lustre/lfsck/lfsck_engine.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_engine.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lu_object.h>
  34 #include <dt_object.h>
  35 #include <lustre_net.h>
  36 #include <lustre_fid.h>
  37 #include <obd_support.h>
  38 #include <lustre_lib.h>
  39
  40 #include "lfsck_internal.h"
  41
  42 int lfsck_unpack_ent(struct lu_dirent *ent, __u64 *cookie, __u16 *type)
  43 {
  44         struct luda_type        *lt;
  45         int                      align = sizeof(*lt) - 1;
  46         int                      len;
  47
  48         fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
  49         *cookie = le64_to_cpu(ent->lde_hash);
  50         ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
  51         ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
  52         ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
  53
  54         if (unlikely(!(ent->lde_attrs & LUDA_TYPE)))
  55                 return -EINVAL;
  56
  57         len = (ent->lde_namelen + align) & ~align;
  58         lt = (struct luda_type *)(ent->lde_name + len);
  59         *type = le16_to_cpu(lt->lt_type);
  60
  61         /* Make sure the name is terminated with '\0'. The data (object type)
  62          * after ent::lde_name maybe broken, but we have stored such data in
  63          * the output parameter @type as above. */
  64         ent->lde_name[ent->lde_namelen] = '\0';
  65
  66         return 0;
  67 }
  68
  69 static void lfsck_di_oit_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  70 {
  71         const struct dt_it_ops  *iops;
  72         struct dt_it            *di;
  73
  74         spin_lock(&lfsck->li_lock);
  75         iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
  76         di = lfsck->li_di_oit;
  77         lfsck->li_di_oit = NULL;
  78         spin_unlock(&lfsck->li_lock);
  79         iops->put(env, di);
  80 }
  81
  82 static void lfsck_di_dir_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  83 {
  84         const struct dt_it_ops  *iops;
  85         struct dt_it            *di;
  86
  87         spin_lock(&lfsck->li_lock);
  88         iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
  89         di = lfsck->li_di_dir;
  90         lfsck->li_di_dir = NULL;
  91         lfsck->li_cookie_dir = 0;
  92         spin_unlock(&lfsck->li_lock);
  93         iops->put(env, di);
  94 }
  95
  96 static int lfsck_update_lma(const struct lu_env *env,
  97                             struct lfsck_instance *lfsck, struct dt_object *obj)
  98 {
  99         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 100         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 101         struct dt_device                *dt     = lfsck->li_bottom;
 102         struct lustre_mdt_attrs         *lma    = &info->lti_lma;
 103         struct lu_buf                   *buf;
 104         struct thandle                  *th;
 105         int                              fl;
 106         int                              rc;
 107         ENTRY;
 108
 109         if (bk->lb_param & LPF_DRYRUN)
 110                 RETURN(0);
 111
 112         buf = lfsck_buf_get(env, info->lti_lma_old, LMA_OLD_SIZE);
 113         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LMA, BYPASS_CAPA);
 114         if (rc < 0) {
 115                 if (rc != -ENODATA)
 116                         RETURN(rc);
 117
 118                 fl = LU_XATTR_CREATE;
 119                 lustre_lma_init(lma, lfsck_dto2fid(obj), LMAC_FID_ON_OST, 0);
 120         } else {
 121                 if (rc != LMA_OLD_SIZE && rc != sizeof(struct lustre_mdt_attrs))
 122                         RETURN(-EINVAL);
 123
 124                 fl = LU_XATTR_REPLACE;
 125                 lustre_lma_swab(lma);
 126                 lustre_lma_init(lma, lfsck_dto2fid(obj),
 127                                 lma->lma_compat | LMAC_FID_ON_OST,
 128                                 lma->lma_incompat);
 129         }
 130         lustre_lma_swab(lma);
 131
 132         th = dt_trans_create(env, dt);
 133         if (IS_ERR(th))
 134                 RETURN(PTR_ERR(th));
 135
 136         buf = lfsck_buf_get(env, lma, sizeof(*lma));
 137         rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
 138         if (rc != 0)
 139                 GOTO(stop, rc);
 140
 141         rc = dt_trans_start(env, dt, th);
 142         if (rc != 0)
 143                 GOTO(stop, rc);
 144
 145         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th, BYPASS_CAPA);
 146
 147         GOTO(stop, rc);
 148
 149 stop:
 150         dt_trans_stop(env, dt, th);
 151         return rc;
 152 }
 153
 154 static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj,
 155                             struct lu_fid *fid)
 156 {
 157         if (unlikely(!S_ISDIR(lfsck_object_type(obj)) ||
 158                      !dt_try_as_dir(env, obj)))
 159                 return -ENOTDIR;
 160
 161         return dt_lookup(env, obj, (struct dt_rec *)fid,
 162                          (const struct dt_key *)"..", BYPASS_CAPA);
 163 }
 164
 165 /**
 166  * Check whether needs to scan the directory or not.
 167  *
 168  * 1) If we are not doing namespace LFSCK, or the given @obj is not directory,
 169  *    then needs not to scan the @obj. Otherwise,
 170  * 2) Global /ROOT needs to be scanned, backend root needs not to be scanned.
 171  * 3) If the @obj is neither IGIF nor normal FID (including .lustre and its
 172  *    sub-directories that have been scanned when the LFSCK engine start),
 173  *    then needs not to be scanned.
 174  * 4) If it is a remote object, then scanning the object will be done on the
 175  *    MDT on which the object really resides.
 176  * 5) If the local object has normal FID, then needs to be scanned. Otherwise,
 177  * 6) If the object has linkEA, then needs to be scanned. Otherwise,
 178  * 7) If none of the previous conditions are true, we need to check the parent
 179  *    directories whether this subdirectory is in a tree that should be scanned.
 180  *    Set the parent as current @obj, repeat 2)-7).
 181  *
 182  * \param[in] env       pointer to the thread context
 183  * \param[in] lfsck     pointer to the lfsck instance
 184  * \param[in] obj       pointer to the object to be checked
 185  *
 186  * \retval              positive number if the directory needs to be scanned
 187  * \retval              0 if the directory needs NOT to be scanned
 188  * \retval              negative error number on failure
 189  */
 190 static int lfsck_needs_scan_dir(const struct lu_env *env,
 191                                 struct lfsck_instance *lfsck,
 192                                 struct dt_object *obj)
 193 {
 194         struct lfsck_thread_info *info    = lfsck_env_info(env);
 195         struct lu_fid            *fid     = &info->lti_fid;
 196         struct lu_seq_range      *range   = &info->lti_range;
 197         struct dt_device         *dev     = lfsck->li_bottom;
 198         struct seq_server_site   *ss      = lu_site2seq(dev->dd_lu_dev.ld_site);
 199         __u32                     idx     = lfsck_dev_idx(dev);
 200         int                       depth   = 0;
 201         int                       rc      = 0;
 202
 203         if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
 204                 return 0;
 205
 206         LASSERT(ss != NULL);
 207
 208         *fid = *lfsck_dto2fid(obj);
 209         while (1) {
 210                 /* Global /ROOT is visible. */
 211                 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
 212                         return 1;
 213
 214                 /* Backend root is invisible. */
 215                 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
 216                         return 0;
 217
 218                 if (!fid_is_norm(fid) && !fid_is_igif(fid))
 219                         return 0;
 220
 221                 fld_range_set_mdt(range);
 222                 rc = fld_local_lookup(env, ss->ss_server_fld,
 223                                       fid_seq(fid), range);
 224                 if (rc != 0 || range->lsr_index != idx)
 225                         /* Current FID should NOT be for the input parameter
 226                          * @obj, because the lfsck_master_oit_engine() has
 227                          * filtered out agent object. So current FID is for
 228                          * the ancestor of the original input parameter @obj.
 229                          * So the ancestor is a remote directory. The input
 230                          * parameter @obj is local directory, and should be
 231                          * scanned under such case. */
 232                         return 1;
 233
 234                 /* normal FID on this target (locally) must be for the
 235                  * client-side visiable object. */
 236                 if (fid_is_norm(fid))
 237                         return 1;
 238
 239                 if (obj == NULL) {
 240                         obj = lfsck_object_find(env, lfsck, fid);
 241                         if (IS_ERR(obj))
 242                                 return PTR_ERR(obj);
 243
 244                         depth++;
 245                         if (!dt_object_exists(obj))
 246                                 GOTO(out, rc = 0);
 247                 }
 248
 249                 dt_read_lock(env, obj, MOR_TGT_CHILD);
 250                 if (unlikely(lfsck_is_dead_obj(obj))) {
 251                         dt_read_unlock(env, obj);
 252
 253                         GOTO(out, rc = 0);
 254                 }
 255
 256                 rc = dt_xattr_get(env, obj,
 257                                   lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK,
 258                                   BYPASS_CAPA);
 259                 dt_read_unlock(env, obj);
 260                 if (rc >= 0)
 261                         GOTO(out, rc = 1);
 262
 263                 if (rc < 0 && rc != -ENODATA)
 264                         GOTO(out, rc);
 265
 266                 rc = lfsck_parent_fid(env, obj, fid);
 267                 if (depth > 0)
 268                         lfsck_object_put(env, obj);
 269
 270                 obj = NULL;
 271                 if (rc != 0)
 272                         return rc;
 273
 274                 if (!fid_is_sane(fid))
 275                         return 0;
 276         }
 277
 278 out:
 279         if (depth > 0 && obj != NULL)
 280                 lfsck_object_put(env, obj);
 281
 282         return rc;
 283 }
 284
 285 static int lfsck_load_stripe_lmv(const struct lu_env *env,
 286                                  struct lfsck_instance *lfsck,
 287                                  struct dt_object *obj)
 288 {
 289         struct lmv_mds_md_v1    *lmv    = &lfsck_env_info(env)->lti_lmv;
 290         struct lfsck_lmv        *llmv;
 291         int                      rc;
 292         ENTRY;
 293
 294         LASSERT(lfsck->li_obj_dir == NULL);
 295         LASSERT(lfsck->li_lmv == NULL);
 296
 297         rc = lfsck_read_stripe_lmv(env, obj, lmv);
 298         if (rc == -ENODATA) {
 299                 lfsck->li_obj_dir = lfsck_object_get(obj);
 300
 301                 RETURN(0);
 302         }
 303
 304         if (rc < 0)
 305                 RETURN(rc);
 306
 307         OBD_ALLOC_PTR(llmv);
 308         if (llmv == NULL)
 309                 RETURN(-ENOMEM);
 310
 311         if (lmv->lmv_magic == LMV_MAGIC) {
 312                 struct lfsck_slave_lmv_rec      *lslr;
 313                 __u32                            stripes;
 314
 315                 llmv->ll_lmv_master = 1;
 316                 if (lmv->lmv_stripe_count < 1)
 317                         stripes = LFSCK_LMV_DEF_STRIPES;
 318                 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
 319                         stripes = LFSCK_LMV_MAX_STRIPES;
 320                 else
 321                         stripes = lmv->lmv_stripe_count;
 322
 323                 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
 324                 if (lslr == NULL) {
 325                         OBD_FREE_PTR(llmv);
 326
 327                         RETURN(-ENOMEM);
 328                 }
 329
 330                 /* Find the object against the bottom device. */
 331                 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
 332                                                lfsck_dto2fid(obj));
 333                 if (IS_ERR(obj)) {
 334                         OBD_FREE_LARGE(lslr, sizeof(*lslr) * stripes);
 335                         OBD_FREE_PTR(llmv);
 336
 337                         RETURN(PTR_ERR(obj));
 338                 }
 339
 340                 llmv->ll_stripes_allocated = stripes;
 341                 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
 342                 llmv->ll_lslr = lslr;
 343                 lfsck->li_obj_dir = obj;
 344         } else {
 345                 llmv->ll_lmv_slave = 1;
 346                 lfsck->li_obj_dir = lfsck_object_get(obj);
 347         }
 348
 349         llmv->ll_lmv = *lmv;
 350         atomic_set(&llmv->ll_ref, 1);
 351         lfsck->li_lmv = llmv;
 352
 353         RETURN(0);
 354 }
 355
 356 /* LFSCK wrap functions */
 357
 358 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
 359                        bool new_checked)
 360 {
 361         struct lfsck_component *com;
 362
 363         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 364                 com->lc_ops->lfsck_fail(env, com, new_checked);
 365         }
 366 }
 367
 368 void lfsck_close_dir(const struct lu_env *env,
 369                      struct lfsck_instance *lfsck, int result)
 370 {
 371         struct lfsck_component *com;
 372         ENTRY;
 373
 374         if (lfsck->li_lmv != NULL) {
 375                 lfsck->li_lmv->ll_exit_value = result;
 376                 if (lfsck->li_obj_dir != NULL) {
 377                         list_for_each_entry(com, &lfsck->li_list_dir,
 378                                             lc_link_dir) {
 379                                 com->lc_ops->lfsck_close_dir(env, com);
 380                         }
 381                 }
 382
 383                 lfsck_lmv_put(env, lfsck->li_lmv);
 384                 lfsck->li_lmv = NULL;
 385         }
 386
 387         if (lfsck->li_di_dir != NULL) {
 388                 const struct dt_it_ops  *dir_iops;
 389                 struct dt_it            *dir_di   = lfsck->li_di_dir;
 390
 391                 LASSERT(lfsck->li_obj_dir != NULL);
 392
 393                 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
 394                 lfsck_di_dir_put(env, lfsck);
 395                 dir_iops->fini(env, dir_di);
 396         }
 397
 398         if (lfsck->li_obj_dir != NULL) {
 399                 struct dt_object        *dir_obj  = lfsck->li_obj_dir;
 400
 401                 lfsck->li_obj_dir = NULL;
 402                 lfsck_object_put(env, dir_obj);
 403         }
 404
 405         EXIT;
 406 }
 407
 408 int lfsck_open_dir(const struct lu_env *env,
 409                    struct lfsck_instance *lfsck, __u64 cookie)
 410 {
 411         struct dt_object        *obj    = lfsck->li_obj_dir;
 412         struct dt_it            *di     = lfsck->li_di_dir;
 413         struct lfsck_component  *com;
 414         const struct dt_it_ops  *iops;
 415         int                      rc     = 0;
 416         ENTRY;
 417
 418         LASSERT(obj != NULL);
 419         LASSERT(di == NULL);
 420
 421         if (unlikely(!dt_try_as_dir(env, obj)))
 422                 GOTO(out, rc = -ENOTDIR);
 423
 424         list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
 425                 rc = com->lc_ops->lfsck_open_dir(env, com);
 426                 if (rc != 0)
 427                         GOTO(out, rc);
 428         }
 429
 430         iops = &obj->do_index_ops->dio_it;
 431         di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
 432         if (IS_ERR(di))
 433                 GOTO(out, rc = PTR_ERR(di));
 434
 435         rc = iops->load(env, di, cookie);
 436         if (rc == 0 || (rc > 0 && cookie > 0))
 437                 rc = iops->next(env, di);
 438         else if (rc > 0)
 439                 rc = 0;
 440
 441         if (rc != 0) {
 442                 iops->put(env, di);
 443                 iops->fini(env, di);
 444         } else {
 445                 lfsck->li_cookie_dir = iops->store(env, di);
 446                 spin_lock(&lfsck->li_lock);
 447                 lfsck->li_di_dir = di;
 448                 spin_unlock(&lfsck->li_lock);
 449         }
 450
 451         GOTO(out, rc);
 452
 453 out:
 454         if (rc != 0)
 455                 lfsck_close_dir(env, lfsck, rc);
 456
 457         return rc;
 458 }
 459
 460 static int lfsck_checkpoint(const struct lu_env *env,
 461                             struct lfsck_instance *lfsck)
 462 {
 463         struct lfsck_component *com;
 464         int                     rc  = 0;
 465         int                     rc1 = 0;
 466
 467         if (likely(cfs_time_beforeq(cfs_time_current(),
 468                                     lfsck->li_time_next_checkpoint)))
 469                 return 0;
 470
 471         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 472         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 473                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
 474                 if (rc != 0)
 475                         rc1 = rc;
 476         }
 477
 478         lfsck->li_time_last_checkpoint = cfs_time_current();
 479         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 480                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 481         return rc1 != 0 ? rc1 : rc;
 482 }
 483
 484 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 485                       struct lfsck_start_param *lsp)
 486 {
 487         struct dt_object       *obj     = NULL;
 488         struct lfsck_component *com;
 489         struct lfsck_component *next;
 490         struct lfsck_position  *pos     = NULL;
 491         const struct dt_it_ops *iops    =
 492                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 493         int                     rc;
 494         ENTRY;
 495
 496         LASSERT(lfsck->li_obj_dir == NULL);
 497         LASSERT(lfsck->li_di_dir == NULL);
 498
 499         lfsck->li_current_oit_processed = 0;
 500         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 501                 com->lc_new_checked = 0;
 502                 rc = com->lc_ops->lfsck_prep(env, com, lsp);
 503                 if (rc != 0)
 504                         GOTO(out, rc);
 505
 506                 if ((pos == NULL) ||
 507                     (!lfsck_pos_is_zero(&com->lc_pos_start) &&
 508                      lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
 509                         pos = &com->lc_pos_start;
 510         }
 511
 512         /* Init otable-based iterator. */
 513         if (pos == NULL) {
 514                 rc = iops->load(env, lfsck->li_di_oit, 0);
 515                 if (rc > 0) {
 516                         lfsck->li_oit_over = 1;
 517                         rc = 0;
 518                 }
 519
 520                 GOTO(out, rc);
 521         }
 522
 523         rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
 524         if (rc < 0)
 525                 GOTO(out, rc);
 526         else if (rc > 0)
 527                 lfsck->li_oit_over = 1;
 528
 529         if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
 530                 GOTO(out, rc = 0);
 531
 532         /* Find the directory for namespace-based traverse. */
 533         obj = lfsck_object_find(env, lfsck, &pos->lp_dir_parent);
 534         if (IS_ERR(obj))
 535                 RETURN(PTR_ERR(obj));
 536
 537         /* Remote directory will be scanned by the LFSCK instance
 538          * on the MDT where the remote object really resides on. */
 539         if (!dt_object_exists(obj) || dt_object_remote(obj) ||
 540             unlikely(!S_ISDIR(lfsck_object_type(obj))))
 541                 GOTO(out, rc = 0);
 542
 543         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 544         if (rc == 0) {
 545                 /* For the master MDT-object of a striped directory,
 546                  * reset the iteration from the directory beginning. */
 547                 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
 548                         pos->lp_dir_cookie = 0;
 549
 550                 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
 551                 if (rc > 0)
 552                         /* The end of the directory. */
 553                         rc = 0;
 554         }
 555
 556         GOTO(out, rc);
 557
 558 out:
 559         if (obj != NULL)
 560                 lfsck_object_put(env, obj);
 561
 562         if (rc != 0) {
 563                 lfsck_close_dir(env, lfsck, rc);
 564                 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 565                                          lc_link) {
 566                         com->lc_ops->lfsck_post(env, com, rc, true);
 567                 }
 568
 569                 return rc;
 570         }
 571
 572         rc = 0;
 573         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
 574         lfsck->li_pos_current = lfsck->li_pos_checkpoint;
 575         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 576                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
 577                 if (rc != 0)
 578                         break;
 579         }
 580
 581         lfsck->li_time_last_checkpoint = cfs_time_current();
 582         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 583                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 584         return rc;
 585 }
 586
 587 static int lfsck_exec_oit(const struct lu_env *env,
 588                           struct lfsck_instance *lfsck, struct dt_object *obj)
 589 {
 590         struct lfsck_component *com;
 591         int                     rc;
 592         ENTRY;
 593
 594         LASSERT(lfsck->li_obj_dir == NULL);
 595
 596         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 597                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
 598                 if (rc != 0)
 599                         RETURN(rc);
 600         }
 601
 602         rc = lfsck_needs_scan_dir(env, lfsck, obj);
 603         if (rc <= 0)
 604                 GOTO(out, rc);
 605
 606         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 607         if (rc == 0)
 608                 rc = lfsck_open_dir(env, lfsck, 0);
 609
 610         GOTO(out, rc);
 611
 612 out:
 613         if (rc < 0)
 614                 lfsck_fail(env, lfsck, false);
 615
 616         if (rc != 0)
 617                 lfsck_close_dir(env, lfsck, rc);
 618
 619         return rc > 0 ? 0 : rc;
 620 }
 621
 622 static int lfsck_exec_dir(const struct lu_env *env,
 623                           struct lfsck_instance *lfsck,
 624                           struct lu_dirent *ent, __u16 type)
 625 {
 626         struct lfsck_component *com;
 627         int                     rc;
 628
 629         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 630                 rc = com->lc_ops->lfsck_exec_dir(env, com, ent, type);
 631                 if (rc != 0)
 632                         return rc;
 633         }
 634         return 0;
 635 }
 636
 637 static int lfsck_master_dir_engine(const struct lu_env *env,
 638                                    struct lfsck_instance *lfsck);
 639
 640 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 641                       int result)
 642 {
 643         struct lfsck_component *com;
 644         struct lfsck_component *next;
 645         int                     rc  = result;
 646
 647         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 648         lfsck_close_dir(env, lfsck, result);
 649
 650         while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
 651                !list_empty(&lfsck->li_list_lmv)) {
 652                 struct lfsck_lmv_unit *llu;
 653
 654                 spin_lock(&lfsck->li_lock);
 655                 llu = list_entry(lfsck->li_list_lmv.next,
 656                                  struct lfsck_lmv_unit, llu_link);
 657                 list_del_init(&llu->llu_link);
 658                 spin_unlock(&lfsck->li_lock);
 659
 660                 lfsck->li_lmv = &llu->llu_lmv;
 661                 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 662                 rc = lfsck_open_dir(env, lfsck, 0);
 663                 if (rc == 0) {
 664                         rc = lfsck_master_dir_engine(env, lfsck);
 665                         lfsck_close_dir(env, lfsck, result);
 666                 }
 667         }
 668
 669         result = rc;
 670
 671         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 672                 rc = com->lc_ops->lfsck_post(env, com, result, false);
 673                 if (rc != 0)
 674                         CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
 675                                "rc = %d\n", lfsck_lfsck2name(lfsck),
 676                                (__u32)com->lc_type, rc);
 677         }
 678
 679         lfsck->li_time_last_checkpoint = cfs_time_current();
 680         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 681                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 682
 683         /* Ignore some component post failure to make other can go ahead. */
 684         return result;
 685 }
 686
 687 static int lfsck_double_scan(const struct lu_env *env,
 688                              struct lfsck_instance *lfsck)
 689 {
 690         struct lfsck_component *com;
 691         struct lfsck_component *next;
 692         struct l_wait_info      lwi = { 0 };
 693         int                     rc  = 0;
 694         int                     rc1 = 0;
 695
 696         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
 697                 rc = com->lc_ops->lfsck_double_scan(env, com);
 698                 if (rc != 0)
 699                         rc1 = rc;
 700         }
 701
 702         l_wait_event(lfsck->li_thread.t_ctl_waitq,
 703                      atomic_read(&lfsck->li_double_scan_count) == 0,
 704                      &lwi);
 705
 706         if (lfsck->li_status != LS_PAUSED &&
 707             lfsck->li_status != LS_CO_PAUSED) {
 708                 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 709                                          lc_link) {
 710                         spin_lock(&lfsck->li_lock);
 711                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 712                         spin_unlock(&lfsck->li_lock);
 713                 }
 714         }
 715
 716         return rc1 != 0 ? rc1 : rc;
 717 }
 718
 719 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
 720 {
 721         struct lfsck_component *com;
 722         struct lfsck_component *next;
 723
 724         list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 725                                  lc_link) {
 726                 if (com->lc_ops->lfsck_quit != NULL)
 727                         com->lc_ops->lfsck_quit(env, com);
 728
 729                 spin_lock(&lfsck->li_lock);
 730                 list_del_init(&com->lc_link_dir);
 731                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 732                 spin_unlock(&lfsck->li_lock);
 733         }
 734
 735         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 736                                  lc_link) {
 737                 if (com->lc_ops->lfsck_quit != NULL)
 738                         com->lc_ops->lfsck_quit(env, com);
 739
 740                 spin_lock(&lfsck->li_lock);
 741                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 742                 spin_unlock(&lfsck->li_lock);
 743         }
 744 }
 745
 746 /* LFSCK engines */
 747
 748 static int lfsck_master_dir_engine(const struct lu_env *env,
 749                                    struct lfsck_instance *lfsck)
 750 {
 751         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 752         struct dt_object                *dir    = lfsck->li_obj_dir;
 753         const struct dt_it_ops          *iops   = &dir->do_index_ops->dio_it;
 754         struct dt_it                    *di     = lfsck->li_di_dir;
 755         struct lu_dirent                *ent    =
 756                         (struct lu_dirent *)info->lti_key;
 757         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 758         struct ptlrpc_thread            *thread = &lfsck->li_thread;
 759         int                              rc;
 760         __u16                            type;
 761         ENTRY;
 762
 763         do {
 764                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY2) &&
 765                     cfs_fail_val > 0) {
 766                         struct l_wait_info lwi;
 767
 768                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
 769                                           NULL, NULL);
 770                         l_wait_event(thread->t_ctl_waitq,
 771                                      !thread_is_running(thread),
 772                                      &lwi);
 773
 774                         if (unlikely(!thread_is_running(thread))) {
 775                                 CDEBUG(D_LFSCK, "%s: scan dir exit for engine "
 776                                        "stop, parent "DFID", cookie "LPX64"\n",
 777                                        lfsck_lfsck2name(lfsck),
 778                                        PFID(lfsck_dto2fid(dir)),
 779                                        lfsck->li_cookie_dir);
 780                                 RETURN(0);
 781                         }
 782                 }
 783
 784                 lfsck->li_new_scanned++;
 785                 rc = iops->rec(env, di, (struct dt_rec *)ent,
 786                                lfsck->li_args_dir);
 787                 if (rc == 0)
 788                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
 789                                               &type);
 790
 791                 if (rc != 0) {
 792                         CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
 793                                "parent "DFID", cookie "LPX64": rc = %d\n",
 794                                lfsck_lfsck2name(lfsck),
 795                                PFID(lfsck_dto2fid(dir)),
 796                                lfsck->li_cookie_dir, rc);
 797                         lfsck_fail(env, lfsck, true);
 798                         if (bk->lb_param & LPF_FAILOUT)
 799                                 RETURN(rc);
 800                         else
 801                                 goto checkpoint;
 802                 }
 803
 804                 if (ent->lde_attrs & LUDA_IGNORE &&
 805                     strcmp(ent->lde_name, dotdot) != 0)
 806                         goto checkpoint;
 807
 808                 /* The type in the @ent structure may has been overwritten,
 809                  * so we need to pass the @type parameter independently. */
 810                 rc = lfsck_exec_dir(env, lfsck, ent, type);
 811                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 812                         RETURN(rc);
 813
 814 checkpoint:
 815                 rc = lfsck_checkpoint(env, lfsck);
 816                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 817                         RETURN(rc);
 818
 819                 /* Rate control. */
 820                 lfsck_control_speed(lfsck);
 821                 if (unlikely(!thread_is_running(thread))) {
 822                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 823                                "parent "DFID", cookie "LPX64"\n",
 824                                lfsck_lfsck2name(lfsck),
 825                                PFID(lfsck_dto2fid(dir)),
 826                                lfsck->li_cookie_dir);
 827                         RETURN(0);
 828                 }
 829
 830                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
 831                         spin_lock(&lfsck->li_lock);
 832                         thread_set_flags(thread, SVC_STOPPING);
 833                         spin_unlock(&lfsck->li_lock);
 834                         RETURN(-EINVAL);
 835                 }
 836
 837                 rc = iops->next(env, di);
 838         } while (rc == 0);
 839
 840         if (rc > 0 && !lfsck->li_oit_over)
 841                 lfsck_close_dir(env, lfsck, rc);
 842
 843         RETURN(rc);
 844 }
 845
 846 /**
 847  * Object-table based iteration engine.
 848  *
 849  * Object-table based iteration is the basic linear engine to scan all the
 850  * objects on current device in turn. For each object, it calls all the
 851  * registered LFSCK component(s)' API to perform related consistency
 852  * verification.
 853  *
 854  * It flushes related LFSCK trace files to disk via making checkpoint
 855  * periodically. Then if the server crashed or the LFSCK is paused, the
 856  * LFSCK can resume from the latest checkpoint.
 857  *
 858  * It also controls the whole LFSCK speed via lfsck_control_speed() to
 859  * avoid the server to become overload.
 860  *
 861  * \param[in] env       pointer to the thread context
 862  * \param[in] lfsck     pointer to the lfsck instance
 863  *
 864  * \retval              positive number if all objects have been scanned
 865  * \retval              0 if the iteration is stopped or paused
 866  * \retval              negative error number on failure
 867  */
 868 static int lfsck_master_oit_engine(const struct lu_env *env,
 869                                    struct lfsck_instance *lfsck)
 870 {
 871         struct lfsck_thread_info *info  = lfsck_env_info(env);
 872         const struct dt_it_ops   *iops  =
 873                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 874         struct dt_it             *di    = lfsck->li_di_oit;
 875         struct lu_fid            *fid   = &info->lti_fid;
 876         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 877         struct ptlrpc_thread     *thread = &lfsck->li_thread;
 878         struct dt_device         *dev   = lfsck->li_bottom;
 879         struct seq_server_site   *ss    = lu_site2seq(dev->dd_lu_dev.ld_site);
 880         __u32                    idx    = lfsck_dev_idx(dev);
 881         int                      rc;
 882         ENTRY;
 883
 884         if (unlikely(ss == NULL))
 885                 RETURN(-EIO);
 886
 887         do {
 888                 struct dt_object *target;
 889                 bool              update_lma = false;
 890
 891                 if (lfsck->li_di_dir != NULL) {
 892                         rc = lfsck_master_dir_engine(env, lfsck);
 893                         if (rc <= 0)
 894                                 RETURN(rc);
 895                 }
 896
 897                 if (unlikely(lfsck->li_oit_over))
 898                         RETURN(1);
 899
 900                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY1) &&
 901                     cfs_fail_val > 0) {
 902                         struct l_wait_info lwi;
 903
 904                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
 905                                           NULL, NULL);
 906                         l_wait_event(thread->t_ctl_waitq,
 907                                      !thread_is_running(thread),
 908                                      &lwi);
 909
 910                         if (unlikely(!thread_is_running(thread))) {
 911                                 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine "
 912                                        "stop, cookie "LPU64"\n",
 913                                        lfsck_lfsck2name(lfsck),
 914                                        iops->store(env, di));
 915                                 RETURN(0);
 916                         }
 917                 }
 918
 919                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
 920                         RETURN(0);
 921
 922                 lfsck->li_current_oit_processed = 1;
 923
 924                 if (!list_empty(&lfsck->li_list_lmv)) {
 925                         struct lfsck_lmv_unit *llu;
 926
 927                         spin_lock(&lfsck->li_lock);
 928                         llu = list_entry(lfsck->li_list_lmv.next,
 929                                          struct lfsck_lmv_unit, llu_link);
 930                         list_del_init(&llu->llu_link);
 931                         spin_unlock(&lfsck->li_lock);
 932
 933                         lfsck->li_lmv = &llu->llu_lmv;
 934                         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 935                         rc = lfsck_open_dir(env, lfsck, 0);
 936                         if (rc == 0)
 937                                 rc = lfsck_master_dir_engine(env, lfsck);
 938
 939                         if (rc <= 0)
 940                                 RETURN(rc);
 941                 }
 942
 943                 lfsck->li_new_scanned++;
 944                 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
 945                 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
 946                 if (rc != 0) {
 947                         CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
 948                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 949                         lfsck_fail(env, lfsck, true);
 950                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
 951                                 RETURN(rc);
 952                         else
 953                                 goto checkpoint;
 954                 }
 955
 956                 if (fid_is_idif(fid)) {
 957                         __u32 idx1 = fid_idif_ost_idx(fid);
 958
 959                         LASSERT(!lfsck->li_master);
 960
 961                         /* It is an old format device, update the LMA. */
 962                         if (idx != idx1) {
 963                                 struct ost_id *oi = &info->lti_oi;
 964
 965                                 fid_to_ostid(fid, oi);
 966                                 ostid_to_fid(fid, oi, idx);
 967                                 update_lma = true;
 968                         }
 969                 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
 970                            !fid_is_last_id(fid) &&
 971                            !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
 972
 973                         /* If the FID/object is only used locally and invisible
 974                          * to external nodes, then LFSCK will not handle it.
 975                          *
 976                          * dot_lustre sequence has been handled specially. */
 977                         goto checkpoint;
 978                 } else {
 979                         struct lu_seq_range *range = &info->lti_range;
 980
 981                         if (lfsck->li_master)
 982                                 fld_range_set_mdt(range);
 983                         else
 984                                 fld_range_set_ost(range);
 985                         rc = fld_local_lookup(env, ss->ss_server_fld,
 986                                               fid_seq(fid), range);
 987                         if (rc != 0 || range->lsr_index != idx) {
 988                                 /* Remote object will be handled by the LFSCK
 989                                  * instance on the MDT where the remote object
 990                                  * really resides on. */
 991                                 rc = 0;
 992                                 goto checkpoint;
 993                         }
 994                 }
 995
 996                 target = lfsck_object_find(env, lfsck, fid);
 997                 if (IS_ERR(target)) {
 998                         CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
 999                                DFID", cookie "LPU64": rc = %d\n",
1000                                lfsck_lfsck2name(lfsck), PFID(fid),
1001                                iops->store(env, di), rc);
1002                         lfsck_fail(env, lfsck, true);
1003                         if (bk->lb_param & LPF_FAILOUT)
1004                                 RETURN(PTR_ERR(target));
1005                         else
1006                                 goto checkpoint;
1007                 }
1008
1009                 if (dt_object_exists(target)) {
1010                         if (update_lma) {
1011                                 rc = lfsck_update_lma(env, lfsck, target);
1012                                 if (rc != 0)
1013                                         CDEBUG(D_LFSCK, "%s: fail to update "
1014                                                "LMA for "DFID": rc = %d\n",
1015                                                lfsck_lfsck2name(lfsck),
1016                                                PFID(lfsck_dto2fid(target)), rc);
1017                         }
1018                         if (rc == 0)
1019                                 rc = lfsck_exec_oit(env, lfsck, target);
1020                 }
1021                 lfsck_object_put(env, target);
1022                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1023                         RETURN(rc);
1024
1025 checkpoint:
1026                 rc = lfsck_checkpoint(env, lfsck);
1027                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1028                         RETURN(rc);
1029
1030                 /* Rate control. */
1031                 lfsck_control_speed(lfsck);
1032
1033                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1034                         spin_lock(&lfsck->li_lock);
1035                         thread_set_flags(thread, SVC_STOPPING);
1036                         spin_unlock(&lfsck->li_lock);
1037                         RETURN(-EINVAL);
1038                 }
1039
1040                 rc = iops->next(env, di);
1041                 if (unlikely(rc > 0))
1042                         lfsck->li_oit_over = 1;
1043                 else if (likely(rc == 0))
1044                         lfsck->li_current_oit_processed = 0;
1045
1046                 if (unlikely(!thread_is_running(thread))) {
1047                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1048                                "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1049                                iops->store(env, di));
1050                         RETURN(0);
1051                 }
1052         } while (rc == 0 || lfsck->li_di_dir != NULL);
1053
1054         RETURN(rc);
1055 }
1056
1057 int lfsck_master_engine(void *args)
1058 {
1059         struct lfsck_thread_args *lta      = args;
1060         struct lu_env            *env      = &lta->lta_env;
1061         struct lfsck_instance    *lfsck    = lta->lta_lfsck;
1062         struct ptlrpc_thread     *thread   = &lfsck->li_thread;
1063         struct dt_object         *oit_obj  = lfsck->li_obj_oit;
1064         const struct dt_it_ops   *oit_iops = &oit_obj->do_index_ops->dio_it;
1065         struct dt_it             *oit_di;
1066         struct l_wait_info        lwi      = { 0 };
1067         int                       rc;
1068         ENTRY;
1069
1070         if (lfsck->li_master &&
1071             (!list_empty(&lfsck->li_list_scan) ||
1072              !list_empty(&lfsck->li_list_double_scan))) {
1073                 rc = lfsck_verify_lpf(env, lfsck);
1074                 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1075                  * fatal, because the .lustre/lost+found/ maybe not accessed
1076                  * by the LFSCK if it does not add orphans or others to such
1077                  * directory. So go ahead until hit failure when really uses
1078                  * the directory. */
1079                 if (rc != 0)
1080                         CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1081                                ".lustre/lost+found/, go ahead: rc = %d\n",
1082                                lfsck_lfsck2name(lfsck), rc);
1083         }
1084
1085         oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit, BYPASS_CAPA);
1086         if (IS_ERR(oit_di)) {
1087                 rc = PTR_ERR(oit_di);
1088                 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1089                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1090
1091                 GOTO(fini_args, rc);
1092         }
1093
1094         spin_lock(&lfsck->li_lock);
1095         lfsck->li_di_oit = oit_di;
1096         spin_unlock(&lfsck->li_lock);
1097         rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1098         if (rc != 0)
1099                 GOTO(fini_oit, rc);
1100
1101         CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1102                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1103                ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1104                lfsck->li_pos_checkpoint.lp_oit_cookie,
1105                lfsck->li_pos_checkpoint.lp_dir_cookie,
1106                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1107                current_pid());
1108
1109         spin_lock(&lfsck->li_lock);
1110         thread_set_flags(thread, SVC_RUNNING);
1111         spin_unlock(&lfsck->li_lock);
1112         wake_up_all(&thread->t_ctl_waitq);
1113
1114         l_wait_event(thread->t_ctl_waitq,
1115                      lfsck->li_start_unplug ||
1116                      !thread_is_running(thread),
1117                      &lwi);
1118         if (!thread_is_running(thread))
1119                 GOTO(fini_oit, rc = 0);
1120
1121         if (!list_empty(&lfsck->li_list_scan) ||
1122             list_empty(&lfsck->li_list_double_scan))
1123                 rc = lfsck_master_oit_engine(env, lfsck);
1124         else
1125                 rc = 1;
1126
1127         CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1128                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1129                ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1130                lfsck->li_pos_checkpoint.lp_oit_cookie,
1131                lfsck->li_pos_checkpoint.lp_dir_cookie,
1132                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1133                current_pid(), rc);
1134
1135         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1136                 rc = lfsck_post(env, lfsck, rc);
1137         else
1138                 lfsck_close_dir(env, lfsck, rc);
1139
1140 fini_oit:
1141         lfsck_di_oit_put(env, lfsck);
1142         oit_iops->fini(env, oit_di);
1143         if (rc == 1) {
1144                 if (!list_empty(&lfsck->li_list_double_scan))
1145                         rc = lfsck_double_scan(env, lfsck);
1146                 else
1147                         rc = 0;
1148         } else {
1149                 lfsck_quit(env, lfsck);
1150         }
1151
1152         /* XXX: Purge the pinned objects in the future. */
1153
1154 fini_args:
1155         spin_lock(&lfsck->li_lock);
1156         thread_set_flags(thread, SVC_STOPPED);
1157         spin_unlock(&lfsck->li_lock);
1158         wake_up_all(&thread->t_ctl_waitq);
1159         lfsck_thread_args_fini(lta);
1160         return rc;
1161 }
1162
1163 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1164 {
1165         bool empty = false;
1166
1167         spin_lock(&lad->lad_lock);
1168         if (list_empty(&lad->lad_req_list))
1169                 empty = true;
1170         spin_unlock(&lad->lad_lock);
1171
1172         return empty;
1173 }
1174
1175 /**
1176  * Query the LFSCK status from the instatnces on remote servers.
1177  *
1178  * The LFSCK assistant thread queries the LFSCK instances on other
1179  * servers (MDT/OST) about their status, such as whether they have
1180  * finished the phase1/phase2 scanning or not, and so on.
1181  *
1182  * \param[in] env       pointer to the thread context
1183  * \param[in] com       pointer to the lfsck component
1184  *
1185  * \retval              0 for success
1186  * \retval              negative error number on failure
1187  */
1188 static int lfsck_assistant_query_others(const struct lu_env *env,
1189                                         struct lfsck_component *com)
1190 {
1191         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1192         struct lfsck_request              *lr    = &info->lti_lr;
1193         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1194         struct lfsck_instance             *lfsck = com->lc_lfsck;
1195         struct lfsck_assistant_data       *lad   = com->lc_data;
1196         struct ptlrpc_request_set         *set;
1197         struct lfsck_tgt_descs            *ltds;
1198         struct lfsck_tgt_desc             *ltd;
1199         struct list_head                  *phase_head;
1200         int                                rc    = 0;
1201         int                                rc1   = 0;
1202         ENTRY;
1203
1204         set = ptlrpc_prep_set();
1205         if (set == NULL)
1206                 RETURN(-ENOMEM);
1207
1208         lad->lad_touch_gen++;
1209         memset(lr, 0, sizeof(*lr));
1210         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1211         lr->lr_event = LE_QUERY;
1212         lr->lr_active = com->lc_type;
1213         laia->laia_com = com;
1214         laia->laia_lr = lr;
1215         laia->laia_shared = 0;
1216
1217         if (!list_empty(&lad->lad_mdt_phase1_list)) {
1218                 ltds = &lfsck->li_mdt_descs;
1219                 lr->lr_flags = 0;
1220                 phase_head = &lad->lad_mdt_phase1_list;
1221         } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1222                 goto out;
1223         } else {
1224
1225 again:
1226                 ltds = &lfsck->li_ost_descs;
1227                 lr->lr_flags = LEF_TO_OST;
1228                 phase_head = &lad->lad_ost_phase1_list;
1229         }
1230
1231         laia->laia_ltds = ltds;
1232         spin_lock(&ltds->ltd_lock);
1233         while (!list_empty(phase_head)) {
1234                 struct list_head *phase_list;
1235                 __u32            *gen;
1236
1237                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1238                         ltd = list_entry(phase_head->next,
1239                                          struct lfsck_tgt_desc,
1240                                          ltd_layout_phase_list);
1241                         phase_list = &ltd->ltd_layout_phase_list;
1242                         gen = &ltd->ltd_layout_gen;
1243                 } else {
1244                         ltd = list_entry(phase_head->next,
1245                                          struct lfsck_tgt_desc,
1246                                          ltd_namespace_phase_list);
1247                         phase_list = &ltd->ltd_namespace_phase_list;
1248                         gen = &ltd->ltd_namespace_gen;
1249                 }
1250
1251                 if (*gen == lad->lad_touch_gen)
1252                         break;
1253
1254                 *gen = lad->lad_touch_gen;
1255                 list_move_tail(phase_list, phase_head);
1256                 atomic_inc(&ltd->ltd_ref);
1257                 laia->laia_ltd = ltd;
1258                 spin_unlock(&ltds->ltd_lock);
1259                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1260                                          lfsck_async_interpret_common,
1261                                          laia, LFSCK_QUERY);
1262                 if (rc != 0) {
1263                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1264                                "%s %x for %s: rc = %d\n",
1265                                lfsck_lfsck2name(lfsck),
1266                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1267                                ltd->ltd_index, lad->lad_name, rc);
1268                         lfsck_tgt_put(ltd);
1269                         rc1 = rc;
1270                 }
1271                 spin_lock(&ltds->ltd_lock);
1272         }
1273         spin_unlock(&ltds->ltd_lock);
1274
1275         rc = ptlrpc_set_wait(set);
1276         if (rc < 0) {
1277                 ptlrpc_set_destroy(set);
1278                 RETURN(rc);
1279         }
1280
1281         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1282             list_empty(&lad->lad_mdt_phase1_list))
1283                 goto again;
1284
1285 out:
1286         ptlrpc_set_destroy(set);
1287
1288         RETURN(rc1 != 0 ? rc1 : rc);
1289 }
1290
1291 /**
1292  * Notify the LFSCK event to the instances on remote servers.
1293  *
1294  * The LFSCK assistant thread notifies the LFSCK instances on other
1295  * servers (MDT/OST) about some events, such as start new scanning,
1296  * stop the scanning, this LFSCK instance will exit, and so on.
1297  *
1298  * \param[in] env       pointer to the thread context
1299  * \param[in] com       pointer to the lfsck component
1300  * \param[in] lr        pointer to the LFSCK event request
1301  *
1302  * \retval              0 for success
1303  * \retval              negative error number on failure
1304  */
1305 static int lfsck_assistant_notify_others(const struct lu_env *env,
1306                                          struct lfsck_component *com,
1307                                          struct lfsck_request *lr)
1308 {
1309         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1310         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1311         struct lfsck_instance             *lfsck = com->lc_lfsck;
1312         struct lfsck_assistant_data       *lad   = com->lc_data;
1313         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1314         struct ptlrpc_request_set         *set;
1315         struct lfsck_tgt_descs            *ltds;
1316         struct lfsck_tgt_desc             *ltd;
1317         struct lfsck_tgt_desc             *next;
1318         __u32                              idx;
1319         int                                rc    = 0;
1320         int                                rc1   = 0;
1321         ENTRY;
1322
1323         set = ptlrpc_prep_set();
1324         if (set == NULL)
1325                 RETURN(-ENOMEM);
1326
1327         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1328         lr->lr_active = com->lc_type;
1329         laia->laia_com = com;
1330         laia->laia_lr = lr;
1331         laia->laia_shared = 0;
1332
1333         switch (lr->lr_event) {
1334         case LE_START:
1335                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1336                         goto next;
1337
1338                 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1339                 lr->lr_speed = bk->lb_speed_limit;
1340                 lr->lr_version = bk->lb_version;
1341                 lr->lr_param |= bk->lb_param;
1342                 lr->lr_async_windows = bk->lb_async_windows;
1343                 lr->lr_flags = LEF_TO_OST;
1344
1345                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1346                 ltds = &lfsck->li_ost_descs;
1347                 laia->laia_ltds = ltds;
1348                 down_read(&ltds->ltd_rw_sem);
1349                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1350                         ltd = lfsck_tgt_get(ltds, idx);
1351                         LASSERT(ltd != NULL);
1352
1353                         laia->laia_ltd = ltd;
1354                         ltd->ltd_layout_done = 0;
1355                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1356                                         lfsck_async_interpret_common,
1357                                         laia, LFSCK_NOTIFY);
1358                         if (rc != 0) {
1359                                 lfsck_lad_set_bitmap(env, com, idx);
1360                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1361                                        "notify OST %x for %s start: rc = %d\n",
1362                                        lfsck_lfsck2name(lfsck), idx,
1363                                        lad->lad_name, rc);
1364                                 lfsck_tgt_put(ltd);
1365                         }
1366                 }
1367                 up_read(&ltds->ltd_rw_sem);
1368
1369                 /* Sync up */
1370                 rc = ptlrpc_set_wait(set);
1371                 if (rc < 0) {
1372                         ptlrpc_set_destroy(set);
1373                         RETURN(rc);
1374                 }
1375
1376 next:
1377                 if (!(bk->lb_param & LPF_ALL_TGT))
1378                         break;
1379
1380                 /* link other MDT targets locallly. */
1381                 ltds = &lfsck->li_mdt_descs;
1382                 spin_lock(&ltds->ltd_lock);
1383                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1384                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1385                                 ltd = LTD_TGT(ltds, idx);
1386                                 LASSERT(ltd != NULL);
1387
1388                                 if (!list_empty(&ltd->ltd_layout_list))
1389                                         continue;
1390
1391                                 list_add_tail(&ltd->ltd_layout_list,
1392                                               &lad->lad_mdt_list);
1393                                 list_add_tail(&ltd->ltd_layout_phase_list,
1394                                               &lad->lad_mdt_phase1_list);
1395                         }
1396                 } else {
1397                         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1398                                 ltd = LTD_TGT(ltds, idx);
1399                                 LASSERT(ltd != NULL);
1400
1401                                 if (!list_empty(&ltd->ltd_namespace_list))
1402                                         continue;
1403
1404                                 list_add_tail(&ltd->ltd_namespace_list,
1405                                               &lad->lad_mdt_list);
1406                                 list_add_tail(&ltd->ltd_namespace_phase_list,
1407                                               &lad->lad_mdt_phase1_list);
1408                         }
1409                 }
1410                 spin_unlock(&ltds->ltd_lock);
1411                 break;
1412         case LE_STOP:
1413         case LE_PHASE2_DONE:
1414         case LE_PEER_EXIT: {
1415                 struct list_head *phase_head;
1416
1417                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1418                 if (bk->lb_param & LPF_ALL_TGT) {
1419                         phase_head = &lad->lad_mdt_list;
1420                         ltds = &lfsck->li_mdt_descs;
1421                         if (lr->lr_event == LE_STOP) {
1422                                 /* unlink other MDT targets locallly. */
1423                                 spin_lock(&ltds->ltd_lock);
1424                                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1425                                         list_for_each_entry_safe(ltd, next,
1426                                                 phase_head, ltd_layout_list) {
1427                                                 list_del_init(
1428                                                 &ltd->ltd_layout_phase_list);
1429                                                 list_del_init(
1430                                                 &ltd->ltd_layout_list);
1431                                         }
1432                                 } else {
1433                                         list_for_each_entry_safe(ltd, next,
1434                                                         phase_head,
1435                                                         ltd_namespace_list) {
1436                                                 list_del_init(
1437                                                 &ltd->ltd_namespace_phase_list);
1438                                                 list_del_init(
1439                                                 &ltd->ltd_namespace_list);
1440                                         }
1441                                 }
1442                                 spin_unlock(&ltds->ltd_lock);
1443
1444                                 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1445                                         break;
1446
1447                                 lr->lr_flags |= LEF_TO_OST;
1448                                 phase_head = &lad->lad_ost_list;
1449                                 ltds = &lfsck->li_ost_descs;
1450                         } else {
1451                                 lr->lr_flags &= ~LEF_TO_OST;
1452                         }
1453                 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1454                         break;
1455                 } else {
1456                         lr->lr_flags |= LEF_TO_OST;
1457                         phase_head = &lad->lad_ost_list;
1458                         ltds = &lfsck->li_ost_descs;
1459                 }
1460
1461 again:
1462                 laia->laia_ltds = ltds;
1463                 spin_lock(&ltds->ltd_lock);
1464                 while (!list_empty(phase_head)) {
1465                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1466                                 ltd = list_entry(phase_head->next,
1467                                                  struct lfsck_tgt_desc,
1468                                                  ltd_layout_list);
1469                                 if (!list_empty(&ltd->ltd_layout_phase_list))
1470                                         list_del_init(
1471                                                 &ltd->ltd_layout_phase_list);
1472                                 list_del_init(&ltd->ltd_layout_list);
1473                         } else {
1474                                 ltd = list_entry(phase_head->next,
1475                                                  struct lfsck_tgt_desc,
1476                                                  ltd_namespace_list);
1477                                 if (!list_empty(&ltd->ltd_namespace_phase_list))
1478                                         list_del_init(
1479                                                 &ltd->ltd_namespace_phase_list);
1480                                 list_del_init(&ltd->ltd_namespace_list);
1481                         }
1482                         atomic_inc(&ltd->ltd_ref);
1483                         laia->laia_ltd = ltd;
1484                         spin_unlock(&ltds->ltd_lock);
1485                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1486                                         lfsck_async_interpret_common,
1487                                         laia, LFSCK_NOTIFY);
1488                         if (rc != 0) {
1489                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1490                                        "notify %s %x for %s stop/phase2_done/"
1491                                        "peer_exit: rc = %d\n",
1492                                        lfsck_lfsck2name(lfsck),
1493                                        (lr->lr_flags & LEF_TO_OST) ?
1494                                        "OST" : "MDT", ltd->ltd_index,
1495                                        lad->lad_name, rc);
1496                                 lfsck_tgt_put(ltd);
1497                         }
1498                         spin_lock(&ltds->ltd_lock);
1499                 }
1500                 spin_unlock(&ltds->ltd_lock);
1501
1502                 rc = ptlrpc_set_wait(set);
1503                 if (rc < 0) {
1504                         ptlrpc_set_destroy(set);
1505                         RETURN(rc);
1506                 }
1507
1508                 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1509                     !(lr->lr_flags & LEF_TO_OST)) {
1510                         lr->lr_flags |= LEF_TO_OST;
1511                         phase_head = &lad->lad_ost_list;
1512                         ltds = &lfsck->li_ost_descs;
1513                         goto again;
1514                 }
1515                 break;
1516         }
1517         case LE_PHASE1_DONE:
1518                 lad->lad_ops->la_sync_failures(env, com, lr);
1519                 lad->lad_touch_gen++;
1520                 ltds = &lfsck->li_mdt_descs;
1521                 laia->laia_ltds = ltds;
1522                 spin_lock(&ltds->ltd_lock);
1523                 while (!list_empty(&lad->lad_mdt_list)) {
1524                         struct list_head *list;
1525                         __u32            *gen;
1526
1527                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1528                                 ltd = list_entry(lad->lad_mdt_list.next,
1529                                                  struct lfsck_tgt_desc,
1530                                                  ltd_layout_list);
1531                                 list = &ltd->ltd_layout_list;
1532                                 gen = &ltd->ltd_layout_gen;
1533                         } else {
1534                                 struct lfsck_namespace *ns = com->lc_file_ram;
1535
1536                                 ltd = list_entry(lad->lad_mdt_list.next,
1537                                                  struct lfsck_tgt_desc,
1538                                                  ltd_namespace_list);
1539                                 list = &ltd->ltd_namespace_list;
1540                                 gen = &ltd->ltd_namespace_gen;
1541                                 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1542                         }
1543
1544                         if (*gen == lad->lad_touch_gen)
1545                                 break;
1546
1547                         *gen = lad->lad_touch_gen;
1548                         list_move_tail(list, &lad->lad_mdt_list);
1549                         if (ltd->ltd_namespace_failed)
1550                                 continue;
1551
1552                         atomic_inc(&ltd->ltd_ref);
1553                         laia->laia_ltd = ltd;
1554                         spin_unlock(&ltds->ltd_lock);
1555                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1556                                         lfsck_async_interpret_common,
1557                                         laia, LFSCK_NOTIFY);
1558                         if (rc != 0) {
1559                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1560                                        "notify MDT %x for %s phase1 done: "
1561                                        "rc = %d\n", lfsck_lfsck2name(lfsck),
1562                                        ltd->ltd_index, lad->lad_name, rc);
1563                                 lfsck_tgt_put(ltd);
1564                         }
1565                         spin_lock(&ltds->ltd_lock);
1566                 }
1567                 spin_unlock(&ltds->ltd_lock);
1568                 break;
1569         default:
1570                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1571                        "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1572                 rc = -EINVAL;
1573                 break;
1574         }
1575
1576         rc1 = ptlrpc_set_wait(set);
1577         ptlrpc_set_destroy(set);
1578
1579         RETURN(rc != 0 ? rc : rc1);
1580 }
1581
1582 /**
1583  * The LFSCK assistant thread is triggered by the LFSCK main engine.
1584  * They co-work together as an asynchronous pipeline: the LFSCK main
1585  * engine scans the system and pre-fetches the objects, attributes,
1586  * or name entries, etc, and pushes them into the pipeline as input
1587  * requests for the LFSCK assistant thread; on the other end of the
1588  * pipeline, the LFSCK assistant thread performs the real check and
1589  * repair for every request from the main engine.
1590  *
1591  * Generally, the assistant engine may be blocked when check/repair
1592  * something, so the LFSCK main engine will run some faster. On the
1593  * other hand, the LFSCK main engine will drive multiple assistant
1594  * threads in parallel, means for each LFSCK component on the master
1595  * (such as layout LFSCK, namespace LFSCK), there is an independent
1596  * LFSCK assistant thread. So under such 1:N multiple asynchronous
1597  * pipelines mode, the whole LFSCK performance will be much better
1598  * than check/repair everything by the LFSCK main engine itself.
1599  */
1600 int lfsck_assistant_engine(void *args)
1601 {
1602         struct lfsck_thread_args          *lta     = args;
1603         struct lu_env                     *env     = &lta->lta_env;
1604         struct lfsck_component            *com     = lta->lta_com;
1605         struct lfsck_instance             *lfsck   = lta->lta_lfsck;
1606         struct lfsck_bookmark             *bk      = &lfsck->li_bookmark_ram;
1607         struct lfsck_position             *pos     = &com->lc_pos_start;
1608         struct lfsck_thread_info          *info    = lfsck_env_info(env);
1609         struct lfsck_request              *lr      = &info->lti_lr;
1610         struct lfsck_assistant_data       *lad     = com->lc_data;
1611         struct ptlrpc_thread              *mthread = &lfsck->li_thread;
1612         struct ptlrpc_thread              *athread = &lad->lad_thread;
1613         struct lfsck_assistant_operations *lao     = lad->lad_ops;
1614         struct lfsck_assistant_req        *lar;
1615         struct l_wait_info                 lwi     = { 0 };
1616         int                                rc      = 0;
1617         int                                rc1     = 0;
1618         int                                rc2;
1619         ENTRY;
1620
1621         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1622                lfsck_lfsck2name(lfsck), lad->lad_name);
1623
1624         memset(lr, 0, sizeof(*lr));
1625         lr->lr_event = LE_START;
1626         if (pos->lp_oit_cookie <= 1)
1627                 lr->lr_param = LPF_RESET;
1628         rc = lfsck_assistant_notify_others(env, com, lr);
1629         if (rc != 0) {
1630                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1631                        "to start %s: rc = %d\n",
1632                        lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1633                 GOTO(fini, rc);
1634         }
1635
1636         spin_lock(&lad->lad_lock);
1637         thread_set_flags(athread, SVC_RUNNING);
1638         spin_unlock(&lad->lad_lock);
1639         wake_up_all(&mthread->t_ctl_waitq);
1640
1641         while (1) {
1642                 while (!list_empty(&lad->lad_req_list)) {
1643                         bool wakeup = false;
1644
1645                         if (unlikely(lad->lad_exit ||
1646                                      !thread_is_running(mthread)))
1647                                 GOTO(cleanup1, rc = lad->lad_post_result);
1648
1649                         lar = list_entry(lad->lad_req_list.next,
1650                                          struct lfsck_assistant_req,
1651                                          lar_list);
1652                         /* Only the lfsck_assistant_engine thread itself can
1653                          * remove the "lar" from the head of the list, LFSCK
1654                          * engine thread only inserts other new "lar" at the
1655                          * end of the list. So it is safe to handle current
1656                          * "lar" without the spin_lock. */
1657                         rc = lao->la_handler_p1(env, com, lar);
1658                         spin_lock(&lad->lad_lock);
1659                         list_del_init(&lar->lar_list);
1660                         lad->lad_prefetched--;
1661                         /* Wake up the main engine thread only when the list
1662                          * is empty or half of the prefetched items have been
1663                          * handled to avoid too frequent thread schedule. */
1664                         if (lad->lad_prefetched == 0 ||
1665                             (bk->lb_async_windows != 0 &&
1666                              bk->lb_async_windows / 2 ==
1667                              lad->lad_prefetched))
1668                                 wakeup = true;
1669                         spin_unlock(&lad->lad_lock);
1670                         if (wakeup)
1671                                 wake_up_all(&mthread->t_ctl_waitq);
1672
1673                         lao->la_req_fini(env, lar);
1674                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1675                                 GOTO(cleanup1, rc);
1676                 }
1677
1678                 l_wait_event(athread->t_ctl_waitq,
1679                              !lfsck_assistant_req_empty(lad) ||
1680                              lad->lad_exit ||
1681                              lad->lad_to_post ||
1682                              lad->lad_to_double_scan,
1683                              &lwi);
1684
1685                 if (unlikely(lad->lad_exit))
1686                         GOTO(cleanup1, rc = lad->lad_post_result);
1687
1688                 if (!list_empty(&lad->lad_req_list))
1689                         continue;
1690
1691                 if (lad->lad_to_post) {
1692                         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1693                                lfsck_lfsck2name(lfsck), lad->lad_name);
1694
1695                         if (unlikely(lad->lad_exit))
1696                                 GOTO(cleanup1, rc = lad->lad_post_result);
1697
1698                         lad->lad_to_post = 0;
1699                         LASSERT(lad->lad_post_result > 0);
1700
1701                         memset(lr, 0, sizeof(*lr));
1702                         lr->lr_event = LE_PHASE1_DONE;
1703                         lr->lr_status = lad->lad_post_result;
1704                         rc = lfsck_assistant_notify_others(env, com, lr);
1705                         if (rc != 0)
1706                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to "
1707                                        "notify others for %s post: rc = %d\n",
1708                                        lfsck_lfsck2name(lfsck),
1709                                        lad->lad_name, rc);
1710
1711                         /* Wakeup the master engine to go ahead. */
1712                         wake_up_all(&mthread->t_ctl_waitq);
1713                 }
1714
1715                 if (lad->lad_to_double_scan) {
1716                         lad->lad_to_double_scan = 0;
1717                         atomic_inc(&lfsck->li_double_scan_count);
1718                         lad->lad_in_double_scan = 1;
1719                         wake_up_all(&mthread->t_ctl_waitq);
1720
1721                         com->lc_new_checked = 0;
1722                         com->lc_new_scanned = 0;
1723                         com->lc_time_last_checkpoint = cfs_time_current();
1724                         com->lc_time_next_checkpoint =
1725                                 com->lc_time_last_checkpoint +
1726                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1727
1728                         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1729                                "the second-stage scaning\n",
1730                                lfsck_lfsck2name(lfsck));
1731
1732                         /* Flush async updates before handling orphan. */
1733                         rc2 = dt_sync(env, lfsck->li_next);
1734
1735                         CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1736                                "scan start, synced: rc = %d\n",
1737                                lfsck_lfsck2name(lfsck), rc2);
1738
1739                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1740                                 GOTO(cleanup2, rc = 0);
1741
1742                         while (lad->lad_in_double_scan) {
1743                                 rc = lfsck_assistant_query_others(env, com);
1744                                 if (lfsck_phase2_next_ready(lad))
1745                                         goto p2_next;
1746
1747                                 if (rc < 0)
1748                                         GOTO(cleanup2, rc);
1749
1750                                 /* Pull LFSCK status on related targets once
1751                                  * per 30 seconds if we are not notified. */
1752                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1753                                                            cfs_time_seconds(1),
1754                                                            NULL, NULL);
1755                                 rc = l_wait_event(athread->t_ctl_waitq,
1756                                         lfsck_phase2_next_ready(lad) ||
1757                                         lad->lad_exit ||
1758                                         !thread_is_running(mthread),
1759                                         &lwi);
1760
1761                                 if (unlikely(lad->lad_exit ||
1762                                              !thread_is_running(mthread)))
1763                                         GOTO(cleanup2, rc = 0);
1764
1765                                 if (rc == -ETIMEDOUT)
1766                                         continue;
1767
1768                                 if (rc < 0)
1769                                         GOTO(cleanup2, rc);
1770
1771 p2_next:
1772                                 rc = lao->la_handler_p2(env, com);
1773                                 if (rc != 0)
1774                                         GOTO(cleanup2, rc);
1775
1776                                 if (unlikely(lad->lad_exit ||
1777                                              !thread_is_running(mthread)))
1778                                         GOTO(cleanup2, rc = 0);
1779                         }
1780                 }
1781         }
1782
1783 cleanup1:
1784         /* Cleanup the unfinished requests. */
1785         spin_lock(&lad->lad_lock);
1786         if (rc < 0)
1787                 lad->lad_assistant_status = rc;
1788
1789         if (lad->lad_exit && lad->lad_post_result <= 0)
1790                 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1791
1792         while (!list_empty(&lad->lad_req_list)) {
1793                 lar = list_entry(lad->lad_req_list.next,
1794                                  struct lfsck_assistant_req,
1795                                  lar_list);
1796                 list_del_init(&lar->lar_list);
1797                 lad->lad_prefetched--;
1798                 spin_unlock(&lad->lad_lock);
1799                 lao->la_req_fini(env, lar);
1800                 spin_lock(&lad->lad_lock);
1801         }
1802         spin_unlock(&lad->lad_lock);
1803
1804         LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1805                  lad->lad_prefetched);
1806
1807 cleanup2:
1808         memset(lr, 0, sizeof(*lr));
1809         if (rc > 0) {
1810                 lr->lr_event = LE_PHASE2_DONE;
1811                 lr->lr_status = rc;
1812         } else if (rc == 0) {
1813                 if (lfsck->li_flags & LPF_ALL_TGT) {
1814                         lr->lr_event = LE_STOP;
1815                         lr->lr_status = LS_STOPPED;
1816                 } else {
1817                         lr->lr_event = LE_PEER_EXIT;
1818                         switch (lfsck->li_status) {
1819                         case LS_PAUSED:
1820                         case LS_CO_PAUSED:
1821                                 lr->lr_status = LS_CO_PAUSED;
1822                                 break;
1823                         case LS_STOPPED:
1824                         case LS_CO_STOPPED:
1825                                 lr->lr_status = LS_CO_STOPPED;
1826                                 break;
1827                         default:
1828                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1829                                        "status: rc = %d\n",
1830                                        lfsck_lfsck2name(lfsck),
1831                                        lfsck->li_status);
1832                                 lr->lr_status = LS_CO_FAILED;
1833                                 break;
1834                         }
1835                 }
1836         } else {
1837                 if (lfsck->li_flags & LPF_ALL_TGT) {
1838                         lr->lr_event = LE_STOP;
1839                         lr->lr_status = LS_FAILED;
1840                 } else {
1841                         lr->lr_event = LE_PEER_EXIT;
1842                         lr->lr_status = LS_CO_FAILED;
1843                 }
1844         }
1845
1846         rc1 = lfsck_assistant_notify_others(env, com, lr);
1847         if (rc1 != 0) {
1848                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1849                        "others for %s quit: rc = %d\n",
1850                        lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1851                 rc = rc1;
1852         }
1853
1854         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1855                lfsck_lfsck2name(lfsck));
1856
1857         /* Flush async updates before exit. */
1858         rc2 = dt_sync(env, lfsck->li_next);
1859
1860         CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1861                lfsck_lfsck2name(lfsck), rc2);
1862
1863         /* Under force exit case, some requests may be just freed without
1864          * verification, those objects should be re-handled when next run.
1865          * So not update the on-disk trace file under such case. */
1866         if (lad->lad_in_double_scan) {
1867                 if (!lad->lad_exit)
1868                         rc1 = lao->la_double_scan_result(env, com, rc);
1869
1870                 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1871                        "finished: rc = %d\n",
1872                        lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1873         }
1874
1875 fini:
1876         if (lad->lad_in_double_scan)
1877                 atomic_dec(&lfsck->li_double_scan_count);
1878
1879         spin_lock(&lad->lad_lock);
1880         lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1881         thread_set_flags(athread, SVC_STOPPED);
1882         wake_up_all(&mthread->t_ctl_waitq);
1883         spin_unlock(&lad->lad_lock);
1884
1885         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1886                lfsck_lfsck2name(lfsck), lad->lad_name,
1887                lad->lad_assistant_status);
1888
1889         lfsck_thread_args_fini(lta);
1890
1891         return rc;
1892 }