lustre/mdt/mdt_coordinator.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
  24  *                          alternatives
  25  *
  26  * Copyright (c) 2013, 2016, Intel Corporation.
  27  * Use is subject to license terms.
  28  */
  29 /*
  30  * lustre/mdt/mdt_coordinator.c
  31  *
  32  * Lustre HSM Coordinator
  33  *
  34  * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
  35  * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
  36  * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <linux/kthread.h>
  42 #include <obd_support.h>
  43 #include <lustre_export.h>
  44 #include <obd.h>
  45 #include <lprocfs_status.h>
  46 #include <lustre_log.h>
  47 #include <lustre_kernelcomm.h>
  48 #include "mdt_internal.h"
  49
  50 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
  51
  52 /**
  53  * get obj and HSM attributes on a fid
  54  * \param mti [IN] context
  55  * \param fid [IN] object fid
  56  * \param hsm [OUT] HSM meta data
  57  * \retval obj or error (-ENOENT if not found)
  58  */
  59 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  60                                       const struct lu_fid *fid,
  61                                       struct md_hsm *hsm)
  62 {
  63         struct md_attr          *ma;
  64         struct mdt_object       *obj;
  65         int                      rc;
  66         ENTRY;
  67
  68         ma = &mti->mti_attr;
  69         ma->ma_need = MA_HSM;
  70         ma->ma_valid = 0;
  71
  72         /* find object by FID */
  73         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  74         if (IS_ERR(obj))
  75                 RETURN(obj);
  76
  77         if (!mdt_object_exists(obj)) {
  78                 /* no more object */
  79                 mdt_object_put(mti->mti_env, obj);
  80                 RETURN(ERR_PTR(-ENOENT));
  81         }
  82
  83         rc = mdt_attr_get_complex(mti, obj, ma);
  84         if (rc) {
  85                 mdt_object_put(mti->mti_env, obj);
  86                 RETURN(ERR_PTR(rc));
  87         }
  88
  89         if (ma->ma_valid & MA_HSM)
  90                 *hsm = ma->ma_hsm;
  91         else
  92                 memset(hsm, 0, sizeof(*hsm));
  93         ma->ma_valid = 0;
  94         RETURN(obj);
  95 }
  96
  97 void mdt_hsm_dump_hal(int level, const char *prefix,
  98                       struct hsm_action_list *hal)
  99 {
 100         int                      i, sz;
 101         struct hsm_action_item  *hai;
 102         char                     buf[12];
 103
 104         CDEBUG(level, "%s: HAL header: version %X count %d compound %#llx"
 105                       " archive_id %d flags %#llx\n",
 106                prefix, hal->hal_version, hal->hal_count,
 107                hal->hal_compound_id, hal->hal_archive_id, hal->hal_flags);
 108
 109         hai = hai_first(hal);
 110         for (i = 0; i < hal->hal_count; i++) {
 111                 sz = hai->hai_len - sizeof(*hai);
 112                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 113                        " compound/cookie=%#llx/%#llx"
 114                        " action=%s extent=%#llx-%#llx gid=%#llx"
 115                        " datalen=%d data=[%s]\n",
 116                        prefix, i,
 117                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 118                        hal->hal_compound_id, hai->hai_cookie,
 119                        hsm_copytool_action2name(hai->hai_action),
 120                        hai->hai_extent.offset,
 121                        hai->hai_extent.length,
 122                        hai->hai_gid, sz,
 123                        hai_dump_data_field(hai, buf, sizeof(buf)));
 124                 hai = hai_next(hai);
 125         }
 126 }
 127
 128 /**
 129  * data passed to llog_cat_process() callback
 130  * to scan requests and take actions
 131  */
 132 struct hsm_scan_request {
 133         int                      hal_sz;
 134         int                      hal_used_sz;
 135         struct hsm_action_list  *hal;
 136 };
 137
 138 struct hsm_scan_data {
 139         struct mdt_thread_info          *mti;
 140         char                             fs_name[MTI_NAME_MAXLEN+1];
 141         /* are we scanning the logs for housekeeping, or just looking
 142          * for new work?
 143          */
 144         bool                             housekeeping;
 145         /* request to be send to agents */
 146         int                              max_requests;  /** vector size */
 147         int                              request_cnt;   /** used count */
 148         struct hsm_scan_request         *request;
 149 };
 150
 151 struct hsm_thread_data {
 152         struct mdt_thread_info  *cdt_mti;
 153         struct hsm_scan_request *request;
 154 };
 155 /**
 156  *  llog_cat_process() callback, used to:
 157  *  - find waiting request and start action
 158  *  - purge canceled and done requests
 159  * \param env [IN] environment
 160  * \param llh [IN] llog handle
 161  * \param hdr [IN] llog record
 162  * \param data [IN/OUT] cb data = struct hsm_scan_data
 163  * \retval 0 success
 164  * \retval -ve failure
 165  */
 166 static int mdt_coordinator_cb(const struct lu_env *env,
 167                               struct llog_handle *llh,
 168                               struct llog_rec_hdr *hdr,
 169                               void *data)
 170 {
 171         struct llog_agent_req_rec       *larr;
 172         struct hsm_scan_data            *hsd;
 173         struct hsm_action_item          *hai;
 174         struct mdt_device               *mdt;
 175         struct coordinator              *cdt;
 176         int                              rc;
 177         ENTRY;
 178
 179         hsd = data;
 180         mdt = hsd->mti->mti_mdt;
 181         cdt = &mdt->mdt_coordinator;
 182
 183         larr = (struct llog_agent_req_rec *)hdr;
 184         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 185         switch (larr->arr_status) {
 186         case ARS_WAITING: {
 187                 int i;
 188                 struct hsm_scan_request *request;
 189
 190                 /* Are agents full? */
 191                 if (atomic_read(&cdt->cdt_request_count) >=
 192                     cdt->cdt_max_requests)
 193                         break;
 194
 195                 /* first search whether the request is found in the
 196                  * list we have built. */
 197                 request = NULL;
 198                 for (i = 0; i < hsd->request_cnt; i++) {
 199                         if (hsd->request[i].hal->hal_compound_id ==
 200                             larr->arr_compound_id) {
 201                                 request = &hsd->request[i];
 202                                 break;
 203                         }
 204                 }
 205
 206                 if (!request) {
 207                         struct hsm_action_list *hal;
 208
 209                         if (hsd->request_cnt == hsd->max_requests) {
 210                                 if (!hsd->housekeeping) {
 211                                         /* The request array is full,
 212                                          * stop here. There might be
 213                                          * more known requests that
 214                                          * could be merged, but this
 215                                          * avoid analyzing too many
 216                                          * llogs for minor gains.
 217                                          */
 218                                         RETURN(LLOG_PROC_BREAK);
 219                                 } else {
 220                                         /* Unknown request and no more room
 221                                          * for a new request. Continue to scan
 222                                          * to find other entries for already
 223                                          * existing requests.
 224                                          */
 225                                         RETURN(0);
 226                                 }
 227                         }
 228
 229                         request = &hsd->request[hsd->request_cnt];
 230
 231                         /* allocates hai vector size just needs to be large
 232                          * enough */
 233                         request->hal_sz =
 234                                 sizeof(*request->hal) +
 235                                 cfs_size_round(MTI_NAME_MAXLEN+1) +
 236                                 2 * cfs_size_round(larr->arr_hai.hai_len);
 237                         OBD_ALLOC(hal, request->hal_sz);
 238                         if (!hal)
 239                                 RETURN(-ENOMEM);
 240                         hal->hal_version = HAL_VERSION;
 241                         strlcpy(hal->hal_fsname, hsd->fs_name,
 242                                 MTI_NAME_MAXLEN + 1);
 243                         hal->hal_compound_id = larr->arr_compound_id;
 244                         hal->hal_archive_id = larr->arr_archive_id;
 245                         hal->hal_flags = larr->arr_flags;
 246                         hal->hal_count = 0;
 247                         request->hal_used_sz = hal_size(hal);
 248                         request->hal = hal;
 249                         hsd->request_cnt++;
 250                         hai = hai_first(hal);
 251                 } else {
 252                         /* request is known */
 253                         /* we check if record archive num is the same as the
 254                          * known request, if not we will serve it in multiple
 255                          * time because we do not know if the agent can serve
 256                          * multiple backend
 257                          * a use case is a compound made of multiple restore
 258                          * where the files are not archived in the same backend
 259                          */
 260                         if (larr->arr_archive_id !=
 261                             request->hal->hal_archive_id)
 262                                 RETURN(0);
 263
 264                         if (request->hal_sz <
 265                             request->hal_used_sz +
 266                             cfs_size_round(larr->arr_hai.hai_len)) {
 267                                 /* Not enough room, need an extension */
 268                                 void *hal_buffer;
 269                                 int sz;
 270
 271                                 sz = 2 * request->hal_sz;
 272                                 OBD_ALLOC(hal_buffer, sz);
 273                                 if (!hal_buffer)
 274                                         RETURN(-ENOMEM);
 275                                 memcpy(hal_buffer, request->hal,
 276                                        request->hal_used_sz);
 277                                 OBD_FREE(request->hal,
 278                                          request->hal_sz);
 279                                 request->hal = hal_buffer;
 280                                 request->hal_sz = sz;
 281                         }
 282                         hai = hai_first(request->hal);
 283                         for (i = 0; i < request->hal->hal_count; i++)
 284                                 hai = hai_next(hai);
 285                 }
 286                 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 287                 hai->hai_cookie = larr->arr_hai.hai_cookie;
 288                 hai->hai_gid = larr->arr_hai.hai_gid;
 289
 290                 request->hal_used_sz += cfs_size_round(hai->hai_len);
 291                 request->hal->hal_count++;
 292
 293                 if (hai->hai_action != HSMA_CANCEL)
 294                         cdt_agent_record_hash_add(cdt, hai->hai_cookie,
 295                                                   llh->lgh_hdr->llh_cat_idx,
 296                                                   hdr->lrh_index);
 297                 break;
 298         }
 299         case ARS_STARTED: {
 300                 struct hsm_progress_kernel pgs;
 301                 struct cdt_agent_req *car;
 302                 time64_t now = ktime_get_real_seconds();
 303                 time64_t last;
 304
 305                 if (!hsd->housekeeping)
 306                         break;
 307
 308                 /* we search for a running request
 309                  * error may happen if coordinator crashes or stopped
 310                  * with running request
 311                  */
 312                 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie);
 313                 if (car == NULL) {
 314                         last = larr->arr_req_change;
 315                 } else {
 316                         last = car->car_req_update;
 317                         mdt_cdt_put_request(car);
 318                 }
 319
 320                 /* test if request too long, if yes cancel it
 321                  * the same way the copy tool acknowledge a cancel request */
 322                 if (now <= last + cdt->cdt_active_req_timeout)
 323                         RETURN(0);
 324
 325                 dump_llog_agent_req_rec("request timed out, start cleaning",
 326                                         larr);
 327                 /* a too old cancel request just needs to be removed
 328                  * this can happen, if copy tool does not support
 329                  * cancel for other requests, we have to remove the
 330                  * running request and notify the copytool */
 331                 pgs.hpk_fid = larr->arr_hai.hai_fid;
 332                 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
 333                 pgs.hpk_extent = larr->arr_hai.hai_extent;
 334                 pgs.hpk_flags = HP_FLAG_COMPLETED;
 335                 pgs.hpk_errval = ENOSYS;
 336                 pgs.hpk_data_version = 0;
 337
 338                 /* update request state, but do not record in llog, to
 339                  * avoid deadlock on cdt_llog_lock */
 340                 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
 341                 if (rc)
 342                         CERROR("%s: cannot cleanup timed out request: "
 343                                DFID" for cookie %#llx action=%s\n",
 344                                mdt_obd_name(mdt),
 345                                PFID(&pgs.hpk_fid), pgs.hpk_cookie,
 346                                hsm_copytool_action2name(
 347                                        larr->arr_hai.hai_action));
 348
 349                 if (rc == -ENOENT) {
 350                         /* The request no longer exists, forget
 351                          * about it, and do not send a cancel request
 352                          * to the client, for which an error will be
 353                          * sent back, leading to an endless cycle of
 354                          * cancellation. */
 355                         cdt_agent_record_hash_del(cdt,
 356                                                   larr->arr_hai.hai_cookie);
 357                         RETURN(LLOG_DEL_RECORD);
 358                 }
 359
 360                 /* XXX A cancel request cannot be cancelled. */
 361                 if (larr->arr_hai.hai_action == HSMA_CANCEL)
 362                         RETURN(0);
 363
 364                 larr->arr_status = ARS_CANCELED;
 365                 larr->arr_req_change = now;
 366                 rc = llog_write(hsd->mti->mti_env, llh, hdr, hdr->lrh_index);
 367                 if (rc < 0)
 368                         CERROR("%s: cannot update agent log: rc = %d\n",
 369                                mdt_obd_name(mdt), rc);
 370                 break;
 371         }
 372         case ARS_FAILED:
 373         case ARS_CANCELED:
 374         case ARS_SUCCEED:
 375                 if (!hsd->housekeeping)
 376                         break;
 377
 378                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 379                     ktime_get_real_seconds()) {
 380                         cdt_agent_record_hash_del(cdt,
 381                                                   larr->arr_hai.hai_cookie);
 382                         RETURN(LLOG_DEL_RECORD);
 383                 }
 384                 break;
 385         }
 386         RETURN(0);
 387 }
 388
 389 /**
 390  * create /proc entries for coordinator
 391  * \param mdt [IN]
 392  * \retval 0 success
 393  * \retval -ve failure
 394  */
 395 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 396 {
 397         struct coordinator      *cdt = &mdt->mdt_coordinator;
 398         int                      rc = 0;
 399         ENTRY;
 400
 401         /* init /proc entries, failure is not critical */
 402         cdt->cdt_proc_dir = lprocfs_register("hsm",
 403                                              mdt2obd_dev(mdt)->obd_proc_entry,
 404                                              lprocfs_mdt_hsm_vars, mdt);
 405         if (IS_ERR(cdt->cdt_proc_dir)) {
 406                 rc = PTR_ERR(cdt->cdt_proc_dir);
 407                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 408                        " rc=%d\n", mdt_obd_name(mdt), rc);
 409                 cdt->cdt_proc_dir = NULL;
 410                 RETURN(rc);
 411         }
 412
 413         RETURN(0);
 414 }
 415
 416 /**
 417  * remove /proc entries for coordinator
 418  * \param mdt [IN]
 419  */
 420 void hsm_cdt_procfs_fini(struct mdt_device *mdt)
 421 {
 422         struct coordinator *cdt = &mdt->mdt_coordinator;
 423
 424         if (cdt->cdt_proc_dir != NULL)
 425                 lprocfs_remove(&cdt->cdt_proc_dir);
 426 }
 427
 428 /**
 429  * get vector of hsm cdt /proc vars
 430  * \param none
 431  * \retval var vector
 432  */
 433 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 434 {
 435         return lprocfs_mdt_hsm_vars;
 436 }
 437
 438 /* Release the ressource used by the coordinator. Called when the
 439  * coordinator is stopping. */
 440 static void mdt_hsm_cdt_cleanup(struct mdt_device *mdt)
 441 {
 442         struct coordinator              *cdt = &mdt->mdt_coordinator;
 443         struct cdt_agent_req            *car, *tmp1;
 444         struct hsm_agent                *ha, *tmp2;
 445         struct cdt_restore_handle       *crh, *tmp3;
 446         struct mdt_thread_info          *cdt_mti;
 447
 448         /* start cleaning */
 449         down_write(&cdt->cdt_request_lock);
 450         list_for_each_entry_safe(car, tmp1, &cdt->cdt_request_list,
 451                                  car_request_list) {
 452                 cfs_hash_del(cdt->cdt_request_cookie_hash,
 453                              &car->car_hai->hai_cookie,
 454                              &car->car_cookie_hash);
 455                 list_del(&car->car_request_list);
 456                 mdt_cdt_put_request(car);
 457         }
 458         up_write(&cdt->cdt_request_lock);
 459
 460         down_write(&cdt->cdt_agent_lock);
 461         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 462                 list_del(&ha->ha_list);
 463                 OBD_FREE_PTR(ha);
 464         }
 465         up_write(&cdt->cdt_agent_lock);
 466
 467         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 468         mutex_lock(&cdt->cdt_restore_lock);
 469         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
 470                 list_del(&crh->crh_list);
 471                 /* give back layout lock */
 472                 mdt_object_unlock(cdt_mti, NULL, &crh->crh_lh, 1);
 473                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 474         }
 475         mutex_unlock(&cdt->cdt_restore_lock);
 476 }
 477
 478 /*
 479  * Coordinator state transition table, indexed on enum cdt_states, taking
 480  * from and to states. For instance since CDT_INIT to CDT_RUNNING is a
 481  * valid transition, cdt_transition[CDT_INIT][CDT_RUNNING] is true.
 482  */
 483 static bool cdt_transition[CDT_STATES_COUNT][CDT_STATES_COUNT] = {
 484         /* from -> to:    stopped init   running disable stopping */
 485         /* stopped */   { true,   true,  false,  false,  false },
 486         /* init */      { true,   false, true,   false,  false },
 487         /* running */   { false,  false, true,   true,   true },
 488         /* disable */   { false,  false, true,   true,   true },
 489         /* stopping */  { true,   false, false,  false,  false }
 490 };
 491
 492 /**
 493  * Change coordinator thread state
 494  * Some combinations are not valid, so catch them here.
 495  *
 496  * Returns 0 on success, with old_state set if not NULL, or -EINVAL if
 497  * the transition was not possible.
 498  */
 499 static int set_cdt_state(struct coordinator *cdt, enum cdt_states new_state,
 500                          enum cdt_states *old_state)
 501 {
 502         int rc;
 503         enum cdt_states state;
 504
 505         spin_lock(&cdt->cdt_state_lock);
 506
 507         state = cdt->cdt_state;
 508
 509         if (cdt_transition[state][new_state]) {
 510                 cdt->cdt_state = new_state;
 511                 spin_unlock(&cdt->cdt_state_lock);
 512                 if (old_state)
 513                         *old_state = state;
 514                 rc = 0;
 515         } else {
 516                 spin_unlock(&cdt->cdt_state_lock);
 517                 CDEBUG(D_HSM,
 518                        "unexpected coordinator transition, from=%s, to=%s\n",
 519                        cdt_mdt_state2str(state), cdt_mdt_state2str(new_state));
 520                 rc = -EINVAL;
 521         }
 522
 523         return rc;
 524 }
 525
 526 /**
 527  * coordinator thread
 528  * \param data [IN] obd device
 529  * \retval 0 success
 530  * \retval -ve failure
 531  */
 532 static int mdt_coordinator(void *data)
 533 {
 534         struct hsm_thread_data  *thread_data = data;
 535         struct mdt_thread_info  *mti = thread_data->cdt_mti;
 536         struct mdt_device       *mdt = mti->mti_mdt;
 537         struct coordinator      *cdt = &mdt->mdt_coordinator;
 538         struct hsm_scan_data     hsd = { NULL };
 539         time64_t                 last_housekeeping = 0;
 540         int                      rc = 0;
 541         int                      request_sz;
 542         ENTRY;
 543
 544         /* set up hsd->request and max_requests */
 545         hsd.max_requests = cdt->cdt_max_requests;
 546         request_sz = hsd.max_requests * sizeof(*hsd.request);
 547         hsd.request = thread_data->request;
 548
 549         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 550                mdt_obd_name(mdt), current_pid());
 551
 552         hsd.mti = mti;
 553         obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
 554
 555         set_cdt_state(cdt, CDT_RUNNING, NULL);
 556
 557         /* Inform mdt_hsm_cdt_start(). */
 558         wake_up_all(&cdt->cdt_waitq);
 559
 560         while (1) {
 561                 int i;
 562                 int update_idx = 0;
 563                 int updates_sz;
 564                 int updates_cnt;
 565                 struct hsm_record_update *updates;
 566
 567                 /* Limit execution of the expensive requests traversal
 568                  * to at most one second. This prevents repeatedly
 569                  * locking/unlocking the catalog for each request
 570                  * and preventing other HSM operations from happening
 571                  */
 572                 wait_event_interruptible_timeout(cdt->cdt_waitq,
 573                                                  kthread_should_stop() ||
 574                                                  cdt->cdt_wakeup_coordinator,
 575                                                  cfs_time_seconds(1));
 576
 577                 cdt->cdt_wakeup_coordinator = false;
 578                 CDEBUG(D_HSM, "coordinator resumes\n");
 579
 580                 if (kthread_should_stop()) {
 581                         CDEBUG(D_HSM, "Coordinator stops\n");
 582                         rc = 0;
 583                         break;
 584                 }
 585
 586                 /* if coordinator is suspended continue to wait */
 587                 if (cdt->cdt_state == CDT_DISABLE) {
 588                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 589                         continue;
 590                 }
 591
 592                 /* If no event, and no housekeeping to do, continue to
 593                  * wait. */
 594                 if (last_housekeeping + cdt->cdt_loop_period <=
 595                     ktime_get_real_seconds()) {
 596                         last_housekeeping = ktime_get_real_seconds();
 597                         hsd.housekeeping = true;
 598                 } else if (cdt->cdt_event) {
 599                         hsd.housekeeping = false;
 600                 } else {
 601                         continue;
 602                 }
 603
 604                 cdt->cdt_event = false;
 605
 606                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 607
 608                 if (hsd.max_requests != cdt->cdt_max_requests) {
 609                         /* cdt_max_requests has changed,
 610                          * we need to allocate a new buffer
 611                          */
 612                         struct hsm_scan_request *tmp = NULL;
 613                         int max_requests = cdt->cdt_max_requests;
 614                         OBD_ALLOC_LARGE(tmp, max_requests *
 615                                         sizeof(struct hsm_scan_request));
 616                         if (!tmp) {
 617                                 CERROR("Failed to resize request buffer, "
 618                                        "keeping it at %d\n",
 619                                        hsd.max_requests);
 620                                 cdt->cdt_max_requests = hsd.max_requests;
 621                         } else {
 622                                 OBD_FREE_LARGE(hsd.request, request_sz);
 623                                 hsd.max_requests = max_requests;
 624                                 request_sz = hsd.max_requests *
 625                                         sizeof(struct hsm_scan_request);
 626                                 hsd.request = tmp;
 627                         }
 628                 }
 629
 630                 hsd.request_cnt = 0;
 631
 632                 rc = cdt_llog_process(mti->mti_env, mdt, mdt_coordinator_cb,
 633                                       &hsd, 0, 0, WRITE);
 634                 if (rc < 0)
 635                         goto clean_cb_alloc;
 636
 637                 CDEBUG(D_HSM, "found %d requests to send\n", hsd.request_cnt);
 638
 639                 if (list_empty(&cdt->cdt_agents)) {
 640                         CDEBUG(D_HSM, "no agent available, "
 641                                       "coordinator sleeps\n");
 642                         goto clean_cb_alloc;
 643                 }
 644
 645                 /* Compute how many HAI we have in all the requests */
 646                 updates_cnt = 0;
 647                 for (i = 0; i < hsd.request_cnt; i++) {
 648                         const struct hsm_scan_request *request =
 649                                 &hsd.request[i];
 650
 651                         updates_cnt += request->hal->hal_count;
 652                 }
 653
 654                 /* Allocate a temporary array to store the cookies to
 655                  * update, and their status. */
 656                 updates_sz = updates_cnt * sizeof(*updates);
 657                 OBD_ALLOC(updates, updates_sz);
 658                 if (updates == NULL) {
 659                         CERROR("%s: Cannot allocate memory (%d o) "
 660                                "for %d updates\n",
 661                                mdt_obd_name(mdt), updates_sz, updates_cnt);
 662                         continue;
 663                 }
 664
 665                 /* here hsd contains a list of requests to be started */
 666                 for (i = 0; i < hsd.request_cnt; i++) {
 667                         struct hsm_scan_request *request = &hsd.request[i];
 668                         struct hsm_action_list  *hal = request->hal;
 669                         struct hsm_action_item  *hai;
 670                         int                      j;
 671
 672                         /* still room for work ? */
 673                         if (atomic_read(&cdt->cdt_request_count) >=
 674                             cdt->cdt_max_requests)
 675                                 break;
 676
 677                         rc = mdt_hsm_agent_send(mti, hal, 0);
 678                         /* if failure, we suppose it is temporary
 679                          * if the copy tool failed to do the request
 680                          * it has to use hsm_progress
 681                          */
 682
 683                         /* set up cookie vector to set records status
 684                          * after copy tools start or failed
 685                          */
 686                         hai = hai_first(hal);
 687                         for (j = 0; j < hal->hal_count; j++) {
 688                                 updates[update_idx].cookie = hai->hai_cookie;
 689                                 updates[update_idx].status =
 690                                         (rc ? ARS_WAITING : ARS_STARTED);
 691                                 hai = hai_next(hai);
 692                                 update_idx++;
 693                         }
 694                 }
 695
 696                 if (update_idx) {
 697                         rc = mdt_agent_record_update(mti->mti_env, mdt,
 698                                                      updates, update_idx);
 699                         if (rc)
 700                                 CERROR("%s: mdt_agent_record_update() failed, "
 701                                        "rc=%d, cannot update records "
 702                                        "for %d cookies\n",
 703                                        mdt_obd_name(mdt), rc, update_idx);
 704                 }
 705
 706                 OBD_FREE(updates, updates_sz);
 707
 708 clean_cb_alloc:
 709                 /* free hal allocated by callback */
 710                 for (i = 0; i < hsd.request_cnt; i++) {
 711                         struct hsm_scan_request *request = &hsd.request[i];
 712
 713                         OBD_FREE(request->hal, request->hal_sz);
 714                 }
 715         }
 716
 717         if (hsd.request)
 718                 OBD_FREE_LARGE(hsd.request, request_sz);
 719
 720         mdt_hsm_cdt_cleanup(mdt);
 721
 722         if (rc != 0)
 723                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 724                        mdt_obd_name(mdt), current_pid(), rc);
 725         else
 726                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 727                               " no error\n",
 728                        mdt_obd_name(mdt), current_pid());
 729
 730         RETURN(rc);
 731 }
 732
 733 /**
 734  * lookup a restore handle by FID
 735  * caller needs to hold cdt_restore_lock
 736  * \param cdt [IN] coordinator
 737  * \param fid [IN] FID
 738  * \retval cdt_restore_handle found
 739  * \retval NULL not found
 740  */
 741 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
 742                                                        const struct lu_fid *fid)
 743 {
 744         struct cdt_restore_handle       *crh;
 745         ENTRY;
 746
 747         list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
 748                 if (lu_fid_eq(&crh->crh_fid, fid))
 749                         RETURN(crh);
 750         }
 751         RETURN(NULL);
 752 }
 753
 754 /**
 755  * data passed to llog_cat_process() callback
 756  * to scan requests and take actions
 757  */
 758 struct hsm_restore_data {
 759         struct mdt_thread_info  *hrd_mti;
 760 };
 761
 762 /**
 763  *  llog_cat_process() callback, used to:
 764  *  - find restore request and allocate the restore handle
 765  * \param env [IN] environment
 766  * \param llh [IN] llog handle
 767  * \param hdr [IN] llog record
 768  * \param data [IN/OUT] cb data = struct hsm_restore_data
 769  * \retval 0 success
 770  * \retval -ve failure
 771  */
 772 static int hsm_restore_cb(const struct lu_env *env,
 773                           struct llog_handle *llh,
 774                           struct llog_rec_hdr *hdr, void *data)
 775 {
 776         struct llog_agent_req_rec       *larr;
 777         struct hsm_restore_data         *hrd;
 778         struct cdt_restore_handle       *crh;
 779         struct hsm_action_item          *hai;
 780         struct mdt_thread_info          *mti;
 781         struct coordinator              *cdt;
 782         struct mdt_object               *child;
 783         int rc;
 784         ENTRY;
 785
 786         hrd = data;
 787         mti = hrd->hrd_mti;
 788         cdt = &mti->mti_mdt->mdt_coordinator;
 789
 790         larr = (struct llog_agent_req_rec *)hdr;
 791         hai = &larr->arr_hai;
 792         if (hai->hai_cookie > cdt->cdt_last_cookie)
 793                 /* update the cookie to avoid collision */
 794                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 795
 796         if (hai->hai_action != HSMA_RESTORE ||
 797             agent_req_in_final_state(larr->arr_status))
 798                 RETURN(0);
 799
 800         /* restore request not in a final state */
 801
 802         /* force replay of restore requests left in started state from previous
 803          * CDT context, to be canceled later if finally found to be incompatible
 804          * when being re-started */
 805         if (larr->arr_status == ARS_STARTED) {
 806                 larr->arr_status = ARS_WAITING;
 807                 larr->arr_req_change = ktime_get_real_seconds();
 808                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 809                 if (rc != 0)
 810                         GOTO(out, rc);
 811         }
 812
 813         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 814         if (crh == NULL)
 815                 RETURN(-ENOMEM);
 816
 817         crh->crh_fid = hai->hai_fid;
 818         /* in V1 all file is restored
 819         crh->extent.start = hai->hai_extent.offset;
 820         crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
 821         */
 822         crh->crh_extent.start = 0;
 823         crh->crh_extent.end = hai->hai_extent.length;
 824         /* get the layout lock */
 825         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 826         child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 827                                      MDS_INODELOCK_LAYOUT);
 828         if (IS_ERR(child))
 829                 GOTO(out, rc = PTR_ERR(child));
 830
 831         rc = 0;
 832         /* we choose to not keep a reference
 833          * on the object during the restore time which can be very long */
 834         mdt_object_put(mti->mti_env, child);
 835
 836         mutex_lock(&cdt->cdt_restore_lock);
 837         list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
 838         mutex_unlock(&cdt->cdt_restore_lock);
 839
 840 out:
 841         RETURN(rc);
 842 }
 843
 844 /**
 845  * restore coordinator state at startup
 846  * the goal is to take a layout lock for each registered restore request
 847  * \param mti [IN] context
 848  */
 849 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 850 {
 851         struct hsm_restore_data  hrd;
 852         int                      rc;
 853         ENTRY;
 854
 855         hrd.hrd_mti = mti;
 856
 857         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, hsm_restore_cb, &hrd,
 858                               0, 0, WRITE);
 859
 860         RETURN(rc);
 861 }
 862
 863 static int hsm_init_ucred(struct lu_ucred *uc)
 864 {
 865         ENTRY;
 866
 867         uc->uc_valid = UCRED_OLD;
 868         uc->uc_o_uid = 0;
 869         uc->uc_o_gid = 0;
 870         uc->uc_o_fsuid = 0;
 871         uc->uc_o_fsgid = 0;
 872         uc->uc_uid = 0;
 873         uc->uc_gid = 0;
 874         uc->uc_fsuid = 0;
 875         uc->uc_fsgid = 0;
 876         uc->uc_suppgids[0] = -1;
 877         uc->uc_suppgids[1] = -1;
 878         uc->uc_cap = CFS_CAP_FS_MASK;
 879         uc->uc_umask = 0777;
 880         uc->uc_ginfo = NULL;
 881         uc->uc_identity = NULL;
 882
 883         RETURN(0);
 884 }
 885
 886 /**
 887  * initialize coordinator struct
 888  * \param mdt [IN] device
 889  * \retval 0 success
 890  * \retval -ve failure
 891  */
 892 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 893 {
 894         struct coordinator      *cdt = &mdt->mdt_coordinator;
 895         struct mdt_thread_info  *cdt_mti = NULL;
 896         int                      rc;
 897         ENTRY;
 898
 899         init_waitqueue_head(&cdt->cdt_waitq);
 900         init_rwsem(&cdt->cdt_llog_lock);
 901         init_rwsem(&cdt->cdt_agent_lock);
 902         init_rwsem(&cdt->cdt_request_lock);
 903         mutex_init(&cdt->cdt_restore_lock);
 904         spin_lock_init(&cdt->cdt_state_lock);
 905         set_cdt_state(cdt, CDT_STOPPED, NULL);
 906
 907         INIT_LIST_HEAD(&cdt->cdt_request_list);
 908         INIT_LIST_HEAD(&cdt->cdt_agents);
 909         INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 910
 911         cdt->cdt_request_cookie_hash = cfs_hash_create("REQUEST_COOKIE_HASH",
 912                                                        CFS_HASH_BITS_MIN,
 913                                                        CFS_HASH_BITS_MAX,
 914                                                        CFS_HASH_BKT_BITS,
 915                                                        0 /* extra bytes */,
 916                                                        CFS_HASH_MIN_THETA,
 917                                                        CFS_HASH_MAX_THETA,
 918                                                 &cdt_request_cookie_hash_ops,
 919                                                        CFS_HASH_DEFAULT);
 920         if (cdt->cdt_request_cookie_hash == NULL)
 921                 RETURN(-ENOMEM);
 922
 923         cdt->cdt_agent_record_hash = cfs_hash_create("AGENT_RECORD_HASH",
 924                                                      CFS_HASH_BITS_MIN,
 925                                                      CFS_HASH_BITS_MAX,
 926                                                      CFS_HASH_BKT_BITS,
 927                                                      0 /* extra bytes */,
 928                                                      CFS_HASH_MIN_THETA,
 929                                                      CFS_HASH_MAX_THETA,
 930                                                      &cdt_agent_record_hash_ops,
 931                                                      CFS_HASH_DEFAULT);
 932         if (cdt->cdt_agent_record_hash == NULL)
 933                 GOTO(out_request_cookie_hash, rc = -ENOMEM);
 934
 935         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 936         if (rc < 0)
 937                 GOTO(out_agent_record_hash, rc);
 938
 939         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
 940         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
 941         if (rc < 0)
 942                 GOTO(out_env, rc);
 943
 944         lu_context_enter(&cdt->cdt_session);
 945         cdt->cdt_env.le_ses = &cdt->cdt_session;
 946
 947         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 948         LASSERT(cdt_mti != NULL);
 949
 950         cdt_mti->mti_env = &cdt->cdt_env;
 951         cdt_mti->mti_mdt = mdt;
 952
 953         hsm_init_ucred(mdt_ucred(cdt_mti));
 954
 955         /* default values for /proc tunnables
 956          * can be override by MGS conf */
 957         cdt->cdt_default_archive_id = 1;
 958         cdt->cdt_grace_delay = 60;
 959         cdt->cdt_loop_period = 10;
 960         cdt->cdt_max_requests = 3;
 961         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 962         cdt->cdt_active_req_timeout = 3600;
 963
 964         /* Initialize cdt_compound_id here to allow its usage for
 965          * delayed requests from RAoLU policy */
 966         atomic_set(&cdt->cdt_compound_id, ktime_get_real_seconds());
 967
 968         /* by default do not remove archives on last unlink */
 969         cdt->cdt_remove_archive_on_last_unlink = false;
 970
 971         RETURN(0);
 972
 973 out_env:
 974         lu_env_fini(&cdt->cdt_env);
 975 out_agent_record_hash:
 976         cfs_hash_putref(cdt->cdt_agent_record_hash);
 977         cdt->cdt_agent_record_hash = NULL;
 978 out_request_cookie_hash:
 979         cfs_hash_putref(cdt->cdt_request_cookie_hash);
 980         cdt->cdt_request_cookie_hash = NULL;
 981
 982         return rc;
 983 }
 984
 985 /**
 986  * free a coordinator thread
 987  * \param mdt [IN] device
 988  */
 989 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
 990 {
 991         struct coordinator *cdt = &mdt->mdt_coordinator;
 992         ENTRY;
 993
 994         lu_context_exit(cdt->cdt_env.le_ses);
 995         lu_context_fini(cdt->cdt_env.le_ses);
 996
 997         lu_env_fini(&cdt->cdt_env);
 998
 999         cfs_hash_putref(cdt->cdt_agent_record_hash);
1000         cdt->cdt_agent_record_hash = NULL;
1001
1002         cfs_hash_putref(cdt->cdt_request_cookie_hash);
1003         cdt->cdt_request_cookie_hash = NULL;
1004
1005         RETURN(0);
1006 }
1007
1008 /**
1009  * start a coordinator thread
1010  * \param mdt [IN] device
1011  * \retval 0 success
1012  * \retval -ve failure
1013  */
1014 static int mdt_hsm_cdt_start(struct mdt_device *mdt)
1015 {
1016         struct coordinator      *cdt = &mdt->mdt_coordinator;
1017         int                      rc;
1018         void                    *ptr;
1019         struct task_struct      *task;
1020         int                      request_sz;
1021         struct hsm_thread_data   thread_data;
1022         ENTRY;
1023
1024         /* functions defined but not yet used
1025          * this avoid compilation warning
1026          */
1027         ptr = dump_requests;
1028
1029         rc = set_cdt_state(cdt, CDT_INIT, NULL);
1030         if (rc) {
1031                 CERROR("%s: Coordinator already started or stopping\n",
1032                        mdt_obd_name(mdt));
1033                 RETURN(-EALREADY);
1034         }
1035
1036         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
1037         cdt->cdt_policy = CDT_DEFAULT_POLICY;
1038
1039         /* just need to be larger than previous one */
1040         /* cdt_last_cookie is protected by cdt_llog_lock */
1041         cdt->cdt_last_cookie = ktime_get_real_seconds();
1042         atomic_set(&cdt->cdt_request_count, 0);
1043         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
1044         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
1045         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
1046
1047         /* to avoid deadlock when start is made through /proc
1048          * /proc entries are created by the coordinator thread */
1049
1050         /* set up list of started restore requests */
1051         thread_data.cdt_mti =
1052                 lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1053         rc = mdt_hsm_pending_restore(thread_data.cdt_mti);
1054         if (rc)
1055                 CERROR("%s: cannot take the layout locks needed"
1056                        " for registered restore: %d\n",
1057                        mdt_obd_name(mdt), rc);
1058
1059         if (mdt->mdt_bottom->dd_rdonly)
1060                 RETURN(0);
1061
1062         /* Allocate the initial hsd.request[] vector*/
1063         request_sz = cdt->cdt_max_requests * sizeof(struct hsm_scan_request);
1064         OBD_ALLOC_LARGE(thread_data.request, request_sz);
1065         if (!thread_data.request) {
1066                 set_cdt_state(cdt, CDT_STOPPED, NULL);
1067                 RETURN(-ENOMEM);
1068         }
1069
1070         task = kthread_run(mdt_coordinator, &thread_data, "hsm_cdtr");
1071         if (IS_ERR(task)) {
1072                 rc = PTR_ERR(task);
1073                 set_cdt_state(cdt, CDT_STOPPED, NULL);
1074                 OBD_FREE(thread_data.request, request_sz);
1075                 CERROR("%s: error starting coordinator thread: %d\n",
1076                        mdt_obd_name(mdt), rc);
1077         } else {
1078                 cdt->cdt_task = task;
1079                 wait_event(cdt->cdt_waitq,
1080                            cdt->cdt_state != CDT_INIT);
1081                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1082                        mdt_obd_name(mdt));
1083                 rc = 0;
1084         }
1085
1086         RETURN(rc);
1087 }
1088
1089 /**
1090  * stop a coordinator thread
1091  * \param mdt [IN] device
1092  */
1093 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1094 {
1095         struct coordinator *cdt = &mdt->mdt_coordinator;
1096         int rc;
1097
1098         ENTRY;
1099         /* stop coordinator thread */
1100         rc = set_cdt_state(cdt, CDT_STOPPING, NULL);
1101         if (rc == 0) {
1102                 kthread_stop(cdt->cdt_task);
1103                 cdt->cdt_task = NULL;
1104                 set_cdt_state(cdt, CDT_STOPPED, NULL);
1105         }
1106
1107         RETURN(rc);
1108 }
1109
1110 static int mdt_hsm_set_exists(struct mdt_thread_info *mti,
1111                               const struct lu_fid *fid,
1112                               u32 archive_id)
1113 {
1114         struct mdt_object *obj;
1115         struct md_hsm mh;
1116         int rc;
1117
1118         obj = mdt_hsm_get_md_hsm(mti, fid, &mh);
1119         if (IS_ERR(obj))
1120                 GOTO(out, rc = PTR_ERR(obj));
1121
1122         if (mh.mh_flags & HS_EXISTS &&
1123             mh.mh_arch_id == archive_id)
1124                 GOTO(out_obj, rc = 0);
1125
1126         mh.mh_flags |= HS_EXISTS;
1127         mh.mh_arch_id = archive_id;
1128         rc = mdt_hsm_attr_set(mti, obj, &mh);
1129
1130 out_obj:
1131         mdt_object_put(mti->mti_env, obj);
1132 out:
1133         return rc;
1134 }
1135
1136 /**
1137  * register all requests from an hal in the memory list
1138  * \param mti [IN] context
1139  * \param hal [IN] request
1140  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1141  *  which is running the CT
1142  * \retval 0 success
1143  * \retval -ve failure
1144  */
1145 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1146                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1147 {
1148         struct mdt_device       *mdt = mti->mti_mdt;
1149         struct coordinator      *cdt = &mdt->mdt_coordinator;
1150         struct hsm_action_item  *hai;
1151         int                      rc = 0, i;
1152         ENTRY;
1153
1154         /* register request in memory list */
1155         hai = hai_first(hal);
1156         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1157                 struct cdt_agent_req *car;
1158
1159                 /* in case of a cancel request, we first mark the ondisk
1160                  * record of the request we want to stop as canceled
1161                  * this does not change the cancel record
1162                  * it will be done when updating the request status
1163                  */
1164                 if (hai->hai_action == HSMA_CANCEL) {
1165                         struct hsm_record_update update = {
1166                                 .cookie = hai->hai_cookie,
1167                                 .status = ARS_CANCELED,
1168                         };
1169
1170                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1171                                                      &update, 1);
1172                         if (rc) {
1173                                 CERROR("%s: mdt_agent_record_update() failed, "
1174                                        "rc=%d, cannot update status to %s "
1175                                        "for cookie %#llx\n",
1176                                        mdt_obd_name(mdt), rc,
1177                                        agent_req_status2name(ARS_CANCELED),
1178                                        hai->hai_cookie);
1179                                 GOTO(out, rc);
1180                         }
1181
1182                         /* find the running request to set it canceled */
1183                         car = mdt_cdt_find_request(cdt, hai->hai_cookie);
1184                         if (car != NULL) {
1185                                 car->car_canceled = 1;
1186                                 /* uuid has to be changed to the one running the
1187                                 * request to cancel */
1188                                 *uuid = car->car_uuid;
1189                                 mdt_cdt_put_request(car);
1190                         }
1191                         /* no need to memorize cancel request
1192                          * this also avoid a deadlock when we receive
1193                          * a purge all requests command
1194                          */
1195                         continue;
1196                 }
1197
1198                 if (hai->hai_action == HSMA_ARCHIVE) {
1199                         rc = mdt_hsm_set_exists(mti, &hai->hai_fid,
1200                                                 hal->hal_archive_id);
1201                         if (rc == -ENOENT)
1202                                 continue;
1203                         else if (rc < 0)
1204                                 GOTO(out, rc);
1205                 }
1206
1207                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1208                                             hal->hal_archive_id, hal->hal_flags,
1209                                             uuid, hai);
1210                 if (IS_ERR(car))
1211                         GOTO(out, rc = PTR_ERR(car));
1212
1213                 rc = mdt_cdt_add_request(cdt, car);
1214                 if (rc != 0)
1215                         mdt_cdt_free_request(car);
1216         }
1217 out:
1218         RETURN(rc);
1219 }
1220
1221 /**
1222  * swap layouts between 2 fids
1223  * \param mti [IN] context
1224  * \param obj [IN]
1225  * \param dfid [IN]
1226  * \param mh_common [IN] MD HSM
1227  */
1228 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1229                             struct mdt_object *obj, const struct lu_fid *dfid,
1230                             struct md_hsm *mh_common)
1231 {
1232         struct mdt_object       *dobj;
1233         struct mdt_lock_handle  *dlh;
1234         int                      rc;
1235         ENTRY;
1236
1237         if (!mdt_object_exists(obj))
1238                 GOTO(out, rc = -ENOENT);
1239
1240         /* we already have layout lock on obj so take only
1241          * on dfid */
1242         dlh = &mti->mti_lh[MDT_LH_OLD];
1243         mdt_lock_reg_init(dlh, LCK_EX);
1244         dobj = mdt_object_find_lock(mti, dfid, dlh, MDS_INODELOCK_LAYOUT);
1245         if (IS_ERR(dobj))
1246                 GOTO(out, rc = PTR_ERR(dobj));
1247
1248         /* if copy tool closes the volatile before sending the final
1249          * progress through llapi_hsm_copy_end(), all the objects
1250          * are removed and mdd_swap_layout LBUG */
1251         if (!mdt_object_exists(dobj)) {
1252                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1253                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1254                 GOTO(out_dobj, rc = -ENOENT);
1255         }
1256         /* Since we only handle restores here, unconditionally use
1257          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1258          * be preserved in case of failure during swap_layout and not
1259          * leave a file in an intermediate but incoherent state.
1260          * But need to setup HSM xattr of data FID before, reuse
1261          * mti and mh presets for FID in hsm_cdt_request_completed(),
1262          * only need to clear RELEASED and DIRTY.
1263          */
1264         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1265         rc = mdt_hsm_attr_set(mti, dobj, mh_common);
1266         if (rc == 0)
1267                 rc = mo_swap_layouts(mti->mti_env,
1268                                      mdt_object_child(obj),
1269                                      mdt_object_child(dobj),
1270                                      SWAP_LAYOUTS_MDS_HSM);
1271
1272 out_dobj:
1273         mdt_object_unlock_put(mti, dobj, dlh, 1);
1274 out:
1275         RETURN(rc);
1276 }
1277
1278 /**
1279  * update status of a completed request
1280  * \param mti [IN] context
1281  * \param pgs [IN] progress of the copy tool
1282  * \param update_record [IN] update llog record
1283  * \retval 0 success
1284  * \retval -ve failure
1285  */
1286 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1287                                      struct hsm_progress_kernel *pgs,
1288                                      const struct cdt_agent_req *car,
1289                                      enum agent_req_status *status)
1290 {
1291         const struct lu_env     *env = mti->mti_env;
1292         struct mdt_device       *mdt = mti->mti_mdt;
1293         struct coordinator      *cdt = &mdt->mdt_coordinator;
1294         struct mdt_object       *obj = NULL;
1295         int                      cl_flags = 0, rc = 0;
1296         struct md_hsm            mh;
1297         bool                     is_mh_changed;
1298         bool                     need_changelog = true;
1299         ENTRY;
1300
1301         /* default is to retry */
1302         *status = ARS_WAITING;
1303
1304         /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
1305          * if error/removed continue anyway to get correct reporting done */
1306         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1307         /* we will update MD HSM only if needed */
1308         is_mh_changed = false;
1309
1310         /* no need to change mh->mh_arch_id
1311          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1312          */
1313         if (pgs->hpk_errval != 0) {
1314                 switch (pgs->hpk_errval) {
1315                 case ENOSYS:
1316                         /* the copy tool does not support cancel
1317                          * so the cancel request is failed
1318                          * As we cannot distinguish a cancel progress
1319                          * from another action progress (they have the
1320                          * same cookie), we suppose here the CT returns
1321                          * ENOSYS only if does not support cancel
1322                          */
1323                         /* this can also happen when cdt calls it to
1324                          * for a timed out request */
1325                         *status = ARS_FAILED;
1326                         /* to have a cancel event in changelog */
1327                         pgs->hpk_errval = ECANCELED;
1328                         break;
1329                 case ECANCELED:
1330                         /* the request record has already been set to
1331                          * ARS_CANCELED, this set the cancel request
1332                          * to ARS_SUCCEED */
1333                         *status = ARS_SUCCEED;
1334                         break;
1335                 default:
1336                         /* retry only if current policy or requested, and
1337                          * object is not on error/removed */
1338                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1339                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ||
1340                                    IS_ERR(obj)) ? ARS_FAILED : ARS_WAITING;
1341                         break;
1342                 }
1343
1344                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1345                         CERROR("%s: Request %#llx on "DFID
1346                                " failed, error code %d too large\n",
1347                                mdt_obd_name(mdt),
1348                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1349                                pgs->hpk_errval);
1350                         hsm_set_cl_error(&cl_flags,
1351                                          CLF_HSM_ERROVERFLOW);
1352                         rc = -EINVAL;
1353                 } else {
1354                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1355                 }
1356
1357                 switch (car->car_hai->hai_action) {
1358                 case HSMA_ARCHIVE:
1359                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1360                         break;
1361                 case HSMA_RESTORE:
1362                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1363                         break;
1364                 case HSMA_REMOVE:
1365                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1366                         break;
1367                 case HSMA_CANCEL:
1368                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1369                         CERROR("%s: Failed request %#llx on "DFID
1370                                " cannot be a CANCEL\n",
1371                                mdt_obd_name(mdt),
1372                                pgs->hpk_cookie,
1373                                PFID(&pgs->hpk_fid));
1374                         break;
1375                 default:
1376                         CERROR("%s: Failed request %#llx on "DFID
1377                                " %d is an unknown action\n",
1378                                mdt_obd_name(mdt),
1379                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1380                                car->car_hai->hai_action);
1381                         rc = -EINVAL;
1382                         break;
1383                 }
1384         } else {
1385                 *status = ARS_SUCCEED;
1386                 switch (car->car_hai->hai_action) {
1387                 case HSMA_ARCHIVE:
1388                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1389                         /* set ARCHIVE keep EXIST and clear LOST and
1390                          * DIRTY */
1391                         mh.mh_arch_ver = pgs->hpk_data_version;
1392                         mh.mh_flags |= HS_ARCHIVED;
1393                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1394                         is_mh_changed = true;
1395                         break;
1396                 case HSMA_RESTORE:
1397                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1398
1399                         /* do not clear RELEASED and DIRTY here
1400                          * this will occur in hsm_swap_layouts()
1401                          */
1402
1403                         /* Restoring has changed the file version on
1404                          * disk. */
1405                         mh.mh_arch_ver = pgs->hpk_data_version;
1406                         is_mh_changed = true;
1407                         break;
1408                 case HSMA_REMOVE:
1409                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1410                         /* clear ARCHIVED EXISTS and LOST */
1411                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1412                         is_mh_changed = true;
1413                         break;
1414                 case HSMA_CANCEL:
1415                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1416                         CERROR("%s: Successful request %#llx on "DFID" cannot be a CANCEL\n",
1417                                mdt_obd_name(mdt),
1418                                pgs->hpk_cookie,
1419                                PFID(&pgs->hpk_fid));
1420                         break;
1421                 default:
1422                         CERROR("%s: Successful request %#llx on "DFID" %d is an unknown action\n",
1423                                mdt_obd_name(mdt),
1424                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1425                                car->car_hai->hai_action);
1426                         rc = -EINVAL;
1427                         break;
1428                 }
1429         }
1430
1431         /* rc != 0 means error when analysing action, it may come from
1432          * a crasy CT no need to manage DIRTY
1433          * and if mdt_hsm_get_md_hsm() has returned an error, mh has not been
1434          * filled
1435          */
1436         if (rc == 0 && !IS_ERR(obj))
1437                 hsm_set_cl_flags(&cl_flags,
1438                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1439
1440         /* unlock is done later, after layout lock management */
1441         if (is_mh_changed && !IS_ERR(obj))
1442                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1443
1444         /* we give back layout lock only if restore was successful or
1445          * if no retry will be attempted and if object is still alive,
1446          * in other cases we just unlock the object */
1447         if (car->car_hai->hai_action == HSMA_RESTORE) {
1448                 struct cdt_restore_handle       *crh;
1449
1450                 /* restore in data FID done, we swap the layouts
1451                  * only if restore is successful */
1452                 if (pgs->hpk_errval == 0 && !IS_ERR(obj)) {
1453                         rc = hsm_swap_layouts(mti, obj, &car->car_hai->hai_dfid,
1454                                               &mh);
1455                         if (rc) {
1456                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1457                                         *status = ARS_FAILED;
1458                                 pgs->hpk_errval = -rc;
1459                         }
1460                 }
1461                 /* we have to retry, so keep layout lock */
1462                 if (*status == ARS_WAITING)
1463                         GOTO(out, rc);
1464
1465                 /* restore special case, need to create ChangeLog record
1466                  * before to give back layout lock to avoid concurrent
1467                  * file updater to post out of order ChangeLog */
1468                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1469                              &car->car_hai->hai_fid);
1470                 need_changelog = false;
1471
1472                 /* give back layout lock */
1473                 mutex_lock(&cdt->cdt_restore_lock);
1474                 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1475                 if (crh != NULL)
1476                         list_del(&crh->crh_list);
1477                 mutex_unlock(&cdt->cdt_restore_lock);
1478                 /* Just give back layout lock, we keep the reference
1479                  * which is given back later with the lock for HSM
1480                  * flags.
1481                  * XXX obj may be invalid so we do not pass it. */
1482                 if (crh != NULL)
1483                         mdt_object_unlock(mti, NULL, &crh->crh_lh, 1);
1484
1485                 if (crh != NULL)
1486                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1487         }
1488
1489         GOTO(out, rc);
1490
1491 out:
1492         /* always add a ChangeLog record */
1493         if (need_changelog)
1494                 mo_changelog(env, CL_HSM, cl_flags, mdt->mdt_child,
1495                              &car->car_hai->hai_fid);
1496
1497         if (!IS_ERR(obj))
1498                 mdt_object_put(mti->mti_env, obj);
1499
1500         RETURN(rc);
1501 }
1502
1503 /**
1504  * update status of a request
1505  * \param mti [IN] context
1506  * \param pgs [IN] progress of the copy tool
1507  * \param update_record [IN] update llog record
1508  * \retval 0 success
1509  * \retval -ve failure
1510  */
1511 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1512                                  struct hsm_progress_kernel *pgs,
1513                                  const int update_record)
1514 {
1515         struct mdt_device       *mdt = mti->mti_mdt;
1516         struct coordinator      *cdt = &mdt->mdt_coordinator;
1517         struct cdt_agent_req    *car;
1518         int                      rc = 0;
1519         ENTRY;
1520
1521         /* no coordinator started, so we cannot serve requests */
1522         if (cdt->cdt_state == CDT_STOPPED)
1523                 RETURN(-EAGAIN);
1524
1525         /* first do sanity checks */
1526         car = mdt_cdt_update_request(cdt, pgs);
1527         if (IS_ERR(car)) {
1528                 CERROR("%s: Cannot find running request for cookie %#llx"
1529                        " on fid="DFID"\n",
1530                        mdt_obd_name(mdt),
1531                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1532
1533                 RETURN(PTR_ERR(car));
1534         }
1535
1536         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie=%#llx"
1537                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1538                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1539                       hsm_copytool_action2name(car->car_hai->hai_action),
1540                       pgs->hpk_flags, pgs->hpk_errval,
1541                       PFID(&car->car_hai->hai_fid),
1542                       PFID(&car->car_hai->hai_dfid));
1543
1544         /* progress is done on FID or data FID depending of the action and
1545          * of the copy progress */
1546         /* for restore progress is used to send back the data FID to cdt */
1547         if (car->car_hai->hai_action == HSMA_RESTORE &&
1548             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1549                 car->car_hai->hai_dfid = pgs->hpk_fid;
1550
1551         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1552              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1553             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1554              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1555                 CERROR("%s: Progress on "DFID" for cookie %#llx"
1556                        " does not match request FID "DFID" nor data FID "
1557                        DFID"\n",
1558                        mdt_obd_name(mdt),
1559                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1560                        PFID(&car->car_hai->hai_fid),
1561                        PFID(&car->car_hai->hai_dfid));
1562                 GOTO(out, rc = -EINVAL);
1563         }
1564
1565         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1566                 CERROR("%s: Progress on "DFID" for cookie %#llx action=%s"
1567                        " is not coherent (err=%d and not completed"
1568                        " (flags=%d))\n",
1569                        mdt_obd_name(mdt),
1570                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1571                        hsm_copytool_action2name(car->car_hai->hai_action),
1572                        pgs->hpk_errval, pgs->hpk_flags);
1573                 GOTO(out, rc = -EINVAL);
1574         }
1575
1576         /* now progress is valid */
1577
1578         /* we use a root like ucred */
1579         hsm_init_ucred(mdt_ucred(mti));
1580
1581         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1582                 enum agent_req_status    status;
1583
1584                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1585
1586                 CDEBUG(D_HSM, "%s record: fid="DFID" cookie=%#llx action=%s "
1587                               "status=%s\n",
1588                        update_record ? "Updating" : "Not updating",
1589                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1590                        hsm_copytool_action2name(car->car_hai->hai_action),
1591                        agent_req_status2name(status));
1592
1593                 /* update record first (LU-9075) */
1594                 if (update_record) {
1595                         int rc1;
1596                         struct hsm_record_update update = {
1597                                 .cookie = pgs->hpk_cookie,
1598                                 .status = status,
1599                         };
1600
1601                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1602                                                       &update, 1);
1603                         if (rc1)
1604                                 CERROR("%s: mdt_agent_record_update() failed,"
1605                                        " rc=%d, cannot update status to %s"
1606                                        " for cookie %#llx\n",
1607                                        mdt_obd_name(mdt), rc1,
1608                                        agent_req_status2name(status),
1609                                        pgs->hpk_cookie);
1610                         rc = (rc != 0 ? rc : rc1);
1611                 }
1612
1613                 /* then remove request from memory list (LU-9075) */
1614                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1615
1616                 /* ct has completed a request, so a slot is available,
1617                  * signal the coordinator to find new work */
1618                 mdt_hsm_cdt_event(cdt);
1619         } else {
1620                 /* if copytool send a progress on a canceled request
1621                  * we inform copytool it should stop
1622                  */
1623                 if (car->car_canceled == 1)
1624                         rc = -ECANCELED;
1625         }
1626         GOTO(out, rc);
1627
1628 out:
1629         /* remove ref got from mdt_cdt_update_request() */
1630         mdt_cdt_put_request(car);
1631
1632         return rc;
1633 }
1634
1635
1636 /**
1637  * data passed to llog_cat_process() callback
1638  * to cancel requests
1639  */
1640 struct hsm_cancel_all_data {
1641         struct mdt_device       *mdt;
1642 };
1643
1644 /**
1645  *  llog_cat_process() callback, used to:
1646  *  - purge all requests
1647  * \param env [IN] environment
1648  * \param llh [IN] llog handle
1649  * \param hdr [IN] llog record
1650  * \param data [IN] cb data = struct hsm_cancel_all_data
1651  * \retval 0 success
1652  * \retval -ve failure
1653  */
1654 static int mdt_cancel_all_cb(const struct lu_env *env,
1655                              struct llog_handle *llh,
1656                              struct llog_rec_hdr *hdr, void *data)
1657 {
1658         struct llog_agent_req_rec       *larr;
1659         struct hsm_cancel_all_data      *hcad;
1660         int                              rc = 0;
1661         ENTRY;
1662
1663         larr = (struct llog_agent_req_rec *)hdr;
1664         hcad = data;
1665         if (larr->arr_status == ARS_WAITING ||
1666             larr->arr_status == ARS_STARTED) {
1667                 larr->arr_status = ARS_CANCELED;
1668                 larr->arr_req_change = ktime_get_real_seconds();
1669                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1670         }
1671
1672         RETURN(rc);
1673 }
1674
1675 /**
1676  * cancel all actions
1677  * \param obd [IN] MDT device
1678  */
1679 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1680 {
1681         struct lu_env                    env;
1682         struct lu_context                session;
1683         struct mdt_thread_info          *mti;
1684         struct coordinator              *cdt = &mdt->mdt_coordinator;
1685         struct cdt_agent_req            *car;
1686         struct hsm_action_list          *hal = NULL;
1687         struct hsm_action_item          *hai;
1688         struct hsm_cancel_all_data       hcad;
1689         int                              hal_sz = 0, hal_len, rc;
1690         enum cdt_states                  old_state;
1691         ENTRY;
1692
1693         rc = lu_env_init(&env, LCT_MD_THREAD);
1694         if (rc < 0)
1695                 RETURN(rc);
1696
1697         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
1698         rc = lu_context_init(&session, LCT_SERVER_SESSION);
1699         if (rc < 0)
1700                 GOTO(out_env, rc);
1701
1702         lu_context_enter(&session);
1703         env.le_ses = &session;
1704
1705         mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
1706         LASSERT(mti != NULL);
1707
1708         mti->mti_env = &env;
1709         mti->mti_mdt = mdt;
1710
1711         hsm_init_ucred(mdt_ucred(mti));
1712
1713         /* disable coordinator */
1714         rc = set_cdt_state(cdt, CDT_DISABLE, &old_state);
1715         if (rc)
1716                 RETURN(rc);
1717
1718         /* send cancel to all running requests */
1719         down_read(&cdt->cdt_request_lock);
1720         list_for_each_entry(car, &cdt->cdt_request_list, car_request_list) {
1721                 mdt_cdt_get_request(car);
1722                 /* request is not yet removed from list, it will be done
1723                  * when copytool will return progress
1724                  */
1725
1726                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1727                         mdt_cdt_put_request(car);
1728                         continue;
1729                 }
1730
1731                 /* needed size */
1732                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1733                           cfs_size_round(car->car_hai->hai_len);
1734
1735                 if (hal_len > hal_sz && hal_sz > 0) {
1736                         /* not enough room, free old buffer */
1737                         OBD_FREE(hal, hal_sz);
1738                         hal = NULL;
1739                 }
1740
1741                 /* empty buffer, allocate one */
1742                 if (hal == NULL) {
1743                         hal_sz = hal_len;
1744                         OBD_ALLOC(hal, hal_sz);
1745                         if (hal == NULL) {
1746                                 mdt_cdt_put_request(car);
1747                                 up_read(&cdt->cdt_request_lock);
1748                                 GOTO(out_cdt_state, rc = -ENOMEM);
1749                         }
1750                 }
1751
1752                 hal->hal_version = HAL_VERSION;
1753                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1754                                 MTI_NAME_MAXLEN);
1755                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1756                 hal->hal_compound_id = car->car_compound_id;
1757                 hal->hal_archive_id = car->car_archive_id;
1758                 hal->hal_flags = car->car_flags;
1759                 hal->hal_count = 0;
1760
1761                 hai = hai_first(hal);
1762                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1763                 hai->hai_action = HSMA_CANCEL;
1764                 hal->hal_count = 1;
1765
1766                 /* it is possible to safely call mdt_hsm_agent_send()
1767                  * (ie without a deadlock on cdt_request_lock), because the
1768                  * write lock is taken only if we are not in purge mode
1769                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1770                  *   nor mdt_cdt_remove_request())
1771                  */
1772                 /* no conflict with cdt thread because cdt is disable and we
1773                  * have the request lock */
1774                 mdt_hsm_agent_send(mti, hal, 1);
1775
1776                 mdt_cdt_put_request(car);
1777         }
1778         up_read(&cdt->cdt_request_lock);
1779
1780         if (hal != NULL)
1781                 OBD_FREE(hal, hal_sz);
1782
1783         /* cancel all on-disk records */
1784         hcad.mdt = mdt;
1785
1786         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt, mdt_cancel_all_cb,
1787                               &hcad, 0, 0, WRITE);
1788 out_cdt_state:
1789         /* Enable coordinator, unless the coordinator was stopping. */
1790         set_cdt_state(cdt, old_state, NULL);
1791         lu_context_exit(&session);
1792         lu_context_fini(&session);
1793 out_env:
1794         lu_env_fini(&env);
1795
1796         RETURN(rc);
1797 }
1798
1799 /**
1800  * check if a request is compatible with file status
1801  * \param hai [IN] request description
1802  * \param archive_id [IN] request archive id
1803  * \param rq_flags [IN] request flags
1804  * \param hsm [IN] file HSM metadata
1805  * \retval boolean
1806  */
1807 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1808                               u32 archive_id, u64 rq_flags,
1809                               const struct md_hsm *hsm)
1810 {
1811         int      is_compat = false;
1812         int      hsm_flags;
1813         ENTRY;
1814
1815         hsm_flags = hsm->mh_flags;
1816         switch (hai->hai_action) {
1817         case HSMA_ARCHIVE:
1818                 if (!(hsm_flags & HS_NOARCHIVE) &&
1819                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1820                         is_compat = true;
1821
1822                 if (hsm_flags & HS_EXISTS &&
1823                     archive_id != 0 &&
1824                     archive_id != hsm->mh_arch_id)
1825                         is_compat = false;
1826
1827                 break;
1828         case HSMA_RESTORE:
1829                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1830                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1831                         is_compat = true;
1832                 break;
1833         case HSMA_REMOVE:
1834                 if (!(hsm_flags & HS_RELEASED) &&
1835                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1836                         is_compat = true;
1837                 break;
1838         case HSMA_CANCEL:
1839                 is_compat = true;
1840                 break;
1841         }
1842         CDEBUG(D_HSM, "fid="DFID" action=%s flags=%#llx"
1843                       " extent=%#llx-%#llx hsm_flags=%.8X %s\n",
1844                       PFID(&hai->hai_fid),
1845                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1846                       hai->hai_extent.offset, hai->hai_extent.length,
1847                       hsm->mh_flags,
1848                       (is_compat ? "compatible" : "uncompatible"));
1849
1850         RETURN(is_compat);
1851 }
1852
1853 /*
1854  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1855  */
1856 static const struct {
1857         __u64            bit;
1858         char            *name;
1859         char            *nickname;
1860 } hsm_policy_names[] = {
1861         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1862         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1863         { 0 },
1864 };
1865
1866 /**
1867  * convert a policy name to a bit
1868  * \param name [IN] policy name
1869  * \retval 0 unknown
1870  * \retval   policy bit
1871  */
1872 static __u64 hsm_policy_str2bit(const char *name)
1873 {
1874         int      i;
1875
1876         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1877                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1878                     strcmp(hsm_policy_names[i].name, name) == 0)
1879                         return hsm_policy_names[i].bit;
1880         return 0;
1881 }
1882
1883 /**
1884  * convert a policy bit field to a string
1885  * \param mask [IN] policy bit field
1886  * \param hexa [IN] print mask before bit names
1887  * \param buffer [OUT] string
1888  * \param count [IN] size of buffer
1889  */
1890 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1891                                 const bool hexa)
1892 {
1893         int      i, j;
1894         __u64    bit;
1895         ENTRY;
1896
1897         if (hexa)
1898                 seq_printf(m, "(%#llx) ", mask);
1899
1900         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1901                 bit = (1ULL << i);
1902
1903                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1904                         if (hsm_policy_names[j].bit == bit)
1905                                 break;
1906                 }
1907                 if (bit & mask)
1908                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1909                 else
1910                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1911         }
1912         /* remove last ' ' */
1913         m->count--;
1914         seq_putc(m, '\n');
1915 }
1916
1917 /* methods to read/write HSM policy flags */
1918 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1919 {
1920         struct mdt_device       *mdt = m->private;
1921         struct coordinator      *cdt = &mdt->mdt_coordinator;
1922         ENTRY;
1923
1924         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1925         RETURN(0);
1926 }
1927
1928 static ssize_t
1929 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1930                          size_t count, loff_t *off)
1931 {
1932         struct seq_file         *m = file->private_data;
1933         struct mdt_device       *mdt = m->private;
1934         struct coordinator      *cdt = &mdt->mdt_coordinator;
1935         char                    *start, *token, sign;
1936         char                    *buf;
1937         __u64                    policy;
1938         __u64                    add_mask, remove_mask, set_mask;
1939         int                      rc;
1940         ENTRY;
1941
1942         if (count + 1 > PAGE_SIZE)
1943                 RETURN(-EINVAL);
1944
1945         OBD_ALLOC(buf, count + 1);
1946         if (buf == NULL)
1947                 RETURN(-ENOMEM);
1948
1949         if (copy_from_user(buf, buffer, count))
1950                 GOTO(out, rc = -EFAULT);
1951
1952         buf[count] = '\0';
1953
1954         start = buf;
1955         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1956                start);
1957
1958         add_mask = remove_mask = set_mask = 0;
1959         do {
1960                 token = strsep(&start, "\n ");
1961                 sign = *token;
1962
1963                 if (sign == '\0')
1964                         continue;
1965
1966                 if (sign == '-' || sign == '+')
1967                         token++;
1968
1969                 policy = hsm_policy_str2bit(token);
1970                 if (policy == 0) {
1971                         CWARN("%s: '%s' is unknown, "
1972                               "supported policies are:\n", mdt_obd_name(mdt),
1973                               token);
1974                         hsm_policy_bit2str(m, 0, false);
1975                         GOTO(out, rc = -EINVAL);
1976                 }
1977                 switch (sign) {
1978                 case '-':
1979                         remove_mask |= policy;
1980                         break;
1981                 case '+':
1982                         add_mask |= policy;
1983                         break;
1984                 default:
1985                         set_mask |= policy;
1986                         break;
1987                 }
1988
1989         } while (start != NULL);
1990
1991         CDEBUG(D_HSM, "%s: new policy: rm=%#llx add=%#llx set=%#llx\n",
1992                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1993
1994         /* if no sign in all string, it is a clear and set
1995          * if some sign found, all unsigned are converted
1996          * to add
1997          * P1 P2 = set to P1 and P2
1998          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1999          */
2000         if (remove_mask == 0 && add_mask == 0) {
2001                 cdt->cdt_policy = set_mask;
2002         } else {
2003                 cdt->cdt_policy |= set_mask | add_mask;
2004                 cdt->cdt_policy &= ~remove_mask;
2005         }
2006
2007         GOTO(out, rc = count);
2008
2009 out:
2010         OBD_FREE(buf, count + 1);
2011         RETURN(rc);
2012 }
2013 LPROC_SEQ_FOPS(mdt_hsm_policy);
2014
2015 #define GENERATE_PROC_METHOD(VAR)                                       \
2016 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
2017 {                                                                       \
2018         struct mdt_device       *mdt = m->private;                      \
2019         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
2020         ENTRY;                                                          \
2021                                                                         \
2022         seq_printf(m, "%llu\n", (__u64)cdt->VAR);                       \
2023         RETURN(0);                                                      \
2024 }                                                                       \
2025 static ssize_t                                                          \
2026 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
2027                           size_t count, loff_t *off)                    \
2028                                                                         \
2029 {                                                                       \
2030         struct seq_file         *m = file->private_data;                \
2031         struct mdt_device       *mdt = m->private;                      \
2032         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
2033         __s64                    val;                                   \
2034         int                      rc;                                    \
2035         ENTRY;                                                          \
2036                                                                         \
2037         rc = lprocfs_str_to_s64(buffer, count, &val);                   \
2038         if (rc)                                                         \
2039                 RETURN(rc);                                             \
2040         if (val > 0 && val < INT_MAX) {                                 \
2041                 cdt->VAR = val;                                         \
2042                 RETURN(count);                                          \
2043         }                                                               \
2044         RETURN(-EINVAL);                                                \
2045 }                                                                       \
2046
2047 GENERATE_PROC_METHOD(cdt_loop_period)
2048 GENERATE_PROC_METHOD(cdt_grace_delay)
2049 GENERATE_PROC_METHOD(cdt_active_req_timeout)
2050 GENERATE_PROC_METHOD(cdt_max_requests)
2051 GENERATE_PROC_METHOD(cdt_default_archive_id)
2052
2053 /*
2054  * procfs write method for MDT/hsm_control
2055  * proc entry is in mdt directory so data is mdt obd_device pointer
2056  */
2057 #define CDT_ENABLE_CMD   "enabled"
2058 #define CDT_STOP_CMD     "shutdown"
2059 #define CDT_DISABLE_CMD  "disabled"
2060 #define CDT_PURGE_CMD    "purge"
2061 #define CDT_HELP_CMD     "help"
2062 #define CDT_MAX_CMD_LEN  10
2063
2064 ssize_t
2065 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
2066                               size_t count, loff_t *off)
2067 {
2068         struct seq_file         *m = file->private_data;
2069         struct obd_device       *obd = m->private;
2070         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
2071         struct coordinator      *cdt = &(mdt->mdt_coordinator);
2072         int                      rc, usage = 0;
2073         char                     kernbuf[CDT_MAX_CMD_LEN];
2074         ENTRY;
2075
2076         if (count == 0 || count >= sizeof(kernbuf))
2077                 RETURN(-EINVAL);
2078
2079         if (copy_from_user(kernbuf, buffer, count))
2080                 RETURN(-EFAULT);
2081         kernbuf[count] = 0;
2082
2083         if (kernbuf[count - 1] == '\n')
2084                 kernbuf[count - 1] = 0;
2085
2086         rc = 0;
2087         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
2088                 if (cdt->cdt_state == CDT_DISABLE) {
2089                         rc = set_cdt_state(cdt, CDT_RUNNING, NULL);
2090                         mdt_hsm_cdt_event(cdt);
2091                         wake_up(&cdt->cdt_waitq);
2092                 } else {
2093                         rc = mdt_hsm_cdt_start(mdt);
2094                 }
2095         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2096                 if ((cdt->cdt_state == CDT_STOPPING) ||
2097                     (cdt->cdt_state == CDT_STOPPED)) {
2098                         CERROR("%s: Coordinator already stopped\n",
2099                                mdt_obd_name(mdt));
2100                         rc = -EALREADY;
2101                 } else {
2102                         rc = mdt_hsm_cdt_stop(mdt);
2103                 }
2104         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2105                 if ((cdt->cdt_state == CDT_STOPPING) ||
2106                     (cdt->cdt_state == CDT_STOPPED)) {
2107                         CERROR("%s: Coordinator is stopped\n",
2108                                mdt_obd_name(mdt));
2109                         rc = -EINVAL;
2110                 } else {
2111                         rc = set_cdt_state(cdt, CDT_DISABLE, NULL);
2112                 }
2113         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2114                 rc = hsm_cancel_all_actions(mdt);
2115         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2116                 usage = 1;
2117         } else {
2118                 usage = 1;
2119                 rc = -EINVAL;
2120         }
2121
2122         if (usage == 1)
2123                 CERROR("%s: Valid coordinator control commands are: "
2124                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2125                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2126                        CDT_PURGE_CMD, CDT_HELP_CMD);
2127
2128         if (rc)
2129                 RETURN(rc);
2130
2131         RETURN(count);
2132 }
2133
2134 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2135 {
2136         struct obd_device       *obd = m->private;
2137         struct coordinator      *cdt;
2138         ENTRY;
2139
2140         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2141
2142         seq_printf(m, "%s\n", cdt_mdt_state2str(cdt->cdt_state));
2143
2144         RETURN(0);
2145 }
2146
2147 static int
2148 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2149 {
2150         bool first = true;
2151         int i;
2152         ENTRY;
2153
2154         for (i = 0; i < 8 * sizeof(mask); i++) {
2155                 if (mask & (1UL << i)) {
2156                         seq_printf(m, "%s%s", first ? "" : " ",
2157                                    hsm_copytool_action2name(i));
2158                         first = false;
2159                 }
2160         }
2161         seq_putc(m, '\n');
2162
2163         RETURN(0);
2164 }
2165
2166 static int
2167 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2168 {
2169         struct mdt_device *mdt = m->private;
2170         struct coordinator *cdt = &mdt->mdt_coordinator;
2171
2172         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2173 }
2174
2175 static int
2176 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2177 {
2178         struct mdt_device *mdt = m->private;
2179         struct coordinator *cdt = &mdt->mdt_coordinator;
2180
2181         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2182 }
2183
2184 static int
2185 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2186 {
2187         struct mdt_device *mdt = m->private;
2188         struct coordinator *cdt = &mdt->mdt_coordinator;
2189
2190         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2191 }
2192
2193 static inline enum hsm_copytool_action
2194 hsm_copytool_name2action(const char *name)
2195 {
2196         if (strcasecmp(name, "NOOP") == 0)
2197                 return HSMA_NONE;
2198         else if (strcasecmp(name, "ARCHIVE") == 0)
2199                 return HSMA_ARCHIVE;
2200         else if (strcasecmp(name, "RESTORE") == 0)
2201                 return HSMA_RESTORE;
2202         else if (strcasecmp(name, "REMOVE") == 0)
2203                 return HSMA_REMOVE;
2204         else if (strcasecmp(name, "CANCEL") == 0)
2205                 return HSMA_CANCEL;
2206         else
2207                 return -1;
2208 }
2209
2210 static ssize_t
2211 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2212                             size_t user_count, __u64 *mask)
2213 {
2214         char *buf, *pos, *name;
2215         size_t buf_size;
2216         __u64 new_mask = 0;
2217         int rc;
2218         ENTRY;
2219
2220         if (!(user_count < 4096))
2221                 RETURN(-ENOMEM);
2222
2223         buf_size = user_count + 1;
2224
2225         OBD_ALLOC(buf, buf_size);
2226         if (buf == NULL)
2227                 RETURN(-ENOMEM);
2228
2229         if (copy_from_user(buf, user_buf, buf_size - 1))
2230                 GOTO(out, rc = -EFAULT);
2231
2232         buf[buf_size - 1] = '\0';
2233
2234         pos = buf;
2235         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2236                 int action;
2237
2238                 if (*name == '\0')
2239                         continue;
2240
2241                 action = hsm_copytool_name2action(name);
2242                 if (action < 0)
2243                         GOTO(out, rc = -EINVAL);
2244
2245                 new_mask |= (1UL << action);
2246         }
2247
2248         *mask = new_mask;
2249         rc = user_count;
2250 out:
2251         OBD_FREE(buf, buf_size);
2252
2253         RETURN(rc);
2254 }
2255
2256 static ssize_t
2257 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2258                                         size_t count, loff_t *off)
2259 {
2260         struct seq_file         *m = file->private_data;
2261         struct mdt_device       *mdt = m->private;
2262         struct coordinator *cdt = &mdt->mdt_coordinator;
2263
2264         return mdt_write_hsm_request_mask(file, buf, count,
2265                                            &cdt->cdt_user_request_mask);
2266 }
2267
2268 static ssize_t
2269 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2270                                         size_t count, loff_t *off)
2271 {
2272         struct seq_file         *m = file->private_data;
2273         struct mdt_device       *mdt = m->private;
2274         struct coordinator      *cdt = &mdt->mdt_coordinator;
2275
2276         return mdt_write_hsm_request_mask(file, buf, count,
2277                                            &cdt->cdt_group_request_mask);
2278 }
2279
2280 static ssize_t
2281 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2282                                         size_t count, loff_t *off)
2283 {
2284         struct seq_file         *m = file->private_data;
2285         struct mdt_device       *mdt = m->private;
2286         struct coordinator      *cdt = &mdt->mdt_coordinator;
2287
2288         return mdt_write_hsm_request_mask(file, buf, count,
2289                                            &cdt->cdt_other_request_mask);
2290 }
2291
2292 static int mdt_hsm_cdt_raolu_seq_show(struct seq_file *m, void *data)
2293 {
2294         struct mdt_device *mdt = m->private;
2295         struct coordinator *cdt = &mdt->mdt_coordinator;
2296         ENTRY;
2297
2298         seq_printf(m, "%d\n", (int)cdt->cdt_remove_archive_on_last_unlink);
2299         RETURN(0);
2300 }
2301
2302 static ssize_t
2303 mdt_hsm_cdt_raolu_seq_write(struct file *file, const char __user *buffer,
2304                             size_t count, loff_t *off)
2305
2306 {
2307         struct seq_file *m = file->private_data;
2308         struct mdt_device *mdt = m->private;
2309         struct coordinator *cdt = &mdt->mdt_coordinator;
2310         __s64 val;
2311         int rc;
2312         ENTRY;
2313
2314         rc = lprocfs_str_to_s64(buffer, count, &val);
2315         if (rc < 0)
2316                 RETURN(rc);
2317
2318         cdt->cdt_remove_archive_on_last_unlink = val;
2319         RETURN(count);
2320 }
2321
2322 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2323 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2324 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2325 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2326 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2327 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2328 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2329 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2330 LPROC_SEQ_FOPS(mdt_hsm_cdt_raolu);
2331
2332 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2333         { .name =       "agents",
2334           .fops =       &mdt_hsm_agent_fops                     },
2335         { .name =       "actions",
2336           .fops =       &mdt_hsm_actions_fops,
2337           .proc_mode =  0444                                    },
2338         { .name =       "default_archive_id",
2339           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2340         { .name =       "grace_delay",
2341           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2342         { .name =       "loop_period",
2343           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2344         { .name =       "max_requests",
2345           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2346         { .name =       "policy",
2347           .fops =       &mdt_hsm_policy_fops                    },
2348         { .name =       "active_request_timeout",
2349           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2350         { .name =       "active_requests",
2351           .fops =       &mdt_hsm_active_requests_fops           },
2352         { .name =       "user_request_mask",
2353           .fops =       &mdt_hsm_user_request_mask_fops,        },
2354         { .name =       "group_request_mask",
2355           .fops =       &mdt_hsm_group_request_mask_fops,       },
2356         { .name =       "other_request_mask",
2357           .fops =       &mdt_hsm_other_request_mask_fops,       },
2358         { .name =       "remove_archive_on_last_unlink",
2359           .fops =       &mdt_hsm_cdt_raolu_fops,                },
2360         { 0 }
2361 };