lustre/lfsck/lfsck_engine.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_engine.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lu_object.h>
  34 #include <dt_object.h>
  35 #include <lustre_net.h>
  36 #include <lustre_fid.h>
  37 #include <obd_support.h>
  38 #include <lustre_lib.h>
  39
  40 #include "lfsck_internal.h"
  41
  42 int lfsck_unpack_ent(struct lu_dirent *ent, __u64 *cookie, __u16 *type)
  43 {
  44         struct luda_type        *lt;
  45         int                      align = sizeof(*lt) - 1;
  46         int                      len;
  47
  48         fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
  49         *cookie = le64_to_cpu(ent->lde_hash);
  50         ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
  51         ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
  52         ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
  53
  54         if (unlikely(!(ent->lde_attrs & LUDA_TYPE)))
  55                 return -EINVAL;
  56
  57         len = (ent->lde_namelen + align) & ~align;
  58         lt = (struct luda_type *)(ent->lde_name + len);
  59         *type = le16_to_cpu(lt->lt_type);
  60
  61         /* Make sure the name is terminated with '\0'. The data (object type)
  62          * after ent::lde_name maybe broken, but we have stored such data in
  63          * the output parameter @type as above. */
  64         ent->lde_name[ent->lde_namelen] = '\0';
  65
  66         return 0;
  67 }
  68
  69 static void lfsck_di_oit_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  70 {
  71         const struct dt_it_ops  *iops;
  72         struct dt_it            *di;
  73
  74         spin_lock(&lfsck->li_lock);
  75         iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
  76         di = lfsck->li_di_oit;
  77         lfsck->li_di_oit = NULL;
  78         spin_unlock(&lfsck->li_lock);
  79         iops->put(env, di);
  80 }
  81
  82 static void lfsck_di_dir_put(const struct lu_env *env, struct lfsck_instance *lfsck)
  83 {
  84         const struct dt_it_ops  *iops;
  85         struct dt_it            *di;
  86
  87         spin_lock(&lfsck->li_lock);
  88         iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
  89         di = lfsck->li_di_dir;
  90         lfsck->li_di_dir = NULL;
  91         lfsck->li_cookie_dir = 0;
  92         spin_unlock(&lfsck->li_lock);
  93         iops->put(env, di);
  94 }
  95
  96 static int lfsck_update_lma(const struct lu_env *env,
  97                             struct lfsck_instance *lfsck, struct dt_object *obj)
  98 {
  99         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 100         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 101         struct dt_device                *dev    = lfsck_obj2dev(obj);
 102         struct lustre_mdt_attrs         *lma    = &info->lti_lma;
 103         struct lu_buf                   *buf;
 104         struct thandle                  *th;
 105         int                              fl;
 106         int                              rc;
 107         ENTRY;
 108
 109         if (bk->lb_param & LPF_DRYRUN)
 110                 RETURN(0);
 111
 112         buf = lfsck_buf_get(env, info->lti_lma_old, LMA_OLD_SIZE);
 113         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LMA);
 114         if (rc < 0) {
 115                 if (rc != -ENODATA)
 116                         RETURN(rc);
 117
 118                 fl = LU_XATTR_CREATE;
 119                 lustre_lma_init(lma, lfsck_dto2fid(obj), LMAC_FID_ON_OST, 0);
 120         } else {
 121                 if (rc != LMA_OLD_SIZE && rc != sizeof(struct lustre_mdt_attrs))
 122                         RETURN(-EINVAL);
 123
 124                 fl = LU_XATTR_REPLACE;
 125                 lustre_lma_swab(lma);
 126                 lustre_lma_init(lma, lfsck_dto2fid(obj),
 127                                 lma->lma_compat | LMAC_FID_ON_OST,
 128                                 lma->lma_incompat);
 129         }
 130         lustre_lma_swab(lma);
 131
 132         th = dt_trans_create(env, dev);
 133         if (IS_ERR(th))
 134                 RETURN(PTR_ERR(th));
 135
 136         buf = lfsck_buf_get(env, lma, sizeof(*lma));
 137         rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
 138         if (rc != 0)
 139                 GOTO(stop, rc);
 140
 141         rc = dt_trans_start_local(env, dev, th);
 142         if (rc != 0)
 143                 GOTO(stop, rc);
 144
 145         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
 146
 147         GOTO(stop, rc);
 148
 149 stop:
 150         dt_trans_stop(env, dev, th);
 151         return rc;
 152 }
 153
 154 static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj,
 155                             struct lu_fid *fid)
 156 {
 157         if (unlikely(!S_ISDIR(lfsck_object_type(obj)) ||
 158                      !dt_try_as_dir(env, obj)))
 159                 return -ENOTDIR;
 160
 161         return dt_lookup(env, obj, (struct dt_rec *)fid,
 162                          (const struct dt_key *)"..");
 163 }
 164
 165 /**
 166  * Check whether needs to scan the directory or not.
 167  *
 168  * 1) If we are not doing namespace LFSCK, or the given @obj is not directory,
 169  *    then needs not to scan the @obj. Otherwise,
 170  * 2) Global /ROOT needs to be scanned, backend root needs not to be scanned.
 171  * 3) If the @obj is neither IGIF nor normal FID (including .lustre and its
 172  *    sub-directories that have been scanned when the LFSCK engine start),
 173  *    then needs not to be scanned.
 174  * 4) If it is a remote object, then scanning the object will be done on the
 175  *    MDT on which the object really resides.
 176  * 5) If the local object has normal FID, then needs to be scanned. Otherwise,
 177  * 6) If the object has linkEA, then needs to be scanned. Otherwise,
 178  * 7) If none of the previous conditions are true, we need to check the parent
 179  *    directories whether this subdirectory is in a tree that should be scanned.
 180  *    Set the parent as current @obj, repeat 2)-7).
 181  *
 182  * \param[in] env       pointer to the thread context
 183  * \param[in] lfsck     pointer to the lfsck instance
 184  * \param[in] obj       pointer to the object to be checked
 185  *
 186  * \retval              positive number if the directory needs to be scanned
 187  * \retval              0 if the directory needs NOT to be scanned
 188  * \retval              negative error number on failure
 189  */
 190 static int lfsck_needs_scan_dir(const struct lu_env *env,
 191                                 struct lfsck_instance *lfsck,
 192                                 struct dt_object *obj)
 193 {
 194         struct lfsck_thread_info *info    = lfsck_env_info(env);
 195         struct lu_fid            *fid     = &info->lti_fid;
 196         struct lu_seq_range      *range   = &info->lti_range;
 197         struct seq_server_site   *ss      = lfsck_dev_site(lfsck);
 198         __u32                     idx     = lfsck_dev_idx(lfsck);
 199         int                       depth   = 0;
 200         int                       rc      = 0;
 201
 202         if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
 203                 return 0;
 204
 205         LASSERT(ss != NULL);
 206
 207         *fid = *lfsck_dto2fid(obj);
 208         while (1) {
 209                 /* Global /ROOT is visible. */
 210                 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
 211                         return 1;
 212
 213                 /* Backend root is invisible. */
 214                 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
 215                         return 0;
 216
 217                 if (!fid_is_norm(fid) && !fid_is_igif(fid))
 218                         return 0;
 219
 220                 fld_range_set_mdt(range);
 221                 rc = fld_local_lookup(env, ss->ss_server_fld,
 222                                       fid_seq(fid), range);
 223                 if (rc != 0 || range->lsr_index != idx)
 224                         /* Current FID should NOT be for the input parameter
 225                          * @obj, because the lfsck_master_oit_engine() has
 226                          * filtered out agent object. So current FID is for
 227                          * the ancestor of the original input parameter @obj.
 228                          * So the ancestor is a remote directory. The input
 229                          * parameter @obj is local directory, and should be
 230                          * scanned under such case. */
 231                         return 1;
 232
 233                 /* normal FID on this target (locally) must be for the
 234                  * client-side visiable object. */
 235                 if (fid_is_norm(fid))
 236                         return 1;
 237
 238                 if (obj == NULL) {
 239                         obj = lfsck_object_find_bottom(env, lfsck, fid);
 240                         if (IS_ERR(obj))
 241                                 return PTR_ERR(obj);
 242
 243                         depth++;
 244                         if (!dt_object_exists(obj))
 245                                 GOTO(out, rc = 0);
 246                 }
 247
 248                 dt_read_lock(env, obj, MOR_TGT_CHILD);
 249                 if (unlikely(lfsck_is_dead_obj(obj))) {
 250                         dt_read_unlock(env, obj);
 251
 252                         GOTO(out, rc = 0);
 253                 }
 254
 255                 rc = dt_xattr_get(env, obj,
 256                                   lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK);
 257                 dt_read_unlock(env, obj);
 258                 if (rc >= 0)
 259                         GOTO(out, rc = 1);
 260
 261                 if (rc < 0 && rc != -ENODATA)
 262                         GOTO(out, rc);
 263
 264                 rc = lfsck_parent_fid(env, obj, fid);
 265                 if (depth > 0)
 266                         lfsck_object_put(env, obj);
 267
 268                 obj = NULL;
 269                 if (rc != 0)
 270                         return rc;
 271
 272                 if (!fid_is_sane(fid))
 273                         return 0;
 274         }
 275
 276 out:
 277         if (depth > 0 && obj != NULL)
 278                 lfsck_object_put(env, obj);
 279
 280         return rc;
 281 }
 282
 283 static int lfsck_load_stripe_lmv(const struct lu_env *env,
 284                                  struct lfsck_instance *lfsck,
 285                                  struct dt_object *obj)
 286 {
 287         struct lmv_mds_md_v1    *lmv    = &lfsck_env_info(env)->lti_lmv;
 288         struct lfsck_lmv        *llmv;
 289         int                      rc;
 290         ENTRY;
 291
 292         LASSERT(lfsck->li_obj_dir == NULL);
 293         LASSERT(lfsck->li_lmv == NULL);
 294
 295         rc = lfsck_read_stripe_lmv(env, obj, lmv);
 296         if (rc == -ENODATA) {
 297                 lfsck->li_obj_dir = lfsck_object_get(obj);
 298
 299                 RETURN(0);
 300         }
 301
 302         if (rc < 0)
 303                 RETURN(rc);
 304
 305         OBD_ALLOC_PTR(llmv);
 306         if (llmv == NULL)
 307                 RETURN(-ENOMEM);
 308
 309         if (lmv->lmv_magic == LMV_MAGIC) {
 310                 struct lfsck_slave_lmv_rec      *lslr;
 311                 __u32                            stripes;
 312
 313                 llmv->ll_lmv_master = 1;
 314                 if (lmv->lmv_stripe_count < 1)
 315                         stripes = LFSCK_LMV_DEF_STRIPES;
 316                 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
 317                         stripes = LFSCK_LMV_MAX_STRIPES;
 318                 else
 319                         stripes = lmv->lmv_stripe_count;
 320
 321                 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
 322                 if (lslr == NULL) {
 323                         OBD_FREE_PTR(llmv);
 324
 325                         RETURN(-ENOMEM);
 326                 }
 327
 328                 llmv->ll_stripes_allocated = stripes;
 329                 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
 330                 llmv->ll_lslr = lslr;
 331         } else {
 332                 llmv->ll_lmv_slave = 1;
 333         }
 334
 335         lfsck->li_obj_dir = lfsck_object_get(obj);
 336         llmv->ll_lmv = *lmv;
 337         atomic_set(&llmv->ll_ref, 1);
 338         lfsck->li_lmv = llmv;
 339
 340         RETURN(0);
 341 }
 342
 343 /* LFSCK wrap functions */
 344
 345 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
 346                        bool new_checked)
 347 {
 348         struct lfsck_component *com;
 349
 350         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 351                 com->lc_ops->lfsck_fail(env, com, new_checked);
 352         }
 353 }
 354
 355 void lfsck_close_dir(const struct lu_env *env,
 356                      struct lfsck_instance *lfsck, int result)
 357 {
 358         struct lfsck_component *com;
 359         ENTRY;
 360
 361         if (lfsck->li_lmv != NULL) {
 362                 lfsck->li_lmv->ll_exit_value = result;
 363                 if (lfsck->li_obj_dir != NULL) {
 364                         list_for_each_entry(com, &lfsck->li_list_dir,
 365                                             lc_link_dir) {
 366                                 com->lc_ops->lfsck_close_dir(env, com);
 367                         }
 368                 }
 369
 370                 lfsck_lmv_put(env, lfsck->li_lmv);
 371                 lfsck->li_lmv = NULL;
 372         }
 373
 374         if (lfsck->li_di_dir != NULL) {
 375                 const struct dt_it_ops  *dir_iops;
 376                 struct dt_it            *dir_di   = lfsck->li_di_dir;
 377
 378                 LASSERT(lfsck->li_obj_dir != NULL);
 379
 380                 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
 381                 lfsck_di_dir_put(env, lfsck);
 382                 dir_iops->fini(env, dir_di);
 383         }
 384
 385         if (lfsck->li_obj_dir != NULL) {
 386                 struct dt_object        *dir_obj  = lfsck->li_obj_dir;
 387
 388                 lfsck->li_obj_dir = NULL;
 389                 lfsck_object_put(env, dir_obj);
 390         }
 391
 392         EXIT;
 393 }
 394
 395 int lfsck_open_dir(const struct lu_env *env,
 396                    struct lfsck_instance *lfsck, __u64 cookie)
 397 {
 398         struct dt_object        *obj    = lfsck->li_obj_dir;
 399         struct dt_it            *di     = lfsck->li_di_dir;
 400         struct lfsck_component  *com;
 401         const struct dt_it_ops  *iops;
 402         int                      rc     = 0;
 403         ENTRY;
 404
 405         LASSERT(obj != NULL);
 406         LASSERT(di == NULL);
 407
 408         if (unlikely(!dt_try_as_dir(env, obj)))
 409                 GOTO(out, rc = -ENOTDIR);
 410
 411         list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
 412                 rc = com->lc_ops->lfsck_open_dir(env, com);
 413                 if (rc != 0)
 414                         GOTO(out, rc);
 415         }
 416
 417         iops = &obj->do_index_ops->dio_it;
 418         di = iops->init(env, obj, lfsck->li_args_dir);
 419         if (IS_ERR(di))
 420                 GOTO(out, rc = PTR_ERR(di));
 421
 422         rc = iops->load(env, di, cookie);
 423         if (rc == 0 || (rc > 0 && cookie > 0))
 424                 rc = iops->next(env, di);
 425         else if (rc > 0)
 426                 rc = 0;
 427
 428         if (rc != 0) {
 429                 iops->put(env, di);
 430                 iops->fini(env, di);
 431         } else {
 432                 lfsck->li_cookie_dir = iops->store(env, di);
 433                 spin_lock(&lfsck->li_lock);
 434                 lfsck->li_di_dir = di;
 435                 spin_unlock(&lfsck->li_lock);
 436         }
 437
 438         GOTO(out, rc);
 439
 440 out:
 441         if (rc != 0)
 442                 lfsck_close_dir(env, lfsck, rc);
 443
 444         return rc;
 445 }
 446
 447 static int lfsck_checkpoint(const struct lu_env *env,
 448                             struct lfsck_instance *lfsck)
 449 {
 450         struct lfsck_component *com;
 451         int                     rc  = 0;
 452         int                     rc1 = 0;
 453
 454         if (likely(cfs_time_beforeq(cfs_time_current(),
 455                                     lfsck->li_time_next_checkpoint)))
 456                 return 0;
 457
 458         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 459         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 460                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
 461                 if (rc != 0)
 462                         rc1 = rc;
 463         }
 464
 465         lfsck->li_time_last_checkpoint = cfs_time_current();
 466         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 467                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 468         return rc1 != 0 ? rc1 : rc;
 469 }
 470
 471 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 472                       struct lfsck_start_param *lsp)
 473 {
 474         struct dt_object       *obj     = NULL;
 475         struct lfsck_component *com;
 476         struct lfsck_component *next;
 477         struct lfsck_position  *pos     = NULL;
 478         const struct dt_it_ops *iops    =
 479                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 480         int                     rc;
 481         ENTRY;
 482
 483         LASSERT(lfsck->li_obj_dir == NULL);
 484         LASSERT(lfsck->li_di_dir == NULL);
 485
 486         lfsck->li_current_oit_processed = 0;
 487         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 488                 com->lc_new_checked = 0;
 489                 rc = com->lc_ops->lfsck_prep(env, com, lsp);
 490                 if (rc != 0)
 491                         GOTO(out, rc);
 492
 493                 if ((pos == NULL) ||
 494                     (!lfsck_pos_is_zero(&com->lc_pos_start) &&
 495                      lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
 496                         pos = &com->lc_pos_start;
 497         }
 498
 499         /* Init otable-based iterator. */
 500         if (pos == NULL) {
 501                 rc = iops->load(env, lfsck->li_di_oit, 0);
 502                 if (rc > 0) {
 503                         lfsck->li_oit_over = 1;
 504                         rc = 0;
 505                 }
 506
 507                 GOTO(out, rc);
 508         }
 509
 510         rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
 511         if (rc < 0)
 512                 GOTO(out, rc);
 513         else if (rc > 0)
 514                 lfsck->li_oit_over = 1;
 515
 516         if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
 517                 GOTO(out, rc = 0);
 518
 519         /* Find the directory for namespace-based traverse. */
 520         obj = lfsck_object_find_bottom(env, lfsck, &pos->lp_dir_parent);
 521         if (IS_ERR(obj))
 522                 RETURN(PTR_ERR(obj));
 523
 524         /* Remote directory will be scanned by the LFSCK instance
 525          * on the MDT where the remote object really resides on. */
 526         if (!dt_object_exists(obj) || dt_object_remote(obj) ||
 527             unlikely(!S_ISDIR(lfsck_object_type(obj))))
 528                 GOTO(out, rc = 0);
 529
 530         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 531         if (rc == 0) {
 532                 /* For the master MDT-object of a striped directory,
 533                  * reset the iteration from the directory beginning. */
 534                 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
 535                         pos->lp_dir_cookie = 0;
 536
 537                 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
 538                 if (rc > 0)
 539                         /* The end of the directory. */
 540                         rc = 0;
 541         }
 542
 543         GOTO(out, rc);
 544
 545 out:
 546         if (obj != NULL)
 547                 lfsck_object_put(env, obj);
 548
 549         if (rc != 0) {
 550                 lfsck_close_dir(env, lfsck, rc);
 551                 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 552                                          lc_link) {
 553                         com->lc_ops->lfsck_post(env, com, rc, true);
 554                 }
 555
 556                 return rc;
 557         }
 558
 559         rc = 0;
 560         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
 561         lfsck->li_pos_current = lfsck->li_pos_checkpoint;
 562         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 563                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
 564                 if (rc != 0)
 565                         break;
 566         }
 567
 568         lfsck->li_time_last_checkpoint = cfs_time_current();
 569         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 570                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 571         return rc;
 572 }
 573
 574 static int lfsck_exec_oit(const struct lu_env *env,
 575                           struct lfsck_instance *lfsck, struct dt_object *obj)
 576 {
 577         struct lfsck_component *com;
 578         int                     rc;
 579         ENTRY;
 580
 581         LASSERT(lfsck->li_obj_dir == NULL);
 582
 583         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 584                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
 585                 if (rc != 0)
 586                         RETURN(rc);
 587         }
 588
 589         rc = lfsck_needs_scan_dir(env, lfsck, obj);
 590         if (rc <= 0)
 591                 GOTO(out, rc);
 592
 593         rc = lfsck_load_stripe_lmv(env, lfsck, obj);
 594         if (rc == 0)
 595                 rc = lfsck_open_dir(env, lfsck, 0);
 596
 597         GOTO(out, rc);
 598
 599 out:
 600         if (rc < 0)
 601                 lfsck_fail(env, lfsck, false);
 602
 603         if (rc != 0)
 604                 lfsck_close_dir(env, lfsck, rc);
 605
 606         return rc > 0 ? 0 : rc;
 607 }
 608
 609 static int lfsck_exec_dir(const struct lu_env *env,
 610                           struct lfsck_instance *lfsck,
 611                           struct lfsck_assistant_object *lso,
 612                           struct lu_dirent *ent, __u16 type)
 613 {
 614         struct lfsck_component *com;
 615         int                     rc;
 616
 617         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 618                 rc = com->lc_ops->lfsck_exec_dir(env, com, lso, ent, type);
 619                 if (rc != 0)
 620                         return rc;
 621         }
 622         return 0;
 623 }
 624
 625 static int lfsck_master_dir_engine(const struct lu_env *env,
 626                                    struct lfsck_instance *lfsck);
 627
 628 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 629                       int result)
 630 {
 631         struct lfsck_component *com;
 632         struct lfsck_component *next;
 633         int                     rc  = result;
 634
 635         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 636         lfsck_close_dir(env, lfsck, result);
 637
 638         while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
 639                !list_empty(&lfsck->li_list_lmv)) {
 640                 struct lfsck_lmv_unit *llu;
 641
 642                 spin_lock(&lfsck->li_lock);
 643                 llu = list_entry(lfsck->li_list_lmv.next,
 644                                  struct lfsck_lmv_unit, llu_link);
 645                 list_del_init(&llu->llu_link);
 646                 spin_unlock(&lfsck->li_lock);
 647
 648                 lfsck->li_lmv = &llu->llu_lmv;
 649                 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 650                 rc = lfsck_open_dir(env, lfsck, 0);
 651                 if (rc == 0) {
 652                         rc = lfsck_master_dir_engine(env, lfsck);
 653                         lfsck_close_dir(env, lfsck, result);
 654                 }
 655         }
 656
 657         result = rc;
 658
 659         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 660                 rc = com->lc_ops->lfsck_post(env, com, result, false);
 661                 if (rc != 0)
 662                         CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
 663                                "rc = %d\n", lfsck_lfsck2name(lfsck),
 664                                (__u32)com->lc_type, rc);
 665         }
 666
 667         lfsck->li_time_last_checkpoint = cfs_time_current();
 668         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
 669                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 670
 671         /* Ignore some component post failure to make other can go ahead. */
 672         return result;
 673 }
 674
 675 static int lfsck_double_scan(const struct lu_env *env,
 676                              struct lfsck_instance *lfsck)
 677 {
 678         struct lfsck_component *com;
 679         struct lfsck_component *next;
 680         struct l_wait_info      lwi = { 0 };
 681         int                     rc  = 0;
 682         int                     rc1 = 0;
 683
 684         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
 685                 rc = com->lc_ops->lfsck_double_scan(env, com);
 686                 if (rc != 0)
 687                         rc1 = rc;
 688         }
 689
 690         l_wait_event(lfsck->li_thread.t_ctl_waitq,
 691                      atomic_read(&lfsck->li_double_scan_count) == 0,
 692                      &lwi);
 693
 694         if (lfsck->li_status != LS_PAUSED &&
 695             lfsck->li_status != LS_CO_PAUSED) {
 696                 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 697                                          lc_link) {
 698                         spin_lock(&lfsck->li_lock);
 699                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 700                         spin_unlock(&lfsck->li_lock);
 701                 }
 702         }
 703
 704         return rc1 != 0 ? rc1 : rc;
 705 }
 706
 707 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
 708 {
 709         struct lfsck_component *com;
 710         struct lfsck_component *next;
 711
 712         list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
 713                                  lc_link) {
 714                 if (com->lc_ops->lfsck_quit != NULL)
 715                         com->lc_ops->lfsck_quit(env, com);
 716
 717                 spin_lock(&lfsck->li_lock);
 718                 list_del_init(&com->lc_link_dir);
 719                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 720                 spin_unlock(&lfsck->li_lock);
 721         }
 722
 723         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 724                                  lc_link) {
 725                 if (com->lc_ops->lfsck_quit != NULL)
 726                         com->lc_ops->lfsck_quit(env, com);
 727
 728                 spin_lock(&lfsck->li_lock);
 729                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
 730                 spin_unlock(&lfsck->li_lock);
 731         }
 732 }
 733
 734 /* LFSCK engines */
 735
 736 static int lfsck_master_dir_engine(const struct lu_env *env,
 737                                    struct lfsck_instance *lfsck)
 738 {
 739         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 740         struct dt_object                *dir    = lfsck->li_obj_dir;
 741         const struct dt_it_ops          *iops   = &dir->do_index_ops->dio_it;
 742         struct dt_it                    *di     = lfsck->li_di_dir;
 743         struct lu_dirent                *ent    =
 744                         (struct lu_dirent *)info->lti_key;
 745         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 746         struct ptlrpc_thread            *thread = &lfsck->li_thread;
 747         struct lfsck_assistant_object   *lso    = NULL;
 748         int                              rc;
 749         __u16                            type;
 750         ENTRY;
 751
 752         do {
 753                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
 754                     unlikely(!thread_is_running(thread))) {
 755                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 756                                "parent "DFID", cookie "LPX64"\n",
 757                                lfsck_lfsck2name(lfsck),
 758                                PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
 759
 760                         GOTO(out, rc = 0);
 761                 }
 762
 763                 lfsck->li_new_scanned++;
 764                 rc = iops->rec(env, di, (struct dt_rec *)ent,
 765                                lfsck->li_args_dir);
 766                 if (rc == 0)
 767                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
 768                                               &type);
 769
 770                 if (rc != 0) {
 771                         CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
 772                                "parent "DFID", cookie "LPX64": rc = %d\n",
 773                                lfsck_lfsck2name(lfsck),
 774                                PFID(lfsck_dto2fid(dir)),
 775                                lfsck->li_cookie_dir, rc);
 776                         lfsck_fail(env, lfsck, true);
 777                         if (bk->lb_param & LPF_FAILOUT)
 778                                 GOTO(out, rc);
 779                         else
 780                                 goto checkpoint;
 781                 }
 782
 783                 if (ent->lde_attrs & LUDA_IGNORE &&
 784                     strcmp(ent->lde_name, dotdot) != 0)
 785                         goto checkpoint;
 786
 787                 /* skip dot entry. */
 788                 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
 789                         goto checkpoint;
 790
 791                 if (lso == NULL) {
 792                         lso = lfsck_assistant_object_init(env,
 793                                 lfsck_dto2fid(dir), NULL,
 794                                 lfsck->li_pos_current.lp_oit_cookie, true);
 795                         if (IS_ERR(lso)) {
 796                                 if (bk->lb_param & LPF_FAILOUT)
 797                                         RETURN(PTR_ERR(lso));
 798
 799                                 lso = NULL;
 800                                 goto checkpoint;
 801                         }
 802                 }
 803
 804                 /* The type in the @ent structure may has been overwritten,
 805                  * so we need to pass the @type parameter independently. */
 806                 rc = lfsck_exec_dir(env, lfsck, lso, ent, type);
 807                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 808                         GOTO(out, rc);
 809
 810 checkpoint:
 811                 rc = lfsck_checkpoint(env, lfsck);
 812                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
 813                         GOTO(out, rc);
 814
 815                 /* Rate control. */
 816                 lfsck_control_speed(lfsck);
 817                 if (unlikely(!thread_is_running(thread))) {
 818                         CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
 819                                "parent "DFID", cookie "LPX64"\n",
 820                                lfsck_lfsck2name(lfsck),
 821                                PFID(lfsck_dto2fid(dir)),
 822                                lfsck->li_cookie_dir);
 823                         GOTO(out, rc = 0);
 824                 }
 825
 826                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
 827                         spin_lock(&lfsck->li_lock);
 828                         thread_set_flags(thread, SVC_STOPPING);
 829                         spin_unlock(&lfsck->li_lock);
 830                         GOTO(out, rc = -EINVAL);
 831                 }
 832
 833                 rc = iops->next(env, di);
 834         } while (rc == 0);
 835
 836         if (rc > 0 && !lfsck->li_oit_over)
 837                 lfsck_close_dir(env, lfsck, rc);
 838
 839         GOTO(out, rc);
 840
 841 out:
 842         if (lso != NULL)
 843                 lfsck_assistant_object_put(env, lso);
 844
 845         return rc;
 846 }
 847
 848 /**
 849  * Object-table based iteration engine.
 850  *
 851  * Object-table based iteration is the basic linear engine to scan all the
 852  * objects on current device in turn. For each object, it calls all the
 853  * registered LFSCK component(s)' API to perform related consistency
 854  * verification.
 855  *
 856  * It flushes related LFSCK trace files to disk via making checkpoint
 857  * periodically. Then if the server crashed or the LFSCK is paused, the
 858  * LFSCK can resume from the latest checkpoint.
 859  *
 860  * It also controls the whole LFSCK speed via lfsck_control_speed() to
 861  * avoid the server to become overload.
 862  *
 863  * \param[in] env       pointer to the thread context
 864  * \param[in] lfsck     pointer to the lfsck instance
 865  *
 866  * \retval              positive number if all objects have been scanned
 867  * \retval              0 if the iteration is stopped or paused
 868  * \retval              negative error number on failure
 869  */
 870 static int lfsck_master_oit_engine(const struct lu_env *env,
 871                                    struct lfsck_instance *lfsck)
 872 {
 873         struct lfsck_thread_info *info  = lfsck_env_info(env);
 874         const struct dt_it_ops   *iops  =
 875                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
 876         struct dt_it             *di    = lfsck->li_di_oit;
 877         struct lu_fid            *fid   = &info->lti_fid;
 878         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 879         struct ptlrpc_thread     *thread = &lfsck->li_thread;
 880         struct seq_server_site   *ss    = lfsck_dev_site(lfsck);
 881         __u32                    idx    = lfsck_dev_idx(lfsck);
 882         int                      rc;
 883         ENTRY;
 884
 885         if (unlikely(ss == NULL))
 886                 RETURN(-EIO);
 887
 888         do {
 889                 struct dt_object *target;
 890                 bool              update_lma = false;
 891
 892                 if (lfsck->li_di_dir != NULL) {
 893                         rc = lfsck_master_dir_engine(env, lfsck);
 894                         if (rc <= 0)
 895                                 RETURN(rc);
 896                 }
 897
 898                 if (unlikely(lfsck->li_oit_over))
 899                         RETURN(1);
 900
 901                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
 902                     unlikely(!thread_is_running(thread))) {
 903                         CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
 904                                "cookie "LPU64"\n",
 905                                lfsck_lfsck2name(lfsck), iops->store(env, di));
 906
 907                         RETURN(0);
 908                 }
 909
 910                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
 911                         RETURN(0);
 912
 913                 lfsck->li_current_oit_processed = 1;
 914
 915                 if (!list_empty(&lfsck->li_list_lmv)) {
 916                         struct lfsck_lmv_unit *llu;
 917
 918                         spin_lock(&lfsck->li_lock);
 919                         llu = list_entry(lfsck->li_list_lmv.next,
 920                                          struct lfsck_lmv_unit, llu_link);
 921                         list_del_init(&llu->llu_link);
 922                         spin_unlock(&lfsck->li_lock);
 923
 924                         lfsck->li_lmv = &llu->llu_lmv;
 925                         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
 926                         rc = lfsck_open_dir(env, lfsck, 0);
 927                         if (rc == 0)
 928                                 rc = lfsck_master_dir_engine(env, lfsck);
 929
 930                         if (rc <= 0)
 931                                 RETURN(rc);
 932                 }
 933
 934                 lfsck->li_new_scanned++;
 935                 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
 936                 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
 937                 if (rc != 0) {
 938                         CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
 939                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 940                         lfsck_fail(env, lfsck, true);
 941                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
 942                                 RETURN(rc);
 943                         else
 944                                 goto checkpoint;
 945                 }
 946
 947                 if (unlikely(!fid_is_sane(fid))) {
 948                         CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
 949                                ", skip it\n",
 950                                lfsck_lfsck2name(lfsck), PFID(fid));
 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_bottom(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);
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         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1126         CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1127                "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1128                ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1129                lfsck->li_pos_checkpoint.lp_oit_cookie,
1130                lfsck->li_pos_checkpoint.lp_dir_cookie,
1131                PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1132                current_pid(), rc);
1133
1134         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1135                 rc = lfsck_post(env, lfsck, rc);
1136         else
1137                 lfsck_close_dir(env, lfsck, rc);
1138
1139 fini_oit:
1140         lfsck_di_oit_put(env, lfsck);
1141         oit_iops->fini(env, oit_di);
1142         if (rc == 1) {
1143                 if (!list_empty(&lfsck->li_list_double_scan))
1144                         rc = lfsck_double_scan(env, lfsck);
1145                 else
1146                         rc = 0;
1147         } else {
1148                 lfsck_quit(env, lfsck);
1149         }
1150
1151         /* XXX: Purge the pinned objects in the future. */
1152
1153 fini_args:
1154         spin_lock(&lfsck->li_lock);
1155         thread_set_flags(thread, SVC_STOPPED);
1156         spin_unlock(&lfsck->li_lock);
1157         wake_up_all(&thread->t_ctl_waitq);
1158         lfsck_thread_args_fini(lta);
1159         return rc;
1160 }
1161
1162 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1163 {
1164         bool empty = false;
1165
1166         spin_lock(&lad->lad_lock);
1167         if (list_empty(&lad->lad_req_list))
1168                 empty = true;
1169         spin_unlock(&lad->lad_lock);
1170
1171         return empty;
1172 }
1173
1174 /**
1175  * Query the LFSCK status from the instatnces on remote servers.
1176  *
1177  * The LFSCK assistant thread queries the LFSCK instances on other
1178  * servers (MDT/OST) about their status, such as whether they have
1179  * finished the phase1/phase2 scanning or not, and so on.
1180  *
1181  * \param[in] env       pointer to the thread context
1182  * \param[in] com       pointer to the lfsck component
1183  *
1184  * \retval              0 for success
1185  * \retval              negative error number on failure
1186  */
1187 static int lfsck_assistant_query_others(const struct lu_env *env,
1188                                         struct lfsck_component *com)
1189 {
1190         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1191         struct lfsck_request              *lr    = &info->lti_lr;
1192         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1193         struct lfsck_instance             *lfsck = com->lc_lfsck;
1194         struct lfsck_assistant_data       *lad   = com->lc_data;
1195         struct ptlrpc_request_set         *set;
1196         struct lfsck_tgt_descs            *ltds;
1197         struct lfsck_tgt_desc             *ltd;
1198         struct list_head                  *phase_head;
1199         int                                rc    = 0;
1200         int                                rc1   = 0;
1201         ENTRY;
1202
1203         set = ptlrpc_prep_set();
1204         if (set == NULL)
1205                 RETURN(-ENOMEM);
1206
1207         lad->lad_touch_gen++;
1208         memset(lr, 0, sizeof(*lr));
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 instances 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);
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                 lr->lr_speed = bk->lb_speed_limit;
1338                 lr->lr_version = bk->lb_version;
1339                 lr->lr_param |= bk->lb_param;
1340                 lr->lr_async_windows = bk->lb_async_windows;
1341                 lr->lr_flags = LEF_TO_OST;
1342
1343                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1344                 ltds = &lfsck->li_ost_descs;
1345                 laia->laia_ltds = ltds;
1346                 down_read(&ltds->ltd_rw_sem);
1347                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1348                         ltd = lfsck_tgt_get(ltds, idx);
1349                         LASSERT(ltd != NULL);
1350
1351                         laia->laia_ltd = ltd;
1352                         ltd->ltd_layout_done = 0;
1353                         ltd->ltd_synced_failures = 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 = lfsck_ltd2tgt(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 = lfsck_ltd2tgt(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_synced_failures)
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         int                                rc2;
1618         ENTRY;
1619
1620         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1621                lfsck_lfsck2name(lfsck), lad->lad_name);
1622
1623         memset(lr, 0, sizeof(*lr));
1624         lr->lr_event = LE_START;
1625         if (pos->lp_oit_cookie <= 1)
1626                 lr->lr_param = LPF_RESET;
1627         rc = lfsck_assistant_notify_others(env, com, lr);
1628         if (rc != 0) {
1629                 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1630                        "to start %s: rc = %d\n",
1631                        lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1632                 GOTO(fini, rc);
1633         }
1634
1635         spin_lock(&lad->lad_lock);
1636         thread_set_flags(athread, SVC_RUNNING);
1637         spin_unlock(&lad->lad_lock);
1638         wake_up_all(&mthread->t_ctl_waitq);
1639
1640         while (1) {
1641                 while (!list_empty(&lad->lad_req_list)) {
1642                         bool wakeup = false;
1643
1644                         if (unlikely(lad->lad_exit ||
1645                                      !thread_is_running(mthread)))
1646                                 GOTO(cleanup1, rc = lad->lad_post_result);
1647
1648                         lar = list_entry(lad->lad_req_list.next,
1649                                          struct lfsck_assistant_req,
1650                                          lar_list);
1651                         /* Only the lfsck_assistant_engine thread itself can
1652                          * remove the "lar" from the head of the list, LFSCK
1653                          * engine thread only inserts other new "lar" at the
1654                          * end of the list. So it is safe to handle current
1655                          * "lar" without the spin_lock. */
1656                         rc = lao->la_handler_p1(env, com, lar);
1657                         spin_lock(&lad->lad_lock);
1658                         list_del_init(&lar->lar_list);
1659                         lad->lad_prefetched--;
1660                         /* Wake up the main engine thread only when the list
1661                          * is empty or half of the prefetched items have been
1662                          * handled to avoid too frequent thread schedule. */
1663                         if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1664                                 wakeup = true;
1665                         spin_unlock(&lad->lad_lock);
1666                         if (wakeup)
1667                                 wake_up_all(&mthread->t_ctl_waitq);
1668
1669                         lao->la_req_fini(env, lar);
1670                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1671                                 GOTO(cleanup1, rc);
1672                 }
1673
1674                 l_wait_event(athread->t_ctl_waitq,
1675                              !lfsck_assistant_req_empty(lad) ||
1676                              lad->lad_exit ||
1677                              lad->lad_to_post ||
1678                              lad->lad_to_double_scan,
1679                              &lwi);
1680
1681                 if (unlikely(lad->lad_exit))
1682                         GOTO(cleanup1, rc = lad->lad_post_result);
1683
1684                 if (!list_empty(&lad->lad_req_list))
1685                         continue;
1686
1687                 if (lad->lad_to_post) {
1688                         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1689                                lfsck_lfsck2name(lfsck), lad->lad_name);
1690
1691                         if (unlikely(lad->lad_exit))
1692                                 GOTO(cleanup1, rc = lad->lad_post_result);
1693
1694                         lad->lad_to_post = 0;
1695                         LASSERT(lad->lad_post_result > 0);
1696
1697                         /* Wakeup the master engine to go ahead. */
1698                         wake_up_all(&mthread->t_ctl_waitq);
1699
1700                         memset(lr, 0, sizeof(*lr));
1701                         lr->lr_event = LE_PHASE1_DONE;
1702                         lr->lr_status = lad->lad_post_result;
1703                         rc = lfsck_assistant_notify_others(env, com, lr);
1704
1705                         CDEBUG(D_LFSCK, "%s: LFSCK assistant notified "
1706                                "others for %s post: rc = %d\n",
1707                                lfsck_lfsck2name(lfsck),
1708                                lad->lad_name, rc);
1709                 }
1710
1711                 if (lad->lad_to_double_scan) {
1712                         lad->lad_to_double_scan = 0;
1713                         atomic_inc(&lfsck->li_double_scan_count);
1714                         lad->lad_in_double_scan = 1;
1715                         wake_up_all(&mthread->t_ctl_waitq);
1716
1717                         com->lc_new_checked = 0;
1718                         com->lc_new_scanned = 0;
1719                         com->lc_time_last_checkpoint = cfs_time_current();
1720                         com->lc_time_next_checkpoint =
1721                                 com->lc_time_last_checkpoint +
1722                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1723
1724                         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1725                                "the second-stage scaning\n",
1726                                lfsck_lfsck2name(lfsck));
1727
1728                         /* Flush async updates before handling orphan. */
1729                         rc2 = dt_sync(env, lfsck->li_next);
1730
1731                         CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1732                                "scan start, synced: rc = %d\n",
1733                                lfsck_lfsck2name(lfsck), rc2);
1734
1735                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1736                                 GOTO(cleanup2, rc = 0);
1737
1738                         while (lad->lad_in_double_scan) {
1739                                 rc = lfsck_assistant_query_others(env, com);
1740                                 if (lfsck_phase2_next_ready(lad))
1741                                         goto p2_next;
1742
1743                                 if (rc < 0)
1744                                         GOTO(cleanup2, rc);
1745
1746                                 /* Pull LFSCK status on related targets once
1747                                  * per 30 seconds if we are not notified. */
1748                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1749                                                            cfs_time_seconds(1),
1750                                                            NULL, NULL);
1751                                 rc = l_wait_event(athread->t_ctl_waitq,
1752                                         lfsck_phase2_next_ready(lad) ||
1753                                         lad->lad_exit ||
1754                                         !thread_is_running(mthread),
1755                                         &lwi);
1756
1757                                 if (unlikely(lad->lad_exit ||
1758                                              !thread_is_running(mthread)))
1759                                         GOTO(cleanup2, rc = 0);
1760
1761                                 if (rc == -ETIMEDOUT)
1762                                         continue;
1763
1764                                 if (rc < 0)
1765                                         GOTO(cleanup2, rc);
1766
1767 p2_next:
1768                                 rc = lao->la_handler_p2(env, com);
1769                                 if (rc != 0)
1770                                         GOTO(cleanup2, rc);
1771
1772                                 if (unlikely(lad->lad_exit ||
1773                                              !thread_is_running(mthread)))
1774                                         GOTO(cleanup2, rc = 0);
1775                         }
1776                 }
1777         }
1778
1779 cleanup1:
1780         /* Cleanup the unfinished requests. */
1781         spin_lock(&lad->lad_lock);
1782         if (rc < 0)
1783                 lad->lad_assistant_status = rc;
1784
1785         if (lad->lad_exit && lad->lad_post_result <= 0)
1786                 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1787
1788         while (!list_empty(&lad->lad_req_list)) {
1789                 lar = list_entry(lad->lad_req_list.next,
1790                                  struct lfsck_assistant_req,
1791                                  lar_list);
1792                 list_del_init(&lar->lar_list);
1793                 lad->lad_prefetched--;
1794                 spin_unlock(&lad->lad_lock);
1795                 lao->la_req_fini(env, lar);
1796                 spin_lock(&lad->lad_lock);
1797         }
1798         spin_unlock(&lad->lad_lock);
1799
1800         LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1801                  lad->lad_prefetched);
1802
1803 cleanup2:
1804         memset(lr, 0, sizeof(*lr));
1805         if (rc > 0) {
1806                 lr->lr_event = LE_PHASE2_DONE;
1807                 lr->lr_status = rc;
1808         } else if (rc == 0) {
1809                 if (lfsck->li_flags & LPF_ALL_TGT) {
1810                         lr->lr_event = LE_STOP;
1811                         lr->lr_status = LS_STOPPED;
1812                 } else {
1813                         lr->lr_event = LE_PEER_EXIT;
1814                         switch (lfsck->li_status) {
1815                         case LS_PAUSED:
1816                         case LS_CO_PAUSED:
1817                                 lr->lr_status = LS_CO_PAUSED;
1818                                 break;
1819                         case LS_STOPPED:
1820                         case LS_CO_STOPPED:
1821                                 lr->lr_status = LS_CO_STOPPED;
1822                                 break;
1823                         default:
1824                                 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1825                                        "status: rc = %d\n",
1826                                        lfsck_lfsck2name(lfsck),
1827                                        lfsck->li_status);
1828                                 lr->lr_status = LS_CO_FAILED;
1829                                 break;
1830                         }
1831                 }
1832         } else {
1833                 if (lfsck->li_flags & LPF_ALL_TGT) {
1834                         lr->lr_event = LE_STOP;
1835                         lr->lr_status = LS_FAILED;
1836                 } else {
1837                         lr->lr_event = LE_PEER_EXIT;
1838                         lr->lr_status = LS_CO_FAILED;
1839                 }
1840         }
1841
1842         rc1 = lfsck_assistant_notify_others(env, com, lr);
1843         if (rc1 != 0) {
1844                 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1845                        "others for %s quit: rc = %d\n",
1846                        lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1847                 rc = rc1;
1848         }
1849
1850         CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1851                lfsck_lfsck2name(lfsck));
1852
1853         /* Flush async updates before exit. */
1854         rc2 = dt_sync(env, lfsck->li_next);
1855
1856         CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1857                lfsck_lfsck2name(lfsck), rc2);
1858
1859         /* Under force exit case, some requests may be just freed without
1860          * verification, those objects should be re-handled when next run.
1861          * So not update the on-disk trace file under such case. */
1862         if (lad->lad_in_double_scan) {
1863                 if (!lad->lad_exit)
1864                         rc1 = lao->la_double_scan_result(env, com, rc);
1865
1866                 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1867                        "finished: rc = %d\n",
1868                        lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1869         }
1870
1871 fini:
1872         if (lad->lad_in_double_scan)
1873                 atomic_dec(&lfsck->li_double_scan_count);
1874
1875         spin_lock(&lad->lad_lock);
1876         lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1877         thread_set_flags(athread, SVC_STOPPED);
1878         wake_up_all(&mthread->t_ctl_waitq);
1879         spin_unlock(&lad->lad_lock);
1880
1881         CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1882                lfsck_lfsck2name(lfsck), lad->lad_name,
1883                lad->lad_assistant_status);
1884
1885         lfsck_thread_args_fini(lta);
1886
1887         return rc;
1888 }