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