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