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, 2014, 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_update_lma(const struct lu_env *env,
  97                             struct lfsck_instance *lfsck, struct dt_object *obj)
  98 {
  99         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 100         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 101         struct dt_device                *dt     = lfsck->li_bottom;
 102         struct lustre_mdt_attrs         *lma    = &info->lti_lma;
 103         struct lu_buf                   *buf;
 104         struct thandle                  *th;
 105         int                              fl;
 106         int                              rc;
 107         ENTRY;
 108
 109         if (bk->lb_param & LPF_DRYRUN)
 110                 RETURN(0);
 111
 112         buf = lfsck_buf_get(env, info->lti_lma_old, LMA_OLD_SIZE);
 113         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LMA, BYPASS_CAPA);
 114         if (rc < 0) {
 115                 if (rc != -ENODATA)
 116                         RETURN(rc);
 117
 118                 fl = LU_XATTR_CREATE;
 119                 lustre_lma_init(lma, lfsck_dto2fid(obj), LMAC_FID_ON_OST, 0);
 120         } else {
 121                 if (rc != LMA_OLD_SIZE && rc != sizeof(struct lustre_mdt_attrs))
 122                         RETURN(-EINVAL);
 123
 124                 fl = LU_XATTR_REPLACE;
 125                 lustre_lma_swab(lma);
 126                 lustre_lma_init(lma, lfsck_dto2fid(obj),
 127                                 lma->lma_compat | LMAC_FID_ON_OST,
 128                                 lma->lma_incompat);
 129         }
 130         lustre_lma_swab(lma);
 131
 132         th = dt_trans_create(env, dt);
 133         if (IS_ERR(th))
 134                 RETURN(PTR_ERR(th));
 135
 136         buf = lfsck_buf_get(env, lma, sizeof(*lma));
 137         rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
 138         if (rc != 0)
 139                 GOTO(stop, rc);
 140
 141         rc = dt_trans_start(env, dt, th);
 142         if (rc != 0)
 143                 GOTO(stop, rc);
 144
 145         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th, BYPASS_CAPA);
 146
 147         GOTO(stop, rc);
 148
 149 stop:
 150         dt_trans_stop(env, dt, th);
 151         return rc;
 152 }
 153
 154 static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj,
 155                             struct lu_fid *fid)
 156 {
 157         if (unlikely(!S_ISDIR(lfsck_object_type(obj)) ||
 158                      !dt_try_as_dir(env, obj)))
 159                 return -ENOTDIR;
 160
 161         return dt_lookup(env, obj, (struct dt_rec *)fid,
 162                          (const struct dt_key *)"..", BYPASS_CAPA);
 163 }
 164
 165 /**
 166  * Check whether needs to scan the directory or not.
 167  *
 168  * 1) If we are not doing namespace LFSCK, or the given @obj is not directory,
 169  *    then needs not to scan the @obj. Otherwise,
 170  * 2) Global /ROOT needs to be scanned, backend root needs not to be scanned.
 171  * 3) If the @obj is neither IGIF nor normal FID (including .lustre and its
 172  *    sub-directories that have been scanned when the LFSCK engine start),
 173  *    then needs not to be scanned.
 174  * 4) If it is a remote object, then scanning the object will be done on the
 175  *    MDT on which the object really resides.
 176  * 5) If the local object has normal FID, then needs to be scanned. Otherwise,
 177  * 6) If the object has linkEA, then needs to be scanned. Otherwise,
 178  * 7) If none of the previous conditions are true, we need to check the parent
 179  *    directories whether this subdirectory is in a tree that should be scanned.
 180  *    Set the parent as current @obj, repeat 2)-7).
 181  *
 182  * \param[in] env       pointer to the thread context
 183  * \param[in] lfsck     pointer to the lfsck instance
 184  * \param[in] obj       pointer to the object to be checked
 185  *
 186  * \retval              positive number if the directory needs to be scanned
 187  * \retval              0 if the directory needs NOT to be scanned
 188  * \retval              negative error number on failure
 189  */
 190 static int lfsck_needs_scan_dir(const struct lu_env *env,
 191                                 struct lfsck_instance *lfsck,
 192                                 struct dt_object *obj)
 193 {
 194         struct lfsck_thread_info *info    = lfsck_env_info(env);
 195         struct lu_fid            *fid     = &info->lti_fid;
 196         struct lu_seq_range      *range   = &info->lti_range;
 197         struct dt_device         *dev     = lfsck->li_bottom;
 198         struct seq_server_site   *ss      = lu_site2seq(dev->dd_lu_dev.ld_site);
 199         __u32                     idx     = lfsck_dev_idx(dev);
 200         int                       depth   = 0;
 201         int                       rc      = 0;
 202
 203         if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
 204                 return 0;
 205
 206         LASSERT(ss != NULL);
 207
 208         *fid = *lfsck_dto2fid(obj);
 209         while (1) {
 210                 /* Global /ROOT is visible. */
 211                 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
 212                         return 1;
 213
 214                 /* Backend root is invisible. */
 215                 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
 216                         return 0;
 217
 218                 if (!fid_is_norm(fid) && !fid_is_igif(fid))
 219                         return 0;
 220
 221                 fld_range_set_mdt(range);
 222                 rc = fld_local_lookup(env, ss->ss_server_fld,
 223                                       fid_seq(fid), range);
 224                 if (rc != 0 || range->lsr_index != idx)
 225                         /* Current FID should NOT be for the input parameter
 226                          * @obj, because the lfsck_master_oit_engine() has
 227                          * filtered out agent object. So current FID is for
 228                          * the ancestor of the original input parameter @obj.
 229                          * So the ancestor is a remote directory. The input
 230                          * parameter @obj is local directory, and should be
 231                          * scanned under such case. */
 232                         return 1;
 233
 234                 /* normal FID on this target (locally) must be for the
 235                  * client-side visiable object. */
 236                 if (fid_is_norm(fid))
 237                         return 1;
 238
 239                 if (obj == NULL) {
 240                         obj = lfsck_object_find(env, lfsck, fid);
 241                         if (IS_ERR(obj))
 242                                 return PTR_ERR(obj);
 243
 244                         depth++;
 245                         if (!dt_object_exists(obj))
 246                                 GOTO(out, rc = 0);
 247                 }
 248
 249                 dt_read_lock(env, obj, MOR_TGT_CHILD);
 250                 if (unlikely(lfsck_is_dead_obj(obj))) {
 251                         dt_read_unlock(env, obj);
 252
 253                         GOTO(out, rc = 0);
 254                 }
 255
 256                 rc = dt_xattr_get(env, obj,
 257                                   lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK,
 258                                   BYPASS_CAPA);
 259                 dt_read_unlock(env, obj);
 260                 if (rc >= 0)
 261                         GOTO(out, rc = 1);
 262
 263                 if (rc < 0 && rc != -ENODATA)
 264                         GOTO(out, rc);
 265
 266                 rc = lfsck_parent_fid(env, obj, fid);
 267                 if (depth > 0)
 268                         lfsck_object_put(env, obj);
 269
 270                 obj = NULL;
 271                 if (rc != 0)
 272                         return rc;
 273
 274                 if (!fid_is_sane(fid))
 275                         return 0;
 276         }
 277
 278 out:
 279         if (depth > 0 && obj != NULL)
 280                 lfsck_object_put(env, obj);
 281
 282         return rc;
 283 }
 284
 285 static int lfsck_load_stripe_lmv(const struct lu_env *env,
 286                                  struct lfsck_instance *lfsck,
 287                                  struct dt_object *obj)
 288 {
 289         struct lmv_mds_md_v1    *lmv    = &lfsck_env_info(env)->lti_lmv;
 290         struct lfsck_lmv        *llmv;
 291         int                      rc;
 292         ENTRY;
 293
 294         LASSERT(lfsck->li_obj_dir == NULL);
 295         LASSERT(lfsck->li_lmv == NULL);
 296
 297         rc = lfsck_read_stripe_lmv(env, obj, lmv);
 298         if (rc == -ENODATA) {
 299                 lfsck->li_obj_dir = lfsck_object_get(obj);
 300
 301                 RETURN(0);
 302         }
 303
 304         if (rc < 0)
 305                 RETURN(rc);
 306
 307         OBD_ALLOC_PTR(llmv);
 308         if (llmv == NULL)
 309                 RETURN(-ENOMEM);
 310
 311         if (lmv->lmv_magic == LMV_MAGIC) {
 312                 struct lfsck_slave_lmv_rec      *lslr;
 313                 __u32                            stripes;
 314
 315                 llmv->ll_lmv_master = 1;
 316                 if (lmv->lmv_stripe_count < 1)
 317                         stripes = LFSCK_LMV_DEF_STRIPES;
 318                 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
 319                         stripes = LFSCK_LMV_MAX_STRIPES;
 320                 else
 321                         stripes = lmv->lmv_stripe_count;
 322
 323                 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
 324                 if (lslr == NULL) {
 325                         OBD_FREE_PTR(llmv);
 326
 327                         RETURN(-ENOMEM);
 328                 }
 329
 330                 /* Find the object against the bottom device. */
 331                 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
 332                                                lfsck_dto2fid(obj));
 333                 if (IS_ERR(obj)) {
 334                         OBD_FREE_LARGE(lslr, sizeof(*lslr) * stripes);
 335                         OBD_FREE_PTR(llmv);
 336
 337                         RETURN(PTR_ERR(obj));
 338                 }
 339
 340                 llmv->ll_stripes_allocated = stripes;
 341                 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
 342                 llmv->ll_lslr = lslr;
 343                 lfsck->li_obj_dir = obj;
 344         } else {
 345                 llmv->ll_lmv_slave = 1;
 346                 lfsck->li_obj_dir = lfsck_object_get(obj);
 347         }
 348
 349         llmv->ll_lmv = *lmv;
 350         atomic_set(&llmv->ll_ref, 1);
 351         lfsck->li_lmv = llmv;
 352
 353         RETURN(0);
 354 }
 355
 356 /* LFSCK wrap functions */
 357
 358 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
 359                        bool new_checked)
 360 {
 361         struct lfsck_component *com;
 362
 363         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 364                 com->lc_ops->lfsck_fail(env, com, new_checked);
 365         }
 366 }
 367
 368 void lfsck_close_dir(const struct lu_env *env,
 369                      struct lfsck_instance *lfsck, int result)
 370 {
 371         struct lfsck_component *com;
 372         ENTRY;
 373
 374         if (lfsck->li_lmv != NULL) {
 375                 lfsck->li_lmv->ll_exit_value = result;
 376                 if (lfsck->li_obj_dir != NULL) {
 377                         list_for_each_entry(com, &lfsck->li_list_dir,
 378                                             lc_link_dir) {
 379                                 com->lc_ops->lfsck_close_dir(env, com);
 380                         }
 381                 }
 382
 383                 lfsck_lmv_put(env, lfsck->li_lmv);
 384                 lfsck->li_lmv = NULL;
 385         }
 386
 387         if (lfsck->li_di_dir != NULL) {
 388                 const struct dt_it_ops  *dir_iops;
 389                 struct dt_it            *dir_di   = lfsck->li_di_dir;
 390
 391                 LASSERT(lfsck->li_obj_dir != NULL);
 392
 393                 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
 394                 lfsck_di_dir_put(env, lfsck);
 395                 dir_iops->fini(env, dir_di);
 396         }
 397
 398         if (lfsck->li_obj_dir != NULL) {
 399                 struct dt_object        *dir_obj  = lfsck->li_obj_dir;
 400
 401                 lfsck->li_obj_dir = NULL;
 402                 lfsck_object_put(env, dir_obj);
 403         }
 404
 405         EXIT;
 406 }
 407
 408 int lfsck_open_dir(const struct lu_env *env,
 409                    struct lfsck_instance *lfsck, __u64 cookie)
 410 {
 411         struct dt_object        *obj    = lfsck->li_obj_dir;
 412         struct dt_it            *di     = lfsck->li_di_dir;
 413         struct lfsck_component  *com;
 414         const struct dt_it_ops  *iops;
 415         int                      rc     = 0;
 416         ENTRY;
 417
 418         LASSERT(obj != NULL);
 419         LASSERT(di == NULL);
 420
 421         if (unlikely(!dt_try_as_dir(env, obj)))
 422                 GOTO(out, rc = -ENOTDIR);
 423
 424         list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
 425                 rc = com->lc_ops->lfsck_open_dir(env, com);
 426                 if (rc != 0)
 427                         GOTO(out, rc);
 428         }
 429
 430         iops = &obj->do_index_ops->dio_it;
 431         di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
 432         if (IS_ERR(di))
 433                 GOTO(out, rc = PTR_ERR(di));
 434
 435         rc = iops->load(env, di, cookie);
 436         if (rc == 0 || (rc > 0 && cookie > 0))
 437                 rc = iops->next(env, di);
 438         else if (rc > 0)
 439                 rc = 0;
 440
 441         if (rc != 0) {
 442                 iops->put(env, di);
 443                 iops->fini(env, di);
 444         } else {
 445                 lfsck->li_cookie_dir = iops->store(env, di);
 446                 spin_lock(&lfsck->li_lock);
 447                 lfsck->li_di_dir = di;
 448                 spin_unlock(&lfsck->li_lock);
 449         }
 450
 451         GOTO(out, rc);
 452
 453 out:
 454         if (rc != 0)
 455                 lfsck_close_dir(env, lfsck, rc);
 456
 457         return rc;
 458 }
 459
 460 static int lfsck_checkpoint(const struct lu_env *env,
 461                             struct lfsck_instance *lfsck)
 462 {
 463         struct lfsck_component *com;
 464         int                     rc  = 0;
 465         int                     rc1 = 0;
 466
 467         if (likely(cfs_time_beforeq(cfs_time_current(),
 468                                     lfsck->li_time_next_checkpoint)))
 469                 return 0;
 470
 471         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 472         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 473                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
 474                 if (rc != 0)
 475                         rc1 = rc;
 476         }
 477
 478         lfsck->li_time_last_checkpoint = cfs_time_current();
 479         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 480                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 481         return rc1 != 0 ? rc1 : rc;
 482 }
 483
 484 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 485                       struct lfsck_start_param *lsp)
 486 {
 487         struct dt_object       *obj     = NULL;
 488         struct lfsck_component *com;
 489         struct lfsck_component *next;
 490         struct lfsck_position  *pos     = NULL;
 491         const struct dt_it_ops *iops    =
 492                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 493         int                     rc;
 494         ENTRY;
 495
 496         LASSERT(lfsck->li_obj_dir == NULL);
 497         LASSERT(lfsck->li_di_dir == NULL);
 498
 499         lfsck->li_current_oit_processed = 0;
 500         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 501                 com->lc_new_checked = 0;
 502                 rc = com->lc_ops->lfsck_prep(env, com, lsp);
 503                 if (rc != 0)
 504                         GOTO(out, rc);
 505
 506                 if ((pos == NULL) ||
 507                     (!lfsck_pos_is_zero(&com->lc_pos_start) &&
 508                      lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
 509                         pos = &com->lc_pos_start;
 510         }
 511
 512         /* Init otable-based iterator. */
 513         if (pos == NULL) {
 514                 rc = iops->load(env, lfsck->li_di_oit, 0);
 515                 if (rc > 0) {
 516                         lfsck->li_oit_over = 1;
 517                         rc = 0;
 518                 }
 519
 520                 GOTO(out, rc);
 521         }
 522
 523         rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
 524         if (rc < 0)
 525                 GOTO(out, rc);
 526         else if (rc > 0)
 527                 lfsck->li_oit_over = 1;
 528
 529         if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
 530                 GOTO(out, rc = 0);
 531
 532         /* Find the directory for namespace-based traverse. */
 533         obj = lfsck_object_find(env, lfsck, &pos->lp_dir_parent);
 534         if (IS_ERR(obj))
 535                 RETURN(PTR_ERR(obj));
 536
 537         /* Remote directory will be scanned by the LFSCK instance
 538          * on the MDT where the remote object really resides on. */
 539         if (!dt_object_exists(obj) || dt_object_remote(obj) ||
 540             unlikely(!S_ISDIR(lfsck_object_type(obj))))
 541                 GOTO(out, rc = 0);
 542
 543         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 544         if (rc == 0) {
 545                 /* For the master MDT-object of a striped directory,
 546                  * reset the iteration from the directory beginning. */
 547                 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
 548                         pos->lp_dir_cookie = 0;
 549
 550                 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
 551                 if (rc > 0)
 552                         /* The end of the directory. */
 553                         rc = 0;
 554         }
 555
 556         GOTO(out, rc);
 557
 558 out:
 559         if (obj != NULL)
 560                 lfsck_object_put(env, obj);
 561
 562         if (rc != 0) {
 563                 lfsck_close_dir(env, lfsck, rc);
 564                 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 565                                          lc_link) {
 566                         com->lc_ops->lfsck_post(env, com, rc, true);
 567                 }
 568
 569                 return rc;
 570         }
 571
 572         rc = 0;
 573         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
 574         lfsck->li_pos_current = lfsck->li_pos_checkpoint;
 575         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 576                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
 577                 if (rc != 0)
 578                         break;
 579         }
 580
 581         lfsck->li_time_last_checkpoint = cfs_time_current();
 582         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 583                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 584         return rc;
 585 }
 586
 587 static int lfsck_exec_oit(const struct lu_env *env,
 588                           struct lfsck_instance *lfsck, struct dt_object *obj)
 589 {
 590         struct lfsck_component *com;
 591         int                     rc;
 592         ENTRY;
 593
 594         LASSERT(lfsck->li_obj_dir == NULL);
 595
 596         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 597                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
 598                 if (rc != 0)
 599                         RETURN(rc);
 600         }
 601
 602         rc = lfsck_needs_scan_dir(env, lfsck, obj);
 603         if (rc <= 0)
 604                 GOTO(out, rc);
 605
 606         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 607         if (rc == 0)
 608                 rc = lfsck_open_dir(env, lfsck, 0);
 609
 610         GOTO(out, rc);
 611
 612 out:
 613         if (rc < 0)
 614                 lfsck_fail(env, lfsck, false);
 615
 616         if (rc != 0)
 617                 lfsck_close_dir(env, lfsck, rc);
 618
 619         return rc > 0 ? 0 : rc;
 620 }
 621
 622 static int lfsck_exec_dir(const struct lu_env *env,
 623                           struct lfsck_instance *lfsck,
 624                           struct lu_dirent *ent, __u16 type)
 625 {
 626         struct lfsck_component *com;
 627         int                     rc;
 628
 629         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 630                 rc = com->lc_ops->lfsck_exec_dir(env, com, ent, type);
 631                 if (rc != 0)
 632                         return rc;
 633         }
 634         return 0;
 635 }
 636
 637 static int lfsck_master_dir_engine(const struct lu_env *env,
 638                                    struct lfsck_instance *lfsck);
 639
 640 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 641                       int result)
 642 {
 643         struct lfsck_component *com;
 644         struct lfsck_component *next;
 645         int                     rc  = result;
 646
 647         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 648         lfsck_close_dir(env, lfsck, result);
 649
 650         while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
 651                !list_empty(&lfsck->li_list_lmv)) {
 652                 struct lfsck_lmv_unit *llu;
 653
 654                 spin_lock(&lfsck->li_lock);
 655                 llu = list_entry(lfsck->li_list_lmv.next,
 656                                  struct lfsck_lmv_unit, llu_link);
 657                 list_del_init(&llu->llu_link);
 658                 spin_unlock(&lfsck->li_lock);
 659
 660                 lfsck->li_lmv = &llu->llu_lmv;
 661                 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 662                 rc = lfsck_open_dir(env, lfsck, 0);
 663                 if (rc == 0) {
 664                         rc = lfsck_master_dir_engine(env, lfsck);
 665                         lfsck_close_dir(env, lfsck, result);
 666                 }
 667         }
 668
 669         result = rc;
 670
 671         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 672                 rc = com->lc_ops->lfsck_post(env, com, result, false);
 673                 if (rc != 0)
 674                         CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
 675                                "rc = %d\n", lfsck_lfsck2name(lfsck),
 676                                (__u32)com->lc_type, rc);
 677         }
 678
 679         lfsck->li_time_last_checkpoint = cfs_time_current();
 680         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 681                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 682
 683         /* Ignore some component post failure to make other can go ahead. */
 684         return result;
 685 }
 686
 687 static int lfsck_double_scan(const struct lu_env *env,
 688                              struct lfsck_instance *lfsck)
 689 {
 690         struct lfsck_component *com;
 691         struct lfsck_component *next;
 692         struct l_wait_info      lwi = { 0 };
 693         int                     rc  = 0;
 694         int                     rc1 = 0;
 695
 696         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
 697                 rc = com->lc_ops->lfsck_double_scan(env, com);
 698                 if (rc != 0)
 699                         rc1 = rc;
 700         }
 701
 702         l_wait_event(lfsck->li_thread.t_ctl_waitq,
 703                      atomic_read(&lfsck->li_double_scan_count) == 0,
 704                      &lwi);
 705
 706         if (lfsck->li_status != LS_PAUSED &&
 707             lfsck->li_status != LS_CO_PAUSED) {
 708                 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 709                                          lc_link) {
 710                         spin_lock(&lfsck->li_lock);
 711                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 712                         spin_unlock(&lfsck->li_lock);
 713                 }
 714         }
 715
 716         return rc1 != 0 ? rc1 : rc;
 717 }
 718
 719 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
 720 {
 721         struct lfsck_component *com;
 722         struct lfsck_component *next;
 723
 724         list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 725                                  lc_link) {
 726                 if (com->lc_ops->lfsck_quit != NULL)
 727                         com->lc_ops->lfsck_quit(env, com);
 728
 729                 spin_lock(&lfsck->li_lock);
 730                 list_del_init(&com->lc_link_dir);
 731                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 732                 spin_unlock(&lfsck->li_lock);
 733         }
 734
 735         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 736                                  lc_link) {
 737                 if (com->lc_ops->lfsck_quit != NULL)
 738                         com->lc_ops->lfsck_quit(env, com);
 739
 740                 spin_lock(&lfsck->li_lock);
 741                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 742                 spin_unlock(&lfsck->li_lock);
 743         }
 744 }
 745
 746 /* LFSCK engines */
 747
 748 static int lfsck_master_dir_engine(const struct lu_env *env,
 749                                    struct lfsck_instance *lfsck)
 750 {
 751         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 752         struct dt_object                *dir    = lfsck->li_obj_dir;
 753         const struct dt_it_ops          *iops   = &dir->do_index_ops->dio_it;
 754         struct dt_it                    *di     = lfsck->li_di_dir;
 755         struct lu_dirent                *ent    =
 756                         (struct lu_dirent *)info->lti_key;
 757         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 758         struct ptlrpc_thread            *thread = &lfsck->li_thread;
 759         int                              rc;
 760         __u16                            type;
 761         ENTRY;
 762
 763         do {
 764                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
 765                     unlikely(!thread_is_running(thread))) {
 766                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 767                                "parent "DFID", cookie "LPX64"\n",
 768                                lfsck_lfsck2name(lfsck),
 769                                PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
 770
 771                         RETURN(0);
 772                 }
 773
 774                 lfsck->li_new_scanned++;
 775                 rc = iops->rec(env, di, (struct dt_rec *)ent,
 776                                lfsck->li_args_dir);
 777                 if (rc == 0)
 778                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
 779                                               &type);
 780
 781                 if (rc != 0) {
 782                         CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
 783                                "parent "DFID", cookie "LPX64": rc = %d\n",
 784                                lfsck_lfsck2name(lfsck),
 785                                PFID(lfsck_dto2fid(dir)),
 786                                lfsck->li_cookie_dir, rc);
 787                         lfsck_fail(env, lfsck, true);
 788                         if (bk->lb_param & LPF_FAILOUT)
 789                                 RETURN(rc);
 790                         else
 791                                 goto checkpoint;
 792                 }
 793
 794                 if (ent->lde_attrs & LUDA_IGNORE &&
 795                     strcmp(ent->lde_name, dotdot) != 0)
 796                         goto checkpoint;
 797
 798                 /* The type in the @ent structure may has been overwritten,
 799                  * so we need to pass the @type parameter independently. */
 800                 rc = lfsck_exec_dir(env, lfsck, ent, type);
 801                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 802                         RETURN(rc);
 803
 804 checkpoint:
 805                 rc = lfsck_checkpoint(env, lfsck);
 806                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 807                         RETURN(rc);
 808
 809                 /* Rate control. */
 810                 lfsck_control_speed(lfsck);
 811                 if (unlikely(!thread_is_running(thread))) {
 812                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 813                                "parent "DFID", cookie "LPX64"\n",
 814                                lfsck_lfsck2name(lfsck),
 815                                PFID(lfsck_dto2fid(dir)),
 816                                lfsck->li_cookie_dir);
 817                         RETURN(0);
 818                 }
 819
 820                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
 821                         spin_lock(&lfsck->li_lock);
 822                         thread_set_flags(thread, SVC_STOPPING);
 823                         spin_unlock(&lfsck->li_lock);
 824                         RETURN(-EINVAL);
 825                 }
 826
 827                 rc = iops->next(env, di);
 828         } while (rc == 0);
 829
 830         if (rc > 0 && !lfsck->li_oit_over)
 831                 lfsck_close_dir(env, lfsck, rc);
 832
 833         RETURN(rc);
 834 }
 835
 836 /**
 837  * Object-table based iteration engine.
 838  *
 839  * Object-table based iteration is the basic linear engine to scan all the
 840  * objects on current device in turn. For each object, it calls all the
 841  * registered LFSCK component(s)' API to perform related consistency
 842  * verification.
 843  *
 844  * It flushes related LFSCK trace files to disk via making checkpoint
 845  * periodically. Then if the server crashed or the LFSCK is paused, the
 846  * LFSCK can resume from the latest checkpoint.
 847  *
 848  * It also controls the whole LFSCK speed via lfsck_control_speed() to
 849  * avoid the server to become overload.
 850  *
 851  * \param[in] env       pointer to the thread context
 852  * \param[in] lfsck     pointer to the lfsck instance
 853  *
 854  * \retval              positive number if all objects have been scanned
 855  * \retval              0 if the iteration is stopped or paused
 856  * \retval              negative error number on failure
 857  */
 858 static int lfsck_master_oit_engine(const struct lu_env *env,
 859                                    struct lfsck_instance *lfsck)
 860 {
 861         struct lfsck_thread_info *info  = lfsck_env_info(env);
 862         const struct dt_it_ops   *iops  =
 863                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 864         struct dt_it             *di    = lfsck->li_di_oit;
 865         struct lu_fid            *fid   = &info->lti_fid;
 866         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 867         struct ptlrpc_thread     *thread = &lfsck->li_thread;
 868         struct dt_device         *dev   = lfsck->li_bottom;
 869         struct seq_server_site   *ss    = lu_site2seq(dev->dd_lu_dev.ld_site);
 870         __u32                    idx    = lfsck_dev_idx(dev);
 871         int                      rc;
 872         ENTRY;
 873
 874         if (unlikely(ss == NULL))
 875                 RETURN(-EIO);
 876
 877         do {
 878                 struct dt_object *target;
 879                 bool              update_lma = false;
 880
 881                 if (lfsck->li_di_dir != NULL) {
 882                         rc = lfsck_master_dir_engine(env, lfsck);
 883                         if (rc <= 0)
 884                                 RETURN(rc);
 885                 }
 886
 887                 if (unlikely(lfsck->li_oit_over))
 888                         RETURN(1);
 889
 890                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
 891                     unlikely(!thread_is_running(thread))) {
 892                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
 893                                "cookie "LPU64"\n",
 894                                lfsck_lfsck2name(lfsck), iops->store(env, di));
 895
 896                         RETURN(0);
 897                 }
 898
 899                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
 900                         RETURN(0);
 901
 902                 lfsck->li_current_oit_processed = 1;
 903
 904                 if (!list_empty(&lfsck->li_list_lmv)) {
 905                         struct lfsck_lmv_unit *llu;
 906
 907                         spin_lock(&lfsck->li_lock);
 908                         llu = list_entry(lfsck->li_list_lmv.next,
 909                                          struct lfsck_lmv_unit, llu_link);
 910                         list_del_init(&llu->llu_link);
 911                         spin_unlock(&lfsck->li_lock);
 912
 913                         lfsck->li_lmv = &llu->llu_lmv;
 914                         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 915                         rc = lfsck_open_dir(env, lfsck, 0);
 916                         if (rc == 0)
 917                                 rc = lfsck_master_dir_engine(env, lfsck);
 918
 919                         if (rc <= 0)
 920                                 RETURN(rc);
 921                 }
 922
 923                 lfsck->li_new_scanned++;
 924                 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
 925                 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
 926                 if (rc != 0) {
 927                         CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
 928                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 929                         lfsck_fail(env, lfsck, true);
 930                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
 931                                 RETURN(rc);
 932                         else
 933                                 goto checkpoint;
 934                 }
 935
 936                 if (fid_is_idif(fid)) {
 937                         __u32 idx1 = fid_idif_ost_idx(fid);
 938
 939                         LASSERT(!lfsck->li_master);
 940
 941                         /* It is an old format device, update the LMA. */
 942                         if (idx != idx1) {
 943                                 struct ost_id *oi = &info->lti_oi;
 944
 945                                 fid_to_ostid(fid, oi);
 946                                 ostid_to_fid(fid, oi, idx);
 947                                 update_lma = true;
 948                         }
 949                 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
 950                            !fid_is_last_id(fid) &&
 951                            !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
 952
 953                         /* If the FID/object is only used locally and invisible
 954                          * to external nodes, then LFSCK will not handle it.
 955                          *
 956                          * dot_lustre sequence has been handled specially. */
 957                         goto checkpoint;
 958                 } else {
 959                         struct lu_seq_range *range = &info->lti_range;
 960
 961                         if (lfsck->li_master)
 962                                 fld_range_set_mdt(range);
 963                         else
 964                                 fld_range_set_ost(range);
 965                         rc = fld_local_lookup(env, ss->ss_server_fld,
 966                                               fid_seq(fid), range);
 967                         if (rc != 0 || range->lsr_index != idx) {
 968                                 /* Remote object will be handled by the LFSCK
 969                                  * instance on the MDT where the remote object
 970                                  * really resides on. */
 971                                 rc = 0;
 972                                 goto checkpoint;
 973                         }
 974                 }
 975
 976                 target = lfsck_object_find(env, lfsck, fid);
 977                 if (IS_ERR(target)) {
 978                         CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
 979                                DFID", cookie "LPU64": rc = %d\n",
 980                                lfsck_lfsck2name(lfsck), PFID(fid),
 981                                iops->store(env, di), rc);
 982                         lfsck_fail(env, lfsck, true);
 983                         if (bk->lb_param & LPF_FAILOUT)
 984                                 RETURN(PTR_ERR(target));
 985                         else
 986                                 goto checkpoint;
 987                 }
 988
 989                 if (dt_object_exists(target)) {
 990                         if (update_lma) {
 991                                 rc = lfsck_update_lma(env, lfsck, target);
 992                                 if (rc != 0)
 993                                         CDEBUG(D_LFSCK, "%s: fail to update "
 994                                                "LMA for "DFID": rc = %d\n",
 995                                                lfsck_lfsck2name(lfsck),
 996                                                PFID(lfsck_dto2fid(target)), rc);
 997                         }
 998                         if (rc == 0)
 999                                 rc = lfsck_exec_oit(env, lfsck, target);
1000                 }
1001                 lfsck_object_put(env, target);
1002                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1003                         RETURN(rc);
1004
1005 checkpoint:
1006                 rc = lfsck_checkpoint(env, lfsck);
1007                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1008                         RETURN(rc);
1009
1010                 /* Rate control. */
1011                 lfsck_control_speed(lfsck);
1012
1013                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1014                         spin_lock(&lfsck->li_lock);
1015                         thread_set_flags(thread, SVC_STOPPING);
1016                         spin_unlock(&lfsck->li_lock);
1017                         RETURN(-EINVAL);
1018                 }
1019
1020                 rc = iops->next(env, di);
1021                 if (unlikely(rc > 0))
1022                         lfsck->li_oit_over = 1;
1023                 else if (likely(rc == 0))
1024                         lfsck->li_current_oit_processed = 0;
1025
1026                 if (unlikely(!thread_is_running(thread))) {
1027                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1028                                "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1029                                iops->store(env, di));
1030                         RETURN(0);
1031                 }
1032         } while (rc == 0 || lfsck->li_di_dir != NULL);
1033
1034         RETURN(rc);
1035 }
1036
1037 int lfsck_master_engine(void *args)
1038 {
1039         struct lfsck_thread_args *lta      = args;
1040         struct lu_env            *env      = &lta->lta_env;
1041         struct lfsck_instance    *lfsck    = lta->lta_lfsck;
1042         struct ptlrpc_thread     *thread   = &lfsck->li_thread;
1043         struct dt_object         *oit_obj  = lfsck->li_obj_oit;
1044         const struct dt_it_ops   *oit_iops = &oit_obj->do_index_ops->dio_it;
1045         struct dt_it             *oit_di;
1046         struct l_wait_info        lwi      = { 0 };
1047         int                       rc;
1048         ENTRY;
1049
1050         if (lfsck->li_master &&
1051             (!list_empty(&lfsck->li_list_scan) ||
1052              !list_empty(&lfsck->li_list_double_scan))) {
1053                 rc = lfsck_verify_lpf(env, lfsck);
1054                 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1055                  * fatal, because the .lustre/lost+found/ maybe not accessed
1056                  * by the LFSCK if it does not add orphans or others to such
1057                  * directory. So go ahead until hit failure when really uses
1058                  * the directory. */
1059                 if (rc != 0)
1060                         CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1061                                ".lustre/lost+found/, go ahead: rc = %d\n",
1062                                lfsck_lfsck2name(lfsck), rc);
1063         }
1064
1065         oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit, BYPASS_CAPA);
1066         if (IS_ERR(oit_di)) {
1067                 rc = PTR_ERR(oit_di);
1068                 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1069                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1070
1071                 GOTO(fini_args, rc);
1072         }
1073
1074         spin_lock(&lfsck->li_lock);
1075         lfsck->li_di_oit = oit_di;
1076         spin_unlock(&lfsck->li_lock);
1077         rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1078         if (rc != 0)
1079                 GOTO(fini_oit, rc);
1080
1081         CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1082                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1083                ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1084                lfsck->li_pos_checkpoint.lp_oit_cookie,
1085                lfsck->li_pos_checkpoint.lp_dir_cookie,
1086                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1087                current_pid());
1088
1089         spin_lock(&lfsck->li_lock);
1090         thread_set_flags(thread, SVC_RUNNING);
1091         spin_unlock(&lfsck->li_lock);
1092         wake_up_all(&thread->t_ctl_waitq);
1093
1094         l_wait_event(thread->t_ctl_waitq,
1095                      lfsck->li_start_unplug ||
1096                      !thread_is_running(thread),
1097                      &lwi);
1098         if (!thread_is_running(thread))
1099                 GOTO(fini_oit, rc = 0);
1100
1101         if (!list_empty(&lfsck->li_list_scan) ||
1102             list_empty(&lfsck->li_list_double_scan))
1103                 rc = lfsck_master_oit_engine(env, lfsck);
1104         else
1105                 rc = 1;
1106
1107         CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1108                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1109                ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1110                lfsck->li_pos_checkpoint.lp_oit_cookie,
1111                lfsck->li_pos_checkpoint.lp_dir_cookie,
1112                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1113                current_pid(), rc);
1114
1115         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1116                 rc = lfsck_post(env, lfsck, rc);
1117         else
1118                 lfsck_close_dir(env, lfsck, rc);
1119
1120 fini_oit:
1121         lfsck_di_oit_put(env, lfsck);
1122         oit_iops->fini(env, oit_di);
1123         if (rc == 1) {
1124                 if (!list_empty(&lfsck->li_list_double_scan))
1125                         rc = lfsck_double_scan(env, lfsck);
1126                 else
1127                         rc = 0;
1128         } else {
1129                 lfsck_quit(env, lfsck);
1130         }
1131
1132         /* XXX: Purge the pinned objects in the future. */
1133
1134 fini_args:
1135         spin_lock(&lfsck->li_lock);
1136         thread_set_flags(thread, SVC_STOPPED);
1137         spin_unlock(&lfsck->li_lock);
1138         wake_up_all(&thread->t_ctl_waitq);
1139         lfsck_thread_args_fini(lta);
1140         return rc;
1141 }
1142
1143 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1144 {
1145         bool empty = false;
1146
1147         spin_lock(&lad->lad_lock);
1148         if (list_empty(&lad->lad_req_list))
1149                 empty = true;
1150         spin_unlock(&lad->lad_lock);
1151
1152         return empty;
1153 }
1154
1155 /**
1156  * Query the LFSCK status from the instatnces on remote servers.
1157  *
1158  * The LFSCK assistant thread queries the LFSCK instances on other
1159  * servers (MDT/OST) about their status, such as whether they have
1160  * finished the phase1/phase2 scanning or not, and so on.
1161  *
1162  * \param[in] env       pointer to the thread context
1163  * \param[in] com       pointer to the lfsck component
1164  *
1165  * \retval              0 for success
1166  * \retval              negative error number on failure
1167  */
1168 static int lfsck_assistant_query_others(const struct lu_env *env,
1169                                         struct lfsck_component *com)
1170 {
1171         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1172         struct lfsck_request              *lr    = &info->lti_lr;
1173         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1174         struct lfsck_instance             *lfsck = com->lc_lfsck;
1175         struct lfsck_assistant_data       *lad   = com->lc_data;
1176         struct ptlrpc_request_set         *set;
1177         struct lfsck_tgt_descs            *ltds;
1178         struct lfsck_tgt_desc             *ltd;
1179         struct list_head                  *phase_head;
1180         int                                rc    = 0;
1181         int                                rc1   = 0;
1182         ENTRY;
1183
1184         set = ptlrpc_prep_set();
1185         if (set == NULL)
1186                 RETURN(-ENOMEM);
1187
1188         lad->lad_touch_gen++;
1189         memset(lr, 0, sizeof(*lr));
1190         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1191         lr->lr_event = LE_QUERY;
1192         lr->lr_active = com->lc_type;
1193         laia->laia_com = com;
1194         laia->laia_lr = lr;
1195         laia->laia_shared = 0;
1196
1197         if (!list_empty(&lad->lad_mdt_phase1_list)) {
1198                 ltds = &lfsck->li_mdt_descs;
1199                 lr->lr_flags = 0;
1200                 phase_head = &lad->lad_mdt_phase1_list;
1201         } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1202                 goto out;
1203         } else {
1204
1205 again:
1206                 ltds = &lfsck->li_ost_descs;
1207                 lr->lr_flags = LEF_TO_OST;
1208                 phase_head = &lad->lad_ost_phase1_list;
1209         }
1210
1211         laia->laia_ltds = ltds;
1212         spin_lock(&ltds->ltd_lock);
1213         while (!list_empty(phase_head)) {
1214                 struct list_head *phase_list;
1215                 __u32            *gen;
1216
1217                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1218                         ltd = list_entry(phase_head->next,
1219                                          struct lfsck_tgt_desc,
1220                                          ltd_layout_phase_list);
1221                         phase_list = &ltd->ltd_layout_phase_list;
1222                         gen = &ltd->ltd_layout_gen;
1223                 } else {
1224                         ltd = list_entry(phase_head->next,
1225                                          struct lfsck_tgt_desc,
1226                                          ltd_namespace_phase_list);
1227                         phase_list = &ltd->ltd_namespace_phase_list;
1228                         gen = &ltd->ltd_namespace_gen;
1229                 }
1230
1231                 if (*gen == lad->lad_touch_gen)
1232                         break;
1233
1234                 *gen = lad->lad_touch_gen;
1235                 list_move_tail(phase_list, phase_head);
1236                 atomic_inc(&ltd->ltd_ref);
1237                 laia->laia_ltd = ltd;
1238                 spin_unlock(&ltds->ltd_lock);
1239                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1240                                          lfsck_async_interpret_common,
1241                                          laia, LFSCK_QUERY);
1242                 if (rc != 0) {
1243                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1244                                "%s %x for %s: rc = %d\n",
1245                                lfsck_lfsck2name(lfsck),
1246                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1247                                ltd->ltd_index, lad->lad_name, rc);
1248                         lfsck_tgt_put(ltd);
1249                         rc1 = rc;
1250                 }
1251                 spin_lock(&ltds->ltd_lock);
1252         }
1253         spin_unlock(&ltds->ltd_lock);
1254
1255         rc = ptlrpc_set_wait(set);
1256         if (rc < 0) {
1257                 ptlrpc_set_destroy(set);
1258                 RETURN(rc);
1259         }
1260
1261         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1262             list_empty(&lad->lad_mdt_phase1_list))
1263                 goto again;
1264
1265 out:
1266         ptlrpc_set_destroy(set);
1267
1268         RETURN(rc1 != 0 ? rc1 : rc);
1269 }
1270
1271 /**
1272  * Notify the LFSCK event to the instances on remote servers.
1273  *
1274  * The LFSCK assistant thread notifies the LFSCK instances on other
1275  * servers (MDT/OST) about some events, such as start new scanning,
1276  * stop the scanning, this LFSCK instance will exit, and so on.
1277  *
1278  * \param[in] env       pointer to the thread context
1279  * \param[in] com       pointer to the lfsck component
1280  * \param[in] lr        pointer to the LFSCK event request
1281  *
1282  * \retval              0 for success
1283  * \retval              negative error number on failure
1284  */
1285 static int lfsck_assistant_notify_others(const struct lu_env *env,
1286                                          struct lfsck_component *com,
1287                                          struct lfsck_request *lr)
1288 {
1289         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1290         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1291         struct lfsck_instance             *lfsck = com->lc_lfsck;
1292         struct lfsck_assistant_data       *lad   = com->lc_data;
1293         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1294         struct ptlrpc_request_set         *set;
1295         struct lfsck_tgt_descs            *ltds;
1296         struct lfsck_tgt_desc             *ltd;
1297         struct lfsck_tgt_desc             *next;
1298         __u32                              idx;
1299         int                                rc    = 0;
1300         int                                rc1   = 0;
1301         ENTRY;
1302
1303         set = ptlrpc_prep_set();
1304         if (set == NULL)
1305                 RETURN(-ENOMEM);
1306
1307         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1308         lr->lr_active = com->lc_type;
1309         laia->laia_com = com;
1310         laia->laia_lr = lr;
1311         laia->laia_shared = 0;
1312
1313         switch (lr->lr_event) {
1314         case LE_START:
1315                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1316                         goto next;
1317
1318                 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1319                 lr->lr_speed = bk->lb_speed_limit;
1320                 lr->lr_version = bk->lb_version;
1321                 lr->lr_param |= bk->lb_param;
1322                 lr->lr_async_windows = bk->lb_async_windows;
1323                 lr->lr_flags = LEF_TO_OST;
1324
1325                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1326                 ltds = &lfsck->li_ost_descs;
1327                 laia->laia_ltds = ltds;
1328                 down_read(&ltds->ltd_rw_sem);
1329                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1330                         ltd = lfsck_tgt_get(ltds, idx);
1331                         LASSERT(ltd != NULL);
1332
1333                         laia->laia_ltd = ltd;
1334                         ltd->ltd_layout_done = 0;
1335                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1336                                         lfsck_async_interpret_common,
1337                                         laia, LFSCK_NOTIFY);
1338                         if (rc != 0) {
1339                                 lfsck_lad_set_bitmap(env, com, idx);
1340                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1341                                        "notify OST %x for %s start: rc = %d\n",
1342                                        lfsck_lfsck2name(lfsck), idx,
1343                                        lad->lad_name, rc);
1344                                 lfsck_tgt_put(ltd);
1345                         }
1346                 }
1347                 up_read(&ltds->ltd_rw_sem);
1348
1349                 /* Sync up */
1350                 rc = ptlrpc_set_wait(set);
1351                 if (rc < 0) {
1352                         ptlrpc_set_destroy(set);
1353                         RETURN(rc);
1354                 }
1355
1356 next:
1357                 if (!(bk->lb_param & LPF_ALL_TGT))
1358                         break;
1359
1360                 /* link other MDT targets locallly. */
1361                 ltds = &lfsck->li_mdt_descs;
1362                 spin_lock(&ltds->ltd_lock);
1363                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1364                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1365                                 ltd = LTD_TGT(ltds, idx);
1366                                 LASSERT(ltd != NULL);
1367
1368                                 if (!list_empty(&ltd->ltd_layout_list))
1369                                         continue;
1370
1371                                 list_add_tail(&ltd->ltd_layout_list,
1372                                               &lad->lad_mdt_list);
1373                                 list_add_tail(&ltd->ltd_layout_phase_list,
1374                                               &lad->lad_mdt_phase1_list);
1375                         }
1376                 } else {
1377                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1378                                 ltd = LTD_TGT(ltds, idx);
1379                                 LASSERT(ltd != NULL);
1380
1381                                 if (!list_empty(&ltd->ltd_namespace_list))
1382                                         continue;
1383
1384                                 list_add_tail(&ltd->ltd_namespace_list,
1385                                               &lad->lad_mdt_list);
1386                                 list_add_tail(&ltd->ltd_namespace_phase_list,
1387                                               &lad->lad_mdt_phase1_list);
1388                         }
1389                 }
1390                 spin_unlock(&ltds->ltd_lock);
1391                 break;
1392         case LE_STOP:
1393         case LE_PHASE2_DONE:
1394         case LE_PEER_EXIT: {
1395                 struct list_head *phase_head;
1396
1397                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1398                 if (bk->lb_param & LPF_ALL_TGT) {
1399                         phase_head = &lad->lad_mdt_list;
1400                         ltds = &lfsck->li_mdt_descs;
1401                         if (lr->lr_event == LE_STOP) {
1402                                 /* unlink other MDT targets locallly. */
1403                                 spin_lock(&ltds->ltd_lock);
1404                                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1405                                         list_for_each_entry_safe(ltd, next,
1406                                                 phase_head, ltd_layout_list) {
1407                                                 list_del_init(
1408                                                 &ltd->ltd_layout_phase_list);
1409                                                 list_del_init(
1410                                                 &ltd->ltd_layout_list);
1411                                         }
1412                                 } else {
1413                                         list_for_each_entry_safe(ltd, next,
1414                                                         phase_head,
1415                                                         ltd_namespace_list) {
1416                                                 list_del_init(
1417                                                 &ltd->ltd_namespace_phase_list);
1418                                                 list_del_init(
1419                                                 &ltd->ltd_namespace_list);
1420                                         }
1421                                 }
1422                                 spin_unlock(&ltds->ltd_lock);
1423
1424                                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1425                                         break;
1426
1427                                 lr->lr_flags |= LEF_TO_OST;
1428                                 phase_head = &lad->lad_ost_list;
1429                                 ltds = &lfsck->li_ost_descs;
1430                         } else {
1431                                 lr->lr_flags &= ~LEF_TO_OST;
1432                         }
1433                 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1434                         break;
1435                 } else {
1436                         lr->lr_flags |= LEF_TO_OST;
1437                         phase_head = &lad->lad_ost_list;
1438                         ltds = &lfsck->li_ost_descs;
1439                 }
1440
1441 again:
1442                 laia->laia_ltds = ltds;
1443                 spin_lock(&ltds->ltd_lock);
1444                 while (!list_empty(phase_head)) {
1445                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1446                                 ltd = list_entry(phase_head->next,
1447                                                  struct lfsck_tgt_desc,
1448                                                  ltd_layout_list);
1449                                 if (!list_empty(&ltd->ltd_layout_phase_list))
1450                                         list_del_init(
1451                                                 &ltd->ltd_layout_phase_list);
1452                                 list_del_init(&ltd->ltd_layout_list);
1453                         } else {
1454                                 ltd = list_entry(phase_head->next,
1455                                                  struct lfsck_tgt_desc,
1456                                                  ltd_namespace_list);
1457                                 if (!list_empty(&ltd->ltd_namespace_phase_list))
1458                                         list_del_init(
1459                                                 &ltd->ltd_namespace_phase_list);
1460                                 list_del_init(&ltd->ltd_namespace_list);
1461                         }
1462                         atomic_inc(&ltd->ltd_ref);
1463                         laia->laia_ltd = ltd;
1464                         spin_unlock(&ltds->ltd_lock);
1465                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1466                                         lfsck_async_interpret_common,
1467                                         laia, LFSCK_NOTIFY);
1468                         if (rc != 0) {
1469                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1470                                        "notify %s %x for %s stop/phase2_done/"
1471                                        "peer_exit: rc = %d\n",
1472                                        lfsck_lfsck2name(lfsck),
1473                                        (lr->lr_flags & LEF_TO_OST) ?
1474                                        "OST" : "MDT", ltd->ltd_index,
1475                                        lad->lad_name, rc);
1476                                 lfsck_tgt_put(ltd);
1477                         }
1478                         spin_lock(&ltds->ltd_lock);
1479                 }
1480                 spin_unlock(&ltds->ltd_lock);
1481
1482                 rc = ptlrpc_set_wait(set);
1483                 if (rc < 0) {
1484                         ptlrpc_set_destroy(set);
1485                         RETURN(rc);
1486                 }
1487
1488                 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1489                     !(lr->lr_flags & LEF_TO_OST)) {
1490                         lr->lr_flags |= LEF_TO_OST;
1491                         phase_head = &lad->lad_ost_list;
1492                         ltds = &lfsck->li_ost_descs;
1493                         goto again;
1494                 }
1495                 break;
1496         }
1497         case LE_PHASE1_DONE:
1498                 lad->lad_ops->la_sync_failures(env, com, lr);
1499                 lad->lad_touch_gen++;
1500                 ltds = &lfsck->li_mdt_descs;
1501                 laia->laia_ltds = ltds;
1502                 spin_lock(&ltds->ltd_lock);
1503                 while (!list_empty(&lad->lad_mdt_list)) {
1504                         struct list_head *list;
1505                         __u32            *gen;
1506
1507                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1508                                 ltd = list_entry(lad->lad_mdt_list.next,
1509                                                  struct lfsck_tgt_desc,
1510                                                  ltd_layout_list);
1511                                 list = &ltd->ltd_layout_list;
1512                                 gen = &ltd->ltd_layout_gen;
1513                         } else {
1514                                 struct lfsck_namespace *ns = com->lc_file_ram;
1515
1516                                 ltd = list_entry(lad->lad_mdt_list.next,
1517                                                  struct lfsck_tgt_desc,
1518                                                  ltd_namespace_list);
1519                                 list = &ltd->ltd_namespace_list;
1520                                 gen = &ltd->ltd_namespace_gen;
1521                                 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1522                         }
1523
1524                         if (*gen == lad->lad_touch_gen)
1525                                 break;
1526
1527                         *gen = lad->lad_touch_gen;
1528                         list_move_tail(list, &lad->lad_mdt_list);
1529                         if (ltd->ltd_namespace_failed)
1530                                 continue;
1531
1532                         atomic_inc(&ltd->ltd_ref);
1533                         laia->laia_ltd = ltd;
1534                         spin_unlock(&ltds->ltd_lock);
1535                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1536                                         lfsck_async_interpret_common,
1537                                         laia, LFSCK_NOTIFY);
1538                         if (rc != 0) {
1539                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1540                                        "notify MDT %x for %s phase1 done: "
1541                                        "rc = %d\n", lfsck_lfsck2name(lfsck),
1542                                        ltd->ltd_index, lad->lad_name, rc);
1543                                 lfsck_tgt_put(ltd);
1544                         }
1545                         spin_lock(&ltds->ltd_lock);
1546                 }
1547                 spin_unlock(&ltds->ltd_lock);
1548                 break;
1549         default:
1550                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1551                        "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1552                 rc = -EINVAL;
1553                 break;
1554         }
1555
1556         rc1 = ptlrpc_set_wait(set);
1557         ptlrpc_set_destroy(set);
1558
1559         RETURN(rc != 0 ? rc : rc1);
1560 }
1561
1562 /**
1563  * The LFSCK assistant thread is triggered by the LFSCK main engine.
1564  * They co-work together as an asynchronous pipeline: the LFSCK main
1565  * engine scans the system and pre-fetches the objects, attributes,
1566  * or name entries, etc, and pushes them into the pipeline as input
1567  * requests for the LFSCK assistant thread; on the other end of the
1568  * pipeline, the LFSCK assistant thread performs the real check and
1569  * repair for every request from the main engine.
1570  *
1571  * Generally, the assistant engine may be blocked when check/repair
1572  * something, so the LFSCK main engine will run some faster. On the
1573  * other hand, the LFSCK main engine will drive multiple assistant
1574  * threads in parallel, means for each LFSCK component on the master
1575  * (such as layout LFSCK, namespace LFSCK), there is an independent
1576  * LFSCK assistant thread. So under such 1:N multiple asynchronous
1577  * pipelines mode, the whole LFSCK performance will be much better
1578  * than check/repair everything by the LFSCK main engine itself.
1579  */
1580 int lfsck_assistant_engine(void *args)
1581 {
1582         struct lfsck_thread_args          *lta     = args;
1583         struct lu_env                     *env     = &lta->lta_env;
1584         struct lfsck_component            *com     = lta->lta_com;
1585         struct lfsck_instance             *lfsck   = lta->lta_lfsck;
1586         struct lfsck_bookmark             *bk      = &lfsck->li_bookmark_ram;
1587         struct lfsck_position             *pos     = &com->lc_pos_start;
1588         struct lfsck_thread_info          *info    = lfsck_env_info(env);
1589         struct lfsck_request              *lr      = &info->lti_lr;
1590         struct lfsck_assistant_data       *lad     = com->lc_data;
1591         struct ptlrpc_thread              *mthread = &lfsck->li_thread;
1592         struct ptlrpc_thread              *athread = &lad->lad_thread;
1593         struct lfsck_assistant_operations *lao     = lad->lad_ops;
1594         struct lfsck_assistant_req        *lar;
1595         struct l_wait_info                 lwi     = { 0 };
1596         int                                rc      = 0;
1597         int                                rc1     = 0;
1598         int                                rc2;
1599         ENTRY;
1600
1601         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1602                lfsck_lfsck2name(lfsck), lad->lad_name);
1603
1604         memset(lr, 0, sizeof(*lr));
1605         lr->lr_event = LE_START;
1606         if (pos->lp_oit_cookie <= 1)
1607                 lr->lr_param = LPF_RESET;
1608         rc = lfsck_assistant_notify_others(env, com, lr);
1609         if (rc != 0) {
1610                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1611                        "to start %s: rc = %d\n",
1612                        lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1613                 GOTO(fini, rc);
1614         }
1615
1616         spin_lock(&lad->lad_lock);
1617         thread_set_flags(athread, SVC_RUNNING);
1618         spin_unlock(&lad->lad_lock);
1619         wake_up_all(&mthread->t_ctl_waitq);
1620
1621         while (1) {
1622                 while (!list_empty(&lad->lad_req_list)) {
1623                         bool wakeup = false;
1624
1625                         if (unlikely(lad->lad_exit ||
1626                                      !thread_is_running(mthread)))
1627                                 GOTO(cleanup1, rc = lad->lad_post_result);
1628
1629                         lar = list_entry(lad->lad_req_list.next,
1630                                          struct lfsck_assistant_req,
1631                                          lar_list);
1632                         /* Only the lfsck_assistant_engine thread itself can
1633                          * remove the "lar" from the head of the list, LFSCK
1634                          * engine thread only inserts other new "lar" at the
1635                          * end of the list. So it is safe to handle current
1636                          * "lar" without the spin_lock. */
1637                         rc = lao->la_handler_p1(env, com, lar);
1638                         spin_lock(&lad->lad_lock);
1639                         list_del_init(&lar->lar_list);
1640                         lad->lad_prefetched--;
1641                         /* Wake up the main engine thread only when the list
1642                          * is empty or half of the prefetched items have been
1643                          * handled to avoid too frequent thread schedule. */
1644                         if (lad->lad_prefetched == 0 ||
1645                             (bk->lb_async_windows != 0 &&
1646                              bk->lb_async_windows / 2 ==
1647                              lad->lad_prefetched))
1648                                 wakeup = true;
1649                         spin_unlock(&lad->lad_lock);
1650                         if (wakeup)
1651                                 wake_up_all(&mthread->t_ctl_waitq);
1652
1653                         lao->la_req_fini(env, lar);
1654                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1655                                 GOTO(cleanup1, rc);
1656                 }
1657
1658                 l_wait_event(athread->t_ctl_waitq,
1659                              !lfsck_assistant_req_empty(lad) ||
1660                              lad->lad_exit ||
1661                              lad->lad_to_post ||
1662                              lad->lad_to_double_scan,
1663                              &lwi);
1664
1665                 if (unlikely(lad->lad_exit))
1666                         GOTO(cleanup1, rc = lad->lad_post_result);
1667
1668                 if (!list_empty(&lad->lad_req_list))
1669                         continue;
1670
1671                 if (lad->lad_to_post) {
1672                         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1673                                lfsck_lfsck2name(lfsck), lad->lad_name);
1674
1675                         if (unlikely(lad->lad_exit))
1676                                 GOTO(cleanup1, rc = lad->lad_post_result);
1677
1678                         lad->lad_to_post = 0;
1679                         LASSERT(lad->lad_post_result > 0);
1680
1681                         memset(lr, 0, sizeof(*lr));
1682                         lr->lr_event = LE_PHASE1_DONE;
1683                         lr->lr_status = lad->lad_post_result;
1684                         rc = lfsck_assistant_notify_others(env, com, lr);
1685                         if (rc != 0)
1686                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to "
1687                                        "notify others for %s post: rc = %d\n",
1688                                        lfsck_lfsck2name(lfsck),
1689                                        lad->lad_name, rc);
1690
1691                         /* Wakeup the master engine to go ahead. */
1692                         wake_up_all(&mthread->t_ctl_waitq);
1693                 }
1694
1695                 if (lad->lad_to_double_scan) {
1696                         lad->lad_to_double_scan = 0;
1697                         atomic_inc(&lfsck->li_double_scan_count);
1698                         lad->lad_in_double_scan = 1;
1699                         wake_up_all(&mthread->t_ctl_waitq);
1700
1701                         com->lc_new_checked = 0;
1702                         com->lc_new_scanned = 0;
1703                         com->lc_time_last_checkpoint = cfs_time_current();
1704                         com->lc_time_next_checkpoint =
1705                                 com->lc_time_last_checkpoint +
1706                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1707
1708                         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1709                                "the second-stage scaning\n",
1710                                lfsck_lfsck2name(lfsck));
1711
1712                         /* Flush async updates before handling orphan. */
1713                         rc2 = dt_sync(env, lfsck->li_next);
1714
1715                         CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1716                                "scan start, synced: rc = %d\n",
1717                                lfsck_lfsck2name(lfsck), rc2);
1718
1719                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1720                                 GOTO(cleanup2, rc = 0);
1721
1722                         while (lad->lad_in_double_scan) {
1723                                 rc = lfsck_assistant_query_others(env, com);
1724                                 if (lfsck_phase2_next_ready(lad))
1725                                         goto p2_next;
1726
1727                                 if (rc < 0)
1728                                         GOTO(cleanup2, rc);
1729
1730                                 /* Pull LFSCK status on related targets once
1731                                  * per 30 seconds if we are not notified. */
1732                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1733                                                            cfs_time_seconds(1),
1734                                                            NULL, NULL);
1735                                 rc = l_wait_event(athread->t_ctl_waitq,
1736                                         lfsck_phase2_next_ready(lad) ||
1737                                         lad->lad_exit ||
1738                                         !thread_is_running(mthread),
1739                                         &lwi);
1740
1741                                 if (unlikely(lad->lad_exit ||
1742                                              !thread_is_running(mthread)))
1743                                         GOTO(cleanup2, rc = 0);
1744
1745                                 if (rc == -ETIMEDOUT)
1746                                         continue;
1747
1748                                 if (rc < 0)
1749                                         GOTO(cleanup2, rc);
1750
1751 p2_next:
1752                                 rc = lao->la_handler_p2(env, com);
1753                                 if (rc != 0)
1754                                         GOTO(cleanup2, rc);
1755
1756                                 if (unlikely(lad->lad_exit ||
1757                                              !thread_is_running(mthread)))
1758                                         GOTO(cleanup2, rc = 0);
1759                         }
1760                 }
1761         }
1762
1763 cleanup1:
1764         /* Cleanup the unfinished requests. */
1765         spin_lock(&lad->lad_lock);
1766         if (rc < 0)
1767                 lad->lad_assistant_status = rc;
1768
1769         if (lad->lad_exit && lad->lad_post_result <= 0)
1770                 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1771
1772         while (!list_empty(&lad->lad_req_list)) {
1773                 lar = list_entry(lad->lad_req_list.next,
1774                                  struct lfsck_assistant_req,
1775                                  lar_list);
1776                 list_del_init(&lar->lar_list);
1777                 lad->lad_prefetched--;
1778                 spin_unlock(&lad->lad_lock);
1779                 lao->la_req_fini(env, lar);
1780                 spin_lock(&lad->lad_lock);
1781         }
1782         spin_unlock(&lad->lad_lock);
1783
1784         LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1785                  lad->lad_prefetched);
1786
1787 cleanup2:
1788         memset(lr, 0, sizeof(*lr));
1789         if (rc > 0) {
1790                 lr->lr_event = LE_PHASE2_DONE;
1791                 lr->lr_status = rc;
1792         } else if (rc == 0) {
1793                 if (lfsck->li_flags & LPF_ALL_TGT) {
1794                         lr->lr_event = LE_STOP;
1795                         lr->lr_status = LS_STOPPED;
1796                 } else {
1797                         lr->lr_event = LE_PEER_EXIT;
1798                         switch (lfsck->li_status) {
1799                         case LS_PAUSED:
1800                         case LS_CO_PAUSED:
1801                                 lr->lr_status = LS_CO_PAUSED;
1802                                 break;
1803                         case LS_STOPPED:
1804                         case LS_CO_STOPPED:
1805                                 lr->lr_status = LS_CO_STOPPED;
1806                                 break;
1807                         default:
1808                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1809                                        "status: rc = %d\n",
1810                                        lfsck_lfsck2name(lfsck),
1811                                        lfsck->li_status);
1812                                 lr->lr_status = LS_CO_FAILED;
1813                                 break;
1814                         }
1815                 }
1816         } else {
1817                 if (lfsck->li_flags & LPF_ALL_TGT) {
1818                         lr->lr_event = LE_STOP;
1819                         lr->lr_status = LS_FAILED;
1820                 } else {
1821                         lr->lr_event = LE_PEER_EXIT;
1822                         lr->lr_status = LS_CO_FAILED;
1823                 }
1824         }
1825
1826         rc1 = lfsck_assistant_notify_others(env, com, lr);
1827         if (rc1 != 0) {
1828                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1829                        "others for %s quit: rc = %d\n",
1830                        lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1831                 rc = rc1;
1832         }
1833
1834         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1835                lfsck_lfsck2name(lfsck));
1836
1837         /* Flush async updates before exit. */
1838         rc2 = dt_sync(env, lfsck->li_next);
1839
1840         CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1841                lfsck_lfsck2name(lfsck), rc2);
1842
1843         /* Under force exit case, some requests may be just freed without
1844          * verification, those objects should be re-handled when next run.
1845          * So not update the on-disk trace file under such case. */
1846         if (lad->lad_in_double_scan) {
1847                 if (!lad->lad_exit)
1848                         rc1 = lao->la_double_scan_result(env, com, rc);
1849
1850                 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1851                        "finished: rc = %d\n",
1852                        lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1853         }
1854
1855 fini:
1856         if (lad->lad_in_double_scan)
1857                 atomic_dec(&lfsck->li_double_scan_count);
1858
1859         spin_lock(&lad->lad_lock);
1860         lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1861         thread_set_flags(athread, SVC_STOPPED);
1862         wake_up_all(&mthread->t_ctl_waitq);
1863         spin_unlock(&lad->lad_lock);
1864
1865         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1866                lfsck_lfsck2name(lfsck), lad->lad_name,
1867                lad->lad_assistant_status);
1868
1869         lfsck_thread_args_fini(lta);
1870
1871         return rc;
1872 }