lustre/lfsck/lfsck_engine.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) 2013, 2016, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_engine.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lu_object.h>
  34 #include <dt_object.h>
  35 #include <lustre_net.h>
  36 #include <lustre_fid.h>
  37 #include <obd_support.h>
  38 #include <lustre_lib.h>
  39
  40 #include "lfsck_internal.h"
  41
  42 int lfsck_unpack_ent(struct lu_dirent *ent, __u64 *cookie, __u16 *type)
  43 {
  44         struct luda_type        *lt;
  45         int                      align = sizeof(*lt) - 1;
  46         int                      len;
  47
  48         fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
  49         *cookie = le64_to_cpu(ent->lde_hash);
  50         ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
  51         ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
  52         ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
  53
  54         if (unlikely(!(ent->lde_attrs & LUDA_TYPE)))
  55                 return -EINVAL;
  56
  57         len = (ent->lde_namelen + align) & ~align;
  58         lt = (struct luda_type *)(ent->lde_name + len);
  59         *type = le16_to_cpu(lt->lt_type);
  60
  61         /* Make sure the name is terminated with '\0'. The data (object type)
  62          * after ent::lde_name maybe broken, but we have stored such data in
  63          * the output parameter @type as above. */
  64         ent->lde_name[ent->lde_namelen] = '\0';
  65
  66         return 0;
  67 }
  68
  69 static void lfsck_di_oit_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  70 {
  71         const struct dt_it_ops  *iops;
  72         struct dt_it            *di;
  73
  74         spin_lock(&lfsck->li_lock);
  75         iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
  76         di = lfsck->li_di_oit;
  77         lfsck->li_di_oit = NULL;
  78         spin_unlock(&lfsck->li_lock);
  79         iops->put(env, di);
  80 }
  81
  82 static void lfsck_di_dir_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  83 {
  84         const struct dt_it_ops  *iops;
  85         struct dt_it            *di;
  86
  87         spin_lock(&lfsck->li_lock);
  88         iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
  89         di = lfsck->li_di_dir;
  90         lfsck->li_di_dir = NULL;
  91         lfsck->li_cookie_dir = 0;
  92         spin_unlock(&lfsck->li_lock);
  93         iops->put(env, di);
  94 }
  95
  96 /**
  97  * Check whether needs to scan the directory or not.
  98  *
  99  * 1) If we are not doing namespace LFSCK, or the given @obj is not directory,
 100  *    then needs not to scan the @obj. Otherwise,
 101  * 2) Global /ROOT needs to be scanned, backend root needs not to be scanned.
 102  * 3) If the @obj is neither IGIF nor normal FID (including .lustre and its
 103  *    sub-directories that have been scanned when the LFSCK engine start),
 104  *    then needs not to be scanned.
 105  * 4) If it is a remote object, then scanning the object will be done on the
 106  *    MDT on which the object really resides.
 107  * 5) If the local object has normal FID, then needs to be scanned. Otherwise,
 108  * 6) If the object has linkEA, then needs to be scanned. Otherwise,
 109  * 7) If none of the previous conditions are true, we need to check the parent
 110  *    directories whether this subdirectory is in a tree that should be scanned.
 111  *    Set the parent as current @obj, repeat 2)-7).
 112  *
 113  * \param[in] env       pointer to the thread context
 114  * \param[in] lfsck     pointer to the lfsck instance
 115  * \param[in] obj       pointer to the object to be checked
 116  *
 117  * \retval              positive number if the directory needs to be scanned
 118  * \retval              0 if the directory needs NOT to be scanned
 119  * \retval              negative error number on failure
 120  */
 121 static int lfsck_needs_scan_dir(const struct lu_env *env,
 122                                 struct lfsck_instance *lfsck,
 123                                 struct dt_object *obj)
 124 {
 125         struct lfsck_thread_info *info    = lfsck_env_info(env);
 126         struct lu_fid            *fid     = &info->lti_fid;
 127         struct lu_seq_range      *range   = &info->lti_range;
 128         struct lu_attr           *la      = &info->lti_la;
 129         struct seq_server_site   *ss      = lfsck_dev_site(lfsck);
 130         __u32                     idx     = lfsck_dev_idx(lfsck);
 131         int                       depth   = 0;
 132         int                       rc      = 0;
 133
 134         if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
 135                 return 0;
 136
 137         *fid = *lfsck_dto2fid(obj);
 138         rc = dt_attr_get(env, obj, la);
 139         if (unlikely(rc || (la->la_valid & LA_FLAGS &&
 140                             la->la_flags & LUSTRE_ORPHAN_FL))) {
 141                 /* Orphan directory is empty, does not need scan. */
 142                 CDEBUG(D_INFO,
 143                        "%s: skip orphan dir "DFID", %llx/%x: rc = %d\n",
 144                        lfsck_lfsck2name(lfsck), PFID(fid),
 145                        la->la_valid, la->la_flags, rc);
 146
 147                 return rc;
 148         }
 149
 150         LASSERT(ss != NULL);
 151
 152         while (1) {
 153                 /* Global /ROOT is visible. */
 154                 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
 155                         return 1;
 156
 157                 /* Backend root is invisible. */
 158                 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
 159                         return 0;
 160
 161                 if (!fid_is_norm(fid) && !fid_is_igif(fid))
 162                         return 0;
 163
 164                 fld_range_set_mdt(range);
 165                 rc = fld_local_lookup(env, ss->ss_server_fld,
 166                                       fid_seq(fid), range);
 167                 if (rc != 0 || range->lsr_index != idx)
 168                         /* Current FID should NOT be for the input parameter
 169                          * @obj, because the lfsck_master_oit_engine() has
 170                          * filtered out agent object. So current FID is for
 171                          * the ancestor of the original input parameter @obj.
 172                          * So the ancestor is a remote directory. The input
 173                          * parameter @obj is local directory, and should be
 174                          * scanned under such case. */
 175                         return 1;
 176
 177                 /* normal FID on this target (locally) must be for the
 178                  * client-side visiable object. */
 179                 if (fid_is_norm(fid))
 180                         return 1;
 181
 182                 /* Only true after "obj = NULL" set below */
 183                 if (obj == NULL) {
 184                         obj = lfsck_object_find_bottom(env, lfsck, fid);
 185                         if (IS_ERR(obj))
 186                                 return PTR_ERR(obj);
 187
 188                         depth++;
 189                         if (!dt_object_exists(obj))
 190                                 GOTO(out, rc = 0);
 191                 }
 192
 193                 dt_read_lock(env, obj, DT_TGT_CHILD);
 194                 if (unlikely(lfsck_is_dead_obj(obj))) {
 195                         dt_read_unlock(env, obj);
 196
 197                         GOTO(out, rc = 0);
 198                 }
 199
 200                 rc = dt_xattr_get(env, obj,
 201                                   lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK);
 202                 dt_read_unlock(env, obj);
 203                 if (rc >= 0)
 204                         GOTO(out, rc = 1);
 205
 206                 if (rc < 0 && rc != -ENODATA)
 207                         GOTO(out, rc);
 208
 209                 rc = dt_lookup_dir(env, obj, dotdot, fid);
 210                 if (depth > 0)
 211                         lfsck_object_put(env, obj);
 212
 213                 obj = NULL;
 214                 if (rc != 0)
 215                         return rc;
 216
 217                 if (!fid_is_sane(fid))
 218                         return 0;
 219         }
 220
 221 out:
 222         if (depth > 0 && obj != NULL)
 223                 lfsck_object_put(env, obj);
 224
 225         return rc;
 226 }
 227
 228 static int lfsck_load_stripe_lmv(const struct lu_env *env,
 229                                  struct lfsck_instance *lfsck,
 230                                  struct dt_object *obj)
 231 {
 232         struct lmv_mds_md_v1    *lmv    = &lfsck_env_info(env)->lti_lmv;
 233         struct lfsck_lmv        *llmv;
 234         int                      rc;
 235         ENTRY;
 236
 237         LASSERT(lfsck->li_obj_dir == NULL);
 238         LASSERT(lfsck->li_lmv == NULL);
 239
 240         rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
 241         if (rc == -ENODATA) {
 242                 lfsck->li_obj_dir = lfsck_object_get(obj);
 243
 244                 RETURN(0);
 245         }
 246
 247         if (rc < 0)
 248                 RETURN(rc);
 249
 250         OBD_ALLOC_PTR(llmv);
 251         if (llmv == NULL)
 252                 RETURN(-ENOMEM);
 253
 254         if (lmv->lmv_magic == LMV_MAGIC) {
 255                 struct lfsck_slave_lmv_rec      *lslr;
 256                 __u32                            stripes;
 257
 258                 llmv->ll_lmv_master = 1;
 259                 if (lmv->lmv_stripe_count < 1)
 260                         stripes = LFSCK_LMV_DEF_STRIPES;
 261                 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
 262                         stripes = LFSCK_LMV_MAX_STRIPES;
 263                 else
 264                         stripes = lmv->lmv_stripe_count;
 265
 266                 OBD_ALLOC_PTR_ARRAY_LARGE(lslr, stripes);
 267                 if (lslr == NULL) {
 268                         OBD_FREE_PTR(llmv);
 269
 270                         RETURN(-ENOMEM);
 271                 }
 272
 273                 llmv->ll_stripes_allocated = stripes;
 274                 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
 275                 llmv->ll_lslr = lslr;
 276         } else {
 277                 llmv->ll_lmv_slave = 1;
 278         }
 279
 280         lfsck->li_obj_dir = lfsck_object_get(obj);
 281         llmv->ll_lmv = *lmv;
 282         atomic_set(&llmv->ll_ref, 1);
 283         lfsck->li_lmv = llmv;
 284
 285         RETURN(0);
 286 }
 287
 288 /* LFSCK wrap functions */
 289
 290 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
 291                        bool new_checked)
 292 {
 293         struct lfsck_component *com;
 294
 295         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 296                 com->lc_ops->lfsck_fail(env, com, new_checked);
 297         }
 298 }
 299
 300 void lfsck_close_dir(const struct lu_env *env,
 301                      struct lfsck_instance *lfsck, int result)
 302 {
 303         struct lfsck_component *com;
 304         ENTRY;
 305
 306         if (lfsck->li_lmv != NULL) {
 307                 lfsck->li_lmv->ll_exit_value = result;
 308                 if (lfsck->li_obj_dir != NULL) {
 309                         list_for_each_entry(com, &lfsck->li_list_dir,
 310                                             lc_link_dir) {
 311                                 com->lc_ops->lfsck_close_dir(env, com);
 312                         }
 313                 }
 314
 315                 lfsck_lmv_put(env, lfsck->li_lmv);
 316                 lfsck->li_lmv = NULL;
 317         }
 318
 319         if (lfsck->li_di_dir != NULL) {
 320                 const struct dt_it_ops  *dir_iops;
 321                 struct dt_it            *dir_di   = lfsck->li_di_dir;
 322
 323                 LASSERT(lfsck->li_obj_dir != NULL);
 324
 325                 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
 326                 lfsck_di_dir_put(env, lfsck);
 327                 dir_iops->fini(env, dir_di);
 328         }
 329
 330         if (lfsck->li_obj_dir != NULL) {
 331                 struct dt_object        *dir_obj  = lfsck->li_obj_dir;
 332
 333                 lfsck->li_obj_dir = NULL;
 334                 lfsck_object_put(env, dir_obj);
 335         }
 336
 337         EXIT;
 338 }
 339
 340 int lfsck_open_dir(const struct lu_env *env,
 341                    struct lfsck_instance *lfsck, __u64 cookie)
 342 {
 343         struct dt_object        *obj    = lfsck->li_obj_dir;
 344         struct dt_it            *di     = lfsck->li_di_dir;
 345         struct lfsck_component  *com;
 346         const struct dt_it_ops  *iops;
 347         int                      rc     = 0;
 348         ENTRY;
 349
 350         LASSERT(obj != NULL);
 351         LASSERT(di == NULL);
 352
 353         if (unlikely(!dt_try_as_dir(env, obj, true)))
 354                 GOTO(out, rc = -ENOTDIR);
 355
 356         list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
 357                 rc = com->lc_ops->lfsck_open_dir(env, com);
 358                 if (rc != 0)
 359                         GOTO(out, rc);
 360         }
 361
 362         iops = &obj->do_index_ops->dio_it;
 363         di = iops->init(env, obj, lfsck->li_args_dir);
 364         if (IS_ERR(di))
 365                 GOTO(out, rc = PTR_ERR(di));
 366
 367         rc = iops->load(env, di, cookie);
 368         if (rc == -ENODATA)
 369                 rc = 1;
 370         else if (rc == 0 || (rc > 0 && cookie > 0))
 371                 rc = iops->next(env, di);
 372         else if (rc > 0)
 373                 rc = 0;
 374
 375         if (rc != 0) {
 376                 iops->put(env, di);
 377                 iops->fini(env, di);
 378         } else {
 379                 lfsck->li_cookie_dir = iops->store(env, di);
 380                 spin_lock(&lfsck->li_lock);
 381                 lfsck->li_di_dir = di;
 382                 spin_unlock(&lfsck->li_lock);
 383         }
 384
 385         GOTO(out, rc);
 386
 387 out:
 388         if (rc != 0)
 389                 lfsck_close_dir(env, lfsck, rc);
 390
 391         return rc;
 392 }
 393
 394 static int lfsck_checkpoint(const struct lu_env *env,
 395                             struct lfsck_instance *lfsck)
 396 {
 397         struct lfsck_component *com;
 398         int                     rc  = 0;
 399         int                     rc1 = 0;
 400
 401         if (likely(ktime_get_seconds() <= lfsck->li_time_next_checkpoint))
 402                 return 0;
 403
 404         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 405         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 406                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
 407                 if (rc != 0)
 408                         rc1 = rc;
 409         }
 410
 411         lfsck->li_time_last_checkpoint = ktime_get_seconds();
 412         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 413                                          LFSCK_CHECKPOINT_INTERVAL;
 414         return rc1 != 0 ? rc1 : rc;
 415 }
 416
 417 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 418                       struct lfsck_start_param *lsp)
 419 {
 420         struct dt_object       *obj     = NULL;
 421         struct lfsck_component *com;
 422         struct lfsck_component *next;
 423         struct lfsck_position  *pos     = NULL;
 424         const struct dt_it_ops *iops    =
 425                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 426         int                     rc;
 427         ENTRY;
 428
 429         LASSERT(lfsck->li_obj_dir == NULL);
 430         LASSERT(lfsck->li_di_dir == NULL);
 431
 432         lfsck->li_current_oit_processed = 0;
 433         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 434                 com->lc_new_checked = 0;
 435                 rc = com->lc_ops->lfsck_prep(env, com, lsp);
 436                 if (rc != 0)
 437                         GOTO(out, rc);
 438
 439                 if ((pos == NULL) ||
 440                     (!lfsck_pos_is_zero(&com->lc_pos_start) &&
 441                      lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
 442                         pos = &com->lc_pos_start;
 443         }
 444
 445         /* Init otable-based iterator. */
 446         if (pos == NULL) {
 447                 rc = iops->load(env, lfsck->li_di_oit, 0);
 448                 if (rc > 0 || unlikely(rc == -ENODATA)) {
 449                         lfsck->li_oit_over = 1;
 450                         rc = 0;
 451                 }
 452
 453                 GOTO(out, rc);
 454         }
 455
 456         rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
 457         if (rc > 0 || unlikely(rc == -ENODATA))
 458                 lfsck->li_oit_over = 1;
 459         else if (rc < 0)
 460                 GOTO(out, rc);
 461
 462         if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
 463                 GOTO(out, rc = 0);
 464
 465         /* Find the directory for namespace-based traverse. */
 466         obj = lfsck_object_find_bottom(env, lfsck, &pos->lp_dir_parent);
 467         if (IS_ERR(obj))
 468                 RETURN(PTR_ERR(obj));
 469
 470         /* Remote directory will be scanned by the LFSCK instance
 471          * on the MDT where the remote object really resides on. */
 472         if (!dt_object_exists(obj) || dt_object_remote(obj) ||
 473             unlikely(!S_ISDIR(lfsck_object_type(obj))))
 474                 GOTO(out, rc = 0);
 475
 476         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 477         if (rc == 0) {
 478                 /* For the master MDT-object of a striped directory,
 479                  * reset the iteration from the directory beginning. */
 480                 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
 481                         pos->lp_dir_cookie = 0;
 482
 483                 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
 484                 if (rc > 0)
 485                         /* The end of the directory. */
 486                         rc = 0;
 487         }
 488
 489         GOTO(out, rc);
 490
 491 out:
 492         if (obj != NULL)
 493                 lfsck_object_put(env, obj);
 494
 495         if (rc != 0) {
 496                 lfsck_close_dir(env, lfsck, rc);
 497                 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 498                                          lc_link) {
 499                         com->lc_ops->lfsck_post(env, com, rc, true);
 500                 }
 501
 502                 return rc;
 503         }
 504
 505         rc = 0;
 506         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
 507         lfsck->li_pos_current = lfsck->li_pos_checkpoint;
 508         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 509                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
 510                 if (rc != 0)
 511                         break;
 512         }
 513
 514         lfsck->li_time_last_checkpoint = ktime_get_seconds();
 515         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 516                                          LFSCK_CHECKPOINT_INTERVAL;
 517         return rc;
 518 }
 519
 520 static int lfsck_exec_oit(const struct lu_env *env,
 521                           struct lfsck_instance *lfsck, struct dt_object *obj)
 522 {
 523         struct lfsck_component *com;
 524         int                     rc;
 525         ENTRY;
 526
 527         LASSERT(lfsck->li_obj_dir == NULL);
 528
 529         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 530                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
 531                 if (rc != 0)
 532                         RETURN(rc);
 533         }
 534
 535         rc = lfsck_needs_scan_dir(env, lfsck, obj);
 536         if (rc <= 0)
 537                 GOTO(out, rc);
 538
 539         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 540         if (rc == 0)
 541                 rc = lfsck_open_dir(env, lfsck, 0);
 542
 543         GOTO(out, rc);
 544
 545 out:
 546         if (rc < 0)
 547                 lfsck_fail(env, lfsck, false);
 548
 549         if (rc != 0)
 550                 lfsck_close_dir(env, lfsck, rc);
 551
 552         return rc > 0 ? 0 : rc;
 553 }
 554
 555 static int lfsck_exec_dir(const struct lu_env *env,
 556                           struct lfsck_instance *lfsck,
 557                           struct lfsck_assistant_object *lso,
 558                           struct lu_dirent *ent, __u16 type)
 559 {
 560         struct lfsck_component *com;
 561         int                     rc;
 562
 563         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 564                 rc = com->lc_ops->lfsck_exec_dir(env, com, lso, ent, type);
 565                 if (rc != 0)
 566                         return rc;
 567         }
 568         return 0;
 569 }
 570
 571 static int lfsck_master_dir_engine(const struct lu_env *env,
 572                                    struct lfsck_instance *lfsck);
 573
 574 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 575                       int result)
 576 {
 577         struct lfsck_component *com;
 578         struct lfsck_component *next;
 579         int                     rc  = result;
 580
 581         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 582         lfsck_close_dir(env, lfsck, result);
 583
 584         while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
 585                !list_empty(&lfsck->li_list_lmv)) {
 586                 struct lfsck_lmv_unit *llu;
 587
 588                 spin_lock(&lfsck->li_lock);
 589                 llu = list_first_entry(&lfsck->li_list_lmv,
 590                                        struct lfsck_lmv_unit, llu_link);
 591                 list_del_init(&llu->llu_link);
 592                 spin_unlock(&lfsck->li_lock);
 593
 594                 lfsck->li_lmv = &llu->llu_lmv;
 595                 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 596                 rc = lfsck_open_dir(env, lfsck, 0);
 597                 if (rc == 0) {
 598                         rc = lfsck_master_dir_engine(env, lfsck);
 599                         lfsck_close_dir(env, lfsck, result);
 600                 }
 601         }
 602
 603         result = rc;
 604
 605         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 606                 rc = com->lc_ops->lfsck_post(env, com, result, false);
 607                 if (rc != 0)
 608                         CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
 609                                "rc = %d\n", lfsck_lfsck2name(lfsck),
 610                                (__u32)com->lc_type, rc);
 611         }
 612
 613         lfsck->li_time_last_checkpoint = ktime_get_seconds();
 614         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 615                                          LFSCK_CHECKPOINT_INTERVAL;
 616
 617         /* Ignore some component post failure to make other can go ahead. */
 618         return result;
 619 }
 620
 621 static int lfsck_double_scan(const struct lu_env *env,
 622                              struct lfsck_instance *lfsck)
 623 {
 624         struct lfsck_component *com;
 625         struct lfsck_component *next;
 626         int                     rc  = 0;
 627         int                     rc1 = 0;
 628
 629         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
 630                 rc = com->lc_ops->lfsck_double_scan(env, com);
 631                 if (rc != 0)
 632                         rc1 = rc;
 633         }
 634
 635         wait_event_idle(lfsck->li_thread.t_ctl_waitq,
 636                         atomic_read(&lfsck->li_double_scan_count) == 0);
 637
 638         if (lfsck->li_status != LS_PAUSED &&
 639             lfsck->li_status != LS_CO_PAUSED) {
 640                 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 641                                          lc_link) {
 642                         spin_lock(&lfsck->li_lock);
 643                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 644                         spin_unlock(&lfsck->li_lock);
 645                 }
 646         }
 647
 648         return rc1 != 0 ? rc1 : rc;
 649 }
 650
 651 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
 652 {
 653         struct lfsck_component *com;
 654         struct lfsck_component *next;
 655
 656         list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 657                                  lc_link) {
 658                 if (com->lc_ops->lfsck_quit != NULL)
 659                         com->lc_ops->lfsck_quit(env, com);
 660
 661                 spin_lock(&lfsck->li_lock);
 662                 list_del_init(&com->lc_link_dir);
 663                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 664                 spin_unlock(&lfsck->li_lock);
 665         }
 666
 667         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 668                                  lc_link) {
 669                 if (com->lc_ops->lfsck_quit != NULL)
 670                         com->lc_ops->lfsck_quit(env, com);
 671
 672                 spin_lock(&lfsck->li_lock);
 673                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 674                 spin_unlock(&lfsck->li_lock);
 675         }
 676 }
 677
 678 /* LFSCK engines */
 679
 680 static int lfsck_master_dir_engine(const struct lu_env *env,
 681                                    struct lfsck_instance *lfsck)
 682 {
 683         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 684         struct dt_object                *dir    = lfsck->li_obj_dir;
 685         const struct dt_it_ops          *iops   = &dir->do_index_ops->dio_it;
 686         struct dt_it                    *di     = lfsck->li_di_dir;
 687         struct lu_dirent                *ent    =
 688                         (struct lu_dirent *)info->lti_key;
 689         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 690         struct ptlrpc_thread            *thread = &lfsck->li_thread;
 691         struct lfsck_assistant_object   *lso    = NULL;
 692         int                              rc;
 693         __u16                            type;
 694         ENTRY;
 695
 696         do {
 697                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
 698                     unlikely(!thread_is_running(thread))) {
 699                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 700                                "parent "DFID", cookie %#llx\n",
 701                                lfsck_lfsck2name(lfsck),
 702                                PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
 703
 704                         GOTO(out, rc = 0);
 705                 }
 706
 707                 lfsck->li_new_scanned++;
 708                 rc = iops->rec(env, di, (struct dt_rec *)ent,
 709                                lfsck->li_args_dir);
 710                 if (rc == 0)
 711                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
 712                                               &type);
 713
 714                 if (rc != 0) {
 715                         CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
 716                                "parent "DFID", cookie %#llx: rc = %d\n",
 717                                lfsck_lfsck2name(lfsck),
 718                                PFID(lfsck_dto2fid(dir)),
 719                                lfsck->li_cookie_dir, rc);
 720                         lfsck_fail(env, lfsck, true);
 721                         if (bk->lb_param & LPF_FAILOUT)
 722                                 GOTO(out, rc);
 723                         else
 724                                 goto checkpoint;
 725                 }
 726
 727                 if (ent->lde_attrs & LUDA_IGNORE)
 728                         goto checkpoint;
 729
 730                 /* skip dot entry. */
 731                 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
 732                         goto checkpoint;
 733
 734                 if (lso == NULL) {
 735                         lso = lfsck_assistant_object_init(env,
 736                                 lfsck_dto2fid(dir), NULL,
 737                                 lfsck->li_pos_current.lp_oit_cookie, true);
 738                         if (IS_ERR(lso)) {
 739                                 if (bk->lb_param & LPF_FAILOUT)
 740                                         RETURN(PTR_ERR(lso));
 741
 742                                 lso = NULL;
 743                                 goto checkpoint;
 744                         }
 745                 }
 746
 747                 /* The type in the @ent structure may has been overwritten,
 748                  * so we need to pass the @type parameter independently. */
 749                 rc = lfsck_exec_dir(env, lfsck, lso, ent, type);
 750                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 751                         GOTO(out, rc);
 752
 753 checkpoint:
 754                 rc = lfsck_checkpoint(env, lfsck);
 755                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 756                         GOTO(out, rc);
 757
 758                 /* Rate control. */
 759                 lfsck_control_speed(lfsck);
 760                 if (unlikely(!thread_is_running(thread))) {
 761                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 762                                "parent "DFID", cookie %#llx\n",
 763                                lfsck_lfsck2name(lfsck),
 764                                PFID(lfsck_dto2fid(dir)),
 765                                lfsck->li_cookie_dir);
 766                         GOTO(out, rc = 0);
 767                 }
 768
 769                 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
 770                         spin_lock(&lfsck->li_lock);
 771                         thread_set_flags(thread, SVC_STOPPING);
 772                         spin_unlock(&lfsck->li_lock);
 773                         GOTO(out, rc = -EINVAL);
 774                 }
 775
 776                 rc = iops->next(env, di);
 777                 if (rc < 0)
 778                         CDEBUG(D_LFSCK, "%s dir engine fail to locate next "
 779                                "for the directory "DFID": rc = %d\n",
 780                                lfsck_lfsck2name(lfsck),
 781                                PFID(&lfsck->li_pos_current.lp_dir_parent), rc);
 782         } while (rc == 0);
 783
 784         if (rc > 0 && !lfsck->li_oit_over)
 785                 lfsck_close_dir(env, lfsck, rc);
 786
 787         GOTO(out, rc);
 788
 789 out:
 790         if (lso != NULL)
 791                 lfsck_assistant_object_put(env, lso);
 792
 793         return rc;
 794 }
 795
 796 /**
 797  * Object-table based iteration engine.
 798  *
 799  * Object-table based iteration is the basic linear engine to scan all the
 800  * objects on current device in turn. For each object, it calls all the
 801  * registered LFSCK component(s)' API to perform related consistency
 802  * verification.
 803  *
 804  * It flushes related LFSCK trace files to disk via making checkpoint
 805  * periodically. Then if the server crashed or the LFSCK is paused, the
 806  * LFSCK can resume from the latest checkpoint.
 807  *
 808  * It also controls the whole LFSCK speed via lfsck_control_speed() to
 809  * avoid the server to become overload.
 810  *
 811  * \param[in] env       pointer to the thread context
 812  * \param[in] lfsck     pointer to the lfsck instance
 813  *
 814  * \retval              positive number if all objects have been scanned
 815  * \retval              0 if the iteration is stopped or paused
 816  * \retval              negative error number on failure
 817  */
 818 static int lfsck_master_oit_engine(const struct lu_env *env,
 819                                    struct lfsck_instance *lfsck)
 820 {
 821         struct lfsck_thread_info *info  = lfsck_env_info(env);
 822         const struct dt_it_ops   *iops  =
 823                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 824         struct dt_it             *di    = lfsck->li_di_oit;
 825         struct lu_fid            *fid   = &info->lti_fid;
 826         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 827         struct ptlrpc_thread     *thread = &lfsck->li_thread;
 828         struct seq_server_site   *ss    = lfsck_dev_site(lfsck);
 829         __u32                    idx    = lfsck_dev_idx(lfsck);
 830         int                      rc;
 831         ENTRY;
 832
 833         if (unlikely(ss == NULL))
 834                 RETURN(-EIO);
 835
 836         do {
 837                 struct dt_object *target;
 838
 839                 if (lfsck->li_di_dir != NULL) {
 840                         rc = lfsck_master_dir_engine(env, lfsck);
 841                         if (rc <= 0)
 842                                 RETURN(rc);
 843                 }
 844
 845                 if (unlikely(lfsck->li_oit_over))
 846                         RETURN(1);
 847
 848                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
 849                     unlikely(!thread_is_running(thread))) {
 850                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
 851                                "cookie %llu\n",
 852                                lfsck_lfsck2name(lfsck), iops->store(env, di));
 853
 854                         RETURN(0);
 855                 }
 856
 857                 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
 858                         RETURN(0);
 859
 860                 lfsck->li_current_oit_processed = 1;
 861
 862                 if (!list_empty(&lfsck->li_list_lmv)) {
 863                         struct lfsck_lmv_unit *llu;
 864
 865                         spin_lock(&lfsck->li_lock);
 866                         llu = list_first_entry(&lfsck->li_list_lmv,
 867                                                struct lfsck_lmv_unit, llu_link);
 868                         list_del_init(&llu->llu_link);
 869                         spin_unlock(&lfsck->li_lock);
 870
 871                         lfsck->li_lmv = &llu->llu_lmv;
 872                         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 873                         rc = lfsck_open_dir(env, lfsck, 0);
 874                         if (rc == 0)
 875                                 rc = lfsck_master_dir_engine(env, lfsck);
 876
 877                         if (rc <= 0)
 878                                 RETURN(rc);
 879                 }
 880
 881                 lfsck->li_new_scanned++;
 882                 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
 883                 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
 884                 if (rc != 0) {
 885                         CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
 886                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 887                         lfsck_fail(env, lfsck, true);
 888                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
 889                                 RETURN(rc);
 890                         else
 891                                 goto checkpoint;
 892                 }
 893
 894                 if (unlikely(!fid_is_sane(fid))) {
 895                         CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
 896                                ", skip it\n",
 897                                lfsck_lfsck2name(lfsck), PFID(fid));
 898                         goto checkpoint;
 899                 }
 900
 901                 if (fid_is_idif(fid)) {
 902                         __u32 idx1 = fid_idif_ost_idx(fid);
 903
 904                         LASSERT(!lfsck->li_master);
 905
 906                         if (idx != idx1) {
 907                                 struct ost_id *oi = &info->lti_oi;
 908
 909                                 if (unlikely(idx1 != 0)) {
 910                                         CDEBUG(D_LFSCK, "%s: invalid IDIF "DFID
 911                                                ", not match device index %u\n",
 912                                                lfsck_lfsck2name(lfsck),
 913                                                PFID(fid), idx);
 914
 915                                         goto checkpoint;
 916                                 }
 917
 918                                 /* rebuild the IDIF with index to
 919                                  * avoid double instances for the
 920                                  * same object. */
 921                                 fid_to_ostid(fid, oi);
 922                                 ostid_to_fid(fid, oi, idx);
 923                         }
 924                 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
 925                            !fid_is_last_id(fid) &&
 926                            !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
 927
 928                         /* If the FID/object is only used locally and invisible
 929                          * to external nodes, then LFSCK will not handle it.
 930                          *
 931                          * dot_lustre sequence has been handled specially. */
 932                         goto checkpoint;
 933                 } else {
 934                         struct lu_seq_range *range = &info->lti_range;
 935
 936                         if (lfsck->li_master)
 937                                 fld_range_set_mdt(range);
 938                         else
 939                                 fld_range_set_ost(range);
 940                         rc = fld_local_lookup(env, ss->ss_server_fld,
 941                                               fid_seq(fid), range);
 942                         if (rc != 0 || range->lsr_index != idx) {
 943                                 /* Remote object will be handled by the LFSCK
 944                                  * instance on the MDT where the remote object
 945                                  * really resides on. */
 946                                 rc = 0;
 947                                 goto checkpoint;
 948                         }
 949                 }
 950
 951                 target = lfsck_object_find_bottom(env, lfsck, fid);
 952                 if (IS_ERR(target)) {
 953                         CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
 954                                DFID", cookie %llu: rc = %d\n",
 955                                lfsck_lfsck2name(lfsck), PFID(fid),
 956                                iops->store(env, di), rc);
 957                         lfsck_fail(env, lfsck, true);
 958                         if (bk->lb_param & LPF_FAILOUT)
 959                                 RETURN(PTR_ERR(target));
 960                         else
 961                                 goto checkpoint;
 962                 }
 963
 964                 if (dt_object_exists(target))
 965                         rc = lfsck_exec_oit(env, lfsck, target);
 966
 967                 lfsck_object_put(env, target);
 968                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 969                         RETURN(rc);
 970
 971 checkpoint:
 972                 rc = lfsck_checkpoint(env, lfsck);
 973                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 974                         RETURN(rc);
 975
 976                 /* Rate control. */
 977                 lfsck_control_speed(lfsck);
 978
 979                 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
 980                         spin_lock(&lfsck->li_lock);
 981                         thread_set_flags(thread, SVC_STOPPING);
 982                         spin_unlock(&lfsck->li_lock);
 983                         RETURN(-EINVAL);
 984                 }
 985
 986                 rc = iops->next(env, di);
 987                 if (unlikely(rc > 0))
 988                         lfsck->li_oit_over = 1;
 989                 else if (likely(rc == 0))
 990                         lfsck->li_current_oit_processed = 0;
 991                 else
 992                         CDEBUG(D_LFSCK, "%s oit engine fail to locate next at "
 993                                "%llu: rc = %d\n", lfsck_lfsck2name(lfsck),
 994                                iops->store(env, di), rc);
 995
 996                 if (unlikely(!thread_is_running(thread))) {
 997                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
 998                                "cookie %llu\n", lfsck_lfsck2name(lfsck),
 999                                iops->store(env, di));
1000                         RETURN(0);
1001                 }
1002         } while (rc == 0 || lfsck->li_di_dir != NULL);
1003
1004         RETURN(rc);
1005 }
1006
1007 int lfsck_master_engine(void *args)
1008 {
1009         struct lfsck_thread_args *lta      = args;
1010         struct lu_env            *env      = &lta->lta_env;
1011         struct lfsck_instance    *lfsck    = lta->lta_lfsck;
1012         struct ptlrpc_thread     *thread   = &lfsck->li_thread;
1013         struct dt_object         *oit_obj  = lfsck->li_obj_oit;
1014         const struct dt_it_ops   *oit_iops = &oit_obj->do_index_ops->dio_it;
1015         struct dt_it             *oit_di;
1016         int                       rc;
1017         ENTRY;
1018
1019         /*
1020          * thread is spawned with all signals set to SIG_IGN, re-enable
1021          * SIGINT for lfsck_stop() to awaken and stop the thread.
1022          */
1023         allow_signal(SIGINT);
1024         spin_lock(&lfsck->li_lock);
1025         lfsck->li_task = current;
1026         spin_unlock(&lfsck->li_lock);
1027
1028         /* There will be some objects verification during the LFSCK start,
1029          * such as the subsequent lfsck_verify_lpf(). Trigger low layer OI
1030          * OI scrub before that to handle the potential inconsistence. */
1031         oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit);
1032         if (IS_ERR(oit_di)) {
1033                 rc = PTR_ERR(oit_di);
1034                 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1035                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1036
1037                 GOTO(fini_args, rc);
1038         }
1039
1040         if (lfsck->li_master &&
1041             (!list_empty(&lfsck->li_list_scan) ||
1042              !list_empty(&lfsck->li_list_double_scan))) {
1043                 rc = lfsck_verify_lpf(env, lfsck);
1044                 /* FIXME: once OI files are missing, this will fail, it should
1045                  * return error, but to satisfy sanity-lfsck test 4 & 5, leave
1046                  * it uninitialized here, and any code dereference it need to
1047                  * check.
1048                  */
1049                 if (rc != 0)
1050                         CERROR("%s: master engine fail to verify the "
1051                                ".lustre/lost+found/, go ahead: rc = %d\n",
1052                                lfsck_lfsck2name(lfsck), rc);
1053         }
1054
1055         spin_lock(&lfsck->li_lock);
1056         lfsck->li_di_oit = oit_di;
1057         spin_unlock(&lfsck->li_lock);
1058         rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1059         if (rc != 0)
1060                 GOTO(fini_oit, rc);
1061
1062         CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1063                "oit_cookie = %llu, dir_cookie = %#llx, parent = "DFID
1064                ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1065                lfsck->li_pos_checkpoint.lp_oit_cookie,
1066                lfsck->li_pos_checkpoint.lp_dir_cookie,
1067                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1068                current->pid);
1069
1070         spin_lock(&lfsck->li_lock);
1071         if (unlikely(!thread_is_starting(thread))) {
1072                 spin_unlock(&lfsck->li_lock);
1073                 GOTO(fini_oit, rc = 0);
1074         }
1075
1076         thread_set_flags(thread, SVC_RUNNING);
1077         spin_unlock(&lfsck->li_lock);
1078         wake_up(&thread->t_ctl_waitq);
1079
1080         wait_event_idle(thread->t_ctl_waitq,
1081                         lfsck->li_start_unplug ||
1082                         !thread_is_running(thread));
1083         if (!thread_is_running(thread))
1084                 GOTO(fini_oit, rc = 0);
1085
1086         if (!list_empty(&lfsck->li_list_scan) ||
1087             list_empty(&lfsck->li_list_double_scan))
1088                 rc = lfsck_master_oit_engine(env, lfsck);
1089         else
1090                 rc = 1;
1091
1092         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1093         CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1094                "oit_cookie = %llu, dir_cookie = %#llx, parent = "DFID
1095                ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1096                lfsck->li_pos_checkpoint.lp_oit_cookie,
1097                lfsck->li_pos_checkpoint.lp_dir_cookie,
1098                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1099                current->pid, rc);
1100
1101         if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1102                 rc = lfsck_post(env, lfsck, rc);
1103         else
1104                 lfsck_close_dir(env, lfsck, rc);
1105
1106 fini_oit:
1107         lfsck_di_oit_put(env, lfsck);
1108         oit_iops->fini(env, oit_di);
1109         if (rc == 1) {
1110                 if (!list_empty(&lfsck->li_list_double_scan))
1111                         rc = lfsck_double_scan(env, lfsck);
1112                 else
1113                         rc = 0;
1114         } else {
1115                 lfsck_quit(env, lfsck);
1116         }
1117
1118         /* XXX: Purge the pinned objects in the future. */
1119
1120 fini_args:
1121         spin_lock(&lfsck->li_lock);
1122         thread_set_flags(thread, SVC_STOPPED);
1123         lfsck->li_task = NULL;
1124         spin_unlock(&lfsck->li_lock);
1125         wake_up(&thread->t_ctl_waitq);
1126         lfsck_thread_args_fini(lta);
1127         return rc;
1128 }
1129
1130 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1131 {
1132         bool empty = false;
1133
1134         spin_lock(&lad->lad_lock);
1135         if (list_empty(&lad->lad_req_list))
1136                 empty = true;
1137         spin_unlock(&lad->lad_lock);
1138
1139         return empty;
1140 }
1141
1142 /**
1143  * Query the LFSCK status from the instatnces on remote servers.
1144  *
1145  * The LFSCK assistant thread queries the LFSCK instances on other
1146  * servers (MDT/OST) about their status, such as whether they have
1147  * finished the phase1/phase2 scanning or not, and so on.
1148  *
1149  * \param[in] env       pointer to the thread context
1150  * \param[in] com       pointer to the lfsck component
1151  *
1152  * \retval              0 for success
1153  * \retval              negative error number on failure
1154  */
1155 static int lfsck_assistant_query_others(const struct lu_env *env,
1156                                         struct lfsck_component *com)
1157 {
1158         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1159         struct lfsck_request              *lr    = &info->lti_lr;
1160         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1161         struct lfsck_instance             *lfsck = com->lc_lfsck;
1162         struct lfsck_assistant_data       *lad   = com->lc_data;
1163         struct ptlrpc_request_set         *set;
1164         struct lfsck_tgt_descs            *ltds;
1165         struct lfsck_tgt_desc             *ltd;
1166         struct list_head                  *phase_head;
1167         int                                rc    = 0;
1168         int                                rc1   = 0;
1169         ENTRY;
1170
1171         set = ptlrpc_prep_set();
1172         if (set == NULL)
1173                 RETURN(-ENOMEM);
1174
1175         lad->lad_touch_gen++;
1176         memset(lr, 0, sizeof(*lr));
1177         lr->lr_event = LE_QUERY;
1178         lr->lr_active = com->lc_type;
1179
1180         memset(laia, 0, sizeof(*laia));
1181         laia->laia_com = com;
1182         laia->laia_lr = lr;
1183
1184         if (!list_empty(&lad->lad_mdt_phase1_list)) {
1185                 ltds = &lfsck->li_mdt_descs;
1186                 lr->lr_flags = 0;
1187                 phase_head = &lad->lad_mdt_phase1_list;
1188         } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1189                 goto out;
1190         } else {
1191
1192 again:
1193                 ltds = &lfsck->li_ost_descs;
1194                 lr->lr_flags = LEF_TO_OST;
1195                 phase_head = &lad->lad_ost_phase1_list;
1196         }
1197
1198         laia->laia_ltds = ltds;
1199         spin_lock(&ltds->ltd_lock);
1200         while (!list_empty(phase_head)) {
1201                 struct list_head *phase_list;
1202                 __u32            *gen;
1203
1204                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1205                         ltd = list_first_entry(phase_head,
1206                                                struct lfsck_tgt_desc,
1207                                                ltd_layout_phase_list);
1208                         phase_list = &ltd->ltd_layout_phase_list;
1209                         gen = &ltd->ltd_layout_gen;
1210                 } else {
1211                         ltd = list_first_entry(phase_head,
1212                                                struct lfsck_tgt_desc,
1213                                                ltd_namespace_phase_list);
1214                         phase_list = &ltd->ltd_namespace_phase_list;
1215                         gen = &ltd->ltd_namespace_gen;
1216                 }
1217
1218                 if (*gen == lad->lad_touch_gen)
1219                         break;
1220
1221                 *gen = lad->lad_touch_gen;
1222                 list_move_tail(phase_list, phase_head);
1223                 atomic_inc(&ltd->ltd_ref);
1224                 laia->laia_ltd = ltd;
1225                 spin_unlock(&ltds->ltd_lock);
1226                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1227                                          lfsck_async_interpret_common,
1228                                          laia, LFSCK_QUERY);
1229                 if (rc != 0) {
1230                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1231                                "%s %x for %s: rc = %d\n",
1232                                lfsck_lfsck2name(lfsck),
1233                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1234                                ltd->ltd_index, lad->lad_name, rc);
1235                         lfsck_tgt_put(ltd);
1236                         rc1 = rc;
1237                 }
1238                 spin_lock(&ltds->ltd_lock);
1239         }
1240         spin_unlock(&ltds->ltd_lock);
1241
1242         rc = ptlrpc_set_wait(env, set);
1243         if (rc < 0) {
1244                 ptlrpc_set_destroy(set);
1245                 RETURN(rc);
1246         }
1247
1248         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1249             list_empty(&lad->lad_mdt_phase1_list))
1250                 goto again;
1251
1252 out:
1253         ptlrpc_set_destroy(set);
1254
1255         RETURN(rc1 != 0 ? rc1 : rc);
1256 }
1257
1258 /**
1259  * Notify the LFSCK event to the instances on remote servers.
1260  *
1261  * The LFSCK assistant thread notifies the LFSCK instances on other
1262  * servers (MDT/OST) about some events, such as start new scanning,
1263  * stop the scanning, this LFSCK instance will exit, and so on.
1264  *
1265  * \param[in] env       pointer to the thread context
1266  * \param[in] com       pointer to the lfsck component
1267  * \param[in] lr        pointer to the LFSCK event request
1268  *
1269  * \retval              0 for success
1270  * \retval              negative error number on failure
1271  */
1272 static int lfsck_assistant_notify_others(const struct lu_env *env,
1273                                          struct lfsck_component *com,
1274                                          struct lfsck_request *lr)
1275 {
1276         struct lfsck_thread_info *info = lfsck_env_info(env);
1277         struct lfsck_async_interpret_args *laia = &info->lti_laia;
1278         struct lfsck_instance *lfsck = com->lc_lfsck;
1279         struct lfsck_assistant_data *lad = com->lc_data;
1280         struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1281         struct ptlrpc_request_set *set;
1282         struct lfsck_tgt_descs *ltds;
1283         struct lfsck_tgt_desc *ltd;
1284         struct lfsck_tgt_desc *next;
1285         int idx;
1286         int rc = 0;
1287         int rc1 = 0;
1288
1289         ENTRY;
1290         set = ptlrpc_prep_set();
1291         if (set == NULL)
1292                 RETURN(-ENOMEM);
1293
1294         lr->lr_index = lfsck_dev_idx(lfsck);
1295         lr->lr_active = com->lc_type;
1296
1297         memset(laia, 0, sizeof(*laia));
1298         laia->laia_com = com;
1299         laia->laia_lr = lr;
1300
1301         switch (lr->lr_event) {
1302         case LE_START:
1303                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1304                         goto next;
1305
1306                 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1307                 lr->lr_speed = bk->lb_speed_limit;
1308                 lr->lr_version = bk->lb_version;
1309                 lr->lr_param |= bk->lb_param;
1310                 lr->lr_async_windows = bk->lb_async_windows;
1311                 lr->lr_flags = LEF_TO_OST;
1312
1313                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1314                 ltds = &lfsck->li_ost_descs;
1315                 laia->laia_ltds = ltds;
1316                 down_read(&ltds->ltd_rw_sem);
1317                 for_each_set_bit(idx, ltds->ltd_tgts_bitmap,
1318                                  ltds->ltd_tgts_mask_len) {
1319                         ltd = lfsck_tgt_get(ltds, idx);
1320                         LASSERT(ltd != NULL);
1321
1322                         laia->laia_ltd = ltd;
1323                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1324                                         lfsck_async_interpret_common,
1325                                         laia, LFSCK_NOTIFY);
1326                         if (rc != 0) {
1327                                 lfsck_lad_set_bitmap(env, com, idx);
1328                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1329                                        "notify OST %x for %s start: rc = %d\n",
1330                                        lfsck_lfsck2name(lfsck), idx,
1331                                        lad->lad_name, rc);
1332                                 lfsck_tgt_put(ltd);
1333                         }
1334                 }
1335                 up_read(&ltds->ltd_rw_sem);
1336
1337                 /* Sync up */
1338                 rc = ptlrpc_set_wait(env, set);
1339                 if (rc < 0) {
1340                         ptlrpc_set_destroy(set);
1341                         RETURN(rc);
1342                 }
1343
1344 next:
1345                 if (!(bk->lb_param & LPF_ALL_TGT))
1346                         break;
1347
1348                 /* link other MDT targets locallly. */
1349                 ltds = &lfsck->li_mdt_descs;
1350                 spin_lock(&ltds->ltd_lock);
1351                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1352                         for_each_set_bit(idx, ltds->ltd_tgts_bitmap,
1353                                          ltds->ltd_tgts_mask_len) {
1354                                 ltd = lfsck_ltd2tgt(ltds, idx);
1355                                 LASSERT(ltd != NULL);
1356
1357                                 if (!list_empty(&ltd->ltd_layout_list))
1358                                         continue;
1359
1360                                 list_add_tail(&ltd->ltd_layout_list,
1361                                               &lad->lad_mdt_list);
1362                                 list_add_tail(&ltd->ltd_layout_phase_list,
1363                                               &lad->lad_mdt_phase1_list);
1364                         }
1365                 } else {
1366                         for_each_set_bit(idx, ltds->ltd_tgts_bitmap,
1367                                          ltds->ltd_tgts_mask_len) {
1368                                 ltd = lfsck_ltd2tgt(ltds, idx);
1369                                 LASSERT(ltd != NULL);
1370
1371                                 if (!list_empty(&ltd->ltd_namespace_list))
1372                                         continue;
1373
1374                                 list_add_tail(&ltd->ltd_namespace_list,
1375                                               &lad->lad_mdt_list);
1376                                 list_add_tail(&ltd->ltd_namespace_phase_list,
1377                                               &lad->lad_mdt_phase1_list);
1378                         }
1379                 }
1380                 spin_unlock(&ltds->ltd_lock);
1381                 break;
1382         case LE_STOP:
1383         case LE_PHASE2_DONE:
1384         case LE_PEER_EXIT: {
1385                 struct list_head *phase_head;
1386
1387                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1388                 if (bk->lb_param & LPF_ALL_TGT) {
1389                         phase_head = &lad->lad_mdt_list;
1390                         ltds = &lfsck->li_mdt_descs;
1391                         if (lr->lr_event == LE_STOP) {
1392                                 /* unlink other MDT targets locallly. */
1393                                 spin_lock(&ltds->ltd_lock);
1394                                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1395                                         list_for_each_entry_safe(ltd, next,
1396                                                 phase_head, ltd_layout_list) {
1397                                                 list_del_init(
1398                                                 &ltd->ltd_layout_phase_list);
1399                                                 list_del_init(
1400                                                 &ltd->ltd_layout_list);
1401                                         }
1402                                 } else {
1403                                         list_for_each_entry_safe(ltd, next,
1404                                                         phase_head,
1405                                                         ltd_namespace_list) {
1406                                                 list_del_init(
1407                                                 &ltd->ltd_namespace_phase_list);
1408                                                 list_del_init(
1409                                                 &ltd->ltd_namespace_list);
1410                                         }
1411                                 }
1412                                 spin_unlock(&ltds->ltd_lock);
1413
1414                                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1415                                         break;
1416
1417                                 lr->lr_flags |= LEF_TO_OST;
1418                                 phase_head = &lad->lad_ost_list;
1419                                 ltds = &lfsck->li_ost_descs;
1420                         } else {
1421                                 lr->lr_flags &= ~LEF_TO_OST;
1422                         }
1423                 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1424                         break;
1425                 } else {
1426                         lr->lr_flags |= LEF_TO_OST;
1427                         phase_head = &lad->lad_ost_list;
1428                         ltds = &lfsck->li_ost_descs;
1429                 }
1430
1431 again:
1432                 laia->laia_ltds = ltds;
1433                 spin_lock(&ltds->ltd_lock);
1434                 while (!list_empty(phase_head)) {
1435                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1436                                 ltd = list_first_entry(phase_head,
1437                                                        struct lfsck_tgt_desc,
1438                                                        ltd_layout_list);
1439                                 if (!list_empty(&ltd->ltd_layout_phase_list))
1440                                         list_del_init(
1441                                                 &ltd->ltd_layout_phase_list);
1442                                 list_del_init(&ltd->ltd_layout_list);
1443                         } else {
1444                                 ltd = list_first_entry(phase_head,
1445                                                        struct lfsck_tgt_desc,
1446                                                        ltd_namespace_list);
1447                                 if (!list_empty(&ltd->ltd_namespace_phase_list))
1448                                         list_del_init(
1449                                                 &ltd->ltd_namespace_phase_list);
1450                                 list_del_init(&ltd->ltd_namespace_list);
1451                         }
1452                         atomic_inc(&ltd->ltd_ref);
1453                         laia->laia_ltd = ltd;
1454                         spin_unlock(&ltds->ltd_lock);
1455                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1456                                         lfsck_async_interpret_common,
1457                                         laia, LFSCK_NOTIFY);
1458                         if (rc != 0) {
1459                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1460                                        "notify %s %x for %s stop/phase2_done/"
1461                                        "peer_exit: rc = %d\n",
1462                                        lfsck_lfsck2name(lfsck),
1463                                        (lr->lr_flags & LEF_TO_OST) ?
1464                                        "OST" : "MDT", ltd->ltd_index,
1465                                        lad->lad_name, rc);
1466                                 lfsck_tgt_put(ltd);
1467                         }
1468                         spin_lock(&ltds->ltd_lock);
1469                 }
1470                 spin_unlock(&ltds->ltd_lock);
1471
1472                 rc = ptlrpc_set_wait(env, set);
1473                 if (rc < 0) {
1474                         ptlrpc_set_destroy(set);
1475                         RETURN(rc);
1476                 }
1477
1478                 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1479                     !(lr->lr_flags & LEF_TO_OST)) {
1480                         lr->lr_flags |= LEF_TO_OST;
1481                         phase_head = &lad->lad_ost_list;
1482                         ltds = &lfsck->li_ost_descs;
1483                         goto again;
1484                 }
1485                 break;
1486         }
1487         case LE_PHASE1_DONE:
1488                 lad->lad_ops->la_sync_failures(env, com, lr);
1489                 lad->lad_touch_gen++;
1490                 ltds = &lfsck->li_mdt_descs;
1491                 laia->laia_ltds = ltds;
1492                 spin_lock(&ltds->ltd_lock);
1493                 while (!list_empty(&lad->lad_mdt_list)) {
1494                         struct list_head *list;
1495                         __u32            *gen;
1496
1497                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1498                                 ltd = list_first_entry(&lad->lad_mdt_list,
1499                                                        struct lfsck_tgt_desc,
1500                                                        ltd_layout_list);
1501                                 list = &ltd->ltd_layout_list;
1502                                 gen = &ltd->ltd_layout_gen;
1503                         } else {
1504                                 struct lfsck_namespace *ns = com->lc_file_ram;
1505
1506                                 ltd = list_first_entry(&lad->lad_mdt_list,
1507                                                        struct lfsck_tgt_desc,
1508                                                        ltd_namespace_list);
1509                                 list = &ltd->ltd_namespace_list;
1510                                 gen = &ltd->ltd_namespace_gen;
1511                                 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1512                         }
1513
1514                         if (*gen == lad->lad_touch_gen)
1515                                 break;
1516
1517                         *gen = lad->lad_touch_gen;
1518                         list_move_tail(list, &lad->lad_mdt_list);
1519                         if (ltd->ltd_synced_failures)
1520                                 continue;
1521
1522                         atomic_inc(&ltd->ltd_ref);
1523                         laia->laia_ltd = ltd;
1524                         spin_unlock(&ltds->ltd_lock);
1525                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1526                                         lfsck_async_interpret_common,
1527                                         laia, LFSCK_NOTIFY);
1528                         if (rc != 0) {
1529                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1530                                        "notify MDT %x for %s phase1 done: "
1531                                        "rc = %d\n", lfsck_lfsck2name(lfsck),
1532                                        ltd->ltd_index, lad->lad_name, rc);
1533                                 lfsck_tgt_put(ltd);
1534                         }
1535                         spin_lock(&ltds->ltd_lock);
1536                 }
1537                 spin_unlock(&ltds->ltd_lock);
1538                 break;
1539         default:
1540                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1541                        "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1542                 rc = -EINVAL;
1543                 break;
1544         }
1545
1546         rc1 = ptlrpc_set_wait(env, set);
1547         ptlrpc_set_destroy(set);
1548
1549         RETURN(rc != 0 ? rc : rc1);
1550 }
1551
1552 /**
1553  * The LFSCK assistant thread is triggered by the LFSCK main engine.
1554  * They co-work together as an asynchronous pipeline: the LFSCK main
1555  * engine scans the system and pre-fetches the objects, attributes,
1556  * or name entries, etc, and pushes them into the pipeline as input
1557  * requests for the LFSCK assistant thread; on the other end of the
1558  * pipeline, the LFSCK assistant thread performs the real check and
1559  * repair for every request from the main engine.
1560  *
1561  * Generally, the assistant engine may be blocked when check/repair
1562  * something, so the LFSCK main engine will run some faster. On the
1563  * other hand, the LFSCK main engine will drive multiple assistant
1564  * threads in parallel, means for each LFSCK component on the master
1565  * (such as layout LFSCK, namespace LFSCK), there is an independent
1566  * LFSCK assistant thread. So under such 1:N multiple asynchronous
1567  * pipelines mode, the whole LFSCK performance will be much better
1568  * than check/repair everything by the LFSCK main engine itself.
1569  */
1570 int lfsck_assistant_engine(void *args)
1571 {
1572         struct lfsck_thread_args          *lta     = args;
1573         struct lu_env                     *env     = &lta->lta_env;
1574         struct lfsck_component            *com     = lta->lta_com;
1575         struct lfsck_instance             *lfsck   = lta->lta_lfsck;
1576         struct lfsck_bookmark             *bk      = &lfsck->li_bookmark_ram;
1577         struct lfsck_position             *pos     = &com->lc_pos_start;
1578         struct lfsck_thread_info          *info    = lfsck_env_info(env);
1579         struct lfsck_request              *lr      = &info->lti_lr;
1580         struct lfsck_assistant_data       *lad     = com->lc_data;
1581         struct ptlrpc_thread              *mthread = &lfsck->li_thread;
1582         struct ptlrpc_thread              *athread = &lad->lad_thread;
1583         const struct lfsck_assistant_operations *lao     = lad->lad_ops;
1584         struct lfsck_assistant_req        *lar;
1585         int                                rc      = 0;
1586         int                                rc1     = 0;
1587         int                                rc2;
1588         ENTRY;
1589
1590         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1591                lfsck_lfsck2name(lfsck), lad->lad_name);
1592
1593         memset(lr, 0, sizeof(*lr));
1594         lr->lr_event = LE_START;
1595         if (pos->lp_oit_cookie <= 1)
1596                 lr->lr_param = LPF_RESET;
1597         rc = lfsck_assistant_notify_others(env, com, lr);
1598         if (rc != 0) {
1599                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1600                        "to start %s: rc = %d\n",
1601                        lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1602                 GOTO(fini, rc);
1603         }
1604
1605         /*
1606          * thread is spawned with all signals set to SIG_IGN, re-enable
1607          * SIGINT for lfsck_stop() to awaken and stop the thread.
1608          */
1609         allow_signal(SIGINT);
1610         spin_lock(&lad->lad_lock);
1611         lad->lad_task = current;
1612         thread_set_flags(athread, SVC_RUNNING);
1613         spin_unlock(&lad->lad_lock);
1614         wake_up(&mthread->t_ctl_waitq);
1615
1616         while (1) {
1617                 while (!list_empty(&lad->lad_req_list)) {
1618                         bool wakeup = false;
1619
1620                         if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags) ||
1621                                      !thread_is_running(mthread)))
1622                                 GOTO(cleanup, rc = lad->lad_post_result);
1623
1624                         lar = list_first_entry(&lad->lad_req_list,
1625                                                struct lfsck_assistant_req,
1626                                                lar_list);
1627                         /* Only the lfsck_assistant_engine thread itself can
1628                          * remove the "lar" from the head of the list, LFSCK
1629                          * engine thread only inserts other new "lar" at the
1630                          * end of the list. So it is safe to handle current
1631                          * "lar" without the spin_lock. */
1632                         rc = lao->la_handler_p1(env, com, lar);
1633                         spin_lock(&lad->lad_lock);
1634                         list_del_init(&lar->lar_list);
1635                         lad->lad_prefetched--;
1636                         /* Wake up the main engine thread only when the list
1637                          * is empty or half of the prefetched items have been
1638                          * handled to avoid too frequent thread schedule. */
1639                         if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1640                                 wakeup = true;
1641                         spin_unlock(&lad->lad_lock);
1642                         if (wakeup)
1643                                 wake_up(&mthread->t_ctl_waitq);
1644
1645                         lao->la_req_fini(env, lar);
1646                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1647                                 GOTO(cleanup, rc);
1648                 }
1649
1650                 wait_event_idle(athread->t_ctl_waitq,
1651                                 !lfsck_assistant_req_empty(lad) ||
1652                                 test_bit(LAD_EXIT, &lad->lad_flags) ||
1653                                 test_bit(LAD_TO_POST, &lad->lad_flags) ||
1654                                 test_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags));
1655
1656                 if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags)))
1657                         GOTO(cleanup, rc = lad->lad_post_result);
1658
1659                 if (!list_empty(&lad->lad_req_list))
1660                         continue;
1661
1662                 if (test_bit(LAD_TO_POST, &lad->lad_flags)) {
1663                         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1664                                lfsck_lfsck2name(lfsck), lad->lad_name);
1665
1666                         if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags)))
1667                                 GOTO(cleanup, rc = lad->lad_post_result);
1668
1669                         clear_bit(LAD_TO_POST, &lad->lad_flags);
1670                         LASSERT(lad->lad_post_result > 0);
1671
1672                         /* Wakeup the master engine to go ahead. */
1673                         wake_up(&mthread->t_ctl_waitq);
1674
1675                         memset(lr, 0, sizeof(*lr));
1676                         lr->lr_event = LE_PHASE1_DONE;
1677                         lr->lr_status = lad->lad_post_result;
1678                         rc = lfsck_assistant_notify_others(env, com, lr);
1679
1680                         CDEBUG(D_LFSCK, "%s: LFSCK assistant notified "
1681                                "others for %s post: rc = %d\n",
1682                                lfsck_lfsck2name(lfsck),
1683                                lad->lad_name, rc);
1684                 }
1685
1686                 if (test_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags)) {
1687                         clear_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags);
1688                         atomic_inc(&lfsck->li_double_scan_count);
1689                         set_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags);
1690                         wake_up(&mthread->t_ctl_waitq);
1691
1692                         com->lc_new_checked = 0;
1693                         com->lc_new_scanned = 0;
1694                         com->lc_time_last_checkpoint = ktime_get_seconds();
1695                         com->lc_time_next_checkpoint =
1696                                 com->lc_time_last_checkpoint +
1697                                 LFSCK_CHECKPOINT_INTERVAL;
1698
1699                         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1700                                "the second-stage scaning\n",
1701                                lfsck_lfsck2name(lfsck));
1702
1703                         /* Flush async updates before handling orphan. */
1704                         rc2 = dt_sync(env, lfsck->li_next);
1705
1706                         CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1707                                "scan start, synced: rc = %d\n",
1708                                lfsck_lfsck2name(lfsck), rc2);
1709
1710                         if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1711                                 GOTO(cleanup, rc = 0);
1712
1713                         while (test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags)) {
1714                                 int seconds = 30;
1715
1716                                 rc = lfsck_assistant_query_others(env, com);
1717                                 if (lfsck_phase2_next_ready(lad))
1718                                         goto p2_next;
1719
1720                                 if (rc < 0)
1721                                         GOTO(cleanup, rc);
1722
1723                                 /* Pull LFSCK status on related targets once
1724                                  * per 30 seconds if we are not notified. */
1725                                 while (seconds > 0 &&
1726                                        wait_event_idle_timeout(
1727                                                athread->t_ctl_waitq,
1728                                                lfsck_phase2_next_ready(lad) ||
1729                                                test_bit(LAD_EXIT,
1730                                                         &lad->lad_flags) ||
1731                                                !thread_is_running(mthread),
1732                                                cfs_time_seconds(1)) == 0)
1733                                         seconds -= 1;
1734
1735                                 if (unlikely(
1736                                         test_bit(LAD_EXIT, &lad->lad_flags) ||
1737                                         !thread_is_running(mthread)))
1738                                         GOTO(cleanup, rc = 0);
1739
1740                                 if (seconds == 0)
1741                                         continue;
1742
1743 p2_next:
1744                                 rc = lao->la_handler_p2(env, com);
1745                                 if (rc != 0)
1746                                         GOTO(cleanup, rc);
1747
1748                                 if (unlikely(
1749                                         test_bit(LAD_EXIT, &lad->lad_flags) ||
1750                                         !thread_is_running(mthread)))
1751                                         GOTO(cleanup, rc = 0);
1752                         }
1753                 }
1754         }
1755
1756 cleanup:
1757         /* Cleanup the unfinished requests. */
1758         spin_lock(&lad->lad_lock);
1759         if (rc < 0)
1760                 lad->lad_assistant_status = rc;
1761
1762         if (test_bit(LAD_EXIT, &lad->lad_flags) && lad->lad_post_result <= 0)
1763                 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1764
1765         thread_set_flags(athread, SVC_STOPPING);
1766         while (!list_empty(&lad->lad_req_list)) {
1767                 lar = list_first_entry(&lad->lad_req_list,
1768                                        struct lfsck_assistant_req,
1769                                        lar_list);
1770                 list_del_init(&lar->lar_list);
1771                 lad->lad_prefetched--;
1772                 spin_unlock(&lad->lad_lock);
1773                 lao->la_req_fini(env, lar);
1774                 spin_lock(&lad->lad_lock);
1775         }
1776         spin_unlock(&lad->lad_lock);
1777
1778         memset(lr, 0, sizeof(*lr));
1779         if (rc > 0) {
1780                 lr->lr_event = LE_PHASE2_DONE;
1781                 lr->lr_status = rc;
1782         } else if (rc == 0) {
1783                 if (lfsck->li_flags & LPF_ALL_TGT) {
1784                         lr->lr_event = LE_STOP;
1785                         lr->lr_status = LS_STOPPED;
1786                 } else {
1787                         lr->lr_event = LE_PEER_EXIT;
1788                         switch (lfsck->li_status) {
1789                         case LS_PAUSED:
1790                         case LS_CO_PAUSED:
1791                                 lr->lr_status = LS_CO_PAUSED;
1792                                 break;
1793                         case LS_STOPPED:
1794                         case LS_CO_STOPPED:
1795                                 lr->lr_status = LS_CO_STOPPED;
1796                                 break;
1797                         default:
1798                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1799                                        "status: rc = %d\n",
1800                                        lfsck_lfsck2name(lfsck),
1801                                        lfsck->li_status);
1802                                 lr->lr_status = LS_CO_FAILED;
1803                                 break;
1804                         }
1805                 }
1806         } else {
1807                 if (lfsck->li_flags & LPF_ALL_TGT) {
1808                         lr->lr_event = LE_STOP;
1809                         lr->lr_status = LS_FAILED;
1810                 } else {
1811                         lr->lr_event = LE_PEER_EXIT;
1812                         lr->lr_status = LS_CO_FAILED;
1813                 }
1814         }
1815
1816         rc1 = lfsck_assistant_notify_others(env, com, lr);
1817         if (rc1 != 0) {
1818                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1819                        "others for %s quit: rc = %d\n",
1820                        lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1821                 rc = rc1;
1822         }
1823
1824         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1825                lfsck_lfsck2name(lfsck));
1826
1827         /* Flush async updates before exit. */
1828         rc2 = dt_sync(env, lfsck->li_next);
1829
1830         CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1831                lfsck_lfsck2name(lfsck), rc2);
1832
1833         /* Under force exit case, some requests may be just freed without
1834          * verification, those objects should be re-handled when next run.
1835          * So not update the on-disk trace file under such case. */
1836         if (test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags)) {
1837                 if (!test_bit(LAD_EXIT, &lad->lad_flags))
1838                         rc1 = lao->la_double_scan_result(env, com, rc);
1839
1840                 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1841                        "finished: rc = %d\n",
1842                        lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1843         }
1844
1845 fini:
1846         if (test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags))
1847                 atomic_dec(&lfsck->li_double_scan_count);
1848
1849         spin_lock(&lad->lad_lock);
1850         lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1851         thread_set_flags(athread, SVC_STOPPED);
1852         lad->lad_task = NULL;
1853         spin_unlock(&lad->lad_lock);
1854
1855         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1856                lfsck_lfsck2name(lfsck), lad->lad_name,
1857                lad->lad_assistant_status);
1858
1859         lfsck_thread_args_fini(lta);
1860         wake_up(&mthread->t_ctl_waitq);
1861
1862         return rc;
1863 }