407                               struct llog_handle *llh,
 408                               struct llog_rec_hdr *hdr,
 409                               void *data)
 410 {
 411         struct llog_agent_req_rec *larr = (struct llog_agent_req_rec *)hdr;
 412         struct hsm_scan_data *hsd = data;
 413         struct mdt_device *mdt = hsd->hsd_mti->mti_mdt;
 414         struct coordinator *cdt = &mdt->mdt_coordinator;
 415         ENTRY;
 416
 417         larr = (struct llog_agent_req_rec *)hdr;
 418         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 419         switch (larr->arr_status) {
 420         case ARS_WAITING:
 421                 RETURN(mdt_cdt_waiting_cb(env, mdt, llh, larr, hsd));
 422         case ARS_STARTED:
 423                 RETURN(mdt_cdt_started_cb(env, mdt, llh, larr, hsd));
 424         default:
 425                 if (!hsd->hsd_housekeeping)
 426                         RETURN(0);
 427
 428                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 429                     ktime_get_real_seconds()) {
 430                         cdt_agent_record_hash_del(cdt,
 431                                                   larr->arr_hai.hai_cookie);
 432                         RETURN(LLOG_DEL_RECORD);
 433                 }
 434
 435                 RETURN(0);
 436         }
 437 }
 438
 439 /**
 440  * create /proc entries for coordinator
 441  * \param mdt [IN]
 442  * \retval 0 success
 443  * \retval -ve failure
 444  */
 445 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 446 {
 447         struct coordinator      *cdt = &mdt->mdt_coordinator;
 448         int                      rc = 0;
 449         ENTRY;
 450
 451         /* init /proc entries, failure is not critical */
 452         cdt->cdt_proc_dir = lprocfs_register("hsm",
 453                                              mdt2obd_dev(mdt)->obd_proc_entry,
 454                                              lprocfs_mdt_hsm_vars, mdt);
 455         if (IS_ERR(cdt->cdt_proc_dir)) {
 456                 rc = PTR_ERR(cdt->cdt_proc_dir);
 457                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 458                        " rc=%d\n", mdt_obd_name(mdt), rc);
 459                 cdt->cdt_proc_dir = NULL;
 460                 RETURN(rc);
 461         }
 462
 463         RETURN(0);
 464 }
 465
 466 /**
 467  * remove /proc entries for coordinator
 468  * \param mdt [IN]
 469  */
 470 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
 471 {
 472         struct coordinator *cdt = &mdt->mdt_coordinator;
 473
 474         if (cdt->cdt_proc_dir != NULL)
 475                 lprocfs_remove(&cdt->cdt_proc_dir);
 476 }
 477
 478 /**
 479  * get vector of hsm cdt /proc vars
 480  * \param none
 481  * \retval var vector
 482  */
 483 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 484 {
 485         return lprocfs_mdt_hsm_vars;
 486 }
 487
 488 /* Release the ressource used by the coordinator. Called when the
 489  * coordinator is stopping. */
 490 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
 491 {
 492         struct coordinator              *cdt = &mdt->mdt_coordinator;
 493         struct cdt_agent_req            *car, *tmp1;
 494         struct hsm_agent                *ha, *tmp2;
 495         struct cdt_restore_handle       *crh, *tmp3;
 496         struct mdt_thread_info          *cdt_mti;
 497
 498         /* start cleaning */
 499         down_write(&cdt->cdt_request_lock);
 500         list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
 501                                  car_request_list) {
 502                 cfs_hash_del(cdt->cdt_request_cookie_hash,
 503                              &car->car_hai->hai_cookie,
 504                              &car->car_cookie_hash);
 505                 list_del(&car->car_request_list);
 506                 mdt_cdt_put_request(car);
 507         }
 508         up_write(&cdt->cdt_request_lock);
 509
 510         down_write(&cdt->cdt_agent_lock);
 511         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 512                 list_del(&ha->ha_list);
 513                 if (ha->ha_archive_cnt != 0)
 514                         OBD_FREE(ha->ha_archive_id, ha->ha_archive_cnt *
 515                                  sizeof(*ha->ha_archive_id));
 516                 OBD_FREE_PTR(ha);
 517         }
 518         up_write(&cdt->cdt_agent_lock);
 519
 520         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 521         mutex_lock(&cdt->cdt_restore_lock);
 522         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_handle_list,
 523                                  crh_list) {
 524                 list_del(&crh->crh_list);
 525                 /* give back layout lock */
 526                 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
 527                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 528         }
 529         mutex_unlock(&cdt->cdt_restore_lock);
 530 }
 531
 532 /*
 533  * Coordinator state transition table, indexed on enum cdt_states, taking
 534  * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
 535  * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
 536  */
 537 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
 538         /* from -> to:    stopped init   running disable stopping */
 539         /* stopped */   { true,   true,  false,  false,  false },
 540         /* init */      { true,   false, true,   false,  false },
 541         /* running */   { false,  false, true,   true,   true },
 542         /* disable */   { false,  false, true,   true,   true },
 543         /* stopping */  { true,   false, false,  false,  false }
 544 };
 545
 546 /**
 547  * Change coordinator thread state
 548  * Some combinations are not valid, so catch them here.
 549  *
 550  * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
 551  * the transition was not possible.
 552  */
 553 static int set_cdt_state_locked(struct coordinator *cdt,
 554                                 enum cdt_states new_state)
 555 {
 556         int rc;
 557         enum cdt_states state;
 558
 559         state = cdt->cdt_state;
 560
 561         if (cdt_transition[state][new_state]) {
 562                 cdt->cdt_state = new_state;
 563                 rc = 0;
 564         } else {
 565                 CDEBUG(D_HSM,
 566                        "unexpected coordinator transition, from=%s, to=%s\n",
 567                        cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
 568                 rc = -EINVAL;
 569         }
 570
 571         return rc;
 572 }
 573
 574 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state)
 575 {
 576         int rc;
 577
 578         mutex_lock(&cdt->cdt_state_lock);
 579         rc = set_cdt_state_locked(cdt, new_state);
 580         mutex_unlock(&cdt->cdt_state_lock);
 581
 582         return rc;
 583 }
 584
 585
 586
 587 /**
 588  * coordinator thread
 589  * \param data [IN] obd device
 590  * \retval 0 success
 591  * \retval -ve failure
 592  */
 593 static int mdt_coordinator(void *data)
 594 {
 595         struct mdt_thread_info  *mti = data;
 596         struct mdt_device       *mdt = mti->mti_mdt;
 597         struct coordinator      *cdt = &mdt->mdt_coordinator;
 598         struct hsm_scan_data     hsd = { NULL };
 599         time64_t                 last_housekeeping = 0;
 600         size_t request_sz = 0;
 601         int rc;
 602         ENTRY;
 603
 604         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 605                mdt_obd_name(mdt), current_pid());
 606
 607         hsd.hsd_mti = mti;
 608         obd_uuid2fsname(hsd.hsd_fsname, mdt_obd_name(mdt),
 609                         sizeof(hsd.hsd_fsname));
 610
 611         set_cdt_state(cdt, CDT_RUNNING);
 612
 613         /* Inform mdt_hsm_cdt_start(). */
 614         wake_up_all(&cdt->cdt_waitq);
 615
 616         while (1) {
 617                 int i;
 618                 int update_idx = 0;
 619                 int updates_sz;
 620                 int updates_cnt;
 621                 struct hsm_record_update *updates;
 622
 623                 /* Limit execution of the expensive requests traversal
 624                  * to at most one second. This prevents repeatedly
 625                  * locking/unlocking the catalog for each request
 626                  * and preventing other HSM operations from happening
 627                  */
 628                 wait_event_interruptible_timeout(cdt->cdt_waitq,
 629                                                  kthread_should_stop() ||
 630                                                  cdt->cdt_wakeup_coordinator,
 631                                                  cfs_time_seconds(1));
 632
 633                 cdt->cdt_wakeup_coordinator = false;
 634                 CDEBUG(D_HSM, "coordinator resumes\n");
 635
 636                 if (kthread_should_stop()) {
 637                         CDEBUG(D_HSM, "Coordinator stops\n");
 638                         rc = 0;
 639                         break;
 640                 }
 641
 642                 /* if coordinator is suspended continue to wait */
 643                 if (cdt->cdt_state == CDT_DISABLE) {
 644                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 645                         continue;
 646                 }
 647
 648                 /* If no event, and no housekeeping to do, continue to
 649                  * wait. */
 650                 if (last_housekeeping + cdt->cdt_loop_period <=
 651                     ktime_get_real_seconds()) {
 652                         last_housekeeping = ktime_get_real_seconds();
 653                         hsd.hsd_housekeeping = true;
 654                 } else if (cdt->cdt_event) {
 655                         hsd.hsd_housekeeping = false;
 656                 } else {
 657                         continue;
 658                 }
 659
 660                 cdt->cdt_event = false;
 661
 662                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 663
 664                 if (hsd.hsd_request_len != cdt->cdt_max_requests) {
 665                         /* cdt_max_requests has changed,
 666                          * we need to allocate a new buffer
 667                          */
 668                         struct hsm_scan_request *tmp = NULL;
 669                         int max_requests = cdt->cdt_max_requests;
 670                         OBD_ALLOC_LARGE(tmp, max_requests *
 671                                         sizeof(struct hsm_scan_request));
 672                         if (!tmp) {
 673                                 CERROR("Failed to resize request buffer, "
 674                                        "keeping it at %d\n",
 675                                        hsd.hsd_request_len);
 676                         } else {
 677                                 if (hsd.hsd_request != NULL)
 678                                         OBD_FREE_LARGE(hsd.hsd_request,
 679                                                        request_sz);
 680
 681                                 hsd.hsd_request_len = max_requests;
 682                                 request_sz = hsd.hsd_request_len *
 683                                         sizeof(struct hsm_scan_request);
 684                                 hsd.hsd_request = tmp;
 685                         }
 686                 }
 687
 688                 hsd.hsd_action_count = 0;
 689                 hsd.hsd_request_count = 0;
 690                 hsd.hsd_one_restore = false;
 691
 692                 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
 693                                       &hsd, 0, 0, WRITE);
 694                 if (rc < 0)
 695                         goto clean_cb_alloc;
 696
 697                 CDEBUG(D_HSM, "found %d requests to send\n",
 698                        hsd.hsd_request_count);
 699
 700                 if (list_empty(&cdt->cdt_agents)) {
 701                         CDEBUG(D_HSM, "no agent available, "
 702                                       "coordinator sleeps\n");
 703                         goto clean_cb_alloc;
 704                 }
 705
 706                 /* Compute how many HAI we have in all the requests */
 707                 updates_cnt = 0;
 708                 for (i = 0; i < hsd.hsd_request_count; i++) {
 709                         const struct hsm_scan_request *request =
 710                                 &hsd.hsd_request[i];
 711
 712                         updates_cnt += request->hal->hal_count;
 713                 }
 714
 715                 /* Allocate a temporary array to store the cookies to
 716                  * update, and their status. */
 717                 updates_sz = updates_cnt * sizeof(*updates);
 718                 OBD_ALLOC_LARGE(updates, updates_sz);
 719                 if (updates == NULL) {
 720                         CERROR("%s: Cannot allocate memory (%d o) "
 721                                "for %d updates\n",
 722                                mdt_obd_name(mdt), updates_sz, updates_cnt);
 723                         continue;
 724                 }
 725
 726                 /* here hsd contains a list of requests to be started */
 727                 for (i = 0; i < hsd.hsd_request_count; i++) {
 728                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 729                         struct hsm_action_list  *hal = request->hal;
 730                         struct hsm_action_item  *hai;
 731                         int                      j;
 732
 733                         /* still room for work ? */
 734                         if (atomic_read(&cdt->cdt_request_count) >=
 735                             cdt->cdt_max_requests)
 736                                 break;
 737
 738                         rc = mdt_hsm_agent_send(mti, hal, 0);
 739                         /* if failure, we suppose it is temporary
 740                          * if the copy tool failed to do the request
 741                          * it has to use hsm_progress
 742                          */
 743
 744                         /* set up cookie vector to set records status
 745                          * after copy tools start or failed
 746                          */
 747                         hai = hai_first(hal);
 748                         for (j = 0; j < hal->hal_count; j++) {
 749                                 updates[update_idx].cookie = hai->hai_cookie;
 750                                 updates[update_idx].status =
 751                                         (rc ? ARS_WAITING : ARS_STARTED);
 752                                 hai = hai_next(hai);
 753                                 update_idx++;
 754                         }
 755                 }
 756
 757                 if (update_idx) {
 758                         rc = mdt_agent_record_update(mti->mti_env, mdt,
 759                                                      updates, update_idx);
 760                         if (rc)
 761                                 CERROR("%s: mdt_agent_record_update() failed, "
 762                                        "rc=%d, cannot update records "
 763                                        "for %d cookies\n",
 764                                        mdt_obd_name(mdt), rc, update_idx);
 765                 }
 766
 767                 OBD_FREE_LARGE(updates, updates_sz);
 768
 769 clean_cb_alloc:
 770                 /* free hal allocated by callback */
 771                 for (i = 0; i < hsd.hsd_request_count; i++) {
 772                         struct hsm_scan_request *request = &hsd.hsd_request[i];
 773
 774                         OBD_FREE_LARGE(request->hal, request->hal_sz);
 775                 }
 776         }
 777
 778         if (hsd.hsd_request != NULL)
 779                 OBD_FREE_LARGE(hsd.hsd_request, request_sz);
 780
 781         mdt_hsm_cdt_cleanup(mdt);
 782
 783         if (rc != 0)
 784                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 785                        mdt_obd_name(mdt), current_pid(), rc);
 786         else
 787                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 788                               " no error\n",
 789                        mdt_obd_name(mdt), current_pid());
 790
 791         RETURN(rc);
 792 }
 793
 794 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
 795                            const struct lu_fid *fid,
 796                            const struct hsm_extent *he)
 797 {
 798         struct cdt_restore_handle *crh;
 799         struct mdt_object *obj;
 800         int rc;
 801         ENTRY;
 802
 803         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 804         if (crh == NULL)
 805                 RETURN(-ENOMEM);
 806
 807         crh->crh_fid = *fid;
 808         /* in V1 all file is restored
 809          * crh->extent.start = he->offset;
 810          * crh->extent.end = he->offset + he->length;
 811          */
 812         crh->crh_extent.start = 0;
 813         crh->crh_extent.end = he->length;
 814         /* get the layout lock */
 815         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 816         obj = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 817                                    MDS_INODELOCK_LAYOUT);
 818         if (IS_ERR(obj))
 819                 GOTO(out_crh, rc = PTR_ERR(obj));
 820
 821         /* We do not keep a reference on the object during the restore
 822          * which can be very long. */
 823         mdt_object_put(mti->mti_env, obj);
 824
 825         mutex_lock(&cdt->cdt_restore_lock);
 826         if (unlikely(cdt->cdt_state == CDT_STOPPED ||
 827                      cdt->cdt_state == CDT_STOPPING)) {
 828                 mutex_unlock(&cdt->cdt_restore_lock);
 829                 GOTO(out_lh, rc = -EAGAIN);
 830         }
 831
 832         list_add_tail(&crh->crh_list, &cdt->cdt_restore_handle_list);
 833         mutex_unlock(&cdt->cdt_restore_lock);
 834
 835         RETURN(0);
 836 out_lh:
 837         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 838 out_crh:
 839         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 840
 841         return rc;
 842 }
 843
 844 /**
 845  * lookup a restore handle by FID
 846  * caller needs to hold cdt_restore_lock
 847  * \param cdt [IN] coordinator
 848  * \param fid [IN] FID
 849  * \retval cdt_restore_handle found
 850  * \retval NULL not found
 851  */
 852 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
 853                                                    const struct lu_fid *fid)
 854 {
 855         struct cdt_restore_handle *crh;
 856         ENTRY;
 857
 858         list_for_each_entry(crh, &cdt->cdt_restore_handle_list, crh_list) {
 859                 if (lu_fid_eq(&crh->crh_fid, fid))
 860                         RETURN(crh);
 861         }
 862
 863         RETURN(NULL);
 864 }
 865
 866 void cdt_restore_handle_del(struct mdt_thread_info *mti,
 867                             struct coordinator *cdt, const struct lu_fid *fid)
 868 {
 869         struct cdt_restore_handle *crh;
 870
 871         /* give back layout lock */
 872         mutex_lock(&cdt->cdt_restore_lock);
 873         crh = cdt_restore_handle_find(cdt, fid);
 874         if (crh != NULL)
 875                 list_del(&crh->crh_list);
 876         mutex_unlock(&cdt->cdt_restore_lock);
 877
 878         if (crh == NULL)
 879                 return;
 880
 881         /* XXX We pass a NULL object since the restore handle does not
 882          * keep a reference on the object being restored. */
 883         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
 884         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 885 }
 886
 887 /**
 888  * data passed to llog_cat_process() callback
 889  * to scan requests and take actions
 890  */
 891 struct hsm_restore_data {
 892         struct mdt_thread_info  *hrd_mti;
 893 };
 894
 895 /**
 896  *  llog_cat_process() callback, used to:
 897  *  - find restore request and allocate the restore handle
 898  * \param env [IN] environment
 899  * \param llh [IN] llog handle
 900  * \param hdr [IN] llog record
 901  * \param data [IN/OUT] cb data = struct hsm_restore_data
 902  * \retval 0 success
 903  * \retval -ve failure
 904  */
 905 static int hsm_restore_cb(const struct lu_env *env,
 906                           struct llog_handle *llh,
 907                           struct llog_rec_hdr *hdr, void *data)
 908 {
 909         struct llog_agent_req_rec       *larr;
 910         struct hsm_restore_data         *hrd;
 911         struct hsm_action_item          *hai;
 912         struct mdt_thread_info          *mti;
 913         struct coordinator              *cdt;
 914         int rc;
 915         ENTRY;
 916
 917         hrd = data;
 918         mti = hrd->hrd_mti;
 919         cdt = &mti->mti_mdt->mdt_coordinator;
 920
 921         larr = (struct llog_agent_req_rec *)hdr;
 922         hai = &larr->arr_hai;
 923         if (hai->hai_cookie > cdt->cdt_last_cookie)
 924                 /* update the cookie to avoid collision */
 925                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 926
 927         if (hai->hai_action != HSMA_RESTORE ||
 928             agent_req_in_final_state(larr->arr_status))
 929                 RETURN(0);
 930
 931         /* restore request not in a final state */
 932
 933         /* force replay of restore requests left in started state from previous
 934          * CDT context, to be canceled later if finally found to be incompatible
 935          * when being re-started */
 936         if (larr->arr_status == ARS_STARTED) {
 937                 larr->arr_status = ARS_WAITING;
 938                 larr->arr_req_change = ktime_get_real_seconds();
 939                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 940                 if (rc != 0)
 941                         GOTO(out, rc);
 942         }
 943
 944         rc = cdt_restore_handle_add(mti, cdt, &hai->hai_fid, &hai->hai_extent);
 945 out:
 946         RETURN(rc);
 947 }
 948
 949 /**
 950  * restore coordinator state at startup
 951  * the goal is to take a layout lock for each registered restore request
 952  * \param mti [IN] context
 953  */
 954 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 955 {
 956         struct hsm_restore_data  hrd;
 957         int                      rc;
 958         ENTRY;
 959
 960         hrd.hrd_mti = mti;
 961
 962         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
 963                               0, 0, WRITE);
 964
 965         RETURN(rc);
 966 }
 967
 968 static int hsm_init_ucred(struct lu_ucred *uc)
 969 {
 970         ENTRY;
 971
 972         uc->uc_valid = UCRED_OLD;
 973         uc->uc_o_uid = 0;
 974         uc->uc_o_gid = 0;
 975         uc->uc_o_fsuid = 0;
 976         uc->uc_o_fsgid = 0;
 977         uc->uc_uid = 0;
 978         uc->uc_gid = 0;
 979         uc->uc_fsuid = 0;
 980         uc->uc_fsgid = 0;
 981         uc->uc_suppgids[0] = -1;
 982         uc->uc_suppgids[1] = -1;
 983         uc->uc_cap = CFS_CAP_FS_MASK;
 984         uc->uc_umask = 0777;
 985         uc->uc_ginfo = NULL;
 986         uc->uc_identity = NULL;
 987         /* always record internal HSM activity if also enabled globally */
 988         uc->uc_enable_audit = 1;
 989
 990         RETURN(0);
 991 }
 992
 993 /**
 994  * initialize coordinator struct
 995  * \param mdt [IN] device
 996  * \retval 0 success
 997  * \retval -ve failure
 998  */
 999 int mdt_hsm_cdt_init(struct mdt_device *mdt)
1000 {
1001         struct coordinator      *cdt = &mdt->mdt_coordinator;
1002         struct mdt_thread_info  *cdt_mti = NULL;
1003         int                      rc;
1004         ENTRY;
1005
1006         init_waitqueue_head(&cdt->cdt_waitq);
1007         init_rwsem(&cdt->cdt_llog_lock);
1008         init_rwsem(&cdt->cdt_agent_lock);
1009         init_rwsem(&cdt->cdt_request_lock);
1010         mutex_init(&cdt->cdt_restore_lock);
1011         mutex_init(&cdt->cdt_state_lock);
1012         set_cdt_state(cdt, CDT_STOPPED);
1013
1014         INIT_LIST_HEAD(&cdt->cdt_request_list);
1015         INIT_LIST_HEAD(&cdt->cdt_agents);
1016         INIT_LIST_HEAD(&cdt->cdt_restore_handle_list);
1017
1018         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
1019                                                        CFS_HASH_BITS_MIN,
1020                                                        CFS_HASH_BITS_MAX,
1021                                                        CFS_HASH_BKT_BITS,
1022                                                        0 /* extra bytes */,
1023                                                        CFS_HASH_MIN_THETA,
1024                                                        CFS_HASH_MAX_THETA,
1025                                                 &cdt_request_cookie_hash_ops,
1026                                                        CFS_HASH_DEFAULT);
1027         if (cdt->cdt_request_cookie_hash == NULL)
1028                 RETURN(-ENOMEM);
1029
1030         cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
1031                                                      CFS_HASH_BITS_MIN,
1032                                                      CFS_HASH_BITS_MAX,
1033                                                      CFS_HASH_BKT_BITS,
1034                                                      0 /* extra bytes */,
1035                                                      CFS_HASH_MIN_THETA,
1036                                                      CFS_HASH_MAX_THETA,
1037                                                      &cdt_agent_record_hash_ops,
1038                                                      CFS_HASH_DEFAULT);
1039         if (cdt->cdt_agent_record_hash == NULL)
1040                 GOTO(out_request_cookie_hash, rc = -ENOMEM);
1041
1042         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
1043         if (rc < 0)
1044                 GOTO(out_agent_record_hash, rc);
1045
1046         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1047         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
1048         if (rc < 0)
1049                 GOTO(out_env, rc);
1050
1051         lu_context_enter(&cdt->cdt_session);
1052         cdt->cdt_env.le_ses = &cdt->cdt_session;
1053
1054         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1055         LASSERT(cdt_mti != NULL);
1056
1057         cdt_mti->mti_env = &cdt->cdt_env;
1058         cdt_mti->mti_mdt = mdt;
1059
1060         hsm_init_ucred(mdt_ucred(cdt_mti));
1061
1062         /* default values for /proc tunnables
1063          * can be override by MGS conf */
1064         cdt->cdt_default_archive_id = 1;
1065         cdt->cdt_grace_delay = 60;
1066         cdt->cdt_loop_period = 10;
1067         cdt->cdt_max_requests = 3;
1068         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1069         cdt->cdt_active_req_timeout = 3600;
1070
1071         /* by default do not remove archives on last unlink */
1072         cdt->cdt_remove_archive_on_last_unlink = false;
1073
1074         RETURN(0);
1075
1076 out_env:
1077         lu_env_fini(&cdt->cdt_env);
1078 out_agent_record_hash:
1079         cfs_hash_putref(cdt->cdt_agent_record_hash);
1080         cdt->cdt_agent_record_hash = NULL;
1081 out_request_cookie_hash:
1082         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1083         cdt->cdt_request_cookie_hash = NULL;
1084
1085         return rc;
1086 }
1087
1088 /**
1089  * free a coordinator thread
1090  * \param mdt [IN] device
1091  */
1092 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
1093 {
1094         struct coordinator *cdt = &mdt->mdt_coordinator;
1095         ENTRY;
1096
1097         lu_context_exit(cdt->cdt_env.le_ses);
1098         lu_context_fini(cdt->cdt_env.le_ses);
1099
1100         lu_env_fini(&cdt->cdt_env);
1101
1102         cfs_hash_putref(cdt->cdt_agent_record_hash);
1103         cdt->cdt_agent_record_hash = NULL;
1104
1105         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1106         cdt->cdt_request_cookie_hash = NULL;
1107
1108         RETURN(0);
1109 }
1110
1111 /**
1112  * start a coordinator thread
1113  * \param mdt [IN] device
1114  * \retval 0 success
1115  * \retval -ve failure
1116  */
1117 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1118 {
1119         struct coordinator      *cdt = &mdt->mdt_coordinator;
1120         struct mdt_thread_info *cdt_mti;
1121         int                      rc;
1122         void                    *ptr;
1123         struct task_struct      *task;
1124         ENTRY;
1125
1126         /* functions defined but not yet used
1127          * this avoid compilation warning
1128          */
1129         ptr = dump_requests;
1130
1131         rc = set_cdt_state(cdt, CDT_INIT);
1132         if (rc) {
1133                 CERROR("%s: Coordinator already started or stopping\n",
1134                        mdt_obd_name(mdt));
1135                 RETURN(-EALREADY);
1136         }
1137
1138         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
1139         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1140
1141         /* just need to be larger than previous one */
1142         /* cdt_last_cookie is protected by cdt_llog_lock */
1143         cdt->cdt_last_cookie = ktime_get_real_seconds();
1144         atomic_set(&cdt->cdt_request_count, 0);
1145         atomic_set(&cdt->cdt_archive_count, 0);
1146         atomic_set(&cdt->cdt_restore_count, 0);
1147         atomic_set(&cdt->cdt_remove_count, 0);
1148         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
1149         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
1150         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
1151
1152         /* to avoid deadlock when start is made through /proc
1153          * /proc entries are created by the coordinator thread */
1154
1155         /* set up list of started restore requests */
1156         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1157         rc = mdt_hsm_pending_restore(cdt_mti);
1158         if (rc)
1159                 CERROR("%s: cannot take the layout locks needed"
1160                        " for registered restore: %d\n",
1161                        mdt_obd_name(mdt), rc);
1162
1163         if (mdt->mdt_bottom->dd_rdonly)
1164                 RETURN(0);
1165
1166         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
1167         if (IS_ERR(task)) {
1168                 rc = PTR_ERR(task);
1169                 set_cdt_state(cdt, CDT_STOPPED);
1170                 CERROR("%s: error starting coordinator thread: %d\n",
1171                        mdt_obd_name(mdt), rc);
1172         } else {
1173                 cdt->cdt_task = task;
1174                 wait_event(cdt->cdt_waitq,
1175                            cdt->cdt_state != CDT_INIT);
1176                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1177                        mdt_obd_name(mdt));
1178                 rc = 0;
1179         }
1180
1181         RETURN(rc);
1182 }
1183
1184 /**
1185  * stop a coordinator thread
1186  * \param mdt [IN] device
1187  */
1188 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1189 {
1190         struct coordinator *cdt = &mdt->mdt_coordinator;
1191         int rc;
1192
1193         ENTRY;
1194         /* stop coordinator thread */
1195         rc = set_cdt_state(cdt, CDT_STOPPING);
1196         if (rc == 0) {
1197                 kthread_stop(cdt->cdt_task);
1198                 cdt->cdt_task = NULL;
1199                 set_cdt_state(cdt, CDT_STOPPED);
1200         }
1201
1202         RETURN(rc);
1203 }
1204
1205 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1206                               const struct lu_fid *fid,
1207                               u32 archive_id)
1208 {
1209         struct mdt_object *obj;
1210         struct md_hsm mh;
1211         int rc;
1212
1213         obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1214         if (IS_ERR(obj))
1215                 GOTO(out, rc = PTR_ERR(obj));
1216
1217         if (mh.mh_flags & HS_EXISTS &&
1218             mh.mh_arch_id == archive_id)
1219                 GOTO(out_obj, rc = 0);
1220
1221         mh.mh_flags |= HS_EXISTS;
1222         mh.mh_arch_id = archive_id;
1223         rc = mdt_hsm_attr_set(mti, obj, &mh);
1224
1225 out_obj:
1226         mdt_object_put(mti->mti_env, obj);
1227 out:
1228         return rc;
1229 }
1230
1231 /**
1232  * register all requests from an hal in the memory list
1233  * \param mti [IN] context
1234  * \param hal [IN] request
1235  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1236  *  which is running the CT
1237  * \retval 0 success
1238  * \retval -ve failure
1239  */
1240 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1241                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1242 {
1243         struct mdt_device       *mdt = mti->mti_mdt;
1244         struct coordinator      *cdt = &mdt->mdt_coordinator;
1245         struct hsm_action_item  *hai;
1246         int                      rc = 0, i;
1247         ENTRY;
1248
1249         /* register request in memory list */
1250         hai = hai_first(hal);
1251         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1252                 struct cdt_agent_req *car;
1253
1254                 /* in case of a cancel request, we first mark the ondisk
1255                  * record of the request we want to stop as canceled
1256                  * this does not change the cancel record
1257                  * it will be done when updating the request status
1258                  */
1259                 if (hai->hai_action == HSMA_CANCEL) {
1260                         struct hsm_record_update update = {
1261                                 .cookie = hai->hai_cookie,
1262                                 .status = ARS_CANCELED,
1263                         };
1264
1265                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1266                                                      &update, 1);
1267                         if (rc) {
1268                                 CERROR("%s: mdt_agent_record_update() failed, "
1269                                        "rc=%d, cannot update status to %s "
1270                                        "for cookie %#llx\n",
1271                                        mdt_obd_name(mdt), rc,
1272                                        agent_req_status2name(ARS_CANCELED),
1273                                        hai->hai_cookie);
1274                                 GOTO(out, rc);
1275                         }
1276
1277                         /* find the running request to set it canceled */
1278                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1279                         if (car != NULL) {
1280                                 car->car_canceled = 1;
1281                                 /* uuid has to be changed to the one running the
1282                                 * request to cancel */
1283                                 *uuid = car->car_uuid;
1284                                 mdt_cdt_put_request(car);
1285                         }
1286                         /* no need to memorize cancel request
1287                          * this also avoid a deadlock when we receive
1288                          * a purge all requests command
1289                          */
1290                         continue;
1291                 }
1292
1293                 if (hai->hai_action == HSMA_ARCHIVE) {
1294                         rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1295                                                 hal->hal_archive_id);
1296                         if (rc == -ENOENT)
1297                                 continue;
1298                         else if (rc < 0)
1299                                 GOTO(out, rc);
1300                 }
1301
1302                 car = mdt_cdt_alloc_request(hal->hal_archive_id, hal->hal_flags,
1303                                             uuid, hai);
1304                 if (IS_ERR(car))
1305                         GOTO(out, rc = PTR_ERR(car));
1306
1307                 rc = mdt_cdt_add_request(cdt, car);
1308                 if (rc != 0)
1309                         mdt_cdt_free_request(car);
1310         }
1311 out:
1312         RETURN(rc);
1313 }
1314
1315 /**
1316  * swap layouts between 2 fids
1317  * \param mti [IN] context
1318  * \param obj [IN]
1319  * \param dfid [IN]
1320  * \param mh_common [IN] MD HSM
1321  */
1322 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1323                             struct mdt_object *obj, const struct lu_fid *dfid,
1324                             struct md_hsm *mh_common)
1325 {
1326         struct mdt_object       *dobj;
1327         struct mdt_lock_handle  *dlh;
1328         int                      rc;
1329         ENTRY;
1330
1331         if (!mdt_object_exists(obj))
1332                 GOTO(out, rc = -ENOENT);
1333
1334         /* we already have layout lock on obj so take only
1335          * on dfid */
1336         dlh = &mti->mti_lh[MDT_LH_OLD];
1337         mdt_lock_reg_init(dlh, LCK_EX);
1338         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1339         if (IS_ERR(dobj))
1340                 GOTO(out, rc = PTR_ERR(dobj));
1341
1342         /* if copy tool closes the volatile before sending the final
1343          * progress through llapi_hsm_copy_end(), all the objects
1344          * are removed and mdd_swap_layout LBUG */
1345         if (!mdt_object_exists(dobj)) {
1346                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1347                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1348                 GOTO(out_dobj, rc = -ENOENT);
1349         }
1350         /* Since we only handle restores here, unconditionally use
1351          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1352          * be preserved in case of failure during swap_layout and not
1353          * leave a file in an intermediate but incoherent state.
1354          * But need to setup HSM xattr of data FID before, reuse
1355          * mti and mh presets for FID in hsm_cdt_request_completed(),
1356          * only need to clear RELEASED and DIRTY.
1357          */
1358         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1359         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1360         if (rc == 0)
1361                 rc = mo_swap_layouts(mti->mti_env,
1362                                      mdt_object_child(obj),
1363                                      mdt_object_child(dobj),
1364                                      SWAP_LAYOUTS_MDS_HSM);
1365         if (rc == 0) {
1366                 rc = mdt_lsom_downgrade(mti, obj);
1367                 if (rc)
1368                         CDEBUG(D_INODE,
1369                                "%s: File fid="DFID" SOM "
1370                                "downgrade failed, rc = %d\n",
1371                                mdt_obd_name(mti->mti_mdt),
1372                                PFID(mdt_object_fid(obj)), rc);
1373         }
1374 out_dobj:
1375         mdt_object_unlock_put(mti, dobj, dlh, 1);
1376 out:
1377         RETURN(rc);
1378 }
1379
1380 /**
1381  * update status of a completed request
1382  * \param mti [IN] context
1383  * \param pgs [IN] progress of the copy tool
1384  * \retval 0 success
1385  * \retval -ve failure
1386  */
1387 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1388                                      struct hsm_progress_kernel *pgs,
1389                                      const struct cdt_agent_req *car,
1390                                      enum agent_req_status *status)
1391 {
1392         const struct lu_env     *env = mti->mti_env;
1393         struct mdt_device       *mdt = mti->mti_mdt;
1394         struct coordinator      *cdt = &mdt->mdt_coordinator;
1395         struct mdt_object       *obj = NULL;
1396         int                      cl_flags = 0, rc = 0;
1397         struct md_hsm            mh;
1398         bool                     is_mh_changed;
1399         bool                     need_changelog = true;
1400         ENTRY;
1401
1402         /* default is to retry */
1403         *status = ARS_WAITING;
1404
1405         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1406          * if error/removed continue anyway to get correct reporting done */
1407         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1408         /* we will update MD HSM only if needed */
1409         is_mh_changed = false;
1410
1411         /* no need to change mh->mh_arch_id
1412          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1413          */
1414         if (pgs->hpk_errval != 0) {
1415                 switch (pgs->hpk_errval) {
1416                 case ENOSYS:
1417                         /* the copy tool does not support cancel
1418                          * so the cancel request is failed
1419                          * As we cannot distinguish a cancel progress
1420                          * from another action progress (they have the
1421                          * same cookie), we suppose here the CT returns
1422                          * ENOSYS only if does not support cancel
1423                          */
1424                         /* this can also happen when cdt calls it to
1425                          * for a timed out request */
1426                         *status = ARS_FAILED;
1427                         /* to have a cancel event in changelog */
1428                         pgs->hpk_errval = ECANCELED;
1429                         break;
1430                 case ECANCELED:
1431                         /* the request record has already been set to
1432                          * ARS_CANCELED, this set the cancel request
1433                          * to ARS_SUCCEED */
1434                         *status = ARS_SUCCEED;
1435                         break;
1436                 default:
1437                         /* retry only if current policy or requested, and
1438                          * object is not on error/removed */
1439                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1440                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1441                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1442                         break;
1443                 }
1444
1445                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1446                         CERROR("%s: Request %#llx on "DFID
1447                                " failed, error code %d too large\n",
1448                                mdt_obd_name(mdt),
1449                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1450                                pgs->hpk_errval);
1451                         hsm_set_cl_error(&cl_flags,
1452                                          CLF_HSM_ERROVERFLOW);
1453                         rc = -EINVAL;
1454                 } else {
1455                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1456                 }
1457
1458                 switch (car->car_hai->hai_action) {
1459                 case HSMA_ARCHIVE:
1460                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1461                         break;
1462                 case HSMA_RESTORE:
1463                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1464                         break;
1465                 case HSMA_REMOVE:
1466                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1467                         break;
1468                 case HSMA_CANCEL:
1469                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1470                         CERROR("%s: Failed request %#llx on "DFID
1471                                " cannot be a CANCEL\n",
1472                                mdt_obd_name(mdt),
1473                                pgs->hpk_cookie,
1474                                PFID(&pgs->hpk_fid));
1475                         break;
1476                 default:
1477                         CERROR("%s: Failed request %#llx on "DFID
1478                                " %d is an unknown action\n",
1479                                mdt_obd_name(mdt),
1480                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1481                                car->car_hai->hai_action);
1482                         rc = -EINVAL;
1483                         break;
1484                 }
1485         } else {
1486                 *status = ARS_SUCCEED;
1487                 switch (car->car_hai->hai_action) {
1488                 case HSMA_ARCHIVE:
1489                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1490                         /* set ARCHIVE keep EXIST and clear LOST and
1491                          * DIRTY */
1492                         mh.mh_arch_ver = pgs->hpk_data_version;
1493                         mh.mh_flags |= HS_ARCHIVED;
1494                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1495                         is_mh_changed = true;
1496                         break;
1497                 case HSMA_RESTORE:
1498                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1499
1500                         /* do not clear RELEASED and DIRTY here
1501                          * this will occur in hsm_swap_layouts()
1502                          */
1503
1504                         /* Restoring has changed the file version on
1505                          * disk. */
1506                         mh.mh_arch_ver = pgs->hpk_data_version;
1507                         is_mh_changed = true;
1508                         break;
1509                 case HSMA_REMOVE:
1510                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1511                         /* clear ARCHIVED EXISTS and LOST */
1512                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1513                         is_mh_changed = true;
1514                         break;
1515                 case HSMA_CANCEL:
1516                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1517                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1518                                mdt_obd_name(mdt),
1519                                pgs->hpk_cookie,
1520                                PFID(&pgs->hpk_fid));
1521                         break;
1522                 default:
1523                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1524                                mdt_obd_name(mdt),
1525                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1526                                car->car_hai->hai_action);
1527                         rc = -EINVAL;
1528                         break;
1529                 }
1530         }
1531
1532         /* rc != 0 means error when analysing action, it may come from
1533          * a crasy CT no need to manage DIRTY
1534          * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1535          * filled
1536          */
1537         if (rc == 0 && !IS_ERR(obj))
1538                 hsm_set_cl_flags(&cl_flags,
1539                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1540
1541         /* unlock is done later, after layout lock management */
1542         if (is_mh_changed && !IS_ERR(obj))
1543                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1544
1545         /* we give back layout lock only if restore was successful or
1546          * if no retry will be attempted and if object is still alive,
1547          * in other cases we just unlock the object */
1548         if (car->car_hai->hai_action == HSMA_RESTORE) {
1549                 /* restore in data FID done, we swap the layouts
1550                  * only if restore is successful */
1551                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1552                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1553                                               &mh);
1554                         if (rc) {
1555                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1556                                         *status = ARS_FAILED;
1557                                 pgs->hpk_errval = -rc;
1558                         }
1559                 }
1560                 /* we have to retry, so keep layout lock */
1561                 if (*status == ARS_WAITING)
1562                         GOTO(out, rc);
1563
1564                 /* restore special case, need to create ChangeLog record
1565                  * before to give back layout lock to avoid concurrent
1566                  * file updater to post out of order ChangeLog */
1567                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1568                              &car->car_hai->hai_fid);
1569                 need_changelog = false;
1570
1571                 cdt_restore_handle_del(mti, cdt, &car->car_hai->hai_fid);
1572         }
1573
1574         GOTO(out, rc);
1575
1576 out:
1577         /* always add a ChangeLog record */
1578         if (need_changelog)
1579                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1580                              &car->car_hai->hai_fid);
1581
1582         if (!IS_ERR(obj))
1583                 mdt_object_put(mti->mti_env, obj);
1584
1585         RETURN(rc);
1586 }
1587
1588 /**
1589  * update status of a request
1590  * \param mti [IN] context
1591  * \param pgs [IN] progress of the copy tool
1592  * \retval 0 success
1593  * \retval -ve failure
1594  */
1595 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1596                                  struct hsm_progress_kernel *pgs)
1597 {
1598         struct mdt_device       *mdt = mti->mti_mdt;
1599         struct coordinator      *cdt = &mdt->mdt_coordinator;
1600         struct cdt_agent_req    *car;
1601         int                      rc = 0;
1602         ENTRY;
1603
1604         /* no coordinator started, so we cannot serve requests */
1605         if (cdt->cdt_state == CDT_STOPPED)
1606                 RETURN(-EAGAIN);
1607
1608         /* first do sanity checks */
1609         car = mdt_cdt_update_request(cdt, pgs);
1610         if (IS_ERR(car)) {
1611                 CERROR("%s: Cannot find running request for cookie %#llx"
1612                        " on fid="DFID"\n",
1613                        mdt_obd_name(mdt),
1614                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1615
1616                 RETURN(PTR_ERR(car));
1617         }
1618
1619         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1620                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1621                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1622                       hsm_copytool_action2name(car->car_hai->hai_action),
1623                       pgs->hpk_flags, pgs->hpk_errval,
1624                       PFID(&car->car_hai->hai_fid),
1625                       PFID(&car->car_hai->hai_dfid));
1626
1627         /* progress is done on FID or data FID depending of the action and
1628          * of the copy progress */
1629         /* for restore progress is used to send back the data FID to cdt */
1630         if (car->car_hai->hai_action == HSMA_RESTORE &&
1631             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1632                 car->car_hai->hai_dfid = pgs->hpk_fid;
1633
1634         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1635              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1636             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1637              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1638                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1639                        " does not match request FID "DFID" nor data FID "
1640                        DFID"\n",
1641                        mdt_obd_name(mdt),
1642                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1643                        PFID(&car->car_hai->hai_fid),
1644                        PFID(&car->car_hai->hai_dfid));
1645                 GOTO(out, rc = -EINVAL);
1646         }
1647
1648         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1649                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1650                        " is not coherent (err=%d and not completed"
1651                        " (flags=%d))\n",
1652                        mdt_obd_name(mdt),
1653                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1654                        hsm_copytool_action2name(car->car_hai->hai_action),
1655                        pgs->hpk_errval, pgs->hpk_flags);
1656                 GOTO(out, rc = -EINVAL);
1657         }
1658
1659         /* now progress is valid */
1660
1661         /* we use a root like ucred */
1662         hsm_init_ucred(mdt_ucred(mti));
1663
1664         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1665                 enum agent_req_status status;
1666                 struct hsm_record_update update;
1667                 int rc1;
1668
1669                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1670
1671                 CDEBUG(D_HSM, "updating record: fid="DFID" cookie=%#llx action=%s "
1672                               "status=%s\n",
1673                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1674                        hsm_copytool_action2name(car->car_hai->hai_action),
1675                        agent_req_status2name(status));
1676
1677                 /* update record first (LU-9075) */
1678                 update.cookie = pgs->hpk_cookie;
1679                 update.status = status;
1680
1681                 rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1682                                               &update, 1);
1683                 if (rc1)
1684                         CERROR("%s: mdt_agent_record_update() failed,"
1685                                " rc=%d, cannot update status to %s"
1686                                " for cookie %#llx\n",
1687                                mdt_obd_name(mdt), rc1,
1688                                agent_req_status2name(status),
1689                                pgs->hpk_cookie);
1690                 rc = (rc != 0 ? rc : rc1);
1691
1692                 /* then remove request from memory list (LU-9075) */
1693                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1694
1695                 /* ct has completed a request, so a slot is available,
1696                  * signal the coordinator to find new work */
1697                 mdt_hsm_cdt_event(cdt);
1698         } else {
1699                 /* if copytool send a progress on a canceled request
1700                  * we inform copytool it should stop
1701                  */
1702                 if (car->car_canceled == 1)
1703                         rc = -ECANCELED;
1704         }
1705         GOTO(out, rc);
1706
1707 out:
1708         /* remove ref got from mdt_cdt_update_request() */
1709         mdt_cdt_put_request(car);
1710
1711         return rc;
1712 }
1713
1714
1715 /**
1716  * data passed to llog_cat_process() callback
1717  * to cancel requests
1718  */
1719 struct hsm_cancel_all_data {
1720         struct mdt_device       *mdt;
1721 };
1722
1723 /**
1724  *  llog_cat_process() callback, used to:
1725  *  - purge all requests
1726  * \param env [IN] environment
1727  * \param llh [IN] llog handle
1728  * \param hdr [IN] llog record
1729  * \param data [IN] cb data = struct hsm_cancel_all_data
1730  * \retval 0 success
1731  * \retval -ve failure
1732  */
1733 static int mdt_cancel_all_cb(const struct lu_env *env,
1734                              struct llog_handle *llh,
1735                              struct llog_rec_hdr *hdr, void *data)
1736 {
1737         struct llog_agent_req_rec       *larr;
1738         struct hsm_cancel_all_data      *hcad;
1739         int                              rc = 0;
1740         ENTRY;
1741
1742         larr = (struct llog_agent_req_rec *)hdr;
1743         hcad = data;
1744         if (larr->arr_status == ARS_WAITING ||
1745             larr->arr_status == ARS_STARTED) {
1746                 larr->arr_status = ARS_CANCELED;
1747                 larr->arr_req_change = ktime_get_real_seconds();
1748                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1749         }
1750
1751         RETURN(rc);
1752 }
1753
1754 /**
1755  * cancel all actions
1756  * \param obd [IN] MDT device
1757  */
1758 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1759 {
1760         struct lu_env                    env;
1761         struct lu_context                session;
1762         struct mdt_thread_info          *mti;
1763         struct coordinator              *cdt = &mdt->mdt_coordinator;
1764         struct cdt_agent_req            *car;
1765         struct hsm_action_list          *hal = NULL;
1766         struct hsm_action_item          *hai;
1767         struct hsm_cancel_all_data       hcad;
1768         int                              hal_sz = 0, hal_len, rc;
1769         enum cdt_states                  old_state;
1770         ENTRY;
1771
1772         rc = lu_env_init(&env, LCT_MD_THREAD);
1773         if (rc < 0)
1774                 RETURN(rc);
1775
1776         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1777         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1778         if (rc < 0)
1779                 GOTO(out_env, rc);
1780
1781         lu_context_enter(&session);
1782         env.le_ses = &session;
1783
1784         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1785         LASSERT(mti != NULL);
1786
1787         mti->mti_env = &env;
1788         mti->mti_mdt = mdt;
1789
1790         hsm_init_ucred(mdt_ucred(mti));
1791
1792         mutex_lock(&cdt->cdt_state_lock);
1793         old_state = cdt->cdt_state;
1794
1795         /* disable coordinator */
1796         rc = set_cdt_state_locked(cdt, CDT_DISABLE);
1797         if (rc)
1798                 GOTO(out_cdt_state_unlock, rc);
1799
1800         /* send cancel to all running requests */
1801         down_read(&cdt->cdt_request_lock);
1802         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1803                 mdt_cdt_get_request(car);
1804                 /* request is not yet removed from list, it will be done
1805                  * when copytool will return progress
1806                  */
1807
1808                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1809                         mdt_cdt_put_request(car);
1810                         continue;
1811                 }
1812
1813                 /* needed size */
1814                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1815                           cfs_size_round(car->car_hai->hai_len);
1816
1817                 if (hal_len > hal_sz && hal_sz > 0) {
1818                         /* not enough room, free old buffer */
1819                         OBD_FREE(hal, hal_sz);
1820                         hal = NULL;
1821                 }
1822
1823                 /* empty buffer, allocate one */
1824                 if (hal == NULL) {
1825                         hal_sz = hal_len;
1826                         OBD_ALLOC(hal, hal_sz);
1827                         if (hal == NULL) {
1828                                 mdt_cdt_put_request(car);
1829                                 up_read(&cdt->cdt_request_lock);
1830                                 GOTO(out_cdt_state, rc = -ENOMEM);
1831                         }
1832                 }
1833
1834                 hal->hal_version = HAL_VERSION;
1835                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1836                                 MTI_NAME_MAXLEN);
1837                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1838                 hal->hal_archive_id = car->car_archive_id;
1839                 hal->hal_flags = car->car_flags;
1840                 hal->hal_count = 0;
1841
1842                 hai = hai_first(hal);
1843                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1844                 hai->hai_action = HSMA_CANCEL;
1845                 hal->hal_count = 1;
1846
1847                 /* it is possible to safely call mdt_hsm_agent_send()
1848                  * (ie without a deadlock on cdt_request_lock), because the
1849                  * write lock is taken only if we are not in purge mode
1850                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1851                  *   nor mdt_cdt_remove_request())
1852                  */
1853                 /* no conflict with cdt thread because cdt is disable and we
1854                  * have the request lock */
1855                 mdt_hsm_agent_send(mti, hal, 1);
1856
1857                 mdt_cdt_put_request(car);
1858         }
1859         up_read(&cdt->cdt_request_lock);
1860
1861         if (hal != NULL)
1862                 OBD_FREE(hal, hal_sz);
1863
1864         /* cancel all on-disk records */
1865         hcad.mdt = mdt;
1866
1867         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1868                               &hcad, 0, 0, WRITE);
1869 out_cdt_state:
1870         /* Enable coordinator, unless the coordinator was stopping. */
1871         set_cdt_state_locked(cdt, old_state);
1872 out_cdt_state_unlock:
1873         mutex_unlock(&cdt->cdt_state_lock);
1874
1875         lu_context_exit(&session);
1876         lu_context_fini(&session);
1877 out_env:
1878         lu_env_fini(&env);
1879
1880         RETURN(rc);
1881 }
1882
1883 /**
1884  * check if a request is compatible with file status
1885  * \param hai [IN] request description
1886  * \param archive_id [IN] request archive id
1887  * \param rq_flags [IN] request flags
1888  * \param hsm [IN] file HSM metadata
1889  * \retval boolean
1890  */
1891 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1892                               u32 archive_id, u64 rq_flags,
1893                               const struct md_hsm *hsm)
1894 {
1895         int      is_compat = false;
1896         int      hsm_flags;
1897         ENTRY;
1898
1899         hsm_flags = hsm->mh_flags;
1900         switch (hai->hai_action) {
1901         case HSMA_ARCHIVE:
1902                 if (!(hsm_flags & HS_NOARCHIVE) &&
1903                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1904                         is_compat = true;
1905
1906                 if (hsm_flags & HS_EXISTS &&
1907                     archive_id != 0 &&
1908                     archive_id != hsm->mh_arch_id)
1909                         is_compat = false;
1910
1911                 break;
1912         case HSMA_RESTORE:
1913                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1914                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1915                         is_compat = true;
1916                 break;
1917         case HSMA_REMOVE:
1918                 if (!(hsm_flags & HS_RELEASED) &&
1919                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1920                         is_compat = true;
1921                 break;
1922         case HSMA_CANCEL:
1923                 is_compat = true;
1924                 break;
1925         }
1926         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1927                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1928                       PFID(&hai->hai_fid),
1929                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1930                       hai->hai_extent.offset, hai->hai_extent.length,
1931                       hsm->mh_flags,
1932                       (is_compat ? "compatible" : "uncompatible"));
1933
1934         RETURN(is_compat);
1935 }
1936
1937 /*
1938  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1939  */
1940 static const struct {
1941         __u64            bit;
1942         char            *name;
1943         char            *nickname;
1944 } hsm_policy_names[] = {
1945         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1946         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1947         { 0 },
1948 };
1949
1950 /**
1951  * convert a policy name to a bit
1952  * \param name [IN] policy name
1953  * \retval 0 unknown
1954  * \retval   policy bit
1955  */
1956 static __u64 hsm_policy_str2bit(const char *name)
1957 {
1958         int      i;
1959
1960         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1961                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1962                     strcmp(hsm_policy_names[i].name, name) == 0)
1963                         return hsm_policy_names[i].bit;
1964         return 0;
1965 }
1966
1967 /**
1968  * convert a policy bit field to a string
1969  * \param mask [IN] policy bit field
1970  * \param hexa [IN] print mask before bit names
1971  * \param buffer [OUT] string
1972  * \param count [IN] size of buffer
1973  */
1974 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1975                                 const bool hexa)
1976 {
1977         int      i, j;
1978         __u64    bit;
1979         ENTRY;
1980
1981         if (hexa)
1982                 seq_printf(m, "(%#llx) ", mask);
1983
1984         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1985                 bit = (1ULL << i);
1986
1987                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1988                         if (hsm_policy_names[j].bit == bit)
1989                                 break;
1990                 }
1991                 if (bit & mask)
1992                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1993                 else
1994                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1995         }
1996         /* remove last ' ' */
1997         m->count--;
1998         seq_putc(m, '\n');
1999 }
2000
2001 /* methods to read/write HSM policy flags */
2002 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
2003 {
2004         struct mdt_device       *mdt = m->private;
2005         struct coordinator      *cdt = &mdt->mdt_coordinator;
2006         ENTRY;
2007
2008         hsm_policy_bit2str(m, cdt->cdt_policy, false);
2009         RETURN(0);
2010 }
2011
2012 static ssize_t
2013 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
2014                          size_t count, loff_t *off)
2015 {
2016         struct seq_file         *m = file->private_data;
2017         struct mdt_device       *mdt = m->private;
2018         struct coordinator      *cdt = &mdt->mdt_coordinator;
2019         char                    *start, *token, sign;
2020         char                    *buf;
2021         __u64                    policy;
2022         __u64                    add_mask, remove_mask, set_mask;
2023         int                      rc;
2024         ENTRY;
2025
2026         if (count + 1 > PAGE_SIZE)
2027                 RETURN(-EINVAL);
2028
2029         OBD_ALLOC(buf, count + 1);
2030         if (buf == NULL)
2031                 RETURN(-ENOMEM);
2032
2033         if (copy_from_user(buf, buffer, count))
2034                 GOTO(out, rc = -EFAULT);
2035
2036         buf[count] = '\0';
2037
2038         start = buf;
2039         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
2040                start);
2041
2042         add_mask = remove_mask = set_mask = 0;
2043         do {
2044                 token = strsep(&start, "\n ");
2045                 sign = *token;
2046
2047                 if (sign == '\0')
2048                         continue;
2049
2050                 if (sign == '-' || sign == '+')
2051                         token++;
2052
2053                 policy = hsm_policy_str2bit(token);
2054                 if (policy == 0) {
2055                         CWARN("%s: '%s' is unknown, "
2056                               "supported policies are:\n", mdt_obd_name(mdt),
2057                               token);
2058                         hsm_policy_bit2str(m, 0, false);
2059                         GOTO(out, rc = -EINVAL);
2060                 }
2061                 switch (sign) {
2062                 case '-':
2063                         remove_mask |= policy;
2064                         break;
2065                 case '+':
2066                         add_mask |= policy;
2067                         break;
2068                 default:
2069                         set_mask |= policy;
2070                         break;
2071                 }
2072
2073         } while (start != NULL);
2074
2075         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
2076                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
2077
2078         /* if no sign in all string, it is a clear and set
2079          * if some sign found, all unsigned are converted
2080          * to add
2081          * P1 P2 = set to P1 and P2
2082          * P1 -P2 = add P1 clear P2 same as +P1 -P2
2083          */
2084         if (remove_mask == 0 && add_mask == 0) {
2085                 cdt->cdt_policy = set_mask;
2086         } else {
2087                 cdt->cdt_policy |= set_mask | add_mask;
2088                 cdt->cdt_policy &= ~remove_mask;
2089         }
2090
2091         GOTO(out, rc = count);
2092
2093 out:
2094         OBD_FREE(buf, count + 1);
2095         RETURN(rc);
2096 }
2097 LPROC_SEQ_FOPS(mdt_hsm_policy);
2098
2099 #define GENERATE_PROC_METHOD(VAR)                                       \
2100 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
2101 {                                                                       \
2102         struct mdt_device       *mdt = m->private;                      \
2103         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
2104         ENTRY;                                                          \
2105                                                                         \
2106         seq_printf(m, "%llu\n", (__u64)cdt->VAR);                       \
2107         RETURN(0);                                                      \
2108 }                                                                       \
2109 static ssize_t                                                          \
2110 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
2111                           size_t count, loff_t *off)                    \
2112                                                                         \
2113 {                                                                       \
2114         struct seq_file         *m = file->private_data;                \
2115         struct mdt_device       *mdt = m->private;                      \
2116         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
2117         unsigned int val;                                               \
2118         int rc;                                                         \
2119                                                                         \
2120         ENTRY;                                                          \
2121         rc = kstrtouint_from_user(buffer, count, 0, &val);              \
2122         if (rc)                                                         \
2123                 RETURN(rc);                                             \
2124                                                                         \
2125         if (val !=  0) {                                                \
2126                 cdt->VAR = val;                                         \
2127                 RETURN(count);                                          \
2128         }                                                               \
2129         RETURN(-EINVAL);                                                \
2130 }                                                                       \
2131
2132 GENERATE_PROC_METHOD(cdt_loop_period)
2133 GENERATE_PROC_METHOD(cdt_grace_delay)
2134 GENERATE_PROC_METHOD(cdt_active_req_timeout)
2135 GENERATE_PROC_METHOD(cdt_max_requests)
2136 GENERATE_PROC_METHOD(cdt_default_archive_id)
2137
2138 /*
2139  * procfs write method for MDT/hsm_control
2140  * proc entry is in mdt directory so data is mdt obd_device pointer
2141  */
2142 #define CDT_ENABLE_CMD   "enabled"
2143 #define CDT_STOP_CMD     "shutdown"
2144 #define CDT_DISABLE_CMD  "disabled"
2145 #define CDT_PURGE_CMD    "purge"
2146 #define CDT_HELP_CMD     "help"
2147 #define CDT_MAX_CMD_LEN  10
2148
2149 ssize_t
2150 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
2151                               size_t count, loff_t *off)
2152 {
2153         struct seq_file         *m = file->private_data;
2154         struct obd_device       *obd = m->private;
2155         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
2156         struct coordinator      *cdt = &(mdt->mdt_coordinator);
2157         int                      rc, usage = 0;
2158         char                     kernbuf[CDT_MAX_CMD_LEN];
2159         ENTRY;
2160
2161         if (count == 0 || count >= sizeof(kernbuf))
2162                 RETURN(-EINVAL);
2163
2164         if (copy_from_user(kernbuf, buffer, count))
2165                 RETURN(-EFAULT);
2166         kernbuf[count] = 0;
2167
2168         if (kernbuf[count - 1] == '\n')
2169                 kernbuf[count - 1] = 0;
2170
2171         rc = 0;
2172         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
2173                 if (cdt->cdt_state == CDT_DISABLE) {
2174                         rc = set_cdt_state(cdt, CDT_RUNNING);
2175                         mdt_hsm_cdt_event(cdt);
2176                         wake_up(&cdt->cdt_waitq);
2177                 } else {
2178                         rc = mdt_hsm_cdt_start(mdt);
2179                 }
2180         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2181                 if ((cdt->cdt_state == CDT_STOPPING) ||
2182                     (cdt->cdt_state == CDT_STOPPED)) {
2183                         CERROR("%s: Coordinator already stopped\n",
2184                                mdt_obd_name(mdt));
2185                         rc = -EALREADY;
2186                 } else {
2187                         rc = mdt_hsm_cdt_stop(mdt);
2188                 }
2189         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2190                 if ((cdt->cdt_state == CDT_STOPPING) ||
2191                     (cdt->cdt_state == CDT_STOPPED)) {
2192                         CERROR("%s: Coordinator is stopped\n",
2193                                mdt_obd_name(mdt));
2194                         rc = -EINVAL;
2195                 } else {
2196                         rc = set_cdt_state(cdt, CDT_DISABLE);
2197                 }
2198         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2199                 rc = hsm_cancel_all_actions(mdt);
2200         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2201                 usage = 1;
2202         } else {
2203                 usage = 1;
2204                 rc = -EINVAL;
2205         }
2206
2207         if (usage == 1)
2208                 CERROR("%s: Valid coordinator control commands are: "
2209                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2210                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2211                        CDT_PURGE_CMD, CDT_HELP_CMD);
2212
2213         if (rc)
2214                 RETURN(rc);
2215
2216         RETURN(count);
2217 }
2218
2219 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2220 {
2221         struct obd_device       *obd = m->private;
2222         struct coordinator      *cdt;
2223         ENTRY;
2224
2225         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2226
2227         seq_printf(m, "%s\n", cdt_mdt_state2str(cdt->cdt_state));
2228
2229         RETURN(0);
2230 }
2231
2232 static int
2233 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2234 {
2235         bool first = true;
2236         int i;
2237         ENTRY;
2238
2239         for (i = 0; i < 8 * sizeof(mask); i++) {
2240                 if (mask & (1UL << i)) {
2241                         seq_printf(m, "%s%s", first ? "" : " ",
2242                                    hsm_copytool_action2name(i));
2243                         first = false;
2244                 }
2245         }
2246         seq_putc(m, '\n');
2247
2248         RETURN(0);
2249 }
2250
2251 static int
2252 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2253 {
2254         struct mdt_device *mdt = m->private;
2255         struct coordinator *cdt = &mdt->mdt_coordinator;
2256
2257         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2258 }
2259
2260 static int
2261 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2262 {
2263         struct mdt_device *mdt = m->private;
2264         struct coordinator *cdt = &mdt->mdt_coordinator;
2265
2266         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2267 }
2268
2269 static int
2270 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2271 {
2272         struct mdt_device *mdt = m->private;
2273         struct coordinator *cdt = &mdt->mdt_coordinator;
2274
2275         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2276 }
2277
2278 static inline enum hsm_copytool_action
2279 hsm_copytool_name2action(const char *name)
2280 {
2281         if (strcasecmp(name, "NOOP") == 0)
2282                 return HSMA_NONE;
2283         else if (strcasecmp(name, "ARCHIVE") == 0)
2284                 return HSMA_ARCHIVE;
2285         else if (strcasecmp(name, "RESTORE") == 0)
2286                 return HSMA_RESTORE;
2287         else if (strcasecmp(name, "REMOVE") == 0)
2288                 return HSMA_REMOVE;
2289         else if (strcasecmp(name, "CANCEL") == 0)
2290                 return HSMA_CANCEL;
2291         else
2292                 return -1;
2293 }
2294
2295 static ssize_t
2296 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2297                             size_t user_count, __u64 *mask)
2298 {
2299         char *buf, *pos, *name;
2300         size_t buf_size;
2301         __u64 new_mask = 0;
2302         int rc;
2303         ENTRY;
2304
2305         if (!(user_count < 4096))
2306                 RETURN(-ENOMEM);
2307
2308         buf_size = user_count + 1;
2309
2310         OBD_ALLOC(buf, buf_size);
2311         if (buf == NULL)
2312                 RETURN(-ENOMEM);
2313
2314         if (copy_from_user(buf, user_buf, buf_size - 1))
2315                 GOTO(out, rc = -EFAULT);
2316
2317         buf[buf_size - 1] = '\0';
2318
2319         pos = buf;
2320         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2321                 int action;
2322
2323                 if (*name == '\0')
2324                         continue;
2325
2326                 action = hsm_copytool_name2action(name);
2327                 if (action < 0)
2328                         GOTO(out, rc = -EINVAL);
2329
2330                 new_mask |= (1UL << action);
2331         }
2332
2333         *mask = new_mask;
2334         rc = user_count;
2335 out:
2336         OBD_FREE(buf, buf_size);
2337
2338         RETURN(rc);
2339 }
2340
2341 static ssize_t
2342 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2343                                         size_t count, loff_t *off)
2344 {
2345         struct seq_file         *m = file->private_data;
2346         struct mdt_device       *mdt = m->private;
2347         struct coordinator *cdt = &mdt->mdt_coordinator;
2348
2349         return mdt_write_hsm_request_mask(file, buf, count,
2350                                            &cdt->cdt_user_request_mask);
2351 }
2352
2353 static ssize_t
2354 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2355                                         size_t count, loff_t *off)
2356 {
2357         struct seq_file         *m = file->private_data;
2358         struct mdt_device       *mdt = m->private;
2359         struct coordinator      *cdt = &mdt->mdt_coordinator;
2360
2361         return mdt_write_hsm_request_mask(file, buf, count,
2362                                            &cdt->cdt_group_request_mask);
2363 }
2364
2365 static ssize_t
2366 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2367                                         size_t count, loff_t *off)
2368 {
2369         struct seq_file         *m = file->private_data;
2370         struct mdt_device       *mdt = m->private;
2371         struct coordinator      *cdt = &mdt->mdt_coordinator;
2372
2373         return mdt_write_hsm_request_mask(file, buf, count,
2374                                            &cdt->cdt_other_request_mask);
2375 }
2376
2377 static int mdt_hsm_cdt_raolu_seq_show(struct seq_file *m, void *data)
2378 {
2379         struct mdt_device *mdt = m->private;
2380         struct coordinator *cdt = &mdt->mdt_coordinator;
2381         ENTRY;
2382
2383         seq_printf(m, "%d\n", (int)cdt->cdt_remove_archive_on_last_unlink);
2384         RETURN(0);
2385 }
2386
2387 static ssize_t
2388 mdt_hsm_cdt_raolu_seq_write(struct file *file, const char __user *buffer,
2389                             size_t count, loff_t *off)
2390
2391 {
2392         struct seq_file *m = file->private_data;
2393         struct mdt_device *mdt = m->private;
2394         struct coordinator *cdt = &mdt->mdt_coordinator;
2395         bool val;
2396         int rc;
2397
2398         ENTRY;
2399         rc = kstrtobool_from_user(buffer, count, &val);
2400         if (rc < 0)
2401                 RETURN(rc);
2402
2403         cdt->cdt_remove_archive_on_last_unlink = val;
2404         RETURN(count);
2405 }
2406
2407 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2408 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2409 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2410 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2411 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2412 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2413 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2414 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2415 LPROC_SEQ_FOPS(mdt_hsm_cdt_raolu);
2416
2417 /* Read-only proc files for request counters */
2418 static int mdt_hsm_cdt_archive_count_seq_show(struct seq_file *m, void *data)
2419 {
2420         struct mdt_device *mdt = m->private;
2421         struct coordinator *cdt = &mdt->mdt_coordinator;
2422         ENTRY;
2423
2424         seq_printf(m, "%d\n", atomic_read(&cdt->cdt_archive_count));
2425         RETURN(0);
2426 }
2427
2428 static int mdt_hsm_cdt_restore_count_seq_show(struct seq_file *m, void *data)
2429 {
2430         struct mdt_device *mdt = m->private;
2431         struct coordinator *cdt = &mdt->mdt_coordinator;
2432         ENTRY;
2433
2434         seq_printf(m, "%d\n", atomic_read(&cdt->cdt_restore_count));
2435         RETURN(0);
2436 }
2437
2438 static int mdt_hsm_cdt_remove_count_seq_show(struct seq_file *m, void *data)
2439 {
2440         struct mdt_device *mdt = m->private;
2441         struct coordinator *cdt = &mdt->mdt_coordinator;
2442         ENTRY;
2443
2444         seq_printf(m, "%d\n", atomic_read(&cdt->cdt_remove_count));
2445         RETURN(0);
2446 }
2447
2448 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_archive_count);
2449 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_restore_count);
2450 LPROC_SEQ_FOPS_RO(mdt_hsm_cdt_remove_count);
2451
2452 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2453         { .name =       "agents",
2454           .fops =       &mdt_hsm_agent_fops                     },
2455         { .name =       "actions",
2456           .fops =       &mdt_hsm_actions_fops,
2457           .proc_mode =  0444                                    },
2458         { .name =       "default_archive_id",
2459           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2460         { .name =       "grace_delay",
2461           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2462         { .name =       "loop_period",
2463           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2464         { .name =       "max_requests",
2465           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2466         { .name =       "policy",
2467           .fops =       &mdt_hsm_policy_fops                    },
2468         { .name =       "active_request_timeout",
2469           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2470         { .name =       "active_requests",
2471           .fops =       &mdt_hsm_active_requests_fops           },
2472         { .name =       "user_request_mask",
2473           .fops =       &mdt_hsm_user_request_mask_fops,        },
2474         { .name =       "group_request_mask",
2475           .fops =       &mdt_hsm_group_request_mask_fops,       },
2476         { .name =       "other_request_mask",
2477           .fops =       &mdt_hsm_other_request_mask_fops,       },
2478         { .name =       "remove_archive_on_last_unlink",
2479           .fops =       &mdt_hsm_cdt_raolu_fops,                },
2480         { .name =       "archive_count",
2481           .fops =       &mdt_hsm_cdt_archive_count_fops,        },
2482         { .name =       "restore_count",
2483           .fops =       &mdt_hsm_cdt_restore_count_fops,        },
2484         { .name =       "remove_count",
2485           .fops =       &mdt_hsm_cdt_remove_count_fops,         },
2486         { 0 }
2487 };