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