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) 2012, 2013, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_engine.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lu_object.h>
  34 #include <dt_object.h>
  35 #include <lustre_net.h>
  36 #include <lustre_fid.h>
  37 #include <obd_support.h>
  38 #include <lustre_lib.h>
  39
  40 #include "lfsck_internal.h"
  41
  42 int lfsck_unpack_ent(struct lu_dirent *ent, __u64 *cookie, __u16 *type)
  43 {
  44         struct luda_type        *lt;
  45         int                      align = sizeof(*lt) - 1;
  46         int                      len;
  47
  48         fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
  49         *cookie = le64_to_cpu(ent->lde_hash);
  50         ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
  51         ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
  52         ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
  53
  54         if (unlikely(!(ent->lde_attrs & LUDA_TYPE)))
  55                 return -EINVAL;
  56
  57         len = (ent->lde_namelen + align) & ~align;
  58         lt = (struct luda_type *)(ent->lde_name + len);
  59         *type = le16_to_cpu(lt->lt_type);
  60
  61         /* Make sure the name is terminated with '\0'. The data (object type)
  62          * after ent::lde_name maybe broken, but we have stored such data in
  63          * the output parameter @type as above. */
  64         ent->lde_name[ent->lde_namelen] = '\0';
  65
  66         return 0;
  67 }
  68
  69 static void lfsck_di_oit_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  70 {
  71         const struct dt_it_ops  *iops;
  72         struct dt_it            *di;
  73
  74         spin_lock(&lfsck->li_lock);
  75         iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
  76         di = lfsck->li_di_oit;
  77         lfsck->li_di_oit = NULL;
  78         spin_unlock(&lfsck->li_lock);
  79         iops->put(env, di);
  80 }
  81
  82 static void lfsck_di_dir_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  83 {
  84         const struct dt_it_ops  *iops;
  85         struct dt_it            *di;
  86
  87         spin_lock(&lfsck->li_lock);
  88         iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
  89         di = lfsck->li_di_dir;
  90         lfsck->li_di_dir = NULL;
  91         lfsck->li_cookie_dir = 0;
  92         spin_unlock(&lfsck->li_lock);
  93         iops->put(env, di);
  94 }
  95
  96 static int lfsck_update_lma(const struct lu_env *env,
  97                             struct lfsck_instance *lfsck, struct dt_object *obj)
  98 {
  99         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 100         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 101         struct dt_device                *dt     = lfsck->li_bottom;
 102         struct lustre_mdt_attrs         *lma    = &info->lti_lma;
 103         struct lu_buf                   *buf;
 104         struct thandle                  *th;
 105         int                              fl;
 106         int                              rc;
 107         ENTRY;
 108
 109         if (bk->lb_param & LPF_DRYRUN)
 110                 RETURN(0);
 111
 112         buf = lfsck_buf_get(env, info->lti_lma_old, LMA_OLD_SIZE);
 113         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LMA, BYPASS_CAPA);
 114         if (rc < 0) {
 115                 if (rc != -ENODATA)
 116                         RETURN(rc);
 117
 118                 fl = LU_XATTR_CREATE;
 119                 lustre_lma_init(lma, lfsck_dto2fid(obj), LMAC_FID_ON_OST, 0);
 120         } else {
 121                 if (rc != LMA_OLD_SIZE && rc != sizeof(struct lustre_mdt_attrs))
 122                         RETURN(-EINVAL);
 123
 124                 fl = LU_XATTR_REPLACE;
 125                 lustre_lma_swab(lma);
 126                 lustre_lma_init(lma, lfsck_dto2fid(obj),
 127                                 lma->lma_compat | LMAC_FID_ON_OST,
 128                                 lma->lma_incompat);
 129         }
 130         lustre_lma_swab(lma);
 131
 132         th = dt_trans_create(env, dt);
 133         if (IS_ERR(th))
 134                 RETURN(PTR_ERR(th));
 135
 136         buf = lfsck_buf_get(env, lma, sizeof(*lma));
 137         rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
 138         if (rc != 0)
 139                 GOTO(stop, rc);
 140
 141         rc = dt_trans_start(env, dt, th);
 142         if (rc != 0)
 143                 GOTO(stop, rc);
 144
 145         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th, BYPASS_CAPA);
 146
 147         GOTO(stop, rc);
 148
 149 stop:
 150         dt_trans_stop(env, dt, th);
 151         return rc;
 152 }
 153
 154 static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj,
 155                             struct lu_fid *fid)
 156 {
 157         if (unlikely(!S_ISDIR(lfsck_object_type(obj)) ||
 158                      !dt_try_as_dir(env, obj)))
 159                 return -ENOTDIR;
 160
 161         return dt_lookup(env, obj, (struct dt_rec *)fid,
 162                          (const struct dt_key *)"..", BYPASS_CAPA);
 163 }
 164
 165 /**
 166  * Check whether needs to scan the directory or not.
 167  *
 168  * 1) If we are not doing namespace LFSCK, or the given @obj is not directory,
 169  *    then needs not to scan the @obj. Otherwise,
 170  * 2) Global /ROOT needs to be scanned, backend root needs not to be scanned.
 171  * 3) If the @obj is neither IGIF nor normal FID (including .lustre and its
 172  *    sub-directories that have been scanned when the LFSCK engine start),
 173  *    then needs not to be scanned.
 174  * 4) If it is a remote object, then scanning the object will be done on the
 175  *    MDT on which the object really resides.
 176  * 5) If the local object has normal FID, then needs to be scanned. Otherwise,
 177  * 6) If the object has linkEA, then needs to be scanned. Otherwise,
 178  * 7) If none of the previous conditions are true, we need to check the parent
 179  *    directories whether this subdirectory is in a tree that should be scanned.
 180  *    Set the parent as current @obj, repeat 2)-7).
 181  *
 182  * \param[in] env       pointer to the thread context
 183  * \param[in] lfsck     pointer to the lfsck instance
 184  * \param[in] obj       pointer to the object to be checked
 185  *
 186  * \retval              positive number if the directory needs to be scanned
 187  * \retval              0 if the directory needs NOT to be scanned
 188  * \retval              negative error number on failure
 189  */
 190 static int lfsck_needs_scan_dir(const struct lu_env *env,
 191                                 struct lfsck_instance *lfsck,
 192                                 struct dt_object *obj)
 193 {
 194         struct lfsck_thread_info *info    = lfsck_env_info(env);
 195         struct lu_fid            *fid     = &info->lti_fid;
 196         struct lu_seq_range      *range   = &info->lti_range;
 197         struct dt_device         *dev     = lfsck->li_bottom;
 198         struct seq_server_site   *ss      = lu_site2seq(dev->dd_lu_dev.ld_site);
 199         __u32                     idx     = lfsck_dev_idx(dev);
 200         int                       depth   = 0;
 201         int                       rc      = 0;
 202
 203         if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
 204                 return 0;
 205
 206         LASSERT(ss != NULL);
 207
 208         *fid = *lfsck_dto2fid(obj);
 209         while (1) {
 210                 /* Global /ROOT is visible. */
 211                 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
 212                         return 1;
 213
 214                 /* Backend root is invisible. */
 215                 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
 216                         return 0;
 217
 218                 if (!fid_is_norm(fid) && !fid_is_igif(fid))
 219                         return 0;
 220
 221                 fld_range_set_mdt(range);
 222                 rc = fld_local_lookup(env, ss->ss_server_fld,
 223                                       fid_seq(fid), range);
 224                 if (rc != 0 || range->lsr_index != idx) {
 225                         /* Current FID should NOT be for the input parameter
 226                          * @obj, because the lfsck_master_oit_engine() has
 227                          * filtered out agent object. So current FID is for
 228                          * the ancestor of the original input parameter @obj.
 229                          * So the ancestor is a remote directory. The input
 230                          * parameter @obj is local directory, and should be
 231                          * scanned under such case. */
 232                         LASSERT(depth > 0);
 233
 234                         return 1;
 235                 }
 236
 237                 /* normal FID on this target (locally) must be for the
 238                  * client-side visiable object. */
 239                 if (fid_is_norm(fid))
 240                         return 1;
 241
 242                 if (obj == NULL) {
 243                         obj = lfsck_object_find(env, lfsck, fid);
 244                         if (IS_ERR(obj))
 245                                 return PTR_ERR(obj);
 246
 247                         depth++;
 248                         if (!dt_object_exists(obj))
 249                                 GOTO(out, rc = 0);
 250                 }
 251
 252                 dt_read_lock(env, obj, MOR_TGT_CHILD);
 253                 if (unlikely(lfsck_is_dead_obj(obj))) {
 254                         dt_read_unlock(env, obj);
 255
 256                         GOTO(out, rc = 0);
 257                 }
 258
 259                 rc = dt_xattr_get(env, obj,
 260                                   lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK,
 261                                   BYPASS_CAPA);
 262                 dt_read_unlock(env, obj);
 263                 if (rc >= 0)
 264                         GOTO(out, rc = 1);
 265
 266                 if (rc < 0 && rc != -ENODATA)
 267                         GOTO(out, rc);
 268
 269                 rc = lfsck_parent_fid(env, obj, fid);
 270                 if (depth > 0)
 271                         lfsck_object_put(env, obj);
 272
 273                 obj = NULL;
 274                 if (rc != 0)
 275                         return rc;
 276
 277                 if (!fid_is_sane(fid))
 278                         return 0;
 279         }
 280
 281 out:
 282         if (depth > 0 && obj != NULL)
 283                 lfsck_object_put(env, obj);
 284
 285         return rc;
 286 }
 287
 288 static int lfsck_load_stripe_lmv(const struct lu_env *env,
 289                                  struct lfsck_instance *lfsck,
 290                                  struct dt_object *obj)
 291 {
 292         struct lmv_mds_md_v1    *lmv    = &lfsck_env_info(env)->lti_lmv;
 293         struct lfsck_lmv        *llmv;
 294         int                      rc;
 295         ENTRY;
 296
 297         LASSERT(lfsck->li_obj_dir == NULL);
 298         LASSERT(lfsck->li_lmv == NULL);
 299
 300         rc = lfsck_read_stripe_lmv(env, obj, lmv);
 301         if (rc == -ENODATA) {
 302                 lfsck->li_obj_dir = lfsck_object_get(obj);
 303
 304                 RETURN(0);
 305         }
 306
 307         if (rc < 0)
 308                 RETURN(rc);
 309
 310         OBD_ALLOC_PTR(llmv);
 311         if (llmv == NULL)
 312                 RETURN(-ENOMEM);
 313
 314         if (lmv->lmv_magic == LMV_MAGIC) {
 315                 struct lfsck_slave_lmv_rec      *lslr;
 316                 __u32                            stripes;
 317
 318                 llmv->ll_lmv_master = 1;
 319                 if (lmv->lmv_stripe_count < 1)
 320                         stripes = LFSCK_LMV_DEF_STRIPES;
 321                 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
 322                         stripes = LFSCK_LMV_MAX_STRIPES;
 323                 else
 324                         stripes = lmv->lmv_stripe_count;
 325
 326                 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
 327                 if (lslr == NULL) {
 328                         OBD_FREE_PTR(llmv);
 329
 330                         RETURN(-ENOMEM);
 331                 }
 332
 333                 /* Find the object against the bottom device. */
 334                 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
 335                                                lfsck_dto2fid(obj));
 336                 if (IS_ERR(obj)) {
 337                         OBD_FREE_LARGE(lslr, sizeof(*lslr) * stripes);
 338                         OBD_FREE_PTR(llmv);
 339
 340                         RETURN(PTR_ERR(obj));
 341                 }
 342
 343                 llmv->ll_stripes_allocated = stripes;
 344                 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
 345                 llmv->ll_lslr = lslr;
 346                 lfsck->li_obj_dir = obj;
 347         } else {
 348                 llmv->ll_lmv_slave = 1;
 349                 lfsck->li_obj_dir = lfsck_object_get(obj);
 350         }
 351
 352         llmv->ll_lmv = *lmv;
 353         atomic_set(&llmv->ll_ref, 1);
 354         lfsck->li_lmv = llmv;
 355
 356         RETURN(0);
 357 }
 358
 359 /* LFSCK wrap functions */
 360
 361 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
 362                        bool new_checked)
 363 {
 364         struct lfsck_component *com;
 365
 366         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 367                 com->lc_ops->lfsck_fail(env, com, new_checked);
 368         }
 369 }
 370
 371 void lfsck_close_dir(const struct lu_env *env,
 372                      struct lfsck_instance *lfsck, int result)
 373 {
 374         struct lfsck_component *com;
 375         ENTRY;
 376
 377         if (lfsck->li_lmv != NULL) {
 378                 lfsck->li_lmv->ll_exit_value = result;
 379                 if (lfsck->li_obj_dir != NULL) {
 380                         list_for_each_entry(com, &lfsck->li_list_dir,
 381                                             lc_link_dir) {
 382                                 com->lc_ops->lfsck_close_dir(env, com);
 383                         }
 384                 }
 385
 386                 lfsck_lmv_put(env, lfsck->li_lmv);
 387                 lfsck->li_lmv = NULL;
 388         }
 389
 390         if (lfsck->li_di_dir != NULL) {
 391                 const struct dt_it_ops  *dir_iops =
 392                                 &lfsck->li_obj_dir->do_index_ops->dio_it;
 393                 struct dt_it            *dir_di   = lfsck->li_di_dir;
 394
 395                 lfsck_di_dir_put(env, lfsck);
 396                 dir_iops->fini(env, dir_di);
 397         }
 398
 399         if (lfsck->li_obj_dir != NULL) {
 400                 struct dt_object        *dir_obj  = lfsck->li_obj_dir;
 401
 402                 lfsck->li_obj_dir = NULL;
 403                 lfsck_object_put(env, dir_obj);
 404         }
 405
 406         EXIT;
 407 }
 408
 409 int lfsck_open_dir(const struct lu_env *env,
 410                    struct lfsck_instance *lfsck, __u64 cookie)
 411 {
 412         struct dt_object        *obj    = lfsck->li_obj_dir;
 413         struct dt_it            *di     = lfsck->li_di_dir;
 414         struct lfsck_component  *com;
 415         const struct dt_it_ops  *iops;
 416         int                      rc     = 0;
 417         ENTRY;
 418
 419         LASSERT(obj != NULL);
 420         LASSERT(di == NULL);
 421
 422         if (unlikely(!dt_try_as_dir(env, obj)))
 423                 GOTO(out, rc = -ENOTDIR);
 424
 425         list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
 426                 rc = com->lc_ops->lfsck_open_dir(env, com);
 427                 if (rc != 0)
 428                         GOTO(out, rc);
 429         }
 430
 431         iops = &obj->do_index_ops->dio_it;
 432         di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
 433         if (IS_ERR(di))
 434                 GOTO(out, rc = PTR_ERR(di));
 435
 436         rc = iops->load(env, di, cookie);
 437         if (rc == 0 || (rc > 0 && cookie > 0))
 438                 rc = iops->next(env, di);
 439         else if (rc > 0)
 440                 rc = 0;
 441
 442         if (rc != 0) {
 443                 iops->put(env, di);
 444                 iops->fini(env, di);
 445         } else {
 446                 lfsck->li_cookie_dir = iops->store(env, di);
 447                 spin_lock(&lfsck->li_lock);
 448                 lfsck->li_di_dir = di;
 449                 spin_unlock(&lfsck->li_lock);
 450         }
 451
 452         GOTO(out, rc);
 453
 454 out:
 455         if (rc != 0)
 456                 lfsck_close_dir(env, lfsck, rc);
 457
 458         return rc;
 459 }
 460
 461 static int lfsck_checkpoint(const struct lu_env *env,
 462                             struct lfsck_instance *lfsck)
 463 {
 464         struct lfsck_component *com;
 465         int                     rc  = 0;
 466         int                     rc1 = 0;
 467
 468         if (likely(cfs_time_beforeq(cfs_time_current(),
 469                                     lfsck->li_time_next_checkpoint)))
 470                 return 0;
 471
 472         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 473         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 474                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
 475                 if (rc != 0)
 476                         rc1 = rc;
 477         }
 478
 479         lfsck->li_time_last_checkpoint = cfs_time_current();
 480         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 481                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 482         return rc1 != 0 ? rc1 : rc;
 483 }
 484
 485 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 486                       struct lfsck_start_param *lsp)
 487 {
 488         struct dt_object       *obj     = NULL;
 489         struct lfsck_component *com;
 490         struct lfsck_component *next;
 491         struct lfsck_position  *pos     = NULL;
 492         const struct dt_it_ops *iops    =
 493                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 494         int                     rc;
 495         ENTRY;
 496
 497         LASSERT(lfsck->li_obj_dir == NULL);
 498         LASSERT(lfsck->li_di_dir == NULL);
 499
 500         lfsck->li_current_oit_processed = 0;
 501         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 502                 com->lc_new_checked = 0;
 503                 rc = com->lc_ops->lfsck_prep(env, com, lsp);
 504                 if (rc != 0)
 505                         GOTO(out, rc);
 506
 507                 if ((pos == NULL) ||
 508                     (!lfsck_pos_is_zero(&com->lc_pos_start) &&
 509                      lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
 510                         pos = &com->lc_pos_start;
 511         }
 512
 513         /* Init otable-based iterator. */
 514         if (pos == NULL) {
 515                 rc = iops->load(env, lfsck->li_di_oit, 0);
 516                 if (rc > 0) {
 517                         lfsck->li_oit_over = 1;
 518                         rc = 0;
 519                 }
 520
 521                 GOTO(out, rc);
 522         }
 523
 524         rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
 525         if (rc < 0)
 526                 GOTO(out, rc);
 527         else if (rc > 0)
 528                 lfsck->li_oit_over = 1;
 529
 530         if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
 531                 GOTO(out, rc = 0);
 532
 533         /* Find the directory for namespace-based traverse. */
 534         obj = lfsck_object_find(env, lfsck, &pos->lp_dir_parent);
 535         if (IS_ERR(obj))
 536                 RETURN(PTR_ERR(obj));
 537
 538         /* Remote directory will be scanned by the LFSCK instance
 539          * on the MDT where the remote object really resides on. */
 540         if (!dt_object_exists(obj) || dt_object_remote(obj) ||
 541             unlikely(!S_ISDIR(lfsck_object_type(obj))))
 542                 GOTO(out, rc = 0);
 543
 544         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 545         if (rc == 0) {
 546                 /* For the master MDT-object of a striped directory,
 547                  * reset the iteration from the directory beginning. */
 548                 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
 549                         pos->lp_dir_cookie = 0;
 550
 551                 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
 552         }
 553
 554         GOTO(out, rc);
 555
 556 out:
 557         if (obj != NULL)
 558                 lfsck_object_put(env, obj);
 559
 560         if (rc != 0) {
 561                 lfsck_close_dir(env, lfsck, rc);
 562                 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 563                                          lc_link) {
 564                         com->lc_ops->lfsck_post(env, com, rc, true);
 565                 }
 566
 567                 return rc;
 568         }
 569
 570         rc = 0;
 571         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
 572         lfsck->li_pos_current = lfsck->li_pos_checkpoint;
 573         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 574                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
 575                 if (rc != 0)
 576                         break;
 577         }
 578
 579         lfsck->li_time_last_checkpoint = cfs_time_current();
 580         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 581                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 582         return rc;
 583 }
 584
 585 static int lfsck_exec_oit(const struct lu_env *env,
 586                           struct lfsck_instance *lfsck, struct dt_object *obj)
 587 {
 588         struct lfsck_component *com;
 589         int                     rc;
 590         ENTRY;
 591
 592         LASSERT(lfsck->li_obj_dir == NULL);
 593
 594         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 595                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
 596                 if (rc != 0)
 597                         RETURN(rc);
 598         }
 599
 600         rc = lfsck_needs_scan_dir(env, lfsck, obj);
 601         if (rc <= 0)
 602                 GOTO(out, rc);
 603
 604         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 605         if (rc == 0)
 606                 rc = lfsck_open_dir(env, lfsck, 0);
 607
 608         GOTO(out, rc);
 609
 610 out:
 611         if (rc < 0)
 612                 lfsck_fail(env, lfsck, false);
 613
 614         if (rc != 0)
 615                 lfsck_close_dir(env, lfsck, rc);
 616
 617         return rc > 0 ? 0 : rc;
 618 }
 619
 620 static int lfsck_exec_dir(const struct lu_env *env,
 621                           struct lfsck_instance *lfsck,
 622                           struct lu_dirent *ent, __u16 type)
 623 {
 624         struct lfsck_component *com;
 625         int                     rc;
 626
 627         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 628                 rc = com->lc_ops->lfsck_exec_dir(env, com, ent, type);
 629                 if (rc != 0)
 630                         return rc;
 631         }
 632         return 0;
 633 }
 634
 635 static int lfsck_master_dir_engine(const struct lu_env *env,
 636                                    struct lfsck_instance *lfsck);
 637
 638 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 639                       int result)
 640 {
 641         struct lfsck_component *com;
 642         struct lfsck_component *next;
 643         int                     rc  = result;
 644
 645         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 646         lfsck_close_dir(env, lfsck, result);
 647
 648         while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
 649                !list_empty(&lfsck->li_list_lmv)) {
 650                 struct lfsck_lmv_unit *llu;
 651
 652                 spin_lock(&lfsck->li_lock);
 653                 llu = list_entry(lfsck->li_list_lmv.next,
 654                                  struct lfsck_lmv_unit, llu_link);
 655                 list_del_init(&llu->llu_link);
 656                 spin_unlock(&lfsck->li_lock);
 657
 658                 lfsck->li_lmv = &llu->llu_lmv;
 659                 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 660                 rc = lfsck_open_dir(env, lfsck, 0);
 661                 if (rc == 0) {
 662                         rc = lfsck_master_dir_engine(env, lfsck);
 663                         lfsck_close_dir(env, lfsck, result);
 664                 }
 665         }
 666
 667         result = rc;
 668
 669         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 670                 rc = com->lc_ops->lfsck_post(env, com, result, false);
 671                 if (rc != 0)
 672                         CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
 673                                "rc = %d\n", lfsck_lfsck2name(lfsck),
 674                                (__u32)com->lc_type, rc);
 675         }
 676
 677         lfsck->li_time_last_checkpoint = cfs_time_current();
 678         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 679                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 680
 681         /* Ignore some component post failure to make other can go ahead. */
 682         return result;
 683 }
 684
 685 static int lfsck_double_scan(const struct lu_env *env,
 686                              struct lfsck_instance *lfsck)
 687 {
 688         struct lfsck_component *com;
 689         struct lfsck_component *next;
 690         struct l_wait_info      lwi = { 0 };
 691         int                     rc  = 0;
 692         int                     rc1 = 0;
 693
 694         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
 695                 rc = com->lc_ops->lfsck_double_scan(env, com);
 696                 if (rc != 0)
 697                         rc1 = rc;
 698         }
 699
 700         l_wait_event(lfsck->li_thread.t_ctl_waitq,
 701                      atomic_read(&lfsck->li_double_scan_count) == 0,
 702                      &lwi);
 703
 704         if (lfsck->li_status != LS_PAUSED &&
 705             lfsck->li_status != LS_CO_PAUSED) {
 706                 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 707                                          lc_link) {
 708                         spin_lock(&lfsck->li_lock);
 709                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 710                         spin_unlock(&lfsck->li_lock);
 711                 }
 712         }
 713
 714         return rc1 != 0 ? rc1 : rc;
 715 }
 716
 717 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
 718 {
 719         struct lfsck_component *com;
 720         struct lfsck_component *next;
 721
 722         list_for_each_entry_safe(com, next, &lfsck->li_list_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_del_init(&com->lc_link_dir);
 729                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 730                 spin_unlock(&lfsck->li_lock);
 731         }
 732
 733         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 734                                  lc_link) {
 735                 if (com->lc_ops->lfsck_quit != NULL)
 736                         com->lc_ops->lfsck_quit(env, com);
 737
 738                 spin_lock(&lfsck->li_lock);
 739                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 740                 spin_unlock(&lfsck->li_lock);
 741         }
 742 }
 743
 744 /* LFSCK engines */
 745
 746 static int lfsck_master_dir_engine(const struct lu_env *env,
 747                                    struct lfsck_instance *lfsck)
 748 {
 749         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 750         struct dt_object                *dir    = lfsck->li_obj_dir;
 751         const struct dt_it_ops          *iops   = &dir->do_index_ops->dio_it;
 752         struct dt_it                    *di     = lfsck->li_di_dir;
 753         struct lu_dirent                *ent    =
 754                         (struct lu_dirent *)info->lti_key;
 755         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 756         struct ptlrpc_thread            *thread = &lfsck->li_thread;
 757         int                              rc;
 758         __u16                            type;
 759         ENTRY;
 760
 761         do {
 762                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY2) &&
 763                     cfs_fail_val > 0) {
 764                         struct l_wait_info lwi;
 765
 766                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
 767                                           NULL, NULL);
 768                         l_wait_event(thread->t_ctl_waitq,
 769                                      !thread_is_running(thread),
 770                                      &lwi);
 771
 772                         if (unlikely(!thread_is_running(thread))) {
 773                                 CDEBUG(D_LFSCK, "%s: scan dir exit for engine "
 774                                        "stop, parent "DFID", cookie "LPX64"\n",
 775                                        lfsck_lfsck2name(lfsck),
 776                                        PFID(lfsck_dto2fid(dir)),
 777                                        lfsck->li_cookie_dir);
 778                                 RETURN(0);
 779                         }
 780                 }
 781
 782                 lfsck->li_new_scanned++;
 783                 rc = iops->rec(env, di, (struct dt_rec *)ent,
 784                                lfsck->li_args_dir);
 785                 if (rc == 0)
 786                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
 787                                               &type);
 788
 789                 if (rc != 0) {
 790                         CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
 791                                "parent "DFID", cookie "LPX64": rc = %d\n",
 792                                lfsck_lfsck2name(lfsck),
 793                                PFID(lfsck_dto2fid(dir)),
 794                                lfsck->li_cookie_dir, rc);
 795                         lfsck_fail(env, lfsck, true);
 796                         if (bk->lb_param & LPF_FAILOUT)
 797                                 RETURN(rc);
 798                         else
 799                                 goto checkpoint;
 800                 }
 801
 802                 if (ent->lde_attrs & LUDA_IGNORE &&
 803                     strcmp(ent->lde_name, dotdot) != 0)
 804                         goto checkpoint;
 805
 806                 /* The type in the @ent structure may has been overwritten,
 807                  * so we need to pass the @type parameter independently. */
 808                 rc = lfsck_exec_dir(env, lfsck, ent, type);
 809                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 810                         RETURN(rc);
 811
 812 checkpoint:
 813                 rc = lfsck_checkpoint(env, lfsck);
 814                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 815                         RETURN(rc);
 816
 817                 /* Rate control. */
 818                 lfsck_control_speed(lfsck);
 819                 if (unlikely(!thread_is_running(thread))) {
 820                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 821                                "parent "DFID", cookie "LPX64"\n",
 822                                lfsck_lfsck2name(lfsck),
 823                                PFID(lfsck_dto2fid(dir)),
 824                                lfsck->li_cookie_dir);
 825                         RETURN(0);
 826                 }
 827
 828                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
 829                         spin_lock(&lfsck->li_lock);
 830                         thread_set_flags(thread, SVC_STOPPING);
 831                         spin_unlock(&lfsck->li_lock);
 832                         RETURN(-EINVAL);
 833                 }
 834
 835                 rc = iops->next(env, di);
 836         } while (rc == 0);
 837
 838         if (rc > 0 && !lfsck->li_oit_over)
 839                 lfsck_close_dir(env, lfsck, rc);
 840
 841         RETURN(rc);
 842 }
 843
 844 /**
 845  * Object-table based iteration engine.
 846  *
 847  * Object-table based iteration is the basic linear engine to scan all the
 848  * objects on current device in turn. For each object, it calls all the
 849  * registered LFSCK component(s)' API to perform related consistency
 850  * verification.
 851  *
 852  * It flushes related LFSCK trace files to disk via making checkpoint
 853  * periodically. Then if the server crashed or the LFSCK is paused, the
 854  * LFSCK can resume from the latest checkpoint.
 855  *
 856  * It also controls the whole LFSCK speed via lfsck_control_speed() to
 857  * avoid the server to become overload.
 858  *
 859  * \param[in] env       pointer to the thread context
 860  * \param[in] lfsck     pointer to the lfsck instance
 861  *
 862  * \retval              positive number if all objects have been scanned
 863  * \retval              0 if the iteration is stopped or paused
 864  * \retval              negative error number on failure
 865  */
 866 static int lfsck_master_oit_engine(const struct lu_env *env,
 867                                    struct lfsck_instance *lfsck)
 868 {
 869         struct lfsck_thread_info *info  = lfsck_env_info(env);
 870         const struct dt_it_ops   *iops  =
 871                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 872         struct dt_it             *di    = lfsck->li_di_oit;
 873         struct lu_fid            *fid   = &info->lti_fid;
 874         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 875         struct ptlrpc_thread     *thread = &lfsck->li_thread;
 876         struct dt_device         *dev   = lfsck->li_bottom;
 877         struct seq_server_site   *ss    = lu_site2seq(dev->dd_lu_dev.ld_site);
 878         __u32                    idx    = lfsck_dev_idx(dev);
 879         int                      rc;
 880         ENTRY;
 881
 882         if (unlikely(ss == NULL))
 883                 RETURN(-EIO);
 884
 885         do {
 886                 struct dt_object *target;
 887                 bool              update_lma = false;
 888
 889                 if (lfsck->li_di_dir != NULL) {
 890                         rc = lfsck_master_dir_engine(env, lfsck);
 891                         if (rc <= 0)
 892                                 RETURN(rc);
 893                 }
 894
 895                 if (unlikely(lfsck->li_oit_over))
 896                         RETURN(1);
 897
 898                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY1) &&
 899                     cfs_fail_val > 0) {
 900                         struct l_wait_info lwi;
 901
 902                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
 903                                           NULL, NULL);
 904                         l_wait_event(thread->t_ctl_waitq,
 905                                      !thread_is_running(thread),
 906                                      &lwi);
 907
 908                         if (unlikely(!thread_is_running(thread))) {
 909                                 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine "
 910                                        "stop, cookie "LPU64"\n",
 911                                        lfsck_lfsck2name(lfsck),
 912                                        iops->store(env, di));
 913                                 RETURN(0);
 914                         }
 915                 }
 916
 917                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
 918                         RETURN(0);
 919
 920                 lfsck->li_current_oit_processed = 1;
 921
 922                 if (!list_empty(&lfsck->li_list_lmv)) {
 923                         struct lfsck_lmv_unit *llu;
 924
 925                         spin_lock(&lfsck->li_lock);
 926                         llu = list_entry(lfsck->li_list_lmv.next,
 927                                          struct lfsck_lmv_unit, llu_link);
 928                         list_del_init(&llu->llu_link);
 929                         spin_unlock(&lfsck->li_lock);
 930
 931                         lfsck->li_lmv = &llu->llu_lmv;
 932                         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 933                         rc = lfsck_open_dir(env, lfsck, 0);
 934                         if (rc == 0)
 935                                 rc = lfsck_master_dir_engine(env, lfsck);
 936
 937                         if (rc <= 0)
 938                                 RETURN(rc);
 939                 }
 940
 941                 lfsck->li_new_scanned++;
 942                 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
 943                 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
 944                 if (rc != 0) {
 945                         CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
 946                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 947                         lfsck_fail(env, lfsck, true);
 948                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
 949                                 RETURN(rc);
 950                         else
 951                                 goto checkpoint;
 952                 }
 953
 954                 if (fid_is_idif(fid)) {
 955                         __u32 idx1 = fid_idif_ost_idx(fid);
 956
 957                         LASSERT(!lfsck->li_master);
 958
 959                         /* It is an old format device, update the LMA. */
 960                         if (idx != idx1) {
 961                                 struct ost_id *oi = &info->lti_oi;
 962
 963                                 fid_to_ostid(fid, oi);
 964                                 ostid_to_fid(fid, oi, idx);
 965                                 update_lma = true;
 966                         }
 967                 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
 968                            !fid_is_last_id(fid) &&
 969                            !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
 970
 971                         /* If the FID/object is only used locally and invisible
 972                          * to external nodes, then LFSCK will not handle it.
 973                          *
 974                          * dot_lustre sequence has been handled specially. */
 975                         goto checkpoint;
 976                 } else {
 977                         struct lu_seq_range *range = &info->lti_range;
 978
 979                         if (lfsck->li_master)
 980                                 fld_range_set_mdt(range);
 981                         else
 982                                 fld_range_set_ost(range);
 983                         rc = fld_local_lookup(env, ss->ss_server_fld,
 984                                               fid_seq(fid), range);
 985                         if (rc != 0 || range->lsr_index != idx) {
 986                                 /* Remote object will be handled by the LFSCK
 987                                  * instance on the MDT where the remote object
 988                                  * really resides on. */
 989                                 rc = 0;
 990                                 goto checkpoint;
 991                         }
 992                 }
 993
 994                 target = lfsck_object_find(env, lfsck, fid);
 995                 if (IS_ERR(target)) {
 996                         CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
 997                                DFID", cookie "LPU64": rc = %d\n",
 998                                lfsck_lfsck2name(lfsck), PFID(fid),
 999                                iops->store(env, di), rc);
1000                         lfsck_fail(env, lfsck, true);
1001                         if (bk->lb_param & LPF_FAILOUT)
1002                                 RETURN(PTR_ERR(target));
1003                         else
1004                                 goto checkpoint;
1005                 }
1006
1007                 if (dt_object_exists(target)) {
1008                         if (update_lma) {
1009                                 rc = lfsck_update_lma(env, lfsck, target);
1010                                 if (rc != 0)
1011                                         CDEBUG(D_LFSCK, "%s: fail to update "
1012                                                "LMA for "DFID": rc = %d\n",
1013                                                lfsck_lfsck2name(lfsck),
1014                                                PFID(lfsck_dto2fid(target)), rc);
1015                         }
1016                         if (rc == 0)
1017                                 rc = lfsck_exec_oit(env, lfsck, target);
1018                 }
1019                 lfsck_object_put(env, target);
1020                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1021                         RETURN(rc);
1022
1023 checkpoint:
1024                 rc = lfsck_checkpoint(env, lfsck);
1025                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1026                         RETURN(rc);
1027
1028                 /* Rate control. */
1029                 lfsck_control_speed(lfsck);
1030
1031                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1032                         spin_lock(&lfsck->li_lock);
1033                         thread_set_flags(thread, SVC_STOPPING);
1034                         spin_unlock(&lfsck->li_lock);
1035                         RETURN(-EINVAL);
1036                 }
1037
1038                 rc = iops->next(env, di);
1039                 if (unlikely(rc > 0))
1040                         lfsck->li_oit_over = 1;
1041                 else if (likely(rc == 0))
1042                         lfsck->li_current_oit_processed = 0;
1043
1044                 if (unlikely(!thread_is_running(thread))) {
1045                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1046                                "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1047                                iops->store(env, di));
1048                         RETURN(0);
1049                 }
1050         } while (rc == 0 || lfsck->li_di_dir != NULL);
1051
1052         RETURN(rc);
1053 }
1054
1055 int lfsck_master_engine(void *args)
1056 {
1057         struct lfsck_thread_args *lta      = args;
1058         struct lu_env            *env      = &lta->lta_env;
1059         struct lfsck_instance    *lfsck    = lta->lta_lfsck;
1060         struct ptlrpc_thread     *thread   = &lfsck->li_thread;
1061         struct dt_object         *oit_obj  = lfsck->li_obj_oit;
1062         const struct dt_it_ops   *oit_iops = &oit_obj->do_index_ops->dio_it;
1063         struct dt_it             *oit_di;
1064         struct l_wait_info        lwi      = { 0 };
1065         int                       rc;
1066         ENTRY;
1067
1068         if (lfsck->li_master &&
1069             (!list_empty(&lfsck->li_list_scan) ||
1070              !list_empty(&lfsck->li_list_double_scan))) {
1071                 rc = lfsck_verify_lpf(env, lfsck);
1072                 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1073                  * fatal, because the .lustre/lost+found/ maybe not accessed
1074                  * by the LFSCK if it does not add orphans or others to such
1075                  * directory. So go ahead until hit failure when really uses
1076                  * the directory. */
1077                 if (rc != 0)
1078                         CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1079                                ".lustre/lost+found/, go ahead: rc = %d\n",
1080                                lfsck_lfsck2name(lfsck), rc);
1081         }
1082
1083         oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit, BYPASS_CAPA);
1084         if (IS_ERR(oit_di)) {
1085                 rc = PTR_ERR(oit_di);
1086                 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1087                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1088
1089                 GOTO(fini_args, rc);
1090         }
1091
1092         spin_lock(&lfsck->li_lock);
1093         lfsck->li_di_oit = oit_di;
1094         spin_unlock(&lfsck->li_lock);
1095         rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1096         if (rc != 0)
1097                 GOTO(fini_oit, rc);
1098
1099         CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1100                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1101                ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1102                lfsck->li_pos_checkpoint.lp_oit_cookie,
1103                lfsck->li_pos_checkpoint.lp_dir_cookie,
1104                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1105                current_pid());
1106
1107         spin_lock(&lfsck->li_lock);
1108         thread_set_flags(thread, SVC_RUNNING);
1109         spin_unlock(&lfsck->li_lock);
1110         wake_up_all(&thread->t_ctl_waitq);
1111
1112         l_wait_event(thread->t_ctl_waitq,
1113                      lfsck->li_start_unplug ||
1114                      !thread_is_running(thread),
1115                      &lwi);
1116         if (!thread_is_running(thread))
1117                 GOTO(fini_oit, rc = 0);
1118
1119         if (!list_empty(&lfsck->li_list_scan) ||
1120             list_empty(&lfsck->li_list_double_scan))
1121                 rc = lfsck_master_oit_engine(env, lfsck);
1122         else
1123                 rc = 1;
1124
1125         CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1126                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1127                ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1128                lfsck->li_pos_checkpoint.lp_oit_cookie,
1129                lfsck->li_pos_checkpoint.lp_dir_cookie,
1130                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1131                current_pid(), rc);
1132
1133         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1134                 rc = lfsck_post(env, lfsck, rc);
1135         else
1136                 lfsck_close_dir(env, lfsck, rc);
1137
1138 fini_oit:
1139         lfsck_di_oit_put(env, lfsck);
1140         oit_iops->fini(env, oit_di);
1141         if (rc == 1) {
1142                 if (!list_empty(&lfsck->li_list_double_scan))
1143                         rc = lfsck_double_scan(env, lfsck);
1144                 else
1145                         rc = 0;
1146         } else {
1147                 lfsck_quit(env, lfsck);
1148         }
1149
1150         /* XXX: Purge the pinned objects in the future. */
1151
1152 fini_args:
1153         spin_lock(&lfsck->li_lock);
1154         thread_set_flags(thread, SVC_STOPPED);
1155         spin_unlock(&lfsck->li_lock);
1156         wake_up_all(&thread->t_ctl_waitq);
1157         lfsck_thread_args_fini(lta);
1158         return rc;
1159 }
1160
1161 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1162 {
1163         bool empty = false;
1164
1165         spin_lock(&lad->lad_lock);
1166         if (list_empty(&lad->lad_req_list))
1167                 empty = true;
1168         spin_unlock(&lad->lad_lock);
1169
1170         return empty;
1171 }
1172
1173 /**
1174  * Query the LFSCK status from the instatnces on remote servers.
1175  *
1176  * The LFSCK assistant thread queries the LFSCK instances on other
1177  * servers (MDT/OST) about their status, such as whether they have
1178  * finished the phase1/phase2 scanning or not, and so on.
1179  *
1180  * \param[in] env       pointer to the thread context
1181  * \param[in] com       pointer to the lfsck component
1182  *
1183  * \retval              0 for success
1184  * \retval              negative error number on failure
1185  */
1186 static int lfsck_assistant_query_others(const struct lu_env *env,
1187                                         struct lfsck_component *com)
1188 {
1189         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1190         struct lfsck_request              *lr    = &info->lti_lr;
1191         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1192         struct lfsck_instance             *lfsck = com->lc_lfsck;
1193         struct lfsck_assistant_data       *lad   = com->lc_data;
1194         struct ptlrpc_request_set         *set;
1195         struct lfsck_tgt_descs            *ltds;
1196         struct lfsck_tgt_desc             *ltd;
1197         struct list_head                  *phase_head;
1198         int                                rc    = 0;
1199         int                                rc1   = 0;
1200         ENTRY;
1201
1202         set = ptlrpc_prep_set();
1203         if (set == NULL)
1204                 RETURN(-ENOMEM);
1205
1206         lad->lad_touch_gen++;
1207         memset(lr, 0, sizeof(*lr));
1208         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1209         lr->lr_event = LE_QUERY;
1210         lr->lr_active = com->lc_type;
1211         laia->laia_com = com;
1212         laia->laia_lr = lr;
1213         laia->laia_shared = 0;
1214
1215         if (!list_empty(&lad->lad_mdt_phase1_list)) {
1216                 ltds = &lfsck->li_mdt_descs;
1217                 lr->lr_flags = 0;
1218                 phase_head = &lad->lad_mdt_phase1_list;
1219         } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1220                 goto out;
1221         } else {
1222
1223 again:
1224                 ltds = &lfsck->li_ost_descs;
1225                 lr->lr_flags = LEF_TO_OST;
1226                 phase_head = &lad->lad_ost_phase1_list;
1227         }
1228
1229         laia->laia_ltds = ltds;
1230         spin_lock(&ltds->ltd_lock);
1231         while (!list_empty(phase_head)) {
1232                 struct list_head *phase_list;
1233                 __u32            *gen;
1234
1235                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1236                         ltd = list_entry(phase_head->next,
1237                                          struct lfsck_tgt_desc,
1238                                          ltd_layout_phase_list);
1239                         phase_list = &ltd->ltd_layout_phase_list;
1240                         gen = &ltd->ltd_layout_gen;
1241                 } else {
1242                         ltd = list_entry(phase_head->next,
1243                                          struct lfsck_tgt_desc,
1244                                          ltd_namespace_phase_list);
1245                         phase_list = &ltd->ltd_namespace_phase_list;
1246                         gen = &ltd->ltd_namespace_gen;
1247                 }
1248
1249                 if (*gen == lad->lad_touch_gen)
1250                         break;
1251
1252                 *gen = lad->lad_touch_gen;
1253                 list_move_tail(phase_list, phase_head);
1254                 atomic_inc(&ltd->ltd_ref);
1255                 laia->laia_ltd = ltd;
1256                 spin_unlock(&ltds->ltd_lock);
1257                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1258                                          lfsck_async_interpret_common,
1259                                          laia, LFSCK_QUERY);
1260                 if (rc != 0) {
1261                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1262                                "%s %x for %s: rc = %d\n",
1263                                lfsck_lfsck2name(lfsck),
1264                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1265                                ltd->ltd_index, lad->lad_name, rc);
1266                         lfsck_tgt_put(ltd);
1267                         rc1 = rc;
1268                 }
1269                 spin_lock(&ltds->ltd_lock);
1270         }
1271         spin_unlock(&ltds->ltd_lock);
1272
1273         rc = ptlrpc_set_wait(set);
1274         if (rc < 0) {
1275                 ptlrpc_set_destroy(set);
1276                 RETURN(rc);
1277         }
1278
1279         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1280             list_empty(&lad->lad_mdt_phase1_list))
1281                 goto again;
1282
1283 out:
1284         ptlrpc_set_destroy(set);
1285
1286         RETURN(rc1 != 0 ? rc1 : rc);
1287 }
1288
1289 /**
1290  * Notify the LFSCK event to the instatnces on remote servers.
1291  *
1292  * The LFSCK assistant thread notifies the LFSCK instances on other
1293  * servers (MDT/OST) about some events, such as start new scanning,
1294  * stop the scanning, this LFSCK instance will exit, and so on.
1295  *
1296  * \param[in] env       pointer to the thread context
1297  * \param[in] com       pointer to the lfsck component
1298  * \param[in] lr        pointer to the LFSCK event request
1299  *
1300  * \retval              0 for success
1301  * \retval              negative error number on failure
1302  */
1303 static int lfsck_assistant_notify_others(const struct lu_env *env,
1304                                          struct lfsck_component *com,
1305                                          struct lfsck_request *lr)
1306 {
1307         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1308         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1309         struct lfsck_instance             *lfsck = com->lc_lfsck;
1310         struct lfsck_assistant_data       *lad   = com->lc_data;
1311         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1312         struct ptlrpc_request_set         *set;
1313         struct lfsck_tgt_descs            *ltds;
1314         struct lfsck_tgt_desc             *ltd;
1315         struct lfsck_tgt_desc             *next;
1316         __u32                              idx;
1317         int                                rc    = 0;
1318         int                                rc1   = 0;
1319         ENTRY;
1320
1321         set = ptlrpc_prep_set();
1322         if (set == NULL)
1323                 RETURN(-ENOMEM);
1324
1325         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1326         lr->lr_active = com->lc_type;
1327         laia->laia_com = com;
1328         laia->laia_lr = lr;
1329         laia->laia_shared = 0;
1330
1331         switch (lr->lr_event) {
1332         case LE_START:
1333                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1334                         goto next;
1335
1336                 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
1337                                LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
1338                 lr->lr_speed = bk->lb_speed_limit;
1339                 lr->lr_version = bk->lb_version;
1340                 lr->lr_param |= bk->lb_param;
1341                 lr->lr_async_windows = bk->lb_async_windows;
1342                 lr->lr_flags = LEF_TO_OST;
1343
1344                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1345                 ltds = &lfsck->li_ost_descs;
1346                 laia->laia_ltds = ltds;
1347                 down_read(&ltds->ltd_rw_sem);
1348                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1349                         ltd = lfsck_tgt_get(ltds, idx);
1350                         LASSERT(ltd != NULL);
1351
1352                         laia->laia_ltd = ltd;
1353                         ltd->ltd_layout_done = 0;
1354                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1355                                         lfsck_async_interpret_common,
1356                                         laia, LFSCK_NOTIFY);
1357                         if (rc != 0) {
1358                                 lfsck_lad_set_bitmap(env, com, idx);
1359                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1360                                        "notify OST %x for %s start: rc = %d\n",
1361                                        lfsck_lfsck2name(lfsck), idx,
1362                                        lad->lad_name, rc);
1363                                 lfsck_tgt_put(ltd);
1364                         }
1365                 }
1366                 up_read(&ltds->ltd_rw_sem);
1367
1368                 /* Sync up */
1369                 rc = ptlrpc_set_wait(set);
1370                 if (rc < 0) {
1371                         ptlrpc_set_destroy(set);
1372                         RETURN(rc);
1373                 }
1374
1375 next:
1376                 if (!(bk->lb_param & LPF_ALL_TGT))
1377                         break;
1378
1379                 /* link other MDT targets locallly. */
1380                 ltds = &lfsck->li_mdt_descs;
1381                 spin_lock(&ltds->ltd_lock);
1382                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1383                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1384                                 ltd = LTD_TGT(ltds, idx);
1385                                 LASSERT(ltd != NULL);
1386
1387                                 if (!list_empty(&ltd->ltd_layout_list))
1388                                         continue;
1389
1390                                 list_add_tail(&ltd->ltd_layout_list,
1391                                               &lad->lad_mdt_list);
1392                                 list_add_tail(&ltd->ltd_layout_phase_list,
1393                                               &lad->lad_mdt_phase1_list);
1394                         }
1395                 } else {
1396                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1397                                 ltd = LTD_TGT(ltds, idx);
1398                                 LASSERT(ltd != NULL);
1399
1400                                 if (!list_empty(&ltd->ltd_namespace_list))
1401                                         continue;
1402
1403                                 list_add_tail(&ltd->ltd_namespace_list,
1404                                               &lad->lad_mdt_list);
1405                                 list_add_tail(&ltd->ltd_namespace_phase_list,
1406                                               &lad->lad_mdt_phase1_list);
1407                         }
1408                 }
1409                 spin_unlock(&ltds->ltd_lock);
1410                 break;
1411         case LE_STOP:
1412         case LE_PHASE2_DONE:
1413         case LE_PEER_EXIT: {
1414                 struct list_head *phase_head;
1415
1416                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1417                 if (bk->lb_param & LPF_ALL_TGT) {
1418                         phase_head = &lad->lad_mdt_list;
1419                         ltds = &lfsck->li_mdt_descs;
1420                         if (lr->lr_event == LE_STOP) {
1421                                 /* unlink other MDT targets locallly. */
1422                                 spin_lock(&ltds->ltd_lock);
1423                                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1424                                         list_for_each_entry_safe(ltd, next,
1425                                                 phase_head, ltd_layout_list) {
1426                                                 list_del_init(
1427                                                 &ltd->ltd_layout_phase_list);
1428                                                 list_del_init(
1429                                                 &ltd->ltd_layout_list);
1430                                         }
1431                                 } else {
1432                                         list_for_each_entry_safe(ltd, next,
1433                                                         phase_head,
1434                                                         ltd_namespace_list) {
1435                                                 list_del_init(
1436                                                 &ltd->ltd_namespace_phase_list);
1437                                                 list_del_init(
1438                                                 &ltd->ltd_namespace_list);
1439                                         }
1440                                 }
1441                                 spin_unlock(&ltds->ltd_lock);
1442
1443                                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1444                                         break;
1445
1446                                 lr->lr_flags |= LEF_TO_OST;
1447                                 phase_head = &lad->lad_ost_list;
1448                                 ltds = &lfsck->li_ost_descs;
1449                         } else {
1450                                 lr->lr_flags &= ~LEF_TO_OST;
1451                         }
1452                 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1453                         break;
1454                 } else {
1455                         lr->lr_flags |= LEF_TO_OST;
1456                         phase_head = &lad->lad_ost_list;
1457                         ltds = &lfsck->li_ost_descs;
1458                 }
1459
1460 again:
1461                 laia->laia_ltds = ltds;
1462                 spin_lock(&ltds->ltd_lock);
1463                 while (!list_empty(phase_head)) {
1464                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1465                                 ltd = list_entry(phase_head->next,
1466                                                  struct lfsck_tgt_desc,
1467                                                  ltd_layout_list);
1468                                 if (!list_empty(&ltd->ltd_layout_phase_list))
1469                                         list_del_init(
1470                                                 &ltd->ltd_layout_phase_list);
1471                                 list_del_init(&ltd->ltd_layout_list);
1472                         } else {
1473                                 ltd = list_entry(phase_head->next,
1474                                                  struct lfsck_tgt_desc,
1475                                                  ltd_namespace_list);
1476                                 if (!list_empty(&ltd->ltd_namespace_phase_list))
1477                                         list_del_init(
1478                                                 &ltd->ltd_namespace_phase_list);
1479                                 list_del_init(&ltd->ltd_namespace_list);
1480                         }
1481                         atomic_inc(&ltd->ltd_ref);
1482                         laia->laia_ltd = ltd;
1483                         spin_unlock(&ltds->ltd_lock);
1484                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1485                                         lfsck_async_interpret_common,
1486                                         laia, LFSCK_NOTIFY);
1487                         if (rc != 0) {
1488                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1489                                        "notify %s %x for %s stop/phase2_done/"
1490                                        "peer_exit: rc = %d\n",
1491                                        lfsck_lfsck2name(lfsck),
1492                                        (lr->lr_flags & LEF_TO_OST) ?
1493                                        "OST" : "MDT", ltd->ltd_index,
1494                                        lad->lad_name, rc);
1495                                 lfsck_tgt_put(ltd);
1496                         }
1497                         spin_lock(&ltds->ltd_lock);
1498                 }
1499                 spin_unlock(&ltds->ltd_lock);
1500
1501                 rc = ptlrpc_set_wait(set);
1502                 if (rc < 0) {
1503                         ptlrpc_set_destroy(set);
1504                         RETURN(rc);
1505                 }
1506
1507                 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1508                     !(lr->lr_flags & LEF_TO_OST)) {
1509                         lr->lr_flags |= LEF_TO_OST;
1510                         phase_head = &lad->lad_ost_list;
1511                         ltds = &lfsck->li_ost_descs;
1512                         goto again;
1513                 }
1514                 break;
1515         }
1516         case LE_PHASE1_DONE:
1517                 lad->lad_ops->la_sync_failures(env, com, lr);
1518                 lad->lad_touch_gen++;
1519                 ltds = &lfsck->li_mdt_descs;
1520                 laia->laia_ltds = ltds;
1521                 spin_lock(&ltds->ltd_lock);
1522                 while (!list_empty(&lad->lad_mdt_list)) {
1523                         struct list_head *list;
1524                         __u32            *gen;
1525
1526                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1527                                 ltd = list_entry(lad->lad_mdt_list.next,
1528                                                  struct lfsck_tgt_desc,
1529                                                  ltd_layout_list);
1530                                 list = &ltd->ltd_layout_list;
1531                                 gen = &ltd->ltd_layout_gen;
1532                         } else {
1533                                 struct lfsck_namespace *ns = com->lc_file_ram;
1534
1535                                 ltd = list_entry(lad->lad_mdt_list.next,
1536                                                  struct lfsck_tgt_desc,
1537                                                  ltd_namespace_list);
1538                                 list = &ltd->ltd_namespace_list;
1539                                 gen = &ltd->ltd_namespace_gen;
1540                                 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1541                         }
1542
1543                         if (*gen == lad->lad_touch_gen)
1544                                 break;
1545
1546                         *gen = lad->lad_touch_gen;
1547                         list_move_tail(list, &lad->lad_mdt_list);
1548                         if (ltd->ltd_namespace_failed)
1549                                 continue;
1550
1551                         atomic_inc(&ltd->ltd_ref);
1552                         laia->laia_ltd = ltd;
1553                         spin_unlock(&ltds->ltd_lock);
1554                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1555                                         lfsck_async_interpret_common,
1556                                         laia, LFSCK_NOTIFY);
1557                         if (rc != 0) {
1558                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1559                                        "notify MDT %x for %s phase1 done: "
1560                                        "rc = %d\n", lfsck_lfsck2name(lfsck),
1561                                        ltd->ltd_index, lad->lad_name, rc);
1562                                 lfsck_tgt_put(ltd);
1563                         }
1564                         spin_lock(&ltds->ltd_lock);
1565                 }
1566                 spin_unlock(&ltds->ltd_lock);
1567                 break;
1568         default:
1569                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1570                        "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1571                 rc = -EINVAL;
1572                 break;
1573         }
1574
1575         rc1 = ptlrpc_set_wait(set);
1576         ptlrpc_set_destroy(set);
1577
1578         RETURN(rc != 0 ? rc : rc1);
1579 }
1580
1581 /**
1582  * The LFSCK assistant thread is triggered by the LFSCK main engine.
1583  * They co-work together as an asynchronous pipeline: the LFSCK main
1584  * engine scans the system and pre-fetches the objects, attributes,
1585  * or name entries, etc, and pushes them into the pipeline as input
1586  * requests for the LFSCK assistant thread; on the other end of the
1587  * pipeline, the LFSCK assistant thread performs the real check and
1588  * repair for every request from the main engine.
1589  *
1590  * Generally, the assistant engine may be blocked when check/repair
1591  * something, so the LFSCK main engine will run some faster. On the
1592  * other hand, the LFSCK main engine will drive multiple assistant
1593  * threads in parallel, means for each LFSCK component on the master
1594  * (such as layout LFSCK, namespace LFSCK), there is an independent
1595  * LFSCK assistant thread. So under such 1:N multiple asynchronous
1596  * pipelines mode, the whole LFSCK performance will be much better
1597  * than check/repair everything by the LFSCK main engine itself.
1598  */
1599 int lfsck_assistant_engine(void *args)
1600 {
1601         struct lfsck_thread_args          *lta     = args;
1602         struct lu_env                     *env     = &lta->lta_env;
1603         struct lfsck_component            *com     = lta->lta_com;
1604         struct lfsck_instance             *lfsck   = lta->lta_lfsck;
1605         struct lfsck_bookmark             *bk      = &lfsck->li_bookmark_ram;
1606         struct lfsck_position             *pos     = &com->lc_pos_start;
1607         struct lfsck_thread_info          *info    = lfsck_env_info(env);
1608         struct lfsck_request              *lr      = &info->lti_lr;
1609         struct lfsck_assistant_data       *lad     = com->lc_data;
1610         struct ptlrpc_thread              *mthread = &lfsck->li_thread;
1611         struct ptlrpc_thread              *athread = &lad->lad_thread;
1612         struct lfsck_assistant_operations *lao     = lad->lad_ops;
1613         struct lfsck_assistant_req        *lar;
1614         struct l_wait_info                 lwi     = { 0 };
1615         int                                rc      = 0;
1616         int                                rc1     = 0;
1617         ENTRY;
1618
1619         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1620                lfsck_lfsck2name(lfsck), lad->lad_name);
1621
1622         memset(lr, 0, sizeof(*lr));
1623         lr->lr_event = LE_START;
1624         if (pos->lp_oit_cookie <= 1)
1625                 lr->lr_param = LPF_RESET;
1626         rc = lfsck_assistant_notify_others(env, com, lr);
1627         if (rc != 0) {
1628                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1629                        "to start %s: rc = %d\n",
1630                        lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1631                 GOTO(fini, rc);
1632         }
1633
1634         spin_lock(&lad->lad_lock);
1635         thread_set_flags(athread, SVC_RUNNING);
1636         spin_unlock(&lad->lad_lock);
1637         wake_up_all(&mthread->t_ctl_waitq);
1638
1639         while (1) {
1640                 while (!list_empty(&lad->lad_req_list)) {
1641                         bool wakeup = false;
1642
1643                         if (unlikely(lad->lad_exit ||
1644                                      !thread_is_running(mthread)))
1645                                 GOTO(cleanup1, rc = lad->lad_post_result);
1646
1647                         lar = list_entry(lad->lad_req_list.next,
1648                                          struct lfsck_assistant_req,
1649                                          lar_list);
1650                         /* Only the lfsck_assistant_engine thread itself can
1651                          * remove the "lar" from the head of the list, LFSCK
1652                          * engine thread only inserts other new "lar" at the
1653                          * end of the list. So it is safe to handle current
1654                          * "lar" without the spin_lock. */
1655                         rc = lao->la_handler_p1(env, com, lar);
1656                         spin_lock(&lad->lad_lock);
1657                         list_del_init(&lar->lar_list);
1658                         lad->lad_prefetched--;
1659                         /* Wake up the main engine thread only when the list
1660                          * is empty or half of the prefetched items have been
1661                          * handled to avoid too frequent thread schedule. */
1662                         if (lad->lad_prefetched == 0 ||
1663                             (bk->lb_async_windows != 0 &&
1664                              bk->lb_async_windows / 2 ==
1665                              lad->lad_prefetched))
1666                                 wakeup = true;
1667                         spin_unlock(&lad->lad_lock);
1668                         if (wakeup)
1669                                 wake_up_all(&mthread->t_ctl_waitq);
1670
1671                         lao->la_req_fini(env, lar);
1672                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1673                                 GOTO(cleanup1, rc);
1674                 }
1675
1676                 l_wait_event(athread->t_ctl_waitq,
1677                              !lfsck_assistant_req_empty(lad) ||
1678                              lad->lad_exit ||
1679                              lad->lad_to_post ||
1680                              lad->lad_to_double_scan,
1681                              &lwi);
1682
1683                 if (unlikely(lad->lad_exit))
1684                         GOTO(cleanup1, rc = lad->lad_post_result);
1685
1686                 if (!list_empty(&lad->lad_req_list))
1687                         continue;
1688
1689                 if (lad->lad_to_post) {
1690                         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1691                                lfsck_lfsck2name(lfsck), lad->lad_name);
1692
1693                         if (unlikely(lad->lad_exit))
1694                                 GOTO(cleanup1, rc = lad->lad_post_result);
1695
1696                         lad->lad_to_post = 0;
1697                         LASSERT(lad->lad_post_result > 0);
1698
1699                         memset(lr, 0, sizeof(*lr));
1700                         lr->lr_event = LE_PHASE1_DONE;
1701                         lr->lr_status = lad->lad_post_result;
1702                         rc = lfsck_assistant_notify_others(env, com, lr);
1703                         if (rc != 0)
1704                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to "
1705                                        "notify others for %s post: rc = %d\n",
1706                                        lfsck_lfsck2name(lfsck),
1707                                        lad->lad_name, rc);
1708
1709                         /* Wakeup the master engine to go ahead. */
1710                         wake_up_all(&mthread->t_ctl_waitq);
1711                 }
1712
1713                 if (lad->lad_to_double_scan) {
1714                         lad->lad_to_double_scan = 0;
1715                         atomic_inc(&lfsck->li_double_scan_count);
1716                         lad->lad_in_double_scan = 1;
1717                         wake_up_all(&mthread->t_ctl_waitq);
1718
1719                         com->lc_new_checked = 0;
1720                         com->lc_new_scanned = 0;
1721                         com->lc_time_last_checkpoint = cfs_time_current();
1722                         com->lc_time_next_checkpoint =
1723                                 com->lc_time_last_checkpoint +
1724                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1725
1726                         /* Flush async updates before handling orphan. */
1727                         dt_sync(env, lfsck->li_next);
1728
1729                         CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1730                                "scan start\n", lfsck_lfsck2name(lfsck));
1731
1732                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1733                                 GOTO(cleanup2, rc = 0);
1734
1735                         while (lad->lad_in_double_scan) {
1736                                 rc = lfsck_assistant_query_others(env, com);
1737                                 if (lfsck_phase2_next_ready(lad))
1738                                         goto p2_next;
1739
1740                                 if (rc < 0)
1741                                         GOTO(cleanup2, rc);
1742
1743                                 /* Pull LFSCK status on related targets once
1744                                  * per 30 seconds if we are not notified. */
1745                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1746                                                            cfs_time_seconds(1),
1747                                                            NULL, NULL);
1748                                 rc = l_wait_event(athread->t_ctl_waitq,
1749                                         lfsck_phase2_next_ready(lad) ||
1750                                         lad->lad_exit ||
1751                                         !thread_is_running(mthread),
1752                                         &lwi);
1753
1754                                 if (unlikely(lad->lad_exit ||
1755                                              !thread_is_running(mthread)))
1756                                         GOTO(cleanup2, rc = 0);
1757
1758                                 if (rc == -ETIMEDOUT)
1759                                         continue;
1760
1761                                 if (rc < 0)
1762                                         GOTO(cleanup2, rc);
1763
1764 p2_next:
1765                                 rc = lao->la_handler_p2(env, com);
1766                                 if (rc != 0)
1767                                         GOTO(cleanup2, rc);
1768
1769                                 if (unlikely(lad->lad_exit ||
1770                                              !thread_is_running(mthread)))
1771                                         GOTO(cleanup2, rc = 0);
1772                         }
1773                 }
1774         }
1775
1776 cleanup1:
1777         /* Cleanup the unfinished requests. */
1778         spin_lock(&lad->lad_lock);
1779         if (rc < 0)
1780                 lad->lad_assistant_status = rc;
1781
1782         if (lad->lad_exit && lad->lad_post_result <= 0)
1783                 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1784
1785         while (!list_empty(&lad->lad_req_list)) {
1786                 lar = list_entry(lad->lad_req_list.next,
1787                                  struct lfsck_assistant_req,
1788                                  lar_list);
1789                 list_del_init(&lar->lar_list);
1790                 lad->lad_prefetched--;
1791                 spin_unlock(&lad->lad_lock);
1792                 lao->la_req_fini(env, lar);
1793                 spin_lock(&lad->lad_lock);
1794         }
1795         spin_unlock(&lad->lad_lock);
1796
1797         LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1798                  lad->lad_prefetched);
1799
1800 cleanup2:
1801         memset(lr, 0, sizeof(*lr));
1802         if (rc > 0) {
1803                 lr->lr_event = LE_PHASE2_DONE;
1804                 lr->lr_status = rc;
1805         } else if (rc == 0) {
1806                 if (lfsck->li_flags & LPF_ALL_TGT) {
1807                         lr->lr_event = LE_STOP;
1808                         lr->lr_status = LS_STOPPED;
1809                 } else {
1810                         lr->lr_event = LE_PEER_EXIT;
1811                         switch (lfsck->li_status) {
1812                         case LS_PAUSED:
1813                         case LS_CO_PAUSED:
1814                                 lr->lr_status = LS_CO_PAUSED;
1815                                 break;
1816                         case LS_STOPPED:
1817                         case LS_CO_STOPPED:
1818                                 lr->lr_status = LS_CO_STOPPED;
1819                                 break;
1820                         default:
1821                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1822                                        "status: rc = %d\n",
1823                                        lfsck_lfsck2name(lfsck),
1824                                        lfsck->li_status);
1825                                 lr->lr_status = LS_CO_FAILED;
1826                                 break;
1827                         }
1828                 }
1829         } else {
1830                 if (lfsck->li_flags & LPF_ALL_TGT) {
1831                         lr->lr_event = LE_STOP;
1832                         lr->lr_status = LS_FAILED;
1833                 } else {
1834                         lr->lr_event = LE_PEER_EXIT;
1835                         lr->lr_status = LS_CO_FAILED;
1836                 }
1837         }
1838
1839         rc1 = lfsck_assistant_notify_others(env, com, lr);
1840         if (rc1 != 0) {
1841                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1842                        "others for %s quit: rc = %d\n",
1843                        lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1844                 rc = rc1;
1845         }
1846
1847         /* Flush async updates before exit. */
1848         dt_sync(env, lfsck->li_next);
1849
1850         /* Under force exit case, some requests may be just freed without
1851          * verification, those objects should be re-handled when next run.
1852          * So not update the on-disk trace file under such case. */
1853         if (lad->lad_in_double_scan) {
1854                 if (!lad->lad_exit)
1855                         rc1 = lao->la_double_scan_result(env, com, rc);
1856
1857                 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1858                        "finished: rc = %d\n",
1859                        lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1860         }
1861
1862 fini:
1863         if (lad->lad_in_double_scan)
1864                 atomic_dec(&lfsck->li_double_scan_count);
1865
1866         spin_lock(&lad->lad_lock);
1867         lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1868         thread_set_flags(athread, SVC_STOPPED);
1869         wake_up_all(&mthread->t_ctl_waitq);
1870         spin_unlock(&lad->lad_lock);
1871
1872         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1873                lfsck_lfsck2name(lfsck), lad->lad_name,
1874                lad->lad_assistant_status);
1875
1876         lfsck_thread_args_fini(lta);
1877
1878         return rc;
1879 }