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, NULL);
 286                 if (car == NULL) {
 287                         last = larr->arr_req_create;
 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_requests);
 780         INIT_LIST_HEAD(&cdt->cdt_agents);
 781         INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 782
 783         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 784         if (rc < 0)
 785                 RETURN(rc);
 786
 787         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
 788         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
 789         if (rc == 0) {
 790                 lu_context_enter(&cdt->cdt_session);
 791                 cdt->cdt_env.le_ses = &cdt->cdt_session;
 792         } else {
 793                 lu_env_fini(&cdt->cdt_env);
 794                 RETURN(rc);
 795         }
 796
 797         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 798         LASSERT(cdt_mti != NULL);
 799
 800         cdt_mti->mti_env = &cdt->cdt_env;
 801         cdt_mti->mti_mdt = mdt;
 802
 803         hsm_init_ucred(mdt_ucred(cdt_mti));
 804
 805         /* default values for /proc tunnables
 806          * can be override by MGS conf */
 807         cdt->cdt_default_archive_id = 1;
 808         cdt->cdt_grace_delay = 60;
 809         cdt->cdt_loop_period = 10;
 810         cdt->cdt_max_requests = 3;
 811         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 812         cdt->cdt_active_req_timeout = 3600;
 813
 814         RETURN(0);
 815 }
 816
 817 /**
 818  * free a coordinator thread
 819  * \param mdt [IN] device
 820  */
 821 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
 822 {
 823         struct coordinator *cdt = &mdt->mdt_coordinator;
 824         ENTRY;
 825
 826         lu_context_exit(cdt->cdt_env.le_ses);
 827         lu_context_fini(cdt->cdt_env.le_ses);
 828
 829         lu_env_fini(&cdt->cdt_env);
 830
 831         RETURN(0);
 832 }
 833
 834 /**
 835  * start a coordinator thread
 836  * \param mdt [IN] device
 837  * \retval 0 success
 838  * \retval -ve failure
 839  */
 840 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
 841 {
 842         struct coordinator      *cdt = &mdt->mdt_coordinator;
 843         int                      rc;
 844         void                    *ptr;
 845         struct mdt_thread_info  *cdt_mti;
 846         struct task_struct      *task;
 847         ENTRY;
 848
 849         /* functions defined but not yet used
 850          * this avoid compilation warning
 851          */
 852         ptr = dump_requests;
 853
 854         if (cdt->cdt_state != CDT_STOPPED) {
 855                 CERROR("%s: Coordinator already started\n",
 856                        mdt_obd_name(mdt));
 857                 RETURN(-EALREADY);
 858         }
 859
 860         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
 861         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 862
 863         cdt->cdt_state = CDT_INIT;
 864
 865         atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
 866         /* just need to be larger than previous one */
 867         /* cdt_last_cookie is protected by cdt_llog_lock */
 868         cdt->cdt_last_cookie = cfs_time_current_sec();
 869         atomic_set(&cdt->cdt_request_count, 0);
 870         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
 871         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
 872         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
 873
 874         /* to avoid deadlock when start is made through /proc
 875          * /proc entries are created by the coordinator thread */
 876
 877         /* set up list of started restore requests */
 878         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 879         rc = mdt_hsm_pending_restore(cdt_mti);
 880         if (rc)
 881                 CERROR("%s: cannot take the layout locks needed"
 882                        " for registered restore: %d\n",
 883                        mdt_obd_name(mdt), rc);
 884
 885         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
 886         if (IS_ERR(task)) {
 887                 rc = PTR_ERR(task);
 888                 cdt->cdt_state = CDT_STOPPED;
 889                 CERROR("%s: error starting coordinator thread: %d\n",
 890                        mdt_obd_name(mdt), rc);
 891                 RETURN(rc);
 892         } else {
 893                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
 894                        mdt_obd_name(mdt));
 895                 rc = 0;
 896         }
 897
 898         wait_event(cdt->cdt_waitq,
 899                        (cdt->cdt_flags & SVC_RUNNING));
 900
 901         cdt->cdt_state = CDT_RUNNING;
 902         mdt->mdt_opts.mo_coordinator = 1;
 903         RETURN(0);
 904 }
 905
 906 /**
 907  * stop a coordinator thread
 908  * \param mdt [IN] device
 909  */
 910 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
 911 {
 912         struct coordinator              *cdt = &mdt->mdt_coordinator;
 913         struct cdt_agent_req            *car, *tmp1;
 914         struct hsm_agent                *ha, *tmp2;
 915         struct cdt_restore_handle       *crh, *tmp3;
 916         struct mdt_thread_info          *cdt_mti;
 917         ENTRY;
 918
 919         if (cdt->cdt_state == CDT_STOPPED) {
 920                 CERROR("%s: Coordinator already stopped\n",
 921                        mdt_obd_name(mdt));
 922                 RETURN(-EALREADY);
 923         }
 924
 925         if (cdt->cdt_state != CDT_STOPPING) {
 926                 /* stop coordinator thread before cleaning */
 927                 cdt->cdt_flags = SVC_STOPPING;
 928                 wake_up(&cdt->cdt_waitq);
 929                 wait_event(cdt->cdt_waitq,
 930                            cdt->cdt_flags & SVC_STOPPED);
 931         }
 932         cdt->cdt_state = CDT_STOPPED;
 933
 934         /* start cleaning */
 935         down_write(&cdt->cdt_request_lock);
 936         list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
 937                                  car_request_list) {
 938                 list_del(&car->car_request_list);
 939                 mdt_cdt_free_request(car);
 940         }
 941         up_write(&cdt->cdt_request_lock);
 942
 943         down_write(&cdt->cdt_agent_lock);
 944         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 945                 list_del(&ha->ha_list);
 946                 OBD_FREE_PTR(ha);
 947         }
 948         up_write(&cdt->cdt_agent_lock);
 949
 950         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 951         mutex_lock(&cdt->cdt_restore_lock);
 952         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
 953                 struct mdt_object       *child;
 954
 955                 /* give back layout lock */
 956                 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
 957                 if (!IS_ERR(child))
 958                         mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
 959
 960                 list_del(&crh->crh_list);
 961
 962                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 963         }
 964         mutex_unlock(&cdt->cdt_restore_lock);
 965
 966         mdt->mdt_opts.mo_coordinator = 0;
 967
 968         RETURN(0);
 969 }
 970
 971 /**
 972  * register all requests from an hal in the memory list
 973  * \param mti [IN] context
 974  * \param hal [IN] request
 975  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
 976  *  which is running the CT
 977  * \retval 0 success
 978  * \retval -ve failure
 979  */
 980 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
 981                     struct hsm_action_list *hal, struct obd_uuid *uuid)
 982 {
 983         struct mdt_device       *mdt = mti->mti_mdt;
 984         struct coordinator      *cdt = &mdt->mdt_coordinator;
 985         struct hsm_action_item  *hai;
 986         int                      rc = 0, i;
 987         ENTRY;
 988
 989         /* register request in memory list */
 990         hai = hai_first(hal);
 991         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
 992                 struct cdt_agent_req *car;
 993
 994                 /* in case of a cancel request, we first mark the ondisk
 995                  * record of the request we want to stop as canceled
 996                  * this does not change the cancel record
 997                  * it will be done when updating the request status
 998                  */
 999                 if (hai->hai_action == HSMA_CANCEL) {
1000                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1001                                                      &hai->hai_cookie,
1002                                                      1, ARS_CANCELED);
1003                         if (rc) {
1004                                 CERROR("%s: mdt_agent_record_update() failed, "
1005                                        "rc=%d, cannot update status to %s "
1006                                        "for cookie %#llx\n",
1007                                        mdt_obd_name(mdt), rc,
1008                                        agent_req_status2name(ARS_CANCELED),
1009                                        hai->hai_cookie);
1010                                 GOTO(out, rc);
1011                         }
1012
1013                         /* find the running request to set it canceled */
1014                         car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1015                         if (car != NULL) {
1016                                 car->car_canceled = 1;
1017                                 /* uuid has to be changed to the one running the
1018                                 * request to cancel */
1019                                 *uuid = car->car_uuid;
1020                                 mdt_cdt_put_request(car);
1021                         }
1022                         /* no need to memorize cancel request
1023                          * this also avoid a deadlock when we receive
1024                          * a purge all requests command
1025                          */
1026                         continue;
1027                 }
1028
1029                 if (hai->hai_action == HSMA_ARCHIVE) {
1030                         struct mdt_object *obj;
1031                         struct md_hsm hsm;
1032
1033                         obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1034                         if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1035                                 continue;
1036                         if (IS_ERR(obj))
1037                                 GOTO(out, rc = PTR_ERR(obj));
1038
1039                         hsm.mh_flags |= HS_EXISTS;
1040                         hsm.mh_arch_id = hal->hal_archive_id;
1041                         rc = mdt_hsm_attr_set(mti, obj, &hsm);
1042                         mdt_object_put(mti->mti_env, obj);
1043                         if (rc)
1044                                 GOTO(out, rc);
1045                 }
1046
1047                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1048                                             hal->hal_archive_id, hal->hal_flags,
1049                                             uuid, hai);
1050                 if (IS_ERR(car))
1051                         GOTO(out, rc = PTR_ERR(car));
1052
1053                 rc = mdt_cdt_add_request(cdt, car);
1054                 if (rc != 0)
1055                         mdt_cdt_free_request(car);
1056         }
1057 out:
1058         RETURN(rc);
1059 }
1060
1061 /**
1062  * swap layouts between 2 fids
1063  * \param mti [IN] context
1064  * \param obj [IN]
1065  * \param dfid [IN]
1066  * \param mh_common [IN] MD HSM
1067  */
1068 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1069                             struct mdt_object *obj, const struct lu_fid *dfid,
1070                             struct md_hsm *mh_common)
1071 {
1072         struct mdt_object       *dobj;
1073         struct mdt_lock_handle  *dlh;
1074         int                      rc;
1075         ENTRY;
1076
1077         if (!mdt_object_exists(obj))
1078                 GOTO(out, rc = -ENOENT);
1079
1080         /* we already have layout lock on obj so take only
1081          * on dfid */
1082         dlh = &mti->mti_lh[MDT_LH_OLD];
1083         mdt_lock_reg_init(dlh, LCK_EX);
1084         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1085         if (IS_ERR(dobj))
1086                 GOTO(out, rc = PTR_ERR(dobj));
1087
1088         /* if copy tool closes the volatile before sending the final
1089          * progress through llapi_hsm_copy_end(), all the objects
1090          * are removed and mdd_swap_layout LBUG */
1091         if (!mdt_object_exists(dobj)) {
1092                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1093                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1094                 GOTO(out_dobj, rc = -ENOENT);
1095         }
1096         /* Since we only handle restores here, unconditionally use
1097          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1098          * be preserved in case of failure during swap_layout and not
1099          * leave a file in an intermediate but incoherent state.
1100          * But need to setup HSM xattr of data FID before, reuse
1101          * mti and mh presets for FID in hsm_cdt_request_completed(),
1102          * only need to clear RELEASED and DIRTY.
1103          */
1104         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1105         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1106         if (rc == 0)
1107                 rc = mo_swap_layouts(mti->mti_env,
1108                                      mdt_object_child(obj),
1109                                      mdt_object_child(dobj),
1110                                      SWAP_LAYOUTS_MDS_HSM);
1111
1112 out_dobj:
1113         mdt_object_unlock_put(mti, dobj, dlh, 1);
1114 out:
1115         RETURN(rc);
1116 }
1117
1118 /**
1119  * update status of a completed request
1120  * \param mti [IN] context
1121  * \param pgs [IN] progress of the copy tool
1122  * \param update_record [IN] update llog record
1123  * \retval 0 success
1124  * \retval -ve failure
1125  */
1126 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1127                                      struct hsm_progress_kernel *pgs,
1128                                      const struct cdt_agent_req *car,
1129                                      enum agent_req_status *status)
1130 {
1131         const struct lu_env     *env = mti->mti_env;
1132         struct mdt_device       *mdt = mti->mti_mdt;
1133         struct coordinator      *cdt = &mdt->mdt_coordinator;
1134         struct mdt_object       *obj = NULL;
1135         int                      cl_flags = 0, rc = 0;
1136         struct md_hsm            mh;
1137         bool                     is_mh_changed;
1138         ENTRY;
1139
1140         /* default is to retry */
1141         *status = ARS_WAITING;
1142
1143         /* find object by FID
1144          * if error/removed continue anyway to get correct reporting done */
1145         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1146         /* we will update MD HSM only if needed */
1147         is_mh_changed = false;
1148
1149         /* no need to change mh->mh_arch_id
1150          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1151          */
1152         if (pgs->hpk_errval != 0) {
1153                 switch (pgs->hpk_errval) {
1154                 case ENOSYS:
1155                         /* the copy tool does not support cancel
1156                          * so the cancel request is failed
1157                          * As we cannot distinguish a cancel progress
1158                          * from another action progress (they have the
1159                          * same cookie), we suppose here the CT returns
1160                          * ENOSYS only if does not support cancel
1161                          */
1162                         /* this can also happen when cdt calls it to
1163                          * for a timed out request */
1164                         *status = ARS_FAILED;
1165                         /* to have a cancel event in changelog */
1166                         pgs->hpk_errval = ECANCELED;
1167                         break;
1168                 case ECANCELED:
1169                         /* the request record has already been set to
1170                          * ARS_CANCELED, this set the cancel request
1171                          * to ARS_SUCCEED */
1172                         *status = ARS_SUCCEED;
1173                         break;
1174                 default:
1175                         /* retry only if current policy or requested, and
1176                          * object is not on error/removed */
1177                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1178                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1179                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1180                         break;
1181                 }
1182
1183                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1184                         CERROR("%s: Request %#llx on "DFID
1185                                " failed, error code %d too large\n",
1186                                mdt_obd_name(mdt),
1187                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1188                                pgs->hpk_errval);
1189                         hsm_set_cl_error(&cl_flags,
1190                                          CLF_HSM_ERROVERFLOW);
1191                         rc = -EINVAL;
1192                 } else {
1193                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1194                 }
1195
1196                 switch (car->car_hai->hai_action) {
1197                 case HSMA_ARCHIVE:
1198                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1199                         break;
1200                 case HSMA_RESTORE:
1201                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1202                         break;
1203                 case HSMA_REMOVE:
1204                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1205                         break;
1206                 case HSMA_CANCEL:
1207                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1208                         CERROR("%s: Failed request %#llx on "DFID
1209                                " cannot be a CANCEL\n",
1210                                mdt_obd_name(mdt),
1211                                pgs->hpk_cookie,
1212                                PFID(&pgs->hpk_fid));
1213                         break;
1214                 default:
1215                         CERROR("%s: Failed request %#llx on "DFID
1216                                " %d is an unknown action\n",
1217                                mdt_obd_name(mdt),
1218                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1219                                car->car_hai->hai_action);
1220                         rc = -EINVAL;
1221                         break;
1222                 }
1223         } else {
1224                 *status = ARS_SUCCEED;
1225                 switch (car->car_hai->hai_action) {
1226                 case HSMA_ARCHIVE:
1227                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1228                         /* set ARCHIVE keep EXIST and clear LOST and
1229                          * DIRTY */
1230                         mh.mh_arch_ver = pgs->hpk_data_version;
1231                         mh.mh_flags |= HS_ARCHIVED;
1232                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1233                         is_mh_changed = true;
1234                         break;
1235                 case HSMA_RESTORE:
1236                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1237
1238                         /* do not clear RELEASED and DIRTY here
1239                          * this will occur in hsm_swap_layouts()
1240                          */
1241
1242                         /* Restoring has changed the file version on
1243                          * disk. */
1244                         mh.mh_arch_ver = pgs->hpk_data_version;
1245                         is_mh_changed = true;
1246                         break;
1247                 case HSMA_REMOVE:
1248                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1249                         /* clear ARCHIVED EXISTS and LOST */
1250                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1251                         is_mh_changed = true;
1252                         break;
1253                 case HSMA_CANCEL:
1254                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1255                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1256                                mdt_obd_name(mdt),
1257                                pgs->hpk_cookie,
1258                                PFID(&pgs->hpk_fid));
1259                         break;
1260                 default:
1261                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1262                                mdt_obd_name(mdt),
1263                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1264                                car->car_hai->hai_action);
1265                         rc = -EINVAL;
1266                         break;
1267                 }
1268         }
1269
1270         /* rc != 0 means error when analysing action, it may come from
1271          * a crasy CT no need to manage DIRTY
1272          */
1273         if (rc == 0)
1274                 hsm_set_cl_flags(&cl_flags,
1275                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1276
1277         /* unlock is done later, after layout lock management */
1278         if (is_mh_changed && !IS_ERR(obj))
1279                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1280
1281         /* we give back layout lock only if restore was successful or
1282          * if no retry will be attempted and if object is still alive,
1283          * in other cases we just unlock the object */
1284         if (car->car_hai->hai_action == HSMA_RESTORE) {
1285                 struct cdt_restore_handle       *crh;
1286
1287                 /* restore in data FID done, we swap the layouts
1288                  * only if restore is successful */
1289                 if (pgs->hpk_errval == 0 && !IS_ERR_OR_NULL(obj)) {
1290                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1291                                               &mh);
1292                         if (rc) {
1293                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1294                                         *status = ARS_FAILED;
1295                                 pgs->hpk_errval = -rc;
1296                         }
1297                 }
1298                 /* we have to retry, so keep layout lock */
1299                 if (*status == ARS_WAITING)
1300                         GOTO(out, rc);
1301
1302                 /* give back layout lock */
1303                 mutex_lock(&cdt->cdt_restore_lock);
1304                 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1305                 if (crh != NULL)
1306                         list_del(&crh->crh_list);
1307                 mutex_unlock(&cdt->cdt_restore_lock);
1308                 /* Just give back layout lock, we keep the reference
1309                  * which is given back later with the lock for HSM
1310                  * flags.
1311                  * XXX obj may be invalid so we do not pass it. */
1312                 if (crh != NULL)
1313                         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
1314
1315                 if (crh != NULL)
1316                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1317         }
1318
1319         GOTO(out, rc);
1320
1321 out:
1322         /* always add a ChangeLog record */
1323         mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1324                      &car->car_hai->hai_fid);
1325
1326         if (!IS_ERR(obj))
1327                 mdt_object_put(mti->mti_env, obj);
1328
1329         RETURN(rc);
1330 }
1331
1332 /**
1333  * update status of a request
1334  * \param mti [IN] context
1335  * \param pgs [IN] progress of the copy tool
1336  * \param update_record [IN] update llog record
1337  * \retval 0 success
1338  * \retval -ve failure
1339  */
1340 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1341                                  struct hsm_progress_kernel *pgs,
1342                                  const int update_record)
1343 {
1344         struct mdt_device       *mdt = mti->mti_mdt;
1345         struct coordinator      *cdt = &mdt->mdt_coordinator;
1346         struct cdt_agent_req    *car;
1347         int                      rc = 0;
1348         ENTRY;
1349
1350         /* no coordinator started, so we cannot serve requests */
1351         if (cdt->cdt_state == CDT_STOPPED)
1352                 RETURN(-EAGAIN);
1353
1354         /* first do sanity checks */
1355         car = mdt_cdt_update_request(cdt, pgs);
1356         if (IS_ERR(car)) {
1357                 CERROR("%s: Cannot find running request for cookie %#llx"
1358                        " on fid="DFID"\n",
1359                        mdt_obd_name(mdt),
1360                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1361
1362                 RETURN(PTR_ERR(car));
1363         }
1364
1365         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1366                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1367                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1368                       hsm_copytool_action2name(car->car_hai->hai_action),
1369                       pgs->hpk_flags, pgs->hpk_errval,
1370                       PFID(&car->car_hai->hai_fid),
1371                       PFID(&car->car_hai->hai_dfid));
1372
1373         /* progress is done on FID or data FID depending of the action and
1374          * of the copy progress */
1375         /* for restore progress is used to send back the data FID to cdt */
1376         if (car->car_hai->hai_action == HSMA_RESTORE &&
1377             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1378                 car->car_hai->hai_dfid = pgs->hpk_fid;
1379
1380         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1381              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1382             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1383              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1384                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1385                        " does not match request FID "DFID" nor data FID "
1386                        DFID"\n",
1387                        mdt_obd_name(mdt),
1388                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1389                        PFID(&car->car_hai->hai_fid),
1390                        PFID(&car->car_hai->hai_dfid));
1391                 GOTO(out, rc = -EINVAL);
1392         }
1393
1394         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1395                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1396                        " is not coherent (err=%d and not completed"
1397                        " (flags=%d))\n",
1398                        mdt_obd_name(mdt),
1399                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1400                        hsm_copytool_action2name(car->car_hai->hai_action),
1401                        pgs->hpk_errval, pgs->hpk_flags);
1402                 GOTO(out, rc = -EINVAL);
1403         }
1404
1405         /* now progress is valid */
1406
1407         /* we use a root like ucred */
1408         hsm_init_ucred(mdt_ucred(mti));
1409
1410         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1411                 enum agent_req_status    status;
1412
1413                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1414
1415                 /* remove request from memory list */
1416                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1417
1418                 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie=%#llx"
1419                               " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1420                        pgs->hpk_cookie,
1421                        hsm_copytool_action2name(car->car_hai->hai_action),
1422                        agent_req_status2name(status));
1423
1424                 if (update_record) {
1425                         int rc1;
1426
1427                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1428                                                      &pgs->hpk_cookie, 1,
1429                                                      status);
1430                         if (rc1)
1431                                 CERROR("%s: mdt_agent_record_update() failed,"
1432                                        " rc=%d, cannot update status to %s"
1433                                        " for cookie %#llx\n",
1434                                        mdt_obd_name(mdt), rc1,
1435                                        agent_req_status2name(status),
1436                                        pgs->hpk_cookie);
1437                         rc = (rc != 0 ? rc : rc1);
1438                 }
1439                 /* ct has completed a request, so a slot is available, wakeup
1440                  * cdt to find new work */
1441                 mdt_hsm_cdt_wakeup(mdt);
1442         } else {
1443                 /* if copytool send a progress on a canceled request
1444                  * we inform copytool it should stop
1445                  */
1446                 if (car->car_canceled == 1)
1447                         rc = -ECANCELED;
1448         }
1449         GOTO(out, rc);
1450
1451 out:
1452         /* remove ref got from mdt_cdt_update_request() */
1453         mdt_cdt_put_request(car);
1454
1455         return rc;
1456 }
1457
1458
1459 /**
1460  * data passed to llog_cat_process() callback
1461  * to cancel requests
1462  */
1463 struct hsm_cancel_all_data {
1464         struct mdt_device       *mdt;
1465 };
1466
1467 /**
1468  *  llog_cat_process() callback, used to:
1469  *  - purge all requests
1470  * \param env [IN] environment
1471  * \param llh [IN] llog handle
1472  * \param hdr [IN] llog record
1473  * \param data [IN] cb data = struct hsm_cancel_all_data
1474  * \retval 0 success
1475  * \retval -ve failure
1476  */
1477 static int mdt_cancel_all_cb(const struct lu_env *env,
1478                              struct llog_handle *llh,
1479                              struct llog_rec_hdr *hdr, void *data)
1480 {
1481         struct llog_agent_req_rec       *larr;
1482         struct hsm_cancel_all_data      *hcad;
1483         int                              rc = 0;
1484         ENTRY;
1485
1486         larr = (struct llog_agent_req_rec *)hdr;
1487         hcad = data;
1488         if (larr->arr_status == ARS_WAITING ||
1489             larr->arr_status == ARS_STARTED) {
1490                 larr->arr_status = ARS_CANCELED;
1491                 larr->arr_req_change = cfs_time_current_sec();
1492                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1493         }
1494
1495         RETURN(rc);
1496 }
1497
1498 /**
1499  * cancel all actions
1500  * \param obd [IN] MDT device
1501  */
1502 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1503 {
1504         struct mdt_thread_info          *mti;
1505         struct coordinator              *cdt = &mdt->mdt_coordinator;
1506         struct cdt_agent_req            *car;
1507         struct hsm_action_list          *hal = NULL;
1508         struct hsm_action_item          *hai;
1509         struct hsm_cancel_all_data       hcad;
1510         int                              hal_sz = 0, hal_len, rc;
1511         enum cdt_states                  save_state;
1512         ENTRY;
1513
1514         /* retrieve coordinator context */
1515         mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1516
1517         /* disable coordinator */
1518         save_state = cdt->cdt_state;
1519         cdt->cdt_state = CDT_DISABLE;
1520
1521         /* send cancel to all running requests */
1522         down_read(&cdt->cdt_request_lock);
1523         list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1524                 mdt_cdt_get_request(car);
1525                 /* request is not yet removed from list, it will be done
1526                  * when copytool will return progress
1527                  */
1528
1529                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1530                         mdt_cdt_put_request(car);
1531                         continue;
1532                 }
1533
1534                 /* needed size */
1535                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1536                           cfs_size_round(car->car_hai->hai_len);
1537
1538                 if (hal_len > hal_sz && hal_sz > 0) {
1539                         /* not enough room, free old buffer */
1540                         OBD_FREE(hal, hal_sz);
1541                         hal = NULL;
1542                 }
1543
1544                 /* empty buffer, allocate one */
1545                 if (hal == NULL) {
1546                         hal_sz = hal_len;
1547                         OBD_ALLOC(hal, hal_sz);
1548                         if (hal == NULL) {
1549                                 mdt_cdt_put_request(car);
1550                                 up_read(&cdt->cdt_request_lock);
1551                                 GOTO(out, rc = -ENOMEM);
1552                         }
1553                 }
1554
1555                 hal->hal_version = HAL_VERSION;
1556                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1557                                 MTI_NAME_MAXLEN);
1558                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1559                 hal->hal_compound_id = car->car_compound_id;
1560                 hal->hal_archive_id = car->car_archive_id;
1561                 hal->hal_flags = car->car_flags;
1562                 hal->hal_count = 0;
1563
1564                 hai = hai_first(hal);
1565                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1566                 hai->hai_action = HSMA_CANCEL;
1567                 hal->hal_count = 1;
1568
1569                 /* it is possible to safely call mdt_hsm_agent_send()
1570                  * (ie without a deadlock on cdt_request_lock), because the
1571                  * write lock is taken only if we are not in purge mode
1572                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1573                  *   nor mdt_cdt_remove_request())
1574                  */
1575                 /* no conflict with cdt thread because cdt is disable and we
1576                  * have the request lock */
1577                 mdt_hsm_agent_send(mti, hal, 1);
1578
1579                 mdt_cdt_put_request(car);
1580         }
1581         up_read(&cdt->cdt_request_lock);
1582
1583         if (hal != NULL)
1584                 OBD_FREE(hal, hal_sz);
1585
1586         /* cancel all on-disk records */
1587         hcad.mdt = mdt;
1588
1589         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1590                               mdt_cancel_all_cb, &hcad);
1591 out:
1592         /* enable coordinator */
1593         cdt->cdt_state = save_state;
1594
1595         RETURN(rc);
1596 }
1597
1598 /**
1599  * check if a request is compatible with file status
1600  * \param hai [IN] request description
1601  * \param hal_an [IN] request archive number (not used)
1602  * \param rq_flags [IN] request flags
1603  * \param hsm [IN] file HSM metadata
1604  * \retval boolean
1605  */
1606 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1607                               const int hal_an, const __u64 rq_flags,
1608                               const struct md_hsm *hsm)
1609 {
1610         int      is_compat = false;
1611         int      hsm_flags;
1612         ENTRY;
1613
1614         hsm_flags = hsm->mh_flags;
1615         switch (hai->hai_action) {
1616         case HSMA_ARCHIVE:
1617                 if (!(hsm_flags & HS_NOARCHIVE) &&
1618                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1619                         is_compat = true;
1620                 break;
1621         case HSMA_RESTORE:
1622                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1623                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1624                         is_compat = true;
1625                 break;
1626         case HSMA_REMOVE:
1627                 if (!(hsm_flags & HS_RELEASED) &&
1628                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1629                         is_compat = true;
1630                 break;
1631         case HSMA_CANCEL:
1632                 is_compat = true;
1633                 break;
1634         }
1635         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1636                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1637                       PFID(&hai->hai_fid),
1638                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1639                       hai->hai_extent.offset, hai->hai_extent.length,
1640                       hsm->mh_flags,
1641                       (is_compat ? "compatible" : "uncompatible"));
1642
1643         RETURN(is_compat);
1644 }
1645
1646 /*
1647  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1648  */
1649 static const struct {
1650         __u64            bit;
1651         char            *name;
1652         char            *nickname;
1653 } hsm_policy_names[] = {
1654         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1655         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1656         { 0 },
1657 };
1658
1659 /**
1660  * convert a policy name to a bit
1661  * \param name [IN] policy name
1662  * \retval 0 unknown
1663  * \retval   policy bit
1664  */
1665 static __u64 hsm_policy_str2bit(const char *name)
1666 {
1667         int      i;
1668
1669         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1670                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1671                     strcmp(hsm_policy_names[i].name, name) == 0)
1672                         return hsm_policy_names[i].bit;
1673         return 0;
1674 }
1675
1676 /**
1677  * convert a policy bit field to a string
1678  * \param mask [IN] policy bit field
1679  * \param hexa [IN] print mask before bit names
1680  * \param buffer [OUT] string
1681  * \param count [IN] size of buffer
1682  */
1683 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1684                                 const bool hexa)
1685 {
1686         int      i, j;
1687         __u64    bit;
1688         ENTRY;
1689
1690         if (hexa)
1691                 seq_printf(m, "(%#llx) ", mask);
1692
1693         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1694                 bit = (1ULL << i);
1695
1696                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1697                         if (hsm_policy_names[j].bit == bit)
1698                                 break;
1699                 }
1700                 if (bit & mask)
1701                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1702                 else
1703                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1704         }
1705         /* remove last ' ' */
1706         m->count--;
1707         seq_putc(m, '\0');
1708 }
1709
1710 /* methods to read/write HSM policy flags */
1711 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1712 {
1713         struct mdt_device       *mdt = m->private;
1714         struct coordinator      *cdt = &mdt->mdt_coordinator;
1715         ENTRY;
1716
1717         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1718         RETURN(0);
1719 }
1720
1721 static ssize_t
1722 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1723                          size_t count, loff_t *off)
1724 {
1725         struct seq_file         *m = file->private_data;
1726         struct mdt_device       *mdt = m->private;
1727         struct coordinator      *cdt = &mdt->mdt_coordinator;
1728         char                    *start, *token, sign;
1729         char                    *buf;
1730         __u64                    policy;
1731         __u64                    add_mask, remove_mask, set_mask;
1732         int                      rc;
1733         ENTRY;
1734
1735         if (count + 1 > PAGE_SIZE)
1736                 RETURN(-EINVAL);
1737
1738         OBD_ALLOC(buf, count + 1);
1739         if (buf == NULL)
1740                 RETURN(-ENOMEM);
1741
1742         if (copy_from_user(buf, buffer, count))
1743                 GOTO(out, rc = -EFAULT);
1744
1745         buf[count] = '\0';
1746
1747         start = buf;
1748         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1749                start);
1750
1751         add_mask = remove_mask = set_mask = 0;
1752         do {
1753                 token = strsep(&start, "\n ");
1754                 sign = *token;
1755
1756                 if (sign == '\0')
1757                         continue;
1758
1759                 if (sign == '-' || sign == '+')
1760                         token++;
1761
1762                 policy = hsm_policy_str2bit(token);
1763                 if (policy == 0) {
1764                         CWARN("%s: '%s' is unknown, "
1765                               "supported policies are:\n", mdt_obd_name(mdt),
1766                               token);
1767                         hsm_policy_bit2str(m, 0, false);
1768                         GOTO(out, rc = -EINVAL);
1769                 }
1770                 switch (sign) {
1771                 case '-':
1772                         remove_mask |= policy;
1773                         break;
1774                 case '+':
1775                         add_mask |= policy;
1776                         break;
1777                 default:
1778                         set_mask |= policy;
1779                         break;
1780                 }
1781
1782         } while (start != NULL);
1783
1784         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
1785                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1786
1787         /* if no sign in all string, it is a clear and set
1788          * if some sign found, all unsigned are converted
1789          * to add
1790          * P1 P2 = set to P1 and P2
1791          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1792          */
1793         if (remove_mask == 0 && add_mask == 0) {
1794                 cdt->cdt_policy = set_mask;
1795         } else {
1796                 cdt->cdt_policy |= set_mask | add_mask;
1797                 cdt->cdt_policy &= ~remove_mask;
1798         }
1799
1800         GOTO(out, rc = count);
1801
1802 out:
1803         OBD_FREE(buf, count + 1);
1804         RETURN(rc);
1805 }
1806 LPROC_SEQ_FOPS(mdt_hsm_policy);
1807
1808 #define GENERATE_PROC_METHOD(VAR)                                       \
1809 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
1810 {                                                                       \
1811         struct mdt_device       *mdt = m->private;                      \
1812         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1813         ENTRY;                                                          \
1814                                                                         \
1815         seq_printf(m, "%llu\n", (__u64)cdt->VAR);                       \
1816         RETURN(0);                                                      \
1817 }                                                                       \
1818 static ssize_t                                                          \
1819 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1820                           size_t count, loff_t *off)                    \
1821                                                                         \
1822 {                                                                       \
1823         struct seq_file         *m = file->private_data;                \
1824         struct mdt_device       *mdt = m->private;                      \
1825         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1826         __s64                    val;                                   \
1827         int                      rc;                                    \
1828         ENTRY;                                                          \
1829                                                                         \
1830         rc = lprocfs_str_to_s64(buffer, count, &val);                   \
1831         if (rc)                                                         \
1832                 RETURN(rc);                                             \
1833         if (val > 0 && val < INT_MAX) {                                 \
1834                 cdt->VAR = val;                                         \
1835                 RETURN(count);                                          \
1836         }                                                               \
1837         RETURN(-EINVAL);                                                \
1838 }                                                                       \
1839
1840 GENERATE_PROC_METHOD(cdt_loop_period)
1841 GENERATE_PROC_METHOD(cdt_grace_delay)
1842 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1843 GENERATE_PROC_METHOD(cdt_max_requests)
1844 GENERATE_PROC_METHOD(cdt_default_archive_id)
1845
1846 /*
1847  * procfs write method for MDT/hsm_control
1848  * proc entry is in mdt directory so data is mdt obd_device pointer
1849  */
1850 #define CDT_ENABLE_CMD   "enabled"
1851 #define CDT_STOP_CMD     "shutdown"
1852 #define CDT_DISABLE_CMD  "disabled"
1853 #define CDT_PURGE_CMD    "purge"
1854 #define CDT_HELP_CMD     "help"
1855 #define CDT_MAX_CMD_LEN  10
1856
1857 ssize_t
1858 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1859                               size_t count, loff_t *off)
1860 {
1861         struct seq_file         *m = file->private_data;
1862         struct obd_device       *obd = m->private;
1863         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
1864         struct coordinator      *cdt = &(mdt->mdt_coordinator);
1865         int                      rc, usage = 0;
1866         char                     kernbuf[CDT_MAX_CMD_LEN];
1867         ENTRY;
1868
1869         if (count == 0 || count >= sizeof(kernbuf))
1870                 RETURN(-EINVAL);
1871
1872         if (copy_from_user(kernbuf, buffer, count))
1873                 RETURN(-EFAULT);
1874         kernbuf[count] = 0;
1875
1876         if (kernbuf[count - 1] == '\n')
1877                 kernbuf[count - 1] = 0;
1878
1879         rc = 0;
1880         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1881                 if (cdt->cdt_state == CDT_DISABLE) {
1882                         cdt->cdt_state = CDT_RUNNING;
1883                         mdt_hsm_cdt_wakeup(mdt);
1884                 } else {
1885                         rc = mdt_hsm_cdt_start(mdt);
1886                 }
1887         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1888                 if ((cdt->cdt_state == CDT_STOPPING) ||
1889                     (cdt->cdt_state == CDT_STOPPED)) {
1890                         CERROR("%s: Coordinator already stopped\n",
1891                                mdt_obd_name(mdt));
1892                         rc = -EALREADY;
1893                 } else {
1894                         cdt->cdt_state = CDT_STOPPING;
1895                 }
1896         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
1897                 if ((cdt->cdt_state == CDT_STOPPING) ||
1898                     (cdt->cdt_state == CDT_STOPPED)) {
1899                         CERROR("%s: Coordinator is stopped\n",
1900                                mdt_obd_name(mdt));
1901                         rc = -EINVAL;
1902                 } else {
1903                         cdt->cdt_state = CDT_DISABLE;
1904                 }
1905         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
1906                 rc = hsm_cancel_all_actions(mdt);
1907         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
1908                 usage = 1;
1909         } else {
1910                 usage = 1;
1911                 rc = -EINVAL;
1912         }
1913
1914         if (usage == 1)
1915                 CERROR("%s: Valid coordinator control commands are: "
1916                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
1917                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1918                        CDT_PURGE_CMD, CDT_HELP_CMD);
1919
1920         if (rc)
1921                 RETURN(rc);
1922
1923         RETURN(count);
1924 }
1925
1926 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
1927 {
1928         struct obd_device       *obd = m->private;
1929         struct coordinator      *cdt;
1930         ENTRY;
1931
1932         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
1933
1934         if (cdt->cdt_state == CDT_INIT)
1935                 seq_printf(m, "init\n");
1936         else if (cdt->cdt_state == CDT_RUNNING)
1937                 seq_printf(m, "enabled\n");
1938         else if (cdt->cdt_state == CDT_STOPPING)
1939                 seq_printf(m, "stopping\n");
1940         else if (cdt->cdt_state == CDT_STOPPED)
1941                 seq_printf(m, "stopped\n");
1942         else if (cdt->cdt_state == CDT_DISABLE)
1943                 seq_printf(m, "disabled\n");
1944         else
1945                 seq_printf(m, "unknown\n");
1946
1947         RETURN(0);
1948 }
1949
1950 static int
1951 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
1952 {
1953         bool first = true;
1954         int i;
1955         ENTRY;
1956
1957         for (i = 0; i < 8 * sizeof(mask); i++) {
1958                 if (mask & (1UL << i)) {
1959                         seq_printf(m, "%s%s", first ? "" : " ",
1960                                    hsm_copytool_action2name(i));
1961                         first = false;
1962                 }
1963         }
1964         seq_putc(m, '\n');
1965
1966         RETURN(0);
1967 }
1968
1969 static int
1970 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
1971 {
1972         struct mdt_device *mdt = m->private;
1973         struct coordinator *cdt = &mdt->mdt_coordinator;
1974
1975         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
1976 }
1977
1978 static int
1979 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
1980 {
1981         struct mdt_device *mdt = m->private;
1982         struct coordinator *cdt = &mdt->mdt_coordinator;
1983
1984         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
1985 }
1986
1987 static int
1988 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
1989 {
1990         struct mdt_device *mdt = m->private;
1991         struct coordinator *cdt = &mdt->mdt_coordinator;
1992
1993         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
1994 }
1995
1996 static inline enum hsm_copytool_action
1997 hsm_copytool_name2action(const char *name)
1998 {
1999         if (strcasecmp(name, "NOOP") == 0)
2000                 return HSMA_NONE;
2001         else if (strcasecmp(name, "ARCHIVE") == 0)
2002                 return HSMA_ARCHIVE;
2003         else if (strcasecmp(name, "RESTORE") == 0)
2004                 return HSMA_RESTORE;
2005         else if (strcasecmp(name, "REMOVE") == 0)
2006                 return HSMA_REMOVE;
2007         else if (strcasecmp(name, "CANCEL") == 0)
2008                 return HSMA_CANCEL;
2009         else
2010                 return -1;
2011 }
2012
2013 static ssize_t
2014 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2015                             size_t user_count, __u64 *mask)
2016 {
2017         char *buf, *pos, *name;
2018         size_t buf_size;
2019         __u64 new_mask = 0;
2020         int rc;
2021         ENTRY;
2022
2023         if (!(user_count < 4096))
2024                 RETURN(-ENOMEM);
2025
2026         buf_size = user_count + 1;
2027
2028         OBD_ALLOC(buf, buf_size);
2029         if (buf == NULL)
2030                 RETURN(-ENOMEM);
2031
2032         if (copy_from_user(buf, user_buf, buf_size - 1))
2033                 GOTO(out, rc = -EFAULT);
2034
2035         buf[buf_size - 1] = '\0';
2036
2037         pos = buf;
2038         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2039                 int action;
2040
2041                 if (*name == '\0')
2042                         continue;
2043
2044                 action = hsm_copytool_name2action(name);
2045                 if (action < 0)
2046                         GOTO(out, rc = -EINVAL);
2047
2048                 new_mask |= (1UL << action);
2049         }
2050
2051         *mask = new_mask;
2052         rc = user_count;
2053 out:
2054         OBD_FREE(buf, buf_size);
2055
2056         RETURN(rc);
2057 }
2058
2059 static ssize_t
2060 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2061                                         size_t count, loff_t *off)
2062 {
2063         struct seq_file         *m = file->private_data;
2064         struct mdt_device       *mdt = m->private;
2065         struct coordinator *cdt = &mdt->mdt_coordinator;
2066
2067         return mdt_write_hsm_request_mask(file, buf, count,
2068                                            &cdt->cdt_user_request_mask);
2069 }
2070
2071 static ssize_t
2072 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2073                                         size_t count, loff_t *off)
2074 {
2075         struct seq_file         *m = file->private_data;
2076         struct mdt_device       *mdt = m->private;
2077         struct coordinator      *cdt = &mdt->mdt_coordinator;
2078
2079         return mdt_write_hsm_request_mask(file, buf, count,
2080                                            &cdt->cdt_group_request_mask);
2081 }
2082
2083 static ssize_t
2084 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2085                                         size_t count, loff_t *off)
2086 {
2087         struct seq_file         *m = file->private_data;
2088         struct mdt_device       *mdt = m->private;
2089         struct coordinator      *cdt = &mdt->mdt_coordinator;
2090
2091         return mdt_write_hsm_request_mask(file, buf, count,
2092                                            &cdt->cdt_other_request_mask);
2093 }
2094
2095 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2096 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2097 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2098 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2099 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2100 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2101 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2102 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2103
2104 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2105         { .name =       "agents",
2106           .fops =       &mdt_hsm_agent_fops                     },
2107         { .name =       "actions",
2108           .fops =       &mdt_hsm_actions_fops,
2109           .proc_mode =  0444                                    },
2110         { .name =       "default_archive_id",
2111           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2112         { .name =       "grace_delay",
2113           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2114         { .name =       "loop_period",
2115           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2116         { .name =       "max_requests",
2117           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2118         { .name =       "policy",
2119           .fops =       &mdt_hsm_policy_fops                    },
2120         { .name =       "active_request_timeout",
2121           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2122         { .name =       "active_requests",
2123           .fops =       &mdt_hsm_active_requests_fops           },
2124         { .name =       "user_request_mask",
2125           .fops =       &mdt_hsm_user_request_mask_fops,        },
2126         { .name =       "group_request_mask",
2127           .fops =       &mdt_hsm_group_request_mask_fops,       },
2128         { .name =       "other_request_mask",
2129           .fops =       &mdt_hsm_other_request_mask_fops,       },
2130         { 0 }
2131 };