lustre/mdt/mdt_coordinator.c

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