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, 2016, 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 compound %#llx"
 105                       " archive_id %d flags %#llx\n",
 106                prefix, hal->hal_version, hal->hal_count,
 107                hal->hal_compound_id, 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                        " compound/cookie=%#llx/%#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                        hal->hal_compound_id, 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          *mti;
 140         char                             fs_name[MTI_NAME_MAXLEN+1];
 141         /* request to be send to agents */
 142         int                              max_requests;  /** vector size */
 143         int                              request_cnt;   /** used count */
 144         struct hsm_scan_request         *request;
 145 };
 146
 147 struct hsm_thread_data {
 148         struct mdt_thread_info  *cdt_mti;
 149         struct hsm_scan_request *request;
 150 };
 151 /**
 152  *  llog_cat_process() callback, used to:
 153  *  - find waiting request and start action
 154  *  - purge canceled and done requests
 155  * \param env [IN] environment
 156  * \param llh [IN] llog handle
 157  * \param hdr [IN] llog record
 158  * \param data [IN/OUT] cb data = struct hsm_scan_data
 159  * \retval 0 success
 160  * \retval -ve failure
 161  */
 162 static int mdt_coordinator_cb(const struct lu_env *env,
 163                               struct llog_handle *llh,
 164                               struct llog_rec_hdr *hdr,
 165                               void *data)
 166 {
 167         struct llog_agent_req_rec       *larr;
 168         struct hsm_scan_data            *hsd;
 169         struct hsm_action_item          *hai;
 170         struct mdt_device               *mdt;
 171         struct coordinator              *cdt;
 172         int                              rc;
 173         ENTRY;
 174
 175         hsd = data;
 176         mdt = hsd->mti->mti_mdt;
 177         cdt = &mdt->mdt_coordinator;
 178
 179         larr = (struct llog_agent_req_rec *)hdr;
 180         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 181         switch (larr->arr_status) {
 182         case ARS_WAITING: {
 183                 int i;
 184                 struct hsm_scan_request *request;
 185
 186                 /* Are agents full? */
 187                 if (atomic_read(&cdt->cdt_request_count) >=
 188                     cdt->cdt_max_requests)
 189                         break;
 190
 191                 /* first search whether the request is found in the
 192                  * list we have built. */
 193                 request = NULL;
 194                 for (i = 0; i < hsd->request_cnt; i++) {
 195                         if (hsd->request[i].hal->hal_compound_id ==
 196                             larr->arr_compound_id) {
 197                                 request = &hsd->request[i];
 198                                 break;
 199                         }
 200                 }
 201
 202                 if (!request) {
 203                         struct hsm_action_list *hal;
 204
 205                         if (hsd->request_cnt == hsd->max_requests)
 206                                 /* Unknown request and no more room
 207                                  * for a new request. Continue to scan
 208                                  * to find other entries for already
 209                                  * existing requests.
 210                                  */
 211                                 RETURN(0);
 212
 213                         request = &hsd->request[hsd->request_cnt];
 214
 215                         /* allocates hai vector size just needs to be large
 216                          * enough */
 217                         request->hal_sz =
 218                                 sizeof(*request->hal) +
 219                                 cfs_size_round(MTI_NAME_MAXLEN+1) +
 220                                 2 * cfs_size_round(larr->arr_hai.hai_len);
 221                         OBD_ALLOC(hal, request->hal_sz);
 222                         if (!hal)
 223                                 RETURN(-ENOMEM);
 224                         hal->hal_version = HAL_VERSION;
 225                         strlcpy(hal->hal_fsname, hsd->fs_name,
 226                                 MTI_NAME_MAXLEN + 1);
 227                         hal->hal_compound_id = larr->arr_compound_id;
 228                         hal->hal_archive_id = larr->arr_archive_id;
 229                         hal->hal_flags = larr->arr_flags;
 230                         hal->hal_count = 0;
 231                         request->hal_used_sz = hal_size(hal);
 232                         request->hal = hal;
 233                         hsd->request_cnt++;
 234                         hai = hai_first(hal);
 235                 } else {
 236                         /* request is known */
 237                         /* we check if record archive num is the same as the
 238                          * known request, if not we will serve it in multiple
 239                          * time because we do not know if the agent can serve
 240                          * multiple backend
 241                          * a use case is a compound made of multiple restore
 242                          * where the files are not archived in the same backend
 243                          */
 244                         if (larr->arr_archive_id !=
 245                             request->hal->hal_archive_id)
 246                                 RETURN(0);
 247
 248                         if (request->hal_sz <
 249                             request->hal_used_sz +
 250                             cfs_size_round(larr->arr_hai.hai_len)) {
 251                                 /* Not enough room, need an extension */
 252                                 void *hal_buffer;
 253                                 int sz;
 254
 255                                 sz = 2 * request->hal_sz;
 256                                 OBD_ALLOC(hal_buffer, sz);
 257                                 if (!hal_buffer)
 258                                         RETURN(-ENOMEM);
 259                                 memcpy(hal_buffer, request->hal,
 260                                        request->hal_used_sz);
 261                                 OBD_FREE(request->hal,
 262                                          request->hal_sz);
 263                                 request->hal = hal_buffer;
 264                                 request->hal_sz = sz;
 265                         }
 266                         hai = hai_first(request->hal);
 267                         for (i = 0; i < request->hal->hal_count; i++)
 268                                 hai = hai_next(hai);
 269                 }
 270                 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 271                 hai->hai_cookie = larr->arr_hai.hai_cookie;
 272                 hai->hai_gid = larr->arr_hai.hai_gid;
 273
 274                 request->hal_used_sz += cfs_size_round(hai->hai_len);
 275                 request->hal->hal_count++;
 276
 277                 if (hai->hai_action != HSMA_CANCEL)
 278                         cdt_agent_record_hash_add(cdt, hai->hai_cookie,
 279                                                   llh->lgh_hdr->llh_cat_idx,
 280                                                   hdr->lrh_index);
 281                 break;
 282         }
 283         case ARS_STARTED: {
 284                 struct hsm_progress_kernel pgs;
 285                 struct cdt_agent_req *car;
 286                 cfs_time_t now = cfs_time_current_sec();
 287                 cfs_time_t last;
 288
 289                 /* we search for a running request
 290                  * error may happen if coordinator crashes or stopped
 291                  * with running request
 292                  */
 293                 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie);
 294                 if (car == NULL) {
 295                         last = larr->arr_req_change;
 296                 } else {
 297                         last = car->car_req_update;
 298                         mdt_cdt_put_request(car);
 299                 }
 300
 301                 /* test if request too long, if yes cancel it
 302                  * the same way the copy tool acknowledge a cancel request */
 303                 if (now <= last + cdt->cdt_active_req_timeout)
 304                         RETURN(0);
 305
 306                 dump_llog_agent_req_rec("request timed out, start cleaning",
 307                                         larr);
 308                 /* a too old cancel request just needs to be removed
 309                  * this can happen, if copy tool does not support
 310                  * cancel for other requests, we have to remove the
 311                  * running request and notify the copytool */
 312                 pgs.hpk_fid = larr->arr_hai.hai_fid;
 313                 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
 314                 pgs.hpk_extent = larr->arr_hai.hai_extent;
 315                 pgs.hpk_flags = HP_FLAG_COMPLETED;
 316                 pgs.hpk_errval = ENOSYS;
 317                 pgs.hpk_data_version = 0;
 318
 319                 /* update request state, but do not record in llog, to
 320                  * avoid deadlock on cdt_llog_lock */
 321                 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
 322                 if (rc)
 323                         CERROR("%s: cannot cleanup timed out request: "
 324                                DFID" for cookie %#llx action=%s\n",
 325                                mdt_obd_name(mdt),
 326                                PFID(&pgs.hpk_fid), pgs.hpk_cookie,
 327                                hsm_copytool_action2name(
 328                                        larr->arr_hai.hai_action));
 329
 330                 if (rc == -ENOENT) {
 331                         /* The request no longer exists, forget
 332                          * about it, and do not send a cancel request
 333                          * to the client, for which an error will be
 334                          * sent back, leading to an endless cycle of
 335                          * cancellation. */
 336                         cdt_agent_record_hash_del(cdt,
 337                                                   larr->arr_hai.hai_cookie);
 338                         RETURN(LLOG_DEL_RECORD);
 339                 }
 340
 341                 /* XXX A cancel request cannot be cancelled. */
 342                 if (larr->arr_hai.hai_action == HSMA_CANCEL)
 343                         RETURN(0);
 344
 345                 larr->arr_status = ARS_CANCELED;
 346                 larr->arr_req_change = now;
 347                 rc = llog_write(hsd->mti->mti_env, llh, hdr, hdr->lrh_index);
 348                 if (rc < 0)
 349                         CERROR("%s: cannot update agent log: rc = %d\n",
 350                                mdt_obd_name(mdt), rc);
 351                 break;
 352         }
 353         case ARS_FAILED:
 354         case ARS_CANCELED:
 355         case ARS_SUCCEED:
 356                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 357                     cfs_time_current_sec()) {
 358                         cdt_agent_record_hash_del(cdt,
 359                                                   larr->arr_hai.hai_cookie);
 360                         RETURN(LLOG_DEL_RECORD);
 361                 }
 362                 break;
 363         }
 364         RETURN(0);
 365 }
 366
 367 /**
 368  * create /proc entries for coordinator
 369  * \param mdt [IN]
 370  * \retval 0 success
 371  * \retval -ve failure
 372  */
 373 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 374 {
 375         struct coordinator      *cdt = &mdt->mdt_coordinator;
 376         int                      rc = 0;
 377         ENTRY;
 378
 379         /* init /proc entries, failure is not critical */
 380         cdt->cdt_proc_dir = lprocfs_register("hsm",
 381                                              mdt2obd_dev(mdt)->obd_proc_entry,
 382                                              lprocfs_mdt_hsm_vars, mdt);
 383         if (IS_ERR(cdt->cdt_proc_dir)) {
 384                 rc = PTR_ERR(cdt->cdt_proc_dir);
 385                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 386                        " rc=%d\n", mdt_obd_name(mdt), rc);
 387                 cdt->cdt_proc_dir = NULL;
 388                 RETURN(rc);
 389         }
 390
 391         RETURN(0);
 392 }
 393
 394 /**
 395  * remove /proc entries for coordinator
 396  * \param mdt [IN]
 397  */
 398 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
 399 {
 400         struct coordinator *cdt = &mdt->mdt_coordinator;
 401
 402         if (cdt->cdt_proc_dir != NULL)
 403                 lprocfs_remove(&cdt->cdt_proc_dir);
 404 }
 405
 406 /**
 407  * get vector of hsm cdt /proc vars
 408  * \param none
 409  * \retval var vector
 410  */
 411 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 412 {
 413         return lprocfs_mdt_hsm_vars;
 414 }
 415
 416 /* Release the ressource used by the coordinator. Called when the
 417  * coordinator is stopping. */
 418 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
 419 {
 420         struct coordinator              *cdt = &mdt->mdt_coordinator;
 421         struct cdt_agent_req            *car, *tmp1;
 422         struct hsm_agent                *ha, *tmp2;
 423         struct cdt_restore_handle       *crh, *tmp3;
 424         struct mdt_thread_info          *cdt_mti;
 425
 426         /* start cleaning */
 427         down_write(&cdt->cdt_request_lock);
 428         list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
 429                                  car_request_list) {
 430                 cfs_hash_del(cdt->cdt_request_cookie_hash,
 431                              &car->car_hai->hai_cookie,
 432                              &car->car_cookie_hash);
 433                 list_del(&car->car_request_list);
 434                 mdt_cdt_put_request(car);
 435         }
 436         up_write(&cdt->cdt_request_lock);
 437
 438         down_write(&cdt->cdt_agent_lock);
 439         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 440                 list_del(&ha->ha_list);
 441                 OBD_FREE_PTR(ha);
 442         }
 443         up_write(&cdt->cdt_agent_lock);
 444
 445         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 446         mutex_lock(&cdt->cdt_restore_lock);
 447         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
 448                 list_del(&crh->crh_list);
 449                 /* give back layout lock */
 450                 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
 451                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 452         }
 453         mutex_unlock(&cdt->cdt_restore_lock);
 454 }
 455
 456 /*
 457  * Coordinator state transition table, indexed on enum cdt_states, taking
 458  * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
 459  * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
 460  */
 461 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
 462         /* from -> to:    stopped init   running disable stopping */
 463         /* stopped */   { true,   true,  false,  false,  false },
 464         /* init */      { true,   false, true,   false,  false },
 465         /* running */   { false,  false, true,   true,   true },
 466         /* disable */   { false,  false, true,   true,   true },
 467         /* stopping */  { true,   false, false,  false,  false }
 468 };
 469
 470 /**
 471  * Change coordinator thread state
 472  * Some combinations are not valid, so catch them here.
 473  *
 474  * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
 475  * the transition was not possible.
 476  */
 477 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state,
 478                          enum cdt_states *old_state)
 479 {
 480         int rc;
 481         enum cdt_states state;
 482
 483         spin_lock(&cdt->cdt_state_lock);
 484
 485         state = cdt->cdt_state;
 486
 487         if (cdt_transition[state][new_state]) {
 488                 cdt->cdt_state = new_state;
 489                 spin_unlock(&cdt->cdt_state_lock);
 490                 if (old_state)
 491                         *old_state = state;
 492                 rc = 0;
 493         } else {
 494                 spin_unlock(&cdt->cdt_state_lock);
 495                 CDEBUG(D_HSM,
 496                        "unexpected coordinator transition, from=%s, to=%s\n",
 497                        cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
 498                 rc = -EINVAL;
 499         }
 500
 501         return rc;
 502 }
 503
 504 /**
 505  * coordinator thread
 506  * \param data [IN] obd device
 507  * \retval 0 success
 508  * \retval -ve failure
 509  */
 510 static int mdt_coordinator(void *data)
 511 {
 512         struct hsm_thread_data  *thread_data = data;
 513         struct mdt_thread_info  *mti = thread_data->cdt_mti;
 514         struct mdt_device       *mdt = mti->mti_mdt;
 515         struct coordinator      *cdt = &mdt->mdt_coordinator;
 516         struct hsm_scan_data     hsd = { NULL };
 517         time64_t                 wait_event_time = 1 * HZ;
 518         time64_t                 last_housekeeping = 0;
 519         int                      rc = 0;
 520         int                      request_sz;
 521         ENTRY;
 522
 523         /* set up hsd->request and max_requests */
 524         hsd.max_requests = cdt->cdt_max_requests;
 525         request_sz = hsd.max_requests * sizeof(*hsd.request);
 526         hsd.request = thread_data->request;
 527
 528         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 529                mdt_obd_name(mdt), current_pid());
 530
 531         hsd.mti = mti;
 532         obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
 533
 534         set_cdt_state(cdt, CDT_RUNNING, NULL);
 535
 536         /* Inform mdt_hsm_cdt_start(). */
 537         wake_up_all(&cdt->cdt_waitq);
 538
 539         while (1) {
 540                 int i;
 541
 542                 /* Limit execution of the expensive requests traversal
 543                  * to at most every "wait_event_time" jiffies. This prevents
 544                  * repeatedly locking/unlocking the catalog for each request
 545                  * and preventing other HSM operations from happening */
 546                 wait_event_interruptible_timeout(cdt->cdt_waitq,
 547                                                  kthread_should_stop(),
 548                                                  wait_event_time);
 549
 550                 CDEBUG(D_HSM, "coordinator resumes\n");
 551
 552                 if (kthread_should_stop()) {
 553                         CDEBUG(D_HSM, "Coordinator stops\n");
 554                         rc = 0;
 555                         break;
 556                 }
 557
 558                 /* if coordinator is suspended continue to wait */
 559                 if (cdt->cdt_state == CDT_DISABLE) {
 560                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 561                         continue;
 562                 }
 563
 564                 /* If no event, and no housekeeping to do, continue to
 565                  * wait. */
 566                 if (last_housekeeping + cdt->cdt_loop_period <= get_seconds())
 567                         last_housekeeping = get_seconds();
 568                 else if (!cdt->cdt_event)
 569                         continue;
 570
 571                 cdt->cdt_event = false;
 572
 573                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 574
 575                 if (hsd.max_requests != cdt->cdt_max_requests) {
 576                         /* cdt_max_requests has changed,
 577                          * we need to allocate a new buffer
 578                          */
 579                         struct hsm_scan_request *tmp = NULL;
 580                         int max_requests = cdt->cdt_max_requests;
 581                         OBD_ALLOC_LARGE(tmp, max_requests *
 582                                         sizeof(struct hsm_scan_request));
 583                         if (!tmp) {
 584                                 CERROR("Failed to resize request buffer, "
 585                                        "keeping it at %d\n",
 586                                        hsd.max_requests);
 587                                 cdt->cdt_max_requests = hsd.max_requests;
 588                         } else {
 589                                 OBD_FREE_LARGE(hsd.request, request_sz);
 590                                 hsd.max_requests = max_requests;
 591                                 request_sz = hsd.max_requests *
 592                                         sizeof(struct hsm_scan_request);
 593                                 hsd.request = tmp;
 594                         }
 595                 }
 596
 597                 hsd.request_cnt = 0;
 598
 599                 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
 600                                       &hsd, 0, 0, WRITE);
 601                 if (rc < 0)
 602                         goto clean_cb_alloc;
 603
 604                 CDEBUG(D_HSM, "found %d requests to send\n", hsd.request_cnt);
 605
 606                 if (list_empty(&cdt->cdt_agents)) {
 607                         CDEBUG(D_HSM, "no agent available, "
 608                                       "coordinator sleeps\n");
 609                         goto clean_cb_alloc;
 610                 }
 611
 612                 /* here hsd contains a list of requests to be started */
 613                 for (i = 0; i < hsd.request_cnt; i++) {
 614                         struct hsm_scan_request *request = &hsd.request[i];
 615                         struct hsm_action_list  *hal = request->hal;
 616                         struct hsm_action_item  *hai;
 617                         __u64                   *cookies;
 618                         int                      sz, j;
 619                         enum agent_req_status    status;
 620
 621                         /* still room for work ? */
 622                         if (atomic_read(&cdt->cdt_request_count) >=
 623                             cdt->cdt_max_requests)
 624                                 break;
 625
 626                         rc = mdt_hsm_agent_send(mti, hal, 0);
 627                         /* if failure, we suppose it is temporary
 628                          * if the copy tool failed to do the request
 629                          * it has to use hsm_progress
 630                          */
 631                         status = (rc ? ARS_WAITING : ARS_STARTED);
 632
 633                         /* set up cookie vector to set records status
 634                          * after copy tools start or failed
 635                          */
 636                         sz = hal->hal_count * sizeof(__u64);
 637                         OBD_ALLOC(cookies, sz);
 638                         if (cookies == NULL)
 639                                 continue;
 640
 641                         hai = hai_first(hal);
 642                         for (j = 0; j < hal->hal_count; j++) {
 643                                 cookies[j] = hai->hai_cookie;
 644                                 hai = hai_next(hai);
 645                         }
 646
 647                         rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
 648                                                      hal->hal_count, status);
 649                         if (rc)
 650                                 CERROR("%s: mdt_agent_record_update() failed, "
 651                                        "rc=%d, cannot update status to %s "
 652                                        "for %d cookies\n",
 653                                        mdt_obd_name(mdt), rc,
 654                                        agent_req_status2name(status),
 655                                        hal->hal_count);
 656
 657                         OBD_FREE(cookies, sz);
 658                 }
 659 clean_cb_alloc:
 660                 /* free hal allocated by callback */
 661                 for (i = 0; i < hsd.request_cnt; i++) {
 662                         struct hsm_scan_request *request = &hsd.request[i];
 663
 664                         OBD_FREE(request->hal, request->hal_sz);
 665                 }
 666         }
 667
 668         if (hsd.request)
 669                 OBD_FREE_LARGE(hsd.request, request_sz);
 670
 671         mdt_hsm_cdt_cleanup(mdt);
 672
 673         if (rc != 0)
 674                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 675                        mdt_obd_name(mdt), current_pid(), rc);
 676         else
 677                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 678                               " no error\n",
 679                        mdt_obd_name(mdt), current_pid());
 680
 681         RETURN(rc);
 682 }
 683
 684 /**
 685  * lookup a restore handle by FID
 686  * caller needs to hold cdt_restore_lock
 687  * \param cdt [IN] coordinator
 688  * \param fid [IN] FID
 689  * \retval cdt_restore_handle found
 690  * \retval NULL not found
 691  */
 692 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
 693                                                        const struct lu_fid *fid)
 694 {
 695         struct cdt_restore_handle       *crh;
 696         ENTRY;
 697
 698         list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
 699                 if (lu_fid_eq(&crh->crh_fid, fid))
 700                         RETURN(crh);
 701         }
 702         RETURN(NULL);
 703 }
 704
 705 /**
 706  * data passed to llog_cat_process() callback
 707  * to scan requests and take actions
 708  */
 709 struct hsm_restore_data {
 710         struct mdt_thread_info  *hrd_mti;
 711 };
 712
 713 /**
 714  *  llog_cat_process() callback, used to:
 715  *  - find restore request and allocate the restore handle
 716  * \param env [IN] environment
 717  * \param llh [IN] llog handle
 718  * \param hdr [IN] llog record
 719  * \param data [IN/OUT] cb data = struct hsm_restore_data
 720  * \retval 0 success
 721  * \retval -ve failure
 722  */
 723 static int hsm_restore_cb(const struct lu_env *env,
 724                           struct llog_handle *llh,
 725                           struct llog_rec_hdr *hdr, void *data)
 726 {
 727         struct llog_agent_req_rec       *larr;
 728         struct hsm_restore_data         *hrd;
 729         struct cdt_restore_handle       *crh;
 730         struct hsm_action_item          *hai;
 731         struct mdt_thread_info          *mti;
 732         struct coordinator              *cdt;
 733         struct mdt_object               *child;
 734         int rc;
 735         ENTRY;
 736
 737         hrd = data;
 738         mti = hrd->hrd_mti;
 739         cdt = &mti->mti_mdt->mdt_coordinator;
 740
 741         larr = (struct llog_agent_req_rec *)hdr;
 742         hai = &larr->arr_hai;
 743         if (hai->hai_cookie > cdt->cdt_last_cookie)
 744                 /* update the cookie to avoid collision */
 745                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 746
 747         if (hai->hai_action != HSMA_RESTORE ||
 748             agent_req_in_final_state(larr->arr_status))
 749                 RETURN(0);
 750
 751         /* restore request not in a final state */
 752
 753         /* force replay of restore requests left in started state from previous
 754          * CDT context, to be canceled later if finally found to be incompatible
 755          * when being re-started */
 756         if (larr->arr_status == ARS_STARTED) {
 757                 larr->arr_status = ARS_WAITING;
 758                 larr->arr_req_change = cfs_time_current_sec();
 759                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 760                 if (rc != 0)
 761                         GOTO(out, rc);
 762         }
 763
 764         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 765         if (crh == NULL)
 766                 RETURN(-ENOMEM);
 767
 768         crh->crh_fid = hai->hai_fid;
 769         /* in V1 all file is restored
 770         crh->extent.start = hai->hai_extent.offset;
 771         crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
 772         */
 773         crh->crh_extent.start = 0;
 774         crh->crh_extent.end = hai->hai_extent.length;
 775         /* get the layout lock */
 776         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 777         child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 778                                      MDS_INODELOCK_LAYOUT);
 779         if (IS_ERR(child))
 780                 GOTO(out, rc = PTR_ERR(child));
 781
 782         rc = 0;
 783         /* we choose to not keep a reference
 784          * on the object during the restore time which can be very long */
 785         mdt_object_put(mti->mti_env, child);
 786
 787         mutex_lock(&cdt->cdt_restore_lock);
 788         list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
 789         mutex_unlock(&cdt->cdt_restore_lock);
 790
 791 out:
 792         RETURN(rc);
 793 }
 794
 795 /**
 796  * restore coordinator state at startup
 797  * the goal is to take a layout lock for each registered restore request
 798  * \param mti [IN] context
 799  */
 800 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 801 {
 802         struct hsm_restore_data  hrd;
 803         int                      rc;
 804         ENTRY;
 805
 806         hrd.hrd_mti = mti;
 807
 808         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
 809                               0, 0, WRITE);
 810
 811         RETURN(rc);
 812 }
 813
 814 static int hsm_init_ucred(struct lu_ucred *uc)
 815 {
 816         ENTRY;
 817
 818         uc->uc_valid = UCRED_OLD;
 819         uc->uc_o_uid = 0;
 820         uc->uc_o_gid = 0;
 821         uc->uc_o_fsuid = 0;
 822         uc->uc_o_fsgid = 0;
 823         uc->uc_uid = 0;
 824         uc->uc_gid = 0;
 825         uc->uc_fsuid = 0;
 826         uc->uc_fsgid = 0;
 827         uc->uc_suppgids[0] = -1;
 828         uc->uc_suppgids[1] = -1;
 829         uc->uc_cap = CFS_CAP_FS_MASK;
 830         uc->uc_umask = 0777;
 831         uc->uc_ginfo = NULL;
 832         uc->uc_identity = NULL;
 833
 834         RETURN(0);
 835 }
 836
 837 /**
 838  * initialize coordinator struct
 839  * \param mdt [IN] device
 840  * \retval 0 success
 841  * \retval -ve failure
 842  */
 843 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 844 {
 845         struct coordinator      *cdt = &mdt->mdt_coordinator;
 846         struct mdt_thread_info  *cdt_mti = NULL;
 847         int                      rc;
 848         ENTRY;
 849
 850         init_waitqueue_head(&cdt->cdt_waitq);
 851         init_rwsem(&cdt->cdt_llog_lock);
 852         init_rwsem(&cdt->cdt_agent_lock);
 853         init_rwsem(&cdt->cdt_request_lock);
 854         mutex_init(&cdt->cdt_restore_lock);
 855         spin_lock_init(&cdt->cdt_state_lock);
 856         set_cdt_state(cdt, CDT_STOPPED, NULL);
 857
 858         INIT_LIST_HEAD(&cdt->cdt_request_list);
 859         INIT_LIST_HEAD(&cdt->cdt_agents);
 860         INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 861
 862         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
 863                                                        CFS_HASH_BITS_MIN,
 864                                                        CFS_HASH_BITS_MAX,
 865                                                        CFS_HASH_BKT_BITS,
 866                                                        0 /* extra bytes */,
 867                                                        CFS_HASH_MIN_THETA,
 868                                                        CFS_HASH_MAX_THETA,
 869                                                 &cdt_request_cookie_hash_ops,
 870                                                        CFS_HASH_DEFAULT);
 871         if (cdt->cdt_request_cookie_hash == NULL)
 872                 RETURN(-ENOMEM);
 873
 874         cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
 875                                                      CFS_HASH_BITS_MIN,
 876                                                      CFS_HASH_BITS_MAX,
 877                                                      CFS_HASH_BKT_BITS,
 878                                                      0 /* extra bytes */,
 879                                                      CFS_HASH_MIN_THETA,
 880                                                      CFS_HASH_MAX_THETA,
 881                                                      &cdt_agent_record_hash_ops,
 882                                                      CFS_HASH_DEFAULT);
 883         if (cdt->cdt_agent_record_hash == NULL)
 884                 GOTO(out_request_cookie_hash, rc = -ENOMEM);
 885
 886         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 887         if (rc < 0)
 888                 GOTO(out_agent_record_hash, rc);
 889
 890         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
 891         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
 892         if (rc < 0)
 893                 GOTO(out_env, rc);
 894
 895         lu_context_enter(&cdt->cdt_session);
 896         cdt->cdt_env.le_ses = &cdt->cdt_session;
 897
 898         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 899         LASSERT(cdt_mti != NULL);
 900
 901         cdt_mti->mti_env = &cdt->cdt_env;
 902         cdt_mti->mti_mdt = mdt;
 903
 904         hsm_init_ucred(mdt_ucred(cdt_mti));
 905
 906         /* default values for /proc tunnables
 907          * can be override by MGS conf */
 908         cdt->cdt_default_archive_id = 1;
 909         cdt->cdt_grace_delay = 60;
 910         cdt->cdt_loop_period = 10;
 911         cdt->cdt_max_requests = 3;
 912         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 913         cdt->cdt_active_req_timeout = 3600;
 914
 915         /* Initialize cdt_compound_id here to allow its usage for
 916          * delayed requests from RAoLU policy */
 917         atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
 918
 919         /* by default do not remove archives on last unlink */
 920         cdt->cdt_remove_archive_on_last_unlink = false;
 921
 922         RETURN(0);
 923
 924 out_env:
 925         lu_env_fini(&cdt->cdt_env);
 926 out_agent_record_hash:
 927         cfs_hash_putref(cdt->cdt_agent_record_hash);
 928         cdt->cdt_agent_record_hash = NULL;
 929 out_request_cookie_hash:
 930         cfs_hash_putref(cdt->cdt_request_cookie_hash);
 931         cdt->cdt_request_cookie_hash = NULL;
 932
 933         return rc;
 934 }
 935
 936 /**
 937  * free a coordinator thread
 938  * \param mdt [IN] device
 939  */
 940 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
 941 {
 942         struct coordinator *cdt = &mdt->mdt_coordinator;
 943         ENTRY;
 944
 945         lu_context_exit(cdt->cdt_env.le_ses);
 946         lu_context_fini(cdt->cdt_env.le_ses);
 947
 948         lu_env_fini(&cdt->cdt_env);
 949
 950         cfs_hash_putref(cdt->cdt_agent_record_hash);
 951         cdt->cdt_agent_record_hash = NULL;
 952
 953         cfs_hash_putref(cdt->cdt_request_cookie_hash);
 954         cdt->cdt_request_cookie_hash = NULL;
 955
 956         RETURN(0);
 957 }
 958
 959 /**
 960  * start a coordinator thread
 961  * \param mdt [IN] device
 962  * \retval 0 success
 963  * \retval -ve failure
 964  */
 965 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
 966 {
 967         struct coordinator      *cdt = &mdt->mdt_coordinator;
 968         int                      rc;
 969         void                    *ptr;
 970         struct task_struct      *task;
 971         int                      request_sz;
 972         struct hsm_thread_data   thread_data;
 973         ENTRY;
 974
 975         /* functions defined but not yet used
 976          * this avoid compilation warning
 977          */
 978         ptr = dump_requests;
 979
 980         rc = set_cdt_state(cdt, CDT_INIT, NULL);
 981         if (rc) {
 982                 CERROR("%s: Coordinator already started or stopping\n",
 983                        mdt_obd_name(mdt));
 984                 RETURN(-EALREADY);
 985         }
 986
 987         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
 988         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 989
 990         /* just need to be larger than previous one */
 991         /* cdt_last_cookie is protected by cdt_llog_lock */
 992         cdt->cdt_last_cookie = cfs_time_current_sec();
 993         atomic_set(&cdt->cdt_request_count, 0);
 994         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
 995         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
 996         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
 997
 998         /* to avoid deadlock when start is made through /proc
 999          * /proc entries are created by the coordinator thread */
1000
1001         /* set up list of started restore requests */
1002         thread_data.cdt_mti =
1003                 lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1004         rc = mdt_hsm_pending_restore(thread_data.cdt_mti);
1005         if (rc)
1006                 CERROR("%s: cannot take the layout locks needed"
1007                        " for registered restore: %d\n",
1008                        mdt_obd_name(mdt), rc);
1009
1010         if (mdt->mdt_bottom->dd_rdonly)
1011                 RETURN(0);
1012
1013         /* Allocate the initial hsd.request[] vector*/
1014         request_sz = cdt->cdt_max_requests * sizeof(struct hsm_scan_request);
1015         OBD_ALLOC_LARGE(thread_data.request, request_sz);
1016         if (!thread_data.request) {
1017                 set_cdt_state(cdt, CDT_STOPPED, NULL);
1018                 RETURN(-ENOMEM);
1019         }
1020
1021         task = kthread_run(mdt_coordinator, &thread_data, "hsm_cdtr");
1022         if (IS_ERR(task)) {
1023                 rc = PTR_ERR(task);
1024                 set_cdt_state(cdt, CDT_STOPPED, NULL);
1025                 OBD_FREE(thread_data.request, request_sz);
1026                 CERROR("%s: error starting coordinator thread: %d\n",
1027                        mdt_obd_name(mdt), rc);
1028         } else {
1029                 cdt->cdt_task = task;
1030                 wait_event(cdt->cdt_waitq,
1031                            cdt->cdt_state != CDT_INIT);
1032                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1033                        mdt_obd_name(mdt));
1034                 rc = 0;
1035         }
1036
1037         RETURN(rc);
1038 }
1039
1040 /**
1041  * stop a coordinator thread
1042  * \param mdt [IN] device
1043  */
1044 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1045 {
1046         struct coordinator *cdt = &mdt->mdt_coordinator;
1047         int rc;
1048
1049         ENTRY;
1050         /* stop coordinator thread */
1051         rc = set_cdt_state(cdt, CDT_STOPPING, NULL);
1052         if (rc == 0) {
1053                 kthread_stop(cdt->cdt_task);
1054                 cdt->cdt_task = NULL;
1055                 set_cdt_state(cdt, CDT_STOPPED, NULL);
1056         }
1057
1058         RETURN(rc);
1059 }
1060
1061 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1062                               const struct lu_fid *fid,
1063                               u32 archive_id)
1064 {
1065         struct mdt_object *obj;
1066         struct md_hsm mh;
1067         int rc;
1068
1069         obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1070         if (IS_ERR(obj))
1071                 GOTO(out, rc = PTR_ERR(obj));
1072
1073         if (mh.mh_flags & HS_EXISTS &&
1074             mh.mh_arch_id == archive_id)
1075                 GOTO(out_obj, rc = 0);
1076
1077         mh.mh_flags |= HS_EXISTS;
1078         mh.mh_arch_id = archive_id;
1079         rc = mdt_hsm_attr_set(mti, obj, &mh);
1080
1081 out_obj:
1082         mdt_object_put(mti->mti_env, obj);
1083 out:
1084         return rc;
1085 }
1086
1087 /**
1088  * register all requests from an hal in the memory list
1089  * \param mti [IN] context
1090  * \param hal [IN] request
1091  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1092  *  which is running the CT
1093  * \retval 0 success
1094  * \retval -ve failure
1095  */
1096 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1097                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1098 {
1099         struct mdt_device       *mdt = mti->mti_mdt;
1100         struct coordinator      *cdt = &mdt->mdt_coordinator;
1101         struct hsm_action_item  *hai;
1102         int                      rc = 0, i;
1103         ENTRY;
1104
1105         /* register request in memory list */
1106         hai = hai_first(hal);
1107         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1108                 struct cdt_agent_req *car;
1109
1110                 /* in case of a cancel request, we first mark the ondisk
1111                  * record of the request we want to stop as canceled
1112                  * this does not change the cancel record
1113                  * it will be done when updating the request status
1114                  */
1115                 if (hai->hai_action == HSMA_CANCEL) {
1116                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1117                                                      &hai->hai_cookie,
1118                                                      1, ARS_CANCELED);
1119                         if (rc) {
1120                                 CERROR("%s: mdt_agent_record_update() failed, "
1121                                        "rc=%d, cannot update status to %s "
1122                                        "for cookie %#llx\n",
1123                                        mdt_obd_name(mdt), rc,
1124                                        agent_req_status2name(ARS_CANCELED),
1125                                        hai->hai_cookie);
1126                                 GOTO(out, rc);
1127                         }
1128
1129                         /* find the running request to set it canceled */
1130                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1131                         if (car != NULL) {
1132                                 car->car_canceled = 1;
1133                                 /* uuid has to be changed to the one running the
1134                                 * request to cancel */
1135                                 *uuid = car->car_uuid;
1136                                 mdt_cdt_put_request(car);
1137                         }
1138                         /* no need to memorize cancel request
1139                          * this also avoid a deadlock when we receive
1140                          * a purge all requests command
1141                          */
1142                         continue;
1143                 }
1144
1145                 if (hai->hai_action == HSMA_ARCHIVE) {
1146                         rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1147                                                 hal->hal_archive_id);
1148                         if (rc == -ENOENT)
1149                                 continue;
1150                         else if (rc < 0)
1151                                 GOTO(out, rc);
1152                 }
1153
1154                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1155                                             hal->hal_archive_id, hal->hal_flags,
1156                                             uuid, hai);
1157                 if (IS_ERR(car))
1158                         GOTO(out, rc = PTR_ERR(car));
1159
1160                 rc = mdt_cdt_add_request(cdt, car);
1161                 if (rc != 0)
1162                         mdt_cdt_free_request(car);
1163         }
1164 out:
1165         RETURN(rc);
1166 }
1167
1168 /**
1169  * swap layouts between 2 fids
1170  * \param mti [IN] context
1171  * \param obj [IN]
1172  * \param dfid [IN]
1173  * \param mh_common [IN] MD HSM
1174  */
1175 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1176                             struct mdt_object *obj, const struct lu_fid *dfid,
1177                             struct md_hsm *mh_common)
1178 {
1179         struct mdt_object       *dobj;
1180         struct mdt_lock_handle  *dlh;
1181         int                      rc;
1182         ENTRY;
1183
1184         if (!mdt_object_exists(obj))
1185                 GOTO(out, rc = -ENOENT);
1186
1187         /* we already have layout lock on obj so take only
1188          * on dfid */
1189         dlh = &mti->mti_lh[MDT_LH_OLD];
1190         mdt_lock_reg_init(dlh, LCK_EX);
1191         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1192         if (IS_ERR(dobj))
1193                 GOTO(out, rc = PTR_ERR(dobj));
1194
1195         /* if copy tool closes the volatile before sending the final
1196          * progress through llapi_hsm_copy_end(), all the objects
1197          * are removed and mdd_swap_layout LBUG */
1198         if (!mdt_object_exists(dobj)) {
1199                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1200                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1201                 GOTO(out_dobj, rc = -ENOENT);
1202         }
1203         /* Since we only handle restores here, unconditionally use
1204          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1205          * be preserved in case of failure during swap_layout and not
1206          * leave a file in an intermediate but incoherent state.
1207          * But need to setup HSM xattr of data FID before, reuse
1208          * mti and mh presets for FID in hsm_cdt_request_completed(),
1209          * only need to clear RELEASED and DIRTY.
1210          */
1211         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1212         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1213         if (rc == 0)
1214                 rc = mo_swap_layouts(mti->mti_env,
1215                                      mdt_object_child(obj),
1216                                      mdt_object_child(dobj),
1217                                      SWAP_LAYOUTS_MDS_HSM);
1218
1219 out_dobj:
1220         mdt_object_unlock_put(mti, dobj, dlh, 1);
1221 out:
1222         RETURN(rc);
1223 }
1224
1225 /**
1226  * update status of a completed request
1227  * \param mti [IN] context
1228  * \param pgs [IN] progress of the copy tool
1229  * \param update_record [IN] update llog record
1230  * \retval 0 success
1231  * \retval -ve failure
1232  */
1233 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1234                                      struct hsm_progress_kernel *pgs,
1235                                      const struct cdt_agent_req *car,
1236                                      enum agent_req_status *status)
1237 {
1238         const struct lu_env     *env = mti->mti_env;
1239         struct mdt_device       *mdt = mti->mti_mdt;
1240         struct coordinator      *cdt = &mdt->mdt_coordinator;
1241         struct mdt_object       *obj = NULL;
1242         int                      cl_flags = 0, rc = 0;
1243         struct md_hsm            mh;
1244         bool                     is_mh_changed;
1245         bool                     need_changelog = true;
1246         ENTRY;
1247
1248         /* default is to retry */
1249         *status = ARS_WAITING;
1250
1251         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1252          * if error/removed continue anyway to get correct reporting done */
1253         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1254         /* we will update MD HSM only if needed */
1255         is_mh_changed = false;
1256
1257         /* no need to change mh->mh_arch_id
1258          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1259          */
1260         if (pgs->hpk_errval != 0) {
1261                 switch (pgs->hpk_errval) {
1262                 case ENOSYS:
1263                         /* the copy tool does not support cancel
1264                          * so the cancel request is failed
1265                          * As we cannot distinguish a cancel progress
1266                          * from another action progress (they have the
1267                          * same cookie), we suppose here the CT returns
1268                          * ENOSYS only if does not support cancel
1269                          */
1270                         /* this can also happen when cdt calls it to
1271                          * for a timed out request */
1272                         *status = ARS_FAILED;
1273                         /* to have a cancel event in changelog */
1274                         pgs->hpk_errval = ECANCELED;
1275                         break;
1276                 case ECANCELED:
1277                         /* the request record has already been set to
1278                          * ARS_CANCELED, this set the cancel request
1279                          * to ARS_SUCCEED */
1280                         *status = ARS_SUCCEED;
1281                         break;
1282                 default:
1283                         /* retry only if current policy or requested, and
1284                          * object is not on error/removed */
1285                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1286                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1287                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1288                         break;
1289                 }
1290
1291                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1292                         CERROR("%s: Request %#llx on "DFID
1293                                " failed, error code %d too large\n",
1294                                mdt_obd_name(mdt),
1295                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1296                                pgs->hpk_errval);
1297                         hsm_set_cl_error(&cl_flags,
1298                                          CLF_HSM_ERROVERFLOW);
1299                         rc = -EINVAL;
1300                 } else {
1301                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1302                 }
1303
1304                 switch (car->car_hai->hai_action) {
1305                 case HSMA_ARCHIVE:
1306                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1307                         break;
1308                 case HSMA_RESTORE:
1309                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1310                         break;
1311                 case HSMA_REMOVE:
1312                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1313                         break;
1314                 case HSMA_CANCEL:
1315                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1316                         CERROR("%s: Failed request %#llx on "DFID
1317                                " cannot be a CANCEL\n",
1318                                mdt_obd_name(mdt),
1319                                pgs->hpk_cookie,
1320                                PFID(&pgs->hpk_fid));
1321                         break;
1322                 default:
1323                         CERROR("%s: Failed request %#llx on "DFID
1324                                " %d is an unknown action\n",
1325                                mdt_obd_name(mdt),
1326                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1327                                car->car_hai->hai_action);
1328                         rc = -EINVAL;
1329                         break;
1330                 }
1331         } else {
1332                 *status = ARS_SUCCEED;
1333                 switch (car->car_hai->hai_action) {
1334                 case HSMA_ARCHIVE:
1335                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1336                         /* set ARCHIVE keep EXIST and clear LOST and
1337                          * DIRTY */
1338                         mh.mh_arch_ver = pgs->hpk_data_version;
1339                         mh.mh_flags |= HS_ARCHIVED;
1340                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1341                         is_mh_changed = true;
1342                         break;
1343                 case HSMA_RESTORE:
1344                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1345
1346                         /* do not clear RELEASED and DIRTY here
1347                          * this will occur in hsm_swap_layouts()
1348                          */
1349
1350                         /* Restoring has changed the file version on
1351                          * disk. */
1352                         mh.mh_arch_ver = pgs->hpk_data_version;
1353                         is_mh_changed = true;
1354                         break;
1355                 case HSMA_REMOVE:
1356                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1357                         /* clear ARCHIVED EXISTS and LOST */
1358                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1359                         is_mh_changed = true;
1360                         break;
1361                 case HSMA_CANCEL:
1362                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1363                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1364                                mdt_obd_name(mdt),
1365                                pgs->hpk_cookie,
1366                                PFID(&pgs->hpk_fid));
1367                         break;
1368                 default:
1369                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1370                                mdt_obd_name(mdt),
1371                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1372                                car->car_hai->hai_action);
1373                         rc = -EINVAL;
1374                         break;
1375                 }
1376         }
1377
1378         /* rc != 0 means error when analysing action, it may come from
1379          * a crasy CT no need to manage DIRTY
1380          * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1381          * filled
1382          */
1383         if (rc == 0 && !IS_ERR(obj))
1384                 hsm_set_cl_flags(&cl_flags,
1385                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1386
1387         /* unlock is done later, after layout lock management */
1388         if (is_mh_changed && !IS_ERR(obj))
1389                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1390
1391         /* we give back layout lock only if restore was successful or
1392          * if no retry will be attempted and if object is still alive,
1393          * in other cases we just unlock the object */
1394         if (car->car_hai->hai_action == HSMA_RESTORE) {
1395                 struct cdt_restore_handle       *crh;
1396
1397                 /* restore in data FID done, we swap the layouts
1398                  * only if restore is successful */
1399                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1400                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1401                                               &mh);
1402                         if (rc) {
1403                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1404                                         *status = ARS_FAILED;
1405                                 pgs->hpk_errval = -rc;
1406                         }
1407                 }
1408                 /* we have to retry, so keep layout lock */
1409                 if (*status == ARS_WAITING)
1410                         GOTO(out, rc);
1411
1412                 /* restore special case, need to create ChangeLog record
1413                  * before to give back layout lock to avoid concurrent
1414                  * file updater to post out of order ChangeLog */
1415                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1416                              &car->car_hai->hai_fid);
1417                 need_changelog = false;
1418
1419                 /* give back layout lock */
1420                 mutex_lock(&cdt->cdt_restore_lock);
1421                 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1422                 if (crh != NULL)
1423                         list_del(&crh->crh_list);
1424                 mutex_unlock(&cdt->cdt_restore_lock);
1425                 /* Just give back layout lock, we keep the reference
1426                  * which is given back later with the lock for HSM
1427                  * flags.
1428                  * XXX obj may be invalid so we do not pass it. */
1429                 if (crh != NULL)
1430                         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
1431
1432                 if (crh != NULL)
1433                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1434         }
1435
1436         GOTO(out, rc);
1437
1438 out:
1439         /* always add a ChangeLog record */
1440         if (need_changelog)
1441                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1442                              &car->car_hai->hai_fid);
1443
1444         if (!IS_ERR(obj))
1445                 mdt_object_put(mti->mti_env, obj);
1446
1447         RETURN(rc);
1448 }
1449
1450 /**
1451  * update status of a request
1452  * \param mti [IN] context
1453  * \param pgs [IN] progress of the copy tool
1454  * \param update_record [IN] update llog record
1455  * \retval 0 success
1456  * \retval -ve failure
1457  */
1458 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1459                                  struct hsm_progress_kernel *pgs,
1460                                  const int update_record)
1461 {
1462         struct mdt_device       *mdt = mti->mti_mdt;
1463         struct coordinator      *cdt = &mdt->mdt_coordinator;
1464         struct cdt_agent_req    *car;
1465         int                      rc = 0;
1466         ENTRY;
1467
1468         /* no coordinator started, so we cannot serve requests */
1469         if (cdt->cdt_state == CDT_STOPPED)
1470                 RETURN(-EAGAIN);
1471
1472         /* first do sanity checks */
1473         car = mdt_cdt_update_request(cdt, pgs);
1474         if (IS_ERR(car)) {
1475                 CERROR("%s: Cannot find running request for cookie %#llx"
1476                        " on fid="DFID"\n",
1477                        mdt_obd_name(mdt),
1478                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1479
1480                 RETURN(PTR_ERR(car));
1481         }
1482
1483         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1484                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1485                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1486                       hsm_copytool_action2name(car->car_hai->hai_action),
1487                       pgs->hpk_flags, pgs->hpk_errval,
1488                       PFID(&car->car_hai->hai_fid),
1489                       PFID(&car->car_hai->hai_dfid));
1490
1491         /* progress is done on FID or data FID depending of the action and
1492          * of the copy progress */
1493         /* for restore progress is used to send back the data FID to cdt */
1494         if (car->car_hai->hai_action == HSMA_RESTORE &&
1495             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1496                 car->car_hai->hai_dfid = pgs->hpk_fid;
1497
1498         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1499              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1500             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1501              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1502                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1503                        " does not match request FID "DFID" nor data FID "
1504                        DFID"\n",
1505                        mdt_obd_name(mdt),
1506                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1507                        PFID(&car->car_hai->hai_fid),
1508                        PFID(&car->car_hai->hai_dfid));
1509                 GOTO(out, rc = -EINVAL);
1510         }
1511
1512         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1513                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1514                        " is not coherent (err=%d and not completed"
1515                        " (flags=%d))\n",
1516                        mdt_obd_name(mdt),
1517                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1518                        hsm_copytool_action2name(car->car_hai->hai_action),
1519                        pgs->hpk_errval, pgs->hpk_flags);
1520                 GOTO(out, rc = -EINVAL);
1521         }
1522
1523         /* now progress is valid */
1524
1525         /* we use a root like ucred */
1526         hsm_init_ucred(mdt_ucred(mti));
1527
1528         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1529                 enum agent_req_status    status;
1530
1531                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1532
1533                 CDEBUG(D_HSM, "%s record: fid="DFID" cookie=%#llx action=%s "
1534                               "status=%s\n",
1535                        update_record ? "Updating" : "Not updating",
1536                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1537                        hsm_copytool_action2name(car->car_hai->hai_action),
1538                        agent_req_status2name(status));
1539
1540                 /* update record first (LU-9075) */
1541                 if (update_record) {
1542                         int rc1;
1543
1544                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1545                                                      &pgs->hpk_cookie, 1,
1546                                                      status);
1547                         if (rc1)
1548                                 CERROR("%s: mdt_agent_record_update() failed,"
1549                                        " rc=%d, cannot update status to %s"
1550                                        " for cookie %#llx\n",
1551                                        mdt_obd_name(mdt), rc1,
1552                                        agent_req_status2name(status),
1553                                        pgs->hpk_cookie);
1554                         rc = (rc != 0 ? rc : rc1);
1555                 }
1556
1557                 /* then remove request from memory list (LU-9075) */
1558                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1559
1560                 /* ct has completed a request, so a slot is available,
1561                  * signal the coordinator to find new work */
1562                 mdt_hsm_cdt_event(cdt);
1563         } else {
1564                 /* if copytool send a progress on a canceled request
1565                  * we inform copytool it should stop
1566                  */
1567                 if (car->car_canceled == 1)
1568                         rc = -ECANCELED;
1569         }
1570         GOTO(out, rc);
1571
1572 out:
1573         /* remove ref got from mdt_cdt_update_request() */
1574         mdt_cdt_put_request(car);
1575
1576         return rc;
1577 }
1578
1579
1580 /**
1581  * data passed to llog_cat_process() callback
1582  * to cancel requests
1583  */
1584 struct hsm_cancel_all_data {
1585         struct mdt_device       *mdt;
1586 };
1587
1588 /**
1589  *  llog_cat_process() callback, used to:
1590  *  - purge all requests
1591  * \param env [IN] environment
1592  * \param llh [IN] llog handle
1593  * \param hdr [IN] llog record
1594  * \param data [IN] cb data = struct hsm_cancel_all_data
1595  * \retval 0 success
1596  * \retval -ve failure
1597  */
1598 static int mdt_cancel_all_cb(const struct lu_env *env,
1599                              struct llog_handle *llh,
1600                              struct llog_rec_hdr *hdr, void *data)
1601 {
1602         struct llog_agent_req_rec       *larr;
1603         struct hsm_cancel_all_data      *hcad;
1604         int                              rc = 0;
1605         ENTRY;
1606
1607         larr = (struct llog_agent_req_rec *)hdr;
1608         hcad = data;
1609         if (larr->arr_status == ARS_WAITING ||
1610             larr->arr_status == ARS_STARTED) {
1611                 larr->arr_status = ARS_CANCELED;
1612                 larr->arr_req_change = cfs_time_current_sec();
1613                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1614         }
1615
1616         RETURN(rc);
1617 }
1618
1619 /**
1620  * cancel all actions
1621  * \param obd [IN] MDT device
1622  */
1623 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1624 {
1625         struct lu_env                    env;
1626         struct lu_context                session;
1627         struct mdt_thread_info          *mti;
1628         struct coordinator              *cdt = &mdt->mdt_coordinator;
1629         struct cdt_agent_req            *car;
1630         struct hsm_action_list          *hal = NULL;
1631         struct hsm_action_item          *hai;
1632         struct hsm_cancel_all_data       hcad;
1633         int                              hal_sz = 0, hal_len, rc;
1634         enum cdt_states                  old_state;
1635         ENTRY;
1636
1637         rc = lu_env_init(&env, LCT_MD_THREAD);
1638         if (rc < 0)
1639                 RETURN(rc);
1640
1641         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1642         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1643         if (rc < 0)
1644                 GOTO(out_env, rc);
1645
1646         lu_context_enter(&session);
1647         env.le_ses = &session;
1648
1649         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1650         LASSERT(mti != NULL);
1651
1652         mti->mti_env = &env;
1653         mti->mti_mdt = mdt;
1654
1655         hsm_init_ucred(mdt_ucred(mti));
1656
1657         /* disable coordinator */
1658         rc = set_cdt_state(cdt, CDT_DISABLE, &old_state);
1659         if (rc)
1660                 RETURN(rc);
1661
1662         /* send cancel to all running requests */
1663         down_read(&cdt->cdt_request_lock);
1664         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1665                 mdt_cdt_get_request(car);
1666                 /* request is not yet removed from list, it will be done
1667                  * when copytool will return progress
1668                  */
1669
1670                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1671                         mdt_cdt_put_request(car);
1672                         continue;
1673                 }
1674
1675                 /* needed size */
1676                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1677                           cfs_size_round(car->car_hai->hai_len);
1678
1679                 if (hal_len > hal_sz && hal_sz > 0) {
1680                         /* not enough room, free old buffer */
1681                         OBD_FREE(hal, hal_sz);
1682                         hal = NULL;
1683                 }
1684
1685                 /* empty buffer, allocate one */
1686                 if (hal == NULL) {
1687                         hal_sz = hal_len;
1688                         OBD_ALLOC(hal, hal_sz);
1689                         if (hal == NULL) {
1690                                 mdt_cdt_put_request(car);
1691                                 up_read(&cdt->cdt_request_lock);
1692                                 GOTO(out_cdt_state, rc = -ENOMEM);
1693                         }
1694                 }
1695
1696                 hal->hal_version = HAL_VERSION;
1697                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1698                                 MTI_NAME_MAXLEN);
1699                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1700                 hal->hal_compound_id = car->car_compound_id;
1701                 hal->hal_archive_id = car->car_archive_id;
1702                 hal->hal_flags = car->car_flags;
1703                 hal->hal_count = 0;
1704
1705                 hai = hai_first(hal);
1706                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1707                 hai->hai_action = HSMA_CANCEL;
1708                 hal->hal_count = 1;
1709
1710                 /* it is possible to safely call mdt_hsm_agent_send()
1711                  * (ie without a deadlock on cdt_request_lock), because the
1712                  * write lock is taken only if we are not in purge mode
1713                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1714                  *   nor mdt_cdt_remove_request())
1715                  */
1716                 /* no conflict with cdt thread because cdt is disable and we
1717                  * have the request lock */
1718                 mdt_hsm_agent_send(mti, hal, 1);
1719
1720                 mdt_cdt_put_request(car);
1721         }
1722         up_read(&cdt->cdt_request_lock);
1723
1724         if (hal != NULL)
1725                 OBD_FREE(hal, hal_sz);
1726
1727         /* cancel all on-disk records */
1728         hcad.mdt = mdt;
1729
1730         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1731                               &hcad, 0, 0, WRITE);
1732 out_cdt_state:
1733         /* Enable coordinator, unless the coordinator was stopping. */
1734         set_cdt_state(cdt, old_state, NULL);
1735         lu_context_exit(&session);
1736         lu_context_fini(&session);
1737 out_env:
1738         lu_env_fini(&env);
1739
1740         RETURN(rc);
1741 }
1742
1743 /**
1744  * check if a request is compatible with file status
1745  * \param hai [IN] request description
1746  * \param archive_id [IN] request archive id
1747  * \param rq_flags [IN] request flags
1748  * \param hsm [IN] file HSM metadata
1749  * \retval boolean
1750  */
1751 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1752                               u32 archive_id, u64 rq_flags,
1753                               const struct md_hsm *hsm)
1754 {
1755         int      is_compat = false;
1756         int      hsm_flags;
1757         ENTRY;
1758
1759         hsm_flags = hsm->mh_flags;
1760         switch (hai->hai_action) {
1761         case HSMA_ARCHIVE:
1762                 if (!(hsm_flags & HS_NOARCHIVE) &&
1763                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1764                         is_compat = true;
1765
1766                 if (hsm_flags & HS_EXISTS &&
1767                     archive_id != 0 &&
1768                     archive_id != hsm->mh_arch_id)
1769                         is_compat = false;
1770
1771                 break;
1772         case HSMA_RESTORE:
1773                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1774                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1775                         is_compat = true;
1776                 break;
1777         case HSMA_REMOVE:
1778                 if (!(hsm_flags & HS_RELEASED) &&
1779                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1780                         is_compat = true;
1781                 break;
1782         case HSMA_CANCEL:
1783                 is_compat = true;
1784                 break;
1785         }
1786         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1787                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1788                       PFID(&hai->hai_fid),
1789                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1790                       hai->hai_extent.offset, hai->hai_extent.length,
1791                       hsm->mh_flags,
1792                       (is_compat ? "compatible" : "uncompatible"));
1793
1794         RETURN(is_compat);
1795 }
1796
1797 /*
1798  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1799  */
1800 static const struct {
1801         __u64            bit;
1802         char            *name;
1803         char            *nickname;
1804 } hsm_policy_names[] = {
1805         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1806         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1807         { 0 },
1808 };
1809
1810 /**
1811  * convert a policy name to a bit
1812  * \param name [IN] policy name
1813  * \retval 0 unknown
1814  * \retval   policy bit
1815  */
1816 static __u64 hsm_policy_str2bit(const char *name)
1817 {
1818         int      i;
1819
1820         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1821                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1822                     strcmp(hsm_policy_names[i].name, name) == 0)
1823                         return hsm_policy_names[i].bit;
1824         return 0;
1825 }
1826
1827 /**
1828  * convert a policy bit field to a string
1829  * \param mask [IN] policy bit field
1830  * \param hexa [IN] print mask before bit names
1831  * \param buffer [OUT] string
1832  * \param count [IN] size of buffer
1833  */
1834 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1835                                 const bool hexa)
1836 {
1837         int      i, j;
1838         __u64    bit;
1839         ENTRY;
1840
1841         if (hexa)
1842                 seq_printf(m, "(%#llx) ", mask);
1843
1844         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1845                 bit = (1ULL << i);
1846
1847                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1848                         if (hsm_policy_names[j].bit == bit)
1849                                 break;
1850                 }
1851                 if (bit & mask)
1852                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1853                 else
1854                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1855         }
1856         /* remove last ' ' */
1857         m->count--;
1858         seq_putc(m, '\n');
1859 }
1860
1861 /* methods to read/write HSM policy flags */
1862 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1863 {
1864         struct mdt_device       *mdt = m->private;
1865         struct coordinator      *cdt = &mdt->mdt_coordinator;
1866         ENTRY;
1867
1868         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1869         RETURN(0);
1870 }
1871
1872 static ssize_t
1873 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1874                          size_t count, loff_t *off)
1875 {
1876         struct seq_file         *m = file->private_data;
1877         struct mdt_device       *mdt = m->private;
1878         struct coordinator      *cdt = &mdt->mdt_coordinator;
1879         char                    *start, *token, sign;
1880         char                    *buf;
1881         __u64                    policy;
1882         __u64                    add_mask, remove_mask, set_mask;
1883         int                      rc;
1884         ENTRY;
1885
1886         if (count + 1 > PAGE_SIZE)
1887                 RETURN(-EINVAL);
1888
1889         OBD_ALLOC(buf, count + 1);
1890         if (buf == NULL)
1891                 RETURN(-ENOMEM);
1892
1893         if (copy_from_user(buf, buffer, count))
1894                 GOTO(out, rc = -EFAULT);
1895
1896         buf[count] = '\0';
1897
1898         start = buf;
1899         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1900                start);
1901
1902         add_mask = remove_mask = set_mask = 0;
1903         do {
1904                 token = strsep(&start, "\n ");
1905                 sign = *token;
1906
1907                 if (sign == '\0')
1908                         continue;
1909
1910                 if (sign == '-' || sign == '+')
1911                         token++;
1912
1913                 policy = hsm_policy_str2bit(token);
1914                 if (policy == 0) {
1915                         CWARN("%s: '%s' is unknown, "
1916                               "supported policies are:\n", mdt_obd_name(mdt),
1917                               token);
1918                         hsm_policy_bit2str(m, 0, false);
1919                         GOTO(out, rc = -EINVAL);
1920                 }
1921                 switch (sign) {
1922                 case '-':
1923                         remove_mask |= policy;
1924                         break;
1925                 case '+':
1926                         add_mask |= policy;
1927                         break;
1928                 default:
1929                         set_mask |= policy;
1930                         break;
1931                 }
1932
1933         } while (start != NULL);
1934
1935         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
1936                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1937
1938         /* if no sign in all string, it is a clear and set
1939          * if some sign found, all unsigned are converted
1940          * to add
1941          * P1 P2 = set to P1 and P2
1942          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1943          */
1944         if (remove_mask == 0 && add_mask == 0) {
1945                 cdt->cdt_policy = set_mask;
1946         } else {
1947                 cdt->cdt_policy |= set_mask | add_mask;
1948                 cdt->cdt_policy &= ~remove_mask;
1949         }
1950
1951         GOTO(out, rc = count);
1952
1953 out:
1954         OBD_FREE(buf, count + 1);
1955         RETURN(rc);
1956 }
1957 LPROC_SEQ_FOPS(mdt_hsm_policy);
1958
1959 #define GENERATE_PROC_METHOD(VAR)                                       \
1960 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
1961 {                                                                       \
1962         struct mdt_device       *mdt = m->private;                      \
1963         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1964         ENTRY;                                                          \
1965                                                                         \
1966         seq_printf(m, "%llu\n", (__u64)cdt->VAR);                       \
1967         RETURN(0);                                                      \
1968 }                                                                       \
1969 static ssize_t                                                          \
1970 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1971                           size_t count, loff_t *off)                    \
1972                                                                         \
1973 {                                                                       \
1974         struct seq_file         *m = file->private_data;                \
1975         struct mdt_device       *mdt = m->private;                      \
1976         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1977         __s64                    val;                                   \
1978         int                      rc;                                    \
1979         ENTRY;                                                          \
1980                                                                         \
1981         rc = lprocfs_str_to_s64(buffer, count, &val);                   \
1982         if (rc)                                                         \
1983                 RETURN(rc);                                             \
1984         if (val > 0 && val < INT_MAX) {                                 \
1985                 cdt->VAR = val;                                         \
1986                 RETURN(count);                                          \
1987         }                                                               \
1988         RETURN(-EINVAL);                                                \
1989 }                                                                       \
1990
1991 GENERATE_PROC_METHOD(cdt_loop_period)
1992 GENERATE_PROC_METHOD(cdt_grace_delay)
1993 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1994 GENERATE_PROC_METHOD(cdt_max_requests)
1995 GENERATE_PROC_METHOD(cdt_default_archive_id)
1996
1997 /*
1998  * procfs write method for MDT/hsm_control
1999  * proc entry is in mdt directory so data is mdt obd_device pointer
2000  */
2001 #define CDT_ENABLE_CMD   "enabled"
2002 #define CDT_STOP_CMD     "shutdown"
2003 #define CDT_DISABLE_CMD  "disabled"
2004 #define CDT_PURGE_CMD    "purge"
2005 #define CDT_HELP_CMD     "help"
2006 #define CDT_MAX_CMD_LEN  10
2007
2008 ssize_t
2009 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
2010                               size_t count, loff_t *off)
2011 {
2012         struct seq_file         *m = file->private_data;
2013         struct obd_device       *obd = m->private;
2014         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
2015         struct coordinator      *cdt = &(mdt->mdt_coordinator);
2016         int                      rc, usage = 0;
2017         char                     kernbuf[CDT_MAX_CMD_LEN];
2018         ENTRY;
2019
2020         if (count == 0 || count >= sizeof(kernbuf))
2021                 RETURN(-EINVAL);
2022
2023         if (copy_from_user(kernbuf, buffer, count))
2024                 RETURN(-EFAULT);
2025         kernbuf[count] = 0;
2026
2027         if (kernbuf[count - 1] == '\n')
2028                 kernbuf[count - 1] = 0;
2029
2030         rc = 0;
2031         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
2032                 if (cdt->cdt_state == CDT_DISABLE) {
2033                         rc = set_cdt_state(cdt, CDT_RUNNING, NULL);
2034                         mdt_hsm_cdt_event(cdt);
2035                         wake_up(&cdt->cdt_waitq);
2036                 } else {
2037                         rc = mdt_hsm_cdt_start(mdt);
2038                 }
2039         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2040                 if ((cdt->cdt_state == CDT_STOPPING) ||
2041                     (cdt->cdt_state == CDT_STOPPED)) {
2042                         CERROR("%s: Coordinator already stopped\n",
2043                                mdt_obd_name(mdt));
2044                         rc = -EALREADY;
2045                 } else {
2046                         rc = mdt_hsm_cdt_stop(mdt);
2047                 }
2048         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2049                 if ((cdt->cdt_state == CDT_STOPPING) ||
2050                     (cdt->cdt_state == CDT_STOPPED)) {
2051                         CERROR("%s: Coordinator is stopped\n",
2052                                mdt_obd_name(mdt));
2053                         rc = -EINVAL;
2054                 } else {
2055                         rc = set_cdt_state(cdt, CDT_DISABLE, NULL);
2056                 }
2057         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2058                 rc = hsm_cancel_all_actions(mdt);
2059         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2060                 usage = 1;
2061         } else {
2062                 usage = 1;
2063                 rc = -EINVAL;
2064         }
2065
2066         if (usage == 1)
2067                 CERROR("%s: Valid coordinator control commands are: "
2068                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2069                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2070                        CDT_PURGE_CMD, CDT_HELP_CMD);
2071
2072         if (rc)
2073                 RETURN(rc);
2074
2075         RETURN(count);
2076 }
2077
2078 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2079 {
2080         struct obd_device       *obd = m->private;
2081         struct coordinator      *cdt;
2082         ENTRY;
2083
2084         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2085
2086         seq_printf(m, "%s\n", cdt_mdt_state2str(cdt->cdt_state));
2087
2088         RETURN(0);
2089 }
2090
2091 static int
2092 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2093 {
2094         bool first = true;
2095         int i;
2096         ENTRY;
2097
2098         for (i = 0; i < 8 * sizeof(mask); i++) {
2099                 if (mask & (1UL << i)) {
2100                         seq_printf(m, "%s%s", first ? "" : " ",
2101                                    hsm_copytool_action2name(i));
2102                         first = false;
2103                 }
2104         }
2105         seq_putc(m, '\n');
2106
2107         RETURN(0);
2108 }
2109
2110 static int
2111 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2112 {
2113         struct mdt_device *mdt = m->private;
2114         struct coordinator *cdt = &mdt->mdt_coordinator;
2115
2116         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2117 }
2118
2119 static int
2120 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2121 {
2122         struct mdt_device *mdt = m->private;
2123         struct coordinator *cdt = &mdt->mdt_coordinator;
2124
2125         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2126 }
2127
2128 static int
2129 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2130 {
2131         struct mdt_device *mdt = m->private;
2132         struct coordinator *cdt = &mdt->mdt_coordinator;
2133
2134         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2135 }
2136
2137 static inline enum hsm_copytool_action
2138 hsm_copytool_name2action(const char *name)
2139 {
2140         if (strcasecmp(name, "NOOP") == 0)
2141                 return HSMA_NONE;
2142         else if (strcasecmp(name, "ARCHIVE") == 0)
2143                 return HSMA_ARCHIVE;
2144         else if (strcasecmp(name, "RESTORE") == 0)
2145                 return HSMA_RESTORE;
2146         else if (strcasecmp(name, "REMOVE") == 0)
2147                 return HSMA_REMOVE;
2148         else if (strcasecmp(name, "CANCEL") == 0)
2149                 return HSMA_CANCEL;
2150         else
2151                 return -1;
2152 }
2153
2154 static ssize_t
2155 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2156                             size_t user_count, __u64 *mask)
2157 {
2158         char *buf, *pos, *name;
2159         size_t buf_size;
2160         __u64 new_mask = 0;
2161         int rc;
2162         ENTRY;
2163
2164         if (!(user_count < 4096))
2165                 RETURN(-ENOMEM);
2166
2167         buf_size = user_count + 1;
2168
2169         OBD_ALLOC(buf, buf_size);
2170         if (buf == NULL)
2171                 RETURN(-ENOMEM);
2172
2173         if (copy_from_user(buf, user_buf, buf_size - 1))
2174                 GOTO(out, rc = -EFAULT);
2175
2176         buf[buf_size - 1] = '\0';
2177
2178         pos = buf;
2179         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2180                 int action;
2181
2182                 if (*name == '\0')
2183                         continue;
2184
2185                 action = hsm_copytool_name2action(name);
2186                 if (action < 0)
2187                         GOTO(out, rc = -EINVAL);
2188
2189                 new_mask |= (1UL << action);
2190         }
2191
2192         *mask = new_mask;
2193         rc = user_count;
2194 out:
2195         OBD_FREE(buf, buf_size);
2196
2197         RETURN(rc);
2198 }
2199
2200 static ssize_t
2201 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2202                                         size_t count, loff_t *off)
2203 {
2204         struct seq_file         *m = file->private_data;
2205         struct mdt_device       *mdt = m->private;
2206         struct coordinator *cdt = &mdt->mdt_coordinator;
2207
2208         return mdt_write_hsm_request_mask(file, buf, count,
2209                                            &cdt->cdt_user_request_mask);
2210 }
2211
2212 static ssize_t
2213 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2214                                         size_t count, loff_t *off)
2215 {
2216         struct seq_file         *m = file->private_data;
2217         struct mdt_device       *mdt = m->private;
2218         struct coordinator      *cdt = &mdt->mdt_coordinator;
2219
2220         return mdt_write_hsm_request_mask(file, buf, count,
2221                                            &cdt->cdt_group_request_mask);
2222 }
2223
2224 static ssize_t
2225 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2226                                         size_t count, loff_t *off)
2227 {
2228         struct seq_file         *m = file->private_data;
2229         struct mdt_device       *mdt = m->private;
2230         struct coordinator      *cdt = &mdt->mdt_coordinator;
2231
2232         return mdt_write_hsm_request_mask(file, buf, count,
2233                                            &cdt->cdt_other_request_mask);
2234 }
2235
2236 static int mdt_hsm_cdt_raolu_seq_show(struct seq_file *m, void *data)
2237 {
2238         struct mdt_device *mdt = m->private;
2239         struct coordinator *cdt = &mdt->mdt_coordinator;
2240         ENTRY;
2241
2242         seq_printf(m, "%d\n", (int)cdt->cdt_remove_archive_on_last_unlink);
2243         RETURN(0);
2244 }
2245
2246 static ssize_t
2247 mdt_hsm_cdt_raolu_seq_write(struct file *file, const char __user *buffer,
2248                             size_t count, loff_t *off)
2249
2250 {
2251         struct seq_file *m = file->private_data;
2252         struct mdt_device *mdt = m->private;
2253         struct coordinator *cdt = &mdt->mdt_coordinator;
2254         __s64 val;
2255         int rc;
2256         ENTRY;
2257
2258         rc = lprocfs_str_to_s64(buffer, count, &val);
2259         if (rc < 0)
2260                 RETURN(rc);
2261
2262         cdt->cdt_remove_archive_on_last_unlink = val;
2263         RETURN(count);
2264 }
2265
2266 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2267 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2268 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2269 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2270 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2271 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2272 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2273 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2274 LPROC_SEQ_FOPS(mdt_hsm_cdt_raolu);
2275
2276 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2277         { .name =       "agents",
2278           .fops =       &mdt_hsm_agent_fops                     },
2279         { .name =       "actions",
2280           .fops =       &mdt_hsm_actions_fops,
2281           .proc_mode =  0444                                    },
2282         { .name =       "default_archive_id",
2283           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2284         { .name =       "grace_delay",
2285           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2286         { .name =       "loop_period",
2287           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2288         { .name =       "max_requests",
2289           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2290         { .name =       "policy",
2291           .fops =       &mdt_hsm_policy_fops                    },
2292         { .name =       "active_request_timeout",
2293           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2294         { .name =       "active_requests",
2295           .fops =       &mdt_hsm_active_requests_fops           },
2296         { .name =       "user_request_mask",
2297           .fops =       &mdt_hsm_user_request_mask_fops,        },
2298         { .name =       "group_request_mask",
2299           .fops =       &mdt_hsm_group_request_mask_fops,       },
2300         { .name =       "other_request_mask",
2301           .fops =       &mdt_hsm_other_request_mask_fops,       },
2302         { .name =       "remove_archive_on_last_unlink",
2303           .fops =       &mdt_hsm_cdt_raolu_fops,                },
2304         { 0 }
2305 };