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_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_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         if (mdt->mdt_bottom->dd_rdonly)
 886                 RETURN(0);
 887
 888         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
 889         if (IS_ERR(task)) {
 890                 rc = PTR_ERR(task);
 891                 cdt->cdt_state = CDT_STOPPED;
 892                 CERROR("%s: error starting coordinator thread: %d\n",
 893                        mdt_obd_name(mdt), rc);
 894                 RETURN(rc);
 895         } else {
 896                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
 897                        mdt_obd_name(mdt));
 898                 rc = 0;
 899         }
 900
 901         wait_event(cdt->cdt_waitq,
 902                        (cdt->cdt_flags & SVC_RUNNING));
 903
 904         cdt->cdt_state = CDT_RUNNING;
 905         mdt->mdt_opts.mo_coordinator = 1;
 906         RETURN(0);
 907 }
 908
 909 /**
 910  * stop a coordinator thread
 911  * \param mdt [IN] device
 912  */
 913 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
 914 {
 915         struct coordinator              *cdt = &mdt->mdt_coordinator;
 916         struct cdt_agent_req            *car, *tmp1;
 917         struct hsm_agent                *ha, *tmp2;
 918         struct cdt_restore_handle       *crh, *tmp3;
 919         struct mdt_thread_info          *cdt_mti;
 920         ENTRY;
 921
 922         if (cdt->cdt_state == CDT_STOPPED) {
 923                 CERROR("%s: Coordinator already stopped\n",
 924                        mdt_obd_name(mdt));
 925                 RETURN(-EALREADY);
 926         }
 927
 928         if (cdt->cdt_state != CDT_STOPPING) {
 929                 /* stop coordinator thread before cleaning */
 930                 cdt->cdt_flags = SVC_STOPPING;
 931                 wake_up(&cdt->cdt_waitq);
 932                 wait_event(cdt->cdt_waitq,
 933                            cdt->cdt_flags & SVC_STOPPED);
 934         }
 935         cdt->cdt_state = CDT_STOPPED;
 936
 937         /* start cleaning */
 938         down_write(&cdt->cdt_request_lock);
 939         list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
 940                                  car_request_list) {
 941                 list_del(&car->car_request_list);
 942                 mdt_cdt_free_request(car);
 943         }
 944         up_write(&cdt->cdt_request_lock);
 945
 946         down_write(&cdt->cdt_agent_lock);
 947         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 948                 list_del(&ha->ha_list);
 949                 OBD_FREE_PTR(ha);
 950         }
 951         up_write(&cdt->cdt_agent_lock);
 952
 953         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 954         mutex_lock(&cdt->cdt_restore_lock);
 955         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
 956                 struct mdt_object       *child;
 957
 958                 /* give back layout lock */
 959                 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
 960                 if (!IS_ERR(child))
 961                         mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
 962
 963                 list_del(&crh->crh_list);
 964
 965                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 966         }
 967         mutex_unlock(&cdt->cdt_restore_lock);
 968
 969         mdt->mdt_opts.mo_coordinator = 0;
 970
 971         RETURN(0);
 972 }
 973
 974 /**
 975  * register all requests from an hal in the memory list
 976  * \param mti [IN] context
 977  * \param hal [IN] request
 978  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
 979  *  which is running the CT
 980  * \retval 0 success
 981  * \retval -ve failure
 982  */
 983 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
 984                     struct hsm_action_list *hal, struct obd_uuid *uuid)
 985 {
 986         struct mdt_device       *mdt = mti->mti_mdt;
 987         struct coordinator      *cdt = &mdt->mdt_coordinator;
 988         struct hsm_action_item  *hai;
 989         int                      rc = 0, i;
 990         ENTRY;
 991
 992         /* register request in memory list */
 993         hai = hai_first(hal);
 994         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
 995                 struct cdt_agent_req *car;
 996
 997                 /* in case of a cancel request, we first mark the ondisk
 998                  * record of the request we want to stop as canceled
 999                  * this does not change the cancel record
1000                  * it will be done when updating the request status
1001                  */
1002                 if (hai->hai_action == HSMA_CANCEL) {
1003                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1004                                                      &hai->hai_cookie,
1005                                                      1, ARS_CANCELED);
1006                         if (rc) {
1007                                 CERROR("%s: mdt_agent_record_update() failed, "
1008                                        "rc=%d, cannot update status to %s "
1009                                        "for cookie %#llx\n",
1010                                        mdt_obd_name(mdt), rc,
1011                                        agent_req_status2name(ARS_CANCELED),
1012                                        hai->hai_cookie);
1013                                 GOTO(out, rc);
1014                         }
1015
1016                         /* find the running request to set it canceled */
1017                         car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1018                         if (car != NULL) {
1019                                 car->car_canceled = 1;
1020                                 /* uuid has to be changed to the one running the
1021                                 * request to cancel */
1022                                 *uuid = car->car_uuid;
1023                                 mdt_cdt_put_request(car);
1024                         }
1025                         /* no need to memorize cancel request
1026                          * this also avoid a deadlock when we receive
1027                          * a purge all requests command
1028                          */
1029                         continue;
1030                 }
1031
1032                 if (hai->hai_action == HSMA_ARCHIVE) {
1033                         struct mdt_object *obj;
1034                         struct md_hsm hsm;
1035
1036                         obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1037                         if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1038                                 continue;
1039                         if (IS_ERR(obj))
1040                                 GOTO(out, rc = PTR_ERR(obj));
1041
1042                         hsm.mh_flags |= HS_EXISTS;
1043                         hsm.mh_arch_id = hal->hal_archive_id;
1044                         rc = mdt_hsm_attr_set(mti, obj, &hsm);
1045                         mdt_object_put(mti->mti_env, obj);
1046                         if (rc)
1047                                 GOTO(out, rc);
1048                 }
1049
1050                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1051                                             hal->hal_archive_id, hal->hal_flags,
1052                                             uuid, hai);
1053                 if (IS_ERR(car))
1054                         GOTO(out, rc = PTR_ERR(car));
1055
1056                 rc = mdt_cdt_add_request(cdt, car);
1057                 if (rc != 0)
1058                         mdt_cdt_free_request(car);
1059         }
1060 out:
1061         RETURN(rc);
1062 }
1063
1064 /**
1065  * swap layouts between 2 fids
1066  * \param mti [IN] context
1067  * \param obj [IN]
1068  * \param dfid [IN]
1069  * \param mh_common [IN] MD HSM
1070  */
1071 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1072                             struct mdt_object *obj, const struct lu_fid *dfid,
1073                             struct md_hsm *mh_common)
1074 {
1075         struct mdt_object       *dobj;
1076         struct mdt_lock_handle  *dlh;
1077         int                      rc;
1078         ENTRY;
1079
1080         if (!mdt_object_exists(obj))
1081                 GOTO(out, rc = -ENOENT);
1082
1083         /* we already have layout lock on obj so take only
1084          * on dfid */
1085         dlh = &mti->mti_lh[MDT_LH_OLD];
1086         mdt_lock_reg_init(dlh, LCK_EX);
1087         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1088         if (IS_ERR(dobj))
1089                 GOTO(out, rc = PTR_ERR(dobj));
1090
1091         /* if copy tool closes the volatile before sending the final
1092          * progress through llapi_hsm_copy_end(), all the objects
1093          * are removed and mdd_swap_layout LBUG */
1094         if (!mdt_object_exists(dobj)) {
1095                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1096                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1097                 GOTO(out_dobj, rc = -ENOENT);
1098         }
1099         /* Since we only handle restores here, unconditionally use
1100          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1101          * be preserved in case of failure during swap_layout and not
1102          * leave a file in an intermediate but incoherent state.
1103          * But need to setup HSM xattr of data FID before, reuse
1104          * mti and mh presets for FID in hsm_cdt_request_completed(),
1105          * only need to clear RELEASED and DIRTY.
1106          */
1107         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1108         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1109         if (rc == 0)
1110                 rc = mo_swap_layouts(mti->mti_env,
1111                                      mdt_object_child(obj),
1112                                      mdt_object_child(dobj),
1113                                      SWAP_LAYOUTS_MDS_HSM);
1114
1115 out_dobj:
1116         mdt_object_unlock_put(mti, dobj, dlh, 1);
1117 out:
1118         RETURN(rc);
1119 }
1120
1121 /**
1122  * update status of a completed request
1123  * \param mti [IN] context
1124  * \param pgs [IN] progress of the copy tool
1125  * \param update_record [IN] update llog record
1126  * \retval 0 success
1127  * \retval -ve failure
1128  */
1129 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1130                                      struct hsm_progress_kernel *pgs,
1131                                      const struct cdt_agent_req *car,
1132                                      enum agent_req_status *status)
1133 {
1134         const struct lu_env     *env = mti->mti_env;
1135         struct mdt_device       *mdt = mti->mti_mdt;
1136         struct coordinator      *cdt = &mdt->mdt_coordinator;
1137         struct mdt_object       *obj = NULL;
1138         int                      cl_flags = 0, rc = 0;
1139         struct md_hsm            mh;
1140         bool                     is_mh_changed;
1141         bool                     need_changelog = true;
1142         ENTRY;
1143
1144         /* default is to retry */
1145         *status = ARS_WAITING;
1146
1147         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1148          * if error/removed continue anyway to get correct reporting done */
1149         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1150         /* we will update MD HSM only if needed */
1151         is_mh_changed = false;
1152
1153         /* no need to change mh->mh_arch_id
1154          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1155          */
1156         if (pgs->hpk_errval != 0) {
1157                 switch (pgs->hpk_errval) {
1158                 case ENOSYS:
1159                         /* the copy tool does not support cancel
1160                          * so the cancel request is failed
1161                          * As we cannot distinguish a cancel progress
1162                          * from another action progress (they have the
1163                          * same cookie), we suppose here the CT returns
1164                          * ENOSYS only if does not support cancel
1165                          */
1166                         /* this can also happen when cdt calls it to
1167                          * for a timed out request */
1168                         *status = ARS_FAILED;
1169                         /* to have a cancel event in changelog */
1170                         pgs->hpk_errval = ECANCELED;
1171                         break;
1172                 case ECANCELED:
1173                         /* the request record has already been set to
1174                          * ARS_CANCELED, this set the cancel request
1175                          * to ARS_SUCCEED */
1176                         *status = ARS_SUCCEED;
1177                         break;
1178                 default:
1179                         /* retry only if current policy or requested, and
1180                          * object is not on error/removed */
1181                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1182                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1183                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1184                         break;
1185                 }
1186
1187                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1188                         CERROR("%s: Request %#llx on "DFID
1189                                " failed, error code %d too large\n",
1190                                mdt_obd_name(mdt),
1191                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1192                                pgs->hpk_errval);
1193                         hsm_set_cl_error(&cl_flags,
1194                                          CLF_HSM_ERROVERFLOW);
1195                         rc = -EINVAL;
1196                 } else {
1197                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1198                 }
1199
1200                 switch (car->car_hai->hai_action) {
1201                 case HSMA_ARCHIVE:
1202                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1203                         break;
1204                 case HSMA_RESTORE:
1205                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1206                         break;
1207                 case HSMA_REMOVE:
1208                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1209                         break;
1210                 case HSMA_CANCEL:
1211                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1212                         CERROR("%s: Failed request %#llx on "DFID
1213                                " cannot be a CANCEL\n",
1214                                mdt_obd_name(mdt),
1215                                pgs->hpk_cookie,
1216                                PFID(&pgs->hpk_fid));
1217                         break;
1218                 default:
1219                         CERROR("%s: Failed request %#llx on "DFID
1220                                " %d is an unknown action\n",
1221                                mdt_obd_name(mdt),
1222                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1223                                car->car_hai->hai_action);
1224                         rc = -EINVAL;
1225                         break;
1226                 }
1227         } else {
1228                 *status = ARS_SUCCEED;
1229                 switch (car->car_hai->hai_action) {
1230                 case HSMA_ARCHIVE:
1231                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1232                         /* set ARCHIVE keep EXIST and clear LOST and
1233                          * DIRTY */
1234                         mh.mh_arch_ver = pgs->hpk_data_version;
1235                         mh.mh_flags |= HS_ARCHIVED;
1236                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1237                         is_mh_changed = true;
1238                         break;
1239                 case HSMA_RESTORE:
1240                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1241
1242                         /* do not clear RELEASED and DIRTY here
1243                          * this will occur in hsm_swap_layouts()
1244                          */
1245
1246                         /* Restoring has changed the file version on
1247                          * disk. */
1248                         mh.mh_arch_ver = pgs->hpk_data_version;
1249                         is_mh_changed = true;
1250                         break;
1251                 case HSMA_REMOVE:
1252                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1253                         /* clear ARCHIVED EXISTS and LOST */
1254                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1255                         is_mh_changed = true;
1256                         break;
1257                 case HSMA_CANCEL:
1258                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1259                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1260                                mdt_obd_name(mdt),
1261                                pgs->hpk_cookie,
1262                                PFID(&pgs->hpk_fid));
1263                         break;
1264                 default:
1265                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1266                                mdt_obd_name(mdt),
1267                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1268                                car->car_hai->hai_action);
1269                         rc = -EINVAL;
1270                         break;
1271                 }
1272         }
1273
1274         /* rc != 0 means error when analysing action, it may come from
1275          * a crasy CT no need to manage DIRTY
1276          */
1277         if (rc == 0)
1278                 hsm_set_cl_flags(&cl_flags,
1279                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1280
1281         /* unlock is done later, after layout lock management */
1282         if (is_mh_changed && !IS_ERR(obj))
1283                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1284
1285         /* we give back layout lock only if restore was successful or
1286          * if no retry will be attempted and if object is still alive,
1287          * in other cases we just unlock the object */
1288         if (car->car_hai->hai_action == HSMA_RESTORE) {
1289                 struct cdt_restore_handle       *crh;
1290
1291                 /* restore in data FID done, we swap the layouts
1292                  * only if restore is successful */
1293                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1294                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1295                                               &mh);
1296                         if (rc) {
1297                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1298                                         *status = ARS_FAILED;
1299                                 pgs->hpk_errval = -rc;
1300                         }
1301                 }
1302                 /* we have to retry, so keep layout lock */
1303                 if (*status == ARS_WAITING)
1304                         GOTO(out, rc);
1305
1306                 /* restore special case, need to create ChangeLog record
1307                  * before to give back layout lock to avoid concurrent
1308                  * file updater to post out of order ChangeLog */
1309                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1310                              &car->car_hai->hai_fid);
1311                 need_changelog = false;
1312
1313                 /* give back layout lock */
1314                 mutex_lock(&cdt->cdt_restore_lock);
1315                 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1316                 if (crh != NULL)
1317                         list_del(&crh->crh_list);
1318                 mutex_unlock(&cdt->cdt_restore_lock);
1319                 /* Just give back layout lock, we keep the reference
1320                  * which is given back later with the lock for HSM
1321                  * flags.
1322                  * XXX obj may be invalid so we do not pass it. */
1323                 if (crh != NULL)
1324                         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
1325
1326                 if (crh != NULL)
1327                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1328         }
1329
1330         GOTO(out, rc);
1331
1332 out:
1333         /* always add a ChangeLog record */
1334         if (need_changelog)
1335                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1336                              &car->car_hai->hai_fid);
1337
1338         if (!IS_ERR(obj))
1339                 mdt_object_put(mti->mti_env, obj);
1340
1341         RETURN(rc);
1342 }
1343
1344 /**
1345  * update status of a request
1346  * \param mti [IN] context
1347  * \param pgs [IN] progress of the copy tool
1348  * \param update_record [IN] update llog record
1349  * \retval 0 success
1350  * \retval -ve failure
1351  */
1352 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1353                                  struct hsm_progress_kernel *pgs,
1354                                  const int update_record)
1355 {
1356         struct mdt_device       *mdt = mti->mti_mdt;
1357         struct coordinator      *cdt = &mdt->mdt_coordinator;
1358         struct cdt_agent_req    *car;
1359         int                      rc = 0;
1360         ENTRY;
1361
1362         /* no coordinator started, so we cannot serve requests */
1363         if (cdt->cdt_state == CDT_STOPPED)
1364                 RETURN(-EAGAIN);
1365
1366         /* first do sanity checks */
1367         car = mdt_cdt_update_request(cdt, pgs);
1368         if (IS_ERR(car)) {
1369                 CERROR("%s: Cannot find running request for cookie %#llx"
1370                        " on fid="DFID"\n",
1371                        mdt_obd_name(mdt),
1372                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1373
1374                 RETURN(PTR_ERR(car));
1375         }
1376
1377         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1378                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1379                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1380                       hsm_copytool_action2name(car->car_hai->hai_action),
1381                       pgs->hpk_flags, pgs->hpk_errval,
1382                       PFID(&car->car_hai->hai_fid),
1383                       PFID(&car->car_hai->hai_dfid));
1384
1385         /* progress is done on FID or data FID depending of the action and
1386          * of the copy progress */
1387         /* for restore progress is used to send back the data FID to cdt */
1388         if (car->car_hai->hai_action == HSMA_RESTORE &&
1389             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1390                 car->car_hai->hai_dfid = pgs->hpk_fid;
1391
1392         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1393              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1394             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1395              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1396                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1397                        " does not match request FID "DFID" nor data FID "
1398                        DFID"\n",
1399                        mdt_obd_name(mdt),
1400                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1401                        PFID(&car->car_hai->hai_fid),
1402                        PFID(&car->car_hai->hai_dfid));
1403                 GOTO(out, rc = -EINVAL);
1404         }
1405
1406         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1407                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1408                        " is not coherent (err=%d and not completed"
1409                        " (flags=%d))\n",
1410                        mdt_obd_name(mdt),
1411                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1412                        hsm_copytool_action2name(car->car_hai->hai_action),
1413                        pgs->hpk_errval, pgs->hpk_flags);
1414                 GOTO(out, rc = -EINVAL);
1415         }
1416
1417         /* now progress is valid */
1418
1419         /* we use a root like ucred */
1420         hsm_init_ucred(mdt_ucred(mti));
1421
1422         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1423                 enum agent_req_status    status;
1424
1425                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1426
1427                 CDEBUG(D_HSM, "%s record: fid="DFID" cookie=%#llx action=%s "
1428                               "status=%s\n",
1429                        update_record ? "Updating" : "Not updating",
1430                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1431                        hsm_copytool_action2name(car->car_hai->hai_action),
1432                        agent_req_status2name(status));
1433
1434                 /* update record first (LU-9075) */
1435                 if (update_record) {
1436                         int rc1;
1437
1438                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1439                                                      &pgs->hpk_cookie, 1,
1440                                                      status);
1441                         if (rc1)
1442                                 CERROR("%s: mdt_agent_record_update() failed,"
1443                                        " rc=%d, cannot update status to %s"
1444                                        " for cookie %#llx\n",
1445                                        mdt_obd_name(mdt), rc1,
1446                                        agent_req_status2name(status),
1447                                        pgs->hpk_cookie);
1448                         rc = (rc != 0 ? rc : rc1);
1449                 }
1450
1451                 /* then remove request from memory list (LU-9075) */
1452                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1453
1454                 /* ct has completed a request, so a slot is available, wakeup
1455                  * cdt to find new work */
1456                 mdt_hsm_cdt_wakeup(mdt);
1457         } else {
1458                 /* if copytool send a progress on a canceled request
1459                  * we inform copytool it should stop
1460                  */
1461                 if (car->car_canceled == 1)
1462                         rc = -ECANCELED;
1463         }
1464         GOTO(out, rc);
1465
1466 out:
1467         /* remove ref got from mdt_cdt_update_request() */
1468         mdt_cdt_put_request(car);
1469
1470         return rc;
1471 }
1472
1473
1474 /**
1475  * data passed to llog_cat_process() callback
1476  * to cancel requests
1477  */
1478 struct hsm_cancel_all_data {
1479         struct mdt_device       *mdt;
1480 };
1481
1482 /**
1483  *  llog_cat_process() callback, used to:
1484  *  - purge all requests
1485  * \param env [IN] environment
1486  * \param llh [IN] llog handle
1487  * \param hdr [IN] llog record
1488  * \param data [IN] cb data = struct hsm_cancel_all_data
1489  * \retval 0 success
1490  * \retval -ve failure
1491  */
1492 static int mdt_cancel_all_cb(const struct lu_env *env,
1493                              struct llog_handle *llh,
1494                              struct llog_rec_hdr *hdr, void *data)
1495 {
1496         struct llog_agent_req_rec       *larr;
1497         struct hsm_cancel_all_data      *hcad;
1498         int                              rc = 0;
1499         ENTRY;
1500
1501         larr = (struct llog_agent_req_rec *)hdr;
1502         hcad = data;
1503         if (larr->arr_status == ARS_WAITING ||
1504             larr->arr_status == ARS_STARTED) {
1505                 larr->arr_status = ARS_CANCELED;
1506                 larr->arr_req_change = cfs_time_current_sec();
1507                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1508         }
1509
1510         RETURN(rc);
1511 }
1512
1513 /**
1514  * cancel all actions
1515  * \param obd [IN] MDT device
1516  */
1517 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1518 {
1519         struct lu_env                    env;
1520         struct lu_context                session;
1521         struct mdt_thread_info          *mti;
1522         struct coordinator              *cdt = &mdt->mdt_coordinator;
1523         struct cdt_agent_req            *car;
1524         struct hsm_action_list          *hal = NULL;
1525         struct hsm_action_item          *hai;
1526         struct hsm_cancel_all_data       hcad;
1527         int                              hal_sz = 0, hal_len, rc;
1528         enum cdt_states                  save_state;
1529         ENTRY;
1530
1531         rc = lu_env_init(&env, LCT_MD_THREAD);
1532         if (rc < 0)
1533                 RETURN(rc);
1534
1535         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1536         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1537         if (rc < 0)
1538                 GOTO(out_env, rc);
1539
1540         lu_context_enter(&session);
1541         env.le_ses = &session;
1542
1543         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1544         LASSERT(mti != NULL);
1545
1546         mti->mti_env = &env;
1547         mti->mti_mdt = mdt;
1548
1549         hsm_init_ucred(mdt_ucred(mti));
1550
1551         /* disable coordinator */
1552         save_state = cdt->cdt_state;
1553         cdt->cdt_state = CDT_DISABLE;
1554
1555         /* send cancel to all running requests */
1556         down_read(&cdt->cdt_request_lock);
1557         list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1558                 mdt_cdt_get_request(car);
1559                 /* request is not yet removed from list, it will be done
1560                  * when copytool will return progress
1561                  */
1562
1563                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1564                         mdt_cdt_put_request(car);
1565                         continue;
1566                 }
1567
1568                 /* needed size */
1569                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1570                           cfs_size_round(car->car_hai->hai_len);
1571
1572                 if (hal_len > hal_sz && hal_sz > 0) {
1573                         /* not enough room, free old buffer */
1574                         OBD_FREE(hal, hal_sz);
1575                         hal = NULL;
1576                 }
1577
1578                 /* empty buffer, allocate one */
1579                 if (hal == NULL) {
1580                         hal_sz = hal_len;
1581                         OBD_ALLOC(hal, hal_sz);
1582                         if (hal == NULL) {
1583                                 mdt_cdt_put_request(car);
1584                                 up_read(&cdt->cdt_request_lock);
1585                                 GOTO(out_cdt_state, rc = -ENOMEM);
1586                         }
1587                 }
1588
1589                 hal->hal_version = HAL_VERSION;
1590                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1591                                 MTI_NAME_MAXLEN);
1592                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1593                 hal->hal_compound_id = car->car_compound_id;
1594                 hal->hal_archive_id = car->car_archive_id;
1595                 hal->hal_flags = car->car_flags;
1596                 hal->hal_count = 0;
1597
1598                 hai = hai_first(hal);
1599                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1600                 hai->hai_action = HSMA_CANCEL;
1601                 hal->hal_count = 1;
1602
1603                 /* it is possible to safely call mdt_hsm_agent_send()
1604                  * (ie without a deadlock on cdt_request_lock), because the
1605                  * write lock is taken only if we are not in purge mode
1606                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1607                  *   nor mdt_cdt_remove_request())
1608                  */
1609                 /* no conflict with cdt thread because cdt is disable and we
1610                  * have the request lock */
1611                 mdt_hsm_agent_send(mti, hal, 1);
1612
1613                 mdt_cdt_put_request(car);
1614         }
1615         up_read(&cdt->cdt_request_lock);
1616
1617         if (hal != NULL)
1618                 OBD_FREE(hal, hal_sz);
1619
1620         /* cancel all on-disk records */
1621         hcad.mdt = mdt;
1622
1623         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1624                               mdt_cancel_all_cb, &hcad);
1625 out_cdt_state:
1626         /* enable coordinator */
1627         cdt->cdt_state = save_state;
1628         lu_context_exit(&session);
1629         lu_context_fini(&session);
1630 out_env:
1631         lu_env_fini(&env);
1632
1633         RETURN(rc);
1634 }
1635
1636 /**
1637  * check if a request is compatible with file status
1638  * \param hai [IN] request description
1639  * \param hal_an [IN] request archive number (not used)
1640  * \param rq_flags [IN] request flags
1641  * \param hsm [IN] file HSM metadata
1642  * \retval boolean
1643  */
1644 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1645                               const int hal_an, const __u64 rq_flags,
1646                               const struct md_hsm *hsm)
1647 {
1648         int      is_compat = false;
1649         int      hsm_flags;
1650         ENTRY;
1651
1652         hsm_flags = hsm->mh_flags;
1653         switch (hai->hai_action) {
1654         case HSMA_ARCHIVE:
1655                 if (!(hsm_flags & HS_NOARCHIVE) &&
1656                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1657                         is_compat = true;
1658                 break;
1659         case HSMA_RESTORE:
1660                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1661                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1662                         is_compat = true;
1663                 break;
1664         case HSMA_REMOVE:
1665                 if (!(hsm_flags & HS_RELEASED) &&
1666                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1667                         is_compat = true;
1668                 break;
1669         case HSMA_CANCEL:
1670                 is_compat = true;
1671                 break;
1672         }
1673         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1674                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1675                       PFID(&hai->hai_fid),
1676                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1677                       hai->hai_extent.offset, hai->hai_extent.length,
1678                       hsm->mh_flags,
1679                       (is_compat ? "compatible" : "uncompatible"));
1680
1681         RETURN(is_compat);
1682 }
1683
1684 /*
1685  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1686  */
1687 static const struct {
1688         __u64            bit;
1689         char            *name;
1690         char            *nickname;
1691 } hsm_policy_names[] = {
1692         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1693         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1694         { 0 },
1695 };
1696
1697 /**
1698  * convert a policy name to a bit
1699  * \param name [IN] policy name
1700  * \retval 0 unknown
1701  * \retval   policy bit
1702  */
1703 static __u64 hsm_policy_str2bit(const char *name)
1704 {
1705         int      i;
1706
1707         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1708                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1709                     strcmp(hsm_policy_names[i].name, name) == 0)
1710                         return hsm_policy_names[i].bit;
1711         return 0;
1712 }
1713
1714 /**
1715  * convert a policy bit field to a string
1716  * \param mask [IN] policy bit field
1717  * \param hexa [IN] print mask before bit names
1718  * \param buffer [OUT] string
1719  * \param count [IN] size of buffer
1720  */
1721 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1722                                 const bool hexa)
1723 {
1724         int      i, j;
1725         __u64    bit;
1726         ENTRY;
1727
1728         if (hexa)
1729                 seq_printf(m, "(%#llx) ", mask);
1730
1731         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1732                 bit = (1ULL << i);
1733
1734                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1735                         if (hsm_policy_names[j].bit == bit)
1736                                 break;
1737                 }
1738                 if (bit & mask)
1739                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1740                 else
1741                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1742         }
1743         /* remove last ' ' */
1744         m->count--;
1745         seq_putc(m, '\n');
1746 }
1747
1748 /* methods to read/write HSM policy flags */
1749 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1750 {
1751         struct mdt_device       *mdt = m->private;
1752         struct coordinator      *cdt = &mdt->mdt_coordinator;
1753         ENTRY;
1754
1755         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1756         RETURN(0);
1757 }
1758
1759 static ssize_t
1760 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1761                          size_t count, loff_t *off)
1762 {
1763         struct seq_file         *m = file->private_data;
1764         struct mdt_device       *mdt = m->private;
1765         struct coordinator      *cdt = &mdt->mdt_coordinator;
1766         char                    *start, *token, sign;
1767         char                    *buf;
1768         __u64                    policy;
1769         __u64                    add_mask, remove_mask, set_mask;
1770         int                      rc;
1771         ENTRY;
1772
1773         if (count + 1 > PAGE_SIZE)
1774                 RETURN(-EINVAL);
1775
1776         OBD_ALLOC(buf, count + 1);
1777         if (buf == NULL)
1778                 RETURN(-ENOMEM);
1779
1780         if (copy_from_user(buf, buffer, count))
1781                 GOTO(out, rc = -EFAULT);
1782
1783         buf[count] = '\0';
1784
1785         start = buf;
1786         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1787                start);
1788
1789         add_mask = remove_mask = set_mask = 0;
1790         do {
1791                 token = strsep(&start, "\n ");
1792                 sign = *token;
1793
1794                 if (sign == '\0')
1795                         continue;
1796
1797                 if (sign == '-' || sign == '+')
1798                         token++;
1799
1800                 policy = hsm_policy_str2bit(token);
1801                 if (policy == 0) {
1802                         CWARN("%s: '%s' is unknown, "
1803                               "supported policies are:\n", mdt_obd_name(mdt),
1804                               token);
1805                         hsm_policy_bit2str(m, 0, false);
1806                         GOTO(out, rc = -EINVAL);
1807                 }
1808                 switch (sign) {
1809                 case '-':
1810                         remove_mask |= policy;
1811                         break;
1812                 case '+':
1813                         add_mask |= policy;
1814                         break;
1815                 default:
1816                         set_mask |= policy;
1817                         break;
1818                 }
1819
1820         } while (start != NULL);
1821
1822         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
1823                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1824
1825         /* if no sign in all string, it is a clear and set
1826          * if some sign found, all unsigned are converted
1827          * to add
1828          * P1 P2 = set to P1 and P2
1829          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1830          */
1831         if (remove_mask == 0 && add_mask == 0) {
1832                 cdt->cdt_policy = set_mask;
1833         } else {
1834                 cdt->cdt_policy |= set_mask | add_mask;
1835                 cdt->cdt_policy &= ~remove_mask;
1836         }
1837
1838         GOTO(out, rc = count);
1839
1840 out:
1841         OBD_FREE(buf, count + 1);
1842         RETURN(rc);
1843 }
1844 LPROC_SEQ_FOPS(mdt_hsm_policy);
1845
1846 #define GENERATE_PROC_METHOD(VAR)                                       \
1847 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
1848 {                                                                       \
1849         struct mdt_device       *mdt = m->private;                      \
1850         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1851         ENTRY;                                                          \
1852                                                                         \
1853         seq_printf(m, "%llu\n", (__u64)cdt->VAR);                       \
1854         RETURN(0);                                                      \
1855 }                                                                       \
1856 static ssize_t                                                          \
1857 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1858                           size_t count, loff_t *off)                    \
1859                                                                         \
1860 {                                                                       \
1861         struct seq_file         *m = file->private_data;                \
1862         struct mdt_device       *mdt = m->private;                      \
1863         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1864         __s64                    val;                                   \
1865         int                      rc;                                    \
1866         ENTRY;                                                          \
1867                                                                         \
1868         rc = lprocfs_str_to_s64(buffer, count, &val);                   \
1869         if (rc)                                                         \
1870                 RETURN(rc);                                             \
1871         if (val > 0 && val < INT_MAX) {                                 \
1872                 cdt->VAR = val;                                         \
1873                 RETURN(count);                                          \
1874         }                                                               \
1875         RETURN(-EINVAL);                                                \
1876 }                                                                       \
1877
1878 GENERATE_PROC_METHOD(cdt_loop_period)
1879 GENERATE_PROC_METHOD(cdt_grace_delay)
1880 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1881 GENERATE_PROC_METHOD(cdt_max_requests)
1882 GENERATE_PROC_METHOD(cdt_default_archive_id)
1883
1884 /*
1885  * procfs write method for MDT/hsm_control
1886  * proc entry is in mdt directory so data is mdt obd_device pointer
1887  */
1888 #define CDT_ENABLE_CMD   "enabled"
1889 #define CDT_STOP_CMD     "shutdown"
1890 #define CDT_DISABLE_CMD  "disabled"
1891 #define CDT_PURGE_CMD    "purge"
1892 #define CDT_HELP_CMD     "help"
1893 #define CDT_MAX_CMD_LEN  10
1894
1895 ssize_t
1896 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1897                               size_t count, loff_t *off)
1898 {
1899         struct seq_file         *m = file->private_data;
1900         struct obd_device       *obd = m->private;
1901         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
1902         struct coordinator      *cdt = &(mdt->mdt_coordinator);
1903         int                      rc, usage = 0;
1904         char                     kernbuf[CDT_MAX_CMD_LEN];
1905         ENTRY;
1906
1907         if (count == 0 || count >= sizeof(kernbuf))
1908                 RETURN(-EINVAL);
1909
1910         if (copy_from_user(kernbuf, buffer, count))
1911                 RETURN(-EFAULT);
1912         kernbuf[count] = 0;
1913
1914         if (kernbuf[count - 1] == '\n')
1915                 kernbuf[count - 1] = 0;
1916
1917         rc = 0;
1918         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1919                 if (cdt->cdt_state == CDT_DISABLE) {
1920                         cdt->cdt_state = CDT_RUNNING;
1921                         mdt_hsm_cdt_wakeup(mdt);
1922                 } else {
1923                         rc = mdt_hsm_cdt_start(mdt);
1924                 }
1925         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1926                 if ((cdt->cdt_state == CDT_STOPPING) ||
1927                     (cdt->cdt_state == CDT_STOPPED)) {
1928                         CERROR("%s: Coordinator already stopped\n",
1929                                mdt_obd_name(mdt));
1930                         rc = -EALREADY;
1931                 } else {
1932                         cdt->cdt_state = CDT_STOPPING;
1933                         mdt_hsm_cdt_wakeup(mdt);
1934                 }
1935         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
1936                 if ((cdt->cdt_state == CDT_STOPPING) ||
1937                     (cdt->cdt_state == CDT_STOPPED)) {
1938                         CERROR("%s: Coordinator is stopped\n",
1939                                mdt_obd_name(mdt));
1940                         rc = -EINVAL;
1941                 } else {
1942                         cdt->cdt_state = CDT_DISABLE;
1943                 }
1944         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
1945                 rc = hsm_cancel_all_actions(mdt);
1946         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
1947                 usage = 1;
1948         } else {
1949                 usage = 1;
1950                 rc = -EINVAL;
1951         }
1952
1953         if (usage == 1)
1954                 CERROR("%s: Valid coordinator control commands are: "
1955                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
1956                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1957                        CDT_PURGE_CMD, CDT_HELP_CMD);
1958
1959         if (rc)
1960                 RETURN(rc);
1961
1962         RETURN(count);
1963 }
1964
1965 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
1966 {
1967         struct obd_device       *obd = m->private;
1968         struct coordinator      *cdt;
1969         ENTRY;
1970
1971         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
1972
1973         if (cdt->cdt_state == CDT_INIT)
1974                 seq_printf(m, "init\n");
1975         else if (cdt->cdt_state == CDT_RUNNING)
1976                 seq_printf(m, "enabled\n");
1977         else if (cdt->cdt_state == CDT_STOPPING)
1978                 seq_printf(m, "stopping\n");
1979         else if (cdt->cdt_state == CDT_STOPPED)
1980                 seq_printf(m, "stopped\n");
1981         else if (cdt->cdt_state == CDT_DISABLE)
1982                 seq_printf(m, "disabled\n");
1983         else
1984                 seq_printf(m, "unknown\n");
1985
1986         RETURN(0);
1987 }
1988
1989 static int
1990 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
1991 {
1992         bool first = true;
1993         int i;
1994         ENTRY;
1995
1996         for (i = 0; i < 8 * sizeof(mask); i++) {
1997                 if (mask & (1UL << i)) {
1998                         seq_printf(m, "%s%s", first ? "" : " ",
1999                                    hsm_copytool_action2name(i));
2000                         first = false;
2001                 }
2002         }
2003         seq_putc(m, '\n');
2004
2005         RETURN(0);
2006 }
2007
2008 static int
2009 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2010 {
2011         struct mdt_device *mdt = m->private;
2012         struct coordinator *cdt = &mdt->mdt_coordinator;
2013
2014         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2015 }
2016
2017 static int
2018 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2019 {
2020         struct mdt_device *mdt = m->private;
2021         struct coordinator *cdt = &mdt->mdt_coordinator;
2022
2023         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2024 }
2025
2026 static int
2027 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2028 {
2029         struct mdt_device *mdt = m->private;
2030         struct coordinator *cdt = &mdt->mdt_coordinator;
2031
2032         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2033 }
2034
2035 static inline enum hsm_copytool_action
2036 hsm_copytool_name2action(const char *name)
2037 {
2038         if (strcasecmp(name, "NOOP") == 0)
2039                 return HSMA_NONE;
2040         else if (strcasecmp(name, "ARCHIVE") == 0)
2041                 return HSMA_ARCHIVE;
2042         else if (strcasecmp(name, "RESTORE") == 0)
2043                 return HSMA_RESTORE;
2044         else if (strcasecmp(name, "REMOVE") == 0)
2045                 return HSMA_REMOVE;
2046         else if (strcasecmp(name, "CANCEL") == 0)
2047                 return HSMA_CANCEL;
2048         else
2049                 return -1;
2050 }
2051
2052 static ssize_t
2053 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2054                             size_t user_count, __u64 *mask)
2055 {
2056         char *buf, *pos, *name;
2057         size_t buf_size;
2058         __u64 new_mask = 0;
2059         int rc;
2060         ENTRY;
2061
2062         if (!(user_count < 4096))
2063                 RETURN(-ENOMEM);
2064
2065         buf_size = user_count + 1;
2066
2067         OBD_ALLOC(buf, buf_size);
2068         if (buf == NULL)
2069                 RETURN(-ENOMEM);
2070
2071         if (copy_from_user(buf, user_buf, buf_size - 1))
2072                 GOTO(out, rc = -EFAULT);
2073
2074         buf[buf_size - 1] = '\0';
2075
2076         pos = buf;
2077         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2078                 int action;
2079
2080                 if (*name == '\0')
2081                         continue;
2082
2083                 action = hsm_copytool_name2action(name);
2084                 if (action < 0)
2085                         GOTO(out, rc = -EINVAL);
2086
2087                 new_mask |= (1UL << action);
2088         }
2089
2090         *mask = new_mask;
2091         rc = user_count;
2092 out:
2093         OBD_FREE(buf, buf_size);
2094
2095         RETURN(rc);
2096 }
2097
2098 static ssize_t
2099 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2100                                         size_t count, loff_t *off)
2101 {
2102         struct seq_file         *m = file->private_data;
2103         struct mdt_device       *mdt = m->private;
2104         struct coordinator *cdt = &mdt->mdt_coordinator;
2105
2106         return mdt_write_hsm_request_mask(file, buf, count,
2107                                            &cdt->cdt_user_request_mask);
2108 }
2109
2110 static ssize_t
2111 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2112                                         size_t count, loff_t *off)
2113 {
2114         struct seq_file         *m = file->private_data;
2115         struct mdt_device       *mdt = m->private;
2116         struct coordinator      *cdt = &mdt->mdt_coordinator;
2117
2118         return mdt_write_hsm_request_mask(file, buf, count,
2119                                            &cdt->cdt_group_request_mask);
2120 }
2121
2122 static ssize_t
2123 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2124                                         size_t count, loff_t *off)
2125 {
2126         struct seq_file         *m = file->private_data;
2127         struct mdt_device       *mdt = m->private;
2128         struct coordinator      *cdt = &mdt->mdt_coordinator;
2129
2130         return mdt_write_hsm_request_mask(file, buf, count,
2131                                            &cdt->cdt_other_request_mask);
2132 }
2133
2134 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2135 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2136 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2137 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2138 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2139 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2140 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2141 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2142
2143 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2144         { .name =       "agents",
2145           .fops =       &mdt_hsm_agent_fops                     },
2146         { .name =       "actions",
2147           .fops =       &mdt_hsm_actions_fops,
2148           .proc_mode =  0444                                    },
2149         { .name =       "default_archive_id",
2150           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2151         { .name =       "grace_delay",
2152           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2153         { .name =       "loop_period",
2154           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2155         { .name =       "max_requests",
2156           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2157         { .name =       "policy",
2158           .fops =       &mdt_hsm_policy_fops                    },
2159         { .name =       "active_request_timeout",
2160           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2161         { .name =       "active_requests",
2162           .fops =       &mdt_hsm_active_requests_fops           },
2163         { .name =       "user_request_mask",
2164           .fops =       &mdt_hsm_user_request_mask_fops,        },
2165         { .name =       "group_request_mask",
2166           .fops =       &mdt_hsm_group_request_mask_fops,       },
2167         { .name =       "other_request_mask",
2168           .fops =       &mdt_hsm_other_request_mask_fops,       },
2169         { 0 }
2170 };