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                *dev    = lfsck_obj2dev(obj);
 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, dev);
 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_local(env, dev, 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, dev, 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 seq_server_site   *ss      = lfsck_dev_site(lfsck);
 198         __u32                     idx     = lfsck_dev_idx(lfsck);
 199         int                       depth   = 0;
 200         int                       rc      = 0;
 201
 202         if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
 203                 return 0;
 204
 205         LASSERT(ss != NULL);
 206
 207         *fid = *lfsck_dto2fid(obj);
 208         while (1) {
 209                 /* Global /ROOT is visible. */
 210                 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
 211                         return 1;
 212
 213                 /* Backend root is invisible. */
 214                 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
 215                         return 0;
 216
 217                 if (!fid_is_norm(fid) && !fid_is_igif(fid))
 218                         return 0;
 219
 220                 fld_range_set_mdt(range);
 221                 rc = fld_local_lookup(env, ss->ss_server_fld,
 222                                       fid_seq(fid), range);
 223                 if (rc != 0 || range->lsr_index != idx)
 224                         /* Current FID should NOT be for the input parameter
 225                          * @obj, because the lfsck_master_oit_engine() has
 226                          * filtered out agent object. So current FID is for
 227                          * the ancestor of the original input parameter @obj.
 228                          * So the ancestor is a remote directory. The input
 229                          * parameter @obj is local directory, and should be
 230                          * scanned under such case. */
 231                         return 1;
 232
 233                 /* normal FID on this target (locally) must be for the
 234                  * client-side visiable object. */
 235                 if (fid_is_norm(fid))
 236                         return 1;
 237
 238                 if (obj == NULL) {
 239                         obj = lfsck_object_find_bottom(env, lfsck, fid);
 240                         if (IS_ERR(obj))
 241                                 return PTR_ERR(obj);
 242
 243                         depth++;
 244                         if (!dt_object_exists(obj))
 245                                 GOTO(out, rc = 0);
 246                 }
 247
 248                 dt_read_lock(env, obj, MOR_TGT_CHILD);
 249                 if (unlikely(lfsck_is_dead_obj(obj))) {
 250                         dt_read_unlock(env, obj);
 251
 252                         GOTO(out, rc = 0);
 253                 }
 254
 255                 rc = dt_xattr_get(env, obj,
 256                                   lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK,
 257                                   BYPASS_CAPA);
 258                 dt_read_unlock(env, obj);
 259                 if (rc >= 0)
 260                         GOTO(out, rc = 1);
 261
 262                 if (rc < 0 && rc != -ENODATA)
 263                         GOTO(out, rc);
 264
 265                 rc = lfsck_parent_fid(env, obj, fid);
 266                 if (depth > 0)
 267                         lfsck_object_put(env, obj);
 268
 269                 obj = NULL;
 270                 if (rc != 0)
 271                         return rc;
 272
 273                 if (!fid_is_sane(fid))
 274                         return 0;
 275         }
 276
 277 out:
 278         if (depth > 0 && obj != NULL)
 279                 lfsck_object_put(env, obj);
 280
 281         return rc;
 282 }
 283
 284 static int lfsck_load_stripe_lmv(const struct lu_env *env,
 285                                  struct lfsck_instance *lfsck,
 286                                  struct dt_object *obj)
 287 {
 288         struct lmv_mds_md_v1    *lmv    = &lfsck_env_info(env)->lti_lmv;
 289         struct lfsck_lmv        *llmv;
 290         int                      rc;
 291         ENTRY;
 292
 293         LASSERT(lfsck->li_obj_dir == NULL);
 294         LASSERT(lfsck->li_lmv == NULL);
 295
 296         rc = lfsck_read_stripe_lmv(env, obj, lmv);
 297         if (rc == -ENODATA) {
 298                 lfsck->li_obj_dir = lfsck_object_get(obj);
 299
 300                 RETURN(0);
 301         }
 302
 303         if (rc < 0)
 304                 RETURN(rc);
 305
 306         OBD_ALLOC_PTR(llmv);
 307         if (llmv == NULL)
 308                 RETURN(-ENOMEM);
 309
 310         if (lmv->lmv_magic == LMV_MAGIC) {
 311                 struct lfsck_slave_lmv_rec      *lslr;
 312                 __u32                            stripes;
 313
 314                 llmv->ll_lmv_master = 1;
 315                 if (lmv->lmv_stripe_count < 1)
 316                         stripes = LFSCK_LMV_DEF_STRIPES;
 317                 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
 318                         stripes = LFSCK_LMV_MAX_STRIPES;
 319                 else
 320                         stripes = lmv->lmv_stripe_count;
 321
 322                 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
 323                 if (lslr == NULL) {
 324                         OBD_FREE_PTR(llmv);
 325
 326                         RETURN(-ENOMEM);
 327                 }
 328
 329                 llmv->ll_stripes_allocated = stripes;
 330                 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
 331                 llmv->ll_lslr = lslr;
 332         } else {
 333                 llmv->ll_lmv_slave = 1;
 334         }
 335
 336         lfsck->li_obj_dir = lfsck_object_get(obj);
 337         llmv->ll_lmv = *lmv;
 338         atomic_set(&llmv->ll_ref, 1);
 339         lfsck->li_lmv = llmv;
 340
 341         RETURN(0);
 342 }
 343
 344 /* LFSCK wrap functions */
 345
 346 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
 347                        bool new_checked)
 348 {
 349         struct lfsck_component *com;
 350
 351         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 352                 com->lc_ops->lfsck_fail(env, com, new_checked);
 353         }
 354 }
 355
 356 void lfsck_close_dir(const struct lu_env *env,
 357                      struct lfsck_instance *lfsck, int result)
 358 {
 359         struct lfsck_component *com;
 360         ENTRY;
 361
 362         if (lfsck->li_lmv != NULL) {
 363                 lfsck->li_lmv->ll_exit_value = result;
 364                 if (lfsck->li_obj_dir != NULL) {
 365                         list_for_each_entry(com, &lfsck->li_list_dir,
 366                                             lc_link_dir) {
 367                                 com->lc_ops->lfsck_close_dir(env, com);
 368                         }
 369                 }
 370
 371                 lfsck_lmv_put(env, lfsck->li_lmv);
 372                 lfsck->li_lmv = NULL;
 373         }
 374
 375         if (lfsck->li_di_dir != NULL) {
 376                 const struct dt_it_ops  *dir_iops;
 377                 struct dt_it            *dir_di   = lfsck->li_di_dir;
 378
 379                 LASSERT(lfsck->li_obj_dir != NULL);
 380
 381                 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
 382                 lfsck_di_dir_put(env, lfsck);
 383                 dir_iops->fini(env, dir_di);
 384         }
 385
 386         if (lfsck->li_obj_dir != NULL) {
 387                 struct dt_object        *dir_obj  = lfsck->li_obj_dir;
 388
 389                 lfsck->li_obj_dir = NULL;
 390                 lfsck_object_put(env, dir_obj);
 391         }
 392
 393         EXIT;
 394 }
 395
 396 int lfsck_open_dir(const struct lu_env *env,
 397                    struct lfsck_instance *lfsck, __u64 cookie)
 398 {
 399         struct dt_object        *obj    = lfsck->li_obj_dir;
 400         struct dt_it            *di     = lfsck->li_di_dir;
 401         struct lfsck_component  *com;
 402         const struct dt_it_ops  *iops;
 403         int                      rc     = 0;
 404         ENTRY;
 405
 406         LASSERT(obj != NULL);
 407         LASSERT(di == NULL);
 408
 409         if (unlikely(!dt_try_as_dir(env, obj)))
 410                 GOTO(out, rc = -ENOTDIR);
 411
 412         list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
 413                 rc = com->lc_ops->lfsck_open_dir(env, com);
 414                 if (rc != 0)
 415                         GOTO(out, rc);
 416         }
 417
 418         iops = &obj->do_index_ops->dio_it;
 419         di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
 420         if (IS_ERR(di))
 421                 GOTO(out, rc = PTR_ERR(di));
 422
 423         rc = iops->load(env, di, cookie);
 424         if (rc == 0 || (rc > 0 && cookie > 0))
 425                 rc = iops->next(env, di);
 426         else if (rc > 0)
 427                 rc = 0;
 428
 429         if (rc != 0) {
 430                 iops->put(env, di);
 431                 iops->fini(env, di);
 432         } else {
 433                 lfsck->li_cookie_dir = iops->store(env, di);
 434                 spin_lock(&lfsck->li_lock);
 435                 lfsck->li_di_dir = di;
 436                 spin_unlock(&lfsck->li_lock);
 437         }
 438
 439         GOTO(out, rc);
 440
 441 out:
 442         if (rc != 0)
 443                 lfsck_close_dir(env, lfsck, rc);
 444
 445         return rc;
 446 }
 447
 448 static int lfsck_checkpoint(const struct lu_env *env,
 449                             struct lfsck_instance *lfsck)
 450 {
 451         struct lfsck_component *com;
 452         int                     rc  = 0;
 453         int                     rc1 = 0;
 454
 455         if (likely(cfs_time_beforeq(cfs_time_current(),
 456                                     lfsck->li_time_next_checkpoint)))
 457                 return 0;
 458
 459         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 460         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 461                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
 462                 if (rc != 0)
 463                         rc1 = rc;
 464         }
 465
 466         lfsck->li_time_last_checkpoint = cfs_time_current();
 467         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 468                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 469         return rc1 != 0 ? rc1 : rc;
 470 }
 471
 472 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 473                       struct lfsck_start_param *lsp)
 474 {
 475         struct dt_object       *obj     = NULL;
 476         struct lfsck_component *com;
 477         struct lfsck_component *next;
 478         struct lfsck_position  *pos     = NULL;
 479         const struct dt_it_ops *iops    =
 480                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 481         int                     rc;
 482         ENTRY;
 483
 484         LASSERT(lfsck->li_obj_dir == NULL);
 485         LASSERT(lfsck->li_di_dir == NULL);
 486
 487         lfsck->li_current_oit_processed = 0;
 488         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 489                 com->lc_new_checked = 0;
 490                 rc = com->lc_ops->lfsck_prep(env, com, lsp);
 491                 if (rc != 0)
 492                         GOTO(out, rc);
 493
 494                 if ((pos == NULL) ||
 495                     (!lfsck_pos_is_zero(&com->lc_pos_start) &&
 496                      lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
 497                         pos = &com->lc_pos_start;
 498         }
 499
 500         /* Init otable-based iterator. */
 501         if (pos == NULL) {
 502                 rc = iops->load(env, lfsck->li_di_oit, 0);
 503                 if (rc > 0) {
 504                         lfsck->li_oit_over = 1;
 505                         rc = 0;
 506                 }
 507
 508                 GOTO(out, rc);
 509         }
 510
 511         rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
 512         if (rc < 0)
 513                 GOTO(out, rc);
 514         else if (rc > 0)
 515                 lfsck->li_oit_over = 1;
 516
 517         if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
 518                 GOTO(out, rc = 0);
 519
 520         /* Find the directory for namespace-based traverse. */
 521         obj = lfsck_object_find_bottom(env, lfsck, &pos->lp_dir_parent);
 522         if (IS_ERR(obj))
 523                 RETURN(PTR_ERR(obj));
 524
 525         /* Remote directory will be scanned by the LFSCK instance
 526          * on the MDT where the remote object really resides on. */
 527         if (!dt_object_exists(obj) || dt_object_remote(obj) ||
 528             unlikely(!S_ISDIR(lfsck_object_type(obj))))
 529                 GOTO(out, rc = 0);
 530
 531         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 532         if (rc == 0) {
 533                 /* For the master MDT-object of a striped directory,
 534                  * reset the iteration from the directory beginning. */
 535                 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
 536                         pos->lp_dir_cookie = 0;
 537
 538                 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
 539                 if (rc > 0)
 540                         /* The end of the directory. */
 541                         rc = 0;
 542         }
 543
 544         GOTO(out, rc);
 545
 546 out:
 547         if (obj != NULL)
 548                 lfsck_object_put(env, obj);
 549
 550         if (rc != 0) {
 551                 lfsck_close_dir(env, lfsck, rc);
 552                 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 553                                          lc_link) {
 554                         com->lc_ops->lfsck_post(env, com, rc, true);
 555                 }
 556
 557                 return rc;
 558         }
 559
 560         rc = 0;
 561         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
 562         lfsck->li_pos_current = lfsck->li_pos_checkpoint;
 563         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 564                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
 565                 if (rc != 0)
 566                         break;
 567         }
 568
 569         lfsck->li_time_last_checkpoint = cfs_time_current();
 570         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 571                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 572         return rc;
 573 }
 574
 575 static int lfsck_exec_oit(const struct lu_env *env,
 576                           struct lfsck_instance *lfsck, struct dt_object *obj)
 577 {
 578         struct lfsck_component *com;
 579         int                     rc;
 580         ENTRY;
 581
 582         LASSERT(lfsck->li_obj_dir == NULL);
 583
 584         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 585                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
 586                 if (rc != 0)
 587                         RETURN(rc);
 588         }
 589
 590         rc = lfsck_needs_scan_dir(env, lfsck, obj);
 591         if (rc <= 0)
 592                 GOTO(out, rc);
 593
 594         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 595         if (rc == 0)
 596                 rc = lfsck_open_dir(env, lfsck, 0);
 597
 598         GOTO(out, rc);
 599
 600 out:
 601         if (rc < 0)
 602                 lfsck_fail(env, lfsck, false);
 603
 604         if (rc != 0)
 605                 lfsck_close_dir(env, lfsck, rc);
 606
 607         return rc > 0 ? 0 : rc;
 608 }
 609
 610 static int lfsck_exec_dir(const struct lu_env *env,
 611                           struct lfsck_instance *lfsck,
 612                           struct lu_dirent *ent, __u16 type)
 613 {
 614         struct lfsck_component *com;
 615         int                     rc;
 616
 617         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 618                 rc = com->lc_ops->lfsck_exec_dir(env, com, ent, type);
 619                 if (rc != 0)
 620                         return rc;
 621         }
 622         return 0;
 623 }
 624
 625 static int lfsck_master_dir_engine(const struct lu_env *env,
 626                                    struct lfsck_instance *lfsck);
 627
 628 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 629                       int result)
 630 {
 631         struct lfsck_component *com;
 632         struct lfsck_component *next;
 633         int                     rc  = result;
 634
 635         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 636         lfsck_close_dir(env, lfsck, result);
 637
 638         while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
 639                !list_empty(&lfsck->li_list_lmv)) {
 640                 struct lfsck_lmv_unit *llu;
 641
 642                 spin_lock(&lfsck->li_lock);
 643                 llu = list_entry(lfsck->li_list_lmv.next,
 644                                  struct lfsck_lmv_unit, llu_link);
 645                 list_del_init(&llu->llu_link);
 646                 spin_unlock(&lfsck->li_lock);
 647
 648                 lfsck->li_lmv = &llu->llu_lmv;
 649                 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 650                 rc = lfsck_open_dir(env, lfsck, 0);
 651                 if (rc == 0) {
 652                         rc = lfsck_master_dir_engine(env, lfsck);
 653                         lfsck_close_dir(env, lfsck, result);
 654                 }
 655         }
 656
 657         result = rc;
 658
 659         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 660                 rc = com->lc_ops->lfsck_post(env, com, result, false);
 661                 if (rc != 0)
 662                         CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
 663                                "rc = %d\n", lfsck_lfsck2name(lfsck),
 664                                (__u32)com->lc_type, rc);
 665         }
 666
 667         lfsck->li_time_last_checkpoint = cfs_time_current();
 668         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 669                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 670
 671         /* Ignore some component post failure to make other can go ahead. */
 672         return result;
 673 }
 674
 675 static int lfsck_double_scan(const struct lu_env *env,
 676                              struct lfsck_instance *lfsck)
 677 {
 678         struct lfsck_component *com;
 679         struct lfsck_component *next;
 680         struct l_wait_info      lwi = { 0 };
 681         int                     rc  = 0;
 682         int                     rc1 = 0;
 683
 684         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
 685                 rc = com->lc_ops->lfsck_double_scan(env, com);
 686                 if (rc != 0)
 687                         rc1 = rc;
 688         }
 689
 690         l_wait_event(lfsck->li_thread.t_ctl_waitq,
 691                      atomic_read(&lfsck->li_double_scan_count) == 0,
 692                      &lwi);
 693
 694         if (lfsck->li_status != LS_PAUSED &&
 695             lfsck->li_status != LS_CO_PAUSED) {
 696                 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 697                                          lc_link) {
 698                         spin_lock(&lfsck->li_lock);
 699                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 700                         spin_unlock(&lfsck->li_lock);
 701                 }
 702         }
 703
 704         return rc1 != 0 ? rc1 : rc;
 705 }
 706
 707 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
 708 {
 709         struct lfsck_component *com;
 710         struct lfsck_component *next;
 711
 712         list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 713                                  lc_link) {
 714                 if (com->lc_ops->lfsck_quit != NULL)
 715                         com->lc_ops->lfsck_quit(env, com);
 716
 717                 spin_lock(&lfsck->li_lock);
 718                 list_del_init(&com->lc_link_dir);
 719                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 720                 spin_unlock(&lfsck->li_lock);
 721         }
 722
 723         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 724                                  lc_link) {
 725                 if (com->lc_ops->lfsck_quit != NULL)
 726                         com->lc_ops->lfsck_quit(env, com);
 727
 728                 spin_lock(&lfsck->li_lock);
 729                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 730                 spin_unlock(&lfsck->li_lock);
 731         }
 732 }
 733
 734 /* LFSCK engines */
 735
 736 static int lfsck_master_dir_engine(const struct lu_env *env,
 737                                    struct lfsck_instance *lfsck)
 738 {
 739         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 740         struct dt_object                *dir    = lfsck->li_obj_dir;
 741         const struct dt_it_ops          *iops   = &dir->do_index_ops->dio_it;
 742         struct dt_it                    *di     = lfsck->li_di_dir;
 743         struct lu_dirent                *ent    =
 744                         (struct lu_dirent *)info->lti_key;
 745         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 746         struct ptlrpc_thread            *thread = &lfsck->li_thread;
 747         int                              rc;
 748         __u16                            type;
 749         ENTRY;
 750
 751         do {
 752                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
 753                     unlikely(!thread_is_running(thread))) {
 754                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 755                                "parent "DFID", cookie "LPX64"\n",
 756                                lfsck_lfsck2name(lfsck),
 757                                PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
 758
 759                         RETURN(0);
 760                 }
 761
 762                 lfsck->li_new_scanned++;
 763                 rc = iops->rec(env, di, (struct dt_rec *)ent,
 764                                lfsck->li_args_dir);
 765                 if (rc == 0)
 766                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
 767                                               &type);
 768
 769                 if (rc != 0) {
 770                         CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
 771                                "parent "DFID", cookie "LPX64": rc = %d\n",
 772                                lfsck_lfsck2name(lfsck),
 773                                PFID(lfsck_dto2fid(dir)),
 774                                lfsck->li_cookie_dir, rc);
 775                         lfsck_fail(env, lfsck, true);
 776                         if (bk->lb_param & LPF_FAILOUT)
 777                                 RETURN(rc);
 778                         else
 779                                 goto checkpoint;
 780                 }
 781
 782                 if (ent->lde_attrs & LUDA_IGNORE &&
 783                     strcmp(ent->lde_name, dotdot) != 0)
 784                         goto checkpoint;
 785
 786                 /* skip dot entry. */
 787                 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
 788                         goto checkpoint;
 789
 790                 /* The type in the @ent structure may has been overwritten,
 791                  * so we need to pass the @type parameter independently. */
 792                 rc = lfsck_exec_dir(env, lfsck, ent, type);
 793                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 794                         RETURN(rc);
 795
 796 checkpoint:
 797                 rc = lfsck_checkpoint(env, lfsck);
 798                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 799                         RETURN(rc);
 800
 801                 /* Rate control. */
 802                 lfsck_control_speed(lfsck);
 803                 if (unlikely(!thread_is_running(thread))) {
 804                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 805                                "parent "DFID", cookie "LPX64"\n",
 806                                lfsck_lfsck2name(lfsck),
 807                                PFID(lfsck_dto2fid(dir)),
 808                                lfsck->li_cookie_dir);
 809                         RETURN(0);
 810                 }
 811
 812                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
 813                         spin_lock(&lfsck->li_lock);
 814                         thread_set_flags(thread, SVC_STOPPING);
 815                         spin_unlock(&lfsck->li_lock);
 816                         RETURN(-EINVAL);
 817                 }
 818
 819                 rc = iops->next(env, di);
 820         } while (rc == 0);
 821
 822         if (rc > 0 && !lfsck->li_oit_over)
 823                 lfsck_close_dir(env, lfsck, rc);
 824
 825         RETURN(rc);
 826 }
 827
 828 /**
 829  * Object-table based iteration engine.
 830  *
 831  * Object-table based iteration is the basic linear engine to scan all the
 832  * objects on current device in turn. For each object, it calls all the
 833  * registered LFSCK component(s)' API to perform related consistency
 834  * verification.
 835  *
 836  * It flushes related LFSCK trace files to disk via making checkpoint
 837  * periodically. Then if the server crashed or the LFSCK is paused, the
 838  * LFSCK can resume from the latest checkpoint.
 839  *
 840  * It also controls the whole LFSCK speed via lfsck_control_speed() to
 841  * avoid the server to become overload.
 842  *
 843  * \param[in] env       pointer to the thread context
 844  * \param[in] lfsck     pointer to the lfsck instance
 845  *
 846  * \retval              positive number if all objects have been scanned
 847  * \retval              0 if the iteration is stopped or paused
 848  * \retval              negative error number on failure
 849  */
 850 static int lfsck_master_oit_engine(const struct lu_env *env,
 851                                    struct lfsck_instance *lfsck)
 852 {
 853         struct lfsck_thread_info *info  = lfsck_env_info(env);
 854         const struct dt_it_ops   *iops  =
 855                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 856         struct dt_it             *di    = lfsck->li_di_oit;
 857         struct lu_fid            *fid   = &info->lti_fid;
 858         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 859         struct ptlrpc_thread     *thread = &lfsck->li_thread;
 860         struct seq_server_site   *ss    = lfsck_dev_site(lfsck);
 861         __u32                    idx    = lfsck_dev_idx(lfsck);
 862         int                      rc;
 863         ENTRY;
 864
 865         if (unlikely(ss == NULL))
 866                 RETURN(-EIO);
 867
 868         do {
 869                 struct dt_object *target;
 870                 bool              update_lma = false;
 871
 872                 if (lfsck->li_di_dir != NULL) {
 873                         rc = lfsck_master_dir_engine(env, lfsck);
 874                         if (rc <= 0)
 875                                 RETURN(rc);
 876                 }
 877
 878                 if (unlikely(lfsck->li_oit_over))
 879                         RETURN(1);
 880
 881                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
 882                     unlikely(!thread_is_running(thread))) {
 883                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
 884                                "cookie "LPU64"\n",
 885                                lfsck_lfsck2name(lfsck), iops->store(env, di));
 886
 887                         RETURN(0);
 888                 }
 889
 890                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
 891                         RETURN(0);
 892
 893                 lfsck->li_current_oit_processed = 1;
 894
 895                 if (!list_empty(&lfsck->li_list_lmv)) {
 896                         struct lfsck_lmv_unit *llu;
 897
 898                         spin_lock(&lfsck->li_lock);
 899                         llu = list_entry(lfsck->li_list_lmv.next,
 900                                          struct lfsck_lmv_unit, llu_link);
 901                         list_del_init(&llu->llu_link);
 902                         spin_unlock(&lfsck->li_lock);
 903
 904                         lfsck->li_lmv = &llu->llu_lmv;
 905                         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 906                         rc = lfsck_open_dir(env, lfsck, 0);
 907                         if (rc == 0)
 908                                 rc = lfsck_master_dir_engine(env, lfsck);
 909
 910                         if (rc <= 0)
 911                                 RETURN(rc);
 912                 }
 913
 914                 lfsck->li_new_scanned++;
 915                 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
 916                 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
 917                 if (rc != 0) {
 918                         CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
 919                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 920                         lfsck_fail(env, lfsck, true);
 921                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
 922                                 RETURN(rc);
 923                         else
 924                                 goto checkpoint;
 925                 }
 926
 927                 if (unlikely(!fid_is_sane(fid))) {
 928                         CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
 929                                ", skip it\n",
 930                                lfsck_lfsck2name(lfsck), PFID(fid));
 931                         goto checkpoint;
 932                 }
 933
 934                 if (fid_is_idif(fid)) {
 935                         __u32 idx1 = fid_idif_ost_idx(fid);
 936
 937                         LASSERT(!lfsck->li_master);
 938
 939                         /* It is an old format device, update the LMA. */
 940                         if (idx != idx1) {
 941                                 struct ost_id *oi = &info->lti_oi;
 942
 943                                 fid_to_ostid(fid, oi);
 944                                 ostid_to_fid(fid, oi, idx);
 945                                 update_lma = true;
 946                         }
 947                 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
 948                            !fid_is_last_id(fid) &&
 949                            !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
 950
 951                         /* If the FID/object is only used locally and invisible
 952                          * to external nodes, then LFSCK will not handle it.
 953                          *
 954                          * dot_lustre sequence has been handled specially. */
 955                         goto checkpoint;
 956                 } else {
 957                         struct lu_seq_range *range = &info->lti_range;
 958
 959                         if (lfsck->li_master)
 960                                 fld_range_set_mdt(range);
 961                         else
 962                                 fld_range_set_ost(range);
 963                         rc = fld_local_lookup(env, ss->ss_server_fld,
 964                                               fid_seq(fid), range);
 965                         if (rc != 0 || range->lsr_index != idx) {
 966                                 /* Remote object will be handled by the LFSCK
 967                                  * instance on the MDT where the remote object
 968                                  * really resides on. */
 969                                 rc = 0;
 970                                 goto checkpoint;
 971                         }
 972                 }
 973
 974                 target = lfsck_object_find_bottom(env, lfsck, fid);
 975                 if (IS_ERR(target)) {
 976                         CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
 977                                DFID", cookie "LPU64": rc = %d\n",
 978                                lfsck_lfsck2name(lfsck), PFID(fid),
 979                                iops->store(env, di), rc);
 980                         lfsck_fail(env, lfsck, true);
 981                         if (bk->lb_param & LPF_FAILOUT)
 982                                 RETURN(PTR_ERR(target));
 983                         else
 984                                 goto checkpoint;
 985                 }
 986
 987                 if (dt_object_exists(target)) {
 988                         if (update_lma) {
 989                                 rc = lfsck_update_lma(env, lfsck, target);
 990                                 if (rc != 0)
 991                                         CDEBUG(D_LFSCK, "%s: fail to update "
 992                                                "LMA for "DFID": rc = %d\n",
 993                                                lfsck_lfsck2name(lfsck),
 994                                                PFID(lfsck_dto2fid(target)), rc);
 995                         }
 996                         if (rc == 0)
 997                                 rc = lfsck_exec_oit(env, lfsck, target);
 998                 }
 999                 lfsck_object_put(env, target);
1000                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1001                         RETURN(rc);
1002
1003 checkpoint:
1004                 rc = lfsck_checkpoint(env, lfsck);
1005                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1006                         RETURN(rc);
1007
1008                 /* Rate control. */
1009                 lfsck_control_speed(lfsck);
1010
1011                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1012                         spin_lock(&lfsck->li_lock);
1013                         thread_set_flags(thread, SVC_STOPPING);
1014                         spin_unlock(&lfsck->li_lock);
1015                         RETURN(-EINVAL);
1016                 }
1017
1018                 rc = iops->next(env, di);
1019                 if (unlikely(rc > 0))
1020                         lfsck->li_oit_over = 1;
1021                 else if (likely(rc == 0))
1022                         lfsck->li_current_oit_processed = 0;
1023
1024                 if (unlikely(!thread_is_running(thread))) {
1025                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1026                                "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1027                                iops->store(env, di));
1028                         RETURN(0);
1029                 }
1030         } while (rc == 0 || lfsck->li_di_dir != NULL);
1031
1032         RETURN(rc);
1033 }
1034
1035 int lfsck_master_engine(void *args)
1036 {
1037         struct lfsck_thread_args *lta      = args;
1038         struct lu_env            *env      = &lta->lta_env;
1039         struct lfsck_instance    *lfsck    = lta->lta_lfsck;
1040         struct ptlrpc_thread     *thread   = &lfsck->li_thread;
1041         struct dt_object         *oit_obj  = lfsck->li_obj_oit;
1042         const struct dt_it_ops   *oit_iops = &oit_obj->do_index_ops->dio_it;
1043         struct dt_it             *oit_di;
1044         struct l_wait_info        lwi      = { 0 };
1045         int                       rc;
1046         ENTRY;
1047
1048         if (lfsck->li_master &&
1049             (!list_empty(&lfsck->li_list_scan) ||
1050              !list_empty(&lfsck->li_list_double_scan))) {
1051                 rc = lfsck_verify_lpf(env, lfsck);
1052                 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1053                  * fatal, because the .lustre/lost+found/ maybe not accessed
1054                  * by the LFSCK if it does not add orphans or others to such
1055                  * directory. So go ahead until hit failure when really uses
1056                  * the directory. */
1057                 if (rc != 0)
1058                         CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1059                                ".lustre/lost+found/, go ahead: rc = %d\n",
1060                                lfsck_lfsck2name(lfsck), rc);
1061         }
1062
1063         oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit, BYPASS_CAPA);
1064         if (IS_ERR(oit_di)) {
1065                 rc = PTR_ERR(oit_di);
1066                 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1067                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1068
1069                 GOTO(fini_args, rc);
1070         }
1071
1072         spin_lock(&lfsck->li_lock);
1073         lfsck->li_di_oit = oit_di;
1074         spin_unlock(&lfsck->li_lock);
1075         rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1076         if (rc != 0)
1077                 GOTO(fini_oit, rc);
1078
1079         CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1080                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1081                ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1082                lfsck->li_pos_checkpoint.lp_oit_cookie,
1083                lfsck->li_pos_checkpoint.lp_dir_cookie,
1084                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1085                current_pid());
1086
1087         spin_lock(&lfsck->li_lock);
1088         thread_set_flags(thread, SVC_RUNNING);
1089         spin_unlock(&lfsck->li_lock);
1090         wake_up_all(&thread->t_ctl_waitq);
1091
1092         l_wait_event(thread->t_ctl_waitq,
1093                      lfsck->li_start_unplug ||
1094                      !thread_is_running(thread),
1095                      &lwi);
1096         if (!thread_is_running(thread))
1097                 GOTO(fini_oit, rc = 0);
1098
1099         if (!list_empty(&lfsck->li_list_scan) ||
1100             list_empty(&lfsck->li_list_double_scan))
1101                 rc = lfsck_master_oit_engine(env, lfsck);
1102         else
1103                 rc = 1;
1104
1105         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1106         CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1107                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1108                ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1109                lfsck->li_pos_checkpoint.lp_oit_cookie,
1110                lfsck->li_pos_checkpoint.lp_dir_cookie,
1111                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1112                current_pid(), rc);
1113
1114         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1115                 rc = lfsck_post(env, lfsck, rc);
1116         else
1117                 lfsck_close_dir(env, lfsck, rc);
1118
1119 fini_oit:
1120         lfsck_di_oit_put(env, lfsck);
1121         oit_iops->fini(env, oit_di);
1122         if (rc == 1) {
1123                 if (!list_empty(&lfsck->li_list_double_scan))
1124                         rc = lfsck_double_scan(env, lfsck);
1125                 else
1126                         rc = 0;
1127         } else {
1128                 lfsck_quit(env, lfsck);
1129         }
1130
1131         /* XXX: Purge the pinned objects in the future. */
1132
1133 fini_args:
1134         spin_lock(&lfsck->li_lock);
1135         thread_set_flags(thread, SVC_STOPPED);
1136         spin_unlock(&lfsck->li_lock);
1137         wake_up_all(&thread->t_ctl_waitq);
1138         lfsck_thread_args_fini(lta);
1139         return rc;
1140 }
1141
1142 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1143 {
1144         bool empty = false;
1145
1146         spin_lock(&lad->lad_lock);
1147         if (list_empty(&lad->lad_req_list))
1148                 empty = true;
1149         spin_unlock(&lad->lad_lock);
1150
1151         return empty;
1152 }
1153
1154 /**
1155  * Query the LFSCK status from the instatnces on remote servers.
1156  *
1157  * The LFSCK assistant thread queries the LFSCK instances on other
1158  * servers (MDT/OST) about their status, such as whether they have
1159  * finished the phase1/phase2 scanning or not, and so on.
1160  *
1161  * \param[in] env       pointer to the thread context
1162  * \param[in] com       pointer to the lfsck component
1163  *
1164  * \retval              0 for success
1165  * \retval              negative error number on failure
1166  */
1167 static int lfsck_assistant_query_others(const struct lu_env *env,
1168                                         struct lfsck_component *com)
1169 {
1170         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1171         struct lfsck_request              *lr    = &info->lti_lr;
1172         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1173         struct lfsck_instance             *lfsck = com->lc_lfsck;
1174         struct lfsck_assistant_data       *lad   = com->lc_data;
1175         struct ptlrpc_request_set         *set;
1176         struct lfsck_tgt_descs            *ltds;
1177         struct lfsck_tgt_desc             *ltd;
1178         struct list_head                  *phase_head;
1179         int                                rc    = 0;
1180         int                                rc1   = 0;
1181         ENTRY;
1182
1183         set = ptlrpc_prep_set();
1184         if (set == NULL)
1185                 RETURN(-ENOMEM);
1186
1187         lad->lad_touch_gen++;
1188         memset(lr, 0, sizeof(*lr));
1189         lr->lr_event = LE_QUERY;
1190         lr->lr_active = com->lc_type;
1191         laia->laia_com = com;
1192         laia->laia_lr = lr;
1193         laia->laia_shared = 0;
1194
1195         if (!list_empty(&lad->lad_mdt_phase1_list)) {
1196                 ltds = &lfsck->li_mdt_descs;
1197                 lr->lr_flags = 0;
1198                 phase_head = &lad->lad_mdt_phase1_list;
1199         } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1200                 goto out;
1201         } else {
1202
1203 again:
1204                 ltds = &lfsck->li_ost_descs;
1205                 lr->lr_flags = LEF_TO_OST;
1206                 phase_head = &lad->lad_ost_phase1_list;
1207         }
1208
1209         laia->laia_ltds = ltds;
1210         spin_lock(&ltds->ltd_lock);
1211         while (!list_empty(phase_head)) {
1212                 struct list_head *phase_list;
1213                 __u32            *gen;
1214
1215                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1216                         ltd = list_entry(phase_head->next,
1217                                          struct lfsck_tgt_desc,
1218                                          ltd_layout_phase_list);
1219                         phase_list = &ltd->ltd_layout_phase_list;
1220                         gen = &ltd->ltd_layout_gen;
1221                 } else {
1222                         ltd = list_entry(phase_head->next,
1223                                          struct lfsck_tgt_desc,
1224                                          ltd_namespace_phase_list);
1225                         phase_list = &ltd->ltd_namespace_phase_list;
1226                         gen = &ltd->ltd_namespace_gen;
1227                 }
1228
1229                 if (*gen == lad->lad_touch_gen)
1230                         break;
1231
1232                 *gen = lad->lad_touch_gen;
1233                 list_move_tail(phase_list, phase_head);
1234                 atomic_inc(&ltd->ltd_ref);
1235                 laia->laia_ltd = ltd;
1236                 spin_unlock(&ltds->ltd_lock);
1237                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1238                                          lfsck_async_interpret_common,
1239                                          laia, LFSCK_QUERY);
1240                 if (rc != 0) {
1241                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1242                                "%s %x for %s: rc = %d\n",
1243                                lfsck_lfsck2name(lfsck),
1244                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1245                                ltd->ltd_index, lad->lad_name, rc);
1246                         lfsck_tgt_put(ltd);
1247                         rc1 = rc;
1248                 }
1249                 spin_lock(&ltds->ltd_lock);
1250         }
1251         spin_unlock(&ltds->ltd_lock);
1252
1253         rc = ptlrpc_set_wait(set);
1254         if (rc < 0) {
1255                 ptlrpc_set_destroy(set);
1256                 RETURN(rc);
1257         }
1258
1259         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1260             list_empty(&lad->lad_mdt_phase1_list))
1261                 goto again;
1262
1263 out:
1264         ptlrpc_set_destroy(set);
1265
1266         RETURN(rc1 != 0 ? rc1 : rc);
1267 }
1268
1269 /**
1270  * Notify the LFSCK event to the instances on remote servers.
1271  *
1272  * The LFSCK assistant thread notifies the LFSCK instances on other
1273  * servers (MDT/OST) about some events, such as start new scanning,
1274  * stop the scanning, this LFSCK instance will exit, and so on.
1275  *
1276  * \param[in] env       pointer to the thread context
1277  * \param[in] com       pointer to the lfsck component
1278  * \param[in] lr        pointer to the LFSCK event request
1279  *
1280  * \retval              0 for success
1281  * \retval              negative error number on failure
1282  */
1283 static int lfsck_assistant_notify_others(const struct lu_env *env,
1284                                          struct lfsck_component *com,
1285                                          struct lfsck_request *lr)
1286 {
1287         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1288         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1289         struct lfsck_instance             *lfsck = com->lc_lfsck;
1290         struct lfsck_assistant_data       *lad   = com->lc_data;
1291         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1292         struct ptlrpc_request_set         *set;
1293         struct lfsck_tgt_descs            *ltds;
1294         struct lfsck_tgt_desc             *ltd;
1295         struct lfsck_tgt_desc             *next;
1296         __u32                              idx;
1297         int                                rc    = 0;
1298         int                                rc1   = 0;
1299         ENTRY;
1300
1301         set = ptlrpc_prep_set();
1302         if (set == NULL)
1303                 RETURN(-ENOMEM);
1304
1305         lr->lr_index = lfsck_dev_idx(lfsck);
1306         lr->lr_active = com->lc_type;
1307         laia->laia_com = com;
1308         laia->laia_lr = lr;
1309         laia->laia_shared = 0;
1310
1311         switch (lr->lr_event) {
1312         case LE_START:
1313                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1314                         goto next;
1315
1316                 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1317                 lr->lr_speed = bk->lb_speed_limit;
1318                 lr->lr_version = bk->lb_version;
1319                 lr->lr_param |= bk->lb_param;
1320                 lr->lr_async_windows = bk->lb_async_windows;
1321                 lr->lr_flags = LEF_TO_OST;
1322
1323                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1324                 ltds = &lfsck->li_ost_descs;
1325                 laia->laia_ltds = ltds;
1326                 down_read(&ltds->ltd_rw_sem);
1327                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1328                         ltd = lfsck_tgt_get(ltds, idx);
1329                         LASSERT(ltd != NULL);
1330
1331                         laia->laia_ltd = ltd;
1332                         ltd->ltd_layout_done = 0;
1333                         ltd->ltd_synced_failures = 0;
1334                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1335                                         lfsck_async_interpret_common,
1336                                         laia, LFSCK_NOTIFY);
1337                         if (rc != 0) {
1338                                 lfsck_lad_set_bitmap(env, com, idx);
1339                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1340                                        "notify OST %x for %s start: rc = %d\n",
1341                                        lfsck_lfsck2name(lfsck), idx,
1342                                        lad->lad_name, rc);
1343                                 lfsck_tgt_put(ltd);
1344                         }
1345                 }
1346                 up_read(&ltds->ltd_rw_sem);
1347
1348                 /* Sync up */
1349                 rc = ptlrpc_set_wait(set);
1350                 if (rc < 0) {
1351                         ptlrpc_set_destroy(set);
1352                         RETURN(rc);
1353                 }
1354
1355 next:
1356                 if (!(bk->lb_param & LPF_ALL_TGT))
1357                         break;
1358
1359                 /* link other MDT targets locallly. */
1360                 ltds = &lfsck->li_mdt_descs;
1361                 spin_lock(&ltds->ltd_lock);
1362                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1363                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1364                                 ltd = LTD_TGT(ltds, idx);
1365                                 LASSERT(ltd != NULL);
1366
1367                                 if (!list_empty(&ltd->ltd_layout_list))
1368                                         continue;
1369
1370                                 list_add_tail(&ltd->ltd_layout_list,
1371                                               &lad->lad_mdt_list);
1372                                 list_add_tail(&ltd->ltd_layout_phase_list,
1373                                               &lad->lad_mdt_phase1_list);
1374                         }
1375                 } else {
1376                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1377                                 ltd = LTD_TGT(ltds, idx);
1378                                 LASSERT(ltd != NULL);
1379
1380                                 if (!list_empty(&ltd->ltd_namespace_list))
1381                                         continue;
1382
1383                                 list_add_tail(&ltd->ltd_namespace_list,
1384                                               &lad->lad_mdt_list);
1385                                 list_add_tail(&ltd->ltd_namespace_phase_list,
1386                                               &lad->lad_mdt_phase1_list);
1387                         }
1388                 }
1389                 spin_unlock(&ltds->ltd_lock);
1390                 break;
1391         case LE_STOP:
1392         case LE_PHASE2_DONE:
1393         case LE_PEER_EXIT: {
1394                 struct list_head *phase_head;
1395
1396                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1397                 if (bk->lb_param & LPF_ALL_TGT) {
1398                         phase_head = &lad->lad_mdt_list;
1399                         ltds = &lfsck->li_mdt_descs;
1400                         if (lr->lr_event == LE_STOP) {
1401                                 /* unlink other MDT targets locallly. */
1402                                 spin_lock(&ltds->ltd_lock);
1403                                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1404                                         list_for_each_entry_safe(ltd, next,
1405                                                 phase_head, ltd_layout_list) {
1406                                                 list_del_init(
1407                                                 &ltd->ltd_layout_phase_list);
1408                                                 list_del_init(
1409                                                 &ltd->ltd_layout_list);
1410                                         }
1411                                 } else {
1412                                         list_for_each_entry_safe(ltd, next,
1413                                                         phase_head,
1414                                                         ltd_namespace_list) {
1415                                                 list_del_init(
1416                                                 &ltd->ltd_namespace_phase_list);
1417                                                 list_del_init(
1418                                                 &ltd->ltd_namespace_list);
1419                                         }
1420                                 }
1421                                 spin_unlock(&ltds->ltd_lock);
1422
1423                                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1424                                         break;
1425
1426                                 lr->lr_flags |= LEF_TO_OST;
1427                                 phase_head = &lad->lad_ost_list;
1428                                 ltds = &lfsck->li_ost_descs;
1429                         } else {
1430                                 lr->lr_flags &= ~LEF_TO_OST;
1431                         }
1432                 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1433                         break;
1434                 } else {
1435                         lr->lr_flags |= LEF_TO_OST;
1436                         phase_head = &lad->lad_ost_list;
1437                         ltds = &lfsck->li_ost_descs;
1438                 }
1439
1440 again:
1441                 laia->laia_ltds = ltds;
1442                 spin_lock(&ltds->ltd_lock);
1443                 while (!list_empty(phase_head)) {
1444                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1445                                 ltd = list_entry(phase_head->next,
1446                                                  struct lfsck_tgt_desc,
1447                                                  ltd_layout_list);
1448                                 if (!list_empty(&ltd->ltd_layout_phase_list))
1449                                         list_del_init(
1450                                                 &ltd->ltd_layout_phase_list);
1451                                 list_del_init(&ltd->ltd_layout_list);
1452                         } else {
1453                                 ltd = list_entry(phase_head->next,
1454                                                  struct lfsck_tgt_desc,
1455                                                  ltd_namespace_list);
1456                                 if (!list_empty(&ltd->ltd_namespace_phase_list))
1457                                         list_del_init(
1458                                                 &ltd->ltd_namespace_phase_list);
1459                                 list_del_init(&ltd->ltd_namespace_list);
1460                         }
1461                         atomic_inc(&ltd->ltd_ref);
1462                         laia->laia_ltd = ltd;
1463                         spin_unlock(&ltds->ltd_lock);
1464                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1465                                         lfsck_async_interpret_common,
1466                                         laia, LFSCK_NOTIFY);
1467                         if (rc != 0) {
1468                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1469                                        "notify %s %x for %s stop/phase2_done/"
1470                                        "peer_exit: rc = %d\n",
1471                                        lfsck_lfsck2name(lfsck),
1472                                        (lr->lr_flags & LEF_TO_OST) ?
1473                                        "OST" : "MDT", ltd->ltd_index,
1474                                        lad->lad_name, rc);
1475                                 lfsck_tgt_put(ltd);
1476                         }
1477                         spin_lock(&ltds->ltd_lock);
1478                 }
1479                 spin_unlock(&ltds->ltd_lock);
1480
1481                 rc = ptlrpc_set_wait(set);
1482                 if (rc < 0) {
1483                         ptlrpc_set_destroy(set);
1484                         RETURN(rc);
1485                 }
1486
1487                 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1488                     !(lr->lr_flags & LEF_TO_OST)) {
1489                         lr->lr_flags |= LEF_TO_OST;
1490                         phase_head = &lad->lad_ost_list;
1491                         ltds = &lfsck->li_ost_descs;
1492                         goto again;
1493                 }
1494                 break;
1495         }
1496         case LE_PHASE1_DONE:
1497                 lad->lad_ops->la_sync_failures(env, com, lr);
1498                 lad->lad_touch_gen++;
1499                 ltds = &lfsck->li_mdt_descs;
1500                 laia->laia_ltds = ltds;
1501                 spin_lock(&ltds->ltd_lock);
1502                 while (!list_empty(&lad->lad_mdt_list)) {
1503                         struct list_head *list;
1504                         __u32            *gen;
1505
1506                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1507                                 ltd = list_entry(lad->lad_mdt_list.next,
1508                                                  struct lfsck_tgt_desc,
1509                                                  ltd_layout_list);
1510                                 list = &ltd->ltd_layout_list;
1511                                 gen = &ltd->ltd_layout_gen;
1512                         } else {
1513                                 struct lfsck_namespace *ns = com->lc_file_ram;
1514
1515                                 ltd = list_entry(lad->lad_mdt_list.next,
1516                                                  struct lfsck_tgt_desc,
1517                                                  ltd_namespace_list);
1518                                 list = &ltd->ltd_namespace_list;
1519                                 gen = &ltd->ltd_namespace_gen;
1520                                 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1521                         }
1522
1523                         if (*gen == lad->lad_touch_gen)
1524                                 break;
1525
1526                         *gen = lad->lad_touch_gen;
1527                         list_move_tail(list, &lad->lad_mdt_list);
1528                         if (ltd->ltd_synced_failures)
1529                                 continue;
1530
1531                         atomic_inc(&ltd->ltd_ref);
1532                         laia->laia_ltd = ltd;
1533                         spin_unlock(&ltds->ltd_lock);
1534                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1535                                         lfsck_async_interpret_common,
1536                                         laia, LFSCK_NOTIFY);
1537                         if (rc != 0) {
1538                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1539                                        "notify MDT %x for %s phase1 done: "
1540                                        "rc = %d\n", lfsck_lfsck2name(lfsck),
1541                                        ltd->ltd_index, lad->lad_name, rc);
1542                                 lfsck_tgt_put(ltd);
1543                         }
1544                         spin_lock(&ltds->ltd_lock);
1545                 }
1546                 spin_unlock(&ltds->ltd_lock);
1547                 break;
1548         default:
1549                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1550                        "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1551                 rc = -EINVAL;
1552                 break;
1553         }
1554
1555         rc1 = ptlrpc_set_wait(set);
1556         ptlrpc_set_destroy(set);
1557
1558         RETURN(rc != 0 ? rc : rc1);
1559 }
1560
1561 /**
1562  * The LFSCK assistant thread is triggered by the LFSCK main engine.
1563  * They co-work together as an asynchronous pipeline: the LFSCK main
1564  * engine scans the system and pre-fetches the objects, attributes,
1565  * or name entries, etc, and pushes them into the pipeline as input
1566  * requests for the LFSCK assistant thread; on the other end of the
1567  * pipeline, the LFSCK assistant thread performs the real check and
1568  * repair for every request from the main engine.
1569  *
1570  * Generally, the assistant engine may be blocked when check/repair
1571  * something, so the LFSCK main engine will run some faster. On the
1572  * other hand, the LFSCK main engine will drive multiple assistant
1573  * threads in parallel, means for each LFSCK component on the master
1574  * (such as layout LFSCK, namespace LFSCK), there is an independent
1575  * LFSCK assistant thread. So under such 1:N multiple asynchronous
1576  * pipelines mode, the whole LFSCK performance will be much better
1577  * than check/repair everything by the LFSCK main engine itself.
1578  */
1579 int lfsck_assistant_engine(void *args)
1580 {
1581         struct lfsck_thread_args          *lta     = args;
1582         struct lu_env                     *env     = &lta->lta_env;
1583         struct lfsck_component            *com     = lta->lta_com;
1584         struct lfsck_instance             *lfsck   = lta->lta_lfsck;
1585         struct lfsck_bookmark             *bk      = &lfsck->li_bookmark_ram;
1586         struct lfsck_position             *pos     = &com->lc_pos_start;
1587         struct lfsck_thread_info          *info    = lfsck_env_info(env);
1588         struct lfsck_request              *lr      = &info->lti_lr;
1589         struct lfsck_assistant_data       *lad     = com->lc_data;
1590         struct ptlrpc_thread              *mthread = &lfsck->li_thread;
1591         struct ptlrpc_thread              *athread = &lad->lad_thread;
1592         struct lfsck_assistant_operations *lao     = lad->lad_ops;
1593         struct lfsck_assistant_req        *lar;
1594         struct l_wait_info                 lwi     = { 0 };
1595         int                                rc      = 0;
1596         int                                rc1     = 0;
1597         int                                rc2;
1598         ENTRY;
1599
1600         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1601                lfsck_lfsck2name(lfsck), lad->lad_name);
1602
1603         memset(lr, 0, sizeof(*lr));
1604         lr->lr_event = LE_START;
1605         if (pos->lp_oit_cookie <= 1)
1606                 lr->lr_param = LPF_RESET;
1607         rc = lfsck_assistant_notify_others(env, com, lr);
1608         if (rc != 0) {
1609                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1610                        "to start %s: rc = %d\n",
1611                        lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1612                 GOTO(fini, rc);
1613         }
1614
1615         spin_lock(&lad->lad_lock);
1616         thread_set_flags(athread, SVC_RUNNING);
1617         spin_unlock(&lad->lad_lock);
1618         wake_up_all(&mthread->t_ctl_waitq);
1619
1620         while (1) {
1621                 while (!list_empty(&lad->lad_req_list)) {
1622                         bool wakeup = false;
1623
1624                         if (unlikely(lad->lad_exit ||
1625                                      !thread_is_running(mthread)))
1626                                 GOTO(cleanup1, rc = lad->lad_post_result);
1627
1628                         lar = list_entry(lad->lad_req_list.next,
1629                                          struct lfsck_assistant_req,
1630                                          lar_list);
1631                         /* Only the lfsck_assistant_engine thread itself can
1632                          * remove the "lar" from the head of the list, LFSCK
1633                          * engine thread only inserts other new "lar" at the
1634                          * end of the list. So it is safe to handle current
1635                          * "lar" without the spin_lock. */
1636                         rc = lao->la_handler_p1(env, com, lar);
1637                         spin_lock(&lad->lad_lock);
1638                         list_del_init(&lar->lar_list);
1639                         lad->lad_prefetched--;
1640                         /* Wake up the main engine thread only when the list
1641                          * is empty or half of the prefetched items have been
1642                          * handled to avoid too frequent thread schedule. */
1643                         if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1644                                 wakeup = true;
1645                         spin_unlock(&lad->lad_lock);
1646                         if (wakeup)
1647                                 wake_up_all(&mthread->t_ctl_waitq);
1648
1649                         lao->la_req_fini(env, lar);
1650                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1651                                 GOTO(cleanup1, rc);
1652                 }
1653
1654                 l_wait_event(athread->t_ctl_waitq,
1655                              !lfsck_assistant_req_empty(lad) ||
1656                              lad->lad_exit ||
1657                              lad->lad_to_post ||
1658                              lad->lad_to_double_scan,
1659                              &lwi);
1660
1661                 if (unlikely(lad->lad_exit))
1662                         GOTO(cleanup1, rc = lad->lad_post_result);
1663
1664                 if (!list_empty(&lad->lad_req_list))
1665                         continue;
1666
1667                 if (lad->lad_to_post) {
1668                         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1669                                lfsck_lfsck2name(lfsck), lad->lad_name);
1670
1671                         if (unlikely(lad->lad_exit))
1672                                 GOTO(cleanup1, rc = lad->lad_post_result);
1673
1674                         lad->lad_to_post = 0;
1675                         LASSERT(lad->lad_post_result > 0);
1676
1677                         memset(lr, 0, sizeof(*lr));
1678                         lr->lr_event = LE_PHASE1_DONE;
1679                         lr->lr_status = lad->lad_post_result;
1680                         rc = lfsck_assistant_notify_others(env, com, lr);
1681
1682                         CDEBUG(D_LFSCK, "%s: LFSCK assistant notified "
1683                                "others for %s post: rc = %d\n",
1684                                lfsck_lfsck2name(lfsck),
1685                                lad->lad_name, rc);
1686
1687                         /* Wakeup the master engine to go ahead. */
1688                         wake_up_all(&mthread->t_ctl_waitq);
1689                 }
1690
1691                 if (lad->lad_to_double_scan) {
1692                         lad->lad_to_double_scan = 0;
1693                         atomic_inc(&lfsck->li_double_scan_count);
1694                         lad->lad_in_double_scan = 1;
1695                         wake_up_all(&mthread->t_ctl_waitq);
1696
1697                         com->lc_new_checked = 0;
1698                         com->lc_new_scanned = 0;
1699                         com->lc_time_last_checkpoint = cfs_time_current();
1700                         com->lc_time_next_checkpoint =
1701                                 com->lc_time_last_checkpoint +
1702                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1703
1704                         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1705                                "the second-stage scaning\n",
1706                                lfsck_lfsck2name(lfsck));
1707
1708                         /* Flush async updates before handling orphan. */
1709                         rc2 = dt_sync(env, lfsck->li_next);
1710
1711                         CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1712                                "scan start, synced: rc = %d\n",
1713                                lfsck_lfsck2name(lfsck), rc2);
1714
1715                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1716                                 GOTO(cleanup2, rc = 0);
1717
1718                         while (lad->lad_in_double_scan) {
1719                                 rc = lfsck_assistant_query_others(env, com);
1720                                 if (lfsck_phase2_next_ready(lad))
1721                                         goto p2_next;
1722
1723                                 if (rc < 0)
1724                                         GOTO(cleanup2, rc);
1725
1726                                 /* Pull LFSCK status on related targets once
1727                                  * per 30 seconds if we are not notified. */
1728                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1729                                                            cfs_time_seconds(1),
1730                                                            NULL, NULL);
1731                                 rc = l_wait_event(athread->t_ctl_waitq,
1732                                         lfsck_phase2_next_ready(lad) ||
1733                                         lad->lad_exit ||
1734                                         !thread_is_running(mthread),
1735                                         &lwi);
1736
1737                                 if (unlikely(lad->lad_exit ||
1738                                              !thread_is_running(mthread)))
1739                                         GOTO(cleanup2, rc = 0);
1740
1741                                 if (rc == -ETIMEDOUT)
1742                                         continue;
1743
1744                                 if (rc < 0)
1745                                         GOTO(cleanup2, rc);
1746
1747 p2_next:
1748                                 rc = lao->la_handler_p2(env, com);
1749                                 if (rc != 0)
1750                                         GOTO(cleanup2, rc);
1751
1752                                 if (unlikely(lad->lad_exit ||
1753                                              !thread_is_running(mthread)))
1754                                         GOTO(cleanup2, rc = 0);
1755                         }
1756                 }
1757         }
1758
1759 cleanup1:
1760         /* Cleanup the unfinished requests. */
1761         spin_lock(&lad->lad_lock);
1762         if (rc < 0)
1763                 lad->lad_assistant_status = rc;
1764
1765         if (lad->lad_exit && lad->lad_post_result <= 0)
1766                 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1767
1768         while (!list_empty(&lad->lad_req_list)) {
1769                 lar = list_entry(lad->lad_req_list.next,
1770                                  struct lfsck_assistant_req,
1771                                  lar_list);
1772                 list_del_init(&lar->lar_list);
1773                 lad->lad_prefetched--;
1774                 spin_unlock(&lad->lad_lock);
1775                 lao->la_req_fini(env, lar);
1776                 spin_lock(&lad->lad_lock);
1777         }
1778         spin_unlock(&lad->lad_lock);
1779
1780         LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1781                  lad->lad_prefetched);
1782
1783 cleanup2:
1784         memset(lr, 0, sizeof(*lr));
1785         if (rc > 0) {
1786                 lr->lr_event = LE_PHASE2_DONE;
1787                 lr->lr_status = rc;
1788         } else if (rc == 0) {
1789                 if (lfsck->li_flags & LPF_ALL_TGT) {
1790                         lr->lr_event = LE_STOP;
1791                         lr->lr_status = LS_STOPPED;
1792                 } else {
1793                         lr->lr_event = LE_PEER_EXIT;
1794                         switch (lfsck->li_status) {
1795                         case LS_PAUSED:
1796                         case LS_CO_PAUSED:
1797                                 lr->lr_status = LS_CO_PAUSED;
1798                                 break;
1799                         case LS_STOPPED:
1800                         case LS_CO_STOPPED:
1801                                 lr->lr_status = LS_CO_STOPPED;
1802                                 break;
1803                         default:
1804                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1805                                        "status: rc = %d\n",
1806                                        lfsck_lfsck2name(lfsck),
1807                                        lfsck->li_status);
1808                                 lr->lr_status = LS_CO_FAILED;
1809                                 break;
1810                         }
1811                 }
1812         } else {
1813                 if (lfsck->li_flags & LPF_ALL_TGT) {
1814                         lr->lr_event = LE_STOP;
1815                         lr->lr_status = LS_FAILED;
1816                 } else {
1817                         lr->lr_event = LE_PEER_EXIT;
1818                         lr->lr_status = LS_CO_FAILED;
1819                 }
1820         }
1821
1822         rc1 = lfsck_assistant_notify_others(env, com, lr);
1823         if (rc1 != 0) {
1824                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1825                        "others for %s quit: rc = %d\n",
1826                        lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1827                 rc = rc1;
1828         }
1829
1830         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1831                lfsck_lfsck2name(lfsck));
1832
1833         /* Flush async updates before exit. */
1834         rc2 = dt_sync(env, lfsck->li_next);
1835
1836         CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1837                lfsck_lfsck2name(lfsck), rc2);
1838
1839         /* Under force exit case, some requests may be just freed without
1840          * verification, those objects should be re-handled when next run.
1841          * So not update the on-disk trace file under such case. */
1842         if (lad->lad_in_double_scan) {
1843                 if (!lad->lad_exit)
1844                         rc1 = lao->la_double_scan_result(env, com, rc);
1845
1846                 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1847                        "finished: rc = %d\n",
1848                        lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1849         }
1850
1851 fini:
1852         if (lad->lad_in_double_scan)
1853                 atomic_dec(&lfsck->li_double_scan_count);
1854
1855         spin_lock(&lad->lad_lock);
1856         lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1857         thread_set_flags(athread, SVC_STOPPED);
1858         wake_up_all(&mthread->t_ctl_waitq);
1859         spin_unlock(&lad->lad_lock);
1860
1861         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1862                lfsck_lfsck2name(lfsck), lad->lad_name,
1863                lad->lad_assistant_status);
1864
1865         lfsck_thread_args_fini(lta);
1866
1867         return rc;
1868 }