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 };
 146
 147 /**
 148  *  llog_cat_process() callback, used to:
 149  *  - find waiting request and start action
 150  *  - purge canceled and done requests
 151  * \param env [IN] environment
 152  * \param llh [IN] llog handle
 153  * \param hdr [IN] llog record
 154  * \param data [IN/OUT] cb data = struct hsm_scan_data
 155  * \retval 0 success
 156  * \retval -ve failure
 157  */
 158 static int mdt_coordinator_cb(const struct lu_env *env,
 159                               struct llog_handle *llh,
 160                               struct llog_rec_hdr *hdr,
 161                               void *data)
 162 {
 163         struct llog_agent_req_rec       *larr;
 164         struct hsm_scan_data            *hsd;
 165         struct hsm_action_item          *hai;
 166         struct mdt_device               *mdt;
 167         struct coordinator              *cdt;
 168         int                              rc;
 169         ENTRY;
 170
 171         hsd = data;
 172         mdt = hsd->mti->mti_mdt;
 173         cdt = &mdt->mdt_coordinator;
 174
 175         larr = (struct llog_agent_req_rec *)hdr;
 176         dump_llog_agent_req_rec("mdt_coordinator_cb(): ", larr);
 177         switch (larr->arr_status) {
 178         case ARS_WAITING: {
 179                 int i, empty_slot, found;
 180
 181                 /* Are agents full? */
 182                 if (atomic_read(&cdt->cdt_request_count) >=
 183                     cdt->cdt_max_requests)
 184                         break;
 185
 186                 /* first search whether the request is found in the list we
 187                  * have built and if there is room in the request vector */
 188                 empty_slot = -1;
 189                 found = -1;
 190                 for (i = 0; i < hsd->max_requests &&
 191                             (empty_slot == -1 || found == -1); i++) {
 192                         if (hsd->request[i].hal == NULL) {
 193                                 empty_slot = i;
 194                                 continue;
 195                         }
 196                         if (hsd->request[i].hal->hal_compound_id ==
 197                                 larr->arr_compound_id) {
 198                                 found = i;
 199                                 continue;
 200                         }
 201                 }
 202                 if (found == -1 && empty_slot == -1)
 203                         /* unknown request and no more room for new request,
 204                          * continue scan for to find other entries for
 205                          * already found request
 206                          */
 207                         RETURN(0);
 208
 209                 if (found == -1) {
 210                         struct hsm_action_list *hal;
 211
 212                         /* request is not already known */
 213                         /* allocates hai vector size just needs to be large
 214                          * enough */
 215                         hsd->request[empty_slot].hal_sz =
 216                                      sizeof(*hsd->request[empty_slot].hal) +
 217                                      cfs_size_round(MTI_NAME_MAXLEN+1) +
 218                                      2 * cfs_size_round(larr->arr_hai.hai_len);
 219                         OBD_ALLOC(hal, hsd->request[empty_slot].hal_sz);
 220                         if (!hal) {
 221                                 CERROR("%s: Cannot allocate memory (%d o)"
 222                                        "for compound "LPX64"\n",
 223                                        mdt_obd_name(mdt),
 224                                        hsd->request[i].hal_sz,
 225                                        larr->arr_compound_id);
 226                                 RETURN(-ENOMEM);
 227                         }
 228                         hal->hal_version = HAL_VERSION;
 229                         strlcpy(hal->hal_fsname, hsd->fs_name,
 230                                 MTI_NAME_MAXLEN + 1);
 231                         hal->hal_compound_id = larr->arr_compound_id;
 232                         hal->hal_archive_id = larr->arr_archive_id;
 233                         hal->hal_flags = larr->arr_flags;
 234                         hal->hal_count = 0;
 235                         hsd->request[empty_slot].hal_used_sz = hal_size(hal);
 236                         hsd->request[empty_slot].hal = hal;
 237                         hsd->request_cnt++;
 238                         found = empty_slot;
 239                         hai = hai_first(hal);
 240                 } else {
 241                         /* request is known */
 242                         /* we check if record archive num is the same as the
 243                          * known request, if not we will serve it in multiple
 244                          * time because we do not know if the agent can serve
 245                          * multiple backend
 246                          * a use case is a compound made of multiple restore
 247                          * where the files are not archived in the same backend
 248                          */
 249                         if (larr->arr_archive_id !=
 250                             hsd->request[found].hal->hal_archive_id)
 251                                 RETURN(0);
 252
 253                         if (hsd->request[found].hal_sz <
 254                             hsd->request[found].hal_used_sz +
 255                              cfs_size_round(larr->arr_hai.hai_len)) {
 256                                 /* Not enough room, need an extension */
 257                                 void *hal_buffer;
 258                                 int sz;
 259
 260                                 sz = 2 * hsd->request[found].hal_sz;
 261                                 OBD_ALLOC(hal_buffer, sz);
 262                                 if (!hal_buffer) {
 263                                         CERROR("%s: Cannot allocate memory "
 264                                                "(%d o) for compound "LPX64"\n",
 265                                                mdt_obd_name(mdt), sz,
 266                                                larr->arr_compound_id);
 267                                         RETURN(-ENOMEM);
 268                                 }
 269                                 memcpy(hal_buffer, hsd->request[found].hal,
 270                                        hsd->request[found].hal_used_sz);
 271                                 OBD_FREE(hsd->request[found].hal,
 272                                          hsd->request[found].hal_sz);
 273                                 hsd->request[found].hal = hal_buffer;
 274                                 hsd->request[found].hal_sz = sz;
 275                         }
 276                         hai = hai_first(hsd->request[found].hal);
 277                         for (i = 0; i < hsd->request[found].hal->hal_count;
 278                              i++)
 279                                 hai = hai_next(hai);
 280                 }
 281                 memcpy(hai, &larr->arr_hai, larr->arr_hai.hai_len);
 282                 hai->hai_cookie = larr->arr_hai.hai_cookie;
 283                 hai->hai_gid = larr->arr_hai.hai_gid;
 284
 285                 hsd->request[found].hal_used_sz +=
 286                                                    cfs_size_round(hai->hai_len);
 287                 hsd->request[found].hal->hal_count++;
 288                 break;
 289         }
 290         case ARS_STARTED: {
 291                 struct hsm_progress_kernel pgs;
 292                 struct cdt_agent_req *car;
 293                 cfs_time_t now = cfs_time_current_sec();
 294                 cfs_time_t last;
 295
 296                 /* we search for a running request
 297                  * error may happen if coordinator crashes or stopped
 298                  * with running request
 299                  */
 300                 car = mdt_cdt_find_request(cdt, larr->arr_hai.hai_cookie, NULL);
 301                 if (car == NULL) {
 302                         last = larr->arr_req_create;
 303                 } else {
 304                         last = car->car_req_update;
 305                         mdt_cdt_put_request(car);
 306                 }
 307
 308                 /* test if request too long, if yes cancel it
 309                  * the same way the copy tool acknowledge a cancel request */
 310                 if (now <= last + cdt->cdt_active_req_timeout)
 311                         RETURN(0);
 312
 313                 dump_llog_agent_req_rec("request timed out, start cleaning",
 314                                         larr);
 315                 /* a too old cancel request just needs to be removed
 316                  * this can happen, if copy tool does not support
 317                  * cancel for other requests, we have to remove the
 318                  * running request and notify the copytool */
 319                 pgs.hpk_fid = larr->arr_hai.hai_fid;
 320                 pgs.hpk_cookie = larr->arr_hai.hai_cookie;
 321                 pgs.hpk_extent = larr->arr_hai.hai_extent;
 322                 pgs.hpk_flags = HP_FLAG_COMPLETED;
 323                 pgs.hpk_errval = ENOSYS;
 324                 pgs.hpk_data_version = 0;
 325
 326                 /* update request state, but do not record in llog, to
 327                  * avoid deadlock on cdt_llog_lock */
 328                 rc = mdt_hsm_update_request_state(hsd->mti, &pgs, 0);
 329                 if (rc)
 330                         CERROR("%s: cannot cleanup timed out request: "
 331                                DFID" for cookie "LPX64" action=%s\n",
 332                                mdt_obd_name(mdt),
 333                                PFID(&pgs.hpk_fid), pgs.hpk_cookie,
 334                                hsm_copytool_action2name(
 335                                        larr->arr_hai.hai_action));
 336
 337                 if (rc == -ENOENT) {
 338                         /* The request no longer exists, forget
 339                          * about it, and do not send a cancel request
 340                          * to the client, for which an error will be
 341                          * sent back, leading to an endless cycle of
 342                          * cancellation. */
 343                         RETURN(LLOG_DEL_RECORD);
 344                 }
 345
 346                 /* XXX A cancel request cannot be cancelled. */
 347                 if (larr->arr_hai.hai_action == HSMA_CANCEL)
 348                         RETURN(0);
 349
 350                 larr->arr_status = ARS_CANCELED;
 351                 larr->arr_req_change = now;
 352                 rc = llog_write(hsd->mti->mti_env, llh, hdr, hdr->lrh_index);
 353                 if (rc < 0)
 354                         CERROR("%s: cannot update agent log: rc = %d\n",
 355                                mdt_obd_name(mdt), rc);
 356                 break;
 357         }
 358         case ARS_FAILED:
 359         case ARS_CANCELED:
 360         case ARS_SUCCEED:
 361                 if ((larr->arr_req_change + cdt->cdt_grace_delay) <
 362                     cfs_time_current_sec())
 363                         RETURN(LLOG_DEL_RECORD);
 364                 break;
 365         }
 366         RETURN(0);
 367 }
 368
 369 /**
 370  * create /proc entries for coordinator
 371  * \param mdt [IN]
 372  * \retval 0 success
 373  * \retval -ve failure
 374  */
 375 int hsm_cdt_procfs_init(struct mdt_device *mdt)
 376 {
 377         struct coordinator      *cdt = &mdt->mdt_coordinator;
 378         int                      rc = 0;
 379         ENTRY;
 380
 381         /* init /proc entries, failure is not critical */
 382         cdt->cdt_proc_dir = lprocfs_register("hsm",
 383                                              mdt2obd_dev(mdt)->obd_proc_entry,
 384                                              lprocfs_mdt_hsm_vars, mdt);
 385         if (IS_ERR(cdt->cdt_proc_dir)) {
 386                 rc = PTR_ERR(cdt->cdt_proc_dir);
 387                 CERROR("%s: Cannot create 'hsm' directory in mdt proc dir,"
 388                        " rc=%d\n", mdt_obd_name(mdt), rc);
 389                 cdt->cdt_proc_dir = NULL;
 390                 RETURN(rc);
 391         }
 392
 393         RETURN(0);
 394 }
 395
 396 /**
 397  * remove /proc entries for coordinator
 398  * \param mdt [IN]
 399  */
 400 void  hsm_cdt_procfs_fini(struct mdt_device *mdt)
 401 {
 402         struct coordinator      *cdt = &mdt->mdt_coordinator;
 403
 404         LASSERT(cdt->cdt_state == CDT_STOPPED);
 405         if (cdt->cdt_proc_dir != NULL)
 406                 lprocfs_remove(&cdt->cdt_proc_dir);
 407 }
 408
 409 /**
 410  * get vector of hsm cdt /proc vars
 411  * \param none
 412  * \retval var vector
 413  */
 414 struct lprocfs_vars *hsm_cdt_get_proc_vars(void)
 415 {
 416         return lprocfs_mdt_hsm_vars;
 417 }
 418
 419 /**
 420  * coordinator thread
 421  * \param data [IN] obd device
 422  * \retval 0 success
 423  * \retval -ve failure
 424  */
 425 static int mdt_coordinator(void *data)
 426 {
 427         struct mdt_thread_info  *mti = data;
 428         struct mdt_device       *mdt = mti->mti_mdt;
 429         struct coordinator      *cdt = &mdt->mdt_coordinator;
 430         struct hsm_scan_data     hsd = { NULL };
 431         int                      rc = 0;
 432         ENTRY;
 433
 434         cdt->cdt_thread.t_flags = SVC_RUNNING;
 435         wake_up(&cdt->cdt_thread.t_ctl_waitq);
 436
 437         CDEBUG(D_HSM, "%s: coordinator thread starting, pid=%d\n",
 438                mdt_obd_name(mdt), current_pid());
 439
 440         /* we use a copy of cdt_max_requests in the cb, so if cdt_max_requests
 441          * increases due to a change from /proc we do not overflow the
 442          * hsd.request[] vector
 443          */
 444         hsd.max_requests = cdt->cdt_max_requests;
 445         hsd.request_sz = hsd.max_requests * sizeof(*hsd.request);
 446         OBD_ALLOC(hsd.request, hsd.request_sz);
 447         if (!hsd.request)
 448                 GOTO(out, rc = -ENOMEM);
 449
 450         hsd.mti = mti;
 451         obd_uuid2fsname(hsd.fs_name, mdt_obd_name(mdt), MTI_NAME_MAXLEN);
 452
 453         while (1) {
 454                 struct l_wait_info lwi;
 455                 int i;
 456
 457                 lwi = LWI_TIMEOUT(cfs_time_seconds(cdt->cdt_loop_period),
 458                                   NULL, NULL);
 459                 l_wait_event(cdt->cdt_thread.t_ctl_waitq,
 460                              (cdt->cdt_thread.t_flags &
 461                               (SVC_STOPPING|SVC_EVENT)),
 462                              &lwi);
 463
 464                 CDEBUG(D_HSM, "coordinator resumes\n");
 465
 466                 if (cdt->cdt_thread.t_flags & SVC_STOPPING ||
 467                     cdt->cdt_state == CDT_STOPPING) {
 468                         cdt->cdt_thread.t_flags &= ~SVC_STOPPING;
 469                         rc = 0;
 470                         break;
 471                 }
 472
 473                 /* wake up before timeout, new work arrives */
 474                 if (cdt->cdt_thread.t_flags & SVC_EVENT)
 475                         cdt->cdt_thread.t_flags &= ~SVC_EVENT;
 476
 477                 /* if coordinator is suspended continue to wait */
 478                 if (cdt->cdt_state == CDT_DISABLE) {
 479                         CDEBUG(D_HSM, "disable state, coordinator sleeps\n");
 480                         continue;
 481                 }
 482
 483                 CDEBUG(D_HSM, "coordinator starts reading llog\n");
 484
 485                 if (hsd.max_requests != cdt->cdt_max_requests) {
 486                         /* cdt_max_requests has changed,
 487                          * we need to allocate a new buffer
 488                          */
 489                         OBD_FREE(hsd.request, hsd.request_sz);
 490                         hsd.max_requests = cdt->cdt_max_requests;
 491                         hsd.request_sz =
 492                                    hsd.max_requests * sizeof(*hsd.request);
 493                         OBD_ALLOC(hsd.request, hsd.request_sz);
 494                         if (!hsd.request) {
 495                                 rc = -ENOMEM;
 496                                 break;
 497                         }
 498                 }
 499
 500                 hsd.request_cnt = 0;
 501
 502                 rc = cdt_llog_process(mti->mti_env, mdt,
 503                                       mdt_coordinator_cb, &hsd);
 504                 if (rc < 0)
 505                         goto clean_cb_alloc;
 506
 507                 CDEBUG(D_HSM, "found %d requests to send\n", hsd.request_cnt);
 508
 509                 if (list_empty(&cdt->cdt_agents)) {
 510                         CDEBUG(D_HSM, "no agent available, "
 511                                       "coordinator sleeps\n");
 512                         goto clean_cb_alloc;
 513                 }
 514
 515                 /* here hsd contains a list of requests to be started */
 516                 for (i = 0; i < hsd.max_requests; i++) {
 517                         struct hsm_action_list  *hal = hsd.request[i].hal;
 518                         struct hsm_action_item  *hai;
 519                         __u64                   *cookies;
 520                         int                      sz, j;
 521                         enum agent_req_status    status;
 522
 523                         /* still room for work ? */
 524                         if (atomic_read(&cdt->cdt_request_count) >=
 525                             cdt->cdt_max_requests)
 526                                 break;
 527
 528                         if (hal == NULL)
 529                                 continue;
 530
 531                         /* found a request, we start it */
 532                         rc = mdt_hsm_agent_send(mti, hal, 0);
 533                         /* if failure, we suppose it is temporary
 534                          * if the copy tool failed to do the request
 535                          * it has to use hsm_progress
 536                          */
 537                         status = (rc ? ARS_WAITING : ARS_STARTED);
 538
 539                         /* set up cookie vector to set records status
 540                          * after copy tools start or failed
 541                          */
 542                         sz = hal->hal_count * sizeof(__u64);
 543                         OBD_ALLOC(cookies, sz);
 544                         if (cookies == NULL) {
 545                                 CERROR("%s: Cannot allocate memory (%d o) "
 546                                        "for cookies vector "LPX64"\n",
 547                                        mdt_obd_name(mdt), sz,
 548                                        hal->hal_compound_id);
 549                                 continue;
 550                         }
 551                         hai = hai_first(hal);
 552                         for (j = 0; j < hal->hal_count; j++) {
 553                                 cookies[j] = hai->hai_cookie;
 554                                 hai = hai_next(hai);
 555                         }
 556
 557                         rc = mdt_agent_record_update(mti->mti_env, mdt, cookies,
 558                                                      hal->hal_count, status);
 559                         if (rc)
 560                                 CERROR("%s: mdt_agent_record_update() failed, "
 561                                        "rc=%d, cannot update status to %s "
 562                                        "for %d cookies\n",
 563                                        mdt_obd_name(mdt), rc,
 564                                        agent_req_status2name(status),
 565                                        hal->hal_count);
 566
 567                         OBD_FREE(cookies, sz);
 568                 }
 569 clean_cb_alloc:
 570                 /* free hal allocated by callback */
 571                 for (i = 0; i < hsd.max_requests; i++) {
 572                         if (hsd.request[i].hal) {
 573                                 OBD_FREE(hsd.request[i].hal,
 574                                          hsd.request[i].hal_sz);
 575                                 hsd.request[i].hal_sz = 0;
 576                                 hsd.request[i].hal = NULL;
 577                                 hsd.request_cnt--;
 578                         }
 579                 }
 580                 LASSERT(hsd.request_cnt == 0);
 581
 582                 /* reset callback data */
 583                 memset(hsd.request, 0, hsd.request_sz);
 584         }
 585         EXIT;
 586 out:
 587         if (hsd.request)
 588                 OBD_FREE(hsd.request, hsd.request_sz);
 589
 590         if (cdt->cdt_state == CDT_STOPPING) {
 591                 /* request comes from /proc path, so we need to clean cdt
 592                  * struct */
 593                  mdt_hsm_cdt_stop(mdt);
 594                  mdt->mdt_opts.mo_coordinator = 0;
 595         } else {
 596                 /* request comes from a thread event, generated
 597                  * by mdt_stop_coordinator(), we have to ack
 598                  * and cdt cleaning will be done by event sender
 599                  */
 600                 cdt->cdt_thread.t_flags = SVC_STOPPED;
 601                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
 602         }
 603
 604         if (rc != 0)
 605                 CERROR("%s: coordinator thread exiting, process=%d, rc=%d\n",
 606                        mdt_obd_name(mdt), current_pid(), rc);
 607         else
 608                 CDEBUG(D_HSM, "%s: coordinator thread exiting, process=%d,"
 609                               " no error\n",
 610                        mdt_obd_name(mdt), current_pid());
 611
 612         return rc;
 613 }
 614
 615 /**
 616  * lookup a restore handle by FID
 617  * caller needs to hold cdt_restore_lock
 618  * \param cdt [IN] coordinator
 619  * \param fid [IN] FID
 620  * \retval cdt_restore_handle found
 621  * \retval NULL not found
 622  */
 623 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
 624                                                        const struct lu_fid *fid)
 625 {
 626         struct cdt_restore_handle       *crh;
 627         ENTRY;
 628
 629         list_for_each_entry(crh, &cdt->cdt_restore_hdl, crh_list) {
 630                 if (lu_fid_eq(&crh->crh_fid, fid))
 631                         RETURN(crh);
 632         }
 633         RETURN(NULL);
 634 }
 635
 636 /**
 637  * data passed to llog_cat_process() callback
 638  * to scan requests and take actions
 639  */
 640 struct hsm_restore_data {
 641         struct mdt_thread_info  *hrd_mti;
 642 };
 643
 644 /**
 645  *  llog_cat_process() callback, used to:
 646  *  - find restore request and allocate the restore handle
 647  * \param env [IN] environment
 648  * \param llh [IN] llog handle
 649  * \param hdr [IN] llog record
 650  * \param data [IN/OUT] cb data = struct hsm_restore_data
 651  * \retval 0 success
 652  * \retval -ve failure
 653  */
 654 static int hsm_restore_cb(const struct lu_env *env,
 655                           struct llog_handle *llh,
 656                           struct llog_rec_hdr *hdr, void *data)
 657 {
 658         struct llog_agent_req_rec       *larr;
 659         struct hsm_restore_data         *hrd;
 660         struct cdt_restore_handle       *crh;
 661         struct hsm_action_item          *hai;
 662         struct mdt_thread_info          *mti;
 663         struct coordinator              *cdt;
 664         struct mdt_object               *child;
 665         int rc;
 666         ENTRY;
 667
 668         hrd = data;
 669         mti = hrd->hrd_mti;
 670         cdt = &mti->mti_mdt->mdt_coordinator;
 671
 672         larr = (struct llog_agent_req_rec *)hdr;
 673         hai = &larr->arr_hai;
 674         if (hai->hai_cookie > cdt->cdt_last_cookie)
 675                 /* update the cookie to avoid collision */
 676                 cdt->cdt_last_cookie = hai->hai_cookie + 1;
 677
 678         if (hai->hai_action != HSMA_RESTORE ||
 679             agent_req_in_final_state(larr->arr_status))
 680                 RETURN(0);
 681
 682         /* restore request not in a final state */
 683
 684         /* force replay of restore requests left in started state from previous
 685          * CDT context, to be canceled later if finally found to be incompatible
 686          * when being re-started */
 687         if (larr->arr_status == ARS_STARTED) {
 688                 larr->arr_status = ARS_WAITING;
 689                 larr->arr_req_change = cfs_time_current_sec();
 690                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
 691                 if (rc != 0)
 692                         GOTO(out, rc);
 693         }
 694
 695         OBD_SLAB_ALLOC_PTR(crh, mdt_hsm_cdt_kmem);
 696         if (crh == NULL)
 697                 RETURN(-ENOMEM);
 698
 699         crh->crh_fid = hai->hai_fid;
 700         /* in V1 all file is restored
 701         crh->extent.start = hai->hai_extent.offset;
 702         crh->extent.end = hai->hai_extent.offset + hai->hai_extent.length;
 703         */
 704         crh->crh_extent.start = 0;
 705         crh->crh_extent.end = hai->hai_extent.length;
 706         /* get the layout lock */
 707         mdt_lock_reg_init(&crh->crh_lh, LCK_EX);
 708         child = mdt_object_find_lock(mti, &crh->crh_fid, &crh->crh_lh,
 709                                      MDS_INODELOCK_LAYOUT);
 710         if (IS_ERR(child))
 711                 GOTO(out, rc = PTR_ERR(child));
 712
 713         rc = 0;
 714         /* we choose to not keep a reference
 715          * on the object during the restore time which can be very long */
 716         mdt_object_put(mti->mti_env, child);
 717
 718         mutex_lock(&cdt->cdt_restore_lock);
 719         list_add_tail(&crh->crh_list, &cdt->cdt_restore_hdl);
 720         mutex_unlock(&cdt->cdt_restore_lock);
 721
 722 out:
 723         RETURN(rc);
 724 }
 725
 726 /**
 727  * restore coordinator state at startup
 728  * the goal is to take a layout lock for each registered restore request
 729  * \param mti [IN] context
 730  */
 731 static int mdt_hsm_pending_restore(struct mdt_thread_info *mti)
 732 {
 733         struct hsm_restore_data  hrd;
 734         int                      rc;
 735         ENTRY;
 736
 737         hrd.hrd_mti = mti;
 738
 739         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
 740                               hsm_restore_cb, &hrd);
 741
 742         RETURN(rc);
 743 }
 744
 745 static int hsm_init_ucred(struct lu_ucred *uc)
 746 {
 747         ENTRY;
 748
 749         uc->uc_valid = UCRED_OLD;
 750         uc->uc_o_uid = 0;
 751         uc->uc_o_gid = 0;
 752         uc->uc_o_fsuid = 0;
 753         uc->uc_o_fsgid = 0;
 754         uc->uc_uid = 0;
 755         uc->uc_gid = 0;
 756         uc->uc_fsuid = 0;
 757         uc->uc_fsgid = 0;
 758         uc->uc_suppgids[0] = -1;
 759         uc->uc_suppgids[1] = -1;
 760         uc->uc_cap = CFS_CAP_FS_MASK;
 761         uc->uc_umask = 0777;
 762         uc->uc_ginfo = NULL;
 763         uc->uc_identity = NULL;
 764
 765         RETURN(0);
 766 }
 767
 768 /**
 769  * wake up coordinator thread
 770  * \param mdt [IN] device
 771  * \retval 0 success
 772  * \retval -ve failure
 773  */
 774 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
 775 {
 776         struct coordinator      *cdt = &mdt->mdt_coordinator;
 777         ENTRY;
 778
 779         if (cdt->cdt_state == CDT_STOPPED)
 780                 RETURN(-ESRCH);
 781
 782         /* wake up coordinator */
 783         cdt->cdt_thread.t_flags = SVC_EVENT;
 784         wake_up(&cdt->cdt_thread.t_ctl_waitq);
 785
 786         RETURN(0);
 787 }
 788
 789 /**
 790  * initialize coordinator struct
 791  * \param mdt [IN] device
 792  * \retval 0 success
 793  * \retval -ve failure
 794  */
 795 int mdt_hsm_cdt_init(struct mdt_device *mdt)
 796 {
 797         struct coordinator      *cdt = &mdt->mdt_coordinator;
 798         struct mdt_thread_info  *cdt_mti = NULL;
 799         int                      rc;
 800         ENTRY;
 801
 802         cdt->cdt_state = CDT_STOPPED;
 803
 804         init_waitqueue_head(&cdt->cdt_thread.t_ctl_waitq);
 805         mutex_init(&cdt->cdt_llog_lock);
 806         init_rwsem(&cdt->cdt_agent_lock);
 807         init_rwsem(&cdt->cdt_request_lock);
 808         mutex_init(&cdt->cdt_restore_lock);
 809
 810         INIT_LIST_HEAD(&cdt->cdt_requests);
 811         INIT_LIST_HEAD(&cdt->cdt_agents);
 812         INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
 813
 814         rc = lu_env_init(&cdt->cdt_env, LCT_MD_THREAD);
 815         if (rc < 0)
 816                 RETURN(rc);
 817
 818         /* for mdt_ucred(), lu_ucred stored in lu_ucred_key */
 819         rc = lu_context_init(&cdt->cdt_session, LCT_SERVER_SESSION);
 820         if (rc == 0) {
 821                 lu_context_enter(&cdt->cdt_session);
 822                 cdt->cdt_env.le_ses = &cdt->cdt_session;
 823         } else {
 824                 lu_env_fini(&cdt->cdt_env);
 825                 RETURN(rc);
 826         }
 827
 828         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 829         LASSERT(cdt_mti != NULL);
 830
 831         cdt_mti->mti_env = &cdt->cdt_env;
 832         cdt_mti->mti_mdt = mdt;
 833
 834         hsm_init_ucred(mdt_ucred(cdt_mti));
 835
 836         /* default values for /proc tunnables
 837          * can be override by MGS conf */
 838         cdt->cdt_default_archive_id = 1;
 839         cdt->cdt_grace_delay = 60;
 840         cdt->cdt_loop_period = 10;
 841         cdt->cdt_max_requests = 3;
 842         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 843         cdt->cdt_active_req_timeout = 3600;
 844
 845         RETURN(0);
 846 }
 847
 848 /**
 849  * free a coordinator thread
 850  * \param mdt [IN] device
 851  */
 852 int  mdt_hsm_cdt_fini(struct mdt_device *mdt)
 853 {
 854         struct coordinator *cdt = &mdt->mdt_coordinator;
 855         ENTRY;
 856
 857         lu_context_exit(cdt->cdt_env.le_ses);
 858         lu_context_fini(cdt->cdt_env.le_ses);
 859
 860         lu_env_fini(&cdt->cdt_env);
 861
 862         RETURN(0);
 863 }
 864
 865 /**
 866  * start a coordinator thread
 867  * \param mdt [IN] device
 868  * \retval 0 success
 869  * \retval -ve failure
 870  */
 871 int mdt_hsm_cdt_start(struct mdt_device *mdt)
 872 {
 873         struct coordinator      *cdt = &mdt->mdt_coordinator;
 874         int                      rc;
 875         void                    *ptr;
 876         struct mdt_thread_info  *cdt_mti;
 877         struct task_struct      *task;
 878         ENTRY;
 879
 880         /* functions defined but not yet used
 881          * this avoid compilation warning
 882          */
 883         ptr = dump_requests;
 884
 885         if (cdt->cdt_state != CDT_STOPPED) {
 886                 CERROR("%s: Coordinator already started\n",
 887                        mdt_obd_name(mdt));
 888                 RETURN(-EALREADY);
 889         }
 890
 891         CLASSERT(1 << (CDT_POLICY_SHIFT_COUNT - 1) == CDT_POLICY_LAST);
 892         cdt->cdt_policy = CDT_DEFAULT_POLICY;
 893
 894         cdt->cdt_state = CDT_INIT;
 895
 896         atomic_set(&cdt->cdt_compound_id, cfs_time_current_sec());
 897         /* just need to be larger than previous one */
 898         /* cdt_last_cookie is protected by cdt_llog_lock */
 899         cdt->cdt_last_cookie = cfs_time_current_sec();
 900         atomic_set(&cdt->cdt_request_count, 0);
 901         cdt->cdt_user_request_mask = (1UL << HSMA_RESTORE);
 902         cdt->cdt_group_request_mask = (1UL << HSMA_RESTORE);
 903         cdt->cdt_other_request_mask = (1UL << HSMA_RESTORE);
 904
 905         /* to avoid deadlock when start is made through /proc
 906          * /proc entries are created by the coordinator thread */
 907
 908         /* set up list of started restore requests */
 909         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 910         rc = mdt_hsm_pending_restore(cdt_mti);
 911         if (rc)
 912                 CERROR("%s: cannot take the layout locks needed"
 913                        " for registered restore: %d\n",
 914                        mdt_obd_name(mdt), rc);
 915
 916         task = kthread_run(mdt_coordinator, cdt_mti, "hsm_cdtr");
 917         if (IS_ERR(task)) {
 918                 rc = PTR_ERR(task);
 919                 cdt->cdt_state = CDT_STOPPED;
 920                 CERROR("%s: error starting coordinator thread: %d\n",
 921                        mdt_obd_name(mdt), rc);
 922                 RETURN(rc);
 923         } else {
 924                 CDEBUG(D_HSM, "%s: coordinator thread started\n",
 925                        mdt_obd_name(mdt));
 926                 rc = 0;
 927         }
 928
 929         wait_event(cdt->cdt_thread.t_ctl_waitq,
 930                        (cdt->cdt_thread.t_flags & SVC_RUNNING));
 931
 932         cdt->cdt_state = CDT_RUNNING;
 933         mdt->mdt_opts.mo_coordinator = 1;
 934         RETURN(0);
 935 }
 936
 937 /**
 938  * stop a coordinator thread
 939  * \param mdt [IN] device
 940  */
 941 int mdt_hsm_cdt_stop(struct mdt_device *mdt)
 942 {
 943         struct coordinator              *cdt = &mdt->mdt_coordinator;
 944         struct cdt_agent_req            *car, *tmp1;
 945         struct hsm_agent                *ha, *tmp2;
 946         struct cdt_restore_handle       *crh, *tmp3;
 947         struct mdt_thread_info          *cdt_mti;
 948         ENTRY;
 949
 950         if (cdt->cdt_state == CDT_STOPPED) {
 951                 CERROR("%s: Coordinator already stopped\n",
 952                        mdt_obd_name(mdt));
 953                 RETURN(-EALREADY);
 954         }
 955
 956         if (cdt->cdt_state != CDT_STOPPING) {
 957                 /* stop coordinator thread before cleaning */
 958                 cdt->cdt_thread.t_flags = SVC_STOPPING;
 959                 wake_up(&cdt->cdt_thread.t_ctl_waitq);
 960                 wait_event(cdt->cdt_thread.t_ctl_waitq,
 961                            cdt->cdt_thread.t_flags & SVC_STOPPED);
 962         }
 963         cdt->cdt_state = CDT_STOPPED;
 964
 965         /* start cleaning */
 966         down_write(&cdt->cdt_request_lock);
 967         list_for_each_entry_safe(car, tmp1, &cdt->cdt_requests,
 968                                  car_request_list) {
 969                 list_del(&car->car_request_list);
 970                 mdt_cdt_free_request(car);
 971         }
 972         up_write(&cdt->cdt_request_lock);
 973
 974         down_write(&cdt->cdt_agent_lock);
 975         list_for_each_entry_safe(ha, tmp2, &cdt->cdt_agents, ha_list) {
 976                 list_del(&ha->ha_list);
 977                 OBD_FREE_PTR(ha);
 978         }
 979         up_write(&cdt->cdt_agent_lock);
 980
 981         cdt_mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
 982         mutex_lock(&cdt->cdt_restore_lock);
 983         list_for_each_entry_safe(crh, tmp3, &cdt->cdt_restore_hdl, crh_list) {
 984                 struct mdt_object       *child;
 985
 986                 /* give back layout lock */
 987                 child = mdt_object_find(&cdt->cdt_env, mdt, &crh->crh_fid);
 988                 if (!IS_ERR(child))
 989                         mdt_object_unlock_put(cdt_mti, child, &crh->crh_lh, 1);
 990
 991                 list_del(&crh->crh_list);
 992
 993                 OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
 994         }
 995         mutex_unlock(&cdt->cdt_restore_lock);
 996
 997         mdt->mdt_opts.mo_coordinator = 0;
 998
 999         RETURN(0);
1000 }
1001
1002 /**
1003  * register all requests from an hal in the memory list
1004  * \param mti [IN] context
1005  * \param hal [IN] request
1006  * \param uuid [OUT] in case of CANCEL, the uuid of the agent
1007  *  which is running the CT
1008  * \retval 0 success
1009  * \retval -ve failure
1010  */
1011 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1012                     struct hsm_action_list *hal, struct obd_uuid *uuid)
1013 {
1014         struct mdt_device       *mdt = mti->mti_mdt;
1015         struct coordinator      *cdt = &mdt->mdt_coordinator;
1016         struct hsm_action_item  *hai;
1017         int                      rc = 0, i;
1018         ENTRY;
1019
1020         /* register request in memory list */
1021         hai = hai_first(hal);
1022         for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
1023                 struct cdt_agent_req *car;
1024
1025                 /* in case of a cancel request, we first mark the ondisk
1026                  * record of the request we want to stop as canceled
1027                  * this does not change the cancel record
1028                  * it will be done when updating the request status
1029                  */
1030                 if (hai->hai_action == HSMA_CANCEL) {
1031                         rc = mdt_agent_record_update(mti->mti_env, mti->mti_mdt,
1032                                                      &hai->hai_cookie,
1033                                                      1, ARS_CANCELED);
1034                         if (rc) {
1035                                 CERROR("%s: mdt_agent_record_update() failed, "
1036                                        "rc=%d, cannot update status to %s "
1037                                        "for cookie "LPX64"\n",
1038                                        mdt_obd_name(mdt), rc,
1039                                        agent_req_status2name(ARS_CANCELED),
1040                                        hai->hai_cookie);
1041                                 GOTO(out, rc);
1042                         }
1043
1044                         /* find the running request to set it canceled */
1045                         car = mdt_cdt_find_request(cdt, hai->hai_cookie, NULL);
1046                         if (car != NULL) {
1047                                 car->car_canceled = 1;
1048                                 /* uuid has to be changed to the one running the
1049                                 * request to cancel */
1050                                 *uuid = car->car_uuid;
1051                                 mdt_cdt_put_request(car);
1052                         }
1053                         /* no need to memorize cancel request
1054                          * this also avoid a deadlock when we receive
1055                          * a purge all requests command
1056                          */
1057                         continue;
1058                 }
1059
1060                 if (hai->hai_action == HSMA_ARCHIVE) {
1061                         struct mdt_object *obj;
1062                         struct md_hsm hsm;
1063
1064                         obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
1065                         if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT))
1066                                 continue;
1067                         if (IS_ERR(obj))
1068                                 GOTO(out, rc = PTR_ERR(obj));
1069
1070                         hsm.mh_flags |= HS_EXISTS;
1071                         hsm.mh_arch_id = hal->hal_archive_id;
1072                         rc = mdt_hsm_attr_set(mti, obj, &hsm);
1073                         mdt_object_put(mti->mti_env, obj);
1074                         if (rc)
1075                                 GOTO(out, rc);
1076                 }
1077
1078                 car = mdt_cdt_alloc_request(hal->hal_compound_id,
1079                                             hal->hal_archive_id, hal->hal_flags,
1080                                             uuid, hai);
1081                 if (IS_ERR(car))
1082                         GOTO(out, rc = PTR_ERR(car));
1083
1084                 rc = mdt_cdt_add_request(cdt, car);
1085                 if (rc != 0)
1086                         mdt_cdt_free_request(car);
1087         }
1088 out:
1089         RETURN(rc);
1090 }
1091
1092 /**
1093  * swap layouts between 2 fids
1094  * \param mti [IN] context
1095  * \param fid1 [IN]
1096  * \param fid2 [IN]
1097  * \param mh_common [IN] MD HSM
1098  */
1099 static int hsm_swap_layouts(struct mdt_thread_info *mti,
1100                             const lustre_fid *fid, const lustre_fid *dfid,
1101                             struct md_hsm *mh_common)
1102 {
1103         struct mdt_device       *mdt = mti->mti_mdt;
1104         struct mdt_object       *child1, *child2;
1105         struct mdt_lock_handle  *lh2;
1106         int                      rc;
1107         ENTRY;
1108
1109         child1 = mdt_object_find(mti->mti_env, mdt, fid);
1110         if (IS_ERR(child1))
1111                 GOTO(out, rc = PTR_ERR(child1));
1112
1113         /* we already have layout lock on FID so take only
1114          * on dfid */
1115         lh2 = &mti->mti_lh[MDT_LH_OLD];
1116         mdt_lock_reg_init(lh2, LCK_EX);
1117         child2 = mdt_object_find_lock(mti, dfid, lh2, MDS_INODELOCK_LAYOUT);
1118         if (IS_ERR(child2))
1119                 GOTO(out_child1, rc = PTR_ERR(child2));
1120
1121         /* if copy tool closes the volatile before sending the final
1122          * progress through llapi_hsm_copy_end(), all the objects
1123          * are removed and mdd_swap_layout LBUG */
1124         if (!mdt_object_exists(child2)) {
1125                 CERROR("%s: Copytool has closed volatile file "DFID"\n",
1126                        mdt_obd_name(mti->mti_mdt), PFID(dfid));
1127                 GOTO(out_child2, rc = -ENOENT);
1128         }
1129         /* Since we only handle restores here, unconditionally use
1130          * SWAP_LAYOUTS_MDS_HSM flag to ensure original layout will
1131          * be preserved in case of failure during swap_layout and not
1132          * leave a file in an intermediate but incoherent state.
1133          * But need to setup HSM xattr of data FID before, reuse
1134          * mti and mh presets for FID in hsm_cdt_request_completed(),
1135          * only need to clear RELEASED and DIRTY.
1136          */
1137         mh_common->mh_flags &= ~(HS_RELEASED | HS_DIRTY);
1138         rc = mdt_hsm_attr_set(mti, child2, mh_common);
1139         if (rc == 0)
1140                 rc = mo_swap_layouts(mti->mti_env,
1141                                      mdt_object_child(child1),
1142                                      mdt_object_child(child2),
1143                                      SWAP_LAYOUTS_MDS_HSM);
1144
1145 out_child2:
1146         mdt_object_unlock_put(mti, child2, lh2, 1);
1147 out_child1:
1148         mdt_object_put(mti->mti_env, child1);
1149 out:
1150         RETURN(rc);
1151 }
1152
1153 /**
1154  * update status of a completed request
1155  * \param mti [IN] context
1156  * \param pgs [IN] progress of the copy tool
1157  * \param update_record [IN] update llog record
1158  * \retval 0 success
1159  * \retval -ve failure
1160  */
1161 static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
1162                                      struct hsm_progress_kernel *pgs,
1163                                      const struct cdt_agent_req *car,
1164                                      enum agent_req_status *status)
1165 {
1166         const struct lu_env     *env = mti->mti_env;
1167         struct mdt_device       *mdt = mti->mti_mdt;
1168         struct coordinator      *cdt = &mdt->mdt_coordinator;
1169         struct mdt_object       *obj = NULL;
1170         int                      cl_flags = 0, rc = 0;
1171         struct md_hsm            mh;
1172         bool                     is_mh_changed;
1173         ENTRY;
1174
1175         /* default is to retry */
1176         *status = ARS_WAITING;
1177
1178         /* find object by FID */
1179         obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
1180         /* we will update MD HSM only if needed */
1181         is_mh_changed = false;
1182         if (IS_ERR(obj)) {
1183                 /* object removed */
1184                 *status = ARS_SUCCEED;
1185                 goto unlock;
1186         }
1187
1188         /* no need to change mh->mh_arch_id
1189          * mdt_hsm_get_md_hsm() got it from disk and it is still valid
1190          */
1191         if (pgs->hpk_errval != 0) {
1192                 switch (pgs->hpk_errval) {
1193                 case ENOSYS:
1194                         /* the copy tool does not support cancel
1195                          * so the cancel request is failed
1196                          * As we cannot distinguish a cancel progress
1197                          * from another action progress (they have the
1198                          * same cookie), we suppose here the CT returns
1199                          * ENOSYS only if does not support cancel
1200                          */
1201                         /* this can also happen when cdt calls it to
1202                          * for a timed out request */
1203                         *status = ARS_FAILED;
1204                         /* to have a cancel event in changelog */
1205                         pgs->hpk_errval = ECANCELED;
1206                         break;
1207                 case ECANCELED:
1208                         /* the request record has already been set to
1209                          * ARS_CANCELED, this set the cancel request
1210                          * to ARS_SUCCEED */
1211                         *status = ARS_SUCCEED;
1212                         break;
1213                 default:
1214                         *status = (cdt->cdt_policy & CDT_NORETRY_ACTION ||
1215                                    !(pgs->hpk_flags & HP_FLAG_RETRY) ?
1216                                    ARS_FAILED : ARS_WAITING);
1217                         break;
1218                 }
1219
1220                 if (pgs->hpk_errval > CLF_HSM_MAXERROR) {
1221                         CERROR("%s: Request "LPX64" on "DFID
1222                                " failed, error code %d too large\n",
1223                                mdt_obd_name(mdt),
1224                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1225                                pgs->hpk_errval);
1226                         hsm_set_cl_error(&cl_flags,
1227                                          CLF_HSM_ERROVERFLOW);
1228                         rc = -EINVAL;
1229                 } else {
1230                         hsm_set_cl_error(&cl_flags, pgs->hpk_errval);
1231                 }
1232
1233                 switch (car->car_hai->hai_action) {
1234                 case HSMA_ARCHIVE:
1235                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1236                         break;
1237                 case HSMA_RESTORE:
1238                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1239                         break;
1240                 case HSMA_REMOVE:
1241                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1242                         break;
1243                 case HSMA_CANCEL:
1244                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1245                         CERROR("%s: Failed request "LPX64" on "DFID
1246                                " cannot be a CANCEL\n",
1247                                mdt_obd_name(mdt),
1248                                pgs->hpk_cookie,
1249                                PFID(&pgs->hpk_fid));
1250                         break;
1251                 default:
1252                         CERROR("%s: Failed request "LPX64" on "DFID
1253                                " %d is an unknown action\n",
1254                                mdt_obd_name(mdt),
1255                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1256                                car->car_hai->hai_action);
1257                         rc = -EINVAL;
1258                         break;
1259                 }
1260         } else {
1261                 *status = ARS_SUCCEED;
1262                 switch (car->car_hai->hai_action) {
1263                 case HSMA_ARCHIVE:
1264                         hsm_set_cl_event(&cl_flags, HE_ARCHIVE);
1265                         /* set ARCHIVE keep EXIST and clear LOST and
1266                          * DIRTY */
1267                         mh.mh_arch_ver = pgs->hpk_data_version;
1268                         mh.mh_flags |= HS_ARCHIVED;
1269                         mh.mh_flags &= ~(HS_LOST|HS_DIRTY);
1270                         is_mh_changed = true;
1271                         break;
1272                 case HSMA_RESTORE:
1273                         hsm_set_cl_event(&cl_flags, HE_RESTORE);
1274
1275                         /* do not clear RELEASED and DIRTY here
1276                          * this will occur in hsm_swap_layouts()
1277                          */
1278
1279                         /* Restoring has changed the file version on
1280                          * disk. */
1281                         mh.mh_arch_ver = pgs->hpk_data_version;
1282                         is_mh_changed = true;
1283                         break;
1284                 case HSMA_REMOVE:
1285                         hsm_set_cl_event(&cl_flags, HE_REMOVE);
1286                         /* clear ARCHIVED EXISTS and LOST */
1287                         mh.mh_flags &= ~(HS_ARCHIVED | HS_EXISTS | HS_LOST);
1288                         is_mh_changed = true;
1289                         break;
1290                 case HSMA_CANCEL:
1291                         hsm_set_cl_event(&cl_flags, HE_CANCEL);
1292                         CERROR("%s: Successful request "LPX64
1293                                " on "DFID
1294                                " cannot be a CANCEL\n",
1295                                mdt_obd_name(mdt),
1296                                pgs->hpk_cookie,
1297                                PFID(&pgs->hpk_fid));
1298                         break;
1299                 default:
1300                         CERROR("%s: Successful request "LPX64
1301                                " on "DFID
1302                                " %d is an unknown action\n",
1303                                mdt_obd_name(mdt),
1304                                pgs->hpk_cookie, PFID(&pgs->hpk_fid),
1305                                car->car_hai->hai_action);
1306                         rc = -EINVAL;
1307                         break;
1308                 }
1309         }
1310
1311         /* rc != 0 means error when analysing action, it may come from
1312          * a crasy CT no need to manage DIRTY
1313          */
1314         if (rc == 0)
1315                 hsm_set_cl_flags(&cl_flags,
1316                                  mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
1317
1318         /* unlock is done later, after layout lock management */
1319         if (is_mh_changed)
1320                 rc = mdt_hsm_attr_set(mti, obj, &mh);
1321
1322 unlock:
1323         /* we give back layout lock only if restore was successful or
1324          * if restore was canceled or if policy is to not retry
1325          * in other cases we just unlock the object */
1326         if (car->car_hai->hai_action == HSMA_RESTORE &&
1327             (pgs->hpk_errval == 0 || pgs->hpk_errval == ECANCELED ||
1328              cdt->cdt_policy & CDT_NORETRY_ACTION)) {
1329                 struct cdt_restore_handle       *crh;
1330
1331                 /* restore in data FID done, we swap the layouts
1332                  * only if restore is successful */
1333                 if (pgs->hpk_errval == 0) {
1334                         rc = hsm_swap_layouts(mti, &car->car_hai->hai_fid,
1335                                               &car->car_hai->hai_dfid, &mh);
1336                         if (rc) {
1337                                 if (cdt->cdt_policy & CDT_NORETRY_ACTION)
1338                                         *status = ARS_FAILED;
1339                                 pgs->hpk_errval = -rc;
1340                         }
1341                 }
1342                 /* we have to retry, so keep layout lock */
1343                 if (*status == ARS_WAITING)
1344                         GOTO(out, rc);
1345
1346                 /* give back layout lock */
1347                 mutex_lock(&cdt->cdt_restore_lock);
1348                 crh = mdt_hsm_restore_hdl_find(cdt, &car->car_hai->hai_fid);
1349                 if (crh != NULL)
1350                         list_del(&crh->crh_list);
1351                 mutex_unlock(&cdt->cdt_restore_lock);
1352                 /* just give back layout lock, we keep
1353                  * the reference which is given back
1354                  * later with the lock for HSM flags */
1355                 if (!IS_ERR(obj) && crh != NULL)
1356                         mdt_object_unlock(mti, obj, &crh->crh_lh, 1);
1357
1358                 if (crh != NULL)
1359                         OBD_SLAB_FREE_PTR(crh, mdt_hsm_cdt_kmem);
1360         }
1361
1362         GOTO(out, rc);
1363
1364 out:
1365         if (obj != NULL && !IS_ERR(obj)) {
1366                 mo_changelog(env, CL_HSM, cl_flags,
1367                              mdt_object_child(obj));
1368                 mdt_object_put(mti->mti_env, obj);
1369         }
1370
1371         RETURN(rc);
1372 }
1373
1374 /**
1375  * update status of a request
1376  * \param mti [IN] context
1377  * \param pgs [IN] progress of the copy tool
1378  * \param update_record [IN] update llog record
1379  * \retval 0 success
1380  * \retval -ve failure
1381  */
1382 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1383                                  struct hsm_progress_kernel *pgs,
1384                                  const int update_record)
1385 {
1386         struct mdt_device       *mdt = mti->mti_mdt;
1387         struct coordinator      *cdt = &mdt->mdt_coordinator;
1388         struct cdt_agent_req    *car;
1389         int                      rc = 0;
1390         ENTRY;
1391
1392         /* no coordinator started, so we cannot serve requests */
1393         if (cdt->cdt_state == CDT_STOPPED)
1394                 RETURN(-EAGAIN);
1395
1396         /* first do sanity checks */
1397         car = mdt_cdt_update_request(cdt, pgs);
1398         if (IS_ERR(car)) {
1399                 CERROR("%s: Cannot find running request for cookie "LPX64
1400                        " on fid="DFID"\n",
1401                        mdt_obd_name(mdt),
1402                        pgs->hpk_cookie, PFID(&pgs->hpk_fid));
1403
1404                 RETURN(PTR_ERR(car));
1405         }
1406
1407         CDEBUG(D_HSM, "Progress received for fid="DFID" cookie="LPX64
1408                       " action=%s flags=%d err=%d fid="DFID" dfid="DFID"\n",
1409                       PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1410                       hsm_copytool_action2name(car->car_hai->hai_action),
1411                       pgs->hpk_flags, pgs->hpk_errval,
1412                       PFID(&car->car_hai->hai_fid),
1413                       PFID(&car->car_hai->hai_dfid));
1414
1415         /* progress is done on FID or data FID depending of the action and
1416          * of the copy progress */
1417         /* for restore progress is used to send back the data FID to cdt */
1418         if (car->car_hai->hai_action == HSMA_RESTORE &&
1419             lu_fid_eq(&car->car_hai->hai_fid, &car->car_hai->hai_dfid))
1420                 car->car_hai->hai_dfid = pgs->hpk_fid;
1421
1422         if ((car->car_hai->hai_action == HSMA_RESTORE ||
1423              car->car_hai->hai_action == HSMA_ARCHIVE) &&
1424             (!lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_dfid) &&
1425              !lu_fid_eq(&pgs->hpk_fid, &car->car_hai->hai_fid))) {
1426                 CERROR("%s: Progress on "DFID" for cookie "LPX64
1427                        " does not match request FID "DFID" nor data FID "
1428                        DFID"\n",
1429                        mdt_obd_name(mdt),
1430                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1431                        PFID(&car->car_hai->hai_fid),
1432                        PFID(&car->car_hai->hai_dfid));
1433                 GOTO(out, rc = -EINVAL);
1434         }
1435
1436         if (pgs->hpk_errval != 0 && !(pgs->hpk_flags & HP_FLAG_COMPLETED)) {
1437                 CERROR("%s: Progress on "DFID" for cookie "LPX64" action=%s"
1438                        " is not coherent (err=%d and not completed"
1439                        " (flags=%d))\n",
1440                        mdt_obd_name(mdt),
1441                        PFID(&pgs->hpk_fid), pgs->hpk_cookie,
1442                        hsm_copytool_action2name(car->car_hai->hai_action),
1443                        pgs->hpk_errval, pgs->hpk_flags);
1444                 GOTO(out, rc = -EINVAL);
1445         }
1446
1447         /* now progress is valid */
1448
1449         /* we use a root like ucred */
1450         hsm_init_ucred(mdt_ucred(mti));
1451
1452         if (pgs->hpk_flags & HP_FLAG_COMPLETED) {
1453                 enum agent_req_status    status;
1454
1455                 rc = hsm_cdt_request_completed(mti, pgs, car, &status);
1456
1457                 /* remove request from memory list */
1458                 mdt_cdt_remove_request(cdt, pgs->hpk_cookie);
1459
1460                 CDEBUG(D_HSM, "Updating record: fid="DFID" cookie="LPX64
1461                               " action=%s status=%s\n", PFID(&pgs->hpk_fid),
1462                        pgs->hpk_cookie,
1463                        hsm_copytool_action2name(car->car_hai->hai_action),
1464                        agent_req_status2name(status));
1465
1466                 if (update_record) {
1467                         int rc1;
1468
1469                         rc1 = mdt_agent_record_update(mti->mti_env, mdt,
1470                                                      &pgs->hpk_cookie, 1,
1471                                                      status);
1472                         if (rc1)
1473                                 CERROR("%s: mdt_agent_record_update() failed,"
1474                                        " rc=%d, cannot update status to %s"
1475                                        " for cookie "LPX64"\n",
1476                                        mdt_obd_name(mdt), rc1,
1477                                        agent_req_status2name(status),
1478                                        pgs->hpk_cookie);
1479                         rc = (rc != 0 ? rc : rc1);
1480                 }
1481                 /* ct has completed a request, so a slot is available, wakeup
1482                  * cdt to find new work */
1483                 mdt_hsm_cdt_wakeup(mdt);
1484         } else {
1485                 /* if copytool send a progress on a canceled request
1486                  * we inform copytool it should stop
1487                  */
1488                 if (car->car_canceled == 1)
1489                         rc = -ECANCELED;
1490         }
1491         GOTO(out, rc);
1492
1493 out:
1494         /* remove ref got from mdt_cdt_update_request() */
1495         mdt_cdt_put_request(car);
1496
1497         return rc;
1498 }
1499
1500
1501 /**
1502  * data passed to llog_cat_process() callback
1503  * to cancel requests
1504  */
1505 struct hsm_cancel_all_data {
1506         struct mdt_device       *mdt;
1507 };
1508
1509 /**
1510  *  llog_cat_process() callback, used to:
1511  *  - purge all requests
1512  * \param env [IN] environment
1513  * \param llh [IN] llog handle
1514  * \param hdr [IN] llog record
1515  * \param data [IN] cb data = struct hsm_cancel_all_data
1516  * \retval 0 success
1517  * \retval -ve failure
1518  */
1519 static int mdt_cancel_all_cb(const struct lu_env *env,
1520                              struct llog_handle *llh,
1521                              struct llog_rec_hdr *hdr, void *data)
1522 {
1523         struct llog_agent_req_rec       *larr;
1524         struct hsm_cancel_all_data      *hcad;
1525         int                              rc = 0;
1526         ENTRY;
1527
1528         larr = (struct llog_agent_req_rec *)hdr;
1529         hcad = data;
1530         if (larr->arr_status == ARS_WAITING ||
1531             larr->arr_status == ARS_STARTED) {
1532                 larr->arr_status = ARS_CANCELED;
1533                 larr->arr_req_change = cfs_time_current_sec();
1534                 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1535         }
1536
1537         RETURN(rc);
1538 }
1539
1540 /**
1541  * cancel all actions
1542  * \param obd [IN] MDT device
1543  */
1544 static int hsm_cancel_all_actions(struct mdt_device *mdt)
1545 {
1546         struct mdt_thread_info          *mti;
1547         struct coordinator              *cdt = &mdt->mdt_coordinator;
1548         struct cdt_agent_req            *car;
1549         struct hsm_action_list          *hal = NULL;
1550         struct hsm_action_item          *hai;
1551         struct hsm_cancel_all_data       hcad;
1552         int                              hal_sz = 0, hal_len, rc;
1553         enum cdt_states                  save_state;
1554         ENTRY;
1555
1556         /* retrieve coordinator context */
1557         mti = lu_context_key_get(&cdt->cdt_env.le_ctx, &mdt_thread_key);
1558
1559         /* disable coordinator */
1560         save_state = cdt->cdt_state;
1561         cdt->cdt_state = CDT_DISABLE;
1562
1563         /* send cancel to all running requests */
1564         down_read(&cdt->cdt_request_lock);
1565         list_for_each_entry(car, &cdt->cdt_requests, car_request_list) {
1566                 mdt_cdt_get_request(car);
1567                 /* request is not yet removed from list, it will be done
1568                  * when copytool will return progress
1569                  */
1570
1571                 if (car->car_hai->hai_action == HSMA_CANCEL) {
1572                         mdt_cdt_put_request(car);
1573                         continue;
1574                 }
1575
1576                 /* needed size */
1577                 hal_len = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN + 1) +
1578                           cfs_size_round(car->car_hai->hai_len);
1579
1580                 if (hal_len > hal_sz && hal_sz > 0) {
1581                         /* not enough room, free old buffer */
1582                         OBD_FREE(hal, hal_sz);
1583                         hal = NULL;
1584                 }
1585
1586                 /* empty buffer, allocate one */
1587                 if (hal == NULL) {
1588                         hal_sz = hal_len;
1589                         OBD_ALLOC(hal, hal_sz);
1590                         if (hal == NULL) {
1591                                 mdt_cdt_put_request(car);
1592                                 up_read(&cdt->cdt_request_lock);
1593                                 GOTO(out, rc = -ENOMEM);
1594                         }
1595                 }
1596
1597                 hal->hal_version = HAL_VERSION;
1598                 obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(mdt),
1599                                 MTI_NAME_MAXLEN);
1600                 hal->hal_fsname[MTI_NAME_MAXLEN] = '\0';
1601                 hal->hal_compound_id = car->car_compound_id;
1602                 hal->hal_archive_id = car->car_archive_id;
1603                 hal->hal_flags = car->car_flags;
1604                 hal->hal_count = 0;
1605
1606                 hai = hai_first(hal);
1607                 memcpy(hai, car->car_hai, car->car_hai->hai_len);
1608                 hai->hai_action = HSMA_CANCEL;
1609                 hal->hal_count = 1;
1610
1611                 /* it is possible to safely call mdt_hsm_agent_send()
1612                  * (ie without a deadlock on cdt_request_lock), because the
1613                  * write lock is taken only if we are not in purge mode
1614                  * (mdt_hsm_agent_send() does not call mdt_cdt_add_request()
1615                  *   nor mdt_cdt_remove_request())
1616                  */
1617                 /* no conflict with cdt thread because cdt is disable and we
1618                  * have the request lock */
1619                 mdt_hsm_agent_send(mti, hal, 1);
1620
1621                 mdt_cdt_put_request(car);
1622         }
1623         up_read(&cdt->cdt_request_lock);
1624
1625         if (hal != NULL)
1626                 OBD_FREE(hal, hal_sz);
1627
1628         /* cancel all on-disk records */
1629         hcad.mdt = mdt;
1630
1631         rc = cdt_llog_process(mti->mti_env, mti->mti_mdt,
1632                               mdt_cancel_all_cb, &hcad);
1633 out:
1634         /* enable coordinator */
1635         cdt->cdt_state = save_state;
1636
1637         RETURN(rc);
1638 }
1639
1640 /**
1641  * check if a request is compatible with file status
1642  * \param hai [IN] request description
1643  * \param hal_an [IN] request archive number (not used)
1644  * \param rq_flags [IN] request flags
1645  * \param hsm [IN] file HSM metadata
1646  * \retval boolean
1647  */
1648 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1649                               const int hal_an, const __u64 rq_flags,
1650                               const struct md_hsm *hsm)
1651 {
1652         int      is_compat = false;
1653         int      hsm_flags;
1654         ENTRY;
1655
1656         hsm_flags = hsm->mh_flags;
1657         switch (hai->hai_action) {
1658         case HSMA_ARCHIVE:
1659                 if (!(hsm_flags & HS_NOARCHIVE) &&
1660                     (hsm_flags & HS_DIRTY || !(hsm_flags & HS_ARCHIVED)))
1661                         is_compat = true;
1662                 break;
1663         case HSMA_RESTORE:
1664                 if (!(hsm_flags & HS_DIRTY) && (hsm_flags & HS_RELEASED) &&
1665                     hsm_flags & HS_ARCHIVED && !(hsm_flags & HS_LOST))
1666                         is_compat = true;
1667                 break;
1668         case HSMA_REMOVE:
1669                 if (!(hsm_flags & HS_RELEASED) &&
1670                     (hsm_flags & (HS_ARCHIVED | HS_EXISTS)))
1671                         is_compat = true;
1672                 break;
1673         case HSMA_CANCEL:
1674                 is_compat = true;
1675                 break;
1676         }
1677         CDEBUG(D_HSM, "fid="DFID" action=%s flags="LPX64
1678                       " extent="LPX64"-"LPX64" hsm_flags=%.8X %s\n",
1679                       PFID(&hai->hai_fid),
1680                       hsm_copytool_action2name(hai->hai_action), rq_flags,
1681                       hai->hai_extent.offset, hai->hai_extent.length,
1682                       hsm->mh_flags,
1683                       (is_compat ? "compatible" : "uncompatible"));
1684
1685         RETURN(is_compat);
1686 }
1687
1688 /*
1689  * /proc interface used to get/set HSM behaviour (cdt->cdt_policy)
1690  */
1691 static const struct {
1692         __u64            bit;
1693         char            *name;
1694         char            *nickname;
1695 } hsm_policy_names[] = {
1696         { CDT_NONBLOCKING_RESTORE,      "NonBlockingRestore",   "NBR"},
1697         { CDT_NORETRY_ACTION,           "NoRetryAction",        "NRA"},
1698         { 0 },
1699 };
1700
1701 /**
1702  * convert a policy name to a bit
1703  * \param name [IN] policy name
1704  * \retval 0 unknown
1705  * \retval   policy bit
1706  */
1707 static __u64 hsm_policy_str2bit(const char *name)
1708 {
1709         int      i;
1710
1711         for (i = 0; hsm_policy_names[i].bit != 0; i++)
1712                 if (strcmp(hsm_policy_names[i].nickname, name) == 0 ||
1713                     strcmp(hsm_policy_names[i].name, name) == 0)
1714                         return hsm_policy_names[i].bit;
1715         return 0;
1716 }
1717
1718 /**
1719  * convert a policy bit field to a string
1720  * \param mask [IN] policy bit field
1721  * \param hexa [IN] print mask before bit names
1722  * \param buffer [OUT] string
1723  * \param count [IN] size of buffer
1724  */
1725 static void hsm_policy_bit2str(struct seq_file *m, const __u64 mask,
1726                                 const bool hexa)
1727 {
1728         int      i, j;
1729         __u64    bit;
1730         ENTRY;
1731
1732         if (hexa)
1733                 seq_printf(m, "("LPX64") ", mask);
1734
1735         for (i = 0; i < CDT_POLICY_SHIFT_COUNT; i++) {
1736                 bit = (1ULL << i);
1737
1738                 for (j = 0; hsm_policy_names[j].bit != 0; j++) {
1739                         if (hsm_policy_names[j].bit == bit)
1740                                 break;
1741                 }
1742                 if (bit & mask)
1743                         seq_printf(m, "[%s] ", hsm_policy_names[j].name);
1744                 else
1745                         seq_printf(m, "%s ", hsm_policy_names[j].name);
1746         }
1747         /* remove last ' ' */
1748         m->count--;
1749         seq_putc(m, '\0');
1750 }
1751
1752 /* methods to read/write HSM policy flags */
1753 static int mdt_hsm_policy_seq_show(struct seq_file *m, void *data)
1754 {
1755         struct mdt_device       *mdt = m->private;
1756         struct coordinator      *cdt = &mdt->mdt_coordinator;
1757         ENTRY;
1758
1759         hsm_policy_bit2str(m, cdt->cdt_policy, false);
1760         RETURN(0);
1761 }
1762
1763 static ssize_t
1764 mdt_hsm_policy_seq_write(struct file *file, const char __user *buffer,
1765                          size_t count, loff_t *off)
1766 {
1767         struct seq_file         *m = file->private_data;
1768         struct mdt_device       *mdt = m->private;
1769         struct coordinator      *cdt = &mdt->mdt_coordinator;
1770         char                    *start, *token, sign;
1771         char                    *buf;
1772         __u64                    policy;
1773         __u64                    add_mask, remove_mask, set_mask;
1774         int                      rc;
1775         ENTRY;
1776
1777         if (count + 1 > PAGE_SIZE)
1778                 RETURN(-EINVAL);
1779
1780         OBD_ALLOC(buf, count + 1);
1781         if (buf == NULL)
1782                 RETURN(-ENOMEM);
1783
1784         if (copy_from_user(buf, buffer, count))
1785                 GOTO(out, rc = -EFAULT);
1786
1787         buf[count] = '\0';
1788
1789         start = buf;
1790         CDEBUG(D_HSM, "%s: receive new policy: '%s'\n", mdt_obd_name(mdt),
1791                start);
1792
1793         add_mask = remove_mask = set_mask = 0;
1794         do {
1795                 token = strsep(&start, "\n ");
1796                 sign = *token;
1797
1798                 if (sign == '\0')
1799                         continue;
1800
1801                 if (sign == '-' || sign == '+')
1802                         token++;
1803
1804                 policy = hsm_policy_str2bit(token);
1805                 if (policy == 0) {
1806                         CWARN("%s: '%s' is unknown, "
1807                               "supported policies are:\n", mdt_obd_name(mdt),
1808                               token);
1809                         hsm_policy_bit2str(m, 0, false);
1810                         GOTO(out, rc = -EINVAL);
1811                 }
1812                 switch (sign) {
1813                 case '-':
1814                         remove_mask |= policy;
1815                         break;
1816                 case '+':
1817                         add_mask |= policy;
1818                         break;
1819                 default:
1820                         set_mask |= policy;
1821                         break;
1822                 }
1823
1824         } while (start != NULL);
1825
1826         CDEBUG(D_HSM, "%s: new policy: rm="LPX64" add="LPX64" set="LPX64"\n",
1827                mdt_obd_name(mdt), remove_mask, add_mask, set_mask);
1828
1829         /* if no sign in all string, it is a clear and set
1830          * if some sign found, all unsigned are converted
1831          * to add
1832          * P1 P2 = set to P1 and P2
1833          * P1 -P2 = add P1 clear P2 same as +P1 -P2
1834          */
1835         if (remove_mask == 0 && add_mask == 0) {
1836                 cdt->cdt_policy = set_mask;
1837         } else {
1838                 cdt->cdt_policy |= set_mask | add_mask;
1839                 cdt->cdt_policy &= ~remove_mask;
1840         }
1841
1842         GOTO(out, rc = count);
1843
1844 out:
1845         OBD_FREE(buf, count + 1);
1846         RETURN(rc);
1847 }
1848 LPROC_SEQ_FOPS(mdt_hsm_policy);
1849
1850 #define GENERATE_PROC_METHOD(VAR)                                       \
1851 static int mdt_hsm_##VAR##_seq_show(struct seq_file *m, void *data)     \
1852 {                                                                       \
1853         struct mdt_device       *mdt = m->private;                      \
1854         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1855         ENTRY;                                                          \
1856                                                                         \
1857         seq_printf(m, LPU64"\n", (__u64)cdt->VAR);                      \
1858         RETURN(0);                                                      \
1859 }                                                                       \
1860 static ssize_t                                                          \
1861 mdt_hsm_##VAR##_seq_write(struct file *file, const char __user *buffer, \
1862                           size_t count, loff_t *off)                    \
1863                                                                         \
1864 {                                                                       \
1865         struct seq_file         *m = file->private_data;                \
1866         struct mdt_device       *mdt = m->private;                      \
1867         struct coordinator      *cdt = &mdt->mdt_coordinator;           \
1868         __s64                    val;                                   \
1869         int                      rc;                                    \
1870         ENTRY;                                                          \
1871                                                                         \
1872         rc = lprocfs_str_to_s64(buffer, count, &val);                   \
1873         if (rc)                                                         \
1874                 RETURN(rc);                                             \
1875         if (val > 0 && val < INT_MAX) {                                 \
1876                 cdt->VAR = val;                                         \
1877                 RETURN(count);                                          \
1878         }                                                               \
1879         RETURN(-EINVAL);                                                \
1880 }                                                                       \
1881
1882 GENERATE_PROC_METHOD(cdt_loop_period)
1883 GENERATE_PROC_METHOD(cdt_grace_delay)
1884 GENERATE_PROC_METHOD(cdt_active_req_timeout)
1885 GENERATE_PROC_METHOD(cdt_max_requests)
1886 GENERATE_PROC_METHOD(cdt_default_archive_id)
1887
1888 /*
1889  * procfs write method for MDT/hsm_control
1890  * proc entry is in mdt directory so data is mdt obd_device pointer
1891  */
1892 #define CDT_ENABLE_CMD   "enabled"
1893 #define CDT_STOP_CMD     "shutdown"
1894 #define CDT_DISABLE_CMD  "disabled"
1895 #define CDT_PURGE_CMD    "purge"
1896 #define CDT_HELP_CMD     "help"
1897 #define CDT_MAX_CMD_LEN  10
1898
1899 ssize_t
1900 mdt_hsm_cdt_control_seq_write(struct file *file, const char __user *buffer,
1901                               size_t count, loff_t *off)
1902 {
1903         struct seq_file         *m = file->private_data;
1904         struct obd_device       *obd = m->private;
1905         struct mdt_device       *mdt = mdt_dev(obd->obd_lu_dev);
1906         struct coordinator      *cdt = &(mdt->mdt_coordinator);
1907         int                      rc, usage = 0;
1908         char                     kernbuf[CDT_MAX_CMD_LEN];
1909         ENTRY;
1910
1911         if (count == 0 || count >= sizeof(kernbuf))
1912                 RETURN(-EINVAL);
1913
1914         if (copy_from_user(kernbuf, buffer, count))
1915                 RETURN(-EFAULT);
1916         kernbuf[count] = 0;
1917
1918         if (kernbuf[count - 1] == '\n')
1919                 kernbuf[count - 1] = 0;
1920
1921         rc = 0;
1922         if (strcmp(kernbuf, CDT_ENABLE_CMD) == 0) {
1923                 if (cdt->cdt_state == CDT_DISABLE) {
1924                         cdt->cdt_state = CDT_RUNNING;
1925                         mdt_hsm_cdt_wakeup(mdt);
1926                 } else {
1927                         rc = mdt_hsm_cdt_start(mdt);
1928                 }
1929         } else if (strcmp(kernbuf, CDT_STOP_CMD) == 0) {
1930                 if ((cdt->cdt_state == CDT_STOPPING) ||
1931                     (cdt->cdt_state == CDT_STOPPED)) {
1932                         CERROR("%s: Coordinator already stopped\n",
1933                                mdt_obd_name(mdt));
1934                         rc = -EALREADY;
1935                 } else {
1936                         cdt->cdt_state = CDT_STOPPING;
1937                 }
1938         } else if (strcmp(kernbuf, CDT_DISABLE_CMD) == 0) {
1939                 if ((cdt->cdt_state == CDT_STOPPING) ||
1940                     (cdt->cdt_state == CDT_STOPPED)) {
1941                         CERROR("%s: Coordinator is stopped\n",
1942                                mdt_obd_name(mdt));
1943                         rc = -EINVAL;
1944                 } else {
1945                         cdt->cdt_state = CDT_DISABLE;
1946                 }
1947         } else if (strcmp(kernbuf, CDT_PURGE_CMD) == 0) {
1948                 rc = hsm_cancel_all_actions(mdt);
1949         } else if (strcmp(kernbuf, CDT_HELP_CMD) == 0) {
1950                 usage = 1;
1951         } else {
1952                 usage = 1;
1953                 rc = -EINVAL;
1954         }
1955
1956         if (usage == 1)
1957                 CERROR("%s: Valid coordinator control commands are: "
1958                        "%s %s %s %s %s\n", mdt_obd_name(mdt),
1959                        CDT_ENABLE_CMD, CDT_STOP_CMD, CDT_DISABLE_CMD,
1960                        CDT_PURGE_CMD, CDT_HELP_CMD);
1961
1962         if (rc)
1963                 RETURN(rc);
1964
1965         RETURN(count);
1966 }
1967
1968 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data)
1969 {
1970         struct obd_device       *obd = m->private;
1971         struct coordinator      *cdt;
1972         ENTRY;
1973
1974         cdt = &(mdt_dev(obd->obd_lu_dev)->mdt_coordinator);
1975
1976         if (cdt->cdt_state == CDT_INIT)
1977                 seq_printf(m, "init\n");
1978         else if (cdt->cdt_state == CDT_RUNNING)
1979                 seq_printf(m, "enabled\n");
1980         else if (cdt->cdt_state == CDT_STOPPING)
1981                 seq_printf(m, "stopping\n");
1982         else if (cdt->cdt_state == CDT_STOPPED)
1983                 seq_printf(m, "stopped\n");
1984         else if (cdt->cdt_state == CDT_DISABLE)
1985                 seq_printf(m, "disabled\n");
1986         else
1987                 seq_printf(m, "unknown\n");
1988
1989         RETURN(0);
1990 }
1991
1992 static int
1993 mdt_hsm_request_mask_show(struct seq_file *m, __u64 mask)
1994 {
1995         bool first = true;
1996         int i;
1997         ENTRY;
1998
1999         for (i = 0; i < 8 * sizeof(mask); i++) {
2000                 if (mask & (1UL << i)) {
2001                         seq_printf(m, "%s%s", first ? "" : " ",
2002                                    hsm_copytool_action2name(i));
2003                         first = false;
2004                 }
2005         }
2006         seq_putc(m, '\n');
2007
2008         RETURN(0);
2009 }
2010
2011 static int
2012 mdt_hsm_user_request_mask_seq_show(struct seq_file *m, void *data)
2013 {
2014         struct mdt_device *mdt = m->private;
2015         struct coordinator *cdt = &mdt->mdt_coordinator;
2016
2017         return mdt_hsm_request_mask_show(m, cdt->cdt_user_request_mask);
2018 }
2019
2020 static int
2021 mdt_hsm_group_request_mask_seq_show(struct seq_file *m, void *data)
2022 {
2023         struct mdt_device *mdt = m->private;
2024         struct coordinator *cdt = &mdt->mdt_coordinator;
2025
2026         return mdt_hsm_request_mask_show(m, cdt->cdt_group_request_mask);
2027 }
2028
2029 static int
2030 mdt_hsm_other_request_mask_seq_show(struct seq_file *m, void *data)
2031 {
2032         struct mdt_device *mdt = m->private;
2033         struct coordinator *cdt = &mdt->mdt_coordinator;
2034
2035         return mdt_hsm_request_mask_show(m, cdt->cdt_other_request_mask);
2036 }
2037
2038 static inline enum hsm_copytool_action
2039 hsm_copytool_name2action(const char *name)
2040 {
2041         if (strcasecmp(name, "NOOP") == 0)
2042                 return HSMA_NONE;
2043         else if (strcasecmp(name, "ARCHIVE") == 0)
2044                 return HSMA_ARCHIVE;
2045         else if (strcasecmp(name, "RESTORE") == 0)
2046                 return HSMA_RESTORE;
2047         else if (strcasecmp(name, "REMOVE") == 0)
2048                 return HSMA_REMOVE;
2049         else if (strcasecmp(name, "CANCEL") == 0)
2050                 return HSMA_CANCEL;
2051         else
2052                 return -1;
2053 }
2054
2055 static ssize_t
2056 mdt_write_hsm_request_mask(struct file *file, const char __user *user_buf,
2057                             size_t user_count, __u64 *mask)
2058 {
2059         char *buf, *pos, *name;
2060         size_t buf_size;
2061         __u64 new_mask = 0;
2062         int rc;
2063         ENTRY;
2064
2065         if (!(user_count < 4096))
2066                 RETURN(-ENOMEM);
2067
2068         buf_size = user_count + 1;
2069
2070         OBD_ALLOC(buf, buf_size);
2071         if (buf == NULL)
2072                 RETURN(-ENOMEM);
2073
2074         if (copy_from_user(buf, user_buf, buf_size - 1))
2075                 GOTO(out, rc = -EFAULT);
2076
2077         buf[buf_size - 1] = '\0';
2078
2079         pos = buf;
2080         while ((name = strsep(&pos, " \t\v\n")) != NULL) {
2081                 int action;
2082
2083                 if (*name == '\0')
2084                         continue;
2085
2086                 action = hsm_copytool_name2action(name);
2087                 if (action < 0)
2088                         GOTO(out, rc = -EINVAL);
2089
2090                 new_mask |= (1UL << action);
2091         }
2092
2093         *mask = new_mask;
2094         rc = user_count;
2095 out:
2096         OBD_FREE(buf, buf_size);
2097
2098         RETURN(rc);
2099 }
2100
2101 static ssize_t
2102 mdt_hsm_user_request_mask_seq_write(struct file *file, const char __user *buf,
2103                                         size_t count, loff_t *off)
2104 {
2105         struct seq_file         *m = file->private_data;
2106         struct mdt_device       *mdt = m->private;
2107         struct coordinator *cdt = &mdt->mdt_coordinator;
2108
2109         return mdt_write_hsm_request_mask(file, buf, count,
2110                                            &cdt->cdt_user_request_mask);
2111 }
2112
2113 static ssize_t
2114 mdt_hsm_group_request_mask_seq_write(struct file *file, const char __user *buf,
2115                                         size_t count, loff_t *off)
2116 {
2117         struct seq_file         *m = file->private_data;
2118         struct mdt_device       *mdt = m->private;
2119         struct coordinator      *cdt = &mdt->mdt_coordinator;
2120
2121         return mdt_write_hsm_request_mask(file, buf, count,
2122                                            &cdt->cdt_group_request_mask);
2123 }
2124
2125 static ssize_t
2126 mdt_hsm_other_request_mask_seq_write(struct file *file, const char __user *buf,
2127                                         size_t count, loff_t *off)
2128 {
2129         struct seq_file         *m = file->private_data;
2130         struct mdt_device       *mdt = m->private;
2131         struct coordinator      *cdt = &mdt->mdt_coordinator;
2132
2133         return mdt_write_hsm_request_mask(file, buf, count,
2134                                            &cdt->cdt_other_request_mask);
2135 }
2136
2137 LPROC_SEQ_FOPS(mdt_hsm_cdt_loop_period);
2138 LPROC_SEQ_FOPS(mdt_hsm_cdt_grace_delay);
2139 LPROC_SEQ_FOPS(mdt_hsm_cdt_active_req_timeout);
2140 LPROC_SEQ_FOPS(mdt_hsm_cdt_max_requests);
2141 LPROC_SEQ_FOPS(mdt_hsm_cdt_default_archive_id);
2142 LPROC_SEQ_FOPS(mdt_hsm_user_request_mask);
2143 LPROC_SEQ_FOPS(mdt_hsm_group_request_mask);
2144 LPROC_SEQ_FOPS(mdt_hsm_other_request_mask);
2145
2146 static struct lprocfs_vars lprocfs_mdt_hsm_vars[] = {
2147         { .name =       "agents",
2148           .fops =       &mdt_hsm_agent_fops                     },
2149         { .name =       "actions",
2150           .fops =       &mdt_hsm_actions_fops,
2151           .proc_mode =  0444                                    },
2152         { .name =       "default_archive_id",
2153           .fops =       &mdt_hsm_cdt_default_archive_id_fops    },
2154         { .name =       "grace_delay",
2155           .fops =       &mdt_hsm_cdt_grace_delay_fops           },
2156         { .name =       "loop_period",
2157           .fops =       &mdt_hsm_cdt_loop_period_fops           },
2158         { .name =       "max_requests",
2159           .fops =       &mdt_hsm_cdt_max_requests_fops          },
2160         { .name =       "policy",
2161           .fops =       &mdt_hsm_policy_fops                    },
2162         { .name =       "active_request_timeout",
2163           .fops =       &mdt_hsm_cdt_active_req_timeout_fops    },
2164         { .name =       "active_requests",
2165           .fops =       &mdt_hsm_active_requests_fops           },
2166         { .name =       "user_request_mask",
2167           .fops =       &mdt_hsm_user_request_mask_fops,        },
2168         { .name =       "group_request_mask",
2169           .fops =       &mdt_hsm_group_request_mask_fops,       },
2170         { .name =       "other_request_mask",
2171           .fops =       &mdt_hsm_other_request_mask_fops,       },
2172         { 0 }
2173 };