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