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_net.h>
  44 #include <lustre_export.h>
  45 #include <obd.h>
  46 #include <lprocfs_status.h>
  47 #include <lustre_log.h>
  48 #include <lustre_kernelcomm.h>
  49 #include "mdt_internal.h"
  50
  51 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
  52
  53 /**
  54  * get obj and HSM attributes on a fid
  55  * \param mti [IN] context
  56  * \param fid [IN] object fid
  57  * \param hsm [OUT] HSM meta data
  58  * \retval obj or error (-ENOENT if not found)
  59  */
  60 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  61                                       const struct lu_fid *fid,
  62                                       struct md_hsm *hsm)
  63 {
  64         struct md_attr          *ma;
  65         struct mdt_object       *obj;
  66         int                      rc;
  67         ENTRY;
  68
  69         ma = &mti->mti_attr;
  70         ma->ma_need = MA_HSM;
  71         ma->ma_valid = 0;
  72
  73         /* find object by FID */
  74         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  75         if (IS_ERR(obj))
  76                 RETURN(obj);
  77
  78         if (!mdt_object_exists(obj)) {
  79                 /* no more object */
  80                 mdt_object_put(mti->mti_env, obj);
  81                 RETURN(ERR_PTR(-ENOENT));
  82         }
  83
  84         rc = mdt_attr_get_complex(mti, obj, ma);
  85         if (rc) {
  86                 mdt_object_put(mti->mti_env, obj);
  87                 RETURN(ERR_PTR(rc));
  88         }
  89
  90         if (ma->ma_valid & MA_HSM)
  91                 *hsm = ma->ma_hsm;
  92         else
  93                 memset(hsm, 0, sizeof(*hsm));
  94         ma->ma_valid = 0;
  95         RETURN(obj);
  96 }
  97
  98 void mdt_hsm_dump_hal(int level, const char *prefix,
  99                       struct hsm_action_list *hal)
 100 {
 101         int                      i, sz;
 102         struct hsm_action_item  *hai;
 103         char                     buf[12];
 104
 105         CDEBUG(level, "%s: HAL header: version %X count %d compound %#llx"
 106                       " archive_id %d flags %#llx\n",
 107                prefix, hal->hal_version, hal->hal_count,
 108                hal->hal_compound_id, hal->hal_archive_id, hal->hal_flags);
 109
 110         hai = hai_first(hal);
 111         for (i = 0; i < hal->hal_count; i++) {
 112                 sz = hai->hai_len - sizeof(*hai);
 113                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 114                        " compound/cookie=%#llx/%#llx"
 115                        " action=%s extent=%#llx-%#llx gid=%#llx"
 116                        " datalen=%d data=[%s]\n",
 117                        prefix, i,
 118                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 119                        hal->hal_compound_id, hai->hai_cookie,
 120                        hsm_copytool_action2name(hai->hai_action),
 121                        hai->hai_extent.offset,
 122                        hai->hai_extent.length,
 123                        hai->hai_gid, sz,
 124                        hai_dump_data_field(hai, buf, sizeof(buf)));
 125                 hai = hai_next(hai);
 126         }
 127 }
 128
 129 /**
 130  * data passed to llog_cat_process() callback
 131  * to scan requests and take actions
 132  */
 133 struct hsm_scan_request {
 134         int                      hal_sz;
 135         int                      hal_used_sz;
 136         struct hsm_action_list  *hal;
 137 };
 138
 139 struct hsm_scan_data {
 140         struct mdt_thread_info          *mti;
 141         char                             fs_name[MTI_NAME_MAXLEN+1];
 142         /* request to be send to agents */
 143         int                              max_requests;  /** vector size */
 144         int                              request_cnt;   /** used count */
 145         struct hsm_scan_request         *request;
 146 };
 147
 148 /**
 149  *  llog_cat_process() callback, used to:
 150  *  - find waiting request and start action
 151  *  - purge canceled and done requests
 152  * \param env [IN] environment
 153  * \param llh [IN] llog handle
 154  * \param hdr [IN] llog record
 155  * \param data [IN/OUT] cb data = struct hsm_scan_data
 156  * \retval 0 success
 157  * \retval -ve failure
 158  */
 159 static int mdt_coordinator_cb(const struct lu_env *env,
 160                               struct llog_handle *llh,
 161                               struct llog_rec_hdr *hdr,
 162                               void *data)
 163 {
 164         struct llog_agent_req_rec       *larr;
 165         struct hsm_scan_data            *hsd;
 166         struct hsm_action_item          *hai;
 167         struct mdt_device               *mdt;
 168         struct coordinator              *cdt;
 169         int                              rc;
 170         ENTRY;
 171
 172         hsd = data;
 173         mdt = hsd->mti->mti_mdt;
 174         cdt = &mdt->mdt_coordinator;
 175
 176         larr = (struct llog_agent_req_rec *)hdr;
 177         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 178         switch (larr->arr_status) {
 179         case ARS_WAITING: {
 180                 int i;
 181                 struct hsm_scan_request *request;
 182
 183                 /* Are agents full? */
 184                 if (atomic_read(&cdt->cdt_request_count) >=
 185                     cdt->cdt_max_requests)
 186                         break;
 187
 188                 /* first search whether the request is found in the
 189                  * list we have built. */
 190                 request = NULL;
 191                 for (i = 0; i < hsd->request_cnt; i++) {
 192                         if (hsd->request[i].hal->hal_compound_id ==
 193                             larr->arr_compound_id) {
 194                                 request = &hsd->request[i];
 195                                 break;
 196                         }
 197                 }
 198
 199                 if (!request) {
 200                         struct hsm_action_list *hal;
 201
 202                         if (hsd->request_cnt == hsd->max_requests)
 203                                 /* Unknown request and no more room
 204                                  * for a new request. Continue to scan
 205                                  * to find other entries for already
 206                                  * existing requests.
 207                                  */
 208                                 RETURN(0);
 209
 210                         request = &hsd->request[hsd->request_cnt];
 211
 212                         /* allocates hai vector size just needs to be large
 213                          * enough */
 214                         request->hal_sz =
 215                                 sizeof(*request->hal) +
 216                                 cfs_size_round(MTI_NAME_MAXLEN+1) +
 217                                 2 * cfs_size_round(larr->arr_hai.hai_len);
 218                         OBD_ALLOC(hal, request->hal_sz);
 219                         if (!hal)
 220                                 RETURN(-ENOMEM);
 221                         hal->hal_version = HAL_VERSION;
 222                         strlcpy(hal->hal_fsname, hsd->fs_name,
 223                                 MTI_NAME_MAXLEN + 1);
 224                         hal->hal_compound_id = larr->arr_compound_id;
 225                         hal->hal_archive_id = larr->arr_archive_id;
 226                         hal->hal_flags = larr->arr_flags;
 227                         hal->hal_count = 0;
 228                         request->hal_used_sz = hal_size(hal);
 229                         request->hal = hal;
 230                         hsd->request_cnt++;
 231                         hai = hai_first(hal);
 232                 } else {
 233                         /* request is known */
 234                         /* we check if record archive num is the same as the
 235                          * known request, if not we will serve it in multiple
 236                          * time because we do not know if the agent can serve
 237                          * multiple backend
 238                          * a use case is a compound made of multiple restore
 239                          * where the files are not archived in the same backend
 240                          */
 241                         if (larr->arr_archive_id !=
 242                             request->hal->hal_archive_id)
 243                                 RETURN(0);
 244
 245                         if (request->hal_sz <
 246                             request->hal_used_sz +
 247                             cfs_size_round(larr->arr_hai.hai_len)) {
 248                                 /* Not enough room, need an extension */
 249                                 void *hal_buffer;
 250                                 int sz;
 251
 252                                 sz = 2 * request->hal_sz;
 253                                 OBD_ALLOC(hal_buffer, sz);
 254                                 if (!hal_buffer)
 255                                         RETURN(-ENOMEM);
 256                                 memcpy(hal_buffer, request->hal,
 257                                        request->hal_used_sz);
 258                                 OBD_FREE(request->hal,
 259                                          request->hal_sz);
 260                                 request->hal = hal_buffer;
 261                                 request->hal_sz = sz;
 262                         }
 263                         hai = hai_first(request->hal);
 264                         for (i = 0; i < request->hal->hal_count; i++)
 265                                 hai = hai_next(hai);
 266                 }
 267                 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 268                 hai->hai_cookie = larr->arr_hai.hai_cookie;
 269                 hai->hai_gid = larr->arr_hai.hai_gid;
 270
 271                 request->hal_used_sz += cfs_size_round(hai->hai_len);
 272                 request->hal->hal_count++;
 273                 break;
 274         }
 275         case ARS_STARTED: {
 276                 struct hsm_progress_kernel pgs;
 277                 struct cdt_agent_req *car;
 278                 cfs_time_t now = cfs_time_current_sec();
 279                 cfs_time_t last;
 280
 281                 /* we search for a running request
 282                  * error may happen if coordinator crashes or stopped
 283                  * with running request
 284                  */
 285                 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie);
 286                 if (car == NULL) {
 287                         last = larr->arr_req_change;
 288                 } else {
 289                         last = car->car_req_update;
 290                         mdt_cdt_put_request(car);
 291                 }
 292
 293                 /* test if request too long, if yes cancel it
 294                  * the same way the copy tool acknowledge a cancel request */
 295                 if (now <= last + cdt->cdt_active_req_timeout)
 296                         RETURN(0);
 297
 298                 dump_llog_agent_req_rec("request timed out, start cleaning",
 299                                         larr);
 300                 /* a too old cancel request just needs to be removed
 301                  * this can happen, if copy tool does not support
 302                  * cancel for other requests, we have to remove the
 303                  * running request and notify the copytool */
 304                 pgs.hpk_fid = larr->arr_hai.hai_fid;
 305                 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
 306                 pgs.hpk_extent = larr->arr_hai.hai_extent;
 307                 pgs.hpk_flags = HP_FLAG_COMPLETED;
 308                 pgs.hpk_errval = ENOSYS;
 309                 pgs.hpk_data_version = 0;
 310
 311                 /* update request state, but do not record in llog, to
 312                  * avoid deadlock on cdt_llog_lock */
 313                 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
 314                 if (rc)
 315                         CERROR("%s: cannot cleanup timed out request: "
 316                                DFID" for cookie %#llx action=%s\n",
 317                                mdt_obd_name(mdt),
 318                                PFID(&pgs.hpk_fid), pgs.hpk_cookie,
 319                                hsm_copytool_action2name(
 320                                        larr->arr_hai.hai_action));
 321
 322                 if (rc == -ENOENT) {
 323                         /* The request no longer exists, forget
 324                          * about it, and do not send a cancel request
 325                          * to the client, for which an error will be
 326                          * sent back, leading to an endless cycle of
 327                          * cancellation. */
 328                         RETURN(LLOG_DEL_RECORD);
 329                 }
 330
 331                 /* XXX A cancel request cannot be cancelled. */
 332                 if (larr->arr_hai.hai_action == HSMA_CANCEL)
 333                         RETURN(0);
 334
 335                 larr->arr_status = ARS_CANCELED;
 336                 larr->arr_req_change = now;
 337                 rc = llog_write(hsd->mti->mti_env, llh, hdr, hdr->lrh_index);
 338                 if (rc < 0)
 339                         CERROR("%s: cannot update agent log: rc = %d\n",
 340                                mdt_obd_name(mdt), rc);
 341                 break;
 342         }
 343         case ARS_FAILED:
 344         case ARS_CANCELED:
 345         case ARS_SUCCEED:
 346                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 347                     cfs_time_current_sec())
 348                         RETURN(LLOG_DEL_RECORD);
 349                 break;
 350         }
 351         RETURN(0);
 352 }
 353
 354 /**
 355  * create /proc entries for coordinator
 356  * \param mdt [IN]
 357  * \retval 0 success
 358  * \retval -ve failure
 359  */
 360 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 361 {
 362         struct coordinator      *cdt = &mdt->mdt_coordinator;
 363         int                      rc = 0;
 364         ENTRY;
 365
 366         /* init /proc entries, failure is not critical */
 367         cdt->cdt_proc_dir = lprocfs_register("hsm",
 368                                              mdt2obd_dev(mdt)->obd_proc_entry,
 369                                              lprocfs_mdt_hsm_vars, mdt);
 370         if (IS_ERR(cdt->cdt_proc_dir)) {
 371                 rc = PTR_ERR(cdt->cdt_proc_dir);
 372                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 373                        " rc=%d\n", mdt_obd_name(mdt), rc);
 374                 cdt->cdt_proc_dir = NULL;
 375                 RETURN(rc);
 376         }
 377
 378         RETURN(0);
 379 }
 380
 381 /**
 382  * remove /proc entries for coordinator
 383  * \param mdt [IN]
 384  */
 385 void  hsm_cdt_procfs_fini(struct mdt_device *mdt)
 386 {
 387         struct coordinator      *cdt = &mdt->mdt_coordinator;
 388
 389         LASSERT(cdt->cdt_state == CDT_STOPPED);
 390         if (cdt->cdt_proc_dir != NULL)
 391                 lprocfs_remove(&cdt->cdt_proc_dir);
 392 }
 393
 394 /**
 395  * get vector of hsm cdt /proc vars
 396  * \param none
 397  * \retval var vector
 398  */
 399 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 400 {
 401         return lprocfs_mdt_hsm_vars;
 402 }
 403
 404 /**
 405  * coordinator thread
 406  * \param data [IN] obd device
 407  * \retval 0 success
 408  * \retval -ve failure
 409  */
 410 static int mdt_coordinator(void *data)
 411 {
 412         struct mdt_thread_info  *mti = data;
 413         struct mdt_device       *mdt = mti->mti_mdt;
 414         struct coordinator      *cdt = &mdt->mdt_coordinator;
 415         struct hsm_scan_data     hsd = { NULL };
 416         int                      rc = 0;
 417         int                      request_sz;
 418         ENTRY;
 419
 420         cdt->cdt_flags = SVC_RUNNING;
 421         wake_up(&cdt->cdt_waitq);
 422
 423         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 424                mdt_obd_name(mdt), current_pid());
 425
 426         /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
 427          * increases due to a change from /proc we do not overflow the
 428          * hsd.request[] vector
 429          */
 430         hsd.max_requests = cdt->cdt_max_requests;
 431         request_sz = hsd.max_requests * sizeof(*hsd.request);
 432         OBD_ALLOC(hsd.request, request_sz);
 433         if (!hsd.request)
 434                 GOTO(out, rc = -ENOMEM);
 435
 436         hsd.mti = mti;
 437         obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
 438
 439         while (1) {
 440                 struct l_wait_info lwi;
 441                 int i;
 442
 443                 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
 444                                   NULL, NULL);
 445                 l_wait_event(cdt->cdt_waitq,
 446                              cdt->cdt_flags & (SVC_STOPPING|SVC_EVENT),
 447                              &lwi);
 448
 449                 CDEBUG(D_HSM, "coordinator resumes\n");
 450
 451                 if (cdt->cdt_flags & SVC_STOPPING ||
 452                     cdt->cdt_state == CDT_STOPPING) {
 453                         cdt->cdt_flags &= ~SVC_STOPPING;
 454                         rc = 0;
 455                         break;
 456                 }
 457
 458                 /* wake up before timeout, new work arrives */
 459                 if (cdt->cdt_flags & SVC_EVENT)
 460                         cdt->cdt_flags &= ~SVC_EVENT;
 461
 462                 /* if coordinator is suspended continue to wait */
 463                 if (cdt->cdt_state == CDT_DISABLE) {
 464                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 465                         continue;
 466                 }
 467
 468                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 469
 470                 if (hsd.max_requests != cdt->cdt_max_requests) {
 471                         /* cdt_max_requests has changed,
 472                          * we need to allocate a new buffer
 473                          */
 474                         OBD_FREE(hsd.request, request_sz);
 475                         hsd.max_requests = cdt->cdt_max_requests;
 476                         request_sz = hsd.max_requests * sizeof(*hsd.request);
 477                         OBD_ALLOC(hsd.request, request_sz);
 478                         if (!hsd.request) {
 479                                 rc = -ENOMEM;
 480                                 break;
 481                         }
 482                 }
 483
 484                 hsd.request_cnt = 0;
 485
 486                 rc = cdt_llog_process(mti->mti_env, mdt,
 487                                       mdt_coordinator_cb, &hsd);
 488                 if (rc < 0)
 489                         goto clean_cb_alloc;
 490
 491                 CDEBUG(D_HSM, "found %d requests to send\n", hsd.request_cnt);
 492
 493                 if (list_empty(&cdt->cdt_agents)) {
 494                         CDEBUG(D_HSM, "no agent available, "
 495                                       "coordinator sleeps\n");
 496                         goto clean_cb_alloc;
 497                 }
 498
 499                 /* here hsd contains a list of requests to be started */
 500                 for (i = 0; i < hsd.request_cnt; i++) {
 501                         struct hsm_scan_request *request = &hsd.request[i];
 502                         struct hsm_action_list  *hal = request->hal;
 503                         struct hsm_action_item  *hai;
 504                         __u64                   *cookies;
 505                         int                      sz, j;
 506                         enum agent_req_status    status;
 507
 508                         /* still room for work ? */
 509                         if (atomic_read(&cdt->cdt_request_count) >=
 510                             cdt->cdt_max_requests)
 511                                 break;
 512
 513                         rc = mdt_hsm_agent_send(mti, hal, 0);
 514                         /* if failure, we suppose it is temporary
 515                          * if the copy tool failed to do the request
 516                          * it has to use hsm_progress
 517                          */
 518                         status = (rc ? ARS_WAITING : ARS_STARTED);
 519
 520                         /* set up cookie vector to set records status
 521                          * after copy tools start or failed
 522                          */
 523                         sz = hal->hal_count * sizeof(__u64);
 524                         OBD_ALLOC(cookies, sz);
 525                         if (cookies == NULL)
 526                                 continue;
 527
 528                         hai = hai_first(hal);
 529                         for (j = 0; j < hal->hal_count; j++) {
 530                                 cookies[j] = hai->hai_cookie;
 531                                 hai = hai_next(hai);
 532                         }
 533
 534                         rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
 535                                                      hal->hal_count, status);
 536                         if (rc)
 537                                 CERROR("%s: mdt_agent_record_update() failed, "
 538                                        "rc=%d, cannot update status to %s "
 539                                        "for %d cookies\n",
 540                                        mdt_obd_name(mdt), rc,
 541                                        agent_req_status2name(status),
 542                                        hal->hal_count);
 543
 544                         OBD_FREE(cookies, sz);
 545                 }
 546 clean_cb_alloc:
 547                 /* free hal allocated by callback */
 548                 for (i = 0; i < hsd.request_cnt; i++) {
 549                         struct hsm_scan_request *request = &hsd.request[i];
 550
 551                         OBD_FREE(request->hal, request->hal_sz);
 552                 }
 553         }
 554         EXIT;
 555 out:
 556         if (hsd.request)
 557                 OBD_FREE(hsd.request, request_sz);
 558
 559         if (cdt->cdt_state == CDT_STOPPING) {
 560                 /* request comes from /proc path, so we need to clean cdt
 561                  * struct */
 562                  mdt_hsm_cdt_stop(mdt);
 563                  mdt->mdt_opts.mo_coordinator = 0;
 564         } else {
 565                 /* request comes from a thread event, generated
 566                  * by mdt_stop_coordinator(), we have to ack
 567                  * and cdt cleaning will be done by event sender
 568                  */
 569                 cdt->cdt_flags = SVC_STOPPED;
 570                 wake_up(&cdt->cdt_waitq);
 571         }
 572
 573         if (rc != 0)
 574                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 575                        mdt_obd_name(mdt), current_pid(), rc);
 576         else
 577                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 578                               " no error\n",
 579                        mdt_obd_name(mdt), current_pid());
 580
 581         return rc;
 582 }
 583
 584 /**
 585  * lookup a restore handle by FID
 586  * caller needs to hold cdt_restore_lock
 587  * \param cdt [IN] coordinator
 588  * \param fid [IN] FID
 589  * \retval cdt_restore_handle found
 590  * \retval NULL not found
 591  */
 592 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
 593                                                        const struct lu_fid *fid)
 594 {
 595         struct cdt_restore_handle       *crh;
 596         ENTRY;
 597
 598         list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
 599                 if (lu_fid_eq(&crh->crh_fid, fid))
 600                         RETURN(crh);
 601         }
 602         RETURN(NULL);
 603 }
 604
 605 /**
 606  * data passed to llog_cat_process() callback
 607  * to scan requests and take actions
 608  */
 609 struct hsm_restore_data {
 610         struct mdt_thread_info  *hrd_mti;
 611 };
 612
 613 /**
 614  *  llog_cat_process() callback, used to:
 615  *  - find restore request and allocate the restore handle
 616  * \param env [IN] environment
 617  * \param llh [IN] llog handle
 618  * \param hdr [IN] llog record
 619  * \param data [IN/OUT] cb data = struct hsm_restore_data
 620  * \retval 0 success
 621  * \retval -ve failure
 622  */
 623 static int hsm_restore_cb(const struct lu_env *env,
 624                           struct llog_handle *llh,
 625                           struct llog_rec_hdr *hdr, void *data)
 626 {
 627         struct llog_agent_req_rec       *larr;
 628         struct hsm_restore_data         *hrd;
 629         struct cdt_restore_handle       *crh;
 630         struct hsm_action_item          *hai;
 631         struct mdt_thread_info          *mti;
 632         struct coordinator              *cdt;
 633         struct mdt_object               *child;
 634         int rc;
 635         ENTRY;
 636
 637         hrd = data;
 638         mti = hrd->hrd_mti;
 639         cdt = &mti->mti_mdt->mdt_coordinator;
 640
 641         larr = (struct llog_agent_req_rec *)hdr;
 642         hai = &larr->arr_hai;
 643         if (hai->hai_cookie > cdt->cdt_last_cookie)
 644                 /* update the cookie to avoid collision */
 645                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 646
 647         if (hai->hai_action != HSMA_RESTORE ||
 648             agent_req_in_final_state(larr->arr_status))
 649                 RETURN(0);
 650
 651         /* restore request not in a final state */
 652
 653         /* force replay of restore requests left in started state from previous
 654          * CDT context, to be canceled later if finally found to be incompatible
 655          * when being re-started */
 656         if (larr->arr_status == ARS_STARTED) {
 657                 larr->arr_status = ARS_WAITING;
 658                 larr->arr_req_change = cfs_time_current_sec();
 659                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 660                 if (rc != 0)
 661                         GOTO(out, rc);
 662         }
 663
 664         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 665         if (crh == NULL)
 666                 RETURN(-ENOMEM);
 667
 668         crh->crh_fid = hai->hai_fid;
 669         /* in V1 all file is restored
 670         crh->extent.start = hai->hai_extent.offset;
 671         crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
 672         */
 673         crh->crh_extent.start = 0;
 674         crh->crh_extent.end = hai->hai_extent.length;
 675         /* get the layout lock */
 676         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 677         child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 678                                      MDS_INODELOCK_LAYOUT);
 679         if (IS_ERR(child))
 680                 GOTO(out, rc = PTR_ERR(child));
 681
 682         rc = 0;
 683         /* we choose to not keep a reference
 684          * on the object during the restore time which can be very long */
 685         mdt_object_put(mti->mti_env, child);
 686
 687         mutex_lock(&cdt->cdt_restore_lock);
 688         list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
 689         mutex_unlock(&cdt->cdt_restore_lock);
 690
 691 out:
 692         RETURN(rc);
 693 }
 694
 695 /**
 696  * restore coordinator state at startup
 697  * the goal is to take a layout lock for each registered restore request
 698  * \param mti [IN] context
 699  */
 700 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 701 {
 702         struct hsm_restore_data  hrd;
 703         int                      rc;
 704         ENTRY;
 705
 706         hrd.hrd_mti = mti;
 707
 708         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
 709                               hsm_restore_cb, &hrd);
 710
 711         RETURN(rc);
 712 }
 713
 714 static int hsm_init_ucred(struct lu_ucred *uc)
 715 {
 716         ENTRY;
 717
 718         uc->uc_valid = UCRED_OLD;
 719         uc->uc_o_uid = 0;
 720         uc->uc_o_gid = 0;
 721         uc->uc_o_fsuid = 0;
 722         uc->uc_o_fsgid = 0;
 723         uc->uc_uid = 0;
 724         uc->uc_gid = 0;
 725         uc->uc_fsuid = 0;
 726         uc->uc_fsgid = 0;
 727         uc->uc_suppgids[0] = -1;
 728         uc->uc_suppgids[1] = -1;
 729         uc->uc_cap = CFS_CAP_FS_MASK;
 730         uc->uc_umask = 0777;
 731         uc->uc_ginfo = NULL;
 732         uc->uc_identity = NULL;
 733
 734         RETURN(0);
 735 }
 736
 737 /**
 738  * wake up coordinator thread
 739  * \param mdt [IN] device
 740  * \retval 0 success
 741  * \retval -ve failure
 742  */
 743 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
 744 {
 745         struct coordinator      *cdt = &mdt->mdt_coordinator;
 746         ENTRY;
 747
 748         if (cdt->cdt_state == CDT_STOPPED)
 749                 RETURN(-ESRCH);
 750
 751         /* wake up coordinator */
 752         cdt->cdt_flags = SVC_EVENT;
 753         wake_up(&cdt->cdt_waitq);
 754
 755         RETURN(0);
 756 }
 757
 758 /**
 759  * initialize coordinator struct
 760  * \param mdt [IN] device
 761  * \retval 0 success
 762  * \retval -ve failure
 763  */
 764 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 765 {
 766         struct coordinator      *cdt = &mdt->mdt_coordinator;
 767         struct mdt_thread_info  *cdt_mti = NULL;
 768         int                      rc;
 769         ENTRY;
 770
 771         cdt->cdt_state = CDT_STOPPED;
 772
 773         init_waitqueue_head(&cdt->cdt_waitq);
 774         mutex_init(&cdt->cdt_llog_lock);
 775         init_rwsem(&cdt->cdt_agent_lock);
 776         init_rwsem(&cdt->cdt_request_lock);
 777         mutex_init(&cdt->cdt_restore_lock);
 778
 779         INIT_LIST_HEAD(&cdt->cdt_request_list);
 780         INIT_LIST_HEAD(&cdt->cdt_agents);
 781         INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 782
 783         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
 784                                                        CFS_HASH_BITS_MIN,
 785                                                        CFS_HASH_BITS_MAX,
 786                                                        CFS_HASH_BKT_BITS,
 787                                                        0 /* extra bytes */,
 788                                                        CFS_HASH_MIN_THETA,
 789                                                        CFS_HASH_MAX_THETA,
 790                                                 &cdt_request_cookie_hash_ops,
 791                                                        CFS_HASH_DEFAULT);
 792         if (cdt->cdt_request_cookie_hash == NULL)
 793                 RETURN(-ENOMEM);
 794
 795         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 796         if (rc < 0)
 797                 GOTO(out_request_cookie_hash, rc);
 798
 799         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
 800         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
 801         if (rc < 0)
 802                 GOTO(out_env, rc);
 803
 804         lu_context_enter(&cdt->cdt_session);
 805         cdt->cdt_env.le_ses = &cdt->cdt_session;
 806
 807         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 808         LASSERT(cdt_mti != NULL);
 809
 810         cdt_mti->mti_env = &cdt->cdt_env;
 811         cdt_mti->mti_mdt = mdt;
 812
 813         hsm_init_ucred(mdt_ucred(cdt_mti));
 814
 815         /* default values for /proc tunnables
 816          * can be override by MGS conf */
 817         cdt->cdt_default_archive_id = 1;
 818         cdt->cdt_grace_delay = 60;
 819         cdt->cdt_loop_period = 10;
 820         cdt->cdt_max_requests = 3;
 821         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 822         cdt->cdt_active_req_timeout = 3600;
 823
 824         RETURN(0);
 825
 826 out_env:
 827         lu_env_fini(&cdt->cdt_env);
 828 out_request_cookie_hash:
 829         cfs_hash_putref(cdt->cdt_request_cookie_hash);
 830         cdt->cdt_request_cookie_hash = NULL;
 831
 832         return rc;
 833 }
 834
 835 /**
 836  * free a coordinator thread
 837  * \param mdt [IN] device
 838  */
 839 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
 840 {
 841         struct coordinator *cdt = &mdt->mdt_coordinator;
 842         ENTRY;
 843
 844         lu_context_exit(cdt->cdt_env.le_ses);
 845         lu_context_fini(cdt->cdt_env.le_ses);
 846
 847         lu_env_fini(&cdt->cdt_env);
 848
 849         cfs_hash_putref(cdt->cdt_request_cookie_hash);
 850         cdt->cdt_request_cookie_hash = NULL;
 851
 852         RETURN(0);
 853 }
 854
 855 /**
 856  * start a coordinator thread
 857  * \param mdt [IN] device
 858  * \retval 0 success
 859  * \retval -ve failure
 860  */
 861 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
 862 {
 863         struct coordinator      *cdt = &mdt->mdt_coordinator;
 864         int                      rc;
 865         void                    *ptr;
 866         struct mdt_thread_info  *cdt_mti;
 867         struct task_struct      *task;
 868         ENTRY;
 869
 870         /* functions defined but not yet used
 871          * this avoid compilation warning
 872          */
 873         ptr = dump_requests;
 874
 875         if (cdt->cdt_state != CDT_STOPPED) {
 876                 CERROR("%s: Coordinator already started\n",
 877                        mdt_obd_name(mdt));
 878                 RETURN(-EALREADY);
 879         }
 880
 881         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
 882         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 883
 884         cdt->cdt_state = CDT_INIT;
 885
 886         atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
 887         /* just need to be larger than previous one */
 888         /* cdt_last_cookie is protected by cdt_llog_lock */
 889         cdt->cdt_last_cookie = cfs_time_current_sec();
 890         atomic_set(&cdt->cdt_request_count, 0);
 891         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
 892         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
 893         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
 894
 895         /* to avoid deadlock when start is made through /proc
 896          * /proc entries are created by the coordinator thread */
 897
 898         /* set up list of started restore requests */
 899         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 900         rc = mdt_hsm_pending_restore(cdt_mti);
 901         if (rc)
 902                 CERROR("%s: cannot take the layout locks needed"
 903                        " for registered restore: %d\n",
 904                        mdt_obd_name(mdt), rc);
 905
 906         if (mdt->mdt_bottom->dd_rdonly)
 907                 RETURN(0);
 908
 909         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
 910         if (IS_ERR(task)) {
 911                 rc = PTR_ERR(task);
 912                 cdt->cdt_state = CDT_STOPPED;
 913                 CERROR("%s: error starting coordinator thread: %d\n",
 914                        mdt_obd_name(mdt), rc);
 915                 RETURN(rc);
 916         } else {
 917                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
 918                        mdt_obd_name(mdt));
 919                 rc = 0;
 920         }
 921
 922         wait_event(cdt->cdt_waitq,
 923                        (cdt->cdt_flags & SVC_RUNNING));
 924
 925         cdt->cdt_state = CDT_RUNNING;
 926         mdt->mdt_opts.mo_coordinator = 1;
 927         RETURN(0);
 928 }
 929
 930 /**
 931  * stop a coordinator thread
 932  * \param mdt [IN] device
 933  */
 934 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
 935 {
 936         struct coordinator              *cdt = &mdt->mdt_coordinator;
 937         struct cdt_agent_req            *car, *tmp1;
 938         struct hsm_agent                *ha, *tmp2;
 939         struct cdt_restore_handle       *crh, *tmp3;
 940         struct mdt_thread_info          *cdt_mti;
 941         ENTRY;
 942
 943         if (cdt->cdt_state == CDT_STOPPED) {
 944                 CERROR("%s: Coordinator already stopped\n",
 945                        mdt_obd_name(mdt));
 946                 RETURN(-EALREADY);
 947         }
 948
 949         if (cdt->cdt_state != CDT_STOPPING) {
 950                 /* stop coordinator thread before cleaning */
 951                 cdt->cdt_flags = SVC_STOPPING;
 952                 wake_up(&cdt->cdt_waitq);
 953                 wait_event(cdt->cdt_waitq,
 954                            cdt->cdt_flags & SVC_STOPPED);
 955         }
 956         cdt->cdt_state = CDT_STOPPED;
 957
 958         /* start cleaning */
 959         down_write(&cdt->cdt_request_lock);
 960         list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
 961                                  car_request_list) {
 962                 cfs_hash_del(cdt->cdt_request_cookie_hash,
 963                              &car->car_hai->hai_cookie,
 964                              &car->car_cookie_hash);
 965                 list_del(&car->car_request_list);
 966                 mdt_cdt_put_request(car);
 967         }
 968         up_write(&cdt->cdt_request_lock);
 969
 970         down_write(&cdt->cdt_agent_lock);
 971         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 972                 list_del(&ha->ha_list);
 973                 OBD_FREE_PTR(ha);
 974         }
 975         up_write(&cdt->cdt_agent_lock);
 976
 977         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 978         mutex_lock(&cdt->cdt_restore_lock);
 979         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
 980                 struct mdt_object       *child;
 981
 982                 /* give back layout lock */
 983                 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
 984                 if (!IS_ERR(child))
 985                         mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
 986
 987                 list_del(&crh->crh_list);
 988
 989                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 990         }
 991         mutex_unlock(&cdt->cdt_restore_lock);
 992
 993         mdt->mdt_opts.mo_coordinator = 0;
 994
 995         RETURN(0);
 996 }
 997
 998 /**
 999  * register all requests from an hal in the memory list
1000  * \param mti [IN] context
1001  * \param hal [IN] request
1002  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1003  *  which is running the CT
1004  * \retval 0 success
1005  * \retval -ve failure
1006  */
1007 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1008                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1009 {
1010         struct mdt_device       *mdt = mti->mti_mdt;
1011         struct coordinator      *cdt = &mdt->mdt_coordinator;
1012         struct hsm_action_item  *hai;
1013         int                      rc = 0, i;
1014         ENTRY;
1015
1016         /* register request in memory list */
1017         hai = hai_first(hal);
1018         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1019                 struct cdt_agent_req *car;
1020
1021                 /* in case of a cancel request, we first mark the ondisk
1022                  * record of the request we want to stop as canceled
1023                  * this does not change the cancel record
1024                  * it will be done when updating the request status
1025                  */
1026                 if (hai->hai_action == HSMA_CANCEL) {
1027                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1028                                                      &hai->hai_cookie,
1029                                                      1, ARS_CANCELED);
1030                         if (rc) {
1031                                 CERROR("%s: mdt_agent_record_update() failed, "
1032                                        "rc=%d, cannot update status to %s "
1033                                        "for cookie %#llx\n",
1034                                        mdt_obd_name(mdt), rc,
1035                                        agent_req_status2name(ARS_CANCELED),
1036                                        hai->hai_cookie);
1037                                 GOTO(out, rc);
1038                         }
1039
1040                         /* find the running request to set it canceled */
1041                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1042                         if (car != NULL) {
1043                                 car->car_canceled = 1;
1044                                 /* uuid has to be changed to the one running the
1045                                 * request to cancel */
1046                                 *uuid = car->car_uuid;
1047                                 mdt_cdt_put_request(car);
1048                         }
1049                         /* no need to memorize cancel request
1050                          * this also avoid a deadlock when we receive
1051                          * a purge all requests command
1052                          */
1053                         continue;
1054                 }
1055
1056                 if (hai->hai_action == HSMA_ARCHIVE) {
1057                         struct mdt_object *obj;
1058                         struct md_hsm hsm;
1059
1060                         obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1061                         if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1062                                 continue;
1063                         if (IS_ERR(obj))
1064                                 GOTO(out, rc = PTR_ERR(obj));
1065
1066                         hsm.mh_flags |= HS_EXISTS;
1067                         hsm.mh_arch_id = hal->hal_archive_id;
1068                         rc = mdt_hsm_attr_set(mti, obj, &hsm);
1069                         mdt_object_put(mti->mti_env, obj);
1070                         if (rc)
1071                                 GOTO(out, rc);
1072                 }
1073
1074                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1075                                             hal->hal_archive_id, hal->hal_flags,
1076                                             uuid, hai);
1077                 if (IS_ERR(car))
1078                         GOTO(out, rc = PTR_ERR(car));
1079
1080                 rc = mdt_cdt_add_request(cdt, car);
1081                 if (rc != 0)
1082                         mdt_cdt_free_request(car);
1083         }
1084 out:
1085         RETURN(rc);
1086 }
1087
1088 /**
1089  * swap layouts between 2 fids
1090  * \param mti [IN] context
1091  * \param obj [IN]
1092  * \param dfid [IN]
1093  * \param mh_common [IN] MD HSM
1094  */
1095 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1096                             struct mdt_object *obj, const struct lu_fid *dfid,
1097                             struct md_hsm *mh_common)
1098 {
1099         struct mdt_object       *dobj;
1100         struct mdt_lock_handle  *dlh;
1101         int                      rc;
1102         ENTRY;
1103
1104         if (!mdt_object_exists(obj))
1105                 GOTO(out, rc = -ENOENT);
1106
1107         /* we already have layout lock on obj so take only
1108          * on dfid */
1109         dlh = &mti->mti_lh[MDT_LH_OLD];
1110         mdt_lock_reg_init(dlh, LCK_EX);
1111         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1112         if (IS_ERR(dobj))
1113                 GOTO(out, rc = PTR_ERR(dobj));
1114
1115         /* if copy tool closes the volatile before sending the final
1116          * progress through llapi_hsm_copy_end(), all the objects
1117          * are removed and mdd_swap_layout LBUG */
1118         if (!mdt_object_exists(dobj)) {
1119                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1120                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1121                 GOTO(out_dobj, rc = -ENOENT);
1122         }
1123         /* Since we only handle restores here, unconditionally use
1124          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1125          * be preserved in case of failure during swap_layout and not
1126          * leave a file in an intermediate but incoherent state.
1127          * But need to setup HSM xattr of data FID before, reuse
1128          * mti and mh presets for FID in hsm_cdt_request_completed(),
1129          * only need to clear RELEASED and DIRTY.
1130          */
1131         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1132         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1133         if (rc == 0)
1134                 rc = mo_swap_layouts(mti->mti_env,
1135                                      mdt_object_child(obj),
1136                                      mdt_object_child(dobj),
1137                                      SWAP_LAYOUTS_MDS_HSM);
1138
1139 out_dobj:
1140         mdt_object_unlock_put(mti, dobj, dlh, 1);
1141 out:
1142         RETURN(rc);
1143 }
1144
1145 /**
1146  * update status of a completed request
1147  * \param mti [IN] context
1148  * \param pgs [IN] progress of the copy tool
1149  * \param update_record [IN] update llog record
1150  * \retval 0 success
1151  * \retval -ve failure
1152  */
1153 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1154                                      struct hsm_progress_kernel *pgs,
1155                                      const struct cdt_agent_req *car,
1156                                      enum agent_req_status *status)
1157 {
1158         const struct lu_env     *env = mti->mti_env;
1159         struct mdt_device       *mdt = mti->mti_mdt;
1160         struct coordinator      *cdt = &mdt->mdt_coordinator;
1161         struct mdt_object       *obj = NULL;
1162         int                      cl_flags = 0, rc = 0;
1163         struct md_hsm            mh;
1164         bool                     is_mh_changed;
1165         bool                     need_changelog = true;
1166         ENTRY;
1167
1168         /* default is to retry */
1169         *status = ARS_WAITING;
1170
1171         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1172          * if error/removed continue anyway to get correct reporting done */
1173         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1174         /* we will update MD HSM only if needed */
1175         is_mh_changed = false;
1176
1177         /* no need to change mh->mh_arch_id
1178          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1179          */
1180         if (pgs->hpk_errval != 0) {
1181                 switch (pgs->hpk_errval) {
1182                 case ENOSYS:
1183                         /* the copy tool does not support cancel
1184                          * so the cancel request is failed
1185                          * As we cannot distinguish a cancel progress
1186                          * from another action progress (they have the
1187                          * same cookie), we suppose here the CT returns
1188                          * ENOSYS only if does not support cancel
1189                          */
1190                         /* this can also happen when cdt calls it to
1191                          * for a timed out request */
1192                         *status = ARS_FAILED;
1193                         /* to have a cancel event in changelog */
1194                         pgs->hpk_errval = ECANCELED;
1195                         break;
1196                 case ECANCELED:
1197                         /* the request record has already been set to
1198                          * ARS_CANCELED, this set the cancel request
1199                          * to ARS_SUCCEED */
1200                         *status = ARS_SUCCEED;
1201                         break;
1202                 default:
1203                         /* retry only if current policy or requested, and
1204                          * object is not on error/removed */
1205                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1206                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1207                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1208                         break;
1209                 }
1210
1211                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1212                         CERROR("%s: Request %#llx on "DFID
1213                                " failed, error code %d too large\n",
1214                                mdt_obd_name(mdt),
1215                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1216                                pgs->hpk_errval);
1217                         hsm_set_cl_error(&cl_flags,
1218                                          CLF_HSM_ERROVERFLOW);
1219                         rc = -EINVAL;
1220                 } else {
1221                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1222                 }
1223
1224                 switch (car->car_hai->hai_action) {
1225                 case HSMA_ARCHIVE:
1226                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1227                         break;
1228                 case HSMA_RESTORE:
1229                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1230                         break;
1231                 case HSMA_REMOVE:
1232                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1233                         break;
1234                 case HSMA_CANCEL:
1235                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1236                         CERROR("%s: Failed request %#llx on "DFID
1237                                " cannot be a CANCEL\n",
1238                                mdt_obd_name(mdt),
1239                                pgs->hpk_cookie,
1240                                PFID(&pgs->hpk_fid));
1241                         break;
1242                 default:
1243                         CERROR("%s: Failed request %#llx on "DFID
1244                                " %d is an unknown action\n",
1245                                mdt_obd_name(mdt),
1246                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1247                                car->car_hai->hai_action);
1248                         rc = -EINVAL;
1249                         break;
1250                 }
1251         } else {
1252                 *status = ARS_SUCCEED;
1253                 switch (car->car_hai->hai_action) {
1254                 case HSMA_ARCHIVE:
1255                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1256                         /* set ARCHIVE keep EXIST and clear LOST and
1257                          * DIRTY */
1258                         mh.mh_arch_ver = pgs->hpk_data_version;
1259                         mh.mh_flags |= HS_ARCHIVED;
1260                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1261                         is_mh_changed = true;
1262                         break;
1263                 case HSMA_RESTORE:
1264                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1265
1266                         /* do not clear RELEASED and DIRTY here
1267                          * this will occur in hsm_swap_layouts()
1268                          */
1269
1270                         /* Restoring has changed the file version on
1271                          * disk. */
1272                         mh.mh_arch_ver = pgs->hpk_data_version;
1273                         is_mh_changed = true;
1274                         break;
1275                 case HSMA_REMOVE:
1276                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1277                         /* clear ARCHIVED EXISTS and LOST */
1278                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1279                         is_mh_changed = true;
1280                         break;
1281                 case HSMA_CANCEL:
1282                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1283                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1284                                mdt_obd_name(mdt),
1285                                pgs->hpk_cookie,
1286                                PFID(&pgs->hpk_fid));
1287                         break;
1288                 default:
1289                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1290                                mdt_obd_name(mdt),
1291                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1292                                car->car_hai->hai_action);
1293                         rc = -EINVAL;
1294                         break;
1295                 }
1296         }
1297
1298         /* rc != 0 means error when analysing action, it may come from
1299          * a crasy CT no need to manage DIRTY
1300          */
1301         if (rc == 0)
1302                 hsm_set_cl_flags(&cl_flags,
1303                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1304
1305         /* unlock is done later, after layout lock management */
1306         if (is_mh_changed && !IS_ERR(obj))
1307                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1308
1309         /* we give back layout lock only if restore was successful or
1310          * if no retry will be attempted and if object is still alive,
1311          * in other cases we just unlock the object */
1312         if (car->car_hai->hai_action == HSMA_RESTORE) {
1313                 struct cdt_restore_handle       *crh;
1314
1315                 /* restore in data FID done, we swap the layouts
1316                  * only if restore is successful */
1317                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1318                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1319                                               &mh);
1320                         if (rc) {
1321                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1322                                         *status = ARS_FAILED;
1323                                 pgs->hpk_errval = -rc;
1324                         }
1325                 }
1326                 /* we have to retry, so keep layout lock */
1327                 if (*status == ARS_WAITING)
1328                         GOTO(out, rc);
1329
1330                 /* restore special case, need to create ChangeLog record
1331                  * before to give back layout lock to avoid concurrent
1332                  * file updater to post out of order ChangeLog */
1333                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1334                              &car->car_hai->hai_fid);
1335                 need_changelog = false;
1336
1337                 /* give back layout lock */
1338                 mutex_lock(&cdt->cdt_restore_lock);
1339                 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1340                 if (crh != NULL)
1341                         list_del(&crh->crh_list);
1342                 mutex_unlock(&cdt->cdt_restore_lock);
1343                 /* Just give back layout lock, we keep the reference
1344                  * which is given back later with the lock for HSM
1345                  * flags.
1346                  * XXX obj may be invalid so we do not pass it. */
1347                 if (crh != NULL)
1348                         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
1349
1350                 if (crh != NULL)
1351                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1352         }
1353
1354         GOTO(out, rc);
1355
1356 out:
1357         /* always add a ChangeLog record */
1358         if (need_changelog)
1359                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1360                              &car->car_hai->hai_fid);
1361
1362         if (!IS_ERR(obj))
1363                 mdt_object_put(mti->mti_env, obj);
1364
1365         RETURN(rc);
1366 }
1367
1368 /**
1369  * update status of a request
1370  * \param mti [IN] context
1371  * \param pgs [IN] progress of the copy tool
1372  * \param update_record [IN] update llog record
1373  * \retval 0 success
1374  * \retval -ve failure
1375  */
1376 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1377                                  struct hsm_progress_kernel *pgs,
1378                                  const int update_record)
1379 {
1380         struct mdt_device       *mdt = mti->mti_mdt;
1381         struct coordinator      *cdt = &mdt->mdt_coordinator;
1382         struct cdt_agent_req    *car;
1383         int                      rc = 0;
1384         ENTRY;
1385
1386         /* no coordinator started, so we cannot serve requests */
1387         if (cdt->cdt_state == CDT_STOPPED)
1388                 RETURN(-EAGAIN);
1389
1390         /* first do sanity checks */
1391         car = mdt_cdt_update_request(cdt, pgs);
1392         if (IS_ERR(car)) {
1393                 CERROR("%s: Cannot find running request for cookie %#llx"
1394                        " on fid="DFID"\n",
1395                        mdt_obd_name(mdt),
1396                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1397
1398                 RETURN(PTR_ERR(car));
1399         }
1400
1401         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1402                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1403                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1404                       hsm_copytool_action2name(car->car_hai->hai_action),
1405                       pgs->hpk_flags, pgs->hpk_errval,
1406                       PFID(&car->car_hai->hai_fid),
1407                       PFID(&car->car_hai->hai_dfid));
1408
1409         /* progress is done on FID or data FID depending of the action and
1410          * of the copy progress */
1411         /* for restore progress is used to send back the data FID to cdt */
1412         if (car->car_hai->hai_action == HSMA_RESTORE &&
1413             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1414                 car->car_hai->hai_dfid = pgs->hpk_fid;
1415
1416         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1417              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1418             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1419              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1420                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1421                        " does not match request FID "DFID" nor data FID "
1422                        DFID"\n",
1423                        mdt_obd_name(mdt),
1424                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1425                        PFID(&car->car_hai->hai_fid),
1426                        PFID(&car->car_hai->hai_dfid));
1427                 GOTO(out, rc = -EINVAL);
1428         }
1429
1430         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1431                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1432                        " is not coherent (err=%d and not completed"
1433                        " (flags=%d))\n",
1434                        mdt_obd_name(mdt),
1435                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1436                        hsm_copytool_action2name(car->car_hai->hai_action),
1437                        pgs->hpk_errval, pgs->hpk_flags);
1438                 GOTO(out, rc = -EINVAL);
1439         }
1440
1441         /* now progress is valid */
1442
1443         /* we use a root like ucred */
1444         hsm_init_ucred(mdt_ucred(mti));
1445
1446         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1447                 enum agent_req_status    status;
1448
1449                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1450
1451                 CDEBUG(D_HSM, "%s record: fid="DFID" cookie=%#llx action=%s "
1452                               "status=%s\n",
1453                        update_record ? "Updating" : "Not updating",
1454                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1455                        hsm_copytool_action2name(car->car_hai->hai_action),
1456                        agent_req_status2name(status));
1457
1458                 /* update record first (LU-9075) */
1459                 if (update_record) {
1460                         int rc1;
1461
1462                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1463                                                      &pgs->hpk_cookie, 1,
1464                                                      status);
1465                         if (rc1)
1466                                 CERROR("%s: mdt_agent_record_update() failed,"
1467                                        " rc=%d, cannot update status to %s"
1468                                        " for cookie %#llx\n",
1469                                        mdt_obd_name(mdt), rc1,
1470                                        agent_req_status2name(status),
1471                                        pgs->hpk_cookie);
1472                         rc = (rc != 0 ? rc : rc1);
1473                 }
1474
1475                 /* then remove request from memory list (LU-9075) */
1476                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1477
1478                 /* ct has completed a request, so a slot is available, wakeup
1479                  * cdt to find new work */
1480                 mdt_hsm_cdt_wakeup(mdt);
1481         } else {
1482                 /* if copytool send a progress on a canceled request
1483                  * we inform copytool it should stop
1484                  */
1485                 if (car->car_canceled == 1)
1486                         rc = -ECANCELED;
1487         }
1488         GOTO(out, rc);
1489
1490 out:
1491         /* remove ref got from mdt_cdt_update_request() */
1492         mdt_cdt_put_request(car);
1493
1494         return rc;
1495 }
1496
1497
1498 /**
1499  * data passed to llog_cat_process() callback
1500  * to cancel requests
1501  */
1502 struct hsm_cancel_all_data {
1503         struct mdt_device       *mdt;
1504 };
1505
1506 /**
1507  *  llog_cat_process() callback, used to:
1508  *  - purge all requests
1509  * \param env [IN] environment
1510  * \param llh [IN] llog handle
1511  * \param hdr [IN] llog record
1512  * \param data [IN] cb data = struct hsm_cancel_all_data
1513  * \retval 0 success
1514  * \retval -ve failure
1515  */
1516 static int mdt_cancel_all_cb(const struct lu_env *env,
1517                              struct llog_handle *llh,
1518                              struct llog_rec_hdr *hdr, void *data)
1519 {
1520         struct llog_agent_req_rec       *larr;
1521         struct hsm_cancel_all_data      *hcad;
1522         int                              rc = 0;
1523         ENTRY;
1524
1525         larr = (struct llog_agent_req_rec *)hdr;
1526         hcad = data;
1527         if (larr->arr_status == ARS_WAITING ||
1528             larr->arr_status == ARS_STARTED) {
1529                 larr->arr_status = ARS_CANCELED;
1530                 larr->arr_req_change = cfs_time_current_sec();
1531                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1532         }
1533
1534         RETURN(rc);
1535 }
1536
1537 /**
1538  * cancel all actions
1539  * \param obd [IN] MDT device
1540  */
1541 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1542 {
1543         struct lu_env                    env;
1544         struct lu_context                session;
1545         struct mdt_thread_info          *mti;
1546         struct coordinator              *cdt = &mdt->mdt_coordinator;
1547         struct cdt_agent_req            *car;
1548         struct hsm_action_list          *hal = NULL;
1549         struct hsm_action_item          *hai;
1550         struct hsm_cancel_all_data       hcad;
1551         int                              hal_sz = 0, hal_len, rc;
1552         enum cdt_states                  save_state;
1553         ENTRY;
1554
1555         rc = lu_env_init(&env, LCT_MD_THREAD);
1556         if (rc < 0)
1557                 RETURN(rc);
1558
1559         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1560         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1561         if (rc < 0)
1562                 GOTO(out_env, rc);
1563
1564         lu_context_enter(&session);
1565         env.le_ses = &session;
1566
1567         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1568         LASSERT(mti != NULL);
1569
1570         mti->mti_env = &env;
1571         mti->mti_mdt = mdt;
1572
1573         hsm_init_ucred(mdt_ucred(mti));
1574
1575         /* disable coordinator */
1576         save_state = cdt->cdt_state;
1577         cdt->cdt_state = CDT_DISABLE;
1578
1579         /* send cancel to all running requests */
1580         down_read(&cdt->cdt_request_lock);
1581         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1582                 mdt_cdt_get_request(car);
1583                 /* request is not yet removed from list, it will be done
1584                  * when copytool will return progress
1585                  */
1586
1587                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1588                         mdt_cdt_put_request(car);
1589                         continue;
1590                 }
1591
1592                 /* needed size */
1593                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1594                           cfs_size_round(car->car_hai->hai_len);
1595
1596                 if (hal_len > hal_sz && hal_sz > 0) {
1597                         /* not enough room, free old buffer */
1598                         OBD_FREE(hal, hal_sz);
1599                         hal = NULL;
1600                 }
1601
1602                 /* empty buffer, allocate one */
1603                 if (hal == NULL) {
1604                         hal_sz = hal_len;
1605                         OBD_ALLOC(hal, hal_sz);
1606                         if (hal == NULL) {
1607                                 mdt_cdt_put_request(car);
1608                                 up_read(&cdt->cdt_request_lock);
1609                                 GOTO(out_cdt_state, rc = -ENOMEM);
1610                         }
1611                 }
1612
1613                 hal->hal_version = HAL_VERSION;
1614                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1615                                 MTI_NAME_MAXLEN);
1616                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1617                 hal->hal_compound_id = car->car_compound_id;
1618                 hal->hal_archive_id = car->car_archive_id;
1619                 hal->hal_flags = car->car_flags;
1620                 hal->hal_count = 0;
1621
1622                 hai = hai_first(hal);
1623                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1624                 hai->hai_action = HSMA_CANCEL;
1625                 hal->hal_count = 1;
1626
1627                 /* it is possible to safely call mdt_hsm_agent_send()
1628                  * (ie without a deadlock on cdt_request_lock), because the
1629                  * write lock is taken only if we are not in purge mode
1630                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1631                  *   nor mdt_cdt_remove_request())
1632                  */
1633                 /* no conflict with cdt thread because cdt is disable and we
1634                  * have the request lock */
1635                 mdt_hsm_agent_send(mti, hal, 1);
1636
1637                 mdt_cdt_put_request(car);
1638         }
1639         up_read(&cdt->cdt_request_lock);
1640
1641         if (hal != NULL)
1642                 OBD_FREE(hal, hal_sz);
1643
1644         /* cancel all on-disk records */
1645         hcad.mdt = mdt;
1646
1647         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1648                               mdt_cancel_all_cb, &hcad);
1649 out_cdt_state:
1650         /* enable coordinator */
1651         cdt->cdt_state = save_state;
1652         lu_context_exit(&session);
1653         lu_context_fini(&session);
1654 out_env:
1655         lu_env_fini(&env);
1656
1657         RETURN(rc);
1658 }
1659
1660 /**
1661  * check if a request is compatible with file status
1662  * \param hai [IN] request description
1663  * \param archive_id [IN] request archive id
1664  * \param rq_flags [IN] request flags
1665  * \param hsm [IN] file HSM metadata
1666  * \retval boolean
1667  */
1668 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1669                               u32 archive_id, u64 rq_flags,
1670                               const struct md_hsm *hsm)
1671 {
1672         int      is_compat = false;
1673         int      hsm_flags;
1674         ENTRY;
1675
1676         hsm_flags = hsm->mh_flags;
1677         switch (hai->hai_action) {
1678         case HSMA_ARCHIVE:
1679                 if (!(hsm_flags & HS_NOARCHIVE) &&
1680                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1681                         is_compat = true;
1682
1683                 if (hsm_flags & HS_EXISTS &&
1684                     archive_id != 0 &&
1685                     archive_id != hsm->mh_arch_id)
1686                         is_compat = false;
1687
1688                 break;
1689         case HSMA_RESTORE:
1690                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1691                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1692                         is_compat = true;
1693                 break;
1694         case HSMA_REMOVE:
1695                 if (!(hsm_flags & HS_RELEASED) &&
1696                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1697                         is_compat = true;
1698                 break;
1699         case HSMA_CANCEL:
1700                 is_compat = true;
1701                 break;
1702         }
1703         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1704                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1705                       PFID(&hai->hai_fid),
1706                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1707                       hai->hai_extent.offset, hai->hai_extent.length,
1708                       hsm->mh_flags,
1709                       (is_compat ? "compatible" : "uncompatible"));
1710
1711         RETURN(is_compat);
1712 }
1713
1714 /*
1715  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1716  */
1717 static const struct {
1718         __u64            bit;
1719         char            *name;
1720         char            *nickname;
1721 } hsm_policy_names[] = {
1722         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1723         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1724         { 0 },
1725 };
1726
1727 /**
1728  * convert a policy name to a bit
1729  * \param name [IN] policy name
1730  * \retval 0 unknown
1731  * \retval   policy bit
1732  */
1733 static __u64 hsm_policy_str2bit(const char *name)
1734 {
1735         int      i;
1736
1737         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1738                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1739                     strcmp(hsm_policy_names[i].name, name) == 0)
1740                         return hsm_policy_names[i].bit;
1741         return 0;
1742 }
1743
1744 /**
1745  * convert a policy bit field to a string
1746  * \param mask [IN] policy bit field
1747  * \param hexa [IN] print mask before bit names
1748  * \param buffer [OUT] string
1749  * \param count [IN] size of buffer
1750  */
1751 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1752                                 const bool hexa)
1753 {
1754         int      i, j;
1755         __u64    bit;
1756         ENTRY;
1757
1758         if (hexa)
1759                 seq_printf(m, "(%#llx) ", mask);
1760
1761         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1762                 bit = (1ULL << i);
1763
1764                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1765                         if (hsm_policy_names[j].bit == bit)
1766                                 break;
1767                 }
1768                 if (bit & mask)
1769                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1770                 else
1771                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1772         }
1773         /* remove last ' ' */
1774         m->count--;
1775         seq_putc(m, '\n');
1776 }
1777
1778 /* methods to read/write HSM policy flags */
1779 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1780 {
1781         struct mdt_device       *mdt = m->private;
1782         struct coordinator      *cdt = &mdt->mdt_coordinator;
1783         ENTRY;
1784
1785         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1786         RETURN(0);
1787 }
1788
1789 static ssize_t
1790 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1791                          size_t count, loff_t *off)
1792 {
1793         struct seq_file         *m = file->private_data;
1794         struct mdt_device       *mdt = m->private;
1795         struct coordinator      *cdt = &mdt->mdt_coordinator;
1796         char                    *start, *token, sign;
1797         char                    *buf;
1798         __u64                    policy;
1799         __u64                    add_mask, remove_mask, set_mask;
1800         int                      rc;
1801         ENTRY;
1802
1803         if (count + 1 > PAGE_SIZE)
1804                 RETURN(-EINVAL);
1805
1806         OBD_ALLOC(buf, count + 1);
1807         if (buf == NULL)
1808                 RETURN(-ENOMEM);
1809
1810         if (copy_from_user(buf, buffer, count))
1811                 GOTO(out, rc = -EFAULT);
1812
1813         buf[count] = '\0';
1814
1815         start = buf;
1816         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1817                start);
1818
1819         add_mask = remove_mask = set_mask = 0;
1820         do {
1821                 token = strsep(&start, "\n ");
1822                 sign = *token;
1823
1824                 if (sign == '\0')
1825                         continue;
1826
1827                 if (sign == '-' || sign == '+')
1828                         token++;
1829
1830                 policy = hsm_policy_str2bit(token);
1831                 if (policy == 0) {
1832                         CWARN("%s: '%s' is unknown, "
1833                               "supported policies are:\n", mdt_obd_name(mdt),
1834                               token);
1835                         hsm_policy_bit2str(m, 0, false);
1836                         GOTO(out, rc = -EINVAL);
1837                 }
1838                 switch (sign) {
1839                 case '-':
1840                         remove_mask |= policy;
1841                         break;
1842                 case '+':
1843                         add_mask |= policy;
1844                         break;
1845                 default:
1846                         set_mask |= policy;
1847                         break;
1848                 }
1849
1850         } while (start != NULL);
1851
1852         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
1853                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1854
1855         /* if no sign in all string, it is a clear and set
1856          * if some sign found, all unsigned are converted
1857          * to add
1858          * P1 P2 = set to P1 and P2
1859          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1860          */
1861         if (remove_mask == 0 && add_mask == 0) {
1862                 cdt->cdt_policy = set_mask;
1863         } else {
1864                 cdt->cdt_policy |= set_mask | add_mask;
1865                 cdt->cdt_policy &= ~remove_mask;
1866         }
1867
1868         GOTO(out, rc = count);
1869
1870 out:
1871         OBD_FREE(buf, count + 1);
1872         RETURN(rc);
1873 }
1874 LPROC_SEQ_FOPS(mdt_hsm_policy);
1875
1876 #define GENERATE_PROC_METHOD(VAR)                                       \
1877 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
1878 {                                                                       \
1879         struct mdt_device       *mdt = m->private;                      \
1880         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1881         ENTRY;                                                          \
1882                                                                         \
1883         seq_printf(m, "%llu\n", (__u64)cdt->VAR);                       \
1884         RETURN(0);                                                      \
1885 }                                                                       \
1886 static ssize_t                                                          \
1887 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1888                           size_t count, loff_t *off)                    \
1889                                                                         \
1890 {                                                                       \
1891         struct seq_file         *m = file->private_data;                \
1892         struct mdt_device       *mdt = m->private;                      \
1893         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1894         __s64                    val;                                   \
1895         int                      rc;                                    \
1896         ENTRY;                                                          \
1897                                                                         \
1898         rc = lprocfs_str_to_s64(buffer, count, &val);                   \
1899         if (rc)                                                         \
1900                 RETURN(rc);                                             \
1901         if (val > 0 && val < INT_MAX) {                                 \
1902                 cdt->VAR = val;                                         \
1903                 RETURN(count);                                          \
1904         }                                                               \
1905         RETURN(-EINVAL);                                                \
1906 }                                                                       \
1907
1908 GENERATE_PROC_METHOD(cdt_loop_period)
1909 GENERATE_PROC_METHOD(cdt_grace_delay)
1910 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1911 GENERATE_PROC_METHOD(cdt_max_requests)
1912 GENERATE_PROC_METHOD(cdt_default_archive_id)
1913
1914 /*
1915  * procfs write method for MDT/hsm_control
1916  * proc entry is in mdt directory so data is mdt obd_device pointer
1917  */
1918 #define CDT_ENABLE_CMD   "enabled"
1919 #define CDT_STOP_CMD     "shutdown"
1920 #define CDT_DISABLE_CMD  "disabled"
1921 #define CDT_PURGE_CMD    "purge"
1922 #define CDT_HELP_CMD     "help"
1923 #define CDT_MAX_CMD_LEN  10
1924
1925 ssize_t
1926 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1927                               size_t count, loff_t *off)
1928 {
1929         struct seq_file         *m = file->private_data;
1930         struct obd_device       *obd = m->private;
1931         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
1932         struct coordinator      *cdt = &(mdt->mdt_coordinator);
1933         int                      rc, usage = 0;
1934         char                     kernbuf[CDT_MAX_CMD_LEN];
1935         ENTRY;
1936
1937         if (count == 0 || count >= sizeof(kernbuf))
1938                 RETURN(-EINVAL);
1939
1940         if (copy_from_user(kernbuf, buffer, count))
1941                 RETURN(-EFAULT);
1942         kernbuf[count] = 0;
1943
1944         if (kernbuf[count - 1] == '\n')
1945                 kernbuf[count - 1] = 0;
1946
1947         rc = 0;
1948         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1949                 if (cdt->cdt_state == CDT_DISABLE) {
1950                         cdt->cdt_state = CDT_RUNNING;
1951                         mdt_hsm_cdt_wakeup(mdt);
1952                 } else {
1953                         rc = mdt_hsm_cdt_start(mdt);
1954                 }
1955         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1956                 if ((cdt->cdt_state == CDT_STOPPING) ||
1957                     (cdt->cdt_state == CDT_STOPPED)) {
1958                         CERROR("%s: Coordinator already stopped\n",
1959                                mdt_obd_name(mdt));
1960                         rc = -EALREADY;
1961                 } else {
1962                         cdt->cdt_state = CDT_STOPPING;
1963                         mdt_hsm_cdt_wakeup(mdt);
1964                 }
1965         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
1966                 if ((cdt->cdt_state == CDT_STOPPING) ||
1967                     (cdt->cdt_state == CDT_STOPPED)) {
1968                         CERROR("%s: Coordinator is stopped\n",
1969                                mdt_obd_name(mdt));
1970                         rc = -EINVAL;
1971                 } else {
1972                         cdt->cdt_state = CDT_DISABLE;
1973                 }
1974         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
1975                 rc = hsm_cancel_all_actions(mdt);
1976         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
1977                 usage = 1;
1978         } else {
1979                 usage = 1;
1980                 rc = -EINVAL;
1981         }
1982
1983         if (usage == 1)
1984                 CERROR("%s: Valid coordinator control commands are: "
1985                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
1986                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1987                        CDT_PURGE_CMD, CDT_HELP_CMD);
1988
1989         if (rc)
1990                 RETURN(rc);
1991
1992         RETURN(count);
1993 }
1994
1995 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
1996 {
1997         struct obd_device       *obd = m->private;
1998         struct coordinator      *cdt;
1999         ENTRY;
2000
2001         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2002
2003         if (cdt->cdt_state == CDT_INIT)
2004                 seq_printf(m, "init\n");
2005         else if (cdt->cdt_state == CDT_RUNNING)
2006                 seq_printf(m, "enabled\n");
2007         else if (cdt->cdt_state == CDT_STOPPING)
2008                 seq_printf(m, "stopping\n");
2009         else if (cdt->cdt_state == CDT_STOPPED)
2010                 seq_printf(m, "stopped\n");
2011         else if (cdt->cdt_state == CDT_DISABLE)
2012                 seq_printf(m, "disabled\n");
2013         else
2014                 seq_printf(m, "unknown\n");
2015
2016         RETURN(0);
2017 }
2018
2019 static int
2020 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2021 {
2022         bool first = true;
2023         int i;
2024         ENTRY;
2025
2026         for (i = 0; i < 8 * sizeof(mask); i++) {
2027                 if (mask & (1UL << i)) {
2028                         seq_printf(m, "%s%s", first ? "" : " ",
2029                                    hsm_copytool_action2name(i));
2030                         first = false;
2031                 }
2032         }
2033         seq_putc(m, '\n');
2034
2035         RETURN(0);
2036 }
2037
2038 static int
2039 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2040 {
2041         struct mdt_device *mdt = m->private;
2042         struct coordinator *cdt = &mdt->mdt_coordinator;
2043
2044         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2045 }
2046
2047 static int
2048 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2049 {
2050         struct mdt_device *mdt = m->private;
2051         struct coordinator *cdt = &mdt->mdt_coordinator;
2052
2053         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2054 }
2055
2056 static int
2057 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2058 {
2059         struct mdt_device *mdt = m->private;
2060         struct coordinator *cdt = &mdt->mdt_coordinator;
2061
2062         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2063 }
2064
2065 static inline enum hsm_copytool_action
2066 hsm_copytool_name2action(const char *name)
2067 {
2068         if (strcasecmp(name, "NOOP") == 0)
2069                 return HSMA_NONE;
2070         else if (strcasecmp(name, "ARCHIVE") == 0)
2071                 return HSMA_ARCHIVE;
2072         else if (strcasecmp(name, "RESTORE") == 0)
2073                 return HSMA_RESTORE;
2074         else if (strcasecmp(name, "REMOVE") == 0)
2075                 return HSMA_REMOVE;
2076         else if (strcasecmp(name, "CANCEL") == 0)
2077                 return HSMA_CANCEL;
2078         else
2079                 return -1;
2080 }
2081
2082 static ssize_t
2083 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2084                             size_t user_count, __u64 *mask)
2085 {
2086         char *buf, *pos, *name;
2087         size_t buf_size;
2088         __u64 new_mask = 0;
2089         int rc;
2090         ENTRY;
2091
2092         if (!(user_count < 4096))
2093                 RETURN(-ENOMEM);
2094
2095         buf_size = user_count + 1;
2096
2097         OBD_ALLOC(buf, buf_size);
2098         if (buf == NULL)
2099                 RETURN(-ENOMEM);
2100
2101         if (copy_from_user(buf, user_buf, buf_size - 1))
2102                 GOTO(out, rc = -EFAULT);
2103
2104         buf[buf_size - 1] = '\0';
2105
2106         pos = buf;
2107         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2108                 int action;
2109
2110                 if (*name == '\0')
2111                         continue;
2112
2113                 action = hsm_copytool_name2action(name);
2114                 if (action < 0)
2115                         GOTO(out, rc = -EINVAL);
2116
2117                 new_mask |= (1UL << action);
2118         }
2119
2120         *mask = new_mask;
2121         rc = user_count;
2122 out:
2123         OBD_FREE(buf, buf_size);
2124
2125         RETURN(rc);
2126 }
2127
2128 static ssize_t
2129 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2130                                         size_t count, loff_t *off)
2131 {
2132         struct seq_file         *m = file->private_data;
2133         struct mdt_device       *mdt = m->private;
2134         struct coordinator *cdt = &mdt->mdt_coordinator;
2135
2136         return mdt_write_hsm_request_mask(file, buf, count,
2137                                            &cdt->cdt_user_request_mask);
2138 }
2139
2140 static ssize_t
2141 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2142                                         size_t count, loff_t *off)
2143 {
2144         struct seq_file         *m = file->private_data;
2145         struct mdt_device       *mdt = m->private;
2146         struct coordinator      *cdt = &mdt->mdt_coordinator;
2147
2148         return mdt_write_hsm_request_mask(file, buf, count,
2149                                            &cdt->cdt_group_request_mask);
2150 }
2151
2152 static ssize_t
2153 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2154                                         size_t count, loff_t *off)
2155 {
2156         struct seq_file         *m = file->private_data;
2157         struct mdt_device       *mdt = m->private;
2158         struct coordinator      *cdt = &mdt->mdt_coordinator;
2159
2160         return mdt_write_hsm_request_mask(file, buf, count,
2161                                            &cdt->cdt_other_request_mask);
2162 }
2163
2164 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2165 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2166 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2167 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2168 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2169 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2170 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2171 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2172
2173 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2174         { .name =       "agents",
2175           .fops =       &mdt_hsm_agent_fops                     },
2176         { .name =       "actions",
2177           .fops =       &mdt_hsm_actions_fops,
2178           .proc_mode =  0444                                    },
2179         { .name =       "default_archive_id",
2180           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2181         { .name =       "grace_delay",
2182           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2183         { .name =       "loop_period",
2184           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2185         { .name =       "max_requests",
2186           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2187         { .name =       "policy",
2188           .fops =       &mdt_hsm_policy_fops                    },
2189         { .name =       "active_request_timeout",
2190           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2191         { .name =       "active_requests",
2192           .fops =       &mdt_hsm_active_requests_fops           },
2193         { .name =       "user_request_mask",
2194           .fops =       &mdt_hsm_user_request_mask_fops,        },
2195         { .name =       "group_request_mask",
2196           .fops =       &mdt_hsm_group_request_mask_fops,       },
2197         { .name =       "other_request_mask",
2198           .fops =       &mdt_hsm_other_request_mask_fops,       },
2199         { 0 }
2200 };