lustre/mdt/mdt_coordinator.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2011, 2012 Commissariat a l'energie atomique et aux energies
  24  *                          alternatives
  25  *
  26  * Copyright (c) 2013, 2014, Intel Corporation.
  27  * Use is subject to license terms.
  28  */
  29 /*
  30  * lustre/mdt/mdt_coordinator.c
  31  *
  32  * Lustre HSM Coordinator
  33  *
  34  * Author: Jacques-Charles Lafoucriere <jacques-charles.lafoucriere@cea.fr>
  35  * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
  36  * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <linux/kthread.h>
  42 #include <obd_support.h>
  43 #include <lustre_net.h>
  44 #include <lustre_export.h>
  45 #include <obd.h>
  46 #include <lprocfs_status.h>
  47 #include <lustre_log.h>
  48 #include <lustre_kernelcomm.h>
  49 #include "mdt_internal.h"
  50
  51 static struct lprocfs_vars lprocfs_mdt_hsm_vars[];
  52
  53 /**
  54  * get obj and HSM attributes on a fid
  55  * \param mti [IN] context
  56  * \param fid [IN] object fid
  57  * \param hsm [OUT] HSM meta data
  58  * \retval obj or error (-ENOENT if not found)
  59  */
  60 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
  61                                       const struct lu_fid *fid,
  62                                       struct md_hsm *hsm)
  63 {
  64         struct md_attr          *ma;
  65         struct mdt_object       *obj;
  66         int                      rc;
  67         ENTRY;
  68
  69         ma = &mti->mti_attr;
  70         ma->ma_need = MA_HSM;
  71         ma->ma_valid = 0;
  72
  73         /* find object by FID */
  74         obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
  75         if (IS_ERR(obj))
  76                 RETURN(obj);
  77
  78         if (!mdt_object_exists(obj)) {
  79                 /* no more object */
  80                 mdt_object_put(mti->mti_env, obj);
  81                 RETURN(ERR_PTR(-ENOENT));
  82         }
  83
  84         rc = mdt_attr_get_complex(mti, obj, ma);
  85         if (rc) {
  86                 mdt_object_put(mti->mti_env, obj);
  87                 RETURN(ERR_PTR(rc));
  88         }
  89
  90         if (ma->ma_valid & MA_HSM)
  91                 *hsm = ma->ma_hsm;
  92         else
  93                 memset(hsm, 0, sizeof(*hsm));
  94         ma->ma_valid = 0;
  95         RETURN(obj);
  96 }
  97
  98 void mdt_hsm_dump_hal(int level, const char *prefix,
  99                       struct hsm_action_list *hal)
 100 {
 101         int                      i, sz;
 102         struct hsm_action_item  *hai;
 103         char                     buf[12];
 104
 105         CDEBUG(level, "%s: HAL header: version %X count %d compound "LPX64
 106                       " archive_id %d flags "LPX64"\n",
 107                prefix, hal->hal_version, hal->hal_count,
 108                hal->hal_compound_id, hal->hal_archive_id, hal->hal_flags);
 109
 110         hai = hai_first(hal);
 111         for (i = 0; i < hal->hal_count; i++) {
 112                 sz = hai->hai_len - sizeof(*hai);
 113                 CDEBUG(level, "%s %d: fid="DFID" dfid="DFID
 114                        " compound/cookie="LPX64"/"LPX64
 115                        " action=%s extent="LPX64"-"LPX64" gid="LPX64
 116                        " datalen=%d data=[%s]\n",
 117                        prefix, i,
 118                        PFID(&hai->hai_fid), PFID(&hai->hai_dfid),
 119                        hal->hal_compound_id, hai->hai_cookie,
 120                        hsm_copytool_action2name(hai->hai_action),
 121                        hai->hai_extent.offset,
 122                        hai->hai_extent.length,
 123                        hai->hai_gid, sz,
 124                        hai_dump_data_field(hai, buf, sizeof(buf)));
 125                 hai = hai_next(hai);
 126         }
 127 }
 128
 129 /**
 130  * data passed to llog_cat_process() callback
 131  * to scan requests and take actions
 132  */
 133 struct hsm_scan_data {
 134         struct mdt_thread_info          *mti;
 135         char                             fs_name[MTI_NAME_MAXLEN+1];
 136         /* request to be send to agents */
 137         int                              request_sz;    /** allocated size */
 138         int                              max_requests;  /** vector size */
 139         int                              request_cnt;   /** used count */
 140         struct {
 141                 int                      hal_sz;
 142                 int                      hal_used_sz;
 143                 struct hsm_action_list  *hal;
 144         } *request;
 145         /* records to be canceled */
 146         int                              max_cookie;    /** vector size */
 147         int                              cookie_cnt;    /** used count */
 148         __u64                           *cookies;
 149 };
 150
 151 /**
 152  *  llog_cat_process() callback, used to:
 153  *  - find waiting request and start action
 154  *  - purge canceled and done requests
 155  * \param env [IN] environment
 156  * \param llh [IN] llog handle
 157  * \param hdr [IN] llog record
 158  * \param data [IN/OUT] cb data = struct hsm_scan_data
 159  * \retval 0 success
 160  * \retval -ve failure
 161  */
 162 static int mdt_coordinator_cb(const struct lu_env *env,
 163                               struct llog_handle *llh,
 164                               struct llog_rec_hdr *hdr,
 165                               void *data)
 166 {
 167         const struct llog_agent_req_rec *larr;
 168         struct hsm_scan_data            *hsd;
 169         struct hsm_action_item          *hai;
 170         struct mdt_device               *mdt;
 171         struct coordinator              *cdt;
 172         int                              rc;
 173         ENTRY;
 174
 175         hsd = data;
 176         mdt = hsd->mti->mti_mdt;
 177         cdt = &mdt->mdt_coordinator;
 178
 179         larr = (struct llog_agent_req_rec *)hdr;
 180         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 181         switch (larr->arr_status) {
 182         case ARS_WAITING: {
 183                 int i, empty_slot, found;
 184
 185                 /* Are agents full? */
 186                 if (atomic_read(&cdt->cdt_request_count) ==
 187                     cdt->cdt_max_requests)
 188                         break;
 189
 190                 /* first search whether the request is found in the list we
 191                  * have built and if there is room in the request vector */
 192                 empty_slot = -1;
 193                 found = -1;
 194                 for (i = 0; i < hsd->max_requests &&
 195                             (empty_slot == -1 || found == -1); i++) {
 196                         if (hsd->request[i].hal == NULL) {
 197                                 empty_slot = i;
 198                                 continue;
 199                         }
 200                         if (hsd->request[i].hal->hal_compound_id ==
 201                                 larr->arr_compound_id) {
 202                                 found = i;
 203                                 continue;
 204                         }
 205                 }
 206                 if (found == -1 && empty_slot == -1)
 207                         /* unknown request and no more room for new request,
 208                          * continue scan for to find other entries for
 209                          * already found request
 210                          */
 211                         RETURN(0);
 212
 213                 if (found == -1) {
 214                         struct hsm_action_list *hal;
 215
 216                         /* request is not already known */
 217                         /* allocates hai vector size just needs to be large
 218                          * enough */
 219                         hsd->request[empty_slot].hal_sz =
 220                                      sizeof(*hsd->request[empty_slot].hal) +
 221                                      cfs_size_round(MTI_NAME_MAXLEN+1) +
 222                                      2 * cfs_size_round(larr->arr_hai.hai_len);
 223                         OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
 224                         if (!hal) {
 225                                 CERROR("%s: Cannot allocate memory (%d o)"
 226                                        "for compound "LPX64"\n",
 227                                        mdt_obd_name(mdt),
 228                                        hsd->request[i].hal_sz,
 229                                        larr->arr_compound_id);
 230                                 RETURN(-ENOMEM);
 231                         }
 232                         hal->hal_version = HAL_VERSION;
 233                         strlcpy(hal->hal_fsname, hsd->fs_name,
 234                                 MTI_NAME_MAXLEN + 1);
 235                         hal->hal_compound_id = larr->arr_compound_id;
 236                         hal->hal_archive_id = larr->arr_archive_id;
 237                         hal->hal_flags = larr->arr_flags;
 238                         hal->hal_count = 0;
 239                         hsd->request[empty_slot].hal_used_sz = hal_size(hal);
 240                         hsd->request[empty_slot].hal = hal;
 241                         hsd->request_cnt++;
 242                         found = empty_slot;
 243                         hai = hai_first(hal);
 244                 } else {
 245                         /* request is known */
 246                         /* we check if record archive num is the same as the
 247                          * known request, if not we will serve it in multiple
 248                          * time because we do not know if the agent can serve
 249                          * multiple backend
 250                          * a use case is a compound made of multiple restore
 251                          * where the files are not archived in the same backend
 252                          */
 253                         if (larr->arr_archive_id !=
 254                             hsd->request[found].hal->hal_archive_id)
 255                                 RETURN(0);
 256
 257                         if (hsd->request[found].hal_sz <
 258                             hsd->request[found].hal_used_sz +
 259                              cfs_size_round(larr->arr_hai.hai_len)) {
 260                                 /* Not enough room, need an extension */
 261                                 void *hal_buffer;
 262                                 int sz;
 263
 264                                 sz = 2 * hsd->request[found].hal_sz;
 265                                 OBD_ALLOC(hal_buffer, sz);
 266                                 if (!hal_buffer) {
 267                                         CERROR("%s: Cannot allocate memory "
 268                                                "(%d o) for compound "LPX64"\n",
 269                                                mdt_obd_name(mdt), sz,
 270                                                larr->arr_compound_id);
 271                                         RETURN(-ENOMEM);
 272                                 }
 273                                 memcpy(hal_buffer, hsd->request[found].hal,
 274                                        hsd->request[found].hal_used_sz);
 275                                 OBD_FREE(hsd->request[found].hal,
 276                                          hsd->request[found].hal_sz);
 277                                 hsd->request[found].hal = hal_buffer;
 278                                 hsd->request[found].hal_sz = sz;
 279                         }
 280                         hai = hai_first(hsd->request[found].hal);
 281                         for (i = 0; i < hsd->request[found].hal->hal_count;
 282                              i++)
 283                                 hai = hai_next(hai);
 284                 }
 285                 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 286                 hai->hai_cookie = larr->arr_hai.hai_cookie;
 287                 hai->hai_gid = larr->arr_hai.hai_gid;
 288
 289                 hsd->request[found].hal_used_sz +=
 290                                                    cfs_size_round(hai->hai_len);
 291                 hsd->request[found].hal->hal_count++;
 292                 break;
 293         }
 294         case ARS_STARTED: {
 295                 struct cdt_agent_req *car;
 296                 cfs_time_t last;
 297
 298                 /* we search for a running request
 299                  * error may happen if coordinator crashes or stopped
 300                  * with running request
 301                  */
 302                 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
 303                 if (car == NULL) {
 304                         last = larr->arr_req_create;
 305                 } else {
 306                         last = car->car_req_update;
 307                         mdt_cdt_put_request(car);
 308                 }
 309
 310                 /* test if request too long, if yes cancel it
 311                  * the same way the copy tool acknowledge a cancel request */
 312                 if ((last + cdt->cdt_active_req_timeout)
 313                      < cfs_time_current_sec()) {
 314                         struct hsm_progress_kernel pgs;
 315
 316                         dump_llog_agent_req_rec("mdt_coordinator_cb(): "
 317                                                 "request timed out, start "
 318                                                 "cleaning", larr);
 319                         /* a too old cancel request just needs to be removed
 320                          * this can happen, if copy tool does not support cancel
 321                          * for other requests, we have to remove the running
 322                          * request and notify the copytool
 323                          */
 324                         pgs.hpk_fid = larr->arr_hai.hai_fid;
 325                         pgs.hpk_cookie = larr->arr_hai.hai_cookie;
 326                         pgs.hpk_extent = larr->arr_hai.hai_extent;
 327                         pgs.hpk_flags = HP_FLAG_COMPLETED;
 328                         pgs.hpk_errval = ENOSYS;
 329                         pgs.hpk_data_version = 0;
 330                         /* update request state, but do not record in llog, to
 331                          * avoid deadlock on cdt_llog_lock
 332                          */
 333                         rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
 334                         if (rc)
 335                                 CERROR("%s: Cannot cleanup timed out request: "
 336                                        DFID" for cookie "LPX64" action=%s\n",
 337                                        mdt_obd_name(mdt),
 338                                        PFID(&pgs.hpk_fid), pgs.hpk_cookie,
 339                                        hsm_copytool_action2name(
 340                                                      larr->arr_hai.hai_action));
 341
 342                         if (rc == -ENOENT) {
 343                                 /* The request no longer exists, forget
 344                                  * about it, and do not send a cancel request
 345                                  * to the client, for which an error will be
 346                                  * sent back, leading to an endless cycle of
 347                                  * cancellation. */
 348                                 RETURN(LLOG_DEL_RECORD);
 349                         }
 350
 351                         /* add the cookie to the list of record to be
 352                          * canceled by caller */
 353                         if (hsd->max_cookie == (hsd->cookie_cnt - 1)) {
 354                                 __u64 *ptr, *old_ptr;
 355                                 int old_sz, new_sz, new_cnt;
 356
 357                                 /* need to increase vector size */
 358                                 old_sz = sizeof(__u64) * hsd->max_cookie;
 359                                 old_ptr = hsd->cookies;
 360
 361                                 new_cnt = 2 * hsd->max_cookie;
 362                                 new_sz = sizeof(__u64) * new_cnt;
 363
 364                                 OBD_ALLOC(ptr, new_sz);
 365                                 if (!ptr) {
 366                                         CERROR("%s: Cannot allocate memory "
 367                                                "(%d o) for cookie vector\n",
 368                                                mdt_obd_name(mdt), new_sz);
 369                                         RETURN(-ENOMEM);
 370                                 }
 371                                 memcpy(ptr, hsd->cookies, old_sz);
 372                                 hsd->cookies = ptr;
 373                                 hsd->max_cookie = new_cnt;
 374                                 OBD_FREE(old_ptr, old_sz);
 375                         }
 376                         hsd->cookies[hsd->cookie_cnt] =
 377                                                        larr->arr_hai.hai_cookie;
 378                         hsd->cookie_cnt++;
 379                 }
 380                 break;
 381         }
 382         case ARS_FAILED:
 383         case ARS_CANCELED:
 384         case ARS_SUCCEED:
 385                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 386                     cfs_time_current_sec())
 387                         RETURN(LLOG_DEL_RECORD);
 388                 break;
 389         }
 390         RETURN(0);
 391 }
 392
 393 /**
 394  * create /proc entries for coordinator
 395  * \param mdt [IN]
 396  * \retval 0 success
 397  * \retval -ve failure
 398  */
 399 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 400 {
 401         struct coordinator      *cdt = &mdt->mdt_coordinator;
 402         int                      rc = 0;
 403         ENTRY;
 404
 405         /* init /proc entries, failure is not critical */
 406         cdt->cdt_proc_dir = lprocfs_register("hsm",
 407                                              mdt2obd_dev(mdt)->obd_proc_entry,
 408                                              lprocfs_mdt_hsm_vars, mdt);
 409         if (IS_ERR(cdt->cdt_proc_dir)) {
 410                 rc = PTR_ERR(cdt->cdt_proc_dir);
 411                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 412                        " rc=%d\n", mdt_obd_name(mdt), rc);
 413                 cdt->cdt_proc_dir = NULL;
 414                 RETURN(rc);
 415         }
 416
 417         RETURN(0);
 418 }
 419
 420 /**
 421  * remove /proc entries for coordinator
 422  * \param mdt [IN]
 423  */
 424 void  hsm_cdt_procfs_fini(struct mdt_device *mdt)
 425 {
 426         struct coordinator      *cdt = &mdt->mdt_coordinator;
 427
 428         LASSERT(cdt->cdt_state == CDT_STOPPED);
 429         if (cdt->cdt_proc_dir != NULL)
 430                 lprocfs_remove(&cdt->cdt_proc_dir);
 431 }
 432
 433 /**
 434  * get vector of hsm cdt /proc vars
 435  * \param none
 436  * \retval var vector
 437  */
 438 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 439 {
 440         return lprocfs_mdt_hsm_vars;
 441 }
 442
 443 /**
 444  * coordinator thread
 445  * \param data [IN] obd device
 446  * \retval 0 success
 447  * \retval -ve failure
 448  */
 449 static int mdt_coordinator(void *data)
 450 {
 451         struct mdt_thread_info  *mti = data;
 452         struct mdt_device       *mdt = mti->mti_mdt;
 453         struct coordinator      *cdt = &mdt->mdt_coordinator;
 454         struct hsm_scan_data     hsd = { NULL };
 455         int                      rc = 0;
 456         ENTRY;
 457
 458         cdt->cdt_thread.t_flags = SVC_RUNNING;
 459         wake_up(&cdt->cdt_thread.t_ctl_waitq);
 460
 461         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 462                mdt_obd_name(mdt), current_pid());
 463
 464         /* timeouted cookie vector initialization */
 465         hsd.max_cookie = 0;
 466         hsd.cookie_cnt = 0;
 467         hsd.cookies = NULL;
 468         /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
 469          * increases due to a change from /proc we do not overflow the
 470          * hsd.request[] vector
 471          */
 472         hsd.max_requests = cdt->cdt_max_requests;
 473         hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
 474         OBD_ALLOC(hsd.request, hsd.request_sz);
 475         if (!hsd.request)
 476                 GOTO(out, rc = -ENOMEM);
 477
 478         hsd.mti = mti;
 479         obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
 480
 481         while (1) {
 482                 struct l_wait_info lwi;
 483                 int i;
 484
 485                 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
 486                                   NULL, NULL);
 487                 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
 488                              (cdt->cdt_thread.t_flags &
 489                               (SVC_STOPPING|SVC_EVENT)),
 490                              &lwi);
 491
 492                 CDEBUG(D_HSM, "coordinator resumes\n");
 493
 494                 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
 495                     cdt->cdt_state == CDT_STOPPING) {
 496                         cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
 497                         rc = 0;
 498                         break;
 499                 }
 500
 501                 /* wake up before timeout, new work arrives */
 502                 if (cdt->cdt_thread.t_flags & SVC_EVENT)
 503                         cdt->cdt_thread.t_flags &= ~SVC_EVENT;
 504
 505                 /* if coordinator is suspended continue to wait */
 506                 if (cdt->cdt_state == CDT_DISABLE) {
 507                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 508                         continue;
 509                 }
 510
 511                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 512
 513                 if (hsd.max_requests != cdt->cdt_max_requests) {
 514                         /* cdt_max_requests has changed,
 515                          * we need to allocate a new buffer
 516                          */
 517                         OBD_FREE(hsd.request, hsd.request_sz);
 518                         hsd.max_requests = cdt->cdt_max_requests;
 519                         hsd.request_sz =
 520                                    hsd.max_requests * sizeof(*hsd.request);
 521                         OBD_ALLOC(hsd.request, hsd.request_sz);
 522                         if (!hsd.request) {
 523                                 rc = -ENOMEM;
 524                                 break;
 525                         }
 526                 }
 527
 528                 /* create canceled cookie vector for an arbitrary size
 529                  * if needed, vector will grow during llog scan
 530                  */
 531                 hsd.max_cookie = 10;
 532                 hsd.cookie_cnt = 0;
 533                 OBD_ALLOC(hsd.cookies, hsd.max_cookie * sizeof(__u64));
 534                 if (!hsd.cookies) {
 535                         rc = -ENOMEM;
 536                         goto clean_cb_alloc;
 537                 }
 538                 hsd.request_cnt = 0;
 539
 540                 rc = cdt_llog_process(mti->mti_env, mdt,
 541                                       mdt_coordinator_cb, &hsd);
 542                 if (rc < 0)
 543                         goto clean_cb_alloc;
 544
 545                 CDEBUG(D_HSM, "Found %d requests to send and %d"
 546                               " requests to cancel\n",
 547                        hsd.request_cnt, hsd.cookie_cnt);
 548                 /* first we cancel llog records of the timed out requests */
 549                 if (hsd.cookie_cnt > 0) {
 550                         rc = mdt_agent_record_update(mti->mti_env, mdt,
 551                                                      hsd.cookies,
 552                                                      hsd.cookie_cnt,
 553                                                      ARS_CANCELED);
 554                         if (rc)
 555                                 CERROR("%s: mdt_agent_record_update() failed, "
 556                                        "rc=%d, cannot update status to %s "
 557                                        "for %d cookies\n",
 558                                        mdt_obd_name(mdt), rc,
 559                                        agent_req_status2name(ARS_CANCELED),
 560                                        hsd.cookie_cnt);
 561                 }
 562
 563                 if (list_empty(&cdt->cdt_agents)) {
 564                         CDEBUG(D_HSM, "no agent available, "
 565                                       "coordinator sleeps\n");
 566                         goto clean_cb_alloc;
 567                 }
 568
 569                 /* here hsd contains a list of requests to be started */
 570                 for (i = 0; i < hsd.max_requests; i++) {
 571                         struct hsm_action_list  *hal;
 572                         struct hsm_action_item  *hai;
 573                         __u64                   *cookies;
 574                         int                      sz, j;
 575                         enum agent_req_status    status;
 576
 577                         /* still room for work ? */
 578                         if (atomic_read(&cdt->cdt_request_count) ==
 579                             cdt->cdt_max_requests)
 580                                 break;
 581
 582                         if (hsd.request[i].hal == NULL)
 583                                 continue;
 584
 585                         /* found a request, we start it */
 586                         /* kuc payload allocation so we avoid an additionnal
 587                          * allocation in mdt_hsm_agent_send()
 588                          */
 589                         hal = kuc_alloc(hsd.request[i].hal_used_sz,
 590                                         KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
 591                         if (IS_ERR(hal)) {
 592                                 CERROR("%s: Cannot allocate memory (%d o) "
 593                                        "for compound "LPX64"\n",
 594                                        mdt_obd_name(mdt),
 595                                        hsd.request[i].hal_used_sz,
 596                                        hsd.request[i].hal->hal_compound_id);
 597                                 continue;
 598                         }
 599                         memcpy(hal, hsd.request[i].hal,
 600                                hsd.request[i].hal_used_sz);
 601
 602                         rc = mdt_hsm_agent_send(mti, hal, 0);
 603                         /* if failure, we suppose it is temporary
 604                          * if the copy tool failed to do the request
 605                          * it has to use hsm_progress
 606                          */
 607                         status = (rc ? ARS_WAITING : ARS_STARTED);
 608
 609                         /* set up cookie vector to set records status
 610                          * after copy tools start or failed
 611                          */
 612                         sz = hsd.request[i].hal->hal_count * sizeof(__u64);
 613                         OBD_ALLOC(cookies, sz);
 614                         if (cookies == NULL) {
 615                                 CERROR("%s: Cannot allocate memory (%d o) "
 616                                        "for cookies vector "LPX64"\n",
 617                                        mdt_obd_name(mdt), sz,
 618                                        hsd.request[i].hal->hal_compound_id);
 619                                 kuc_free(hal, hsd.request[i].hal_used_sz);
 620                                 continue;
 621                         }
 622                         hai = hai_first(hal);
 623                         for (j = 0; j < hsd.request[i].hal->hal_count; j++) {
 624                                 cookies[j] = hai->hai_cookie;
 625                                 hai = hai_next(hai);
 626                         }
 627
 628                         rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
 629                                                 hsd.request[i].hal->hal_count,
 630                                                 status);
 631                         if (rc)
 632                                 CERROR("%s: mdt_agent_record_update() failed, "
 633                                        "rc=%d, cannot update status to %s "
 634                                        "for %d cookies\n",
 635                                        mdt_obd_name(mdt), rc,
 636                                        agent_req_status2name(status),
 637                                        hsd.request[i].hal->hal_count);
 638
 639                         OBD_FREE(cookies, sz);
 640                         kuc_free(hal, hsd.request[i].hal_used_sz);
 641                 }
 642 clean_cb_alloc:
 643                 /* free cookie vector allocated for/by callback */
 644                 if (hsd.cookies) {
 645                         OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
 646                         hsd.max_cookie = 0;
 647                         hsd.cookie_cnt = 0;
 648                         hsd.cookies = NULL;
 649                 }
 650
 651                 /* free hal allocated by callback */
 652                 for (i = 0; i < hsd.max_requests; i++) {
 653                         if (hsd.request[i].hal) {
 654                                 OBD_FREE(hsd.request[i].hal,
 655                                          hsd.request[i].hal_sz);
 656                                 hsd.request[i].hal_sz = 0;
 657                                 hsd.request[i].hal = NULL;
 658                                 hsd.request_cnt--;
 659                         }
 660                 }
 661                 LASSERT(hsd.request_cnt == 0);
 662
 663                 /* reset callback data */
 664                 memset(hsd.request, 0, hsd.request_sz);
 665         }
 666         EXIT;
 667 out:
 668         if (hsd.request)
 669                 OBD_FREE(hsd.request, hsd.request_sz);
 670
 671         if (hsd.cookies)
 672                 OBD_FREE(hsd.cookies, hsd.max_cookie * sizeof(__u64));
 673
 674         if (cdt->cdt_state == CDT_STOPPING) {
 675                 /* request comes from /proc path, so we need to clean cdt
 676                  * struct */
 677                  mdt_hsm_cdt_stop(mdt);
 678                  mdt->mdt_opts.mo_coordinator = 0;
 679         } else {
 680                 /* request comes from a thread event, generated
 681                  * by mdt_stop_coordinator(), we have to ack
 682                  * and cdt cleaning will be done by event sender
 683                  */
 684                 cdt->cdt_thread.t_flags = SVC_STOPPED;
 685                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
 686         }
 687
 688         if (rc != 0)
 689                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 690                        mdt_obd_name(mdt), current_pid(), rc);
 691         else
 692                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 693                               " no error\n",
 694                        mdt_obd_name(mdt), current_pid());
 695
 696         return rc;
 697 }
 698
 699 /**
 700  * lookup a restore handle by FID
 701  * caller needs to hold cdt_restore_lock
 702  * \param cdt [IN] coordinator
 703  * \param fid [IN] FID
 704  * \retval cdt_restore_handle found
 705  * \retval NULL not found
 706  */
 707 static struct cdt_restore_handle *hsm_restore_hdl_find(struct coordinator *cdt,
 708                                                        const struct lu_fid *fid)
 709 {
 710         struct cdt_restore_handle       *crh;
 711         ENTRY;
 712
 713         list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
 714                 if (lu_fid_eq(&crh->crh_fid, fid))
 715                         RETURN(crh);
 716         }
 717         RETURN(NULL);
 718 }
 719
 720 /**
 721  * data passed to llog_cat_process() callback
 722  * to scan requests and take actions
 723  */
 724 struct hsm_restore_data {
 725         struct mdt_thread_info  *hrd_mti;
 726 };
 727
 728 /**
 729  *  llog_cat_process() callback, used to:
 730  *  - find restore request and allocate the restore handle
 731  * \param env [IN] environment
 732  * \param llh [IN] llog handle
 733  * \param hdr [IN] llog record
 734  * \param data [IN/OUT] cb data = struct hsm_restore_data
 735  * \retval 0 success
 736  * \retval -ve failure
 737  */
 738 static int hsm_restore_cb(const struct lu_env *env,
 739                           struct llog_handle *llh,
 740                           struct llog_rec_hdr *hdr, void *data)
 741 {
 742         struct llog_agent_req_rec       *larr;
 743         struct hsm_restore_data         *hrd;
 744         struct cdt_restore_handle       *crh;
 745         struct hsm_action_item          *hai;
 746         struct mdt_thread_info          *mti;
 747         struct coordinator              *cdt;
 748         struct mdt_object               *child;
 749         int rc;
 750         ENTRY;
 751
 752         hrd = data;
 753         mti = hrd->hrd_mti;
 754         cdt = &mti->mti_mdt->mdt_coordinator;
 755
 756         larr = (struct llog_agent_req_rec *)hdr;
 757         hai = &larr->arr_hai;
 758         if (hai->hai_cookie > cdt->cdt_last_cookie)
 759                 /* update the cookie to avoid collision */
 760                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 761
 762         if (hai->hai_action != HSMA_RESTORE ||
 763             agent_req_in_final_state(larr->arr_status))
 764                 RETURN(0);
 765
 766         /* restore request not in a final state */
 767
 768         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 769         if (crh == NULL)
 770                 RETURN(-ENOMEM);
 771
 772         crh->crh_fid = hai->hai_fid;
 773         /* in V1 all file is restored
 774         crh->extent.start = hai->hai_extent.offset;
 775         crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
 776         */
 777         crh->crh_extent.start = 0;
 778         crh->crh_extent.end = hai->hai_extent.length;
 779         /* get the layout lock */
 780         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 781         child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 782                                      MDS_INODELOCK_LAYOUT);
 783         if (IS_ERR(child))
 784                 GOTO(out, rc = PTR_ERR(child));
 785
 786         rc = 0;
 787         /* we choose to not keep a reference
 788          * on the object during the restore time which can be very long */
 789         mdt_object_put(mti->mti_env, child);
 790
 791         mutex_lock(&cdt->cdt_restore_lock);
 792         list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
 793         mutex_unlock(&cdt->cdt_restore_lock);
 794
 795 out:
 796         RETURN(rc);
 797 }
 798
 799 /**
 800  * restore coordinator state at startup
 801  * the goal is to take a layout lock for each registered restore request
 802  * \param mti [IN] context
 803  */
 804 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 805 {
 806         struct hsm_restore_data  hrd;
 807         int                      rc;
 808         ENTRY;
 809
 810         hrd.hrd_mti = mti;
 811
 812         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
 813                               hsm_restore_cb, &hrd);
 814
 815         RETURN(rc);
 816 }
 817
 818 static int hsm_init_ucred(struct lu_ucred *uc)
 819 {
 820         ENTRY;
 821
 822         uc->uc_valid = UCRED_OLD;
 823         uc->uc_o_uid = 0;
 824         uc->uc_o_gid = 0;
 825         uc->uc_o_fsuid = 0;
 826         uc->uc_o_fsgid = 0;
 827         uc->uc_uid = 0;
 828         uc->uc_gid = 0;
 829         uc->uc_fsuid = 0;
 830         uc->uc_fsgid = 0;
 831         uc->uc_suppgids[0] = -1;
 832         uc->uc_suppgids[1] = -1;
 833         uc->uc_cap = CFS_CAP_FS_MASK;
 834         uc->uc_umask = 0777;
 835         uc->uc_ginfo = NULL;
 836         uc->uc_identity = NULL;
 837
 838         RETURN(0);
 839 }
 840
 841 /**
 842  * wake up coordinator thread
 843  * \param mdt [IN] device
 844  * \retval 0 success
 845  * \retval -ve failure
 846  */
 847 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
 848 {
 849         struct coordinator      *cdt = &mdt->mdt_coordinator;
 850         ENTRY;
 851
 852         if (cdt->cdt_state == CDT_STOPPED)
 853                 RETURN(-ESRCH);
 854
 855         /* wake up coordinator */
 856         cdt->cdt_thread.t_flags = SVC_EVENT;
 857         wake_up(&cdt->cdt_thread.t_ctl_waitq);
 858
 859         RETURN(0);
 860 }
 861
 862 /**
 863  * initialize coordinator struct
 864  * \param mdt [IN] device
 865  * \retval 0 success
 866  * \retval -ve failure
 867  */
 868 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 869 {
 870         struct coordinator      *cdt = &mdt->mdt_coordinator;
 871         struct mdt_thread_info  *cdt_mti = NULL;
 872         int                      rc;
 873         ENTRY;
 874
 875         cdt->cdt_state = CDT_STOPPED;
 876
 877         init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
 878         mutex_init(&cdt->cdt_llog_lock);
 879         init_rwsem(&cdt->cdt_agent_lock);
 880         init_rwsem(&cdt->cdt_request_lock);
 881         mutex_init(&cdt->cdt_restore_lock);
 882
 883         INIT_LIST_HEAD(&cdt->cdt_requests);
 884         INIT_LIST_HEAD(&cdt->cdt_agents);
 885         INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 886
 887         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 888         if (rc < 0)
 889                 RETURN(rc);
 890
 891         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
 892         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
 893         if (rc == 0) {
 894                 lu_context_enter(&cdt->cdt_session);
 895                 cdt->cdt_env.le_ses = &cdt->cdt_session;
 896         } else {
 897                 lu_env_fini(&cdt->cdt_env);
 898                 RETURN(rc);
 899         }
 900
 901         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 902         LASSERT(cdt_mti != NULL);
 903
 904         cdt_mti->mti_env = &cdt->cdt_env;
 905         cdt_mti->mti_mdt = mdt;
 906
 907         hsm_init_ucred(mdt_ucred(cdt_mti));
 908
 909         /* default values for /proc tunnables
 910          * can be override by MGS conf */
 911         cdt->cdt_default_archive_id = 1;
 912         cdt->cdt_grace_delay = 60;
 913         cdt->cdt_loop_period = 10;
 914         cdt->cdt_max_requests = 3;
 915         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 916         cdt->cdt_active_req_timeout = 3600;
 917
 918         /* Initialize cdt_compound_id here to allow its usage for
 919          * delayed requests from RAoLU policy */
 920         atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
 921
 922         /* by default do not remove archives on last unlink */
 923         cdt->cdt_remove_archive_on_last_unlink = false;
 924
 925         RETURN(0);
 926 }
 927
 928 /**
 929  * free a coordinator thread
 930  * \param mdt [IN] device
 931  */
 932 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
 933 {
 934         struct coordinator *cdt = &mdt->mdt_coordinator;
 935         ENTRY;
 936
 937         lu_context_exit(cdt->cdt_env.le_ses);
 938         lu_context_fini(cdt->cdt_env.le_ses);
 939
 940         lu_env_fini(&cdt->cdt_env);
 941
 942         RETURN(0);
 943 }
 944
 945 /**
 946  * start a coordinator thread
 947  * \param mdt [IN] device
 948  * \retval 0 success
 949  * \retval -ve failure
 950  */
 951 int mdt_hsm_cdt_start(struct mdt_device *mdt)
 952 {
 953         struct coordinator      *cdt = &mdt->mdt_coordinator;
 954         int                      rc;
 955         void                    *ptr;
 956         struct mdt_thread_info  *cdt_mti;
 957         struct task_struct      *task;
 958         ENTRY;
 959
 960         /* functions defined but not yet used
 961          * this avoid compilation warning
 962          */
 963         ptr = dump_requests;
 964
 965         if (cdt->cdt_state != CDT_STOPPED) {
 966                 CERROR("%s: Coordinator already started\n",
 967                        mdt_obd_name(mdt));
 968                 RETURN(-EALREADY);
 969         }
 970
 971         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
 972         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 973
 974         cdt->cdt_state = CDT_INIT;
 975
 976         /* just need to be larger than previous one */
 977         /* cdt_last_cookie is protected by cdt_llog_lock */
 978         cdt->cdt_last_cookie = cfs_time_current_sec();
 979         atomic_set(&cdt->cdt_request_count, 0);
 980         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
 981         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
 982         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
 983
 984         /* to avoid deadlock when start is made through /proc
 985          * /proc entries are created by the coordinator thread */
 986
 987         /* set up list of started restore requests */
 988         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 989         rc = mdt_hsm_pending_restore(cdt_mti);
 990         if (rc)
 991                 CERROR("%s: cannot take the layout locks needed"
 992                        " for registered restore: %d\n",
 993                        mdt_obd_name(mdt), rc);
 994
 995         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
 996         if (IS_ERR(task)) {
 997                 rc = PTR_ERR(task);
 998                 cdt->cdt_state = CDT_STOPPED;
 999                 CERROR("%s: error starting coordinator thread: %d\n",
1000                        mdt_obd_name(mdt), rc);
1001                 RETURN(rc);
1002         } else {
1003                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
1004                        mdt_obd_name(mdt));
1005                 rc = 0;
1006         }
1007
1008         wait_event(cdt->cdt_thread.t_ctl_waitq,
1009                        (cdt->cdt_thread.t_flags & SVC_RUNNING));
1010
1011         cdt->cdt_state = CDT_RUNNING;
1012         mdt->mdt_opts.mo_coordinator = 1;
1013         RETURN(0);
1014 }
1015
1016 /**
1017  * stop a coordinator thread
1018  * \param mdt [IN] device
1019  */
1020 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1021 {
1022         struct coordinator              *cdt = &mdt->mdt_coordinator;
1023         struct cdt_agent_req            *car, *tmp1;
1024         struct hsm_agent                *ha, *tmp2;
1025         struct cdt_restore_handle       *crh, *tmp3;
1026         struct mdt_thread_info          *cdt_mti;
1027         ENTRY;
1028
1029         if (cdt->cdt_state == CDT_STOPPED) {
1030                 CERROR("%s: Coordinator already stopped\n",
1031                        mdt_obd_name(mdt));
1032                 RETURN(-EALREADY);
1033         }
1034
1035         if (cdt->cdt_state != CDT_STOPPING) {
1036                 /* stop coordinator thread before cleaning */
1037                 cdt->cdt_thread.t_flags = SVC_STOPPING;
1038                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
1039                 wait_event(cdt->cdt_thread.t_ctl_waitq,
1040                            cdt->cdt_thread.t_flags & SVC_STOPPED);
1041         }
1042         cdt->cdt_state = CDT_STOPPED;
1043
1044         /* start cleaning */
1045         down_write(&cdt->cdt_request_lock);
1046         list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
1047                                  car_request_list) {
1048                 list_del(&car->car_request_list);
1049                 mdt_cdt_free_request(car);
1050         }
1051         up_write(&cdt->cdt_request_lock);
1052
1053         down_write(&cdt->cdt_agent_lock);
1054         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
1055                 list_del(&ha->ha_list);
1056                 OBD_FREE_PTR(ha);
1057         }
1058         up_write(&cdt->cdt_agent_lock);
1059
1060         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1061         mutex_lock(&cdt->cdt_restore_lock);
1062         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
1063                 struct mdt_object       *child;
1064
1065                 /* give back layout lock */
1066                 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
1067                 if (!IS_ERR(child))
1068                         mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
1069
1070                 list_del(&crh->crh_list);
1071
1072                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1073         }
1074         mutex_unlock(&cdt->cdt_restore_lock);
1075
1076         mdt->mdt_opts.mo_coordinator = 0;
1077
1078         RETURN(0);
1079 }
1080
1081 /**
1082  * register all requests from an hal in the memory list
1083  * \param mti [IN] context
1084  * \param hal [IN] request
1085  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1086  *  which is running the CT
1087  * \retval 0 success
1088  * \retval -ve failure
1089  */
1090 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1091                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1092 {
1093         struct mdt_device       *mdt = mti->mti_mdt;
1094         struct coordinator      *cdt = &mdt->mdt_coordinator;
1095         struct hsm_action_item  *hai;
1096         int                      rc = 0, i;
1097         ENTRY;
1098
1099         /* register request in memory list */
1100         hai = hai_first(hal);
1101         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1102                 struct cdt_agent_req *car;
1103
1104                 /* in case of a cancel request, we first mark the ondisk
1105                  * record of the request we want to stop as canceled
1106                  * this does not change the cancel record
1107                  * it will be done when updating the request status
1108                  */
1109                 if (hai->hai_action == HSMA_CANCEL) {
1110                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1111                                                      &hai->hai_cookie,
1112                                                      1, ARS_CANCELED);
1113                         if (rc) {
1114                                 CERROR("%s: mdt_agent_record_update() failed, "
1115                                        "rc=%d, cannot update status to %s "
1116                                        "for cookie "LPX64"\n",
1117                                        mdt_obd_name(mdt), rc,
1118                                        agent_req_status2name(ARS_CANCELED),
1119                                        hai->hai_cookie);
1120                                 GOTO(out, rc);
1121                         }
1122
1123                         /* find the running request to set it canceled */
1124                         car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1125                         if (car != NULL) {
1126                                 car->car_canceled = 1;
1127                                 /* uuid has to be changed to the one running the
1128                                 * request to cancel */
1129                                 *uuid = car->car_uuid;
1130                                 mdt_cdt_put_request(car);
1131                         }
1132                         /* no need to memorize cancel request
1133                          * this also avoid a deadlock when we receive
1134                          * a purge all requests command
1135                          */
1136                         continue;
1137                 }
1138
1139                 if (hai->hai_action == HSMA_ARCHIVE) {
1140                         struct mdt_object *obj;
1141                         struct md_hsm hsm;
1142
1143                         obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1144                         if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1145                                 continue;
1146                         if (IS_ERR(obj))
1147                                 GOTO(out, rc = PTR_ERR(obj));
1148
1149                         hsm.mh_flags |= HS_EXISTS;
1150                         hsm.mh_arch_id = hal->hal_archive_id;
1151                         rc = mdt_hsm_attr_set(mti, obj, &hsm);
1152                         mdt_object_put(mti->mti_env, obj);
1153                         if (rc)
1154                                 GOTO(out, rc);
1155                 }
1156
1157                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1158                                             hal->hal_archive_id, hal->hal_flags,
1159                                             uuid, hai);
1160                 if (IS_ERR(car))
1161                         GOTO(out, rc = PTR_ERR(car));
1162
1163                 rc = mdt_cdt_add_request(cdt, car);
1164                 if (rc != 0)
1165                         mdt_cdt_free_request(car);
1166         }
1167 out:
1168         RETURN(rc);
1169 }
1170
1171 /**
1172  * swap layouts between 2 fids
1173  * \param mti [IN] context
1174  * \param fid1 [IN]
1175  * \param fid2 [IN]
1176  * \param mh_common [IN] MD HSM
1177  */
1178 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1179                             const lustre_fid *fid, const lustre_fid *dfid,
1180                             struct md_hsm *mh_common)
1181 {
1182         struct mdt_device       *mdt = mti->mti_mdt;
1183         struct mdt_object       *child1, *child2;
1184         struct mdt_lock_handle  *lh2;
1185         int                      rc;
1186         ENTRY;
1187
1188         child1 = mdt_object_find(mti->mti_env, mdt, fid);
1189         if (IS_ERR(child1))
1190                 GOTO(out, rc = PTR_ERR(child1));
1191
1192         /* we already have layout lock on FID so take only
1193          * on dfid */
1194         lh2 = &mti->mti_lh[MDT_LH_OLD];
1195         mdt_lock_reg_init(lh2, LCK_EX);
1196         child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1197         if (IS_ERR(child2))
1198                 GOTO(out_child1, rc = PTR_ERR(child2));
1199
1200         /* if copy tool closes the volatile before sending the final
1201          * progress through llapi_hsm_copy_end(), all the objects
1202          * are removed and mdd_swap_layout LBUG */
1203         if (!mdt_object_exists(child2)) {
1204                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1205                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1206                 GOTO(out_child2, rc = -ENOENT);
1207         }
1208         /* Since we only handle restores here, unconditionally use
1209          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1210          * be preserved in case of failure during swap_layout and not
1211          * leave a file in an intermediate but incoherent state.
1212          * But need to setup HSM xattr of data FID before, reuse
1213          * mti and mh presets for FID in hsm_cdt_request_completed(),
1214          * only need to clear RELEASED and DIRTY.
1215          */
1216         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1217         rc = mdt_hsm_attr_set(mti, child2, mh_common);
1218         if (rc == 0)
1219                 rc = mo_swap_layouts(mti->mti_env,
1220                                      mdt_object_child(child1),
1221                                      mdt_object_child(child2),
1222                                      SWAP_LAYOUTS_MDS_HSM);
1223
1224 out_child2:
1225         mdt_object_unlock_put(mti, child2, lh2, 1);
1226 out_child1:
1227         mdt_object_put(mti->mti_env, child1);
1228 out:
1229         RETURN(rc);
1230 }
1231
1232 /**
1233  * update status of a completed request
1234  * \param mti [IN] context
1235  * \param pgs [IN] progress of the copy tool
1236  * \param update_record [IN] update llog record
1237  * \retval 0 success
1238  * \retval -ve failure
1239  */
1240 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1241                                      struct hsm_progress_kernel *pgs,
1242                                      const struct cdt_agent_req *car,
1243                                      enum agent_req_status *status)
1244 {
1245         const struct lu_env     *env = mti->mti_env;
1246         struct mdt_device       *mdt = mti->mti_mdt;
1247         struct coordinator      *cdt = &mdt->mdt_coordinator;
1248         struct mdt_object       *obj = NULL;
1249         int                      cl_flags = 0, rc = 0;
1250         struct md_hsm            mh;
1251         bool                     is_mh_changed;
1252         ENTRY;
1253
1254         /* default is to retry */
1255         *status = ARS_WAITING;
1256
1257         /* find object by FID */
1258         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1259         /* we will update MD HSM only if needed */
1260         is_mh_changed = false;
1261         if (IS_ERR(obj)) {
1262                 /* object removed */
1263                 *status = ARS_SUCCEED;
1264                 goto unlock;
1265         }
1266
1267         /* no need to change mh->mh_arch_id
1268          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1269          */
1270         if (pgs->hpk_errval != 0) {
1271                 switch (pgs->hpk_errval) {
1272                 case ENOSYS:
1273                         /* the copy tool does not support cancel
1274                          * so the cancel request is failed
1275                          * As we cannot distinguish a cancel progress
1276                          * from another action progress (they have the
1277                          * same cookie), we suppose here the CT returns
1278                          * ENOSYS only if does not support cancel
1279                          */
1280                         /* this can also happen when cdt calls it to
1281                          * for a timed out request */
1282                         *status = ARS_FAILED;
1283                         /* to have a cancel event in changelog */
1284                         pgs->hpk_errval = ECANCELED;
1285                         break;
1286                 case ECANCELED:
1287                         /* the request record has already been set to
1288                          * ARS_CANCELED, this set the cancel request
1289                          * to ARS_SUCCEED */
1290                         *status = ARS_SUCCEED;
1291                         break;
1292                 default:
1293                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1294                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1295                                    ARS_FAILED : ARS_WAITING);
1296                         break;
1297                 }
1298
1299                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1300                         CERROR("%s: Request "LPX64" on "DFID
1301                                " failed, error code %d too large\n",
1302                                mdt_obd_name(mdt),
1303                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1304                                pgs->hpk_errval);
1305                         hsm_set_cl_error(&cl_flags,
1306                                          CLF_HSM_ERROVERFLOW);
1307                         rc = -EINVAL;
1308                 } else {
1309                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1310                 }
1311
1312                 switch (car->car_hai->hai_action) {
1313                 case HSMA_ARCHIVE:
1314                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1315                         break;
1316                 case HSMA_RESTORE:
1317                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1318                         break;
1319                 case HSMA_REMOVE:
1320                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1321                         break;
1322                 case HSMA_CANCEL:
1323                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1324                         CERROR("%s: Failed request "LPX64" on "DFID
1325                                " cannot be a CANCEL\n",
1326                                mdt_obd_name(mdt),
1327                                pgs->hpk_cookie,
1328                                PFID(&pgs->hpk_fid));
1329                         break;
1330                 default:
1331                         CERROR("%s: Failed request "LPX64" on "DFID
1332                                " %d is an unknown action\n",
1333                                mdt_obd_name(mdt),
1334                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1335                                car->car_hai->hai_action);
1336                         rc = -EINVAL;
1337                         break;
1338                 }
1339         } else {
1340                 *status = ARS_SUCCEED;
1341                 switch (car->car_hai->hai_action) {
1342                 case HSMA_ARCHIVE:
1343                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1344                         /* set ARCHIVE keep EXIST and clear LOST and
1345                          * DIRTY */
1346                         mh.mh_arch_ver = pgs->hpk_data_version;
1347                         mh.mh_flags |= HS_ARCHIVED;
1348                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1349                         is_mh_changed = true;
1350                         break;
1351                 case HSMA_RESTORE:
1352                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1353
1354                         /* do not clear RELEASED and DIRTY here
1355                          * this will occur in hsm_swap_layouts()
1356                          */
1357
1358                         /* Restoring has changed the file version on
1359                          * disk. */
1360                         mh.mh_arch_ver = pgs->hpk_data_version;
1361                         is_mh_changed = true;
1362                         break;
1363                 case HSMA_REMOVE:
1364                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1365                         /* clear ARCHIVED EXISTS and LOST */
1366                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1367                         is_mh_changed = true;
1368                         break;
1369                 case HSMA_CANCEL:
1370                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1371                         CERROR("%s: Successful request "LPX64
1372                                " on "DFID
1373                                " cannot be a CANCEL\n",
1374                                mdt_obd_name(mdt),
1375                                pgs->hpk_cookie,
1376                                PFID(&pgs->hpk_fid));
1377                         break;
1378                 default:
1379                         CERROR("%s: Successful request "LPX64
1380                                " on "DFID
1381                                " %d is an unknown action\n",
1382                                mdt_obd_name(mdt),
1383                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1384                                car->car_hai->hai_action);
1385                         rc = -EINVAL;
1386                         break;
1387                 }
1388         }
1389
1390         /* rc != 0 means error when analysing action, it may come from
1391          * a crasy CT no need to manage DIRTY
1392          */
1393         if (rc == 0)
1394                 hsm_set_cl_flags(&cl_flags,
1395                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1396
1397         /* unlock is done later, after layout lock management */
1398         if (is_mh_changed)
1399                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1400
1401 unlock:
1402         /* we give back layout lock only if restore was successful or
1403          * if restore was canceled or if policy is to not retry
1404          * in other cases we just unlock the object */
1405         if (car->car_hai->hai_action == HSMA_RESTORE &&
1406             (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1407              cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1408                 struct cdt_restore_handle       *crh;
1409
1410                 /* restore in data FID done, we swap the layouts
1411                  * only if restore is successful */
1412                 if (pgs->hpk_errval == 0) {
1413                         rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1414                                               &car->car_hai->hai_dfid, &mh);
1415                         if (rc) {
1416                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1417                                         *status = ARS_FAILED;
1418                                 pgs->hpk_errval = -rc;
1419                         }
1420                 }
1421                 /* we have to retry, so keep layout lock */
1422                 if (*status == ARS_WAITING)
1423                         GOTO(out, rc);
1424
1425                 /* give back layout lock */
1426                 mutex_lock(&cdt->cdt_restore_lock);
1427                 crh = hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1428                 if (crh != NULL)
1429                         list_del(&crh->crh_list);
1430                 mutex_unlock(&cdt->cdt_restore_lock);
1431                 /* just give back layout lock, we keep
1432                  * the reference which is given back
1433                  * later with the lock for HSM flags */
1434                 if (!IS_ERR(obj) && crh != NULL)
1435                         mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1436
1437                 if (crh != NULL)
1438                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1439         }
1440
1441         GOTO(out, rc);
1442
1443 out:
1444         if (obj != NULL && !IS_ERR(obj)) {
1445                 mo_changelog(env, CL_HSM, cl_flags,
1446                              mdt_object_child(obj));
1447                 mdt_object_put(mti->mti_env, obj);
1448         }
1449
1450         RETURN(rc);
1451 }
1452
1453 /**
1454  * update status of a request
1455  * \param mti [IN] context
1456  * \param pgs [IN] progress of the copy tool
1457  * \param update_record [IN] update llog record
1458  * \retval 0 success
1459  * \retval -ve failure
1460  */
1461 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1462                                  struct hsm_progress_kernel *pgs,
1463                                  const int update_record)
1464 {
1465         struct mdt_device       *mdt = mti->mti_mdt;
1466         struct coordinator      *cdt = &mdt->mdt_coordinator;
1467         struct cdt_agent_req    *car;
1468         int                      rc = 0;
1469         ENTRY;
1470
1471         /* no coordinator started, so we cannot serve requests */
1472         if (cdt->cdt_state == CDT_STOPPED)
1473                 RETURN(-EAGAIN);
1474
1475         /* first do sanity checks */
1476         car = mdt_cdt_update_request(cdt, pgs);
1477         if (IS_ERR(car)) {
1478                 CERROR("%s: Cannot find running request for cookie "LPX64
1479                        " on fid="DFID"\n",
1480                        mdt_obd_name(mdt),
1481                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1482                 if (car == NULL)
1483                         RETURN(-ENOENT);
1484                 RETURN(PTR_ERR(car));
1485         }
1486
1487         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1488                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1489                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1490                       hsm_copytool_action2name(car->car_hai->hai_action),
1491                       pgs->hpk_flags, pgs->hpk_errval,
1492                       PFID(&car->car_hai->hai_fid),
1493                       PFID(&car->car_hai->hai_dfid));
1494
1495         /* progress is done on FID or data FID depending of the action and
1496          * of the copy progress */
1497         /* for restore progress is used to send back the data FID to cdt */
1498         if (car->car_hai->hai_action == HSMA_RESTORE &&
1499             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1500                 car->car_hai->hai_dfid = pgs->hpk_fid;
1501
1502         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1503              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1504             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1505              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1506                 CERROR("%s: Progress on "DFID" for cookie "LPX64
1507                        " does not match request FID "DFID" nor data FID "
1508                        DFID"\n",
1509                        mdt_obd_name(mdt),
1510                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1511                        PFID(&car->car_hai->hai_fid),
1512                        PFID(&car->car_hai->hai_dfid));
1513                 GOTO(out, rc = -EINVAL);
1514         }
1515
1516         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1517                 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1518                        " is not coherent (err=%d and not completed"
1519                        " (flags=%d))\n",
1520                        mdt_obd_name(mdt),
1521                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1522                        hsm_copytool_action2name(car->car_hai->hai_action),
1523                        pgs->hpk_errval, pgs->hpk_flags);
1524                 GOTO(out, rc = -EINVAL);
1525         }
1526
1527         /* now progress is valid */
1528
1529         /* we use a root like ucred */
1530         hsm_init_ucred(mdt_ucred(mti));
1531
1532         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1533                 enum agent_req_status    status;
1534
1535                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1536
1537                 /* remove request from memory list */
1538                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1539
1540                 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1541                               " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1542                        pgs->hpk_cookie,
1543                        hsm_copytool_action2name(car->car_hai->hai_action),
1544                        agent_req_status2name(status));
1545
1546                 if (update_record) {
1547                         int rc1;
1548
1549                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1550                                                      &pgs->hpk_cookie, 1,
1551                                                      status);
1552                         if (rc1)
1553                                 CERROR("%s: mdt_agent_record_update() failed,"
1554                                        " rc=%d, cannot update status to %s"
1555                                        " for cookie "LPX64"\n",
1556                                        mdt_obd_name(mdt), rc1,
1557                                        agent_req_status2name(status),
1558                                        pgs->hpk_cookie);
1559                         rc = (rc != 0 ? rc : rc1);
1560                 }
1561                 /* ct has completed a request, so a slot is available, wakeup
1562                  * cdt to find new work */
1563                 mdt_hsm_cdt_wakeup(mdt);
1564         } else {
1565                 /* if copytool send a progress on a canceled request
1566                  * we inform copytool it should stop
1567                  */
1568                 if (car->car_canceled == 1)
1569                         rc = -ECANCELED;
1570         }
1571         GOTO(out, rc);
1572
1573 out:
1574         /* remove ref got from mdt_cdt_update_request() */
1575         mdt_cdt_put_request(car);
1576
1577         return rc;
1578 }
1579
1580
1581 /**
1582  * data passed to llog_cat_process() callback
1583  * to cancel requests
1584  */
1585 struct hsm_cancel_all_data {
1586         struct mdt_device       *mdt;
1587 };
1588
1589 /**
1590  *  llog_cat_process() callback, used to:
1591  *  - purge all requests
1592  * \param env [IN] environment
1593  * \param llh [IN] llog handle
1594  * \param hdr [IN] llog record
1595  * \param data [IN] cb data = struct hsm_cancel_all_data
1596  * \retval 0 success
1597  * \retval -ve failure
1598  */
1599 static int mdt_cancel_all_cb(const struct lu_env *env,
1600                              struct llog_handle *llh,
1601                              struct llog_rec_hdr *hdr, void *data)
1602 {
1603         struct llog_agent_req_rec       *larr;
1604         struct hsm_cancel_all_data      *hcad;
1605         int                              rc = 0;
1606         ENTRY;
1607
1608         larr = (struct llog_agent_req_rec *)hdr;
1609         hcad = data;
1610         if (larr->arr_status == ARS_WAITING ||
1611             larr->arr_status == ARS_STARTED) {
1612                 larr->arr_status = ARS_CANCELED;
1613                 larr->arr_req_change = cfs_time_current_sec();
1614                 rc = mdt_agent_llog_update_rec(env, hcad->mdt, llh, larr);
1615                 if (rc == 0)
1616                         RETURN(LLOG_DEL_RECORD);
1617         }
1618         RETURN(rc);
1619 }
1620
1621 /**
1622  * cancel all actions
1623  * \param obd [IN] MDT device
1624  */
1625 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1626 {
1627         struct mdt_thread_info          *mti;
1628         struct coordinator              *cdt = &mdt->mdt_coordinator;
1629         struct cdt_agent_req            *car;
1630         struct hsm_action_list          *hal = NULL;
1631         struct hsm_action_item          *hai;
1632         struct hsm_cancel_all_data       hcad;
1633         int                              hal_sz = 0, hal_len, rc;
1634         enum cdt_states                  save_state;
1635         ENTRY;
1636
1637         /* retrieve coordinator context */
1638         mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1639
1640         /* disable coordinator */
1641         save_state = cdt->cdt_state;
1642         cdt->cdt_state = CDT_DISABLE;
1643
1644         /* send cancel to all running requests */
1645         down_read(&cdt->cdt_request_lock);
1646         list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1647                 mdt_cdt_get_request(car);
1648                 /* request is not yet removed from list, it will be done
1649                  * when copytool will return progress
1650                  */
1651
1652                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1653                         mdt_cdt_put_request(car);
1654                         continue;
1655                 }
1656
1657                 /* needed size */
1658                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1659                           cfs_size_round(car->car_hai->hai_len);
1660
1661                 if (hal_len > hal_sz && hal_sz > 0) {
1662                         /* not enough room, free old buffer */
1663                         OBD_FREE(hal, hal_sz);
1664                         hal = NULL;
1665                 }
1666
1667                 /* empty buffer, allocate one */
1668                 if (hal == NULL) {
1669                         hal_sz = hal_len;
1670                         OBD_ALLOC(hal, hal_sz);
1671                         if (hal == NULL) {
1672                                 mdt_cdt_put_request(car);
1673                                 up_read(&cdt->cdt_request_lock);
1674                                 GOTO(out, rc = -ENOMEM);
1675                         }
1676                 }
1677
1678                 hal->hal_version = HAL_VERSION;
1679                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1680                                 MTI_NAME_MAXLEN);
1681                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1682                 hal->hal_compound_id = car->car_compound_id;
1683                 hal->hal_archive_id = car->car_archive_id;
1684                 hal->hal_flags = car->car_flags;
1685                 hal->hal_count = 0;
1686
1687                 hai = hai_first(hal);
1688                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1689                 hai->hai_action = HSMA_CANCEL;
1690                 hal->hal_count = 1;
1691
1692                 /* it is possible to safely call mdt_hsm_agent_send()
1693                  * (ie without a deadlock on cdt_request_lock), because the
1694                  * write lock is taken only if we are not in purge mode
1695                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1696                  *   nor mdt_cdt_remove_request())
1697                  */
1698                 /* no conflict with cdt thread because cdt is disable and we
1699                  * have the request lock */
1700                 mdt_hsm_agent_send(mti, hal, 1);
1701
1702                 mdt_cdt_put_request(car);
1703         }
1704         up_read(&cdt->cdt_request_lock);
1705
1706         if (hal != NULL)
1707                 OBD_FREE(hal, hal_sz);
1708
1709         /* cancel all on-disk records */
1710         hcad.mdt = mdt;
1711
1712         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1713                               mdt_cancel_all_cb, &hcad);
1714 out:
1715         /* enable coordinator */
1716         cdt->cdt_state = save_state;
1717
1718         RETURN(rc);
1719 }
1720
1721 /**
1722  * check if a request is compatible with file status
1723  * \param hai [IN] request description
1724  * \param hal_an [IN] request archive number (not used)
1725  * \param rq_flags [IN] request flags
1726  * \param hsm [IN] file HSM metadata
1727  * \retval boolean
1728  */
1729 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1730                               const int hal_an, const __u64 rq_flags,
1731                               const struct md_hsm *hsm)
1732 {
1733         int      is_compat = false;
1734         int      hsm_flags;
1735         ENTRY;
1736
1737         hsm_flags = hsm->mh_flags;
1738         switch (hai->hai_action) {
1739         case HSMA_ARCHIVE:
1740                 if (!(hsm_flags & HS_NOARCHIVE) &&
1741                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1742                         is_compat = true;
1743                 break;
1744         case HSMA_RESTORE:
1745                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1746                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1747                         is_compat = true;
1748                 break;
1749         case HSMA_REMOVE:
1750                 if (!(hsm_flags & HS_RELEASED) &&
1751                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1752                         is_compat = true;
1753                 break;
1754         case HSMA_CANCEL:
1755                 is_compat = true;
1756                 break;
1757         }
1758         CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1759                       " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1760                       PFID(&hai->hai_fid),
1761                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1762                       hai->hai_extent.offset, hai->hai_extent.length,
1763                       hsm->mh_flags,
1764                       (is_compat ? "compatible" : "uncompatible"));
1765
1766         RETURN(is_compat);
1767 }
1768
1769 /*
1770  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1771  */
1772 static const struct {
1773         __u64            bit;
1774         char            *name;
1775         char            *nickname;
1776 } hsm_policy_names[] = {
1777         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1778         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1779         { 0 },
1780 };
1781
1782 /**
1783  * convert a policy name to a bit
1784  * \param name [IN] policy name
1785  * \retval 0 unknown
1786  * \retval   policy bit
1787  */
1788 static __u64 hsm_policy_str2bit(const char *name)
1789 {
1790         int      i;
1791
1792         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1793                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1794                     strcmp(hsm_policy_names[i].name, name) == 0)
1795                         return hsm_policy_names[i].bit;
1796         return 0;
1797 }
1798
1799 /**
1800  * convert a policy bit field to a string
1801  * \param mask [IN] policy bit field
1802  * \param hexa [IN] print mask before bit names
1803  * \param buffer [OUT] string
1804  * \param count [IN] size of buffer
1805  */
1806 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1807                                 const bool hexa)
1808 {
1809         int      i, j;
1810         __u64    bit;
1811         ENTRY;
1812
1813         if (hexa)
1814                 seq_printf(m, "("LPX64") ", mask);
1815
1816         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1817                 bit = (1ULL << i);
1818
1819                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1820                         if (hsm_policy_names[j].bit == bit)
1821                                 break;
1822                 }
1823                 if (bit & mask)
1824                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1825                 else
1826                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1827         }
1828         /* remove last ' ' */
1829         m->count--;
1830         seq_putc(m, '\0');
1831 }
1832
1833 /* methods to read/write HSM policy flags */
1834 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1835 {
1836         struct mdt_device       *mdt = m->private;
1837         struct coordinator      *cdt = &mdt->mdt_coordinator;
1838         ENTRY;
1839
1840         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1841         RETURN(0);
1842 }
1843
1844 static ssize_t
1845 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1846                          size_t count, loff_t *off)
1847 {
1848         struct seq_file         *m = file->private_data;
1849         struct mdt_device       *mdt = m->private;
1850         struct coordinator      *cdt = &mdt->mdt_coordinator;
1851         char                    *start, *token, sign;
1852         char                    *buf;
1853         __u64                    policy;
1854         __u64                    add_mask, remove_mask, set_mask;
1855         int                      rc;
1856         ENTRY;
1857
1858         if (count + 1 > PAGE_SIZE)
1859                 RETURN(-EINVAL);
1860
1861         OBD_ALLOC(buf, count + 1);
1862         if (buf == NULL)
1863                 RETURN(-ENOMEM);
1864
1865         if (copy_from_user(buf, buffer, count))
1866                 GOTO(out, rc = -EFAULT);
1867
1868         buf[count] = '\0';
1869
1870         start = buf;
1871         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1872                start);
1873
1874         add_mask = remove_mask = set_mask = 0;
1875         do {
1876                 token = strsep(&start, "\n ");
1877                 sign = *token;
1878
1879                 if (sign == '\0')
1880                         continue;
1881
1882                 if (sign == '-' || sign == '+')
1883                         token++;
1884
1885                 policy = hsm_policy_str2bit(token);
1886                 if (policy == 0) {
1887                         CWARN("%s: '%s' is unknown, "
1888                               "supported policies are:\n", mdt_obd_name(mdt),
1889                               token);
1890                         hsm_policy_bit2str(m, 0, false);
1891                         GOTO(out, rc = -EINVAL);
1892                 }
1893                 switch (sign) {
1894                 case '-':
1895                         remove_mask |= policy;
1896                         break;
1897                 case '+':
1898                         add_mask |= policy;
1899                         break;
1900                 default:
1901                         set_mask |= policy;
1902                         break;
1903                 }
1904
1905         } while (start != NULL);
1906
1907         CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1908                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1909
1910         /* if no sign in all string, it is a clear and set
1911          * if some sign found, all unsigned are converted
1912          * to add
1913          * P1 P2 = set to P1 and P2
1914          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1915          */
1916         if (remove_mask == 0 && add_mask == 0) {
1917                 cdt->cdt_policy = set_mask;
1918         } else {
1919                 cdt->cdt_policy |= set_mask | add_mask;
1920                 cdt->cdt_policy &= ~remove_mask;
1921         }
1922
1923         GOTO(out, rc = count);
1924
1925 out:
1926         OBD_FREE(buf, count + 1);
1927         RETURN(rc);
1928 }
1929 LPROC_SEQ_FOPS(mdt_hsm_policy);
1930
1931 #define GENERATE_PROC_METHOD(VAR)                                       \
1932 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
1933 {                                                                       \
1934         struct mdt_device       *mdt = m->private;                      \
1935         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1936         ENTRY;                                                          \
1937                                                                         \
1938         seq_printf(m, LPU64"\n", (__u64)cdt->VAR);                      \
1939         RETURN(0);                                                      \
1940 }                                                                       \
1941 static ssize_t                                                          \
1942 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1943                           size_t count, loff_t *off)                    \
1944                                                                         \
1945 {                                                                       \
1946         struct seq_file         *m = file->private_data;                \
1947         struct mdt_device       *mdt = m->private;                      \
1948         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1949         int                      val;                                   \
1950         int                      rc;                                    \
1951         ENTRY;                                                          \
1952                                                                         \
1953         rc = lprocfs_write_helper(buffer, count, &val);                 \
1954         if (rc)                                                         \
1955                 RETURN(rc);                                             \
1956         if (val > 0) {                                                  \
1957                 cdt->VAR = val;                                         \
1958                 RETURN(count);                                          \
1959         }                                                               \
1960         RETURN(-EINVAL);                                                \
1961 }                                                                       \
1962
1963 GENERATE_PROC_METHOD(cdt_loop_period)
1964 GENERATE_PROC_METHOD(cdt_grace_delay)
1965 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1966 GENERATE_PROC_METHOD(cdt_max_requests)
1967 GENERATE_PROC_METHOD(cdt_default_archive_id)
1968
1969 /*
1970  * procfs write method for MDT/hsm_control
1971  * proc entry is in mdt directory so data is mdt obd_device pointer
1972  */
1973 #define CDT_ENABLE_CMD   "enabled"
1974 #define CDT_STOP_CMD     "shutdown"
1975 #define CDT_DISABLE_CMD  "disabled"
1976 #define CDT_PURGE_CMD    "purge"
1977 #define CDT_HELP_CMD     "help"
1978 #define CDT_MAX_CMD_LEN  10
1979
1980 ssize_t
1981 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1982                               size_t count, loff_t *off)
1983 {
1984         struct seq_file         *m = file->private_data;
1985         struct obd_device       *obd = m->private;
1986         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
1987         struct coordinator      *cdt = &(mdt->mdt_coordinator);
1988         int                      rc, usage = 0;
1989         char                     kernbuf[CDT_MAX_CMD_LEN];
1990         ENTRY;
1991
1992         if (count == 0 || count >= sizeof(kernbuf))
1993                 RETURN(-EINVAL);
1994
1995         if (copy_from_user(kernbuf, buffer, count))
1996                 RETURN(-EFAULT);
1997         kernbuf[count] = 0;
1998
1999         if (kernbuf[count - 1] == '\n')
2000                 kernbuf[count - 1] = 0;
2001
2002         rc = 0;
2003         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
2004                 if (cdt->cdt_state == CDT_DISABLE) {
2005                         cdt->cdt_state = CDT_RUNNING;
2006                         mdt_hsm_cdt_wakeup(mdt);
2007                 } else {
2008                         rc = mdt_hsm_cdt_start(mdt);
2009                 }
2010         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
2011                 if ((cdt->cdt_state == CDT_STOPPING) ||
2012                     (cdt->cdt_state == CDT_STOPPED)) {
2013                         CERROR("%s: Coordinator already stopped\n",
2014                                mdt_obd_name(mdt));
2015                         rc = -EALREADY;
2016                 } else {
2017                         cdt->cdt_state = CDT_STOPPING;
2018                 }
2019         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
2020                 if ((cdt->cdt_state == CDT_STOPPING) ||
2021                     (cdt->cdt_state == CDT_STOPPED)) {
2022                         CERROR("%s: Coordinator is stopped\n",
2023                                mdt_obd_name(mdt));
2024                         rc = -EINVAL;
2025                 } else {
2026                         cdt->cdt_state = CDT_DISABLE;
2027                 }
2028         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
2029                 rc = hsm_cancel_all_actions(mdt);
2030         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
2031                 usage = 1;
2032         } else {
2033                 usage = 1;
2034                 rc = -EINVAL;
2035         }
2036
2037         if (usage == 1)
2038                 CERROR("%s: Valid coordinator control commands are: "
2039                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
2040                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
2041                        CDT_PURGE_CMD, CDT_HELP_CMD);
2042
2043         if (rc)
2044                 RETURN(rc);
2045
2046         RETURN(count);
2047 }
2048
2049 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
2050 {
2051         struct obd_device       *obd = m->private;
2052         struct coordinator      *cdt;
2053         ENTRY;
2054
2055         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
2056
2057         if (cdt->cdt_state == CDT_INIT)
2058                 seq_printf(m, "init\n");
2059         else if (cdt->cdt_state == CDT_RUNNING)
2060                 seq_printf(m, "enabled\n");
2061         else if (cdt->cdt_state == CDT_STOPPING)
2062                 seq_printf(m, "stopping\n");
2063         else if (cdt->cdt_state == CDT_STOPPED)
2064                 seq_printf(m, "stopped\n");
2065         else if (cdt->cdt_state == CDT_DISABLE)
2066                 seq_printf(m, "disabled\n");
2067         else
2068                 seq_printf(m, "unknown\n");
2069
2070         RETURN(0);
2071 }
2072
2073 static int
2074 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
2075 {
2076         int i, rc = 0;
2077         ENTRY;
2078
2079         for (i = 0; i < 8 * sizeof(mask); i++) {
2080                 if (mask & (1UL << i))
2081                         rc += seq_printf(m, "%s%s", rc == 0 ? "" : " ",
2082                                         hsm_copytool_action2name(i));
2083         }
2084         rc += seq_printf(m, "\n");
2085
2086         RETURN(rc);
2087 }
2088
2089 static int
2090 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2091 {
2092         struct mdt_device *mdt = m->private;
2093         struct coordinator *cdt = &mdt->mdt_coordinator;
2094
2095         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2096 }
2097
2098 static int
2099 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2100 {
2101         struct mdt_device *mdt = m->private;
2102         struct coordinator *cdt = &mdt->mdt_coordinator;
2103
2104         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2105 }
2106
2107 static int
2108 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2109 {
2110         struct mdt_device *mdt = m->private;
2111         struct coordinator *cdt = &mdt->mdt_coordinator;
2112
2113         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2114 }
2115
2116 static inline enum hsm_copytool_action
2117 hsm_copytool_name2action(const char *name)
2118 {
2119         if (strcasecmp(name, "NOOP") == 0)
2120                 return HSMA_NONE;
2121         else if (strcasecmp(name, "ARCHIVE") == 0)
2122                 return HSMA_ARCHIVE;
2123         else if (strcasecmp(name, "RESTORE") == 0)
2124                 return HSMA_RESTORE;
2125         else if (strcasecmp(name, "REMOVE") == 0)
2126                 return HSMA_REMOVE;
2127         else if (strcasecmp(name, "CANCEL") == 0)
2128                 return HSMA_CANCEL;
2129         else
2130                 return -1;
2131 }
2132
2133 static ssize_t
2134 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2135                             size_t user_count, __u64 *mask)
2136 {
2137         char *buf, *pos, *name;
2138         size_t buf_size;
2139         __u64 new_mask = 0;
2140         int rc;
2141         ENTRY;
2142
2143         if (!(user_count < 4096))
2144                 RETURN(-ENOMEM);
2145
2146         buf_size = user_count + 1;
2147
2148         OBD_ALLOC(buf, buf_size);
2149         if (buf == NULL)
2150                 RETURN(-ENOMEM);
2151
2152         if (copy_from_user(buf, user_buf, buf_size - 1))
2153                 GOTO(out, rc = -EFAULT);
2154
2155         buf[buf_size - 1] = '\0';
2156
2157         pos = buf;
2158         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2159                 int action;
2160
2161                 if (*name == '\0')
2162                         continue;
2163
2164                 action = hsm_copytool_name2action(name);
2165                 if (action < 0)
2166                         GOTO(out, rc = -EINVAL);
2167
2168                 new_mask |= (1UL << action);
2169         }
2170
2171         *mask = new_mask;
2172         rc = user_count;
2173 out:
2174         OBD_FREE(buf, buf_size);
2175
2176         RETURN(rc);
2177 }
2178
2179 static ssize_t
2180 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2181                                         size_t count, loff_t *off)
2182 {
2183         struct seq_file         *m = file->private_data;
2184         struct mdt_device       *mdt = m->private;
2185         struct coordinator *cdt = &mdt->mdt_coordinator;
2186
2187         return mdt_write_hsm_request_mask(file, buf, count,
2188                                            &cdt->cdt_user_request_mask);
2189 }
2190
2191 static ssize_t
2192 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2193                                         size_t count, loff_t *off)
2194 {
2195         struct seq_file         *m = file->private_data;
2196         struct mdt_device       *mdt = m->private;
2197         struct coordinator      *cdt = &mdt->mdt_coordinator;
2198
2199         return mdt_write_hsm_request_mask(file, buf, count,
2200                                            &cdt->cdt_group_request_mask);
2201 }
2202
2203 static ssize_t
2204 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2205                                         size_t count, loff_t *off)
2206 {
2207         struct seq_file         *m = file->private_data;
2208         struct mdt_device       *mdt = m->private;
2209         struct coordinator      *cdt = &mdt->mdt_coordinator;
2210
2211         return mdt_write_hsm_request_mask(file, buf, count,
2212                                            &cdt->cdt_other_request_mask);
2213 }
2214
2215 static int mdt_hsm_cdt_raolu_seq_show(struct seq_file *m, void *data)
2216 {
2217         struct mdt_device       *mdt = m->private;
2218         struct coordinator      *cdt = &mdt->mdt_coordinator;
2219         ENTRY;
2220
2221         seq_printf(m, "%d\n", (int)cdt->cdt_remove_archive_on_last_unlink);
2222         RETURN(0);
2223 }
2224
2225 static ssize_t
2226 mdt_hsm_cdt_raolu_seq_write(struct file *file, const char __user *buffer,
2227                           size_t count, loff_t *off)
2228
2229 {
2230         struct seq_file         *m = file->private_data;
2231         struct mdt_device       *mdt = m->private;
2232         struct coordinator      *cdt = &mdt->mdt_coordinator;
2233         int                      val;
2234         int                      rc;
2235         ENTRY;
2236
2237         rc = lprocfs_write_helper(buffer, count, &val);
2238         if (rc < 0)
2239                 RETURN(rc);
2240
2241         cdt->cdt_remove_archive_on_last_unlink = val;
2242         RETURN(count);
2243 }
2244
2245 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2246 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2247 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2248 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2249 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2250 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2251 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2252 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2253 LPROC_SEQ_FOPS(mdt_hsm_cdt_raolu);
2254
2255 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2256         { .name =       "agents",
2257           .fops =       &mdt_hsm_agent_fops                     },
2258         { .name =       "actions",
2259           .fops =       &mdt_hsm_actions_fops,
2260           .proc_mode =  0444                                    },
2261         { .name =       "default_archive_id",
2262           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2263         { .name =       "grace_delay",
2264           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2265         { .name =       "loop_period",
2266           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2267         { .name =       "max_requests",
2268           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2269         { .name =       "policy",
2270           .fops =       &mdt_hsm_policy_fops                    },
2271         { .name =       "active_request_timeout",
2272           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2273         { .name =       "active_requests",
2274           .fops =       &mdt_hsm_active_requests_fops           },
2275         { .name =       "user_request_mask",
2276           .fops =       &mdt_hsm_user_request_mask_fops,        },
2277         { .name =       "group_request_mask",
2278           .fops =       &mdt_hsm_group_request_mask_fops,       },
2279         { .name =       "other_request_mask",
2280           .fops =       &mdt_hsm_other_request_mask_fops,       },
2281         { .name =       "remove_archive_on_last_unlink",
2282           .fops =       &mdt_hsm_cdt_raolu_fops,                },
2283         { 0 }
2284 };