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) 2012, 2013, Intel Corporation.
  24  * Use is subject to license terms.
  25  * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
  26  *                          alternatives
  27  */
  28 /*
  29  * lustre/mdt/mdt_coordinator.c
  30  *
  31  * Lustre HSM Coordinator
  32  *
  33  * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
  34  * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
  35  * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_MDS
  39
  40 #include <obd_support.h>
  41 #include <lustre_net.h>
  42 #include <lustre_export.h>
  43 #include <obd.h>
  44 #include <obd_lov.h>
  45 #include <lprocfs_status.h>
  46 #include <lustre_log.h>
  47 #include "mdt_internal.h"
  48
  49 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
  50
  51 /**
  52  * get obj and HSM attributes on a fid
  53  * \param mti [IN] context
  54  * \param fid [IN] object fid
  55  * \param hsm [OUT] HSM meta data
  56  * \retval obj or error (-ENOENT if not found)
  57  */
  58 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  59                                       const struct lu_fid *fid,
  60                                       struct md_hsm *hsm)
  61 {
  62         struct md_attr          *ma;
  63         struct mdt_object       *obj;
  64         int                      rc;
  65         ENTRY;
  66
  67         ma = &mti->mti_attr;
  68         ma->ma_need = MA_HSM;
  69         ma->ma_valid = 0;
  70
  71         /* find object by FID */
  72         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  73         if (IS_ERR(obj))
  74                 RETURN(obj);
  75
  76         if (!mdt_object_exists(obj)) {
  77                 /* no more object */
  78                 mdt_object_put(mti->mti_env, obj);
  79                 RETURN(ERR_PTR(-ENOENT));
  80         }
  81
  82         rc = mdt_attr_get_complex(mti, obj, ma);
  83         if (rc) {
  84                 mdt_object_put(mti->mti_env, obj);
  85                 RETURN(ERR_PTR(rc));
  86         }
  87
  88         if (ma->ma_valid & MA_HSM)
  89                 *hsm = ma->ma_hsm;
  90         else
  91                 memset(hsm, 0, sizeof(*hsm));
  92         ma->ma_valid = 0;
  93         RETURN(obj);
  94 }
  95
  96 void mdt_hsm_dump_hal(int level, const char *prefix,
  97                       struct hsm_action_list *hal)
  98 {
  99         int                      i, sz;
 100         struct hsm_action_item  *hai;
 101         char                     buf[12];
 102
 103         CDEBUG(level, "%s: HAL header: version %X count %d compound "LPX64
 104                       " archive_id %d flags "LPX64"\n",
 105                prefix, hal->hal_version, hal->hal_count,
 106                hal->hal_compound_id, hal->hal_archive_id, hal->hal_flags);
 107
 108         hai = hai_first(hal);
 109         for (i = 0; i < hal->hal_count; i++) {
 110                 sz = hai->hai_len - sizeof(*hai);
 111                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 112                        " compound/cookie="LPX64"/"LPX64
 113                        " action=%s extent="LPX64"-"LPX64" gid="LPX64
 114                        " datalen=%d data=[%s]\n",
 115                        prefix, i,
 116                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 117                        hal->hal_compound_id, hai->hai_cookie,
 118                        hsm_copytool_action2name(hai->hai_action),
 119                        hai->hai_extent.offset,
 120                        hai->hai_extent.length,
 121                        hai->hai_gid, sz,
 122                        hai_dump_data_field(hai, buf, sizeof(buf)));
 123                 hai = hai_next(hai);
 124         }
 125 }
 126
 127 /**
 128  * data passed to llog_cat_process() callback
 129  * to scan requests and take actions
 130  */
 131 struct hsm_scan_data {
 132         struct mdt_thread_info          *mti;
 133         char                             fs_name[MTI_NAME_MAXLEN+1];
 134         /* request to be send to agents */
 135         int                              request_sz;    /** allocated size */
 136         int                              max_requests;  /** vector size */
 137         int                              request_cnt;   /** used count */
 138         struct {
 139                 int                      hal_sz;
 140                 int                      hal_used_sz;
 141                 struct hsm_action_list  *hal;
 142         } *request;
 143         /* records to be canceled */
 144         int                              max_cookie;    /** vector size */
 145         int                              cookie_cnt;    /** used count */
 146         __u64                           *cookies;
 147 };
 148
 149 /**
 150  *  llog_cat_process() callback, used to:
 151  *  - find waiting request and start action
 152  *  - purge canceled and done requests
 153  * \param env [IN] environment
 154  * \param llh [IN] llog handle
 155  * \param hdr [IN] llog record
 156  * \param data [IN/OUT] cb data = struct hsm_scan_data
 157  * \retval 0 success
 158  * \retval -ve failure
 159  */
 160 static int mdt_coordinator_cb(const struct lu_env *env,
 161                               struct llog_handle *llh,
 162                               struct llog_rec_hdr *hdr,
 163                               void *data)
 164 {
 165         const struct llog_agent_req_rec *larr;
 166         struct hsm_scan_data            *hsd;
 167         struct hsm_action_item          *hai;
 168         struct mdt_device               *mdt;
 169         struct coordinator              *cdt;
 170         int                              rc;
 171         ENTRY;
 172
 173         hsd = data;
 174         mdt = hsd->mti->mti_mdt;
 175         cdt = &mdt->mdt_coordinator;
 176
 177         larr = (struct llog_agent_req_rec *)hdr;
 178         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 179         switch (larr->arr_status) {
 180         case ARS_WAITING: {
 181                 int i, empty_slot, found;
 182
 183                 /* Are agents full? */
 184                 if (atomic_read(&cdt->cdt_request_count) ==
 185                     cdt->cdt_max_requests)
 186                         break;
 187
 188                 /* first search if the request if known in the list we have
 189                  * build and if there is room in the request vector */
 190                 empty_slot = -1;
 191                 found = -1;
 192                 for (i = 0; i < hsd->max_requests &&
 193                             (empty_slot == -1 || found == -1); i++) {
 194                         if (hsd->request[i].hal == NULL) {
 195                                 empty_slot = i;
 196                                 continue;
 197                         }
 198                         if (hsd->request[i].hal->hal_compound_id ==
 199                                 larr->arr_compound_id) {
 200                                 found = i;
 201                                 continue;
 202                         }
 203                 }
 204                 if (found == -1 && empty_slot == -1)
 205                         /* unknown request and no more room for new request,
 206                          * continue scan for to find other entries for
 207                          * already found request
 208                          */
 209                         RETURN(0);
 210
 211                 if (found == -1) {
 212                         struct hsm_action_list *hal;
 213
 214                         /* request is not already known */
 215                         /* allocates hai vector size just needs to be large
 216                          * enough */
 217                         hsd->request[empty_slot].hal_sz =
 218                                      sizeof(*hsd->request[empty_slot].hal) +
 219                                      cfs_size_round(MTI_NAME_MAXLEN+1) +
 220                                      2 * cfs_size_round(larr->arr_hai.hai_len);
 221                         OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
 222                         if (!hal) {
 223                                 CERROR("%s: Cannot allocate memory (%d o)"
 224                                        "for compound "LPX64"\n",
 225                                        mdt_obd_name(mdt),
 226                                        hsd->request[i].hal_sz,
 227                                        larr->arr_compound_id);
 228                                 RETURN(-ENOMEM);
 229                         }
 230                         hal->hal_version = HAL_VERSION;
 231                         strncpy(hal->hal_fsname, hsd->fs_name,
 232                                 MTI_NAME_MAXLEN);
 233                         hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
 234                         hal->hal_compound_id = larr->arr_compound_id;
 235                         hal->hal_archive_id = larr->arr_archive_id;
 236                         hal->hal_flags = larr->arr_flags;
 237                         hal->hal_count = 0;
 238                         hsd->request[empty_slot].hal_used_sz = hal_size(hal);
 239                         hsd->request[empty_slot].hal = hal;
 240                         hsd->request_cnt++;
 241                         found = empty_slot;
 242                         hai = hai_first(hal);
 243                 } else {
 244                         /* request is known */
 245                         /* we check if record archive num is the same as the
 246                          * known request, if not we will serve it in multiple
 247                          * time because we do not know if the agent can serve
 248                          * multiple backend
 249                          * a use case is a compound made of multiple restore
 250                          * where the files are not archived in the same backend
 251                          */
 252                         if (larr->arr_archive_id !=
 253                             hsd->request[found].hal->hal_archive_id)
 254                                 RETURN(0);
 255
 256                         if (hsd->request[found].hal_sz <
 257                             hsd->request[found].hal_used_sz +
 258                              cfs_size_round(larr->arr_hai.hai_len)) {
 259                                 /* Not enough room, need an extension */
 260                                 void *hal_buffer;
 261                                 int sz;
 262
 263                                 sz = 2 * hsd->request[found].hal_sz;
 264                                 OBD_ALLOC(hal_buffer, sz);
 265                                 if (!hal_buffer) {
 266                                         CERROR("%s: Cannot allocate memory "
 267                                                "(%d o) for compound "LPX64"\n",
 268                                                mdt_obd_name(mdt), sz,
 269                                                larr->arr_compound_id);
 270                                         RETURN(-ENOMEM);
 271                                 }
 272                                 memcpy(hal_buffer, hsd->request[found].hal,
 273                                        hsd->request[found].hal_used_sz);
 274                                 OBD_FREE(hsd->request[found].hal,
 275                                          hsd->request[found].hal_sz);
 276                                 hsd->request[found].hal = hal_buffer;
 277                                 hsd->request[found].hal_sz = sz;
 278                         }
 279                         hai = hai_first(hsd->request[found].hal);
 280                         for (i = 0; i < hsd->request[found].hal->hal_count;
 281                              i++)
 282                                 hai = hai_next(hai);
 283                 }
 284                 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 285                 hai->hai_cookie = larr->arr_hai.hai_cookie;
 286                 hai->hai_gid = larr->arr_hai.hai_gid;
 287
 288                 hsd->request[found].hal_used_sz +=
 289                                                    cfs_size_round(hai->hai_len);
 290                 hsd->request[found].hal->hal_count++;
 291                 break;
 292         }
 293         case ARS_STARTED: {
 294                 struct cdt_agent_req *car;
 295                 cfs_time_t last;
 296
 297                 /* we search for a running request
 298                  * error may happen if coordinator crashes or stopped
 299                  * with running request
 300                  */
 301                 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
 302                 if (car == NULL) {
 303                         last = larr->arr_req_create;
 304                 } else {
 305                         last = car->car_req_update;
 306                         mdt_cdt_put_request(car);
 307                 }
 308
 309                 /* test if request too long, if yes cancel it
 310                  * the same way the copy tool acknowledge a cancel request */
 311                 if ((last + cdt->cdt_active_req_timeout)
 312                      < cfs_time_current_sec()) {
 313                         struct hsm_progress_kernel pgs;
 314
 315                         dump_llog_agent_req_rec("mdt_coordinator_cb(): "
 316                                                 "request timeouted, start "
 317                                                 "cleaning", larr);
 318                         /* a too old cancel request just needs to be removed
 319                          * this can happen, if copy tool does not support cancel
 320                          * for other requests, we have to remove the running
 321                          * request and notify the copytool
 322                          */
 323                         pgs.hpk_fid = larr->arr_hai.hai_fid;
 324                         pgs.hpk_cookie = larr->arr_hai.hai_cookie;
 325                         pgs.hpk_extent = larr->arr_hai.hai_extent;
 326                         pgs.hpk_flags = HP_FLAG_COMPLETED;
 327                         pgs.hpk_errval = ENOSYS;
 328                         pgs.hpk_data_version = 0;
 329                         /* update request state, but do not record in llog, to
 330                          * avoid deadlock on cdt_llog_lock
 331                          */
 332                         rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
 333                         if (rc)
 334                                 CERROR("%s: Cannot cleanup timeouted request: "
 335                                        DFID" for cookie "LPX64" action=%s\n",
 336                                        mdt_obd_name(mdt),
 337                                        PFID(&pgs.hpk_fid), pgs.hpk_cookie,
 338                                        hsm_copytool_action2name(
 339                                                      larr->arr_hai.hai_action));
 340
 341                         /* add the cookie to the list of record to be
 342                          * canceled by caller */
 343                         if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
 344                                 __u64 *ptr, *old_ptr;
 345                                 int old_sz, new_sz, new_cnt;
 346
 347                                 /* need to increase vector size */
 348                                 old_sz = sizeof(__u64) * hsd->max_cookie;
 349                                 old_ptr = hsd->cookies;
 350
 351                                 new_cnt = 2 * hsd->max_cookie;
 352                                 new_sz = sizeof(__u64) * new_cnt;
 353
 354                                 OBD_ALLOC(ptr, new_sz);
 355                                 if (!ptr) {
 356                                         CERROR("%s: Cannot allocate memory "
 357                                                "(%d o) for cookie vector\n",
 358                                                mdt_obd_name(mdt), new_sz);
 359                                         RETURN(-ENOMEM);
 360                                 }
 361                                 memcpy(ptr, hsd->cookies, old_sz);
 362                                 hsd->cookies = ptr;
 363                                 hsd->max_cookie = new_cnt;
 364                                 OBD_FREE(old_ptr, old_sz);
 365                         }
 366                         hsd->cookies[hsd->cookie_cnt] =
 367                                                        larr->arr_hai.hai_cookie;
 368                         hsd->cookie_cnt++;
 369                 }
 370                 break;
 371         }
 372         case ARS_FAILED:
 373         case ARS_CANCELED:
 374         case ARS_SUCCEED:
 375                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 376                     cfs_time_current_sec())
 377                         RETURN(LLOG_DEL_RECORD);
 378                 break;
 379         }
 380         RETURN(0);
 381 }
 382
 383 /**
 384  * create /proc entries for coordinator
 385  * \param mdt [IN]
 386  * \retval 0 success
 387  * \retval -ve failure
 388  */
 389 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 390 {
 391         struct coordinator      *cdt = &mdt->mdt_coordinator;
 392         int                      rc = 0;
 393         ENTRY;
 394
 395         /* init /proc entries, failure is not critical */
 396         cdt->cdt_proc_dir = lprocfs_register("hsm",
 397                                              mdt2obd_dev(mdt)->obd_proc_entry,
 398                                              lprocfs_mdt_hsm_vars, mdt);
 399         if (IS_ERR(cdt->cdt_proc_dir)) {
 400                 rc = PTR_ERR(cdt->cdt_proc_dir);
 401                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 402                        " rc=%d\n", mdt_obd_name(mdt), rc);
 403                 cdt->cdt_proc_dir = NULL;
 404                 RETURN(rc);
 405         }
 406
 407         RETURN(0);
 408 }
 409
 410 /**
 411  * remove /proc entries for coordinator
 412  * \param mdt [IN]
 413  */
 414 void  hsm_cdt_procfs_fini(struct mdt_device *mdt)
 415 {
 416         struct coordinator      *cdt = &mdt->mdt_coordinator;
 417
 418         LASSERT(cdt->cdt_state == CDT_STOPPED);
 419         if (cdt->cdt_proc_dir != NULL)
 420                 lprocfs_remove(&cdt->cdt_proc_dir);
 421 }
 422
 423 /**
 424  * get vector of hsm cdt /proc vars
 425  * \param none
 426  * \retval var vector
 427  */
 428 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 429 {
 430         return lprocfs_mdt_hsm_vars;
 431 }
 432
 433 /**
 434  * coordinator thread
 435  * \param data [IN] obd device
 436  * \retval 0 success
 437  * \retval -ve failure
 438  */
 439 static int mdt_coordinator(void *data)
 440 {
 441         struct mdt_thread_info  *mti = data;
 442         struct mdt_device       *mdt = mti->mti_mdt;
 443         struct coordinator      *cdt = &mdt->mdt_coordinator;
 444         struct hsm_scan_data     hsd = { 0 };
 445         int                      rc = 0;
 446         ENTRY;
 447
 448         cdt->cdt_thread.t_flags = SVC_RUNNING;
 449         wake_up(&cdt->cdt_thread.t_ctl_waitq);
 450
 451         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 452                mdt_obd_name(mdt), current_pid());
 453
 454         /* timeouted cookie vector initialization */
 455         hsd.max_cookie = 0;
 456         hsd.cookie_cnt = 0;
 457         hsd.cookies = NULL;
 458         /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
 459          * increases due to a change from /proc we do not overflow the
 460          * hsd.request[] vector
 461          */
 462         hsd.max_requests = cdt->cdt_max_requests;
 463         hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
 464         OBD_ALLOC(hsd.request, hsd.request_sz);
 465         if (!hsd.request)
 466                 GOTO(out, rc = -ENOMEM);
 467
 468         hsd.mti = mti;
 469         obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
 470
 471         while (1) {
 472                 struct l_wait_info lwi;
 473                 int i;
 474
 475                 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
 476                                   NULL, NULL);
 477                 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
 478                              (cdt->cdt_thread.t_flags &
 479                               (SVC_STOPPING|SVC_EVENT)),
 480                              &lwi);
 481
 482                 CDEBUG(D_HSM, "coordinator resumes\n");
 483
 484                 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
 485                     cdt->cdt_state == CDT_STOPPING) {
 486                         cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
 487                         rc = 0;
 488                         break;
 489                 }
 490
 491                 /* wake up before timeout, new work arrives */
 492                 if (cdt->cdt_thread.t_flags & SVC_EVENT)
 493                         cdt->cdt_thread.t_flags &= ~SVC_EVENT;
 494
 495                 /* if coordinator is suspended continue to wait */
 496                 if (cdt->cdt_state == CDT_DISABLE) {
 497                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 498                         continue;
 499                 }
 500
 501                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 502
 503                 if (hsd.max_requests != cdt->cdt_max_requests) {
 504                         /* cdt_max_requests has changed,
 505                          * we need to allocate a new buffer
 506                          */
 507                         OBD_FREE(hsd.request, hsd.request_sz);
 508                         hsd.max_requests = cdt->cdt_max_requests;
 509                         hsd.request_sz =
 510                                    hsd.max_requests * sizeof(*hsd.request);
 511                         OBD_ALLOC(hsd.request, hsd.request_sz);
 512                         if (!hsd.request) {
 513                                 rc = -ENOMEM;
 514                                 break;
 515                         }
 516                 }
 517
 518                 /* create canceled cookie vector for an arbitrary size
 519                  * if needed, vector will grow during llog scan
 520                  */
 521                 hsd.max_cookie = 10;
 522                 hsd.cookie_cnt = 0;
 523                 OBD_ALLOC(hsd.cookies, hsd.max_cookie * sizeof(__u64));
 524                 if (!hsd.cookies) {
 525                         rc = -ENOMEM;
 526                         goto clean_cb_alloc;
 527                 }
 528                 hsd.request_cnt = 0;
 529
 530                 rc = cdt_llog_process(mti->mti_env, mdt,
 531                                       mdt_coordinator_cb, &hsd);
 532                 if (rc < 0)
 533                         goto clean_cb_alloc;
 534
 535                 CDEBUG(D_HSM, "Found %d requests to send and %d"
 536                               " requests to cancel\n",
 537                        hsd.request_cnt, hsd.cookie_cnt);
 538                 /* first we cancel llog records of the timeouted requests */
 539                 if (hsd.cookie_cnt > 0) {
 540                         rc = mdt_agent_record_update(mti->mti_env, mdt,
 541                                                      hsd.cookies,
 542                                                      hsd.cookie_cnt,
 543                                                      ARS_CANCELED);
 544                         if (rc)
 545                                 CERROR("%s: mdt_agent_record_update() failed, "
 546                                        "rc=%d, cannot update status to %s "
 547                                        "for %d cookies\n",
 548                                        mdt_obd_name(mdt), rc,
 549                                        agent_req_status2name(ARS_CANCELED),
 550                                        hsd.cookie_cnt);
 551                 }
 552
 553                 if (list_empty(&cdt->cdt_agents)) {
 554                         CDEBUG(D_HSM, "no agent available, "
 555                                       "coordinator sleeps\n");
 556                         goto clean_cb_alloc;
 557                 }
 558
 559                 /* here hsd contains a list of requests to be started */
 560                 for (i = 0; i < hsd.max_requests; i++) {
 561                         struct hsm_action_list  *hal;
 562                         struct hsm_action_item  *hai;
 563                         __u64                   *cookies;
 564                         int                      sz, j;
 565                         enum agent_req_status    status;
 566
 567                         /* still room for work ? */
 568                         if (atomic_read(&cdt->cdt_request_count) ==
 569                             cdt->cdt_max_requests)
 570                                 break;
 571
 572                         if (hsd.request[i].hal == NULL)
 573                                 continue;
 574
 575                         /* found a request, we start it */
 576                         /* kuc payload allocation so we avoid an additionnal
 577                          * allocation in mdt_hsm_agent_send()
 578                          */
 579                         hal = kuc_alloc(hsd.request[i].hal_used_sz,
 580                                         KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
 581                         if (IS_ERR(hal)) {
 582                                 CERROR("%s: Cannot allocate memory (%d o) "
 583                                        "for compound "LPX64"\n",
 584                                        mdt_obd_name(mdt),
 585                                        hsd.request[i].hal_used_sz,
 586                                        hsd.request[i].hal->hal_compound_id);
 587                                 continue;
 588                         }
 589                         memcpy(hal, hsd.request[i].hal,
 590                                hsd.request[i].hal_used_sz);
 591
 592                         rc = mdt_hsm_agent_send(mti, hal, 0);
 593                         /* if failure, we suppose it is temporary
 594                          * if the copy tool failed to do the request
 595                          * it has to use hsm_progress
 596                          */
 597                         status = (rc ? ARS_WAITING : ARS_STARTED);
 598
 599                         /* set up cookie vector to set records status
 600                          * after copy tools start or failed
 601                          */
 602                         sz = hsd.request[i].hal->hal_count * sizeof(__u64);
 603                         OBD_ALLOC(cookies, sz);
 604                         if (cookies == NULL) {
 605                                 CERROR("%s: Cannot allocate memory (%d o) "
 606                                        "for cookies vector "LPX64"\n",
 607                                        mdt_obd_name(mdt), sz,
 608                                        hsd.request[i].hal->hal_compound_id);
 609                                 kuc_free(hal, hsd.request[i].hal_used_sz);
 610                                 continue;
 611                         }
 612                         hai = hai_first(hal);
 613                         for (j = 0; j < hsd.request[i].hal->hal_count; j++) {
 614                                 cookies[j] = hai->hai_cookie;
 615                                 hai = hai_next(hai);
 616                         }
 617
 618                         rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
 619                                                 hsd.request[i].hal->hal_count,
 620                                                 status);
 621                         if (rc)
 622                                 CERROR("%s: mdt_agent_record_update() failed, "
 623                                        "rc=%d, cannot update status to %s "
 624                                        "for %d cookies\n",
 625                                        mdt_obd_name(mdt), rc,
 626                                        agent_req_status2name(status),
 627                                        hsd.request[i].hal->hal_count);
 628
 629                         OBD_FREE(cookies, sz);
 630                         kuc_free(hal, hsd.request[i].hal_used_sz);
 631                 }
 632 clean_cb_alloc:
 633                 /* free cookie vector allocated for/by callback */
 634                 if (hsd.cookies) {
 635                         OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
 636                         hsd.max_cookie = 0;
 637                         hsd.cookie_cnt = 0;
 638                         hsd.cookies = NULL;
 639                 }
 640
 641                 /* free hal allocated by callback */
 642                 for (i = 0; i < hsd.max_requests; i++) {
 643                         if (hsd.request[i].hal) {
 644                                 OBD_FREE(hsd.request[i].hal,
 645                                          hsd.request[i].hal_sz);
 646                                 hsd.request[i].hal_sz = 0;
 647                                 hsd.request[i].hal = NULL;
 648                                 hsd.request_cnt--;
 649                         }
 650                 }
 651                 LASSERT(hsd.request_cnt == 0);
 652
 653                 /* reset callback data */
 654                 memset(hsd.request, 0, hsd.request_sz);
 655         }
 656         EXIT;
 657 out:
 658         if (hsd.request)
 659                 OBD_FREE(hsd.request, hsd.request_sz);
 660
 661         if (hsd.cookies)
 662                 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
 663
 664         if (cdt->cdt_state == CDT_STOPPING) {
 665                 /* request comes from /proc path, so we need to clean cdt
 666                  * struct */
 667                  mdt_hsm_cdt_stop(mdt);
 668                  mdt->mdt_opts.mo_coordinator = 0;
 669         } else {
 670                 /* request comes from a thread event, generated
 671                  * by mdt_stop_coordinator(), we have to ack
 672                  * and cdt cleaning will be done by event sender
 673                  */
 674                 cdt->cdt_thread.t_flags = SVC_STOPPED;
 675                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
 676         }
 677
 678         if (rc != 0)
 679                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 680                        mdt_obd_name(mdt), current_pid(), rc);
 681         else
 682                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 683                               " no error\n",
 684                        mdt_obd_name(mdt), current_pid());
 685
 686         return rc;
 687 }
 688
 689 /**
 690  * lookup a restore handle by FID
 691  * caller needs to hold cdt_restore_lock
 692  * \param cdt [IN] coordinator
 693  * \param fid [IN] FID
 694  * \retval cdt_restore_handle found
 695  * \retval NULL not found
 696  */
 697 static struct cdt_restore_handle *hsm_restore_hdl_find(struct coordinator *cdt,
 698                                                        const struct lu_fid *fid)
 699 {
 700         struct cdt_restore_handle       *crh;
 701         ENTRY;
 702
 703         list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
 704                 if (lu_fid_eq(&crh->crh_fid, fid))
 705                         RETURN(crh);
 706         }
 707         RETURN(NULL);
 708 }
 709
 710 /**
 711  * data passed to llog_cat_process() callback
 712  * to scan requests and take actions
 713  */
 714 struct hsm_restore_data {
 715         struct mdt_thread_info  *hrd_mti;
 716 };
 717
 718 /**
 719  *  llog_cat_process() callback, used to:
 720  *  - find restore request and allocate the restore handle
 721  * \param env [IN] environment
 722  * \param llh [IN] llog handle
 723  * \param hdr [IN] llog record
 724  * \param data [IN/OUT] cb data = struct hsm_restore_data
 725  * \retval 0 success
 726  * \retval -ve failure
 727  */
 728 static int hsm_restore_cb(const struct lu_env *env,
 729                           struct llog_handle *llh,
 730                           struct llog_rec_hdr *hdr, void *data)
 731 {
 732         struct llog_agent_req_rec       *larr;
 733         struct hsm_restore_data         *hrd;
 734         struct cdt_restore_handle       *crh;
 735         struct hsm_action_item          *hai;
 736         struct mdt_thread_info          *mti;
 737         struct coordinator              *cdt;
 738         struct mdt_object               *child;
 739         int rc;
 740         ENTRY;
 741
 742         hrd = data;
 743         mti = hrd->hrd_mti;
 744         cdt = &mti->mti_mdt->mdt_coordinator;
 745
 746         larr = (struct llog_agent_req_rec *)hdr;
 747         hai = &larr->arr_hai;
 748         if (hai->hai_cookie > cdt->cdt_last_cookie)
 749                 /* update the cookie to avoid collision */
 750                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 751
 752         if (hai->hai_action != HSMA_RESTORE ||
 753             agent_req_in_final_state(larr->arr_status))
 754                 RETURN(0);
 755
 756         /* restore request not in a final state */
 757
 758         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 759         if (crh == NULL)
 760                 RETURN(-ENOMEM);
 761
 762         crh->crh_fid = hai->hai_fid;
 763         /* in V1 all file is restored
 764         crh->extent.start = hai->hai_extent.offset;
 765         crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
 766         */
 767         crh->crh_extent.start = 0;
 768         crh->crh_extent.end = hai->hai_extent.length;
 769         /* get the layout lock */
 770         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 771         child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 772                                      MDS_INODELOCK_LAYOUT);
 773         if (IS_ERR(child))
 774                 GOTO(out, rc = PTR_ERR(child));
 775
 776         rc = 0;
 777         /* we choose to not keep a reference
 778          * on the object during the restore time which can be very long */
 779         mdt_object_put(mti->mti_env, child);
 780
 781         mutex_lock(&cdt->cdt_restore_lock);
 782         list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
 783         mutex_unlock(&cdt->cdt_restore_lock);
 784
 785 out:
 786         RETURN(rc);
 787 }
 788
 789 /**
 790  * restore coordinator state at startup
 791  * the goal is to take a layout lock for each registered restore request
 792  * \param mti [IN] context
 793  */
 794 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 795 {
 796         struct hsm_restore_data  hrd;
 797         int                      rc;
 798         ENTRY;
 799
 800         hrd.hrd_mti = mti;
 801
 802         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
 803                               hsm_restore_cb, &hrd);
 804
 805         RETURN(rc);
 806 }
 807
 808 static int hsm_init_ucred(struct lu_ucred *uc)
 809 {
 810         ENTRY;
 811
 812         uc->uc_valid = UCRED_OLD;
 813         uc->uc_o_uid = 0;
 814         uc->uc_o_gid = 0;
 815         uc->uc_o_fsuid = 0;
 816         uc->uc_o_fsgid = 0;
 817         uc->uc_uid = 0;
 818         uc->uc_gid = 0;
 819         uc->uc_fsuid = 0;
 820         uc->uc_fsgid = 0;
 821         uc->uc_suppgids[0] = -1;
 822         uc->uc_suppgids[1] = -1;
 823         uc->uc_cap = CFS_CAP_FS_MASK;
 824         uc->uc_umask = 0777;
 825         uc->uc_ginfo = NULL;
 826         uc->uc_identity = NULL;
 827
 828         RETURN(0);
 829 }
 830
 831 /**
 832  * wake up coordinator thread
 833  * \param mdt [IN] device
 834  * \retval 0 success
 835  * \retval -ve failure
 836  */
 837 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
 838 {
 839         struct coordinator      *cdt = &mdt->mdt_coordinator;
 840         ENTRY;
 841
 842         if (cdt->cdt_state == CDT_STOPPED)
 843                 RETURN(-ESRCH);
 844
 845         /* wake up coordinator */
 846         cdt->cdt_thread.t_flags = SVC_EVENT;
 847         wake_up(&cdt->cdt_thread.t_ctl_waitq);
 848
 849         RETURN(0);
 850 }
 851
 852 /**
 853  * initialize coordinator struct
 854  * \param mdt [IN] device
 855  * \retval 0 success
 856  * \retval -ve failure
 857  */
 858 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 859 {
 860         struct coordinator      *cdt = &mdt->mdt_coordinator;
 861         struct mdt_thread_info  *cdt_mti = NULL;
 862         int                      rc;
 863         ENTRY;
 864
 865         cdt->cdt_state = CDT_STOPPED;
 866
 867         init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
 868         mutex_init(&cdt->cdt_llog_lock);
 869         init_rwsem(&cdt->cdt_agent_lock);
 870         init_rwsem(&cdt->cdt_request_lock);
 871         mutex_init(&cdt->cdt_restore_lock);
 872
 873         CFS_INIT_LIST_HEAD(&cdt->cdt_requests);
 874         CFS_INIT_LIST_HEAD(&cdt->cdt_agents);
 875         CFS_INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 876
 877         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 878         if (rc < 0)
 879                 RETURN(rc);
 880
 881         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
 882         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
 883         if (rc == 0) {
 884                 lu_context_enter(&cdt->cdt_session);
 885                 cdt->cdt_env.le_ses = &cdt->cdt_session;
 886         } else {
 887                 lu_env_fini(&cdt->cdt_env);
 888                 RETURN(rc);
 889         }
 890
 891         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 892         LASSERT(cdt_mti != NULL);
 893
 894         cdt_mti->mti_env = &cdt->cdt_env;
 895         cdt_mti->mti_mdt = mdt;
 896
 897         hsm_init_ucred(mdt_ucred(cdt_mti));
 898
 899         /* default values for /proc tunnables
 900          * can be override by MGS conf */
 901         cdt->cdt_default_archive_id = 1;
 902         cdt->cdt_grace_delay = 60;
 903         cdt->cdt_loop_period = 10;
 904         cdt->cdt_max_requests = 3;
 905         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 906         cdt->cdt_active_req_timeout = 3600;
 907
 908         RETURN(0);
 909 }
 910
 911 /**
 912  * free a coordinator thread
 913  * \param mdt [IN] device
 914  */
 915 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
 916 {
 917         struct coordinator *cdt = &mdt->mdt_coordinator;
 918         ENTRY;
 919
 920         lu_context_exit(cdt->cdt_env.le_ses);
 921         lu_context_fini(cdt->cdt_env.le_ses);
 922
 923         lu_env_fini(&cdt->cdt_env);
 924
 925         RETURN(0);
 926 }
 927
 928 /**
 929  * start a coordinator thread
 930  * \param mdt [IN] device
 931  * \retval 0 success
 932  * \retval -ve failure
 933  */
 934 int mdt_hsm_cdt_start(struct mdt_device *mdt)
 935 {
 936         struct coordinator      *cdt = &mdt->mdt_coordinator;
 937         int                      rc;
 938         void                    *ptr;
 939         struct mdt_thread_info  *cdt_mti;
 940         struct task_struct      *task;
 941         ENTRY;
 942
 943         /* functions defined but not yet used
 944          * this avoid compilation warning
 945          */
 946         ptr = dump_requests;
 947
 948         if (cdt->cdt_state != CDT_STOPPED) {
 949                 CERROR("%s: Coordinator already started\n",
 950                        mdt_obd_name(mdt));
 951                 RETURN(-EALREADY);
 952         }
 953
 954         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
 955         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 956
 957         cdt->cdt_state = CDT_INIT;
 958
 959         atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
 960         /* just need to be larger than previous one */
 961         /* cdt_last_cookie is protected by cdt_llog_lock */
 962         cdt->cdt_last_cookie = cfs_time_current_sec();
 963         atomic_set(&cdt->cdt_request_count, 0);
 964         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
 965         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
 966         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
 967
 968         /* to avoid deadlock when start is made through /proc
 969          * /proc entries are created by the coordinator thread */
 970
 971         /* set up list of started restore requests */
 972         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 973         rc = mdt_hsm_pending_restore(cdt_mti);
 974         if (rc)
 975                 CERROR("%s: cannot take the layout locks needed"
 976                        " for registered restore: %d",
 977                        mdt_obd_name(mdt), rc);
 978
 979         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
 980         if (IS_ERR(task)) {
 981                 rc = PTR_ERR(task);
 982                 cdt->cdt_state = CDT_STOPPED;
 983                 CERROR("%s: error starting coordinator thread: %d\n",
 984                        mdt_obd_name(mdt), rc);
 985                 RETURN(rc);
 986         } else {
 987                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
 988                        mdt_obd_name(mdt));
 989                 rc = 0;
 990         }
 991
 992         wait_event(cdt->cdt_thread.t_ctl_waitq,
 993                        (cdt->cdt_thread.t_flags & SVC_RUNNING));
 994
 995         cdt->cdt_state = CDT_RUNNING;
 996         mdt->mdt_opts.mo_coordinator = 1;
 997         RETURN(0);
 998 }
 999
1000 /**
1001  * stop a coordinator thread
1002  * \param mdt [IN] device
1003  */
1004 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1005 {
1006         struct coordinator              *cdt = &mdt->mdt_coordinator;
1007         struct cdt_agent_req            *car, *tmp1;
1008         struct hsm_agent                *ha, *tmp2;
1009         struct cdt_restore_handle       *crh, *tmp3;
1010         struct mdt_thread_info          *cdt_mti;
1011         ENTRY;
1012
1013         if (cdt->cdt_state == CDT_STOPPED) {
1014                 CERROR("%s: Coordinator already stopped\n",
1015                        mdt_obd_name(mdt));
1016                 RETURN(-EALREADY);
1017         }
1018
1019         if (cdt->cdt_state != CDT_STOPPING) {
1020                 /* stop coordinator thread before cleaning */
1021                 cdt->cdt_thread.t_flags = SVC_STOPPING;
1022                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
1023                 wait_event(cdt->cdt_thread.t_ctl_waitq,
1024                            cdt->cdt_thread.t_flags & SVC_STOPPED);
1025         }
1026         cdt->cdt_state = CDT_STOPPED;
1027
1028         /* start cleaning */
1029         down_write(&cdt->cdt_request_lock);
1030         list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
1031                                  car_request_list) {
1032                 list_del(&car->car_request_list);
1033                 mdt_cdt_free_request(car);
1034         }
1035         up_write(&cdt->cdt_request_lock);
1036
1037         down_write(&cdt->cdt_agent_lock);
1038         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
1039                 list_del(&ha->ha_list);
1040                 OBD_FREE_PTR(ha);
1041         }
1042         up_write(&cdt->cdt_agent_lock);
1043
1044         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1045         mutex_lock(&cdt->cdt_restore_lock);
1046         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
1047                 struct mdt_object       *child;
1048
1049                 /* give back layout lock */
1050                 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
1051                 if (!IS_ERR(child))
1052                         mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
1053
1054                 list_del(&crh->crh_list);
1055
1056                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1057         }
1058         mutex_unlock(&cdt->cdt_restore_lock);
1059
1060         mdt->mdt_opts.mo_coordinator = 0;
1061
1062         RETURN(0);
1063 }
1064
1065 /**
1066  * register all requests from an hal in the memory list
1067  * \param mti [IN] context
1068  * \param hal [IN] request
1069  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1070  *  which is running the CT
1071  * \retval 0 success
1072  * \retval -ve failure
1073  */
1074 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1075                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1076 {
1077         struct mdt_device       *mdt = mti->mti_mdt;
1078         struct coordinator      *cdt = &mdt->mdt_coordinator;
1079         struct hsm_action_item  *hai;
1080         int                      rc = 0, i;
1081         ENTRY;
1082
1083         /* register request in memory list */
1084         hai = hai_first(hal);
1085         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1086                 struct cdt_agent_req *car;
1087
1088                 /* in case of a cancel request, we first mark the ondisk
1089                  * record of the request we want to stop as canceled
1090                  * this does not change the cancel record
1091                  * it will be done when updating the request status
1092                  */
1093                 if (hai->hai_action == HSMA_CANCEL) {
1094                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1095                                                      &hai->hai_cookie,
1096                                                      1, ARS_CANCELED);
1097                         if (rc) {
1098                                 CERROR("%s: mdt_agent_record_update() failed, "
1099                                        "rc=%d, cannot update status to %s "
1100                                        "for cookie "LPX64"\n",
1101                                        mdt_obd_name(mdt), rc,
1102                                        agent_req_status2name(ARS_CANCELED),
1103                                        hai->hai_cookie);
1104                                 GOTO(out, rc);
1105                         }
1106
1107                         /* find the running request to set it canceled */
1108                         car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1109                         if (car != NULL) {
1110                                 car->car_canceled = 1;
1111                                 /* uuid has to be changed to the one running the
1112                                 * request to cancel */
1113                                 *uuid = car->car_uuid;
1114                                 mdt_cdt_put_request(car);
1115                         }
1116                         /* no need to memorize cancel request
1117                          * this also avoid a deadlock when we receive
1118                          * a purge all requests command
1119                          */
1120                         continue;
1121                 }
1122
1123                 if (hai->hai_action == HSMA_ARCHIVE) {
1124                         struct mdt_object *obj;
1125                         struct md_hsm hsm;
1126
1127                         obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1128                         if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1129                                 continue;
1130                         if (IS_ERR(obj))
1131                                 GOTO(out, rc = PTR_ERR(obj));
1132
1133                         hsm.mh_flags |= HS_EXISTS;
1134                         hsm.mh_arch_id = hal->hal_archive_id;
1135                         rc = mdt_hsm_attr_set(mti, obj, &hsm);
1136                         mdt_object_put(mti->mti_env, obj);
1137                         if (rc)
1138                                 GOTO(out, rc);
1139                 }
1140
1141                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1142                                             hal->hal_archive_id, hal->hal_flags,
1143                                             uuid, hai);
1144                 if (IS_ERR(car))
1145                         GOTO(out, rc = PTR_ERR(car));
1146
1147                 rc = mdt_cdt_add_request(cdt, car);
1148                 if (rc != 0)
1149                         mdt_cdt_free_request(car);
1150         }
1151 out:
1152         RETURN(rc);
1153 }
1154
1155 /**
1156  * swap layouts between 2 fids
1157  * \param mti [IN] context
1158  * \param fid1 [IN]
1159  * \param fid2 [IN]
1160  */
1161 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1162                             const lustre_fid *fid, const lustre_fid *dfid)
1163 {
1164         struct mdt_device       *mdt = mti->mti_mdt;
1165         struct mdt_object       *child1, *child2;
1166         struct mdt_lock_handle  *lh2;
1167         int                      rc;
1168         ENTRY;
1169
1170         child1 = mdt_object_find(mti->mti_env, mdt, fid);
1171         if (IS_ERR(child1))
1172                 GOTO(out, rc = PTR_ERR(child1));
1173
1174         /* we already have layout lock on FID so take only
1175          * on dfid */
1176         lh2 = &mti->mti_lh[MDT_LH_OLD];
1177         mdt_lock_reg_init(lh2, LCK_EX);
1178         child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1179         if (IS_ERR(child2))
1180                 GOTO(out_child1, rc = PTR_ERR(child2));
1181
1182         /* if copy tool closes the volatile before sending the final
1183          * progress through llapi_hsm_copy_end(), all the objects
1184          * are removed and mdd_swap_layout LBUG */
1185         if (mdt_object_exists(child2)) {
1186                 rc = mo_swap_layouts(mti->mti_env, mdt_object_child(child1),
1187                                      mdt_object_child(child2), 0);
1188         } else {
1189                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1190                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1191                 rc = -ENOENT;
1192         }
1193
1194         mdt_object_unlock_put(mti, child2, lh2, 1);
1195 out_child1:
1196         mdt_object_put(mti->mti_env, child1);
1197 out:
1198         RETURN(rc);
1199 }
1200
1201 /**
1202  * update status of a completed request
1203  * \param mti [IN] context
1204  * \param pgs [IN] progress of the copy tool
1205  * \param update_record [IN] update llog record
1206  * \retval 0 success
1207  * \retval -ve failure
1208  */
1209 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1210                                      struct hsm_progress_kernel *pgs,
1211                                      const struct cdt_agent_req *car,
1212                                      enum agent_req_status *status)
1213 {
1214         const struct lu_env     *env = mti->mti_env;
1215         struct mdt_device       *mdt = mti->mti_mdt;
1216         struct coordinator      *cdt = &mdt->mdt_coordinator;
1217         struct mdt_object       *obj = NULL;
1218         int                      cl_flags = 0, rc = 0;
1219         struct md_hsm            mh;
1220         bool                     is_mh_changed;
1221         ENTRY;
1222
1223         /* default is to retry */
1224         *status = ARS_WAITING;
1225
1226         /* find object by FID */
1227         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1228         /* we will update MD HSM only if needed */
1229         is_mh_changed = false;
1230         if (IS_ERR(obj)) {
1231                 /* object removed */
1232                 *status = ARS_SUCCEED;
1233                 goto unlock;
1234         }
1235
1236         /* no need to change mh->mh_arch_id
1237          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1238          */
1239         if (pgs->hpk_errval != 0) {
1240                 switch (pgs->hpk_errval) {
1241                 case ENOSYS:
1242                         /* the copy tool does not support cancel
1243                          * so the cancel request is failed
1244                          * As we cannot distinguish a cancel progress
1245                          * from another action progress (they have the
1246                          * same cookie), we suppose here the CT returns
1247                          * ENOSYS only if does not support cancel
1248                          */
1249                         /* this can also happen when cdt calls it to
1250                          * for a timeouted request */
1251                         *status = ARS_FAILED;
1252                         /* to have a cancel event in changelog */
1253                         pgs->hpk_errval = ECANCELED;
1254                         break;
1255                 case ECANCELED:
1256                         /* the request record has already been set to
1257                          * ARS_CANCELED, this set the cancel request
1258                          * to ARS_SUCCEED */
1259                         *status = ARS_SUCCEED;
1260                         break;
1261                 default:
1262                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1263                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1264                                    ARS_FAILED : ARS_WAITING);
1265                         break;
1266                 }
1267
1268                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1269                         CERROR("%s: Request "LPX64" on "DFID
1270                                " failed, error code %d too large\n",
1271                                mdt_obd_name(mdt),
1272                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1273                                pgs->hpk_errval);
1274                         hsm_set_cl_error(&cl_flags,
1275                                          CLF_HSM_ERROVERFLOW);
1276                         rc = -EINVAL;
1277                 } else {
1278                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1279                 }
1280
1281                 switch (car->car_hai->hai_action) {
1282                 case HSMA_ARCHIVE:
1283                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1284                         break;
1285                 case HSMA_RESTORE:
1286                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1287                         break;
1288                 case HSMA_REMOVE:
1289                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1290                         break;
1291                 case HSMA_CANCEL:
1292                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1293                         CERROR("%s: Failed request "LPX64" on "DFID
1294                                " cannot be a CANCEL\n",
1295                                mdt_obd_name(mdt),
1296                                pgs->hpk_cookie,
1297                                PFID(&pgs->hpk_fid));
1298                         break;
1299                 default:
1300                         CERROR("%s: Failed request "LPX64" on "DFID
1301                                " %d is an unknown action\n",
1302                                mdt_obd_name(mdt),
1303                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1304                                car->car_hai->hai_action);
1305                         rc = -EINVAL;
1306                         break;
1307                 }
1308         } else {
1309                 *status = ARS_SUCCEED;
1310                 switch (car->car_hai->hai_action) {
1311                 case HSMA_ARCHIVE:
1312                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1313                         /* set ARCHIVE keep EXIST and clear LOST and
1314                          * DIRTY */
1315                         mh.mh_arch_ver = pgs->hpk_data_version;
1316                         mh.mh_flags |= HS_ARCHIVED;
1317                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1318                         is_mh_changed = true;
1319                         break;
1320                 case HSMA_RESTORE:
1321                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1322
1323                         /* clear RELEASED and DIRTY */
1324                         mh.mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1325                         /* Restoring has changed the file version on
1326                          * disk. */
1327                         mh.mh_arch_ver = pgs->hpk_data_version;
1328                         is_mh_changed = true;
1329                         break;
1330                 case HSMA_REMOVE:
1331                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1332                         /* clear ARCHIVED EXISTS and LOST */
1333                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1334                         is_mh_changed = true;
1335                         break;
1336                 case HSMA_CANCEL:
1337                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1338                         CERROR("%s: Successful request "LPX64
1339                                " on "DFID
1340                                " cannot be a CANCEL\n",
1341                                mdt_obd_name(mdt),
1342                                pgs->hpk_cookie,
1343                                PFID(&pgs->hpk_fid));
1344                         break;
1345                 default:
1346                         CERROR("%s: Successful request "LPX64
1347                                " on "DFID
1348                                " %d is an unknown action\n",
1349                                mdt_obd_name(mdt),
1350                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1351                                car->car_hai->hai_action);
1352                         rc = -EINVAL;
1353                         break;
1354                 }
1355         }
1356
1357         /* rc != 0 means error when analysing action, it may come from
1358          * a crasy CT no need to manage DIRTY
1359          */
1360         if (rc == 0)
1361                 hsm_set_cl_flags(&cl_flags,
1362                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1363
1364         /* unlock is done later, after layout lock management */
1365         if (is_mh_changed)
1366                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1367
1368 unlock:
1369         /* we give back layout lock only if restore was successful or
1370          * if restore was canceled or if policy is to not retry
1371          * in other cases we just unlock the object */
1372         if (car->car_hai->hai_action == HSMA_RESTORE &&
1373             (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1374              cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1375                 struct cdt_restore_handle       *crh;
1376
1377                 /* restore in data FID done, we swap the layouts
1378                  * only if restore is successfull */
1379                 if (pgs->hpk_errval == 0) {
1380                         rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1381                                               &car->car_hai->hai_dfid);
1382                         if (rc) {
1383                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1384                                         *status = ARS_FAILED;
1385                                 pgs->hpk_errval = -rc;
1386                         }
1387                 }
1388                 /* we have to retry, so keep layout lock */
1389                 if (*status == ARS_WAITING)
1390                         GOTO(out, rc);
1391
1392                 /* give back layout lock */
1393                 mutex_lock(&cdt->cdt_restore_lock);
1394                 crh = hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1395                 if (crh != NULL)
1396                         list_del(&crh->crh_list);
1397                 mutex_unlock(&cdt->cdt_restore_lock);
1398                 /* just give back layout lock, we keep
1399                  * the reference which is given back
1400                  * later with the lock for HSM flags */
1401                 if (!IS_ERR(obj) && crh != NULL)
1402                         mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1403
1404                 if (crh != NULL)
1405                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1406         }
1407
1408         GOTO(out, rc);
1409
1410 out:
1411         if (obj != NULL && !IS_ERR(obj)) {
1412                 mo_changelog(env, CL_HSM, cl_flags,
1413                              mdt_object_child(obj));
1414                 mdt_object_put(mti->mti_env, obj);
1415         }
1416
1417         RETURN(rc);
1418 }
1419
1420 /**
1421  * update status of a request
1422  * \param mti [IN] context
1423  * \param pgs [IN] progress of the copy tool
1424  * \param update_record [IN] update llog record
1425  * \retval 0 success
1426  * \retval -ve failure
1427  */
1428 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1429                                  struct hsm_progress_kernel *pgs,
1430                                  const int update_record)
1431 {
1432         struct mdt_device       *mdt = mti->mti_mdt;
1433         struct coordinator      *cdt = &mdt->mdt_coordinator;
1434         struct cdt_agent_req    *car;
1435         int                      rc = 0;
1436         ENTRY;
1437
1438         /* no coordinator started, so we cannot serve requests */
1439         if (cdt->cdt_state == CDT_STOPPED)
1440                 RETURN(-EAGAIN);
1441
1442         /* first do sanity checks */
1443         car = mdt_cdt_update_request(cdt, pgs);
1444         if (IS_ERR(car)) {
1445                 CERROR("%s: Cannot find running request for cookie "LPX64
1446                        " on fid="DFID"\n",
1447                        mdt_obd_name(mdt),
1448                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1449                 if (car == NULL)
1450                         RETURN(-ENOENT);
1451                 RETURN(PTR_ERR(car));
1452         }
1453
1454         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1455                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1456                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1457                       hsm_copytool_action2name(car->car_hai->hai_action),
1458                       pgs->hpk_flags, pgs->hpk_errval,
1459                       PFID(&car->car_hai->hai_fid),
1460                       PFID(&car->car_hai->hai_dfid));
1461
1462         /* progress is done on FID or data FID depending of the action and
1463          * of the copy progress */
1464         /* for restore progress is used to send back the data FID to cdt */
1465         if (car->car_hai->hai_action == HSMA_RESTORE &&
1466             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1467                 car->car_hai->hai_dfid = pgs->hpk_fid;
1468
1469         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1470              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1471             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1472              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1473                 CERROR("%s: Progress on "DFID" for cookie "LPX64
1474                        " does not match request FID "DFID" nor data FID "
1475                        DFID"\n",
1476                        mdt_obd_name(mdt),
1477                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1478                        PFID(&car->car_hai->hai_fid),
1479                        PFID(&car->car_hai->hai_dfid));
1480                 GOTO(out, rc = -EINVAL);
1481         }
1482
1483         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1484                 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1485                        " is not coherent (err=%d and not completed"
1486                        " (flags=%d))\n",
1487                        mdt_obd_name(mdt),
1488                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1489                        hsm_copytool_action2name(car->car_hai->hai_action),
1490                        pgs->hpk_errval, pgs->hpk_flags);
1491                 GOTO(out, rc = -EINVAL);
1492         }
1493
1494         /* now progress is valid */
1495
1496         /* we use a root like ucred */
1497         hsm_init_ucred(mdt_ucred(mti));
1498
1499         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1500                 enum agent_req_status    status;
1501
1502                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1503
1504                 /* remove request from memory list */
1505                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1506
1507                 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1508                               " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1509                        pgs->hpk_cookie,
1510                        hsm_copytool_action2name(car->car_hai->hai_action),
1511                        agent_req_status2name(status));
1512
1513                 if (update_record) {
1514                         int rc1;
1515
1516                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1517                                                      &pgs->hpk_cookie, 1,
1518                                                      status);
1519                         if (rc1)
1520                                 CERROR("%s: mdt_agent_record_update() failed,"
1521                                        " rc=%d, cannot update status to %s"
1522                                        " for cookie "LPX64"\n",
1523                                        mdt_obd_name(mdt), rc1,
1524                                        agent_req_status2name(status),
1525                                        pgs->hpk_cookie);
1526                         rc = (rc != 0 ? rc : rc1);
1527                 }
1528                 /* ct has completed a request, so a slot is available, wakeup
1529                  * cdt to find new work */
1530                 mdt_hsm_cdt_wakeup(mdt);
1531         } else {
1532                 /* if copytool send a progress on a canceled request
1533                  * we inform copytool it should stop
1534                  */
1535                 if (car->car_canceled == 1)
1536                         rc = -ECANCELED;
1537         }
1538         GOTO(out, rc);
1539
1540 out:
1541         /* remove ref got from mdt_cdt_update_request() */
1542         mdt_cdt_put_request(car);
1543
1544         return rc;
1545 }
1546
1547
1548 /**
1549  * data passed to llog_cat_process() callback
1550  * to cancel requests
1551  */
1552 struct hsm_cancel_all_data {
1553         struct mdt_device       *mdt;
1554 };
1555
1556 /**
1557  *  llog_cat_process() callback, used to:
1558  *  - purge all requests
1559  * \param env [IN] environment
1560  * \param llh [IN] llog handle
1561  * \param hdr [IN] llog record
1562  * \param data [IN] cb data = struct hsm_cancel_all_data
1563  * \retval 0 success
1564  * \retval -ve failure
1565  */
1566 static int mdt_cancel_all_cb(const struct lu_env *env,
1567                              struct llog_handle *llh,
1568                              struct llog_rec_hdr *hdr, void *data)
1569 {
1570         struct llog_agent_req_rec       *larr;
1571         struct hsm_cancel_all_data      *hcad;
1572         int                              rc = 0;
1573         ENTRY;
1574
1575         larr = (struct llog_agent_req_rec *)hdr;
1576         hcad = data;
1577         if (larr->arr_status == ARS_WAITING ||
1578             larr->arr_status == ARS_STARTED) {
1579                 larr->arr_status = ARS_CANCELED;
1580                 larr->arr_req_change = cfs_time_current_sec();
1581                 rc = mdt_agent_llog_update_rec(env, hcad->mdt, llh, larr);
1582                 if (rc == 0)
1583                         RETURN(LLOG_DEL_RECORD);
1584         }
1585         RETURN(rc);
1586 }
1587
1588 /**
1589  * cancel all actions
1590  * \param obd [IN] MDT device
1591  */
1592 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1593 {
1594         struct mdt_thread_info          *mti;
1595         struct coordinator              *cdt = &mdt->mdt_coordinator;
1596         struct cdt_agent_req            *car;
1597         struct hsm_action_list          *hal = NULL;
1598         struct hsm_action_item          *hai;
1599         struct hsm_cancel_all_data       hcad;
1600         int                              hal_sz = 0, hal_len, rc;
1601         enum cdt_states                  save_state;
1602         ENTRY;
1603
1604         /* retrieve coordinator context */
1605         mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1606
1607         /* disable coordinator */
1608         save_state = cdt->cdt_state;
1609         cdt->cdt_state = CDT_DISABLE;
1610
1611         /* send cancel to all running requests */
1612         down_read(&cdt->cdt_request_lock);
1613         list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1614                 mdt_cdt_get_request(car);
1615                 /* request is not yet removed from list, it will be done
1616                  * when copytool will return progress
1617                  */
1618
1619                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1620                         mdt_cdt_put_request(car);
1621                         continue;
1622                 }
1623
1624                 /* needed size */
1625                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1626                           cfs_size_round(car->car_hai->hai_len);
1627
1628                 if (hal_len > hal_sz && hal_sz > 0) {
1629                         /* not enough room, free old buffer */
1630                         OBD_FREE(hal, hal_sz);
1631                         hal = NULL;
1632                 }
1633
1634                 /* empty buffer, allocate one */
1635                 if (hal == NULL) {
1636                         hal_sz = hal_len;
1637                         OBD_ALLOC(hal, hal_sz);
1638                         if (hal == NULL) {
1639                                 mdt_cdt_put_request(car);
1640                                 up_read(&cdt->cdt_request_lock);
1641                                 GOTO(out, rc = -ENOMEM);
1642                         }
1643                 }
1644
1645                 hal->hal_version = HAL_VERSION;
1646                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1647                                 MTI_NAME_MAXLEN);
1648                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1649                 hal->hal_compound_id = car->car_compound_id;
1650                 hal->hal_archive_id = car->car_archive_id;
1651                 hal->hal_flags = car->car_flags;
1652                 hal->hal_count = 0;
1653
1654                 hai = hai_first(hal);
1655                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1656                 hai->hai_action = HSMA_CANCEL;
1657                 hal->hal_count = 1;
1658
1659                 /* it is possible to safely call mdt_hsm_agent_send()
1660                  * (ie without a deadlock on cdt_request_lock), because the
1661                  * write lock is taken only if we are not in purge mode
1662                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1663                  *   nor mdt_cdt_remove_request())
1664                  */
1665                 /* no conflict with cdt thread because cdt is disable and we
1666                  * have the request lock */
1667                 mdt_hsm_agent_send(mti, hal, 1);
1668
1669                 mdt_cdt_put_request(car);
1670         }
1671         up_read(&cdt->cdt_request_lock);
1672
1673         if (hal != NULL)
1674                 OBD_FREE(hal, hal_sz);
1675
1676         /* cancel all on-disk records */
1677         hcad.mdt = mdt;
1678
1679         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1680                               mdt_cancel_all_cb, &hcad);
1681 out:
1682         /* enable coordinator */
1683         cdt->cdt_state = save_state;
1684
1685         RETURN(rc);
1686 }
1687
1688 /**
1689  * check if a request is comptaible with file status
1690  * \param hai [IN] request description
1691  * \param hal_an [IN] request archive number (not used)
1692  * \param rq_flags [IN] request flags
1693  * \param hsm [IN] file HSM metadata
1694  * \retval boolean
1695  */
1696 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1697                               const int hal_an, const __u64 rq_flags,
1698                               const struct md_hsm *hsm)
1699 {
1700         int      is_compat = false;
1701         int      hsm_flags;
1702         ENTRY;
1703
1704         hsm_flags = hsm->mh_flags;
1705         switch (hai->hai_action) {
1706         case HSMA_ARCHIVE:
1707                 if (!(hsm_flags & HS_NOARCHIVE) &&
1708                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1709                         is_compat = true;
1710                 break;
1711         case HSMA_RESTORE:
1712                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1713                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1714                         is_compat = true;
1715                 break;
1716         case HSMA_REMOVE:
1717                 if (!(hsm_flags & HS_RELEASED) &&
1718                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1719                         is_compat = true;
1720                 break;
1721         case HSMA_CANCEL:
1722                 is_compat = true;
1723                 break;
1724         }
1725         CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1726                       " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1727                       PFID(&hai->hai_fid),
1728                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1729                       hai->hai_extent.offset, hai->hai_extent.length,
1730                       hsm->mh_flags,
1731                       (is_compat ? "compatible" : "uncompatible"));
1732
1733         RETURN(is_compat);
1734 }
1735
1736 /*
1737  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1738  */
1739 static const struct {
1740         __u64            bit;
1741         char            *name;
1742         char            *nickname;
1743 } hsm_policy_names[] = {
1744         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1745         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1746         { 0 },
1747 };
1748
1749 /**
1750  * convert a policy name to a bit
1751  * \param name [IN] policy name
1752  * \retval 0 unknown
1753  * \retval   policy bit
1754  */
1755 static __u64 hsm_policy_str2bit(const char *name)
1756 {
1757         int      i;
1758
1759         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1760                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1761                     strcmp(hsm_policy_names[i].name, name) == 0)
1762                         return hsm_policy_names[i].bit;
1763         return 0;
1764 }
1765
1766 /**
1767  * convert a policy bit field to a string
1768  * \param mask [IN] policy bit field
1769  * \param hexa [IN] print mask before bit names
1770  * \param buffer [OUT] string
1771  * \param count [IN] size of buffer
1772  * \retval size filled in buffer
1773  */
1774 static int hsm_policy_bit2str(const __u64 mask, const bool hexa, char *buffer,
1775                               int count)
1776 {
1777         int      i, j, sz;
1778         char    *ptr;
1779         __u64    bit;
1780         ENTRY;
1781
1782         ptr = buffer;
1783         if (hexa) {
1784                 sz = snprintf(buffer, count, "("LPX64") ", mask);
1785                 ptr += sz;
1786                 count -= sz;
1787         }
1788         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1789                 bit = (1ULL << i);
1790
1791                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1792                         if (hsm_policy_names[j].bit == bit)
1793                                 break;
1794                 }
1795                 if (bit & mask)
1796                         sz = snprintf(ptr, count, "[%s] ",
1797                                       hsm_policy_names[j].name);
1798                 else
1799                         sz = snprintf(ptr, count, "%s ",
1800                                       hsm_policy_names[j].name);
1801
1802                 ptr += sz;
1803                 count -= sz;
1804         }
1805         /* remove last ' ' */
1806         *ptr = '\0';
1807         ptr--;
1808         RETURN(ptr - buffer);
1809 }
1810
1811 /* methods to read/write HSM policy flags */
1812 static int lprocfs_rd_hsm_policy(char *page, char **start, off_t off,
1813                                  int count, int *eof, void *data)
1814 {
1815         struct mdt_device       *mdt = data;
1816         struct coordinator      *cdt = &mdt->mdt_coordinator;
1817         int                      sz;
1818         ENTRY;
1819
1820         sz = hsm_policy_bit2str(cdt->cdt_policy, false, page, count);
1821         page[sz] = '\n';
1822         sz++;
1823         page[sz] = '\0';
1824         *eof = 1;
1825         RETURN(sz);
1826 }
1827
1828 static int lprocfs_wr_hsm_policy(struct file *file, const char *buffer,
1829                                  unsigned long count, void *data)
1830 {
1831         struct mdt_device       *mdt = data;
1832         struct coordinator      *cdt = &mdt->mdt_coordinator;
1833         char                    *start, *token, sign;
1834         char                    *buf;
1835         __u64                    policy;
1836         __u64                    add_mask, remove_mask, set_mask;
1837         int                      sz;
1838         int                      rc;
1839         ENTRY;
1840
1841         if (count + 1 > PAGE_SIZE)
1842                 RETURN(-EINVAL);
1843
1844         OBD_ALLOC(buf, count + 1);
1845         if (buf == NULL)
1846                 RETURN(-ENOMEM);
1847
1848         if (copy_from_user(buf, buffer, count))
1849                 RETURN(-EFAULT);
1850         buf[count] = '\0';
1851
1852         start = buf;
1853         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1854                start);
1855
1856         add_mask = remove_mask = set_mask = 0;
1857         do {
1858                 token = strsep(&start, "\n ");
1859                 sign = *token;
1860
1861                 if (sign == '\0')
1862                         continue;
1863
1864                 if (sign == '-' || sign == '+')
1865                         token++;
1866
1867                 policy = hsm_policy_str2bit(token);
1868                 if (policy == 0) {
1869                         char *msg;
1870
1871                         sz = PAGE_SIZE;
1872                         OBD_ALLOC(msg, sz);
1873                         if (!msg)
1874                                 RETURN(-ENOMEM);
1875
1876                         hsm_policy_bit2str(0, false, msg, sz);
1877                         CWARN("%s: '%s' is unknown, "
1878                               "supported policies are: %s\n", mdt_obd_name(mdt),
1879                               token, msg);
1880                         OBD_FREE(msg, sz);
1881                         GOTO(out, rc = -EINVAL);
1882                 }
1883                 switch (sign) {
1884                 case '-':
1885                         remove_mask |= policy;
1886                         break;
1887                 case '+':
1888                         add_mask |= policy;
1889                         break;
1890                 default:
1891                         set_mask |= policy;
1892                         break;
1893                 }
1894
1895         } while (start != NULL);
1896
1897         CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1898                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1899
1900         /* if no sign in all string, it is a clear and set
1901          * if some sign found, all unsigned are converted
1902          * to add
1903          * P1 P2 = set to P1 and P2
1904          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1905          */
1906         if (remove_mask == 0 && add_mask == 0) {
1907                 cdt->cdt_policy = set_mask;
1908         } else {
1909                 cdt->cdt_policy |= set_mask | add_mask;
1910                 cdt->cdt_policy &= ~remove_mask;
1911         }
1912
1913         GOTO(out, rc = count);
1914
1915 out:
1916         OBD_FREE(buf, count + 1);
1917         RETURN(rc);
1918 }
1919
1920 #define GENERATE_PROC_METHOD(VAR)                                       \
1921 static int lprocfs_rd_hsm_##VAR(char *page, char **start, off_t off,    \
1922                                 int count, int *eof, void *data)        \
1923 {                                                                       \
1924         struct mdt_device       *mdt = data;                            \
1925         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1926         int                      sz;                                    \
1927         ENTRY;                                                          \
1928                                                                         \
1929         sz = snprintf(page, count, LPU64"\n", (__u64)cdt->VAR);         \
1930         *eof = 1;                                                       \
1931         RETURN(sz);                                                     \
1932 }                                                                       \
1933 static int lprocfs_wr_hsm_##VAR(struct file *file, const char *buffer,  \
1934                                 unsigned long count, void *data)        \
1935                                                                         \
1936 {                                                                       \
1937         struct mdt_device       *mdt = data;                            \
1938         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1939         int                      val;                                   \
1940         int                      rc;                                    \
1941         ENTRY;                                                          \
1942                                                                         \
1943         rc = lprocfs_write_helper(buffer, count, &val);                 \
1944         if (rc)                                                         \
1945                 RETURN(rc);                                             \
1946         if (val > 0) {                                                  \
1947                 cdt->VAR = val;                                         \
1948                 RETURN(count);                                          \
1949         }                                                               \
1950         RETURN(-EINVAL);                                                \
1951 }
1952
1953 GENERATE_PROC_METHOD(cdt_loop_period)
1954 GENERATE_PROC_METHOD(cdt_grace_delay)
1955 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1956 GENERATE_PROC_METHOD(cdt_max_requests)
1957 GENERATE_PROC_METHOD(cdt_default_archive_id)
1958
1959 /*
1960  * procfs write method for MDT/hsm_control
1961  * proc entry is in mdt directory so data is mdt obd_device pointer
1962  */
1963 #define CDT_ENABLE_CMD   "enabled"
1964 #define CDT_STOP_CMD     "shutdown"
1965 #define CDT_DISABLE_CMD  "disabled"
1966 #define CDT_PURGE_CMD    "purge"
1967 #define CDT_HELP_CMD     "help"
1968
1969 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
1970                                unsigned long count, void *data)
1971 {
1972         struct obd_device       *obd = data;
1973         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
1974         struct coordinator      *cdt = &(mdt->mdt_coordinator);
1975         int                      rc, usage = 0;
1976         ENTRY;
1977
1978         rc = 0;
1979         if (strncmp(buffer, CDT_ENABLE_CMD, strlen(CDT_ENABLE_CMD)) == 0) {
1980                 if (cdt->cdt_state == CDT_DISABLE) {
1981                         cdt->cdt_state = CDT_RUNNING;
1982                         mdt_hsm_cdt_wakeup(mdt);
1983                 } else {
1984                         rc = mdt_hsm_cdt_start(mdt);
1985                 }
1986         } else if (strncmp(buffer, CDT_STOP_CMD, strlen(CDT_STOP_CMD)) == 0) {
1987                 if ((cdt->cdt_state == CDT_STOPPING) ||
1988                     (cdt->cdt_state == CDT_STOPPED)) {
1989                         CERROR("%s: Coordinator already stopped\n",
1990                                mdt_obd_name(mdt));
1991                         rc = -EALREADY;
1992                 } else {
1993                         cdt->cdt_state = CDT_STOPPING;
1994                 }
1995         } else if (strncmp(buffer, CDT_DISABLE_CMD,
1996                            strlen(CDT_DISABLE_CMD)) == 0) {
1997                 if ((cdt->cdt_state == CDT_STOPPING) ||
1998                     (cdt->cdt_state == CDT_STOPPED)) {
1999                         CERROR("%s: Coordinator is stopped\n",
2000                                mdt_obd_name(mdt));
2001                         rc = -EINVAL;
2002                 } else {
2003                         cdt->cdt_state = CDT_DISABLE;
2004                 }
2005         } else if (strncmp(buffer, CDT_PURGE_CMD, strlen(CDT_PURGE_CMD)) == 0) {
2006                 rc = hsm_cancel_all_actions(mdt);
2007         } else if (strncmp(buffer, CDT_HELP_CMD, strlen(CDT_HELP_CMD)) == 0) {
2008                 usage = 1;
2009         } else {
2010                 usage = 1;
2011                 rc = -EINVAL;
2012         }
2013
2014         if (usage == 1)
2015                 CERROR("%s: Valid coordinator control commands are: "
2016                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2017                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2018                        CDT_PURGE_CMD, CDT_HELP_CMD);
2019
2020         if (rc)
2021                 RETURN(rc);
2022
2023         RETURN(count);
2024 }
2025
2026 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
2027                                int count, int *eof, void *data)
2028 {
2029         struct obd_device       *obd = data;
2030         struct coordinator      *cdt;
2031         int                      sz;
2032         ENTRY;
2033
2034         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2035         *eof = 1;
2036
2037         if (cdt->cdt_state == CDT_INIT)
2038                 sz = snprintf(page, count, "init\n");
2039         else if (cdt->cdt_state == CDT_RUNNING)
2040                 sz = snprintf(page, count, "enabled\n");
2041         else if (cdt->cdt_state == CDT_STOPPING)
2042                 sz = snprintf(page, count, "stopping\n");
2043         else if (cdt->cdt_state == CDT_STOPPED)
2044                 sz = snprintf(page, count, "stopped\n");
2045         else if (cdt->cdt_state == CDT_DISABLE)
2046                 sz = snprintf(page, count, "disabled\n");
2047         else
2048                 sz = snprintf(page, count, "unknown\n");
2049
2050         RETURN(sz);
2051 }
2052
2053 static int
2054 lprocfs_rd_hsm_request_mask(char *page, char **start, off_t off,
2055                             int count, int *eof, __u64 mask)
2056 {
2057         int i, rc = 0;
2058         ENTRY;
2059
2060         for (i = 0; i < 8 * sizeof(mask); i++) {
2061                 if (mask & (1UL << i))
2062                         rc += snprintf(page + rc, count - rc, "%s%s",
2063                                        rc == 0 ? "" : " ",
2064                                        hsm_copytool_action2name(i));
2065         }
2066
2067         rc += snprintf(page + rc, count - rc, "\n");
2068
2069         RETURN(rc);
2070 }
2071
2072 static int
2073 lprocfs_rd_hsm_user_request_mask(char *page, char **start, off_t off,
2074                                  int count, int *eof, void *data)
2075 {
2076         struct mdt_device *mdt = data;
2077         struct coordinator *cdt = &mdt->mdt_coordinator;
2078
2079         return lprocfs_rd_hsm_request_mask(page, start, off, count, eof,
2080                                            cdt->cdt_user_request_mask);
2081 }
2082
2083 static int
2084 lprocfs_rd_hsm_group_request_mask(char *page, char **start, off_t off,
2085                                   int count, int *eof, void *data)
2086 {
2087         struct mdt_device *mdt = data;
2088         struct coordinator *cdt = &mdt->mdt_coordinator;
2089
2090         return lprocfs_rd_hsm_request_mask(page, start, off, count, eof,
2091                                            cdt->cdt_group_request_mask);
2092 }
2093
2094 static int
2095 lprocfs_rd_hsm_other_request_mask(char *page, char **start, off_t off,
2096                                   int count, int *eof, void *data)
2097 {
2098         struct mdt_device *mdt = data;
2099         struct coordinator *cdt = &mdt->mdt_coordinator;
2100
2101         return lprocfs_rd_hsm_request_mask(page, start, off, count, eof,
2102                                            cdt->cdt_other_request_mask);
2103 }
2104
2105 static inline enum hsm_copytool_action
2106 hsm_copytool_name2action(const char *name)
2107 {
2108         if (strcasecmp(name, "NOOP") == 0)
2109                 return HSMA_NONE;
2110         else if (strcasecmp(name, "ARCHIVE") == 0)
2111                 return HSMA_ARCHIVE;
2112         else if (strcasecmp(name, "RESTORE") == 0)
2113                 return HSMA_RESTORE;
2114         else if (strcasecmp(name, "REMOVE") == 0)
2115                 return HSMA_REMOVE;
2116         else if (strcasecmp(name, "CANCEL") == 0)
2117                 return HSMA_CANCEL;
2118         else
2119                 return -1;
2120 }
2121
2122 static int
2123 lprocfs_wr_hsm_request_mask(struct file *file, const char __user *user_buf,
2124                             unsigned long user_count, __u64 *mask)
2125 {
2126         char *buf, *pos, *name;
2127         size_t buf_size;
2128         __u64 new_mask = 0;
2129         int rc;
2130         ENTRY;
2131
2132         if (!(user_count < 4096))
2133                 RETURN(-ENOMEM);
2134
2135         buf_size = user_count + 1;
2136
2137         OBD_ALLOC(buf, buf_size);
2138         if (buf == NULL)
2139                 RETURN(-ENOMEM);
2140
2141         if (copy_from_user(buf, user_buf, buf_size - 1))
2142                 GOTO(out, rc = -EFAULT);
2143
2144         buf[buf_size - 1] = '\0';
2145
2146         pos = buf;
2147         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2148                 int action;
2149
2150                 if (*name == '\0')
2151                         continue;
2152
2153                 action = hsm_copytool_name2action(name);
2154                 if (action < 0)
2155                         GOTO(out, rc = -EINVAL);
2156
2157                 new_mask |= (1UL << action);
2158         }
2159
2160         *mask = new_mask;
2161         rc = user_count;
2162 out:
2163         OBD_FREE(buf, buf_size);
2164
2165         RETURN(rc);
2166 }
2167
2168 static int
2169 lprocfs_wr_hsm_user_request_mask(struct file *file, const char __user *buf,
2170                                  unsigned long count, void *data)
2171 {
2172         struct mdt_device *mdt = data;
2173         struct coordinator *cdt = &mdt->mdt_coordinator;
2174
2175         return lprocfs_wr_hsm_request_mask(file, buf, count,
2176                                            &cdt->cdt_user_request_mask);
2177 }
2178
2179 static int
2180 lprocfs_wr_hsm_group_request_mask(struct file *file, const char __user *buf,
2181                                   unsigned long count, void *data)
2182 {
2183         struct mdt_device *mdt = data;
2184         struct coordinator *cdt = &mdt->mdt_coordinator;
2185
2186         return lprocfs_wr_hsm_request_mask(file, buf, count,
2187                                            &cdt->cdt_group_request_mask);
2188 }
2189
2190 static int
2191 lprocfs_wr_hsm_other_request_mask(struct file *file, const char __user *buf,
2192                                   unsigned long count, void *data)
2193 {
2194         struct mdt_device *mdt = data;
2195         struct coordinator *cdt = &mdt->mdt_coordinator;
2196
2197         return lprocfs_wr_hsm_request_mask(file, buf, count,
2198                                            &cdt->cdt_other_request_mask);
2199 }
2200
2201 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2202         { "agents",                     NULL, NULL, NULL, &mdt_hsm_agent_fops,
2203                                         0 },
2204         { "actions",                    NULL, NULL, NULL, &mdt_hsm_actions_fops,
2205                                         0444 },
2206         { "default_archive_id",         lprocfs_rd_hsm_cdt_default_archive_id,
2207                                         lprocfs_wr_hsm_cdt_default_archive_id,
2208                                         NULL, NULL, 0 },
2209         { "grace_delay",                lprocfs_rd_hsm_cdt_grace_delay,
2210                                         lprocfs_wr_hsm_cdt_grace_delay,
2211                                         NULL, NULL, 0 },
2212         { "loop_period",                lprocfs_rd_hsm_cdt_loop_period,
2213                                         lprocfs_wr_hsm_cdt_loop_period,
2214                                         NULL, NULL, 0 },
2215         { "max_requests",               lprocfs_rd_hsm_cdt_max_requests,
2216                                         lprocfs_wr_hsm_cdt_max_requests,
2217                                         NULL, NULL, 0 },
2218         { "policy",                     lprocfs_rd_hsm_policy,
2219                                         lprocfs_wr_hsm_policy,
2220                                         NULL, NULL, 0 },
2221         { "active_request_timeout",     lprocfs_rd_hsm_cdt_active_req_timeout,
2222                                         lprocfs_wr_hsm_cdt_active_req_timeout,
2223                                         NULL, NULL, 0 },
2224         { "active_requests",            NULL, NULL, NULL,
2225                                         &mdt_hsm_active_requests_fops, 0 },
2226         { "user_request_mask",          lprocfs_rd_hsm_user_request_mask,
2227                                         lprocfs_wr_hsm_user_request_mask, },
2228         { "group_request_mask",         lprocfs_rd_hsm_group_request_mask,
2229                                         lprocfs_wr_hsm_group_request_mask, },
2230         { "other_request_mask",         lprocfs_rd_hsm_other_request_mask,
2231                                         lprocfs_wr_hsm_other_request_mask, },
2232         { 0 }
2233 };